Re: [path][rfc] add PR_DETACH prctl command

[Oleg Nesterov <oleg@redhat.com>]

On 04/08, Stas Sergeev wrote:
>
> I updated patch to fix the race between wait_task_detached()
> and wait_task_zombie() by using is_detaching flag.
> Here's the patch.
> What problems do remain here?

Stas, sorry, I simply have no time to read this patch. Not to mention,
you know that I dislike it in any case.

I only quickly looked at the changes in sys_prctl and they are definitely
wrong.

> +			me->real_parent = pid_ns->child_reaper;
> +			if (thread_group_leader(me)) {
> +				list_move_tail(&me->sibling,
> +						&me->real_parent->children);
> +				/* reparent threads */
> +				p = me;
> +				while_each_thread(me, p) {
> +					if (!task_ptrace(p))
> +						p->parent = pid_ns->child_reaper;
> +					p->real_parent = pid_ns->child_reaper;

Why do you check thread_group_leader() ???? What if the caller is not
the leader?

I told this many times, and I got lost. If you reparent, you should
always reparent the whole group.

Also, you are doing do_signal_parent(me). I do not know what it does
(once again, I didn't read the patch), but this looks suspicious.
A sub-thread shouldn't notify the parent.

Oleg.