net/mlx5e: Support SRIOV TC encapsulation offloads for IPv6 tunnels
authorOr Gerlitz <ogerlitz@mellanox.com>
Sun, 11 Dec 2016 19:28:28 +0000 (21:28 +0200)
committerSaeed Mahameed <saeedm@mellanox.com>
Tue, 24 Jan 2017 19:14:02 +0000 (21:14 +0200)
Add the missing parts for offloading IPv6 tunnels. This includes
route and neigh lookups and construnction of the IPv6 tunnel headers.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c

index 3d0dbfb..640f10f 100644 (file)
@@ -718,6 +718,47 @@ static int mlx5e_route_lookup_ipv4(struct mlx5e_priv *priv,
        return 0;
 }
 
+static int mlx5e_route_lookup_ipv6(struct mlx5e_priv *priv,
+                                  struct net_device *mirred_dev,
+                                  struct net_device **out_dev,
+                                  struct flowi6 *fl6,
+                                  struct neighbour **out_n,
+                                  int *out_ttl)
+{
+       struct neighbour *n = NULL;
+       struct dst_entry *dst;
+
+#if IS_ENABLED(CONFIG_INET) && IS_ENABLED(CONFIG_IPV6)
+       struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
+       int ret;
+
+       dst = ip6_route_output(dev_net(mirred_dev), NULL, fl6);
+       if (dst->error) {
+               ret = dst->error;
+               dst_release(dst);
+               return ret;
+       }
+
+       *out_ttl = ip6_dst_hoplimit(dst);
+
+       /* if the egress device isn't on the same HW e-switch, we use the uplink */
+       if (!switchdev_port_same_parent_id(priv->netdev, dst->dev))
+               *out_dev = mlx5_eswitch_get_uplink_netdev(esw);
+       else
+               *out_dev = dst->dev;
+#else
+       return -EOPNOTSUPP;
+#endif
+
+       n = dst_neigh_lookup(dst, &fl6->daddr);
+       dst_release(dst);
+       if (!n)
+               return -ENOMEM;
+
+       *out_n = n;
+       return 0;
+}
+
 static int gen_vxlan_header_ipv4(struct net_device *out_dev,
                                 char buf[],
                                 unsigned char h_dest[ETH_ALEN],
@@ -754,6 +795,41 @@ static int gen_vxlan_header_ipv4(struct net_device *out_dev,
        return encap_size;
 }
 
+static int gen_vxlan_header_ipv6(struct net_device *out_dev,
+                                char buf[],
+                                unsigned char h_dest[ETH_ALEN],
+                                int ttl,
+                                struct in6_addr *daddr,
+                                struct in6_addr *saddr,
+                                __be16 udp_dst_port,
+                                __be32 vx_vni)
+{
+       int encap_size = VXLAN_HLEN + sizeof(struct ipv6hdr) + ETH_HLEN;
+       struct ethhdr *eth = (struct ethhdr *)buf;
+       struct ipv6hdr *ip6h = (struct ipv6hdr *)((char *)eth + sizeof(struct ethhdr));
+       struct udphdr *udp = (struct udphdr *)((char *)ip6h + sizeof(struct ipv6hdr));
+       struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
+
+       memset(buf, 0, encap_size);
+
+       ether_addr_copy(eth->h_dest, h_dest);
+       ether_addr_copy(eth->h_source, out_dev->dev_addr);
+       eth->h_proto = htons(ETH_P_IPV6);
+
+       ip6_flow_hdr(ip6h, 0, 0);
+       /* the HW fills up ipv6 payload len */
+       ip6h->nexthdr     = IPPROTO_UDP;
+       ip6h->hop_limit   = ttl;
+       ip6h->daddr       = *daddr;
+       ip6h->saddr       = *saddr;
+
+       udp->dest = udp_dst_port;
+       vxh->vx_flags = VXLAN_HF_VNI;
+       vxh->vx_vni = vxlan_vni_field(vx_vni);
+
+       return encap_size;
+}
+
 static int mlx5e_create_encap_header_ipv4(struct mlx5e_priv *priv,
                                          struct net_device *mirred_dev,
                                          struct mlx5_encap_entry *e,
@@ -821,6 +897,75 @@ out:
        return err;
 }
 
