Re: sk->sk_socket seems to disappear before connection termination

[Eric Dumazet <eric.dumazet@gmail.com>]

Le mercredi 10 novembre 2010 à 18:17 +0100, Jan Engelhardt a écrit :
> On Wednesday 2010-11-10 17:44, Eric Dumazet wrote:
> >
> >[ 9920.234680] ipt_LOG: sk=ffff880118bd32c0 sk->sk_socket=ffff88011d0d8c00 file=ffff88011cd4e100
> >[ 9920.234731] IN= OUT=eth1 SRC=192.168.20.108 DST=192.168.20.110 LEN=52 TOS=0x10 PREC=0x00 TTL=64 ID=63704 DF PROTO=TCP SPT=60088 DPT=22 WINDOW=35 RES=0x00 ACK FIN URGP=0 UID=0 GID=0 
> >[ 9920.235221] ipt_LOG: sk=ffff880078998000 sk->sk_socket=ffff880078c58300 file=          (null)
> >[ 9920.235271] IN= OUT=eth1 SRC=192.168.20.108 DST=192.168.20.110 LEN=52 TOS=0x00 PREC=0x00 TTL=64 ID=0 DF PROTO=TCP SPT=60088 DPT=22 WINDOW=35 RES=0x00 ACK URGP=0 
> >
> >You can see in my log, that the last packet seems to be from a different
> >socket ! (sk pointer changed to ffff880078998000 !)
> 
> Yes, that's it.
> 
> >Well well well, thats an ACK, in answer to FIN packet received from remote
> >side.
> 
> But why is it not handled by sk ffff880118bd32c0 anymore?
> It does have, after all, the same (addr,port) tuple.
> And it is sort of a hiccup for xt_owner users.

Its because of TIMEWAIT state : no more socket

We use a special tcp socket (net->ipv4.tcp_sock) in tcp_v4_send_ack()

Its yet another contention point on SMP machines :(