UDP broadcast packets not looped back

[Benoit PAPILLAULT <benoit.papillault@free.fr>]

[-- Attachment #1: Type: text/plain, Size: 1596 bytes --]

Hi there,

I've already sent this email to lkml (see 
http://lkml.org/lkml/2007/8/13/1144) and got no feedback so far. I was 
told that netdev might be a more appropriate list, so here I am!

I wrote a small C program using the BSD socket API to send a UDP
broadcast packet to all interfaces on my machine. To do so, i sendto()
it to 255.255.255.255 after using setsockopt(SO_BINDTODEVICE) on the
correct interface. It works perfectly.

However, in the same program, either using the same socket or another on
the same UDP port, I found out that I just received packets i'm sending 
as well. And do not want it. I perfectly understand that some program 
might be interested in broadcast packets they sent, but i don't.

I have seen that in the multicast world, every program can choose this
behaviour throught the setsockopt IP_MULTICAST_LOOP. I do not want to
use multicast since I don't want my packets to be routed (if there was
ever a multicast router on the link). But I admit I could use a TTL=1
packet in this case.

Anyway, I dig five minutes into the kernel source code I have at hand
and wrote a small patch to add IP_MULTICAST_LOOP to broadcast packets.
Since it might be interesting to other people, I just submit it here. It
has been testing on Ubuntu and kernel 2.6.23-rc1 from the wireless.git
repository. It should be very easy to adapt to other kernel version.

I have not seen this feature as being standard throught the OpenGroup
specification
(http://www.opengroup.org/onlinepubs/009695399/functions/setsockopt.html),
but it might be usefull anyway.

Comments welcome,
Benoit

[-- Attachment #2: wireless-dev-udp-broadcast.diff --]
[-- Type: text/x-patch, Size: 805 bytes --]

diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index c9e2b5e..1052a6e 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -232,6 +232,9 @@ int ip_mc_output(struct sk_buff *skb)
 	skb->dev = dev;
 	skb->protocol = htons(ETH_P_IP);
 
+	printk("rt->rt_flags = %x mc_loop=%d\n",
+		rt->rt_flags, sk ? inet_sk(sk)->mc_loop : 0);
+
 	/*
 	 *	Multicasts are looped back for other local users
 	 */
@@ -266,10 +269,12 @@ int ip_mc_output(struct sk_buff *skb)
 	}
 
 	if (rt->rt_flags&RTCF_BROADCAST) {
+		if (!sk || inet_sk(sk)->mc_loop) {
 		struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 		if (newskb)
 			NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
 				newskb->dev, ip_dev_loopback_xmit);
+		}
 	}
 
 	return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, skb->dev,