ipv6: use a stronger hash for tcp
authorEric Dumazet <edumazet@google.com>
Thu, 21 Feb 2013 12:18:52 +0000 (12:18 +0000)
committerBen Hutchings <ben@decadent.org.uk>
Wed, 6 Mar 2013 03:24:21 +0000 (03:24 +0000)
commit0fd0ff7e1fcc4b4bc5d17ab1d200f23dea7c681d
treeb0633b26fe93a7d93385ce2744fcb19cecda82cd
parent52430c06469c05c36dd688c8daff25e5bcfde8e9
ipv6: use a stronger hash for tcp

[ Upstream commit 08dcdbf6a7b9d14c2302c5bd0c5390ddf122f664 ]

It looks like its possible to open thousands of TCP IPv6
sessions on a server, all landing in a single slot of TCP hash
table. Incoming packets have to lookup sockets in a very
long list.

We should hash all bits from foreign IPv6 addresses, using
a salt and hash mix, not a simple XOR.

inet6_ehashfn() can also separately use the ports, instead
of xoring them.

Reported-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
include/net/inet6_hashtables.h
include/net/inet_sock.h
include/net/ipv6.h
net/ipv4/af_inet.c