Btrfs qgroup reserve codes lacks check for rewrite dirty page, causing
every write, even rewriting a uncommitted dirty page, to reserve space.
But only written data will free the reserved space, causing reserved
space leaking.
The bug exists almost from the beginning of btrfs qgroup codes, but
nobody found it.
For example:
1)Write [0, 12K) into file A
reserve 12K space
File A:
0 4K 8K 12K
|<--------dirty-------->|
reserved: 12K
2)Write [0,4K) into file A
0 4K 8K 12K
|<--------dirty-------->|
reserved: 16K <<< Should be 12K
3) Commit transaction
Dirty pages [0,12) written to disk.
Free 12K reserved space.
reserved: 4K <<< Should be 0
This testcase will test such problem.
Kernel fix will need some huge change, so won't be soon.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
For busy-box systems that don't support the extended format options.
Signed-off-by: Ari Sundholm <ari@tuxera.com>
Reviewed-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Add cleanup for seqres.full for new test case template, as sometimes
new testcase may forgot to cleanup seqres.full.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Test renaming of various entry types in directories of various sizes.
Check that filesystem didn't get corrupted.
[dchinner: fixed missing bits from new test template, removed
checking of scratch as harness does that. ]
Signed-off-by: Jan Kara <jack@suse.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The current _require_meta_uuid() test looks for a failure return
code from xfs_db -x -c "uuid generate" but in fact this exits
with success. (In fact uuid_f always exits with success; perhaps
this needs fixing, but that's in the wild now).
So grep for the string(s) stating that it failed, instead.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
ext4/305 sleeps 3 minutes and does mount/umount loop in background,
which produces lots of logs in dmesg and 3 minutes is not necessary.
Ted pointed out that 30 mount/umount cycles is enough to crash a buggy
kernel, so just limit the mount/umount loop to reduce the runtime. And
now the runtime is about 2s.
Reported-by: Theodore Ts'o <tytso@mit.edu>
Suggested-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Eryu Guan <eguan@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
_filter_uuid() get updated and changed output from:
uuid: <UUID>
->
uuid: <UUID>
It is a typo introduced by xfs/077, this patch fixed this.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Test that if we fsync a file that got one extent partially cloned into a
lower file offset, after a power failure our file has the same content it
had before the power failure and after the extent cloning operation.
This test is motivated by an issue found in btrfs that is fixed by the
linux kernel patch titled:
"Btrfs: fix file read corruption after extent cloning and fsync"
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The below command in "Test 4":
xfs_io -c "pwrite -S 0x33 -b 512 `expr $blksize \* 2` 512"
will run failed on 4k sector drives. So I use sector size to
replace the hard-code 512.
And we won't run this case when $sector_size > $page_size / 8.
Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
This case use hard-code 512, but in 4k sector size device,
it will fail.
So I call _min_dio_alignment() to get the sector size, then
replace `512`.
Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
xfs/020 need -f option, or it'll be fail on 4k sector device.
Add -f option for xfs/032 for safe and better.
There're some cases use _check_xfs_filesystem(), or others
function which call this function to check a regular file.
That's will fail when the regular file on a 4k sector device.
For example xfs/250.
So I change _check_xfs_filesystem(), add -f option to xfs_repair,
when the $device is a file.
Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
generic/084 try to run 'tail' command, tail will use inotify.
There're some limit about the number of inotify. For example
fs.inotify.max_user_instances specifies an upper limit on
the number of inotify instances that can be created per real
user ID.
When I test on a machine with 154 cpu cores, this case run
failed, and hit many warning likes:
+tail: inotify cannot be used, reverting to polling: Too many open files
Because the fs.inotify.max_user_instances is 128, so if we
try to tail 154 files, it will be failed.
So use src/multi_open_unlink to instead of tail will avoid
this problem.
Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Targeted fuzzing tests which destroy various pieces of file and
symlink metadata; the tests look for (a) kernel detection of
corruption, (b) xfs_repair repair of said corruption, and (c)
post-repair fs usability.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Targeted fuzzing tests which destroy various pieces of directory
metadata; the tests look for (a) kernel detection of corruption, (b)
xfs_repair repair of said corruption, and (c) post-repair fs
usability.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Targeted fuzzing tests which destroy various pieces of filesystem or
allocation group metadata; the tests look for (a) kernel detection of
corruption, (b) xfs_repair repair of said corruption, and (c)
post-repair fs usability.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Targeted fuzzing tests which destroy various pieces of file,
directory, and symlink metadata; the tests look for (a) kernel
detection of corruption, (b) e2fsck repair of said corruption, and (c)
post-repair fs usability.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Targeted fuzzing tests which destroy various pieces of filesystem or
block group metadata; the tests look for (a) kernel detection of
corruption, (b) e2fsck repair of said corruption, and (c) post-repair
fs usability.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Introduce tests for XFS and ext4 which format a filesystem, populate
it, then uses blocktrash and e2fuzz to corrupt the metadata. The FS
is remounted, modified, and unmounted. Following that, xfs_repair or
e2fsck are run until it no longer finds errors to correct, after which
the FS is mounted yet again and exercised to see if there are any
errors remaining.
The XFS test requires an xfs_db that can handle blocktrash and v5
filesystems.
The ext4 test requires metadata_csum support in e2fsprogs.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
When testing 512 block size xfs, xfs/074 fails as
QA output created by 074
+fallocate: No space left on device
Silence is golden
That's because 40051712*512=20G < 30G.
And quote from Dave:
That was sized to give AGs of a specific size, which originally
contributed to the problem being exposed. You should change the block
count specification to a size specification so the filesysetm being
made on 4k block size filesystems remains unchanged.
size = 40051712b
= 40051712 * 4096
= 164,051,812,352
= 156452m
So set the filesystem size to 156452m explicitly.
Suggested-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Eryu Guan <eguan@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Per Dave Chinner's suggestion, modify generic/064 so that it won't fail
if it finds a few more extents than it expects in its test file after
inserting ranges. When 064's test file is first created, some file
systems may use more than the ideal minimum single extent to represent
it, and this can lead to a mismatch between the actual and expected
extent count after the ranges have been inserted. Ext4 file systems
mounted with delayed allocation disabled can exhibit this behavior
if a test file's blocks happen to be allocated across regions of file
system metadata.
Also, replace the open coded counting of extents and holes with a
simpler call to _count_extents(), and clarify some comments.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Reviewed-by: Namjae Jeon <namjae.jeon@samsung.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
xfs/042 is really an xfs_fsr test, and it only writes about 60MB of
data. however, because it is trying to write lots of data in ENOSPC
conditions, it can take a long time as XFS does flushes to maximise
space usage at ENOSPC. e.g. on a slow 1p VM:
xfs/042 426s ... 425s
It takes a long time to write a small amount of data. To avoid this
slow flushing problem, create the fragmented files with fallocate()
rather than write(), which fails much faster as there is no dirty
data to flush before retrying the allocation and failing. THis
results in:
xfs/042 425s ... 11s
A massive reduction in runtime, such that we can consider putting
xfs/042 into the quick group....
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Test that when we have a file with multiple hard links belonging to
different parent directories, if we remove one of those links, fsync the
file using one of its other links (that has a parent directory different
from the one we removed a link from), power fail and then replay the
fsync log/journal, the hard link we removed is not available anymore and
all the filesystem metadata is in a consistent state.
This test is motivated by an issue found in btrfs, where the test fails
with:
generic/107 2s ... - output mismatch (see .../results/generic/107.out.bad)
--- tests/generic/107.out 2015-08-04 09:47:46.922131256 +0100
+++ /home/fdmanana/git/hub/xfstests/results//generic/107.out.bad
@@ -1,3 +1,5 @@
QA output created by 107
Entries in testdir:
foo2
+foo3
+rmdir: failed to remove '/home/fdmanana/btrfs-tests/scratch_1/testdir': Directory not empty
...
