Skip to contentSkip to conversation

btrfs: Add a test for leaking root crash at unmount time

PATCH New Qu Wenruo 2 participants 3 messages May 20, 2020 Btrfs file system

lore
Share & tools
lore

Qu Wenruo

author
#0 lore raw

PATCH v2 1/2 btrfs: Add a test for leaking root crash at unmount time

Test if canceled balance could lead to root leakage.
If the kernel has CONFIG_BTRFS_DEBUG compiled, unmount time root leakge check would detect it, and cause NULL pointer dereference as the pages of the leaked root are already freed.

The fix is titled "btrfs: relocation: Fix reloc root leakage and the NULL
pointer reference caused by the leakage".

---
Changelog:
v2:
- Fix gramma and typos
- Use _run_btrfs_balance_start()

---
 tests/btrfs/212     | 85 +++++++++++++++++++++++++++++++++++++++++++++
 tests/btrfs/212.out |  2 ++
 tests/btrfs/group   |  1 +
 3 files changed, 88 insertions(+)
 create mode 100755 tests/btrfs/212
 create mode 100644 tests/btrfs/212.out
tests/btrfs/212 new file +85 -0
new file mode 100755
index 00000000..d2a7b0d7
@@ -0,0 +1,85 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (C) 2020 SUSE Linux Products GmbH. All Rights Reserved.
+#
+# FS QA Test 212
+#
+# Test if unmounting a fs with balance canceled can lead to crash.
+# This needs CONFIG_BTRFS_DEBUG compiled, which adds extra unmount time self-test
+#
+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.*
+	kill $balance_pid &> /dev/null
+	kill $cancel_pid &> /dev/null
+	"$KILLALL_PROG" -q $FSSTRESS_PROG &> /dev/null
+	$BTRFS_UTIL_PROG balance cancel $SCRATCH_MNT &> /dev/null
+	wait
+}
+
+# get standard environment, filters and checks
+. ./common/rc
+. ./common/filter
+
+# remove previous $seqres.full before test
+rm -f $seqres.full
+
+# real QA test starts here
+
+# Modify as appropriate.
+_supported_fs btrfs 
+_supported_os Linux
+_require_scratch
+_require_command "$KILLALL_PROG" killall
+
+_scratch_mkfs >> $seqres.full
+_scratch_mount
+
+runtime=15
+
+balance_workload()
+{
+	trap "wait; exit" SIGTERM
+	while true; do
+		_run_btrfs_balance_start &> /dev/null
+	done
+}
+
+cancel_workload()
+{
+	trap "wait; exit" SIGTERM
+	while true; do
+		$BTRFS_UTIL_PROG balance cancel $SCRATCH_MNT &> /dev/null
+		sleep 2
+	done
+}
+
+$FSSTRESS_PROG -d $SCRATCH_MNT -w -n 100000  >> $seqres.full 2>/dev/null &
+balance_workload &
+balance_pid=$!
+
+cancel_workload &
+cancel_pid=$!
+
+sleep $runtime
+
+kill $balance_pid
+kill $cancel_pid
+"$KILLALL_PROG" -q $FSSTRESS_PROG &> /dev/null
+$BTRFS_UTIL_PROG balance cancel $SCRATCH_MNT &> /dev/null
+wait
+
+echo "Silence is golden"
+# success, all done
+status=0
+exit
tests/btrfs/212.out new file +2 -0
new file mode 100644
index 00000000..32d11390
@@ -0,0 +1,2 @@
+QA output created by 212
+Silence is golden
tests/btrfs/group +1 -0
index 66b1beac..8d65bddd 100644
@@ -214,3 +214,4 @@
 209 auto quick log
 210 auto quick qgroup snapshot
 211 auto quick log prealloc
+212 auto balance dangerous
Signature · 1 line
2.26.2
author
#1 lore raw

PATCH 2/2 btrfs: Add a test for dead looping balance after balance cancel

Test if canceling a running balance can cause later balance to dead loop.

The fix is titled "btrfs: relocation: Clear the DEAD_RELOC_TREE bit for
orphan roots to prevent runaway balance".

---
Changelog:
v2:
- Remove lsof debug output
v3:
- Remove ps debug output
v4:
- Use $XFS_IO_PROG directly to avoid wrapped dd command This allows us to kill the writer and wait it correctly, other than killing the bash process running the wrapper function.
- Fix typos
- Use _run_btrfs_balance_start() wrapper

