On 2021/11/12 18:51, Filipe Manana wrote:
On Fri, Nov 12, 2021 at 2:36 AM Qu Wenruo <wqu@xxxxxxxx> wrote:Since kernel commit d4088803f511 ("btrfs: subpage: make lzo_compress_pages() compatible"), lzo compression no longer respects the max compressed page limit, and can cause kernel crash. The fix is titled "btrfs: fix a out-of-boundary access for copy_compressed_data_to_page()". This patch will add such regression test for compress-force=lzo mount option against incompressible data. Signed-off-by: Qu Wenruo <wqu@xxxxxxxx> --- tests/btrfs/049 | 41 +++++++++++++++++++++++++++++++++++++++++ tests/btrfs/049.out | 1 + 2 files changed, 42 insertions(+) create mode 100755 tests/btrfs/049 diff --git a/tests/btrfs/049 b/tests/btrfs/049 new file mode 100755 index 00000000..5a73f738 --- /dev/null +++ b/tests/btrfs/049 @@ -0,0 +1,41 @@ +#! /bin/bash +# SPDX-License-Identifier: GPL-2.0 +# Copyright (C) 2021 SUSE Linux Products GmbH. All Rights Reserved. +# +# FS QA Test 049 +# +# Test if btrfs will crash when compress-force=lzo hits incompressible data +# +. ./common/preamble +_begin_fstest auto quick compress dangerous + +# Override the default cleanup function. +# _cleanup() +# { +# cd / +# rm -r -f $tmp.* +# } + +# Import common functions. +# . ./common/filter + +# real QA test starts here + +# Modify as appropriate. +_supported_fs btrfs +_require_scratch + +_scratch_mkfs >> $seqres.full +_scratch_mount -o compress-force=lzo + +$XFS_IO_PROG -f -c "pwrite -i /dev/urandom 0 4k" $SCRATCH_MNT/file > /dev/nullWhile at it, instead of just having the minimum to trigger that specific bug, it would also be good to: 1) Try with zlib and zstd, instead of only lzo. This way it will help more easily catch any future regression with those algorithms; 2) Instead of redirecting to /dev/null and ignoring if we are actually writing what we are supposed to do, can we verify that we write all the bytes - i.e. that there's no short write? 3) After unmount, we could also mount the fs again and check if we are able to read exactly what we tried to write - that the write was persisted and we're not reading from the page cache.
That would be a pretty good generic advice for all regression tests. Would try to keep these advices in mind for all future tests. Thanks, Qu
Thanks.+ +# With kernel commit d4088803f511 ("btrfs: subpage: make lzo_compress_pages() +# compatible"), the kernel would crash when writing back above data. +_scratch_unmount + +echo "Silence is golden" + +# success, all done +status=0 +exit diff --git a/tests/btrfs/049.out b/tests/btrfs/049.out index cb0061b3..c69568ad 100644 --- a/tests/btrfs/049.out +++ b/tests/btrfs/049.out @@ -1 +1,2 @@ QA output created by 049 +Silence is golden -- 2.33.0
