Aren
/
NetBSD
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
NetBSD/sys/nfs/nfs_node.c

356 lines
8.9 KiB
C
Raw Normal View History

nfs_readdirplusrpc: fix a deadlock problem. don't wait for vnode lock to load attributes. otherwise, because READDIRPLUS returns DOTDOT entry as well, we violate locking order.
2004-04-05 14:44:09 +04:00
/* $NetBSD: nfs_node.c,v 1.75 2004/04/05 10:44:09 yamt Exp $ */
New RCS ID's, take two. they're more aesthecially pleasant, and use 'NetBSD'
1994-06-29 10:39:25 +04:00
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
/*
Update to 4.4-Lite fs code, with local changes.
1994-06-08 15:33:09 +04:00
* Copyright (c) 1989, 1993
* The Regents of the University of California. All rights reserved.
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
*
* This code is derived from software contributed to Berkeley by
* Rick Macklem at The University of Guelph.
*
* 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.
Move UCB-licensed code from 4-clause to 3-clause licence. Patches provided by Joel Baker in PR 22364, verified by myself.
2003-08-07 20:26:28 +04:00
* 3. Neither the name of the University nor the names of its contributors
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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.
*
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
* @(#)nfs_node.c 8.6 (Berkeley) 5/22/95
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
*/
add RCSIDs
2001-11-10 13:59:08 +03:00
#include <sys/cdefs.h>
nfs_readdirplusrpc: fix a deadlock problem. don't wait for vnode lock to load attributes. otherwise, because READDIRPLUS returns DOTDOT entry as well, we violate locking order.
2004-04-05 14:44:09 +04:00
__KERNEL_RCSID(0, "$NetBSD: nfs_node.c,v 1.75 2004/04/05 10:44:09 yamt Exp $");
add RCSIDs
2001-11-10 13:59:08 +03:00
New kernel option, NFS_V2_ONLY, which aims to reduce the NFS client to just that required to support NFSv2 mounts. Not finished yet, but already provides some 44k of saving in code size on arm26. More savings, and some documentation, are still to come.
2000-09-19 21:04:50 +04:00
#include "opt_nfs.h"
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
Canonicalize all #includes.
1993-12-18 03:40:47 +03:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/mount.h>
#include <sys/namei.h>
#include <sys/vnode.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
Use the pool allocator and the "nointr" pool page allocator for NFS nodes and vattr structures.
1998-09-01 07:11:36 +04:00
#include <sys/pool.h>
Do locking around nfsnode hashing (perhaps even right this time!)
1997-07-08 03:34:55 +04:00
#include <sys/lock.h>
Replace nfs_hash() (with its extremely bad hash) with a macro to call hash32_buf() to obtain a 32 bit hash. On some tests I ran I obtained a 30x improvement in hash distribution and a 6x reduction in time spent in nfs_nget().
2001-12-06 04:26:36 +03:00
#include <sys/hash.h>
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
Update to 4.4-Lite fs code, with local changes.
1994-06-08 15:33:09 +04:00
#include <nfs/rpcv2.h>
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
#include <nfs/nfsproto.h>
Canonicalize all #includes.
1993-12-18 03:40:47 +03:00
#include <nfs/nfs.h>
#include <nfs/nfsnode.h>
#include <nfs/nfsmount.h>
Update to 4.4-Lite fs code, with local changes.
1994-06-08 15:33:09 +04:00
#include <nfs/nqnfs.h>
nfs prototype changes
1996-02-10 00:48:19 +03:00
#include <nfs/nfs_var.h>
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
Make sure these all agree on the same definitons of various variables.
2002-12-02 02:02:10 +03:00
struct nfsnodehashhead *nfsnodehashtbl;
More LIST/CIRCLEQ migration.
1994-08-19 02:47:43 +04:00
u_long nfsnodehash;
Do locking around nfsnode hashing (perhaps even right this time!)
1997-07-08 03:34:55 +04:00
struct lock nfs_hashlock;
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
Use the pool allocator and the "nointr" pool page allocator for NFS nodes and vattr structures.
1998-09-01 07:11:36 +04:00
struct pool nfs_node_pool; /* memory pool for nfs nodes */
struct pool nfs_vattr_pool; /* memory pool for nfs vattrs */
Add extensible malloc types, adapted from FreeBSD. This turns malloc types into a structure, a pointer to which is passed around, instead of an int constant. Allow the limit to be adjusted when the malloc type is defined, or with a function call, as suggested by Jonathan Stone.
2003-02-01 09:23:35 +03:00
MALLOC_DEFINE(M_NFSBIGFH, "NFS bigfh", "NFS big filehandle");
MALLOC_DEFINE(M_NFSNODE, "NFS node", "NFS vnode private part");
Fix nested extern declaration of prtactive.
2001-02-07 15:40:43 +03:00
extern int prtactive;
Replace nfs_hash() (with its extremely bad hash) with a macro to call hash32_buf() to obtain a 32 bit hash. On some tests I ran I obtained a 30x improvement in hash distribution and a 6x reduction in time spent in nfs_nget().
