When dealing with high I/O throughput, we could run out of buffer
space if the filesystem was not consuming requests fast enough.
Here we slightly raise the buffer size, and we make it tunable
through the PERFUSE_BUFSIZE environment variable so that we can
cope with higher requirement later.
While there, document PERFUSE_OPTIONS environment variable.
FUSE filesystems do not expect to get metadata updates for [amc]time
and size, they updates the value on their own after operations.
The PUFFS PUFFS_KFLAG_NOFLUSH_META option prevents regular metadata cache
flushes to the filesystem , and libperfuse uses it to match Linux FUSE
behavior.
While there, fix a bug in SETATTR: do not update kernel metadata cache
from SETATTR reply when the request is asynchronous, as we do not have
the reply yet.
- 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.
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.
parent, keeping them active, and allowing to lookup .. without sending
a request to the filesystem.
Enable the featuure for perfused, as this is how FUSE works.
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.
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()
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.
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
and the amount of data memory involved is not easy to forcast. We therefore
raise the limit to the maximum.
Patch from Manuel Bouyer. It helps completing a cvs update on a glusterfs
colume.
perfused stays in foreground. The difference is a mistake and was not
intended.
There is still a bug ready to bite here, since SOCK_STREAM is not reliable.
We just hope that buffers are big enough to hold all packets, but if they
are overflown, we loose a packet and a file operation gets stuck.
We really nee SOCk_SEQPACKET here, but unfortunately it is very broken at
that time.
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()
- 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
- 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.
- 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
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)
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...)
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
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
FUSE filesystem must be patched to #include <perfuse.h> in the source
files that open /dev/fuse and perform the mount(2) system call. The
FUSE filesystem must be linked with -lperfuse.
libperfuse(3) implements the FUSE kernel interface, on which libfuse or
any FUSE filesystem that opens /dev/fuse directly can be used.
For now, an external daemon called perfused(8) is used. This may change
in the future.
christos