This test originally used zero range operations to reproduce problematic indirect delalloc reservations on XFS. Zero range has since been updated to include a pagecache flush before it updates extents, which means delalloc extents are converted to real extents and indlen reservations are drained before extents are modified. As a result, zero range is no longer an effective tool to reproduce the original problem. Instead, a debug mode failure injection mechanism has been added to XFS to cause dropped writes and trigger the delalloc extent hole punch behavior required to reproduce this problem. Update the test to use this new mechanism. Signed-off-by: Brian Foster <bfoster@xxxxxxxxxx> --- tests/xfs/289 | 28 +++++++++++++++++++--------- tests/xfs/289.out | 4 +--- tests/xfs/group | 2 +- 3 files changed, 21 insertions(+), 13 deletions(-) diff --git a/tests/xfs/289 b/tests/xfs/289 index 33cf060..3aa53b9 100755 --- a/tests/xfs/289 +++ b/tests/xfs/289 @@ -44,6 +44,7 @@ _cleanup() # get standard environment, filters and checks . ./common/rc +. ./common/punch # real QA test starts here rm -f $seqres.full @@ -52,33 +53,42 @@ rm -f $seqres.full _supported_fs generic _supported_os Linux _require_scratch -_require_xfs_io_command "fzero" +_require_xfs_sysfs $(_short_dev $TEST_DEV)/drop_writes _scratch_mkfs >/dev/null 2>&1 _scratch_mount +sdev=$(_short_dev $SCRATCH_DEV) file=$SCRATCH_MNT/file.$seq bytes=$((64 * 1024)) # create sequential delayed allocation $XFS_IO_PROG -f -c "pwrite 0 $bytes" $file >> $seqres.full 2>&1 -# Zero every other 4k range to split the larger delalloc extent into many more -# smaller extents. Use zero instead of hole punch because the former does not -# force writeback (and hence delalloc conversion). It can simply discard -# delalloc blocks and convert the ranges to unwritten. +# Enable write drops. All buffered writes are dropped from this point on. +echo 1 > /sys/fs/xfs/$sdev/drop_writes + +# Write every other 4k range to split the larger delalloc extent into many more +# smaller extents. Use pwrite because with write failures enabled, all +# preexisting delalloc blocks in the range of the I/O are tossed without +# discretion. This allows manipulation of the delalloc extent without conversion +# to real blocks (and thus releasing the indirect reservation). endoff=$((bytes - 4096)) for i in $(seq 0 8192 $endoff); do - $XFS_IO_PROG -c "fzero -k $i 4k" $file >> $seqres.full 2>&1 + $XFS_IO_PROG -c "pwrite $i 4k" $file >> $seqres.full 2>&1 done -# now zero the opposite set to remove remaining delalloc extents +# now pwrite the opposite set to remove remaining delalloc extents for i in $(seq 4096 8192 $endoff); do - $XFS_IO_PROG -c "fzero -k $i 4k" $file >> $seqres.full 2>&1 + $XFS_IO_PROG -c "pwrite $i 4k" $file >> $seqres.full 2>&1 done +echo 0 > /sys/fs/xfs/$sdev/drop_writes + +echo "Silence is golden." + _scratch_cycle_mount -hexdump $file +$XFS_IO_PROG -c 'bmap -vp' $file | _filter_bmap status=0 exit diff --git a/tests/xfs/289.out b/tests/xfs/289.out index bdcf195..72e60f9 100644 --- a/tests/xfs/289.out +++ b/tests/xfs/289.out @@ -1,4 +1,2 @@ QA output created by 289 -0000000 0000 0000 0000 0000 0000 0000 0000 0000 -* -0010000 +Silence is golden. diff --git a/tests/xfs/group b/tests/xfs/group index 5c1ba16..7c188fe 100644 --- a/tests/xfs/group +++ b/tests/xfs/group @@ -280,7 +280,7 @@ 283 dump ioctl auto quick 284 auto quick dump copy db mkfs repair 287 auto dump quota quick -289 auto quick rw zero +289 auto quick rw 290 auto rw prealloc quick ioctl zero 291 auto repair 292 auto mkfs quick -- 2.7.4 -- To unsubscribe from this list: send the line "unsubscribe linux-xfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
