L2 over MPLS

[support for VPWS/VPLS]
- switch to using dpo_proto_t rather than fib_protocol_t in fib_paths so that we can describe L2 paths
- VLIB nodes to handle pop/push of MPLS labels to L2

Change-Id: Id050d06a11fd2c9c1c81ce5a0654e6c5ae6afa6e
Signed-off-by: Neale Ranns <nranns@cisco.com>

5 files changed, 94 insertions, 80 deletions

diff --git a/src/vnet/mpls/mpls.api b/src/vnet/mpls/mpls.api
index 67f1045d57a..5973a0a6cbd 100644
--- a/src/vnet/mpls/mpls.api
+++ b/src/vnet/mpls/mpls.api
@@ -156,7 +156,7 @@ manual_endian manual_print define mpls_tunnel_details
     @param mr_is_interface_rx - Interface Receive path
     @param mr_is_interface_rx - RPF-ID Receive path. The next-hop interface
                                 is used as the RPF-ID
-    @param mr_next_hop_proto_is_ip4 - The next-hop is IPV4
+    @param mr_next_hop_proto - The next-hop protocol, of type dpo_proto_t
     @param mr_next_hop_weight - The weight, for UCMP
     @param mr_next_hop[16] - the nextop address
     @param mr_next_hop_sw_if_index - the next-hop SW interface
@@ -182,7 +182,7 @@ autoreply define mpls_route_add_del
   u8 mr_is_resolve_attached;
   u8 mr_is_interface_rx;
   u8 mr_is_rpf_id;
-  u8 mr_next_hop_proto_is_ip4;
+  u8 mr_next_hop_proto;
   u8 mr_next_hop_weight;
   u8 mr_next_hop_preference;
   u8 mr_next_hop[16];
diff --git a/src/vnet/mpls/mpls.c b/src/vnet/mpls/mpls.c
index 068d31f4c8e..266ba42c94f 100644
--- a/src/vnet/mpls/mpls.c
+++ b/src/vnet/mpls/mpls.c
@@ -261,7 +261,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
 			 &rpath.frp_sw_if_index,
 			 &rpath.frp_weight))
       {
-	  rpath.frp_proto = FIB_PROTOCOL_IP4;
+	  rpath.frp_proto = DPO_PROTO_IP4;
 	  vec_add1(rpaths, rpath);
       }
 
@@ -272,7 +272,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
 			 &rpath.frp_sw_if_index,
 			 &rpath.frp_weight))
       {
-	  rpath.frp_proto = FIB_PROTOCOL_IP6;
+	  rpath.frp_proto = DPO_PROTO_IP6;
 	  vec_add1(rpaths, rpath);
       }
 
