L2 over LISP and GRE (VPP-457)

Change-Id: I0d7f9c7f41a9f9e0acb0950adedb90d45df08c2a
Signed-off-by: Neale Ranns <nranns@cisco.com>

4 files changed, 118 insertions, 256 deletions

diff --git a/vnet/vnet/mpls/interface.c b/vnet/vnet/mpls/interface.c
index 9ef4c293..dd40fc29 100644
--- a/vnet/vnet/mpls/interface.c
+++ b/vnet/vnet/mpls/interface.c
@@ -536,190 +536,21 @@ static const int dpo_next_2_mpls_post_rewrite[DPO_LAST] = {
     [DPO_LOAD_BALANCE] = IP_LOOKUP_NEXT_LOAD_BALANCE,
 };
 
-static uword
-mpls_post_rewrite (vlib_main_t * vm,
-                   vlib_node_runtime_t * node,
-                   vlib_frame_t * from_frame)
+static void
+mpls_gre_fixup (vlib_main_t *vm,
+		ip_adjacency_t *adj,
+		vlib_buffer_t * b0)
 {
-  ip4_main_t * im = &ip4_main;
-  ip_lookup_main_t * lm = &im->lookup_main;
-  u32 n_left_from, next_index, * from, * to_next;
-  u16 old_l0 = 0; //, old_l1 = 0;
-
-  from = vlib_frame_vector_args (from_frame);
-  n_left_from = from_frame->n_vectors;
-
-  next_index = node->cached_next_index;
-
-  while (n_left_from > 0)
-    {
-      u32 n_left_to_next;
-
-      vlib_get_next_frame (vm, node, next_index,
-			   to_next, n_left_to_next);
-
-      /* while (n_left_from >= 4 && n_left_to_next >= 2) */
-      /* 	{ */
-      /* 	  u32 bi0, bi1; */
-      /* 	  vlib_buffer_t * b0, * b1; */
-      /*     ip4_header_t * ip0, * ip1; */
-      /* 	  u32 next0; */
-      /* 	  u32 next1; */
-      /*     u16 new_l0, new_l1, adj_index0, adj_index1; */
-      /*     ip_csum_t sum0, sum1; */
-      /* 	  ip_adjacency_t *adj0, *adj1; */
-
-      /* 	  /\* Prefetch next iteration. *\/ */
-      /* 	  { */
-      /* 	    vlib_buffer_t * p2, * p3; */
-
-      /* 	    p2 = vlib_get_buffer (vm, from[2]); */
-      /* 	    p3 = vlib_get_buffer (vm, from[3]); */
-
-      /* 	    vlib_prefetch_buffer_header (p2, LOAD); */
-      /* 	    vlib_prefetch_buffer_header (p3, LOAD); */
-
-      /* 	    CLIB_PREFETCH (p2->data, 2*CLIB_CACHE_LINE_BYTES, LOAD); */
-      /* 	    CLIB_PREFETCH (p3->data, 2*CLIB_CACHE_LINE_BYTES, LOAD); */
-      /* 	  } */
-
-      /* 	  bi0 = from[0]; */
-      /* 	  bi1 = from[1]; */
-      /* 	  to_next[0] = bi0; */
-      /* 	  to_next[1] = bi1; */
-      /* 	  from += 2; */
-      /* 	  to_next += 2; */
-      /* 	  n_left_to_next -= 2; */
-      /* 	  n_left_from -= 2; */
-
-
-      /* 	  b0 = vlib_get_buffer (vm, bi0); */
-      /* 	  b1 = vlib_get_buffer (vm, bi1); */
-      /*     ip0 = vlib_buffer_get_current (b0); */
-      /*     ip1 = vlib_buffer_get_current (b1); */
-          
-      /*     /\* Note: the tunnel rewrite sets up sw_if_index[VLIB_TX] *\/ */
-
-      /*     /\* set the GRE (outer) ip packet length, fix the bloody checksum *\/ */
-      /*     sum0 = ip0->checksum; */
-      /*     sum1 = ip1->checksum; */
-
-      /*     /\* old_l0, old_l1 always 0, see the rewrite setup *\/ */
-      /*     new_l0 =  */
-      /*       clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)); */
-      /*     new_l1 =  */
-      /*       clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1)); */
-          
-      /*     sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t, */
-      /*                            length /\* changed member *\/); */
-      /*     sum1 = ip_csum_update (sum1, old_l1, new_l1, ip4_header_t, */
-      /*                            length /\* changed member *\/); */
-      /*     ip0->checksum = ip_csum_fold (sum0); */
-      /*     ip1->checksum = ip_csum_fold (sum1); */
-      /*     ip0->length = new_l0; */
-      /*     ip1->length = new_l1; */
+    ip4_header_t * ip0;
 
