vxlan: Fix L3 mode

Type: fix

Partially revert fix SEGV reported in VPP-1962
[commit a4b0541f64eef02fa0d003d8f831cfdeb45d3668]

This adds an is_l3 option to choose between L2 & L3
mode add tunnel creation time

Change-Id: Ia2c91a1099074b7d23fc031b78ed0f68628eeabe
Signed-off-by: Nathan Skrzypczak <nathan.skrzypczak@gmail.com>

5 files changed, 158 insertions, 103 deletions

diff --git a/src/vnet/vxlan/test/vpp_vxlan_tunnel.py b/src/vnet/vxlan/test/vpp_vxlan_tunnel.py
index dce5ab86c07..d7e087da6f8 100644
--- a/src/vnet/vxlan/test/vpp_vxlan_tunnel.py
+++ b/src/vnet/vxlan/test/vpp_vxlan_tunnel.py
@@ -38,7 +38,7 @@ class VppVxlanTunnel(VppInterface):
                  mcast_itf=None,
                  mcast_sw_if_index=INDEX_INVALID,
                  decap_next_index=INDEX_INVALID,
-                 encap_vrf_id=None, instance=0xffffffff):
+                 encap_vrf_id=None, instance=0xffffffff, is_l3=False):
         """ Create VXLAN Tunnel interface """
         super(VppVxlanTunnel, self).__init__(test)
         self.src = src
@@ -51,16 +51,17 @@ class VppVxlanTunnel(VppInterface):
         self.encap_vrf_id = encap_vrf_id
         self.decap_next_index = decap_next_index
         self.instance = instance
+        self.is_l3 = is_l3
 
         if (self.mcast_itf):
             self.mcast_sw_if_index = self.mcast_itf.sw_if_index
 
     def add_vpp_config(self):
-        reply = self.test.vapi.vxlan_add_del_tunnel_v2(
+        reply = self.test.vapi.vxlan_add_del_tunnel_v3(
             is_add=1, src_address=self.src, dst_address=self.dst, vni=self.vni,
             src_port=self.src_port, dst_port=self.dst_port,
             mcast_sw_if_index=self.mcast_sw_if_index,
-            encap_vrf_id=self.encap_vrf_id,
+            encap_vrf_id=self.encap_vrf_id, is_l3=self.is_l3,
             instance=self.instance, decap_next_index=self.decap_next_index)
         self.set_sw_if_index(reply.sw_if_index)
         self._test.registry.register(self, self._test.logger)
diff --git a/src/vnet/vxlan/vxlan.api b/src/vnet/vxlan/vxlan.api
index 2370fb5157c..b7e678595d8 100644
--- a/src/vnet/vxlan/vxlan.api
+++ b/src/vnet/vxlan/vxlan.api
@@ -74,6 +74,38 @@ define vxlan_add_del_tunnel_v2
   u32 vni;
 };
 
+/** \brief Create or delete a VXLAN tunnel
+    @param client_index - opaque cookie to identify the sender
+    @param context - sender context, to match reply w/ request
+    @param is_add - Use 1 to create the tunnel, 0 to remove it
+    @param instance - optional unique custom device instance, else ~0.
+    @param src_address - Source IP address
+    @param dst_address - Destination IP address, can be multicast
+    @param src_port - Source UDP port. It is not included in sent packets. Used only for port registration
+    @param dst_port - Destination UDP port
+    @param mcast_sw_if_index - Interface for multicast destination
+    @param encap_vrf_id - Encap route table FIB index
+    @param decap_next_index - index of decap next graph node
+    @param vni - The VXLAN Network Identifier, uint24
+    @param is_l3 - if true, create the interface in L3 mode, w/o MAC
+*/
+define vxlan_add_del_tunnel_v3
+{
+  u32 client_index;
+  u32 context;
+  bool is_add [default=true];
+  u32 instance [default=0xffffffff];	/* If non-~0, specifies a custom dev instance */
+  vl_api_address_t src_address;
+  vl_api_address_t dst_address;
+  u16 src_port;
+  u16 dst_port;
+  vl_api_interface_index_t mcast_sw_if_index;
+  u32 encap_vrf_id;
+  u32 decap_next_index;
+  u32 vni;
+  bool is_l3 [default=false];
+};
+
 define vxlan_add_del_tunnel_reply
 {
   u32 context;
@@ -86,6 +118,12 @@ define vxlan_add_del_tunnel_v2_reply
   i32 retval;
   vl_api_interface_index_t sw_if_index;
 };
+define vxlan_add_del_tunnel_v3_reply
+{
+  u32 context;
+  i32 retval;
+  vl_api_interface_index_t sw_if_index;
+};
 
