ipv4: mpls: fix mpls_xmit for iptunnel

When using mpls over gre/gre6 setup, rt->rt_gw4 address is not set, the
same for rt->rt_gw_family.  Therefore, when rt->rt_gw_family is checked
in mpls_xmit(), neigh_xmit() call is skipped. As a result, such setup
doesn't work anymore.

This issue was found with LTP mpls03 tests.

Fixes: 1550c171 ("ipv4: Prepare rtable for IPv6 gateway")
Signed-off-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/mpls/mpls_iptunnel.c

@@ -133,12 +133,12 @@ static int mpls_xmit(struct sk_buff *skb)
 	mpls_stats_inc_outucastpkts(out_dev, skb);
 
 	if (rt) {
-		if (rt->rt_gw_family == AF_INET)
-			err = neigh_xmit(NEIGH_ARP_TABLE, out_dev, &rt->rt_gw4,
-					 skb);
-		else if (rt->rt_gw_family == AF_INET6)
+		if (rt->rt_gw_family == AF_INET6)
 			err = neigh_xmit(NEIGH_ND_TABLE, out_dev, &rt->rt_gw6,
 					 skb);
+		else
+			err = neigh_xmit(NEIGH_ARP_TABLE, out_dev, &rt->rt_gw4,
+					 skb);
 	} else if (rt6) {
 		if (ipv6_addr_v4mapped(&rt6->rt6i_gateway)) {
 			/* 6PE (RFC 4798) */