-      /* 	  /\* replace the TX adj in the packet with the next in the chain *\/ */
-      /* 	  adj_index0 = vnet_buffer (b0)->ip.adj_index[VLIB_TX]; */
-      /* 	  adj_index1 = vnet_buffer (b1)->ip.adj_index[VLIB_TX]; */
+    ip0 = vlib_buffer_get_current (b0);
 
-      /* 	  adj0 = ip_get_adjacency (lm, adj_index0); */
-      /* 	  adj1 = ip_get_adjacency (lm, adj_index1); */
-
-      /* 	  ASSERT(adj0->sub_type.midchain.adj_index != ADJ_INDEX_INVALID); */
-      /* 	  ASSERT(adj1->sub_type.midchain.adj_index != ADJ_INDEX_INVALID); */
-
-      /* 	  adj_index0 = adj0->sub_type.midchain.adj_index; */
-      /* 	  adj_index1 = adj1->sub_type.midchain.adj_index; */
-
-      /* 	  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = adj_index0; */
-      /* 	  vnet_buffer (b1)->ip.adj_index[VLIB_TX] = adj_index1; */
-
-      /* 	  /\* get the next adj in the chain to determine the next graph node *\/ */
-      /* 	  adj0 = ip_get_adjacency (lm, adj_index0); */
-      /* 	  adj1 = ip_get_adjacency (lm, adj_index1); */
-
-      /* 	  next0 = adj0->lookup_next_index; */
-      /* 	  next1 = adj1->lookup_next_index; */
-
-      /* 	  vlib_validate_buffer_enqueue_x2 (vm, node, next_index, */
-      /* 					   to_next, n_left_to_next, */
-      /* 					   bi0, bi1, next0, next1); */
-      /* 	} */
-
-      while (n_left_from > 0 && n_left_to_next > 0)
-	{
-	  ip_adjacency_t * adj0;
-	  u32 bi0;
-	  vlib_buffer_t * b0;
-          ip4_header_t * ip0;
-	  u32 next0;
-          u16 new_l0, adj_index0;
-          ip_csum_t sum0;
-
-	  bi0 = from[0];
-	  to_next[0] = bi0;
-	  from += 1;
-	  to_next += 1;
-	  n_left_from -= 1;
-	  n_left_to_next -= 1;
-
-	  b0 = vlib_get_buffer (vm, bi0);
-          ip0 = vlib_buffer_get_current (b0);
-          
-          /* Note: the tunnel rewrite sets up sw_if_index[VLIB_TX] */
-
-          /* set the GRE (outer) ip packet length, fix the bloody checksum */
-          sum0 = ip0->checksum;
-          /* old_l0 always 0, see the rewrite setup */
-          new_l0 = 
-            clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
-          
-          sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t,
-                                 length /* changed member */);
-          ip0->checksum = ip_csum_fold (sum0);
-          ip0->length = new_l0;
-
-	  /* replace the TX adj in the packet with the next in the chain */
-	  adj_index0 = vnet_buffer (b0)->ip.adj_index[VLIB_TX];
-
-          ASSERT(adj_index0);
-
-	  adj0 = ip_get_adjacency (lm, adj_index0);
-	  ASSERT(adj0->sub_type.midchain.next_dpo.dpoi_index != ADJ_INDEX_INVALID);
-	  adj_index0 = adj0->sub_type.midchain.next_dpo.dpoi_index;
-	  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = adj_index0;
-
-	  /* get the next adj in the chain to determine the next graph node */
-	  ASSERT(0);
-	  next0 = 0; //adj0->sub_type.midchain.next_dpo.dpoi_next;
-
-	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
-					   to_next, n_left_to_next,
-					   bi0, next0);
-	}
-
-      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
-    }
-  vlib_node_increment_counter (vm, mpls_input_node.index,
-                               MPLS_ERROR_PKTS_ENCAP, from_frame->n_vectors);
-  return from_frame->n_vectors;
+    /* Fixup the checksum and len fields in the GRE tunnel encap
+     * that was applied at the midchain node */
+    ip0->length = clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
+    ip0->checksum = ip4_header_checksum (ip0);
 }
 
