[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20070901114126.GA243@tv-sign.ru>
Date: Sat, 1 Sep 2007 15:41:26 +0400
From: Oleg Nesterov <oleg@...sign.ru>
To: Roland McGrath <roland@...hat.com>
Cc: Andrew Morton <akpm@...ux-foundation.org>,
linux-kernel@...r.kernel.org
Subject: Re: PATCH? fix SIGNAL_STOP_DEQUEUED vs SIGCONT race
On 09/01, Roland McGrath wrote:
>
> > However, this changes the behaviour when the task is ptraced. If the debugger
> > doesn't clear ->exit_code, SIGSTOP always succeeds after ptrace_stop(), even
> > if SIGCONT was sent in between. I can't decide whether this change is good
> > or bad, hopefully Roland can clarify.
>
> Hmm. I think this is bad.
>
> First, considering only the single-threaded case, there are debugger vs
> SIGCONT races. Someone does kill(pid,SIGSTOP);kill(pid,SIGCONT); while pid
> is debugged. The mandate for end user behavior here is that pid cannot
> wind up sitting in job control stop in the end. Say the debugger is
> e.g. strace, simply printing every signal and passing it through.
> So say it goes:
> T K D
> merrily running ... blocked in wait4
> kill(K, SIGSTOP)
> dequeue SIGSTOP
> -> ptrace_stop
> wait4 -> K,{SIGSTOP}
> kill(K, SIGCONT)
> PTRACE_CONT,K,SIGSTOP
> do_signal_stop(SIGSTOP)
> wait4 -> K,{SIGSTOP}
Thanks Roland.
Yes, that is what I was worrying about.
> It's still probably a worthwhile cleanup to have the logic only in
> get_signal_to_deliver, and to fix the problem you cited. It will take only
> a little extra code to handle the ptrace case too, i.e.
> if (sig_kernel_stop(signr) &&
> current->sighand->action[signr-1] == SIG_DFL &&
> !(current->signal->flags & SIGNAL_GROUP_EXIT))
> current->signal->flags |= SIGNAL_STOP_DEQUEUED;
> ptrace_stop(signr, signr, info);
> if (sig_kernel_stop(signr) && current->exit_code == signr &&
> !(current->signal->flags & SIGNAL_STOP_DEQUEUED) &&
> current->sighand->action[signr-1] == SIG_DFL)
> current->exit_code = 0;
Yes, I also thought about something like this, but tried to avoid because
it adds some complications. OTOH, this is not the fast path.
I'll try to think a bit more about this, and update the patch according
to your comments. Looks like we don't need to check SIGNAL_GROUP_EXIT
here, we are doing this later anyway.
Thanks!
Oleg.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists