xfstests-dev.git
5 years agogeneric: unaligned direct AIO write test
Zorro Lang [Sat, 30 Mar 2019 15:14:47 +0000 (23:14 +0800)]
generic: unaligned direct AIO write test

A simply reproducer from Frank Sorenson:

  ftruncate(fd, 65012224)
  io_prep_pwrite(iocbs[0], fd, buf[0], 104857663963648);
  io_prep_pwrite(iocbs[1], fd, buf[1], 104857665012224);

  io_submit(io_ctx, 1, &iocbs[0]);
  io_submit(io_ctx, 1, &iocbs[1]);

  io_getevents(io_ctx, 2, 2, events, NULL)

help to find an ext4 corruption:
           **************** **************** ****************
           *    page 1    * *    page 2    * *    page 3    *
           **************** **************** ****************
  existing 0000000000000000 0000000000000000 0000000000000000
  write 1    AAAAAAAAAAAAAA AA
  write 2                     BBBBBBBBBBBBBB BB

  result   00AAAAAAAAAAAAAA 00BBBBBBBBBBBBBB BB00000000000000
  desired  00AAAAAAAAAAAAAA AABBBBBBBBBBBBBB BB00000000000000

This issue remind us we might miss unaligned AIO test for long time.
We thought fsx cover this part, but looks like it's not. So this case
trys to cover unaligned direct AIO write test on file with different
initial truncate i_size.

The following patches fix the issue on xfs and ext4.

xfs: serialize unaligned dio writes against all other dio writes
ext4: Fix data corruption caused by unaligned direct AIO

Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofsx: fix errors due to unsupported FIDEDUPERANGE
Jan Kara [Thu, 28 Mar 2019 10:51:19 +0000 (11:51 +0100)]
fsx: fix errors due to unsupported FIDEDUPERANGE

Older kernels (prior commit 494633fac7896 "vfs: vfs_dedupe_file_range()
doesn't return EOPNOTSUPP") will return EINVAL when operation is not
supported. Make fsx treat this error as a sign of unsupported
deduplication as well to make it usable with these older kernels.

Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoxfs/065: make sure SCRATCH_DEV is unmount before mkfs
Zorro Lang [Wed, 27 Mar 2019 07:45:09 +0000 (15:45 +0800)]
xfs/065: make sure SCRATCH_DEV is unmount before mkfs

Commit "8309b39a fstests: fix broken _require_scratch usage" did
below change on xfs/065:

  -_scratch_unmount 2>/dev/null
  +_scratch_mkfs_xfs >> $seqres.full

It cause xfs/065 always fails now, as:
  QA output created by 065
  mkfs.xfs: /dev/sdb2 contains a mounted filesystem
  ...

So use _require_scratch, to make sure the SCRATCH_DEV is unmounted
before mkfs.

Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric/077: Don't delete $seqres.full file after test
Nikolay Borisov [Tue, 26 Mar 2019 15:01:58 +0000 (17:01 +0200)]
generic/077: Don't delete $seqres.full file after test

When this test finishes there is no 077.full file with output from
commands.  Sometimes this information is useful for post mortem so
stop deleting the file upon test completion.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric: prohibit fstrim on journalled filesystems with norecovery
Darrick J. Wong [Sat, 23 Mar 2019 00:35:32 +0000 (17:35 -0700)]
generic: prohibit fstrim on journalled filesystems with norecovery

This test makes sure that we can't use stale unrecovered fs metadata to
drive a DISCARD festival on a disk and thereby destroy user data by
accident.

The following patches fixed the bug on ext4, xfs and btrfs
ext4: prohibit fstrim in norecovery mode
xfs: prohibit fstrim in norecovery mode
Btrfs: do not allow trimming when a fs is mounted with the nologreplay option

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofstests: Verify that removed device has its superblocks deleted
Nikolay Borisov [Mon, 25 Mar 2019 14:07:13 +0000 (16:07 +0200)]
fstests: Verify that removed device has its superblocks deleted

When a device is removed from a btrfs filesystem its superblock copies
must be deleted. This test ensures this is indeed the case.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric: test stale data exposure after writeback crash
Brian Foster [Mon, 25 Mar 2019 17:17:25 +0000 (13:17 -0400)]
generic: test stale data exposure after writeback crash

XFS has historically had a stale data exposure window if a crash
occurs after a delalloc->physical extent conversion but before
writeback completes to the associated extent. While this should be a
rare occurrence in production environments due to typical writeback
ordering and such, it is not guaranteed in all cases until data
extents are initialized as unwritten (or otherwise zeroed) before
they are written.

Add a test that performs selective writeback ordering to reproduce
stale data exposure after a crash. Note that this test currently
fails on XFS.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofstests: Add missing call to _scratch_dev_pool_put
Nikolay Borisov [Mon, 25 Mar 2019 14:52:21 +0000 (16:52 +0200)]
fstests: Add missing call to _scratch_dev_pool_put

Every call to _scratch_dev_pool_get must be paired with call to
_scratch_dev_pool_put otherwise the SCRATCH_POOL variable will have
less devices than it actually must.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoxfs: test xfs_copy and xfs_mdrestore on the populate images
Darrick J. Wong [Sun, 24 Mar 2019 06:48:24 +0000 (23:48 -0700)]
xfs: test xfs_copy and xfs_mdrestore on the populate images

Make sure that copy, metadump, and mdrestore work on a filesystem with
all known metadata types.

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>
5 years ago_require_prjquota: Disable tests only when using realtime fs
Chandan Rajendra [Wed, 20 Mar 2019 14:21:12 +0000 (19:51 +0530)]
_require_prjquota: Disable tests only when using realtime fs

$USE_EXTERNAL needs to be set when using external log devices. In such a
setup, tests which have "_require_prjquota
$SCRATCH_DEV" (e.g. generic/383)  incorrectly end up being marked as
"not run" since the test "[ "$USE_EXTERNAL" = yes -a ! -z "$_dev" ]"
evaluates to true.

This commit fixes the bug by marking the test as "not run" only when
$USE_EXTERNAL is set and one of $TEST_RTDEV or $SCRATCH_RTDEV is set.

Signed-off-by: Chandan Rajendra <chandan@linux.ibm.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofstests: fix broken _require_scratch usage
Darrick J. Wong [Wed, 20 Mar 2019 00:46:01 +0000 (17:46 -0700)]
fstests: fix broken _require_scratch usage

_require_scratch doesn't actually format the scratch device with
anything, which means that tests are required to format them before
using them.  Fix tests that don't do this 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>
5 years agocommon/populate: support multiple cached images
Darrick J. Wong [Wed, 20 Mar 2019 00:45:35 +0000 (17:45 -0700)]
common/populate: support multiple cached images

Enhance the populated fs metadump image cache to support multiple
configurations per filesystem so that we reduce the image creation
overhead even further.