-VLIB_REGISTER_NODE (mpls_post_rewrite_node) = {
-  .function = mpls_post_rewrite,
-  .name = "mpls-post-rewrite",
-  /* Takes a vector of packets. */
-  .vector_size = sizeof (u32),
-
-  .runtime_data_bytes = 0,
-
-  .n_next_nodes = IP_LOOKUP_N_NEXT,
-  .next_nodes = IP4_LOOKUP_NEXT_NODES,
-};
-
-VLIB_NODE_FUNCTION_MULTIARCH (mpls_post_rewrite_node, mpls_post_rewrite)
-
 static u8 * mpls_gre_rewrite (mpls_main_t *mm, mpls_gre_tunnel_t * t)
 {
   ip4_header_t * ip0;
@@ -780,7 +611,7 @@ mpls_sw_interface_enable_disable (mpls_main_t * mm,
 {
   mpls_interface_state_change_callback_t *callback;
   vlib_main_t * vm = vlib_get_main();
-  ip_config_main_t * cm = &mm->rx_config_mains;
+  ip_config_main_t * cm = &mm->feature_config_mains[VNET_IP_RX_UNICAST_FEAT];
   vnet_config_main_t * vcm = &cm->config_main;
   u32 lookup_feature_index;
   fib_node_index_t lfib_index;
@@ -1101,7 +932,8 @@ int mpls_gre_tunnel_add (ip4_address_t *src,
 					    hi->sw_if_index);
 
 	adj_nbr_midchain_update_rewrite(tp->adj_index,
-					mpls_post_rewrite_node.index,
+					mpls_gre_fixup,
+					ADJ_MIDCHAIN_FLAG_NONE,
 					rewrite_data);
 	mpls_gre_tunnel_stack(tp);
 
diff --git a/vnet/vnet/mpls/mpls.h b/vnet/vnet/mpls/mpls.h
index 2aeae49d..e7fed192 100644
--- a/vnet/vnet/mpls/mpls.h
+++ b/vnet/vnet/mpls/mpls.h
@@ -127,11 +127,12 @@ typedef struct {
   uword *fib_index_by_table_id;
 
   /* rx/tx interface/feature configuration. */
-  ip_config_main_t rx_config_mains, tx_config_main;
+  ip_config_main_t feature_config_mains[VNET_N_IP_FEAT];
 
   /* Built-in unicast feature path indices, see ip_feature_init_cast(...)  */
   u32 mpls_rx_feature_lookup;
   u32 mpls_rx_feature_not_enabled;
+  u32 mpls_tx_feature_interface_output;
 
   /* pool of gre tunnel instances */
   mpls_gre_tunnel_t *gre_tunnels;
@@ -154,10 +155,10 @@ typedef struct {
   u32 ip6_classify_mpls_policy_encap_next_index;
 
