[PATCH blktests 0/7] Skip tests if scsi_debug module is already loaded and in use

[PATCH blktests 0/7] Skip tests if scsi_debug module is already loaded and in use

[Shin'ichiro Kawasaki]

When blktests users have scsi_debug devices configured for purposes other than
blktests, running blktests currently unloads scsi_debug unconditionally and
breaks the existing setup. To avoid disrupting such configurations, this patch
series improves blktests to check whether the scsi_debug module is already
loaded and in use, and to skip test cases that require scsi_debug in that
situation.

This series was originally proposed as a GitHub Pull Request [1]. Disha Goel and
I worked on the patches. The first four patches fix pre-existing problems found
while working on this series. The fifth patch implements the improvement
described above. The last two patches fix two additional problems that were
found during this work.

[1] https://github.com/linux-blktests/blktests/pull/218

Disha Goel (1):
  Skip tests if scsi_debug module is already loaded and in use

Shin'ichiro Kawasaki (6):
  check: move fallback device handling into
    _check_and_call_test_device()
  check: call _unload_module() after fallback_device_cleanup()
  check: do not create fallback device when other requirements are not
    met
  check: call fallback_device() in the same bash context
  common/scsi_debug: ensure scsi_debug is loadable before loading it
  zbd/012: add missing scsi_debug check

 check             | 84 ++++++++++++++++++++++++-----------------------
 common/scsi_debug | 31 ++++++++++++++---
 tests/block/009   |  2 +-
 tests/block/025   |  2 +-
 tests/block/028   |  2 +-
 tests/block/032   |  2 +-
 tests/loop/004    |  2 +-
 tests/md/002      |  2 +-
 tests/scsi/007    |  2 +-
 tests/scsi/009    |  2 +-
 tests/srp/rc      |  2 +-
 tests/zbd/008     |  2 +-
 tests/zbd/009     |  2 +-
 tests/zbd/010     |  2 +-
 tests/zbd/012     |  1 +
 15 files changed, 83 insertions(+), 57 deletions(-)

-- 
2.49.0

[PATCH blktests 1/7] check: move fallback device handling into _check_and_call_test_device()

[Shin'ichiro Kawasaki]

Currently, when a test case has callback functions for fallback device,
the fallback device is created in run_test(). Meanwhile,
_unload_module() is called in _call_test(). Then,
cleanup_fallback_device() is called after _unload_module().

  _run_test
    fallback_device
      _check_and_call_test_device
        _call_test
          _unload_modules
    cleanup_fallback_device

This is not desired since module unload can affect
cleanup_fallback_device(). For example, scsi_debug is used as a fallback
device, _unload_modules() can unload scsi_debug before calling
cleanup_fallback_device().

To resolve this, make two changes. The first change is to move fallback
operations from _run_test() to _check_and_call_test_device(). This
commit does it. After this commit, the call chain will be as follows.

  _run_test
    _check_and_call_test_device
      fallback_device
        _call_test
          _unload_modules
      cleanup_fallback_device

As the second change, the following commit will move the
_unload_modules() call.

While at this change, rename the local variable FALLBACK_DEVICE to
'fallback', since capital letters imply that the variable would be
a global variable.

Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
---
 check | 57 ++++++++++++++++++++++++++++-----------------------------
 1 file changed, 28 insertions(+), 29 deletions(-)

diff --git a/check b/check
index f170458..3fa3e97 100755
--- a/check
+++ b/check
@@ -716,14 +716,34 @@ _check_and_call_test() {
 _check_and_call_test_device() {
 	local postfix
 	local unset_skip_reason
-	local ret
+	local ret=0
+	local fallback=0
 
