[3/3] various: test is not appropriate for always_cow mode

Commit Message

Darrick J. Wong Jan. 18, 2023, 12:43 a.m. UTC

From: Darrick J. Wong <[e-mail]>

When always_cow mode is enabled, thes tests cannot set up the
preconditions for the functionality that they wants to test and should
be skipped.

Signed-off-by: Darrick J. Wong <[e-mail]>
---
 tests/generic/392 |   10 ++++++++++
 tests/xfs/326     |   12 ++++++++++++
 tests/xfs/558     |    6 ++++++
 3 files changed, 28 insertions(+)

Comments

Christoph Hellwig Jan. 18, 2023, 5:43 a.m. UTC | #1

Looks good:

Reviewed-by: Christoph Hellwig <[e-mail]>

Patch

diff --git a/tests/generic/392 b/tests/generic/392
index ac4014ab0c..c4bb3f4b92 100755
--- a/tests/generic/392
+++ b/tests/generic/392
@@ -28,6 +28,16 @@  _scratch_mkfs >/dev/null 2>&1
 _require_metadata_journaling $SCRATCH_DEV
 _scratch_mount
 
+# This test requires that i_blocks remains unchanged from the start of the
+# check_inode_metadata call until after recovery is complete.  fpunch calls
+# turn into pagecache writes if the arguments are not aligned to the fs
+# blocksize.  If the range being punched is already mapped to a written extent
+# and alwayscow is enabled, i_blocks will increase by the size of the COW
+# staging extent.  This causes stat to report different numbers for %b, which
+# results in a test failure.  Hence do not run this test if XFS is in alwayscow
+# mode.
+test "$FSTYP" = "xfs" && _require_no_xfs_always_cow
+
 testfile=$SCRATCH_MNT/testfile
 
 # check inode metadata after shutdown
diff --git a/tests/xfs/326 b/tests/xfs/326
index 8ab60684bf..ac620fc433 100755
--- a/tests/xfs/326
+++ b/tests/xfs/326
@@ -43,6 +43,18 @@  _scratch_mount >> $seqres.full
 _require_congruent_file_oplen $SCRATCH_MNT $blksz
 $XFS_IO_PROG -c "cowextsize $sz" $SCRATCH_MNT
 
+# This test uses a very large cowextszhint to manipulate the COW fork to
+# contain a large unwritten extent before injecting the error.  The goal is
+# that the write() will succeed, writeback will flush the dirty data to disk,
+# and writeback completion will shut down the filesystem when it tries to
+# remove the staging extent record from the refcount btree.  In other words,
+# this test ensures that XFS always finishes a COW completion it has started.
+#
+# This test isn't compatible with always_cow mode because the hole in the COW
+# fork left by the first write means that writeback tries to allocate a COW
+# staging extent for an unshared extent and trips over the injected error.
+_require_no_xfs_always_cow
+
 echo "Create files"
 _pwrite_byte 0x66 0 $sz $SCRATCH_MNT/file1 >> $seqres.full
 _cp_reflink $SCRATCH_MNT/file1 $SCRATCH_MNT/file2
diff --git a/tests/xfs/558 b/tests/xfs/558
index 819faf22bc..9e9b3be867 100755
--- a/tests/xfs/558
+++ b/tests/xfs/558
@@ -114,6 +114,12 @@  _require_xfs_io_error_injection "wb_delay_ms"
 _require_scratch_reflink
 _require_cp_reflink
 
+# This test races writeback of a pure overwrite of a data fork extent against
+# the creation of a speculative COW preallocation.  In alwayscow mode, there
+# are no pure overwrites, which means that a precondition of the test is not
+# satisfied, and this test should be skipped.
+_require_no_xfs_always_cow
+
 _scratch_mkfs >> $seqres.full
 _scratch_mount >> $seqres.full