diff mbox series

generic: test reflinked file corruption after short COW

Message ID b63354c6-795d-78e2-4002-83c08a373171@redhat.com (mailing list archive)
State Superseded, archived
Headers show
Series generic: test reflinked file corruption after short COW | expand

Commit Message

Eric Sandeen Sept. 24, 2020, 6:19 p.m. UTC 
This test essentially creates an existing COW extent which
covers the first 1M, and then does another IO that overlaps it,
but extends beyond it.  The bug was that we did not trim the
new IO to the end of the existing COW extent, and so the IO
extended past the COW blocks and corrupted the reflinked files(s).

The bug came and went upstream; it will be hopefully fixed in the
5.4.y stable series via:

https://lore.kernel.org/stable/e7fe7225-4f2b-d13e-bb4b-c7db68f63124@redhat.com/

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
---

Comments

Darrick J. Wong Sept. 24, 2020, 8:17 p.m. UTC | #1 
On Thu, Sep 24, 2020 at 01:19:49PM -0500, Eric Sandeen wrote:
> This test essentially creates an existing COW extent which
> covers the first 1M, and then does another IO that overlaps it,
> but extends beyond it.  The bug was that we did not trim the
> new IO to the end of the existing COW extent, and so the IO
> extended past the COW blocks and corrupted the reflinked files(s).
> 
> The bug came and went upstream; it will be hopefully fixed in the
> 5.4.y stable series via:
> 
> https://lore.kernel.org/stable/e7fe7225-4f2b-d13e-bb4b-c7db68f63124@redhat.com/
> 
> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
> ---
> 
> diff --git a/tests/generic/612 b/tests/generic/612
> new file mode 100755
> index 00000000..5a765a0c
> --- /dev/null
> +++ b/tests/generic/612
> @@ -0,0 +1,83 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2020 Red Hat, Inc.  All Rights Reserved.
> +#
> +# FS QA Test 612
> +#
> +# Regression test for reflink corruption present as of:
> +# 78f0cc9d55cb "xfs: don't use delalloc extents for COW on files with extsize hints"
> +# and (inadvertently) fixed as of:
> +# 36adcbace24e "xfs: fill out the srcmap in iomap_begin"

This probably should list the name of the patch that fixes it for 5.4.

With that added,
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>

--D

> +#
> +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.*
> +}
> +
> +# get standard environment, filters and checks
> +. ./common/rc
> +. ./common/filter
> +. ./common/reflink
> +
> +# remove previous $seqres.full before test
> +rm -f $seqres.full
> +
> +# real QA test starts here
> +
> +# Modify as appropriate.
> +_supported_fs generic
> +_require_test
> +_require_test_reflink
> +
> +DIR=$TEST_DIR/dir.$seq
> +mkdir -p $DIR
> +rm -f $DIR/a $DIR/b
> +
> +# This test essentially creates an existing COW extent which
> +# covers the first 1M, and then does another IO that overlaps it,
> +# but extends beyond it.  The bug was that we did not trim the
> +# new IO to the end of the existing COW extent, and so the IO
> +# extended past the COW blocks and corrupted the reflinked files(s).
> +
> +# Make all files w/ 1m hints; create original 2m file
> +$XFS_IO_PROG -c "extsize 1048576" $DIR | _filter_xfs_io
> +$XFS_IO_PROG -c "cowextsize 1048576" $DIR | _filter_xfs_io
> +
> +echo "Create file b"
> +$XFS_IO_PROG -f -c "pwrite -S 0x0 0 2m" -c fsync $DIR/b | _filter_xfs_io
> +
> +# Make a reflinked copy
> +echo "Reflink copy from b to a"
> +cp --reflink=always $DIR/b $DIR/a
> +
> +echo "Contents of b"
> +hexdump -C $DIR/b
> +
> +# Cycle mount to get stuff out of cache
> +_test_cycle_mount
> +
> +# Create a 1m-hinted IO at offset 0, then
> +# do another IO that overlaps but extends past the 1m hint
> +echo "Write to a"
> +$XFS_IO_PROG -c "pwrite -S 0xa 0k -b 4k 4k" \
> +       -c "pwrite -S 0xa 4k -b 1m 1m" \
> +       $DIR/a | _filter_xfs_io
> +
> +$XFS_IO_PROG -c fsync $DIR/a
> +
> +echo "Contents of b now:"
> +hexdump -C $DIR/b
> +
> +# success, all done
> +status=0
> +exit
> diff --git a/tests/generic/612.out b/tests/generic/612.out
> new file mode 100644
> index 00000000..237a9638
> --- /dev/null
> +++ b/tests/generic/612.out
> @@ -0,0 +1,18 @@
> +QA output created by 612
> +Create file b
> +wrote 2097152/2097152 bytes at offset 0
> +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
> +Reflink copy from b to a
> +Contents of b
> +00000000  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
> +*
> +00200000
> +Write to a
> +wrote 4096/4096 bytes at offset 0
> +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
> +wrote 1048576/1048576 bytes at offset 4096
> +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
> +Contents of b now:
> +00000000  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
> +*
> +00200000
> diff --git a/tests/generic/group b/tests/generic/group
> index 4af4b494..bc115f21 100644
> --- a/tests/generic/group
> +++ b/tests/generic/group
> @@ -614,3 +614,4 @@
>  609 auto quick rw
>  610 auto quick prealloc zero
>  611 auto quick attr
> +612 auto quick clone
>
Eric Sandeen Sept. 24, 2020, 8:42 p.m. UTC | #2 
On 9/24/20 3:17 PM, Darrick J. Wong wrote:
> On Thu, Sep 24, 2020 at 01:19:49PM -0500, Eric Sandeen wrote:
>> This test essentially creates an existing COW extent which
>> covers the first 1M, and then does another IO that overlaps it,
>> but extends beyond it.  The bug was that we did not trim the
>> new IO to the end of the existing COW extent, and so the IO
>> extended past the COW blocks and corrupted the reflinked files(s).
>>
>> The bug came and went upstream; it will be hopefully fixed in the
>> 5.4.y stable series via:
>>
>> https://lore.kernel.org/stable/e7fe7225-4f2b-d13e-bb4b-c7db68f63124@redhat.com/
>>
>> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
>> ---
>>
>> diff --git a/tests/generic/612 b/tests/generic/612
>> new file mode 100755
>> index 00000000..5a765a0c
>> --- /dev/null
>> +++ b/tests/generic/612
>> @@ -0,0 +1,83 @@
>> +#! /bin/bash
>> +# SPDX-License-Identifier: GPL-2.0
>> +# Copyright (c) 2020 Red Hat, Inc.  All Rights Reserved.
>> +#
>> +# FS QA Test 612
>> +#
>> +# Regression test for reflink corruption present as of:
>> +# 78f0cc9d55cb "xfs: don't use delalloc extents for COW on files with extsize hints"
>> +# and (inadvertently) fixed as of:
>> +# 36adcbace24e "xfs: fill out the srcmap in iomap_begin"
> 
> This probably should list the name of the patch that fixes it for 5.4.
> 
> With that added,
> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>

