[1/1] xfs: test xfs_scrub phase 6 media error reporting
diff mbox series

Message ID 158086094326.1990427.7286270181411420127.stgit@magnolia
State New
Headers show
Series
  • xfs: test xfs_scrub media scan
Related show

Commit Message

Darrick J. Wong Feb. 5, 2020, 12:02 a.m. UTC
From: Darrick J. Wong <darrick.wong@oracle.com>

Add new helpers to dmerror to provide for marking selected ranges
totally bad -- both reads and writes will fail.  Create a new test for
xfs_scrub to check that it reports media errors correctly.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
 common/dmerror    |  107 +++++++++++++++++++++++++++++++++++++++++-
 tests/xfs/747     |  136 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 tests/xfs/747.out |   12 +++++
 tests/xfs/748     |  105 +++++++++++++++++++++++++++++++++++++++++
 tests/xfs/748.out |    5 ++
 tests/xfs/group   |    2 +
 6 files changed, 366 insertions(+), 1 deletion(-)
 create mode 100755 tests/xfs/747
 create mode 100644 tests/xfs/747.out
 create mode 100755 tests/xfs/748
 create mode 100644 tests/xfs/748.out

Comments

Eryu Guan March 1, 2020, 4 p.m. UTC | #1
On Tue, Feb 04, 2020 at 04:02:23PM -0800, Darrick J. Wong wrote:
> From: Darrick J. Wong <darrick.wong@oracle.com>
> 
> Add new helpers to dmerror to provide for marking selected ranges
> totally bad -- both reads and writes will fail.  Create a new test for
> xfs_scrub to check that it reports media errors correctly.
> 
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>

So is this expected to fail with latest xfsprogs for-next branch? I got
failures like:

 QA output created by 515
  Scrub for injected media error
 -Corruption: disk offset NNN: media error in inodes. (!)
 -SCRATCH_MNT: Unmount and run xfs_repair.
  Scrub after removing injected media error

Thanks,
Eryu
Darrick J. Wong March 3, 2020, 6:06 p.m. UTC | #2
On Mon, Mar 02, 2020 at 12:00:52AM +0800, Eryu Guan wrote:
> On Tue, Feb 04, 2020 at 04:02:23PM -0800, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@oracle.com>
> > 
> > Add new helpers to dmerror to provide for marking selected ranges
> > totally bad -- both reads and writes will fail.  Create a new test for
> > xfs_scrub to check that it reports media errors correctly.
> > 
> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> 
> So is this expected to fail with latest xfsprogs for-next branch? I got
> failures like:
> 
>  QA output created by 515
>   Scrub for injected media error
>  -Corruption: disk offset NNN: media error in inodes. (!)
>  -SCRATCH_MNT: Unmount and run xfs_repair.

The test should pass ... and I can't reproduce it all here.  What are
you MKFS_OPTIONS and MOUNT_OPTIONS and kernel?  Here's mine:

--D

FSTYP         -- xfs (debug)
PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
MKFS_OPTIONS  -- -f -m reflink=1,rmapbt=1, -i sparse=1, /dev/sdf
MOUNT_OPTIONS -- -o usrquota,grpquota,prjquota, /dev/sdf /opt

xfs/747  3s
Ran: xfs/747
Passed all 1 tests

-------------------

FSTYP         -- xfs (debug)
PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
MKFS_OPTIONS  -- -f -m crc=0,reflink=0,rmapbt=0, -i sparse=0, /dev/sdf
MOUNT_OPTIONS -- -o usrquota,grpquota, /dev/sdf /opt

xfs/747 [not run] crc feature not supported by this filesystem
Ran: xfs/747
Not run: xfs/747
Passed all 1 tests