---
 tests/btrfs/213     | 65 +++++++++++++++++++++++++++++++++++++++++++++
 tests/btrfs/213.out |  2 ++
 tests/btrfs/group   |  1 +
 3 files changed, 68 insertions(+)
 create mode 100755 tests/btrfs/213
 create mode 100644 tests/btrfs/213.out
tests/btrfs/213 new file +65 -0
new file mode 100755
index 00000000..a3a2afe0
@@ -0,0 +1,65 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (C) 2020 SUSE Linux Products GmbH. All Rights Reserved.
+#
+# FS QA Test 213
+#
+# Test if canceling a running balance can lead to dead looping balance
+#
+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
+
+# remove previous $seqres.full before test
+rm -f $seqres.full
+
+# Modify as appropriate.
+_supported_fs btrfs
+_supported_os Linux
+_require_scratch
+_require_xfs_io_command pwrite -D
+
+_scratch_mkfs >> $seqres.full
+_scratch_mount
+
+runtime=4
+
+# Create enough IO so that we need around $runtime seconds to relocate it.
+#
+# Here we don't want any wrapper, as we want full control of the process.
+$XFS_IO_PROG -f -c "pwrite -D -b 1M 0 1024T" "$SCRATCH_MNT/file" &> /dev/null &
+write_pid=$!
+sleep $runtime
+kill $write_pid
+wait $write_pid
+
+# Now balance should take at least $runtime seconds, we can cancel it at
+# $runtime/2 to ensure a success cancel.
+_run_btrfs_balance_start -d --bg "$SCRATCH_MNT"
+sleep $(($runtime / 2))
+$BTRFS_UTIL_PROG balance cancel "$SCRATCH_MNT"
+
+# Now check if we can finish relocating metadata, which should finish very
+# quickly.
+$BTRFS_UTIL_PROG balance start -m "$SCRATCH_MNT" >> $seqres.full
+
+echo "Silence is golden"
+
+# success, all done
+status=0
+exit
tests/btrfs/213.out new file +2 -0
new file mode 100644
index 00000000..bd8f2430
@@ -0,0 +1,2 @@
+QA output created by 213
+Silence is golden
tests/btrfs/group +1 -0
index 8d65bddd..59e8ecce 100644
@@ -215,3 +215,4 @@
 210 auto quick qgroup snapshot
 211 auto quick log prealloc
 212 auto balance dangerous
+213 auto quick balance dangerous
Signature · 1 line
2.26.2
Reviewed-by on v1 2/2
#2 lore raw

Re: [PATCH 2/2] btrfs: Add a test for dead looping balance after balance cancel

Quoted 71 lines · Qu Wenruo

Test if canceling a running balance can cause later balance to dead loop.

The fix is titled "btrfs: relocation: Clear the DEAD_RELOC_TREE bit for
orphan roots to prevent runaway balance".

---
Changelog:
v2:
- Remove lsof debug output
v3:
- Remove ps debug output
v4:
- Use $XFS_IO_PROG directly to avoid wrapped dd command This allows us to kill the writer and wait it correctly, other than killing the bash process running the wrapper function.
- Fix typos
- Use _run_btrfs_balance_start() wrapper

---
 tests/btrfs/213     | 65 +++++++++++++++++++++++++++++++++++++++++++++
 tests/btrfs/213.out |  2 ++
 tests/btrfs/group   |  1 +
 3 files changed, 68 insertions(+)
 create mode 100755 tests/btrfs/213
 create mode 100644 tests/btrfs/213.out
tests/btrfs/213 new file +37 -0
new file mode 100755
index 00000000..a3a2afe0
@@ -0,0 +1,65 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (C) 2020 SUSE Linux Products GmbH. All Rights Reserved.
+#
+# FS QA Test 213
+#
+# Test if canceling a running balance can lead to dead looping balance
+#
+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
+
+# remove previous $seqres.full before test
+rm -f $seqres.full
+
+# Modify as appropriate.
+_supported_fs btrfs
+_supported_os Linux
+_require_scratch
+_require_xfs_io_command pwrite -D
+
+_scratch_mkfs >> $seqres.full

tests/btrfs/213 line 47

Quoted 10 lines
tests/btrfs/213 new file +10 -0
+_scratch_mount
+
+runtime=4
+
+# Create enough IO so that we need around $runtime seconds to relocate it.
+#
+# Here we don't want any wrapper, as we want full control of the process.
+$XFS_IO_PROG -f -c "pwrite -D -b 1M 0 1024T" "$SCRATCH_MNT/file" &> /dev/null &
+write_pid=$!
+sleep $runtime

Probably you forgot, but as I said before, we should make sure the xfs_io process is killed in _cleanup() too, in case we abort the test while it is in that sleep above.

With that added,

Thanks.

Quoted 38 lines
tests/btrfs/213 new file +18 -0
+kill $write_pid
+wait $write_pid
+
+# Now balance should take at least $runtime seconds, we can cancel it at
+# $runtime/2 to ensure a success cancel.
+_run_btrfs_balance_start -d --bg "$SCRATCH_MNT"
+sleep $(($runtime / 2))
+$BTRFS_UTIL_PROG balance cancel "$SCRATCH_MNT"
+
+# Now check if we can finish relocating metadata, which should finish very
+# quickly.
+$BTRFS_UTIL_PROG balance start -m "$SCRATCH_MNT" >> $seqres.full
+
+echo "Silence is golden"
+
+# success, all done
+status=0
+exit
tests/btrfs/213.out new file +2 -0
new file mode 100644
index 00000000..bd8f2430
@@ -0,0 +1,2 @@
+QA output created by 213
+Silence is golden
tests/btrfs/group +1 -1
index 8d65bddd..59e8ecce 100644
@@ -215,3 +215,4 @@
 210 auto quick qgroup snapshot
 211 auto quick log prealloc
 212 auto balance dangerous
+213 auto quick balance dangerous
--

2.26.2

Signature · 3 lines
Filipe David Manana,

“Whether you think you can, or you think you can't — you're right.”