Re: [PATCH] fix de_thread() vs do_coredump() deadlock

[Oleg Nesterov]

On 04/09, Roland McGrath wrote:
>
>                                                                  Now, the
> coredump case matches the non-coredump fatal signal: the signal is dropped
> on the floor.

A fatal signal is placed to ->shared_pending in any (non tkill) case, so I
think it is not lost (but may be unnoticed for a while).

sig_kernel_coredump() is different. It could be stealed by one of sub-threads
while another one does de_thread(), that is why it could be lost.

What do you think about something like this untested patch instead?
I am far from sure it is correct, I need a sleep ...

Oleg.

--- fs/exec.c~	2006-04-10 22:15:06.000000000 +0400
+++ fs/exec.c	2006-04-11 00:56:52.000000000 +0400
@@ -657,6 +657,7 @@ static int de_thread(struct task_struct 
 	}
 	sig->group_exit_task = NULL;
 	sig->notify_count = 0;
+	recalc_sigpending();
 	spin_unlock_irq(lock);
 
 	/*
@@ -1478,9 +1479,15 @@ int do_coredump(long signr, int exit_cod
 	}
 	mm->dumpable = 0;
 
-	retval = -EAGAIN;
 	spin_lock_irq(&current->sighand->siglock);
-	if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
+	if (current->signal->flags & SIGNAL_GROUP_EXIT) {
+		// Re-add the signal. This does not matter
+		// if we are doing do_group_exit().
+		// If it was de_thread(), this signal will be
+		// received again after sys_exec() succeeds.
+		sigaddset(&current->signal->shared_pending.signal, signr);
+		retval = -EAGAIN;
+	} else {
 		current->signal->flags = SIGNAL_GROUP_EXIT;
 		current->signal->group_exit_code = exit_code;
 		current->signal->group_stop_count = 0;

-
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/