On Mon, 2024-12-23 at 10:39 +0800, Zhang Yi wrote: > From: Zhang Yi <yi.zhang@xxxxxxxxxx> > > This addresses a data corruption issue encountered during partial > page > zeroing in ext4 which the block size is smaller than the page size > [1]. > Expand this test to include a zeroing range test that spans two > partial > pages to cover this case. > > Link: > https://lore.kernel.org/linux-ext4/20241220011637.1157197-2-yi.zhang@xxxxxxxxxxxxxxx/ > [1] > Signed-off-by: Zhang Yi <yi.zhang@xxxxxxxxxx> > --- > tests/generic/567 | 50 +++++++++++++++++++++++++-------------- > ---- > tests/generic/567.out | 18 ++++++++++++++++ > 2 files changed, 47 insertions(+), 21 deletions(-) > > diff --git a/tests/generic/567 b/tests/generic/567 > index fc109d0d..756280e8 100755 > --- a/tests/generic/567 > +++ b/tests/generic/567 > @@ -4,43 +4,51 @@ > # > # FS QA Test No. generic/567 > # > -# Test mapped writes against punch-hole to ensure we get the data > -# correctly written. This can expose data corruption bugs on > filesystems > -# where the block size is smaller than the page size. > +# Test mapped writes against punch-hole and zero-range to ensure we > get > +# the data correctly written. This can expose data corruption bugs > on > +# filesystems where the block size is smaller than the page size. > # > # (generic/029 is a similar test but for truncate.) > # > . ./common/preamble > -_begin_fstest auto quick rw punch > +_begin_fstest auto quick rw punch zero > > # Import common functions. > . ./common/filter > > _require_scratch > _require_xfs_io_command "fpunch" > +_require_xfs_io_command "fzero" > > testfile=$SCRATCH_MNT/testfile > > _scratch_mkfs > /dev/null 2>&1 Since this test requires block size < page size, do you think it is a good idea to hard code the _scratch_mkfs parameters to explicitly pass the block size to < less than zero? This will require less manipulation with the local.config file. Or maybe have a _notrun to _notrun the test if the block size is not less than the page size? > _scratch_mount > > -# Punch a hole straddling two pages to check that the mapped write > after the > -# hole-punching is correctly handled. > - > -$XFS_IO_PROG -t -f \ > --c "pwrite -S 0x58 0 12288" \ > --c "mmap -rw 0 12288" \ > --c "mwrite -S 0x5a 2048 8192" \ > --c "fpunch 2048 8192" \ > --c "mwrite -S 0x59 2048 8192" \ > --c "close" Minor: isn't the close command redundant? xfs_io will in any case close the file right? > \ > -$testfile | _filter_xfs_io > - > -echo "==== Pre-Remount ===" > -_hexdump $testfile > -_scratch_cycle_mount > -echo "==== Post-Remount ==" > -_hexdump $testfile > +# Punch a hole and zero out straddling two pages to check that the > mapped > +# write after the hole-punching and range-zeroing are correctly > handled. > +_straddling_test() > +{ > + local test_cmd=$1 > + > + $XFS_IO_PROG -t -f \ > + -c "pwrite -S 0x58 0 12288" \ > + -c "mmap -rw 0 12288" \ > + -c "mwrite -S 0x5a 2048 8192" \ > + -c "$test_cmd 2048 8192" \ > + -c "mwrite -S 0x59 2048 8192" \ > + -c "close" \ > + $testfile | _filter_xfs_io > + > + echo "==== Pre-Remount ===" > + _hexdump $testfile > + _scratch_cycle_mount > + echo "==== Post-Remount ==" > + _hexdump $testfile Just guessing here: Do you think it is makes sense to test with both delayed and non-delayed allocation? I mean with and without "msync"? > +} > + > +_straddling_test "fpunch" > +_straddling_test "fzero" Minor: Since we are running 2 independant sub-tests, isn't it better to use 2 different files? --NR > > status=0 > exit > diff --git a/tests/generic/567.out b/tests/generic/567.out > index 0e826ed3..df89b8f3 100644 > --- a/tests/generic/567.out > +++ b/tests/generic/567.out > @@ -17,3 +17,21 @@ XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX > ops/sec) > 002800 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 > 58 >XXXXXXXXXXXXXXXX< > * > 003000 > +wrote 12288/12288 bytes at offset 0 > +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) > +==== Pre-Remount === > +000000 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 > 58 >XXXXXXXXXXXXXXXX< > +* > +000800 59 59 59 59 59 59 59 59 59 59 59 59 59 59 59 > 59 >YYYYYYYYYYYYYYYY< > +* > +002800 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 > 58 >XXXXXXXXXXXXXXXX< > +* > +003000 > +==== Post-Remount == > +000000 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 > 58 >XXXXXXXXXXXXXXXX< > +* > +000800 59 59 59 59 59 59 59 59 59 59 59 59 59 59 59 > 59 >YYYYYYYYYYYYYYYY< > +* > +002800 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 > 58 >XXXXXXXXXXXXXXXX< > +* > +003000
