[LTP] [PATCH] pause01: race condition

[LTP] [PATCH] pause01: race condition

[Yury Norov]

In my environment (aarch64 + ilp32 emulated by QEMU) I see race
condition here:

[...]
static void setup(void);

int main(int ac, char **av)
{
	int lc;
	struct itimerval it = {
		.it_interval = {.tv_sec = 0, .tv_usec = 0},
		.it_value = {.tv_sec = 0, .tv_usec = 1000},
	};

	tst_parse_opts(ac, av, NULL, NULL);

	setup();

	for (lc = 0; TEST_LOOPING(lc); lc++) {
		tst_count = 0;

		if (setitimer(ITIMER_REAL, &it, NULL))
			tst_brkm(TBROK | TERRNO, NULL, "setitimer() failed");

		TEST(pause());

		if (TEST_RETURN != -1) {
[...]
	}

	tst_exit();
}

static void go(int sig)
{
	(void)sig;
}

void setup(void)
{
	tst_sig(NOFORK, DEF_HANDLER, NULL);
	(void)signal(SIGALRM, go);

	TEST_PAUSE;
}

Alarm may come before pause() is called, and be hiddenly dropped by go()
handler. Next pause() call hangs test forever.

Next hack helps me, though it's not a proper fix. Frankly, I don't
realise how to fix the test properly because sequence of settimer() and
pause() is not atomic, and any il_value we choose is not safe.

Yury.

Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
---
 testcases/kernel/syscalls/pause/pause01.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/testcases/kernel/syscalls/pause/pause01.c b/testcases/kernel/syscalls/pause/pause01.c
index c967d8e..f4b2dcd 100644
--- a/testcases/kernel/syscalls/pause/pause01.c
+++ b/testcases/kernel/syscalls/pause/pause01.c
@@ -48,7 +48,7 @@ int main(int ac, char **av)
 	int lc;
 	struct itimerval it = {
 		.it_interval = {.tv_sec = 0, .tv_usec = 0},
-		.it_value = {.tv_sec = 0, .tv_usec = 1000},
+		.it_value = {.tv_sec = 0, .tv_usec = 10000},
 	};
 
 	tst_parse_opts(ac, av, NULL, NULL);
-- 
2.5.0

[LTP] [PATCH] pause01: race condition

[Jan Stancek]

----- Original Message -----
> From: "Yury Norov" <ynorov@caviumnetworks.com>
> To: ltp@lists.linux.it
> Sent: Tuesday, 5 April, 2016 9:11:40 PM
> Subject: [LTP] [PATCH] pause01: race condition
> 
> In my environment (aarch64 + ilp32 emulated by QEMU) I see race
> condition here:
> 
> [...]
> static void setup(void);
> 
> int main(int ac, char **av)
> {
> 	int lc;
> 	struct itimerval it = {
> 		.it_interval = {.tv_sec = 0, .tv_usec = 0},
> 		.it_value = {.tv_sec = 0, .tv_usec = 1000},
> 	};
> 
> 	tst_parse_opts(ac, av, NULL, NULL);
> 
> 	setup();
> 
> 	for (lc = 0; TEST_LOOPING(lc); lc++) {
> 		tst_count = 0;
> 
> 		if (setitimer(ITIMER_REAL, &it, NULL))
> 			tst_brkm(TBROK | TERRNO, NULL, "setitimer() failed");
> 
> 		TEST(pause());
> 
> 		if (TEST_RETURN != -1) {
> [...]
> 	}
> 
> 	tst_exit();
> }
> 
> static void go(int sig)
> {
> 	(void)sig;
> }
> 
> void setup(void)
> {
> 	tst_sig(NOFORK, DEF_HANDLER, NULL);
> 	(void)signal(SIGALRM, go);
> 
> 	TEST_PAUSE;
> }
> 
> Alarm may come before pause() is called, and be hiddenly dropped by go()
> handler. Next pause() call hangs test forever.
> 
> Next hack helps me, though it's not a proper fix. Frankly, I don't
> realise how to fix the test properly because sequence of settimer() and
> pause() is not atomic, and any il_value we choose is not safe.

I've seen this hang once in past. At the time I was thinking
we use 2 processes. Parent will fork a child and wait.
Just before child calls pause() we signal parent and
parent will call tst_process_state_wait() to wait until
child process falls asleep (on pause() call).
Then parent can send a signal to child and check it wakes up.

Regards,
Jan

> 
> Yury.
> 
> Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
> ---
>  testcases/kernel/syscalls/pause/pause01.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/testcases/kernel/syscalls/pause/pause01.c
> b/testcases/kernel/syscalls/pause/pause01.c
> index c967d8e..f4b2dcd 100644
> --- a/testcases/kernel/syscalls/pause/pause01.c
> +++ b/testcases/kernel/syscalls/pause/pause01.c
> @@ -48,7 +48,7 @@ int main(int ac, char **av)
>  	int lc;
>  	struct itimerval it = {
>  		.it_interval = {.tv_sec = 0, .tv_usec = 0},
> -		.it_value = {.tv_sec = 0, .tv_usec = 1000},
> +		.it_value = {.tv_sec = 0, .tv_usec = 10000},
>  	};
>  
>  	tst_parse_opts(ac, av, NULL, NULL);
> --
> 2.5.0
> 
> 
> --
> Mailing list info: https://lists.linux.it/listinfo/ltp
> 

[LTP] [PATCH] pause01: race condition

[Cyril Hrubis]

Hi!
> > Alarm may come before pause() is called, and be hiddenly dropped by go()
> > handler. Next pause() call hangs test forever.
> > 
> > Next hack helps me, though it's not a proper fix. Frankly, I don't
> > realise how to fix the test properly because sequence of settimer() and
> > pause() is not atomic, and any il_value we choose is not safe.
> 
> I've seen this hang once in past. At the time I was thinking
> we use 2 processes. Parent will fork a child and wait.
> Just before child calls pause() we signal parent and
> parent will call tst_process_state_wait() to wait until
> child process falls asleep (on pause() call).
> Then parent can send a signal to child and check it wakes up.

I was about to suggest the same.

-- 
Cyril Hrubis
chrubis@suse.cz