-------------------

FSTYP         -- xfs (debug)
PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
MKFS_OPTIONS  -- -f -m reflink=1,rmapbt=0, -i sparse=1, /dev/sdf
MOUNT_OPTIONS -- -o usrquota,grpquota,prjquota, /dev/sdf /opt

xfs/747  2s
Ran: xfs/747
Passed all 1 tests

--------------------

FSTYP         -- xfs (debug)
PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
MKFS_OPTIONS  -- -f -m reflink=0,rmapbt=0, -i sparse=1, /dev/sdf
MOUNT_OPTIONS -- -o usrquota,grpquota,prjquota, /dev/sdf /opt

xfs/747  3s
Ran: xfs/747
Passed all 1 tests

--D
Darrick J. Wong March 4, 2020, 3:50 a.m. UTC | #3
On Wed, Mar 04, 2020 at 10:51:32AM +0800, Eryu Guan wrote:
> On Tue, Mar 03, 2020 at 10:06:26AM -0800, Darrick J. Wong wrote:
> > On Mon, Mar 02, 2020 at 12:00:52AM +0800, Eryu Guan wrote:
> > > On Tue, Feb 04, 2020 at 04:02:23PM -0800, Darrick J. Wong wrote:
> > > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > > > 
> > > > Add new helpers to dmerror to provide for marking selected ranges
> > > > totally bad -- both reads and writes will fail.  Create a new test for
> > > > xfs_scrub to check that it reports media errors correctly.
> > > > 
> > > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > 
> > > So is this expected to fail with latest xfsprogs for-next branch? I got
> > > failures like:
> > > 
> > >  QA output created by 515
> > >   Scrub for injected media error
> > >  -Corruption: disk offset NNN: media error in inodes. (!)
> > >  -SCRATCH_MNT: Unmount and run xfs_repair.
> > 
> > The test should pass ... and I can't reproduce it all here.  What are
> > you MKFS_OPTIONS and MOUNT_OPTIONS and kernel?  Here's mine:
> 
> FSTYP         -- xfs (debug)
> PLATFORM      -- Linux/x86_64 fedoravm 5.6.0-rc2 #46 SMP Mon Feb 17 11:37:03 CST 2020
> MKFS_OPTIONS  -- -f -f -b size=4k -m reflink=1,rmapbt=1 /dev/mapper/testvg-lv2
> MOUNT_OPTIONS -- /dev/mapper/testvg-lv2 /mnt/scratch
> 
> xfs/515 - output mismatch (see /root/workspace/xfstests/results//xfs_4k_reflink/xfs/515.out.bad)
>     --- tests/xfs/515.out       2020-03-01 22:42:19.569613781 +0800
>     +++ /root/workspace/xfstests/results//xfs_4k_reflink/xfs/515.out.bad        2020-03-01 23:06:33.546230712 +0800
>     @@ -1,5 +1,3 @@
>      QA output created by 515
>      Scrub for injected media error
>     -Corruption: disk offset NNN: media error in inodes. (!)
>     -SCRATCH_MNT: Unmount and run xfs_repair.
>      Scrub after removing injected media error
> 
> And I'm using xfsprogs for-next branch, HEAD is
> 
> commit fbbb184b189c62beed2a694d14e83bd316fd4140
> Author: Eric Sandeen <sandeen@redhat.com>
> Date:   Thu Feb 27 23:20:42 2020 -0500
> 
>     xfs_repair: join realtime inodes to transaction only once

Hmm, that's really odd.  Can you please send me a metadump of the
scratch fs after the test runs?  I tried your exact mkfs/mount options
and it ran just fine here:

FSTYP         -- xfs (debug)
PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
MKFS_OPTIONS  -- -f -f -b size=4k -m reflink=1,rmapbt=1 /dev/sdf
MOUNT_OPTIONS -- /dev/sdf /opt