2001-12-06 04:26:36 +03:00
#define nfs_hash(x,y) hash32_buf((x), (y), HASH32_BUF_INIT)
Add code to UBCify LFS. This is still behind "#ifdef LFS_UBC" for now (there are still some details to work out) but expect that to go away soon. To support these basic changes (creation of lfs_putpages, lfs_gop_write, mods to lfs_balloc) several other changes were made, to wit: * Create a writer daemon kernel thread whose purpose is to handle page writes for the pagedaemon, but which also takes over some of the functions of lfs_check(). This thread is started the first time an LFS is mounted. * Add a "flags" parameter to GOP_SIZE. Current values are GOP_SIZE_READ, meaning that the call should return the size of the in-core version of the file, and GOP_SIZE_WRITE, meaning that it should return the on-disk size. One of GOP_SIZE_READ or GOP_SIZE_WRITE must be specified. * Instead of using malloc(...M_WAITOK) for everything, reserve enough resources to get by and use malloc(...M_NOWAIT), using the reserves if necessary. Use the pool subsystem for structures small enough that this is feasible. This also obsoletes LFS_THROTTLE. And a few that are not strictly necessary: * Moves the LFS inode extensions off onto a separately allocated structure; getting closer to LFS as an LKM. "Welcome to 1.6O." * Unified GOP_ALLOC between FFS and LFS. * Update LFS copyright headers to correct values. * Actually cast to unsigned in lfs_shellsort, like the comment says. * Keep track of which segments were empty before the previous checkpoint; any segments that pass two checkpoints both dirty and empty can be summarily cleaned. Do this. Right now lfs_segclean still works, but this should be turned into an effectless compatibility syscall.
2003-02-18 02:48:08 +03:00
void nfs_gop_size(struct vnode *, off_t, off_t *, int);
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
int nfs_gop_alloc(struct vnode *, off_t, off_t, int, struct ucred *);
make sure that if NMODIFIED is clear, all pages attached to the vnode are clean and without writable mappings. if we try to flush dirty pages past EOF to the server when NMODIFIED is clear, we'll update the attrcache before doing the write, which will try to free the pages past EOF and deadlock. to deal with this, we write-protect pages before we send them to the server, and restrict ourselves to creating read-only mappings if NMODIFIED isn't set. score another one for enami.
2002-03-17 02:05:25 +03:00
int nfs_gop_write(struct vnode *, struct vm_page **, int, int);
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
struct genfs_ops nfs_genfsops = {
nfs_gop_size,
nfs_gop_alloc,
make sure that if NMODIFIED is clear, all pages attached to the vnode are clean and without writable mappings. if we try to flush dirty pages past EOF to the server when NMODIFIED is clear, we'll update the attrcache before doing the write, which will try to free the pages past EOF and deadlock. to deal with this, we write-protect pages before we send them to the server, and restrict ourselves to creating read-only mappings if NMODIFIED isn't set. score another one for enami.
2002-03-17 02:05:25 +03:00
nfs_gop_write,
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
};
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
/*
* Initialize hash links for nfsnodes
* and build nfsnode free list.
*/
nfs prototype changes
1996-02-10 00:48:19 +03:00
void
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
nfs_nhinit()
{
Update for hashinit() change.
2000-11-08 17:28:12 +03:00
nfsnodehashtbl = hashinit(desiredvnodes, HASH_LIST, M_NFSNODE,
M_WAITOK, &nfsnodehash);
Do locking around nfsnode hashing (perhaps even right this time!)
1997-07-08 03:34:55 +04:00
lockinit(&nfs_hashlock, PINOD, "nfs_hashlock", 0, 0);
Use the pool allocator and the "nointr" pool page allocator for NFS nodes and vattr structures.
1998-09-01 07:11:36 +04:00
pool_init(&nfs_node_pool, sizeof(struct nfsnode), 0, 0, 0, "nfsnodepl",
Pool deals fairly well with physical memory shortage, but it doesn't deal with shortages of the VM maps where the backing pages are mapped (usually kmem_map). Try to deal with this: * Group all information about the backend allocator for a pool in a separate structure. The pool references this structure, rather than the individual fields. * Change the pool_init() API accordingly, and adjust all callers. * Link all pools using the same backend allocator on a list. * The backend allocator is responsible for waiting for physical memory to become available, but will still fail if it cannot callocate KVA space for the pages. If this happens, carefully drain all pools using the same backend allocator, so that some KVA space can be freed. * Change pool_reclaim() to indicate if it actually succeeded in freeing some pages, and use that information to make draining easier and more efficient. * Get rid of PR_URGENT. There was only one use of it, and it could be dealt with by the caller. From art@openbsd.org.
2002-03-08 23:48:27 +03:00
&pool_allocator_nointr);
Use the pool allocator and the "nointr" pool page allocator for NFS nodes and vattr structures.
1998-09-01 07:11:36 +04:00
pool_init(&nfs_vattr_pool, sizeof(struct vattr), 0, 0, 0, "nfsvapl",
Pool deals fairly well with physical memory shortage, but it doesn't deal with shortages of the VM maps where the backing pages are mapped (usually kmem_map). Try to deal with this: * Group all information about the backend allocator for a pool in a separate structure. The pool references this structure, rather than the individual fields. * Change the pool_init() API accordingly, and adjust all callers. * Link all pools using the same backend allocator on a list. * The backend allocator is responsible for waiting for physical memory to become available, but will still fail if it cannot callocate KVA space for the pages. If this happens, carefully drain all pools using the same backend allocator, so that some KVA space can be freed. * Change pool_reclaim() to indicate if it actually succeeded in freeing some pages, and use that information to make draining easier and more efficient. * Get rid of PR_URGENT. There was only one use of it, and it could be dealt with by the caller. From art@openbsd.org.
