Aren
/
NetBSD
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
NetBSD/lib/libperfuse/ops.c

3634 lines
88 KiB
C
Raw Normal View History

rmdir dir/.. must return an error. Use ENOTEMPRY like FFS does.
2014-09-05 19:20:16 +04:00
/* $NetBSD: ops.c,v 1.73 2014/09/05 15:20:16 manu Exp $ */
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*-
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
* Copyright (c) 2010-2011 Emmanuel Dreyfus. All rights reserved.
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
* ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
* TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <libgen.h>
#include <errno.h>
#include <err.h>
#include <sysexits.h>
#include <syslog.h>
#include <puffs.h>
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
#include <sys/socket.h>
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#include <sys/socket.h>
Add support for extended attributes
2011-06-28 20:19:16 +04:00
#include <sys/extattr.h>
Add a FUSE trace facility, with statistics collection. This should help tracking bugs and performance issues
2011-12-28 21:33:52 +04:00
#include <sys/time.h>
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#include <machine/vmparam.h>
#include "perfuse_priv.h"
#include "fuse.h"
- Do not checkfor peer credentials when perfused is autostarted and therefore runs with filesystem privileges - shut up warnings and debug messages when perfused is autostarted - make perfused patch modifiable with CFLAGS for easier pkgsrc integration - Fix build warnings
2010-09-07 06:11:04 +04:00
extern int perfuse_diagflags;
- Add missing ENOENT or ESTALL when accessing deleted node - Fix a warning, fix style (80 chars for a line)
2011-11-28 09:33:33 +04:00
#if 0
static void print_node(const char *, puffs_cookie_t);
#endif
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
#ifdef PUFFS_KFLAG_CACHE_FS_TTL
static void perfuse_newinfo_setttl(struct puffs_newinfo *,
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
struct puffs_node *, struct fuse_entry_out *, struct fuse_attr_out *);
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
#endif /* PUFFS_KFLAG_CACHE_FS_TTL */
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
static int xchg_msg(struct puffs_usermount *, puffs_cookie_t,
perfuse_msg_t *, size_t, enum perfuse_xchg_pb_reply);
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
static int mode_access(puffs_cookie_t, const struct puffs_cred *, mode_t);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
static int sticky_access(puffs_cookie_t, struct puffs_node *,
const struct puffs_cred *);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
static void fuse_attr_to_vap(struct perfuse_state *,
struct vattr *, struct fuse_attr *);
static int node_lookup_common(struct puffs_usermount *, puffs_cookie_t,
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
struct puffs_newinfo *, const char *, const struct puffs_cred *,
struct puffs_node **);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
static int node_mk_common(struct puffs_usermount *, puffs_cookie_t,
- only remove queued requests once they are executed, not when they are set to be scheduled later - remove an unused argument to make lint happy
2010-09-02 12:58:06 +04:00
struct puffs_newinfo *, const struct puffs_cn *pcn, perfuse_msg_t *);
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
static uint64_t readdir_last_cookie(struct fuse_dirent *, size_t);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
static ssize_t fuse_to_dirent(struct puffs_usermount *, puffs_cookie_t,
struct fuse_dirent *, size_t);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
static void readdir_buffered(puffs_cookie_t, struct dirent *, off_t *,
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
size_t *);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
static void node_ref(puffs_cookie_t);
static void node_rele(puffs_cookie_t);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
static void requeue_request(struct puffs_usermount *,
puffs_cookie_t opc, enum perfuse_qtype);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
static int dequeue_requests(puffs_cookie_t opc, enum perfuse_qtype, int);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#define DEQUEUE_ALL 0
/*
* From <sys/vnode>, inside #ifdef _KERNEL section
*/
#define IO_SYNC (0x40|IO_DSYNC)
#define IO_DSYNC 0x00200
#define IO_DIRECT 0x02000
/*
* From <fcntl>, inside #ifdef _KERNEL section
*/
#define F_WAIT 0x010
#define F_FLOCK 0x020
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
#define OFLAGS(fflags) ((fflags) - 1)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* Borrowed from src/sys/kern/vfs_subr.c and src/sys/sys/vnode.h
*/
const enum vtype iftovt_tab[16] = {
VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
};
const int vttoif_tab[9] = {
0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
S_IFSOCK, S_IFIFO, S_IFMT,
};
#define IFTOVT(mode) (iftovt_tab[((mode) & S_IFMT) >> 12])
#define VTTOIF(indx) (vttoif_tab[(int)(indx)])
- Add missing ENOENT or ESTALL when accessing deleted node - Fix a warning, fix style (80 chars for a line)
2011-11-28 09:33:33 +04:00
#if 0
static void
Use C89 function definition
2012-03-21 14:10:36 +04:00
print_node(const char *func, puffs_cookie_t opc)
- Add missing ENOENT or ESTALL when accessing deleted node - Fix a warning, fix style (80 chars for a line)
2011-11-28 09:33:33 +04:00
{
struct puffs_node *pn;
struct perfuse_node_data *pnd;
struct vattr *vap;
pn = (struct puffs_node *)opc;
pnd = PERFUSE_NODE_DATA(opc);
vap = &pn->pn_va;
printf("%s: \"%s\", opc = %p, nodeid = 0x%"PRIx64" ino = %"PRIu64"\n",
func, pnd->pnd_name, opc, pnd->pnd_nodeid, vap->va_fileid);
return;
}
#endif /* PERFUSE_DEBUG */
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_close_common(struct puffs_usermount *pu, puffs_cookie_t opc,
int mode)
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
{
struct perfuse_state *ps;
perfuse_msg_t *pm;
int op;
uint64_t fh;
struct fuse_release_in *fri;
struct perfuse_node_data *pnd;
struct puffs_node *pn;
int error;
ps = puffs_getspecific(pu);
pn = (struct puffs_node *)opc;
pnd = PERFUSE_NODE_DATA(pn);
if (puffs_pn_getvap(pn)->va_type == VDIR) {
op = FUSE_RELEASEDIR;
mode = FREAD;
} else {
op = FUSE_RELEASE;
}
/*
* Destroy the filehandle before sending the
* request to the FUSE filesystem, otherwise
* we may get a second close() while we wait
* for the reply, and we would end up closing
* the same fh twice instead of closng both.
*/
fh = perfuse_get_fh(opc, mode);
perfuse_destroy_fh(pn, fh);
/*
* release_flags may be set to FUSE_RELEASE_FLUSH
* to flush locks. lock_owner must be set in that case
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
*
* ps_new_msg() is called with NULL creds, which will
* be interpreted as FUSE superuser. We come here from the
* inactive method, which provides no creds, but obviously
* runs with kernel privilege.
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
*/
pm = ps->ps_new_msg(pu, opc, op, sizeof(*fri), NULL);
fri = GET_INPAYLOAD(ps, pm, fuse_release_in);
fri->fh = fh;
fri->flags = 0;
fri->release_flags = 0;
fri->lock_owner = pnd->pnd_lock_owner;
fri->flags = (fri->lock_owner != 0) ? FUSE_RELEASE_FLUSH : 0;
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FH)
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
__func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
#endif
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm,
NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
DERRX(EX_SOFTWARE, "%s: freed fh = 0x%"PRIx64" but filesystem "
"returned error = %d", __func__, fh, error);
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
ps->ps_destroy_msg(pm);
return 0;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
static int
Use C89 function definition
2012-03-21 14:10:36 +04:00
xchg_msg(struct puffs_usermount *pu, puffs_cookie_t opc, perfuse_msg_t *pm,
size_t len, enum perfuse_xchg_pb_reply wait)
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
{
struct perfuse_state *ps;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
struct perfuse_node_data *pnd;
Add a FUSE trace facility, with statistics collection. This should help tracking bugs and performance issues
2011-12-28 21:33:52 +04:00
struct perfuse_trace *pt = NULL;
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
int error;
ps = puffs_getspecific(pu);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
pnd = NULL;
if ((struct puffs_node *)opc != NULL)
pnd = PERFUSE_NODE_DATA(opc);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
#ifdef PERFUSE_DEBUG
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
if ((perfuse_diagflags & PDF_FILENAME) && (opc != 0))
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
DPRINTF("file = \"%s\", ino = %"PRIu64" flags = 0x%x\n",
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_node_path(ps, opc),
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
((struct puffs_node *)opc)->pn_va.va_fileid,
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
PERFUSE_NODE_DATA(opc)->pnd_flags);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
#endif
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
ps->ps_xchgcount++;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (pnd)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
pnd->pnd_inxchg++;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
Add a FUSE trace facility, with statistics collection. This should help tracking bugs and performance issues
2011-12-28 21:33:52 +04:00
/*
* Record FUSE call start if requested
*/
Improve FUSE trace facility
2012-01-29 10:22:01 +04:00
if (perfuse_diagflags & PDF_TRACE)
pt = perfuse_trace_begin(ps, opc, pm);
Add a FUSE trace facility, with statistics collection. This should help tracking bugs and performance issues
2011-12-28 21:33:52 +04:00
/*
* Do actual FUSE exchange
*/
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
if ((error = ps->ps_xchg_msg(pu, pm, len, wait)) != 0)
ps->ps_destroy_msg(pm);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
Add a FUSE trace facility, with statistics collection. This should help tracking bugs and performance issues
2011-12-28 21:33:52 +04:00
/*
* Record FUSE call end if requested
*/
Improve FUSE trace facility
2012-01-29 10:22:01 +04:00
if (pt != NULL)
perfuse_trace_end(ps, pt, error);
Add a FUSE trace facility, with statistics collection. This should help tracking bugs and performance issues
2011-12-28 21:33:52 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
ps->ps_xchgcount--;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (pnd) {
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
pnd->pnd_inxchg--;
(void)dequeue_requests(opc, PCQ_AFTERXCHG, DEQUEUE_ALL);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
}
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
return error;
}
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
static int
Use C89 function definition
2012-03-21 14:10:36 +04:00
mode_access(puffs_cookie_t opc, const struct puffs_cred *pcr, mode_t mode)
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
{
struct puffs_node *pn;
struct vattr *va;
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
/*
* pcr is NULL for self open through fsync or readdir.
* In both case, access control is useless, as it was
* done before, at open time.
*/
if (pcr == NULL)
return 0;
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
pn = (struct puffs_node *)opc;
va = puffs_pn_getvap(pn);
return puffs_access(va->va_type, va->va_mode,
va->va_uid, va->va_gid,
mode, pcr);
}
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
static int
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
sticky_access(puffs_cookie_t opc, struct puffs_node *targ,
const struct puffs_cred *pcr)
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
{
uid_t uid;
Improve POSIX compliance of FUSE filesystems through PERUSE - access denied is EPERM and not EACCES - access to file owned by someone else in a sticy-bit directory should be allowed for the sticy-bit directory owner - setting sticky-bit on a non directory should produce EFTYPE - implement PATHCONF method as much as we can.
2014-09-03 20:01:45 +04:00
int sticky, owner, parent_owner;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
/*
* This covers the case where the kernel requests a DELETE
* or RENAME on its own, and where puffs_cred_getuid would
* return -1. While such a situation should not happen,
* we allow it here.
*
* This also allows root to tamper with other users' files
* that have the sticky bit.
*/
if (puffs_cred_isjuggernaut(pcr))
return 0;
if (puffs_cred_getuid(pcr, &uid) != 0)
DERRX(EX_SOFTWARE, "puffs_cred_getuid fails in %s", __func__);
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
sticky = puffs_pn_getvap(opc)->va_mode & S_ISTXT;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
owner = puffs_pn_getvap(targ)->va_uid == uid;
Improve POSIX compliance of FUSE filesystems through PERUSE - access denied is EPERM and not EACCES - access to file owned by someone else in a sticy-bit directory should be allowed for the sticy-bit directory owner - setting sticky-bit on a non directory should produce EFTYPE - implement PATHCONF method as much as we can.
2014-09-03 20:01:45 +04:00
parent_owner = puffs_pn_getvap(opc)->va_uid == uid;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
Improve POSIX compliance of FUSE filesystems through PERUSE - access denied is EPERM and not EACCES - access to file owned by someone else in a sticy-bit directory should be allowed for the sticy-bit directory owner - setting sticky-bit on a non directory should produce EFTYPE - implement PATHCONF method as much as we can.
2014-09-03 20:01:45 +04:00
if (sticky && !owner && !parent_owner)
return EPERM;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
Fix uninitiaized variable usage (never though lint would miss that when used by return statement) that caused unprivilegied user to fail on unlink(2) and rename(2) operations.
2011-08-09 13:06:52 +04:00
return 0;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
static void
Use C89 function definition
2012-03-21 14:10:36 +04:00
fuse_attr_to_vap(struct perfuse_state *ps, struct vattr *vap,
struct fuse_attr *fa)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
vap->va_type = IFTOVT(fa->mode);
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
vap->va_mode = fa->mode & ALLPERMS;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
vap->va_nlink = fa->nlink;
vap->va_uid = fa->uid;
vap->va_gid = fa->gid;
- Proper permission checks when doing directory traversal. e.g.: run rm dir/file while dir was never looked up since the mount. In that situation, we get lookup with pcn_nameiop NAMEI_DELETE for dir before we get it for file. But for dir we are just looking for PUFFS_VEXEC. This is solved by honouring NAMEI_ISLASTCN, which is set for the last element only - do not send O_EXCL to FUSE as documentation forbids it. - fix warning
2011-05-18 19:28:12 +04:00
vap->va_fsid = (long)ps->ps_fsid;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
vap->va_fileid = fa->ino;
vap->va_size = fa->size;
vap->va_blocksize = fa->blksize;
Build fixes for LP64
2010-09-01 18:57:24 +04:00
vap->va_atime.tv_sec = (time_t)fa->atime;
vap->va_atime.tv_nsec = (long) fa->atimensec;
vap->va_mtime.tv_sec = (time_t)fa->mtime;
vap->va_mtime.tv_nsec = (long)fa->mtimensec;
vap->va_ctime.tv_sec = (time_t)fa->ctime;
vap->va_ctime.tv_nsec = (long)fa->ctimensec;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
vap->va_birthtime.tv_sec = 0;
vap->va_birthtime.tv_nsec = 0;
vap->va_gen = 0;
vap->va_flags = 0;
vap->va_rdev = fa->rdev;
vap->va_bytes = fa->size;
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
vap->va_filerev = (u_quad_t)PUFFS_VNOVAL;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
vap->va_vaflags = 0;
if (vap->va_blocksize == 0)
vap->va_blocksize = DEV_BSIZE;
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
if (vap->va_size == (size_t)PUFFS_VNOVAL) /* XXX */
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
vap->va_size = 0;
return;
}
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
#ifdef PUFFS_KFLAG_CACHE_FS_TTL
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
static void
perfuse_newinfo_setttl(struct puffs_newinfo *pni,
struct puffs_node *pn, struct fuse_entry_out *feo,
struct fuse_attr_out *fao)
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
{
#ifdef PERFUSE_DEBUG
if ((feo == NULL) && (fao == NULL))
DERRX(EX_SOFTWARE, "%s: feo and fao NULL", __func__);
if ((feo != NULL) && (fao != NULL))
DERRX(EX_SOFTWARE, "%s: feo and fao != NULL", __func__);
#endif /* PERFUSE_DEBUG */
if (fao != NULL) {
struct timespec va_ttl;
va_ttl.tv_sec = fao->attr_valid;
va_ttl.tv_nsec = fao->attr_valid_nsec;
puffs_newinfo_setvattl(pni, &va_ttl);
}
if (feo != NULL) {
struct timespec va_ttl;
struct timespec cn_ttl;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
struct timespec now;
struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(pn);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
va_ttl.tv_sec = feo->attr_valid;
va_ttl.tv_nsec = feo->attr_valid_nsec;
cn_ttl.tv_sec = feo->entry_valid;
cn_ttl.tv_nsec = feo->entry_valid_nsec;
puffs_newinfo_setvattl(pni, &va_ttl);
puffs_newinfo_setcnttl(pni, &cn_ttl);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (clock_gettime(CLOCK_REALTIME, &now) != 0)
DERR(EX_OSERR, "clock_gettime failed");
timespecadd(&now, &cn_ttl, &pnd->pnd_cn_expire);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
}
return;
}
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
#endif /* PUFFS_KFLAG_CACHE_FS_TTL */
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
static int
Use C89 function definition
2012-03-21 14:10:36 +04:00
node_lookup_common(struct puffs_usermount *pu, puffs_cookie_t opc,
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
struct puffs_newinfo *pni, const char *path,
const struct puffs_cred *pcr, struct puffs_node **pnp)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
struct perfuse_node_data *oldpnd;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
perfuse_msg_t *pm;
struct fuse_entry_out *feo;
struct puffs_node *pn;
size_t len;
int error;
- Add missing ENOENT or ESTALL when accessing deleted node - Fix a warning, fix style (80 chars for a line)
2011-11-28 09:33:33 +04:00
/*
* Prevent further lookups if the parent was removed
*/
if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
return ESTALE;
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
if (pnp == NULL)
DERRX(EX_SOFTWARE, "pnp must be != NULL");
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FILENAME)
DPRINTF("%s: opc = %p, file = \"%s\" looking up \"%s\"\n",
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
__func__, (void *)opc,
perfuse_node_path(ps, opc), path);
if (strcmp(path, ".") == 0)
DERRX(EX_SOFTWARE, "unexpected dot-lookup");
if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_RECLAIMED)
DERRX(EX_SOFTWARE,
"looking up reclaimed node opc = %p, name = \"%s\"",
opc, path);
if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_INVALID)
DERRX(EX_SOFTWARE,
"looking up freed node opc = %p, name = \"%s\"",
opc, path);
#endif /* PERFUSE_DEBUG */
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
len = strlen(path) + 1;
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_LOOKUP, len, pcr);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
(void)strlcpy(_GET_INPAYLOAD(ps, pm, char *), path, len);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
When lookup returns a node with null inode number, it means the ENOENT, with negative caching. We do not implement negative caching yet, but we honour the ENOENT.
