net: vrf: Implement get_saddr for IPv6
authorDavid Ahern <dsa@cumulusnetworks.com>
Thu, 16 Jun 2016 23:24:25 +0000 (16:24 -0700)
committerDavid S. Miller <davem@davemloft.net>
Sat, 18 Jun 2016 04:25:29 +0000 (21:25 -0700)
IPv6 source address selection needs to consider the real egress route.
Similar to IPv4 implement a get_saddr6 method which is called if
source address has not been set.  The get_saddr6 method does a full
lookup which means pulling a route from the VRF FIB table and properly
considering linklocal/multicast destination addresses. Lookup failures
(eg., unreachable) then cause the source address selection to fail
which gets propagated back to the caller.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/vrf.c
include/net/l3mdev.h
net/ipv6/ip6_output.c
net/l3mdev/l3mdev.c

index 32173aa..b376282 100644 (file)
@@ -999,6 +999,46 @@ static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
 
        return dst;
 }
+
+/* called under rcu_read_lock */
+static int vrf_get_saddr6(struct net_device *dev, const struct sock *sk,
+                         struct flowi6 *fl6)
+{
+       struct net *net = dev_net(dev);
+       struct dst_entry *dst;
+       struct rt6_info *rt;
+       int err;
+
+       if (rt6_need_strict(&fl6->daddr)) {
+               rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif,
+                                         RT6_LOOKUP_F_IFACE);
+               if (unlikely(!rt))
+                       return 0;
+
+               dst = &rt->dst;
+       } else {
+               __u8 flags = fl6->flowi6_flags;
+
+               fl6->flowi6_flags |= FLOWI_FLAG_L3MDEV_SRC;
+               fl6->flowi6_flags |= FLOWI_FLAG_SKIP_NH_OIF;
+
+               dst = ip6_route_output(net, sk, fl6);
+               rt = (struct rt6_info *)dst;
+
+               fl6->flowi6_flags = flags;
+       }
+
+       err = dst->error;
+       if (!err) {
+               err = ip6_route_get_saddr(net, rt, &fl6->daddr,
+                                         sk ? inet6_sk(sk)->srcprefs : 0,
+                                         &fl6->saddr);
+       }
+
+       dst_release(dst);
+
+       return err;
+}
 #endif
 
 static const struct l3mdev_ops vrf_l3mdev_ops = {
@@ -1008,6 +1048,7 @@ static const struct l3mdev_ops vrf_l3mdev_ops = {
        .l3mdev_l3_rcv          = vrf_l3_rcv,
 #if IS_ENABLED(CONFIG_IPV6)
        .l3mdev_get_rt6_dst     = vrf_get_rt6_dst,
+       .l3mdev_get_saddr6      = vrf_get_saddr6,
 #endif
 };
 
index f8a416e..818fd4f 100644 (file)
@@ -39,6 +39,9 @@ struct l3mdev_ops {
        /* IPv6 ops */
        struct dst_entry * (*l3mdev_get_rt6_dst)(const struct net_device *dev,
                                                 struct flowi6 *fl6);
+       int                (*l3mdev_get_saddr6)(struct net_device *dev,
+                                               const struct sock *sk,
+                                               struct flowi6 *fl6);
 };
 
 #ifdef CONFIG_NET_L3_MASTER_DEV
@@ -140,6 +143,8 @@ static inline bool netif_index_is_l3_master(struct net *net, int ifindex)
 int l3mdev_get_saddr(struct net *net, int ifindex, struct flowi4 *fl4);
 
 struct dst_entry *l3mdev_get_rt6_dst(struct net *net, struct flowi6 *fl6);
+int l3mdev_get_saddr6(struct net *net, const struct sock *sk,
+                     struct flowi6 *fl6);
 
 static inline
 struct sk_buff *l3mdev_l3_rcv(struct sk_buff *skb, u16 proto)
@@ -230,6 +235,12 @@ struct dst_entry *l3mdev_get_rt6_dst(struct net *net, struct flowi6 *fl6)
        return NULL;
 }
 
+static inline int l3mdev_get_saddr6(struct net *net, const struct sock *sk,
+                                   struct flowi6 *fl6)
+{
+       return 0;
+}
+
 static inline
 struct sk_buff *l3mdev_ip_rcv(struct sk_buff *skb)
 {
index fd32175..1dfc402 100644 (file)
@@ -910,6 +910,13 @@ static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
        int err;
        int flags = 0;
 
+       if (ipv6_addr_any(&fl6->saddr) && fl6->flowi6_oif &&
+           (!*dst || !(*dst)->error)) {
+               err = l3mdev_get_saddr6(net, sk, fl6);
+               if (err)
+                       goto out_err;
+       }
+
        /* The correct way to handle this would be to do
         * ip6_route_get_saddr, and then ip6_route_output; however,
         * the route-specific preferred source forces the
@@ -999,10 +1006,11 @@ static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
        return 0;
 
 out_err_release:
-       if (err == -ENETUNREACH)
-               IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
        dst_release(*dst);
        *dst = NULL;
+out_err:
+       if (err == -ENETUNREACH)
+               IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
        return err;
 }
 
index d90e4ef..c4a1c3e 100644 (file)
@@ -162,6 +162,30 @@ int l3mdev_get_saddr(struct net *net, int ifindex, struct flowi4 *fl4)
 }
 EXPORT_SYMBOL_GPL(l3mdev_get_saddr);
 
+int l3mdev_get_saddr6(struct net *net, const struct sock *sk,
+                     struct flowi6 *fl6)
+{
+       struct net_device *dev;
+       int rc = 0;
+
+       if (fl6->flowi6_oif) {
+               rcu_read_lock();
+
+               dev = dev_get_by_index_rcu(net, fl6->flowi6_oif);
+               if (dev && netif_is_l3_slave(dev))
+                       dev = netdev_master_upper_dev_get_rcu(dev);
+
+               if (dev && netif_is_l3_master(dev) &&
+                   dev->l3mdev_ops->l3mdev_get_saddr6)
+                       rc = dev->l3mdev_ops->l3mdev_get_saddr6(dev, sk, fl6);
+
+               rcu_read_unlock();
+       }
+
+       return rc;
+}
+EXPORT_SYMBOL_GPL(l3mdev_get_saddr6);
+
 /**
  *     l3mdev_fib_rule_match - Determine if flowi references an
  *                             L3 master device