Add GTP-U plugin. VPP-694

Basic GTP-U feature

Change-Id: I31226f890a92c5303ac06e112ed7820cae52d9bd
Signed-off-by: Hongjun Ni <hongjun.ni@intel.com>

20 files changed, 4743 insertions, 0 deletions

diff --git a/src/configure.ac b/src/configure.ac
index 581b0968..2a907b2f 100644
--- a/src/configure.ac
+++ b/src/configure.ac
@@ -147,6 +147,7 @@ AC_SUBST(AR_FLAGS)
 PLUGIN_ENABLED(acl)
 PLUGIN_ENABLED(dpdk)
 PLUGIN_ENABLED(flowperpkt)
+PLUGIN_ENABLED(gtpu)
 PLUGIN_ENABLED(ila)
 PLUGIN_ENABLED(ioam)
 PLUGIN_ENABLED(ixge)
diff --git a/src/plugins/Makefile.am b/src/plugins/Makefile.am
index 623892e7..388c9ad2 100644
--- a/src/plugins/Makefile.am
+++ b/src/plugins/Makefile.am
@@ -41,6 +41,11 @@ if ENABLE_FLOWPERPKT_PLUGIN
 include flowperpkt.am
 endif
 
+
+if ENABLE_GTPU_PLUGIN
+include gtpu.am
+endif
+
 if ENABLE_ILA_PLUGIN
 include ila.am
 endif
diff --git a/src/plugins/gtpu.am b/src/plugins/gtpu.am
new file mode 100644
index 00000000..9e5805b7
--- /dev/null
+++ b/src/plugins/gtpu.am
@@ -0,0 +1,38 @@
+# Copyright (c) 2016 Intel and/or its affiliates.
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at:
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+vppapitestplugins_LTLIBRARIES += gtpu_test_plugin.la
+vppplugins_LTLIBRARIES += gtpu_plugin.la
+
+gtpu_plugin_la_SOURCES =		\
+	gtpu/gtpu_decap.c		    \
+    gtpu/gtpu_encap.c		    \
+	gtpu/gtpu.c				    \
+	gtpu/gtpu_api.c
+
+BUILT_SOURCES +=			\
+	gtpu/gtpu.api.h		    \
+	gtpu/gtpu.api.json
+
+API_FILES += gtpu/gtpu.api
+
+noinst_HEADERS +=			\
+  gtpu/gtpu_all_api_h.h		\
+  gtpu/gtpu_msg_enum.h		\
+  gtpu/gtpu.api.h
+
+gtpu_test_plugin_la_SOURCES = \
+  gtpu/gtpu_test.c            \
+  gtpu/gtpu_plugin.api.h
+
+# vi:syntax=automake
diff --git a/src/plugins/gtpu/gtpu.api b/src/plugins/gtpu/gtpu.api
new file mode 100644
index 00000000..b11670b3
--- /dev/null
+++ b/src/plugins/gtpu/gtpu.api
@@ -0,0 +1,120 @@
+/*
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/** \brief /** \brief Set or delete an GTPU tunnel
+    @param client_index - opaque cookie to identify the sender
+    @param context - sender context, to match reply w/ request
+    @param is_add - add address if non-zero, else delete
+    @param is_ipv6 - src_address and dst_address is ipv6 or not
+    @param src_address - GTPU tunnel's source address.
+    @param dst_address - GTPU tunnel's destination address.
+    @param mcast_sw_if_index - version, O-bit and C-bit (see nsh_packet.h)
+    @param encap_vrf_id - fib identifier used for outgoing encapsulated packets
+    @param decap_next_index - the index of the next node if success
+    @param teid - Local Tunnel Endpoint Identifier
+*/
+define gtpu_add_del_tunnel
+{
+  u32 client_index;
+  u32 context;
+  u8 is_add;
+  u8 is_ipv6;
+  u8 src_address[16];
+  u8 dst_address[16];
+  u32 mcast_sw_if_index;
+  u32 encap_vrf_id;
+  u32 decap_next_index;
+  u32 teid;
+};
+
+/** \brief reply for set or delete an GTPU tunnel
+    @param context - sender context, to match reply w/ request
+    @param retval - return code
+    @param sw_if_index - software index of the interface
+*/
+define gtpu_add_del_tunnel_reply
+{
+  u32 context;
+  i32 retval;
+  u32 sw_if_index;
+};
+
+/** \brief Dump GTPU tunnel
+    @param client_index - opaque cookie to identify the sender
+    @param context - sender context, to match reply w/ request
+    @param sw_if_index - software index of the interface
+*/
+define gtpu_tunnel_dump
+{
+  u32 client_index;
+  u32 context;
+  u32 sw_if_index;
+};
+
+/** \brief /** \brief dump details of an GTPU tunnel
+    @param context - sender context, to match reply w/ request
+    @param sw_if_index - software index of the interface
+    @param is_ipv6 - src_address and dst_address is ipv6 or not
+    @param src_address - GTPU tunnel's source address.
+    @param dst_address - GTPU tunnel's destination address.
+    @param mcast_sw_if_index - version, O-bit and C-bit (see nsh_packet.h)
+    @param encap_vrf_id - fib identifier used for outgoing encapsulated packets
+    @param decap_next_index - the index of the next node if success
+    @param teid - Local Tunnel Endpoint Identifier
+*/
+define gtpu_tunnel_details
+{
+  u32 context;
+  u32 sw_if_index;
+  u8 is_ipv6;
+  u8 src_address[16];
+  u8 dst_address[16];
+  u32 mcast_sw_if_index;
+  u32 encap_vrf_id;
+  u32 decap_next_index;
+  u32 teid;
+};
+
+/** \brief Interface set gtpu-bypass request
+    @param client_index - opaque cookie to identify the sender
+    @param context - sender context, to match reply w/ request
+    @param sw_if_index - interface used to reach neighbor
+    @param is_ipv6 - if non-zero, enable ipv6-gtpu-bypass, else ipv4-gtpu-bypass
+    @param enable - if non-zero enable, else disable
+*/
+define sw_interface_set_gtpu_bypass
+{
+  u32 client_index;
+  u32 context;
+  u32 sw_if_index;
+  u8 is_ipv6;
+  u8 enable;
+};
+
+/** \brief Interface set gtpu-bypass response
+    @param context - sender context, to match reply w/ request
+    @param retval - return code for the request
+*/
+define sw_interface_set_gtpu_bypass_reply
+{
+  u32 context;
+  i32 retval;
+};
+
+/*
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/gtpu/gtpu.c b/src/plugins/gtpu/gtpu.c
new file mode 100644
index 00000000..6a5c5a34
--- /dev/null
+++ b/src/plugins/gtpu/gtpu.c
@@ -0,0 +1,1128 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2016 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+#include <stdint.h>
+#include <net/if.h>
+#include <sys/ioctl.h>
+#include <inttypes.h>
+
+#include <vlib/vlib.h>
+#include <vlib/unix/unix.h>
+#include <vnet/ethernet/ethernet.h>
+#include <vnet/fib/fib_entry.h>
+#include <vnet/fib/fib_table.h>
+#include <vnet/mfib/mfib_table.h>
+#include <vnet/adj/adj_mcast.h>
+#include <vnet/dpo/dpo.h>
+#include <vnet/plugin/plugin.h>
+#include <vpp/app/version.h>
+#include <gtpu/gtpu.h>
+
+
+gtpu_main_t gtpu_main;
+
+static u8 *
+format_decap_next (u8 * s, va_list * args)
+{
+  u32 next_index = va_arg (*args, u32);
+
+  switch (next_index)
+    {
+    case GTPU_INPUT_NEXT_DROP:
+      return format (s, "drop");
+    case GTPU_INPUT_NEXT_L2_INPUT:
+      return format (s, "l2");
+    default:
+      return format (s, "index %d", next_index);
+    }
+  return s;
+}
+
+u8 *
+format_gtpu_tunnel (u8 * s, va_list * args)
+{
+  gtpu_tunnel_t *t = va_arg (*args, gtpu_tunnel_t *);
+  gtpu_main_t *ngm = &gtpu_main;
+
+  s = format (s, "[%d] src %U dst %U teid %d sw_if_index %d ",
+	      t - ngm->tunnels,
+	      format_ip46_address, &t->src, IP46_TYPE_ANY,
+	      format_ip46_address, &t->dst, IP46_TYPE_ANY,
+	      t->teid, t->sw_if_index);
+
+  if (ip46_address_is_multicast (&t->dst))
+    s = format (s, "mcast_sw_if_index %d ", t->mcast_sw_if_index);
+
+  s = format (s, "encap_fib_index %d fib_entry_index %d decap_next %U\n",
+	      t->encap_fib_index, t->fib_entry_index,
+	      format_decap_next, t->decap_next_index);
+  return s;
+}
+
+static u8 *
+format_gtpu_name (u8 * s, va_list * args)
+{
+  u32 dev_instance = va_arg (*args, u32);
+  return format (s, "gtpu_tunnel%d", dev_instance);
+}
+
+static uword
+dummy_interface_tx (vlib_main_t * vm,
+		    vlib_node_runtime_t * node, vlib_frame_t * frame)
+{
+  clib_warning ("you shouldn't be here, leaking buffers...");
+  return frame->n_vectors;
+}
+
+static clib_error_t *
+gtpu_interface_admin_up_down (vnet_main_t * vnm, u32 hw_if_index, u32 flags)
+{
+  u32 hw_flags = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP) ?
+    VNET_HW_INTERFACE_FLAG_LINK_UP : 0;
+  vnet_hw_interface_set_flags (vnm, hw_if_index, hw_flags);
+
+  return /* no error */ 0;
+}
+
+/* *INDENT-OFF* */
+VNET_DEVICE_CLASS (gtpu_device_class,static) = {
+  .name = "GTPU",
+  .format_device_name = format_gtpu_name,
+  .format_tx_trace = format_gtpu_encap_trace,
+  .tx_function = dummy_interface_tx,
+  .admin_up_down_function = gtpu_interface_admin_up_down,
+};
+/* *INDENT-ON* */
+
+static u8 *
+format_gtpu_header_with_length (u8 * s, va_list * args)
+{
+  u32 dev_instance = va_arg (*args, u32);
+  s = format (s, "unimplemented dev %u", dev_instance);
+  return s;
+}
+
+/* *INDENT-OFF* */
+VNET_HW_INTERFACE_CLASS (gtpu_hw_class) =
+{
+  .name = "GTPU",
+  .format_header = format_gtpu_header_with_length,
+  .build_rewrite = default_build_rewrite,
+};
+/* *INDENT-ON* */
+
+static void
+gtpu_tunnel_restack_dpo (gtpu_tunnel_t * t)
+{
+  dpo_id_t dpo = DPO_INVALID;
+  u32 encap_index = ip46_address_is_ip4 (&t->dst) ?
+    gtpu4_encap_node.index : gtpu6_encap_node.index;
+  fib_forward_chain_type_t forw_type = ip46_address_is_ip4 (&t->dst) ?
+    FIB_FORW_CHAIN_TYPE_UNICAST_IP4 : FIB_FORW_CHAIN_TYPE_UNICAST_IP6;
+
+  fib_entry_contribute_forwarding (t->fib_entry_index, forw_type, &dpo);
+  dpo_stack_from_node (encap_index, &t->next_dpo, &dpo);
+  dpo_reset (&dpo);
+}
+
+static gtpu_tunnel_t *
+gtpu_tunnel_from_fib_node (fib_node_t * node)
+{
+  return ((gtpu_tunnel_t *) (((char *) node) -
+			     STRUCT_OFFSET_OF (gtpu_tunnel_t, node)));
+}
+
+/**
+ * Function definition to backwalk a FIB node -
+ * Here we will restack the new dpo of GTPU DIP to encap node.
+ */
+static fib_node_back_walk_rc_t
+gtpu_tunnel_back_walk (fib_node_t * node, fib_node_back_walk_ctx_t * ctx)
+{
+  gtpu_tunnel_restack_dpo (gtpu_tunnel_from_fib_node (node));
+  return (FIB_NODE_BACK_WALK_CONTINUE);
+}
+
+/**
+ * Function definition to get a FIB node from its index
+ */
+static fib_node_t *
+gtpu_tunnel_fib_node_get (fib_node_index_t index)
+{
+  gtpu_tunnel_t *t;
+  gtpu_main_t *gtm = &gtpu_main;
+
+  t = pool_elt_at_index (gtm->tunnels, index);
+
+  return (&t->node);
+}
+
+/**
+ * Function definition to inform the FIB node that its last lock has gone.
+ */
+static void
+gtpu_tunnel_last_lock_gone (fib_node_t * node)
+{
+  /*
+   * The GTPU tunnel is a root of the graph. As such
+   * it never has children and thus is never locked.
+   */
+  ASSERT (0);
+}
+
+/*
+ * Virtual function table registered by GTPU tunnels
+ * for participation in the FIB object graph.
+ */
+const static fib_node_vft_t gtpu_vft = {
+  .fnv_get = gtpu_tunnel_fib_node_get,
+  .fnv_last_lock = gtpu_tunnel_last_lock_gone,
+  .fnv_back_walk = gtpu_tunnel_back_walk,
+};
+
+
+#define foreach_copy_field                      \
+_(teid)                                          \
+_(mcast_sw_if_index)                            \
+_(encap_fib_index)                              \
+_(decap_next_index)                             \
+_(src)                                          \
+_(dst)
+
+static void
+ip_udp_gtpu_rewrite (gtpu_tunnel_t * t, bool is_ip6)
+{
+  union
+  {
+    ip4_gtpu_header_t *h4;
+    ip6_gtpu_header_t *h6;
+    u8 *rw;
+  } r =
+  {
+  .rw = 0};
+  int len = is_ip6 ? sizeof *r.h6 : sizeof *r.h4;
+
+  vec_validate_aligned (r.rw, len - 1, CLIB_CACHE_LINE_BYTES);
+
+  udp_header_t *udp;
+  gtpu_header_t *gtpu;
+  /* Fixed portion of the (outer) ip header */
+  if (!is_ip6)
+    {
+      ip4_header_t *ip = &r.h4->ip4;
+      udp = &r.h4->udp;
+      gtpu = &r.h4->gtpu;
+      ip->ip_version_and_header_length = 0x45;
+      ip->ttl = 254;
+      ip->protocol = IP_PROTOCOL_UDP;
+
+      ip->src_address = t->src.ip4;
+      ip->dst_address = t->dst.ip4;
+
+      /* we fix up the ip4 header length and checksum after-the-fact */
+      ip->checksum = ip4_header_checksum (ip);
+    }
+  else
+    {
+      ip6_header_t *ip = &r.h6->ip6;
+      udp = &r.h6->udp;
+      gtpu = &r.h6->gtpu;
+      ip->ip_version_traffic_class_and_flow_label =
+	clib_host_to_net_u32 (6 << 28);
+      ip->hop_limit = 255;
+      ip->protocol = IP_PROTOCOL_UDP;
+
+      ip->src_address = t->src.ip6;
+      ip->dst_address = t->dst.ip6;
+    }
+
+  /* UDP header, randomize src port on something, maybe? */
+  udp->src_port = clib_host_to_net_u16 (2152);
+  udp->dst_port = clib_host_to_net_u16 (UDP_DST_PORT_GTPU);
+
+  /* GTPU header */
+  gtpu->ver_flags = GTPU_V1_VER | GTPU_PT_GTP;
+  gtpu->type = GTPU_TYPE_GTPU;
+  gtpu->teid = clib_host_to_net_u32 (t->teid);
+
+  t->rewrite = r.rw;
+  /* Now only support 8-byte gtpu header. TBD */
+  _vec_len (t->rewrite) = sizeof (ip4_gtpu_header_t) - 4;
+
+  return;
+}
+
+static bool
+gtpu_decap_next_is_valid (gtpu_main_t * gtm, u32 is_ip6, u32 decap_next_index)
+{
+  vlib_main_t *vm = gtm->vlib_main;
+  u32 input_idx = (!is_ip6) ? gtpu4_input_node.index : gtpu6_input_node.index;
+  vlib_node_runtime_t *r = vlib_node_get_runtime (vm, input_idx);
+
+  return decap_next_index < r->n_next_nodes;
+}
+
+static void
+hash_set_key_copy (uword ** h, void *key, uword v)
+{
+  size_t ksz = hash_header (*h)->user;
+  void *copy = clib_mem_alloc (ksz);
+  clib_memcpy (copy, key, ksz);
+  hash_set_mem (*h, copy, v);
+}
+
+static void
+hash_unset_key_free (uword ** h, void *key)
+{
+  hash_pair_t *hp = hash_get_pair_mem (*h, key);
+  ASSERT (hp);
+  key = uword_to_pointer (hp->key, void *);
+  hash_unset_mem (*h, key);
+  clib_mem_free (key);
+}
+
+static uword
+vtep_addr_ref (ip46_address_t * ip)
+{
+  uword *vtep = ip46_address_is_ip4 (ip) ?