 define vxlan_tunnel_dump
 {
diff --git a/src/vnet/vxlan/vxlan.c b/src/vnet/vxlan/vxlan.c
index 97098b83774..dcf480578a7 100644
--- a/src/vnet/vxlan/vxlan.c
+++ b/src/vnet/vxlan/vxlan.c
@@ -442,29 +442,34 @@ int vnet_vxlan_add_del_tunnel
 	  return VNET_API_ERROR_INSTANCE_IN_USE;
 	}
 
-      f64 now = vlib_time_now (vm);
-      u32 rnd;
-      rnd = (u32) (now * 1e6);
-      rnd = random_u32 (&rnd);
-
-      memcpy (hw_addr + 2, &rnd, sizeof (rnd));
-      hw_addr[0] = 2;
-      hw_addr[1] = 0xfe;
-
       hash_set (vxm->instance_used, user_instance, 1);
 
       t->dev_instance = dev_instance;	/* actual */
       t->user_instance = user_instance; /* name */
       t->flow_index = ~0;
 
-      if (ethernet_register_interface (vnm, vxlan_device_class.index,
-				       dev_instance, hw_addr, &t->hw_if_index,
-				       vxlan_eth_flag_change))
+      if (a->is_l3)
+	t->hw_if_index =
+	  vnet_register_interface (vnm, vxlan_device_class.index, dev_instance,
+				   vxlan_hw_class.index, dev_instance);
+      else
 	{
-	  hash_unset (vxm->instance_used, t->user_instance);
+	  f64 now = vlib_time_now (vm);
+	  u32 rnd;
+	  rnd = (u32) (now * 1e6);
+	  rnd = random_u32 (&rnd);
+	  memcpy (hw_addr + 2, &rnd, sizeof (rnd));
+	  hw_addr[0] = 2;
+	  hw_addr[1] = 0xfe;
+	  if (ethernet_register_interface (
+		vnm, vxlan_device_class.index, dev_instance, hw_addr,
+		&t->hw_if_index, vxlan_eth_flag_change))
+	    {
+	      hash_unset (vxm->instance_used, t->user_instance);
 
-	  pool_put (vxm->tunnels, t);
-	  return VNET_API_ERROR_SYSCALL_ERROR_2;
+	      pool_put (vxm->tunnels, t);
+	      return VNET_API_ERROR_SYSCALL_ERROR_2;
+	    }
 	}
 
       vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, t->hw_if_index);
@@ -498,7 +503,10 @@ int vnet_vxlan_add_del_tunnel
 
       if (add_failed)
 	{
-	  ethernet_delete_interface (vnm, t->hw_if_index);
+	  if (a->is_l3)
+	    vnet_delete_hw_interface (vnm, t->hw_if_index);
+	  else
+	    ethernet_delete_interface (vnm, t->hw_if_index);
 	  hash_unset (vxm->instance_used, t->user_instance);
 	  pool_put (vxm->tunnels, t);
 	  return VNET_API_ERROR_INVALID_REGISTRATION;
@@ -647,7 +655,11 @@ int vnet_vxlan_add_del_tunnel
 	  mcast_shared_remove (&t->dst);
 	}
 
-      ethernet_delete_interface (vnm, t->hw_if_index);
+      vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, t->hw_if_index);
+      if (hw->dev_class_index == vxlan_device_class.index)
+	vnet_delete_hw_interface (vnm, t->hw_if_index);
+      else
+	ethernet_delete_interface (vnm, t->hw_if_index);
       hash_unset (vxm->instance_used, t->user_instance);
 
       fib_node_deinit (&t->node);
@@ -717,6 +729,7 @@ vxlan_add_del_tunnel_command_fn (vlib_main_t * vm,
   u8 grp_set = 0;
   u8 ipv4_set = 0;
   u8 ipv6_set = 0;
+  u8 is_l3 = 0;
   u32 instance = ~0;
   u32 encap_fib_index = 0;
   u32 mcast_sw_if_index = ~0;
@@ -764,6 +777,8 @@ vxlan_add_del_tunnel_command_fn (vlib_main_t * vm,
 	  encap_fib_index =
 	    fib_table_find (fib_ip_proto (ipv6_set), table_id);
 	}
+      else if (unformat (line_input, "l3"))
+	is_l3 = 1;
       else if (unformat (line_input, "decap-next %U", unformat_decap_next,
 			 &decap_next_index, ipv4_set))
 	;
@@ -786,6 +801,13 @@ vxlan_add_del_tunnel_command_fn (vlib_main_t * vm,
   if (parse_error)
     return parse_error;
 
