| Message ID | 20181023181638.43407-1-bfoster@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | tests/xfs: writepage map error unmount test | expand |
On Tue, Oct 23, 2018 at 02:16:38PM -0400, Brian Foster wrote: > Certain failures in the page writeback codepath can cause a broader > writeback (i.e., a sync) sequence to exit prematurely. If this > occurs during an unmount, it's likely the filesystem will reclaim > inodes without having cleared all delayed allocation blocks. This > produces a warning and leaves the filesystem inconsistent. > > This test reproduces this scenario using the 'writepage_map' > errortag. It performs delayed allocation, injects writeback errors > and immediately unmounts the filesystem. > > Signed-off-by: Brian Foster <bfoster@redhat.com> > --- > > Note that this depends on a not yet available error tag. I'll wait for the "error tag" patch goes in first, or gets some ACKs first. One really minor issue below. > > Brian > > tests/xfs/494 | 57 +++++++++++++++++++++++++++++++++++++++++++++++ > tests/xfs/494.out | 2 ++ > tests/xfs/group | 1 + > 3 files changed, 60 insertions(+) > create mode 100755 tests/xfs/494 > create mode 100644 tests/xfs/494.out > > diff --git a/tests/xfs/494 b/tests/xfs/494 > new file mode 100755 > index 00000000..30877d38 > --- /dev/null > +++ b/tests/xfs/494 > @@ -0,0 +1,57 @@ > +#! /bin/bash > +# SPDX-License-Identifier: GPL-2.0 > +# Copyright (c) 2018 Red Hat, Inc. All Rights Reserved. > +# > +# FS QA Test 494 > +# > +# Simulate a serialization problem between page writeback error handling and > +# unmount. If ->writepages() fails and returns an error back to the mm > +# subsystem, writeback of subsequent pages is interrupted and the filesystem > +# unmounts before processing all dirty pages. This results in unmounting a > +# filesystem with dirty+delalloc pages, which in turn causes unmount time > +# warnings and filesystem inconsistency. > +# > +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/inject > + > +# remove previous $seqres.full before test > +rm -f $seqres.full > + > +# real QA test starts here > + > +# Modify as appropriate. > +_supported_fs generic Supported fs should be 'xfs' here. Thanks, Eryu > +_supported_os Linux > +_require_error_injection > +_require_xfs_io_error_injection "writepage_map" > +_require_scratch > + > +_scratch_mkfs >> $seqres.full 2>&1 > + > +_scratch_mount > +_scratch_inject_error "writepage_map" 1 > +# use 512k to dirty multiple pages on large page size systems > +$XFS_IO_PROG -fc "pwrite 0 512k" $SCRATCH_MNT/file >> $seqres.full 2>&1 > +_scratch_unmount > + > +echo Silence is golden > + > +# success, all done > +status=0 > +exit > diff --git a/tests/xfs/494.out b/tests/xfs/494.out > new file mode 100644 > index 00000000..827c239a > --- /dev/null > +++ b/tests/xfs/494.out > @@ -0,0 +1,2 @@ > +QA output created by 494 > +Silence is golden > diff --git a/tests/xfs/group b/tests/xfs/group > index 2cec0585..136d3707 100644 > --- a/tests/xfs/group > +++ b/tests/xfs/group > @@ -491,3 +491,4 @@ > 491 auto quick fuzz > 492 auto quick fuzz > 493 auto quick fuzz > +494 auto quick > -- > 2.17.2 >
On Sun, Oct 28, 2018 at 09:56:33PM +0800, Eryu Guan wrote: > On Tue, Oct 23, 2018 at 02:16:38PM -0400, Brian Foster wrote: > > Certain failures in the page writeback codepath can cause a broader > > writeback (i.e., a sync) sequence to exit prematurely. If this > > occurs during an unmount, it's likely the filesystem will reclaim > > inodes without having cleared all delayed allocation blocks. This > > produces a warning and leaves the filesystem inconsistent. > > > > This test reproduces this scenario using the 'writepage_map' > > errortag. It performs delayed allocation, injects writeback errors > > and immediately unmounts the filesystem. > > > > Signed-off-by: Brian Foster <bfoster@redhat.com> > > --- > > > > Note that this depends on a not yet available error tag. > > I'll wait for the "error tag" patch goes in first, or gets some ACKs > first. One really minor issue below. > Yep. I probably should have been more clear and/or set this RFC. This test is wholly dependent on if/whether/how to deal with the associated error handling issue, which atm is still up in the air. The tag/test is primarily to demonstrate the problem and may not be independently useful until there's a supporting fix. I may float a patch for that soon.. > > > > Brian > > > > tests/xfs/494 | 57 +++++++++++++++++++++++++++++++++++++++++++++++ > > tests/xfs/494.out | 2 ++ > > tests/xfs/group | 1 + > > 3 files changed, 60 insertions(+) > > create mode 100755 tests/xfs/494 > > create mode 100644 tests/xfs/494.out > > > > diff --git