| Message ID | 20230606105200.129464-1-wqu@suse.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2] btrfs: add a test case to verify read-only scrub | expand |
On Tue, Jun 6, 2023 at 12:03 PM Qu Wenruo <wqu@suse.com> wrote: > > There is a regression on btrfs read-only scrub behavior. > > The commit e02ee89baa66 ("btrfs: scrub: switch scrub_simple_mirror() to > scrub_stripe infrastructure") makes btrfs scrub to ignore the read-only > flag completely, causing scrub to always fix the corruption. > > This test case would create an fs with repairable corruptions, then run > a read-only scrub, and finally to make sure the corruption is not > repaired. > > Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Looks good, thanks. > --- > Changelog: > v2: > - Add _fixed_by_kernel_commit > > - Reduce the golden output > Instead of the first 512 bytes, the first 16 bytes are more than > enough. > > - Better golden output > Add two more steps explaining what the test is doing. > > - Output the offset for the file operation inside the fs > The offset is fixed, no need to use _filter_xfs_io_offset. > > - Remove the confusing comments on common/filter > --- > tests/btrfs/288 | 70 +++++++++++++++++++++++++++++++++++++++++++++ > tests/btrfs/288.out | 9 ++++++ > 2 files changed, 79 insertions(+) > create mode 100755 tests/btrfs/288 > create mode 100644 tests/btrfs/288.out > > diff --git a/tests/btrfs/288 b/tests/btrfs/288 > new file mode 100755 > index 00000000..52245895 > --- /dev/null > +++ b/tests/btrfs/288 > @@ -0,0 +1,70 @@ > +#! /bin/bash > +# SPDX-License-Identifier: GPL-2.0 > +# Copyright (C) 2023 SUSE Linux Products GmbH. All Rights Reserved. > +# > +# FS QA Test 288 > +# > +# Make sure btrfs-scrub respects the read-only flag. > +# > +. ./common/preamble > +_begin_fstest auto repair quick volume scrub > + > +. ./common/filter > + > +# real QA test starts here > + > +# Modify as appropriate. > +_supported_fs btrfs > +_require_scratch_dev_pool 2 > + > + > +_require_odirect > +# Overwriting data is forbidden on a zoned block device > +_require_non_zoned_device "${SCRATCH_DEV}" > + > +_fixed_by_kernel_commit xxxxxxxxxxxx \ > + "btrfs: scrub: respect the read-only flag during repair" > + > +_scratch_dev_pool_get 2 > + > +# Step 1, create a raid btrfs with one 128K file > +echo "step 1......mkfs.btrfs" > +_scratch_pool_mkfs -d raid1 -b 1G >> $seqres.full 2>&1 > +_scratch_mount > + > +$XFS_IO_PROG -f -d -c "pwrite -S 0xaa -b 128K 0 128K" "$SCRATCH_MNT/foobar" |\ > + _filter_xfs_io > + > +# Step 2, corrupt one mirror so we can still repair the fs. > +echo "step 2......corrupt one mirror" > +# ensure btrfs-map-logical sees the tree updates > +sync > + > +logical=$(_btrfs_get_first_logical $SCRATCH_MNT/foobar) > + > +physical1=$(_btrfs_get_physical ${logical} 1) > +devpath1=$(_btrfs_get_device_path ${logical} 1) > + > +_scratch_unmount > + > +echo " corrupt stripe #1, devpath $devpath1 physical $physical1" \ > + >> $seqres.full > +$XFS_IO_PROG -d -c "pwrite -S 0xf1 -b 64K $physical1 64K" $devpath1 \ > + > /dev/null > + > +# Step 3, do a read-only scrub, which should not fix the corruption. > +echo "step 3......do a read-only scrub" > +_scratch_mount -o ro > +$BTRFS_UTIL_PROG scrub start -BRrd $SCRATCH_MNT >> $seqres.full 2>&1 > +_scratch_unmount > + > +# Step 4, make sure the corruption is still there > +echo "step 4......verify the corruption is not repaired" > +echo " the first 16 bytes of the extent at mirror 1:" > +$XFS_IO_PROG -c "pread -qv $physical1 16" $devpath1 |\ > + _filter_xfs_io_offset > + > +_scratch_dev_pool_put > +# success, all done > +status=0 > +exit > diff --git a/tests/btrfs/288.out b/tests/btrfs/288.out > new file mode 100644 > index 00000000..452bdc67 > --- /dev/null > +++ b/tests/btrfs/288.out > @@ -0,0 +1,9 @@ > +QA output created by 288 > +step 1......mkfs.btrfs > +wrote 131072/131072 bytes at offset 0 > +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) > +step 2......corrupt one mirror > +step 3......do a read-only scrub > +step 4......verify the corruption is not repaired > + the first 16 bytes of the extent at mirror 1: > +XXXXXXXX: f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 ................ > -- > 2.39.0 >
