Distributed Virtual Router Support

A distributed virtual router works by attmpeting to switch a packet, but on failing to find a local consumer (i.e. the packet is destined to a locally attached host) then the packet is sent unmodified 'upstream' to where the rest of the 'distributed' router is present. When L3 switching a packet this means the L2 header must not be modifed. This patch adds a 'l2-bridge' object to the L3 FIB which re-injects packets from the L3 path back into the L2 path - use with extreme caution.

Change-Id: I069724eb45956647d7980cbe40a80a788ee6ee82
Signed-off-by: Neale Ranns <nranns@cisco.com>

5 files changed, 100 insertions, 100 deletions

diff --git a/src/vnet/fib/fib_api.h b/src/vnet/fib/fib_api.h
index f5a107ca352..c369e8f8588 100644
--- a/src/vnet/fib/fib_api.h
+++ b/src/vnet/fib/fib_api.h
@@ -40,6 +40,7 @@ add_del_route_t_handler (u8 is_multipath,
 			 u8 is_resolve_attached,
 			 u8 is_interface_rx,
                          u8 is_rpf_id,
+                         u8 is_l2_bridged,
 			 u32 fib_index,
 			 const fib_prefix_t * prefix,
 			 dpo_proto_t next_hop_proto,
diff --git a/src/vnet/fib/fib_path.c b/src/vnet/fib/fib_path.c
index f126333425a..889d17def9c 100644
--- a/src/vnet/fib/fib_path.c
+++ b/src/vnet/fib/fib_path.c
@@ -23,6 +23,7 @@
 #include <vnet/dpo/lookup_dpo.h>
 #include <vnet/dpo/interface_rx_dpo.h>
 #include <vnet/dpo/mpls_disposition.h>
+#include <vnet/dpo/l2_bridge_dpo.h>
 
 #include <vnet/adj/adj.h>
 #include <vnet/adj/adj_mcast.h>
@@ -771,11 +772,18 @@ fib_path_unresolve (fib_path_t *path)
 	}
 	break;
     case FIB_PATH_TYPE_ATTACHED_NEXT_HOP:
-    case FIB_PATH_TYPE_ATTACHED:
 	adj_child_remove(path->fp_dpo.dpoi_index,
 			 path->fp_sibling);
         adj_unlock(path->fp_dpo.dpoi_index);
         break;
+    case FIB_PATH_TYPE_ATTACHED:
+        if (DPO_PROTO_ETHERNET != path->fp_nh_proto)
+        {
+            adj_child_remove(path->fp_dpo.dpoi_index,
+                             path->fp_sibling);
+            adj_unlock(path->fp_dpo.dpoi_index);
+        }
+        break;
     case FIB_PATH_TYPE_EXCLUSIVE:
 	dpo_reset(&path->exclusive.fp_ex_dpo);
         break;
@@ -1594,28 +1602,35 @@ fib_path_resolve (fib_node_index_t path_index)
 	fib_path_attached_next_hop_set(path);
 	break;
     case FIB_PATH_TYPE_ATTACHED:
-	/*
-	 * path->attached.fp_interface
-	 */
-	if (!vnet_sw_interface_is_admin_up(vnet_get_main(),
-					   path->attached.fp_interface))
-	{
-	    path->fp_oper_flags &= ~FIB_PATH_OPER_FLAG_RESOLVED;
-	}
-        dpo_set(&path->fp_dpo,
-                DPO_ADJACENCY,
-                path->fp_nh_proto,
-                fib_path_attached_get_adj(path,
-                                          dpo_proto_to_link(path->fp_nh_proto)));
-
-	/*
-	 * become a child of the adjacency so we receive updates
-	 * when the interface state changes
-	 */
-	path->fp_sibling = adj_child_add(path->fp_dpo.dpoi_index,
-					 FIB_NODE_TYPE_PATH,
-					 fib_path_get_index(path));
+        if (DPO_PROTO_ETHERNET == path->fp_nh_proto)
+        {
+            l2_bridge_dpo_add_or_lock(path->attached.fp_interface,
+                                      &path->fp_dpo);
+        }
+        else
+        {
+            /*
+             * path->attached.fp_interface
+             */
+            if (!vnet_sw_interface_is_admin_up(vnet_get_main(),
+                                               path->attached.fp_interface))
+            {
+                path->fp_oper_flags &= ~FIB_PATH_OPER_FLAG_RESOLVED;
+            }
+            dpo_set(&path->fp_dpo,
+                    DPO_ADJACENCY,
+                    path->fp_nh_proto,
+                    fib_path_attached_get_adj(path,
+                                              dpo_proto_to_link(path->fp_nh_proto)));
 