   /* feature path configuration lists */
-  vnet_ip_feature_registration_t * next_feature;
+  vnet_ip_feature_registration_t * next_feature[VNET_N_IP_FEAT];
 
   /* Save feature results for show command */
-  char **feature_nodes;
+  char **feature_nodes[VNET_N_IP_FEAT];
 
   /* IP4 enabled count by software interface */
   u8 * mpls_enabled_by_sw_if_index;
@@ -179,11 +180,23 @@ static void __vnet_add_feature_registration_uc_##x (void)       \
 static void __vnet_add_feature_registration_uc_##x (void)       \
 {                                                               \
   mpls_main_t * mm = &mpls_main;                                \
-  uc_##x.next = mm->next_feature;                               \
-  mm->next_feature = &uc_##x;                                   \
+  uc_##x.next = mm->next_feature[VNET_IP_RX_UNICAST_FEAT];      \
+  mm->next_feature[VNET_IP_RX_UNICAST_FEAT] = &uc_##x;          \
 }                                                               \
 __VA_ARGS__ vnet_ip_feature_registration_t uc_##x
 
+#define VNET_MPLS_TX_FEATURE_INIT(x,...)                        \
+  __VA_ARGS__ vnet_ip_feature_registration_t tx_##x;            \
+static void __vnet_add_feature_registration_tx_##x (void)       \
+  __attribute__((__constructor__)) ;                            \
+static void __vnet_add_feature_registration_tx_##x (void)       \
+{                                                               \
+  mpls_main_t * mm = &mpls_main;                                \
+  tx_##x.next = mm->next_feature[VNET_IP_TX_FEAT];              \
+  mm->next_feature[VNET_IP_TX_FEAT] = &tx_##x;                  \
+}                                                               \
+__VA_ARGS__ vnet_ip_feature_registration_t tx_##x 
+
 extern clib_error_t * mpls_feature_init(vlib_main_t * vm);
 
 format_function_t format_mpls_protocol;
diff --git a/vnet/vnet/mpls/mpls_features.c b/vnet/vnet/mpls/mpls_features.c
index d3a726af..80002fcb 100644
--- a/vnet/vnet/mpls/mpls_features.c
+++ b/vnet/vnet/mpls/mpls_features.c
@@ -112,22 +112,49 @@ VNET_MPLS_FEATURE_INIT (mpls_not_enabled, static) = {
   .feature_index = &mpls_main.mpls_rx_feature_not_enabled,
 };
 
-static char * feature_start_nodes[] =
+/* Built-in ip4 tx feature path definition */
+VNET_MPLS_TX_FEATURE_INIT (interface_output, static) = {
+  .node_name = "interface-output",
+  .runs_before = 0, /* not before any other features */
+  .feature_index = &mpls_main.mpls_tx_feature_interface_output,
+};
+
+
+static char * rx_feature_start_nodes[] =
 {
     "mpls-input",
 };
+static char * tx_feature_start_nodes[] = 
+{
+    "mpls-output",
+    "mpls-midchain",
+};
 
 clib_error_t *
 mpls_feature_init (vlib_main_t * vm)
 {
-  ip_config_main_t * cm = &mpls_main.rx_config_mains;
+  ip_config_main_t * cm = &mpls_main.feature_config_mains[VNET_IP_RX_UNICAST_FEAT];
   vnet_config_main_t * vcm = &cm->config_main;
+  clib_error_t *error;
+
+  if ((error = ip_feature_init_cast (vm, cm, vcm,
+				     rx_feature_start_nodes,
+				     ARRAY_LEN(rx_feature_start_nodes),
+				     mpls_main.next_feature[VNET_IP_RX_UNICAST_FEAT],
+				     &mpls_main.feature_nodes[VNET_IP_RX_UNICAST_FEAT])))
+      return error;
 