@@ -283,7 +283,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
 			 &rpath.frp_sw_if_index))
       {
 	  rpath.frp_weight = 1;
-	  rpath.frp_proto = FIB_PROTOCOL_IP4;
+	  rpath.frp_proto = DPO_PROTO_IP4;
 	  vec_add1(rpaths, rpath);
       }
       else if (unformat (line_input, "rx-ip4 %U",
@@ -291,7 +291,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
 			 &rpath.frp_sw_if_index))
       {
 	  rpath.frp_weight = 1;
-	  rpath.frp_proto = FIB_PROTOCOL_IP4;
+	  rpath.frp_proto = DPO_PROTO_IP4;
           rpath.frp_flags = FIB_ROUTE_PATH_INTF_RX;
 	  vec_add1(rpaths, rpath);
       }
@@ -302,7 +302,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
 			 &rpath.frp_sw_if_index))
       {
 	  rpath.frp_weight = 1;
-	  rpath.frp_proto = FIB_PROTOCOL_IP6;
+	  rpath.frp_proto = DPO_PROTO_IP6;
 	  vec_add1(rpaths, rpath);
       }
       else if (unformat (line_input, "via %U next-hop-table %d",
@@ -312,7 +312,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
       {
 	  rpath.frp_weight = 1;
 	  rpath.frp_sw_if_index = ~0;
-	  rpath.frp_proto = FIB_PROTOCOL_IP4;
+	  rpath.frp_proto = DPO_PROTO_IP4;
 	  vec_add1(rpaths, rpath);
       }
       else if (unformat (line_input, "via %U next-hop-table %d",
@@ -322,7 +322,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
       {
 	  rpath.frp_weight = 1;
 	  rpath.frp_sw_if_index = ~0;
-	  rpath.frp_proto = FIB_PROTOCOL_IP6;
+	  rpath.frp_proto = DPO_PROTO_IP6;
 	  vec_add1(rpaths, rpath);
       }
       else if (unformat (line_input, "via %U",
@@ -336,7 +336,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
 	  rpath.frp_fib_index = table_id;
 	  rpath.frp_weight = 1;
 	  rpath.frp_sw_if_index = ~0;
-	  rpath.frp_proto = FIB_PROTOCOL_IP4;
+	  rpath.frp_proto = DPO_PROTO_IP4;
 	  vec_add1(rpaths, rpath);
       }
       else if (unformat (line_input, "via %U",
@@ -346,7 +346,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
 	  rpath.frp_fib_index = table_id;
 	  rpath.frp_weight = 1;
 	  rpath.frp_sw_if_index = ~0;
-	  rpath.frp_proto = FIB_PROTOCOL_IP6;
+	  rpath.frp_proto = DPO_PROTO_IP6;
 	  vec_add1(rpaths, rpath);
       }
       else if (unformat (line_input, "%d", &local_label))
@@ -355,7 +355,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
 			 "ip4-lookup-in-table %d",
 			 &rpath.frp_fib_index))
       {
-          rpath.frp_proto = FIB_PROTOCOL_IP4;
+          rpath.frp_proto = DPO_PROTO_IP4;
           rpath.frp_sw_if_index = FIB_NODE_INDEX_INVALID;
 	  pfx.fp_payload_proto = DPO_PROTO_IP4;
 	  vec_add1(rpaths, rpath);
@@ -364,7 +364,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
 			 "ip6-lookup-in-table %d",
 			 &rpath.frp_fib_index))
       {
-          rpath.frp_proto = FIB_PROTOCOL_IP6;
+          rpath.frp_proto = DPO_PROTO_IP6;
           rpath.frp_sw_if_index = FIB_NODE_INDEX_INVALID;
 	  vec_add1(rpaths, rpath);
 	  pfx.fp_payload_proto = DPO_PROTO_IP6;
@@ -373,11 +373,21 @@ vnet_mpls_local_label (vlib_main_t * vm,
 			 "mpls-lookup-in-table %d",
 			 &rpath.frp_fib_index))
       {
-          rpath.frp_proto = FIB_PROTOCOL_MPLS;
+          rpath.frp_proto = DPO_PROTO_MPLS;
           rpath.frp_sw_if_index = FIB_NODE_INDEX_INVALID;
 	  pfx.fp_payload_proto = DPO_PROTO_MPLS;
 	  vec_add1(rpaths, rpath);
       }
+      else if (unformat (line_input,
+			 "l2-input-on %U",
+			 unformat_vnet_sw_interface, vnm,
+			 &rpath.frp_sw_if_index))
+      {
+          rpath.frp_proto = DPO_PROTO_ETHERNET;
+	  pfx.fp_payload_proto = DPO_PROTO_ETHERNET;
+          rpath.frp_flags = FIB_ROUTE_PATH_INTF_RX;
+	  vec_add1(rpaths, rpath);
+      }
       else if (unformat (line_input, "out-label %U",
                          unformat_mpls_unicast_label,
 			 &out_label))
@@ -440,7 +450,7 @@ vnet_mpls_local_label (vlib_main_t * vm,
       pfx.fp_proto = FIB_PROTOCOL_MPLS;
       pfx.fp_len = 21;
       pfx.fp_label = local_label;
-      pfx.fp_payload_proto = fib_proto_to_dpo(rpaths[0].frp_proto);
+      pfx.fp_payload_proto = rpaths[0].frp_proto;
 
