Re: [PATCH] fix sigqueue accounting for posix-timers broken by new RLIMIT_SIGPENDING tracking code

[Chris Wright <chrisw@osdl.org>]

* Chris Wright (chrisw@osdl.org) wrote:
> * Roland McGrath (roland@redhat.com) wrote:
> > 
> > The introduction of RLIMIT_SIGPENDING and the associated tracking code was
> > broken for the case of preallocated sigqueue elements, i.e. posix-timers.
> > It wrongly includes the timer's preallocated sigqueue structs in the count
> > towards the per-user when allocating them, but (rightly) does not decrement
> > the count when they are freed.  
> 
> Are you sure?  IOW, are you seeing a leak of the count?  The sigqueue
> structure has a lifetime the same as the timer, IIRC.  So each time the
> signal is sent/received there's no accounting, because it's reused.
> But timer create/delete does sigqueue_alloc -> __sigqueue_alloc which
> does the inc. and sigqueue_free -> __sigqueue_free which does the dec.
> I'm fairly sure I had tested this.

Yeah, I re-read the code and re-ran my tests just now to verify.  Current
code is correct.  In fact, this patch is wrong.  It unbalances the inc/dec
such that sigqueue_alloc() won't inc, yet final sigqueue_free (at timer
destruction) will still dec.  Please don't apply this patch.

thanks,
-chris
-- 
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net