Hi Jan, Jan Kasprzak wrote: > Hello, netfilter developers! > > I am trying to delploy a conntrackd-based HA router and (if possible) a > flow statistics collector. I have discovered that nf_conntrack treats each > ICMP echo request/reply pair as a separate connection (in > net/ipv4/netfilter/nf_conntrack_proto_icmp.c:icmp_packet() function). > This has several problems: > > - excessive conntrackd traffic when the ping is running over the router > (one new "connection" per echo request/reply pair). Indeed, one event per new ICMP echo packet is too much. This can be also a way to exhaust resources of the conntrack subsystem. > - should there be a duplicated ICMP echo reply (such as when when pinging > a cluster IP address), only the first echo reply is seen as > ESTABLISHED, the rest is INVALID. > > - no "per-flow" statistics available, as there is no notion of the "flow" at all. Indeed, no packet accounting at all. > I think it would be better to keep the default timeout of > nf_ct_icmp_timeout even after the echo reply is received. Feel free > to correct me why early deleting of ICMP conntrack entries is needed, > or consider applying the following patch. The only problem that I see is that you patch relaxes the current checking that we're doing. I mean, for every packet in one direction we only accept one ICMP reply packet. With your patch, we can accept more than one packet in the reply direction. I have a similar patch in my internal tree. It allows only one reply packet per original and many packet in the original direction as you want. I'll post it asap so we can discuss on it. -- "Los honestos son inadaptados sociales" -- Les Luthiers -- To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
