[PATCH 0/2] xfs: test for readahead use after free panic

[PATCH 0/2] xfs: test for readahead use after free panic

[Brian Foster]

This introduces some generic dm-delay infrastructure, which is basically
copied from the existing dm-flakey bits, and a test that causes current
upstream XFS to panic.

A kernel fix will be posted shortly...

Brian

Brian Foster (2):
  xfstests: support dm-delay to introduce I/O delays
  tests/xfs: test for post umount readahead completion panic

 common/dmdelay    | 89 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
 tests/xfs/999     | 90 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 tests/xfs/999.out |  2 ++
 tests/xfs/group   |  1 +
 4 files changed, 182 insertions(+)
 create mode 100644 common/dmdelay
 create mode 100755 tests/xfs/999
 create mode 100644 tests/xfs/999.out

-- 
2.5.5

[PATCH 1/2] xfstests: support dm-delay to introduce I/O delays

[Brian Foster]

Add some infrastructure in common/dmdelay to support use of the dm-delay
device-mapper module within tests. This is effectively copied from the
existing infrastructure in common/dmflakey. This provides the ability to
delay I/O. It only supports delaying read I/O as that is all that is
required at this point in time.

Signed-off-by: Brian Foster <bfoster@redhat.com>
---
 common/dmdelay | 89 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 89 insertions(+)
 create mode 100644 common/dmdelay

diff --git a/common/dmdelay b/common/dmdelay
new file mode 100644
index 0000000..c53e2dd
--- /dev/null
+++ b/common/dmdelay
@@ -0,0 +1,89 @@
+##/bin/bash
+#
+# Copyright (c) 2016 Red Hat, Inc.  All Rights Reserved.
+#
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License as
+# published by the Free Software Foundation.
+#
+# This program is distributed in the hope that it would be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write the Free Software Foundation,
+# Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+#
+#
+# common functions for setting up and tearing down a dmdelay device
+
+DELAY_NONE=0
+DELAY_READ=1
+
+echo $MOUNT_OPTIONS | grep -q dax
+if [ $? -eq 0 ]; then
+	_notrun "Cannot run tests with DAX on dmdelay devices"
+fi
+
+_init_delay()
+{
+	local BLK_DEV_SIZE=`blockdev --getsz $SCRATCH_DEV`
+	DELAY_DEV=/dev/mapper/delay-test
+	DELAY_TABLE="0 $BLK_DEV_SIZE delay $SCRATCH_DEV 0 0"
+	DELAY_TABLE_RDELAY="0 $BLK_DEV_SIZE delay $SCRATCH_DEV 0 10000 $SCRATCH_DEV 0 0"
+	$DMSETUP_PROG create delay-test --table "$DELAY_TABLE" || \
+		_fatal "failed to create delay device"
+	$DMSETUP_PROG mknodes > /dev/null 2>&1
+}
+
+_mount_delay()
+{
+	_scratch_options mount
+	mount -t $FSTYP $SCRATCH_OPTIONS $MOUNT_OPTIONS $DELAY_DEV $SCRATCH_MNT
+}
+
+_unmount_delay()
+{
+	$UMOUNT_PROG $SCRATCH_MNT
+}
+
+_cleanup_delay()
+{
+	# If dmsetup load fails then we need to make sure to do resume here
+	# otherwise the umount will hang
+	$DMSETUP_PROG resume delay-test > /dev/null 2>&1
+	$UMOUNT_PROG $SCRATCH_MNT > /dev/null 2>&1
+	# wait for device to be fully settled so that 'dmsetup remove' doesn't
+	# fail due to EBUSY
+	$UDEV_SETTLE_PROG >/dev/null 2>&1
+	$DMSETUP_PROG remove delay-test > /dev/null 2>&1
+	$DMSETUP_PROG mknodes > /dev/null 2>&1
+}
+
+# _load_delay_table <table> [lockfs]
+#
+# This defaults to --nolockfs, which doesn't freeze_fs() before loading the new
+# table, so it simulates power failure.
+_load_delay_table()
+{
+	table="$DELAY_TABLE"
+	[ $1 -eq $DELAY_READ ] && table="$DELAY_TABLE_RDELAY"
+
+	suspend_opt="--nolockfs"
+	[ $# -gt 1 ] && [ $2 -eq 1 ] && suspend_opt=""
+
+	# run a suspend/resume cycle to avoid excessive resume delays once a
+	# delay is introduced below
+	$DMSETUP_PROG suspend $suspend_opt delay-test
+	$DMSETUP_PROG resume $suspend_opt delay-test
+
+	$DMSETUP_PROG suspend $suspend_opt delay-test
+	[ $? -ne 0 ] && _fatal "failed to suspend delay-test"
+
+	$DMSETUP_PROG load delay-test --table "$table"
+	[ $? -ne 0 ] && _fatal "failed to load table into delay-test"
+
+	$DMSETUP_PROG resume delay-test
+	[ $? -ne 0 ] && _fatal  "failed to resume delay-test"
+}
-- 
2.5.5

