Skip to contentSkip to conversation

block: add regression test for null-blk concurrently power/submit_queues test

PATCH In review Yi Zhang 3 participants 5 messages Jun 5, 2024 – Jun 18, 2024 Block layer

lore
Share & tools
lore

Yi Zhang

author
#0 lore raw

PATCH v2 1/1 block: add regression test for null-blk concurrently power/submit_queues test

null-blk currently power/submit_queues operations which will lead kernel null-ptr-dereference[1], add one regression test for it and the fix has
been merged to v6.10-rc1 by [2].
[1]
https://lore.kernel.org/linux-block/CAHj4cs9LgsHLnjg8z06LQ3Pr5cax-+Ps+xT7AP7TPnEjStuwZA@mail.gmail.com/
https://lore.kernel.org/linux-block/[email protected]/
[2]
commit a2db328b0839 ("null_blk: fix null-ptr-dereference while configuring 'power' and 'submit_queues'")

---
 tests/block/038     | 61 +++++++++++++++++++++++++++++++++++++++++++++
 tests/block/038.out |  2 ++
 2 files changed, 63 insertions(+)
 create mode 100755 tests/block/038
 create mode 100644 tests/block/038.out
tests/block/038 new file +61 -0
new file mode 100755
index 0000000..fe3c7cd
@@ -0,0 +1,61 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-3.0+
+# Copyright (C) 2024 Yi Zhang <[email protected]>
+#
+# Regression test for commit a2db328b0839 ("null_blk: fix null-ptr-dereference
+# while configuring 'power' and 'submit_queues'").
+
+. tests/block/rc
+. common/null_blk
+
+DESCRIPTION="Test null-blk concurrent power/submit_queues operations"
+QUICK=1
+
+requires() {
+	_have_module null_blk
+	_have_module_param null_blk nr_devices
+	_have_module_param null_blk submit_queues
+}
+
+null_blk_power_loop() {
+	local nullb="$1"
+	for ((i = 1; i <= 200; i++)); do
+		echo 1 > "/sys/kernel/config/nullb/${nullb}/power"
+		echo 0 > "/sys/kernel/config/nullb/${nullb}/power"
+	done
+}
+
+null_blk_submit_queues_loop() {
+	local nullb="$1"
+	for ((i = 1; i <= 200; i++)); do
+		echo 1 > "/sys/kernel/config/nullb/${nullb}/submit_queues"
+		echo 4 > "/sys/kernel/config/nullb/${nullb}/submit_queues"
+	done
+}
+
+test() {
+	echo "Running ${TEST_NAME}"
+
+	local nullb_params=(
+		nr_devices=0
+	)
+	if ! _init_null_blk "${nullb_params[@]}"; then
+		echo "Loading null_blk failed"
+		return 1
+	fi
+
+	if ! _configure_null_blk nullb0; then
+		echo "Configuring null_blk nullb0 failed"
+		return 1
+	fi
+
+	# fire off two null-blk power/submit_queues concurrently and wait
+	# for them to complete...
+	null_blk_power_loop nullb0 &
+	null_blk_submit_queues_loop nullb0 &
+	wait
+
+	_exit_null_blk
+
+	echo "Test complete"
+}
tests/block/038.out new file +2 -0
new file mode 100644
index 0000000..aebde64
@@ -0,0 +1,2 @@
+Running block/038
+Test complete
Signature · 1 line
2.45.1
Reviewed-by on v2 1/1
#1 lore raw
Quoted 10 lines · Yi Zhang

null-blk currently power/submit_queues operations which will lead kernel null-ptr-dereference[1], add one regression test for it and the fix has
been merged to v6.10-rc1 by [2].
[1]
https://lore.kernel.org/linux-block/CAHj4cs9LgsHLnjg8z06LQ3Pr5cax-+Ps+xT7AP7TPnEjStuwZA@mail.gmail.com/
https://lore.kernel.org/linux-block/[email protected]/
[2]
commit a2db328b0839 ("null_blk: fix null-ptr-dereference while configuring 'power' and 'submit_queues'")

Thanks for the test, looks good, I wonder what other drivers might need such test ...

