Several tests happen to make use of loop device support without the
requisite pre-test checks. This results in spurious failures for systems
that might not have loop device support. Add _require_loop checks to
shared/298, xfs/206 and xfs/259.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Kill any lingering fsstress processes and wait properly should we abort
the test. This prevents the workload from inadvertently affecting
subsequent tests.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Create a stress test for the free inode btree. Allocate a set of inodes
sequentually and run a hard link clone and random replacement algorithm
across the set. Background removal of the oldest directories creates a
sparse set of free inodes over time. Run an fsstress workload
concurrently to exercise the fs.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The finobt creates a duplicate subset of inode allocation metadata from
the inobt. xfs_repair should detect and repair inconsistencies in the
finobt that could be caused by bugs or corruption. This test uses xfs_db
to cause targeted corruptions in the finobt and verify repair detects
and corrects the filesystem.
In particular, the test corrupts individual finobt records to cause
inconsistency between the inode allocation count fields as well as
causing the finobt to contain a record with no free inodes.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
xfs/030 nukes various on-disk data structures to test for repair. This
can result in extra output when testing finobt enabled filesystems. For
example, xfs_repair detects an invalid free inode btree root block when
the agi is zeroed.
Filter this output directly in xfs/030 such that the test passes for
finobt and non-finobt filesystems.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Number of helpers for checking xfs_io functionality is slowly
growing. But it's as easy to simply use _require_xfs_io_command()
directly and just specify the command we want to check. It will also
avoid the need to create helper every time we need to check a new
command in xfs_io.
Remove all the helpers and use _require_xfs_io_command() in the
tests.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
There is no mainline kernel support for DMAPI in XFS, and so every
time the xfstests auto group is run it throws a large number of
"[not run] Assuming DMAPI modules are not loaded". I've noted that
people are excluding the dmapi group to avoid this, so rather than
inflict pain on everyone, make hose few that need to do dmapi
testing include it specifically.
IOWs, remove the DMAPI tests from the auto group.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
A couple of tests leave behind large files or directory structures
when they complete, which leads to small TEST_DEVs running out of
space during other tests. Make those space hogs clean up after
themselves so that random tests don't fail with ENOSPC errors.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
When running on a ramdisk, the fsstress background workload consumes
a GB of disk space every 5 seconds. This leads to the test failing
with ENOSPC because the test file cannot be created due otthe
background load cosuming it all. Hence don't run this test unless
the scratch device is large enough not to hit ENOSPC conditions.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The Q_XQUOTARM quotactl was not working properly, because
we weren't passing in proper flags. The xfs_fs_set_xstate()
ioctl handler used the same flags for Q_XQUOTAON/OFF as
well as Q_XQUOTARM, but Q_XQUOTAON/OFF look for
XFS_UQUOTA_ACCT, XFS_UQUOTA_ENFD, XFS_GQUOTA_ACCT etc,
i.e. quota type + state, while Q_XQUOTARM looks only for
the type of quota, i.e. XFS_DQ_USER, XFS_DQ_GROUP etc.
Unfortunately these flag spaces overlap a bit, so we
got semi-random results for Q_XQUOTARM; i.e. the value
for XFS_DQ_USER == XFS_UQUOTA_ACCT, etc. yeargh.
Anyway, here's a simple test that demonstrates it,
kernel patch to fix it will follow.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
This test was written before a solution was in place, I think,
and so the expected output wasn't well tested.
The test does a loop of sparse writes from 6 to 0, but the
.out file expects 6 (not 7) extents. Fix it.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Create new function _test_block_boundaries() which is testing content of
the blocks after the operation such as zero, or punch hole. The test is
doing the operation around block boundaries to assure correct behaviour
of the operation on block unaligned ranges.
This has been based on test xfs/290 which has been changed to use this
new function. A small change to the output file was required.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
When running on a ramdisk, the fsstress background workload consumes
a GB of disk space every 5 seconds. This leads to the test failing
with ENOSPC because the test file cannot be created due otthe
background load cosuming it all. Hence don't run this test unless
the scratch device is large enough not to hit ENOSPC conditions.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
I'm running xfstests against a ramdisk, so I'm limited in size of
the test and scratch devices. While there are large enough to hold a
filesystem image with a 2GB log, the way the log changes position in
an image file as the size of the filesystem increases means that the
aggregated disk space of xfs/217 is more than enough to run a 4GB
TEST_DEV out of space and hence fail the test.
