MPLS Mcast

 1 - interface-DPO
        Used in the Data-plane to change a packet's input interface
 2 - MPLS multicast FIB entry
        Same as a unicast entry but it links to a replicate not a load-balance DPO
 3 - Multicast MPLS tunnel
        Update MPLS tunnels to use a FIB path-list to describe the endpoint[s]. Use the path-list to generate the forwarding chain (DPOs) to link to .
 4 - Resolve a path via a local label (of an mLDP LSP)
        For IP multicast entries to use an LSP in the replication list, we need to decribe the 'resolve-via-label' where the label is that of a multicast LSP.
 5 - MPLS disposition path sets RPF-ID
        For a interface-less LSP (i.e. mLDP not RSVP-TE) at the tail of the LSP we still need to perform an RPF check. An MPLS disposition DPO performs the MPLS pop validation checks and sets the RPF-ID in the packet.
 6 - RPF check with per-entry RPF-ID
       An RPF-ID is used instead of a real interface SW if index in the case the IP traffic arrives from an LSP that does not have an associated interface.

Change-Id: Ib92e177be919147bafeb599729abf3d1abc2f4b3
Signed-off-by: Neale Ranns <nranns@cisco.com>

7 files changed, 226 insertions, 32 deletions

diff --git a/src/vnet/adj/adj.c b/src/vnet/adj/adj.c
index 90182006f60..36dfe500eb3 100644
--- a/src/vnet/adj/adj.c
+++ b/src/vnet/adj/adj.c
@@ -67,6 +67,10 @@ adj_alloc (fib_protocol_t proto)
     adj->lookup_next_index = 0;
     adj->ia_delegates = NULL;
 
+    /* lest it become a midchain in the future */
+    memset(&adj->sub_type.midchain.next_dpo, 0,
+           sizeof(adj->sub_type.midchain.next_dpo));
+
     ip4_main.lookup_main.adjacency_heap = adj_pool;
     ip6_main.lookup_main.adjacency_heap = adj_pool;
 
@@ -118,6 +122,9 @@ format_ip_adjacency (u8 * s, va_list * args)
     case IP_LOOKUP_NEXT_MCAST:
 	s = format (s, "%U", format_adj_mcast, adj_index, 0);
 	break;
+    case IP_LOOKUP_NEXT_MCAST_MIDCHAIN:
+	s = format (s, "%U", format_adj_mcast_midchain, adj_index, 0);
+	break;
     default:
 	break;
     }
@@ -180,6 +187,7 @@ adj_last_lock_gone (ip_adjacency_t *adj)
 			 adj->rewrite_header.sw_if_index);
 	break;
     case IP_LOOKUP_NEXT_MCAST:
+    case IP_LOOKUP_NEXT_MCAST_MIDCHAIN:
 	adj_mcast_remove(adj->ia_nh_proto,
 			 adj->rewrite_header.sw_if_index);
 	break;
@@ -338,6 +346,7 @@ adj_walk (u32 sw_if_index,
     FOR_EACH_FIB_IP_PROTOCOL(proto)
     {
         adj_nbr_walk(sw_if_index, proto, cb, ctx);
+        adj_mcast_walk(sw_if_index, proto, cb, ctx);
     }
 }
 
@@ -544,9 +553,9 @@ adj_show (vlib_main_t * vm,
  * [@0]
  * [@1]  glean: loop0
  * [@2] ipv4 via 1.0.0.2 loop0: IP4: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc
- * [@3] mpls via 1.0.0.2 loop0: MPLS_UNICAST: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc
+ * [@3] mpls via 1.0.0.2 loop0: MPLS: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc
  * [@4] ipv4 via 1.0.0.3 loop0: IP4: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc
- * [@5] mpls via 1.0.0.3 loop0: MPLS_UNICAST: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc
+ * [@5] mpls via 1.0.0.3 loop0: MPLS: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc
  * @cliexend
  ?*/
 VLIB_CLI_COMMAND (adj_show_command, static) = {
diff --git a/src/vnet/adj/adj.h b/src/vnet/adj/adj.h
index 32997c91f92..ed5eb1f1e88 100644
--- a/src/vnet/adj/adj.h
+++ b/src/vnet/adj/adj.h
@@ -81,6 +81,10 @@ typedef enum
   /** Multicast Adjacency. */
   IP_LOOKUP_NEXT_MCAST,
 