       /*
        * the CLI parsing stored table Ids, swap to FIB indicies
diff --git a/src/vnet/mpls/mpls_api.c b/src/vnet/mpls/mpls_api.c
index 92fb24a6e88..737299e6a2b 100644
--- a/src/vnet/mpls/mpls_api.c
+++ b/src/vnet/mpls/mpls_api.c
@@ -144,14 +144,7 @@ mpls_route_add_del_t_handler (vnet_main_t * vnm,
   };
   if (pfx.fp_eos)
     {
-      if (mp->mr_next_hop_proto_is_ip4)
-	{
-	  pfx.fp_payload_proto = DPO_PROTO_IP4;
-	}
-      else
-	{
-	  pfx.fp_payload_proto = DPO_PROTO_IP6;
-	}
+      pfx.fp_payload_proto = mp->mr_next_hop_proto;
     }
   else
     {
@@ -161,7 +154,7 @@ mpls_route_add_del_t_handler (vnet_main_t * vnm,
   rv = add_del_route_check (FIB_PROTOCOL_MPLS,
 			    mp->mr_table_id,
 			    mp->mr_next_hop_sw_if_index,
-			    dpo_proto_to_fib (pfx.fp_payload_proto),
+			    pfx.fp_payload_proto,
 			    mp->mr_next_hop_table_id,
 			    mp->mr_create_table_if_needed,
 			    mp->mr_is_rpf_id,
@@ -173,9 +166,9 @@ mpls_route_add_del_t_handler (vnet_main_t * vnm,
   ip46_address_t nh;
   memset (&nh, 0, sizeof (nh));
 
-  if (mp->mr_next_hop_proto_is_ip4)
+  if (DPO_PROTO_IP4 == mp->mr_next_hop_proto)
     memcpy (&nh.ip4, mp->mr_next_hop, sizeof (nh.ip4));
-  else
+  else if (DPO_PROTO_IP6 == mp->mr_next_hop_proto)
     memcpy (&nh.ip6, mp->mr_next_hop, sizeof (nh.ip6));
 
   n_labels = mp->mr_next_hop_n_out_labels;
@@ -202,7 +195,7 @@ mpls_route_add_del_t_handler (vnet_main_t * vnm,
 				   mp->mr_is_interface_rx,
 				   mp->mr_is_rpf_id,
 				   fib_index, &pfx,
-				   mp->mr_next_hop_proto_is_ip4,
+				   mp->mr_next_hop_proto,
 				   &nh, ntohl (mp->mr_next_hop_sw_if_index),
 				   next_hop_fib_index,
 				   mp->mr_next_hop_weight,
@@ -243,13 +236,13 @@ vl_api_mpls_tunnel_add_del_t_handler (vl_api_mpls_tunnel_add_del_t * mp)
 
   if (mp->mt_next_hop_proto_is_ip4)
     {
-      rpath.frp_proto = FIB_PROTOCOL_IP4;
+      rpath.frp_proto = DPO_PROTO_IP4;
       clib_memcpy (&rpath.frp_addr.ip4,
 		   mp->mt_next_hop, sizeof (rpath.frp_addr.ip4));
     }
   else
     {
-      rpath.frp_proto = FIB_PROTOCOL_IP6;
+      rpath.frp_proto = DPO_PROTO_IP6;
       clib_memcpy (&rpath.frp_addr.ip6,
 		   mp->mt_next_hop, sizeof (rpath.frp_addr.ip6));
     }
diff --git a/src/vnet/mpls/mpls_tunnel.c b/src/vnet/mpls/mpls_tunnel.c
index c025cc58033..6452a60bfe2 100644
--- a/src/vnet/mpls/mpls_tunnel.c
+++ b/src/vnet/mpls/mpls_tunnel.c
@@ -171,7 +171,7 @@ mpls_tunnel_mk_lb (mpls_tunnel_t *mt,
     vec_validate(ctx.next_hops, fib_path_list_get_n_paths(mt->mt_path_list));
     vec_reset_length(ctx.next_hops);
 