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>
5 years agocommon/populate: refactor _scratch_populate_cached
Darrick J. Wong [Wed, 20 Mar 2019 00:45:28 +0000 (17:45 -0700)]
common/populate: refactor _scratch_populate_cached

Refactor _scratch_populate_cached into smaller helper functions.

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>
5 years agoext4/023: don't require scrub for ext4 populated image creation
Darrick J. Wong [Wed, 20 Mar 2019 00:45:22 +0000 (17:45 -0700)]
ext4/023: don't require scrub for ext4 populated image creation

Don't require scrub for ext4's populated fs creation test because there
is no general online scrub program for ext*.

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>
5 years agogeneric/454: stop the test if we run out of space
Darrick J. Wong [Wed, 20 Mar 2019 00:45:07 +0000 (17:45 -0700)]
generic/454: stop the test if we run out of space

Certain filesystems (ext4 w/ 1k block size) can run out of space while
running this test because they have very limited xattr storage
capabilities.  If we run out of space while setting an attr, don't
bother continuing the test.

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>
5 years agogeneric/032: fix unwritten extent checks
Darrick J. Wong [Wed, 20 Mar 2019 00:45:01 +0000 (17:45 -0700)]
generic/032: fix unwritten extent checks

Fix the unwritten extent detector in this test to ignore post-eof
allocations because those are harmless.

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>
5 years agogeneric/042: fix stale disk contents check
Darrick J. Wong [Wed, 20 Mar 2019 00:44:55 +0000 (17:44 -0700)]
generic/042: fix stale disk contents check

This test doesn't call fsync or sync to force writeback of the first 60k
of the file, which means that we could end up with a file full of
zeroes or an empty file.  Since this is a regression test that looks for
stale disk contents slipping through, change the test to look for the
stale bytes.

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>
5 years agocheck: really improve test list randomization
Darrick J. Wong [Wed, 20 Mar 2019 00:44:49 +0000 (17:44 -0700)]
check: really improve test list randomization

coreutils provides the shuf(1) utility that randomizes the order of a
list and seeds its random number generator with /dev/urandom.  It's a
bit speedier than awk, so use it if available.

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>
5 years agocheck: improve test list randomization
Darrick J. Wong [Wed, 20 Mar 2019 00:44:42 +0000 (17:44 -0700)]
check: improve test list randomization

awk doesn't have a particularly good random number generator -- it seeds
from the Unix epoch time in seconds, which means that the run order
across a bunch of VMs started at exactly the same time are unsettlingly
predictable.  Therefore, at least try to seed it with bash's $RANDOM,
which is slightly less predictable.

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>
5 years agogeneric/131: Create server.out manually.
zhangxiaoxu 00467499 [Tue, 19 Mar 2019 06:27:55 +0000 (14:27 +0800)]
generic/131: Create server.out manually.

When test generic/131 on nfs, the 'server.out' maybe create
later than expect. Because the server is running on background,
we should ensure the 'server.out' is exist before 'cat' it.

So, let's create the server.out manually.

Signed-off-by: ZhangXiaoxu <zhangxiaoxu5@huawei.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric/398: accept EXDEV for rename or link into encrypted dir
Eric Biggers [Fri, 15 Mar 2019 00:05:47 +0000 (17:05 -0700)]
generic/398: accept EXDEV for rename or link into encrypted dir

Update generic/398 to pass after kernel commit f5e55e777cc9 ("fscrypt:
return -EXDEV for incompatible rename or link into encrypted dir"),
which intentionally changed some error codes from EPERM to EXDEV in
order to allow standard tools like 'mv' to move files into an encrypted
directory.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoxfs/42[01]: don't disturb unwritten status with md5sum
Christoph Hellwig [Mon, 18 Feb 2019 09:19:51 +0000 (10:19 +0100)]
xfs/42[01]: don't disturb unwritten status with md5sum

The way we decided if an unwritten extent is considered a hole or
data is by checking if the page and/or blocks are marked uptodate,
that is contain valid data in the page cache.

xfs/420 and xfs/421 try to exercise SEEK_HOLE / SEEK_DATA in the
presence of cowextsize preallocations over holes in the data fork.
The current XFS code never actually uses those for buffer writes,
but a pending patch changes that.  For SEEK_HOLE / SEEK_DATA to work
properly in that case we also need to look at the COW fork in their
implementations and thus have to rely on the unwritten extent page
cache probing.  But the tests for it ensure we do have valid data in
the pagecache by calling md5sum on the test files, and thus reading
their contents (including the zero-filled holes) in, and thus making
them all valid data.

Fix that by dropping the page cache content after the md5sum calls.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric: test i_mode recovery after power failure
Chao Yu [Tue, 5 Mar 2019 11:47:44 +0000 (19:47 +0800)]
generic: test i_mode recovery after power failure

After fsync, filesystem should guarantee inode metadata including
permission info being persisted, so even after sudden power-cut,
during mount, we should recover i_mode fields correctly, in order
to not loss those meta info.

So adding this testcase to check whether generic filesystem can
guarantee that.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric: add test for fsync after shrinking truncate and rename
Filipe Manana [Mon, 4 Mar 2019 14:06:22 +0000 (14:06 +0000)]
generic: add test for fsync after shrinking truncate and rename

Test that if we truncate a file to reduce its size, rename it and then
fsync it, after a power failure the file has a correct size and name.

This test is motivated by a bug found in btrfs, which is fixed by a
patch for the linux kernel titled:

  "Btrfs: fix incorrect file size after shrinking truncate and fsync"

This test currently passes on ext4, xfs, f2fs and patched btrfs.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric/097: split user.* xattr tests to new test
Ronnie Sahlberg [Fri, 8 Mar 2019 03:34:38 +0000 (13:34 +1000)]
generic/097: split user.* xattr tests to new test

Split out most of the user.* tests from 097 and move them to a new
test that only tests user.* xattrs.

This makes it possible to use this test on filesystems that can only
provide user.* xattrs such as CIFS.

Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agopopulate: force large finobt creation on xfs
Darrick J. Wong [Mon, 4 Mar 2019 21:19:41 +0000 (13:19 -0800)]
populate: force large finobt creation on xfs

Teach the populate routines to create enough inodes that we end up with
multi-level inode btrees.