2012-11-03 19:43:20 +04:00
/*
* Starting with ABI 7.4, inode number 0 means ENOENT,
* with entry_valid / entry_valid_nsec giving negative
* cache timeout (which we do not implement yet).
*/
if (feo->attr.ino == 0) {
ps->ps_destroy_msg(pm);
return ENOENT;
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* Check for a known node, not reclaimed, with another name.
* It may have been moved, or we can lookup ../
*/
if (((oldpnd = perfuse_node_bynodeid(ps, feo->nodeid)) != NULL) &&
!(oldpnd->pnd_flags & PND_RECLAIMED)) {
/*
* Save the new node name if not ..
*/
if (strncmp(path, "..", len) != 0)
(void)strlcpy(oldpnd->pnd_name,
path, MAXPATHLEN);
pn = oldpnd->pnd_pn;
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
} else {
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
pn = perfuse_new_pn(pu, path, opc);
PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_node_cache(ps, pn);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
#ifdef PERFUSE_DEBUG
if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_RECLAIMED)
DERRX(EX_SOFTWARE,
"reclaimed in lookup opc = %p, name = \"%s\", ck = %p",
opc, path, pn);
if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_INVALID)
DERRX(EX_SOFTWARE,
"freed in lookup opc = %p, name = \"%s\", ck = %p",
opc, path, pn);
#endif /* PERFUSE_DEBUG */
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
pn->pn_va.va_gen = (u_long)(feo->generation);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
PERFUSE_NODE_DATA(pn)->pnd_fuse_nlookup++;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
*pnp = pn;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FILENAME)
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
DPRINTF("%s: opc = %p, looked up opc = %p, "
"nodeid = 0x%"PRIx64" file = \"%s\"\n", __func__,
(void *)opc, pn, feo->nodeid, path);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#endif
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
if (pni != NULL) {
#ifdef PUFFS_KFLAG_CACHE_FS_TTL
puffs_newinfo_setva(pni, &pn->pn_va);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_newinfo_setttl(pni, pn, feo, NULL);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
#endif /* PUFFS_KFLAG_CACHE_FS_TTL */
puffs_newinfo_setcookie(pni, pn);
puffs_newinfo_setvtype(pni, pn->pn_va.va_type);
puffs_newinfo_setsize(pni, (voff_t)pn->pn_va.va_size);
puffs_newinfo_setrdev(pni, pn->pn_va.va_rdev);
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_NODELEAK) {
PERFUSE_NODE_DATA(pn)->pnd_flags &= ~PND_NODELEAK;
ps->ps_nodeleakcount--;
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps->ps_destroy_msg(pm);
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
return 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
/*
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
* Common code for methods that create objects:
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
* perfuse_node_mkdir
* perfuse_node_mknod
* perfuse_node_symlink
*/
static int
Use C89 function definition
2012-03-21 14:10:36 +04:00
node_mk_common(struct puffs_usermount *pu, puffs_cookie_t opc,
struct puffs_newinfo *pni, const struct puffs_cn *pcn,
perfuse_msg_t *pm)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
struct puffs_node *pn;
struct fuse_entry_out *feo;
int error;
ps = puffs_getspecific(pu);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
pn = perfuse_new_pn(pu, pcn->pcn_name, opc);
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
perfuse_node_cache(ps, pn);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
pn->pn_va.va_gen = (u_long)(feo->generation);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
puffs_newinfo_setcookie(pni, pn);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
#ifdef PUFFS_KFLAG_CACHE_FS_TTL
puffs_newinfo_setva(pni, &pn->pn_va);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_newinfo_setttl(pni, pn, feo, NULL);
#endif /* PUFFS_KFLAG_CACHE_FS_TTL */
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FILENAME)
DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
"nodeid = 0x%"PRIx64"\n",
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
__func__, (void *)pn, pcn->pcn_name,
PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid);
#endif
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
ps->ps_destroy_msg(pm);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
/* Parents is now dirty */
- call FSYNCDIR for directories - directories can be open R/W (for FSYNCDIR) - do not skip calls to FSYNC or FSYNCDIR if the filesystem returned ENOSYS: it may change its mind, and it may also actually do something when retunring ENOSYS - When FSYNC and FSYNCDIR return ENOSYS, do not report it to kernel (silent failure)
2010-09-09 13:12:35 +04:00
PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
return 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
static uint64_t
Use C89 function definition
2012-03-21 14:10:36 +04:00
readdir_last_cookie(struct fuse_dirent *fd, size_t fd_len)
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
{
size_t len;
size_t seen = 0;
char *ndp;
do {
len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
seen += len;
if (seen >= fd_len)
break;
ndp = (char *)(void *)fd + (size_t)len;
fd = (struct fuse_dirent *)(void *)ndp;
} while (1 /* CONSTCOND */);
return fd->off;
}
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
static ssize_t
Use C89 function definition
2012-03-21 14:10:36 +04:00
fuse_to_dirent(struct puffs_usermount *pu, puffs_cookie_t opc,
struct fuse_dirent *fd, size_t fd_len)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct dirent *dents;
size_t dents_len;
ssize_t written;
uint64_t fd_offset;
struct fuse_dirent *fd_base;
size_t len;
fd_base = fd;
fd_offset = 0;
written = 0;
dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
build fixes for LP64
2010-09-06 05:17:05 +04:00
dents_len = (size_t)PERFUSE_NODE_DATA(opc)->pnd_dirent_len;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
do {
char *ndp;
size_t reclen;
reclen = _DIRENT_RECLEN(dents, fd->namelen);
/*
* Check we do not overflow the output buffer
* struct fuse_dirent is bigger than struct dirent,
* so we should always use fd_len and never reallocate
* later.
* If we have to reallocate,try to double the buffer
* each time so that we do not have to do it too often.
*/
if (written + reclen > dents_len) {
if (dents_len == 0)
dents_len = fd_len;
else
dents_len =
MAX(2 * dents_len, written + reclen);
dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
if ((dents = realloc(dents, dents_len)) == NULL)
- fix warn/err confusiog - fix debugging printf - add func arguments to simple formats
2011-08-14 03:12:15 +04:00
DERR(EX_OSERR, "%s: malloc failed", __func__);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
PERFUSE_NODE_DATA(opc)->pnd_dirent = dents;
PERFUSE_NODE_DATA(opc)->pnd_dirent_len = dents_len;
/*
* (void *) for delint
*/
ndp = (char *)(void *)dents + written;
dents = (struct dirent *)(void *)ndp;
}
/*
* Filesystem was mounted without -o use_ino
* Perform a lookup to find it.
*/
if (fd->ino == PERFUSE_UNKNOWN_INO) {
struct puffs_node *pn;
For filesystems mounted without -o use_ino, readdir is not able to fetch inode number. We perfom an addtional lookup on each file to get it. In that case, do not lookup .. from root, as it breaks out of the filesystem and hits NULL pointers.
2014-01-06 12:56:34 +04:00
struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
For filesystems mounted without -o use_ino, readdir is not able to fetch inode number. We perfom an addtional lookup on each file to get it. In that case, do not lookup .. from root, as it breaks out of the filesystem and hits NULL pointers.
2014-01-06 12:56:34 +04:00
/*
* Avoid breaking out of fs
* by lookup to .. on root
*/
if ((strcmp(fd->name, "..") == 0) &&
(pnd->pnd_nodeid == FUSE_ROOT_ID)) {
fd->ino = FUSE_ROOT_ID;
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
} else {
For filesystems mounted without -o use_ino, readdir is not able to fetch inode number. We perfom an addtional lookup on each file to get it. In that case, do not lookup .. from root, as it breaks out of the filesystem and hits NULL pointers.
2014-01-06 12:56:34 +04:00
if (node_lookup_common(pu, opc, NULL, fd->name,
NULL, &pn) != 0) {
DWARNX("node_lookup_common failed");
} else {
fd->ino = pn->pn_va.va_fileid;
(void)perfuse_node_reclaim(pu, pn);
}
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
dents->d_fileno = fd->ino;
Build fixes for LP64
2010-09-01 18:57:24 +04:00
dents->d_reclen = (unsigned short)reclen;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
dents->d_namlen = fd->namelen;
dents->d_type = fd->type;
strlcpy(dents->d_name, fd->name, fd->namelen + 1);
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_READDIR)
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
DPRINTF("%s: translated \"%s\" ino = %"PRIu64"\n",
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
__func__, dents->d_name, dents->d_fileno);
#endif
dents = _DIRENT_NEXT(dents);
written += reclen;
/*
* Move to the next record.
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
* fd->off is not the offset, it is an opaque cookie
* given by the filesystem to keep state across multiple
* readdir() operation.
* Use record alignement instead.
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_READDIR)
Build fixes for LP64
2010-09-01 18:57:24 +04:00
DPRINTF("%s: record at %"PRId64"/0x%"PRIx64" "
"length = %zd/0x%zx. "
"next record at %"PRId64"/0x%"PRIx64" "
"max %zd/0x%zx\n",
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
__func__, fd_offset, fd_offset, len, len,
fd_offset + len, fd_offset + len,
fd_len, fd_len);
#endif
fd_offset += len;
/*
* Check if next record is still within the packet
* If it is not, we reached the end of the buffer.
*/
if (fd_offset >= fd_len)
break;
/*
* (void *) for delint
*/
ndp = (char *)(void *)fd_base + (size_t)fd_offset;
fd = (struct fuse_dirent *)(void *)ndp;
} while (1 /* CONSTCOND */);
/*
* Adjust the dirent output length
*/
if (written != -1)
PERFUSE_NODE_DATA(opc)->pnd_dirent_len = written;
return written;
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
static void
Use C89 function definition
2012-03-21 14:10:36 +04:00
readdir_buffered(puffs_cookie_t opc, struct dirent *dent, off_t *readoff,
size_t *reslen)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct dirent *fromdent;
struct perfuse_node_data *pnd;
char *ndp;
pnd = PERFUSE_NODE_DATA(opc);
while (*readoff < pnd->pnd_dirent_len) {
/*
* (void *) for delint
*/
ndp = (char *)(void *)pnd->pnd_dirent + (size_t)*readoff;
fromdent = (struct dirent *)(void *)ndp;
if (*reslen < _DIRENT_SIZE(fromdent))
break;
memcpy(dent, fromdent, _DIRENT_SIZE(fromdent));
*readoff += _DIRENT_SIZE(fromdent);
*reslen -= _DIRENT_SIZE(fromdent);
dent = _DIRENT_NEXT(dent);
}
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_READDIR)
build fixes for LP64
2010-09-06 05:17:05 +04:00
DPRINTF("%s: readoff = %"PRId64", "
"pnd->pnd_dirent_len = %"PRId64"\n",
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
__func__, *readoff, pnd->pnd_dirent_len);
#endif
if (*readoff >= pnd->pnd_dirent_len) {
free(pnd->pnd_dirent);
pnd->pnd_dirent = NULL;
pnd->pnd_dirent_len = 0;
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
return;
}
static void
node_ref(puffs_cookie_t opc)
{
struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
#ifdef PERFUSE_DEBUG
if (pnd->pnd_flags & PND_INVALID)
DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
#endif /* PERFUSE_DEBUG */
pnd->pnd_ref++;
return;
}
static void
node_rele(puffs_cookie_t opc)
{
struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
#ifdef PERFUSE_DEBUG
if (pnd->pnd_flags & PND_INVALID)
DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
#endif /* PERFUSE_DEBUG */
pnd->pnd_ref--;
if (pnd->pnd_ref == 0)
(void)dequeue_requests(opc, PCQ_REF, DEQUEUE_ALL);
return;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
static void
Use C89 function definition
2012-03-21 14:10:36 +04:00
requeue_request(struct puffs_usermount *pu, puffs_cookie_t opc,
enum perfuse_qtype type)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_cc_queue pcq;
struct perfuse_node_data *pnd;
#ifdef PERFUSE_DEBUG
struct perfuse_state *ps;
ps = perfuse_getspecific(pu);
#endif
pnd = PERFUSE_NODE_DATA(opc);
pcq.pcq_type = type;
pcq.pcq_cc = puffs_cc_getcc(pu);
TAILQ_INSERT_TAIL(&pnd->pnd_pcq, &pcq, pcq_next);
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_REQUEUE)
- Use SOCK_DGRAM instead of SOCK_STREAM, as the filesystem seems to assume datagram semantics: when using SOCK_STREAM, if perfused sends frames faster than the filesystem consumes them, it will grab multiple frames at once and discard anything beyond the first one. For now the code can work both with SOCK_DGRAM and SOCK_STREAM, but SOCK_STREAM support will probably have to be removed for the sake of readability. - Remeber to sync parent directories when moving a node - In debug output, display the requeue type (readdir, write, etc...)
2010-09-15 05:51:43 +04:00
DPRINTF("%s: REQUEUE opc = %p, pcc = %p (%s)\n",
__func__, (void *)opc, pcq.pcq_cc,
perfuse_qtypestr[type]);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#endif
puffs_cc_yield(pcq.pcq_cc);
- only remove queued requests once they are executed, not when they are set to be scheduled later - remove an unused argument to make lint happy
2010-09-02 12:58:06 +04:00
TAILQ_REMOVE(&pnd->pnd_pcq, &pcq, pcq_next);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_REQUEUE)
- Use SOCK_DGRAM instead of SOCK_STREAM, as the filesystem seems to assume datagram semantics: when using SOCK_STREAM, if perfused sends frames faster than the filesystem consumes them, it will grab multiple frames at once and discard anything beyond the first one. For now the code can work both with SOCK_DGRAM and SOCK_STREAM, but SOCK_STREAM support will probably have to be removed for the sake of readability. - Remeber to sync parent directories when moving a node - In debug output, display the requeue type (readdir, write, etc...)
2010-09-15 05:51:43 +04:00
DPRINTF("%s: RESUME opc = %p, pcc = %p (%s)\n",
__func__, (void *)opc, pcq.pcq_cc,
perfuse_qtypestr[type]);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#endif
return;
}
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
static int
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
dequeue_requests(puffs_cookie_t opc, enum perfuse_qtype type, int max)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_cc_queue *pcq;
struct perfuse_node_data *pnd;
int dequeued;
pnd = PERFUSE_NODE_DATA(opc);
dequeued = 0;
TAILQ_FOREACH(pcq, &pnd->pnd_pcq, pcq_next) {
if (pcq->pcq_type != type)
continue;
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_REQUEUE)
- Use SOCK_DGRAM instead of SOCK_STREAM, as the filesystem seems to assume datagram semantics: when using SOCK_STREAM, if perfused sends frames faster than the filesystem consumes them, it will grab multiple frames at once and discard anything beyond the first one. For now the code can work both with SOCK_DGRAM and SOCK_STREAM, but SOCK_STREAM support will probably have to be removed for the sake of readability. - Remeber to sync parent directories when moving a node - In debug output, display the requeue type (readdir, write, etc...)
2010-09-15 05:51:43 +04:00
DPRINTF("%s: SCHEDULE opc = %p, pcc = %p (%s)\n",
__func__, (void *)opc, pcq->pcq_cc,
perfuse_qtypestr[type]);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#endif
- only remove queued requests once they are executed, not when they are set to be scheduled later - remove an unused argument to make lint happy
2010-09-02 12:58:06 +04:00
puffs_cc_schedule(pcq->pcq_cc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
if (++dequeued == max)
break;
}
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_REQUEUE)
DPRINTF("%s: DONE opc = %p\n", __func__, (void *)opc);
#endif
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
return dequeued;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
void
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_fs_init(struct puffs_usermount *pu)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
perfuse_msg_t *pm;
struct fuse_init_in *fii;
struct fuse_init_out *fio;
int error;
ps = puffs_getspecific(pu);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
if (puffs_mount(pu, ps->ps_target, ps->ps_mountflags, ps->ps_root) != 0)
- fix warn/err confusiog - fix debugging printf - add func arguments to simple formats
2011-08-14 03:12:15 +04:00
DERR(EX_OSERR, "%s: puffs_mount failed", __func__);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* Linux 2.6.34.1 sends theses flags:
* FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC
* FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK
*
* Linux also sets max_readahead at 32 pages (128 kB)
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
*
* ps_new_msg() is called with NULL creds, which will
* be interpreted as FUSE superuser.
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
pm = ps->ps_new_msg(pu, 0, FUSE_INIT, sizeof(*fii), NULL);
fii = GET_INPAYLOAD(ps, pm, fuse_init_in);
fii->major = FUSE_KERNEL_VERSION;
fii->minor = FUSE_KERNEL_MINOR_VERSION;
make this build on amd64 and remove redundant and unused code.
2011-09-10 02:51:44 +04:00
fii->max_readahead = (unsigned int)(32 * sysconf(_SC_PAGESIZE));
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fii->flags = (FUSE_ASYNC_READ|FUSE_POSIX_LOCKS|FUSE_ATOMIC_O_TRUNC);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, 0, pm, sizeof(*fio), wait_reply)) != 0)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
DERRX(EX_SOFTWARE, "init message exchange failed (%d)", error);
fio = GET_OUTPAYLOAD(ps, pm, fuse_init_out);
ps->ps_max_readahead = fio->max_readahead;
ps->ps_max_write = fio->max_write;
ps->ps_destroy_msg(pm);
return;
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_fs_unmount(struct puffs_usermount *pu, int flags)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
perfuse_msg_t *pm;
struct perfuse_state *ps;
puffs_cookie_t opc;
int error;
ps = puffs_getspecific(pu);
opc = (puffs_cookie_t)puffs_getroot(pu);
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
/*
* ps_new_msg() is called with NULL creds, which will
* be interpreted as FUSE superuser.
