Re: cn_queue.c

[Andrew Morton <akpm@osdl.org>]

Evgeniy Polyakov <johnpol@2ka.mipt.ru> wrote:
>
>  New object has 0 reference counter when created.
>  If some work is appointed to the object, then it's counter is atomically
>  incremented. It is decremented when the work is finished.
>  If object is supposed to be removed while some work
>  may be appointed to it, core ensures that no work _is_ appointed, 
>  and atomically disallows[for example removing workqueue, removing
>  callback, all with appropriate locks being hold] 
>  any other work appointment for the given object.
>  After it [when no work can be appointed to the object] if object
>  still has pending work [and thus has it's refcounter not zero], 
>  removing path waits untill appropriate refcnt hits zero. 
>  Since no _new_ work can be appointed at that level it is just
>  while (atomic_read(refcnt) != 0)
>    msleep();

More like:

	while (atomic_read(&obj->refcnt))
		msleep();
	kfree(obj);

which introduces the possibility of someone grabbing a new ref on the
object just before the kfree().  If there is no means by which any other
actor can acquire a ref to this object then OK, no race.

But it's rather surprising that such a thing can be achieved without any
locking.  What happens if another CPU has just entered
cn_queue_del_callback(), for example?  It has a live cn_callback_entry in
`cbq' which has a zero refcount - cn_queue_free_dev() can throw it away.

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