Skip to contentSkip to conversation

fstests: more random fixes from Ted

PATCH In review Theodore Ts'o 3 participants 9 messages Jul 20, 2022 – Jul 21, 2022

lore
Share & tools
lore
#0 lore raw

cover fstests: more random fixes from Ted

Here are few more hopefully non-controvial patches from my tytso-local branch on https://github.com/tytso/xfstests. The xuniit-quet patch has previously been posted; this has a minor clarification in the commit description, but is otherwise unchanged from the previous v2 version on the list.

The other fixes fixes some problems found when running tests with an external log, and to address generic/556 failing on older LTS kernels.

Theodore Ts'o (4):
report: add support for the xunit-quiet format generic/556: add a check to make sure ext4 supports encrypted
casefolding
dmflakey: don't run dmflakey tests with an external log device generic/475: skip test when using an external log device

 check             |  2 +-
 common/casefold   | 17 +++++++++++++++++
 common/dmflakey   |  4 ++++
 common/report     | 21 ++++++++++++++-------
 tests/generic/475 |  4 ++++
 tests/generic/556 |  1 +
 6 files changed, 41 insertions(+), 8 deletions(-)
Signature · 1 line
2.31.0
#1 lore raw

PATCH 1/4 report: add support for the xunit-quiet format

The xunit-quiet format excludes the NNN.{full,dmesg,bad} files in <system-out> and <system-err> nodes which are included in the xunit report format.

For test runners that save the entire results directory to preserve all of the test artifacts, capturing the NNN.{full,dmesg,bad} in the results.xml file is redundant. In addition, if the NNN.bad is too large, it can cause the junitparser python library to refuse to parse the XML file to prevent potential denial of service attacks[1]. A simple way to avoid this problem is to simply to omit the <system-out> and <system-err> nodes in the results.xml file.

[1] https://gitlab.com/gitlab-org/gitlab/-/issues/268035

---
 check         |  2 +-
 common/report | 21 ++++++++++++++-------
 2 files changed, 15 insertions(+), 8 deletions(-)
check +1 -1
index 5f6d86b4..0b2f10ed 100755
@@ -75,7 +75,7 @@ check options
     -I <n>		iterate the test list <n> times, but stops iterating further in case of any test failure
     -d			dump test output to stdout
     -b			brief test summary
-    -R fmt[,fmt]	generate report in formats specified. Supported format: [xunit]
+    -R fmt[,fmt]	generate report in formats specified. Supported formats: xunit, xunit-quiet
     --large-fs		optimise scratch device for large filesystems
     -s section		run only specified section from config file
     -S section		exclude the specified section from the config file
common/report +14 -7
index 5ca41bc4..64f9c866 100644
@@ -71,11 +71,16 @@ _xunit_make_testcase_report()
 	local test_name="$2"
 	local test_status="$3"
 	local test_time="$4"
+	local report_format="$5"
+	local quiet
+
+	if [ "$report_format" = xunit-quiet ]; then
+		quiet=yes
+	fi
 
 	# TODO: other places may also win if no-section mode will be named like 'default/global'
 	if [ $sect_name == '-no-sections-' ]; then
 		sect_name='global'
-
 	fi
 	local report=$tmp.report.xunit.$sect_name.xml
 
@@ -104,14 +109,16 @@ _xunit_make_testcase_report()
 			_err_msg="Test $test_name failed, reason unknown"
 		fi
 		echo -e "\t\t<failure message=\"$_err_msg\" type=\"TestFail\" />" >> $report
-		if [ -s "$full_file" ]; then
+		if [ -z "$quiet" -a -s "$full_file" ]; then
 			echo -e "\t\t<system-out>" >> $report
 			printf	'<![CDATA[\n' >>$report
 			cat "$full_file" | tr -dc '[:print:][:space:]' | encode_xml >>$report
 			printf ']]>\n'	>>$report
 			echo -e "\t\t</system-out>" >> $report
 		fi
-		if [ -f "$dmesg_file" ]; then
+		if [ -n "$quiet" ]; then
+		   :
+		elif [ -f "$dmesg_file" ]; then
 			echo -e "\t\t<system-err>" >> $report
 			printf	'<![CDATA[\n' >>$report
 			cat "$dmesg_file" | tr -dc '[:print:][:space:]' | encode_xml >>$report
@@ -144,7 +151,7 @@ _make_section_report()
 	local sect_time="$5"
 	for report in $REPORT_LIST; do
 		case "$report" in
-		"xunit")
+		"xunit"|"xunit-quiet")
 			_xunit_make_section_report "$sect_name" "$tests_count" \
 						   "$bad_count" "$notrun_count" \
 						   "$sect_time"
@@ -164,9 +171,9 @@ _make_testcase_report()
 	local test_time="$4"
 	for report in $REPORT_LIST; do
 		case "$report" in
-		"xunit")
+		"xunit"|"xunit-quiet")
 			_xunit_make_testcase_report "$sect_name" "$test_seq" \
-						    "$test_status" "$test_time"
+						    "$test_status" "$test_time" "$report"
 			;;
 		*)
 			_dump_err "report format '$report' is not supported"
@@ -178,7 +185,7 @@ _make_testcase_report()
 _assert_report_list() {
 	for report in $REPORT_LIST; do
 		case "$report" in
-		"xunit")
+		"xunit"|"xunit-quiet")
 			;;
 		*)
 			_fatal "report format '$report' is not supported"
Signature · 1 line
2.31.0
Reviewed-by on v1 1/4
#2 lore raw