*/
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_DESTROY, 0, NULL);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0){
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
DWARN("unmount %s", ps->ps_target);
if (!(flags & MNT_FORCE))
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
return error;
else
error = 0;
} else {
ps->ps_destroy_msg(pm);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
Use SOCK_SEQPACKET in perfuse if available. This fix file operations hangs where the FUSE filesyste replied to an operation and got an ENOBUFS it did not handle. We now are also able to cleanly unmount
2011-05-30 18:50:08 +04:00
ps->ps_umount(pu);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
Use SOCK_SEQPACKET in perfuse if available. This fix file operations hangs where the FUSE filesyste replied to an operation and got an ENOBUFS it did not handle. We now are also able to cleanly unmount
2011-05-30 18:50:08 +04:00
if (perfuse_diagflags & PDF_MISC)
DPRINTF("%s unmounted, exit\n", ps->ps_target);
return 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_fs_statvfs(struct puffs_usermount *pu, struct statvfs *svfsb)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
perfuse_msg_t *pm;
puffs_cookie_t opc;
struct fuse_statfs_out *fso;
int error;
ps = puffs_getspecific(pu);
opc = (puffs_cookie_t)puffs_getroot(pu);
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
/*
* ps_new_msg() is called with NULL creds, which will
* be interpreted as FUSE superuser.
*/
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply)) != 0)
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
svfsb->f_flag = ps->ps_mountflags;
svfsb->f_bsize = fso->st.bsize;
svfsb->f_frsize = fso->st.frsize;
svfsb->f_iosize = ((struct puffs_node *)opc)->pn_va.va_blocksize;
svfsb->f_blocks = fso->st.blocks;
svfsb->f_bfree = fso->st.bfree;
svfsb->f_bavail = fso->st.bavail;
svfsb->f_bresvd = fso->st.bfree - fso->st.bavail;
svfsb->f_files = fso->st.files;
svfsb->f_ffree = fso->st.ffree;
svfsb->f_favail = fso->st.ffree;/* files not reserved for root */
svfsb->f_fresvd = 0; /* files reserved for root */
svfsb->f_syncreads = ps->ps_syncreads;
svfsb->f_syncwrites = ps->ps_syncwrites;
svfsb->f_asyncreads = ps->ps_asyncreads;
svfsb->f_asyncwrites = ps->ps_asyncwrites;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
(void)memcpy(&svfsb->f_fsidx, &ps->ps_fsid, sizeof(ps->ps_fsid));
svfsb->f_fsid = (unsigned long)ps->ps_fsid;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
svfsb->f_namemax = MAXPATHLEN; /* XXX */
- if perfused is not already started (cannot connect to /dev/fuse), FUSE filesystems will attempt to start it on their own, and will communicate using a socketpair - do not advertise NULL file handle as being valid when sending themback to the FUSE filesystem. - unmount if we cannot talk to the FUSE process anymore - set calling process gid properly - debug message cleanup
2010-08-27 13:58:17 +04:00
svfsb->f_owner = ps->ps_owner_uid;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
(void)strlcpy(svfsb->f_mntonname, ps->ps_target, _VFS_NAMELEN);
if (ps->ps_filesystemtype != NULL)
(void)strlcpy(svfsb->f_fstypename,
ps->ps_filesystemtype, _VFS_NAMELEN);
else
(void)strlcpy(svfsb->f_fstypename, "fuse", _VFS_NAMELEN);
if (ps->ps_source != NULL)
strlcpy(svfsb->f_mntfromname, ps->ps_source, _VFS_NAMELEN);
else
strlcpy(svfsb->f_mntfromname, _PATH_FUSE, _VFS_NAMELEN);
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps->ps_destroy_msg(pm);
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
return 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_fs_sync(struct puffs_usermount *pu, int waitfor,
const struct puffs_cred *pcr)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
/*
* FUSE does not seem to have a FS sync callback.
* Maybe do not even register this callback
*/
return puffs_fsnop_sync(pu, waitfor, pcr);
}
/* ARGSUSED0 */
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_fs_fhtonode(struct puffs_usermount *pu, void *fid, size_t fidsize,
struct puffs_newinfo *pni)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
return 0;
}
/* ARGSUSED0 */
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_fs_nodetofh(struct puffs_usermount *pu, puffs_cookie_t cookie,
void *fid, size_t *fidsize)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
return 0;
}
#if 0
/* ARGSUSED0 */
void
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_fs_extattrctl(struct puffs_usermount *pu, int cmd,
puffs_cookie_t *cookie, int flags, int namespace, const char *attrname)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
return 0;
}
#endif /* 0 */
/* ARGSUSED0 */
void
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_fs_suspend(struct puffs_usermount *pu, int status)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
return;
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_lookup(struct puffs_usermount *pu, puffs_cookie_t opc,
struct puffs_newinfo *pni, const struct puffs_cn *pcn)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
struct perfuse_state *ps;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
struct puffs_node *pn;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
mode_t mode;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
int error;
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
ps = puffs_getspecific(pu);
node_ref(opc);
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
/*
* Check permissions
*/
switch(pcn->pcn_nameiop) {
case NAMEI_DELETE: /* FALLTHROUGH */
case NAMEI_RENAME: /* FALLTHROUGH */
case NAMEI_CREATE:
- Proper permission checks when doing directory traversal. e.g.: run rm dir/file while dir was never looked up since the mount. In that situation, we get lookup with pcn_nameiop NAMEI_DELETE for dir before we get it for file. But for dir we are just looking for PUFFS_VEXEC. This is solved by honouring NAMEI_ISLASTCN, which is set for the last element only - do not send O_EXCL to FUSE as documentation forbids it. - fix warning
2011-05-18 19:28:12 +04:00
if (pcn->pcn_flags & NAMEI_ISLASTCN)
mode = PUFFS_VEXEC|PUFFS_VWRITE;
else
mode = PUFFS_VEXEC;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
break;
case NAMEI_LOOKUP: /* FALLTHROUGH */
default:
mode = PUFFS_VEXEC;
break;
}
if ((error = mode_access(opc, pcn->pcn_cred, mode)) != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
error = node_lookup_common(pu, (puffs_cookie_t)opc, pni,
pcn->pcn_name, pcn->pcn_cred, &pn);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (error != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Add missing ENOENT or ESTALL when accessing deleted node - Fix a warning, fix style (80 chars for a line)
2011-11-28 09:33:33 +04:00
/*
* Kernel would kill us if the filesystem returned the parent
* itself. If we want to live, hide that!
*/
if ((opc == (puffs_cookie_t)pn) && (strcmp(pcn->pcn_name, ".") != 0)) {
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
DERRX(EX_SOFTWARE, "lookup \"%s\" in \"%s\" returned parent",
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
pcn->pcn_name, perfuse_node_path(ps, opc));
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
/* NOTREACHED */
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = ESTALE;
goto out;
- Add missing ENOENT or ESTALL when accessing deleted node - Fix a warning, fix style (80 chars for a line)
2011-11-28 09:33:33 +04:00
}
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
/*
* Removed node
*/
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_REMOVED) {
error = ENOENT;
goto out;
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
/*
* Check for sticky bit. Unfortunately there is no way to
* do this before creating the puffs_node, since we require
* this operation to get the node owner.
*/
switch (pcn->pcn_nameiop) {
case NAMEI_DELETE: /* FALLTHROUGH */
case NAMEI_RENAME:
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = sticky_access(opc, pn, pcn->pcn_cred);
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
if (error != 0) {
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
(void)perfuse_node_reclaim(pu, pn);
goto out;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
}
break;
default:
break;
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_create(struct puffs_usermount *pu, puffs_cookie_t opc,
struct puffs_newinfo *pni, const struct puffs_cn *pcn,
const struct vattr *vap)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
perfuse_msg_t *pm;
struct perfuse_state *ps;
struct fuse_create_in *fci;
struct fuse_entry_out *feo;
struct fuse_open_out *foo;
struct puffs_node *pn;
const char *name;
size_t namelen;
size_t len;
int error;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
return ENOENT;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* If create is unimplemented: Check that it does not
* already exists, and if not, do mknod and open
*/
ps = puffs_getspecific(pu);
if (ps->ps_flags & PS_NO_CREAT) {
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pcn->pcn_cred, &pn);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (error == 0) {
(void)perfuse_node_reclaim(pu, pn);
error = EEXIST;
goto out;
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
error = perfuse_node_mknod(pu, opc, pni, pcn, vap);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (error != 0)
goto out;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pcn->pcn_cred, &pn);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (error != 0)
goto out;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
/*
* FUSE does the open at create time, while
* NetBSD will open in a subsequent operation.
* We need to open now, in order to retain FUSE
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
* semantics. The calling process will not get
* a file descriptor before the kernel sends
* the open operation.
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
*/
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = perfuse_node_open(pu, (puffs_cookie_t)pn,
FWRITE, pcn->pcn_cred);
goto out;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
name = pcn->pcn_name;
namelen = pcn->pcn_namelen + 1;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
len = sizeof(*fci) + namelen;
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
/*
* flags should use O_WRONLY instead of O_RDWR, but it
* breaks when the caller tries to read from file.
Mode argument must contain the file type (S_* items) for create and mknod
2010-09-07 20:58:13 +04:00
*
* mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
*/
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_CREATE, len, pcn->pcn_cred);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fci = GET_INPAYLOAD(ps, pm, fuse_create_in);
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
fci->flags = O_CREAT | O_TRUNC | O_RDWR;
Mode argument must contain the file type (S_* items) for create and mknod
2010-09-07 20:58:13 +04:00
fci->mode = vap->va_mode | VTTOIF(vap->va_type);
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
fci->umask = 0; /* Seems unused by libfuse */
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
(void)strlcpy((char*)(void *)(fci + 1), name, namelen);
len = sizeof(*feo) + sizeof(*foo);
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
if ((error = xchg_msg(pu, opc, pm, len, wait_reply)) != 0) {
/*
* create is unimplmented, remember it for later,
* and start over using mknod and open instead.
*/
if (error == ENOSYS) {
ps->ps_flags |= PS_NO_CREAT;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = perfuse_node_create(pu, opc, pni, pcn, vap);
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
foo = (struct fuse_open_out *)(void *)(feo + 1);
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* Save the file handle and inode in node private data
* so that we can reuse it later
*/
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
pn = perfuse_new_pn(pu, name, opc);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
perfuse_new_fh((puffs_cookie_t)pn, foo->fh, FWRITE);
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
perfuse_node_cache(ps, pn);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
pn->pn_va.va_gen = (u_long)(feo->generation);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
puffs_newinfo_setcookie(pni, pn);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
#ifdef PUFFS_KFLAG_CACHE_FS_TTL
puffs_newinfo_setva(pni, &pn->pn_va);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_newinfo_setttl(pni, pn, feo, NULL);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
#endif /* PUFFS_KFLAG_CACHE_FS_TTL */
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
"nodeid = 0x%"PRIx64", wfh = 0x%"PRIx64"\n",
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
__func__, (void *)pn, pcn->pcn_name,
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid,
foo->fh);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#endif
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_mknod(struct puffs_usermount *pu, puffs_cookie_t opc,
struct puffs_newinfo *pni, const struct puffs_cn *pcn,
const struct vattr *vap)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
perfuse_msg_t *pm;
struct fuse_mknod_in *fmi;
const char* path;
size_t len;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
int error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
return ENOENT;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
/*
* Only superuser can mknod objects other than
* directories, files, socks, fifo and links.
*
* Create an object require -WX permission in the parent directory
*/
switch (vap->va_type) {
case VDIR: /* FALLTHROUGH */
case VREG: /* FALLTHROUGH */
case VFIFO: /* FALLTHROUGH */
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
case VSOCK:
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
break;
default: /* VNON, VBLK, VCHR, VBAD */
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (!puffs_cred_isjuggernaut(pcn->pcn_cred)) {
Improve POSIX compliance of FUSE filesystems through PERUSE - access denied is EPERM and not EACCES - access to file owned by someone else in a sticy-bit directory should be allowed for the sticy-bit directory owner - setting sticky-bit on a non directory should produce EFTYPE - implement PATHCONF method as much as we can.
2014-09-03 20:01:45 +04:00
error = EPERM;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
}
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
break;
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
path = pcn->pcn_name;
len = sizeof(*fmi) + pcn->pcn_namelen + 1;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
/*
Mode argument must contain the file type (S_* items) for create and mknod
2010-09-07 20:58:13 +04:00
* mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
*/
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_MKNOD, len, pcn->pcn_cred);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fmi = GET_INPAYLOAD(ps, pm, fuse_mknod_in);
fmi->mode = vap->va_mode | VTTOIF(vap->va_type);
build fixes for LP64
2010-09-06 05:17:05 +04:00
fmi->rdev = (uint32_t)vap->va_rdev;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fmi->umask = 0; /* Seems unused bu libfuse */
(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = node_mk_common(pu, opc, pni, pcn, pm);
out:
node_rele(opc);
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_open(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
const struct puffs_cred *pcr)
Use just introduced open2 PUFFS method and its PUFFS_OPEN_IO_DIRECT oflag to implement FUSE's OPEN_IO_DIRECT, by which the filesystem tells the kernel that read/write to the file should bypass the page cache. Remove a warning about read beyond EOF which will now normally appear when page cache is bypassed.
2014-08-16 20:28:43 +04:00
{
return perfuse_node_open2(pu, opc, mode, pcr, NULL);
}
int
perfuse_node_open2(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
const struct puffs_cred *pcr, int *oflags)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
struct perfuse_node_data *pnd;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
perfuse_msg_t *pm;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
mode_t fmode;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
int op;
struct fuse_open_in *foi;
struct fuse_open_out *foo;
struct puffs_node *pn;
int error;
ps = puffs_getspecific(pu);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
pn = (struct puffs_node *)opc;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
pnd = PERFUSE_NODE_DATA(opc);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
error = 0;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (pnd->pnd_flags & PND_REMOVED)
return ENOENT;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
if (puffs_pn_getvap(pn)->va_type == VDIR)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
op = FUSE_OPENDIR;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
else
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
op = FUSE_OPEN;
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
/*
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
* libfuse docs says
- Proper permission checks when doing directory traversal. e.g.: run rm dir/file while dir was never looked up since the mount. In that situation, we get lookup with pcn_nameiop NAMEI_DELETE for dir before we get it for file. But for dir we are just looking for PUFFS_VEXEC. This is solved by honouring NAMEI_ISLASTCN, which is set for the last element only - do not send O_EXCL to FUSE as documentation forbids it. - fix warning
2011-05-18 19:28:12 +04:00
* - O_CREAT and O_EXCL should never be set.
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
* - O_TRUNC may be used if mount option atomic_o_trunc is used XXX
*
* O_APPEND makes no sense since FUSE always sends
* the file offset for write operations. If the
* filesystem uses pwrite(), O_APPEND would cause
* the offset to be ignored and cause file corruption.
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
- Proper permission checks when doing directory traversal. e.g.: run rm dir/file while dir was never looked up since the mount. In that situation, we get lookup with pcn_nameiop NAMEI_DELETE for dir before we get it for file. But for dir we are just looking for PUFFS_VEXEC. This is solved by honouring NAMEI_ISLASTCN, which is set for the last element only - do not send O_EXCL to FUSE as documentation forbids it. - fix warning
2011-05-18 19:28:12 +04:00
mode &= ~(O_CREAT|O_EXCL|O_APPEND);
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
/*
* Do not open twice, and do not reopen for reading
- delete an obsoelte comment about inactive - remove a test for getattr return field that was never filled - correctly send filehandle and filehandle flags for getaattr
2010-10-04 07:56:24 +04:00
* if we already have write handle.
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
*/
One more explicit error log, and two bug fixes 1) with recent FUSE, when lookup returns a null ino, it means ENOENT 2) odd corner case that caused a bug on dd if=test of=test conv=notrunc This caused the file to be open first ro, then rw. A logic bug in perfuse_node_open caused it to skip the second operation, whereas it should open for writing, and store the write FH without touching the read FH.
2013-07-18 13:01:20 +04:00
switch (mode & (FREAD|FWRITE)) {
case FREAD:
if (pnd->pnd_flags & (PND_RFH|PND_WFH))
goto out;
break;
case FWRITE:
if (pnd->pnd_flags & PND_WFH)
goto out;
break;
case FREAD|FWRITE:
if (pnd->pnd_flags & PND_WFH)
goto out;
/*
* Corner case: if already open for reading (PND_RFH)
* and re-opening FREAD|FWRITE, we need to reopen,
* but only for writing. Note the change on mode
* will only affect perfuse_new_fh()
*/
if (pnd->pnd_flags & PND_RFH)
mode &= ~FREAD;
break;
Catch open without FREAD|FWRITE (it should not happen)
2013-07-19 11:32:35 +04:00
default:
DWARNX("open without either FREAD nor FWRITE");
error = EPERM;
goto out;
One more explicit error log, and two bug fixes 1) with recent FUSE, when lookup returns a null ino, it means ENOENT 2) odd corner case that caused a bug on dd if=test of=test conv=notrunc This caused the file to be open first ro, then rw. A logic bug in perfuse_node_open caused it to skip the second operation, whereas it should open for writing, and store the write FH without touching the read FH.