+  /** Multicast Midchain Adjacency. An Adjacency for sending macst packets
+   *  on a tunnel/virtual interface */
+  IP_LOOKUP_NEXT_MCAST_MIDCHAIN,
+
   IP_LOOKUP_N_NEXT,
 } __attribute__ ((packed)) ip_lookup_next_t;
 
@@ -107,6 +111,7 @@ typedef enum
     [IP_LOOKUP_NEXT_REWRITE] = "ip4-rewrite",    		\
     [IP_LOOKUP_NEXT_MCAST] = "ip4-rewrite-mcast",	        \
     [IP_LOOKUP_NEXT_MIDCHAIN] = "ip4-midchain",		        \
+    [IP_LOOKUP_NEXT_MCAST_MIDCHAIN] = "ip4-mcast-midchain",     \
     [IP_LOOKUP_NEXT_ICMP_ERROR] = "ip4-icmp-error",		\
 }
 
@@ -119,6 +124,7 @@ typedef enum
     [IP_LOOKUP_NEXT_REWRITE] = "ip6-rewrite",			\
     [IP_LOOKUP_NEXT_MCAST] = "ip6-rewrite-mcast",		\
     [IP_LOOKUP_NEXT_MIDCHAIN] = "ip6-midchain",			\
+    [IP_LOOKUP_NEXT_MCAST_MIDCHAIN] = "ip6-mcast-midchain",     \
     [IP_LOOKUP_NEXT_ICMP_ERROR] = "ip6-icmp-error",		\
     [IP6_LOOKUP_NEXT_HOP_BY_HOP] = "ip6-hop-by-hop",		\
     [IP6_LOOKUP_NEXT_ADD_HOP_BY_HOP] = "ip6-add-hop-by-hop",	\
diff --git a/src/vnet/adj/adj_internal.h b/src/vnet/adj/adj_internal.h
index 306686255b8..2c123c542a1 100644
--- a/src/vnet/adj/adj_internal.h
+++ b/src/vnet/adj/adj_internal.h
@@ -17,6 +17,7 @@
 #define __ADJ_INTERNAL_H__
 
 #include <vnet/adj/adj.h>
+#include <vnet/adj/adj_mcast.h>
 #include <vnet/ip/ip.h>
 #include <vnet/mpls/mpls.h>
 #include <vnet/adj/adj_l2.h>
@@ -87,11 +88,14 @@ adj_get_index (ip_adjacency_t *adj)
     return (adj - adj_pool);
 }
 
-extern void adj_nbr_update_rewrite_internal (ip_adjacency_t *adj,
-					     ip_lookup_next_t adj_next_index,
-					     u32 complete_next_index,
-					     u32 next_index,
-					     u8 *rewrite);
+extern void adj_nbr_update_rewrite_internal(ip_adjacency_t *adj,
+                                            ip_lookup_next_t adj_next_index,
+                                            u32 complete_next_index,
+                                            u32 next_index,
+                                            u8 *rewrite);
+extern void adj_midchain_setup(adj_index_t adj_index,
+                               adj_midchain_fixup_t fixup,
+                               adj_flags_t flags);
 
 extern ip_adjacency_t * adj_alloc(fib_protocol_t proto);
 
diff --git a/src/vnet/adj/adj_mcast.c b/src/vnet/adj/adj_mcast.c
index 4f678e43e80..755abfd4639 100644
--- a/src/vnet/adj/adj_mcast.c
+++ b/src/vnet/adj/adj_mcast.c
@@ -13,7 +13,7 @@
  * limitations under the License.
  */
 
-#include <vnet/adj/adj.h>
+#include <vnet/adj/adj_mcast.h>
 #include <vnet/adj/adj_internal.h>
 #include <vnet/fib/fib_walk.h>
 #include <vnet/ip/ip.h>
@@ -129,6 +129,59 @@ adj_mcast_update_rewrite (adj_index_t adj_index,
     adj->rewrite_header.dst_mcast_mask = clib_host_to_net_u32(mask);
 }
 
