On Wed, Sep 02, 2020 at 12:00:44PM -0400, Josef Bacik wrote:
> We had a problem recently where btrfs would deadlock with
> O_DIRECT|O_DSYNC because of an unexpected dependency on ->fsync in
> iomap. This was only caught by chance with aiostress, because weirdly
> we don't actually test this particular configuration anywhere in
> xfstests. Fix this by adding a basic test that just does
> O_DIRECT|O_DSYNC writes. With this test the box deadlocks right away
> with Btrfs, which would have been helpful in finding this issue before
> the patches were merged.
>
> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@xxxxxxx>
> Signed-off-by: Josef Bacik <josef@xxxxxxxxxxxxxx>
> ---
> v2->v3:
> - This time with 609.out added, verified it passed with xfs.
>
> tests/generic/609 | 43 +++++++++++++++++++++++++++++++++++++++++++
> tests/generic/609.out | 3 +++
> tests/generic/group | 1 +
> 3 files changed, 47 insertions(+)
> create mode 100755 tests/generic/609
> create mode 100644 tests/generic/609.out
>
> diff --git a/tests/generic/609 b/tests/generic/609
> new file mode 100755
> index 00000000..3d1c97b2
> --- /dev/null
> +++ b/tests/generic/609
> @@ -0,0 +1,43 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2020 Josef Bacik. All Rights Reserved.
> +#
> +# FS QA Test 609
> +#
> +# iomap can call generic_write_sync() if we're O_DSYNC, so write a basic test to
> +# exercise O_DSYNC so any unsuspecting file systems will get lockdep warnings if
> +# their locking isn't compatible.
> +#
> +seq=`basename $0`
> +seqres=$RESULT_DIR/$seq
> +echo "QA output created by $seq"
> +
> +here=`pwd`
> +tmp=/tmp/$$
> +status=1 # failure is the default!
> +trap "_cleanup; exit \$status" 0 1 2 3 15
> +
> +_cleanup()
> +{
> + cd /
> + rm -f $tmp.*
> + rm -rf $TEST_DIR/file
> +}
> +
> +# get standard environment, filters and checks
> +. ./common/rc
> +. ./common/filter
> +
> +# remove previous $seqres.full before test
> +rm -f $seqres.full
> +
> +# Modify as appropriate.
> +_supported_fs generic
> +_supported_os Linux
> +_require_test
> +_require_xfs_io_command "pwrite" "-DV"
> +
> +$XFS_IO_PROG -f -d -c "pwrite -D -V 1 0 4k" $TEST_DIR/file | _filter_xfs_io
I wonder, does this also work if you did:
$XFS_IO_PROG -f -d -s -c 'pwrite 0 4k' $TEST_DIR/file
In other words, can you reproduce the problem with good old pwrite() and
a file descriptor opened O_SYNC? Or do you specifically have to have
pwritev2 with RWF_DSYNC?
(I might also write 64k to future proof this testcase for the day when
someone builds an fs that can only do 64k direct writes, but maybe
that's crazy...)
> +
> +status=0
> +exit
> diff --git a/tests/generic/609.out b/tests/generic/609.out
> new file mode 100644
> index 00000000..db3242cb
> --- /dev/null
> +++ b/tests/generic/609.out
> @@ -0,0 +1,3 @@
> +QA output created by 609
> +wrote 4096/4096 bytes at offset 0
> +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
> diff --git a/tests/generic/group b/tests/generic/group
> index aa969bcb..ae2567a0 100644
> --- a/tests/generic/group
> +++ b/tests/generic/group
> @@ -611,3 +611,4 @@
> 606 auto attr quick dax
> 607 auto attr quick dax
> 608 auto attr quick dax
> +609 auto quick
This probably ought to be 'auto quick rw' since it's a write test.
The rest of the logic looks sound to me though.
--D
> --
> 2.28.0
>