2013-07-18 13:01:20 +04:00
}
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
/*
* Queue open on a node so that we do not open
* twice. This would be better with read and
* write distinguished.
*/
while (pnd->pnd_flags & PND_INOPEN)
requeue_request(pu, opc, PCQ_OPEN);
pnd->pnd_flags |= PND_INOPEN;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
/*
* Convert PUFFS mode to FUSE mode: convert FREAD/FWRITE
* to O_RDONLY/O_WRONLY while perserving the other options.
*/
fmode = mode & ~(FREAD|FWRITE);
fmode |= (mode & FWRITE) ? O_RDWR : O_RDONLY;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
pm = ps->ps_new_msg(pu, opc, op, sizeof(*foi), pcr);
foi = GET_INPAYLOAD(ps, pm, fuse_open_in);
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
foi->flags = fmode;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
foi->unused = 0;
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm, sizeof(*foo), wait_reply)) != 0)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
goto out;
foo = GET_OUTPAYLOAD(ps, pm, fuse_open_out);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* Save the file handle in node private data
* so that we can reuse it later
*/
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
perfuse_new_fh(opc, foo->fh, mode);
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
Use just introduced open2 PUFFS method and its PUFFS_OPEN_IO_DIRECT oflag to implement FUSE's OPEN_IO_DIRECT, by which the filesystem tells the kernel that read/write to the file should bypass the page cache. Remove a warning about read beyond EOF which will now normally appear when page cache is bypassed.
2014-08-16 20:28:43 +04:00
/*
* Set direct I/O if the filesystems forces it
*/
if ((foo->open_flags & FUSE_FOPEN_DIRECT_IO) && (oflags != NULL))
*oflags |= PUFFS_OPEN_IO_DIRECT;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#ifdef PERFUSE_DEBUG
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
DPRINTF("%s: opc = %p, file = \"%s\", "
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
"nodeid = 0x%"PRIx64", %s%sfh = 0x%"PRIx64"\n",
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
__func__, (void *)opc, perfuse_node_path(ps, opc),
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
pnd->pnd_nodeid, mode & FREAD ? "r" : "",
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
mode & FWRITE ? "w" : "", foo->fh);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#endif
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
ps->ps_destroy_msg(pm);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
out:
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
pnd->pnd_flags &= ~PND_INOPEN;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_rele(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
return error;
}
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
/* ARGSUSED0 */
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_close(struct puffs_usermount *pu, puffs_cookie_t opc, int flags,
const struct puffs_cred *pcr)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
struct perfuse_node_data *pnd;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
pnd = PERFUSE_NODE_DATA(opc);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
if (!(pnd->pnd_flags & PND_OPEN))
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
return EBADF;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
/*
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
* Actual close is postponed at inactive time.
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
return 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_access(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
const struct puffs_cred *pcr)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
perfuse_msg_t *pm;
struct perfuse_state *ps;
struct fuse_access_in *fai;
int error;
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
return ENOENT;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* If we previously detected the filesystem does not
* implement access(), short-circuit the call and skip
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
* to libpuffs access() emulation.
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
ps = puffs_getspecific(pu);
if (ps->ps_flags & PS_NO_ACCESS) {
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
const struct vattr *vap;
- delete an obsoelte comment about inactive - remove a test for getattr return field that was never filled - correctly send filehandle and filehandle flags for getaattr
2010-10-04 07:56:24 +04:00
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
vap = puffs_pn_getvap((struct puffs_node *)opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
error = puffs_access(IFTOVT(vap->va_mode),
vap->va_mode & ACCESSPERMS,
vap->va_uid, vap->va_gid,
(mode_t)mode, pcr);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
/*
* Plain access call
*/
pm = ps->ps_new_msg(pu, opc, FUSE_ACCESS, sizeof(*fai), pcr);
fai = GET_INPAYLOAD(ps, pm, fuse_access_in);
fai->mask = 0;
fai->mask |= (mode & PUFFS_VREAD) ? R_OK : 0;
fai->mask |= (mode & PUFFS_VWRITE) ? W_OK : 0;
fai->mask |= (mode & PUFFS_VEXEC) ? X_OK : 0;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
ps->ps_destroy_msg(pm);
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
/*
* If unimplemented, start over with emulation
*/
if (error == ENOSYS) {
ps->ps_flags |= PS_NO_ACCESS;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = perfuse_node_access(pu, opc, mode, pcr);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
return error;
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_getattr(struct puffs_usermount *pu, puffs_cookie_t opc,
struct vattr *vap, const struct puffs_cred *pcr)
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
{
return perfuse_node_getattr_ttl(pu, opc, vap, pcr, NULL);
}
int
perfuse_node_getattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
struct vattr *vap, const struct puffs_cred *pcr,
struct timespec *va_ttl)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
perfuse_msg_t *pm = NULL;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
struct perfuse_state *ps;
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
struct fuse_getattr_in *fgi;
struct fuse_attr_out *fao;
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
int error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
Send GETATTR to filesystem for removed but still-open files, as it is the expected behavior (bug caught by glusterFS regression tests)
2014-08-09 07:17:11 +04:00
if ((pnd->pnd_flags & PND_REMOVED) && !(pnd->pnd_flags & PND_OPEN))
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
return ENOENT;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
/*
* Serialize size access, see comment in perfuse_node_setattr().
*/
while (pnd->pnd_flags & PND_INRESIZE)
requeue_request(pu, opc, PCQ_RESIZE);
pnd->pnd_flags |= PND_INRESIZE;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* FUSE_GETATTR_FH must be set in fgi->flags
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
* if we use for fgi->fh
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
pm = ps->ps_new_msg(pu, opc, FUSE_GETATTR, sizeof(*fgi), pcr);
fgi = GET_INPAYLOAD(ps, pm, fuse_getattr_in);
fgi->getattr_flags = 0;
fgi->dummy = 0;
- delete an obsoelte comment about inactive - remove a test for getattr return field that was never filled - correctly send filehandle and filehandle flags for getaattr
2010-10-04 07:56:24 +04:00
fgi->fh = 0;
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
- delete an obsoelte comment about inactive - remove a test for getattr return field that was never filled - correctly send filehandle and filehandle flags for getaattr
2010-10-04 07:56:24 +04:00
if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_OPEN) {
fgi->fh = perfuse_get_fh(opc, FREAD);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
fgi->getattr_flags |= FUSE_GETATTR_FH;
- delete an obsoelte comment about inactive - remove a test for getattr return field that was never filled - correctly send filehandle and filehandle flags for getaattr
2010-10-04 07:56:24 +04:00
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_RESIZE)
make this build on amd64 and remove redundant and unused code.
2011-09-10 02:51:44 +04:00
DPRINTF(">> %s %p %" PRIu64 "\n", __func__, (void *)opc,
vap->va_size);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
#endif
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
goto out;
fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_RESIZE)
make this build on amd64 and remove redundant and unused code.
2011-09-10 02:51:44 +04:00
DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
(void *)opc, vap->va_size, fao->attr.size);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
#endif
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
* We set birthtime, flags, filerev,vaflags to 0.
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
* This seems the best bet, since the information is
* not available from filesystem.
*/
fuse_attr_to_vap(ps, vap, &fao->attr);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
if (va_ttl != NULL) {
va_ttl->tv_sec = fao->attr_valid;
va_ttl->tv_nsec = fao->attr_valid_nsec;
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
out:
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
pnd->pnd_flags &= ~PND_INRESIZE;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_rele(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
return error;
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_setattr(struct puffs_usermount *pu, puffs_cookie_t opc,
const struct vattr *vap, const struct puffs_cred *pcr)
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
{
return perfuse_node_setattr_ttl(pu, opc,
Fix the build by adding (unused) flags argument to perfuse_node_setattr_ttl
2012-06-28 17:53:13 +04:00
__UNCONST(vap), pcr, NULL, 0);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
}
int
perfuse_node_setattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
struct vattr *vap, const struct puffs_cred *pcr,
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
struct timespec *va_ttl, int xflag)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
perfuse_msg_t *pm;
uint64_t fh;
struct perfuse_state *ps;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
struct perfuse_node_data *pnd;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
struct fuse_setattr_in *fsi;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
struct fuse_attr_out *fao;
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
struct vattr *old_va;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
enum perfuse_xchg_pb_reply reply;
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
int error;
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
#ifdef PERFUSE_DEBUG
struct vattr *old_vap;
int resize_debug = 0;
#endif
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
ps = puffs_getspecific(pu);
pnd = PERFUSE_NODE_DATA(opc);
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
/*
* The only operation we can do once the file is removed
* is to resize it, and we can do it only if it is open.
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
* Do not even send the operation to the filesystem: the
* file is not there anymore.
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
*/
if (pnd->pnd_flags & PND_REMOVED) {
if (!(pnd->pnd_flags & PND_OPEN))
return ENOENT;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
return 0;
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
}
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
old_va = puffs_pn_getvap((struct puffs_node *)opc);
/*
* Check for permission to change size
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
* It is always allowed if we already have a write file handle
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
*/
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
!(pnd->pnd_flags & PND_WFH) &&
(error = mode_access(opc, pcr, PUFFS_VWRITE)) != 0)
return error;
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
/*
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
* Check for permission to change dates
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
*/
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (((vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) ||
(vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL)) &&
(puffs_access_times(old_va->va_uid, old_va->va_gid,
old_va->va_mode, 0, pcr) != 0))
Improve POSIX compliance of FUSE filesystems through PERUSE - access denied is EPERM and not EACCES - access to file owned by someone else in a sticy-bit directory should be allowed for the sticy-bit directory owner - setting sticky-bit on a non directory should produce EFTYPE - implement PATHCONF method as much as we can.
2014-09-03 20:01:45 +04:00
return EPERM;
Fix memory leak when we discard a voided setattr operation
2012-06-14 09:58:22 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* Check for permission to change owner and group
*/
if (((vap->va_uid != (uid_t)PUFFS_VNOVAL) ||
(vap->va_gid != (gid_t)PUFFS_VNOVAL)) &&
(puffs_access_chown(old_va->va_uid, old_va->va_gid,
vap->va_uid, vap->va_gid, pcr)) != 0)
Improve POSIX compliance of FUSE filesystems through PERUSE - access denied is EPERM and not EACCES - access to file owned by someone else in a sticy-bit directory should be allowed for the sticy-bit directory owner - setting sticky-bit on a non directory should produce EFTYPE - implement PATHCONF method as much as we can.
2014-09-03 20:01:45 +04:00
return EPERM;
/*
* Check for sticky bit on non-directory by non root user
*/
if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
(vap->va_mode & S_ISTXT) && (old_va->va_type != VDIR) &&
!puffs_cred_isjuggernaut(pcr))
return EFTYPE;
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
/*
Fix memory leak when we discard a voided setattr operation
2012-06-14 09:58:22 +04:00
* Check for permission to change permissions
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
*/
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
(puffs_access_chmod(old_va->va_uid, old_va->va_gid,
old_va->va_type, vap->va_mode, pcr)) != 0)
Improve POSIX compliance of FUSE filesystems through PERUSE - access denied is EPERM and not EACCES - access to file owned by someone else in a sticy-bit directory should be allowed for the sticy-bit directory owner - setting sticky-bit on a non directory should produce EFTYPE - implement PATHCONF method as much as we can.
2014-09-03 20:01:45 +04:00
return EPERM;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
if (pnd->pnd_flags & PND_WFH)
fh = perfuse_get_fh(opc, FWRITE);
else
fh = FUSE_UNKNOWN_FH;
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
/*
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
* fchmod() sets mode and fh, and it may carry
* a resize as well. That may break if the
* filesystem does chmod then resize, and fails
* because it does not have permission anymore.
* We work this around by splitting into two setattr.
- set user/group ownership after object creation. - enforce permissios checks. This needs to be reviewed.
2010-08-28 07:46:21 +04:00
*/
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
(vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
(fh != FUSE_UNKNOWN_FH)) {
struct vattr resize_va;
Fix memory leak when we discard a voided setattr operation
2012-06-14 09:58:22 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
(void)memcpy(&resize_va, vap, sizeof(resize_va));
resize_va.va_mode = (mode_t)PUFFS_VNOVAL;
if ((error = perfuse_node_setattr_ttl(pu, opc, &resize_va,
pcr, va_ttl, xflag)) != 0)
goto out2;
Fix memory leak when we discard a voided setattr operation
2012-06-14 09:58:22 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
vap->va_size = (u_quad_t)PUFFS_VNOVAL;
Fix memory leak when we discard a voided setattr operation
2012-06-14 09:58:22 +04:00
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_SETATTR, sizeof(*fsi), pcr);
fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
fsi->valid = 0;
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
/*
* Get a fh if the node is open for writing
*/
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (fh != FUSE_UNKNOWN_FH) {
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fsi->fh = fh;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
fsi->valid |= FUSE_FATTR_FH;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (vap->va_size != (u_quad_t)PUFFS_VNOVAL) {
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
fsi->size = vap->va_size;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fsi->valid |= FUSE_FATTR_SIZE;
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
/*
* Serialize anything that can touch file size
* to avoid reordered GETATTR and SETATTR.
* Out of order SETATTR can report stale size,
* which will cause the kernel to truncate the file.
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
* XXX Probably useless now we have a lock on GETATTR
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
*/
while (pnd->pnd_flags & PND_INRESIZE)
requeue_request(pu, opc, PCQ_RESIZE);
pnd->pnd_flags |= PND_INRESIZE;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
/*
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
* Setting mtime without atime or vice versa leads to
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
* dates being reset to Epoch on glusterfs. If one
* is missing, use the old value.
*/
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if ((vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) ||
(vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL)) {
if (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) {
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
fsi->atime = vap->va_atime.tv_sec;
fsi->atimensec = (uint32_t)vap->va_atime.tv_nsec;
} else {
fsi->atime = old_va->va_atime.tv_sec;
fsi->atimensec = (uint32_t)old_va->va_atime.tv_nsec;
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) {
- fix access control: pcn->pcn_cred is not user credentials - Keep track of file generation - remove size tracking in pnd_size, we have it in pn_va.va_size
2010-10-11 05:08:26 +04:00
fsi->mtime = vap->va_mtime.tv_sec;
fsi->mtimensec = (uint32_t)vap->va_mtime.tv_nsec;
} else {
fsi->mtime = old_va->va_mtime.tv_sec;
fsi->mtimensec = (uint32_t)old_va->va_mtime.tv_nsec;
}
fsi->valid |= (FUSE_FATTR_MTIME|FUSE_FATTR_ATIME);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (vap->va_mode != (mode_t)PUFFS_VNOVAL) {
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
fsi->mode = vap->va_mode;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fsi->valid |= FUSE_FATTR_MODE;
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (vap->va_uid != (uid_t)PUFFS_VNOVAL) {
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fsi->uid = vap->va_uid;
fsi->valid |= FUSE_FATTR_UID;
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (vap->va_gid != (gid_t)PUFFS_VNOVAL) {
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fsi->gid = vap->va_gid;
fsi->valid |= FUSE_FATTR_GID;
}
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
if (pnd->pnd_lock_owner != 0) {
fsi->lock_owner = pnd->pnd_lock_owner;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fsi->valid |= FUSE_FATTR_LOCKOWNER;
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* ftruncate() sends only va_size, and metadata cache
* flush adds va_atime and va_mtime. Some FUSE
* filesystems will attempt to detect ftruncate by
* checking for FATTR_SIZE being set without
* FATTR_UID|FATTR_GID|FATTR_ATIME|FATTR_MTIME|FATTR_MODE
*
* Try to adapt and remove FATTR_ATIME|FATTR_MTIME
* if we suspect a ftruncate().
*/
if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
((vap->va_mode == (mode_t)PUFFS_VNOVAL) &&
(vap->va_uid == (uid_t)PUFFS_VNOVAL) &&
(vap->va_gid == (gid_t)PUFFS_VNOVAL))) {
fsi->atime = 0;
fsi->atimensec = 0;
fsi->mtime = 0;
fsi->mtimensec = 0;
fsi->valid &= ~(FUSE_FATTR_ATIME|FUSE_FATTR_MTIME);
}
/*
* If nothing remain, discard the operation.
*/
if (!(fsi->valid & (FUSE_FATTR_SIZE|FUSE_FATTR_ATIME|FUSE_FATTR_MTIME|
FUSE_FATTR_MODE|FUSE_FATTR_UID|FUSE_FATTR_GID))) {
error = 0;
ps->ps_destroy_msg(pm);
goto out;
}
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
#ifdef PERFUSE_DEBUG
old_vap = puffs_pn_getvap((struct puffs_node *)opc);
if ((perfuse_diagflags & PDF_RESIZE) &&
(old_vap->va_size != (u_quad_t)PUFFS_VNOVAL)) {
resize_debug = 1;
make this build on amd64 and remove redundant and unused code.
2011-09-10 02:51:44 +04:00
DPRINTF(">> %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
(void *)opc,
puffs_pn_getvap((struct puffs_node *)opc)->va_size,
fsi->size);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
}
#endif
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* Do not honour FAF when changing size. How do
* you want such a thing to work?
*/
reply = wait_reply;
#ifdef PUFFS_SETATTR_FAF
if ((xflag & PUFFS_SETATTR_FAF) && !(fsi->valid & FUSE_FATTR_SIZE))
reply = no_reply;
#endif
if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), reply)) != 0)
goto out;
if (reply == no_reply)
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
goto out;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
* Copy back the new values
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
#ifdef PERFUSE_DEBUG
if (resize_debug)
make this build on amd64 and remove redundant and unused code.
2011-09-10 02:51:44 +04:00
DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
(void *)opc, old_vap->va_size, fao->attr.size);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
#endif
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
fuse_attr_to_vap(ps, old_va, &fao->attr);
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
if (va_ttl != NULL) {
va_ttl->tv_sec = fao->attr_valid;
va_ttl->tv_nsec = fao->attr_valid_nsec;
(void)memcpy(vap, old_va, sizeof(*vap));
}
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
out:
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
if (pnd->pnd_flags & PND_INRESIZE) {
pnd->pnd_flags &= ~PND_INRESIZE;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
}
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out2:
node_rele(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
return error;
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_poll(struct puffs_usermount *pu, puffs_cookie_t opc, int *events)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
perfuse_msg_t *pm;
struct fuse_poll_in *fpi;
struct fuse_poll_out *fpo;
int error;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
/*
* kh is set if FUSE_POLL_SCHEDULE_NOTIFY is set.
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
*
* XXX ps_new_msg() is called with NULL creds, which will
* be interpreted as FUSE superuser. We have no way to
* know the requesting process' credential, but since poll
* is supposed to operate on a file that has been open,
* permission should have already been checked at open time.
* That still may breaks on filesystems that provides odd
* semantics.
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
pm = ps->ps_new_msg(pu, opc, FUSE_POLL, sizeof(*fpi), NULL);
fpi = GET_INPAYLOAD(ps, pm, fuse_poll_in);
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
fpi->fh = perfuse_get_fh(opc, FREAD);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fpi->kh = 0;
fpi->flags = 0;
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FH)
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
"fh = 0x%"PRIx64"\n", __func__, (void *)opc,
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
PERFUSE_NODE_DATA(opc)->pnd_nodeid, fpi->fh);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#endif
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm, sizeof(*fpo), wait_reply)) != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fpo = GET_OUTPAYLOAD(ps, pm, fuse_poll_out);
*events = fpo->revents;
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
/* ARGSUSED2 */
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_fsync(struct puffs_usermount *pu, puffs_cookie_t opc,
const struct puffs_cred *pcr, int flags, off_t offlo, off_t offhi)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
- call FSYNCDIR for directories - directories can be open R/W (for FSYNCDIR) - do not skip calls to FSYNC or FSYNCDIR if the filesystem returned ENOSYS: it may change its mind, and it may also actually do something when retunring ENOSYS - When FSYNC and FSYNCDIR return ENOSYS, do not report it to kernel (silent failure)
2010-09-09 13:12:35 +04:00
int op;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
perfuse_msg_t *pm;
struct perfuse_state *ps;
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
struct perfuse_node_data *pnd;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
struct fuse_fsync_in *ffi;
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
uint64_t fh;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
int error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
pm = NULL;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
pnd = PERFUSE_NODE_DATA(opc);
/*
* No need to sync a removed node
*/
if (pnd->pnd_flags & PND_REMOVED)
return 0;
/*
* We do not sync closed files. They have been
* sync at inactive time already.
*/
if (!(pnd->pnd_flags & PND_OPEN))
return 0;
- call FSYNCDIR for directories - directories can be open R/W (for FSYNCDIR) - do not skip calls to FSYNC or FSYNCDIR if the filesystem returned ENOSYS: it may change its mind, and it may also actually do something when retunring ENOSYS - When FSYNC and FSYNCDIR return ENOSYS, do not report it to kernel (silent failure)
2010-09-09 13:12:35 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
- call FSYNCDIR for directories - directories can be open R/W (for FSYNCDIR) - do not skip calls to FSYNC or FSYNCDIR if the filesystem returned ENOSYS: it may change its mind, and it may also actually do something when retunring ENOSYS - When FSYNC and FSYNCDIR return ENOSYS, do not report it to kernel (silent failure)
2010-09-09 13:12:35 +04:00
if (puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
op = FUSE_FSYNCDIR;
else /* VREG but also other types such as VLNK */
op = FUSE_FSYNC;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
/*
* Do not sync if there are no change to sync
- call FSYNCDIR for directories - directories can be open R/W (for FSYNCDIR) - do not skip calls to FSYNC or FSYNCDIR if the filesystem returned ENOSYS: it may change its mind, and it may also actually do something when retunring ENOSYS - When FSYNC and FSYNCDIR return ENOSYS, do not report it to kernel (silent failure)
2010-09-09 13:12:35 +04:00
* XXX remove that test on files if we implement mmap
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
*/
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_SYNC)
DPRINTF("%s: TEST opc = %p, file = \"%s\" is %sdirty\n",
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
__func__, (void*)opc, perfuse_node_path(ps, opc),
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
pnd->pnd_flags & PND_DIRTY ? "" : "not ");
#endif
if (!(pnd->pnd_flags & PND_DIRTY))
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
/*
* It seems NetBSD can call fsync without open first
* glusterfs complain in such a situation:
* "FSYNC() ERR => -1 (Invalid argument)"
* The file will be closed at inactive time.
*
* We open the directory for reading in order to sync.
* This sounds rather counterintuitive, but it works.
*/
if (!(pnd->pnd_flags & PND_WFH)) {
if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
goto out;
}
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
if (op == FUSE_FSYNCDIR)
fh = perfuse_get_fh(opc, FREAD);
else
fh = perfuse_get_fh(opc, FWRITE);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* If fsync_flags is set, meta data should not be flushed.
*/
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pm = ps->ps_new_msg(pu, opc, op, sizeof(*ffi), pcr);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ffi = GET_INPAYLOAD(ps, pm, fuse_fsync_in);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
ffi->fh = fh;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ffi->fsync_flags = (flags & FFILESYNC) ? 0 : 1;
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FH)
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
__func__, (void *)opc,
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
PERFUSE_NODE_DATA(opc)->pnd_nodeid, ffi->fh);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#endif
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm,
NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
goto out;
/*
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
* No reply beyond fuse_out_header: nothing to do on success
* just clear the dirty flag
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
pnd->pnd_flags &= ~PND_DIRTY;
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_SYNC)
DPRINTF("%s: CLEAR opc = %p, file = \"%s\"\n",
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
__func__, (void*)opc, perfuse_node_path(ps, opc));
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#endif
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
out:
- call FSYNCDIR for directories - directories can be open R/W (for FSYNCDIR) - do not skip calls to FSYNC or FSYNCDIR if the filesystem returned ENOSYS: it may change its mind, and it may also actually do something when retunring ENOSYS - When FSYNC and FSYNCDIR return ENOSYS, do not report it to kernel (silent failure)
2010-09-09 13:12:35 +04:00
/*
* ENOSYS is not returned to kernel,
*/
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
if (error == ENOSYS)
- call FSYNCDIR for directories - directories can be open R/W (for FSYNCDIR) - do not skip calls to FSYNC or FSYNCDIR if the filesystem returned ENOSYS: it may change its mind, and it may also actually do something when retunring ENOSYS - When FSYNC and FSYNCDIR return ENOSYS, do not report it to kernel (silent failure)
2010-09-09 13:12:35 +04:00
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_rele(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
return error;
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_remove(struct puffs_usermount *pu, puffs_cookie_t opc,
puffs_cookie_t targ, const struct puffs_cn *pcn)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
struct perfuse_node_data *pnd;
perfuse_msg_t *pm;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
char *path;
const char *name;
size_t len;
int error;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
pnd = PERFUSE_NODE_DATA(opc);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if ((pnd->pnd_flags & PND_REMOVED) ||
(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
return ENOENT;
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#ifdef PERFUSE_DEBUG
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
if (targ == NULL)
DERRX(EX_SOFTWARE, "%s: targ is NULL", __func__);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
DPRINTF("%s: opc = %p, remove opc = %p, file = \"%s\"\n",
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
__func__, (void *)opc, (void *)targ, pcn->pcn_name);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#endif
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
node_ref(targ);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
/*
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
* Await for all operations on the deleted node to drain,
* as the filesystem may be confused to have it deleted
* during a getattr
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
*/
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
requeue_request(pu, targ, PCQ_AFTERXCHG);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
ps = puffs_getspecific(pu);
pnd = PERFUSE_NODE_DATA(opc);
name = pcn->pcn_name;
len = pcn->pcn_namelen + 1;
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_UNLINK, len, pcn->pcn_cred);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
path = _GET_INPAYLOAD(ps, pm, char *);
(void)strlcpy(path, name, len);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_cache_flush(targ);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
- call FSYNCDIR for directories - directories can be open R/W (for FSYNCDIR) - do not skip calls to FSYNC or FSYNCDIR if the filesystem returned ENOSYS: it may change its mind, and it may also actually do something when retunring ENOSYS - When FSYNC and FSYNCDIR return ENOSYS, do not report it to kernel (silent failure)
2010-09-09 13:12:35 +04:00
/*
* The parent directory needs a sync
*/
PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FILENAME)
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
pcn->pcn_name);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#endif
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
node_rele(targ);
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_link(struct puffs_usermount *pu, puffs_cookie_t opc,
puffs_cookie_t targ, const struct puffs_cn *pcn)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
perfuse_msg_t *pm;
const char *name;
size_t len;
struct puffs_node *pn;
struct fuse_link_in *fli;
int error;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
return ENOENT;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
node_ref(targ);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
pn = (struct puffs_node *)targ;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
name = pcn->pcn_name;
len = sizeof(*fli) + pcn->pcn_namelen + 1;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_LINK, len, pcn->pcn_cred);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fli = GET_INPAYLOAD(ps, pm, fuse_link_in);
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
fli->oldnodeid = PERFUSE_NODE_DATA(pn)->pnd_nodeid;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
(void)strlcpy((char *)(void *)(fli + 1), name, len - sizeof(*fli));
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
node_rele(targ);
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_rename(struct puffs_usermount *pu, puffs_cookie_t opc,
puffs_cookie_t src, const struct puffs_cn *pcn_src,
puffs_cookie_t targ_dir, puffs_cookie_t targ,
const struct puffs_cn *pcn_targ)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
struct perfuse_node_data *dstdir_pnd;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
perfuse_msg_t *pm;
struct fuse_rename_in *fri;
const char *newname;
const char *oldname;
char *np;
int error;
size_t len;
size_t newname_len;
size_t oldname_len;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED) ||
(PERFUSE_NODE_DATA(src)->pnd_flags & PND_REMOVED) ||
(PERFUSE_NODE_DATA(targ_dir)->pnd_flags & PND_REMOVED))
return ENOENT;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
node_ref(src);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
/*
* Await for all operations on the deleted node to drain,
* as the filesystem may be confused to have it deleted
* during a getattr
*/
if ((struct puffs_node *)targ != NULL) {
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(targ);
while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
requeue_request(pu, targ, PCQ_AFTERXCHG);
} else {
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
while (PERFUSE_NODE_DATA(src)->pnd_inxchg)
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
requeue_request(pu, src, PCQ_AFTERXCHG);
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
newname = pcn_targ->pcn_name;
newname_len = pcn_targ->pcn_namelen + 1;
oldname = pcn_src->pcn_name;
oldname_len = pcn_src->pcn_namelen + 1;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
len = sizeof(*fri) + oldname_len + newname_len;
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_RENAME, len, pcn_targ->pcn_cred);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fri = GET_INPAYLOAD(ps, pm, fuse_rename_in);
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
fri->newdir = PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
np = (char *)(void *)(fri + 1);
(void)strlcpy(np, oldname, oldname_len);
np += oldname_len;
(void)strlcpy(np, newname, newname_len);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* Record new parent nodeid
*/
dstdir_pnd = PERFUSE_NODE_DATA(targ_dir);
PERFUSE_NODE_DATA(src)->pnd_parent_nodeid = dstdir_pnd->pnd_nodeid;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (opc != targ_dir)
dstdir_pnd->pnd_flags |= PND_DIRTY;
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (strcmp(newname, "..") != 0)
(void)strlcpy(PERFUSE_NODE_DATA(src)->pnd_name,
newname, MAXPATHLEN);
else
PERFUSE_NODE_DATA(src)->pnd_name[0] = 0; /* forget name */
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if ((struct puffs_node *)targ != NULL) {
perfuse_cache_flush(targ);
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
}
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FILENAME)
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
DPRINTF("%s: nodeid = 0x%"PRIx64" file = \"%s\" renamed \"%s\" "
"nodeid = 0x%"PRIx64" -> nodeid = 0x%"PRIx64" \"%s\"\n",
__func__, PERFUSE_NODE_DATA(src)->pnd_nodeid,
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
pcn_src->pcn_name, pcn_targ->pcn_name,
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
PERFUSE_NODE_DATA(opc)->pnd_nodeid,
PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid,
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_node_path(ps, targ_dir));
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#endif
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
node_rele(src);
if ((struct puffs_node *)targ != NULL)
node_rele(targ);
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_mkdir(struct puffs_usermount *pu, puffs_cookie_t opc,
struct puffs_newinfo *pni, const struct puffs_cn *pcn,
const struct vattr *vap)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
perfuse_msg_t *pm;
struct fuse_mkdir_in *fmi;
const char *path;
size_t len;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
int error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
return ENOENT;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
path = pcn->pcn_name;
len = sizeof(*fmi) + pcn->pcn_namelen + 1;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_MKDIR, len, pcn->pcn_cred);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fmi = GET_INPAYLOAD(ps, pm, fuse_mkdir_in);
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
fmi->mode = vap->va_mode;
fmi->umask = 0; /* Seems unused by libfuse? */
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = node_mk_common(pu, opc, pni, pcn, pm);
node_rele(opc);
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_rmdir(struct puffs_usermount *pu, puffs_cookie_t opc,
puffs_cookie_t targ, const struct puffs_cn *pcn)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
struct perfuse_node_data *pnd;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
perfuse_msg_t *pm;
char *path;
const char *name;
size_t len;
int error;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
pnd = PERFUSE_NODE_DATA(opc);
- Add missing ENOENT or ESTALL when accessing deleted node - Fix a warning, fix style (80 chars for a line)
2011-11-28 09:33:33 +04:00
if ((pnd->pnd_flags & PND_REMOVED) ||
(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
return ENOENT;
rmdir dir/.. must return an error. Use ENOTEMPRY like FFS does.
2014-09-05 19:20:16 +04:00
/*
* Attempt to rmdir dir/.. shoud raise ENOTEMPTY
*/
if (PERFUSE_NODE_DATA(targ)->pnd_nodeid == pnd->pnd_parent_nodeid)
return ENOTEMPTY;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
node_ref(targ);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
/*
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
* Await for all operations on the deleted node to drain,
* as the filesystem may be confused to have it deleted
* during a getattr
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
*/
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
requeue_request(pu, targ, PCQ_AFTERXCHG);
ps = puffs_getspecific(pu);
name = pcn->pcn_name;
len = pcn->pcn_namelen + 1;
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_RMDIR, len, pcn->pcn_cred);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
path = _GET_INPAYLOAD(ps, pm, char *);
(void)strlcpy(path, name, len);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_cache_flush(targ);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
- call FSYNCDIR for directories - directories can be open R/W (for FSYNCDIR) - do not skip calls to FSYNC or FSYNCDIR if the filesystem returned ENOSYS: it may change its mind, and it may also actually do something when retunring ENOSYS - When FSYNC and FSYNCDIR return ENOSYS, do not report it to kernel (silent failure)
2010-09-09 13:12:35 +04:00
/*
* The parent directory needs a sync
*/
PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FILENAME)
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_node_path(ps, targ));
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
#endif
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
node_rele(targ);
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
/* vap is unused */
/* ARGSUSED4 */
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_symlink(struct puffs_usermount *pu, puffs_cookie_t opc,
struct puffs_newinfo *pni, const struct puffs_cn *pcn_src,
const struct vattr *vap, const char *link_target)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
perfuse_msg_t *pm;
char *np;
const char *path;
size_t path_len;
size_t linkname_len;
size_t len;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
int error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
return ENOENT;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
path = pcn_src->pcn_name;
path_len = pcn_src->pcn_namelen + 1;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
linkname_len = strlen(link_target) + 1;
len = path_len + linkname_len;
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
pm = ps->ps_new_msg(pu, opc, FUSE_SYMLINK, len, pcn_src->pcn_cred);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
np = _GET_INPAYLOAD(ps, pm, char *);
(void)strlcpy(np, path, path_len);
np += path_len;
(void)strlcpy(np, link_target, linkname_len);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = node_mk_common(pu, opc, pni, pcn_src, pm);
node_rele(opc);
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
/* ARGSUSED4 */
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_readdir(struct puffs_usermount *pu, puffs_cookie_t opc,
struct dirent *dent, off_t *readoff, size_t *reslen,
const struct puffs_cred *pcr, int *eofflag, off_t *cookies,
size_t *ncookies)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
perfuse_msg_t *pm;
uint64_t fh;
struct perfuse_state *ps;
struct perfuse_node_data *pnd;
struct fuse_read_in *fri;
struct fuse_out_header *foh;
struct fuse_dirent *fd;
size_t foh_len;
int error;
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
size_t fd_maxlen;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
error = 0;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
/*
* readdir state is kept at node level, and several readdir
* requests can be issued at the same time on the same node.
* We need to queue requests so that only one is in readdir
* code at the same time.
*/
pnd = PERFUSE_NODE_DATA(opc);
while (pnd->pnd_flags & PND_INREADDIR)
requeue_request(pu, opc, PCQ_READDIR);
pnd->pnd_flags |= PND_INREADDIR;
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_READDIR)
DPRINTF("%s: READDIR opc = %p enter critical section\n",
__func__, (void *)opc);
#endif
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
/*
* Re-initialize pnd->pnd_fd_cookie on the first readdir for a node
*/
if (*readoff == 0)
pnd->pnd_fd_cookie = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* Do we already have the data bufered?