Re: [PATCH 1/4] report: add support for the xunit-quiet format

Quoted 12 lines · Theodore Ts'o

The xunit-quiet format excludes the NNN.{full,dmesg,bad} files in <system-out> and <system-err> nodes which are included in the xunit report format.

For test runners that save the entire results directory to preserve all of the test artifacts, capturing the NNN.{full,dmesg,bad} in the results.xml file is redundant. In addition, if the NNN.bad is too large, it can cause the junitparser python library to refuse to parse the XML file to prevent potential denial of service attacks[1]. A simple way to avoid this problem is to simply to omit the <system-out> and <system-err> nodes in the results.xml file.

Quoted 4 lines

[1] https://gitlab.com/gitlab-org/gitlab/-/issues/268035

Looks fine to me, though I wonder if we should document what these report formats actually do?

I really dislike the "xunit" name since AFAICT it's really the junit xml format, not the xunit xml format, and this trips me up **every** single time I have to go look at the fstests reporting code.

For this bit though,

--D

Quoted 92 lines
---
 check         |  2 +-
 common/report | 21 ++++++++++++++-------
 2 files changed, 15 insertions(+), 8 deletions(-)
check +1 -1
index 5f6d86b4..0b2f10ed 100755
@@ -75,7 +75,7 @@ check options
     -I <n>		iterate the test list <n> times, but stops iterating further in case of any test failure
     -d			dump test output to stdout
     -b			brief test summary
-    -R fmt[,fmt]	generate report in formats specified. Supported format: [xunit]
+    -R fmt[,fmt]	generate report in formats specified. Supported formats: xunit, xunit-quiet
     --large-fs		optimise scratch device for large filesystems
     -s section		run only specified section from config file
     -S section		exclude the specified section from the config file
common/report +14 -7
index 5ca41bc4..64f9c866 100644
@@ -71,11 +71,16 @@ _xunit_make_testcase_report()
 	local test_name="$2"
 	local test_status="$3"
 	local test_time="$4"
+	local report_format="$5"
+	local quiet
+
+	if [ "$report_format" = xunit-quiet ]; then
+		quiet=yes
+	fi
 
 	# TODO: other places may also win if no-section mode will be named like 'default/global'
 	if [ $sect_name == '-no-sections-' ]; then
 		sect_name='global'
-
 	fi
 	local report=$tmp.report.xunit.$sect_name.xml
 
@@ -104,14 +109,16 @@ _xunit_make_testcase_report()
 			_err_msg="Test $test_name failed, reason unknown"
 		fi
 		echo -e "\t\t<failure message=\"$_err_msg\" type=\"TestFail\" />" >> $report
-		if [ -s "$full_file" ]; then
+		if [ -z "$quiet" -a -s "$full_file" ]; then
 			echo -e "\t\t<system-out>" >> $report
 			printf	'<![CDATA[\n' >>$report
 			cat "$full_file" | tr -dc '[:print:][:space:]' | encode_xml >>$report
 			printf ']]>\n'	>>$report
 			echo -e "\t\t</system-out>" >> $report
 		fi
-		if [ -f "$dmesg_file" ]; then
+		if [ -n "$quiet" ]; then
+		   :
+		elif [ -f "$dmesg_file" ]; then
 			echo -e "\t\t<system-err>" >> $report
 			printf	'<![CDATA[\n' >>$report
 			cat "$dmesg_file" | tr -dc '[:print:][:space:]' | encode_xml >>$report
@@ -144,7 +151,7 @@ _make_section_report()
 	local sect_time="$5"
 	for report in $REPORT_LIST; do
 		case "$report" in
-		"xunit")
+		"xunit"|"xunit-quiet")
 			_xunit_make_section_report "$sect_name" "$tests_count" \
 						   "$bad_count" "$notrun_count" \
 						   "$sect_time"
@@ -164,9 +171,9 @@ _make_testcase_report()
 	local test_time="$4"
 	for report in $REPORT_LIST; do
 		case "$report" in
-		"xunit")
+		"xunit"|"xunit-quiet")
 			_xunit_make_testcase_report "$sect_name" "$test_seq" \
-						    "$test_status" "$test_time"
+						    "$test_status" "$test_time" "$report"
 			;;
 		*)
 			_dump_err "report format '$report' is not supported"
@@ -178,7 +185,7 @@ _make_testcase_report()
 _assert_report_list() {
 	for report in $REPORT_LIST; do
 		case "$report" in
-		"xunit")
+		"xunit"|"xunit-quiet")
 			;;
 		*)
 			_fatal "report format '$report' is not supported"
Signature · 1 line
2.31.0
#3 lore raw

PATCH 2/4 generic/556: add a check to make sure ext4 supports encrypted casefolding

In reply to Theodore Ts'o

Some older kernels support ext4 file systems with encryption enabled, and with casefold enabled, but not file systems have both encryption *and* casefolding enabled. On those kernels, generic/556 will fail. Fortunately, we can test if ext4 supports encrypted casefold via the presence of /sys/fs/ext4/features/encrypted_casefold.

---
 common/casefold   | 17 +++++++++++++++++
 tests/generic/556 |  1 +
 2 files changed, 18 insertions(+)
common/casefold +17 -0
index d9126f4c..e3e89508 100644
@@ -43,6 +43,23 @@ _require_scratch_casefold()
 	_require_command "$LSATTR_PROG" lsattr
 }
 
+_require_encrypted_casefold ()
+{
+    case $FSTYP in
+	ext4)
+	    if test ! -f /sys/fs/ext4/features/casefold ; then
+		_notrun "casefolding not supported"
+	    fi
+	    if test ! -f /sys/fs/ext4/features/encryption ; then
+		_notrun "file system encryption not supported"
+	    fi
+	    if test ! -f /sys/fs/ext4/features/encrypted_casefold ; then
+		_notrun "encrypted casefolding not supported"
+	    fi
+	    ;;
+	esac
+}
+
 _scratch_mkfs_casefold()
 {
 	case $FSTYP in
tests/generic/556 +1 -0
index 404a3243..80563502 100755
@@ -13,6 +13,7 @@ _begin_fstest auto quick casefold
 . ./common/attr
 
 _supported_fs generic
+_require_encrypted_casefold
 _require_scratch_nocheck
 _require_scratch_casefold
 _require_symlinks
Signature · 1 line
2.31.0
#4 lore raw

Re: [PATCH 2/4] generic/556: add a check to make sure ext4 supports encrypted casefolding

Hi Ted,

Quoted 7 lines · Theodore Ts'o

Some older kernels support ext4 file systems with encryption enabled, and with casefold enabled, but not file systems have both encryption *and* casefolding enabled. On those kernels, generic/556 will fail. Fortunately, we can test if ext4 supports encrypted casefold via the presence of /sys/fs/ext4/features/encrypted_casefold.

Please see my previous comments on this same patch:

* https://lore.kernel.org/fstests/[email protected]
* https://lore.kernel.org/fstests/[email protected]

- Eric

#5 lore raw

PATCH 3/4 dmflakey: don't run dmflakey tests with an external log device

In reply to Theodore Ts'o

dmflakey works by dropping all writes before unmounting to simulate a crash/power loss. This doesn't work if there is an external log device, since we only drop writes to the primary block device, and not the external log device. Fixing this for real would require somehow arranging to atomically loading a new dmflakey table for two block devices at the same time, so for now, just skip tests using dmflakey if the external log device is enabled.

---
 common/dmflakey | 4 ++++
 1 file changed, 4 insertions(+)
common/dmflakey +4 -0
index 52da3b10..c2be78e9 100644
@@ -12,6 +12,10 @@ _init_flakey()
 {
 	# Scratch device
 	local BLK_DEV_SIZE=`blockdev --getsz $SCRATCH_DEV`
+
+	if test "$USE_EXTERNAL" = yes -a ! -z "$SCRATCH_LOGDEV" ; then
+		_notrun "dmflakey tests don't work with an external log device"
+	fi
 	FLAKEY_DEV=/dev/mapper/flakey-test
 	FLAKEY_TABLE="0 $BLK_DEV_SIZE flakey $SCRATCH_DEV 0 180 0"
 	FLAKEY_TABLE_DROP="0 $BLK_DEV_SIZE flakey $SCRATCH_DEV 0 0 180 1 drop_writes"
Signature · 1 line
2.31.0
#6 lore raw