[PATCH 2/2] tests/xfs: test for post umount readahead completion panic

[Brian Foster]

XFS has a bug where directory readahead completions can occur after
unmount. This can lead to a crash or panic because metadata read
verification attempts to access core XFS data structures (e.g., the log)
after they have been freed and certain pointers have been reset.

Add a test that triggers directory readahead, delays the readahead I/O
and immediately unmounts the filesystem. This test is part of the
dangerous group as it will cause kernels affected by the bug to crash.

Signed-off-by: Brian Foster <bfoster@redhat.com>
---
 tests/xfs/999     | 90 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 tests/xfs/999.out |  2 ++
 tests/xfs/group   |  1 +
 3 files changed, 93 insertions(+)
 create mode 100755 tests/xfs/999
 create mode 100644 tests/xfs/999.out

diff --git a/tests/xfs/999 b/tests/xfs/999
new file mode 100755
index 0000000..041d719
--- /dev/null
+++ b/tests/xfs/999
@@ -0,0 +1,90 @@
+#! /bin/bash
+# FS QA Test No. 999
+#
+# Test to reproduce an XFS unmount crash due to races with directory readahead.
+# XFS had a bug in which unmount would proceed with a readahead I/O in flight.
+# If the unmount deconstructed the log by the time I/O completion occurs,
+# certain metadata read verifier checks could access invalid memory and cause a
+# panic.
+#
+#-----------------------------------------------------------------------
+# Copyright (c) 2016 Red Hat, Inc.  All Rights Reserved.
+#
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License as
+# published by the Free Software Foundation.
+#
+# This program is distributed in the hope that it would be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write the Free Software Foundation,
+# Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+#-----------------------------------------------------------------------
+#
+
+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.*
+	_scratch_unmount > /dev/null 2>&1
+	_cleanup_delay > /dev/null 2>&1
+}
+
+# get standard environment, filters and checks
+. ./common/rc
+. ./common/dmdelay
+
+# Modify as appropriate.
+_supported_fs xfs
+_supported_os Linux
+
+_require_scratch
+_require_dm_target delay
+
+rm -f $seqres.full
+
+echo "Silence is golden."
+
+_scratch_mkfs_xfs >> $seqres.full 2>&1
+
+_init_delay
+_mount_delay
+
+# create a directory large enough for extent format
+mkdir $SCRATCH_MNT/dir
+for i in $(seq 0 999); do
+	touch $SCRATCH_MNT/dir/$i
+done
+
+# remount to clear the buffer cache
+_unmount_delay
+_mount_delay
+
+# introduce a read I/O delay
+_load_delay_table $DELAY_READ
+
+# Map the directory and immediately unmount. This should invoke an asynchronous
+# readahead on the first block of the directory. The readahead is delayed by
+# dm-delay. If the unmount doesn't properly wait for the readahead completion,
+# the read verifier can run after core data structures have been freed and lead
+# to a crash/panic.
+$XFS_IO_PROG -c "bmap -v" $SCRATCH_MNT/dir >> $seqres.full 2>&1
+_unmount_delay
+
+_cleanup_delay
+
+# success, all done
+status=0
+exit
diff --git a/tests/xfs/999.out b/tests/xfs/999.out
new file mode 100644
index 0000000..d254382
--- /dev/null
+++ b/tests/xfs/999.out
@@ -0,0 +1,2 @@
+QA output created by 999
+Silence is golden.
diff --git a/tests/xfs/group b/tests/xfs/group
index 5a35a76..6752cc0 100644
--- a/tests/xfs/group
+++ b/tests/xfs/group
@@ -290,3 +290,4 @@
 308 auto quick clone
 309 auto clone
 310 auto clone rmap
+999 auto dangerous quick
-- 
2.5.5