-ck

#2 lore raw
In reply to Yi Zhang
Quoted 8 lines · Yi Zhang

null-blk currently power/submit_queues operations which will lead kernel null-ptr-dereference[1], add one regression test for it and the fix has
been merged to v6.10-rc1 by [2].
[1]
https://lore.kernel.org/linux-block/CAHj4cs9LgsHLnjg8z06LQ3Pr5cax-+Ps+xT7AP7TPnEjStuwZA@mail.gmail.com/
https://lore.kernel.org/linux-block/[email protected]/
[2]
commit a2db328b0839 ("null_blk: fix null-ptr-dereference while configuring 'power' and 'submit_queues'")

Thank you Yi. I ran the test case and it looks working good. It passes with the kernel v6.10-rc2. It causes the hang with the kernel v6.9. To not confuse blktests users with the hang, I will wait for the commit a2db328b0839 to land on the stable kernels before I apply this patch.

One more thing I noticed is that your current patch requires loadable null_blk. To allow it run with built-in null_blk, I would like to suggest additional change below on top of your patch. It,

- calls _have_null_blk instead of _have_module null_blk,
- checks submit_queues parameter with _have_null_blk_feature instead of _have_module_param,
- does not call _init_null_blk, and
- uses nullb1 instead for nullb0.

Please let me know your thought about these changes. If you are ok with them, I can fold them in the commit.

tests/block/038 +8 -15
index fe3c7cd..56272be 100755
@@ -12,9 +12,10 @@ DESCRIPTION="Test null-blk concurrent power/submit_queues operations"
 QUICK=1
 
 requires() {
-	_have_module null_blk
-	_have_module_param null_blk nr_devices
-	_have_module_param null_blk submit_queues
+	_have_null_blk
+	if ! _have_null_blk_feature submit_queues; then
+		SKIP_REASONS+=("null_blk does not support submit_queues")
+	fi
 }
 
 null_blk_power_loop() {
@@ -36,23 +37,15 @@ null_blk_submit_queues_loop() {
 test() {
 	echo "Running ${TEST_NAME}"
 
-	local nullb_params=(
-		nr_devices=0
-	)
-	if ! _init_null_blk "${nullb_params[@]}"; then
-		echo "Loading null_blk failed"
-		return 1
-	fi
-
-	if ! _configure_null_blk nullb0; then
-		echo "Configuring null_blk nullb0 failed"
+	if ! _configure_null_blk nullb1; then
+		echo "Configuring null_blk nullb1 failed"
 		return 1
 	fi
 
 	# fire off two null-blk power/submit_queues concurrently and wait
 	# for them to complete...
-	null_blk_power_loop nullb0 &
-	null_blk_submit_queues_loop nullb0 &
+	null_blk_power_loop nullb1 &
+	null_blk_submit_queues_loop nullb1 &
 	wait
 
 	_exit_null_blk
author
#3 lore raw
Quoted 20 lines · Shinichiro Kawasaki
Quoted 8 lines · Yi Zhang

null-blk currently power/submit_queues operations which will lead kernel null-ptr-dereference[1], add one regression test for it and the fix has
been merged to v6.10-rc1 by [2].
[1]
https://lore.kernel.org/linux-block/CAHj4cs9LgsHLnjg8z06LQ3Pr5cax-+Ps+xT7AP7TPnEjStuwZA@mail.gmail.com/
https://lore.kernel.org/linux-block/[email protected]/
[2]
commit a2db328b0839 ("null_blk: fix null-ptr-dereference while configuring 'power' and 'submit_queues'")

Thank you Yi. I ran the test case and it looks working good. It passes with the kernel v6.10-rc2. It causes the hang with the kernel v6.9. To not confuse blktests users with the hang, I will wait for the commit a2db328b0839 to land on the stable kernels before I apply this patch.

One more thing I noticed is that your current patch requires loadable null_blk. To allow it run with built-in null_blk, I would like to suggest additional change below on top of your patch. It,

Quoted 8 lines

- calls _have_null_blk instead of _have_module null_blk,
- checks submit_queues parameter with _have_null_blk_feature instead of _have_module_param,
- does not call _init_null_blk, and
- uses nullb1 instead for nullb0.

Please let me know your thought about these changes. If you are ok with them, I can fold them in the commit.

Sure, I'm OK with the change, thanks. :)

Quoted 48 lines
tests/block/038 +8 -15
index fe3c7cd..56272be 100755
@@ -12,9 +12,10 @@ DESCRIPTION="Test null-blk concurrent power/submit_queues operations"
 QUICK=1
 
 requires() {
-       _have_module null_blk
-       _have_module_param null_blk nr_devices
-       _have_module_param null_blk submit_queues
+       _have_null_blk
+       if ! _have_null_blk_feature submit_queues; then
+               SKIP_REASONS+=("null_blk does not support submit_queues")
+       fi
 }
 
 null_blk_power_loop() {
@@ -36,23 +37,15 @@ null_blk_submit_queues_loop() {
 test() {
        echo "Running ${TEST_NAME}"
 
-       local nullb_params=(
-               nr_devices=0
-       )
-       if ! _init_null_blk "${nullb_params[@]}"; then
-               echo "Loading null_blk failed"
-               return 1
-       fi
-
-       if ! _configure_null_blk nullb0; then
-               echo "Configuring null_blk nullb0 failed"
+       if ! _configure_null_blk nullb1; then
+               echo "Configuring null_blk nullb1 failed"
                return 1
        fi
 
        # fire off two null-blk power/submit_queues concurrently and wait
        # for them to complete...
-       null_blk_power_loop nullb0 &
-       null_blk_submit_queues_loop nullb0 &
+       null_blk_power_loop nullb1 &
+       null_blk_submit_queues_loop nullb1 &
        wait
 
        _exit_null_blk
Signature · 2 lines
Best Regards,
  Yi Zhang
#4 lore raw
Quoted 28 lines · Yi Zhang
Quoted 26 lines · Shinichiro Kawasaki
Quoted 8 lines · Yi Zhang

null-blk currently power/submit_queues operations which will lead kernel null-ptr-dereference[1], add one regression test for it and the fix has
been merged to v6.10-rc1 by [2].
[1]
https://lore.kernel.org/linux-block/CAHj4cs9LgsHLnjg8z06LQ3Pr5cax-+Ps+xT7AP7TPnEjStuwZA@mail.gmail.com/
https://lore.kernel.org/linux-block/[email protected]/
[2]
commit a2db328b0839 ("null_blk: fix null-ptr-dereference while configuring 'power' and 'submit_queues'")

Thank you Yi. I ran the test case and it looks working good. It passes with the kernel v6.10-rc2. It causes the hang with the kernel v6.9. To not confuse blktests users with the hang, I will wait for the commit a2db328b0839 to land on the stable kernels before I apply this patch.

One more thing I noticed is that your current patch requires loadable null_blk. To allow it run with built-in null_blk, I would like to suggest additional change below on top of your patch. It,

- calls _have_null_blk instead of _have_module null_blk,
- checks submit_queues parameter with _have_null_blk_feature instead of _have_module_param,
- does not call _init_null_blk, and
- uses nullb1 instead for nullb0.

Quoted 4 lines
Quoted 2 lines

Please let me know your thought about these changes. If you are ok with them, I can fold them in the commit.

Sure, I'm OK with the change, thanks. :)

Now the kernel side fix is in v6.9.4 kernel. I've applied this blktests patch. Thanks! FYI, I added my modification as a following commit [*].

[*] https://github.com/osandov/blktests/commit/eba35aaf32ceb9516a7dbc856c9db4d0ae3268d0