RE: [PATCH v2] rcu: fix a race in hlist_nulls_for_each_entry_rcu macro

[Eric Dumazet <eric.dumazet@gmail.com>]

On Wed, 2013-05-22 at 14:27 +0100, David Laight wrote:
> > So yes, the patch appears to fix the bug, but it sounds not logical to
> > me.
> 
> I was confused because the copy of the code I found was different
> (it has some checks for reusaddr - which force a function call in the
> loop).
> 
> The code being compiled is:
> 
> begin:
> 	result = NULL;
> 	badness = -1;
> 	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {

Here this loops begin by

 someptr = rcu_dereference(somelocation);

May claim is rcu_dereference() should force the compiler to read again
somelocation. Its done thanks to ACCESS_ONCE(). But apparently in the
specific case of &hslot->head, it doesnt work.

When using head directly (ie the same value), the ACCESS_ONCE() does
correctly its job.


> 		score = compute_score2(sk, net, saddr, sport,
> 				      daddr, hnum, dif);
> 		if (score > badness) {
> 			result = sk;
> 			badness = score;
> 			if (score == SCORE2_MAX)
> 				goto exact_match;
> 		}
> 	}
> 	/*
> 	 * if the nulls value we got at the end of this lookup is
> 	 * not the expected one, we must restart lookup.
> 	 * We probably met an item that was moved to another chain.
> 	 */
> 	if (get_nulls_value(node) != slot2)
> 		goto begin;
> 
> Which is entirely inlined - so the compiler is allowed to assume
> that no other code modifies any of the data.

You missed the rcu_dereference()/ACCESS_ONCE() we have in this loop.