Re: [PATCH 1/2] xfstests: support dm-delay to introduce I/O delays

[Eryu Guan]

On Thu, Jun 30, 2016 at 08:49:33AM -0400, Brian Foster wrote:
> Add some infrastructure in common/dmdelay to support use of the dm-delay
> device-mapper module within tests. This is effectively copied from the
> existing infrastructure in common/dmflakey. This provides the ability to
> delay I/O. It only supports delaying read I/O as that is all that is
> required at this point in time.
> 
> Signed-off-by: Brian Foster <bfoster@redhat.com>
> ---
>  common/dmdelay | 89 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 89 insertions(+)
>  create mode 100644 common/dmdelay
> 
> diff --git a/common/dmdelay b/common/dmdelay
> new file mode 100644
> index 0000000..c53e2dd
> --- /dev/null
> +++ b/common/dmdelay
> @@ -0,0 +1,89 @@
> +##/bin/bash
> +#
> +# Copyright (c) 2016 Red Hat, Inc.  All Rights Reserved.
> +#
> +# This program is free software; you can redistribute it and/or
> +# modify it under the terms of the GNU General Public License as
> +# published by the Free Software Foundation.
> +#
> +# This program is distributed in the hope that it would be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program; if not, write the Free Software Foundation,
> +# Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
> +#
> +#
> +# common functions for setting up and tearing down a dmdelay device
> +
> +DELAY_NONE=0
> +DELAY_READ=1
> +
> +echo $MOUNT_OPTIONS | grep -q dax
> +if [ $? -eq 0 ]; then
> +	_notrun "Cannot run tests with DAX on dmdelay devices"
> +fi
> +
> +_init_delay()
> +{
> +	local BLK_DEV_SIZE=`blockdev --getsz $SCRATCH_DEV`
> +	DELAY_DEV=/dev/mapper/delay-test
> +	DELAY_TABLE="0 $BLK_DEV_SIZE delay $SCRATCH_DEV 0 0"
> +	DELAY_TABLE_RDELAY="0 $BLK_DEV_SIZE delay $SCRATCH_DEV 0 10000 $SCRATCH_DEV 0 0"
> +	$DMSETUP_PROG create delay-test --table "$DELAY_TABLE" || \
> +		_fatal "failed to create delay device"
> +	$DMSETUP_PROG mknodes > /dev/null 2>&1
> +}
> +
> +_mount_delay()
> +{
> +	_scratch_options mount
> +	mount -t $FSTYP $SCRATCH_OPTIONS $MOUNT_OPTIONS $DELAY_DEV $SCRATCH_MNT

I replaced mount with $MOUNT_PROG :)

> +}
> +
> +_unmount_delay()
> +{
> +	$UMOUNT_PROG $SCRATCH_MNT
> +}
> +
> +_cleanup_delay()
> +{
> +	# If dmsetup load fails then we need to make sure to do resume here
> +	# otherwise the umount will hang
> +	$DMSETUP_PROG resume delay-test > /dev/null 2>&1
> +	$UMOUNT_PROG $SCRATCH_MNT > /dev/null 2>&1
> +	# wait for device to be fully settled so that 'dmsetup remove' doesn't
> +	# fail due to EBUSY
> +	$UDEV_SETTLE_PROG >/dev/null 2>&1
> +	$DMSETUP_PROG remove delay-test > /dev/null 2>&1
> +	$DMSETUP_PROG mknodes > /dev/null 2>&1
> +}
> +
> +# _load_delay_table <table> [lockfs]
> +#
> +# This defaults to --nolockfs, which doesn't freeze_fs() before loading the new
> +# table, so it simulates power failure.

