Re: [PATCH 3/3] posix timers: use SIGQUEUE_CANCELLED when the timer is destroyed

[Oleg Nesterov <oleg@tv-sign.ru>]

On 05/18, Linus Torvalds wrote:
> 
> On Sun, 18 May 2008, Oleg Nesterov wrote:
> > 
> > Initially, I did
> > 
> > 	q->flags |= SIGQUEUE_CANCELLED;
> > 	spin_lock_irqsave(lock, flags);
> > 	q->flags &= ~SIGQUEUE_PREALLOC;
> > 
> > to document the fact that SIGQUEUE_CANCELLED can be set lockless, but
> > then "optimized" the code, couldn't help myself... Besides, the code
> > above looks really confusing without the fat comment.
> 
> Oh, and the above is *wrong*.
> 
> Why?
> 
> Becayse if SIGQUEUE_PREALLOC setting needs the lock, then setting any 
> *other* bit in that word will also need the lock!
> 
> That's because
> 
> 	q->flags |= SIGQUEUE_CANCELLED;
> 
> writes those other bits too - admittedly with the value they were read 
> just before, but if it races with something setting SIGQUEUE_PREALLOC that 
> doesn't matter - the newly written version will simply be wrong.
> 
> So the rule is that if one bit of a word needs locking, then they *all* 
> do.

Ah. I wasn't clear.

Clearing of SIGQUEUE_PREALLOC needs ->siglock, yes. But not because anybody
else can write to q->flags. Nobody can, we (the timer) "own" this sigqueue.

Once we clear SIGQUEUE_PREALLOC, "q" can be freed by the receiver (it doesn't
writes to q->flags, it only reads ->flags). After that we can't trust the
list_empty() check, we just can't dereference this "struct sigqueue *".

Taking ->siglock before "&= ~SIGQUEUE_PREALLOC" ensures that "q" can't be
be freed if it is queued, nothing more.

Oleg.