The ocfs2 write code loops through a page much like the block code, except
that ocfs2 allocation units can be any size, including larger than page
size. Typically it's equal to or larger than page size - most kernels run 4k
pages, the minimum ocfs2 allocation (cluster) size.
Some changes introduced during 2.6.23 changed the way writes to pages are
handled, and inadvertantly broke support for > 4k page size. Instead of just
writing one cluster at a time, we now handle the whole page in one pass.
This means that multiple (small) seperate allocations might happen in the
same pass. The allocation code howver typically optimizes by getting the
maximum which was reserved. This triggered a BUG_ON in the extend code where
it'd ask for a single bit (for one part of a > 4k page) and get back more
than it asked for.
Fix this by providing a variant of the high level allocation function which
allows the caller to specify a maximum. The traditional function remains and
just calls the new one with a maximum determined from the initial
reservation.
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
* 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6:
[XFS] Avoid replaying inode buffer initialisation log items if on-disk version is newer.
[XFS] Ensure file size updates have been completed before writing inode to disk.
[XFS] On-demand reaping of the MRU cache
The do_split() function for htree dir blocks is intended to split a leaf
block to make room for a new entry. It sorts the entries in the original
block by hash value, then moves the last half of the entries to the new
block - without accounting for how much space this actually moves. (IOW,
it moves half of the entry *count* not half of the entry *space*). If by
chance we have both large & small entries, and we move only the smallest
entries, and we have a large new entry to insert, we may not have created
enough space for it.
The patch below stores each record size when calculating the dx_map, and
then walks the hash-sorted dx_map, calculating how many entries must be
moved to more evenly split the existing entries between the old block and
the new block, guaranteeing enough space for the new entry.
The dx_map "offs" member is reduced to u16 so that the overall map size
does not change - it is temporarily stored at the end of the new block, and
if it grows too large it may be overwritten. By making offs and size both
u16, we won't grow the map size.
Also add a few comments to the functions involved.
This fixes the testcase reported by hooanon05@yahoo.co.jp on the
linux-ext4 list, "ext3 dir_index causes an error"
Thanks to Andreas Dilger for discussing the problem & solution with me.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Andreas Dilger <adilger@clusterfs.com>
Tested-by: Junjiro Okajima <hooanon05@yahoo.co.jp>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: <linux-ext4@vger.kernel.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Instead of running the mru cache reaper all the time based on a timeout,
we should only run it when the cache has active objects. This allows CPUs
to sleep when there is no activity rather than be woken repeatedly just to
check if there is anything to do.
SGI-PV: 968554
SGI-Modid: xfs-linux-melb:xfs-kern:29305a
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Donald Douwsma <donaldd@sgi.com>
Signed-off-by: Tim Shimmin <tes@sgi.com>
* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2:
ocfs2: Fix calculation of i_blocks during truncate
[PATCH] ocfs2: Fix a wrong cluster calculation.
[PATCH] ocfs2: fix mount option parsing
ocfs2: update docs for new features
The inode->i_flock list contains the leases, flocks and posix
locks in the specified order. However, the flocks are added in
the head of this list thus hiding the leases from F_GETLEASE
command, from time_out_leases() and other code that expects
the leases to come first.
The following example will demonstrate this:
#define _GNU_SOURCE
#include <unistd.h>
#include <fcntl.h>
#include <stdio.h>
#include <sys/file.h>
static void show_lease(int fd)
{
int res;
res = fcntl(fd, F_GETLEASE);
switch (res) {
case F_RDLCK:
printf("Read lease\n");
break;
case F_WRLCK:
printf("Write lease\n");
break;
case F_UNLCK:
printf("No leases\n");
break;
default:
printf("Some shit\n");
break;
}
}
int main(int argc, char **argv)
{
int fd, res;
fd = open(argv[1], O_RDONLY);
if (fd == -1) {
perror("Can't open file");
return 1;
}
res = fcntl(fd, F_SETLEASE, F_WRLCK);
if (res == -1) {
perror("Can't set lease");
return 1;
}
show_lease(fd);
if (flock(fd, LOCK_SH) == -1) {
perror("Can't flock shared");
return 1;
}
show_lease(fd);
return 0;
}
The first call to show_lease() will show the write lease set, but
the second will show no leases.
Fix the flock adding so that the leases always stay in the head
of this list.
Found during making the flocks pid-namespaces aware.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Acked-by: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Taneli Vähäkangas <vahakang@cs.helsinki.fi> reported that commit
786d7e1612 aka "Fix rmmod/read/write races
in /proc entries" broke SBCL + SLIME combo.
