I think this possibly explains why I ended up routing ipv6 everywhere rather than bridging it.
---------- Forwarded message ---------- From: Chuck Anderson <[email protected]> Date: Tue, Sep 13, 2011 at 1:00 PM Subject: bridge should flood non-IPv4-multicast ethernet frames To: [email protected] When the bridge code grew multicast snooping capability (currently IPv4/IGMPv2-only as I understand it), it stopped flooding non-IPv4 multicast ethernet frames. This breaks the capability to bridge any non-IPv4 protocols that also use multicast ethernet frames, such as IPv6 and IS-IS, while the bridge snooping capability remains enabled (it appears to be default enabled at least in the RHEL 6 vendor kernel). I noticed this when IPv6 neighbor discovery (ND) and router advertisement (RA) packets weren't making it to a KVM guest via br0 on the host, breaking IPv6 connectivity to the guest. This type of thing is a common bug with vendor's multicast snooping implementations, but I was surprised to discover that Linux has this same bug. See RFC 4541, section 1, last paragraph, and section 2.1.2, paragraph 4: http://tools.ietf.org/html/rfc4541.html I believe the relevent code is in br_device.c: if (is_broadcast_ether_addr(dest)) br_flood_deliver(br, skb); else if (is_multicast_ether_addr(dest)) { if (unlikely(netpoll_tx_running(dev))) { br_flood_deliver(br, skb); goto out; } if (br_multicast_rcv(br, NULL, skb)) { kfree_skb(skb); goto out; } mdst = br_mdb_get(br, skb); if (mdst || BR_INPUT_SKB_CB_MROUTERS_ONLY(skb)) br_multicast_deliver(mdst, skb); else br_flood_deliver(br, skb); } else if ((dst = __br_fdb_get(br, dest)) != NULL) br_deliver(dst->dst, skb); else br_flood_deliver(br, skb); is_multicast_ether_addr() only checks to see if the lowest bit is 1--i.e. any multicast ethernet address. That check alone isn't sufficient. There also needs to be a check that the ethernet frame is in one of the well-known formats for the particular protocol for which snooping is supported, IPv4 being the only one supported by Linux bridging so far. I see a few ways to fix this: 1. IPv4 Multicast always uses multicast ethernet addresses in the format 01:00:5E:xx:xx:xx. Insert a check that the dest address matches 01:00:5E:xx:xx:xx, otherwise always flood the frame so we don't break non-IPv4-multicast frames from being bridged. Something like this pseudocode: if (is_broadcast_ether_addr(dest)) br_flood_deliver(br, skb); else if (is_ipv4_multicast_ether_addr(dest)) { ... static inline int is_ipv4_multicast_ether_addr(const u8 *addr) { return (addr[0] == 0x01 && addr[1] == 0x00 && addr[2] == 0x5e); } 2. Check that the Ethertype is 0x800 (IPv4), and if it is not, always flood the frame so we don't break non-IPv6-multicast frames being bridged. 3. Do both of the above, the key point being that IPv6 multicast frames (33:33:xx:xx:xx:xx), along with any other ethernet multicast frames that aren't supported by the current bridge snooping code, should always be flooded unconditionally. IS-IS for example uses 01:80:C2:00:00:14 and 01:80:C2:00:00:15. Thoughts? -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html -- Dave Täht SKYPE: davetaht US Tel: 1-239-829-5608 http://the-edge.blogspot.com _______________________________________________ Bloat mailing list [email protected] https://lists.bufferbloat.net/listinfo/bloat
