From the kernel patch that this test examines ("xfs: detect agfl
count corruption and reset agfl"):
"The struct xfs_agfl v5 header was originally introduced with
unexpected padding that caused the AGFL to operate with one less
slot than intended. The header has since been packed, but the fix
left an incompatibility for users who upgrade from an old kernel
with the unpacked header to a newer kernel with the packed header
while the AGFL happens to wrap around the end. The newer kernel
recognizes one extra slot at the physical end of the AGFL that the
previous kernel did not. The new kernel will eventually attempt to
allocate a block from that slot, which contains invalid data, and
cause a crash.
"This condition can be detected by comparing the active range of the
AGFL to the count. While this detects a padding mismatch, it can
also trigger false positives for unrelated flcount corruption. Since
we cannot distinguish a size mismatch due to padding from unrelated
corruption, we can't trust the AGFL enough to simply repopulate the
empty slot.
"Instead, avoid unnecessarily complex detection logic and and use a
solution that can handle any form of flcount corruption that slips
through read verifiers: distrust the entire AGFL and reset it to an
empty state. Any valid blocks within the AGFL are intentionally
leaked. This requires xfs_repair to rectify (which was already
necessary based on the state the AGFL was found in). The reset
mitigates the side effect of the padding mismatch problem from a
filesystem crash to a free space accounting inconsistency."
This test exercises the reset code by mutating a fresh filesystem to
contain an agfl with various list configurations of correctly wrapped,
incorrectly wrapped, not wrapped, and actually corrupt free lists; then
checks the success of the reset operation by fragmenting the free space
btrees to exercise the agfl. Kernels without this reset fix will shut
down the filesystem with corruption errors.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
Reduce semtimedop timeout to 5s, 15s is too long if something get
tangled up. Add retry counting to getlk routine, infinite loop is
dangerous. If something goes wrong unexpextedly, test is blocked and
wasting time.
Signed-off-by: Xiong Zhou <xzhou@redhat.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
The _supports_xfs_scrub helper is called with a mountpoint (a working
mountpoint is required for scrub) and a block device (used to detect
norecovery mounts). If either of these conditions aren't satisfied we
should return failure status to the caller, not unilaterally decide to
bail out of the test. In particular, the -b test doesn't work if the
fs has already shutdown on us.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
xfs_scrub takes an xfs mountpoint as its argument, not a block
device. Therefore, fix _check_xfs_filesystem to call it correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
SUSE systems export the NDBM interface via /usr/include/ndbm.h
This means that dbtest wasn't built on SUSE systems until commit
2353022 (build: update AC_PACKAGE_WANT_GDBM() and src/dbtest.c to
build). That change was incompatible with older SLES releases due to
<ndbm.h> and <gdbm.h> both defining the datum type, resulting in
build failures.
Nothing has ever set HAVE_GDBM_H or checked for the <gdbm.h> header
prior to the above commit, and it's not required on SUSE systems
either, so just remove that entirely.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
The AC_PACKAGE_WANT_GDBM macro is not easily read. It's not doing
anything particularly complex other than working through a set of
alternatives for headers and libraries.
This patch cleans it up to be more readable. We also only attempt
to check in libgdbm_compat if the checks in libgdbm fail.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
We currently check if libgdbm_compat contains the required symbols
even if libgdbm does. Let's fall back only when necessary (which is
pretty much always anyway).
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
This test fails on btrfs due to the presence of delayed processing
of file deletes if the file is smaller than 32mb. Initially commit
97575acd74 tried to fix a similar
failure by bumping the size of the filesystem. However that change
had a knock-on effect in that the scratch filesystem created is
larger than 100mb and thus not created in mixed mode. This in turn
causes the fs to have only 20mb for file data (rest is taken by DUP
metadata). Naturally, this leads to file freeing taking up to
"transaction commit interval" (default 30 s) time to properly account
the freed space.
Not standards define when unlink operations should be accounted so
btrfs is well within its right to be implemented in that way. So
to avoid this edge case just issue a sync before taking the 2nd
free space reading.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
Typically, when following absolute redirect, if an opauqe dentry is
found, lookup in further lower directories is stopped. But if a child
dentry has another absolute redirect, then lookup in further lower
layers should continue.
Say, following is example setup.
upper: /redirect (redirect=/a/b/c)
lower1: /a/[b]/c ([b] is opaque) (c has absolute redirect=/a/b/d/)
lower0: /a/b/d/foo
"redirect" directory in upper should merge with lower1:/a/b/c/ and
lower0:/a/b/d/, despite lower1:/a/b/ being opaque.
This example and kernel fix has come from Amir Goldstein. I am just
putting a test for it to make sure its not broken down the line.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
The regression is introduced to btrfs in linux v4.4 and it refuses to
create new files after log replay by returning -EEXIST.
Although the problem is on btrfs only, there is no btrfs stuff in terms
of test, so this makes it generic.
The kernel fix is
Btrfs: fix unexpected -EEXIST when creating new inode
[Eryu: add _require_metadata_journaling rule and 'log' 'metadata' group]
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
In the case of compression, each 128K input data chunk will be
compressed to 4K (because of the characters written are duplicate).
Therefore we have to write (128K * 16) to make sure every stripe can be
hit.
Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
Because of commit e76e13ce8c ("fsstress: implement the
clonerange/deduperange ioctls"), dedupe makes the number of references
to the same extent item increase so much that the default 4K buffer of
logical-resolve is no longer sufficient.
Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
Test that if we have a file with two hard links in the same parent
directory, then remove of the links, create a new file in the same
parent directory and with the name of the link removed, fsync the new
file and have a power loss, mounting the filesystem succeeds.
