Re: [Patch net] bridge: do not expire mdb entry when bridge still uses it

[Cong Wang <amwang@redhat.com>]

On Fri, 2013-03-08 at 10:47 +0800, Herbert Xu wrote:
> On Fri, Mar 08, 2013 at 10:44:06AM +0800, Cong Wang wrote:
> >
> > This is a good question. It is due to inside br_handle_frame_finish() we
> > use 'skb2' to decide if we deliver packets to bridge itself, and in this
> > case 'skb2' is non-NULL only when:
> > 
> >                 if (mdst || BR_INPUT_SKB_CB_MROUTERS_ONLY(skb)) {
> >                         if ((mdst && mdst->mglist) ||
> >                             br_multicast_is_router(br))
> >                                 skb2 = skb; //        <======== HERE
> >                         br_multicast_forward(mdst, skb, skb2);
> > 
> > For other ports, br_multicast_forward() will always forward 'skb' to
> > them.
> 
> Yes but the point is that other ports should also expire after the
> timeout and thus be removed from mdst.  So why is this only happening
> for the bridge aka mdst->mglist?

The mdst may be removed, however the BR_INPUT_SKB_CB_MROUTERS_ONLY(skb)
is still set, because in br_multicast_ipv4_rcv():

        if (iph->protocol != IPPROTO_IGMP) {
                if ((iph->daddr & IGMP_LOCAL_GROUP_MASK) !=
IGMP_LOCAL_GROUP)
                        BR_INPUT_SKB_CB(skb)->mrouters_only = 1;
                return 0;
        }