 	if declare -fF requires >/dev/null; then
 		requires
 	fi
 
+	if [[ ${#TEST_DEVS[@]} -eq 0 ]] && \
+		   declare -fF fallback_device >/dev/null; then
+		TEST_DEVS[0]=fallback
+	fi
+
 	[[ -n $COND_DESC ]] && postfix=_${COND_DESC//[ =]/_}
 	for TEST_DEV in "${TEST_DEVS[@]}"; do
+		if [[ $TEST_DEV == fallback ]]; then
+			if ! TEST_DEV=$(fallback_device); then
+				_warning "$TEST_NAME: fallback_device call failure"
+				return 0
+			fi
+
+			if ! _find_sysfs_dirs "$TEST_DEV"; then
+				_warning "$TEST_NAME: could not find sysfs directory for ${TEST_DEV}"
+				cleanup_fallback_device
+				return 0
+			fi
+			fallback=1
+		fi
+
 		TEST_DEV_SYSFS="${TEST_DEV_SYSFS_DIRS["$TEST_DEV"]}"
 		TEST_DEV_PART_SYSFS="${TEST_DEV_PART_SYSFS_DIRS["$TEST_DEV"]}"
 
@@ -744,6 +764,13 @@ _check_and_call_test_device() {
 		if (( unset_skip_reason )); then
 			unset SKIP_REASONS
 		fi
+
+		if (( fallback )); then
+			cleanup_fallback_device
+			unset "TEST_DEV_SYSFS_DIRS[${TEST_DEV}]]"
+			unset "TEST_DEV_PART_SYSFS_DIRS[${TEST_DEV}]"
+			TEST_DEVS=()
+		fi
 	done
 
 	return $ret
@@ -825,7 +852,6 @@ _run_test() {
 	DMESG_FILTER="cat"
 	RUN_FOR_ZONED=0
 	COND_DESC=""
-	FALLBACK_DEVICE=0
 	MODULES_TO_UNLOAD=()
 
 	local nr_conds cond_i
@@ -855,26 +881,6 @@ _run_test() {
 			ret=$?
 		fi
 	elif declare -fF test_device >/dev/null; then
-		if [[ ${#TEST_DEVS[@]} -eq 0 ]] && \
-			declare -fF fallback_device >/dev/null; then
-			if ! test_dev=$(fallback_device); then
-				_warning "$TEST_NAME: fallback_device call failure"
-				return 0
-			fi
-
-			if ! _find_sysfs_dirs "$test_dev"; then
-				_warning "$TEST_NAME: could not find sysfs directory for ${test_dev}"
-				cleanup_fallback_device
-				return 0
-			fi
-			TEST_DEVS=( "${test_dev}" )
-			FALLBACK_DEVICE=1
-		fi
-
-		if [[ ${#TEST_DEVS[@]} -eq 0 ]]; then
-			return 0
-		fi
-
 		if declare -fF set_conditions >/dev/null; then
 			nr_conds=$(set_conditions)
 			for ((cond_i = 0; cond_i < nr_conds; cond_i++)); do
@@ -887,13 +893,6 @@ _run_test() {
 			_check_and_call_test_device
 			ret=$?
 		fi
-
-		if (( FALLBACK_DEVICE )); then
-			cleanup_fallback_device
-			unset "TEST_DEV_SYSFS_DIRS[${TEST_DEVS[0]}]"
-			unset "TEST_DEV_PART_SYSFS_DIRS[${TEST_DEVS[0]}]"
-			TEST_DEVS=()
-		fi
 	else
 		_check_and_call_test_device_array
 		ret=$?
-- 
2.49.0

[PATCH blktests 2/7] check: call _unload_module() after fallback_device_cleanup()

[Shin'ichiro Kawasaki]

As described in the commit message of the previous commit,
_unload_module() is called before calling cleanup_fallback_device().
Then, modules that is not cleaned up can be unloaded by _unload_module()
and can cause unload failures.

To avoid the failures, move _unload_module() call into
_check_and_call_test_device() so that it is called after the
cleanup_fallback_device() call. After this change, the call chain
becomes as follows:

  _run_test
    _check_and_call_test_device
      fallback_device
      _call_test
      cleanup_fallback_device
      _unload_modules

This change requires to add _unload_modules() call to
_check_and_call_test() and _check_and_call_test_device_array() also.