xfs/747  3s
Ran: xfs/747
Passed all 1 tests

--D

> > --D
> > 
> > FSTYP         -- xfs (debug)
> > PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
> > MKFS_OPTIONS  -- -f -m reflink=1,rmapbt=1, -i sparse=1, /dev/sdf
> > MOUNT_OPTIONS -- -o usrquota,grpquota,prjquota, /dev/sdf /opt
> 
> I think the problem is the mount option, adding the quota related
> options to my config then test passed as well.
> 
> Thanks,
> Eryu
> 
> > 
> > xfs/747  3s
> > Ran: xfs/747
> > Passed all 1 tests
> > 
> > -------------------
> > 
> > FSTYP         -- xfs (debug)
> > PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
> > MKFS_OPTIONS  -- -f -m crc=0,reflink=0,rmapbt=0, -i sparse=0, /dev/sdf
> > MOUNT_OPTIONS -- -o usrquota,grpquota, /dev/sdf /opt
> > 
> > xfs/747 [not run] crc feature not supported by this filesystem
> > Ran: xfs/747
> > Not run: xfs/747
> > Passed all 1 tests
> > 
> > -------------------
> > 
> > FSTYP         -- xfs (debug)
> > PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
> > MKFS_OPTIONS  -- -f -m reflink=1,rmapbt=0, -i sparse=1, /dev/sdf
> > MOUNT_OPTIONS -- -o usrquota,grpquota,prjquota, /dev/sdf /opt
> > 
> > xfs/747  2s
> > Ran: xfs/747
> > Passed all 1 tests
> > 
> > --------------------
> > 
> > FSTYP         -- xfs (debug)
> > PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
> > MKFS_OPTIONS  -- -f -m reflink=0,rmapbt=0, -i sparse=1, /dev/sdf
> > MOUNT_OPTIONS -- -o usrquota,grpquota,prjquota, /dev/sdf /opt
> > 
> > xfs/747  3s
> > Ran: xfs/747
> > Passed all 1 tests
> > 
> > --D
Eryu Guan March 4, 2020, 6:13 a.m. UTC | #4
On Tue, Mar 03, 2020 at 07:50:57PM -0800, Darrick J. Wong wrote:
> On Wed, Mar 04, 2020 at 10:51:32AM +0800, Eryu Guan wrote:
> > On Tue, Mar 03, 2020 at 10:06:26AM -0800, Darrick J. Wong wrote:
> > > On Mon, Mar 02, 2020 at 12:00:52AM +0800, Eryu Guan wrote:
> > > > On Tue, Feb 04, 2020 at 04:02:23PM -0800, Darrick J. Wong wrote:
> > > > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > > > > 
> > > > > Add new helpers to dmerror to provide for marking selected ranges
> > > > > totally bad -- both reads and writes will fail.  Create a new test for
> > > > > xfs_scrub to check that it reports media errors correctly.
> > > > > 
> > > > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > > 
> > > > So is this expected to fail with latest xfsprogs for-next branch? I got
> > > > failures like:
> > > > 
> > > >  QA output created by 515
> > > >   Scrub for injected media error
> > > >  -Corruption: disk offset NNN: media error in inodes. (!)
> > > >  -SCRATCH_MNT: Unmount and run xfs_repair.
> > > 
> > > The test should pass ... and I can't reproduce it all here.  What are
> > > you MKFS_OPTIONS and MOUNT_OPTIONS and kernel?  Here's mine:
> > 
> > FSTYP         -- xfs (debug)
> > PLATFORM      -- Linux/x86_64 fedoravm 5.6.0-rc2 #46 SMP Mon Feb 17 11:37:03 CST 2020
> > MKFS_OPTIONS  -- -f -f -b size=4k -m reflink=1,rmapbt=1 /dev/mapper/testvg-lv2
> > MOUNT_OPTIONS -- /dev/mapper/testvg-lv2 /mnt/scratch
> > 
> > xfs/515 - output mismatch (see /root/workspace/xfstests/results//xfs_4k_reflink/xfs/515.out.bad)
> >     --- tests/xfs/515.out       2020-03-01 22:42:19.569613781 +0800
> >     +++ /root/workspace/xfstests/results//xfs_4k_reflink/xfs/515.out.bad        2020-03-01 23:06:33.546230712 +0800
> >     @@ -1,5 +1,3 @@
> >      QA output created by 515
> >      Scrub for injected media error
> >     -Corruption: disk offset NNN: media error in inodes. (!)
> >     -SCRATCH_MNT: Unmount and run xfs_repair.
> >      Scrub after removing injected media error
> > 
> > And I'm using xfsprogs for-next branch, HEAD is
> > 
> > commit fbbb184b189c62beed2a694d14e83bd316fd4140
> > Author: Eric Sandeen <sandeen@redhat.com>
> > Date:   Thu Feb 27 23:20:42 2020 -0500
> > 
> >     xfs_repair: join realtime inodes to transaction only once
> 
> Hmm, that's really odd.  Can you please send me a metadump of the
> scratch fs after the test runs?  I tried your exact mkfs/mount options

Sure, please see attachment.