-    lb_proto = vnet_link_to_dpo_proto(linkt);
+    lb_proto = fib_forw_chain_type_to_dpo_proto(fct);
 
     fib_path_list_walk(mt->mt_path_list,
                        mpls_tunnel_collect_forwarding,
@@ -313,12 +313,34 @@ mpls_tunnel_restack (mpls_tunnel_t *mt)
     /*
      * walk all the adjacencies on the MPLS interface and restack them
      */
-    FOR_EACH_FIB_PROTOCOL(proto)
+    if (mt->mt_flags & MPLS_TUNNEL_FLAG_L2)
     {
-        adj_nbr_walk(mt->mt_sw_if_index,
-                     proto,
-                     mpls_adj_walk_cb,
-                     NULL);
+        /*
+         * Stack a load-balance that drops, whilst we have no paths
+         */
+        vnet_hw_interface_t * hi;
+        dpo_id_t dpo = DPO_INVALID;
+
+        mpls_tunnel_mk_lb(mt,
+                          VNET_LINK_MPLS,
+                          FIB_FORW_CHAIN_TYPE_ETHERNET,
+                          &dpo);
+
+        hi = vnet_get_hw_interface(vnet_get_main(), mt->mt_hw_if_index);
+        dpo_stack_from_node(hi->tx_node_index,
+                            &mt->mt_l2_lb,
+                            &dpo);
+        dpo_reset(&dpo);
+    }
+    else
+    {
+        FOR_EACH_FIB_PROTOCOL(proto)
+        {
+            adj_nbr_walk(mt->mt_sw_if_index,
+                         proto,
+                         mpls_adj_walk_cb,
+                         NULL);
+        }
     }
 }
 
@@ -495,7 +517,7 @@ mpls_tunnel_tx (vlib_main_t * vm,
 
           b0 = vlib_get_buffer(vm, bi0);
 
-          vnet_buffer(b0)->ip.adj_index[VLIB_TX] = mt->mt_l2_adj;
+          vnet_buffer(b0)->ip.adj_index[VLIB_TX] = mt->mt_l2_lb.dpoi_index;
 
           if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
             {
@@ -506,7 +528,7 @@ mpls_tunnel_tx (vlib_main_t * vm,
 
           vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
                                            to_next, n_left_to_next,
-                                           bi0, mt->mt_l2_tx_arc);
+                                           bi0, mt->mt_l2_lb.dpoi_next_node);
         }
 
       vlib_put_next_frame (vm, node, next_index, n_left_to_next);
@@ -565,8 +587,7 @@ vnet_mpls_tunnel_del (u32 sw_if_index)
     if (FIB_NODE_INDEX_INVALID != mt->mt_path_list)
         fib_path_list_child_remove(mt->mt_path_list,
                                    mt->mt_sibling_index);
-    if (ADJ_INDEX_INVALID != mt->mt_l2_adj)
-        adj_unlock(mt->mt_l2_adj);
+    dpo_reset(&mt->mt_l2_lb);
 
     vec_add1 (mpls_tunnel_free_hw_if_indices, mt->mt_hw_if_index);
     pool_put(mpls_tunnel_pool, mt);
@@ -587,12 +608,13 @@ vnet_mpls_tunnel_create (u8 l2_only,
     memset (mt, 0, sizeof (*mt));
     mti = mt - mpls_tunnel_pool;
     fib_node_init(&mt->mt_node, FIB_NODE_TYPE_MPLS_TUNNEL);
-    mt->mt_l2_adj = ADJ_INDEX_INVALID;
     mt->mt_path_list = FIB_NODE_INDEX_INVALID;
     mt->mt_sibling_index = FIB_NODE_INDEX_INVALID;
 
     if (is_multicast)
         mt->mt_flags |= MPLS_TUNNEL_FLAG_MCAST;
+    if (l2_only)
+        mt->mt_flags |= MPLS_TUNNEL_FLAG_L2;
 