a/tests/xfs/494 b/tests/xfs/494 > > new file mode 100755 > > index 00000000..30877d38 > > --- /dev/null > > +++ b/tests/xfs/494 > > @@ -0,0 +1,57 @@ > > +#! /bin/bash > > +# SPDX-License-Identifier: GPL-2.0 > > +# Copyright (c) 2018 Red Hat, Inc. All Rights Reserved. > > +# > > +# FS QA Test 494 > > +# > > +# Simulate a serialization problem between page writeback error handling and > > +# unmount. If ->writepages() fails and returns an error back to the mm > > +# subsystem, writeback of subsequent pages is interrupted and the filesystem > > +# unmounts before processing all dirty pages. This results in unmounting a > > +# filesystem with dirty+delalloc pages, which in turn causes unmount time > > +# warnings and filesystem inconsistency. > > +# > > +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/inject > > + > > +# remove previous $seqres.full before test > > +rm -f $seqres.full > > + > > +# real QA test starts here > > + > > +# Modify as appropriate. > > +_supported_fs generic > > Supported fs should be 'xfs' here. > Will fix, thanks. Brian > Thanks, > Eryu > > > +_supported_os Linux > > +_require_error_injection > > +_require_xfs_io_error_injection "writepage_map" > > +_require_scratch > > + > > +_scratch_mkfs >> $seqres.full 2>&1 > > + > > +_scratch_mount > > +_scratch_inject_error "writepage_map" 1 > > +# use 512k to dirty multiple pages on large page size systems > > +$XFS_IO_PROG -fc "pwrite 0 512k" $SCRATCH_MNT/file >> $seqres.full 2>&1 > > +_scratch_unmount > > + > > +echo Silence is golden > > + > > +# success, all done > > +status=0 > > +exit > > diff --git a/tests/xfs/494.out b/tests/xfs/494.out > > new file mode 100644 > > index 00000000..827c239a > > --- /dev/null > > +++ b/tests/xfs/494.out > > @@ -0,0 +1,2 @@ > > +QA output created by 494 > > +Silence is golden > > diff --git a/tests/xfs/group b/tests/xfs/group > > index 2cec0585..136d3707 100644 > > --- a/tests/xfs/group > > +++ b/tests/xfs/group > > @@ -491,3 +491,4 @@ > > 491 auto quick fuzz > > 492 auto quick fuzz > > 493 auto quick fuzz > > +494 auto quick > > -- > > 2.17.2 > >
diff --git a/tests/xfs/494 b/tests/xfs/494 new file mode 100755 index 00000000..30877d38 --- /dev/null +++ b/tests/xfs/494 @@ -0,0 +1,57 @@ +#! /bin/bash +# SPDX-License-Identifier: GPL-2.0 +# Copyright (c) 2018 Red Hat, Inc. All Rights Reserved. +# +# FS QA Test 494 +# +# Simulate a serialization problem between page writeback error handling and +# unmount. If ->writepages() fails and returns an error back to the mm +# subsystem, writeback of subsequent pages is interrupted and the filesystem +# unmounts before processing all dirty pages. This results in unmounting a +# filesystem with dirty+delalloc pages, which in turn causes unmount time +# warnings and filesystem inconsistency. +# +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/inject + +# remove previous $seqres.full before test +rm -f $seqres.full + +# real QA test starts here + +# Modify as appropriate. +_supported_fs generic +_supported_os Linux +_require_error_injection +_require_xfs_io_error_injection "writepage_map" +_require_scratch + +_scratch_mkfs >> $seqres.full 2>&1 + +_scratch_mount +_scratch_inject_error "writepage_map" 1 +# use 512k to dirty multiple pages on large page size systems +$XFS_IO_PROG -fc "pwrite 0 512k" $SCRATCH_MNT/file >> $seqres.full 2>&1 +_scratch_unmount + +echo Silence is golden + +# success, all done +status=0 +exit diff --git a/tests/xfs/494.out b/tests/xfs/494.out new file mode 100644 index 00000000..827c239a --- /dev/null +++ b/tests/xfs/494.out @@ -0,0 +1,2 @@ +QA output created by 494 +Silence is golden diff --git a/tests/xfs/group b/tests/xfs/group index 2cec0585..136d3707 100644 --- a/tests/xfs/group +++ b/tests/xfs/group @@ -491,3 +491,4 @@ 491 auto quick fuzz 492 auto quick fuzz 493 auto quick fuzz +494 auto quick
Certain failures in the page writeback codepath can cause a broader writeback (i.e., a sync) sequence to exit prematurely. If this occurs during an unmount, it's likely the filesystem will reclaim inodes without having cleared all delayed allocation blocks. This produces a warning and leaves the filesystem inconsistent. This test reproduces this scenario using the 'writepage_map' errortag. It performs delayed allocation, injects writeback errors and immediately unmounts the filesystem. Signed-off-by: Brian Foster <bfoster@redhat.com> --- Note that this depends on a not yet available error tag. Brian tests/xfs/494 | 57 +++++++++++++++++++++++++++++++++++++++++++++++ tests/xfs/494.out | 2 ++ tests/xfs/group | 1 + 3 files changed, 60 insertions(+) create mode 100755 tests/xfs/494 create mode 100644 tests/xfs/494.out