Synchronization between two processes

The child-process wait SIGUSR1 which would be sended by parent-process, if the child-process execute sigtimedwait() after parent-process send the signal, it would never receive the SIGUSR1 from parent-process. I cann't make sure the SIGUSR1 be sended after child-process execute sigtimedwait() with 100 percent, and i try my best.

Signed-off-by Gui Xiaohua <[email protected]>

kernel/containers/pidns/pidns12.c renamed from kernel/containers/pidns/pidns12-old.c +4 -2

@@ -109,7 +109,7 @@ int child_fn(void *arg)
 	}
 
 	/* Set timeout for sigtimedwait */
-	timeout.tv_sec = 3;
+	timeout.tv_sec = 10;
 	timeout.tv_nsec = 0;
 
 	/* Set mask to wait for SIGUSR1 signal */

@@ -182,6 +182,9 @@ int main(int argc, char *argv[])
 		cleanup();
 	}
 
+	/*Try best to make sure the SIGUSR1 be sended after child-process execute sigtimedwait*/
+	sleep(5);
+
 	/* Send SIGUSR1 to container init */
 	if (kill(cpid, SIGUSR1) == -1) {
 		tst_resm(TBROK, "parent: kill() failed(%s).", strerror(errno));
 
 
------------------------------------------------------------------------------

Crystal Reports - New Free Runtime and 30 Day Trial

Check out the new simplified licensing option that enables unlimited royalty-free distribution of the report engine for externally facing server and web deployment.
http://p.sf.net/sfu/businessobjects
_______________________________________________
Ltp-list mailing list [email protected]
https://lists.sourceforge.net/lists/listinfo/ltp-list

The child-process wait SIGUSR1 which would be sended by parent-process, if the child-process execute sigtimedwait() after parent-process send the signal, it would never receive the SIGUSR1 from parent-process. I cann't make sure the SIGUSR1 be sended after child-process execute sigtimedwait() with 100 percent, and i try my best.

Signed-off-by Gui Xiaohua <[email protected]>

kernel/containers/pidns/pidns12.c renamed from kernel/containers/pidns/pidns12-old.c +4 -2

@@ -109,7 +109,7 @@ int child_fn(void *arg)
 	}
 
 	/* Set timeout for sigtimedwait */
-	timeout.tv_sec = 3;
+	timeout.tv_sec = 10;
 	timeout.tv_nsec = 0;
 
 	/* Set mask to wait for SIGUSR1 signal */

@@ -182,6 +182,9 @@ int main(int argc, char *argv[])
 		cleanup();
 	}
 
+	/*Try best to make sure the SIGUSR1 be sended after child-process execute sigtimedwait*/
+	sleep(5);
+
 	/* Send SIGUSR1 to container init */
 	if (kill(cpid, SIGUSR1) == -1) {
 		tst_resm(TBROK, "parent: kill() failed(%s).", strerror(errno));
 
 
------------------------------------------------------------------------------

Crystal Reports - New Free Runtime and 30 Day Trial

Check out the new simplified licensing option that enables unlimited royalty-free distribution of the report engine for externally facing server and web deployment.
http://p.sf.net/sfu/businessobjects
_______________________________________________
Ltp-list mailing list [email protected]
https://lists.sourceforge.net/lists/listinfo/ltp-list

Quoted 5 lines · Gui Xiaohua

The child-process wait SIGUSR1 which would be sended by parent-process, if the child-process execute sigtimedwait() after parent-process send the signal, it would never receive the SIGUSR1 from parent-process. I cann't make sure the SIGUSR1 be sended after child-process execute sigtimedwait() with 100 percent, and i try my best.

Well, in theory I suppose this could happen, but you'd have to have a pretty bad scheduler if the parent can do a strcmp(buf, "c:go") between the pipe read and signal send, while the child goes straight from pipe write to sigtimedwait.

Have you seen this signal be missed? If not, then I'd rather assume things are reasonable. If you have seen this happen, then why not instead set up a SIGUSR1 handler in the child before doing the pipe write, then just sleep for 3 seconds instead of doing sigtimedwait?