Eryu

> and it ran just fine here:
> 
> FSTYP         -- xfs (debug)
> PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
> MKFS_OPTIONS  -- -f -f -b size=4k -m reflink=1,rmapbt=1 /dev/sdf
> MOUNT_OPTIONS -- /dev/sdf /opt
> 
> xfs/747  3s
> Ran: xfs/747
> Passed all 1 tests
> 
> --D
> 
> > > --D
> > > 
> > > FSTYP         -- xfs (debug)
> > > PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
> > > MKFS_OPTIONS  -- -f -m reflink=1,rmapbt=1, -i sparse=1, /dev/sdf
> > > MOUNT_OPTIONS -- -o usrquota,grpquota,prjquota, /dev/sdf /opt
> > 
> > I think the problem is the mount option, adding the quota related
> > options to my config then test passed as well.
> > 
> > Thanks,
> > Eryu
> > 
> > > 
> > > xfs/747  3s
> > > Ran: xfs/747
> > > Passed all 1 tests
> > > 
> > > -------------------
> > > 
> > > FSTYP         -- xfs (debug)
> > > PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
> > > MKFS_OPTIONS  -- -f -m crc=0,reflink=0,rmapbt=0, -i sparse=0, /dev/sdf
> > > MOUNT_OPTIONS -- -o usrquota,grpquota, /dev/sdf /opt
> > > 
> > > xfs/747 [not run] crc feature not supported by this filesystem
> > > Ran: xfs/747
> > > Not run: xfs/747
> > > Passed all 1 tests
> > > 
> > > -------------------
> > > 
> > > FSTYP         -- xfs (debug)
> > > PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
> > > MKFS_OPTIONS  -- -f -m reflink=1,rmapbt=0, -i sparse=1, /dev/sdf
> > > MOUNT_OPTIONS -- -o usrquota,grpquota,prjquota, /dev/sdf /opt
> > > 
> > > xfs/747  2s
> > > Ran: xfs/747
> > > Passed all 1 tests
> > > 
> > > --------------------
> > > 
> > > FSTYP         -- xfs (debug)
> > > PLATFORM      -- Linux/x86_64 magnolia-mtr00 5.6.0-rc4-xfsx #rc4 SMP PREEMPT Mon Mar 2 21:02:17 PST 2020
> > > MKFS_OPTIONS  -- -f -m reflink=0,rmapbt=0, -i sparse=1, /dev/sdf
> > > MOUNT_OPTIONS -- -o usrquota,grpquota,prjquota, /dev/sdf /opt
> > > 
> > > xfs/747  3s
> > > Ran: xfs/747
> > > Passed all 1 tests
> > > 
> > > --D

Patch
diff mbox series

diff --git a/common/dmerror b/common/dmerror
index 426f1e96..3eb7a2d7 100644
--- a/common/dmerror
+++ b/common/dmerror
@@ -65,7 +65,7 @@  _dmerror_load_error_table()
 	$DMSETUP_PROG suspend $suspend_opt error-test
 	[ $? -ne 0 ] && _fail  "dmsetup suspend failed"
 
-	$DMSETUP_PROG load error-test --table "$DMERROR_TABLE"
+	echo "$DMERROR_TABLE" | $DMSETUP_PROG load error-test
 	load_res=$?
 