+            /*
+             * become a child of the adjacency so we receive updates
+             * when the interface state changes
+             */
+            path->fp_sibling = adj_child_add(path->fp_dpo.dpoi_index,
+                                             FIB_NODE_TYPE_PATH,
+                                             fib_path_get_index(path));
+        }
 	break;
     case FIB_PATH_TYPE_RECURSIVE:
     {
@@ -1996,6 +2011,11 @@ fib_path_contribute_forwarding (fib_node_index_t path_index,
 	    dpo_copy(dpo, &path->exclusive.fp_ex_dpo);
 	    break;
         case FIB_PATH_TYPE_ATTACHED:
+            if (DPO_PROTO_ETHERNET == path->fp_nh_proto)
+            {
+                dpo_copy(dpo, &path->fp_dpo);
+                break;
+            }
 	    switch (fct)
 	    {
 	    case FIB_FORW_CHAIN_TYPE_MPLS_NON_EOS:
diff --git a/src/vnet/fib/fib_table.c b/src/vnet/fib/fib_table.c
index 6daa61c2612..d5625d83673 100644
--- a/src/vnet/fib/fib_table.c
+++ b/src/vnet/fib/fib_table.c
@@ -481,7 +481,8 @@ fib_table_route_path_fixup (const fib_prefix_t *prefix,
     }
     if (fib_prefix_is_host(prefix) &&
 	ip46_address_is_zero(&path->frp_addr) &&
-	path->frp_sw_if_index != ~0)
+	path->frp_sw_if_index != ~0 &&
+        path->frp_proto != DPO_PROTO_ETHERNET)
     {
 	path->frp_addr = prefix->fp_addr;
         path->frp_flags |= FIB_ROUTE_PATH_ATTACHED;
diff --git a/src/vnet/fib/fib_test.c b/src/vnet/fib/fib_test.c
index 64d9047163b..03c9ee75f48 100644
--- a/src/vnet/fib/fib_test.c
+++ b/src/vnet/fib/fib_test.c
@@ -27,9 +27,11 @@
 #include <vnet/bfd/bfd_main.h>
 #include <vnet/dpo/interface_rx_dpo.h>
 #include <vnet/dpo/replicate_dpo.h>
+#include <vnet/dpo/l2_bridge_dpo.h>
 
 #include <vnet/mpls/mpls.h>
 
+#include <vnet/fib/fib_test.h>
 #include <vnet/fib/fib_path_list.h>
 #include <vnet/fib/fib_entry_src.h>
 #include <vnet/fib/fib_walk.h>
@@ -266,83 +268,6 @@ fib_test_build_rewrite (u8 *eth_addr)
     return (rewrite);
 }
 
-typedef enum fib_test_lb_bucket_type_t_ {
-    FT_LB_LABEL_O_ADJ,
-    FT_LB_LABEL_STACK_O_ADJ,
-    FT_LB_LABEL_O_LB,
-    FT_LB_O_LB,
-    FT_LB_SPECIAL,
-    FT_LB_ADJ,
-    FT_LB_INTF,
-} fib_test_lb_bucket_type_t;
-
-typedef struct fib_test_lb_bucket_t_ {
-    fib_test_lb_bucket_type_t type;
-
-    union
-    {
-	struct
-	{
-	    mpls_eos_bit_t eos;
-	    mpls_label_t label;
-	    u8 ttl;
-	    adj_index_t adj;
-	} label_o_adj;
-	struct
-	{
-	    mpls_eos_bit_t eos;
-	    mpls_label_t label_stack[8];
-	    u8 label_stack_size;
-	    u8 ttl;
-	    adj_index_t adj;
-	} label_stack_o_adj;
-	struct
-	{
-	    mpls_eos_bit_t eos;
-	    mpls_label_t label;
-	    u8 ttl;
-	    index_t lb;
-	} label_o_lb;
-	struct
-	{
-	    index_t adj;
-	} adj;
-	struct
-	{
-	    index_t lb;
-	} lb;
-	struct
-	{
-	    index_t adj;
-	} special;
-    };
-} fib_test_lb_bucket_t;
-
-typedef enum fib_test_rep_bucket_type_t_ {
-    FT_REP_LABEL_O_ADJ,
-    FT_REP_DISP_MFIB_LOOKUP,
-    FT_REP_INTF,
-} fib_test_rep_bucket_type_t;
-
-typedef struct fib_test_rep_bucket_t_ {
-    fib_test_rep_bucket_type_t type;
-
-    union
-    {
-	struct
-	{
-	    mpls_eos_bit_t eos;
-	    mpls_label_t label;
-	    u8 ttl;
-	    adj_index_t adj;
-	} label_o_adj;
- 	struct
-	{
-	    adj_index_t adj;
-	} adj;
-   };
-} fib_test_rep_bucket_t;
-
 #define FIB_TEST_LB(_cond, _comment, _args...)			\
 {								\
     if (!FIB_TEST_I(_cond, _comment, ##_args)) {		\
@@ -598,6 +523,16 @@ fib_test_validate_lb_v (const load_balance_t *lb,
 			bucket,
 			exp->adj.adj);
 	    break;
+	case FT_LB_L2:
+	    FIB_TEST_I((DPO_L2_BRIDGE == dpo->dpoi_type),
+		       "bucket %d stacks on %U",
+		       bucket,
+		       format_dpo_type, dpo->dpoi_type);
+	    FIB_TEST_LB((exp->adj.adj == dpo->dpoi_index),
+			"bucket %d stacks on adj %d",
+			bucket,
+			exp->adj.adj);
+	    break;
 	case FT_LB_O_LB:
 	    FIB_TEST_I((DPO_LOAD_BALANCE == dpo->dpoi_type),
                        "bucket %d stacks on %U",
@@ -4067,6 +4002,45 @@ fib_test_v4 (void)
              format_ip_flow_hash_config, lb->lb_hash_config);
 
