bpf/selftests: Fix namespace mount setup in tc_redirect
Subsystems (5)
PATCH 1/1 bpf/selftests: Fix namespace mount setup in tc_redirect
The tc_redirect umounts /sys in the new namespace, which can be mounted as shared and cause global umount. The lazy umount also takes down mounted trees under /sys like debugfs, which won't be available after sysfs mounts again and could cause fails in other tests.
# cat /proc/self/mountinfo | grep debugfs 34 23 0:7 / /sys/kernel/debug rw,nosuid,nodev,noexec,relatime shared:14 - debugfs debugfs rw # cat /proc/self/mountinfo | grep sysfs 23 86 0:22 / /sys rw,nosuid,nodev,noexec,relatime shared:2 - sysfs sysfs rw # mount | grep debugfs debugfs on /sys/kernel/debug type debugfs (rw,nosuid,nodev,noexec,relatime)
# ./test_progs -t tc_redirect #164 tc_redirect:OK Summary: 1/4 PASSED, 0 SKIPPED, 0 FAILED
# mount | grep debugfs # cat /proc/self/mountinfo | grep debugfs # cat /proc/self/mountinfo | grep sysfs 25 86 0:22 / /sys rw,relatime shared:2 - sysfs sysfs rw
Making the sysfs private under the new namespace so the umount won't trigger the global sysfs umount.
--- tools/testing/selftests/bpf/prog_tests/tc_redirect.c | 7 +++++++ 1 file changed, 7 insertions(+)
tools/testing/selftests/bpf/prog_tests/tc_redirect.c +7 -0
@@ -105,6 +105,13 @@ static int setns_by_fd(int nsfd)if (!ASSERT_OK (err,"unshare" ))return err; +/* Make our /sys mount private, so the following umount won't +* trigger the global umount in case it's shared. +*/ + err =mount ("none" ,"/sys" ,NULL ,MS_PRIVATE ,NULL ); +if (!ASSERT_OK (err,"remount private /sys" )) +return err; + err =umount2 ("/sys" ,MNT_DETACH );if (!ASSERT_OK (err,"umount2 /sys" ))return err;
Signature · 1 line
2.33.1
This patch was applied to bpf/bpf-next.git (master) by Daniel Borkmann <[email protected]>:
Hello:
This patch was applied to bpf/bpf-next.git (master) by Daniel Borkmann <[email protected]>:
Quoted 6 lines · On Tue, 4 Jan 2022 13:10:30 +0100 you wrote:
The
tc_redirectumounts /sys in the new namespace, which can be mounted as shared and cause global umount. The lazy umount also takes down mounted trees under /sys like debugfs, which won't be available after sysfs mounts again and could cause fails in other tests.
Quoted 8 lines
# cat /proc/self/mountinfo | grep debugfs 34 23 0:7 / /sys/kernel/debug rw,nosuid,nodev,noexec,relatime shared:14 - debugfs debugfs rw # cat /proc/self/mountinfo | grep sysfs 23 86 0:22 / /sys rw,nosuid,nodev,noexec,relatime shared:2 - sysfs sysfs rw # mount | grep debugfs debugfs on /sys/kernel/debug type debugfs (rw,nosuid,nodev,noexec,relatime)[...]
Here is the summary with links:
- bpf/selftests: Fix namespace mount setup in tc_redirect
https://git.kernel.org/bpf/bpf-next/c/5e22dd186267
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html
Quoted 18 lines · Jiri Olsa
The
tc_redirectumounts /sys in the new namespace, which can be mounted as shared and cause global umount. The lazy umount also takes down mounted trees under /sys like debugfs, which won't be available after sysfs mounts again and could cause fails in other tests.# cat /proc/self/mountinfo | grep debugfs 34 23 0:7 / /sys/kernel/debug rw,nosuid,nodev,noexec,relatime shared:14 - debugfs debugfs rw # cat /proc/self/mountinfo | grep sysfs 23 86 0:22 / /sys rw,nosuid,nodev,noexec,relatime shared:2 - sysfs sysfs rw # mount | grep debugfs debugfs on /sys/kernel/debug type debugfs (rw,nosuid,nodev,noexec,relatime)# ./test_progs -t tc_redirect #164 tc_redirect:OK Summary: 1/4 PASSED, 0 SKIPPED, 0 FAILED
Quoted 7 lines
# mount | grep debugfs # cat /proc/self/mountinfo | grep debugfs # cat /proc/self/mountinfo | grep sysfs 25 86 0:22 / /sys rw,relatime shared:2 - sysfs sysfs rwMaking the sysfs private under the new namespace so the umount won't trigger the global sysfs umount.