-  return (ip_feature_init_cast (vm, cm, vcm,
-                                feature_start_nodes,
-                                ARRAY_LEN(feature_start_nodes),
-                                VNET_IP_RX_UNICAST_FEAT,
-                                VNET_L3_PACKET_TYPE_MPLS_UNICAST));
+  cm  = &mpls_main.feature_config_mains[VNET_IP_TX_FEAT];
+  vcm = &cm->config_main;
+
+  if ((error = ip_feature_init_cast (vm, cm, vcm,
+				     tx_feature_start_nodes,
+				     ARRAY_LEN(tx_feature_start_nodes),
+				     mpls_main.next_feature[VNET_IP_TX_FEAT],
+				     &mpls_main.feature_nodes[VNET_IP_TX_FEAT])))
+      return error;
+
+  return error;
 }
 
 static clib_error_t *
@@ -137,39 +164,53 @@ mpls_sw_interface_add_del (vnet_main_t * vnm,
 {
   vlib_main_t * vm = vnm->vlib_main;
   mpls_main_t * mm = &mpls_main;
-  ip_config_main_t * cm = &mm->rx_config_mains;
-  vnet_config_main_t * vcm = &cm->config_main;
-  u32 drop_feature_index;
-  u32 ci;
-
-  vec_validate_init_empty (mm->mpls_enabled_by_sw_if_index, sw_if_index, 0);
-  vec_validate_init_empty (mm->fib_index_by_sw_if_index, sw_if_index, 0);
-  vec_validate_init_empty (cm->config_index_by_sw_if_index, sw_if_index, ~0);
-  ci = cm->config_index_by_sw_if_index[sw_if_index];
-
-  drop_feature_index = mm->mpls_rx_feature_not_enabled;
-
-  if (is_add)
-    ci = vnet_config_add_feature (vm, vcm, ci,
-                                  drop_feature_index,
-                                  /* config data */ 0,
-                                  /* # bytes of config data */ 0);
-  else
-   {
-     ci = vnet_config_del_feature (vm, vcm, ci,
-                                   drop_feature_index,
-                                   /* config data */ 0,
-                                   /* # bytes of config data */ 0);
-     mm->mpls_enabled_by_sw_if_index[sw_if_index] = 0;;
-   }
-
-  cm->config_index_by_sw_if_index[sw_if_index] = ci;
+  u32 feature_index;
+  u32 ci, cast;
+
+  for (cast = 0; cast < VNET_N_IP_FEAT; cast++)
+  {
+      ip_config_main_t * cm = &mm->feature_config_mains[cast];
+      vnet_config_main_t * vcm = &cm->config_main;
+
+      if (VNET_IP_RX_MULTICAST_FEAT == cast)
+	  continue;
+
+      vec_validate_init_empty (mm->mpls_enabled_by_sw_if_index, sw_if_index, 0);
+      vec_validate_init_empty (mm->fib_index_by_sw_if_index, sw_if_index, 0);
+      vec_validate_init_empty (cm->config_index_by_sw_if_index, sw_if_index, ~0);
+      ci = cm->config_index_by_sw_if_index[sw_if_index];
+
+       if (cast == VNET_IP_RX_UNICAST_FEAT)
+	   feature_index = mm->mpls_rx_feature_not_enabled;
+       else
+	   feature_index = mm->mpls_tx_feature_interface_output;
+
+
+      if (is_add)
+	  ci = vnet_config_add_feature (vm, vcm, ci,
+					feature_index,
+					/* config data */ 0,
+					/* # bytes of config data */ 0);
+      else
+      {
+	  ci = vnet_config_del_feature (vm, vcm, ci,
+					feature_index,
+					/* config data */ 0,
+					/* # bytes of config data */ 0);
+	  mm->mpls_enabled_by_sw_if_index[sw_if_index] = 0;;
+      }
+      cm->config_index_by_sw_if_index[sw_if_index] = ci;
+  }
 
   return /* no error */ 0;
 }
 
 VNET_SW_INTERFACE_ADD_DEL_FUNCTION (mpls_sw_interface_add_del);
 