Seems this comment is for dmflakey and needs update. Can you please
confirm? I can fix it at commit time if needs update, no v2 is needed.

Thanks,
Eryu

> +_load_delay_table()
> +{
> +	table="$DELAY_TABLE"
> +	[ $1 -eq $DELAY_READ ] && table="$DELAY_TABLE_RDELAY"
> +
> +	suspend_opt="--nolockfs"
> +	[ $# -gt 1 ] && [ $2 -eq 1 ] && suspend_opt=""
> +
> +	# run a suspend/resume cycle to avoid excessive resume delays once a
> +	# delay is introduced below
> +	$DMSETUP_PROG suspend $suspend_opt delay-test
> +	$DMSETUP_PROG resume $suspend_opt delay-test
> +
> +	$DMSETUP_PROG suspend $suspend_opt delay-test
> +	[ $? -ne 0 ] && _fatal "failed to suspend delay-test"
> +
> +	$DMSETUP_PROG load delay-test --table "$table"
> +	[ $? -ne 0 ] && _fatal "failed to load table into delay-test"
> +
> +	$DMSETUP_PROG resume delay-test
> +	[ $? -ne 0 ] && _fatal  "failed to resume delay-test"
> +}
> -- 
> 2.5.5
> 
> --
> To unsubscribe from this list: send the line "unsubscribe fstests" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH 2/2] tests/xfs: test for post umount readahead completion panic

[Eryu Guan]

On Thu, Jun 30, 2016 at 08:49:34AM -0400, Brian Foster wrote:
> XFS has a bug where directory readahead completions can occur after
> unmount. This can lead to a crash or panic because metadata read
> verification attempts to access core XFS data structures (e.g., the log)
> after they have been freed and certain pointers have been reset.
> 
> Add a test that triggers directory readahead, delays the readahead I/O
> and immediately unmounts the filesystem. This test is part of the
> dangerous group as it will cause kernels affected by the bug to crash.
> 
> Signed-off-by: Brian Foster <bfoster@redhat.com>

It crashed my 4.7-rc5 kernel quickly.

Reviewed-by: Eryu Guan <eguan@redhat.com>

Re: [PATCH 1/2] xfstests: support dm-delay to introduce I/O delays

[Brian Foster]

On Thu, Jun 30, 2016 at 09:19:22PM +0800, Eryu Guan wrote:
> On Thu, Jun 30, 2016 at 08:49:33AM -0400, Brian Foster wrote:
> > Add some infrastructure in common/dmdelay to support use of the dm-delay
> > device-mapper module within tests. This is effectively copied from the
> > existing infrastructure in common/dmflakey. This provides the ability to
> > delay I/O. It only supports delaying read I/O as that is all that is
> > required at this point in time.
> > 
> > Signed-off-by: Brian Foster <bfoster@redhat.com>
> > ---
> >  common/dmdelay | 89 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  1 file changed, 89 insertions(+)
> >  create mode 100644 common/dmdelay
> > 
> > diff --git a/common/dmdelay b/common/dmdelay
> > new file mode 100644
> > index 0000000..c53e2dd
> > --- /dev/null
> > +++ b/common/dmdelay
> > @@ -0,0 +1,89 @@
> > +##/bin/bash
> > +#
> > +# Copyright (c) 2016 Red Hat, Inc.  All Rights Reserved.
> > +#
> > +# This program is free software; you can redistribute it and/or
> > +# modify it under the terms of the GNU General Public License as
> > +# published by the Free Software Foundation.
> > +#
> > +# This program is distributed in the hope that it would be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public License
> > +# along with this program; if not, write the Free Software Foundation,
> > +# Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
> > +#
> > +#
> > +# common functions for setting up and tearing down a dmdelay device
> > +
> > +DELAY_NONE=0
> > +DELAY_READ=1
> > +
> > +echo $MOUNT_OPTIONS | grep -q dax
> > +if [ $? -eq 0 ]; then
> > +	_notrun "Cannot run tests with DAX on dmdelay devices"
> > +fi
> > +
> > +_init_delay()
> > +{
> > +	local BLK_DEV_SIZE=`blockdev --getsz $SCRATCH_DEV`
> > +	DELAY_DEV=/dev/mapper/delay-test
> > +	DELAY_TABLE="0 $BLK_DEV_SIZE delay $SCRATCH_DEV 0 0"
> > +	DELAY_TABLE_RDELAY="0 $BLK_DEV_SIZE delay $SCRATCH_DEV 0 10000 $SCRATCH_DEV 0 0"
> > +	$DMSETUP_PROG create delay-test --table "$DELAY_TABLE" || \
> > +		_fatal "failed to create delay device"
> > +	$DMSETUP_PROG mknodes > /dev/null 2>&1
> > +}
> > +
> > +_mount_delay()
> > +{
> > +	_scratch_options mount
> > +	mount -t $FSTYP $SCRATCH_OPTIONS $MOUNT_OPTIONS $DELAY_DEV $SCRATCH_MNT
> 
> I replaced mount with $MOUNT_PROG :)
> 