This test is motivated by a bug found in btrfs, which is fixed by
the linux kernel patch titled:
"Btrfs: fix log replay failure after unlink and link combination"
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
Test that when a fsync journal/log exists, if we rename a special file
(fifo, symbolic link or device), create a hard link for it with its old
name and then commit the journal/log, if a power loss happens the
filesystem will not fail to replay the journal/log when it is mounted
the next time.
This test is motivated by a bug found in btrfs, which is fixed by the
following patch for the linux kernel:
"Btrfs: fix log replay failure after linking special file and fsync"
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
Tests that use _overlay_scratch_mount_dirs instead of _scratch_mount
should use _require_scratch_nocheck instead of _require_scratch
because these tests are either mounting with multiple lower dirs or
mounting with non-default lower/upper/work dir, so
_check_overlay_scratch_fs won't handle these cases correctly. So we
introduce _overlay_check_scratch_dirs helper and should call this
helper with the correct dir arguments for these non-default cases.
This patch modify these tests to optionally call
_overlay_check_scratch_dirs at the end of the test or after
_scratch_umount to mount base filesystem only and run the checker.
Signed-off-by: zhangyi (F) <yi.zhang@huawei.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
No post-test check of the overlay dirs is required if case leaves
corrupt filesystem after test. We shoud use _require_scratch_nocheck()
instead of _require_scratch() in these cases.
Signed-off-by: zhangyi (F) <yi.zhang@huawei.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
_check_overlay_scratch_fs() will check lowerdir of overlay filesystem,
this case remove this directory after test will lead to check failure,
and it is not really necessary to remove this directory, so keep this
directory.
Signed-off-by: zhangyi (F) <yi.zhang@huawei.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
Hook filesystem check helper to _check_test_fs and _check_scratch_fs
for checking consistency of underlying dirs of overlay filesystem.
These helpers works only if fsck.overlay exists.
This patch introduce OVERLAY_FSCK_OPTIONS use for check overlayfs like
OVERLAY_MOUNT_OPTIONS, and also introduce a mount point check helper in
common/rc to detect a dir is a mount point or not.
[ _check_test_fs/_check_scratch_fs part picked from Amir's patch, thanks ]
Signed-off-by: zhangyi (F) <yi.zhang@huawei.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
There is a problem of missing fstype check in _is_mounted() helper,
it will return the mountpoint if only the device arguments matches.
For example:
Base mounted filesystem:
/dev/sda2 on /boot type ext4 (rw,relatime,data=ordered)
FSTYPE=xfs
mountpoint=`_is_mounted /dev/sda1`
echo "$mountpoint"
Output: /boot
This patch rename _is_mounted to _is_dev_mounted because it check
the given device only (not mount dir), and add an optional "fstype"
parameter, let user specify file system type instead of default
FSTYPE. Finally, use findmnt instead of mount to avoid complex
processing of mount info and fix this problem simply.
Signed-off-by: zhangyi (F) <yi.zhang@huawei.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
Evidently ceph will report a 4M blocksize, which trips clonerange_f's
clumsy attempt to avoid reflinking an extent on top of itself. The
original code assumed that "pick a random destination up to 1MB past the
end of the file" would suffice, but that clearly won't with a 4M
blocksize. Instead, we'll change it to 1024*blksize.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Tested-by: Luis Henriques <lhenriques@suse.com>
Reviewed-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Eryu Guan <eguan@redhat.com>
generic/472 is for changes that are not upstream and seem dead in
the water at the moment, so remove this test from the auto and quick
groups until it's been resolved upstream and the changes merged.
generic/473 really doesn't seme useful. FIEMAP is a debugging
interface, xfs_io is a debugging tool and so trying to make every
filesytem report exactly the same thing for a ranged query just
strikes me as the wrong thing to be doing. This fails on XFS, and
there's no apparent resolution to that in sight, so remove the test
from the auto and quick, too.
Signed-Off-By: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Eryu Guan <eguan@redhat.com>
According to the bash man page:
OPTIND is initialized to 1 each time the shell or a shell
script is invoked.
This doesn't appear to be true - in tests scripts with no other
getopts calls, I'm seeing the getopts loop in _xfs_check to fail to
parse input parameters correctly. Tracing shows the parameters are
being passed to _xfs_check correctly, but on occassion getopts
simply doesn't see them.
Hence when running tests with both external log and real time
devices, tests are failing at random because xfs_check is
mis-parsing the parameters passed to it and not configuring the
external log correctly:
_check_xfs_filesystem: filesystem on /dev/sdg is inconsistent (c)
*** xfs_check output ***
aborting - no external log specified for FS with an external log
*** end xfs_check output
Fix this by ensuring OPTIND is correctly initialised before using
getopts. Do it for all places that call getopts that don't already
set OPTIND=1 before starting their parsing loop.
Signed-Off-By: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Eryu Guan <eguan@redhat.com>
For 64k block size, the require code fails because the block range
[4k, 12k] would cause the fcollapse syscall to return -EINVAL. Hence
the tests using them are not executed.
This commit fixes the issue by calculating file offset ranges based
on the block size of the underlying filesystem.
Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
Reviewed-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Eryu Guan <eguan@redhat.com>
Darrick J. Wong
Amir Goldstein
Xiong Zhou
Jeff Mahoney
Nikolay Borisov
Vivek Goyal
Liu Bo
Lu Fengqi
Filipe Manana
zhangyi (F)
Dave Chinner
Chandan Rajendra