To avoid this problem, punch out the image file between every mkfs
iteration so that it only consumes the space needed by each
individual mkfs tests, not an aggregation of them all.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Version 5 filesystems always have attr2 format enabled, and it
cannot be turned off via the noattr2 mount option. As such, attempts
to mount with noattr2 will be rejected and this causes cascading
failures within the test.
Hence detect if we've created a CRC enabled filesystem, and if this
is the case _notrun the test.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
CRC enabled filesystems emit different errors on corruption.
Specifically, inode corruption is picked up much earlier due to
verifier failures (e.g. incorrect inode identifier) and so
xfs_repair throws errors sufficiently different that filtering
cannot hide the differences. Hence simply add a new golden output
file and link it appropriately once we know what type of filesystem
we are testing.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
the commit:
10e6e65 xfs: be more forgiving of a v4 secondary sb w/ junk in v5 fields
broke primary sb CRC validation, not erroring out the mount
if the crc was bad.
This tests that it's fixed, and properly fails the mount on
a bad crc.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Test that we properly ignore old growfs-induced junk in the unused
portion of secondary V4 superblocks; at one point this would
trip up the verifiers, and cause a subsequent growfs to fail.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Test the setting of the XFS BMBT fields via xfs_db. Runs through the
valid bit values for each field and tests an illegal value.
[dchinner: added _require_xfs_mkfs_crc and turned off crcs so that
the test doesn't just fail on CRC enabled test runs.]
[dchinner: added hex block values to check they don't get endian
swapped.]
Signed-off-by: Mark Tinguely <tinguely@sgi.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The xfs_db output is different for v5 filesystem metadata, and so
the test fails due to golden image mismatches rather than an actual
test failure. Improve the filter to hide the differences between the
metadata format outputs.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
xfs/299 has failed for me for a long time. In fact, looking at my
logs it has never passed on any of my test machines. IOWs, the
test that was committed was fundamentally broken.
The reason is that it tests project quotas before it tests user or
group quotas and so creates a bunch of files that are owned by root
or privileged users. It think tries to manipulate them as a user,
and, unsurprisingly, it fails to do so. This then causes the test to
throw an error.
The reason it has always failed is the error that is thrown
hardcodes a uid/gid into an error message. This uid/gid is what
causes the golden output mismatch (nobody is 65534 on my machines,
not 99):
*** push past the hard block limit (expect EDQUOT)
[ROOT] 0 0 0 00 [--------] 12 0 0 00 [--------] 0 0 0 00 [--------]
[NAME] =OK= 100 500 00 [--------] 7 4 10 00 [7 days] 0 0 0 00 [--------]
- URK 99: 0 is out of range! [425,500]
+ URK 65534: 0 is out of range! [425,500]
It wasn't until I looked at the xfs/299.full file when trying to
understand why the error was being thrown and whether it shoul dhave
been in the golden output in the first place that I saw the real
problem. That is, All the user/group quota modifications were
failing because of not having permissions to write the files left
behind by the quota test, and that user and group quotas were not
being tested at all by the test.
So, firstly $SCRATCH_MNT needs to be world writeable, and secondly
each test needs to remove the files it created during the test so
they don't impact on furture test iterations.
This then exercises the user and group quotas appropriately, and so
the golden output changes completely to reflect that changes under
user quotas are actually being accounted to the correct user.
Further, the error message that I originally saw errors on goes
away, because everything is now accounted correctly.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Ever since commit 7e2a19504 ("ls -l reports different file size
depending on platform and user.") xfs/066 has been running stat on
the dump/restore directory instead of the large file that the test
is checking can be dumped and restored correctly. IOWs, it's not
been checking the correct thing for almost 10 years.
This test fails on CRC enabled filesystems because the shortform
directory entry size is different (an extra byte for the filetype
filed), and this is where tracking down the failure has lead me.
Fix this by using the correct target file, and improve it by dumping
an md5sum of the source and target files to ensure they contain the
same data.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
This test never passed, so the golden output was never properly
verified as correct. Now that the bug is fixed, fix the golden
output to match the actual test output.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The first test may start with the file from the previous test, and
that is in an unknown state. Hence always remove the test file
before the first test so that it doesn't have extents inside the
test range as it is supposed to be testing into a hole.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Brian Foster
Lukas Czerner
Dave Chinner
Eric Sandeen
Hannes Frederic Sowa
Eric Sandeen
Mark Tinguely