2002-03-08 23:48:27 +03:00
&pool_allocator_nointr);
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
}
add a new VFS op, vfs_reinit, which is called when desiredvnodes is adjusted via sysctl. file systems that have hash tables which are sized based on the value of this variable now resize those hash tables using the new value. the max number of FFS softdeps is also recalculated. convert various file systems to use the <sys/queue.h> macros for their hash tables.
2001-09-15 20:12:54 +04:00
/*
* Reinitialize inode hash table.
*/
void
nfs_nhreinit()
{
struct nfsnode *np;
struct nfsnodehashhead *oldhash, *hash;
u_long oldmask, mask, val;
int i;
hash = hashinit(desiredvnodes, HASH_LIST, M_NFSNODE, M_WAITOK,
&mask);
lockmgr(&nfs_hashlock, LK_EXCLUSIVE, NULL);
oldhash = nfsnodehashtbl;
oldmask = nfsnodehash;
nfsnodehashtbl = hash;
nfsnodehash = mask;
for (i = 0; i <= oldmask; i++) {
while ((np = LIST_FIRST(&oldhash[i])) != NULL) {
LIST_REMOVE(np, n_hash);
val = NFSNOHASH(nfs_hash(np->n_fhp, np->n_fhsize));
LIST_INSERT_HEAD(&hash[val], np, n_hash);
}
}
lockmgr(&nfs_hashlock, LK_RELEASE, NULL);
hashdone(oldhash, M_NFSNODE);
}
Add new VFS op routine - vfs_done and call it on filesystem detach in vfs_detach(). vfs_done may free global filesystem's resources, typically those allocated in respective filesystem's init function. Needed so those filesystems which went in via LKM have a chance to clean after themselves before unloading. This fixes random panics when LKM for filesystem using pools was loaded and unloaded several times. For each leaf filesystem, add appropriate vfs_done routine.
2000-03-16 21:08:17 +03:00
/*
* Free resources previoslu allocated in nfs_nhinit().
*/
void
nfs_nhdone()
{
hashdone(nfsnodehashtbl, M_NFSNODE);
pool_destroy(&nfs_node_pool);
pool_destroy(&nfs_vattr_pool);
}
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
/*
* Look up a vnode/nfsnode by file handle.
* Callers must check for mount points!!
* In all cases, a pointer to a
* nfsnode structure is returned.
*/
nfs prototype changes
1996-02-10 00:48:19 +03:00
int
nfs_readdirplusrpc: fix a deadlock problem. don't wait for vnode lock to load attributes. otherwise, because READDIRPLUS returns DOTDOT entry as well, we violate locking order.
2004-04-05 14:44:09 +04:00
nfs_nget1(mntp, fhp, fhsize, npp, lkflags)
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
struct mount *mntp;
Remove register declarations.
2000-03-30 16:51:13 +04:00
nfsfh_t *fhp;
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
int fhsize;
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
struct nfsnode **npp;
nfs_readdirplusrpc: fix a deadlock problem. don't wait for vnode lock to load attributes. otherwise, because READDIRPLUS returns DOTDOT entry as well, we violate locking order.
2004-04-05 14:44:09 +04:00
int lkflags;
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
{
Remove register declarations.
2000-03-30 16:51:13 +04:00
struct nfsnode *np;
More LIST/CIRCLEQ migration.
1994-08-19 02:47:43 +04:00
struct nfsnodehashhead *nhpp;
Remove register declarations.
2000-03-30 16:51:13 +04:00
struct vnode *vp;
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
int error;
add a new VFS op, vfs_reinit, which is called when desiredvnodes is adjusted via sysctl. file systems that have hash tables which are sized based on the value of this variable now resize those hash tables using the new value. the max number of FFS softdeps is also recalculated. convert various file systems to use the <sys/queue.h> macros for their hash tables.
2001-09-15 20:12:54 +04:00
nhpp = &nfsnodehashtbl[NFSNOHASH(nfs_hash(fhp, fhsize))];
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
loop:
add a new VFS op, vfs_reinit, which is called when desiredvnodes is adjusted via sysctl. file systems that have hash tables which are sized based on the value of this variable now resize those hash tables using the new value. the max number of FFS softdeps is also recalculated. convert various file systems to use the <sys/queue.h> macros for their hash tables.
2001-09-15 20:12:54 +04:00
LIST_FOREACH(np, nhpp, n_hash) {
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
if (mntp != NFSTOV(np)->v_mount || np->n_fhsize != fhsize ||
Initial integration of the Unified Buffer Cache project.
2000-11-27 11:39:39 +03:00
memcmp(fhp, np->n_fhp, fhsize))
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
continue;
vp = NFSTOV(np);
nfs_readdirplusrpc: fix a deadlock problem. don't wait for vnode lock to load attributes. otherwise, because READDIRPLUS returns DOTDOT entry as well, we violate locking order.
2004-04-05 14:44:09 +04:00
error = vget(vp, LK_EXCLUSIVE | lkflags);
if (error == EBUSY)
return error;
if (error)
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
goto loop;
*npp = np;
return(0);
}
Merge with Lite2 + local changes
1998-03-01 05:20:01 +03:00
if (lockmgr(&nfs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0))
Do locking around nfsnode hashing (perhaps even right this time!)
1997-07-08 03:34:55 +04:00
goto loop;
shrink sizeof struct nfsnode by putting exclusive members into union.
2004-03-12 19:52:37 +03:00
error = getnewvnode(VT_NFS, mntp, nfsv2_vnodeop_p, &vp);
nfs prototype changes
1996-02-10 00:48:19 +03:00
if (error) {
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
*npp = 0;
Merge with Lite2 + local changes
1998-03-01 05:20:01 +03:00
lockmgr(&nfs_hashlock, LK_RELEASE, 0);
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
return (error);
}
Use the pool allocator and the "nointr" pool page allocator for NFS nodes and vattr structures.
1998-09-01 07:11:36 +04:00
np = pool_get(&nfs_node_pool, PR_WAITOK);
Initial integration of the Unified Buffer Cache project.
2000-11-27 11:39:39 +03:00
memset(np, 0, sizeof *np);
Update to 4.4-Lite fs code, with local changes.
1994-06-08 15:33:09 +04:00
vp->v_data = np;
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
np->n_vnode = vp;
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
genfs_node_init(vp, &nfs_genfsops);
Initial integration of the Unified Buffer Cache project.
2000-11-27 11:39:39 +03:00
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
/*
* Insert the nfsnode in the hash queue for its new file handle
*/
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
More LIST/CIRCLEQ migration.
1994-08-19 02:47:43 +04:00
LIST_INSERT_HEAD(nhpp, np, n_hash);
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
if (fhsize > NFS_SMALLFH) {
MALLOC()/FREE() are not to be used for variable size allocations.
2000-08-03 10:15:02 +04:00
np->n_fhp = malloc(fhsize, M_NFSBIGFH, M_WAITOK);
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
} else
np->n_fhp = &np->n_fh;
Initial integration of the Unified Buffer Cache project.
2000-11-27 11:39:39 +03:00
memcpy(np->n_fhp, fhp, fhsize);
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
np->n_fhsize = fhsize;
Insert an extra VOP_ACCESS check in nfs_lookup, to avoid cached access mishaps for lookup and getattr. Closes PR 8884. While at it, cache access RPCs.
1999-11-30 02:34:00 +03:00
np->n_accstamp = -1;
Use the pool allocator and the "nointr" pool page allocator for NFS nodes and vattr structures.
1998-09-01 07:11:36 +04:00
np->n_vattr = pool_get(&nfs_vattr_pool, PR_WAITOK);
Unix semantics dictate that access checks for files are done when it is opened. An open file can always be read from and/or written to, depending on how it was opened. Therefore, the read/write/commit RPCs should never return EACCESS, as they are only performed on files that have been successfully opened already. This change improves the current situation and works in most cases. It simply always uses the most recently known owner/group of the file, iff the authentication mechanism is AUTH_UNIX (in other cases, the creds for a succesful open are used, but note that no other cases are currently implemented). A retry mechanism can be used to catch a few more cases, but this is a good improvement for now.
2003-12-08 00:15:46 +03:00
/*
* Initalize read/write creds to useful values. VOP_OPEN will
* overwrite these.
*/
np->n_rcred = curproc->p_ucred;
crhold(np->n_rcred);
np->n_wcred = curproc->p_ucred;
crhold(np->n_wcred);
add a new VFS op, vfs_reinit, which is called when desiredvnodes is adjusted via sysctl. file systems that have hash tables which are sized based on the value of this variable now resize those hash tables using the new value. the max number of FFS softdeps is also recalculated. convert various file systems to use the <sys/queue.h> macros for their hash tables.
2001-09-15 20:12:54 +04:00
lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
lockmgr(&nfs_hashlock, LK_RELEASE, NULL);
don't issue VOP_GETATTR blindly in nfs_nget(). in many cases, GETATTR RPCs here is redundant because the caller has postop_attr. instead, make sure the resulted vnode have a valid attribute in nfs_lookup().
2004-04-05 14:40:56 +04:00
NFS_INVALIDATE_ATTRCACHE(np);
uvm_vnp_setsize(vp, 0);
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
*npp = np;
return (0);
}
nfs prototype changes
1996-02-10 00:48:19 +03:00
int
nfs_inactive(v)
void *v;
{
Update to 4.4-Lite fs code, with local changes.
1994-06-08 15:33:09 +04:00
struct vop_inactive_args /* {
struct vnode *a_vp;
Back out the lwp/ktrace changes. They contained a lot of colateral damage, and need to be examined and discussed more.
2003-06-30 02:28:00 +04:00
struct proc *a_p;
nfs prototype changes
1996-02-10 00:48:19 +03:00
} */ *ap = v;
Remove register declarations.
2000-03-30 16:51:13 +04:00
struct nfsnode *np;
struct sillyrename *sp;
Back out the lwp/ktrace changes. They contained a lot of colateral damage, and need to be examined and discussed more.
2003-06-30 02:28:00 +04:00
struct proc *p = ap->a_p;
In nfs_inactive there's no need anymore for an extra refcount around nfs_vinvalbuf, since it has a real lock on the vnode now, so getnewvnode will not hijack it.
