MPLS Unifom mode

- support both pipe and uniform modes for all MPLS LSP
- all API programming for output-labels requires that the mode (and associated data) is specificed
   - API changes in MPLS, BIER and IP are involved
- new DPO [sub] types for MPLS labels to handle the two modes.

Change-Id: I87b76401e996f10dfbdbe4552ff6b19af958783c
Signed-off-by: Neale Ranns <nranns@cisco.com>

7 files changed, 84 insertions, 92 deletions

diff --git a/src/vnet/mpls/mpls.api b/src/vnet/mpls/mpls.api
index 572ac915a4a..6047d255aa7 100644
--- a/src/vnet/mpls/mpls.api
+++ b/src/vnet/mpls/mpls.api
@@ -13,7 +13,8 @@
  * limitations under the License.
  */
 
-option version = "1.0.1";
+option version = "1.1.0";
+import "vnet/fib/fib_types.api";
 
 /** \brief Bind/Unbind an MPLS local label to an IP prefix. i.e. create
            a per-prefix label entry.
@@ -40,21 +41,6 @@ autoreply define mpls_ip_bind_unbind
   u8 mb_address[16];
 };
 
-/** \brief MPLS tunnel Add / del route
-    @param client_index - opaque cookie to identify the sender
-    @param context - sender context, to match reply w/ request
-    @param mt_is_add - Is this a route add or delete
-    @param mt_sw_if_index - The SW interface index of the tunnel to delete
-    @param mt_is_multicast - Is the tunnel's underlying LSP multicast
-    @param mt_next_hop_proto_is_ip4 - The next-hop is IPV4
-    @param mt_next_hop_weight - The weight, for UCMP
-    @param mt_next_hop_preference - The preference
-    @param mt_next_hop[16] - the nextop address
-    @param mt_next_hop_sw_if_index - the next-hop SW interface
-    @param mt_next_hop_table_id - the next-hop table-id (if appropriate)
-    @param mt_next_hop_n_out_labels - the number of next-hop output labels
-    @param mt_next_hop_out_label_stack - the next-hop output label stack,  outer most first
-*/
 define mpls_tunnel_add_del
 {
   u32 client_index;
@@ -70,7 +56,7 @@ define mpls_tunnel_add_del
   u8 mt_next_hop_n_out_labels;
   u32 mt_next_hop_sw_if_index;
   u32 mt_next_hop_table_id;
-  u32 mt_next_hop_out_label_stack[mt_next_hop_n_out_labels];
+  vl_api_fib_mpls_label_t mt_next_hop_out_label_stack[mt_next_hop_n_out_labels];
 };
 
 /** \brief Reply for MPLS tunnel add / del request
@@ -96,34 +82,6 @@ define mpls_tunnel_dump
   i32 tunnel_index;
 };
 
-/** \brief FIB path
-    @param sw_if_index - index of the interface
-    @param weight - The weight, for UCMP
-    @param is_local - local if non-zero, else remote
-    @param is_drop - Drop the packet
-    @param is_unreach - Drop the packet and rate limit send ICMP unreachable
-    @param is_prohibit - Drop the packet and rate limit send ICMP prohibited
-    @param afi - the afi of the next hop, IP46_TYPE_IP4=1, IP46_TYPE_IP6=2
-    @param next_hop[16] - the next hop address
-
-    WARNING: this type is replicated, pending cleanup completion
-
-*/
-typeonly manual_print manual_endian define fib_path2
-{
-  u32 sw_if_index;
-  u32 table_id;
-  u8 weight;
-  u8 preference;
-  u8 is_local;
-  u8 is_drop;
-  u8 is_unreach;
-  u8 is_prohibit;
-  u8 afi;
-  u8 next_hop[16];
-  u32 labels[16];
-};
-
 /** \brief mpls tunnel details
 */
 manual_endian manual_print define mpls_tunnel_details
@@ -134,7 +92,7 @@ manual_endian manual_print define mpls_tunnel_details
   u8 mt_l2_only;
   u8 mt_is_multicast;
   u32 mt_count;
-  vl_api_fib_path2_t mt_paths[mt_count];
+  vl_api_fib_path_t mt_paths[mt_count];
 };
 
 /** \brief MPLS Route Add / del route
@@ -207,7 +165,7 @@ autoreply define mpls_route_add_del
   u32 mr_next_hop_sw_if_index;
   u32 mr_next_hop_table_id;
   u32 mr_next_hop_via_label;
-  u32 mr_next_hop_out_label_stack[mr_next_hop_n_out_labels];
+  vl_api_fib_mpls_label_t mr_next_hop_out_label_stack[mr_next_hop_n_out_labels];
 };
 
