XFS torn log write detection includes a mechanism to inject CRC errors
into log records at runtime and shutdown the fs accordingly. This
ensures that the CRC verification pass on the subsequent mount discovers
an invalid record near the head of the log and considers it a torn
write.
This test runs a workload with error injection enabled and verifies that
the subsequent mount is successful. The test repeats for several
iterations using a random frequency factor for the error event each
time.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Once upon a time, ext4 encounter NULL pointer dereference under this
situaiton due to jump to a wrong label. Part of this commit fixed
this Oops:
744692d ext4: use ext4_get_block_write in buffer write
We can also run this test on ext2/3.
This case is based on a script from Monakhov Dmitriy @ openvz.
Cc: Monakhov Dmitriy <dmonakhov@openvz.org>
Signed-off-by: Xiong Zhou <xzhou@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Occasionally scsi_debug cannot be removed because it's still in use and
causes xfs/279 to fail.
Now dryrun the removal by modprobe firstly then do the real rmmod.
Signed-off-by: Eryu Guan <eguan@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
If the test fails after removing a device and before adding it back, it
attempts to add back the device in its _cleanup() function. However this
is broken because the device identifier is stored in a variable local to
the function _test_replace() and not in a global variable. So make the
variable global instead of local.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The test was using $SCRATCH_MNT as a mountpoint for $SCRATCH_DEV, which
is counter intuitive and not expected by the fstests framework - this
made the test fail after commit 27d077ec0b (common: use mount/umount
helpers everywhere). So rewrite the test to use the scratch device for
all data and use a test specific directory inside $TEST_DIR to use as a
mount point for a cross mount of $SCRATCH_DEV.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The test was using $SCRATCH_MNT as a mountpoint for $SCRATCH_DEV, which
is counter intuitive and not expected by the fstests framework - this
made the test fail after commit 27d077ec0b (common: use mount/umount
helpers everywhere). So rewrite the test to use the scratch device for
all data and use a test specific directory inside $TEST_DIR to use as a
mount point for a cross mount of $SCRATCH_DEV.
This test was also overriding $seqres.full, through the redirect ">"
operator, if a call to cp failed. Fix that by using instead the operator
">>". Also make the test use the function _mount() instead of calling
the mount program directly.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Create a $name.fsxops file next to $test.fsxlog. When a test fails,
dump the operations in the log into that file in a simple, parseable
format like:
fallocate 0x2e0f2 0xf04a 0x0 keep_size
truncate 0x0 0x11e00 0x0 *
write 0x73400 0x6c00 0x11e00
skip punch_hole 0x71539913 0xdf76 0x7a000 close_open
mapread 0x56000 0x16d08 0x7a000
Here, each operation is on a separate line. When the first word is
"skip", the operation will be skipped. The next parameters are offset,
length, and the current file size, followed by optional flags like
keep_size and clode_open. A trailing asterisk indicates that the
operation overlaps with the operation that has failed.
Add a --replay-ops option that allows to replay the operations recorded
in such a $name.fsxops file. (The log can be modified to easily narrow
down which operations are causing the failure.)
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Add FL_SKIPPED, FL_CLOSE_OPEN, and FL_KEEP_SIZE flags to the log
entries. Use FL_SKIPPED to indicate that an operation was skipped. Use
FL_CLOSE_OPEN to encode when an operation is followed by a close/open
cycle. Use FL_KEEP_SIZE to indicate when the OP_ZERO_RANGE and
OP_FALLOCATE operations should keep the file size unchanged and put the
current file size into args[2] so that we can tell which operation was
actually called from the log.
After that, arg2 of log4 is always either unused or the current file size, so
remove it and unconditionally remember the current file size.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Generate all test parameters in test(), including keep_size.
The code is slightly more complicated than it could be to produce the
same sequence of operations for the same random seed.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Some hex numbers are prefixed with "0x" and right-aligned with spaces,
leading to output like "0x beef". Make that "0x0beef" instead.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Move the run_fsx shell function into common/rc. Fix it to avoid
duplicate output on errors. Write the actual fsx parameters used into
$seqres.full instead of the BSIZE and PSIZE placeholders.
Include the symbolic fallocate mode in fsx error messages instead of the
numeric value. Use fprintf(stderr, ...) instead of warn() when
including strerror(errno) doesn't make sense.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
xfs/242 fails if the mapping flags show unaligned extents;
fix up the regexp to allow this, we really only care about
the unwritten flag.
Signed-off-by: eric Sandeen <sandeen@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Test what happens when we send largeish buffers to CoW all at once.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Test CoW operations when blocksize < pagesize and the only reflink
block is in the middle of the page.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
These tests examine the behavior of advanced and tricky copy on write
situations.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
See what happens when we ENOSPC while growing a btree on behalf of
some reflink operation.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Add more tests for unaligned copy-on-write things, and explicitly
test the ability to pass "len == 0" to mean reflink/dedupe all
the way to the end of the file".
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The free blocks count can vary from our calculations by up to 8% on a
1k-block filesystem, so permit that.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Fix the error messages in the golden output for generic/15[78], which
examine the responses to invalid inputs as returned by the
clone/clone_range/extent_same ioctls. Also fix a filtering omission.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Don't leave cruft behind on the test device's filesystem, so as to
avoid filling it with debris.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Add two scripts: "nextid" finds the next available test ID number in a
group, and "mvtest" relocates a test, fixes the golden output, and
moves the group entry for that test.
v2: sorting group files should preserve group order; nextid should use
the same algorithm as new; move both tools to tools/.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Brian Foster
Xiao Yang
Xiong Zhou
Eryu Guan
Filipe Manana
Andreas Gruenbacher
Eric Sandeen
Darrick J. Wong