Fixes: 756d18d5fef0 ("check: call _unload_modules for each test run")
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
---
 check | 17 +++++------------
 1 file changed, 5 insertions(+), 12 deletions(-)

diff --git a/check b/check
index 3fa3e97..91064d7 100755
--- a/check
+++ b/check
@@ -417,7 +417,7 @@ _cleanup() {
 	_exit_cgroup2
 }
 
-__call_test() {
+_call_test() {
 	local test_func="$1"
 	local seqres="${RESULTS_DIR}/${TEST_NAME}"
 	# shellcheck disable=SC2034
@@ -537,17 +537,6 @@ __call_test() {
 	fi
 }
 
-_call_test() {
-	local ret
-
-	__call_test "$@"
-	ret=$?
-
-	_unload_modules
-
-	return $ret
-}
-
 _test_dev_is_zoned() {
 	[[ -e "${TEST_DEV_SYSFS}/queue/zoned" &&
 	   $(cat "${TEST_DEV_SYSFS}/queue/zoned") != none ]]
@@ -711,6 +700,7 @@ _check_and_call_test() {
 	[[ -n $COND_DESC ]] && postfix=_${COND_DESC//[ =]/_}
 	RESULTS_DIR="$OUTPUT/nodev${postfix}"
 	_call_test test
+	_unload_modules
 }
 
 _check_and_call_test_device() {
@@ -771,6 +761,8 @@ _check_and_call_test_device() {
 			unset "TEST_DEV_PART_SYSFS_DIRS[${TEST_DEV}]"
 			TEST_DEVS=()
 		fi
+
+		_unload_modules
 	done
 
 	return $ret
@@ -842,6 +834,7 @@ _check_and_call_test_device_array() {
 		unset "TEST_DEV_ARRAY_SYSFS_DIRS[${devs[i]}]"
 	done
 
+	_unload_modules
 	return $ret
 }
 
-- 
2.49.0

[PATCH blktests 3/7] check: do not create fallback device when other requirements are not met

[Shin'ichiro Kawasaki]

Currently, fallback devices are prepared regardless of whether other
requirements are fulfilled. This is not always desired. For example,
when users already created scsi_debug devices, it is not good to create
a fallback scsi_debug device, because it unloads scsi_debug module and
remove the scsi_debug devices users created. To prevent such scenario,
confirm requirements by checking skip reasons before creating fallback
devices.

Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
---
 check | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/check b/check
index 91064d7..fdc7ba7 100755
--- a/check
+++ b/check
@@ -721,6 +721,11 @@ _check_and_call_test_device() {
 	[[ -n $COND_DESC ]] && postfix=_${COND_DESC//[ =]/_}
 	for TEST_DEV in "${TEST_DEVS[@]}"; do
 		if [[ $TEST_DEV == fallback ]]; then
+			if [[ -v SKIP_REASONS ]]; then
+				_output_notrun "$TEST_NAME"
+				return 0
+			fi
+
 			if ! TEST_DEV=$(fallback_device); then
 				_warning "$TEST_NAME: fallback_device call failure"
 				return 0
-- 
2.49.0

[PATCH blktests 4/7] check: call fallback_device() in the same bash context

[Shin'ichiro Kawasaki]

Currently, the callback function fallback_device() is called with bash
$() feature. This runs the function call in a new bash context. Then,
global variable changes in the function call are not reflected in the
caller context. This is not handy and does not allow checking
MODULES_TO_UNLOAD status change in the caller, which will be added in
the following commit.

To share the same bash context between the caller and the callee of
fallback_device(), do not use $(). Instead, use a temporary file to keep
the output of fallback_device().

Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
---
 check | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/check b/check
index fdc7ba7..00213cb 100755
--- a/check
+++ b/check
@@ -708,6 +708,7 @@ _check_and_call_test_device() {
 	local unset_skip_reason
 	local ret=0
 	local fallback=0
+	local fb_dev_file
 
 	if declare -fF requires >/dev/null; then
 		requires
@@ -726,10 +727,14 @@ _check_and_call_test_device() {
 				return 0
 			fi
 
-			if ! TEST_DEV=$(fallback_device); then
+			# Use a tmp file to keep bash context in fallback_device
+			fb_dev_file="${OUTPUT}/${TEST_NAME/\//_}_fallback_dev"
+			if ! fallback_device > "$fb_dev_file"; then
 				_warning "$TEST_NAME: fallback_device call failure"
 				return 0
 			fi
+			TEST_DEV=$(<"$fb_dev_file")
+			rm -f "$fb_dev_file"
 
 			if ! _find_sysfs_dirs "$TEST_DEV"; then
 				_warning "$TEST_NAME: could not find sysfs directory for ${TEST_DEV}"
-- 
2.49.0

[PATCH blktests 5/7] Skip tests if scsi_debug module is already loaded and in use

[Shin'ichiro Kawasaki]

From: Disha Goel <disgoel@linux.ibm.com>

Several tests across block/, scsi/, dm/, md/, zbd/, nvme/ require
exclusive access to the scsi_debug module because they load, unload
or reconfigure it. When scsi_debug is loadable and already loaded
by the environment (e.g., by another driver or a previous setup),
these tests fail with:

    modprobe: FATAL: Module scsi_debug is in use.
    Unloading scsi_debug failed

    scsi_debug            327680  4

To prevent these failures, this patch introduces a new helper function
_have_loadable_scsi_debug(). It verifies if the module is already loaded
by checking the ${MODULES_TO_UNLOAD[*]} array. If the module exists but
is not in the array, it indicates the module was loaded before the test
started, and the test is skipped.

Additionally, for cases where scsi_debug is built-in, the environment may
have already created additional hosts. To prevent the tests from disrupting
these hosts, _have_scsi_debug() now checks the add_host attribute. If the
number of hosts is greater than 1, the test is skipped.

Link: https://github.com/linux-blktests/blktests/pull/218
Signed-off-by: Disha Goel <disgoel@linux.ibm.com>
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
---
 common/scsi_debug | 29 +++++++++++++++++++++++++----
 tests/block/009   |  2 +-
 tests/block/025   |  2 +-
 tests/block/028   |  2 +-
 tests/block/032   |  2 +-
 tests/loop/004    |  2 +-
 tests/md/002      |  2 +-
 tests/scsi/007    |  2 +-
 tests/scsi/009    |  2 +-
 tests/srp/rc      |  2 +-
 tests/zbd/008     |  2 +-
 tests/zbd/009     |  2 +-
 tests/zbd/010     |  2 +-
 13 files changed, 37 insertions(+), 16 deletions(-)

diff --git a/common/scsi_debug b/common/scsi_debug
index 8964558..97a65c2 100644
--- a/common/scsi_debug
+++ b/common/scsi_debug
@@ -4,13 +4,34 @@
 #
 # scsi_debug helper functions.
 
-_have_scsi_debug() {
-	_have_driver scsi_debug
-}
-
 SD_PARAM_PATH=/sys/module/scsi_debug/parameters
 SD_PSEUDO_PATH=/sys/bus/pseudo/drivers/scsi_debug
 
+_have_scsi_debug() {
+	_have_driver scsi_debug
+
+	if _module_file_exists scsi_debug; then
+		_have_loadable_scsi_debug
+		return $?
+	fi
+
+	if [[ -e "$SD_PSEUDO_PATH/add_host" ]] && \
+		[[ $(cat "$SD_PSEUDO_PATH/add_host") -gt 1 ]]; then
+		SKIP_REASONS+=("scsi_debug already has multiple hosts configured")
+		return 1
+	fi
+}
+
+_have_loadable_scsi_debug() {
+	_have_module scsi_debug || return 1
+
+	if [[ -d /sys/module/scsi_debug ]] && \
+		[[ ! ${MODULES_TO_UNLOAD[*]} =~ scsi_debug ]]; then
+		SKIP_REASONS+=("scsi_debug is already loaded before test")
+		return 1
+	fi
+}
+
 _scsi_debug_key_path() {
 	local key=${1}
 