 /** \brief Dump MPLS fib table
@@ -234,7 +192,7 @@ manual_endian manual_print define mpls_fib_details
   u8  eos_bit;
   u32 label;
   u32 count;
-  vl_api_fib_path2_t path[count];
+  vl_api_fib_path_t path[count];
 };
 
 /** \brief Enable or Disable MPLS on and interface
diff --git a/src/vnet/mpls/mpls.c b/src/vnet/mpls/mpls.c
index 25957fb3fea..be72d3f829f 100644
--- a/src/vnet/mpls/mpls.c
+++ b/src/vnet/mpls/mpls.c
@@ -47,6 +47,9 @@ u8 * format_mpls_unicast_label (u8 * s, va_list * args)
   case MPLS_IETF_GAL_LABEL:
       s = format (s, "%s", MPLS_IETF_GAL_STRING);
       break;
+  case MPLS_LABEL_POP:
+      s = format (s, "pop");
+      break;
   default:
       s = format (s, "%d", label);
       break;
diff --git a/src/vnet/mpls/mpls_api.c b/src/vnet/mpls/mpls_api.c
index 36fa610e8ca..169ee406a91 100644
--- a/src/vnet/mpls/mpls_api.c
+++ b/src/vnet/mpls/mpls_api.c
@@ -170,8 +170,8 @@ static int
 mpls_route_add_del_t_handler (vnet_main_t * vnm,
 			      vl_api_mpls_route_add_del_t * mp)
 {
+  fib_mpls_label_t *label_stack = NULL;
   u32 fib_index, next_hop_fib_index;
-  mpls_label_t *label_stack = NULL;
   int rv, ii, n_labels;;
 
   fib_prefix_t pfx = {
@@ -211,13 +211,19 @@ mpls_route_add_del_t_handler (vnet_main_t * vnm,
   n_labels = mp->mr_next_hop_n_out_labels;
   if (n_labels == 0)
     ;
-  else if (1 == n_labels)
-    vec_add1 (label_stack, ntohl (mp->mr_next_hop_out_label_stack[0]));
   else
     {
       vec_validate (label_stack, n_labels - 1);
       for (ii = 0; ii < n_labels; ii++)
-	label_stack[ii] = ntohl (mp->mr_next_hop_out_label_stack[ii]);
+	{
+	  label_stack[ii].fml_value =
+	    ntohl (mp->mr_next_hop_out_label_stack[ii].label);
+	  label_stack[ii].fml_ttl = mp->mr_next_hop_out_label_stack[ii].ttl;
+	  label_stack[ii].fml_exp = mp->mr_next_hop_out_label_stack[ii].exp;
+	  label_stack[ii].fml_mode =
+	    (mp->mr_next_hop_out_label_stack[ii].is_uniform ?
+	     FIB_MPLS_LSP_MODE_UNIFORM : FIB_MPLS_LSP_MODE_PIPE);
+	}
     }
 
   /* *INDENT-OFF* */
@@ -323,8 +329,16 @@ vl_api_mpls_tunnel_add_del_t_handler (vl_api_mpls_tunnel_add_del_t * mp)
   if (mp->mt_is_add)
     {
       for (ii = 0; ii < mp->mt_next_hop_n_out_labels; ii++)
-	vec_add1 (rpath.frp_label_stack,
-		  ntohl (mp->mt_next_hop_out_label_stack[ii]));
+	{
+	  fib_mpls_label_t fml = {
+	    .fml_value = ntohl (mp->mt_next_hop_out_label_stack[ii].label),
+	    .fml_ttl = mp->mt_next_hop_out_label_stack[ii].ttl,
+	    .fml_exp = mp->mt_next_hop_out_label_stack[ii].exp,
+	    .fml_mode = (mp->mt_next_hop_out_label_stack[ii].is_uniform ?
+			 FIB_MPLS_LSP_MODE_UNIFORM : FIB_MPLS_LSP_MODE_PIPE),
+	  };
+	  vec_add1 (rpath.frp_label_stack, fml);
+	}
     }
 
   vec_add1 (rpaths, rpath);
@@ -388,7 +402,7 @@ send_mpls_tunnel_entry (u32 mti, void *arg)
   mpls_tunnel_send_walk_ctx_t *ctx;
   vl_api_mpls_tunnel_details_t *mp;
   const mpls_tunnel_t *mt;
-  vl_api_fib_path2_t *fp;
+  vl_api_fib_path_t *fp;
   u32 n;
 
   ctx = arg;
@@ -399,8 +413,8 @@ send_mpls_tunnel_entry (u32 mti, void *arg)
   mt = mpls_tunnel_get (mti);
   n = fib_path_list_get_n_paths (mt->mt_path_list);
 
