[PATCH] Make SO_BINDTODEVICE bind very tightly

[PATCH] Make SO_BINDTODEVICE bind very tightly

[Casey Carter]

[-- Attachment #1: Type: text/plain, Size: 497 bytes --]

In the presence of both sockets that are and sockets that are not bound 
to a specific interface, those bound to the interface should always be 
favored for packet delivery over those not so bound.  Attached patch 
(against 2.6.0-test6) enforces these semantics by a slight adjustment to 
udp_v4_lookup_longway in net/ipv4/udp.c  The consequence of this change 
is to make SO_BINDTODEVICE bind more tightly than simple bind().

-- 
Casey Carter
Casey@Carter.net
ccarter@cs.uiuc.edu
AIM: cartec69


[-- Attachment #2: udp.c.diff --]
[-- Type: text/plain, Size: 398 bytes --]

--- linux-2.6.0-test6/net/ipv4/udp.c	2003-09-27 19:50:10.000000000 -0500
+++ linux-2.6.0-test6-patch/net/ipv4/udp.c	2003-10-07 17:08:10.000000000 -0500
@@ -246,9 +246,9 @@
 			if (sk->sk_bound_dev_if) {
 				if (sk->sk_bound_dev_if != dif)
 					continue;
-				score+=2;
+				score+=8;
 			}
-			if(score == 9) {
+			if(score == 15) {
 				result = sk;
 				break;
 			} else if(score > badness) {

Re: [PATCH] Make SO_BINDTODEVICE bind very tightly

[David S. Miller]

On Tue, 07 Oct 2003 17:16:25 -0500
Casey Carter <Casey@Carter.net> wrote:

> In the presence of both sockets that are and sockets that are not bound 
> to a specific interface, those bound to the interface should always be 
> favored for packet delivery over those not so bound.

No, the fix made in the original posting of this thread is the
correct one.  One may not have some sockets with SO_BINDTODEVICE
and some with not bound at the same time to the same identity.
Therefore the fix should be a bind() time.