+    hash_get (gtpu_main.vtep4, ip->ip4.as_u32) :
+    hash_get_mem (gtpu_main.vtep6, &ip->ip6);
+  if (vtep)
+    return ++(*vtep);
+  ip46_address_is_ip4 (ip) ?
+    hash_set (gtpu_main.vtep4, ip->ip4.as_u32, 1) :
+    hash_set_key_copy (&gtpu_main.vtep6, &ip->ip6, 1);
+  return 1;
+}
+
+static uword
+vtep_addr_unref (ip46_address_t * ip)
+{
+  uword *vtep = ip46_address_is_ip4 (ip) ?
+    hash_get (gtpu_main.vtep4, ip->ip4.as_u32) :
+    hash_get_mem (gtpu_main.vtep6, &ip->ip6);
+  ASSERT (vtep);
+  if (--(*vtep) != 0)
+    return *vtep;
+  ip46_address_is_ip4 (ip) ?
+    hash_unset (gtpu_main.vtep4, ip->ip4.as_u32) :
+    hash_unset_key_free (&gtpu_main.vtep6, &ip->ip6);
+  return 0;
+}
+
+typedef CLIB_PACKED (union
+		     {
+		     struct
+		     {
+		     fib_node_index_t mfib_entry_index;
+		     adj_index_t mcast_adj_index;
+		     }; u64 as_u64;
+		     }) mcast_shared_t;
+
+static inline mcast_shared_t
+mcast_shared_get (ip46_address_t * ip)
+{
+  ASSERT (ip46_address_is_multicast (ip));
+  uword *p = hash_get_mem (gtpu_main.mcast_shared, ip);
+  ASSERT (p);
+  return (mcast_shared_t)
+  {
+  .as_u64 = *p};
+}
+
+static inline void
+mcast_shared_add (ip46_address_t * dst, fib_node_index_t mfei, adj_index_t ai)
+{
+  mcast_shared_t new_ep = {
+    .mcast_adj_index = ai,
+    .mfib_entry_index = mfei,
+  };
+
+  hash_set_key_copy (&gtpu_main.mcast_shared, dst, new_ep.as_u64);
+}
+
+static inline void
+mcast_shared_remove (ip46_address_t * dst)
+{
+  mcast_shared_t ep = mcast_shared_get (dst);
+
+  adj_unlock (ep.mcast_adj_index);
+  mfib_table_entry_delete_index (ep.mfib_entry_index, MFIB_SOURCE_GTPU);
+
+  hash_unset_key_free (&gtpu_main.mcast_shared, dst);
+}
+
+static inline fib_protocol_t
+fib_ip_proto (bool is_ip6)
+{
+  return (is_ip6) ? FIB_PROTOCOL_IP6 : FIB_PROTOCOL_IP4;
+}
+
+int vnet_gtpu_add_del_tunnel
+  (vnet_gtpu_add_del_tunnel_args_t * a, u32 * sw_if_indexp)
+{
+  gtpu_main_t *gtm = &gtpu_main;
+  gtpu_tunnel_t *t = 0;
+  vnet_main_t *vnm = gtm->vnet_main;
+  uword *p;
+  u32 hw_if_index = ~0;
+  u32 sw_if_index = ~0;
+  gtpu4_tunnel_key_t key4;
+  gtpu6_tunnel_key_t key6;
+  u32 is_ip6 = a->is_ip6;
+
+  if (!is_ip6)
+    {
+      key4.src = a->dst.ip4.as_u32;	/* decap src in key is encap dst in config */
+      key4.teid = clib_host_to_net_u32 (a->teid);
+      p = hash_get (gtm->gtpu4_tunnel_by_key, key4.as_u64);
+    }
+  else
+    {
+      key6.src = a->dst.ip6;
+      key6.teid = clib_host_to_net_u32 (a->teid);
+      p = hash_get_mem (gtm->gtpu6_tunnel_by_key, &key6);
+    }
+
+  if (a->is_add)
+    {
+      l2input_main_t *l2im = &l2input_main;
+
+      /* adding a tunnel: tunnel must not already exist */
+      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 = GTPU_INPUT_NEXT_L2_INPUT;
+      if (!gtpu_decap_next_is_valid (gtm, is_ip6, a->decap_next_index))
+	return VNET_API_ERROR_INVALID_DECAP_NEXT;
+
+      pool_get_aligned (gtm->tunnels, t, CLIB_CACHE_LINE_BYTES);
+      memset (t, 0, sizeof (*t));
+
+      /* copy from arg structure */
+#define _(x) t->x = a->x;
+      foreach_copy_field;
+#undef _
+
+      ip_udp_gtpu_rewrite (t, is_ip6);
+
+      /* copy the key */
+      if (is_ip6)
+	hash_set_key_copy (&gtm->gtpu6_tunnel_by_key, &key6,
+			   t - gtm->tunnels);
+      else
+	hash_set (gtm->gtpu4_tunnel_by_key, key4.as_u64, t - gtm->tunnels);
+
+      vnet_hw_interface_t *hi;
+      if (vec_len (gtm->free_gtpu_tunnel_hw_if_indices) > 0)
+	{
+	  vnet_interface_main_t *im = &vnm->interface_main;
+	  hw_if_index = gtm->free_gtpu_tunnel_hw_if_indices
+	    [vec_len (gtm->free_gtpu_tunnel_hw_if_indices) - 1];
+	  _vec_len (gtm->free_gtpu_tunnel_hw_if_indices) -= 1;
+
+	  hi = vnet_get_hw_interface (vnm, hw_if_index);
+	  hi->dev_instance = t - gtm->tunnels;
+	  hi->hw_instance = hi->dev_instance;
+
+	  /* clear old stats of freed tunnel before reuse */
+	  sw_if_index = hi->sw_if_index;
+	  vnet_interface_counter_lock (im);
+	  vlib_zero_combined_counter
+	    (&im->combined_sw_if_counters[VNET_INTERFACE_COUNTER_TX],
+	     sw_if_index);
+	  vlib_zero_combined_counter (&im->combined_sw_if_counters
+				      [VNET_INTERFACE_COUNTER_RX],
+				      sw_if_index);
+	  vlib_zero_simple_counter (&im->sw_if_counters
+				    [VNET_INTERFACE_COUNTER_DROP],
+				    sw_if_index);
+	  vnet_interface_counter_unlock (im);
+	}
+      else
+	{
+	  hw_if_index = vnet_register_interface
+	    (vnm, gtpu_device_class.index, t - gtm->tunnels,
+	     gtpu_hw_class.index, t - gtm->tunnels);
+	  hi = vnet_get_hw_interface (vnm, hw_if_index);
+	}
+
+      t->hw_if_index = hw_if_index;
+      t->sw_if_index = sw_if_index = hi->sw_if_index;
+
+      vec_validate_init_empty (gtm->tunnel_index_by_sw_if_index, sw_if_index,
+			       ~0);
+      gtm->tunnel_index_by_sw_if_index[sw_if_index] = t - gtm->tunnels;
+
+      /* setup l2 input config with l2 feature and bd 0 to drop packet */
+      vec_validate (l2im->configs, sw_if_index);
+      l2im->configs[sw_if_index].feature_bitmap = L2INPUT_FEAT_DROP;
+      l2im->configs[sw_if_index].bd_index = 0;
+
+      vnet_sw_interface_t *si = vnet_get_sw_interface (vnm, sw_if_index);
+      si->flags &= ~VNET_SW_INTERFACE_FLAG_HIDDEN;
+      vnet_sw_interface_set_flags (vnm, sw_if_index,
+				   VNET_SW_INTERFACE_FLAG_ADMIN_UP);
+
+      fib_node_init (&t->node, gtm->fib_node_type);
+      fib_prefix_t tun_dst_pfx;
+      u32 encap_index = !is_ip6 ?
+	gtpu4_encap_node.index : gtpu6_encap_node.index;
+      vnet_flood_class_t flood_class = VNET_FLOOD_CLASS_TUNNEL_NORMAL;
+
+      fib_prefix_from_ip46_addr (&t->dst, &tun_dst_pfx);
+      if (!ip46_address_is_multicast (&t->dst))
+	{
+	  /* Unicast tunnel -
+	   * source the FIB entry for the tunnel's destination
+	   * and become a child thereof. The tunnel will then get poked
+	   * when the forwarding for the entry updates, and the tunnel can
+	   * re-stack accordingly
+	   */
+	  vtep_addr_ref (&t->src);
+	  t->fib_entry_index = fib_table_entry_special_add
+	    (t->encap_fib_index, &tun_dst_pfx, FIB_SOURCE_RR,
+	     FIB_ENTRY_FLAG_NONE);
+	  t->sibling_index = fib_entry_child_add
+	    (t->fib_entry_index, gtm->fib_node_type, t - gtm->tunnels);
+	  gtpu_tunnel_restack_dpo (t);
+	}
+      else
+	{
+	  /* Multicast tunnel -
+	   * as the same mcast group can be used for mutiple mcast tunnels
+	   * with different VNIs, create the output fib adjecency only if
+	   * it does not already exist
+	   */
+	  fib_protocol_t fp = fib_ip_proto (is_ip6);
+
+	  if (vtep_addr_ref (&t->dst) == 1)
+	    {
+	      fib_node_index_t mfei;
+	      adj_index_t ai;
+	      fib_route_path_t path = {
+		.frp_proto = fp,
+		.frp_addr = zero_addr,
+		.frp_sw_if_index = 0xffffffff,
+		.frp_fib_index = ~0,
+		.frp_weight = 0,
+		.frp_flags = FIB_ROUTE_PATH_LOCAL,
+	      };
+	      const mfib_prefix_t mpfx = {
+		.fp_proto = fp,
+		.fp_len = (is_ip6 ? 128 : 32),
+		.fp_grp_addr = tun_dst_pfx.fp_addr,
+	      };
+
+	      /*
+	       * Setup the (*,G) to receive traffic on the mcast group
+	       *  - the forwarding interface is for-us
+	       *  - the accepting interface is that from the API
+	       */
+	      mfib_table_entry_path_update (t->encap_fib_index,
+					    &mpfx,
+					    MFIB_SOURCE_GTPU,
+					    &path, MFIB_ITF_FLAG_FORWARD);
+
+	      path.frp_sw_if_index = a->mcast_sw_if_index;
+	      path.frp_flags = FIB_ROUTE_PATH_FLAG_NONE;
+	      mfei = mfib_table_entry_path_update (t->encap_fib_index,
+						   &mpfx,
+						   MFIB_SOURCE_GTPU,
+						   &path,
+						   MFIB_ITF_FLAG_ACCEPT);
+
+	      /*
+	       * Create the mcast adjacency to send traffic to the group
+	       */
+	      ai = adj_mcast_add_or_lock (fp,
+					  fib_proto_to_link (fp),
+					  a->mcast_sw_if_index);
+
+	      /*
+	       * create a new end-point
+	       */
+	      mcast_shared_add (&t->dst, mfei, ai);
+	    }
+
+	  dpo_id_t dpo = DPO_INVALID;
+	  mcast_shared_t ep = mcast_shared_get (&t->dst);
+
+	  /* Stack shared mcast dst mac addr rewrite on encap */
+	  dpo_set (&dpo, DPO_ADJACENCY_MCAST,
+		   fib_proto_to_dpo (fp), ep.mcast_adj_index);
+
+	  dpo_stack_from_node (encap_index, &t->next_dpo, &dpo);
+
+	  dpo_reset (&dpo);
+	  flood_class = VNET_FLOOD_CLASS_TUNNEL_MASTER;
+	}
+
+      /* Set gtpu tunnel output node */
+      hi->output_node_index = encap_index;
+
+      vnet_get_sw_interface (vnet_get_main (), sw_if_index)->flood_class =
+	flood_class;
+    }
+  else
+    {
+      /* deleting a tunnel: tunnel must exist */
+      if (!p)
+	return VNET_API_ERROR_NO_SUCH_ENTRY;
+
+      t = pool_elt_at_index (gtm->tunnels, p[0]);
+      sw_if_index = t->sw_if_index;
+
+      vnet_sw_interface_set_flags (vnm, t->sw_if_index, 0 /* down */ );
+      vnet_sw_interface_t *si = vnet_get_sw_interface (vnm, t->sw_if_index);
+      si->flags |= VNET_SW_INTERFACE_FLAG_HIDDEN;
+
+      /* make sure tunnel is removed from l2 bd or xconnect */
+      set_int_l2_mode (gtm->vlib_main, vnm, MODE_L3, t->sw_if_index, 0, 0, 0,
+		       0);
+      vec_add1 (gtm->free_gtpu_tunnel_hw_if_indices, t->hw_if_index);
+
+      gtm->tunnel_index_by_sw_if_index[t->sw_if_index] = ~0;
+
+      if (!is_ip6)
+	hash_unset (gtm->gtpu4_tunnel_by_key, key4.as_u64);
+      else
+	hash_unset_key_free (&gtm->gtpu6_tunnel_by_key, &key6);
+
+      if (!ip46_address_is_multicast (&t->dst))
+	{
+	  vtep_addr_unref (&t->src);
+	  fib_entry_child_remove (t->fib_entry_index, t->sibling_index);
+	  fib_table_entry_delete_index (t->fib_entry_index, FIB_SOURCE_RR);
+	}
+      else if (vtep_addr_unref (&t->dst) == 0)
+	{
+	  mcast_shared_remove (&t->dst);
+	}
+
+      fib_node_deinit (&t->node);
+      vec_free (t->rewrite);
+      pool_put (gtm->tunnels, t);
+    }
+
+  if (sw_if_indexp)
+    *sw_if_indexp = sw_if_index;
+
+  return 0;
+}
+
+static uword
+get_decap_next_for_node (u32 node_index, u32 ipv4_set)
+{
+  gtpu_main_t *gtm = &gtpu_main;
+  vlib_main_t *vm = gtm->vlib_main;
+  uword input_node = (ipv4_set) ? gtpu4_input_node.index :
+    gtpu6_input_node.index;
+
+  return vlib_node_add_next (vm, input_node, node_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);
+  gtpu_main_t *gtm = &gtpu_main;
+  vlib_main_t *vm = gtm->vlib_main;
+  u32 node_index;
+  u32 tmp;
+
+  if (unformat (input, "l2"))
+    *result = GTPU_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
+    return 0;
+  return 1;
+}
+
+static clib_error_t *
+gtpu_add_del_tunnel_command_fn (vlib_main_t * vm,
+				unformat_input_t * input,
+				vlib_cli_command_t * cmd)
+{
+  unformat_input_t _line_input, *line_input = &_line_input;
+  ip46_address_t src, dst;
+  u8 is_add = 1;
+  u8 src_set = 0;
+  u8 dst_set = 0;
+  u8 grp_set = 0;
+  u8 ipv4_set = 0;
+  u8 ipv6_set = 0;
+  u32 encap_fib_index = 0;
+  u32 mcast_sw_if_index = ~0;
+  u32 decap_next_index = GTPU_INPUT_NEXT_L2_INPUT;
+  u32 teid = 0;
+  u32 tmp;
+  int rv;
+  vnet_gtpu_add_del_tunnel_args_t _a, *a = &_a;
+  u32 tunnel_sw_if_index;
+  clib_error_t *error = NULL;
+
+  /* Cant "universally zero init" (={0}) due to GCC bug 53119 */
+  memset (&src, 0, sizeof src);
+  memset (&dst, 0, sizeof dst);
+
+  /* Get a line of input. */