     /*
      * Create a new, or re=use and old, tunnel HW interface
@@ -614,7 +636,7 @@ vnet_mpls_tunnel_create (u8 l2_only,
                                  mti,
                                  mpls_tunnel_hw_interface_class.index,
                                  mti);
-        hi = vnet_get_hw_interface(vnm, mt->mt_hw_if_index);
+        hi = vnet_get_hw_interface (vnm, mt->mt_hw_if_index);
     }
 
     /*
@@ -624,19 +646,6 @@ vnet_mpls_tunnel_create (u8 l2_only,
     vec_validate_init_empty(mpls_tunnel_db, mt->mt_sw_if_index, ~0);
     mpls_tunnel_db[mt->mt_sw_if_index] = mti;
 
-    if (l2_only)
-    {
-        mt->mt_l2_adj =
-            adj_nbr_add_or_lock(fib_path_list_get_proto(mt->mt_path_list),
-                                VNET_LINK_ETHERNET,
-                                &zero_addr,
-                                mt->mt_sw_if_index);
-
-        mt->mt_l2_tx_arc = vlib_node_add_named_next(vlib_get_main(),
-                                                    hi->tx_node_index,
-                                                    "adj-l2-midchain");
-    }
-
     return (mt->mt_sw_if_index);
 }
 
@@ -803,7 +812,7 @@ vnet_create_mpls_tunnel_command_fn (vlib_main_t * vm,
                            &rpath.frp_sw_if_index))
         {
             rpath.frp_weight = 1;
-            rpath.frp_proto = FIB_PROTOCOL_IP4;
+            rpath.frp_proto = DPO_PROTO_IP4;
         }
 
         else if (unformat (line_input, "via %U %U",
@@ -813,7 +822,7 @@ vnet_create_mpls_tunnel_command_fn (vlib_main_t * vm,
                            &rpath.frp_sw_if_index))
         {
             rpath.frp_weight = 1;
-            rpath.frp_proto = FIB_PROTOCOL_IP6;
+            rpath.frp_proto = DPO_PROTO_IP6;
         }
         else if (unformat (line_input, "via %U",
                            unformat_ip6_address,
@@ -822,7 +831,7 @@ vnet_create_mpls_tunnel_command_fn (vlib_main_t * vm,
             rpath.frp_fib_index = 0;
             rpath.frp_weight = 1;
             rpath.frp_sw_if_index = ~0;
-            rpath.frp_proto = FIB_PROTOCOL_IP6;
+            rpath.frp_proto = DPO_PROTO_IP6;
         }
         else if (unformat (line_input, "via %U",
                            unformat_ip4_address,
@@ -831,7 +840,7 @@ vnet_create_mpls_tunnel_command_fn (vlib_main_t * vm,
             rpath.frp_fib_index = 0;
             rpath.frp_weight = 1;
             rpath.frp_sw_if_index = ~0;
-            rpath.frp_proto = FIB_PROTOCOL_IP4;
+            rpath.frp_proto = DPO_PROTO_IP4;
         }
         else if (unformat (line_input, "l2-only"))
             l2_only = 1;
@@ -915,6 +924,14 @@ format_mpls_tunnel (u8 * s, va_list * args)
     s = format(s, "%U", format_fib_path_ext_list, &mt->mt_path_exts);
     s = format(s, "\n");
 
