Bugfix: Signal Trace mode may send SIGURG to wrong thread #3039

CodeZHXS · 2025-07-24T11:37:26Z

Bugfix: Use rt_tgsigqueueinfo to send the signal to correct worker thread

What problem does this PR solve?

Issue Number: fix #3038

Problem Summary:

What is changed and the side effects?

Changed:

Side effects:

Performance effects:
Breaking backward compatibility:

Check List:

Please make sure your changes are compilable.
When providing us with a new feature, it is best to add related tests.
Please follow Contributor Covenant Code of Conduct.

chenBright · 2025-07-26T07:24:27Z

https://man7.org/linux/man-pages/man2/sigqueue.2.html

On Linux, sigqueue() is implemented using the rt_sigqueueinfo(2) system call.

chenBright · 2025-07-26T07:27:10Z

Perhaps we can use pthread_sigqueue which has better compatibility.

https://man7.org/linux/man-pages/man3/pthread_sigqueue.3.html

The pthread_sigqueue() function performs a similar task to
sigqueue(3), but, rather than sending a signal to a process, it
sends a signal to a thread in the same process as the calling thread.

CodeZHXS · 2025-07-26T07:35:19Z

Perhaps we can use pthread_sigqueue which has better compatibility.

https://man7.org/linux/man-pages/man3/pthread_sigqueue.3.html

The pthread_sigqueue() function performs a similar task to
sigqueue(3), but, rather than sending a signal to a process, it
sends a signal to a thread in the same process as the calling thread.

It's okay. So I need to change the type of worker_tid from pid_t to pthread_t, right?

chenBright · 2025-07-26T07:41:54Z

So I need to change the type of worker_tid from pid_t to pthread_t, right?

Yes.

CodeZHXS · 2025-07-26T08:16:24Z

@chenBright Please review again.

Copilot

Pull Request Overview

This PR fixes a signal handling bug in the Signal Trace mode where SIGURG was being sent to the wrong thread. The core issue was using process IDs (pid_t) instead of pthread IDs (pthread_t) for thread identification.

Key changes:

Changed thread identification from pid_t to pthread_t throughout the codebase
Replaced sigqueue() with pthread_sigqueue() for proper thread-specific signal delivery
Updated initialization values from -1 to 0 for pthread compatibility

Reviewed Changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/bthread/task_tracer.h	Updated function signatures and member variables to use `pthread_t`
src/bthread/task_tracer.cpp	Implemented thread safety check and switched to `pthread_sigqueue()`
src/bthread/task_meta.h	Changed worker thread ID type from `pid_t` to `pthread_t`
src/bthread/task_group.h	Updated TaskGroup thread ID member and accessor method
src/bthread/task_group.cpp	Updated local variable type for consistency
src/bthread/task_control.cpp	Changed thread ID assignment from `syscall(SYS_gettid)` to `pthread_self()`

Copilot · 2025-07-26T08:17:06Z

src/bthread/task_tracer.cpp

    // Therefore, use async-signal-safe absl::DefaultStackUnwinder instead of libunwind.

+    if (worker_tid == pthread_self()) {
+        return Result::MakeErrorResult("Forbid to trace self=%d", worker_tid);


The format specifier %d is incorrect for pthread_t. pthread_t is an opaque type that may not be an integer on all platforms. Consider using %p to format it as a pointer or convert it appropriately for display.

Suggested change

return Result::MakeErrorResult("Forbid to trace self=%d", worker_tid);

return Result::MakeErrorResult("Forbid to trace self=%p", (void*)worker_tid);

On other platforms, pthread_t is not an integer, but a pointer.

Copilot · 2025-07-26T08:17:06Z

src/bthread/task_tracer.cpp

            // Remove reference for SignalHandler.
            signal_sync->RemoveRefManually();
-            return Result::MakeErrorResult("Fail to sigqueue: %s, tid: %d", berror(), tid);
+            return Result::MakeErrorResult("Fail to pthread_sigqueue: %s, tid: %d", berror(), worker_tid);


The format specifier %d is incorrect for pthread_t. pthread_t is an opaque type that may not be an integer on all platforms. Consider using %p to format it as a pointer or convert it appropriately for display.