     /*
+     * A route via an L2 Bridge
+     */
+    fei = fib_table_entry_path_add(fib_index,
+                                   &pfx_10_10_10_3_s_32,
+                                   FIB_SOURCE_API,
+                                   FIB_ENTRY_FLAG_NONE,
+                                   DPO_PROTO_ETHERNET,
+                                   &zero_addr,
+                                   tm->hw[0]->sw_if_index,
+                                   ~0,
+                                   1,
+                                   NULL,
+                                   FIB_ROUTE_PATH_FLAG_NONE);
+    dpo_id_t l2_dpo = DPO_INVALID;
+    l2_bridge_dpo_add_or_lock(tm->hw[0]->sw_if_index, &l2_dpo);
+    fib_test_lb_bucket_t ip_o_l2 = {
+        .type = FT_LB_L2,
+        .adj = {
+            .adj = l2_dpo.dpoi_index,
+        },
+    };
+
+    FIB_TEST(fib_test_validate_entry(fei,
+                                     FIB_FORW_CHAIN_TYPE_UNICAST_IP4,
+                                     1,
+                                     &ip_o_l2),
+             "10.10.10.3 via L2 on Eth0");
+    fib_table_entry_path_remove(fib_index,
+                                &pfx_10_10_10_3_s_32,
+                                FIB_SOURCE_API,
+                                DPO_PROTO_ETHERNET,
+                                &zero_addr,
+                                tm->hw[0]->sw_if_index,
+                                fib_index,
+                                1,
+                                FIB_ROUTE_PATH_FLAG_NONE);
+    dpo_reset(&l2_dpo);
+
+    /*
      * CLEANUP
      *    remove adj-fibs: 
      */
@@ -4165,6 +4139,8 @@ fib_test_v4 (void)
     	     pool_elts(load_balance_map_pool));
     FIB_TEST((lb_count == pool_elts(load_balance_pool)), "LB pool size is %d",
              pool_elts(load_balance_pool));
+    FIB_TEST((0 == pool_elts(l2_bridge_dpo_pool)), "L2 DPO pool size is %d",
+             pool_elts(l2_bridge_dpo_pool));
 
     return 0;
 }
diff --git a/src/vnet/fib/fib_test.h b/src/vnet/fib/fib_test.h
index 3692f57386d..5adc52ec658 100644
--- a/src/vnet/fib/fib_test.h
+++ b/src/vnet/fib/fib_test.h
@@ -26,6 +26,7 @@ typedef enum fib_test_lb_bucket_type_t_ {
     FT_LB_SPECIAL,
     FT_LB_ADJ,
     FT_LB_INTF,
+    FT_LB_L2,
 } fib_test_lb_bucket_type_t;
 
 typedef struct fib_test_lb_bucket_t_ {
@@ -72,6 +73,7 @@ typedef struct fib_test_lb_bucket_t_ {
 
 typedef enum fib_test_rep_bucket_type_t_ {
     FT_REP_LABEL_O_ADJ,
+    FT_REP_DISP_MFIB_LOOKUP,
     FT_REP_INTF,
 } fib_test_rep_bucket_type_t;