+/**
+ * adj_mcast_midchain_update_rewrite
+ *
+ * Update the adjacency's rewrite string. A NULL string implies the
+ * rewirte is reset (i.e. when ARP/ND etnry is gone).
+ * NB: the adj being updated may be handling traffic in the DP.
+ */
+void
+adj_mcast_midchain_update_rewrite (adj_index_t adj_index,
+                                   adj_midchain_fixup_t fixup,
+                                   adj_flags_t flags,
+                                   u8 *rewrite,
+                                   u8 offset,
+                                   u32 mask)
+{
+    ip_adjacency_t *adj;
+
+    ASSERT(ADJ_INDEX_INVALID != adj_index);
+
+    adj = adj_get(adj_index);
+
+    /*
+     * one time only update. since we don't support chainging the tunnel
+     * src,dst, this is all we need.
+     */
+    ASSERT(adj->lookup_next_index == IP_LOOKUP_NEXT_MCAST);
+    /*
+     * tunnels can always provide a rewrite.
+     */
+    ASSERT(NULL != rewrite);
+
+    adj_midchain_setup(adj_index, fixup, flags);
+
+    /*
+     * update the adj's rewrite string and build the arc
+     * from the rewrite node to the interface's TX node
+     */
+    adj_nbr_update_rewrite_internal(adj, IP_LOOKUP_NEXT_MCAST_MIDCHAIN,
+                                    adj_get_mcast_node(adj->ia_nh_proto),
+                                    vnet_tx_node_index_for_sw_interface(
+                                        vnet_get_main(),
+                                        adj->rewrite_header.sw_if_index),
+                                    rewrite);
+
+    /*
+     * set the fields corresponding to the mcast IP address rewrite
+     * The mask must be stored in network byte order, since the packet's
+     * IP address will also be in network order.
+     */
+    adj->rewrite_header.dst_mcast_offset = offset;
+    adj->rewrite_header.dst_mcast_mask = clib_host_to_net_u32(mask);
+}
+
 void
 adj_mcast_remove (fib_protocol_t proto,
 		  u32 sw_if_index)
@@ -260,6 +313,24 @@ adj_mcast_interface_delete (vnet_main_t * vnm,
 
 VNET_SW_INTERFACE_ADD_DEL_FUNCTION(adj_mcast_interface_delete);
 
+/**
+ * @brief Walk the multicast Adjacencies on a given interface
+ */
+void
+adj_mcast_walk (u32 sw_if_index,
+                fib_protocol_t proto,
+                adj_walk_cb_t cb,
+                void *ctx)
+{
+    if (vec_len(adj_mcasts[proto]) > sw_if_index)
+    {
+        if (ADJ_INDEX_INVALID != adj_mcasts[proto][sw_if_index])
+        {
+            cb(adj_mcasts[proto][sw_if_index], ctx);
+        }
+    }
+}
+
 u8*
 format_adj_mcast (u8* s, va_list *ap)
 {
@@ -269,6 +340,8 @@ format_adj_mcast (u8* s, va_list *ap)
 
     s = format(s, "%U-mcast: ",
                format_fib_protocol, adj->ia_nh_proto);
+    if (adj->rewrite_header.flags & VNET_REWRITE_HAS_FEATURES)
+        s = format(s, "[features] ");
     s = format (s, "%U",
 		format_vnet_rewrite,
                 &adj->rewrite_header, sizeof (adj->rewrite_data), 0);
@@ -276,6 +349,28 @@ format_adj_mcast (u8* s, va_list *ap)
     return (s);
 }
 
+u8*
+format_adj_mcast_midchain (u8* s, va_list *ap)
+{
+    index_t index = va_arg(*ap, index_t);
+    CLIB_UNUSED(u32 indent) = va_arg(*ap, u32);
+    vnet_main_t * vnm = vnet_get_main();
+    ip_adjacency_t * adj = adj_get(index);
+
+    s = format(s, "%U-mcast-midchain: ",
+               format_fib_protocol, adj->ia_nh_proto);
+    s = format (s, "%U",
+		format_vnet_rewrite,
+		vnm->vlib_main, &adj->rewrite_header,
+                sizeof (adj->rewrite_data), 0);
+    s = format (s, "\n%Ustacked-on:\n%U%U",
+		format_white_space, indent,
+		format_white_space, indent+2,
+		format_dpo_id, &adj->sub_type.midchain.next_dpo, indent+2);
+
+    return (s);
+}
+
 
 static void
 adj_dpo_lock (dpo_id_t *dpo)
@@ -293,6 +388,11 @@ const static dpo_vft_t adj_mcast_dpo_vft = {
     .dv_unlock = adj_dpo_unlock,
     .dv_format = format_adj_mcast,
 };
+const static dpo_vft_t adj_mcast_midchain_dpo_vft = {
+    .dv_lock = adj_dpo_lock,
+    .dv_unlock = adj_dpo_unlock,
+    .dv_format = format_adj_mcast_midchain,
+};
 