+  if (is_l3 && decap_next_index == VXLAN_INPUT_NEXT_L2_INPUT)
+    {
+      vlib_node_t *node = vlib_get_node_by_name (
+	vm, (u8 *) (ipv4_set ? "ip4-input" : "ip6-input"));
+      decap_next_index = get_decap_next_for_node (node->index, ipv4_set);
+    }
+
   if (encap_fib_index == ~0)
     return clib_error_return (0, "nonexistent encap-vrf-id %d", table_id);
 
@@ -819,12 +841,12 @@ vxlan_add_del_tunnel_command_fn (vlib_main_t * vm,
   if (vni >> 24)
     return clib_error_return (0, "vni %d out of range", vni);
 
-  vnet_vxlan_add_del_tunnel_args_t a = {
-    .is_add = is_add,
-    .is_ip6 = ipv6_set,
-    .instance = instance,
+  vnet_vxlan_add_del_tunnel_args_t a = { .is_add = is_add,
+					 .is_ip6 = ipv6_set,
+					 .is_l3 = is_l3,
+					 .instance = instance,
 #define _(x) .x = x,
-    foreach_copy_field
+					 foreach_copy_field
 #undef _
   };
 
@@ -893,7 +915,7 @@ VLIB_CLI_COMMAND (create_vxlan_tunnel_command, static) = {
     "create vxlan tunnel src <local-vtep-addr>"
     " {dst <remote-vtep-addr>|group <mcast-vtep-addr> <intf-name>} vni <nn>"
     " [instance <id>]"
-    " [encap-vrf-id <nn>] [decap-next [l2|node <name>]] [del]"
+    " [encap-vrf-id <nn>] [decap-next [l2|node <name>]] [del] [l3]"
     " [src_port <local-vtep-udp-port>] [dst_port <remote-vtep-udp-port>]",
   .function = vxlan_add_del_tunnel_command_fn,
 };
diff --git a/src/vnet/vxlan/vxlan.h b/src/vnet/vxlan/vxlan.h
index 129bb43291b..be819ab1069 100644
--- a/src/vnet/vxlan/vxlan.h
+++ b/src/vnet/vxlan/vxlan.h
@@ -214,6 +214,7 @@ typedef struct
   /* we normally use is_ip4, but since this adds to the
    * structure, this seems less of a breaking change */
   u8 is_ip6;
+  u8 is_l3;
   u32 instance;
   ip46_address_t src, dst;
   u32 mcast_sw_if_index;
diff --git a/src/vnet/vxlan/vxlan_api.c b/src/vnet/vxlan/vxlan_api.c
index 6975d834e2a..381bb832f7a 100644
--- a/src/vnet/vxlan/vxlan_api.c
+++ b/src/vnet/vxlan/vxlan_api.c
@@ -52,6 +52,7 @@
   _ (VXLAN_ADD_DEL_TUNNEL, vxlan_add_del_tunnel)                              \
   _ (VXLAN_TUNNEL_DUMP, vxlan_tunnel_dump)                                    \
   _ (VXLAN_ADD_DEL_TUNNEL_V2, vxlan_add_del_tunnel_v2)                        \
+  _ (VXLAN_ADD_DEL_TUNNEL_V3, vxlan_add_del_tunnel_v3)                        \
   _ (VXLAN_TUNNEL_V2_DUMP, vxlan_tunnel_v2_dump)                              \
   _ (VXLAN_OFFLOAD_RX, vxlan_offload_rx)
 
