This will also improve handling of ipv6 tcp socket request
backlog when syncookies are not enabled.  When backlog
becomes very deep, last quarter of backlog is limited to
validated destinations.  Previously only ipv4 implemented
this logic, but now ipv6 does too.

Now we are only one step away from enabling timewait
recycling for ipv6, and that step is simply filling in
the implementation of tcp_v6_get_peer() and
tcp_v6_tw_get_peer().
Signed-off-by: David S. Miller <davem@davemloft.net>