Here is a couple of patches to fix issues. I think mount_options.tc's one is a real bug(I'm not sure how it worked), but another one is an enhancement for (my) execution environment. Anyway, those should go through kselftests tree.
Thank you, ---
Masami Hiramatsu (Google) (2): selftests/ftrace: Fix to use remount when testing mount GID option selftests/ftrace: Make uprobe test more robust against binary name
.../ftrace/test.d/00basic/mount_options.tc | 8 ++++---- .../ftrace/test.d/dynevent/add_remove_uprobe.tc | 4 +++- 2 files changed, 7 insertions(+), 5 deletions(-)
-- Masami Hiramatsu (Google) mhiramat@kernel.org
From: Masami Hiramatsu (Google) mhiramat@kernel.org
Fix mount_options.tc to use remount option to mount the tracefs. Since the current implementation does not umount the tracefs, this test always fails because of -EBUSY error. Using remount option will allow us to change the mount option.
Fixes: 8b55572e5180 ("tracing/selftests: Add tracefs mount options test") Signed-off-by: Masami Hiramatsu (Google) mhiramat@kernel.org Cc: stable@vger.kernel.org --- .../ftrace/test.d/00basic/mount_options.tc | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/tools/testing/selftests/ftrace/test.d/00basic/mount_options.tc b/tools/testing/selftests/ftrace/test.d/00basic/mount_options.tc index 35e8d47d6072..8a7ce647a60d 100644 --- a/tools/testing/selftests/ftrace/test.d/00basic/mount_options.tc +++ b/tools/testing/selftests/ftrace/test.d/00basic/mount_options.tc @@ -15,11 +15,11 @@ find_alternate_gid() { tac /etc/group | grep -v ":$original_gid:" | head -1 | cut -d: -f3 }
-mount_tracefs_with_options() { +remount_tracefs_with_options() { local mount_point="$1" local options="$2"
- mount -t tracefs -o "$options" nodev "$mount_point" + mount -t tracefs -o "remount,$options" nodev "$mount_point"
setup } @@ -81,7 +81,7 @@ test_gid_mount_option() {
# Unmount existing tracefs instance and mount with new GID unmount_tracefs "$mount_point" - mount_tracefs_with_options "$mount_point" "$new_options" + remount_tracefs_with_options "$mount_point" "$new_options"
check_gid "$mount_point" "$other_group"
@@ -92,7 +92,7 @@ test_gid_mount_option() {
# Unmount and remount with the original GID unmount_tracefs "$mount_point" - mount_tracefs_with_options "$mount_point" "$mount_options" + remount_tracefs_with_options "$mount_point" "$mount_options" check_gid "$mount_point" "$original_group" }
From: Masami Hiramatsu (Google) mhiramat@kernel.org
Make add_remove_uprobe test case more robust against various real binary name. Current add_remove_uprobe.tc test expects the real binary of /bin/sh is '*/bin/*sh', but it does not work on busybox environment. Instead of using fixed pattern, use readlink to identify real binary name.
Signed-off-by: Masami Hiramatsu (Google) mhiramat@kernel.org --- .../ftrace/test.d/dynevent/add_remove_uprobe.tc | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_uprobe.tc b/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_uprobe.tc index a275decdc880..86c76679c56e 100644 --- a/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_uprobe.tc +++ b/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_uprobe.tc @@ -6,8 +6,10 @@ echo 0 > events/enable echo > dynamic_events
+REALBIN=`readlink -f /bin/sh` + echo 'cat /proc/$$/maps' | /bin/sh | \ - grep "r-xp .*/bin/.*sh$" | \ + grep "r-xp .*${REALBIN}$" | \ awk '{printf "p:myevent %s:0x%s\n", $6,$3 }' >> uprobe_events
grep -q myevent uprobe_events
On Tue, Jan 07, 2025 at 09:11:16PM +0900, Masami Hiramatsu (Google) wrote:
From: Masami Hiramatsu (Google) mhiramat@kernel.org
Make add_remove_uprobe test case more robust against various real binary name. Current add_remove_uprobe.tc test expects the real binary of /bin/sh is '*/bin/*sh', but it does not work on busybox environment. Instead of using fixed pattern, use readlink to identify real binary name.
Signed-off-by: Masami Hiramatsu (Google) mhiramat@kernel.org
.../ftrace/test.d/dynevent/add_remove_uprobe.tc | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_uprobe.tc b/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_uprobe.tc index a275decdc880..86c76679c56e 100644 --- a/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_uprobe.tc +++ b/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_uprobe.tc @@ -6,8 +6,10 @@ echo 0 > events/enable echo > dynamic_events +REALBIN=`readlink -f /bin/sh`
echo 'cat /proc/$$/maps' | /bin/sh | \
- grep "r-xp .*/bin/.*sh$" | \
- grep "r-xp .*${REALBIN}$" | \ awk '{printf "p:myevent %s:0x%s\n", $6,$3 }' >> uprobe_events
This test fails on s390, not because of this change, but since the test seems to assume that the entry point of /bin/sh is at offset 0; or at least there is some code that is executed at least once at offset 0; but that is not the case (at least) for s390.
Not sure how this can be addressed. Using "readelf -f" to get the entry point address of the executable and use that for the uprobe would work, but it would also add a dependency to binutils for the ftrace selftests, which doesn't exist yet, as far as I can tell.
Would it maybe an option for architectures to opt-in to use readelf and skip the test (exit_unsupported) if readelf is not available?
On Mon, 17 Feb 2025 14:42:44 +0100 Heiko Carstens hca@linux.ibm.com wrote:
+REALBIN=`readlink -f /bin/sh`
echo 'cat /proc/$$/maps' | /bin/sh | \
- grep "r-xp .*/bin/.*sh$" | \
- grep "r-xp .*${REALBIN}$" | \ awk '{printf "p:myevent %s:0x%s\n", $6,$3 }' >> uprobe_events
This test fails on s390, not because of this change, but since the test seems to assume that the entry point of /bin/sh is at offset 0; or at least there is some code that is executed at least once at offset 0; but that is not the case (at least) for s390.
Not sure how this can be addressed. Using "readelf -f" to get the entry point address of the executable and use that for the uprobe would work, but it would also add a dependency to binutils for the ftrace selftests, which doesn't exist yet, as far as I can tell.
Would it maybe an option for architectures to opt-in to use readelf and skip the test (exit_unsupported) if readelf is not available?
I'm OK for adding a dependency on readelf existing on the test system, as it is part of binutils.
Masami and Shuah,
Are you OK with that?
-- Steve
On 1/7/25 05:10, Masami Hiramatsu (Google) wrote:
Here is a couple of patches to fix issues. I think mount_options.tc's one is a real bug(I'm not sure how it worked), but another one is an enhancement for (my) execution environment. Anyway, those should go through kselftests tree.
Thank you,
Masami Hiramatsu (Google) (2): selftests/ftrace: Fix to use remount when testing mount GID option selftests/ftrace: Make uprobe test more robust against binary name
.../ftrace/test.d/00basic/mount_options.tc | 8 ++++---- .../ftrace/test.d/dynevent/add_remove_uprobe.tc | 4 +++- 2 files changed, 7 insertions(+), 5 deletions(-)
-- Masami Hiramatsu (Google) mhiramat@kernel.org
I applied these two patches to linux-kselftest next fro Linux 6.14-rc1
thanks, -- Shuah
linux-kselftest-mirror@lists.linaro.org
-
Heiko Carstens -
Masami Hiramatsu (Google) -
Shuah Khan -
Steven Rostedt