Augment Vxlan to support NSH-Proxy

PatchSet 9: add range check for decap_next_index
PatchSet 6: delete runtime range check

Change-Id: I415e156d05d09d2ff34f6578924f0b243058f464
Signed-off-by: Hongjun Ni <hongjun.ni@intel.com>

6 files changed, 87 insertions, 13 deletions

diff --git a/vnet/vnet/vxlan-gpe/encap.c b/vnet/vnet/vxlan-gpe/encap.c
index 3f6f43c564c..9cd2c722aa1 100644
--- a/vnet/vnet/vxlan-gpe/encap.c
+++ b/vnet/vnet/vxlan-gpe/encap.c
@@ -407,7 +407,7 @@ VLIB_REGISTER_NODE (vxlan_gpe_encap_node) = {
 
   .next_nodes = {
     [VXLAN_GPE_ENCAP_NEXT_IP4_LOOKUP] = "ip4-lookup",
-		[VXLAN_GPE_ENCAP_NEXT_IP6_LOOKUP] = "ip6-lookup",
+    [VXLAN_GPE_ENCAP_NEXT_IP6_LOOKUP] = "ip6-lookup",
     [VXLAN_GPE_ENCAP_NEXT_DROP] = "error-drop",
   },
 };
diff --git a/vnet/vnet/vxlan/decap.c b/vnet/vnet/vxlan/decap.c
index 05565504669..22f2f85b2bc 100644
--- a/vnet/vnet/vxlan/decap.c
+++ b/vnet/vnet/vxlan/decap.c
@@ -217,7 +217,7 @@ vxlan_input (vlib_main_t * vm,
           }
 
           t0 = pool_elt_at_index (vxm->tunnels, tunnel_index0);
-
+          next0 = t0->decap_next_index;
           sw_if_index0 = t0->sw_if_index;
           len0 = vlib_buffer_length_in_chain (vm, b0);
 
@@ -312,7 +312,7 @@ vxlan_input (vlib_main_t * vm,
           }
 
           t1 = pool_elt_at_index (vxm->tunnels, tunnel_index1);
-
+          next1 = t1->decap_next_index;
           sw_if_index1 = t1->sw_if_index;
           len1 = vlib_buffer_length_in_chain (vm, b1);
 
@@ -464,7 +464,7 @@ vxlan_input (vlib_main_t * vm,
           }
 
           t0 = pool_elt_at_index (vxm->tunnels, tunnel_index0);
-
+          next0 = t0->decap_next_index;
           sw_if_index0 = t0->sw_if_index;
           len0 = vlib_buffer_length_in_chain (vm, b0);
 
diff --git a/vnet/vnet/vxlan/vxlan.c b/vnet/vnet/vxlan/vxlan.c
index bda0bbc968a..a2b89208116 100644
--- a/vnet/vnet/vxlan/vxlan.c
+++ b/vnet/vnet/vxlan/vxlan.c
@@ -17,6 +17,7 @@
 #include <vnet/fib/fib_entry.h>
 #include <vnet/fib/fib_table.h>
 #include <vnet/dpo/receive_dpo.h>
+#include <vlib/vlib.h>
 
 /**
  * @file
@@ -37,6 +38,22 @@
 
 vxlan_main_t vxlan_main;
 
+static u8 * format_decap_next (u8 * s, va_list * args)
+{
+  u32 next_index = va_arg (*args, u32);
+
+  switch (next_index)
+    {
+    case VXLAN_INPUT_NEXT_DROP:
+      return format (s, "drop");
+    case VXLAN_INPUT_NEXT_L2_INPUT:
+      return format (s, "l2");
+    default:
+      return format (s, "next-index %d", next_index);
+    }
+  return s;
+}
+
 u8 * format_vxlan_tunnel (u8 * s, va_list * args)
 {
   vxlan_tunnel_t * t = va_arg (*args, vxlan_tunnel_t *);
@@ -44,11 +61,12 @@ u8 * format_vxlan_tunnel (u8 * s, va_list * args)
 
   s = format (s, 
 	      "[%d] src %U dst %U vni %d encap_fib_index %d sw_if_index %d "
-	      "fib_entry_index %d\n",
+	      "fib_entry_index %d",
               t - ngm->tunnels,
               format_ip46_address, &t->src, IP46_TYPE_ANY,
               format_ip46_address, &t->dst, IP46_TYPE_ANY,
               t->vni,  t->encap_fib_index, t->sw_if_index, t->fib_entry_index);
+  s = format (s, " decap_next %U\n", format_decap_next, t->decap_next_index);
   return s;
 }
 