+  if (!unformat_user (input, unformat_line_input, line_input))
+    return 0;
+
+  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (line_input, "del"))
+	{
+	  is_add = 0;
+	}
+      else if (unformat (line_input, "src %U",
+			 unformat_ip4_address, &src.ip4))
+	{
+	  src_set = 1;
+	  ipv4_set = 1;
+	}
+      else if (unformat (line_input, "dst %U",
+			 unformat_ip4_address, &dst.ip4))
+	{
+	  dst_set = 1;
+	  ipv4_set = 1;
+	}
+      else if (unformat (line_input, "src %U",
+			 unformat_ip6_address, &src.ip6))
+	{
+	  src_set = 1;
+	  ipv6_set = 1;
+	}
+      else if (unformat (line_input, "dst %U",
+			 unformat_ip6_address, &dst.ip6))
+	{
+	  dst_set = 1;
+	  ipv6_set = 1;
+	}
+      else if (unformat (line_input, "group %U %U",
+			 unformat_ip4_address, &dst.ip4,
+			 unformat_vnet_sw_interface,
+			 vnet_get_main (), &mcast_sw_if_index))
+	{
+	  grp_set = dst_set = 1;
+	  ipv4_set = 1;
+	}
+      else if (unformat (line_input, "group %U %U",
+			 unformat_ip6_address, &dst.ip6,
+			 unformat_vnet_sw_interface,
+			 vnet_get_main (), &mcast_sw_if_index))
+	{
+	  grp_set = dst_set = 1;
+	  ipv6_set = 1;
+	}
+      else if (unformat (line_input, "encap-vrf-id %d", &tmp))
+	{
+	  encap_fib_index = fib_table_find (fib_ip_proto (ipv6_set), tmp);
+	  if (encap_fib_index == ~0)
+	    {
+	      error =
+		clib_error_return (0, "nonexistent encap-vrf-id %d", tmp);
+	      goto done;
+	    }
+	}
+      else if (unformat (line_input, "decap-next %U", unformat_decap_next,
+			 &decap_next_index, ipv4_set))
+	;
+      else if (unformat (line_input, "teid %d", &teid))
+	;
+      else
+	{
+	  error = clib_error_return (0, "parse error: '%U'",
+				     format_unformat_error, line_input);
+	  goto done;
+	}
+    }
+
+  if (src_set == 0)
+    {
+      error = clib_error_return (0, "tunnel src address not specified");
+      goto done;
+    }
+
+  if (dst_set == 0)
+    {
+      error = clib_error_return (0, "tunnel dst address not specified");
+      goto done;
+    }
+
+  if (grp_set && !ip46_address_is_multicast (&dst))
+    {
+      error = clib_error_return (0, "tunnel group address not multicast");
+      goto done;
+    }
+
+  if (grp_set == 0 && ip46_address_is_multicast (&dst))
+    {
+      error = clib_error_return (0, "dst address must be unicast");
+      goto done;
+    }
+
+  if (grp_set && mcast_sw_if_index == ~0)
+    {
+      error = clib_error_return (0, "tunnel nonexistent multicast device");
+      goto done;
+    }
+
+  if (ipv4_set && ipv6_set)
+    {
+      error = clib_error_return (0, "both IPv4 and IPv6 addresses specified");
+      goto done;
+    }
+
+  if (ip46_address_cmp (&src, &dst) == 0)
+    {
+      error = clib_error_return (0, "src and dst addresses are identical");
+      goto done;
+    }
+
+  if (decap_next_index == ~0)
+    {
+      error = clib_error_return (0, "next node not found");
+      goto done;
+    }
+
+  memset (a, 0, sizeof (*a));
+
+  a->is_add = is_add;
+  a->is_ip6 = ipv6_set;
+
+#define _(x) a->x = x;
+  foreach_copy_field;
+#undef _
+
+  rv = vnet_gtpu_add_del_tunnel (a, &tunnel_sw_if_index);
+
+  switch (rv)
+    {
+    case 0:
+      if (is_add)
+	vlib_cli_output (vm, "%U\n", format_vnet_sw_if_index_name,
+			 vnet_get_main (), tunnel_sw_if_index);
+      break;
+
+    case VNET_API_ERROR_TUNNEL_EXIST:
+      error = clib_error_return (0, "tunnel already exists...");
+      goto done;
+
+    case VNET_API_ERROR_NO_SUCH_ENTRY:
+      error = clib_error_return (0, "tunnel does not exist...");
+      goto done;
+
+    default:
+      error = clib_error_return
+	(0, "vnet_gtpu_add_del_tunnel returned %d", rv);
+      goto done;
+    }
+
+done:
+  unformat_free (line_input);
+
+  return error;
+}
+
+/*?
+ * Add or delete a GTPU Tunnel.
+ *
+ * GTPU provides the features needed to allow L2 bridge domains (BDs)
+ * to span multiple servers. This is done by building an L2 overlay on
+ * top of an L3 network underlay using GTPU tunnels.
+ *
+ * This makes it possible for servers to be co-located in the same data
+ * center or be separated geographically as long as they are reachable
+ * through the underlay L3 network.
+ *
+ * You can refer to this kind of L2 overlay bridge domain as a GTPU
+ * (Virtual eXtensible VLAN) segment.
+ *
+ * @cliexpar
+ * Example of how to create a GTPU Tunnel:
+ * @cliexcmd{create gtpu tunnel src 10.0.3.1 dst 10.0.3.3 teid 13 encap-vrf-id 7}
+ * Example of how to delete a GTPU Tunnel:
+ * @cliexcmd{create gtpu tunnel src 10.0.3.1 dst 10.0.3.3 teid 13 del}
+ ?*/
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (create_gtpu_tunnel_command, static) = {
+  .path = "create gtpu tunnel",
+  .short_help =
+  "create gtpu tunnel src <local-vtep-addr>"
+  " {dst <remote-vtep-addr>|group <mcast-vtep-addr> <intf-name>} teid <nn>"
+  " [encap-vrf-id <nn>] [decap-next [l2|node <name>]] [del]",
+  .function = gtpu_add_del_tunnel_command_fn,
+};
+/* *INDENT-ON* */
+
+static clib_error_t *
+show_gtpu_tunnel_command_fn (vlib_main_t * vm,
+			     unformat_input_t * input,
+			     vlib_cli_command_t * cmd)
+{
+  gtpu_main_t *gtm = &gtpu_main;
+  gtpu_tunnel_t *t;
+
+  if (pool_elts (gtm->tunnels) == 0)
+    vlib_cli_output (vm, "No gtpu tunnels configured...");
+
+  pool_foreach (t, gtm->tunnels, (
+				   {
+				   vlib_cli_output (vm, "%U",
+						    format_gtpu_tunnel, t);
+				   }
+		));
+
+  return 0;
+}
+
+/*?
+ * Display all the GTPU Tunnel entries.
+ *
+ * @cliexpar
+ * Example of how to display the GTPU Tunnel entries:
+ * @cliexstart{show gtpu tunnel}
+ * [0] src 10.0.3.1 dst 10.0.3.3 teid 13 encap_fib_index 0 sw_if_index 5 decap_next l2
+ * @cliexend
+ ?*/
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (show_gtpu_tunnel_command, static) = {
+    .path = "show gtpu tunnel",
+    .short_help = "show gtpu tunnel",
+    .function = show_gtpu_tunnel_command_fn,
+};
+/* *INDENT-ON* */
+
+void
+vnet_int_gtpu_bypass_mode (u32 sw_if_index, u8 is_ip6, u8 is_enable)
+{
+  if (is_ip6)
+    vnet_feature_enable_disable ("ip6-unicast", "ip6-gtpu-bypass",
+				 sw_if_index, is_enable, 0, 0);
+  else
+    vnet_feature_enable_disable ("ip4-unicast", "ip4-gtpu-bypass",
+				 sw_if_index, is_enable, 0, 0);
+}
+
+static clib_error_t *
+set_ip_gtpu_bypass (u32 is_ip6,
+		    unformat_input_t * input, vlib_cli_command_t * cmd)
+{
+  unformat_input_t _line_input, *line_input = &_line_input;
+  vnet_main_t *vnm = vnet_get_main ();
+  clib_error_t *error = 0;
+  u32 sw_if_index, is_enable;
+
+  sw_if_index = ~0;
+  is_enable = 1;
+
+  if (!unformat_user (input, unformat_line_input, line_input))
+    return 0;
+
+  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat_user
+	  (line_input, unformat_vnet_sw_interface, vnm, &sw_if_index))
+	;
+      else if (unformat (line_input, "del"))
+	is_enable = 0;
+      else
+	{
+	  error = unformat_parse_error (line_input);
+	  goto done;
+	}
+    }
+
+  if (~0 == sw_if_index)
+    {
+      error = clib_error_return (0, "unknown interface `%U'",
+				 format_unformat_error, line_input);
+      goto done;
+    }
+
+  vnet_int_gtpu_bypass_mode (sw_if_index, is_ip6, is_enable);
+
+done:
+  unformat_free (line_input);
+
+  return error;
+}
+
+static clib_error_t *
+set_ip4_gtpu_bypass (vlib_main_t * vm,
+		     unformat_input_t * input, vlib_cli_command_t * cmd)
+{
+  return set_ip_gtpu_bypass (0, input, cmd);
+}
+
+/*?
+ * This command adds the 'ip4-gtpu-bypass' graph node for a given interface.
+ * By adding the IPv4 gtpu-bypass graph node to an interface, the node checks
+ *  for and validate input gtpu packet and bypass ip4-lookup, ip4-local,
+ * ip4-udp-lookup nodes to speedup gtpu packet forwarding. This node will
+ * cause extra overhead to for non-gtpu packets which is kept at a minimum.
+ *
+ * @cliexpar
+ * @parblock
+ * Example of graph node before ip4-gtpu-bypass is enabled:
+ * @cliexstart{show vlib graph ip4-gtpu-bypass}
+ *            Name                      Next                    Previous
+ * ip4-gtpu-bypass                error-drop [0]
+ *                                gtpu4-input [1]
+ *                                 ip4-lookup [2]
+ * @cliexend
+ *
+ * Example of how to enable ip4-gtpu-bypass on an interface:
+ * @cliexcmd{set interface ip gtpu-bypass GigabitEthernet2/0/0}
+ *
+ * Example of graph node after ip4-gtpu-bypass is enabled:
+ * @cliexstart{show vlib graph ip4-gtpu-bypass}
+ *            Name                      Next                    Previous
+ * ip4-gtpu-bypass                error-drop [0]               ip4-input
+ *                                gtpu4-input [1]        ip4-input-no-checksum
+ *                                 ip4-lookup [2]
+ * @cliexend
+ *
+ * Example of how to display the feature enabed on an interface:
+ * @cliexstart{show ip interface features GigabitEthernet2/0/0}
+ * IP feature paths configured on GigabitEthernet2/0/0...
+ * ...
+ * ipv4 unicast:
+ *   ip4-gtpu-bypass
+ *   ip4-lookup
+ * ...
+ * @cliexend
+ *
+ * Example of how to disable ip4-gtpu-bypass on an interface:
+ * @cliexcmd{set interface ip gtpu-bypass GigabitEthernet2/0/0 del}
+ * @endparblock
+?*/
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (set_interface_ip_gtpu_bypass_command, static) = {
+  .path = "set interface ip gtpu-bypass",
+  .function = set_ip4_gtpu_bypass,
+  .short_help = "set interface ip gtpu-bypass <interface> [del]",
+};
+/* *INDENT-ON* */
+
+static clib_error_t *
+set_ip6_gtpu_bypass (vlib_main_t * vm,
+		     unformat_input_t * input, vlib_cli_command_t * cmd)
+{
+  return set_ip_gtpu_bypass (1, input, cmd);
+}
+
+/*?
+ * This command adds the 'ip6-gtpu-bypass' graph node for a given interface.
+ * By adding the IPv6 gtpu-bypass graph node to an interface, the node checks
+ *  for and validate input gtpu packet and bypass ip6-lookup, ip6-local,
+ * ip6-udp-lookup nodes to speedup gtpu packet forwarding. This node will
+ * cause extra overhead to for non-gtpu packets which is kept at a minimum.
+ *
+ * @cliexpar
+ * @parblock
+ * Example of graph node before ip6-gtpu-bypass is enabled:
+ * @cliexstart{show vlib graph ip6-gtpu-bypass}
+ *            Name                      Next                    Previous
+ * ip6-gtpu-bypass                error-drop [0]
+ *                                gtpu6-input [1]
+ *                                 ip6-lookup [2]
+ * @cliexend
+ *
+ * Example of how to enable ip6-gtpu-bypass on an interface:
+ * @cliexcmd{set interface ip6 gtpu-bypass GigabitEthernet2/0/0}
+ *
+ * Example of graph node after ip6-gtpu-bypass is enabled:
+ * @cliexstart{show vlib graph ip6-gtpu-bypass}
+ *            Name                      Next                    Previous
+ * ip6-gtpu-bypass                error-drop [0]               ip6-input
+ *                                gtpu6-input [1]        ip4-input-no-checksum
+ *                                 ip6-lookup [2]
+ * @cliexend
+ *
+ * Example of how to display the feature enabed on an interface:
+ * @cliexstart{show ip interface features GigabitEthernet2/0/0}
+ * IP feature paths configured on GigabitEthernet2/0/0...
+ * ...
+ * ipv6 unicast:
+ *   ip6-gtpu-bypass
+ *   ip6-lookup
+ * ...
