Re: [PATCH] ip_tunnel: fix ip_tunnel_lookup

On Tue, 08 Jul 2014 15:12:10 -0700 (PDT)
David Miller [off-list ref] wrote:

From: Dmitry Popov <redacted>
Date: Sat, 5 Jul 2014 02:26:37 +0400

quoted

@@ -205,6 +207,8 @@ struct ip_tunnel *ip_tunnel_lookup(struct ip_tunnel_net *itn,
 
 	hlist_for_each_entry_rcu(t, head, hash_node) {
 		if (t->parms.i_key != key ||
+		    t->parms.iph.saddr != 0 ||
+		    t->parms.iph.daddr != 0 ||
 		    !(t->dev->flags & IFF_UP))
 			continue;
 

I don't really understand the logic of these tests.

Usually the canonical way to test these kinds of things is:

	if (parms->saddr && parms->saddr != saddr)
		goto no_match;

But you are signalling a non-match any time the address is not a
wildcard.

Why?

Because that's exactly what I want: to skip any non-wildcard tunnels.

How I see ip_tunnel_lookup logic:
1) try to find exact match (and if found return this tunnel):
tunnel.saddr == iph.daddr && tunnel.daddr == iph.saddr && key_matched()
2) try to find matched (local) wildcard tunnel:
tunnel.saddr == any && tunnel.daddr == iph.saddr && key_matched()
3) try to find matched (remote) wildcard tunnel:
tunnel.saddr == iph.daddr && tunnel.daddr == any && key_matched()
(there is also a test for multicast tunnel, but let's skip it for simplicity)
4) try to find matched (full) wildcard tunnel:
tunnel.saddr == any && tunnel.daddr == any && key_matched()
5) if nothing found return default tunnel.

According to this logic, in 4th loop (the one you quoted) we have to test that
tunnel.daddr == any && tunnel.saddr == any. In my opinion those two new lines
are the best way to achieve it.