*/
if (pnd->pnd_dirent != NULL)
goto out;
pnd->pnd_dirent_len = 0;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
/*
* It seems NetBSD can call readdir without open first
* libfuse will crash if it is done that way, hence open first.
*/
if (!(pnd->pnd_flags & PND_OPEN)) {
if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
goto out;
}
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
fh = perfuse_get_fh(opc, FREAD);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FH)
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
"rfh = 0x%"PRIx64"\n", __func__, (void *)opc,
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
PERFUSE_NODE_DATA(opc)->pnd_nodeid, fh);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#endif
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
pnd->pnd_all_fd = NULL;
pnd->pnd_all_fd_len = 0;
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
fd_maxlen = ps->ps_max_readahead - sizeof(*foh);
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
do {
size_t fd_len;
char *afdp;
pm = ps->ps_new_msg(pu, opc, FUSE_READDIR, sizeof(*fri), pcr);
/*
* read_flags, lock_owner and flags are unused in libfuse
*/
fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
fri->fh = fh;
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
fri->offset = pnd->pnd_fd_cookie;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
fri->size = (uint32_t)fd_maxlen;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fri->read_flags = 0;
fri->lock_owner = 0;
fri->flags = 0;
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm,
UNSPEC_REPLY_LEN, wait_reply)) != 0)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
goto out;
/*
* There are many puffs_framebufs calls later,
* therefore foh will not be valid for a long time.
* Just get the length and forget it.
*/
foh = GET_OUTHDR(ps, pm);
foh_len = foh->len;
/*
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
* Empty read: we reached the end of the buffer.
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
if (foh_len == sizeof(*foh)) {
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
ps->ps_destroy_msg(pm);
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
*eofflag = 1;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
break;
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
* Check for corrupted message.
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
if (foh_len < sizeof(*foh) + sizeof(*fd)) {
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
ps->ps_destroy_msg(pm);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
DWARNX("readdir reply too short");
error = EIO;
goto out;
}
fd = GET_OUTPAYLOAD(ps, pm, fuse_dirent);
fd_len = foh_len - sizeof(*foh);
pnd->pnd_all_fd = realloc(pnd->pnd_all_fd,
pnd->pnd_all_fd_len + fd_len);
if (pnd->pnd_all_fd == NULL)
- fix warn/err confusiog - fix debugging printf - add func arguments to simple formats
2011-08-14 03:12:15 +04:00
DERR(EX_OSERR, "%s: malloc failed", __func__);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
afdp = (char *)(void *)pnd->pnd_all_fd + pnd->pnd_all_fd_len;
(void)memcpy(afdp, fd, fd_len);
pnd->pnd_all_fd_len += fd_len;
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
/*
* The fd->off field is used as a cookie for
* resuming the next readdir() where this one was left.
*/
pnd->pnd_fd_cookie = readdir_last_cookie(fd, fd_len);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps->ps_destroy_msg(pm);
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
} while (1 /* CONSTCOND */);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
FUSE struct dirent's off is not the offset in the buffer, it is an opaque cookie that the filesystem passes us, and that we need to send back on the next READDIR. Most filesystem just ignore the value and send the next chunk of buffer, but not all of them. Fixing this allows glusterfs distributed volume to work.
2011-07-14 19:37:32 +04:00
if (pnd->pnd_all_fd != NULL) {
if (fuse_to_dirent(pu, opc, pnd->pnd_all_fd,
pnd->pnd_all_fd_len) == -1)
error = EIO;
}
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
out:
if (pnd->pnd_all_fd != NULL) {
free(pnd->pnd_all_fd);
pnd->pnd_all_fd = NULL;
pnd->pnd_all_fd_len = 0;
}
if (error == 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
readdir_buffered(opc, dent, readoff, reslen);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* Schedule queued readdir requests
*/
pnd->pnd_flags &= ~PND_INREADDIR;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
(void)dequeue_requests(opc, PCQ_READDIR, DEQUEUE_ALL);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_READDIR)
DPRINTF("%s: READDIR opc = %p exit critical section\n",
__func__, (void *)opc);
#endif
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_rele(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
return error;
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_readlink(struct puffs_usermount *pu, puffs_cookie_t opc,
const struct puffs_cred *pcr, char *linkname, size_t *linklen)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
perfuse_msg_t *pm;
int error;
size_t len;
struct fuse_out_header *foh;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
return ENOENT;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
pm = ps->ps_new_msg(pu, opc, FUSE_READLINK, 0, pcr);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
foh = GET_OUTHDR(ps, pm);
FUSE filesystems' readlink returns a resolved link with a NUL trailing character, and PUFFS do not want it. This fixes this bug, that returned stat the informations for x instead of reporting ENOENT: mkdir x && ln x z && stat -x z/whatever/you/want
2010-10-11 05:52:05 +04:00
len = foh->len - sizeof(*foh);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
if (len > *linklen)
Build fixes for LP64
2010-09-01 18:57:24 +04:00
DERRX(EX_PROTOCOL, "path len = %zd too long", len);
FUSE filesystems' readlink returns a resolved link with a NUL trailing character, and PUFFS do not want it. This fixes this bug, that returned stat the informations for x instead of reporting ENOENT: mkdir x && ln x z && stat -x z/whatever/you/want
2010-10-11 05:52:05 +04:00
if (len == 0)
DERRX(EX_PROTOCOL, "path len = %zd too short", len);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
We used to remove the trailing zeros in FUSE readlink replies, but it seems it does not always happen. Just remove them if present.
2014-08-29 08:58:40 +04:00
(void)memcpy(linkname, _GET_OUTPAYLOAD(ps, pm, char *), len);
FUSE filesystems' readlink returns a resolved link with a NUL trailing character, and PUFFS do not want it. This fixes this bug, that returned stat the informations for x instead of reporting ENOENT: mkdir x && ln x z && stat -x z/whatever/you/want
2010-10-11 05:52:05 +04:00
/*
* FUSE filesystems return a NUL terminated string, we
We used to remove the trailing zeros in FUSE readlink replies, but it seems it does not always happen. Just remove them if present.
2014-08-29 08:58:40 +04:00
* do not want the trailing \0
FUSE filesystems' readlink returns a resolved link with a NUL trailing character, and PUFFS do not want it. This fixes this bug, that returned stat the informations for x instead of reporting ENOENT: mkdir x && ln x z && stat -x z/whatever/you/want
2010-10-11 05:52:05 +04:00
*/
We used to remove the trailing zeros in FUSE readlink replies, but it seems it does not always happen. Just remove them if present.
2014-08-29 08:58:40 +04:00
while (len > 0 && linkname[len - 1] == '\0')
len--;
*linklen = len;
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_reclaim(struct puffs_usermount *pu, puffs_cookie_t opc)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
perfuse_msg_t *pm;
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
struct perfuse_node_data *pnd;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
struct fuse_forget_in *ffi;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
int nlookup;
struct timespec now;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
if (opc == 0)
return 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
pnd = PERFUSE_NODE_DATA(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* Never forget the root.
*/
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
if (pnd->pnd_nodeid == FUSE_ROOT_ID)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
return 0;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* There is a race condition between reclaim and lookup.
* When looking up an already known node, the kernel cannot
* hold a reference on the result until it gets the PUFFS
* reply. It mayy therefore reclaim the node after the
* userland looked it up, and before it gets the reply.
* On rely, the kernel re-creates the node, but at that
* time the node has been reclaimed in userland.
*
* In order to avoid this, we refuse reclaiming nodes that
* are too young since the last lookup - and that we do
* not have removed on our own, of course.
*/
if (clock_gettime(CLOCK_REALTIME, &now) != 0)
DERR(EX_OSERR, "clock_gettime failed");
if (timespeccmp(&pnd->pnd_cn_expire, &now, >) &&
!(pnd->pnd_flags & PND_REMOVED)) {
if (!(pnd->pnd_flags & PND_NODELEAK)) {
ps->ps_nodeleakcount++;
pnd->pnd_flags |= PND_NODELEAK;
}
DWARNX("possible leaked node:: opc = %p \"%s\"",
opc, pnd->pnd_name);
return 0;
}
node_ref(opc);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
pnd->pnd_flags |= PND_RECLAIMED;
The kernel can lookup the same node multiple time and will reclaim as many times it looked up. All reclaims but the last one must be ignored, otherwise we discard a node which will still get operations. We therefore have to keep track of lookup/reclaim count and hnour reclaims only when the count reaches zero.
2012-03-08 18:58:57 +04:00
pnd->pnd_puffs_nlookup--;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
nlookup = pnd->pnd_puffs_nlookup;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_RECLAIM)
Build fixes for LP64
2010-09-01 18:57:24 +04:00
DPRINTF("%s (nodeid %"PRId64") reclaimed\n",
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_node_path(ps, opc), pnd->pnd_nodeid);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#endif
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_RECLAIM)
The kernel can lookup the same node multiple time and will reclaim as many times it looked up. All reclaims but the last one must be ignored, otherwise we discard a node which will still get operations. We therefore have to keep track of lookup/reclaim count and hnour reclaims only when the count reaches zero.
2012-03-08 18:58:57 +04:00
DPRINTF("%s (nodeid %"PRId64") is %sreclaimed, nlookup = %d "
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
"%s%s%s%s, pending ops:%s%s%s\n",
perfuse_node_path(ps, opc), pnd->pnd_nodeid,
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
pnd->pnd_flags & PND_RECLAIMED ? "" : "not ",
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
pnd->pnd_puffs_nlookup,
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
pnd->pnd_flags & PND_OPEN ? "open " : "not open",
pnd->pnd_flags & PND_RFH ? "r" : "",
pnd->pnd_flags & PND_WFH ? "w" : "",
- Use SOCK_DGRAM instead of SOCK_STREAM, as the filesystem seems to assume datagram semantics: when using SOCK_STREAM, if perfused sends frames faster than the filesystem consumes them, it will grab multiple frames at once and discard anything beyond the first one. For now the code can work both with SOCK_DGRAM and SOCK_STREAM, but SOCK_STREAM support will probably have to be removed for the sake of readability. - Remeber to sync parent directories when moving a node - In debug output, display the requeue type (readdir, write, etc...)
2010-09-15 05:51:43 +04:00
pnd->pnd_flags & PND_BUSY ? "" : " none",
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
pnd->pnd_flags & PND_INREADDIR ? " readdir" : "",
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
pnd->pnd_flags & PND_INWRITE ? " write" : "",
pnd->pnd_flags & PND_INOPEN ? " open" : "");
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
#endif
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* Make sure it is not looked up again
*/
if (!(pnd->pnd_flags & PND_REMOVED))
perfuse_cache_flush(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* Purge any activity on the node, while checking
* that it remains eligible for a reclaim.
*/
while (pnd->pnd_ref > 1)
requeue_request(pu, opc, PCQ_REF);
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* reclaim cancel?
*/
if (pnd->pnd_puffs_nlookup > nlookup) {
pnd->pnd_flags &= ~PND_RECLAIMED;
perfuse_node_cache(ps, opc);
node_rele(opc);
return 0;
}
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
#ifdef PERFUSE_DEBUG
if ((pnd->pnd_flags & PND_OPEN) ||
!TAILQ_EMPTY(&pnd->pnd_pcq))
One more explicit error log, and two bug fixes 1) with recent FUSE, when lookup returns a null ino, it means ENOENT 2) odd corner case that caused a bug on dd if=test of=test conv=notrunc This caused the file to be open first ro, then rw. A logic bug in perfuse_node_open caused it to skip the second operation, whereas it should open for writing, and store the write FH without touching the read FH.
2013-07-18 13:01:20 +04:00
DERRX(EX_SOFTWARE, "%s: opc = %p \"%s\": still open",
__func__, opc, pnd->pnd_name);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if ((pnd->pnd_flags & PND_BUSY) ||
!TAILQ_EMPTY(&pnd->pnd_pcq))
DERRX(EX_SOFTWARE, "%s: opc = %p: queued operations",
__func__, opc);
if (pnd->pnd_inxchg != 0)
DERRX(EX_SOFTWARE, "%s: opc = %p: ongoing operations",
__func__, opc);
#endif
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* Send the FORGET message
*
* ps_new_msg() is called with NULL creds, which will
* be interpreted as FUSE superuser. This is obviously
* fine since we operate with kernel creds here.
*/
pm = ps->ps_new_msg(pu, opc, FUSE_FORGET,
sizeof(*ffi), NULL);
ffi = GET_INPAYLOAD(ps, pm, fuse_forget_in);
ffi->nlookup = pnd->pnd_fuse_nlookup;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* No reply is expected, pm is freed in xchg_msg
*/
(void)xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, no_reply);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_destroy_pn(pu, opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
return 0;
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_inactive(struct puffs_usermount *pu, puffs_cookie_t opc)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
struct perfuse_node_data *pnd;
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
int error;
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
- When using PUFFS_KFLAG_CACHE_FS_TTL, do not use puffs_node to carry attribute and TTL fora newly created node. Instead extend puffs_newinfo and add puffs_newinfo_setva() and puffs_newinfo_setttl() - Remove node_mk_common_final in libperfuse. It used to set uid/gid for a newly created vnode but has been made redundant along time ago since uid and gid are properly set in FUSE header. - In libperfuse, check for corner case where opc = 0 on INACTIVE and RECLAIM (how is it possible? Check for it to avoid a crash anyway) - In libperfuse, make sure we unlimit RLIMIT_AS and RLIMIT_DATA so that we do notrun out of memory because the kernel is lazy at reclaiming vnodes. - In libperfuse, cleanup style of perfuse_destroy_pn()
2012-04-18 04:57:21 +04:00
if (opc == 0)
return 0;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
pnd = PERFUSE_NODE_DATA(opc);
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (!(pnd->pnd_flags & (PND_OPEN|PND_REMOVED)))
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
/*
* Make sure all operation are finished
* There can be an ongoing write. Other
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
* operation wait for all data before
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
* the close/inactive.
*/
while (pnd->pnd_flags & PND_INWRITE)
requeue_request(pu, opc, PCQ_AFTERWRITE);
/*
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
* The inactive operation may be cancelled,
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
* If no open is in progress, set PND_INOPEN
* so that a new open will be queued.
*/
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
if (pnd->pnd_flags & PND_INOPEN)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
pnd->pnd_flags |= PND_INOPEN;
/*
* Sync data
*/
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
if (pnd->pnd_flags & PND_DIRTY) {
if ((error = perfuse_node_fsync(pu, opc, NULL, 0, 0, 0)) != 0)
DWARN("%s: perfuse_node_fsync failed error = %d",
__func__, error);
}
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
/*
* Close handles
*/
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
if (pnd->pnd_flags & PND_WFH) {
if ((error = perfuse_node_close_common(pu, opc, FWRITE)) != 0)
DWARN("%s: close write FH failed error = %d",
__func__, error);
}
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
Use new PUFFS_KFLAG_CACHE_FS_TTL option to puffs_init(3) so that FUSE TTL on name and attributes are used. This save many PUFFS operations and improves performances. PUFFS_KFLAG_CACHE_FS_TTL is #ifdef'ed in many places for now so that libperfuse can still be used on netbsd-5.
2012-04-08 19:13:06 +04:00
if (pnd->pnd_flags & PND_RFH) {
if ((error = perfuse_node_close_common(pu, opc, FREAD)) != 0)
DWARN("%s: close read FH failed error = %d",
__func__, error);
}
== file close operations == - use PUFFS_KFLAG_WTCACHE to puffs_init so that all writes are immediatly send to the filesystem, and we do not have anymore write after inactive. As a consequence, we can close files at inactive stage, and there is not any concern left with files opened at create time. We also do not have anymore to open ourselves in readdir and fsync. - Fsync on close (inactive stage). That makes sure we will not need to do these operations once the file is closed (FUSE want an open file). short sircuit the request that come after the close, bu not fsinc'ing closed files, - Use PUFFS_KFLAG_IAONDEMAND to get less inactive calls == Removed nodes == - more ENOENT retunred for operations on removed node (but there are probably some still missing): getattr, ooen, setattr, fsync - set PND_REMOVE before sending the UNLINK/RMDIR operations so that we avoid races during UNLINK completion. Also set PND_REMOVED on node we overwirte in rename == Filehandle fixes == - queue open operation to avoid getting two fh for one file - set FH in getattr, if the file is open - Just requires a read FH for fsyncdir, as we always opendir in read mode. Ok, this is misleading :-) == Misc == - do not set FUSE_FATTR_ATIME_NOW in setattr, as we provide the time - short circuit nilpotent operations in setattr - add a filename diagnostic flag to dump file names
2010-09-23 20:02:34 +04:00
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
/*
* This will cause a reclaim to be sent
*/
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
if (pnd->pnd_flags & PND_REMOVED)
puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N1);
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
/*
* Schedule awaiting operations
*/
pnd->pnd_flags &= ~PND_INOPEN;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
- Correctly handle rename whith overwritten destination - Keep track of file name to avoid lookups when we can. This makes sure we do not have two cookies for the same inode, a situation that cause wreak havoc when we come to remove or rename a node. - Do not use PUFFS_FLAG_BUILDPATH at all, since we now track file names - In open, queue requests after checking for access, as there is no merit to queue a will-be-denied request while we can deny it immediatly - request reclaim of removed nodes at inactive stage
2010-10-03 09:46:47 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* errors are ignored, since the kernel ignores the return code.
