Attachment 155366 Details for Bug 240961 – Bugfix updated according to the Roland's comments.

[patch] Bugfix updated according to the Roland's comments.

strace-4.5.15-detach-zombie.patch (text/plain), 2.21 KB, created by Jan Kratochvil on 2007-05-24 17:05:25 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Jan Kratochvil

Created: 2007-05-24 17:05:25 UTC

Size: 2.21 KB

patch

obsolete

>2007-05-24  Jan Kratochvil  <jan.kratochvil@redhat.com>
>	    Roland McGrath  <roland@redhat.com>
>
>	* strace.c (detach): New prototype.
>	[LINUX] (detach): Extended function comment.
>	Call droptcb() instead of the wrong parametrized detach() call.
>	(handle_group_exit): Call droptcb() instead of the wrong parametrized
>	detach() call and instead of the inappropriate detach() call 
>
>diff -u -rup strace-4.5.15-orig/strace.c strace-4.5.15-detach-zombie/strace.c
>--- strace-4.5.15-orig/strace.c	2006-12-13 22:55:39.000000000 +0100
>+++ strace-4.5.15-detach-zombie/strace.c	2007-05-24 18:11:05.000000000 +0200
>@@ -83,6 +83,7 @@ unsigned int nprocs, tcbtabsize;
> char *progname;
> extern char **environ;
> 
>+static int detach P((struct tcb *tcp, int sig));
> static int trace P((void));
> static void cleanup P((void));
> static void interrupt P((int sig));
>@@ -1282,7 +1283,10 @@ struct tcb *tcp;
> 
> #endif /* !USE_PROCFS */
> 
>-/* detach traced process; continue with sig */
>+/* detach traced process; continue with sig
>+   Never call DETACH twice on the same process as both unattached and
>+   attached-unstopped processes give the same ESRCH.  For unattached process we
>+   would SIGSTOP it and wait for its SIGSTOP notification forever.  */
> 
> static int
> detach(tcp, sig)
>@@ -1460,8 +1464,10 @@ int sig;
> 	droptcb(tcp);
> 
> #ifdef LINUX
>-	if (zombie != NULL)
>-		error = detach(zombie) || error;
>+	if (zombie != NULL) {
>+		/* TCP no longer exists therefore you must not detach () it.  */
>+		droptcb(zombie);
>+	}
> #endif
> 
> 	return error;
>@@ -2035,7 +2041,8 @@ handle_group_exit(struct tcb *tcp, int s
> 			fprintf(stderr,
> 				"PANIC: handle_group_exit: %d leader %d\n",
> 				tcp->pid, leader ? leader->pid : -1);
>-		detach(tcp);	/* Already died.  */
>+		/* TCP no longer exists therefore you must not detach () it.  */
>+		droptcb(tcp);	/* Already died.  */
> 	}
> 	else {
> 		/* Mark that we are taking the process down.  */
>@@ -2060,7 +2067,9 @@ handle_group_exit(struct tcb *tcp, int s
> 						fprintf(stderr,
> 							" [%d exit %d kills %d]\n",
> 							tcp->pid, sig, leader->pid);
>-					detach(leader, sig);
>+					/* LEADER no longer exists therefore you
>+					   must not detach () it.  */
>+					droptcb(leader);
> 				}
> 				else
> 					leader->flags |= TCB_GROUP_EXITING;

Actions: View | Diff

Attachments on bug 240961: 155241 | 155245 | 155252 | 155366 | 155367 | 156669