On Mon, Nov 07, 2022 at 12:56:18PM +0800, Zorro Lang wrote:
> If the physical sector size is 4096, but the logical sector size
> is 512, the 512b dio write/read should be allowed.
Huh, did we all completely forget to review this patch?
> Signed-off-by: Zorro Lang <zlang@xxxxxxxxxx>
> ---
>
> Hi,
>
> This reproducer was written for xfs, I try to make it to be a generic
> test case for localfs. Current it test passed on xfs, extN and btrfs,
> the bug can be reproduced on old rhel-6.6 [1]. If it's not right for
> someone fs, please feel free to tell me.
>
> Thanks,
> Zorro
>
> [1]
> # ./check generic/888
> FSTYP -- xfs (non-debug)
> PLATFORM -- Linux/x86_64 xxx-xxxxx-xxxxxx 2.6.32-504.el6.x86_64
> MKFS_OPTIONS -- -f -bsize=4096 /dev/loop1
> MOUNT_OPTIONS -- -o context=system_u:object_r:nfs_t:s0 /dev/loop1 /mnt/scratch
>
> generic/888 - output mismatch (see /root/xfstests-dev/results//generic/888.out.bad)
> --- tests/generic/888.out 2022-11-06 23:42:44.683040977 -0500
> +++ /root/xfstests-dev/results//generic/888.out.bad 2022-11-06 23:48:33.986481844 -0500
> @@ -4,3 +4,4 @@
> 512
> mkfs and mount
> DIO read/write 512 bytes
> +pwrite64: Invalid argument
> ...
> (Run 'diff -u tests/generic/888.out /root/xfstests-dev/results//generic/888.out.bad' to see the entire diff)
> Ran: generic/888
> Failures: generic/888
> Failed 1 of 1 tests
>
> tests/generic/888 | 52 +++++++++++++++++++++++++++++++++++++++++++
> tests/generic/888.out | 6 +++++
> 2 files changed, 58 insertions(+)
> create mode 100755 tests/generic/888
> create mode 100644 tests/generic/888.out
>
> diff --git a/tests/generic/888 b/tests/generic/888
> new file mode 100755
> index 00000000..b5075d1e
> --- /dev/null
> +++ b/tests/generic/888
> @@ -0,0 +1,52 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2022 Red Hat, Inc. All Rights Reserved.
> +#
> +# FS QA Test No. 888
> +#
> +# Make sure logical-sector sized O_DIRECT write is allowed
> +#
> +. ./common/preamble
> +_begin_fstest auto quick
> +
> +# Override the default cleanup function.
> +_cleanup()
> +{
> + cd /
> + rm -r -f $tmp.*
> + [ -d "$SCSI_DEBUG_MNT" ] && $UMOUNT_PROG $SCSI_DEBUG_MNT 2>/dev/null
> + _put_scsi_debug_dev
> +}
> +
> +# Import common functions.
> +. ./common/scsi_debug
> +
> +# real QA test starts here
> +_supported_fs generic
> +_fixed_by_kernel_commit 7c71ee78031c "xfs: allow logical-sector sized O_DIRECT"
> +_require_scsi_debug
> +# If TEST_DEV is block device, make sure current fs is a localfs which can be
> +# written on scsi_debug device
> +_require_test
> +_require_block_device $TEST_DEV
_require_odirect?
> +
> +echo "Get a device with 4096 physical sector size and 512 logical sector size"
> +SCSI_DEBUG_DEV=`_get_scsi_debug_dev 4096 512 0 256`
> +blockdev --getpbsz --getss $SCSI_DEBUG_DEV
> +
> +echo "mkfs and mount"
> +_mkfs_dev $SCSI_DEBUG_DEV || _fail "Can't make $FSTYP on scsi_debug device"
> +SCSI_DEBUG_MNT="$TEST_DIR/scsi_debug_$seq"
> +rm -rf $SCSI_DEBUG_MNT
> +mkdir $SCSI_DEBUG_MNT
> +run_check _mount $SCSI_DEBUG_DEV $SCSI_DEBUG_MNT
/me wonders, should we try to use $MKFS_OPTIONS and $MOUNT_OPTIONS
on the scsidebug device? To catch cases where the config actually
matters for that kind of thing?
--D
> +
> +echo "DIO read/write 512 bytes"
> +# This dio write should succeed, even the physical sector size is 4096, but
> +# the logical sector size is 512
> +$XFS_IO_PROG -d -f -c "pwrite 0 512" $SCSI_DEBUG_MNT/testfile >> $seqres.full
> +$XFS_IO_PROG -d -c "pread 0 512" $SCSI_DEBUG_MNT/testfile >> $seqres.full
> +
> +# success, all done
> +status=0
> +exit
> diff --git a/tests/generic/888.out b/tests/generic/888.out
> new file mode 100644
> index 00000000..0f142ce9
> --- /dev/null
> +++ b/tests/generic/888.out
> @@ -0,0 +1,6 @@
> +QA output created by 888
> +Get a device with 4096 physical sector size and 512 logical sector size
> +4096
> +512
> +mkfs and mount
> +DIO read/write 512 bytes
> --
> 2.31.1
>