+ * @cliexend
+ *
+ * Example of how to disable ip6-gtpu-bypass on an interface:
+ * @cliexcmd{set interface ip6 gtpu-bypass GigabitEthernet2/0/0 del}
+ * @endparblock
+?*/
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (set_interface_ip6_gtpu_bypass_command, static) = {
+  .path = "set interface ip6 gtpu-bypass",
+  .function = set_ip6_gtpu_bypass,
+  .short_help = "set interface ip gtpu-bypass <interface> [del]",
+};
+/* *INDENT-ON* */
+
+clib_error_t *
+gtpu_init (vlib_main_t * vm)
+{
+  gtpu_main_t *gtm = &gtpu_main;
+
+  gtm->vnet_main = vnet_get_main ();
+  gtm->vlib_main = vm;
+
+  /* initialize the ip6 hash */
+  gtm->gtpu6_tunnel_by_key = hash_create_mem (0,
+					      sizeof (gtpu6_tunnel_key_t),
+					      sizeof (uword));
+  gtm->vtep6 = hash_create_mem (0, sizeof (ip6_address_t), sizeof (uword));
+  gtm->mcast_shared = hash_create_mem (0,
+				       sizeof (ip46_address_t),
+				       sizeof (mcast_shared_t));
+
+  udp_register_dst_port (vm, UDP_DST_PORT_GTPU,
+			 gtpu4_input_node.index, /* is_ip4 */ 1);
+  udp_register_dst_port (vm, UDP_DST_PORT_GTPU6,
+			 gtpu6_input_node.index, /* is_ip4 */ 0);
+
+  gtm->fib_node_type = fib_node_register_new_type (&gtpu_vft);
+
+  return 0;
+}
+
+VLIB_INIT_FUNCTION (gtpu_init);
+
+/* *INDENT-OFF* */
+VLIB_PLUGIN_REGISTER () = {
+    .version = VPP_BUILD_VER,
+    .description = "GTPv1-U",
+};
+/* *INDENT-ON* */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/gtpu/gtpu.h b/src/plugins/gtpu/gtpu.h
new file mode 100644
index 00000000..86861974
--- /dev/null
+++ b/src/plugins/gtpu/gtpu.h
@@ -0,0 +1,262 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2016 Intel and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#ifndef included_vnet_gtpu_h
+#define included_vnet_gtpu_h
+
+#include <vppinfra/lock.h>
+#include <vppinfra/error.h>
+#include <vppinfra/hash.h>
+#include <vnet/vnet.h>
+#include <vnet/ip/ip.h>
+#include <vnet/l2/l2_input.h>
+#include <vnet/l2/l2_output.h>
+#include <vnet/l2/l2_bd.h>
+#include <vnet/ethernet/ethernet.h>
+#include <vnet/ip/ip4_packet.h>
+#include <vnet/ip/ip6_packet.h>
+#include <vnet/udp/udp.h>
+#include <vnet/dpo/dpo.h>
+#include <vnet/adj/adj_types.h>
+#include <vnet/fib/fib_table.h>
+
+/**
+ *		Bits
+ * Octets	8	7	6	5	4	3	2	1
+ * 1		          Version	PT	(*)	E	S	PN
+ * 2		Message Type
+ * 3		Length (1st Octet)
+ * 4		Length (2nd Octet)
+ * 5		Tunnel Endpoint Identifier (1st Octet)
+ * 6		Tunnel Endpoint Identifier (2nd Octet)
+ * 7		Tunnel Endpoint Identifier (3rd Octet)
+ * 8		Tunnel Endpoint Identifier (4th Octet)
+ * 9		Sequence Number (1st Octet)1) 4)
+ * 10		Sequence Number (2nd Octet)1) 4)
+ * 11		N-PDU Number2) 4)
+ * 12		Next Extension Header Type3) 4)
+**/
+
+typedef struct
+{
+  u8 ver_flags;
+  u8 type;
+  u16 length;			/* length in octets of the payload */
+  u32 teid;
+  u16 sequence;
+  u8 pdu_number;
+  u8 next_ext_type;
+} gtpu_header_t;
+
+#define GTPU_VER_MASK (7<<5)
+#define GTPU_PT_BIT   (1<<4)
+#define GTPU_E_BIT    (1<<2)
+#define GTPU_S_BIT    (1<<1)
+#define GTPU_PN_BIT   (1<<0)
+#define GTPU_E_S_PN_BIT  (7<<0)
+
+#define GTPU_V1_VER   (1<<5)
+
+#define GTPU_PT_GTP    (1<<4)
+#define GTPU_TYPE_GTPU  255
+
+/* *INDENT-OFF* */
+typedef CLIB_PACKED(struct
+{
+  ip4_header_t ip4;            /* 20 bytes */
+  udp_header_t udp;            /* 8 bytes */
+  gtpu_header_t gtpu;	       /* 8 bytes */
+}) ip4_gtpu_header_t;
+/* *INDENT-ON* */
+
+/* *INDENT-OFF* */
+typedef CLIB_PACKED(struct
+{
+  ip6_header_t ip6;            /* 40 bytes */
+  udp_header_t udp;            /* 8 bytes */
+  gtpu_header_t gtpu;     /* 8 bytes */
+}) ip6_gtpu_header_t;
+/* *INDENT-ON* */
+
+/* *INDENT-OFF* */
+typedef CLIB_PACKED
+(struct {
+  /*
+   * Key fields: ip src and gtpu teid on incoming gtpu packet
+   * all fields in NET byte order
+   */
+  union {
+    struct {
+      u32 src;
+      u32 teid;
+    };
+    u64 as_u64;
+  };
+}) gtpu4_tunnel_key_t;
+/* *INDENT-ON* */
+
+/* *INDENT-OFF* */
+typedef CLIB_PACKED
+(struct {
+  /*
+   * Key fields: ip src and gtpu teid on incoming gtpu packet
+   * all fields in NET byte order
+   */
+  ip6_address_t src;
+  u32 teid;
+}) gtpu6_tunnel_key_t;
+/* *INDENT-ON* */
+
+typedef struct
+{
+  /* Rewrite string */
+  u8 *rewrite;
+
+  /* FIB DPO for IP forwarding of gtpu encap packet */
+  dpo_id_t next_dpo;
+
+  /* gtpu teid in HOST byte order */
+  u32 teid;
+
+  /* tunnel src and dst addresses */
+  ip46_address_t src;
+  ip46_address_t dst;
+
+  /* mcast packet output intf index (used only if dst is mcast) */
+  u32 mcast_sw_if_index;
+
+  /* decap next index */
+  u32 decap_next_index;
+
+  /* The FIB index for src/dst addresses */
+  u32 encap_fib_index;
+
+  /* vnet intfc index */
+  u32 sw_if_index;
+  u32 hw_if_index;
+
+  /**
+   * Linkage into the FIB object graph
+   */
+  fib_node_t node;
+
+  /*
+   * The FIB entry for (depending on gtpu tunnel is unicast or mcast)
+   * sending unicast gtpu encap packets or receiving mcast gtpu packets
+   */
+  fib_node_index_t fib_entry_index;
+  adj_index_t mcast_adj_index;
+
+  /**
+   * The tunnel is a child of the FIB entry for its destination. This is
+   * so it receives updates when the forwarding information for that entry
+   * changes.
+   * The tunnels sibling index on the FIB entry's dependency list.
+   */
+  u32 sibling_index;
+} gtpu_tunnel_t;
+
+#define foreach_gtpu_input_next        \
+_(DROP, "error-drop")                  \
+_(L2_INPUT, "l2-input")
+
+typedef enum
+{
+#define _(s,n) GTPU_INPUT_NEXT_##s,
+  foreach_gtpu_input_next
+#undef _
+    GTPU_INPUT_N_NEXT,
+} gtpu_input_next_t;
+
+typedef enum
+{
+#define gtpu_error(n,s) GTPU_ERROR_##n,
+#include <gtpu/gtpu_error.def>
+#undef gtpu_error
+  GTPU_N_ERROR,
+} gtpu_input_error_t;
+
+typedef struct
+{
+  /* vector of encap tunnel instances */
+  gtpu_tunnel_t *tunnels;
+
+  /* lookup tunnel by key */
+  uword *gtpu4_tunnel_by_key;	/* keyed on ipv4.dst + teid */
+  uword *gtpu6_tunnel_by_key;	/* keyed on ipv6.dst + teid */
+
+  /* local VTEP IPs ref count used by gtpu-bypass node to check if
+     received gtpu packet DIP matches any local VTEP address */
+  uword *vtep4;			/* local ip4 VTEPs keyed on their ip4 addr */
+  uword *vtep6;			/* local ip6 VTEPs keyed on their ip6 addr */
+
+  /* mcast shared info */
+  uword *mcast_shared;		/* keyed on mcast ip46 addr */
+
+  /* Free vlib hw_if_indices */
+  u32 *free_gtpu_tunnel_hw_if_indices;
+
+  /* Mapping from sw_if_index to tunnel index */
+  u32 *tunnel_index_by_sw_if_index;
+
+  /**
+   * Node type for registering to fib changes.
+   */
+  fib_node_type_t fib_node_type;
+
+  /* API message ID base */
+  u16 msg_id_base;
+
+  /* convenience */
+  vlib_main_t *vlib_main;
+  vnet_main_t *vnet_main;
+} gtpu_main_t;
+
+gtpu_main_t gtpu_main;
+
+extern vlib_node_registration_t gtpu4_input_node;
+extern vlib_node_registration_t gtpu6_input_node;
+extern vlib_node_registration_t gtpu4_encap_node;
+extern vlib_node_registration_t gtpu6_encap_node;
+
+u8 *format_gtpu_encap_trace (u8 * s, va_list * args);
+
+typedef struct
+{
+  u8 is_add;
+  u8 is_ip6;
+  ip46_address_t src, dst;
+  u32 mcast_sw_if_index;
+  u32 encap_fib_index;
+  u32 decap_next_index;
+  u32 teid;
+} vnet_gtpu_add_del_tunnel_args_t;
+
+int vnet_gtpu_add_del_tunnel
+  (vnet_gtpu_add_del_tunnel_args_t * a, u32 * sw_if_indexp);
+
+void vnet_int_gtpu_bypass_mode (u32 sw_if_index, u8 is_ip6, u8 is_enable);
+#endif /* included_vnet_gtpu_h */
+
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/gtpu/gtpu_all_api_h.h b/src/plugins/gtpu/gtpu_all_api_h.h
new file mode 100644
index 00000000..a4181ae3
--- /dev/null
+++ b/src/plugins/gtpu/gtpu_all_api_h.h
@@ -0,0 +1,18 @@
+/*
+ * gtpu_all_api_h.h - plug-in api #include file
+ *
+ * Copyright (c) Intel and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/* Include the generated file, see BUILT_SOURCES in Makefile.am */
+#include <gtpu/gtpu.api.h>
diff --git a/src/plugins/gtpu/gtpu_api.c b/src/plugins/gtpu/gtpu_api.c
new file mode 100644
index 00000000..58f48c71
--- /dev/null
+++ b/src/plugins/gtpu/gtpu_api.c
@@ -0,0 +1,256 @@
+/*
+ *------------------------------------------------------------------
+ * gtpu_api.c - gtpu api
+ *
+ * Copyright (c) 2016 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <vnet/interface.h>
+#include <vnet/api_errno.h>
+#include <vnet/feature/feature.h>
+#include <vnet/fib/fib_table.h>
+
+#include <vppinfra/byte_order.h>
+#include <vlibmemory/api.h>
+#include <vlibsocket/api.h>
+
+#include <gtpu/gtpu.h>
+
+
+#define vl_msg_id(n,h) n,
+typedef enum
+{
+#include <gtpu/gtpu.api.h>
+  /* We'll want to know how many messages IDs we need... */
+  VL_MSG_FIRST_AVAILABLE,
+} vl_msg_id_t;
+#undef vl_msg_id
+
+/* define message structures */
+#define vl_typedefs
+#include <gtpu/gtpu.api.h>
+#undef vl_typedefs
+
+/* define generated endian-swappers */
+#define vl_endianfun
+#include <gtpu/gtpu.api.h>
+#undef vl_endianfun
+
+/* instantiate all the print functions we know about */
+#define vl_print(handle, ...) vlib_cli_output (handle, __VA_ARGS__)
+#define vl_printfun
+#include <gtpu/gtpu.api.h>
+#undef vl_printfun
+
+/* Get the API version number */
+#define vl_api_version(n,v) static u32 api_version=(v);
+#include <gtpu/gtpu.api.h>
+#undef vl_api_version
+
+#define vl_msg_name_crc_list
+#include <gtpu/gtpu.api.h>
+#undef vl_msg_name_crc_list
+
+#define REPLY_MSG_ID_BASE gtm->msg_id_base
+#include <vlibapi/api_helper_macros.h>
+
+static void
+setup_message_id_table (gtpu_main_t * gtm, api_main_t * am)
+{
+#define _(id,n,crc) \
+  vl_msg_api_add_msg_name_crc (am, #n "_" #crc, id + gtm->msg_id_base);
+  foreach_vl_msg_name_crc_gtpu;
+#undef _
+}
+
+#define foreach_gtpu_plugin_api_msg                             \
+_(SW_INTERFACE_SET_GTPU_BYPASS, sw_interface_set_gtpu_bypass)         \
+_(GTPU_ADD_DEL_TUNNEL, gtpu_add_del_tunnel)                           \
+_(GTPU_TUNNEL_DUMP, gtpu_tunnel_dump)
+
+static void
+  vl_api_sw_interface_set_gtpu_bypass_t_handler
+  (vl_api_sw_interface_set_gtpu_bypass_t * mp)
+{
+  vl_api_sw_interface_set_gtpu_bypass_reply_t *rmp;
+  int rv = 0;
+  u32 sw_if_index = ntohl (mp->sw_if_index);
+  gtpu_main_t *gtm = &gtpu_main;
+
+  VALIDATE_SW_IF_INDEX (mp);
+
+  vnet_int_gtpu_bypass_mode (sw_if_index, mp->is_ipv6, mp->enable);
+  BAD_SW_IF_INDEX_LABEL;
+
+  REPLY_MACRO (VL_API_SW_INTERFACE_SET_GTPU_BYPASS_REPLY);
+}
+
+static void vl_api_gtpu_add_del_tunnel_t_handler
+  (vl_api_gtpu_add_del_tunnel_t * mp)
+{
+  vl_api_gtpu_add_del_tunnel_reply_t *rmp;
+  int rv = 0;
+  ip4_main_t *im = &ip4_main;
+  gtpu_main_t *gtm = &gtpu_main;
+
+  uword *p = hash_get (im->fib_index_by_table_id, ntohl (mp->encap_vrf_id));
+  if (!p)
+    {
+      rv = VNET_API_ERROR_NO_SUCH_FIB;
+      goto out;
+    }
+
+  vnet_gtpu_add_del_tunnel_args_t a = {
+    .is_add = mp->is_add,
+    .is_ip6 = mp->is_ipv6,
+    .mcast_sw_if_index = ntohl (mp->mcast_sw_if_index),
+    .encap_fib_index = p[0],
+    .decap_next_index = ntohl (mp->decap_next_index),
+    .teid = ntohl (mp->teid),
+    .dst = to_ip46 (mp->is_ipv6, mp->dst_address),
+    .src = to_ip46 (mp->is_ipv6, mp->src_address),
+  };
+
+  /* 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;
+  rv = vnet_gtpu_add_del_tunnel (&a, &sw_if_index);
+
+out:
+  /* *INDENT-OFF* */
+  REPLY_MACRO2(VL_API_GTPU_ADD_DEL_TUNNEL_REPLY,
+  ({
+    rmp->sw_if_index = ntohl (sw_if_index);
+  }));
+  /* *INDENT-ON* */
+}
+
+static void send_gtpu_tunnel_details
+  (gtpu_tunnel_t * t, unix_shared_memory_queue_t * q, u32 context)
+{
+  vl_api_gtpu_tunnel_details_t *rmp;
+  ip4_main_t *im4 = &ip4_main;
+  ip6_main_t *im6 = &ip6_main;
+  u8 is_ipv6 = !ip46_address_is_ip4 (&t->dst);
+
+  rmp = vl_msg_api_alloc (sizeof (*rmp));
+  memset (rmp, 0, sizeof (*rmp));
+  rmp->_vl_msg_id = ntohs (VL_API_GTPU_TUNNEL_DETAILS);
+  if (is_ipv6)
+    {
+      memcpy (rmp->src_address, t->src.ip6.as_u8, 16);
+      memcpy (rmp->dst_address, t->dst.ip6.as_u8, 16);
+      rmp->encap_vrf_id = htonl (im6->fibs[t->encap_fib_index].ft_table_id);
+    }
+  else
+    {
+      memcpy (rmp->src_address, t->src.ip4.as_u8, 4);
+      memcpy (rmp->dst_address, t->dst.ip4.as_u8, 4);
+      rmp->encap_vrf_id = htonl (im4->fibs[t->encap_fib_index].ft_table_id);
+    }
+  rmp->mcast_sw_if_index = htonl (t->mcast_sw_if_index);
+  rmp->teid = htonl (t->teid);
+  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;
+
+  vl_msg_api_send_shmem (q, (u8 *) & rmp);
+}
+
+static void
+vl_api_gtpu_tunnel_dump_t_handler (vl_api_gtpu_tunnel_dump_t * mp)
+{
+  unix_shared_memory_queue_t *q;
+  gtpu_main_t *gtm = &gtpu_main;
+  gtpu_tunnel_t *t;
+  u32 sw_if_index;
+
+  q = vl_api_client_index_to_input_queue (mp->client_index);
+  if (q == 0)
+    {
+      return;
+    }
+
+  sw_if_index = ntohl (mp->sw_if_index);
+
+  if (~0 == sw_if_index)
+    {
+      /* *INDENT-OFF* */
+      pool_foreach (t, gtm->tunnels,
+      ({
+        send_gtpu_tunnel_details(t, q, mp->context);
+      }));
+      /* *INDENT-ON* */
+    }
+  else
+    {
+      if ((sw_if_index >= vec_len (gtm->tunnel_index_by_sw_if_index)) ||
+	  (~0 == gtm->tunnel_index_by_sw_if_index[sw_if_index]))