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>
5 years agogeneric: test statx attribute_mask setting
Darrick J. Wong [Mon, 4 Mar 2019 21:19:35 +0000 (13:19 -0800)]
generic: test statx attribute_mask setting

Make sure the filesystem reports attribute_mask for the attributes it
supports.

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>
5 years agoext4/032: SCRATCH_DIR -> SCRATCH_MNT
Darrick J. Wong [Tue, 5 Mar 2019 19:42:42 +0000 (11:42 -0800)]
ext4/032: SCRATCH_DIR -> SCRATCH_MNT

Use SCRATCH_MNT, not SCRATCH_DIR.

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>
5 years agoshared/298: unmount filesystem before examining underlying storage
Darrick J. Wong [Mon, 4 Mar 2019 21:19:29 +0000 (13:19 -0800)]
shared/298: unmount filesystem before examining underlying storage

This test does some weird things with live filesystems -- it seems to be
validating the behavior of fstrim by comparing the filesystem's free
space map to holes in the file image that backs the filesystem.
However, this doesn't account for the fact that some filesystems
maintain in-core preallocations and/or can perturb the free space data
during unmount.  This causes sporadic test failures when the two become
out of sync.

Therefore, make sure we unmount the filesystem before we start running
tools against the filesystem image file to eliminate the possibility of
changes to the free space map.  This was found by running shared/298 on
xfs with a 1k block size.

cc: enwlinux@gmail.com
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>
5 years agoxfs/010: use correct type for finobt corrupting
Darrick J. Wong [Mon, 4 Mar 2019 21:19:22 +0000 (13:19 -0800)]
xfs/010: use correct type for finobt corrupting

Use 'type finobt' for corrupting the finobt.

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>
5 years agocommon/rc: fix get_max_lfs_filesize
Darrick J. Wong [Mon, 4 Mar 2019 21:19:08 +0000 (13:19 -0800)]
common/rc: fix get_max_lfs_filesize

Helper functions are supposed to have a leading underscore ('_') in the
function name, but this one doesn't have it.  Unfortunately, the calling
test cases (generic/349-351) /do/ have the leading underscore, so now
they're broken.

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>
5 years agogeneric/{436,445}: check falloc support
Amir Goldstein [Tue, 26 Feb 2019 14:09:01 +0000 (16:09 +0200)]
generic/{436,445}: check falloc support

The sanity test case in those tests (i.e. 13..17)
are all skipped in fs with no falloc support, but the tests
are reported to pass.

For example, from 445.full:

 File system supports the default behavior.
 File system does not support fallocate.
 Allocation size: 4096
 17. Test file with unwritten extents, data-hole-data inside page
 Test skipped as fs doesn't support unwritten extents.

Explicitly check for falloc support before running those tests
so they would be properly reported as skipped.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric/482: use thin volume as data device
Brian Foster [Thu, 28 Feb 2019 14:41:28 +0000 (09:41 -0500)]
generic/482: use thin volume as data device

The dm-log-writes replay mechanism issues discards to provide
zeroing functionality to prevent out-of-order replay issues. These
discards don't always result in zeroing bevavior, however, depending
on the underlying physical device. In turn, this causes test
failures on XFS v5 filesystems that enforce metadata log recovery
ordering if the filesystem ends up with stale data from the future
with respect to the active log at a particular recovery point.

To ensure reliable discard zeroing behavior, use a thinly
provisioned volume as the data device instead of using the scratch
device directly. This slows the test down slightly, but provides
reliable functional behavior at a reduced cost from active snapshot
management or forced zeroing.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agocommon/dmlogwrites: genericize log writes target device
Brian Foster [Thu, 28 Feb 2019 14:41:27 +0000 (09:41 -0500)]
common/dmlogwrites: genericize log writes target device

The dm-log-writes infrastructure is currently implemented to use
SCRATCH_DEV as a hardcoded data device. In preparation to allow use
of specialized devices in certain circumstances, genericize the code
to allow an arbitrary data device. This requires passing the target
device as a parameter to several helper functions from various
tests. No functional changes.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agot_attr_corruption: fix this yet again
Darrick J. Wong [Tue, 26 Feb 2019 02:35:35 +0000 (18:35 -0800)]
t_attr_corruption: fix this yet again

Jeff Mahoney pointed out that 'security.evm' actually has an expected
value format, which breaks the test if EVM is enabled.  It turns out
that the 'security.evm' setxattr call in the original syzkaller report
was a total red herring, as this bug can be reproduced without it.

Fix the test case to do the minimum amount of work needed to reproduce
the corruption.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofstests: don't oom the box opening tmpfiles
Darrick J. Wong [Tue, 26 Feb 2019 02:35:29 +0000 (18:35 -0800)]
fstests: don't oom the box opening tmpfiles

For the t_open_tmpfiles tests, limit ourselves to half of file-max
so that we don't OOM the test machine.

[Eryu: fix comments too to match the new limit]

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofsstress: show correct offsets and length for copy_file_range
Rostislav Skudnov [Tue, 26 Feb 2019 14:29:02 +0000 (14:29 +0000)]
fsstress: show correct offsets and length for copy_file_range

Copy original offsets and length and use them for logging as in
splice_f. Fix grammar mistakes in the comment about them.

Signed-off-by: Rostislav Skudnov <rostislav@tuxera.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agocommon/rc: add _get_max_lfs_filesize to return MAX_LFS_FILESIZE
Yufen Yu [Tue, 26 Feb 2019 14:11:54 +0000 (22:11 +0800)]
common/rc: add _get_max_lfs_filesize to return MAX_LFS_FILESIZE

Pick up the common function _get_max_lfs_filesize() to return
MAX_LFS_FILESIZE.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric/299: limit max file size
Yufen Yu [Tue, 26 Feb 2019 14:11:53 +0000 (22:11 +0800)]
generic/299: limit max file size

For some filesystem, such as vfat, the max support file size is 4G.
We limit the max size and let the test go on running.

Fix it by moving the function get_max_file_size() of generci/485 to
common/rc, and add the max filesize limit to generic/299.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoshared/298: Wire btrfs support in get_free_sectors
Nikolay Borisov [Fri, 22 Feb 2019 10:04:25 +0000 (12:04 +0200)]
shared/298: Wire btrfs support in get_free_sectors

Add support for btrfs in shared/298. Achieve this by introducing 2
new awk scripts that parse relevant btrfs structures and print holes.
Additionally modify the test to create larger - 3gb filesystem in the
case of btrfs. This is needed so that distinct block groups are used
for data and metadata.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agosrc/Makefile: Link clock_gettime(2) with -lrt
Xiao Yang [Thu, 21 Feb 2019 07:39:01 +0000 (15:39 +0800)]
src/Makefile: Link clock_gettime(2) with -lrt

Compiling t_open_tmpfiles.c failed on older glibc(before glibc v2.17)
because clock_gettime(2) was not linked with -lrt, as below:
--------------------------------------------------------------------
/home/yangxiao/xfstests/src/t_open_tmpfiles.c:36: undefined reference to `clock_gettime'
--------------------------------------------------------------------

According to clock_gettime(2) manpage, we should link clock_gettime(2)
with -lrt on older glibc.

