Re: Notifier chains are unsafe

[Joe Seigh]

Herbert Xu wrote:
> Andreas Kleen <ak@xxxxxxx> wrote:
>
> > Am Mi 26.10.2005 02:01 schrieb Chandra Seetharaman
> > <sekharan@xxxxxxxxxx>:
> >
> >
> > > > Better would be likely to use RCU.
> > >
> > > RCU will be a problem if the registered notifiers need to block.
> >
> > ?
> > Actually blocking should be ok, as long as the blocking notifier doesn't
> > unregister
> > itself. The current next pointer will be always reloaded after the
> > blocking.
>
>
> Blocking would be OK as long as you reference count the objects.


With or without RCU?  I.e., it's just a straighforward reference counting
solution and RCU is being used to allow incrementing the reference counts
safely.  Otherwise, without RCU you'd need a lock to safely increment the
reference counts.  Also note that with reference counting, the deletes of
the objects can occur any time a reference count is decremented.  So that
would include the notify_call threads as well.

--
Joe Seigh

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/