Quoted 27 lines

Signed-off-by Gui Xiaohua <[email protected]>

kernel/containers/pidns/pidns12.c renamed from kernel/containers/pidns/pidns12-old.c +4 -2

@@ -109,7 +109,7 @@ int child_fn(void *arg)
 	}
 
 	/* Set timeout for sigtimedwait */
-	timeout.tv_sec = 3;
+	timeout.tv_sec = 10;
 	timeout.tv_nsec = 0;
 
 	/* Set mask to wait for SIGUSR1 signal */

@@ -182,6 +182,9 @@ int main(int argc, char *argv[])
 		cleanup();
 	}
 
+	/*Try best to make sure the SIGUSR1 be sended after child-process execute sigtimedwait*/
+	sleep(5);
+
 	/* Send SIGUSR1 to container init */
 	if (kill(cpid, SIGUSR1) == -1) {
 		tst_resm(TBROK, "parent: kill() failed(%s).", strerror(errno));
 
 
------------------------------------------------------------------------------

Crystal Reports - New Free Runtime and 30 Day Trial

Quoted 8 lines

Check out the new simplified licensing option that enables unlimited royalty-free distribution of the report engine for externally facing server and web deployment.
http://p.sf.net/sfu/businessobjects
_______________________________________________
Ltp-list mailing list [email protected]
https://lists.sourceforge.net/lists/listinfo/ltp-list

------------------------------------------------------------------------------
Crystal Reports - New Free Runtime and 30 Day Trial

Check out the new simplified licensing option that enables unlimited royalty-free distribution of the report engine for externally facing server and web deployment.
http://p.sf.net/sfu/businessobjects
_______________________________________________
Ltp-list mailing list [email protected]
https://lists.sourceforge.net/lists/listinfo/ltp-list

Quoted 12 lines · Serge E. Hallyn

Quoted 5 lines · Gui Xiaohua

The child-process wait SIGUSR1 which would be sended by parent-process, if the child-process execute sigtimedwait() after parent-process send the signal, it would never receive the SIGUSR1 from parent-process. I cann't make sure the SIGUSR1 be sended after child-process execute sigtimedwait() with 100 percent, and i try my best.

Well, in theory I suppose this could happen, but you'd have to have a pretty bad scheduler if the parent can do a strcmp(buf, "c:go") between the pipe read and signal send, while the child goes straight from pipe write to sigtimedwait.

Quoted 4 lines

Have you seen this signal be missed? If not, then I'd rather assume things are reasonable. If you have seen this happen, then why not instead set up a SIGUSR1 handler in the child before doing the pipe write, then just sleep for 3 seconds instead of doing sigtimedwait?

Thanks Serge.

Gui,

Are you planning any further patch(s) for this ?

Quoted 20 lines · Subrata Modak

Quoted 18 lines · Serge E. Hallyn

Quoted 6 lines · Gui Xiaohua

The child-process wait SIGUSR1 which would be sended by parent-process, if the child-process execute sigtimedwait() after parent-process send the signal, it would never receive the SIGUSR1 from parent-process. I cann't make sure the SIGUSR1 be sended after child-process execute sigtimedwait() with 100 percent, and i try my best.

Well, in theory I suppose this could happen, but you'd have to have a pretty bad scheduler if the parent can do a strcmp(buf, "c:go") between the pipe read and signal send, while the child goes straight from pipe write to sigtimedwait.

Have you seen this signal be missed? If not, then I'd rather assume things are reasonable. If you have seen this happen, then why not instead set up a SIGUSR1 handler in the child before doing the pipe write, then just sleep for 3 seconds instead of doing sigtimedwait?

Quoted 2 lines

Thanks Serge.

Or, if I'm being unreasonable, then at least have the parent only wait for at most 1 second, and leave the child alone. Every every ltp test is going to hang for 5 seconds...

But I prefer setting the signal handler ahead of time.

Quoted 22 lines · Serge E. Hallyn

Quoted 20 lines · Subrata Modak

Quoted 18 lines · Serge E. Hallyn

Quoted 6 lines · Gui Xiaohua

The child-process wait SIGUSR1 which would be sended by parent-process, if the child-process execute sigtimedwait() after parent-process send the signal, it would never receive the SIGUSR1 from parent-process. I cann't make sure the SIGUSR1 be sended after child-process execute sigtimedwait() with 100 percent, and i try my best.

Well, in theory I suppose this could happen, but you'd have to have a pretty bad scheduler if the parent can do a strcmp(buf, "c:go") between the pipe read and signal send, while the child goes straight from pipe write to sigtimedwait.

Have you seen this signal be missed? If not, then I'd rather assume things are reasonable. If you have seen this happen, then why not instead set up a SIGUSR1 handler in the child before doing the pipe write, then just sleep for 3 seconds instead of doing sigtimedwait?

Quoted 10 lines

Quoted 2 lines

Thanks Serge.

Or, if I'm being unreasonable, then at least have the parent only wait for at most 1 second, and leave the child alone. Every every ltp test is going to hang for 5 seconds...

But I prefer setting the signal handler ahead of time.

Sorry to relpy late.

As your last advice, i change the test case, and it work will.