Hey Jiri,
Thanks for the fix. Did you try making tc_redirect non-serial again (s/serial_test_tc_redirect/test_tc_redirect/) and doing parallelized test_progs run (./test_progs -j) in a tight loop for a while? I suspect this might have been an issue forcing us to make this test serial in the first place, so now that it's fixed, we can make parallel test_progs a bit faster.
Quoted 29 lines
Cc Jussi Maki <[email protected]>Reported-by Hangbin Liu <[email protected]>Signed-off-by Jiri Olsa <[email protected]>--- tools/testing/selftests/bpf/prog_tests/tc_redirect.c | 7 +++++++ 1 file changed, 7 insertions(+)tools/testing/selftests/bpf/prog_tests/tc_redirect.c +7
-1@@ -105,6 +105,13 @@ static int setns_by_fd(int nsfd)if (!ASSERT_OK (err,"unshare" ))return err; +/* Make our /sys mount private, so the following umount won't +* trigger the global umount in case it's shared. +*/ + err =mount ("none" ,"/sys" ,NULL ,MS_PRIVATE ,NULL ); +if (!ASSERT_OK (err,"remount private /sys" )) +return err; + err =umount2 ("/sys" ,MNT_DETACH );if (!ASSERT_OK (err,"umount2 /sys" ))return err; --2.33.1
Quoted 27 lines · Andrii Nakryiko
Quoted 18 lines · Jiri Olsa
The
tc_redirectumounts /sys in the new namespace, which can be mounted as shared and cause global umount. The lazy umount also takes down mounted trees under /sys like debugfs, which won't be available after sysfs mounts again and could cause fails in other tests.# cat /proc/self/mountinfo | grep debugfs 34 23 0:7 / /sys/kernel/debug rw,nosuid,nodev,noexec,relatime shared:14 - debugfs debugfs rw # cat /proc/self/mountinfo | grep sysfs 23 86 0:22 / /sys rw,nosuid,nodev,noexec,relatime shared:2 - sysfs sysfs rw # mount | grep debugfs debugfs on /sys/kernel/debug type debugfs (rw,nosuid,nodev,noexec,relatime)# ./test_progs -t tc_redirect #164 tc_redirect:OK Summary: 1/4 PASSED, 0 SKIPPED, 0 FAILEDQuoted 7 lines
# mount | grep debugfs # cat /proc/self/mountinfo | grep debugfs # cat /proc/self/mountinfo | grep sysfs 25 86 0:22 / /sys rw,relatime shared:2 - sysfs sysfs rwMaking the sysfs private under the new namespace so the umount won't trigger the global sysfs umount.
Quoted 8 lines
Hey Jiri,
Thanks for the fix. Did you try making
tc_redirectnon-serial again (s/serial_test_tc_redirect/test_tc_redirect/) and doing parallelizedtest_progsrun (./test_progs -j) in a tight loop for a while? I suspect this might have been an issue forcing us to make this test serial in the first place, so now that it's fixed, we can make paralleltest_progsa bit faster.
