RE: A (dumb?) waitpid(2) question

[usvyatsky, ilya]

Thanks a lot, I suspected something of this sort...

-----Original Message-----
From: Steven Rostedt [mailto:rostedt@xxxxxxxxxxx] 
Sent: Tuesday, December 21, 2004 9:38 PM
To: usvyatsky, ilya
Cc: LKML
Subject: Re: A (dumb?) waitpid(2) question


On Tue, 2004-12-21 at 19:26 -0500, usvyatsky, ilya wrote:
> As dumb as it seems, I am seing a weird behavior on my RH3.0 box 
> (2.4.21-20.Elsmp kernel).
> 
> It looks like (contrary to the man page and POSIX .1) waitpid(2) does 
> not return upon a SIGALRM.
> 
> I am porting an old piece of Solaris userland code (stripped from all 
> useful
> functionality):

I changed your code with the following:

@@ -25,11 +25,15 @@
     if (child_pid) {
         /* parent */
         int status = 0, timeout = 1;
+       struct sigaction sa;
+       sa.sa_handler = sig_handler;
+       sa.sa_flags = SA_NOMASK;

         do {
             printf("Parent: setting an alarm for %d seconds\n", timeout);

-            signal(SIGALRM, sig_handler);
+           sigaction(SIGALRM, &sa,NULL);
+//            signal(SIGALRM, sig_handler);
             alarm(timeout);

             printf("Parent: waiting for a child %d\n", child_pid);


And got the following result:

$ ./alarmit
Child:  sleeping for 6 seconds
Parent: setting an alarm for 1 seconds
Parent: waiting for a child 25926
Alarm!!!
Parent: wait was interrupted by a signalParent: setting an alarm for 1
seconds
Parent: waiting for a child -1
Alarm!!!
Parent: wait was interrupted by a signalParent: setting an alarm for 1
seconds
Parent: waiting for a child -1
Alarm!!!
Parent: wait was interrupted by a signalParent: setting an alarm for 1
seconds
Parent: waiting for a child -1
Alarm!!!
Parent: wait was interrupted by a signalParent: setting an alarm for 1
seconds
Parent: waiting for a child -1
Alarm!!!
Parent: wait was interrupted by a signalParent: setting an alarm for 1
seconds
Parent: waiting for a child -1
Child:  exiting
Parent: child 25926 terminated normally with status 0


> Is it a bug or a feature?

I guess it's a feature, and the default signal function must have SA_RESTART
set.

> Any suggestions would be greatly appreciated...
> 

Use sigaction instead of signal.


-- Steve

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/