Merge branch 'rt6i_gateway'

Julian Anastasov says:

====================
ipv6: use rt6i_gateway as nexthop

	The following patchset makes sure that rt6i_gateway
contains valid nexthop information in all cases, so that
we can use different nexthop for sending.

	The first patch is a simple fix that makes IPVS, TEE,
RAW(hdrincl) and RTF_DYNAMIC(without RTF_GATEWAY) work as
before 3.9. There is a single corner case not solved by
this patch: RAW(hdrincl) or TEE using local address for
nexthop, a silly feature, I guess. In this case we
see zeroes in rt6i_gateway because we get route that is not
cloned. This is solved only with patch 2.

	The second patch is an optimization that makes sure
all resulting routes have rt6i_gateway filled, so that we
can avoid the complex ipv6_addr_any() call added to rt6_nexthop()
by patch 1. And it sets rt6i_gateway for local routes, a case
not handled by patch 1.

	The third patch uses the new rt6_nexthop() function to fix
the matching of gateways in the same way as commit bbb5823cf742a7
("netfilter: nf_conntrack: fix rt_gateway checks for H.323 helper")
fixes nf_conntrack_h323_main.c for IPv4. Currently, it depends on
the new definition of rt6_nexthop() in patch 2. Actually, if
patch 2 is applied, patch 3 becomes a cosmetic change.

	I see the following two alternatives for applying these
patches:

1. Linger patch 2 in net-next to avoid surprises in the upcoming
release. In this case patch 3 can be reworked not to depend on
the new rt6_nexthop() definition in patch 2. I guess this is a
better option, so that patch 2 can be reviewed and tested for
longer time.

2. Include all 3 patches in net tree - more risky because this
is my first attempt to change IPv6.

	Here is the situation as handled by patch 2:

	In IPv6 the resolved routes are always host routes (/128
with DST_HOST), mostly cloned ones. We allow routes in FIB
to contain rt6i_gateway with zeroes (eg. for local subnets) but
on cloning we can fill the rt6i_gateway field in result.
This works even without this patchset.

	There is a single special case where dst is provided as
skb_dst directly without a routing call: icmp6_dst_alloc(). It is a
private dst allocated just for the particular ICMP packet. Patch 2
fills rt6i_gateway in this case, needed for the new rt6_nexthop()
simplification.

	The last case is addrconf_dst_alloc(), it can put in
FIB local/anycast routes when addresses are added. Patch 2
needs to fill rt6i_gateway in this case because such routes
are returned without cloning.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 2 insertions, 2 deletions

diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index 975624b8d2ea..91fb4e8212f5 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c
@@ -105,7 +105,7 @@ static int ip6_finish_output2(struct sk_buff *skb)
         }
 
         rcu_read_lock_bh();
-        nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
+        nexthop = rt6_nexthop((struct rt6_info *)dst);
         neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
         if (unlikely(!neigh))
                 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
@@ -874,7 +874,7 @@ static int ip6_dst_lookup_tail(struct sock *sk,
          */
         rt = (struct rt6_info *) *dst;
         rcu_read_lock_bh();
-        n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt, &fl6->daddr));
+        n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
         err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
         rcu_read_unlock_bh();