 	$DMSETUP_PROG resume error-test
@@ -100,3 +100,108 @@  _dmerror_load_working_table()
 	[ $load_res -ne 0 ] && _fail "dmsetup failed to load error table"
 	[ $resume_res -ne 0 ] && _fail  "dmsetup resume failed"
 }
+
+# Given a list of (start, length) tuples on stdin, combine adjacent tuples into
+# larger ones and write the new list to stdout.
+__dmerror_combine_extents()
+{
+	awk 'BEGIN{start = 0; len = 0;}{
+if (start + len == $1) {
+	len += $2;
+} else {
+	if (len > 0)
+		printf("%d %d\n", start, len);
+	start = $1;
+	len = $2;
+}
+} END {
+	if (len > 0)
+		printf("%d %d\n", start, len);
+}'
+}
+
+# Given a block device, the name of a preferred dm target, the name of an
+# implied dm target, and a list of (start, len) tuples on stdin, create a new
+# dm table which maps each of the tuples to the preferred target and all other
+# areas to the implied dm target.
+__dmerror_recreate_map()
+{
+	local device="$1"
+	local preferred_tgt="$2"
+	local implied_tgt="$3"
+	local size=$(blockdev --getsz "$device")
+
+	awk -v device="$device" -v size=$size -v implied_tgt="$implied_tgt" \
+		-v preferred_tgt="$preferred_tgt" 'BEGIN{implied_start = 0;}{
+	extent_start = $1;
+	extent_len = $2;
+
+	if (extent_start > size) {
+		extent_start = size;
+		extent_len = 0;
+	} else if (extent_start + extent_len > size) {
+		extent_len = size - extent_start;
+	}
+
+	if (implied_start < extent_start)
+		printf("%d %d %s %s %d\n", implied_start,
+				extent_start - implied_start, implied_tgt,
+				device, implied_start);
+	printf("%d %d %s %s %d\n", extent_start, extent_len, preferred_tgt,
+			device, extent_start);
+	implied_start = extent_start + extent_len;
+}END{
+	if (implied_start < size)
+		printf("%d %d %s %s %d\n", implied_start, size - implied_start,
+				implied_tgt, device, implied_start);
+}'
+}
+
+# Update the dm error table so that the range (start, len) maps to the
+# preferred dm target, overriding anything that maps to the implied dm target.
+# This assumes that the only desired targets for this dm device are the
+# preferred and and implied targets.  The optional fifth argument can be used
+# to change the underlying device.
+__dmerror_change()
+{
+	local start="$1"
+	local len="$2"
+	local preferred_tgt="$3"
+	local implied_tgt="$4"
+	local dm_backing_dev="$5"
+	test -z "$dm_backing_dev" && dm_backing_dev="$SCRATCH_DEV"
+
+	DMERROR_TABLE="$( (echo "$DMERROR_TABLE"; echo "$start $len $preferred_tgt") | \
+		awk -v type="$preferred_tgt" '{if ($3 == type) print $0;}' | \
+		sort -g | \
+		__dmerror_combine_extents | \
+		__dmerror_recreate_map "$dm_backing_dev" "$preferred_tgt" \
+				"$implied_tgt" )"
+}
+
+# Reset the dm error table to everything ok.  The dm device itself must be
+# remapped by calling _dmerror_load_error_table.
+_dmerror_reset_table()
+{
+	DMERROR_TABLE="$DMLINEAR_TABLE"
+}
+
+# Update the dm error table so that IOs to the given range will return EIO.
+# The dm device itself must be remapped by calling _dmerror_load_error_table.
+_dmerror_mark_range_bad()
+{
+	local start="$1"
+	local len="$2"
+
+	__dmerror_change "$start" "$len" error linear
+}
+
+# Update the dm error table so that IOs to the given range will succeed.
+# The dm device itself must be remapped by calling _dmerror_load_error_table.
+_dmerror_mark_range_good()
+{
+	local start="$1"
+	local len="$2"
+
+	__dmerror_change "$start" "$len" linear error
+}
diff --git a/tests/xfs/747 b/tests/xfs/747
new file mode 100755
index 00000000..0fd666c3
--- /dev/null
+++ b/tests/xfs/747
@@ -0,0 +1,136 @@ 
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0-or-later
+# Copyright (c) 2020, Oracle and/or its affiliates.  All Rights Reserved.