+static int mlx5e_create_encap_header_ipv6(struct mlx5e_priv *priv,
+                                         struct net_device *mirred_dev,
+                                         struct mlx5_encap_entry *e,
+                                         struct net_device **out_dev)
+
+{
+       int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
+       struct ip_tunnel_key *tun_key = &e->tun_info.key;
+       int encap_size, err, ttl = 0;
+       struct neighbour *n = NULL;
+       struct flowi6 fl6 = {};
+       char *encap_header;
+
+       encap_header = kzalloc(max_encap_size, GFP_KERNEL);
+       if (!encap_header)
+               return -ENOMEM;
+
+       switch (e->tunnel_type) {
+       case MLX5_HEADER_TYPE_VXLAN:
+               fl6.flowi6_proto = IPPROTO_UDP;
+               fl6.fl6_dport = tun_key->tp_dst;
+               break;
+       default:
+               err = -EOPNOTSUPP;
+               goto out;
+       }
+
+       fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tun_key->tos), tun_key->label);
+       fl6.daddr = tun_key->u.ipv6.dst;
+       fl6.saddr = tun_key->u.ipv6.src;
+
+       err = mlx5e_route_lookup_ipv6(priv, mirred_dev, out_dev,
+                                     &fl6, &n, &ttl);
+       if (err)
+               goto out;
+
+       if (!(n->nud_state & NUD_VALID)) {
+               pr_warn("%s: can't offload, neighbour to %pI6 invalid\n", __func__, &fl6.daddr);
+               err = -EOPNOTSUPP;
+               goto out;
+       }
+
+       e->n = n;
+       e->out_dev = *out_dev;
+
+       neigh_ha_snapshot(e->h_dest, n, *out_dev);
+
+       switch (e->tunnel_type) {
+       case MLX5_HEADER_TYPE_VXLAN:
+               encap_size = gen_vxlan_header_ipv6(*out_dev, encap_header,
+                                                  e->h_dest, ttl,
+                                                  &fl6.daddr,
+                                                  &fl6.saddr, tun_key->tp_dst,
+                                                  tunnel_id_to_key32(tun_key->tun_id));
+               break;
+       default:
+               err = -EOPNOTSUPP;
+               goto out;
+       }
+
+       err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
+                              encap_size, encap_header, &e->encap_id);
+out:
+       if (err && n)
+               neigh_release(n);
+       kfree(encap_header);
+       return err;
+}
+
 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
                              struct ip_tunnel_info *tun_info,
                              struct net_device *mirred_dev,
@@ -831,7 +976,7 @@ static int mlx5e_attach_encap(struct mlx5e_priv *priv,
        struct ip_tunnel_key *key = &tun_info->key;
        struct mlx5_encap_entry *e;
        struct net_device *out_dev;
-       int tunnel_type, err;
+       int tunnel_type, err = -EOPNOTSUPP;
        uintptr_t hash_key;
        bool found = false;
 
@@ -856,12 +1001,6 @@ vxlan_encap_offload_err:
                return -EOPNOTSUPP;
        }
 
-       if (family == AF_INET6) {
-               netdev_warn(priv->netdev,
-                           "IPv6 tunnel encap offload isn't supported\n");
-               return -EOPNOTSUPP;
-       }
-
        hash_key = hash_encap_info(key);
 
        hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
@@ -885,7 +1024,11 @@ vxlan_encap_offload_err:
        e->tunnel_type = tunnel_type;
        INIT_LIST_HEAD(&e->flows);
 
-       err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e, &out_dev);
+       if (family == AF_INET)
+               err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e, &out_dev);
+       else if (family == AF_INET6)
+               err = mlx5e_create_encap_header_ipv6(priv, mirred_dev, e, &out_dev);
+
        if (err)
                goto out_err;