+	{
+	  return;
+	}
+      t = &gtm->tunnels[gtm->tunnel_index_by_sw_if_index[sw_if_index]];
+      send_gtpu_tunnel_details (t, q, mp->context);
+    }
+}
+
+
+static clib_error_t *
+gtpu_api_hookup (vlib_main_t * vm)
+{
+  gtpu_main_t *gtm = &gtpu_main;
+
+  u8 *name = format (0, "gtpu_%08x%c", api_version, 0);
+  gtm->msg_id_base = vl_msg_api_get_msg_ids
+    ((char *) name, VL_MSG_FIRST_AVAILABLE);
+
+#define _(N,n)                                                  \
+    vl_msg_api_set_handlers((VL_API_##N + gtm->msg_id_base),     \
+                           #n,                  \
+                           vl_api_##n##_t_handler,              \
+                           vl_noop_handler,                     \
+                           vl_api_##n##_t_endian,               \
+                           vl_api_##n##_t_print,                \
+                           sizeof(vl_api_##n##_t), 1);
+  foreach_gtpu_plugin_api_msg;
+#undef _
+
+  /* Add our API messages to the global name_crc hash table */
+  setup_message_id_table (gtm, &api_main);
+
+  return 0;
+}
+
+VLIB_API_INIT_FUNCTION (gtpu_api_hookup);
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/gtpu/gtpu_decap.c b/src/plugins/gtpu/gtpu_decap.c
new file mode 100644
index 00000000..b0be32ec
--- /dev/null
+++ b/src/plugins/gtpu/gtpu_decap.c
@@ -0,0 +1,1305 @@
+/*
+ * decap.c: gtpu tunnel decap packet processing
+ *
+ * Copyright (c) 2013 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <vlib/vlib.h>
+#include <vnet/pg/pg.h>
+#include <gtpu/gtpu.h>
+
+vlib_node_registration_t gtpu4_input_node;
+vlib_node_registration_t gtpu6_input_node;
+
+typedef struct {
+  u32 next_index;
+  u32 tunnel_index;
+  u32 error;
+  u32 teid;
+} gtpu_rx_trace_t;
+
+static u8 * format_gtpu_rx_trace (u8 * s, va_list * args)
+{
+  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
+  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
+  gtpu_rx_trace_t * t = va_arg (*args, gtpu_rx_trace_t *);
+
+  if (t->tunnel_index != ~0)
+    {
+      s = format (s, "GTPU decap from gtpu_tunnel%d teid %d next %d error %d",
+                  t->tunnel_index, t->teid, t->next_index, t->error);
+    }
+  else
+    {
+      s = format (s, "GTPU decap error - tunnel for teid %d does not exist",
+		  t->teid);
+    }
+  return s;
+}
+
+always_inline u32
+validate_gtpu_fib (vlib_buffer_t *b, gtpu_tunnel_t *t, u32 is_ip4)
+{
+  u32 fib_index, sw_if_index;
+
+  sw_if_index = vnet_buffer (b)->sw_if_index[VLIB_RX];
+
+  if (is_ip4)
+    fib_index = (vnet_buffer (b)->sw_if_index[VLIB_TX] == (u32) ~ 0) ?
+	vec_elt (ip4_main.fib_index_by_sw_if_index, sw_if_index) :
+	vnet_buffer (b)->sw_if_index[VLIB_TX];
+  else
+    fib_index = (vnet_buffer (b)->sw_if_index[VLIB_TX] == (u32) ~ 0) ?
+	vec_elt (ip6_main.fib_index_by_sw_if_index, sw_if_index) :
+	vnet_buffer (b)->sw_if_index[VLIB_TX];
+
+  return (fib_index == t->encap_fib_index);
+}
+
+always_inline uword
+gtpu_input (vlib_main_t * vm,
+             vlib_node_runtime_t * node,
+             vlib_frame_t * from_frame,
+             u32 is_ip4)
+{
+  u32 n_left_from, next_index, * from, * to_next;
+  gtpu_main_t * gtm = &gtpu_main;
+  vnet_main_t * vnm = gtm->vnet_main;
+  vnet_interface_main_t * im = &vnm->interface_main;
+  u32 last_tunnel_index = ~0;
+  gtpu4_tunnel_key_t last_key4;
+  gtpu6_tunnel_key_t last_key6;
+  u32 pkts_decapsulated = 0;
+  u32 thread_index = vlib_get_thread_index();
+  u32 stats_sw_if_index, stats_n_packets, stats_n_bytes;
+
+  if (is_ip4)
+    last_key4.as_u64 = ~0;
+  else
+    memset (&last_key6, 0xff, sizeof (last_key6));
+
+  from = vlib_frame_vector_args (from_frame);
+  n_left_from = from_frame->n_vectors;
+
+  next_index = node->cached_next_index;
+  stats_sw_if_index = node->runtime_data[0];
+  stats_n_packets = stats_n_bytes = 0;
+
+  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;
+	  u32 next0, next1;
+          ip4_header_t * ip4_0, * ip4_1;
+          ip6_header_t * ip6_0, * ip6_1;
+          gtpu_header_t * gtpu0, * gtpu1;
+          u32 gtpu_hdr_len0 = 0, gtpu_hdr_len1 =0 ;
+	  uword * p0, * p1;
+          u32 tunnel_index0, tunnel_index1;
+          gtpu_tunnel_t * t0, * t1, * mt0 = NULL, * mt1 = NULL;
+          gtpu4_tunnel_key_t key4_0, key4_1;
+          gtpu6_tunnel_key_t key6_0, key6_1;
+          u32 error0, error1;
+	  u32 sw_if_index0, sw_if_index1, len0, len1;
+
+	  /* 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);
+
+          /* udp leaves current_data pointing at the gtpu header */
+          gtpu0 = vlib_buffer_get_current (b0);
+          gtpu1 = vlib_buffer_get_current (b1);
+          if (is_ip4) {
+	    vlib_buffer_advance
+	      (b0, -(word)(sizeof(udp_header_t)+sizeof(ip4_header_t)));
+	    vlib_buffer_advance
+	      (b1, -(word)(sizeof(udp_header_t)+sizeof(ip4_header_t)));
+            ip4_0 = vlib_buffer_get_current (b0);
+            ip4_1 = vlib_buffer_get_current (b1);
+	  } else {
+            vlib_buffer_advance
+              (b0, -(word)(sizeof(udp_header_t)+sizeof(ip6_header_t)));
+            vlib_buffer_advance
+              (b1, -(word)(sizeof(udp_header_t)+sizeof(ip6_header_t)));
+            ip6_0 = vlib_buffer_get_current (b0);
+            ip6_1 = vlib_buffer_get_current (b1);
+	  }
+
+          /* pop (ip, udp, gtpu) */
+          if (is_ip4) {
+            vlib_buffer_advance
+              (b0, sizeof(*ip4_0)+sizeof(udp_header_t));
+	    vlib_buffer_advance
+              (b1, sizeof(*ip4_1)+sizeof(udp_header_t));
+          } else {
+	    vlib_buffer_advance
+              (b0, sizeof(*ip6_0)+sizeof(udp_header_t));
+            vlib_buffer_advance
+              (b1, sizeof(*ip6_1)+sizeof(udp_header_t));
+          }
+
+          tunnel_index0 = ~0;
+          error0 = 0;
+
+          tunnel_index1 = ~0;
+          error1 = 0;
+
+	  if (PREDICT_FALSE ((gtpu0->ver_flags & GTPU_VER_MASK) != GTPU_V1_VER))
+	    {
+	      error0 = GTPU_ERROR_BAD_VER;
+	      next0 = GTPU_INPUT_NEXT_DROP;
+	      goto trace0;
+	    }
+
+	  /* Manipulate packet 0 */
+          if (is_ip4) {
+            key4_0.src = ip4_0->src_address.as_u32;
+            key4_0.teid = gtpu0->teid;
+
+ 	    /* Make sure GTPU tunnel exist according to packet SIP and teid
+ 	     * SIP identify a GTPU path, and teid identify a tunnel in a given GTPU path */
+           if (PREDICT_FALSE (key4_0.as_u64 != last_key4.as_u64))
+              {
+                p0 = hash_get (gtm->gtpu4_tunnel_by_key, key4_0.as_u64);
+                if (PREDICT_FALSE (p0 == NULL))
+                  {
+                    error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+                    next0 = GTPU_INPUT_NEXT_DROP;
+                    goto trace0;
+                  }
+                last_key4.as_u64 = key4_0.as_u64;
+                tunnel_index0 = last_tunnel_index = p0[0];
+              }
+            else
+              tunnel_index0 = last_tunnel_index;
+	    t0 = pool_elt_at_index (gtm->tunnels, tunnel_index0);
+
+	    /* Validate GTPU tunnel encap-fib index agaist packet */
+	    if (PREDICT_FALSE (validate_gtpu_fib (b0, t0, is_ip4) == 0))
+	      {
+		error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+		next0 = GTPU_INPUT_NEXT_DROP;
+		goto trace0;
+	      }
+
+	    /* Validate GTPU tunnel SIP against packet DIP */
+	    if (PREDICT_TRUE (ip4_0->dst_address.as_u32 == t0->src.ip4.as_u32))
+	      goto next0; /* valid packet */
+	    if (PREDICT_FALSE (ip4_address_is_multicast (&ip4_0->dst_address)))
+	      {
+		key4_0.src = ip4_0->dst_address.as_u32;
+		key4_0.teid = gtpu0->teid;
+		/* Make sure mcast GTPU tunnel exist by packet DIP and teid */
+		p0 = hash_get (gtm->gtpu4_tunnel_by_key, key4_0.as_u64);
+		if (PREDICT_TRUE (p0 != NULL))
+		  {
+		    mt0 = pool_elt_at_index (gtm->tunnels, p0[0]);
+		    goto next0; /* valid packet */
+		  }
+	      }
+	    error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+	    next0 = GTPU_INPUT_NEXT_DROP;
+	    goto trace0;
+
+         } else /* !is_ip4 */ {
+            key6_0.src.as_u64[0] = ip6_0->src_address.as_u64[0];
+            key6_0.src.as_u64[1] = ip6_0->src_address.as_u64[1];
+            key6_0.teid = gtpu0->teid;
+
+ 	    /* Make sure GTPU tunnel exist according to packet SIP and teid
+ 	     * SIP identify a GTPU path, and teid identify a tunnel in a given GTPU path */
+            if (PREDICT_FALSE (memcmp(&key6_0, &last_key6, sizeof(last_key6)) != 0))
+              {
+                p0 = hash_get_mem (gtm->gtpu6_tunnel_by_key, &key6_0);
+                if (PREDICT_FALSE (p0 == NULL))
+                  {
+                    error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+                    next0 = GTPU_INPUT_NEXT_DROP;
+                    goto trace0;
+                  }
+                clib_memcpy (&last_key6, &key6_0, sizeof(key6_0));
+                tunnel_index0 = last_tunnel_index = p0[0];
+              }
+            else
+              tunnel_index0 = last_tunnel_index;
+	    t0 = pool_elt_at_index (gtm->tunnels, tunnel_index0);
+
+	    /* Validate GTPU tunnel encap-fib index agaist packet */
+	    if (PREDICT_FALSE (validate_gtpu_fib (b0, t0, is_ip4) == 0))
+	      {
+		error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+		next0 = GTPU_INPUT_NEXT_DROP;
+		goto trace0;
+	      }
+
+	    /* Validate GTPU tunnel SIP against packet DIP */
+	    if (PREDICT_TRUE (ip6_address_is_equal (&ip6_0->dst_address,
+						    &t0->src.ip6)))
+		goto next0; /* valid packet */
+	    if (PREDICT_FALSE (ip6_address_is_multicast (&ip6_0->dst_address)))
+	      {
+		key6_0.src.as_u64[0] = ip6_0->dst_address.as_u64[0];
+		key6_0.src.as_u64[1] = ip6_0->dst_address.as_u64[1];
+		key6_0.teid = gtpu0->teid;
+		p0 = hash_get_mem (gtm->gtpu6_tunnel_by_key, &key6_0);
+		if (PREDICT_TRUE (p0 != NULL))
+		  {
+		    mt0 = pool_elt_at_index (gtm->tunnels, p0[0]);
+		    goto next0; /* valid packet */
+		  }
+	      }
+	    error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+	    next0 = GTPU_INPUT_NEXT_DROP;
+	    goto trace0;
+          }
+
+	next0:
+	  /* Manipulate gtpu header */
+	  if (PREDICT_FALSE((gtpu0->ver_flags & GTPU_E_S_PN_BIT) != 0))
+	    {
+	      gtpu_hdr_len0 = sizeof(gtpu_header_t);
+
+	      /* Manipulate Sequence Number and N-PDU Number */
+	      /* TBD */
+
+	      /* Manipulate Next Extension Header */
+	      /* TBD */
+	    }
+	  else
+	    {
+	      gtpu_hdr_len0 = sizeof(gtpu_header_t) - 4;
+	    }
+
+	  /* Pop gtpu header */
+	  vlib_buffer_advance (b0, gtpu_hdr_len0);
+
+          next0 = t0->decap_next_index;
+          sw_if_index0 = t0->sw_if_index;
+          len0 = vlib_buffer_length_in_chain (vm, b0);
+
+          /* Required to make the l2 tag push / pop code work on l2 subifs */
+          if (PREDICT_TRUE(next0 == GTPU_INPUT_NEXT_L2_INPUT))
+            vnet_update_l2_len (b0);
+
+          /* Set packet input sw_if_index to unicast GTPU tunnel for learning */
+          vnet_buffer(b0)->sw_if_index[VLIB_RX] = sw_if_index0;
+	  sw_if_index0 = (mt0) ? mt0->sw_if_index : sw_if_index0;
+
+          pkts_decapsulated ++;
+          stats_n_packets += 1;
+          stats_n_bytes += len0;
+
+	  /* Batch stats increment on the same gtpu tunnel so counter
+	     is not incremented per packet */
+	  if (PREDICT_FALSE (sw_if_index0 != stats_sw_if_index))
+	    {
+	      stats_n_packets -= 1;
+	      stats_n_bytes -= len0;
+	      if (stats_n_packets)
+		vlib_increment_combined_counter
+		  (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_RX,
+		   thread_index, stats_sw_if_index,
+		   stats_n_packets, stats_n_bytes);
+	      stats_n_packets = 1;
+	      stats_n_bytes = len0;
+	      stats_sw_if_index = sw_if_index0;
+	    }
+
+        trace0:
+          b0->error = error0 ? node->errors[error0] : 0;
+
+          if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
+            {
+              gtpu_rx_trace_t *tr
+                = vlib_add_trace (vm, node, b0, sizeof (*tr));
+              tr->next_index = next0;
+              tr->error = error0;
+              tr->tunnel_index = tunnel_index0;
+              tr->teid = clib_net_to_host_u32(gtpu0->teid);
+            }
+
+          if (PREDICT_FALSE ((gtpu1->ver_flags & GTPU_VER_MASK) != GTPU_V1_VER))
+	    {
+	      error1 = GTPU_ERROR_BAD_VER;
+	      next1 = GTPU_INPUT_NEXT_DROP;
+	      goto trace1;
+	    }
+
+          /* Manipulate packet 1 */
+          if (is_ip4) {
+            key4_1.src = ip4_1->src_address.as_u32;
+            key4_1.teid = gtpu1->teid;
+
+ 	    /* Make sure GTPU tunnel exist according to packet SIP and teid
+ 	     * SIP identify a GTPU path, and teid identify a tunnel in a given GTPU path */
+	    if (PREDICT_FALSE (key4_1.as_u64 != last_key4.as_u64))
+              {
+                p1 = hash_get (gtm->gtpu4_tunnel_by_key, key4_1.as_u64);
+                if (PREDICT_FALSE (p1 == NULL))
+                  {
+                    error1 = GTPU_ERROR_NO_SUCH_TUNNEL;
+                    next1 = GTPU_INPUT_NEXT_DROP;
+                    goto trace1;
+                  }
