Re: [patch] Re: Problem with exiting threads under NPTL

[Ingo Molnar]

here's yet another variant of the patch. Changes relative to yours:

- it sets zap_leader to 0 outside the critical section.

- it doesnt set leader to TASK_DEAD, exit_signal == -1 should really 
  protect it from sys_wait4() i believe.

- it sets 'p = leader' within the unlikely branch - slightly faster common
  case.

the patch testbooted fine here.

	Ingo

--- linux/kernel/exit.c.orig	
+++ linux/kernel/exit.c	
@@ -50,8 +50,11 @@ static void __unhash_process(struct task
 void release_task(struct task_struct * p)
 {
 	task_t *leader;
+	int zap_leader;
 	struct dentry *proc_dentry;
- 
+
+repeat:
+	zap_leader = 0;
 	BUG_ON(p->state < TASK_ZOMBIE);
  
 	atomic_dec(&p->user->processes);
@@ -72,8 +75,16 @@ void release_task(struct task_struct * p
 	 */
 	leader = p->group_leader;
 	if (leader != p && thread_group_empty(leader) &&
-		    leader->state == TASK_ZOMBIE && leader->exit_signal != -1)
+		    leader->state == TASK_ZOMBIE && leader->exit_signal != -1) {
 		do_notify_parent(leader, leader->exit_signal);
+		/*
+		 * If we were the last child thread and the leader has
+		 * exited already, and the leader's parent ignores SIGCHLD,
+		 * then we are the one who should release the leader.
+		 */
+		if (leader->exit_signal == -1)
+			zap_leader = 1;
+	}
 
 	p->parent->cutime += p->utime + p->cutime;
 	p->parent->cstime += p->stime + p->cstime;
@@ -88,6 +99,16 @@ void release_task(struct task_struct * p
 	proc_pid_flush(proc_dentry);
 	release_thread(p);
 	put_task_struct(p);
+
+	/*
+	 * Do this outside the tasklist lock. The reference to the
+	 * leader is safe. There's no recursion possible, since
+	 * the leader of the leader is itself:
+	 */
+	if (unlikely(zap_leader)) {
+		p = leader;
+		goto repeat;
+	}
 }
 
 /* we are using it only for SMP init */
-
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/