Re: [PATCH v2] exec: don't force_sigsegv processes with a pending fatal signal

From: Ivan Delalande
Date: Fri Feb 08 2019 - 19:16:41 EST

Hi Eric,

On Thu, Feb 07, 2019 at 11:13:59PM -0600, Eric W. Biederman wrote:
> I just noticed this. From my patch queue that I intend to send to
> Linus tomorrow. I think this change fixes your issue of getting
> the SIGSEGV instead of the already pending fatal signal.
> So I think this fixes your issue without any other code changes.
> Ivan can you verify that the patch below is enough?

I was having issues with just this patch applied on top of v5.0-rc5 or
the latest master: defunct processes accumulating, exiting processes
that would hang forever, and some kernel functions eating all the CPU
(setup_sigcontext, common_interrupt, __clear_user, do_signalâ).

But using your user-namespace.git/for-linus worked great and I've been
running my reproducer for a few hours now without issue. I'll probably
keep it running over the week-end as it has been unreliable at times,
but it looks promising so far.

A difference I've noticed with your tree (unrelated to my issue here but
that you may want to look at) is when I run my reproducer under
strace -f, I'm now getting quite a lot of "Exit of unknown pid 12345
ignored" warnings from strace, which I've never seen with mainline.
My reproducer simply fork-exec tail processes in a loop, and tries to
sigkill them in the parent with a variable delay.

Thank you,

> diff --git a/kernel/signal.c b/kernel/signal.c
> index 9ca8e5278c8e..5424cb0006bc 100644
> --- a/kernel/signal.c
> +++ b/kernel/signal.c
> @@ -2393,6 +2393,11 @@ bool get_signal(struct ksignal *ksig)
> goto relock;
> }
> + /* Has this task already been marked for death? */
> + ksig->info.si_signo = signr = SIGKILL;
> + if (signal_group_exit(signal))
> + goto fatal;
> +
> for (;;) {
> struct k_sigaction *ka;
> @@ -2488,6 +2493,7 @@ bool get_signal(struct ksignal *ksig)
> continue;
> }
> + fatal:
> spin_unlock_irq(&sighand->siglock);

Ivan Delalande
Arista Networks