diff --git a/tests/block/009 b/tests/block/009
index 7256afc..8995646 100755
--- a/tests/block/009
+++ b/tests/block/009
@@ -12,7 +12,7 @@
 DESCRIPTION="check page-cache coherency after BLKDISCARD"
 
 requires() {
-	_have_module scsi_debug
+	_have_loadable_scsi_debug
 	_have_program xfs_io
 }
 
diff --git a/tests/block/025 b/tests/block/025
index 4c48e9f..6582fa3 100755
--- a/tests/block/025
+++ b/tests/block/025
@@ -12,7 +12,7 @@
 DESCRIPTION="do a huge discard with 4k sector size"
 
 requires() {
-	_have_module scsi_debug
+	_have_loadable_scsi_debug
 }
 
 test() {
diff --git a/tests/block/028 b/tests/block/028
index 13b3278..c03ad6c 100755
--- a/tests/block/028
+++ b/tests/block/028
@@ -12,7 +12,7 @@ DESCRIPTION="do I/O on scsi_debug with DIF/DIX enabled"
 DMESG_FILTER="sed -r 's/(guard tag error at sector|ref tag error at location)/blktests failure: \\1/'"
 
 requires() {
-	_have_module scsi_debug
+	_have_loadable_scsi_debug
 }
 
 test_pi() {
diff --git a/tests/block/032 b/tests/block/032
index 5e25d0b..7f89a15 100755
--- a/tests/block/032
+++ b/tests/block/032
@@ -14,7 +14,7 @@ QUICK=1
 
 requires() {
 	_have_xfs
-	_have_module scsi_debug
+	_have_loadable_scsi_debug
 }
 
 test() {
diff --git a/tests/loop/004 b/tests/loop/004
index ca52d80..fc7e4b0 100755
--- a/tests/loop/004
+++ b/tests/loop/004
@@ -12,7 +12,7 @@ QUICK=1
 
 requires() {
 	_have_program xfs_io
-	_have_module scsi_debug
+	_have_loadable_scsi_debug
 	_have_src_program loblksize
 	_have_loop_set_block_size
 }
diff --git a/tests/md/002 b/tests/md/002
index 1349a18..e1da8b6 100755
--- a/tests/md/002
+++ b/tests/md/002
@@ -12,7 +12,7 @@ DESCRIPTION="test md atomic writes"
 QUICK=1
 
 requires() {
-	_have_driver scsi_debug
+	_have_scsi_debug
 	_stacked_atomic_test_requires
 }
 
diff --git a/tests/scsi/007 b/tests/scsi/007
index a638e20..8a8d4ea 100755
--- a/tests/scsi/007
+++ b/tests/scsi/007
@@ -12,7 +12,7 @@ DESCRIPTION="Trigger the SCSI error handler"
 QUICK=1
 
 requires() {
-	_have_module scsi_debug
+	_have_loadable_scsi_debug
 }
 
 start_tracing() {
diff --git a/tests/scsi/009 b/tests/scsi/009
index 8860e1c..41a5152 100755
--- a/tests/scsi/009
+++ b/tests/scsi/009
@@ -12,7 +12,7 @@ DESCRIPTION="test scsi atomic writes"
 QUICK=1
 
 requires() {
-	_have_driver scsi_debug
+	_have_scsi_debug
 	_have_xfs_io_atomic_write
 }
 
diff --git a/tests/srp/rc b/tests/srp/rc
index 8585272..03fe55a 100755
--- a/tests/srp/rc
+++ b/tests/srp/rc
@@ -58,7 +58,7 @@ group_requires() {
 		_have_kver 5 5
 		_have_iproute2 190404
 	fi
-	_have_module scsi_debug
+	_have_loadable_scsi_debug
 	_have_module target_core_iblock
 	_have_module target_core_mod
 	_module_not_in_use scsi_transport_srp
diff --git a/tests/zbd/008 b/tests/zbd/008
index 55b5b6c..cdfbedf 100755
--- a/tests/zbd/008
+++ b/tests/zbd/008
@@ -13,7 +13,7 @@ DESCRIPTION="check no stale page cache after BLKZONERESET and data read race"
 TIMED=1
 