Thanks.

> > +}
> > +
> > +_unmount_delay()
> > +{
> > +	$UMOUNT_PROG $SCRATCH_MNT
> > +}
> > +
> > +_cleanup_delay()
> > +{
> > +	# If dmsetup load fails then we need to make sure to do resume here
> > +	# otherwise the umount will hang
> > +	$DMSETUP_PROG resume delay-test > /dev/null 2>&1
> > +	$UMOUNT_PROG $SCRATCH_MNT > /dev/null 2>&1
> > +	# wait for device to be fully settled so that 'dmsetup remove' doesn't
> > +	# fail due to EBUSY
> > +	$UDEV_SETTLE_PROG >/dev/null 2>&1
> > +	$DMSETUP_PROG remove delay-test > /dev/null 2>&1
> > +	$DMSETUP_PROG mknodes > /dev/null 2>&1
> > +}
> > +
> > +# _load_delay_table <table> [lockfs]
> > +#
> > +# This defaults to --nolockfs, which doesn't freeze_fs() before loading the new
> > +# table, so it simulates power failure.
> 
> Seems this comment is for dmflakey and needs update. Can you please
> confirm? I can fix it at commit time if needs update, no v2 is needed.
> 

Hmm, the command itself is part of the dmsetup suspend command, but the
power failure bit is probably not relevant since we aren't dropping
I/Os. We could just kill that part of the comment (everything after the
comma)..?

Brian

> Thanks,
> Eryu
> 
> > +_load_delay_table()
> > +{
> > +	table="$DELAY_TABLE"
> > +	[ $1 -eq $DELAY_READ ] && table="$DELAY_TABLE_RDELAY"
> > +
> > +	suspend_opt="--nolockfs"
> > +	[ $# -gt 1 ] && [ $2 -eq 1 ] && suspend_opt=""
> > +
> > +	# run a suspend/resume cycle to avoid excessive resume delays once a
> > +	# delay is introduced below
> > +	$DMSETUP_PROG suspend $suspend_opt delay-test
> > +	$DMSETUP_PROG resume $suspend_opt delay-test
> > +
> > +	$DMSETUP_PROG suspend $suspend_opt delay-test
> > +	[ $? -ne 0 ] && _fatal "failed to suspend delay-test"
> > +
> > +	$DMSETUP_PROG load delay-test --table "$table"
> > +	[ $? -ne 0 ] && _fatal "failed to load table into delay-test"
> > +
> > +	$DMSETUP_PROG resume delay-test
> > +	[ $? -ne 0 ] && _fatal  "failed to resume delay-test"
> > +}
> > -- 
> > 2.5.5
> > 
> > --
> > To unsubscribe from this list: send the line "unsubscribe fstests" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH 1/2] xfstests: support dm-delay to introduce I/O delays

[Eryu Guan]