@@ -176,6 +194,7 @@ const static fib_node_vft_t vxlan_vft = {
 _(vni)                                          \
 _(mcast_sw_if_index)                            \
 _(encap_fib_index)                              \
+_(decap_next_index)                             \
 _(src)                                          \
 _(dst)
 
@@ -245,6 +264,27 @@ static int vxlan6_rewrite (vxlan_tunnel_t * t)
   return (0);
 }
 
+static int vxlan_check_decap_next(vxlan_main_t * vxm, u32 is_ip6, u32 decap_next_index)
+{
+  vlib_main_t * vm = vxm->vlib_main;
+  vlib_node_runtime_t *r;
+
+  if(!is_ip6)
+    {
+      r = vlib_node_get_runtime (vm, vxlan4_input_node.index);
+      if(decap_next_index >= r->n_next_nodes)
+	return 1;
+    }
+  else
+    {
+      r = vlib_node_get_runtime (vm, vxlan6_input_node.index);
+      if(decap_next_index >= r->n_next_nodes)
+	return 1;
+    }
+
+  return 0;
+}
+
 int vnet_vxlan_add_del_tunnel 
 (vnet_vxlan_add_del_tunnel_args_t *a, u32 * sw_if_indexp)
 {
@@ -283,6 +323,12 @@ int vnet_vxlan_add_del_tunnel
       if (p)
         return VNET_API_ERROR_TUNNEL_EXIST;
 
+      /*if not set explicitly, default to l2 */
+      if(a->decap_next_index == ~0)
+	a->decap_next_index = VXLAN_INPUT_NEXT_L2_INPUT;
+      if (vxlan_check_decap_next(vxm, is_ip6, a->decap_next_index))
+	  return VNET_API_ERROR_INVALID_DECAP_NEXT;
+
       pool_get_aligned (vxm->tunnels, t, CLIB_CACHE_LINE_BYTES);
       memset (t, 0, sizeof (*t));
       
@@ -496,13 +542,37 @@ static u32 fib6_index_from_fib_id (u32 fib_id)
   return p[0];
 }
 
+static uword get_decap_next_for_node(u32 node_index, u32 ipv4_set)
+{
+  vxlan_main_t * vxm = &vxlan_main;
+  vlib_main_t * vm = vxm->vlib_main;
+  uword next_index = ~0;
+
+  if (ipv4_set)
+    {
+      next_index = vlib_node_add_next (vm, vxlan4_input_node.index, node_index);
+    }
+  else
+    {
+      next_index = vlib_node_add_next (vm, vxlan6_input_node.index, node_index);
+    }
+
+  return next_index;
+}
+
 static uword unformat_decap_next (unformat_input_t * input, va_list * args)
 {
   u32 * result = va_arg (*args, u32 *);
+  u32 ipv4_set = va_arg (*args, int);
+  vxlan_main_t * vxm = &vxlan_main;
+  vlib_main_t * vm = vxm->vlib_main;
+  u32 node_index;
   u32 tmp;
-  
+
   if (unformat (input, "l2"))
     *result = VXLAN_INPUT_NEXT_L2_INPUT;
+  else if (unformat (input, "node %U", unformat_vlib_node, vm, &node_index))
+    *result = get_decap_next_for_node(node_index, ipv4_set);
   else if (unformat (input, "%d", &tmp))
     *result = tmp;
   else
@@ -525,7 +595,7 @@ vxlan_add_del_tunnel_command_fn (vlib_main_t * vm,
   u8 ipv6_set = 0;
   u32 encap_fib_index = 0;
   u32 mcast_sw_if_index = ~0;
-  u32 decap_next_index = ~0;
+  u32 decap_next_index = VXLAN_INPUT_NEXT_L2_INPUT;
   u32 vni = 0;
   u32 tmp;
   int rv;
@@ -595,7 +665,7 @@ vxlan_add_del_tunnel_command_fn (vlib_main_t * vm,
           return clib_error_return (0, "nonexistent encap-vrf-id %d", tmp);
       }
     else if (unformat (line_input, "decap-next %U", unformat_decap_next, 
-                       &decap_next_index))
+                       &decap_next_index, ipv4_set))
       ;
     else if (unformat (line_input, "vni %d", &vni))
       {
@@ -627,6 +697,9 @@ vxlan_add_del_tunnel_command_fn (vlib_main_t * vm,
   if (ip46_address_cmp(&src, &dst) == 0)
     return clib_error_return (0, "src and dst addresses are identical");
 
+  if (decap_next_index == ~0)
+    return clib_error_return (0, "next node not found");
+
   if (vni == 0)
     return clib_error_return (0, "vni not specified");
 
@@ -689,7 +762,7 @@ VLIB_CLI_COMMAND (create_vxlan_tunnel_command, static) = {
   .short_help = 
   "create vxlan tunnel src <local-vtep-addr>"
   " {dst <remote-vtep-addr>|group <mcast-vtep-addr> <intf-name>} vni <nn>"
-  " [encap-vrf-id <nn>]",
+  " [encap-vrf-id <nn>] [decap-next [l2|node <name>]] [del]",
   .function = vxlan_add_del_tunnel_command_fn,
 };
 /* *INDENT-ON* */
@@ -719,7 +792,7 @@ show_vxlan_tunnel_command_fn (vlib_main_t * vm,
  * @cliexpar
  * Example of how to display the VXLAN Tunnel entries:
  * @cliexstart{show vxlan tunnel}
- * [0] src 10.0.3.1 dst 10.0.3.3 vni 13 encap_fib_index 0 sw_if_index 5 
+ * [0] src 10.0.3.1 dst 10.0.3.3 vni 13 encap_fib_index 0 sw_if_index 5 decap_next l2
  * @cliexend
  ?*/
 /* *INDENT-OFF* */
diff --git a/vnet/vnet/vxlan/vxlan.h b/vnet/vnet/vxlan/vxlan.h
index 3041df27dcc..fe31ce160b8 100644
--- a/vnet/vnet/vxlan/vxlan.h
+++ b/vnet/vnet/vxlan/vxlan.h
@@ -86,6 +86,8 @@ typedef struct {
   ip46_address_t dst;
 
   u32 mcast_sw_if_index;
+  /* decap next index */
+  u32 decap_next_index;
 
   /* The FIB index for src/dst addresses */
   u32 encap_fib_index;
diff --git a/vpp-api-test/vat/api_format.c b/vpp-api-test/vat/api_format.c
index 9d80c42aba2..2925c9bf26e 100644
--- a/vpp-api-test/vat/api_format.c
+++ b/vpp-api-test/vat/api_format.c
@@ -16808,7 +16808,7 @@ _(sw_if_l2tpv3_tunnel_dump, "")                                         \
 _(vxlan_add_del_tunnel,                                                 \
   "src <ip-addr> { dst <ip-addr> | group <mcast-ip-addr>\n"             \
   "{ <intfc> | mcast_sw_if_index <nn> } }\n"                            \
-  "vni <vni> [encap-vrf-id <nn>] [decap-next l2|ip4|ip6] [del]")        \
+  "vni <vni> [encap-vrf-id <nn>] [decap-next <l2|nn>] [del]")           \
 _(vxlan_tunnel_dump, "[<intfc> | sw_if_index <nn>]")                    \
 _(gre_add_del_tunnel,                                                   \
   "src <ip4-addr> dst <ip4-addr> [outer-fib-id <nn>] [teb] [del]\n")    \
diff --git a/vpp/vpp-api/api.c b/vpp/vpp-api/api.c
index a011c650b8e..c110769b620 100644
--- a/vpp/vpp-api/api.c
+++ b/vpp/vpp-api/api.c
@@ -4545,8 +4545,7 @@ static void send_vxlan_tunnel_details
     }
   rmp->mcast_sw_if_index = htonl (t->mcast_sw_if_index);
   rmp->vni = htonl (t->vni);
-  /* decap_next_index is deprecated, hard code to l2-input */
-  rmp->decap_next_index = htonl (VXLAN_INPUT_NEXT_L2_INPUT);
+  rmp->decap_next_index = htonl (t->decap_next_index);
   rmp->sw_if_index = htonl (t->sw_if_index);
   rmp->is_ipv6 = is_ipv6;
   rmp->context = context;