@@ -124,62 +125,58 @@ static void
   REPLY_MACRO (VL_API_SW_INTERFACE_SET_VXLAN_BYPASS_REPLY);
 }
 
-static void vl_api_vxlan_add_del_tunnel_t_handler
-  (vl_api_vxlan_add_del_tunnel_t * mp)
+static int
+vxlan_add_del_tunnel_clean_input (vnet_vxlan_add_del_tunnel_args_t *a,
+				  u32 encap_vrf_id)
 {
-  vl_api_vxlan_add_del_tunnel_reply_t *rmp;
-  int rv = 0;
-  bool is_ipv6;
-  u32 fib_index;
-  ip46_address_t src, dst;
+  a->is_ip6 = !ip46_address_is_ip4 (&a->src);
 
-  ip_address_decode (&mp->src_address, &src);
-  ip_address_decode (&mp->dst_address, &dst);
-
-  if (ip46_address_is_ip4 (&src) != ip46_address_is_ip4 (&dst))
+  a->encap_fib_index = fib_table_find (fib_ip_proto (a->is_ip6), encap_vrf_id);
+  if (a->encap_fib_index == ~0)
     {
-      rv = VNET_API_ERROR_INVALID_VALUE;
-      goto out;
+      return VNET_API_ERROR_NO_SUCH_FIB;
     }
 
-  is_ipv6 = !ip46_address_is_ip4 (&src);
+  if (ip46_address_is_ip4 (&a->src) != ip46_address_is_ip4 (&a->dst))
+    {
+      return VNET_API_ERROR_INVALID_VALUE;
+    }
 
-  fib_index = fib_table_find (fib_ip_proto (is_ipv6),
-			      ntohl (mp->encap_vrf_id));
-  if (fib_index == ~0)
+  /* Check src & dst are different */
+  if (ip46_address_cmp (&a->dst, &a->src) == 0)
     {
-      rv = VNET_API_ERROR_NO_SUCH_FIB;
-      goto out;
+      return VNET_API_ERROR_SAME_SRC_DST;
     }
+  if (ip46_address_is_multicast (&a->dst) &&
+      !vnet_sw_if_index_is_api_valid (a->mcast_sw_if_index))
+    {
+      return VNET_API_ERROR_INVALID_SW_IF_INDEX;
+    }
+  return 0;
+}
+
+static void
+vl_api_vxlan_add_del_tunnel_t_handler (vl_api_vxlan_add_del_tunnel_t *mp)
+{
+  vl_api_vxlan_add_del_tunnel_reply_t *rmp;
+  u32 sw_if_index = ~0;
+  int rv = 0;
 
   vnet_vxlan_add_del_tunnel_args_t a = {
     .is_add = mp->is_add,
-    .is_ip6 = is_ipv6,
     .instance = ntohl (mp->instance),
     .mcast_sw_if_index = ntohl (mp->mcast_sw_if_index),
-    .encap_fib_index = fib_index,
     .decap_next_index = ntohl (mp->decap_next_index),
     .vni = ntohl (mp->vni),
-    .dst = dst,
-    .src = src,
-    .dst_port = is_ipv6 ? UDP_DST_PORT_vxlan6 : UDP_DST_PORT_vxlan,
-    .src_port = is_ipv6 ? UDP_DST_PORT_vxlan6 : UDP_DST_PORT_vxlan,
   };
-
-  /* Check src & dst are different */
-  if (ip46_address_cmp (&a.dst, &a.src) == 0)
-    {
-      rv = VNET_API_ERROR_SAME_SRC_DST;
-      goto out;
-    }
-  if (ip46_address_is_multicast (&a.dst) &&
-      !vnet_sw_if_index_is_api_valid (a.mcast_sw_if_index))
-    {
-      rv = VNET_API_ERROR_INVALID_SW_IF_INDEX;
-      goto out;
-    }
-
-  u32 sw_if_index = ~0;
+  ip_address_decode (&mp->src_address, &a.src);
+  ip_address_decode (&mp->dst_address, &a.dst);
+
+  rv = vxlan_add_del_tunnel_clean_input (&a, ntohl (mp->encap_vrf_id));
+  if (rv)
+    goto out;
+  a.dst_port = a.is_ip6 ? UDP_DST_PORT_vxlan6 : UDP_DST_PORT_vxlan,
+  a.src_port = a.is_ip6 ? UDP_DST_PORT_vxlan6 : UDP_DST_PORT_vxlan,
   rv = vnet_vxlan_add_del_tunnel (&a, &sw_if_index);
 
 out:
@@ -193,62 +190,58 @@ static void
 vl_api_vxlan_add_del_tunnel_v2_t_handler (vl_api_vxlan_add_del_tunnel_v2_t *mp)
 {
   vl_api_vxlan_add_del_tunnel_v2_reply_t *rmp;
+  u32 sw_if_index = ~0;
   int rv = 0;
-  bool is_ipv6;
-  u32 fib_index;
-  ip46_address_t src, dst;
 