+                last_key4.as_u64 = key4_1.as_u64;
+                tunnel_index1 = last_tunnel_index = p1[0];
+              }
+            else
+              tunnel_index1 = last_tunnel_index;
+ 	    t1 = pool_elt_at_index (gtm->tunnels, tunnel_index1);
+
+	    /* Validate GTPU tunnel encap-fib index agaist packet */
+	    if (PREDICT_FALSE (validate_gtpu_fib (b1, t1, is_ip4) == 0))
+	      {
+		error1 = GTPU_ERROR_NO_SUCH_TUNNEL;
+		next1 = GTPU_INPUT_NEXT_DROP;
+		goto trace1;
+	      }
+
+	    /* Validate GTPU tunnel SIP against packet DIP */
+	    if (PREDICT_TRUE (ip4_1->dst_address.as_u32 == t1->src.ip4.as_u32))
+	      goto next1; /* valid packet */
+	    if (PREDICT_FALSE (ip4_address_is_multicast (&ip4_1->dst_address)))
+	      {
+		key4_1.src = ip4_1->dst_address.as_u32;
+		key4_1.teid = gtpu1->teid;
+		/* Make sure mcast GTPU tunnel exist by packet DIP and teid */
+		p1 = hash_get (gtm->gtpu4_tunnel_by_key, key4_1.as_u64);
+		if (PREDICT_TRUE (p1 != NULL))
+		  {
+		    mt1 = pool_elt_at_index (gtm->tunnels, p1[0]);
+		    goto next1; /* valid packet */
+		  }
+	      }
+	    error1 = GTPU_ERROR_NO_SUCH_TUNNEL;
+	    next1 = GTPU_INPUT_NEXT_DROP;
+	    goto trace1;
+
+         } else /* !is_ip4 */ {
+            key6_1.src.as_u64[0] = ip6_1->src_address.as_u64[0];
+            key6_1.src.as_u64[1] = ip6_1->src_address.as_u64[1];
+            key6_1.teid = gtpu1->teid;
+
+ 	    /* Make sure GTPU tunnel exist according to packet SIP and teid
+ 	     * SIP identify a GTPU path, and teid identify a tunnel in a given GTPU path */
+            if (PREDICT_FALSE (memcmp(&key6_1, &last_key6, sizeof(last_key6)) != 0))
+              {
+                p1 = hash_get_mem (gtm->gtpu6_tunnel_by_key, &key6_1);
+
+                if (PREDICT_FALSE (p1 == NULL))
+                  {
+                    error1 = GTPU_ERROR_NO_SUCH_TUNNEL;
+                    next1 = GTPU_INPUT_NEXT_DROP;
+                    goto trace1;
+                  }
+
+                clib_memcpy (&last_key6, &key6_1, sizeof(key6_1));
+                tunnel_index1 = last_tunnel_index = p1[0];
+              }
+            else
+              tunnel_index1 = last_tunnel_index;
+ 	    t1 = pool_elt_at_index (gtm->tunnels, tunnel_index1);
+
+	    /* Validate GTPU tunnel encap-fib index agaist packet */
+	    if (PREDICT_FALSE (validate_gtpu_fib (b1, t1, is_ip4) == 0))
+	      {
+		error1 = GTPU_ERROR_NO_SUCH_TUNNEL;
+		next1 = GTPU_INPUT_NEXT_DROP;
+		goto trace1;
+	      }
+
+	    /* Validate GTPU tunnel SIP against packet DIP */
+	    if (PREDICT_TRUE (ip6_address_is_equal (&ip6_1->dst_address,
+						    &t1->src.ip6)))
+		goto next1; /* valid packet */
+	    if (PREDICT_FALSE (ip6_address_is_multicast (&ip6_1->dst_address)))
+	      {
+		key6_1.src.as_u64[0] = ip6_1->dst_address.as_u64[0];
+		key6_1.src.as_u64[1] = ip6_1->dst_address.as_u64[1];
+		key6_1.teid = gtpu1->teid;
+		p1 = hash_get_mem (gtm->gtpu6_tunnel_by_key, &key6_1);
+		if (PREDICT_TRUE (p1 != NULL))
+		  {
+		    mt1 = pool_elt_at_index (gtm->tunnels, p1[0]);
+		    goto next1; /* valid packet */
+		  }
+	      }
+	    error1 = GTPU_ERROR_NO_SUCH_TUNNEL;
+	    next1 = GTPU_INPUT_NEXT_DROP;
+	    goto trace1;
+	  }
+
+	next1:
+	  /* Manipulate gtpu header */
+	  if (PREDICT_FALSE((gtpu1->ver_flags & GTPU_E_S_PN_BIT) != 0))
+	    {
+	      gtpu_hdr_len1 = sizeof(gtpu_header_t);
+
+	      /* Manipulate Sequence Number and N-PDU Number */
+	      /* TBD */
+
+	      /* Manipulate Next Extension Header */
+	      /* TBD */
+	    }
+	  else
+	    {
+	      gtpu_hdr_len1 = sizeof(gtpu_header_t) - 4;
+	    }
+
+	  /* Pop gtpu header */
+	  vlib_buffer_advance (b1, gtpu_hdr_len1);
+
+          next1 = t1->decap_next_index;
+          sw_if_index1 = t1->sw_if_index;
+          len1 = vlib_buffer_length_in_chain (vm, b1);
+
+          /* Required to make the l2 tag push / pop code work on l2 subifs */
+          if (PREDICT_TRUE(next1 == GTPU_INPUT_NEXT_L2_INPUT))
+            vnet_update_l2_len (b1);
+
+          /* Set packet input sw_if_index to unicast GTPU tunnel for learning */
+          vnet_buffer(b1)->sw_if_index[VLIB_RX] = sw_if_index1;
+	  sw_if_index1 = (mt1) ? mt1->sw_if_index : sw_if_index1;
+
+          pkts_decapsulated ++;
+          stats_n_packets += 1;
+          stats_n_bytes += len1;
+
+	  /* Batch stats increment on the same gtpu tunnel so counter
+	     is not incremented per packet */
+	  if (PREDICT_FALSE (sw_if_index1 != stats_sw_if_index))
+	    {
+	      stats_n_packets -= 1;
+	      stats_n_bytes -= len1;
+	      if (stats_n_packets)
+		vlib_increment_combined_counter
+		  (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_RX,
+		   thread_index, stats_sw_if_index,
+		   stats_n_packets, stats_n_bytes);
+	      stats_n_packets = 1;
+	      stats_n_bytes = len1;
+	      stats_sw_if_index = sw_if_index1;
+	    }
+
+        trace1:
+          b1->error = error1 ? node->errors[error1] : 0;
+
+          if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED))
+            {
+              gtpu_rx_trace_t *tr
+                = vlib_add_trace (vm, node, b1, sizeof (*tr));
+              tr->next_index = next1;
+              tr->error = error1;
+              tr->tunnel_index = tunnel_index1;
+              tr->teid = clib_net_to_host_u32(gtpu1->teid);
+            }
+
+	  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)
+	{
+	  u32 bi0;
+	  vlib_buffer_t * b0;
+	  u32 next0;
+          ip4_header_t * ip4_0;
+          ip6_header_t * ip6_0;
+          gtpu_header_t * gtpu0;
+          u32 gtpu_hdr_len0 = 0;
+	  uword * p0;
+          u32 tunnel_index0;
+          gtpu_tunnel_t * t0, * mt0 = NULL;
+          gtpu4_tunnel_key_t key4_0;
+          gtpu6_tunnel_key_t key6_0;
+          u32 error0;
+	  u32 sw_if_index0, len0;
+
+	  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);
+
+          /* udp leaves current_data pointing at the gtpu header */
+          gtpu0 = vlib_buffer_get_current (b0);
+          if (is_ip4) {
+	    vlib_buffer_advance
+	      (b0, -(word)(sizeof(udp_header_t)+sizeof(ip4_header_t)));
+            ip4_0 = vlib_buffer_get_current (b0);
+          } else {
+            vlib_buffer_advance
+              (b0, -(word)(sizeof(udp_header_t)+sizeof(ip6_header_t)));
+            ip6_0 = vlib_buffer_get_current (b0);
+          }
+
+          /* pop (ip, udp) */
+          if (is_ip4) {
+            vlib_buffer_advance
+              (b0, sizeof(*ip4_0)+sizeof(udp_header_t));
+          } else {
+	    vlib_buffer_advance
+              (b0, sizeof(*ip6_0)+sizeof(udp_header_t));
+          }
+
+          tunnel_index0 = ~0;
+          error0 = 0;
+          if (PREDICT_FALSE ((gtpu0->ver_flags & GTPU_VER_MASK) != GTPU_V1_VER))
+	    {
+	      error0 = GTPU_ERROR_BAD_VER;
+	      next0 = GTPU_INPUT_NEXT_DROP;
+	      goto trace00;
+	    }
+
+          if (is_ip4) {
+            key4_0.src = ip4_0->src_address.as_u32;
+            key4_0.teid = gtpu0->teid;
+
+ 	    /* Make sure GTPU tunnel exist according to packet SIP and teid
+ 	     * SIP identify a GTPU path, and teid identify a tunnel in a given GTPU path */
+            if (PREDICT_FALSE (key4_0.as_u64 != last_key4.as_u64))
+              {
+                p0 = hash_get (gtm->gtpu4_tunnel_by_key, key4_0.as_u64);
+                if (PREDICT_FALSE (p0 == NULL))
+                  {
+                    error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+                    next0 = GTPU_INPUT_NEXT_DROP;
+                    goto trace00;
+                  }
+                last_key4.as_u64 = key4_0.as_u64;
+                tunnel_index0 = last_tunnel_index = p0[0];
+              }
+            else
+              tunnel_index0 = last_tunnel_index;
+	    t0 = pool_elt_at_index (gtm->tunnels, tunnel_index0);
+
+	    /* Validate GTPU tunnel encap-fib index agaist packet */
+	    if (PREDICT_FALSE (validate_gtpu_fib (b0, t0, is_ip4) == 0))
+	      {
+		error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+		next0 = GTPU_INPUT_NEXT_DROP;
+		goto trace00;
+	      }
+
+	    /* Validate GTPU tunnel SIP against packet DIP */
+	    if (PREDICT_TRUE (ip4_0->dst_address.as_u32 == t0->src.ip4.as_u32))
+	      goto next00; /* valid packet */
+	    if (PREDICT_FALSE (ip4_address_is_multicast (&ip4_0->dst_address)))
+	      {
+		key4_0.src = ip4_0->dst_address.as_u32;
+		key4_0.teid = gtpu0->teid;
+		/* Make sure mcast GTPU tunnel exist by packet DIP and teid */
+		p0 = hash_get (gtm->gtpu4_tunnel_by_key, key4_0.as_u64);
+		if (PREDICT_TRUE (p0 != NULL))
+		  {
+		    mt0 = pool_elt_at_index (gtm->tunnels, p0[0]);
+		    goto next00; /* valid packet */
+		  }
+	      }
+	    error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+	    next0 = GTPU_INPUT_NEXT_DROP;
+	    goto trace00;
+
+          } else /* !is_ip4 */ {
+            key6_0.src.as_u64[0] = ip6_0->src_address.as_u64[0];
+            key6_0.src.as_u64[1] = ip6_0->src_address.as_u64[1];
+            key6_0.teid = gtpu0->teid;
+
+ 	    /* Make sure GTPU tunnel exist according to packet SIP and teid
+ 	     * SIP identify a GTPU path, and teid identify a tunnel in a given GTPU path */
+            if (PREDICT_FALSE (memcmp(&key6_0, &last_key6, sizeof(last_key6)) != 0))
+              {
+                p0 = hash_get_mem (gtm->gtpu6_tunnel_by_key, &key6_0);
+                if (PREDICT_FALSE (p0 == NULL))
+                  {
+                    error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+                    next0 = GTPU_INPUT_NEXT_DROP;
+                    goto trace00;
+                  }
+                clib_memcpy (&last_key6, &key6_0, sizeof(key6_0));
+                tunnel_index0 = last_tunnel_index = p0[0];
+              }
+            else
+              tunnel_index0 = last_tunnel_index;
+	    t0 = pool_elt_at_index (gtm->tunnels, tunnel_index0);
+
+	    /* Validate GTPU tunnel encap-fib index agaist packet */
+	    if (PREDICT_FALSE (validate_gtpu_fib (b0, t0, is_ip4) == 0))
+	      {
+		error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+		next0 = GTPU_INPUT_NEXT_DROP;
+		goto trace00;
+	      }
+
+	    /* Validate GTPU tunnel SIP against packet DIP */
+	    if (PREDICT_TRUE (ip6_address_is_equal (&ip6_0->dst_address,
+						    &t0->src.ip6)))
+		goto next00; /* valid packet */
+	    if (PREDICT_FALSE (ip6_address_is_multicast (&ip6_0->dst_address)))
+	      {
+		key6_0.src.as_u64[0] = ip6_0->dst_address.as_u64[0];
+		key6_0.src.as_u64[1] = ip6_0->dst_address.as_u64[1];
+		key6_0.teid = gtpu0->teid;
+		p0 = hash_get_mem (gtm->gtpu6_tunnel_by_key, &key6_0);
+		if (PREDICT_TRUE (p0 != NULL))
+		  {
+		    mt0 = pool_elt_at_index (gtm->tunnels, p0[0]);
+		    goto next00; /* valid packet */
+		  }
+	      }
+	    error0 = GTPU_ERROR_NO_SUCH_TUNNEL;
+	    next0 = GTPU_INPUT_NEXT_DROP;
+	    goto trace00;
+          }
+
+	next00:
+	  /* Manipulate gtpu header */
+	  if (PREDICT_FALSE((gtpu0->ver_flags & GTPU_E_S_PN_BIT) != 0))
+	    {
+	      gtpu_hdr_len0 = sizeof(gtpu_header_t);
+
+	      /* Manipulate Sequence Number and N-PDU Number */
+	      /* TBD */
+
+	      /* Manipulate Next Extension Header */
+	      /* TBD */
+	    }
+	  else
+	    {
+	      gtpu_hdr_len0 = sizeof(gtpu_header_t) - 4;
+	    }
+
+	  /* Pop gtpu header */
+	  vlib_buffer_advance (b0, gtpu_hdr_len0);
+
+	  next0 = t0->decap_next_index;
+	  sw_if_index0 = t0->sw_if_index;
+	  len0 = vlib_buffer_length_in_chain (vm, b0);
+
+          /* Required to make the l2 tag push / pop code work on l2 subifs */
+          if (PREDICT_TRUE(next0 == GTPU_INPUT_NEXT_L2_INPUT))
+            vnet_update_l2_len (b0);
+
+          /* Set packet input sw_if_index to unicast GTPU tunnel for learning */
+          vnet_buffer(b0)->sw_if_index[VLIB_RX] = sw_if_index0;
+	  sw_if_index0 = (mt0) ? mt0->sw_if_index : sw_if_index0;
+
+          pkts_decapsulated ++;
+          stats_n_packets += 1;
+          stats_n_bytes += len0;
+
+	  /* Batch stats increment on the same gtpu tunnel so counter
+	     is not incremented per packet */
+	  if (PREDICT_FALSE (sw_if_index0 != stats_sw_if_index))
+	    {
+	      stats_n_packets -= 1;
+	      stats_n_bytes -= len0;
+	      if (stats_n_packets)
+		vlib_increment_combined_counter
+		  (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_RX,
+		   thread_index, stats_sw_if_index,
+		   stats_n_packets, stats_n_bytes);
+	      stats_n_packets = 1;
+	      stats_n_bytes = len0;
+	      stats_sw_if_index = sw_if_index0;
+	    }
+
+        trace00:
+          b0->error = error0 ? node->errors[error0] : 0;
+
+          if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
+            {
+              gtpu_rx_trace_t *tr
+                = vlib_add_trace (vm, node, b0, sizeof (*tr));
+              tr->next_index = next0;
+              tr->error = error0;
+              tr->tunnel_index = tunnel_index0;
+              tr->teid = clib_net_to_host_u32(gtpu0->teid);
+            }
+	  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);
+    }
+  /* Do we still need this now that tunnel tx stats is kept? */
+  vlib_node_increment_counter (vm, is_ip4?
