net: Introduce sk_route_nocaps
authorEric Dumazet <eric.dumazet@gmail.com>
Sun, 16 May 2010 07:36:33 +0000 (00:36 -0700)
committerDavid S. Miller <davem@davemloft.net>
Sun, 16 May 2010 07:36:33 +0000 (00:36 -0700)
commita465419b1febb603821f924805529cff89cafeed
tree5131fa2dbf624ebeb6cf61bf4dc1bc9464fe0bbd
parent3b098e2d7c693796cc4dffb07caa249fc0f70771
net: Introduce sk_route_nocaps

TCP-MD5 sessions have intermittent failures, when route cache is
invalidated. ip_queue_xmit() has to find a new route, calls
sk_setup_caps(sk, &rt->u.dst), destroying the

sk->sk_route_caps &= ~NETIF_F_GSO_MASK

that MD5 desperately try to make all over its way (from
tcp_transmit_skb() for example)

So we send few bad packets, and everything is fine when
tcp_transmit_skb() is called again for this socket.

Since ip_queue_xmit() is at a lower level than TCP-MD5, I chose to use a
socket field, sk_route_nocaps, containing bits to mask on sk_route_caps.

Reported-by: Bhaskar Dutta <bhaskie@gmail.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/sock.h
net/core/sock.c
net/ipv4/tcp_ipv4.c
net/ipv4/tcp_output.c
net/ipv6/tcp_ipv6.c