 requires() {
-	_have_module scsi_debug
+	_have_loadable_scsi_debug
 	_have_module_param scsi_debug zbc
 	_have_program xfs_io
 }
diff --git a/tests/zbd/009 b/tests/zbd/009
index b268483..a36ad71 100755
--- a/tests/zbd/009
+++ b/tests/zbd/009
@@ -36,7 +36,7 @@ requires() {
 	_have_driver btrfs
 	_have_module_param scsi_debug zone_cap_mb
 	_have_program mkfs.btrfs && have_good_mkfs_btrfs
-	_have_module scsi_debug
+	_have_loadable_scsi_debug
 }
 
 test() {
diff --git a/tests/zbd/010 b/tests/zbd/010
index c5cb76a..cb7202c 100755
--- a/tests/zbd/010
+++ b/tests/zbd/010
@@ -15,7 +15,7 @@ requires() {
 	_have_module null_blk
 	_have_module_param scsi_debug zone_cap_mb
 	_have_program mkfs.f2fs
-	_have_module scsi_debug
+	_have_loadable_scsi_debug
 }
 
 test() {
-- 
2.49.0

[PATCH blktests 6/7] common/scsi_debug: ensure scsi_debug is loadable before loading it

[Shin'ichiro Kawasaki]

Currently, _have_scsi_debug_group_number_stats() assumes that scsi_debug
is loadable. This does not work when scsi_debug is built-in. Before
loading scsi_debug in the function, ensure that scsi_debug is loadable.

Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
---
 common/scsi_debug | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/common/scsi_debug b/common/scsi_debug
index 97a65c2..6aa7420 100644
--- a/common/scsi_debug
+++ b/common/scsi_debug
@@ -49,6 +49,8 @@ _scsi_debug_key_path() {
 _have_scsi_debug_group_number_stats() {
 	local ret=0
 
+	_have_loadable_scsi_debug || return 1
+
 	modprobe -qr scsi_debug >&/dev/null
 	modprobe -q scsi_debug delay=0 >&/dev/null
 	if ! [[ -e ${SD_PSEUDO_PATH}/group_number_stats ]]; then
-- 
2.49.0

[PATCH blktests 7/7] zbd/012: add missing scsi_debug check

[Shin'ichiro Kawasaki]

The test case uses loadable scsi_debug, but it misses the check for it.
Add the missing check.

Fixes: 3ca2657f94b8 ("zbd/012: Test requeuing of zoned writes and queue freezing")
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
---
 tests/zbd/012 | 1 +
 1 file changed, 1 insertion(+)

diff --git a/tests/zbd/012 b/tests/zbd/012
index b2bda5c..c2e47f3 100755
--- a/tests/zbd/012
+++ b/tests/zbd/012
@@ -13,6 +13,7 @@ TIMED=1
 
 requires() {
 	_have_fio_zbd_zonemode
+	_have_loadable_scsi_debug
 }
 
 toggle_iosched() {
-- 
2.49.0

Re: [PATCH blktests 0/7] Skip tests if scsi_debug module is already loaded and in use

[Shinichiro Kawasaki]

On Mar 24, 2026 / 14:49, Shin'ichiro Kawasaki wrote:
> When blktests users have scsi_debug devices configured for purposes other than
> blktests, running blktests currently unloads scsi_debug unconditionally and
> breaks the existing setup. To avoid disrupting such configurations, this patch
> series improves blktests to check whether the scsi_debug module is already
> loaded and in use, and to skip test cases that require scsi_debug in that
> situation.
> 
> This series was originally proposed as a GitHub Pull Request [1]. Disha Goel and
> I worked on the patches. The first four patches fix pre-existing problems found
> while working on this series. The fifth patch implements the improvement
> described above. The last two patches fix two additional problems that were
> found during this work.

FYI, I applied this series.