+			       gtpu4_input_node.index:gtpu6_input_node.index,
+                               GTPU_ERROR_DECAPSULATED,
+                               pkts_decapsulated);
+
+  /* Increment any remaining batch stats */
+  if (stats_n_packets)
+    {
+      vlib_increment_combined_counter
+	(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_RX,
+	 thread_index, stats_sw_if_index, stats_n_packets, stats_n_bytes);
+      node->runtime_data[0] = stats_sw_if_index;
+    }
+
+  return from_frame->n_vectors;
+}
+
+static uword
+gtpu4_input (vlib_main_t * vm,
+             vlib_node_runtime_t * node,
+             vlib_frame_t * from_frame)
+{
+	return gtpu_input(vm, node, from_frame, /* is_ip4 */ 1);
+}
+
+static uword
+gtpu6_input (vlib_main_t * vm,
+             vlib_node_runtime_t * node,
+             vlib_frame_t * from_frame)
+{
+	return gtpu_input(vm, node, from_frame, /* is_ip4 */ 0);
+}
+
+static char * gtpu_error_strings[] = {
+#define gtpu_error(n,s) s,
+#include <gtpu/gtpu_error.def>
+#undef gtpu_error
+#undef _
+};
+
+VLIB_REGISTER_NODE (gtpu4_input_node) = {
+  .function = gtpu4_input,
+  .name = "gtpu4-input",
+  /* Takes a vector of packets. */
+  .vector_size = sizeof (u32),
+
+  .n_errors = GTPU_N_ERROR,
+  .error_strings = gtpu_error_strings,
+
+  .n_next_nodes = GTPU_INPUT_N_NEXT,
+  .next_nodes = {
+#define _(s,n) [GTPU_INPUT_NEXT_##s] = n,
+    foreach_gtpu_input_next
+#undef _
+  },
+
+//temp  .format_buffer = format_gtpu_header,
+  .format_trace = format_gtpu_rx_trace,
+  // $$$$ .unformat_buffer = unformat_gtpu_header,
+};
+
+VLIB_NODE_FUNCTION_MULTIARCH (gtpu4_input_node, gtpu4_input)
+
+VLIB_REGISTER_NODE (gtpu6_input_node) = {
+  .function = gtpu6_input,
+  .name = "gtpu6-input",
+  /* Takes a vector of packets. */
+  .vector_size = sizeof (u32),
+
+  .n_errors = GTPU_N_ERROR,
+  .error_strings = gtpu_error_strings,
+
+  .n_next_nodes = GTPU_INPUT_N_NEXT,
+  .next_nodes = {
+#define _(s,n) [GTPU_INPUT_NEXT_##s] = n,
+    foreach_gtpu_input_next
+#undef _
+  },
+
+//temp  .format_buffer = format_gtpu_header,
+  .format_trace = format_gtpu_rx_trace,
+  // $$$$ .unformat_buffer = unformat_gtpu_header,
+};
+
+VLIB_NODE_FUNCTION_MULTIARCH (gtpu6_input_node, gtpu6_input)
+
+
+typedef enum {
+  IP_GTPU_BYPASS_NEXT_DROP,
+  IP_GTPU_BYPASS_NEXT_GTPU,
+  IP_GTPU_BYPASS_N_NEXT,
+} ip_vxan_bypass_next_t;
+
+always_inline uword
+ip_gtpu_bypass_inline (vlib_main_t * vm,
+			vlib_node_runtime_t * node,
+			vlib_frame_t * frame,
+			u32 is_ip4)
+{
+  gtpu_main_t * gtm = &gtpu_main;
+  u32 * from, * to_next, n_left_from, n_left_to_next, next_index;
+  vlib_node_runtime_t * error_node = vlib_node_get_runtime (vm, ip4_input_node.index);
+  ip4_address_t addr4; /* last IPv4 address matching a local VTEP address */
+  ip6_address_t addr6; /* last IPv6 address matching a local VTEP address */
+
+  from = vlib_frame_vector_args (frame);
+  n_left_from = frame->n_vectors;
+  next_index = node->cached_next_index;
+
+  if (node->flags & VLIB_NODE_FLAG_TRACE)
+    ip4_forward_next_trace (vm, node, frame, VLIB_TX);
+
+  if (is_ip4) addr4.data_u32 = ~0;
+  else ip6_address_set_zero (&addr6);
+
+  while (n_left_from > 0)
+    {
+      vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
+
+      while (n_left_from >= 4 && n_left_to_next >= 2)
+      	{
+      	  vlib_buffer_t * b0, * b1;
+      	  ip4_header_t * ip40, * ip41;
+      	  ip6_header_t * ip60, * ip61;
+      	  udp_header_t * udp0, * udp1;
+      	  u32 bi0, ip_len0, udp_len0, flags0, next0;
+      	  u32 bi1, ip_len1, udp_len1, flags1, next1;
+      	  i32 len_diff0, len_diff1;
+      	  u8 error0, good_udp0, proto0;
+      	  u8 error1, good_udp1, proto1;
+
+	  /* 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 = to_next[0] = from[0];
+      	  bi1 = to_next[1] = from[1];
+      	  from += 2;
+      	  n_left_from -= 2;
+      	  to_next += 2;
+      	  n_left_to_next -= 2;
+
+	  b0 = vlib_get_buffer (vm, bi0);
+	  b1 = vlib_get_buffer (vm, bi1);
+	  if (is_ip4)
+	    {
+	      ip40 = vlib_buffer_get_current (b0);
+	      ip41 = vlib_buffer_get_current (b1);
+	    }
+	  else
+	    {
+	      ip60 = vlib_buffer_get_current (b0);
+	      ip61 = vlib_buffer_get_current (b1);
+	    }
+
+	  /* Setup packet for next IP feature */
+	  vnet_feature_next(vnet_buffer(b0)->sw_if_index[VLIB_RX], &next0, b0);
+	  vnet_feature_next(vnet_buffer(b1)->sw_if_index[VLIB_RX], &next1, b1);
+
+	  if (is_ip4)
+	    {
+	      /* Treat IP frag packets as "experimental" protocol for now
+		 until support of IP frag reassembly is implemented */
+	      proto0 = ip4_is_fragment(ip40) ? 0xfe : ip40->protocol;
+	      proto1 = ip4_is_fragment(ip41) ? 0xfe : ip41->protocol;
+	    }
+	  else
+	    {
+	      proto0 = ip60->protocol;
+	      proto1 = ip61->protocol;
+	    }
+
+	  /* Process packet 0 */
+	  if (proto0 != IP_PROTOCOL_UDP)
+	    goto exit0; /* not UDP packet */
+
+	  if (is_ip4)
+	    udp0 = ip4_next_header (ip40);
+	  else
+	    udp0 = ip6_next_header (ip60);
+
+	  if (udp0->dst_port != clib_host_to_net_u16 (UDP_DST_PORT_GTPU))
+	    goto exit0; /* not GTPU packet */
+
+	  /* Validate DIP against VTEPs*/
+	  if (is_ip4)
+	    {
+	      if (addr4.as_u32 != ip40->dst_address.as_u32)
+	        {
+		  if (!hash_get (gtm->vtep4, ip40->dst_address.as_u32))
+		      goto exit0; /* no local VTEP for GTPU packet */
+		  addr4 = ip40->dst_address;
+	        }
+	    }
+	  else
+	    {
+	      if (!ip6_address_is_equal (&addr6, &ip60->dst_address))
+	        {
+		  if (!hash_get_mem (gtm->vtep6, &ip60->dst_address))
+		      goto exit0; /* no local VTEP for GTPU packet */
+		  addr6 = ip60->dst_address;
+	        }
+	    }
+
+	  flags0 = b0->flags;
+	  good_udp0 = (flags0 & IP_BUFFER_L4_CHECKSUM_CORRECT) != 0;
+
+	  /* Don't verify UDP checksum for packets with explicit zero checksum. */
+	  good_udp0 |= udp0->checksum == 0;
+
+	  /* Verify UDP length */
+	  if (is_ip4)
+	    ip_len0 = clib_net_to_host_u16 (ip40->length);
+	  else
+	    ip_len0 = clib_net_to_host_u16 (ip60->payload_length);
+	  udp_len0 = clib_net_to_host_u16 (udp0->length);
+	  len_diff0 = ip_len0 - udp_len0;
+
+	  /* Verify UDP checksum */
+	  if (PREDICT_FALSE (!good_udp0))
+	    {
+	      if ((flags0 & IP_BUFFER_L4_CHECKSUM_COMPUTED) == 0)
+	        {
+		  if (is_ip4)
+		    flags0 = ip4_tcp_udp_validate_checksum (vm, b0);
+		  else
+		    flags0 = ip6_tcp_udp_icmp_validate_checksum (vm, b0);
+		  good_udp0 =
+		    (flags0 & IP_BUFFER_L4_CHECKSUM_CORRECT) != 0;
+	        }
+	    }
+
+	  if (is_ip4)
+	    {
+	      error0 = good_udp0 ? 0 : IP4_ERROR_UDP_CHECKSUM;
+	      error0 = (len_diff0 >= 0) ? error0 : IP4_ERROR_UDP_LENGTH;
+	    }
+	  else
+	    {
+	      error0 = good_udp0 ? 0 : IP6_ERROR_UDP_CHECKSUM;
+	      error0 = (len_diff0 >= 0) ? error0 : IP6_ERROR_UDP_LENGTH;
+	    }
+
+	  next0 = error0 ?
+	    IP_GTPU_BYPASS_NEXT_DROP : IP_GTPU_BYPASS_NEXT_GTPU;
+	  b0->error = error0 ? error_node->errors[error0] : 0;
+
+	  /* gtpu-input node expect current at GTPU header */
+	  if (is_ip4)
+	    vlib_buffer_advance (b0, sizeof(ip4_header_t)+sizeof(udp_header_t));
+	  else
+	    vlib_buffer_advance (b0, sizeof(ip6_header_t)+sizeof(udp_header_t));
+
+	exit0:
+	  /* Process packet 1 */
+	  if (proto1 != IP_PROTOCOL_UDP)
+	    goto exit1; /* not UDP packet */
+
+	  if (is_ip4)
+	    udp1 = ip4_next_header (ip41);
+	  else
+	    udp1 = ip6_next_header (ip61);
+
+	  if (udp1->dst_port != clib_host_to_net_u16 (UDP_DST_PORT_GTPU))
+	    goto exit1; /* not GTPU packet */
+
+	  /* Validate DIP against VTEPs*/
+	  if (is_ip4)
+	    {
+	      if (addr4.as_u32 != ip41->dst_address.as_u32)
+	        {
+		  if (!hash_get (gtm->vtep4, ip41->dst_address.as_u32))
+		      goto exit1; /* no local VTEP for GTPU packet */
+		  addr4 = ip41->dst_address;
+		}
+	    }
+	  else
+	    {
+	      if (!ip6_address_is_equal (&addr6, &ip61->dst_address))
+	        {
+		  if (!hash_get_mem (gtm->vtep6, &ip61->dst_address))
+		      goto exit1; /* no local VTEP for GTPU packet */
+		  addr6 = ip61->dst_address;
+		}
+	    }
+
+	  flags1 = b1->flags;
+	  good_udp1 = (flags1 & IP_BUFFER_L4_CHECKSUM_CORRECT) != 0;
+
+	  /* Don't verify UDP checksum for packets with explicit zero checksum. */
+	  good_udp1 |= udp1->checksum == 0;
+
+	  /* Verify UDP length */
+	  if (is_ip4)
+	    ip_len1 = clib_net_to_host_u16 (ip41->length);
+	  else
+	    ip_len1 = clib_net_to_host_u16 (ip61->payload_length);
+	  udp_len1 = clib_net_to_host_u16 (udp1->length);
+	  len_diff1 = ip_len1 - udp_len1;
+
+	  /* Verify UDP checksum */
+	  if (PREDICT_FALSE (!good_udp1))
+	    {
+	      if ((flags1 & IP_BUFFER_L4_CHECKSUM_COMPUTED) == 0)
+	        {
+		  if (is_ip4)
+		    flags1 = ip4_tcp_udp_validate_checksum (vm, b1);
+		  else
+		    flags1 = ip6_tcp_udp_icmp_validate_checksum (vm, b1);
+		  good_udp1 =
+		    (flags1 & IP_BUFFER_L4_CHECKSUM_CORRECT) != 0;
+	        }
+	    }
+
+	  if (is_ip4)
+	    {
+	      error1 = good_udp1 ? 0 : IP4_ERROR_UDP_CHECKSUM;
+	      error1 = (len_diff1 >= 0) ? error1 : IP4_ERROR_UDP_LENGTH;
+	    }
+	  else
+	    {
+	      error1 = good_udp1 ? 0 : IP6_ERROR_UDP_CHECKSUM;
+	      error1 = (len_diff1 >= 0) ? error1 : IP6_ERROR_UDP_LENGTH;
+	    }
+
+	  next1 = error1 ?