On Thu, Jun 30, 2016 at 09:44:02AM -0400, Brian Foster wrote:
> On Thu, Jun 30, 2016 at 09:19:22PM +0800, Eryu Guan wrote:
> > On Thu, Jun 30, 2016 at 08:49:33AM -0400, Brian Foster wrote:
> > > Add some infrastructure in common/dmdelay to support use of the dm-delay
> > > device-mapper module within tests. This is effectively copied from the
> > > existing infrastructure in common/dmflakey. This provides the ability to
> > > delay I/O. It only supports delaying read I/O as that is all that is
> > > required at this point in time.
> > > 
> > > Signed-off-by: Brian Foster <bfoster@redhat.com>
> > > ---
> > >  common/dmdelay | 89 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> > >  1 file changed, 89 insertions(+)
> > >  create mode 100644 common/dmdelay
> > > 
> > > diff --git a/common/dmdelay b/common/dmdelay
> > > new file mode 100644
> > > index 0000000..c53e2dd
> > > --- /dev/null
> > > +++ b/common/dmdelay
> > > @@ -0,0 +1,89 @@
> > > +##/bin/bash
> > > +#
> > > +# Copyright (c) 2016 Red Hat, Inc.  All Rights Reserved.
> > > +#
> > > +# This program is free software; you can redistribute it and/or
> > > +# modify it under the terms of the GNU General Public License as
> > > +# published by the Free Software Foundation.
> > > +#
> > > +# This program is distributed in the hope that it would be useful,
> > > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> > > +# GNU General Public License for more details.
> > > +#
> > > +# You should have received a copy of the GNU General Public License
> > > +# along with this program; if not, write the Free Software Foundation,
> > > +# Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
> > > +#
> > > +#
> > > +# common functions for setting up and tearing down a dmdelay device
> > > +
> > > +DELAY_NONE=0
> > > +DELAY_READ=1
> > > +
> > > +echo $MOUNT_OPTIONS | grep -q dax
> > > +if [ $? -eq 0 ]; then
> > > +	_notrun "Cannot run tests with DAX on dmdelay devices"
> > > +fi
> > > +
> > > +_init_delay()
> > > +{
> > > +	local BLK_DEV_SIZE=`blockdev --getsz $SCRATCH_DEV`
> > > +	DELAY_DEV=/dev/mapper/delay-test
> > > +	DELAY_TABLE="0 $BLK_DEV_SIZE delay $SCRATCH_DEV 0 0"
> > > +	DELAY_TABLE_RDELAY="0 $BLK_DEV_SIZE delay $SCRATCH_DEV 0 10000 $SCRATCH_DEV 0 0"
> > > +	$DMSETUP_PROG create delay-test --table "$DELAY_TABLE" || \
> > > +		_fatal "failed to create delay device"
> > > +	$DMSETUP_PROG mknodes > /dev/null 2>&1
> > > +}
> > > +
> > > +_mount_delay()
> > > +{
> > > +	_scratch_options mount
> > > +	mount -t $FSTYP $SCRATCH_OPTIONS $MOUNT_OPTIONS $DELAY_DEV $SCRATCH_MNT
> > 
> > I replaced mount with $MOUNT_PROG :)
> > 
> 
> Thanks.
> 
> > > +}
> > > +
> > > +_unmount_delay()
> > > +{
> > > +	$UMOUNT_PROG $SCRATCH_MNT
> > > +}
> > > +
> > > +_cleanup_delay()
> > > +{
> > > +	# If dmsetup load fails then we need to make sure to do resume here
> > > +	# otherwise the umount will hang
> > > +	$DMSETUP_PROG resume delay-test > /dev/null 2>&1
> > > +	$UMOUNT_PROG $SCRATCH_MNT > /dev/null 2>&1
> > > +	# wait for device to be fully settled so that 'dmsetup remove' doesn't
> > > +	# fail due to EBUSY
> > > +	$UDEV_SETTLE_PROG >/dev/null 2>&1
> > > +	$DMSETUP_PROG remove delay-test > /dev/null 2>&1
> > > +	$DMSETUP_PROG mknodes > /dev/null 2>&1
> > > +}
> > > +
> > > +# _load_delay_table <table> [lockfs]
> > > +#
> > > +# This defaults to --nolockfs, which doesn't freeze_fs() before loading the new
> > > +# table, so it simulates power failure.
> > 
> > Seems this comment is for dmflakey and needs update. Can you please
> > confirm? I can fix it at commit time if needs update, no v2 is needed.
> > 
> 
> Hmm, the command itself is part of the dmsetup suspend command, but the
> power failure bit is probably not relevant since we aren't dropping
> I/Os. We could just kill that part of the comment (everything after the
> comma)..?