*/
out:
node_rele(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
return 0;
}
/* ARGSUSED0 */
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_print(struct puffs_usermount *pu, puffs_cookie_t opc)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
return 0;
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_pathconf(struct puffs_usermount *pu, puffs_cookie_t opc,
Fix build failure on amd64.
2014-09-04 03:59:58 +04:00
int name, register_t *retval)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
Improve POSIX compliance of FUSE filesystems through PERUSE - access denied is EPERM and not EACCES - access to file owned by someone else in a sticy-bit directory should be allowed for the sticy-bit directory owner - setting sticky-bit on a non directory should produce EFTYPE - implement PATHCONF method as much as we can.
2014-09-03 20:01:45 +04:00
perfuse_msg_t *pm;
struct perfuse_state *ps;
struct fuse_statfs_out *fso;
int error = 0;
/*
* Static values copied from UFS
* in src/sys/ufs/ufs/ufs_vnops.c
*/
switch (name) {
case _PC_LINK_MAX:
*retval = LINK_MAX;
break;
case _PC_PATH_MAX:
*retval = PATH_MAX;
break;
case _PC_PIPE_BUF:
*retval = PIPE_BUF;
break;
case _PC_CHOWN_RESTRICTED:
*retval = 1;
break;
case _PC_NO_TRUNC:
*retval = 1;
break;
case _PC_SYNC_IO:
*retval = 1;
break;
case _PC_FILESIZEBITS:
*retval = 42;
break;
case _PC_SYMLINK_MAX:
*retval = MAXPATHLEN;
break;
case _PC_2_SYMLINKS:
*retval = 1;
break;
case _PC_NAME_MAX:
ps = puffs_getspecific(pu);
pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply);
if (error != 0)
return error;
fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
*retval = fso->st.namelen;
ps->ps_destroy_msg(pm);
break;
default:
DWARN("Unimplemented pathconf for name = %d", name);
error = ENOSYS;
break;
}
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_advlock(struct puffs_usermount *pu, puffs_cookie_t opc,
void *id, int op, struct flock *fl, int flags)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
int fop;
perfuse_msg_t *pm;
Correctly pass the advlock owner id from kernel to filesystem, instead of using process PID. Allow the usage of the read filehandle for advlock, in order to support shared locks on read-only files
2011-11-16 08:52:40 +04:00
uint64_t fh;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
struct fuse_lk_in *fli;
Fixes for the advlock method. It can now sustain pkgsrc/devel/locktests with glusterfs as backend
2011-05-03 17:19:50 +04:00
struct fuse_out_header *foh;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
struct fuse_lk_out *flo;
Fixes for the advlock method. It can now sustain pkgsrc/devel/locktests with glusterfs as backend
2011-05-03 17:19:50 +04:00
uint32_t owner;
size_t len;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
int error;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
Correctly pass the advlock owner id from kernel to filesystem, instead of using process PID. Allow the usage of the read filehandle for advlock, in order to support shared locks on read-only files
2011-11-16 08:52:40 +04:00
/*
* Make sure we do have a filehandle, as the FUSE filesystem
* expect one. E.g.: if we provide none, GlusterFS logs an error
* "0-glusterfs-fuse: xl is NULL"
*
* We need the read file handle if the file is open read only,
* in order to support shared locks on read-only files.
* NB: The kernel always sends advlock for read-only
* files at exit time when the process used lock, see
* sys_exit -> exit1 -> fd_free -> fd_close -> VOP_ADVLOCK
*/
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if ((fh = perfuse_get_fh(opc, FREAD)) == FUSE_UNKNOWN_FH) {
error = EBADF;
goto out;
}
Correctly pass the advlock owner id from kernel to filesystem, instead of using process PID. Allow the usage of the read filehandle for advlock, in order to support shared locks on read-only files
2011-11-16 08:52:40 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps = puffs_getspecific(pu);
if (op == F_GETLK)
fop = FUSE_GETLK;
else
fop = (flags & F_WAIT) ? FUSE_SETLKW : FUSE_SETLK;
Fix creds passed to FUSE when requests are done on behalf of the kernel. We previously sent uid/gid set to -1, we now set it to 0.
2011-08-02 18:53:38 +04:00
/*
* XXX ps_new_msg() is called with NULL creds, which will
* be interpreted as FUSE superuser. We have no way to
* know the requesting process' credential, but since advlock()
* is supposed to operate on a file that has been open(),
* permission should have already been checked at open() time.
*/
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
pm = ps->ps_new_msg(pu, opc, fop, sizeof(*fli), NULL);
fli = GET_INPAYLOAD(ps, pm, fuse_lk_in);
Correctly pass the advlock owner id from kernel to filesystem, instead of using process PID. Allow the usage of the read filehandle for advlock, in order to support shared locks on read-only files
2011-11-16 08:52:40 +04:00
fli->fh = fh;
fli->owner = (uint64_t)(vaddr_t)id;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fli->lk.start = fl->l_start;
fli->lk.end = fl->l_start + fl->l_len;
fli->lk.type = fl->l_type;
fli->lk.pid = fl->l_pid;
fli->lk_flags = (flags & F_FLOCK) ? FUSE_LK_FLOCK : 0;
- Add missing ENOENT or ESTALL when accessing deleted node - Fix a warning, fix style (80 chars for a line)
2011-11-28 09:33:33 +04:00
owner = (uint32_t)(vaddr_t)id;
Fixes for the advlock method. It can now sustain pkgsrc/devel/locktests with glusterfs as backend
2011-05-03 17:19:50 +04:00
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FH)
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
__func__, (void *)opc,
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
PERFUSE_NODE_DATA(opc)->pnd_nodeid, fli->fh);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#endif
Fixes for the advlock method. It can now sustain pkgsrc/devel/locktests with glusterfs as backend
2011-05-03 17:19:50 +04:00
if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
Fixes for the advlock method. It can now sustain pkgsrc/devel/locktests with glusterfs as backend
2011-05-03 17:19:50 +04:00
foh = GET_OUTHDR(ps, pm);
len = foh->len - sizeof(*foh);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* Save or clear the lock
*/
switch (op) {
Fixes for the advlock method. It can now sustain pkgsrc/devel/locktests with glusterfs as backend
2011-05-03 17:19:50 +04:00
case F_GETLK:
if (len != sizeof(*flo))
DERRX(EX_SOFTWARE,
"%s: Unexpected lock reply len %zd",
__func__, len);
flo = GET_OUTPAYLOAD(ps, pm, fuse_lk_out);
fl->l_start = flo->lk.start;
fl->l_len = flo->lk.end - flo->lk.start;
fl->l_pid = flo->lk.pid;
fl->l_type = flo->lk.type;
fl->l_whence = SEEK_SET; /* libfuse hardcodes it */
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
PERFUSE_NODE_DATA(opc)->pnd_lock_owner = flo->lk.pid;
break;
case F_UNLCK:
Fixes for the advlock method. It can now sustain pkgsrc/devel/locktests with glusterfs as backend
2011-05-03 17:19:50 +04:00
owner = 0;
/* FALLTHROUGH */
case F_SETLK:
/* FALLTHROUGH */
case F_SETLKW:
if (error != 0)
PERFUSE_NODE_DATA(opc)->pnd_lock_owner = owner;
if (len != 0)
DERRX(EX_SOFTWARE,
"%s: Unexpected unlock reply len %zd",
__func__, len);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
break;
default:
Fixes for the advlock method. It can now sustain pkgsrc/devel/locktests with glusterfs as backend
2011-05-03 17:19:50 +04:00
DERRX(EX_SOFTWARE, "%s: Unexpected op %d", __func__, op);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
break;
}
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_read(struct puffs_usermount *pu, puffs_cookie_t opc, uint8_t *buf,
off_t offset, size_t *resid, const struct puffs_cred *pcr, int ioflag)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
struct perfuse_node_data *pnd;
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
const struct vattr *vap;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
perfuse_msg_t *pm;
struct fuse_read_in *fri;
struct fuse_out_header *foh;
size_t readen;
int error;
ps = puffs_getspecific(pu);
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
pnd = PERFUSE_NODE_DATA(opc);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
vap = puffs_pn_getvap((struct puffs_node *)opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
Return EISDIR for read/write to directories. NetBSD directory read should instead return a getent(2) output, but is that really used?
2011-11-10 20:21:09 +04:00
/*
* NetBSD turns that into a getdents(2) output
* We just do a EISDIR as this feature is of little use.
*/
if (vap->va_type == VDIR)
return EISDIR;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
do {
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
size_t max_read;
max_read = ps->ps_max_readahead - sizeof(*foh);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
* flags may be set to FUSE_READ_LOCKOWNER
* if lock_owner is provided.
*/
pm = ps->ps_new_msg(pu, opc, FUSE_READ, sizeof(*fri), pcr);
fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
fri->fh = perfuse_get_fh(opc, FREAD);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fri->offset = offset;
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
fri->size = (uint32_t)MIN(*resid, max_read);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fri->read_flags = 0; /* XXX Unused by libfuse? */
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
fri->lock_owner = pnd->pnd_lock_owner;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fri->flags = 0;
fri->flags |= (fri->lock_owner != 0) ? FUSE_READ_LOCKOWNER : 0;
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_FH)
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
__func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#endif
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
if (error != 0)
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
return error;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
foh = GET_OUTHDR(ps, pm);
readen = foh->len - sizeof(*foh);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
#ifdef PERFUSE_DEBUG
if (readen > *resid)
DERRX(EX_SOFTWARE, "%s: Unexpected big read %zd",
__func__, readen);
#endif
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
(void)memcpy(buf, _GET_OUTPAYLOAD(ps, pm, char *), readen);
buf += readen;
offset += readen;
*resid -= readen;
ps->ps_destroy_msg(pm);
} while ((*resid != 0) && (readen != 0));
if (ioflag & (IO_SYNC|IO_DSYNC))
ps->ps_syncreads++;
else
ps->ps_asyncreads++;
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
return 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
}
int
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
perfuse_node_write(struct puffs_usermount *pu, puffs_cookie_t opc,
uint8_t *buf, off_t offset, size_t *resid,
const struct puffs_cred *pcr, int ioflag)
{
return perfuse_node_write2(pu, opc, buf, offset, resid, pcr, ioflag, 0);
}
/* ARGSUSED7 */
int
perfuse_node_write2(struct puffs_usermount *pu, puffs_cookie_t opc,
uint8_t *buf, off_t offset, size_t *resid,
const struct puffs_cred *pcr, int ioflag, int xflag)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
struct perfuse_state *ps;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
struct perfuse_node_data *pnd;
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
struct vattr *vap;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
perfuse_msg_t *pm;
struct fuse_write_in *fwi;
struct fuse_write_out *fwo;
size_t data_len;
size_t payload_len;
size_t written;
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
int inresize;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
int error;
ps = puffs_getspecific(pu);
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
pnd = PERFUSE_NODE_DATA(opc);
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
vap = puffs_pn_getvap((struct puffs_node *)opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
written = 0;
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
inresize = 0;
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
error = 0;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
if (vap->va_type == VDIR)
Return EISDIR for read/write to directories. NetBSD directory read should instead return a getent(2) output, but is that really used?
2011-11-10 20:21:09 +04:00
return EISDIR;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
/*
* We need to queue write requests in order to avoid
* dequeueing PCQ_AFTERWRITE when there are pending writes.
*/
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
while (pnd->pnd_flags & PND_INWRITE)
requeue_request(pu, opc, PCQ_WRITE);
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
pnd->pnd_flags |= PND_INWRITE;
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
/*
* Serialize size access, see comment in perfuse_node_setattr().
*/
Avoid comparison between signed and unsigned integer expressions by casting the offset to a unsigned type. This fixes the NetBSD/i386 and hopefully the NetBSD/amd64 build.
2011-09-10 14:06:10 +04:00
if ((u_quad_t)offset + *resid > vap->va_size) {
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
while (pnd->pnd_flags & PND_INRESIZE)
requeue_request(pu, opc, PCQ_RESIZE);
pnd->pnd_flags |= PND_INRESIZE;
inresize = 1;
}
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
/*
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
* append flag: re-read the file size so that
* we get the latest value.
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
*/
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
if (ioflag & PUFFS_IO_APPEND) {
if ((error = perfuse_node_getattr(pu, opc, vap, pcr)) != 0)
goto out;
offset = vap->va_size;
}
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_RESIZE)
make this build on amd64 and remove redundant and unused code.
2011-09-10 02:51:44 +04:00
DPRINTF(">> %s %p %" PRIu64 "\n", __func__,
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
(void *)opc, vap->va_size);
#endif
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
do {
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
size_t max_write;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
/*
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
* There is a writepage flag when data
Use sysconf(_SC_PAGESIZE) instead of PAGE_SIZE.
2011-05-11 18:52:48 +04:00
* is aligned to page size. Use it for
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
* everything but the data after the last
* page boundary.
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*/
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
max_write = ps->ps_max_write - sizeof(*fwi);
data_len = MIN(*resid, max_write);
Use sysconf(_SC_PAGESIZE) instead of PAGE_SIZE.
2011-05-11 18:52:48 +04:00
if (data_len > (size_t)sysconf(_SC_PAGESIZE))
data_len = data_len & ~(sysconf(_SC_PAGESIZE) - 1);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
payload_len = data_len + sizeof(*fwi);
/*
* flags may be set to FUSE_WRITE_CACHE (XXX usage?)
* or FUSE_WRITE_LOCKOWNER, if lock_owner is provided.
* write_flags is set to 1 for writepage.
*/
pm = ps->ps_new_msg(pu, opc, FUSE_WRITE, payload_len, pcr);
fwi = GET_INPAYLOAD(ps, pm, fuse_write_in);
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
fwi->fh = perfuse_get_fh(opc, FWRITE);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fwi->offset = offset;
Build fixes for LP64
2010-09-01 18:57:24 +04:00
fwi->size = (uint32_t)data_len;
Use sysconf(_SC_PAGESIZE) instead of PAGE_SIZE.
2011-05-11 18:52:48 +04:00
fwi->write_flags = (fwi->size % sysconf(_SC_PAGESIZE)) ? 0 : 1;
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
fwi->lock_owner = pnd->pnd_lock_owner;
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
fwi->flags = 0;
fwi->flags |= (fwi->lock_owner != 0) ? FUSE_WRITE_LOCKOWNER : 0;
fwi->flags |= (ioflag & IO_DIRECT) ? 0 : FUSE_WRITE_CACHE;
= Open files = - Restore open on our own in fsycn and readdir, as the node may not already be open, and FUSE really wants it to be. No need to close immediatly, it can be done at inactive time. = Write operations = - fix a nasty bug that corrupted files on write (written added twice) - Keep track of file size in order to honour PUFFS_IO_APPEND = many fixes in rename = - handler overwritten nodes correctly - wait for all operations on the node to drain before doing rename, as filesystems may not cope with operations on a moving file. - setback PUFFS_SETBACK_INACT_N1 cannot be used from rename, we therefore miss the inactive time for an overwritten node. This bounds us to give up PUFFS_KFLAG_IAONDEMAND. = Removed files = - forbid most operations on a removed node, return ENOENT - setback PUFFS_SETBACK_NOREF_N1 at inactive stage to cause removed file reclaim = Misc = - Update outdated ARGSUSED for lint - Fix a memory leak (puffs_pn_remove instead of puffs_pn_put) - Do not use PUFFS_FLAG_BUILDPATH except for debug output. It makes the lookup code much simplier.
2010-09-29 12:01:10 +04:00
(void)memcpy((fwi + 1), buf, data_len);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#ifdef PERFUSE_DEBUG
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
if (perfuse_diagflags & PDF_FH)
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
"fh = 0x%"PRIx64"\n", __func__,
- Fix the confusion between fileno (opaque FUSE reference) and inode numbers. fileno must be used when exchanging FUSE messages. - Do not use kernel name cache anymore, as it caused modification from other machines to be invisible. - Honour name and attribute cache directive from FUSE filesystem
2011-10-30 09:11:37 +04:00
(void *)opc, pnd->pnd_nodeid, fwi->fh);
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#endif
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
if ((error = xchg_msg(pu, opc, pm,
sizeof(*fwo), wait_reply)) != 0)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
goto out;
fwo = GET_OUTPAYLOAD(ps, pm, fuse_write_out);
written = fwo->size;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
ps->ps_destroy_msg(pm);
- performance improvement for read, readdir and write. Now we use SOCK_DGRAM, we can send many pages at once without hitting any bug - when creating a file, it is open for FUSE, but not for the kernel. If the kernel does not do a subsequent open, we have a leak. We fight against this by trying to close such file that the kernel left unopen for some time. - some code refactoring to make message exchange debug easier (more to come)
2010-09-20 11:00:21 +04:00
#ifdef PERFUSE_DEBUG
if (written > *resid)
DERRX(EX_SOFTWARE, "%s: Unexpected big write %zd",
__func__, written);
#endif
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
*resid -= written;
offset += written;
buf += written;
} while (*resid != 0);
/*
* puffs_ops(3) says
* "everything must be written or an error will be generated"
*/
if (*resid != 0)
error = EFBIG;
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_RESIZE) {
if (offset > (off_t)vap->va_size)
make this build on amd64 and remove redundant and unused code.
