Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 2620:137:e000::3:2 as permitted sender) client-ip=2620:137:e000::3:2;
Date:   Fri, 22 Sep 2023 21:39:24 +0530
Message-Id: <87v8c25gwb.fsf@doe.com>
From:   Ritesh Harjani (IBM) <ritesh.list@gmail.com>
To:     Gao Xiang <hsiangkao@linux.alibaba.com>, fstests@vger.kernel.org
Cc:     linux-ext4@vger.kernel.org, Jan Kara <jack@suse.cz>,
        Theodore Ts'o <tytso@mit.edu>
Subject: Re: [PATCH] generic: Add integrity tests with synchronous directio
In-Reply-To: <7ce321e3-3250-5627-18f8-230249fbf78f@linux.alibaba.com>
Precedence: bulk

Gao Xiang <hsiangkao@linux.alibaba.com> writes:

> Hi Ritesh,
>
> On 2023/9/22 20:10, Ritesh Harjani (IBM) wrote:
>> This test covers data & metadata integrity check with directio with
>> o_sync flag and checks the file contents & size after sudden fileystem
>> shutdown once the directio write is completed. ext4 directio after iomap
>> conversion was broken in the sense that if the FS crashes after
>> synchronous directio write, it's file size is not properly updated.
>> This test adds a testcase to cover such scenario.
>> 
>> Man page of open says that -
>> O_SYNC provides synchronized I/O file integrity completion, meaning write
>> operations will flush data and all associated metadata to the underlying
>> hardware
>> 
>> Reported-by: Gao Xiang <hsiangkao@linux.alibaba.com>
>> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
>> ---
>>   tests/generic/471     | 45 +++++++++++++++++++++++++++++++++++++++++++
>>   tests/generic/471.out |  8 ++++++++
>>   2 files changed, 53 insertions(+)
>>   create mode 100755 tests/generic/471
>>   create mode 100644 tests/generic/471.out
>> 
>> diff --git a/tests/generic/471 b/tests/generic/471
>> new file mode 100755
>> index 00000000..6c31cff8
>> --- /dev/null
>> +++ b/tests/generic/471
>> @@ -0,0 +1,45 @@
>> +#! /bin/bash
>> +# SPDX-License-Identifier: GPL-2.0
>> +# Copyright (c) 2023 IBM Corporation.  All Rights Reserved.
>> +#
>> +# FS QA Test 471
>> +#
>> +# Integrity test with DIRECT_IO & O_SYNC with sudden shutdown
>> +#
>> +. ./common/preamble
>> +_begin_fstest auto quick shutdown
>> +
>> +# 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 generic
>> +_require_scratch
>> +_require_scratch_shutdown
>> +
>> +_scratch_mkfs > $seqres.full 2>&1
>> +_scratch_mount
>> +
>> +echo "Create a 1M file using O_DIRECT & O_SYNC"
>> +xfs_io -fsd -c "pwrite -S 0x5a 0 1M" $SCRATCH_MNT/testfile > /dev/null 2>&1
>
> Thanks for the time on this.
>
> I'm fine with this as it's the exact regression test to
> my report.
>
> Although the original issue from our guest real workload
> is actually aio + O_SYNC, but that doesn't matter for
> ext4 since it will serialize the whole process of DIO
> write beyond i_size with inode lock.

Yes, even if we do AIO but since it is an extending write we will pass
IOMAP_DIO_FORCE_WAIT to iomap which means it will wait for the
completion anyway.

But on second thoughts, we can still add both synchronous direct-io
writes and buffered-io writes to this test. The man page of "open" tells
about O_SYNC flag, so the test should make sure that "data" and
"metadata" gets written to disk for both buffered-io and direct-io.
I will enhance that in second revision to cover buffered-io case as well.

>
> Yet if my understanding is correct, some other fses (e.g.
> XFS) seem to be more relaxed than this, see
> xfs_file_dio_write_aligned() and xfs_file_write_checks(),
> so I'm not sure if we need to cover AIO cases as well,
> anyway.

It's O_SYNC flag of open which mandates both data and metadata
integrity after "write" (or similar) completion. 
So, I think it will be better to cover AIO case as well.

There are a bunch of AIO tests present. I will see if either of it can
be enhanced to do basic aiodio writes. If not, I will add a basic
integrity test.

Thanks
-ritesh

>
> Thanks,
> Gao Xiang
>
>> +
>> +echo "Shutdown the fs suddenly"
>> +_scratch_shutdown
>> +
>> +echo "Cycle mount"
>> +_scratch_cycle_mount
>> +
>> +echo "File contents after cycle mount"
>> +_hexdump $SCRATCH_MNT/testfile
>> +
>> +status=0
>> +exit
>> diff --git a/tests/generic/471.out b/tests/generic/471.out
>> new file mode 100644
>> index 00000000..ae279b79
>> --- /dev/null
>> +++ b/tests/generic/471.out
>> @@ -0,0 +1,8 @@
>> +QA output created by 471
>> +Create a 1M file using O_DIRECT & O_SYNC
>> +Shutdown the fs suddenly
>> +Cycle mount
>> +File contents after cycle mount
>> +000000 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a  >ZZZZZZZZZZZZZZZZ<
>> +*
>> +100000