[BUG] af_packet: loop at reception when using fanout
From: Eric Leblond <hidden>
Date: 2012-08-06 14:45:02
Subsystem:
networking [general], the rest · Maintainers:
"David S. Miller", Eric Dumazet, Jakub Kicinski, Paolo Abeni, Linus Torvalds
Hello, When using fanout mode on a AF_PACKET socket, the packet sent via one of the fanout socket are sent back to one of the socket in the fanout set. I've read the code and the error seems to be in dev.c. The code check if skb->sk is not equal to ptype->af_packet_priv but in the fanout case, we should check that skb->sk is not in the sockets of the fanout set. Here's how could look a possible implementation of the fix:
diff --git a/net/core/dev.c b/net/core/dev.c
index 1cb0d8a..63d144f 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c@@ -1654,7 +1654,12 @@ static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) */ if ((ptype->dev == dev || !ptype->dev) && (ptype->af_packet_priv == NULL || - (struct sock *)ptype->af_packet_priv != skb->sk)) { + (((struct sock *)ptype->af_packet_priv != skb->sk) + && + /* MISSING CHECK: af_packet_priv skb->sk one of the socket in fanout id of skb */ + ! packet_sk_in_fanout(ptype, skb->sk) + ) + ) { if (pt_prev) { deliver_skb(skb2, pt_prev, skb->dev); pt_prev = ptype;
If this is done, we will have to export one function from the af_packet module and I don't like the idea. If it is correct way to do it I can implement this and propose a patch. BR, -- Eric Leblond Blog: http://home.regit.org/ - Portfolio: http://regit.500px.com/
Attachments
- signature.asc [application/pgp-signature] 198 bytes