Will have to wait for a merge for that, I guess.

Especially with the typo fixed (or not)

-Eric
diff mbox series

Patch

diff --git a/tests/generic/612 b/tests/generic/612
new file mode 100755
index 00000000..5a765a0c
--- /dev/null
+++ b/tests/generic/612
@@ -0,0 +1,83 @@ 
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2020 Red Hat, Inc.  All Rights Reserved.
+#
+# FS QA Test 612
+#
+# Regression test for reflink corruption present as of:
+# 78f0cc9d55cb "xfs: don't use delalloc extents for COW on files with extsize hints"
+# and (inadvertently) fixed as of:
+# 36adcbace24e "xfs: fill out the srcmap in iomap_begin"
+#
+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.*
+}
+
+# get standard environment, filters and checks
+. ./common/rc
+. ./common/filter
+. ./common/reflink
+
+# remove previous $seqres.full before test
+rm -f $seqres.full
+
+# real QA test starts here
+
+# Modify as appropriate.
+_supported_fs generic
+_require_test
+_require_test_reflink
+
+DIR=$TEST_DIR/dir.$seq
+mkdir -p $DIR
+rm -f $DIR/a $DIR/b
+
+# This test essentially creates an existing COW extent which
+# covers the first 1M, and then does another IO that overlaps it,
+# but extends beyond it.  The bug was that we did not trim the
+# new IO to the end of the existing COW extent, and so the IO
+# extended past the COW blocks and corrupted the reflinked files(s).
+
+# Make all files w/ 1m hints; create original 2m file
+$XFS_IO_PROG -c "extsize 1048576" $DIR | _filter_xfs_io
+$XFS_IO_PROG -c "cowextsize 1048576" $DIR | _filter_xfs_io
+
+echo "Create file b"
+$XFS_IO_PROG -f -c "pwrite -S 0x0 0 2m" -c fsync $DIR/b | _filter_xfs_io
+
+# Make a reflinked copy
+echo "Reflink copy from b to a"
+cp --reflink=always $DIR/b $DIR/a
+
+echo "Contents of b"
+hexdump -C $DIR/b
+
+# Cycle mount to get stuff out of cache
+_test_cycle_mount
+
+# Create a 1m-hinted IO at offset 0, then
+# do another IO that overlaps but extends past the 1m hint
+echo "Write to a"
+$XFS_IO_PROG -c "pwrite -S 0xa 0k -b 4k 4k" \
+       -c "pwrite -S 0xa 4k -b 1m 1m" \
+       $DIR/a | _filter_xfs_io
+
+$XFS_IO_PROG -c fsync $DIR/a
+
+echo "Contents of b now:"
+hexdump -C $DIR/b
+
+# success, all done
+status=0
+exit
diff --git a/tests/generic/612.out b/tests/generic/612.out
new file mode 100644
index 00000000..237a9638
--- /dev/null
+++ b/tests/generic/612.out
@@ -0,0 +1,18 @@ 
+QA output created by 612
+Create file b
+wrote 2097152/2097152 bytes at offset 0
+XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
+Reflink copy from b to a
+Contents of b
+00000000  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
+*
+00200000
+Write to a
+wrote 4096/4096 bytes at offset 0
+XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
+wrote 1048576/1048576 bytes at offset 4096
+XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
+Contents of b now:
+00000000  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
+*
+00200000
diff --git a/tests/generic/group b/tests/generic/group
index 4af4b494..bc115f21 100644
--- a/tests/generic/group
+++ b/tests/generic/group
@@ -614,3 +614,4 @@ 
 609 auto quick rw
 610 auto quick prealloc zero
 611 auto quick attr
+612 auto quick clone