-  mp = vl_msg_api_alloc (sizeof (*mp) + n * sizeof (vl_api_fib_path2_t));
-  memset (mp, 0, sizeof (*mp) + n * sizeof (vl_api_fib_path2_t));
+  mp = vl_msg_api_alloc (sizeof (*mp) + n * sizeof (vl_api_fib_path_t));
+  memset (mp, 0, sizeof (*mp) + n * sizeof (vl_api_fib_path_t));
 
   mp->_vl_msg_id = ntohs (VL_API_MPLS_TUNNEL_DETAILS);
   mp->context = ctx->context;
@@ -456,7 +470,7 @@ send_mpls_fib_details (vpe_api_main_t * am,
 {
   vl_api_mpls_fib_details_t *mp;
   fib_route_path_encode_t *api_rpath;
-  vl_api_fib_path2_t *fp;
+  vl_api_fib_path_t *fp;
   int path_count;
 
   path_count = vec_len (api_rpaths);
diff --git a/src/vnet/mpls/mpls_input.c b/src/vnet/mpls/mpls_input.c
index 86ad8bba270..d1881d4050b 100644
--- a/src/vnet/mpls/mpls_input.c
+++ b/src/vnet/mpls/mpls_input.c
@@ -51,10 +51,11 @@ format_mpls_input_trace (u8 * s, va_list * args)
   foreach_mpls_input_next;
 #undef _
   
-  s = format (s, "MPLS: next %s[%d]  label %d ttl %d", 
+  s = format (s, "MPLS: next %s[%d]  label %d ttl %d exp %d", 
               next_name, t->next_index,
 	      vnet_mpls_uc_get_label(label),
-	      vnet_mpls_uc_get_ttl(label));
+	      vnet_mpls_uc_get_ttl(label),
+	      vnet_mpls_uc_get_exp(label));
 
   return s;
 }
@@ -74,21 +75,13 @@ mpls_input_inline (vlib_main_t * vm,
                    vlib_frame_t * from_frame)
 {
   u32 n_left_from, next_index, * from, * to_next;
-  mpls_input_runtime_t * rt;
-  mpls_main_t * mm;
+  mpls_main_t * mm = &mpls_main;
   u32 thread_index = vlib_get_thread_index();
   vlib_simple_counter_main_t * cm;
   vnet_main_t * vnm = vnet_get_main();
 
   from = vlib_frame_vector_args (from_frame);
   n_left_from = from_frame->n_vectors;
-  rt = vlib_node_get_runtime_data (vm, mpls_input_node.index);
-  mm = rt->mpls_main;
-  /* 
-   * Force an initial lookup every time, in case the control-plane
-   * changed the label->FIB mapping.
-   */
-  rt->last_label = ~0;
 
   next_index = node->cached_next_index;
 
@@ -279,18 +272,11 @@ VLIB_NODE_FUNCTION_MULTIARCH (mpls_input_node, mpls_input)
 static void
 mpls_setup_nodes (vlib_main_t * vm)
 {
-  mpls_input_runtime_t * rt;
   pg_node_t * pn;
 
   pn = pg_get_node (mpls_input_node.index);
   pn->unformat_edit = unformat_pg_mpls_header;
 
-  rt = vlib_node_get_runtime_data (vm, mpls_input_node.index);
-  rt->last_label = (u32) ~0;
-  rt->last_inner_fib_index = 0;
-  rt->last_outer_fib_index = 0;
-  rt->mpls_main = &mpls_main;
-
   ethernet_register_input_type (vm, ETHERNET_TYPE_MPLS,
                                 mpls_input_node.index);
 }
@@ -309,16 +295,3 @@ static clib_error_t * mpls_input_init (vlib_main_t * vm)
 }
 