2011-09-10 02:51:44 +04:00
DPRINTF("<< %s %p %" PRIu64 " -> %lld\n", __func__,
(void *)opc, vap->va_size, (long long)offset);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
else
DPRINTF("<< %s %p \n", __func__, (void *)opc);
}
#endif
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
/*
* Update file size if we wrote beyond the end
*/
if (offset > (off_t)vap->va_size)
vap->va_size = offset;
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
if (inresize) {
#ifdef PERFUSE_DEBUG
if (!(pnd->pnd_flags & PND_INRESIZE))
DERRX(EX_SOFTWARE, "file write grow without resize");
#endif
pnd->pnd_flags &= ~PND_INRESIZE;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
Serialize access to file size. We already have such a thing in the kernel, where it fixes race for PUFFS filesystems, but we need it again in perfused since FUSE filesystems are allowed to reorder requests. The huge issue is in the asyncrhonous SETATTR sent by fsync. It is followed by a syncrhnous FSYNC, so if the filesystem does not reorder requests, once the FSYNC returns, we are confident the SETATTR is done. But since FUSE can reorder, we need to implement sync in perfused.
2011-09-09 19:45:28 +04:00
}
- Implement proper unprivilegied user permission verifications Verification is now done in the lookup method, as it is the way to go. Of course there are corner cases, such as the sticky bit which need special handling in the remove method. - Set full fsidx in vftstat method - Do not pass O_APPEND to the filesystem. FUSE always sends the write offset, so setting O_APPEND is useless. If the filesystem uses it in an open(2) system call, it will even cause file corruptions, since offsets given to pwrite(2) will be ignored. This fix allows glusterfs to host a NetBSD ./build.sh -o build - Do not use the FUSE access method, use getattr and check for permission on our own. The problem is that a FUSE filesystem will typically use the Linux-specific setfsuid() to perform access control. If that is missing, any chack is likely to occur on behalf of the user running the filesystem (typically root), causing access method to return wrong information. - When possible, avoid performing a getattr method call and use cached value in puffs_node instead. We still retreive the latest value by calling getattr when performing append write operation, to minimize the chances that another writer appended since the last time we did. - Update puffs_node cached file size in write method - Remove unused argument to perfuse_destroy_pn()
2011-04-25 08:54:53 +04:00
/*
* Statistics
*/
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
if (ioflag & (IO_SYNC|IO_DSYNC))
ps->ps_syncwrites++;
else
ps->ps_asyncwrites++;
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
/*
* Remember to sync the file
*/
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
pnd->pnd_flags |= PND_DIRTY;
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#ifdef PERFUSE_DEBUG
if (perfuse_diagflags & PDF_SYNC)
DPRINTF("%s: DIRTY opc = %p, file = \"%s\"\n",
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
__func__, (void*)opc, perfuse_node_path(ps, opc));
- open the file before doing fsync: glusterfs wants that. - Automatically call fsync on close for files. If we just close, fsync will come later and we will have to reopen - Add a PND_DIRTY flag to keep track of files that really need a sync. perfuse_node_fsync only calls the FUSE fsync method if there are data to push. - Add a PND_OPEN flag to keep track of open files. Checking non NULL fh is not enough, as some filesystems will always set fh to 0. - Add a sync diagnostic flag, to watch fsync and dirty flag activity. Make the fh diagnostic flag more verbose - Send the fh in setattr (it was hardcoded to 0) I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
2010-08-26 17:29:01 +04:00
#endif
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
out:
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
/*
* VOP_PUTPAGE causes FAF write where kernel does not
* check operation result. At least warn if it failed.
*/
#ifdef PUFFS_WRITE_FAF
if (error && (xflag & PUFFS_WRITE_FAF))
DWARN("Data loss caused by FAF write failed on \"%s\"",
pnd->pnd_name);
#endif /* PUFFS_WRITE_FAF */
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
/*
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
* If there are no more queued write, we can resume
* an operation awaiting write completion.
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
*/
- correctly set flags for CREATE - after a node is deleted, some operations should return ENOENT, some should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other operation may also need a fix. - At reclaim time, there is no need to wait for READDIR and READ completion, since the caller will never close a file before getting readir() and read() replies. Waiting for WRITE completion is still mandatory, but we must ensure that no queued WRITE is awaiting to be scheduled. Once the queue is drained, we must check that the reclaim operation was not canceled by a new file LOOKUP. - At reclaim time, fixed a mix up between read and write fh to close - Fixed permission checks for RENAME (it tested the node itself instead of the source) - When seting file mode, only MKNOD needs the filetype (S_* fields). It is probably a bug to set it for other operations.
2010-09-05 10:49:13 +04:00
pnd->pnd_flags &= ~PND_INWRITE;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
if (dequeue_requests(opc, PCQ_WRITE, 1) == 0)
(void)dequeue_requests(opc, PCQ_AFTERWRITE, DEQUEUE_ALL);
- Postpone file close at reclaim time, since NetBSD sends fsync and setattr(mtime, ctime) after close, while FUSE expects the file to be open for these operations - remove unused argument to node_mk_common() - remove requeued requests when they are executed, not when they are tagged for schedule - try to make filehandle management simplier, by keeping track of only one read and one write filehandle (the latter being really read/write). - when CREATE is not available, we use the MKNOD/OPEN path. Fix a bug here where we opened the parent directory instead of the node: add the missing lookup of the mknod'ed node. - lookup file we just created: glusterfs does not really see them otherwise. - open file when doing setattr(mtime, ctime) on non open files, as some filesystems seems to require it. - Do not flush pagecache for removed nodes - Keep track of read/write operations in progress, and at reclaim time, make sure they are over before closing and forgeting the file.
2010-09-03 11:15:18 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_rele(opc);
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
return error;
}
/* ARGSUSED0 */
void
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_cache_write(struct puffs_usermount *pu, puffs_cookie_t opc, size_t size,
struct puffs_cacherun *runs)
libperfuse(3) is a PUFFS relay to FUSE. In order to use it, FUSE filesystem must be patched to #include <perfuse.h> in the source files that open /dev/fuse and perform the mount(2) system call. The FUSE filesystem must be linked with -lperfuse. libperfuse(3) implements the FUSE kernel interface, on which libfuse or any FUSE filesystem that opens /dev/fuse directly can be used. For now, an external daemon called perfused(8) is used. This may change in the future.
2010-08-25 11:16:00 +04:00
{
return;
}
Add support for extended attributes
2011-06-28 20:19:16 +04:00
/* ARGSUSED4 */
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_getextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
int attrns, const char *attrname, size_t *attrsize, uint8_t *attr,
size_t *resid, const struct puffs_cred *pcr)
Add support for extended attributes
2011-06-28 20:19:16 +04:00
{
struct perfuse_state *ps;
char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
perfuse_msg_t *pm;
struct fuse_getxattr_in *fgi;
struct fuse_getxattr_out *fgo;
struct fuse_out_header *foh;
size_t attrnamelen;
size_t len;
char *np;
int error;
- Make sure non root users cannot access system namespace attributes - honour namespace specification when listing attributes - Also fix message memory leak introduced by previous commit
2014-08-10 07:22:33 +04:00
/* system namespace attrs are not accessible to non root users */
if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
return EPERM;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
Add support for extended attributes
2011-06-28 20:19:16 +04:00
ps = puffs_getspecific(pu);
attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
attrnamelen = strlen(attrname) + 1;
len = sizeof(*fgi) + attrnamelen;
pm = ps->ps_new_msg(pu, opc, FUSE_GETXATTR, len, pcr);
fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
make this build on amd64 and remove redundant and unused code.
2011-09-10 02:51:44 +04:00
fgi->size = (unsigned int)((resid != NULL) ? *resid : 0);
Add support for extended attributes
2011-06-28 20:19:16 +04:00
np = (char *)(void *)(fgi + 1);
(void)strlcpy(np, attrname, attrnamelen);
if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
/*
* We just get fuse_getattr_out with list size if we requested
* a null size.
*/
if (resid == NULL) {
fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
if (attrsize != NULL)
*attrsize = fgo->size;
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
goto out;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
}
/*
* And with a non null requested size, we get the list just
* after the header
*/
foh = GET_OUTHDR(ps, pm);
np = (char *)(void *)(foh + 1);
getextattr: fix attribute length being reported to caller. If buffer is too small, return ENORANGE. Caught by glusterFS regression tests
2014-08-09 23:06:50 +04:00
len = foh->len - sizeof(*foh);
if (attrsize != NULL)
*attrsize = len;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
if (resid != NULL) {
getextattr: fix attribute length being reported to caller. If buffer is too small, return ENORANGE. Caught by glusterFS regression tests
2014-08-09 23:06:50 +04:00
if (*resid < len) {
error = ERANGE;
- Make sure non root users cannot access system namespace attributes - honour namespace specification when listing attributes - Also fix message memory leak introduced by previous commit
2014-08-10 07:22:33 +04:00
ps->ps_destroy_msg(pm);
getextattr: fix attribute length being reported to caller. If buffer is too small, return ENORANGE. Caught by glusterFS regression tests
2014-08-09 23:06:50 +04:00
goto out;
}
Add support for extended attributes
2011-06-28 20:19:16 +04:00
(void)memcpy(attr, np, len);
*resid -= len;
}
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
return error;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_setextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
int attrns, const char *attrname, uint8_t *attr, size_t *resid,
const struct puffs_cred *pcr)
Add support for extended attributes
2011-06-28 20:19:16 +04:00
{
struct perfuse_state *ps;
char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
perfuse_msg_t *pm;
struct fuse_setxattr_in *fsi;
size_t attrnamelen;
size_t len;
char *np;
int error;
- Make sure non root users cannot access system namespace attributes - honour namespace specification when listing attributes - Also fix message memory leak introduced by previous commit
2014-08-10 07:22:33 +04:00
/* system namespace attrs are not accessible to non root users */
if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
return EPERM;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
Add support for extended attributes
2011-06-28 20:19:16 +04:00
ps = puffs_getspecific(pu);
attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
attrnamelen = strlen(attrname) + 1;
len = sizeof(*fsi) + attrnamelen + *resid;
pm = ps->ps_new_msg(pu, opc, FUSE_SETXATTR, len, pcr);
fsi = GET_INPAYLOAD(ps, pm, fuse_setxattr_in);
make this build on amd64 and remove redundant and unused code.
2011-09-10 02:51:44 +04:00
fsi->size = (unsigned int)*resid;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
fsi->flags = 0;
np = (char *)(void *)(fsi + 1);
(void)strlcpy(np, attrname, attrnamelen);
np += attrnamelen;
(void)memcpy(np, (char *)attr, *resid);
if ((error = xchg_msg(pu, opc, pm,
NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
*resid = 0;
error = 0;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
return error;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
}
/* ARGSUSED2 */
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_listextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
int attrns, size_t *attrsize, uint8_t *attrs, size_t *resid, int flag,
const struct puffs_cred *pcr)
Add support for extended attributes
2011-06-28 20:19:16 +04:00
{
struct perfuse_state *ps;
perfuse_msg_t *pm;
struct fuse_getxattr_in *fgi;
struct fuse_getxattr_out *fgo;
struct fuse_out_header *foh;
char *np;
- Make sure non root users cannot access system namespace attributes - honour namespace specification when listing attributes - Also fix message memory leak introduced by previous commit
2014-08-10 07:22:33 +04:00
size_t len, puffs_len, i, attrlen, outlen;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
int error;
- Make sure non root users cannot access system namespace attributes - honour namespace specification when listing attributes - Also fix message memory leak introduced by previous commit
2014-08-10 07:22:33 +04:00
/* system namespace attrs are not accessible to non root users */
if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
return EPERM;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
Add support for extended attributes
2011-06-28 20:19:16 +04:00
ps = puffs_getspecific(pu);
len = sizeof(*fgi);
pm = ps->ps_new_msg(pu, opc, FUSE_LISTXATTR, len, pcr);
fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
if (resid != NULL)
make this build on amd64 and remove redundant and unused code.
2011-09-10 02:51:44 +04:00
fgi->size = (unsigned int)*resid;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
else
fgi->size = 0;
if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
goto out;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
/*
* We just get fuse_getattr_out with list size if we requested
* a null size.
*/
if (resid == NULL) {
fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
if (attrsize != NULL)
*attrsize = fgo->size;
Rework puffs_framebuf management toremove leaks and abusive reuses. On exchange error, the puffs_framebuf is now freed immediatly, before requeuing outstanding requests.
2011-12-16 09:34:54 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
goto out;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
}
/*
* And with a non null requested size, we get the list just
* after the header
*/
foh = GET_OUTHDR(ps, pm);
np = (char *)(void *)(foh + 1);
puffs_len = foh->len - sizeof(*foh);
- Make sure non root users cannot access system namespace attributes - honour namespace specification when listing attributes - Also fix message memory leak introduced by previous commit
2014-08-10 07:22:33 +04:00
if (attrsize != NULL)
*attrsize = puffs_len;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
if (attrs != NULL) {
- Make sure non root users cannot access system namespace attributes - honour namespace specification when listing attributes - Also fix message memory leak introduced by previous commit
2014-08-10 07:22:33 +04:00
if (*resid < puffs_len) {
error = ERANGE;
ps->ps_destroy_msg(pm);
goto out;
Add a flag to VOP_LISTEXTATTR(9) so that the vnode interface can tell the filesystem in which format extended attribute shall be listed. There are currently two formats: - NUL-terminated strings, used for listxattr(2), this is the default. - one byte length-pprefixed, non NUL-terminated strings, used for extattr_list_file(2), which is obtanined by setting the EXTATTR_LIST_PREFIXLEN flag to VOP_LISTEXTATTR(9) This approach avoid the need for converting the list back and forth, except in libperfuse, since FUSE uses NUL-terminated strings, and the kernel may have requested EXTATTR_LIST_PREFIXLEN.
2011-07-04 12:07:29 +04:00
}
- Make sure non root users cannot access system namespace attributes - honour namespace specification when listing attributes - Also fix message memory leak introduced by previous commit
2014-08-10 07:22:33 +04:00
outlen = 0;
for (i = 0; i < puffs_len; i += attrlen + 1) {
attrlen = strlen(np + i);
/*
* Filter attributes per namespace
*/
if (!perfuse_ns_match(attrns, np + i))
continue;
#ifdef PUFFS_EXTATTR_LIST_LENPREFIX
/*
* Convert the FUSE reply to length prefixed strings
* if this is what the kernel wants.
*/
if (flag & PUFFS_EXTATTR_LIST_LENPREFIX) {
(void)memcpy(attrs + outlen + 1,
np + i, attrlen);
*(attrs + outlen) = (uint8_t)attrlen;
} else
Make sure libperfuse still builds on netbsd-5.1
2011-07-19 11:29:39 +04:00
#endif /* PUFFS_EXTATTR_LIST_LENPREFIX */
- Make sure non root users cannot access system namespace attributes - honour namespace specification when listing attributes - Also fix message memory leak introduced by previous commit
2014-08-10 07:22:33 +04:00
(void)memcpy(attrs + outlen, np + i, attrlen + 1);
outlen += attrlen + 1;
}
Add support for extended attributes
2011-06-28 20:19:16 +04:00
- Make sure non root users cannot access system namespace attributes - honour namespace specification when listing attributes - Also fix message memory leak introduced by previous commit
2014-08-10 07:22:33 +04:00
*resid -= outlen;
}
Add support for extended attributes
2011-06-28 20:19:16 +04:00
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
error = 0;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
out:
node_rele(opc);
return error;
Add support for extended attributes
2011-06-28 20:19:16 +04:00
}
int
Use C89 function definition
2012-03-21 14:10:36 +04:00
perfuse_node_deleteextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
int attrns, const char *attrname, const struct puffs_cred *pcr)
Add support for extended attributes
2011-06-28 20:19:16 +04:00
{
struct perfuse_state *ps;
char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
perfuse_msg_t *pm;
size_t attrnamelen;
char *np;
int error;
- Make sure non root users cannot access system namespace attributes - honour namespace specification when listing attributes - Also fix message memory leak introduced by previous commit
2014-08-10 07:22:33 +04:00
/* system namespace attrs are not accessible to non root users */
if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
return EPERM;
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_ref(opc);
Add support for extended attributes
2011-06-28 20:19:16 +04:00
ps = puffs_getspecific(pu);
attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
attrnamelen = strlen(attrname) + 1;
pm = ps->ps_new_msg(pu, opc, FUSE_REMOVEXATTR, attrnamelen, pcr);
np = _GET_INPAYLOAD(ps, pm, char *);
(void)strlcpy(np, attrname, attrnamelen);
error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
ps->ps_destroy_msg(pm);
- Fix same vnodes associated with multiple cookies The scheme used to retreive known nodes on lookup was flawed, as it only used parent and name. This produced a different cookie for the same file if it was renamed, when looking up ../ or when dealing with multiple files associated with the same name through link(2). We therefore abandon the use of node name and introduce hashed lists of inodes. This causes a huge rewrite of reclaim code, which do not attempt to keep parents allocated until all their children are reclaimed - Fix race conditions in reclaim There are a few situations where we issue multiple FUSE operations for a PUFFS operation. On reclaim, we therefore have to wait for all FUSE operation to complete, not just the current exchanges. We do this by introducing node reference count with node_ref() and node_rele(). - Detect data loss caused by FAF VOP_PUTPAGES causes FAF writes where the kernel does not check the operation result. At least issue a warning on error. - Enjoy FAF shortcut on setattr No need to wait for the result if the kernel does not want it. There is however an exception for setattr that touch the size, we need to wait for completion because we have other operations queued for after the resize. - Fix fchmod() on write-open file fchmod() on a node open with write privilege will send setattr with both mode and size set. This confuses some FUSE filesystem. Therefore we send two FUSE operations, one for mode, and one for size. - Remove node TTL handling for netbsd-5 for simplicity sake. The code still builds on netbsd-5 but does not have the node TTL feature anymore. It works fine with kernel support on netbsd-6.
2012-07-21 09:49:42 +04:00
node_rele(opc);
Add support for extended attributes
2011-06-28 20:19:16 +04:00
return error;
}