Suggested change

return Result::MakeErrorResult("Fail to pthread_sigqueue: %s, tid: %d", berror(), worker_tid);

return Result::MakeErrorResult("Fail to pthread_sigqueue: %s, tid: %p", berror(), (void*)worker_tid);

On other platforms, pthread_t is not an integer, but a pointer.

src/bthread/task_group.cpp

src/bthread/task_group.h

chenBright · 2025-07-26T11:45:02Z

src/bthread/task_tracer.cpp

    // Therefore, use async-signal-safe absl::DefaultStackUnwinder instead of libunwind.

+    if (worker_tid == pthread_self()) {
+        return Result::MakeErrorResult("Forbid to trace self=%d", worker_tid);


On other platforms, pthread_t is not an integer, but a pointer.

chenBright · 2025-07-26T11:45:10Z

src/bthread/task_tracer.cpp

            // Remove reference for SignalHandler.
            signal_sync->RemoveRefManually();
-            return Result::MakeErrorResult("Fail to sigqueue: %s, tid: %d", berror(), tid);
+            return Result::MakeErrorResult("Fail to pthread_sigqueue: %s, tid: %d", berror(), worker_tid);


On other platforms, pthread_t is not an integer, but a pointer.

* Bugfix: Use `pthread_sigqueue` to send the signal to correct worker thread' * Change the type of `TaskGroup::_tid` to `pthread_t`

CodeZHXS · 2025-07-31T14:48:25Z

@chenBright ping

chenBright

LGTM

CodeZHXS force-pushed the fix_signal_trace_syscall branch 2 times, most recently from 4abae57 to 8a48105 Compare July 24, 2025 13:20

wwbmmm requested review from chenBright and Copilot July 25, 2025 11:34

This comment was marked as outdated.

Sign in to view

CodeZHXS force-pushed the fix_signal_trace_syscall branch 3 times, most recently from 4ad9917 to eecd82a Compare July 26, 2025 08:11

CodeZHXS requested a review from Copilot July 26, 2025 08:16

Copilot AI reviewed Jul 26, 2025

View reviewed changes

CodeZHXS force-pushed the fix_signal_trace_syscall branch from eecd82a to ef2d16e Compare July 26, 2025 08:23

chenBright reviewed Jul 26, 2025

View reviewed changes

CodeZHXS force-pushed the fix_signal_trace_syscall branch 2 times, most recently from e002e3a to 825625b Compare July 28, 2025 14:58

Bugfix: Signal Trace mode may send SIGURG to wrong thread (apache#3038)

5fe1a06

* Bugfix: Use `pthread_sigqueue` to send the signal to correct worker thread' * Change the type of `TaskGroup::_tid` to `pthread_t`

CodeZHXS force-pushed the fix_signal_trace_syscall branch from 825625b to 5fe1a06 Compare July 28, 2025 15:01

chenBright approved these changes Aug 26, 2025

View reviewed changes

chenBright merged commit b6ae7bf into apache:master Aug 26, 2025
15 checks passed

	return Result::MakeErrorResult("Forbid to trace self=%d", worker_tid);
	return Result::MakeErrorResult("Forbid to trace self=%p", (void*)worker_tid);

	return Result::MakeErrorResult("Fail to pthread_sigqueue: %s, tid: %d", berror(), worker_tid);
	return Result::MakeErrorResult("Fail to pthread_sigqueue: %s, tid: %p", berror(), (void*)worker_tid);

Bugfix: Signal Trace mode may send SIGURG to wrong thread #3039

Bugfix: Signal Trace mode may send SIGURG to wrong thread #3039

Uh oh!

Conversation

CodeZHXS commented Jul 24, 2025

What problem does this PR solve?

What is changed and the side effects?

Check List:

Uh oh!

This comment was marked as outdated.

Uh oh!

chenBright commented Jul 26, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chenBright commented Jul 26, 2025

Uh oh!

CodeZHXS commented Jul 26, 2025 • edited by chenBright Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chenBright commented Jul 26, 2025

Uh oh!

CodeZHXS commented Jul 26, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Copilot AI Jul 26, 2025

Choose a reason for hiding this comment

Uh oh!

chenBright Jul 26, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jul 26, 2025

Choose a reason for hiding this comment

Uh oh!

chenBright Jul 26, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

chenBright Jul 26, 2025

Choose a reason for hiding this comment

Uh oh!

chenBright Jul 26, 2025

Choose a reason for hiding this comment

Uh oh!

CodeZHXS commented Jul 31, 2025

Uh oh!

chenBright left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

chenBright commented Jul 26, 2025 •

edited

Loading

CodeZHXS commented Jul 26, 2025 •

edited by chenBright

Loading