Re: [PATCH v2 5/7] clone4: Add a CLONE_AUTOREAP flag to automatically reap the child process

[Josh Triplett <josh-iaAMLnmF4UmaiuxdJuQwMA@public.gmane.org>]

On Sun, Mar 15, 2015 at 03:52:23PM +0100, Oleg Nesterov wrote:
> On 03/15, Josh Triplett wrote:
> > Add a CLONE_AUTOREAP flag to request this behavior unconditionally,
> 
> Yes, CLONE_AUTOREAP is much better. And I agree (mostly) with that
> we should rely on do_notify_parent().
> 
> Howver the patch still doesn't look right. First of all, ->autoreap
> should be per-process, not per-thread.

Ah, you're thinking of the case where the parent process launches a
child with CLONE_AUTOREAP, that child process launches siblings with
CLONE_THREAD and without CLONE_AUTOREAP, and one of those siblings is
the last to exit?  That seems easy enough to handle: instead of setting
->autoreap unconditionally in copy_process, I can set it only in the
non-CLONE_THREAD case, and otherwise let it inherit.  Then every task in
the group will have the same value for autoreap.

(As an aside, what *is* the use case for CLONE_PARENT without
CLONE_THREAD?)

> And there are ptrace/mt issues,
> it seems. Just for example, we should avoid EXIT_TRACE if autoreap in
> wait_task_zombie() even if we are going to re-notify parent.

I don't see how EXIT_TRACE can happen in wait_task_zombie if autoreap is
set.  wait_task_zombie does a cmpxchg with exit_state and doesn't
proceed unless exit_state was EXIT_ZOMBIE, and I don't see how we can
ever reach the EXIT_ZOMBIE state if autoreap.

> EXCEPT: do we really want SIGCHLD from the exiting child? I think we
> do not. I won't really argue though, but this should be discussed and
> documented. IIUC, with your patch it is still sent.

I think we do, yes.  The caller of clone can already specify what signal
they want, including no signal at all.  If they specify a signal
(SIGCHLD or otherwise) along with CLONE_AUTOREAP, we can send that
signal.  I don't think that causes any particular problem.

That's the same semantic you'd get if you have a SIGCHLD handler with
SA_NOCLDWAIT: you'd still get the signal, even though you don't need to
(and can't) wait on the child process.

> Josh, please give me some time to think and re-check, I'll write another
> email next week. I am not sure this is really needed, but it seems to
> me that we need the preparation patch to make this change clear/simple.

I'd appreciate any feedback you can offer on this series, including any
potential subtle interactions with ptrace.

- Josh Triplett