+#
+# FS QA Test No. 747
+#
+# Check xfs_scrub's media scan can actually return diagnostic information for
+# media errors in file data extents.
+
+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.*
+	_dmerror_cleanup
+}
+
+# get standard environment, filters and checks
+. ./common/rc
+. ./common/fuzzy
+. ./common/filter
+. ./common/dmerror
+
+# real QA test starts here
+_supported_fs xfs
+_supported_os Linux
+_require_dm_target error
+_require_scratch_xfs_crc
+_require_scrub
+
+rm -f $seqres.full
+
+filter_scrub_errors() {
+	_filter_scratch | sed -e "s/offset $((blksz * 2)) /offset 2FSB /g" \
+		-e "s/length $blksz.*/length 1FSB./g"
+}
+
+_scratch_mkfs > $tmp.mkfs
+_dmerror_init
+_dmerror_mount >> $seqres.full 2>&1
+
+_supports_xfs_scrub $SCRATCH_MNT $SCRATCH_DEV || _notrun "Scrub not supported"
+
+victim=$SCRATCH_MNT/a
+$XFS_IO_PROG -f -c "pwrite -S 0x58 0 1m" -c "fsync" $victim >> $seqres.full
+bmap_str="$($XFS_IO_PROG -c "bmap -elpv" $victim | grep "^[[:space:]]*0:")"
+echo "$bmap_str" >> $seqres.full
+
+phys="$(echo "$bmap_str" | $AWK_PROG '{print $3}')"
+len="$(echo "$bmap_str" | $AWK_PROG '{print $6}')"
+blksz=$(_get_file_block_size $SCRATCH_MNT)
+sectors_per_block=$((blksz / 512))
+
+# Did we get at least 4 fs blocks worth of extent?
+min_len_sectors=$(( 4 * sectors_per_block ))
+test "$len" -lt $min_len_sectors && \
+	_fail "could not format a long enough extent on an empty fs??"
+
+phys_start=$(echo "$phys" | sed -e 's/\.\..*//g')
+
+
+echo ":$phys:$len:$blksz:$phys_start" >> $seqres.full
+echo "victim file:" >> $seqres.full
+od -tx1 -Ad -c $victim >> $seqres.full
+
+# Reset the dmerror table so that all IO will pass through.
+_dmerror_reset_table
+
+cat >> $seqres.full << ENDL
+dmerror before:
+$DMERROR_TABLE
+<end table>
+ENDL
+
+# Now mark /only/ the middle of the extent bad.
+_dmerror_mark_range_bad $(( phys_start + (2 * sectors_per_block) + 1 )) 1
+
+cat >> $seqres.full << ENDL
+dmerror after marking bad:
+$DMERROR_TABLE
+<end table>
+ENDL
+
+_dmerror_load_error_table
+
+# See if the media scan picks it up.
+echo "Scrub for injected media error (single threaded)"
+
+# Once in single-threaded mode
+_scratch_scrub -b -x >> $seqres.full 2> $tmp.error
+cat $tmp.error | filter_scrub_errors
+
+# Once in parallel mode
+echo "Scrub for injected media error (multi threaded)"
+_scratch_scrub -x >> $seqres.full 2> $tmp.error
+cat $tmp.error | filter_scrub_errors
+
+# Remount to flush the page cache and reread to see the IO error
+_dmerror_unmount
+_dmerror_mount
+echo "victim file:" >> $seqres.full
+od -tx1 -Ad -c $victim >> $seqres.full 2> $tmp.error
+cat $tmp.error | _filter_scratch
+
+# Scrub again to re-confirm the media error across a remount
+echo "Scrub for injected media error (after remount)"
+_scratch_scrub -x >> $seqres.full 2> $tmp.error
+cat $tmp.error | filter_scrub_errors
+
+# Now mark the bad range good.
+_dmerror_mark_range_good $(( phys_start + (2 * sectors_per_block) + 1 )) 1
+_dmerror_load_error_table
+
+cat >> $seqres.full << ENDL
+dmerror after marking good:
+$DMERROR_TABLE
+<end table>
+ENDL
+
+echo "Scrub after removing injected media error"
+
+# Scrub one last time to make sure the error's gone.
+_scratch_scrub -x >> $seqres.full 2> $tmp.error
+cat $tmp.error | filter_scrub_errors
+
+# success, all done
+status=0
+exit
diff --git a/tests/xfs/747.out b/tests/xfs/747.out
new file mode 100644
index 00000000..f85f1753
--- /dev/null
+++ b/tests/xfs/747.out
@@ -0,0 +1,12 @@ 
+QA output created by 747
+Scrub for injected media error (single threaded)
+Unfixable Error: SCRATCH_MNT/a: media error at data offset 2FSB length 1FSB.
+SCRATCH_MNT: unfixable errors found: 1