2001-02-06 19:25:48 +03:00
struct vnode *vp = ap->a_vp;
Move purging the dircache and removing a vnode from the nqnfs timer queue into nfs_inactive, this is a better place for it. This doesn't actually solve the actual problem, which appears to be a race condition with unmounting and vnode recycling somewhere, but it fixes it in the sense that nfs_reclaim will not reference a bad v_mount anymore.
2003-02-12 17:50:52 +03:00
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
vrecycle removed nfs vnodes. not perfect, but enough for most cases.
2003-07-30 16:25:39 +04:00
boolean_t removed;
Adjust sillyrename cleanup code to deal with the parent vnode already being locked by our thread. VOP_INACTIVATE() makes no statement as to the lock state of the parent, yet this code assumed we had it unlocked. With this change, we let vn_lock() fail with EDEADLK if we already have the parent locked. We then handle the rename cleanup, and on the way out just vrele() the parent vnode, not vput() it. Fixes a case seen by Steve Woodford at Wasabisystems dot com where we'd panic while running a pkgsrc configure test that verified fork() functionality. I expect the problem is a result of the recent exit() changes and the performance of the machines he tested on. Specifically we would crash during an nfs_remove(). As best I can tell, when nfs_remove() tested to see if we should rename or we should remove, v_usecount was > 1 and vattr.va_nlink was 1. Thus we did the sillyrename in nfs_remove(). However by the time we got down to the vput(vp), v_usecount had dropped to one and thus vput() triggered the VOP_INACTIVATE() code path. nfs_inactive() tries to lock the parent to undo the sillyrename, and deadlocks as we still have it locked.
2004-01-24 01:20:20 +03:00
int err;
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
In nfs_inactive there's no need anymore for an extra refcount around nfs_vinvalbuf, since it has a real lock on the vnode now, so getnewvnode will not hijack it.
2001-02-06 19:25:48 +03:00
np = VTONFS(vp);
if (prtactive && vp->v_usecount != 0)
vprint("nfs_inactive: pushing active", vp);
if (vp->v_type != VDIR) {
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
sp = np->n_sillyrename;
Don't set sillyrename field to 0 for directories, as it's in a union with the head of the cookie list. Fixes PR 3215, fix supplied by Hiroshi Tezuka <tezuka@trc.rwcp.or.jp>. Should also fix M_NFSDIROFF memory leak.
1997-02-12 20:26:56 +03:00
np->n_sillyrename = (struct sillyrename *)0;
} else
Drop vnode lock before removing the sillyrename file, to avoid a lock-o-death.
2001-05-03 19:53:04 +04:00
sp = NULL;
if (sp != NULL)
Back out the lwp/ktrace changes. They contained a lot of colateral damage, and need to be examined and discussed more.
2003-06-30 02:28:00 +04:00
nfs_vinvalbuf(vp, 0, sp->s_cred, p, 1);
vrecycle removed nfs vnodes. not perfect, but enough for most cases.
2003-07-30 16:25:39 +04:00
removed = (np->n_flag & NREMOVED) != 0;
Drop vnode lock before removing the sillyrename file, to avoid a lock-o-death.
2001-05-03 19:53:04 +04:00
np->n_flag &= (NMODIFIED | NFLUSHINPROG | NFLUSHWANT | NQNFSEVICTED |
NQNFSNONCACHE | NQNFSWRITE);
VOP_UNLOCK(vp, 0);
if (sp != NULL) {
Fixes from BSDI (thanks go to Keith Bostic). Original RCS messages: date: 1996/09/06 03:00:31; author: donn; state: Exp; lines: +1 -2 Because NFS doesn't implement vnode locking, nfs_inactive() doesn't really have the vnode locked and hence it can't reliably access the vnode after it performs a blocking operation. We remove one blocking call and push the no-op VOP_UNLOCK higher so that we don't access the vnode after we delete the sillyrename file. This should prevent crashes we've seen in which the vnode turned into a UFS vnode and caused a panic in ufs_unlock() when we tried to 'unlock' it. date: 1996/09/25 19:15:21; author: cp; state: Exp; lines: +4 -0 Kirk's change to not corrupt files after a delete. date: 1996/11/08 19:53:45; author: donn; state: Exp; lines: +16 -4 Krik's change to solve the paradox that vclean() calls nfs_inactive() with VXLOCK set on the vnode, and nfs_inactive() was calling vget() to get a reference on the vnode, which in turn hung on VXLOCK. Nfs_inactive() now checks v_usecount to make sure that the vnode is not coming from vclean() before it does a vget().
1997-02-22 05:45:48 +03:00
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
/*
* Remove the silly file that was rename'd earlier
Adjust sillyrename cleanup code to deal with the parent vnode already being locked by our thread. VOP_INACTIVATE() makes no statement as to the lock state of the parent, yet this code assumed we had it unlocked. With this change, we let vn_lock() fail with EDEADLK if we already have the parent locked. We then handle the rename cleanup, and on the way out just vrele() the parent vnode, not vput() it. Fixes a case seen by Steve Woodford at Wasabisystems dot com where we'd panic while running a pkgsrc configure test that verified fork() functionality. I expect the problem is a result of the recent exit() changes and the performance of the machines he tested on. Specifically we would crash during an nfs_remove(). As best I can tell, when nfs_remove() tested to see if we should rename or we should remove, v_usecount was > 1 and vattr.va_nlink was 1. Thus we did the sillyrename in nfs_remove(). However by the time we got down to the vput(vp), v_usecount had dropped to one and thus vput() triggered the VOP_INACTIVATE() code path. nfs_inactive() tries to lock the parent to undo the sillyrename, and deadlocks as we still have it locked.