[PATCH] fix sync problem between two processes of pidns12 test

The child-process wait SIGUSR1 which would be sended by parent-process, if the child-process execute sigtimedwait() after parent-process send the signal, it would never receive the SIGUSR1 from parent-process.

This patch rewrite child_signal_handler() to handle to SIGUSR1 and instead sigtimedwait() with sleep for 3 seconds.

Signed-off-by Gui Xiaohua <[email protected]>

---
 testcases/kernel/containers/pidns/pidns12.c |   35 ++++++--------------------
 1 files changed, 8 insertions(+), 27 deletions(-)

testcases/kernel/containers/pidns/pidns12.c +7 -21

index f893021..9120e82 100644
@@ -72,8 +72,13 @@ void cleanup()
  */
 static void child_signal_handler(int sig, siginfo_t *si, void *unused)
 {
-	/* sigtimedwait() traps siginfo details, so this wont be called */
-	tst_resm(TWARN, "cinit: control should have not reached here!");
+	/* Recieved SIGUSR1. Check sender pid */
+	if (si->si_pid == 0)
+		tst_resm(TPASS, "cinit: signalling PID (from other namespace)"\
+				" is 0 as expected");
+	else
+		tst_resm(TFAIL, "cinit: signalling PID (from other namespace)"\
+				" is not 0, but %d.", si->si_pid);
 }
 
 /*

@@ -82,9 +87,6 @@ static void child_signal_handler(int sig, siginfo_t *si, void *unused)
 int child_fn(void *arg)
 {
 	struct sigaction sa;
-	sigset_t newset;
-	siginfo_t info;
-	struct timespec timeout;
 	pid_t pid, ppid;
 
 	/* Set process id and parent pid */

@@ -108,34 +110,13 @@ int child_fn(void *arg)
 		cleanup();
 	}
 
-	/* Set timeout for sigtimedwait */
-	timeout.tv_sec = 3;
-	timeout.tv_nsec = 0;
-
-	/* Set mask to wait for SIGUSR1 signal */
-	sigemptyset(&newset);
-	sigaddset(&newset, SIGUSR1);
-
 	/* Let parent to signal SIGUSR1 */
 	if (write(pipefd[1], "c:go\0", 5) != 5) {
 		tst_resm(TBROK, "cinit: pipe is broken to write");
 		cleanup();
 	}
 
-	/* Wait for SIGUSR1 */
-	if (sigtimedwait(&newset, &info, &timeout) != SIGUSR1) {
-		tst_resm(TBROK, "cinit: sigtimedwait() failed(%s).",
-				strerror(errno));
-		cleanup();
-	}
-
-	/* Recieved SIGUSR1. Check sender pid */

Quoted 1 line

Signed-off-by Gui Xiaohua <[email protected]>

Serge,

Your view on this new solution ?

Quoted 5 lines · Subrata Modak

Quoted 1 line

Signed-off-by Gui Xiaohua <[email protected]>

Serge,

Your view on this new solution ?

Oops, i forgot to officially note:

Acked-by Serge Hallyn <[email protected]>

thanks,
-serge

Quoted 22 lines · Serge E. Hallyn

Quoted 20 lines · Subrata Modak

Quoted 19 lines · Serge E. Hallyn

Quoted 7 lines · Gui Xiaohua

The child-process wait SIGUSR1 which would be sended by parent-process, if the child-process execute sigtimedwait() after parent-process send the signal, it would never receive the SIGUSR1 from parent-process.
I cann't make sure the SIGUSR1 be sended after child-process execute sigtimedwait() with 100 percent, and i try my best.

Well, in theory I suppose this could happen, but you'd have to have a pretty bad scheduler if the parent can do a strcmp(buf, "c:go") between the pipe read and signal send, while the child goes straight from pipe write to sigtimedwait.

Have you seen this signal be missed? If not, then I'd rather assume things are reasonable. If you have seen this happen, then why not instead set up a SIGUSR1 handler in the child before doing the pipe write, then just sleep for 3 seconds instead of doing sigtimedwait?

Quoted 10 lines

Quoted 2 lines

Thanks Serge.

Or, if I'm being unreasonable, then at least have the parent only wait for at most 1 second, and leave the child alone. Every every ltp test is going to hang for 5 seconds...

But I prefer setting the signal handler ahead of time.

Sorry to relpy late.

As your last advice, i change the test case, and it work will.

[PATCH] fix sync problem between two processes of pidns12 test

The child-process wait SIGUSR1 which would be sended by parent-process, if the child-process execute sigtimedwait() after parent-process send the signal, it would never receive the SIGUSR1 from parent-process.

This patch rewrite child_signal_handler() to handle to SIGUSR1 and instead sigtimedwait() with sleep for 3 seconds.