Re: [PATCH 3/4] dmflakey: don't run dmflakey tests with an external log device

Quoted 18 lines · Theodore Ts'o

dmflakey works by dropping all writes before unmounting to simulate a crash/power loss. This doesn't work if there is an external log device, since we only drop writes to the primary block device, and not the external log device. Fixing this for real would require somehow arranging to atomically loading a new dmflakey table for two block devices at the same time, so for now, just skip tests using dmflakey if the external log device is enabled.

---
 common/dmflakey | 4 ++++
 1 file changed, 4 insertions(+)
common/dmflakey +0 -0
index 52da3b10..c2be78e9 100644

common/dmflakey _init_flakey() · line 18

Quoted 7 lines
common/dmflakey +4 -0
@@ -12,6 +12,10 @@ _init_flakey()
 {
 	# Scratch device
 	local BLK_DEV_SIZE=`blockdev --getsz $SCRATCH_DEV`
+
+	if test "$USE_EXTERNAL" = yes -a ! -z "$SCRATCH_LOGDEV" ; then
+		_notrun "dmflakey tests don't work with an external log device"
+	fi

I'm kinda surprised this still doesn't work -- last year Eryu merged a patch[1] from me that's supposed to make common/dmflakey work with external log devices. What kind of failures do you see?

--D

[1] 9c1f3149 ("dmflakey: support external log and realtime devices")

Quoted 6 lines
common/dmflakey +0 -0
 	FLAKEY_DEV=/dev/mapper/flakey-test
 	FLAKEY_TABLE="0 $BLK_DEV_SIZE flakey $SCRATCH_DEV 0 180 0"
 	FLAKEY_TABLE_DROP="0 $BLK_DEV_SIZE flakey $SCRATCH_DEV 0 0 180 1 drop_writes"
Signature · 1 line
2.31.0
#7 lore raw

PATCH 4/4 generic/475: skip test when using an external log device

In reply to Theodore Ts'o

This test tests log recovery by kicking off fstress on the scratch fs, then switching out the underlying the device with dm-error to see what happens when the disk goes down. But when there is an external log device writes to the log device still succeed, while writes to the rest of the file system fail, and this leads to file system corruption. So skip this test when there is an external log.

---
 tests/generic/475 | 4 ++++
 1 file changed, 4 insertions(+)
tests/generic/475 +4 -0
index c426402e..43c26303 100755
@@ -34,6 +34,10 @@ _require_scratch
 _require_dm_target error
 _require_command "$KILLALL_PROG" "killall"
 
+if test "$USE_EXTERNAL" = yes -a ! -z "$SCRATCH_LOGDEV" ; then
+    _notrun "Cannot have external log device"
+fi
+
 echo "Silence is golden."
 
 _scratch_mkfs >> $seqres.full 2>&1
Signature · 1 line
2.31.0
#8 lore raw

Re: [PATCH 4/4] generic/475: skip test when using an external log device

Quoted 6 lines · Theodore Ts'o

This test tests log recovery by kicking off fstress on the scratch fs, then switching out the underlying the device with dm-error to see what happens when the disk goes down. But when there is an external log device writes to the log device still succeed, while writes to the rest of the file system fail, and this leads to file system corruption. So skip this test when there is an external log.

Hm, could you review the congruent operation length series[1], please?

Once that's done, I can move on to the next push, which is fixing up dmerror[2] and fail-make-request[3] to handle external devices:

[1] https://lore.kernel.org/fstests/165826663647.3249494.13640199673218669145.stgit@magnolia/T/#t
[2] https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfstests-dev.git/log/?h=dmerror-on-rt-devices
[3] https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfstests-dev.git/log/?h=fix-fail-make-reqest

--D

Quoted 23 lines
---
 tests/generic/475 | 4 ++++
 1 file changed, 4 insertions(+)
tests/generic/475 +4 -0
index c426402e..43c26303 100755
@@ -34,6 +34,10 @@ _require_scratch
 _require_dm_target error
 _require_command "$KILLALL_PROG" "killall"
 
+if test "$USE_EXTERNAL" = yes -a ! -z "$SCRATCH_LOGDEV" ; then
+    _notrun "Cannot have external log device"
+fi
+
 echo "Silence is golden."
 
 _scratch_mkfs >> $seqres.full 2>&1
Signature · 1 line
2.31.0