Signed-off-by: Xiao Yang <yangx.jy@cn.fujitsu.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agosrc/t_attr_corruption: fix xattr.h include problems
Darrick J. Wong [Wed, 20 Feb 2019 17:20:05 +0000 (09:20 -0800)]
src/t_attr_corruption: fix xattr.h include problems

Apparently newer versions of libattr (which haven't yet been picked
up by Debian or Ubuntu) don't ship xattr.h anymore, because we're
supposed to use the libc version in sys/xattr.h.  So do that.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofstests: add a seek group
Christoph Hellwig [Mon, 18 Feb 2019 09:05:03 +0000 (10:05 +0100)]
fstests: add a seek group

This groups all tests exercising SEEK_DATA / SEEK_HOLE behavior.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoxfs/42[01]: remove from the dedup group
Christoph Hellwig [Mon, 18 Feb 2019 09:05:02 +0000 (10:05 +0100)]
xfs/42[01]: remove from the dedup group

No dedup functionality is exercised by these tests.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agosrc/t_open_tmpfiles: flush log when shutting down filesystem
Darrick J. Wong [Mon, 18 Feb 2019 18:13:59 +0000 (10:13 -0800)]
src/t_open_tmpfiles: flush log when shutting down filesystem

If the caller of t_open_tmpfiles wants to shut down the filesystem,
be sure to flush the log when we shut down so that log recovery will
have to process all the unlinked temporary files.

This is apparently needed to force ext4 to flush updated inode
blocks through the journal at all.

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>
5 years agoxfs/500: fix reflink support detection and add new groups
Darrick J. Wong [Tue, 19 Feb 2019 05:29:05 +0000 (21:29 -0800)]
xfs/500: fix reflink support detection and add new groups

Fix some problems detecting reflink support in the test.

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>
5 years agoxfs/450: require working falloc command
Darrick J. Wong [Tue, 19 Feb 2019 18:32:38 +0000 (10:32 -0800)]
xfs/450: require working falloc command

This test needs to check for working falloc command before using it.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoshared: cgroup aware writeback accounting test
Brian Foster [Tue, 12 Feb 2019 15:22:03 +0000 (10:22 -0500)]
shared: cgroup aware writeback accounting test

A test to perform reads/writes under various cgroups and verify that
I/Os are accounted properly according to cgroup aware writeback.
This is a generic test, but not all commonly used local filesystems
support cgroup aware writeback at the moment (i.e., XFS). Therefore,
this test currently requires ext4 or btrfs for the time being.

The common/cgroup2 file is copied from a separate cgroup related
patch from Shaohua Li that never made it upstream.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Shaohua Li <shli@fb.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofsstress: avoid infinite zero byte reading
Zorro Lang [Sat, 16 Feb 2019 07:25:40 +0000 (15:25 +0800)]
fsstress: avoid infinite zero byte reading

copyrange_f and splice_f functions use a while loop to read a file,
it's fine if there's only one fsstress process(and its children),
but if some third part testing processes remove the file in the
middle phase of copyrange_f running, copyrange_f maybe always return
0, and the while loop can't be end. As below:

root     47184  xxxxxx S+ ./fsstress -R -d /mnt/scratch -n 10000 -p 20 -v
root     47187  xxxxxx R+ ./fsstress -d /mnt/scratch -n 10000 -p 20 -v
root     47199  xxxxxx R+ ./fsstress -d /mnt/scratch -n 10000 -p 20 -v
root     47314  xxxxxx S+ grep --color=auto fsstress
...
...
copy_file_range(3, [372258], 4, [2658770], 71179, 0) = 0
copy_file_range(3, [372258], 4, [2658770], 71179, 0) = 0
copy_file_range(3, [372258], 4, [2658770], 71179, 0) = 0
copy_file_range(3, [372258], 4, [2658770], 71179, 0) = 0
...
...
lr-x------. 1 root root 64 Jan 28 11:34 /proc/47187/fd/3 -> '/mnt/scratch/p2/f2 (deleted)'

Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric: check the behavior of programs opening a lot of O_TMPFILE files
Darrick J. Wong [Wed, 13 Feb 2019 20:49:05 +0000 (12:49 -0800)]
generic: check the behavior of programs opening a lot of O_TMPFILE files

Create a test (+ helper program) that opens as many unlinked files as it
possibly can on the scratch filesystem, then closes all the files at
once to stress-test unlinked file cleanup.  Add an xfs-specific test to
make sure that the fallback code doesn't bitrot.

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>
5 years agoinject: skip tests when knob dir exists but knob doesn't
Darrick J. Wong [Tue, 12 Feb 2019 02:17:48 +0000 (18:17 -0800)]
inject: skip tests when knob dir exists but knob doesn't

If the XFS error injection knob directory exists but the knob itself
doesn't, then we know that this kernel doesn't support the knob and
can skip the test.

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>
5 years agogeneric: posix acl extended attribute memory corruption test
Darrick J. Wong [Wed, 13 Feb 2019 20:48:14 +0000 (12:48 -0800)]
generic: posix acl extended attribute memory corruption test

XFS had a use-after-free bug when xfs_xattr_put_listent runs out of
listxattr buffer space while trying to store the name
"system.posix_acl_access" and then corrupts memory by not checking
the seen_enough state and then trying to shove
"trusted.SGI_ACL_FILE" into the buffer as well.

In order to tickle the bug in a user visible way we must have
already put a name in the buffer, so we take advantage of the fact
that "security.evm" sorts before "system.posix_acl_access" to make
sure this happens.

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>
5 years agogeneric: check for reasonable inode creation time
Darrick J. Wong [Tue, 12 Feb 2019 02:17:39 +0000 (18:17 -0800)]
generic: check for reasonable inode creation time

If statx returns inode creation time (aka btime), check it to make
sure that the filesystem is setting a creation time that's
reasonably close to when it creates a file.

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>
5 years agocommon: fix _require_btime for lazy filesystems
Darrick J. Wong [Tue, 12 Feb 2019 02:17:32 +0000 (18:17 -0800)]
common: fix _require_btime for lazy filesystems

Filesystems are not required to try to fill the statx btime field
unless the caller actually sets STATX_BTIME.  They're allowed to
volunteer that information "if it's cheap", but XFS doesn't
volunteer and there may be filesystems that support btime but not
cheaply.

