Handles the masked signal when the thread exits

Handles the masked signal when the thread exits

[Gibeom Gwon]

Hello,

I am experiencing strange(unintended?) behavior when using pthread with 
signals. If I set the signal mask with pthread_sigmask() in the thread 
function and the process has pending signal when thread is exiting, 
signal handler executed in thread.

It looks like glibc restores original signal mask at the end of the 
thread. So I suspect this eventually triggers masked signal handler. But 
I'm not sure it is intended or not.

Code samples:

sleep.c
-------
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <signal.h>
#include <sys/wait.h>
#include <pthread.h>

pthread_t thread;
pid_t pid;

void stop_threads() {
		pthread_cancel(thread);
		pthread_join(thread, NULL);
}

void sigint_handler(int signum) {
	printf("sigint: %ld\n",pthread_self());
	stop_threads();
	exit(0);
}

void sigchld_handler(int signum) {
	printf("sigchld: %ld\n",pthread_self());
	pid_t pid;
	int status;
	while((pid = waitpid(-1,&status,WNOHANG)) > 0) {}
}

void spawn_sleep() {
	pid = fork();
	if(pid == 0)
		execl("./sleep","sleep",NULL);
}

void* worker(void *arg) {
	sigset_t mask;
	sigfillset(&mask);
	pthread_sigmask(SIG_SETMASK,&mask,NULL);

	pthread_setcancelstate(PTHREAD_CANCEL_DISABLE,NULL);
	int i = 10;
	while(i) {
		printf("worker...\n");
		sleep(1);
		i--;
	}
	pthread_setcancelstate(PTHREAD_CANCEL_ENABLE,NULL);
     	printf("thread close\n");
	pthread_exit(0);
}

void spawn_thread() {
     pthread_create(&thread,NULL,worker,NULL);
     printf("child thread: %ld\n",thread);
}

int main() {
	printf("main thread: %ld\n",pthread_self());
	spawn_sleep();	

	struct sigaction sa1;
	sa1.sa_handler = sigint_handler;
	sigemptyset(&sa1.sa_mask);
	sigaddset(&sa1.sa_mask,SIGCHLD);
	sa1.sa_flags = 0;
	sigaction(SIGINT, &sa1, 0);

	struct sigaction sa2;
	sa2.sa_handler = sigchld_handler;
	sigemptyset(&sa2.sa_mask);
	sa2.sa_flags = 0;
	sigaction(SIGCHLD, &sa2, 0);

	spawn_thread();

	while(1)
		sleep(1);

	return 0;
}


result
======
$ ./signal
main thread: 139882152073024
child thread: 139882152068672
worker...
worker...
worker...
^Csigint: 139882152073024
worker...
worker...
worker...
worker...
worker...
worker...
worker...
thread close
sigchld: 139882152068672

Comprehensive signal mask set in worker() but SIGCHLD handler executed 
after thread function exited.

Regards,
Gibeom Gwon

Re: Handles the masked signal when the thread exits

[Florian Weimer]

* Gibeom Gwon:

> I am experiencing strange(unintended?) behavior when using pthread
> with signals. If I set the signal mask with pthread_sigmask() in the
> thread function and the process has pending signal when thread is
> exiting, signal handler executed in thread.
>
> It looks like glibc restores original signal mask at the end of the
> thread. So I suspect this eventually triggers masked signal
> handler. But I'm not sure it is intended or not.

Which glibc version are you using?

Thanks,
Florian

Re: Handles the masked signal when the thread exits

[Gibeom Gwon]

On 11/19/21 03:08, Florian Weimer wrote:
> * Gibeom Gwon:
> 
>> I am experiencing strange(unintended?) behavior when using pthread
>> with signals. If I set the signal mask with pthread_sigmask() in the
>> thread function and the process has pending signal when thread is
>> exiting, signal handler executed in thread.
>>
>> It looks like glibc restores original signal mask at the end of the
>> thread. So I suspect this eventually triggers masked signal
>> handler. But I'm not sure it is intended or not.
> 
> Which glibc version are you using?
> 
> Thanks,
> Florian
> 

Ah, I forgot to write the version. I'm using 2.33 and linux distribution
  is Arch Linux.

Re: Handles the masked signal when the thread exits

[Florian Weimer]

* Gibeom Gwon:

> On 11/19/21 03:08, Florian Weimer wrote:
>> * Gibeom Gwon:
>> 
>>> I am experiencing strange(unintended?) behavior when using pthread
>>> with signals. If I set the signal mask with pthread_sigmask() in the
>>> thread function and the process has pending signal when thread is
>>> exiting, signal handler executed in thread.
>>>
>>> It looks like glibc restores original signal mask at the end of the
>>> thread. So I suspect this eventually triggers masked signal
>>> handler. But I'm not sure it is intended or not.
>> Which glibc version are you using?

> Ah, I forgot to write the version. I'm using 2.33 and linux
> distribution is Arch Linux.