(Run 'diff -u tests/generic/107.out .../generic/107.out.bad' to see the entire diff)
_check_btrfs_filesystem: filesystem on /dev/sdc is inconsistent (see .../generic/107.full)
_check_dmesg: something found in dmesg (see .../generic/107.dmesg)
$ cat /home/fdmanana/git/hub/xfstests/results//generic/107.full
_check_btrfs_filesystem: filesystem on /dev/sdc is inconsistent
*** fsck.btrfs output ***
checking extents
checking free space cache
checking fs roots
root 5 inode 257 errors 200, dir isize wrong
unresolved ref dir 257 index 3 namelen 4 name foo3 filetype 1 \
errors 5, no dir item, no inode ref
$ cat /home/fdmanana/git/hub/xfstests/results//generic/107.dmesg
(...)
[188897.707311] BTRFS info (device dm-0): failed to delete reference to \
foo3, inode 258 parent 257
[188897.711345] ------------[ cut here ]------------
[188897.713369] WARNING: CPU: 10 PID: 19452 at fs/btrfs/inode.c:3956 \
__btrfs_unlink_inode+0x182/0x35a [btrfs]()
[188897.717661] BTRFS: Transaction aborted (error -2)
(...)
[188897.747898] Call Trace:
[188897.748519] [<ffffffff8145f077>] dump_stack+0x4f/0x7b
[188897.749602] [<ffffffff81095de5>] ? console_unlock+0x356/0x3a2
[188897.750682] [<ffffffff8104b3b0>] warn_slowpath_common+0xa1/0xbb
[188897.751936] [<ffffffffa04c5d09>] ? __btrfs_unlink_inode+0x182/0x35a [btrfs]
[188897.753485] [<ffffffff8104b410>] warn_slowpath_fmt+0x46/0x48
[188897.754781] [<ffffffffa04c5d09>] __btrfs_unlink_inode+0x182/0x35a [btrfs]
[188897.756295] [<ffffffffa04c6e8f>] btrfs_unlink_inode+0x1e/0x40 [btrfs]
[188897.757692] [<ffffffffa04c6f11>] btrfs_unlink+0x60/0x9b [btrfs]
[188897.758978] [<ffffffff8116fb48>] vfs_unlink+0x9c/0xed
[188897.760151] [<ffffffff81173481>] do_unlinkat+0x12b/0x1fb
[188897.761354] [<ffffffff81253855>] ? lockdep_sys_exit_thunk+0x12/0x14
[188897.762692] [<ffffffff81174056>] SyS_unlinkat+0x29/0x2b
[188897.763741] [<ffffffff81465197>] system_call_fastpath+0x12/0x6f
[188897.764894] ---[ end trace bbfddacb7aaada8c ]---
[188897.765801] BTRFS warning (device dm-0): __btrfs_unlink_inode:3956: \
Aborting unused transaction(No such entry).
Tested against ext3/4, xfs, reiserfs and f2fs too, and all these
filesystems currently pass this test (on a 4.1 linux kernel at least).
The btrfs issue is fixed by the linux kernel patch titled:
"Btrfs: fix stale dir entries after removing a link and fsync".
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Now that generic/038 is running on my test machine, I notice how
slow it is:
generic/038 692s
11-12 minutes for a single test is way too long.
The test is creating
400,000 single block files, which can be easily parallelised and
hence run much faster than the test is currently doing.
Split the file creation up into 4 threads that create 100,000 files
each. 4 is chosen because XFS defaults to 4AGs, ext4 still has decent
speedups at 4 concurrent creates, and other filesystems aren't hurt
by excessive concurrency. The result:
generic/038 237s
on the same machine, which is roughly 3x faster and so it (just)
fast enough to to be considered acceptible.
[Eryu Guan: reduced number of files to minimum needed to reproduce
btrfs problem reliably, added $LOAD_FACTOR scaling for longer
running.]
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Ari Sundholm
Qu Wenruo
Jan Kara
Eric Sandeen
Eryu Guan
Zhao Lei
Filipe Manana
Zorro Lang
Darrick J. Wong
Eric Whitney
Dave Chinner