Looks fine to me. Updated and committed. Thanks for the test!

Eryu

Re: [PATCH 1/2] xfstests: support dm-delay to introduce I/O delays

[Dave Chinner]

On Thu, Jun 30, 2016 at 09:19:22PM +0800, Eryu Guan wrote:
> On Thu, Jun 30, 2016 at 08:49:33AM -0400, Brian Foster wrote:
> > Add some infrastructure in common/dmdelay to support use of the dm-delay
> > device-mapper module within tests. This is effectively copied from the
> > existing infrastructure in common/dmflakey. This provides the ability to
> > delay I/O. It only supports delaying read I/O as that is all that is
> > required at this point in time.
> > 
> > Signed-off-by: Brian Foster <bfoster@redhat.com>
> > ---
> >  common/dmdelay | 89 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  1 file changed, 89 insertions(+)
> >  create mode 100644 common/dmdelay
> > 
> > diff --git a/common/dmdelay b/common/dmdelay
> > new file mode 100644
> > index 0000000..c53e2dd
> > --- /dev/null
> > +++ b/common/dmdelay
> > @@ -0,0 +1,89 @@
> > +##/bin/bash
> > +#
> > +# Copyright (c) 2016 Red Hat, Inc.  All Rights Reserved.
> > +#
> > +# This program is free software; you can redistribute it and/or
> > +# modify it under the terms of the GNU General Public License as
> > +# published by the Free Software Foundation.
> > +#
> > +# This program is distributed in the hope that it would be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public License
> > +# along with this program; if not, write the Free Software Foundation,
> > +# Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
> > +#
> > +#
> > +# common functions for setting up and tearing down a dmdelay device
> > +
> > +DELAY_NONE=0
> > +DELAY_READ=1
> > +
> > +echo $MOUNT_OPTIONS | grep -q dax
> > +if [ $? -eq 0 ]; then
> > +	_notrun "Cannot run tests with DAX on dmdelay devices"
> > +fi
> > +
> > +_init_delay()
> > +{
> > +	local BLK_DEV_SIZE=`blockdev --getsz $SCRATCH_DEV`
> > +	DELAY_DEV=/dev/mapper/delay-test
> > +	DELAY_TABLE="0 $BLK_DEV_SIZE delay $SCRATCH_DEV 0 0"
> > +	DELAY_TABLE_RDELAY="0 $BLK_DEV_SIZE delay $SCRATCH_DEV 0 10000 $SCRATCH_DEV 0 0"
> > +	$DMSETUP_PROG create delay-test --table "$DELAY_TABLE" || \
> > +		_fatal "failed to create delay device"
> > +	$DMSETUP_PROG mknodes > /dev/null 2>&1
> > +}
> > +
> > +_mount_delay()
> > +{
> > +	_scratch_options mount
> > +	mount -t $FSTYP $SCRATCH_OPTIONS $MOUNT_OPTIONS $DELAY_DEV $SCRATCH_MNT
> 
> I replaced mount with $MOUNT_PROG :)

Shouldn't this also be using _common_dev_mount_options()?, as per
_scratch_mount_options?

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH 2/2] tests/xfs: test for post umount readahead completion panic

[Dave Chinner]

On Thu, Jun 30, 2016 at 08:49:34AM -0400, Brian Foster wrote:
> XFS has a bug where directory readahead completions can occur after
> unmount. This can lead to a crash or panic because metadata read
> verification attempts to access core XFS data structures (e.g., the log)
> after they have been freed and certain pointers have been reset.
> 
> Add a test that triggers directory readahead, delays the readahead I/O
> and immediately unmounts the filesystem. This test is part of the
> dangerous group as it will cause kernels affected by the bug to crash.
.....
> +
> +# create a directory large enough for extent format
> +mkdir $SCRATCH_MNT/dir
> +for i in $(seq 0 999); do
> +	touch $SCRATCH_MNT/dir/$i
> +done

minor quibble - what's an "extent format" directory? I think you
mean a directory inode whose data fork is in extent or btree format,
not inline. i.e. not a short-form directory, but rather a block,
leaf or node format directory.