We had some issues with signals and thread exit in 2.34, but they are
exclusive to that release and cannot happen in 2.33.  I can't reproduce
the behavior you see with upstream 2.33, either.

What's your kernel version?  I've tried 5.14.13 and 5.14.17.

Thanks,
Florian

Re: Handles the masked signal when the thread exits

[Gibeom Gwon]

On 11/19/21 03:19, Florian Weimer wrote:
> * Gibeom Gwon:
> 
>> On 11/19/21 03:08, Florian Weimer wrote:
>>> * Gibeom Gwon:
>>>
>>>> I am experiencing strange(unintended?) behavior when using pthread
>>>> with signals. If I set the signal mask with pthread_sigmask() in the
>>>> thread function and the process has pending signal when thread is
>>>> exiting, signal handler executed in thread.
>>>>
>>>> It looks like glibc restores original signal mask at the end of the
>>>> thread. So I suspect this eventually triggers masked signal
>>>> handler. But I'm not sure it is intended or not.
>>> Which glibc version are you using?
> 
>> Ah, I forgot to write the version. I'm using 2.33 and linux
>> distribution is Arch Linux.
> 
> We had some issues with signals and thread exit in 2.34, but they are
> exclusive to that release and cannot happen in 2.33.  I can't reproduce
> the behavior you see with upstream 2.33, either.
> 
> What's your kernel version?  I've tried 5.14.13 and 5.14.17.
> 
> Thanks,
> Florian
> 

Kernel version is 5.15.2.

Well, I accidently dropped sleep.c code sample. Here are the code
samples again. Sorry to bother you.

sleep.c
-------
#include <unistd.h>

int main() {
	sleep(50);
	return 0;
}

signal.c
--------
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <signal.h>
#include <sys/wait.h>
#include <pthread.h>

pthread_t thread;
pid_t pid;

void stop_threads() {
		pthread_cancel(thread);
		pthread_join(thread, NULL);
}

void sigint_handler(int signum) {
	printf("sigint: %ld\n",pthread_self());
	stop_threads();
	exit(0);
}

void sigchld_handler(int signum) {
	printf("sigchld: %ld\n",pthread_self());
	pid_t pid;
	int status;
	while((pid = waitpid(-1,&status,WNOHANG)) > 0) {}
}

void spawn_sleep() {
	pid = fork();
	if(pid == 0)
		execl("./sleep","sleep",NULL);
}

void* worker(void *arg) {
	sigset_t mask;
	sigfillset(&mask);
	pthread_sigmask(SIG_SETMASK,&mask,NULL);

	pthread_setcancelstate(PTHREAD_CANCEL_DISABLE,NULL);
	int i = 10;
	while(i) {
		printf("worker...\n");
		sleep(1);
		i--;
	}
	pthread_setcancelstate(PTHREAD_CANCEL_ENABLE,NULL);
     	printf("thread close\n");
	pthread_exit(0);
}

void spawn_thread() {
     pthread_create(&thread,NULL,worker,NULL);
     printf("child thread: %ld\n",thread);
}

int main() {
	printf("main thread: %ld\n",pthread_self());
	spawn_sleep();	

	struct sigaction sa1;
	sa1.sa_handler = sigint_handler;
	sigemptyset(&sa1.sa_mask);
	sigaddset(&sa1.sa_mask,SIGCHLD);
	sa1.sa_flags = 0;
	sigaction(SIGINT, &sa1, 0);

	struct sigaction sa2;
	sa2.sa_handler = sigchld_handler;
	sigemptyset(&sa2.sa_mask);
	sa2.sa_flags = 0;
	sigaction(SIGCHLD, &sa2, 0);

	spawn_thread();

	while(1)
		sleep(1);

	return 0;
}


Regards,
Gibeom Gwon

Re: Handles the masked signal when the thread exits

[Florian Weimer]

* Gibeom Gwon:

> On 11/19/21 03:19, Florian Weimer wrote:
>> * Gibeom Gwon:
>> 
>>> On 11/19/21 03:08, Florian Weimer wrote:
>>>> * Gibeom Gwon:
>>>>
>>>>> I am experiencing strange(unintended?) behavior when using pthread
>>>>> with signals. If I set the signal mask with pthread_sigmask() in the
>>>>> thread function and the process has pending signal when thread is
>>>>> exiting, signal handler executed in thread.
>>>>>
>>>>> It looks like glibc restores original signal mask at the end of the
>>>>> thread. So I suspect this eventually triggers masked signal
>>>>> handler. But I'm not sure it is intended or not.
>>>> Which glibc version are you using?
>> 
>>> Ah, I forgot to write the version. I'm using 2.33 and linux
>>> distribution is Arch Linux.
>> We had some issues with signals and thread exit in 2.34, but they
>> are
>> exclusive to that release and cannot happen in 2.33.  I can't reproduce
>> the behavior you see with upstream 2.33, either.
>> What's your kernel version?  I've tried 5.14.13 and 5.14.17.
>> Thanks,
>> Florian
>> 
>
> Kernel version is 5.15.2.
>
> Well, I accidently dropped sleep.c code sample. Here are the code
> samples again. Sorry to bother you.