2004-01-24 01:20:20 +03:00
*
* Just in case our thread also has the parent node locked,
* we let vn_lock() fail.
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
*/
In nfs_inactive there's no need anymore for an extra refcount around nfs_vinvalbuf, since it has a real lock on the vnode now, so getnewvnode will not hijack it.
2001-02-06 19:25:48 +03:00
Adjust sillyrename cleanup code to deal with the parent vnode already being locked by our thread. VOP_INACTIVATE() makes no statement as to the lock state of the parent, yet this code assumed we had it unlocked. With this change, we let vn_lock() fail with EDEADLK if we already have the parent locked. We then handle the rename cleanup, and on the way out just vrele() the parent vnode, not vput() it. Fixes a case seen by Steve Woodford at Wasabisystems dot com where we'd panic while running a pkgsrc configure test that verified fork() functionality. I expect the problem is a result of the recent exit() changes and the performance of the machines he tested on. Specifically we would crash during an nfs_remove(). As best I can tell, when nfs_remove() tested to see if we should rename or we should remove, v_usecount was > 1 and vattr.va_nlink was 1. Thus we did the sillyrename in nfs_remove(). However by the time we got down to the vput(vp), v_usecount had dropped to one and thus vput() triggered the VOP_INACTIVATE() code path. nfs_inactive() tries to lock the parent to undo the sillyrename, and deadlocks as we still have it locked.
2004-01-24 01:20:20 +03:00
err = vn_lock(sp->s_dvp, LK_EXCLUSIVE | LK_RETRY
| LK_RECURSEFAIL);
Update to 4.4-Lite fs code, with local changes.
1994-06-08 15:33:09 +04:00
nfs_removeit(sp);
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
crfree(sp->s_cred);
Adjust sillyrename cleanup code to deal with the parent vnode already being locked by our thread. VOP_INACTIVATE() makes no statement as to the lock state of the parent, yet this code assumed we had it unlocked. With this change, we let vn_lock() fail with EDEADLK if we already have the parent locked. We then handle the rename cleanup, and on the way out just vrele() the parent vnode, not vput() it. Fixes a case seen by Steve Woodford at Wasabisystems dot com where we'd panic while running a pkgsrc configure test that verified fork() functionality. I expect the problem is a result of the recent exit() changes and the performance of the machines he tested on. Specifically we would crash during an nfs_remove(). As best I can tell, when nfs_remove() tested to see if we should rename or we should remove, v_usecount was > 1 and vattr.va_nlink was 1. Thus we did the sillyrename in nfs_remove(). However by the time we got down to the vput(vp), v_usecount had dropped to one and thus vput() triggered the VOP_INACTIVATE() code path. nfs_inactive() tries to lock the parent to undo the sillyrename, and deadlocks as we still have it locked.
2004-01-24 01:20:20 +03:00
if (err != EDEADLK)
vput(sp->s_dvp);
else
vrele(sp->s_dvp);
Initial integration of the Unified Buffer Cache project.
2000-11-27 11:39:39 +03:00
FREE(sp, M_NFSREQ);
Update to 4.4-Lite fs code, with local changes.
1994-06-08 15:33:09 +04:00
}
Move purging the dircache and removing a vnode from the nqnfs timer queue into nfs_inactive, this is a better place for it. This doesn't actually solve the actual problem, which appears to be a race condition with unmounting and vnode recycling somewhere, but it fixes it in the sense that nfs_reclaim will not reference a bad v_mount anymore.
2003-02-12 17:50:52 +03:00
use queue manipulation macros.
2003-04-02 19:14:19 +04:00
if ((nmp->nm_flag & NFSMNT_NQNFS) && CIRCLEQ_NEXT(np, n_timer) != 0) {
Move purging the dircache and removing a vnode from the nqnfs timer queue into nfs_inactive, this is a better place for it. This doesn't actually solve the actual problem, which appears to be a race condition with unmounting and vnode recycling somewhere, but it fixes it in the sense that nfs_reclaim will not reference a bad v_mount anymore.
2003-02-12 17:50:52 +03:00
CIRCLEQ_REMOVE(&nmp->nm_timerhead, np, n_timer);
}
if (vp->v_type == VDIR && np->n_dircache)
nfs_invaldircache(vp, 1);
vrecycle removed nfs vnodes. not perfect, but enough for most cases.
2003-07-30 16:25:39 +04:00
if (removed)
vrecycle(vp, NULL, p);
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
return (0);
}
/*
* Reclaim an nfsnode so that it can be used for other purposes.
*/
nfs prototype changes
1996-02-10 00:48:19 +03:00
int
nfs_reclaim(v)
void *v;
{
Update to 4.4-Lite fs code, with local changes.
1994-06-08 15:33:09 +04:00
struct vop_reclaim_args /* {
struct vnode *a_vp;
nfs prototype changes
1996-02-10 00:48:19 +03:00
} */ *ap = v;
Remove register declarations.