Yeah, I'm being pedantic, but we should use the correct terminology
so when someone reads it in 5 years time....

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH 2/2] tests/xfs: test for post umount readahead completion panic

[Brian Foster]

On Fri, Jul 01, 2016 at 11:40:02AM +1000, Dave Chinner wrote:
> On Thu, Jun 30, 2016 at 08:49:34AM -0400, Brian Foster wrote:
> > XFS has a bug where directory readahead completions can occur after
> > unmount. This can lead to a crash or panic because metadata read
> > verification attempts to access core XFS data structures (e.g., the log)
> > after they have been freed and certain pointers have been reset.
> > 
> > Add a test that triggers directory readahead, delays the readahead I/O
> > and immediately unmounts the filesystem. This test is part of the
> > dangerous group as it will cause kernels affected by the bug to crash.
> .....
> > +
> > +# create a directory large enough for extent format
> > +mkdir $SCRATCH_MNT/dir
> > +for i in $(seq 0 999); do
> > +	touch $SCRATCH_MNT/dir/$i
> > +done
> 
> minor quibble - what's an "extent format" directory? I think you
> mean a directory inode whose data fork is in extent or btree format,
> not inline. i.e. not a short-form directory, but rather a block,
> leaf or node format directory.
> 

Indeed. So would you prefer the comment refers to the data fork or
directory format? This really cares more about whether the directory has
an extent count than the format per se, because that's what triggers the
readahead.

Perhaps something like: "insert entries to grow the directory to at
least one extent, which is what triggers readahead on dir open" ?

Eryu,

I'm assuming you can fix this up since you fixed up the previous patch.
Let me know if you want me to send another version.

Brian

> Yeah, I'm being pedantic, but we should use the correct terminology
> so when someone reads it in 5 years time....
> 
> Cheers,
> 
> Dave.
> -- 
> Dave Chinner
> david@fromorbit.com

Re: [PATCH 2/2] tests/xfs: test for post umount readahead completion panic

[Eryu Guan]

On Fri, Jul 01, 2016 at 09:03:09AM -0400, Brian Foster wrote:
> On Fri, Jul 01, 2016 at 11:40:02AM +1000, Dave Chinner wrote:
> > On Thu, Jun 30, 2016 at 08:49:34AM -0400, Brian Foster wrote:
> > > XFS has a bug where directory readahead completions can occur after
> > > unmount. This can lead to a crash or panic because metadata read
> > > verification attempts to access core XFS data structures (e.g., the log)
> > > after they have been freed and certain pointers have been reset.
> > > 
> > > Add a test that triggers directory readahead, delays the readahead I/O
> > > and immediately unmounts the filesystem. This test is part of the
> > > dangerous group as it will cause kernels affected by the bug to crash.
> > .....
> > > +
> > > +# create a directory large enough for extent format
> > > +mkdir $SCRATCH_MNT/dir
> > > +for i in $(seq 0 999); do
> > > +	touch $SCRATCH_MNT/dir/$i
> > > +done
> > 
> > minor quibble - what's an "extent format" directory? I think you
> > mean a directory inode whose data fork is in extent or btree format,
> > not inline. i.e. not a short-form directory, but rather a block,
> > leaf or node format directory.
> > 
> 
> Indeed. So would you prefer the comment refers to the data fork or
> directory format? This really cares more about whether the directory has
> an extent count than the format per se, because that's what triggers the
> readahead.
> 
> Perhaps something like: "insert entries to grow the directory to at
> least one extent, which is what triggers readahead on dir open" ?
> 
> Eryu,
> 
> I'm assuming you can fix this up since you fixed up the previous patch.
> Let me know if you want me to send another version.

Sure, I can fix it. And I fixed the mount option issue Dave pointed out
in the dm-delay patch.

Thanks,
Eryu