Ah, now the test makes a little bit more sense.  But I still get the
expected output:

main thread: 140317152449984
child thread: 140317150295616
worker...
worker...
worker...
worker...
worker...
worker...
worker...
worker...
worker...
worker...
thread close
sigchld: 140317152449984

So the signal gets delivered on the main thread, as expected.

Thanks,
Florian

Re: Handles the masked signal when the thread exits

[Gibeom Gwon]

On 11/19/21 03:31, Florian Weimer wrote:
> * Gibeom Gwon:
> 
>> On 11/19/21 03:19, Florian Weimer wrote:
>>> * Gibeom Gwon:
>>>
>>>> On 11/19/21 03:08, Florian Weimer wrote:
>>>>> * Gibeom Gwon:
>>>>>
>>>>>> I am experiencing strange(unintended?) behavior when using pthread
>>>>>> with signals. If I set the signal mask with pthread_sigmask() in the
>>>>>> thread function and the process has pending signal when thread is
>>>>>> exiting, signal handler executed in thread.
>>>>>>
>>>>>> It looks like glibc restores original signal mask at the end of the
>>>>>> thread. So I suspect this eventually triggers masked signal
>>>>>> handler. But I'm not sure it is intended or not.
>>>>> Which glibc version are you using?
>>>
>>>> Ah, I forgot to write the version. I'm using 2.33 and linux
>>>> distribution is Arch Linux.
>>> We had some issues with signals and thread exit in 2.34, but they
>>> are
>>> exclusive to that release and cannot happen in 2.33.  I can't reproduce
>>> the behavior you see with upstream 2.33, either.
>>> What's your kernel version?  I've tried 5.14.13 and 5.14.17.
>>> Thanks,
>>> Florian
>>>
>>
>> Kernel version is 5.15.2.
>>
>> Well, I accidently dropped sleep.c code sample. Here are the code
>> samples again. Sorry to bother you.
> 
> Ah, now the test makes a little bit more sense.  But I still get the
> expected output:
> 
> main thread: 140317152449984
> child thread: 140317150295616
> worker...
> worker...
> worker...
> worker...
> worker...
> worker...
> worker...
> worker...
> worker...
> worker...
> thread close
> sigchld: 140317152449984
> 
> So the signal gets delivered on the main thread, as expected.
> 
> Thanks,
> Florian
> 

I hit Ctrl+C in the middle of running signal program. You can see
^C and sigint:.. printed in my result.

Regards,
Gibeom Gwon

Re: Handles the masked signal when the thread exits

[Florian Weimer]

* Gibeom Gwon:

> I hit Ctrl+C in the middle of running signal program. You can see
> ^C and sigint:.. printed in my result.

I see it now.  It's a real bug.  It only happens with cancellation.

Would yo umind filing a bug here?

  <https://sourceware.org/bugzilla/enter_bug.cgi?product=glibc&component=nptl>

The fix should look something like this:

diff --git a/nptl/pthread_create.c b/nptl/pthread_create.c
index d6ea43a754..bad9eeb52f 100644
--- a/nptl/pthread_create.c
+++ b/nptl/pthread_create.c
@@ -406,8 +406,6 @@ start_thread (void *arg)
   unwind_buf.priv.data.prev = NULL;
   unwind_buf.priv.data.cleanup = NULL;
 
-  __libc_signal_restore_set (&pd->sigmask);
-
   /* Allow setxid from now onwards.  */
   if (__glibc_unlikely (atomic_exchange_acq (&pd->setxid_futex, 0) == -2))
     futex_wake (&pd->setxid_futex, 1, FUTEX_PRIVATE);
@@ -417,6 +415,8 @@ start_thread (void *arg)
       /* Store the new cleanup handler info.  */
       THREAD_SETMEM (pd, cleanup_jmp_buf, &unwind_buf);
 
+      __libc_signal_restore_set (&pd->sigmask);
+
       LIBC_PROBE (pthread_start, 3, (pthread_t) pd, pd->start_routine, pd->arg);
 
       /* Run the code the user provided.  */

Thanks,
Florian

Re: Handles the masked signal when the thread exits

[Gibeom Gwon]

On 11/19/21 04:10, Florian Weimer wrote:
> * Gibeom Gwon:
> 
>> I hit Ctrl+C in the middle of running signal program. You can see
>> ^C and sigint:.. printed in my result.
> 
> I see it now.  It's a real bug.  It only happens with cancellation.
> 
> Would yo umind filing a bug here?
> 
>    <https://sourceware.org/bugzilla/enter_bug.cgi?product=glibc&component=nptl>
> 

Opened the bug. <https://sourceware.org/bugzilla/show_bug.cgi?id=28607>
Thank you!

Regards,
Gibeom Gwon