The old code in do_select() used DEFAULT_POLLMASK, if couldn't find
->poll handler. The new code makes ->poll always there and returns 0 by
default, which is not correct. Return DEFAULT_POLLMASK instead.
Steps to reproduce:
install emacs, SBCL, SLIME
emacs
M-x slime in *inferior-lisp* buffer
[watch it doing "Connecting to Swank on port X.."]
Please, apply before 2.6.23.
P.S.: why SBCL can't just read(2) /proc/cpuinfo is a mystery.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: T Taneli Vahakangas <vahakang@cs.helsinki.fi>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If we fail to start a transaction when releasing dquot, we have to call
dquot_release() anyway to mark dquot structure as inactive. Otherwise we
end in an infinite loop inside dqput().
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: xb <xavier.bru@bull.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We were setting i_blocks too early - before truncating any allocation.
Correct things to set i_blocks after the allocation change.
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
In ocfs2_alloc_write_write_ctxt, the written clusters length is calculated
by the byte length only. This may cause some problems if we start to write
at some position in the end of one cluster and last to a second cluster
while the "len" is smaller than a cluster size. In that case, we have to
write 2 clusters actually.
So we have to take the start position into consideration also.
Signed-off-by: Tao Ma <tao.ma@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
For some mount option types, ocfs2_parse_options() will try to access
sb->s_fs_info to get at the ocfs2 private superblock. Unfortunately, that
hasn't been allocated yet and will cause a kernel crash.
Fix this by storing options in a struct which can then get pushed into the
ocfs2_super once it's been allocated later. If we need more options which
store to the ocfs2_super in the future, we can just fields to this struct.
Signed-off-by: Tiger Yang <tiger.yang@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
Update documentation listing ocfs2 features to reflect the current state of
the file system. Add missing descriptions for some mount options which ocfs2
supports.
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
fsid_source decided where to get the 'fsid' number to
return for a GETATTR based on the type of filehandle.
It can be from the device, from the fsid, or from the
UUID.
It is possible for the filehandle to be inconsistent
with the export information, so make sure the export information
actually has the info implied by the value returned by
fsid_source.
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: "Luiz Fernando N. Capitulino" <lcapitulino@gmail.com>
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Recent changes in NFSd cause a directory which is mounted-on
to not appear properly when the filesystem containing it is exported.
*exp_get* now returns -ENOENT rather than NULL and when
commit 5d3dbbeaf5
removed the NULL checks, it didn't add a check for -ENOENT.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This git mod: 77e4635ae1
converted to a "greedy" allocation interface, but for the quota hashtables
it switched from allocating XFS_QM_HASHSIZE (nr of elements)
xfs_dqhash_t's to allocating only XFS_QM_HASHSIZE *bytes* - quite a lot
smaller! Then when we converted hsize "back" to nr of elements (the
division line) hsize went to 0. This was leading to oopses when running
any quota tests on the Fedora 8 test kernel, but the problem has been
there for almost a year.
SGI-PV: 968837
SGI-Modid: xfs-linux-melb:xfs-kern:29354a
Signed-off-by: Eric Sandeen <sandeen@sandeen.net>
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Tim Shimmin <tes@sgi.com>
- in xfs_probe_cluster rename the inner len to pg_len. There's no harm
here because the outer len isn't used after the inner len comes into
existence but it keeps the code clean.
- in xfs_da_do_buf remove the inner i because they don't overlap
and they are both the same type.
SGI-PV: 968555
SGI-Modid: xfs-linux-melb:xfs-kern:29311a
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Tim Shimmin <tes@sgi.com>
- remove the != 0 inside the unlikely in ASSERT_ALWAYS because sparse now
complains about comparisons between pointers and 0
- add a standalone ASSERT implementation because defining it to
ASSERT_ALWAYS means the string is expanded before the token passing
stringification. This way we get the actual content of the
assertion in the assfail message and don't overflow sparse's
stringification buffer leading to sparse error messages.
SGI-PV: 968555
SGI-Modid: xfs-linux-melb:xfs-kern:29310a
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Tim Shimmin <tes@sgi.com>
We can't return a masked result of a __bitwise type. Compare it to 0 first
to keep the behaviour without the warning.
SGI-PV: 968555
SGI-Modid: xfs-linux-melb:xfs-kern:29309a
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Tim Shimmin <tes@sgi.com>
Mark Fasheh
Linus Torvalds
Eric Sandeen
Alexey Dobriyan
Lachlan McIlroy
David Chinner
Jeff Garzik
Masakazu Mokuno
Pavel Emelyanov
Andreas Gruenbacher
Jan Kara
tao.ma@oracle.com
Tiger Yang
Neil Brown
Eric Sandeen
Christoph Hellwig