2000-03-30 16:51:13 +04:00
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
if (prtactive && vp->v_usecount != 0)
vprint("nfs_reclaim: pushing active", vp);
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
Don't remove the nfsnode from the hash chain in nfs_inactive. It will never get back... it will not be found in nfs_nget, a new nfsnode+vnode is allocated instead, which causes a node leak, and also makes the mountpointness of the vnode to be forgotten, breaking filesystem crossing lookups through this vnode.
2003-02-15 21:00:25 +03:00
LIST_REMOVE(np, n_hash);
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
/*
* Free up any directory cookie structures and
* large file handle structures that might be associated with
* this nfs node.
*/
Move purging the dircache and removing a vnode from the nqnfs timer queue into nfs_inactive, this is a better place for it. This doesn't actually solve the actual problem, which appears to be a race condition with unmounting and vnode recycling somewhere, but it fixes it in the sense that nfs_reclaim will not reference a bad v_mount anymore.
2003-02-12 17:50:52 +03:00
if (vp->v_type == VDIR && np->n_dircache)
use hashdone to free hashinit'ed memory.
2003-05-07 17:16:35 +04:00
hashdone(np->n_dircache, M_NFSDIROFF);
avoid double free with xlatecookie.
2003-05-22 18:14:02 +04:00
KASSERT(np->n_dirgens == NULL);
Move purging the dircache and removing a vnode from the nqnfs timer queue into nfs_inactive, this is a better place for it. This doesn't actually solve the actual problem, which appears to be a race condition with unmounting and vnode recycling somewhere, but it fixes it in the sense that nfs_reclaim will not reference a bad v_mount anymore.
2003-02-12 17:50:52 +03:00
if (np->n_fhsize > NFS_SMALLFH)
Initial integration of the Unified Buffer Cache project.
2000-11-27 11:39:39 +03:00
free(np->n_fhp, M_NFSBIGFH);
Bring in a merge of Rick Macklem's NFSv3 code from Lite2
1996-02-18 14:53:36 +03:00
Use the pool allocator and the "nointr" pool page allocator for NFS nodes and vattr structures.
1998-09-01 07:11:36 +04:00
pool_put(&nfs_vattr_pool, np->n_vattr);
Move purging the dircache and removing a vnode from the nqnfs timer queue into nfs_inactive, this is a better place for it. This doesn't actually solve the actual problem, which appears to be a race condition with unmounting and vnode recycling somewhere, but it fixes it in the sense that nfs_reclaim will not reference a bad v_mount anymore.
2003-02-12 17:50:52 +03:00
if (np->n_rcred)
Initial integration of the Unified Buffer Cache project.
2000-11-27 11:39:39 +03:00
crfree(np->n_rcred);
Move purging the dircache and removing a vnode from the nqnfs timer queue into nfs_inactive, this is a better place for it. This doesn't actually solve the actual problem, which appears to be a race condition with unmounting and vnode recycling somewhere, but it fixes it in the sense that nfs_reclaim will not reference a bad v_mount anymore.
2003-02-12 17:50:52 +03:00
if (np->n_wcred)
Initial integration of the Unified Buffer Cache project.
2000-11-27 11:39:39 +03:00
crfree(np->n_wcred);
Move purging the dircache and removing a vnode from the nqnfs timer queue into nfs_inactive, this is a better place for it. This doesn't actually solve the actual problem, which appears to be a race condition with unmounting and vnode recycling somewhere, but it fixes it in the sense that nfs_reclaim will not reference a bad v_mount anymore.
2003-02-12 17:50:52 +03:00
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
cache_purge(vp);
Use the pool allocator and the "nointr" pool page allocator for NFS nodes and vattr structures.
1998-09-01 07:11:36 +04:00
pool_put(&nfs_node_pool, vp->v_data);
Initial integration of the Unified Buffer Cache project.
2000-11-27 11:39:39 +03:00
vp->v_data = NULL;
initial import of 386bsd-0.1 sources
1993-03-21 12:45:37 +03:00
return (0);
}
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
void
Add code to UBCify LFS. This is still behind "#ifdef LFS_UBC" for now (there are still some details to work out) but expect that to go away soon. To support these basic changes (creation of lfs_putpages, lfs_gop_write, mods to lfs_balloc) several other changes were made, to wit: * Create a writer daemon kernel thread whose purpose is to handle page writes for the pagedaemon, but which also takes over some of the functions of lfs_check(). This thread is started the first time an LFS is mounted. * Add a "flags" parameter to GOP_SIZE. Current values are GOP_SIZE_READ, meaning that the call should return the size of the in-core version of the file, and GOP_SIZE_WRITE, meaning that it should return the on-disk size. One of GOP_SIZE_READ or GOP_SIZE_WRITE must be specified. * Instead of using malloc(...M_WAITOK) for everything, reserve enough resources to get by and use malloc(...M_NOWAIT), using the reserves if necessary. Use the pool subsystem for structures small enough that this is feasible. This also obsoletes LFS_THROTTLE. And a few that are not strictly necessary: * Moves the LFS inode extensions off onto a separately allocated structure; getting closer to LFS as an LKM. "Welcome to 1.6O." * Unified GOP_ALLOC between FFS and LFS. * Update LFS copyright headers to correct values. * Actually cast to unsigned in lfs_shellsort, like the comment says. * Keep track of which segments were empty before the previous checkpoint; any segments that pass two checkpoints both dirty and empty can be summarily cleaned. Do this. Right now lfs_segclean still works, but this should be turned into an effectless compatibility syscall.