 /**
  * @brief The per-protocol VLIB graph nodes that are assigned to a mcast
@@ -320,6 +420,31 @@ const static char* const * const adj_mcast_nodes[DPO_PROTO_NUM] =
 };
 
 /**
+ * @brief The per-protocol VLIB graph nodes that are assigned to a mcast
+ *        object.
+ *
+ * this means that these graph nodes are ones from which a mcast is the
+ * parent object in the DPO-graph.
+ */
+const static char* const adj_mcast_midchain_ip4_nodes[] =
+{
+    "ip4-mcast-midchain",
+    NULL,
+};
+const static char* const adj_mcast_midchain_ip6_nodes[] =
+{
+    "ip6-mcast-midchain",
+    NULL,
+};
+
+const static char* const * const adj_mcast_midchain_nodes[DPO_PROTO_NUM] =
+{
+    [DPO_PROTO_IP4]  = adj_mcast_midchain_ip4_nodes,
+    [DPO_PROTO_IP6]  = adj_mcast_midchain_ip6_nodes,
+    [DPO_PROTO_MPLS] = NULL,
+};
+
+/**
  * @brief Return the size of the adj DB.
  * This is only for testing purposes so an efficient implementation is not needed
  */
@@ -349,5 +474,10 @@ adj_mcast_db_size (void)
 void
 adj_mcast_module_init (void)
 {
-    dpo_register(DPO_ADJACENCY_MCAST, &adj_mcast_dpo_vft, adj_mcast_nodes);
+    dpo_register(DPO_ADJACENCY_MCAST,
+                 &adj_mcast_dpo_vft,
+                 adj_mcast_nodes);
+    dpo_register(DPO_ADJACENCY_MCAST_MIDCHAIN,
+                 &adj_mcast_midchain_dpo_vft,
+                 adj_mcast_midchain_nodes);
 }
diff --git a/src/vnet/adj/adj_mcast.h b/src/vnet/adj/adj_mcast.h
index 40d44313357..bfb0d6f6d11 100644
--- a/src/vnet/adj/adj_mcast.h
+++ b/src/vnet/adj/adj_mcast.h
@@ -26,6 +26,7 @@
 #define __ADJ_MCAST_H__
 
 #include <vnet/adj/adj_types.h>
+#include <vnet/adj/adj_midchain.h>
 
 /**
  * @brief
@@ -69,9 +70,35 @@ extern void adj_mcast_update_rewrite(adj_index_t adj_index,
                                      u32 mask);
 
 /**
+ * @brief
+ *  Update the rewrite string for an existing adjacecny and
+ *  Convert the adjacency into a midchain
+ *
+ * @param
+ *  The index of the adj to update
+ *
+ * @param
+ *  The new rewrite
+ */
+extern void adj_mcast_midchain_update_rewrite(adj_index_t adj_index,
+                                              adj_midchain_fixup_t fixup,
+                                              adj_flags_t flags,
+                                              u8 *rewrite,
+                                              u8 offset,
+                                              u32 mask);
+/**
+ * @brief Walk the multicast Adjacencies on a given interface
+ */
+extern void adj_mcast_walk (u32 sw_if_index,
+                            fib_protocol_t adj_nh_proto,
+                            adj_walk_cb_t cb,
+                            void *ctx);
+
+/**
  * @brief Format/display a mcast adjacency.
  */
 extern u8* format_adj_mcast(u8* s, va_list *ap);
+extern u8* format_adj_mcast_midchain(u8* s, va_list *ap);
 
 /**
  * @brief Get the sze of the mcast adj DB. Test purposes only.
diff --git a/src/vnet/adj/adj_midchain.c b/src/vnet/adj/adj_midchain.c
index 5756de4393d..a93a1c3e1b7 100644
--- a/src/vnet/adj/adj_midchain.c
+++ b/src/vnet/adj/adj_midchain.c
@@ -346,7 +346,7 @@ adj_get_midchain_node (vnet_link_t link)
 static u8
 adj_midchain_get_feature_arc_index_for_link_type (const ip_adjacency_t *adj)
 {
-  u8 arc = (u8) ~0;
+    u8 arc = (u8) ~0;
     switch (adj->ia_link)
     {
     case VNET_LINK_IP4:
@@ -393,17 +393,14 @@ adj_nbr_midchain_get_tx_node (ip_adjacency_t *adj)
 }
 