Either way, we want to test btime on any filesystem that supports
it, cheaply or otherwise, so set STATX_BTIME when we're trying to
detect support for it.

[Eryu: fix _require_scratch_btime too]

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>
5 years agocommon: fix kmemleak to work with sections
Darrick J. Wong [Tue, 12 Feb 2019 02:17:26 +0000 (18:17 -0800)]
common: fix kmemleak to work with sections

Refactor the kmemleak code to work correctly with sections.  This
requires changing the location of the "is kmemleak enabled?" flag to
use /tmp instead of RESULT_BASE, scanning for leaks after every
test, and clarifying which functions get used when.

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>
5 years agoxfs: test mkfs extent size hint validation
Darrick J. Wong [Sat, 16 Feb 2019 00:41:19 +0000 (16:41 -0800)]
xfs: test mkfs extent size hint validation

Make sure mkfs won't format filesystems that fail extent size hint
validation.

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>
5 years agogeneric/075,112: detect preallocation support for fsx tests
Christoph Hellwig [Fri, 15 Feb 2019 12:41:40 +0000 (13:41 +0100)]
generic/075,112: detect preallocation support for fsx tests

Currently generic/075 and generic/112 have two extra fsx passes each
that exercise fsx with preallocation, which are only enabled for
XFS.

These tests can also be run with other file systems, given that the
XFS prealloc ioctls are implemented in generic code since the
addition of the fallocate system call.  This also means a version of
XFS that does not support preallocation (e.g. because it always
writes out of place) can skip the prealloc tests while still
completing the normal fsx tests just fine.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs: test for corruption when reading compressed files
Filipe Manana [Thu, 14 Feb 2019 15:18:39 +0000 (15:18 +0000)]
btrfs: test for corruption when reading compressed files

Regression test for read corruption of compressed and shared extents
after punching holes into a file. The same extent is shared by the
same file in consecutive ranges (without other extents in between).

This is motivated by a bug recently found in btrfs for which there
is a patch for the linux kernel titled:

  "Btrfs: fix corruption reading shared and compressed extents after hole
   punching"

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric: test fsync after succession of renames and unlink
Filipe Manana [Tue, 12 Feb 2019 18:08:20 +0000 (18:08 +0000)]
generic: test fsync after succession of renames and unlink

Test that after a combination of file renames, linking and creating
a new file with the old name of a renamed file, if we fsync the new
file, after a power failure we are able to mount the filesystem and
all file names correspond to the correct inodes.

This test is motivated by a bug found in btrfs, which is fixed by
applying the following two patches to the linux kernel:

 "[PATCH 1/2] Btrfs: fix fsync after succession of renames of different files"
 "[PATCH 2/2] Btrfs: fix fsync after succession of renames and unlink/rmdir"

The test passes on ext4, xfs and patched btrfs, however at least in
a 5.0-rc5 linux kernel, it fails on f2fs.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric: test fsync after succession of file renames
Filipe Manana [Tue, 12 Feb 2019 18:08:07 +0000 (18:08 +0000)]
generic: test fsync after succession of file renames

Test that after a combination of file renames, linking and creating
a new file with the old name of a renamed file, if we fsync the new
file, after a power failure we are able to mount the filesystem and
all file names correspond to the correct inodes.

This test is motivated by a bug found in btrfs which is fixed by a
patch for the linux kernel titled:

  "Btrfs: fix fsync after succession of renames of different files"

The test passes on ext4, xfs and patched btrfs, however at least in
a 5.0-rc5 linux kernel, it fails on f2fs.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agosrc/t_mtab: Add error check for unlock_mtab()
Cui Yue [Tue, 12 Feb 2019 11:12:30 +0000 (19:12 +0800)]
src/t_mtab: Add error check for unlock_mtab()

When unlink() fails, that is, when the lock file is not deleted
successfully, variable we_created_lockfile is still set to 0.

On the next iteration, the 3 processes will not be able to
successfully create the lock file.

Signed-off-by: Cui Yue <cuiyue-fnst@cn.fujitsu.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoxfs/068: Verify actual file count instead of reported file count
Amir Goldstein [Tue, 5 Feb 2019 19:31:18 +0000 (21:31 +0200)]
xfs/068: Verify actual file count instead of reported file count

This test has the number of files/dirs created by xfsrestore hardcoded
in golden output.

When fsstress is added new ops, the number of files/dirs created with
the same random seed changes and this regularly breaks this test,
so when new fsstress ops are added they should be either added to the
dump test blacklist or golden output of this test needs to be ammended
to reflect the change.

The golden output includes only the file count reported by xfsrestore
and test does not even verify that this is the correct file count.
Instead, leave the golden output neutral and explicitly verify that
file count before and after the test are the same.

With this change, the test becomes agnostic to fsstress ops and we
could also stop blacklisting clone/dedup/copy ops if we want.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofstests: Check that high-offset reads and writes work on non-blockdev fs
Amir Goldstein [Tue, 5 Feb 2019 19:42:02 +0000 (21:42 +0200)]
fstests: Check that high-offset reads and writes work on non-blockdev fs

This is a variant of test generic/466 for filesystems that
do not support mkfs_sized

It is needed for testing high-offset reads and writes with overlayfs
over a basefs that supports huge files.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoxfs/252: requires fallocate support for preallocation
Christoph Hellwig [Tue, 5 Feb 2019 18:48:16 +0000 (19:48 +0100)]
xfs/252: requires fallocate support for preallocation

xfs/252 has a few feature tests, but misses checking for preallocation
support.  Because of that it will fail instead of not being run for
and XFS file system in always COW mode.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofsstress: avoid splice_f generating too large sparse file
Zorro Lang [Mon, 4 Feb 2019 15:48:09 +0000 (23:48 +0800)]
fsstress: avoid splice_f generating too large sparse file

Thanks to Darrick J. Wong find this issue! Current splice_f generates
file offset as below:

  lr = ((int64_t)random() << 32) + random();
  off2 = (off64_t)(lr % maxfsize);

It generates a pseudorandom 64-bit candidate offset for the
destination file where we'll land the splice data, and then caps the
offset at maxfsize (which is 2^63- 1 on x64), which effectively means
that the data will appear at a very high file offset which creates
large (sparse) files very quickly.

That's not what we want, and some case likes shared/009 will take
forever to run md5sum on lots of huge files.

Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agooverlay/060: Use falloc to make sure a meta copy file got copied up
Vivek Goyal [Mon, 4 Feb 2019 18:34:54 +0000 (13:34 -0500)]
overlay/060: Use falloc to make sure a meta copy file got copied up

Overlayfs might copy up data of file on first write of file (and
not necessarily upon open of file). So call falloc file opened
with O_RDWR and after that data must have been copied up.

[Eryu: add _require_xfs_io_command "falloc" to make sure underlying
fs have fallocate(2) support]

Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agocommon/dump: do not override test cleanup trap
Amir Goldstein [Sun, 27 Jan 2019 07:50:56 +0000 (09:50 +0200)]
common/dump: do not override test cleanup trap

Instead, call _cleanup_dump explicitly from a private _cleanup.
Remove the generic cleanup bits (rm $tmp.*) from _cleanup_dump.

The only xfs/dump test that had anything other than rm $tmp.* in
_cleanup in xfs/287, but that was _scratch_unmount, which is not
needed anyway.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoxfs/138: format the scratch device before using it
Darrick J. Wong [Tue, 29 Jan 2019 16:17:19 +0000 (08:17 -0800)]
xfs/138: format the scratch device before using it

Format the scratch device before using it, or else xfs_db will fail,
particularly if the previous test left a corrupt fs behind.

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>
5 years agoxfs/093: make sure the scratch directory still exists after repair
Darrick J. Wong [Tue, 29 Jan 2019 16:17:12 +0000 (08:17 -0800)]
xfs/093: make sure the scratch directory still exists after repair

Make sure that we still have the scratch directory after repairing our
corrupted filesystem, because repair could have nuked it.

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>
5 years agooverlay/061: enhance mmap ro/rw inconsistencies test
Amir Goldstein [Sun, 27 Jan 2019 08:38:50 +0000 (10:38 +0200)]
overlay/061: enhance mmap ro/rw inconsistencies test

overlay/061 is currently the only overlay test that is expected to
fail on upstream kernel.

It is a flavor of test overlay/016 with mread in stead of pread.
The ro/rw inconsistencies related to file read()/write() API were
fixed with stacked file operations in v4.19, but the ro/rw
inconsistencies related to shared mmap read/write remain to be
fixed.

The test currently checks cache coherency between mmap read and file
write(), but this sort of cache coherency is a Linux implementation
detail not a requirement of the API.

Instead of mread vs. pwrite, check consistency of mread vs. mwrite
to shared mmap, which is required by the MMAP_SHARED API.

Because we can, perform the test on shared memory that maps files
that are already close and check also that mwrite after file is
closed is persistent. This adds test coverage for future overlayfs
writeback code.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agooverlay: Do not lose security.capability xattr over metadata only file copy-up
Vivek Goyal [Wed, 30 Jan 2019 18:59:37 +0000 (13:59 -0500)]
overlay: Do not lose security.capability xattr over metadata only file copy-up

Extend test 064 to check security.capability xattr is not lost over
copy-up of a metadata only file. This requires mounting overlay with
option metacopy=on and first trigger metadata only copy-up and then
trigger data copy-up.

Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs: Test if btrfs will report false ENOSPC error balancing small metadata chunk
Qu Wenruo [Tue, 29 Jan 2019 12:32:22 +0000 (20:32 +0800)]
btrfs: Test if btrfs will report false ENOSPC error balancing small metadata chunk

This is a test case for a long existing bug, caused by
over-estimated metadata space_info::bytes_may_use.

There is one proposed patch for btrfs-progs to fix it, titled:
"btrfs-progs: balance: Sync the fs before balancing metadata chunks"

The test case itself is almost the same as btrfs/181, which uses
small files to bump the reserved space to trigger the false alert.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs: Test if btrfs will commit too many transactions for balance
Qu Wenruo [Tue, 29 Jan 2019 12:32:21 +0000 (20:32 +0800)]
btrfs: Test if btrfs will commit too many transactions for balance

Kernel commit 64403612b73a ("btrfs: rework
btrfs_check_space_for_delayed_refs") is introducing a regression for
btrfs balance performance.

Since that commit will cause btrfs to commit too many transactions
for nothing during balance/relocation, it will slow balance
dramatically even we only need to relocate several megabytes.

This test case will catch the problem by using super block
generation as failure criteria.

For small chunk relocated, we will commit 6 transactions for each
block group, and the test case should only have 2 block groups, it
should only commit 12 transactions.

This test case will use 120 as the threshold to detect the failure.

And in my test environment, with kernel fix btrfs committed 14
transactions. While without the fix btrfs committed 209
transactions.

So the test case should be enough to detect the regression, while still
keep the runtime small enough for failure.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs: Test if btrfs hits EDQUOT without trying to reclaim some space
Qu Wenruo [Tue, 29 Jan 2019 12:32:20 +0000 (20:32 +0800)]
btrfs: Test if btrfs hits EDQUOT without trying to reclaim some space

Commit a514d63882c3 ("btrfs: qgroup: Commit transaction in advance
to reduce early EDQUOT") is no longer forcing transaction commit to
reclaim space, and only commits transaction asynchronously in
advance to address it.

However the criteria used in async transaction commit is not
comprehensive, thus it doesn't reclaim space automatically.

This test case will check the behavior by:
1) Falloc a large padding file
   This file will take 90% of the qgroup limit

2) Sync the fs
   To reflect the qgroup changes

3) Delete the file
   Qgroup won't reclaim the space until transaction committed.

4) Try to write a file
   If kernel not fixed, qgroup will not automatically commit transaction
   to reclaim the freed space and hit EDQUOT.

This bug is going to be fixed by a patch for kernel titled:
"btrfs: qgroup: Make qgroup async transaction commit more aggressive".