2003-02-18 02:48:08 +03:00
nfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags)
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
{
Add code to UBCify LFS. This is still behind "#ifdef LFS_UBC" for now (there are still some details to work out) but expect that to go away soon. To support these basic changes (creation of lfs_putpages, lfs_gop_write, mods to lfs_balloc) several other changes were made, to wit: * Create a writer daemon kernel thread whose purpose is to handle page writes for the pagedaemon, but which also takes over some of the functions of lfs_check(). This thread is started the first time an LFS is mounted. * Add a "flags" parameter to GOP_SIZE. Current values are GOP_SIZE_READ, meaning that the call should return the size of the in-core version of the file, and GOP_SIZE_WRITE, meaning that it should return the on-disk size. One of GOP_SIZE_READ or GOP_SIZE_WRITE must be specified. * Instead of using malloc(...M_WAITOK) for everything, reserve enough resources to get by and use malloc(...M_NOWAIT), using the reserves if necessary. Use the pool subsystem for structures small enough that this is feasible. This also obsoletes LFS_THROTTLE. And a few that are not strictly necessary: * Moves the LFS inode extensions off onto a separately allocated structure; getting closer to LFS as an LKM. "Welcome to 1.6O." * Unified GOP_ALLOC between FFS and LFS. * Update LFS copyright headers to correct values. * Actually cast to unsigned in lfs_shellsort, like the comment says. * Keep track of which segments were empty before the previous checkpoint; any segments that pass two checkpoints both dirty and empty can be summarily cleaned. Do this. Right now lfs_segclean still works, but this should be turned into an effectless compatibility syscall.
2003-02-18 02:48:08 +03:00
KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE));
KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE))
!= (GOP_SIZE_READ | GOP_SIZE_WRITE));
a whole bunch of changes to improve performance and robustness under load: - remove special treatment of pager_map mappings in pmaps. this is required now, since I've removed the globals that expose the address range. pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's no longer any need to special-case it. - eliminate struct uvm_vnode by moving its fields into struct vnode. - rewrite the pageout path. the pager is now responsible for handling the high-level requests instead of only getting control after a bunch of work has already been done on its behalf. this will allow us to UBCify LFS, which needs tighter control over its pages than other filesystems do. writing a page to disk no longer requires making it read-only, which allows us to write wired pages without causing all kinds of havoc. - use a new PG_PAGEOUT flag to indicate that a page should be freed on behalf of the pagedaemon when it's unlocked. this flag is very similar to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the pageout fails due to eg. an indirect-block buffer being locked. this allows us to remove the "version" field from struct vm_page, and together with shrinking "loan_count" from 32 bits to 16, struct vm_page is now 4 bytes smaller. - no longer use PG_RELEASED for swap-backed pages. if the page is busy because it's being paged out, we can't release the swap slot to be reallocated until that write is complete, but unlike with vnodes we don't keep a count of in-progress writes so there's no good way to know when the write is done. instead, when we need to free a busy swap-backed page, just sleep until we can get it busy ourselves. - implement a fast-path for extending writes which allows us to avoid zeroing new pages. this substantially reduces cpu usage. - encapsulate the data used by the genfs code in a struct genfs_node, which must be the first element of the filesystem-specific vnode data for filesystems which use genfs_{get,put}pages(). - eliminate many of the UVM pagerops, since they aren't needed anymore now that the pager "put" operation is a higher-level operation. - enhance the genfs code to allow NFS to use the genfs_{get,put}pages instead of a modified copy. - clean up struct vnode by removing all the fields that used to be used by the vfs_cluster.c code (which we don't use anymore with UBC). - remove kmem_object and mb_object since they were useless. instead of allocating pages to these objects, we now just allocate pages with no object. such pages are mapped in the kernel until they are freed, so we can use the mapping to find the page to free it. this allows us to remove splvm() protection in several places. The sum of all these changes improves write throughput on my decstation 5000/200 to within 1% of the rate of NetBSD 1.5 and reduces the elapsed time for "make release" of a NetBSD 1.5 source tree on my 128MB pc to 10% less than a 1.5 kernel took.
2001-09-16 00:36:31 +04:00
*eobp = MAX(size, vp->v_size);
}
int
nfs_gop_alloc(struct vnode *vp, off_t off, off_t len, int flags,
struct ucred *cred)
{
return 0;
}
make sure that if NMODIFIED is clear, all pages attached to the vnode are clean and without writable mappings. if we try to flush dirty pages past EOF to the server when NMODIFIED is clear, we'll update the attrcache before doing the write, which will try to free the pages past EOF and deadlock. to deal with this, we write-protect pages before we send them to the server, and restrict ourselves to creating read-only mappings if NMODIFIED isn't set. score another one for enami.
2002-03-17 02:05:25 +03:00
int
nfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags)
{
int i;
for (i = 0; i < npages; i++) {
pmap_page_protect(pgs[i], VM_PROT_READ);
}
return genfs_gop_write(vp, pgs, npages, flags);
}