Re: 2.6.14-rc1 wait()/SIG_CHILD bevahiour

[Badari Pulavarty <pbadari@us.ibm.com>]

On Mon, 2005-09-19 at 14:01 -0400, Kyle Moffett wrote:
> On Sep 19, 2005, at 13:39:33, Badari Pulavarty wrote:
> > Hi,
> >
> > I am looking at a problem where the parent process doesn't seem to  
> > cleanup the exited children (with a webserver). We narrowed it down  
> > to a simple testcase. Seems more like a lost SIG_CHILD.
> 
> You don't get one SIG_CHLD per child that quits.  The kernel may and  
> probably will merge SIG_CHLD signals together if it has several  
> queued before it gets a chance to deliver them to your process.  This  
> is true of _all_ signals.  If you "kill -STOP 1234", then "kill -QUIT  
> 1234", "kill -QUIT 1234", "kill -QUIT 1234", "kill -CONT 1234", the  
> PID 1234 will have 3 signals delivered:  The original untrappable  
> SIGSTOP, the SIGCONT that causes it to resume, and a single SIGQUIT  

True.

> immediately following it.  The correct and portable way to handle  
> this is to put a loop in your SIGCHLD signal handler:
> 
> #include <sys/types.h>
> #include <sys/wait.h>
> 
> void sigchld_handler(int signal) {
>      pid_t pid;
>      int status;
> 
>      while( -1 != (child = waitpid(-1, &status, WNOHANG)) ) {
>          /*
>           * Now "status" is the exit status of the child and
>           * "pid" is its pid.  See the waitpid() manpage for
>           * macros you can use to get information from the
>           * status variable.
>           */
>          do_some_processing_of_exited_child(pid,status);
>      }
> }


Thats what I already do, in my current testcase.

    while(1) {
        pid = wait4(-1, &status, WNOHANG, &ru);
        if (pid <= 0) break;
//        printf("SIGCHLD received for pid %d\n", pid);
        pid_exited(pid);
    }


Thanks,
Badari

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