]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commit
ipv4: set transport header earlier
authorEric Dumazet <edumazet@google.com>
Tue, 16 Jul 2013 03:03:19 +0000 (20:03 -0700)
committerDavid S. Miller <davem@davemloft.net>
Tue, 16 Jul 2013 19:59:28 +0000 (12:59 -0700)
commit21d1196a35f5686c4323e42a62fdb4b23b0ab4a3
tree559c3cf32d2da124090fd6d256948d0d59eeb7e0
parent9a0f06fee1f453370e87b6568dc1d39d676f53fc
ipv4: set transport header earlier

commit 45f00f99d6e ("ipv4: tcp: clean up tcp_v4_early_demux()") added a
performance regression for non GRO traffic, basically disabling
IP early demux.

IPv6 stack resets transport header in ip6_rcv() before calling
IP early demux in ip6_rcv_finish(), while IPv4 does this only in
ip_local_deliver_finish(), _after_ IP early demux.

GRO traffic happened to enable IP early demux because transport header
is also set in inet_gro_receive()

Instead of reverting the faulty commit, we can make IPv4/IPv6 behave the
same : transport_header should be set in ip_rcv() instead of
ip_local_deliver_finish()

ip_local_deliver_finish() can also use skb_network_header_len() which is
faster than ip_hdrlen()

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/ip_input.c