-  ip_address_decode (&mp->src_address, &src);
-  ip_address_decode (&mp->dst_address, &dst);
+  vnet_vxlan_add_del_tunnel_args_t a = {
+    .is_add = mp->is_add,
+    .instance = ntohl (mp->instance),
+    .mcast_sw_if_index = ntohl (mp->mcast_sw_if_index),
+    .decap_next_index = ntohl (mp->decap_next_index),
+    .vni = ntohl (mp->vni),
+    .dst_port = ntohs (mp->dst_port),
+    .src_port = ntohs (mp->src_port),
+  };
 
-  if (ip46_address_is_ip4 (&src) != ip46_address_is_ip4 (&dst))
-    {
-      rv = VNET_API_ERROR_INVALID_VALUE;
-      goto out;
-    }
+  ip_address_decode (&mp->src_address, &a.src);
+  ip_address_decode (&mp->dst_address, &a.dst);
 
-  is_ipv6 = !ip46_address_is_ip4 (&src);
+  rv = vxlan_add_del_tunnel_clean_input (&a, ntohl (mp->encap_vrf_id));
+  if (rv)
+    goto out;
+  rv = vnet_vxlan_add_del_tunnel (&a, &sw_if_index);
+out:
+  REPLY_MACRO2 (VL_API_VXLAN_ADD_DEL_TUNNEL_V2_REPLY,
+		({ rmp->sw_if_index = ntohl (sw_if_index); }));
+}
 
-  fib_index =
-    fib_table_find (fib_ip_proto (is_ipv6), ntohl (mp->encap_vrf_id));
-  if (fib_index == ~0)
-    {
-      rv = VNET_API_ERROR_NO_SUCH_FIB;
-      goto out;
-    }
+static void
+vl_api_vxlan_add_del_tunnel_v3_t_handler (vl_api_vxlan_add_del_tunnel_v3_t *mp)
+{
+  vl_api_vxlan_add_del_tunnel_v3_reply_t *rmp;
+  u32 sw_if_index = ~0;
+  int rv = 0;
 
   vnet_vxlan_add_del_tunnel_args_t a = {
     .is_add = mp->is_add,
-    .is_ip6 = is_ipv6,
     .instance = ntohl (mp->instance),
     .mcast_sw_if_index = ntohl (mp->mcast_sw_if_index),
-    .encap_fib_index = fib_index,
     .decap_next_index = ntohl (mp->decap_next_index),
     .vni = ntohl (mp->vni),
-    .dst = dst,
-    .src = src,
     .dst_port = ntohs (mp->dst_port),
     .src_port = ntohs (mp->src_port),
+    .is_l3 = mp->is_l3,
   };
 
-  /* Check src & dst are different */
-  if (ip46_address_cmp (&a.dst, &a.src) == 0)
-    {
-      rv = VNET_API_ERROR_SAME_SRC_DST;
-      goto out;
-    }
-  if (ip46_address_is_multicast (&a.dst) &&
-      !vnet_sw_if_index_is_api_valid (a.mcast_sw_if_index))
-    {
-      rv = VNET_API_ERROR_INVALID_SW_IF_INDEX;
-      goto out;
-    }
+  ip_address_decode (&mp->src_address, &a.src);
+  ip_address_decode (&mp->dst_address, &a.dst);
 
-  u32 sw_if_index = ~0;
+  rv = vxlan_add_del_tunnel_clean_input (&a, ntohl (mp->encap_vrf_id));
+  if (rv)
+    goto out;
   rv = vnet_vxlan_add_del_tunnel (&a, &sw_if_index);
-
 out:
-  REPLY_MACRO2 (VL_API_VXLAN_ADD_DEL_TUNNEL_V2_REPLY,
+  REPLY_MACRO2 (VL_API_VXLAN_ADD_DEL_TUNNEL_V3_REPLY,
 		({ rmp->sw_if_index = ntohl (sw_if_index); }));
 }