Re: [PATCH] generic: check direct IO writes with io_uring and O_DSYNC are durable

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]



On Mon, Nov 07, 2022 at 09:38:58AM +0000, fdmanana@xxxxxxxxxx wrote:
> From: Filipe Manana <fdmanana@xxxxxxxx>
> 
> Test that direct IO writes with io_uring and O_DSYNC are durable if a power
> failure happens after they complete.
> 
> This is motivated by a regression on btrfs, affecting 5.15 stable kernels
> and kernels up to 6.0, where often the writes were not persisted (same
> behaviour as if O_DSYNC was not provided). This was recently fixed by the
> following commit:
> 
> 51bd9563b678 ("btrfs: fix deadlock due to page faults during direct IO reads and writes")
> 
> Signed-off-by: Filipe Manana <fdmanana@xxxxxxxx>
> ---
>  tests/generic/703     | 104 ++++++++++++++++++++++++++++++++++++++++++
>  tests/generic/703.out |   2 +
>  2 files changed, 106 insertions(+)
>  create mode 100755 tests/generic/703
>  create mode 100644 tests/generic/703.out
> 
> diff --git a/tests/generic/703 b/tests/generic/703
> new file mode 100755
> index 00000000..39ae3773
> --- /dev/null
> +++ b/tests/generic/703
> @@ -0,0 +1,104 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (C) 2022 SUSE Linux Products GmbH. All Rights Reserved.
> +#
> +# FS QA Test 703
> +#
> +# Test that direct IO writes with io_uring and O_DSYNC are durable if a power
> +# failure happens after they complete.
> +#
> +. ./common/preamble
> +_begin_fstest auto quick log prealloc io_uring
> +
> +_cleanup()
> +{
> +	_cleanup_flakey
> +	cd /
> +	rm -r -f $tmp.*
> +}
> +
> +. ./common/filter

This patch looks very good to me, it even doesn't miss any _require_* helpers
or group names, and its comments are good enough. I can't pick up any problems
from my side, just two tiny review points (as you will change the commit log
at least, so might change them by the way:).

1) This case doesn't use any filter functions, so don't need "common/filter".

> +. ./common/dmflakey
> +
> +fio_config=$tmp.fio
> +fio_out=$tmp.fio.out
> +test_file="${SCRATCH_MNT}/foo"
> +
> +[ $FSTYP == "btrfs" ] &&
> +	_fixed_by_kernel_commit 8184620ae212 \
> +	"btrfs: fix lost file sync on direct IO write with nowait and dsync iocb"
> +
> +_supported_fs generic
> +# We allocate 256M of data for the test file, so require a higher size of 512M
> +# which gives a margin of safety for a COW filesystem like btrfs (where metadata
> +# is always COWed).
> +_require_scratch_size $((512 * 1024))

2) I think nearly no one use a SCRATCH_DEV < 512M to run fstests, but I can't
say this's wrong, so you can decide to keep or remove this line by yourself.
Both are good to me.

With these:
Reviewed-by: Zorro Lang <zlang@xxxxxxxxxx>

> +_require_odirect
> +_require_io_uring
> +_require_dm_target flakey
> +_require_xfs_io_command "falloc"
> +
> +cat >$fio_config <<EOF
> +[test_io_uring_dio_dsync]
> +ioengine=io_uring
> +direct=1
> +bs=64K
> +sync=1
> +filename=$test_file
> +rw=randwrite
> +time_based
> +runtime=10
> +EOF
> +
> +_require_fio $fio_config
> +
> +_scratch_mkfs >>$seqres.full 2>&1
> +_require_metadata_journaling $SCRATCH_DEV
> +_init_flakey
> +_mount_flakey
> +
> +# We do 64K writes in the fio job.
> +_require_congruent_file_oplen $SCRATCH_MNT $((64 * 1024))
> +
> +touch $test_file
> +
> +# On btrfs IOCB_NOWAIT writes can only be done on NOCOW files, so enable
> +# nodatacow on the file if we are running on btrfs.
> +if [ $FSTYP == "btrfs" ]; then
> +	_require_chattr C
> +	$CHATTR_PROG +C $test_file
> +fi
> +
> +$XFS_IO_PROG -c "falloc 0 256M" $test_file
> +
> +# Persist everything, make sure the file exists after power failure.
> +sync
> +
> +echo -e "Running fio with config:\n" >> $seqres.full
> +cat $fio_config >> $seqres.full
> +
> +$FIO_PROG $fio_config --output=$fio_out
> +
> +echo -e "\nOutput from fio:\n" >> $seqres.full
> +cat $fio_out >> $seqres.full
> +
> +digest_before=$(_md5_checksum $test_file)
> +
> +# Simulate a power failure and mount the filesystem to check that all the data
> +# previously written are available.
> +_flakey_drop_and_remount
> +
> +digest_after=$(_md5_checksum $test_file)
> +
> +if [ "$digest_after" != "$digest_before" ]; then
> +	echo "Error: not all file data got persisted."
> +	echo "Digest before power failure: $digest_before"
> +	echo "Digest after power failure:  $digest_after"
> +fi
> +
> +_unmount_flakey
> +
> +# success, all done
> +echo "Silence is golden"
> +status=0
> +exit
> diff --git a/tests/generic/703.out b/tests/generic/703.out
> new file mode 100644
> index 00000000..fba62571
> --- /dev/null
> +++ b/tests/generic/703.out
> @@ -0,0 +1,2 @@
> +QA output created by 703
> +Silence is golden
> -- 
> 2.35.1
> 




[Index of Archives]     [Linux Filesystems Development]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux