Re: [GIT PULL] sigqueue cache fix

[Ingo Molnar]

* Ingo Molnar <mingo@xxxxxxxxxx> wrote:

> 
> * Ingo Molnar <mingo@xxxxxxxxxx> wrote:
> 
> >  - Producer <-> consumer: this is the most interesting race, and I think 
> >    it's unsafe in theory, because the producer doesn't make sure that any 
> >    previous writes to the actual queue entry (struct sigqueue *q) have 
> >    reached storage before the new 'free' entry is advertised to consumers.
> > 
> >    So in principle CPU#0 could see a new sigqueue entry and use it, before 
> >    it's fully freed.
> > 
> >    In *practice* it's probably safe by accident (or by undocumented 
> >    intent), because there's an atomic op we have shortly before putting the 
> >    queue entry into the sigqueue_cache, in __sigqueue_free():
> > 
> >          if (atomic_dec_and_test(&q->user->sigpending))
> >                 free_uid(q->user);
> > 
> >    And atomic_dec_and_test() implies a full barrier - although I haven't 
> >    found the place where we document it and 
> >    Documentation/memory-ordering.txt is silent on it. We should probably 
> >    fix that too.
> > 
> > At minimum the patch adding the ->sigqueue_cache should include a 
> > well-documented race analysis firmly documenting the implicit barrier after 
> > the atomic_dec_and_test().
> 
> I just realized that even with that implicit full barrier it's not safe: 
> the producer uses q->user after the atomic_dec_and_test(). That access is 
> not serialized with the later write to ->sigqueue_cache - and another CPU 
> might see that entry and use the ->sigqueue_cache and corrupt q->user ...
> 
> So I think this code might have a real race on LL/SC platforms and we'll 
> need an smp_mb() in sigqueue_cache_or_free()?

pps. free_uid() happens to have an implicit barrier via 
refcount_dec_and_lock_irqsave():

  void free_uid(struct user_struct *up)
  {
        unsigned long flags;

        if (!up)
                return;

        if (refcount_dec_and_lock_irqsave(&up->__count, &uidhash_lock, &flags))
                free_user(up, flags);


So the q->user read in __sigqueue_free() appears to be implicitly 
serialized by free_uid() with the later write of 'q' to ->sigqueue_cache.

This needs to be robustly documented though.

Thanks,

	Ingo