+#define foreach_af_cast                         \
+_(VNET_IP_RX_UNICAST_FEAT, "mpls input")        \
+_(VNET_IP_TX_FEAT, "mpls output")               \
+
 static clib_error_t *
 show_mpls_features_command_fn (vlib_main_t * vm,
                                unformat_input_t * input,
@@ -181,11 +222,15 @@ show_mpls_features_command_fn (vlib_main_t * vm,
 
   vlib_cli_output (vm, "Available MPLS feature nodes");
 
-  do {
-    features = mm->feature_nodes;
-    for (i = 0; i < vec_len(features); i++)
-      vlib_cli_output (vm, "  %s\n", features[i]);
+#define _(c,s)                                          \
+  do {                                                  \
+    features = mm->feature_nodes[c];                    \
+    vlib_cli_output (vm, "%s:", s);                     \
+    for (i = 0; i < vec_len(features); i++)             \
+      vlib_cli_output (vm, "  %s\n", features[i]);      \
   } while(0);
+  foreach_af_cast;
+#undef _
 
   return 0;
 }
@@ -202,18 +247,7 @@ show_mpls_interface_features_command_fn (vlib_main_t * vm,
                                          vlib_cli_command_t * cmd)
 {
   vnet_main_t * vnm = vnet_get_main();
-  mpls_main_t * mm = &mpls_main;
-
-  ip_config_main_t * cm;
-  vnet_config_main_t * vcm;
-  vnet_config_t * cfg;
-  u32 cfg_index;
-  vnet_config_feature_t * feat;
-  vlib_node_t * n;
   u32 sw_if_index;
-  u32 node_index;
-  u32 current_config_index;
-  int i;
 
   if (! unformat (input, "%U", unformat_vnet_sw_interface,
                   vnm, &sw_if_index))
@@ -222,26 +256,9 @@ show_mpls_interface_features_command_fn (vlib_main_t * vm,
   vlib_cli_output (vm, "MPLS feature paths configured on %U...",
                    format_vnet_sw_if_index_name, vnm, sw_if_index);
 
-  cm = &mm->rx_config_mains;
-  vcm = &cm->config_main;
-
-  current_config_index = vec_elt (cm->config_index_by_sw_if_index,
-                                  sw_if_index);
-
-  ASSERT(current_config_index
-         < vec_len (vcm->config_pool_index_by_user_index));
-
-  cfg_index =
-      vcm->config_pool_index_by_user_index[current_config_index];
-  cfg = pool_elt_at_index (vcm->config_pool, cfg_index);
-
-  for (i = 0; i < vec_len(cfg->features); i++)
-  {
-      feat = cfg->features + i;
-      node_index = feat->node_index;
-      n = vlib_get_node (vm, node_index);
-      vlib_cli_output (vm, "  %v", n->name);
-  }
+  ip_interface_features_show (vm, "MPLS", 
+			      mpls_main.feature_config_mains,
+			      sw_if_index);
 
   return 0;
 }
diff --git a/vnet/vnet/mpls/node.c b/vnet/vnet/mpls/node.c
index 6801cc7b..1b435f31 100644
--- a/vnet/vnet/mpls/node.c
+++ b/vnet/vnet/mpls/node.c
@@ -110,7 +110,7 @@ mpls_input_inline (vlib_main_t * vm,
           h0 = vlib_buffer_get_current (b0);
 	  sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
 
-	  cm0 = &mm->rx_config_mains;
+	  cm0 = &mm->feature_config_mains[VNET_IP_RX_UNICAST_FEAT];
 	  b0->current_config_index = vec_elt (cm0->config_index_by_sw_if_index,
 					      sw_if_index0);