+Scrub for injected media error (multi threaded)
+Unfixable Error: SCRATCH_MNT/a: media error at data offset 2FSB length 1FSB.
+SCRATCH_MNT: unfixable errors found: 1
+od: SCRATCH_MNT/a: read error: Input/output error
+Scrub for injected media error (after remount)
+Unfixable Error: SCRATCH_MNT/a: media error at data offset 2FSB length 1FSB.
+SCRATCH_MNT: unfixable errors found: 1
+Scrub after removing injected media error
diff --git a/tests/xfs/748 b/tests/xfs/748
new file mode 100755
index 00000000..0168b1ee
--- /dev/null
+++ b/tests/xfs/748
@@ -0,0 +1,105 @@ 
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0-or-later
+# Copyright (c) 2020, Oracle and/or its affiliates.  All Rights Reserved.
+#
+# FS QA Test No. 748
+#
+# Check xfs_scrub's media scan can actually return diagnostic information for
+# media errors in filesystem metadata.
+
+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.*
+	_dmerror_cleanup
+}
+
+# get standard environment, filters and checks
+. ./common/rc
+. ./common/fuzzy
+. ./common/filter
+. ./common/dmerror
+
+# real QA test starts here
+_supported_fs xfs
+_supported_os Linux
+_require_dm_target error
+
+# rmapbt is required to enable reporting of what metadata was lost.
+_require_xfs_scratch_rmapbt
+
+_require_scrub
+
+rm -f $seqres.full
+
+filter_scrub_errors() {
+	_filter_scratch | sed -e "s/disk offset [0-9]*: /disk offset NNN: /g" \
+		-e "/errors found:/d" -e 's/phase6.c line [0-9]*/!/g' \
+		-e "/corruptions found:/d" | uniq
+}
+
+_scratch_mkfs > $tmp.mkfs
+_dmerror_init
+_dmerror_mount >> $seqres.full 2>&1
+
+_supports_xfs_scrub $SCRATCH_MNT $SCRATCH_DEV || _notrun "Scrub not supported"
+
+# Create a bunch of metadata so that we can mark them bad in the next step.
+victim=$SCRATCH_MNT/a
+$FSSTRESS_PROG -z -n 200 -p 10 \
+	       -f creat=10 \
+	       -f resvsp=1 \
+	       -f truncate=1 \
+	       -f punch=1 \
+	       -f chown=5 \
+	       -f mkdir=5 \
+	       -f mknod=1 \
+	       -d $victim >> $seqres.full 2>&1
+
+# Mark all the metadata bad
+_dmerror_reset_table
+$XFS_IO_PROG -c "fsmap -n100 -vvv" $victim | grep inodes > $tmp.fsmap
+while read a b c crap; do
+	phys="$(echo $c | sed -e 's/^.\([0-9]*\)\.\.\([0-9]*\).*$/\1:\2/g')"
+	target_begin="$(echo "$phys" | cut -d ':' -f 1)"
+	target_end="$(echo "$phys" | cut -d ':' -f 2)"
+
+	_dmerror_mark_range_bad $target_begin $((target_end - target_begin))
+done < $tmp.fsmap
+cat $tmp.fsmap >> $seqres.full
+
+cat >> $seqres.full << ENDL
+dmerror after marking bad:
+$DMERROR_TABLE
+<end table>
+ENDL
+
+_dmerror_load_error_table
+
+# See if the media scan picks it up.
+echo "Scrub for injected media error"
+
+XFS_SCRUB_PHASE=6 _scratch_scrub -x >> $seqres.full 2> $tmp.error
+cat $tmp.error | filter_scrub_errors
+
+# Make the disk work again
+_dmerror_load_working_table
+
+echo "Scrub after removing injected media error"
+
+# Scrub one last time to make sure the error's gone.
+XFS_SCRUB_PHASE=6 _scratch_scrub -x >> $seqres.full 2> $tmp.error
+cat $tmp.error | filter_scrub_errors
+
+# success, all done
+status=0
+exit
diff --git a/tests/xfs/748.out b/tests/xfs/748.out
new file mode 100644
index 00000000..49dc2d7a
--- /dev/null
+++ b/tests/xfs/748.out
@@ -0,0 +1,5 @@ 
+QA output created by 748
+Scrub for injected media error
+Corruption: disk offset NNN: media error in inodes. (!)
+SCRATCH_MNT: Unmount and run xfs_repair.
+Scrub after removing injected media error
diff --git a/tests/xfs/group b/tests/xfs/group
index 45dd8868..edffef9a 100644
--- a/tests/xfs/group
+++ b/tests/xfs/group
@@ -510,3 +510,5 @@ 
 510 auto ioctl quick
 511 auto quick quota
 512 auto quick acl attr
+747 auto quick scrub
+748 auto quick scrub