Re: [PATCH 5/5] xfs: Check correctness of metadump/mdrestore's ability to work with dirty log

[Chandan Babu R <chandanbabu@kernel.org>]

On Tue, Jan 02, 2024 at 09:44:00 PM -0800, Darrick J. Wong wrote:
> On Tue, Jan 02, 2024 at 02:13:52PM +0530, Chandan Babu R wrote:
>> Add a new test to verify if metadump/mdrestore are able to dump and restore
>> the contents of a dirty log.
>> 
>> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
>> ---
>>  tests/xfs/801     | 115 ++++++++++++++++++++++++++++++++++++++++++++++
>>  tests/xfs/801.out |  14 ++++++
>>  2 files changed, 129 insertions(+)
>>  create mode 100755 tests/xfs/801
>>  create mode 100644 tests/xfs/801.out
>> 
>> diff --git a/tests/xfs/801 b/tests/xfs/801
>> new file mode 100755
>> index 00000000..3ed559df
>> --- /dev/null
>> +++ b/tests/xfs/801
>> @@ -0,0 +1,115 @@
>> +#! /bin/bash
>> +# SPDX-License-Identifier: GPL-2.0
>> +# Copyright (c) 2024 Oracle, Inc.  All Rights Reserved.
>> +#
>> +# FS QA Test 801
>> +#
>> +# Test metadump/mdrestore's ability to dump a dirty log and restore it
>> +# correctly.
>> +#
>> +. ./common/preamble
>> +_begin_fstest auto quick metadump log logprint punch
>> +
>> +# Override the default cleanup function.
>> +_cleanup()
>> +{
>> +	cd /
>> +	rm -r -f $tmp.*
>> +	_scratch_unmount > /dev/null 2>&1
>> +}
>> +
>> +# Import common functions.
>> +. ./common/dmflakey
>> +. ./common/inject
>> +
>> +# real QA test starts here
>> +_supported_fs xfs
>> +_require_scratch
>> +_require_test
>> +_require_xfs_debug
>> +_require_xfs_io_error_injection log_item_pin
>> +_require_dm_target flakey
>> +_require_xfs_io_command "pwrite"
>> +_require_test_program "punch-alternating"
>> +
>> +metadump_file=${TEST_DIR}/${seq}.md
>> +testfile=${SCRATCH_MNT}/testfile
>> +
>> +echo "Format filesystem on scratch device"
>> +_scratch_mkfs >> $seqres.full 2>&1
>> +
>> +max_md_version=1
>> +_scratch_metadump_v2_supported && max_md_version=2
>> +
>> +external_log=0
>> +if [[ $USE_EXTERNAL = yes && -n "$SCRATCH_LOGDEV" ]]; then
>> +	external_log=1
>> +fi
>> +
>> +if [[ $max_md_version == 1 && $external_log == 1 ]]; then
>> +	_notrun "metadump v1 does not support external log device"
>> +fi
>> +
>> +echo "Initialize and mount filesystem on flakey device"
>> +_init_flakey
>> +_load_flakey_table $FLAKEY_ALLOW_WRITES
>> +_mount_flakey
>> +
>> +echo "Create test file"
>> +$XFS_IO_PROG -s -f -c "pwrite 0 5M" $testfile >> $seqres.full
>> +
>> +echo "Punch alternative blocks of test file"
>> +$here/src/punch-alternating $testfile
>> +
>> +echo "Mount cycle the filesystem on flakey device"
>> +_unmount_flakey
>> +_mount_flakey
>> +
>> +device=$(readlink -f $FLAKEY_DEV)
>> +device=$(_short_dev $device)
>> +
>> +echo "Pin log items in the AIL"
>> +echo 1 > /sys/fs/xfs/${device}/errortag/log_item_pin
>
> _scratch_inject_error log_item_pin 1
>

_scratch_inject_error() internally uses $SCRATCH_DEV. This is not suitable for
the test since we are working with $FLAKEY_DEV.

>> +
>> +echo "Create two checkpoint transactions on ondisk log"
>> +for ct in $(seq 1 2); do
>> +	offset=$(xfs_io -c 'fiemap' $testfile | tac |  grep -v hole | \
>
> $XFS_IO_PROG
>

I will fix it up.

>> +			 head -n 1 | awk -F '[\\[.]' '{ print $2 * 512; }')
>> +	$XFS_IO_PROG -c "truncate $offset" -c fsync $testfile
>> +done
>> +
>> +echo "Drop writes to filesystem from here onwards"
>> +_load_flakey_table $FLAKEY_DROP_WRITES
>> +
>> +echo "Unpin log items in AIL"
>> +echo 0 > /sys/fs/xfs/${device}/errortag/log_item_pin
>
> _scratch_inject_error log_item_pin 0
>
>> +
>> +echo "Unmount filesystem on flakey device"
>> +_unmount_flakey
>> +
>> +echo "Clean up flakey device"
>> +_cleanup_flakey
>> +
>> +echo -n "Filesystem has a "
>> +_print_logstate
>> +
>> +echo "Create metadump file, restore it and check restored fs"
>> +for md_version in $(seq 1 $max_md_version); do
>> +	[[ $md_version == 1 && $external_log == 1 ]] && continue
>> +
>> +	version=""
>> +	if [[ $max_md_version == 2 ]]; then
>> +		version="-v $md_version"
>> +	fi
>> +
>> +	_scratch_xfs_metadump $metadump_file -a -o $version
>> +	_scratch_xfs_mdrestore $metadump_file
>
> Don't you want to mdrestore to a file or something so that the next
> iteration of the loop will still be dumping from the exact same fs?

Yes, you are right. I will make the required changes.

-- 
Chandan