 /**
- * adj_nbr_midchain_update_rewrite
+ * adj_midchain_setup
  *
- * Update the adjacency's rewrite string. A NULL string implies the
- * rewrite is reset (i.e. when ARP/ND etnry is gone).
- * NB: the adj being updated may be handling traffic in the DP.
+ * Setup the adj as a mid-chain
  */
 void
-adj_nbr_midchain_update_rewrite (adj_index_t adj_index,
-				 adj_midchain_fixup_t fixup,
-				 adj_flags_t flags,
-				 u8 *rewrite)
+adj_midchain_setup (adj_index_t adj_index,
+                    adj_midchain_fixup_t fixup,
+                    adj_flags_t flags)
 {
     u32 feature_index, tx_node;
     ip_adjacency_t *adj;
@@ -413,16 +410,6 @@ adj_nbr_midchain_update_rewrite (adj_index_t adj_index,
 
     adj = adj_get(adj_index);
 
-    /*
-     * one time only update. since we don't support chainging the tunnel
-     * src,dst, this is all we need.
-     */
-    ASSERT(adj->lookup_next_index == IP_LOOKUP_NEXT_ARP);
-    /*
-     * tunnels can always provide a rewrite.
-     */
-    ASSERT(NULL != rewrite);
-
     adj->sub_type.midchain.fixup_func = fixup;
     adj->ia_flags |= flags;
 
@@ -447,6 +434,38 @@ adj_nbr_midchain_update_rewrite (adj_index_t adj_index,
     dpo_stack_from_node(tx_node,
 			&adj->sub_type.midchain.next_dpo,
 			drop_dpo_get(vnet_link_to_dpo_proto(adj->ia_link)));
+}
+
+/**
+ * adj_nbr_midchain_update_rewrite
+ *
+ * Update the adjacency's rewrite string. A NULL string implies the
+ * rewrite is reset (i.e. when ARP/ND etnry is gone).
+ * NB: the adj being updated may be handling traffic in the DP.
+ */
+void
+adj_nbr_midchain_update_rewrite (adj_index_t adj_index,
+				 adj_midchain_fixup_t fixup,
+				 adj_flags_t flags,
+				 u8 *rewrite)
+{
+    ip_adjacency_t *adj;
+
+    ASSERT(ADJ_INDEX_INVALID != adj_index);
+
+    adj = adj_get(adj_index);
+
+    /*
+     * one time only update. since we don't support chainging the tunnel
+     * src,dst, this is all we need.
+     */
+    ASSERT(adj->lookup_next_index == IP_LOOKUP_NEXT_ARP);
+    /*
+     * tunnels can always provide a rewrite.
+     */
+    ASSERT(NULL != rewrite);
+
+    adj_midchain_setup(adj_index, fixup, flags);
 
     /*
      * update the rewirte with the workers paused.
@@ -454,7 +473,7 @@ adj_nbr_midchain_update_rewrite (adj_index_t adj_index,
     adj_nbr_update_rewrite_internal(adj,
 				    IP_LOOKUP_NEXT_MIDCHAIN,
 				    adj_get_midchain_node(adj->ia_link),
-				    tx_node,
+				    adj_nbr_midchain_get_tx_node(adj),
 				    rewrite);
 }
 
@@ -496,7 +515,8 @@ adj_nbr_midchain_stack (adj_index_t adj_index,
 
     adj = adj_get(adj_index);
 
-    ASSERT(IP_LOOKUP_NEXT_MIDCHAIN == adj->lookup_next_index);
+    ASSERT((IP_LOOKUP_NEXT_MIDCHAIN == adj->lookup_next_index) ||
+           (IP_LOOKUP_NEXT_MCAST_MIDCHAIN == adj->lookup_next_index));
 
     dpo_stack_from_node(adj_nbr_midchain_get_tx_node(adj),
 			&adj->sub_type.midchain.next_dpo,
diff --git a/src/vnet/adj/adj_nbr.c b/src/vnet/adj/adj_nbr.c
index ddacb030f1d..3d450d1fba3 100644
--- a/src/vnet/adj/adj_nbr.c
+++ b/src/vnet/adj/adj_nbr.c
@@ -195,8 +195,6 @@ adj_nbr_alloc (fib_protocol_t nh_proto,
     adj->ia_link = link_type;
     adj->ia_nh_proto = nh_proto;
     adj->rewrite_header.sw_if_index = sw_if_index;
-    memset(&adj->sub_type.midchain.next_dpo, 0,
-           sizeof(adj->sub_type.midchain.next_dpo));
 
     adj_nbr_evaluate_feature (adj_get_index(adj));
     return (adj);