+    if (mt->mt_flags & MPLS_TUNNEL_FLAG_L2)
+    {
+        s = format(s, " forwarding: %U\n",
+                   format_fib_forw_chain_type,
+                   FIB_FORW_CHAIN_TYPE_ETHERNET);
+        s = format(s, " %U\n", format_dpo_id, &mt->mt_l2_lb, 2);
+    }
+
     return (s);
 }
 
diff --git a/src/vnet/mpls/mpls_tunnel.h b/src/vnet/mpls/mpls_tunnel.h
index 4cb0a860802..285817c33c4 100644
--- a/src/vnet/mpls/mpls_tunnel.h
+++ b/src/vnet/mpls/mpls_tunnel.h
@@ -23,14 +23,19 @@ typedef enum mpls_tunnel_attribute_t_
 {
     MPLS_TUNNEL_ATTRIBUTE_FIRST = 0,
     /**
+     * @brief The tunnel is L2 only
+     */
+    MPLS_TUNNEL_ATTRIBUTE_L2 = MPLS_TUNNEL_ATTRIBUTE_FIRST,
+    /**
      * @brief The tunnel has an underlying multicast LSP
      */
-    MPLS_TUNNEL_ATTRIBUTE_MCAST = MPLS_TUNNEL_ATTRIBUTE_FIRST,
+    MPLS_TUNNEL_ATTRIBUTE_MCAST,
     MPLS_TUNNEL_ATTRIBUTE_LAST = MPLS_TUNNEL_ATTRIBUTE_MCAST,
 } mpls_tunnel_attribute_t;
 
 #define MPLS_TUNNEL_ATTRIBUTES {		  \
     [MPLS_TUNNEL_ATTRIBUTE_MCAST]  = "multicast", \
+    [MPLS_TUNNEL_ATTRIBUTE_L2]     = "L2",   \
 }
 #define FOR_EACH_MPLS_TUNNEL_ATTRIBUTE(_item)		\
     for (_item = MPLS_TUNNEL_ATTRIBUTE_FIRST;		\
@@ -39,6 +44,7 @@ typedef enum mpls_tunnel_attribute_t_
 
 typedef enum mpls_tunnel_flag_t_ {
     MPLS_TUNNEL_FLAG_NONE   = 0,
+    MPLS_TUNNEL_FLAG_L2     = (1 << MPLS_TUNNEL_ATTRIBUTE_L2),
     MPLS_TUNNEL_FLAG_MCAST  = (1 << MPLS_TUNNEL_ATTRIBUTE_MCAST),
 } __attribute__ ((packed)) mpls_tunnel_flags_t;
 
@@ -60,14 +66,19 @@ typedef struct mpls_tunnel_t_
 
     /**
      * @brief If the tunnel is an L2 tunnel, this is the link type ETHERNET
-     * adjacency
+     * load-balance
+     */
+    dpo_id_t mt_l2_lb;
+
+    /**
+     * @brief The HW interface index of the tunnel interfaces
      */
-    adj_index_t mt_l2_adj;
+    u32 mt_hw_if_index;
 
     /**
-     * @brief on a L2 tunnel this is the VLIB arc from the L2-tx to the l2-midchain
+     * @brief The SW interface index of the tunnel interfaces
      */
-    u32 mt_l2_tx_arc;
+    u32 mt_sw_if_index;
 
     /**
      * @brief The path-list over which the tunnel's destination is reachable
@@ -83,23 +94,6 @@ typedef struct mpls_tunnel_t_
      * A vector of path extensions o hold the label stack for each path
      */
     fib_path_ext_list_t mt_path_exts;
-
-    /**
-     * @brief Flag to indicate the tunnel is only for L2 traffic, that is
-     * this tunnel belongs in a bridge domain.
-     */
-    u8 mt_l2_only;
-
-    /**
-     * @brief The HW interface index of the tunnel interfaces
-     */
-    u32 mt_hw_if_index;
-
-    /**
-     * @brief The SW interface index of the tunnel interfaces
-     */
-    u32 mt_sw_if_index;
-
 } mpls_tunnel_t;
 
 /**