diff --git a/tests/btrfs/288 b/tests/btrfs/288 new file mode 100755 index 00000000..52245895 --- /dev/null +++ b/tests/btrfs/288 @@ -0,0 +1,70 @@ +#! /bin/bash +# SPDX-License-Identifier: GPL-2.0 +# Copyright (C) 2023 SUSE Linux Products GmbH. All Rights Reserved. +# +# FS QA Test 288 +# +# Make sure btrfs-scrub respects the read-only flag. +# +. ./common/preamble +_begin_fstest auto repair quick volume scrub + +. ./common/filter + +# real QA test starts here + +# Modify as appropriate. +_supported_fs btrfs +_require_scratch_dev_pool 2 + + +_require_odirect +# Overwriting data is forbidden on a zoned block device +_require_non_zoned_device "${SCRATCH_DEV}" + +_fixed_by_kernel_commit xxxxxxxxxxxx \ + "btrfs: scrub: respect the read-only flag during repair" + +_scratch_dev_pool_get 2 + +# Step 1, create a raid btrfs with one 128K file +echo "step 1......mkfs.btrfs" +_scratch_pool_mkfs -d raid1 -b 1G >> $seqres.full 2>&1 +_scratch_mount + +$XFS_IO_PROG -f -d -c "pwrite -S 0xaa -b 128K 0 128K" "$SCRATCH_MNT/foobar" |\ + _filter_xfs_io + +# Step 2, corrupt one mirror so we can still repair the fs. +echo "step 2......corrupt one mirror" +# ensure btrfs-map-logical sees the tree updates +sync + +logical=$(_btrfs_get_first_logical $SCRATCH_MNT/foobar) + +physical1=$(_btrfs_get_physical ${logical} 1) +devpath1=$(_btrfs_get_device_path ${logical} 1) + +_scratch_unmount + +echo " corrupt stripe #1, devpath $devpath1 physical $physical1" \ + >> $seqres.full +$XFS_IO_PROG -d -c "pwrite -S 0xf1 -b 64K $physical1 64K" $devpath1 \ + > /dev/null + +# Step 3, do a read-only scrub, which should not fix the corruption. +echo "step 3......do a read-only scrub" +_scratch_mount -o ro +$BTRFS_UTIL_PROG scrub start -BRrd $SCRATCH_MNT >> $seqres.full 2>&1 +_scratch_unmount + +# Step 4, make sure the corruption is still there +echo "step 4......verify the corruption is not repaired" +echo " the first 16 bytes of the extent at mirror 1:" +$XFS_IO_PROG -c "pread -qv $physical1 16" $devpath1 |\ + _filter_xfs_io_offset + +_scratch_dev_pool_put +# success, all done +status=0 +exit diff --git a/tests/btrfs/288.out b/tests/btrfs/288.out new file mode 100644 index 00000000..452bdc67 --- /dev/null +++ b/tests/btrfs/288.out @@ -0,0 +1,9 @@ +QA output created by 288 +step 1......mkfs.btrfs +wrote 131072/131072 bytes at offset 0 +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) +step 2......corrupt one mirror +step 3......do a read-only scrub +step 4......verify the corruption is not repaired + the first 16 bytes of the extent at mirror 1: +XXXXXXXX: f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 f1 ................
There is a regression on btrfs read-only scrub behavior. The commit e02ee89baa66 ("btrfs: scrub: switch scrub_simple_mirror() to scrub_stripe infrastructure") makes btrfs scrub to ignore the read-only flag completely, causing scrub to always fix the corruption. This test case would create an fs with repairable corruptions, then run a read-only scrub, and finally to make sure the corruption is not repaired. Signed-off-by: Qu Wenruo <wqu@suse.com> --- Changelog: v2: - Add _fixed_by_kernel_commit - Reduce the golden output Instead of the first 512 bytes, the first 16 bytes are more than enough. - Better golden output Add two more steps explaining what the test is doing. - Output the offset for the file operation inside the fs The offset is fixed, no need to use _filter_xfs_io_offset. - Remove the confusing comments on common/filter --- tests/btrfs/288 | 70 +++++++++++++++++++++++++++++++++++++++++++++ tests/btrfs/288.out | 9 ++++++ 2 files changed, 79 insertions(+) create mode 100755 tests/btrfs/288 create mode 100644 tests/btrfs/288.out