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs/131: require support for free-space-tree
Jeff Mahoney [Mon, 21 Jan 2019 16:33:15 +0000 (11:33 -0500)]
btrfs/131: require support for free-space-tree

btrfs/131 tests the free space tree, which older kernels won't have.
We shouldn't run there.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs/023: skip trying to test raid56 without kernel support
Jeff Mahoney [Mon, 21 Jan 2019 16:33:14 +0000 (11:33 -0500)]
btrfs/023: skip trying to test raid56 without kernel support

Older kernels don't support raid56.  This test is still valid for
other profiles, so skip raid56 if the kernel doesn't support it.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs: require feature raid56 for raid56 tests
Jeff Mahoney [Mon, 21 Jan 2019 16:33:13 +0000 (11:33 -0500)]
btrfs: require feature raid56 for raid56 tests

btrfs/125, btrfs/148, btrfs/157, and btrfs/158 test for raid56
behavior.  We shouldn't run if the kernel doesn't have support for
them.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs/010: don't run without /sys/fs/btrfs
Jeff Mahoney [Mon, 21 Jan 2019 16:33:10 +0000 (11:33 -0500)]
btrfs/010: don't run without /sys/fs/btrfs

Older kernels don't have /sys/fs/btrfs.  btrfs/010 will happily run
until it goes to check its work against sysfs and finds those files
don't exist.  This patch introduces a require check to ensure that
the sysfs files are present before running.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs/081: fix killing of reader loop subshell
Filipe Manana [Fri, 25 Jan 2019 11:46:24 +0000 (11:46 +0000)]
btrfs/081: fix killing of reader loop subshell

The test creates a subshell that keeps running the 'cat' command
against a test file in an infinite loop, and after it kills the
subshell it unmounts the filesystem, after which point any 'cat'
subcommand that runs after or at that time will fail resulting in an
unexpected golden output:

  $ ./check btrfs/081
  btrfs/081 3s ... - output mismatch (see /home/fdmanana/git/hub/xfstests/results//btrfs/081.out.bad)
      --- tests/btrfs/081.out 2018-09-16 21:30:48.501104179 +0100
      +++ /home/fdmanana/git/hub/xfstests/results//btrfs/081.out.bad 2019-01-24 20:36:18.989746185 +0000
      @@ -206,5 +206,6 @@
       Verifying file digests after cloning
       14968c092c68e32fa35e776392d14523  SCRATCH_MNT/foo
       14968c092c68e32fa35e776392d14523  SCRATCH_MNT/bar
      +cat: /mnt/scratch/bar: No such file or directory
       Verifying target file digest after umount + mount
       14968c092c68e32fa35e776392d14523  SCRATCH_MNT/bar
      ...
      (Run 'diff -u /home/fdmanana/git/hub/xfstests/tests/btrfs/081.out /home/fdmanana/git/hub/xfstests/results//btrfs/081.out.bad'  to see the entire diff)
  Ran: btrfs/081
  Failures: btrfs/081
  Failed 1 of 1 tests

Fix that by adding a proper trap to the reader loop function so that
the subshell waits for executed 'cat' commands when it receives
SIGTERM.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs/081: declare local variables as local
Filipe Manana [Fri, 25 Jan 2019 11:46:11 +0000 (11:46 +0000)]
btrfs/081: declare local variables as local

Some variables inside the test's functions were used as local but
were not being declared as such. Add the local declaration for them.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agocommon/dump: disable splice from FSSTRESS_AVOID
Zorro Lang [Wed, 23 Jan 2019 07:34:55 +0000 (15:34 +0800)]
common/dump: disable splice from FSSTRESS_AVOID

New fsstress operation breaks fs dump/restore testing which use
fsstress, e.g xfs/068.

In _create_dumpdir_stress_num, disable splice in fsstress so that we
dump exactly the same set of files and directories.

Quote Dave's comments for future reference

"
fsstress is just creating regular files differently. It has no
impact on xfsdump does except to change the number of files created
and the directory layout.

If this new functionality were creating a new type of file that
xfsdump has to handle, or adding new attributes or changing the
metadata of the existing files, then we want to make sure xfsdump is
tested against that, and so we'd be changing the golden output after
careful checking that both xfsdump and xfs_restore are working
correctly and the file count is correct.

But when all we are doing is creating normal, regular files just
with a different syscall, it makes no sense to perturb the existing
test then we have to go and validate that the new set of files being
tested is actually scanned correctly, is complete and correct. Using
a blacklist to avoid unnecessary perturbation such as in cases like
this is the right thing to do because we've had to determine if the
new functionality is a useful addition to xfsdump/restore test
coverage or not.
"

[Eryu: add Dave's comments in commit log for future reference]

Signed-off-by: Zorro Lang <zlang@redhat.com>
Suggested-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofsstress: add splice support
Zorro Lang [Sat, 19 Jan 2019 02:42:57 +0000 (10:42 +0800)]
fsstress: add splice support

Support the splice syscall in fsstress.

Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agooverlay: File capabilities should not be lost over copy-up
Vivek Goyal [Wed, 16 Jan 2019 21:04:15 +0000 (16:04 -0500)]
overlay: File capabilities should not be lost over copy-up

Make sure file capabilities are not lost over copy-up when file is
opened for WRITE but nothing is actually written to it.

Following commit introduced regression where if a lower file with
CAP_SETUID is opened for writing, and capability is cleared over copy up.

bd64e57586d3 ("ovl: During copy up, first copy up metadata and then data")

A later kernel patch will fix it. This test will help avoid introducing
such regressions again.

Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric: test writepage cached mapping validity
Brian Foster [Tue, 15 Jan 2019 14:09:45 +0000 (09:09 -0500)]
generic: test writepage cached mapping validity

XFS has a bug where page writeback can end up sending data to the
wrong location due to a stale, cached file mapping. Add a test to
trigger this problem by racing background writeback with a
truncate/rewrite of the final page of the file.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs: test for deadlock between snapshot delete and other read-write operations
Qu Wenruo [Fri, 11 Jan 2019 05:01:51 +0000 (13:01 +0800)]
btrfs: test for deadlock between snapshot delete and other read-write operations

Commit fb235dc06fac ("btrfs: qgroup: Move half of the qgroup
accounting time out of commit trans") could cause ABBA deadlock
between backref lookup with write lock hold (subvolume deletion) and
other read/write operations.

It's going to be fixed by "btrfs: qgroup: Don't trigger backref walk
at delayed ref insert time".

This test will generate pwrite background workload, along with
constant subvolume creation and deletion to trigger the bug.

It needs some time to generate enough files to bump the tree height
to trigger the bug.

In my test environment, with 'unsafe' cache mode for the VM, it
triggers the bug at around 70~90 seconds. So I leave the default
runtime to 120s to make sure the bug will be triggered.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agobtrfs: Make seed device test cases into their own group
Qu Wenruo [Thu, 10 Jan 2019 06:14:14 +0000 (14:14 +0800)]
btrfs: Make seed device test cases into their own group

btrfs/16[123] are all seed device related test cases, make them into
'seed' group.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric: test that xattrs can have slashes in their names
Darrick J. Wong [Fri, 4 Jan 2019 23:14:48 +0000 (15:14 -0800)]
generic: test that xattrs can have slashes in their names

Eric Sandeen recently found a bug in xfs_repair that flagged extended
attribute names containing "/" as corrupt and purged them.  There's
nothing in the IRIX or Linux manuals that say anything about slashes not
being allowed (and Linux certainly allows this) so let's make sure this
continues to work.

