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

[Stas Sergeev <stsp@aknet.ru>]

05.04.2011 19:15, Oleg Nesterov wrote:
> I don't understand the "and not is a child on init's subthread".
> If the child was created by init's sub-thread, it is the child
> of the whole thread group.
>
> ->real_parent points to thread, yes.
OK, sorry, I thought you mean the CLONE_PARENT case, where
the real_parent points _not_ to thread. That's why the confusion.

>   But the parent is the whole
> process, not thread. The only reason for this oddity is __WNOTHREAD.
OK, now I see what problem you are pointing to.
Will fix.

>> How? I think I prevented that with this:
>> ---
>>
>> +			p->detaching = 0;
>> +			continue;
> Yes, thanks, I didn't notice "continue". But then this is wrong again.
> This can race with wait_task_detached() called by our sub-thread, it
> can clear ->detaching before we check it.
But the above code is under write_lock_irq(&tasklist_lock), so why
would it race?