hi,
right, will try
jirka
Quoted 31 lines
Quoted 30 lines
Cc Jussi Maki <[email protected]>Reported-by Hangbin Liu <[email protected]>Signed-off-by Jiri Olsa <[email protected]>--- tools/testing/selftests/bpf/prog_tests/tc_redirect.c | 7 +++++++ 1 file changed, 7 insertions(+)tools/testing/selftests/bpf/prog_tests/tc_redirect.c +7
-1@@ -105,6 +105,13 @@ static int setns_by_fd(int nsfd)if (!ASSERT_OK (err,"unshare" ))return err; +/* Make our /sys mount private, so the following umount won't +* trigger the global umount in case it's shared. +*/ + err =mount ("none" ,"/sys" ,NULL ,MS_PRIVATE ,NULL ); +if (!ASSERT_OK (err,"remount private /sys" )) +return err; + err =umount2 ("/sys" ,MNT_DETACH );if (!ASSERT_OK (err,"umount2 /sys" ))return err; --2.33.1
Quoted 37 lines · Jiri Olsa
Quoted 27 lines · Andrii Nakryiko
Quoted 18 lines · Jiri Olsa
The
tc_redirectumounts /sys in the new namespace, which can be mounted as shared and cause global umount. The lazy umount also takes down mounted trees under /sys like debugfs, which won't be available after sysfs mounts again and could cause fails in other tests.# cat /proc/self/mountinfo | grep debugfs 34 23 0:7 / /sys/kernel/debug rw,nosuid,nodev,noexec,relatime shared:14 - debugfs debugfs rw # cat /proc/self/mountinfo | grep sysfs 23 86 0:22 / /sys rw,nosuid,nodev,noexec,relatime shared:2 - sysfs sysfs rw # mount | grep debugfs debugfs on /sys/kernel/debug type debugfs (rw,nosuid,nodev,noexec,relatime)# ./test_progs -t tc_redirect #164 tc_redirect:OK Summary: 1/4 PASSED, 0 SKIPPED, 0 FAILEDQuoted 7 lines
# mount | grep debugfs # cat /proc/self/mountinfo | grep debugfs # cat /proc/self/mountinfo | grep sysfs 25 86 0:22 / /sys rw,relatime shared:2 - sysfs sysfs rwMaking the sysfs private under the new namespace so the umount won't trigger the global sysfs umount.
Quoted 8 lines
Hey Jiri,
Thanks for the fix. Did you try making
tc_redirectnon-serial again (s/serial_test_tc_redirect/test_tc_redirect/) and doing parallelizedtest_progsrun (./test_progs -j) in a tight loop for a while? I suspect this might have been an issue forcing us to make this test serial in the first place, so now that it's fixed, we can make paralleltest_progsa bit faster.
Quoted 2 lines
hi,
right, will try
so I can't reproduce the issue in the first place - that means without my fix and with reverted serial_test_tc_redirect change - by running parallelized test_progs, could you guys try it?
jirka
Quoted 39 lines · Jiri Olsa
Quoted 37 lines · Jiri Olsa
Quoted 35 lines · Andrii Nakryiko
Quoted 18 lines · Jiri Olsa
The
tc_redirectumounts /sys in the new namespace, which can be mounted as shared and cause global umount. The lazy umount also takes down mounted trees under /sys like debugfs, which won't be available after sysfs mounts again and could cause fails in other tests.# cat /proc/self/mountinfo | grep debugfs 34 23 0:7 / /sys/kernel/debug rw,nosuid,nodev,noexec,relatime shared:14 - debugfs debugfs rw # cat /proc/self/mountinfo | grep sysfs 23 86 0:22 / /sys rw,nosuid,nodev,noexec,relatime shared:2 - sysfs sysfs rw # mount | grep debugfs debugfs on /sys/kernel/debug type debugfs (rw,nosuid,nodev,noexec,relatime)# ./test_progs -t tc_redirect #164 tc_redirect:OK Summary: 1/4 PASSED, 0 SKIPPED, 0 FAILEDQuoted 7 lines
# mount | grep debugfs # cat /proc/self/mountinfo | grep debugfs # cat /proc/self/mountinfo | grep sysfs 25 86 0:22 / /sys rw,relatime shared:2 - sysfs sysfs rwMaking the sysfs private under the new namespace so the umount won't trigger the global sysfs umount.
Hey Jiri,
Thanks for the fix. Did you try making
tc_redirectnon-serial again (s/serial_test_tc_redirect/test_tc_redirect/) and doing parallelizedtest_progsrun (./test_progs -j) in a tight loop for a while? I suspect this might have been an issue forcing us to make this test serial in the first place, so now that it's fixed, we can make paralleltest_progsa bit faster.
Quoted 7 lines
Quoted 2 lines
hi,
right, will tryso I can't reproduce the issue in the first place - that means without my fix and with reverted
serial_test_tc_redirectchange - by running parallelizedtest_progs, could you guys try it?
I tried it for a bunch of iterations and didn't repro it either. Doesn't mean much as CI's environment is different, but I think it might be worth making it parallel again and see if it reproes in CI again.
Quoted 2 lines
jirka