[Eryu: use $SETFATTR and _getfattr helper]

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric/423: statx mask of the reference file is different from the original file
Cui Yue [Thu, 10 Jan 2019 03:14:59 +0000 (22:14 -0500)]
generic/423: statx mask of the reference file is different from the original file

When running xfstests generic/423 to test system call statx() on
hard link files of NFS, it fails.  error message:

[!] attr 'stx_mask' differs from ref file, 7ff != e0

The values of parameter "mask" between the original file and the
reference file are different.  One is STATX_ALL;
The other is STATX_ATIME | STATX_BTIME | STATX_CTIME | STATX_MTIME.

Modify the function get_reference() to pass the "mask" in, and
change STATX_ATIME | STATX_BTIME | STATX_CTIME | STATX_MTIME to
"mask".

Signed-off-by: Cui Yue <cuiyue-fnst@cn.fujitsu.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoRevert "common/config: create $RESULT_BASE before dumping kmemleak leaks"
Nikolay Borisov [Fri, 4 Jan 2019 14:31:16 +0000 (16:31 +0200)]
Revert "common/config: create $RESULT_BASE before dumping kmemleak leaks"

This commit tried to fix the brokennes of the kmemleak support but it
inadvertently broke the creation of the RESULT_BASE directory which lead to
problems creating check.time file. Turns out kmemleak support in xfstests has
more problems and it needs to be majorly refactor and this commit doesn't
really solve the problem. For the time being just revert to at least allow
older configuration files, which have explicitly set RESULT_BASE to work.

This reverts commit 7fc034868d5dce9aaf3517c9c2197221b9a772e4.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric/131: wait until the server is ready or timeout
Hou Tao [Fri, 4 Jan 2019 08:19:26 +0000 (16:19 +0800)]
generic/131: wait until the server is ready or timeout

When running xfstests under KVM VM and the load of host is high,
only delaying 1s and checking the readiness of server are not
enough, and the test case will fail early.

Fix it by repeatedly checking the readiness signal until it's found,
or timeout is triggered.

[Eryu: check if lock server died or not, like v1 patch did]

Signed-off-by: Hou Tao <houtao1@huawei.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agofsx: check ENOSYS in test_copy_range() & test_fallocate()
Hou Tao [Fri, 4 Jan 2019 08:19:27 +0000 (16:19 +0800)]
fsx: check ENOSYS in test_copy_range() & test_fallocate()

In configure script, we only check whether or not the build of test
program succeeds, but that doesn't mean the kernel has implemented
the syscall, so checking for this case.

Signed-off-by: Hou Tao <houtao1@huawei.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoxfs/139-140: skip testing on large scratch dev
Zorro Lang [Wed, 2 Jan 2019 04:03:19 +0000 (12:03 +0800)]
xfs/139-140: skip testing on large scratch dev

x/139 and x/140 makes XFS with very small agsize. That agsize is too
small for a large fs. And it's not necessary to test on large fs, so
skip it directly if scratch dev is large dev.

Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric/474: shift target directory to a sub-dir of SCRATCH_MNT
Zorro Lang [Wed, 2 Jan 2019 04:03:18 +0000 (12:03 +0800)]
generic/474: shift target directory to a sub-dir of SCRATCH_MNT

If testing on large fs (--large-fs option), there's a huge size
.use_space file in $SCRATCH_MNT, then `fssum $SCRATCH_MNT` trys to
read whole huge file. That's wasting time, so change the target path
to a sub-dir of $SCRATCH_MNT.

Signed-off-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoxfs: only set XFS_MKFS_HAS_NO_META_SUPPORT for XFS
Hou Tao [Fri, 4 Jan 2019 08:56:10 +0000 (16:56 +0800)]
xfs: only set XFS_MKFS_HAS_NO_META_SUPPORT for XFS

No need to set XFS_MKFS_HAS_NO_META_SUPPORT for all filesystems.

Cc: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Hou Tao <houtao1@huawei.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric/466: explicitly request $SCRATCH_DEV to be a block device
Hou Tao [Sat, 29 Dec 2018 02:34:51 +0000 (10:34 +0800)]
generic/466: explicitly request $SCRATCH_DEV to be a block device

so "blockdev --getsize64 $SCRATCH_DEV" will succeed.

Signed-off-by: Hou Tao <houtao1@huawei.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agogeneric/019: require scratch device to be a block device
Hou Tao [Sat, 29 Dec 2018 02:34:49 +0000 (10:34 +0800)]
generic/019: require scratch device to be a block device

To ensure "blockdev --getsz $SCRATCH_DEV" will succeed.

Signed-off-by: Hou Tao <houtao1@huawei.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agocheck: use _try_scratch_mount instead of _scratch_mount to mount SCRATCH_DEV
Hou Tao [Sat, 29 Dec 2018 02:34:48 +0000 (10:34 +0800)]
check: use _try_scratch_mount instead of _scratch_mount to mount SCRATCH_DEV

Else there won't be any error messages when mounting SCRATCH_DEV
failed, because _scratch_mount exits early by invoking _fail.

Signed-off-by: Hou Tao <houtao1@huawei.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoext4/034: adjust commit which fixes the problem tested by ext4/034
Theodore Ts'o [Wed, 26 Dec 2018 16:08:05 +0000 (11:08 -0500)]
ext4/034: adjust commit which fixes the problem tested by ext4/034

Also add a requirment that fallocate and fiemap is supported.
(Fallocate isn't the case when we are emulating ext3, for example.)

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: Liu Bo <obuil.liubo@gmail.com>
Reviewed-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
5 years agoxfs: look for stringified constants in ftrace formats
Darrick J. Wong [Tue, 18 Dec 2018 20:36:09 +0000 (12:36 -0800)]
xfs: look for stringified constants in ftrace formats

Look for uninterpretable stringified constants in the ftrace format
description for xfs tracepoints.

[Eryu: add $CC_PROG definition and require it in test, also use
$DEBUGFS_MNT instead of hard coded path]

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>
5 years agoxfs: filter out mount options that don't work on v4 filesystems
Darrick J. Wong [Tue, 18 Dec 2018 20:36:16 +0000 (12:36 -0800)]
xfs: filter out mount options that don't work on v4 filesystems

A few tests require v4 filesystems and enforce this by disabling
crc's in the _scratch_mkfs call.  However, if the user specified
MOUNT_OPTIONS that only work with v5 filesystems, these tests fail.
If we detect a test creating a v4 scratch filesystem, filter out
incompatible mount options that don't work on v4, such as
simultaneous group/project quota.

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>