 VLIB_INIT_FUNCTION (mpls_input_init);
-
-static clib_error_t * mpls_input_worker_init (vlib_main_t * vm)
-{
-  mpls_input_runtime_t * rt;
-  rt = vlib_node_get_runtime_data (vm, mpls_input_node.index);
-  rt->last_label = (u32) ~0;
-  rt->last_inner_fib_index = 0;
-  rt->last_outer_fib_index = 0;
-  rt->mpls_main = &mpls_main;
-  return 0;
-}
-
-VLIB_WORKER_INIT_FUNCTION (mpls_input_worker_init);
diff --git a/src/vnet/mpls/mpls_tunnel.c b/src/vnet/mpls/mpls_tunnel.c
index 8ed2c409ce8..c2067d81b00 100644
--- a/src/vnet/mpls/mpls_tunnel.c
+++ b/src/vnet/mpls/mpls_tunnel.c
@@ -123,6 +123,12 @@ mpls_tunnel_collect_forwarding (fib_node_index_t pl_index,
     path_ext = fib_path_ext_list_find_by_path_index(&ctx->mt->mt_path_exts,
                                                     path_index);
 
+    /*
+     * we don't want IP TTL decrements for packets hitting the MPLS labels
+     * we stack on, since the IP TTL decrement is done by the adj
+     */
+    path_ext->fpe_mpls_flags |= FIB_PATH_EXT_MPLS_FLAG_NO_IP_TTL_DECR;
+
     if (NULL != path_ext)
     {
         /*
@@ -273,9 +279,8 @@ mpls_tunnel_stack (adj_index_t ai)
 
         mpls_tunnel_mk_lb(mt,
                           adj->ia_link,
-                          (VNET_LINK_MPLS == adj_get_link_type(ai) ?
-                           FIB_FORW_CHAIN_TYPE_MPLS_NON_EOS:
-                           FIB_FORW_CHAIN_TYPE_MPLS_EOS),
+                          fib_forw_chain_type_from_link_type(
+                              adj_get_link_type(ai)),
                           &dpo);
 
         adj_nbr_midchain_stack(ai, &dpo);
@@ -521,6 +526,11 @@ mpls_tunnel_tx (vlib_main_t * vm,
           b0 = vlib_get_buffer(vm, bi0);
 
           vnet_buffer(b0)->ip.adj_index[VLIB_TX] = mt->mt_l2_lb.dpoi_index;
+          /* since we are coming out of the L2 world, where the vlib_buffer
+           * union is used for other things, make sure it is clean for
+           * MPLS from now on.
+           */
+          vnet_buffer(b0)->mpls.first = 0;
 
           if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
             {
diff --git a/src/vnet/mpls/mpls_types.h b/src/vnet/mpls/mpls_types.h
index f1c3191e00c..c21bdf1eace 100644
--- a/src/vnet/mpls/mpls_types.h
+++ b/src/vnet/mpls/mpls_types.h
@@ -47,6 +47,14 @@
 
 #define MPLS_LABEL_INVALID (MPLS_IETF_MAX_LABEL+1)
 
+/**
+ * A value that is explicit about the end of the LSP. Specifying
+ * a label value is needed when the mode configuration (pipe/uniform)
+ * is also requested.
+ * imp-null implies a label swap. pop can be used for a deag.
+ */
+#define MPLS_LABEL_POP (MPLS_IETF_MAX_LABEL+2)
+
 #define MPLS_LABEL_IS_REAL(_lbl) \
     (((_lbl) > MPLS_IETF_MIN_UNRES_LABEL) &&	\
      ((_lbl) <= MPLS_IETF_MAX_UNRES_LABEL))
diff --git a/src/vnet/mpls/packet.h b/src/vnet/mpls/packet.h
index bc67445be89..ca6ac407686 100644
--- a/src/vnet/mpls/packet.h
+++ b/src/vnet/mpls/packet.h
@@ -42,6 +42,32 @@ typedef enum mpls_eos_bit_t_
     [MPLS_EOS] = "eos",				\
 }
 
+/**
+ * The Default TTL added to MPLS label headers when no other value is available
+ */
+#define MPLS_LABEL_DEFAULT_TTL 64
+
+/**
+ * The Default EXP added to MPLS label headers when no other value is available
+ */
+#define MPLS_LABEL_DEFAULT_EXP 0
+
+/**
+ * When in uniform mode convert an IPv[46] DSCP value to an MPLS EXP value
+ */
+static inline u8 ip_dscp_to_mpls_exp (u8 tos)
+{
+    return (tos >> 5);
+}
+
+/**
+ * When in uniform mode convert an MPLS EXP value to an IPv[46] DSCP value
+ */
+static inline u8 mpls_exp_to_ip_dscp (u8 exp)
+{
+    return (exp << 5);
+}
+
 #define FOR_EACH_MPLS_EOS_BIT(_eos) \
     for (_eos = MPLS_NON_EOS; _eos <= MPLS_EOS; _eos++)