+	    IP_GTPU_BYPASS_NEXT_DROP : IP_GTPU_BYPASS_NEXT_GTPU;
+	  b1->error = error1 ? error_node->errors[error1] : 0;
+
+	  /* gtpu-input node expect current at GTPU header */
+	  if (is_ip4)
+	    vlib_buffer_advance (b1, sizeof(ip4_header_t)+sizeof(udp_header_t));
+	  else
+	    vlib_buffer_advance (b1, sizeof(ip6_header_t)+sizeof(udp_header_t));
+
+	exit1:
+	  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)
+	{
+	  vlib_buffer_t * b0;
+	  ip4_header_t * ip40;
+	  ip6_header_t * ip60;
+	  udp_header_t * udp0;
+      	  u32 bi0, ip_len0, udp_len0, flags0, next0;
+	  i32 len_diff0;
+	  u8 error0, good_udp0, proto0;
+
+	  bi0 = to_next[0] = from[0];
+	  from += 1;
+	  n_left_from -= 1;
+	  to_next += 1;
+	  n_left_to_next -= 1;
+
+	  b0 = vlib_get_buffer (vm, bi0);
+	  if (is_ip4)
+	    ip40 = vlib_buffer_get_current (b0);
+	  else
+	    ip60 = vlib_buffer_get_current (b0);
+
+	  /* Setup packet for next IP feature */
+	  vnet_feature_next(vnet_buffer(b0)->sw_if_index[VLIB_RX], &next0, b0);
+
+	  if (is_ip4)
+	    /* Treat IP4 frag packets as "experimental" protocol for now
+	       until support of IP frag reassembly is implemented */
+	    proto0 = ip4_is_fragment(ip40) ? 0xfe : ip40->protocol;
+	  else
+	    proto0 = ip60->protocol;
+
+	  if (proto0 != IP_PROTOCOL_UDP)
+	    goto exit; /* not UDP packet */
+
+	  if (is_ip4)
+	    udp0 = ip4_next_header (ip40);
+	  else
+	    udp0 = ip6_next_header (ip60);
+
+	  if (udp0->dst_port != clib_host_to_net_u16 (UDP_DST_PORT_GTPU))
+	    goto exit; /* not GTPU packet */
+
+	  /* Validate DIP against VTEPs*/
+	  if (is_ip4)
+	    {
+	      if (addr4.as_u32 != ip40->dst_address.as_u32)
+	        {
+		  if (!hash_get (gtm->vtep4, ip40->dst_address.as_u32))
+		      goto exit; /* no local VTEP for GTPU packet */
+		  addr4 = ip40->dst_address;
+		}
+	    }
+	  else
+	    {
+	      if (!ip6_address_is_equal (&addr6, &ip60->dst_address))
+	        {
+		  if (!hash_get_mem (gtm->vtep6, &ip60->dst_address))
+		      goto exit; /* no local VTEP for GTPU packet */
+		  addr6 = ip60->dst_address;
+		}
+	    }
+
+	  flags0 = b0->flags;
+	  good_udp0 = (flags0 & IP_BUFFER_L4_CHECKSUM_CORRECT) != 0;
+
+	  /* Don't verify UDP checksum for packets with explicit zero checksum. */
+	  good_udp0 |= udp0->checksum == 0;
+
+	  /* Verify UDP length */
+	  if (is_ip4)
+	    ip_len0 = clib_net_to_host_u16 (ip40->length);
+	  else
+	    ip_len0 = clib_net_to_host_u16 (ip60->payload_length);
+	  udp_len0 = clib_net_to_host_u16 (udp0->length);
+	  len_diff0 = ip_len0 - udp_len0;
+
+	  /* Verify UDP checksum */
+	  if (PREDICT_FALSE (!good_udp0))
+	    {
+	      if ((flags0 & IP_BUFFER_L4_CHECKSUM_COMPUTED) == 0)
+	        {
+		  if (is_ip4)
+		    flags0 = ip4_tcp_udp_validate_checksum (vm, b0);
+		  else
+		    flags0 = ip6_tcp_udp_icmp_validate_checksum (vm, b0);
+		  good_udp0 =
+		    (flags0 & IP_BUFFER_L4_CHECKSUM_CORRECT) != 0;
+	        }
+	    }
+
+	  if (is_ip4)
+	    {
+	      error0 = good_udp0 ? 0 : IP4_ERROR_UDP_CHECKSUM;
+	      error0 = (len_diff0 >= 0) ? error0 : IP4_ERROR_UDP_LENGTH;
+	    }
+	  else
+	    {
+	      error0 = good_udp0 ? 0 : IP6_ERROR_UDP_CHECKSUM;
+	      error0 = (len_diff0 >= 0) ? error0 : IP6_ERROR_UDP_LENGTH;
+	    }
+
+	  next0 = error0 ?
+	    IP_GTPU_BYPASS_NEXT_DROP : IP_GTPU_BYPASS_NEXT_GTPU;
+	  b0->error = error0 ? error_node->errors[error0] : 0;
+
+	  /* gtpu-input node expect current at GTPU header */
+	  if (is_ip4)
+	    vlib_buffer_advance (b0, sizeof(ip4_header_t)+sizeof(udp_header_t));
+	  else
+	    vlib_buffer_advance (b0, sizeof(ip6_header_t)+sizeof(udp_header_t));
+
+	exit:
+	  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);
+    }
+
+  return frame->n_vectors;
+}
+
+static uword
+ip4_gtpu_bypass (vlib_main_t * vm,
+		  vlib_node_runtime_t * node,
+		  vlib_frame_t * frame)
+{
+  return ip_gtpu_bypass_inline (vm, node, frame, /* is_ip4 */ 1);
+}
+
+VLIB_REGISTER_NODE (ip4_gtpu_bypass_node) = {
+  .function = ip4_gtpu_bypass,
+  .name = "ip4-gtpu-bypass",
+  .vector_size = sizeof (u32),
+
+  .n_next_nodes = IP_GTPU_BYPASS_N_NEXT,
+  .next_nodes = {
+    [IP_GTPU_BYPASS_NEXT_DROP] = "error-drop",
+    [IP_GTPU_BYPASS_NEXT_GTPU] = "gtpu4-input",
+  },
+
+  .format_buffer = format_ip4_header,
+  .format_trace = format_ip4_forward_next_trace,
+};
+
+VLIB_NODE_FUNCTION_MULTIARCH (ip4_gtpu_bypass_node,ip4_gtpu_bypass)
+
+/* Dummy init function to get us linked in. */
+clib_error_t * ip4_gtpu_bypass_init (vlib_main_t * vm)
+{ return 0; }
+
+VLIB_INIT_FUNCTION (ip4_gtpu_bypass_init);
+
+static uword
+ip6_gtpu_bypass (vlib_main_t * vm,
+		  vlib_node_runtime_t * node,
+		  vlib_frame_t * frame)
+{
+  return ip_gtpu_bypass_inline (vm, node, frame, /* is_ip4 */ 0);
+}
+
+VLIB_REGISTER_NODE (ip6_gtpu_bypass_node) = {
+  .function = ip6_gtpu_bypass,
+  .name = "ip6-gtpu-bypass",
+  .vector_size = sizeof (u32),
+
+  .n_next_nodes = IP_GTPU_BYPASS_N_NEXT,
+  .next_nodes = {
+    [IP_GTPU_BYPASS_NEXT_DROP] = "error-drop",
+    [IP_GTPU_BYPASS_NEXT_GTPU] = "gtpu6-input",
+  },
+
+  .format_buffer = format_ip6_header,
+  .format_trace = format_ip6_forward_next_trace,
+};
+
+VLIB_NODE_FUNCTION_MULTIARCH (ip6_gtpu_bypass_node,ip6_gtpu_bypass)
+
+/* Dummy init function to get us linked in. */
+clib_error_t * ip6_gtpu_bypass_init (vlib_main_t * vm)
+{ return 0; }
+
+VLIB_INIT_FUNCTION (ip6_gtpu_bypass_init);
diff --git a/src/plugins/gtpu/gtpu_encap.c b/src/plugins/gtpu/gtpu_encap.c
new file mode 100644
index 00000000..a2cfc646
--- /dev/null
+++ b/src/plugins/gtpu/gtpu_encap.c
@@ -0,0 +1,569 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <vppinfra/error.h>
+#include <vppinfra/hash.h>
+#include <vnet/vnet.h>
+#include <vnet/ip/ip.h>
+#include <vnet/ethernet/ethernet.h>
+#include <gtpu/gtpu.h>
+
+/* Statistics (not all errors) */
+#define foreach_gtpu_encap_error    \
+_(ENCAPSULATED, "good packets encapsulated")
+
+static char * gtpu_encap_error_strings[] = {
+#define _(sym,string) string,
+  foreach_gtpu_encap_error
+#undef _
+};
+
+typedef enum {
+#define _(sym,str) GTPU_ENCAP_ERROR_##sym,
+    foreach_gtpu_encap_error
+#undef _
+    GTPU_ENCAP_N_ERROR,
+} gtpu_encap_error_t;
+
+typedef enum {
+    GTPU_ENCAP_NEXT_DROP,
+    GTPU_ENCAP_N_NEXT,
+} gtpu_encap_next_t;
+
+typedef struct {
+  u32 tunnel_index;
+  u32 teid;
+} gtpu_encap_trace_t;
+
+u8 * format_gtpu_encap_trace (u8 * s, va_list * args)
+{
+  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
+  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
+  gtpu_encap_trace_t * t
+      = va_arg (*args, gtpu_encap_trace_t *);
+
+  s = format (s, "GTPU encap to gtpu_tunnel%d teid %d",
+	      t->tunnel_index, t->teid);
+  return s;
+}
+
+
+#define foreach_fixed_header4_offset            \
+    _(0) _(1) _(2) _(3)
+
+#define foreach_fixed_header6_offset            \
+    _(0) _(1) _(2) _(3) _(4) _(5) _(6)
+
+always_inline uword
+gtpu_encap_inline (vlib_main_t * vm,
+		    vlib_node_runtime_t * node,
+		    vlib_frame_t * from_frame,
+		    u32 is_ip4)
+{
+  u32 n_left_from, next_index, * from, * to_next;
+  gtpu_main_t * gtm = &gtpu_main;
+  vnet_main_t * vnm = gtm->vnet_main;
+  vnet_interface_main_t * im = &vnm->interface_main;
+  u32 pkts_encapsulated = 0;
+  u16 old_l0 = 0, old_l1 = 0;
+  u32 thread_index = vlib_get_thread_index();
+  u32 stats_sw_if_index, stats_n_packets, stats_n_bytes;
+  u32 sw_if_index0 = 0, sw_if_index1 = 0;
+  u32 next0 = 0, next1 = 0;
+  vnet_hw_interface_t * hi0, * hi1;
+  gtpu_tunnel_t * t0 = NULL, * t1 = NULL;
+
+  from = vlib_frame_vector_args (from_frame);
+  n_left_from = from_frame->n_vectors;
+
+  next_index = node->cached_next_index;
+  stats_sw_if_index = node->runtime_data[0];
+  stats_n_packets = stats_n_bytes = 0;
+
+  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;
+          u32 flow_hash0, flow_hash1;
+	  u32 len0, len1;
+          ip4_header_t * ip4_0, * ip4_1;
+          ip6_header_t * ip6_0, * ip6_1;
+          udp_header_t * udp0, * udp1;
+          gtpu_header_t * gtpu0, * gtpu1;
+          u64 * copy_src0, * copy_dst0;
+          u64 * copy_src1, * copy_dst1;
+          u32 * copy_src_last0, * copy_dst_last0;
+          u32 * copy_src_last1, * copy_dst_last1;
+          u16 new_l0, new_l1;
+          ip_csum_t sum0, sum1;
+
+	  /* 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);
+
+          flow_hash0 = vnet_l2_compute_flow_hash (b0);
+          flow_hash1 = vnet_l2_compute_flow_hash (b1);
+
+	  /* Get next node index and adj index from tunnel next_dpo */
+	  if (sw_if_index0 != vnet_buffer(b0)->sw_if_index[VLIB_TX])
+	    {
+	      sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX];
+	      hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0);
+	      t0 = &gtm->tunnels[hi0->dev_instance];
+	      /* Note: change to always set next0 if it may be set to drop */
+	      next0 = t0->next_dpo.dpoi_next_node;
+	    }
+          vnet_buffer(b0)->ip.adj_index[VLIB_TX] = t0->next_dpo.dpoi_index;
+
+	  /* Get next node index and adj index from tunnel next_dpo */
+	  if (sw_if_index1 != vnet_buffer(b1)->sw_if_index[VLIB_TX])
+	    {
+	      sw_if_index1 = vnet_buffer(b1)->sw_if_index[VLIB_TX];
+	      hi1 = vnet_get_sup_hw_interface (vnm, sw_if_index1);
+	      t1 = &gtm->tunnels[hi1->dev_instance];
+	      /* Note: change to always set next1 if it may be set to drop */
+	      next1 = t1->next_dpo.dpoi_next_node;
+	    }
+          vnet_buffer(b1)->ip.adj_index[VLIB_TX] = t1->next_dpo.dpoi_index;
+
+          /* Apply the rewrite string. $$$$ vnet_rewrite? */
+          vlib_buffer_advance (b0, -(word)_vec_len(t0->rewrite));
+          vlib_buffer_advance (b1, -(word)_vec_len(t1->rewrite));
+
+	  if (is_ip4)
+	    {
+	      ip4_0 = vlib_buffer_get_current(b0);
+	      ip4_1 = vlib_buffer_get_current(b1);
+
+	      /* Copy the fixed header */
+	      copy_dst0 = (u64 *) ip4_0;
+	      copy_src0 = (u64 *) t0->rewrite;
+	      copy_dst1 = (u64 *) ip4_1;
+	      copy_src1 = (u64 *) t1->rewrite;
+	      /* Copy first 32 octets 8-bytes at a time */
+#define _(offs) copy_dst0[offs] = copy_src0[offs];
+	      foreach_fixed_header4_offset;
+#undef _
+#define _(offs) copy_dst1[offs] = copy_src1[offs];
+	      foreach_fixed_header4_offset;
+#undef _
+	      /* Last 4 octets. Hopefully gcc will be our friend */
+              copy_dst_last0 = (u32 *)(&copy_dst0[4]);
+              copy_src_last0 = (u32 *)(&copy_src0[4]);
+              copy_dst_last0[0] = copy_src_last0[0];
+              copy_dst_last1 = (u32 *)(&copy_dst1[4]);
+              copy_src_last1 = (u32 *)(&copy_src1[4]);
+              copy_dst_last1[0] = copy_src_last1[0];
+
+	      /* Fix the IP4 checksum and length */
+	      sum0 = ip4_0->checksum;
+	      new_l0 = /* old_l0 always 0, see the rewrite setup */
+                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 */);
+	      ip4_0->checksum = ip_csum_fold (sum0);
+	      ip4_0->length = new_l0;
+	      sum1 = ip4_1->checksum;
+	      new_l1 = /* old_l1 always 0, see the rewrite setup */
+                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1));
+              sum1 = ip_csum_update (sum1, old_l1, new_l1, ip4_header_t,
+				     length /* changed member */);
+	      ip4_1->checksum = ip_csum_fold (sum1);
+	      ip4_1->length = new_l1;
+
+	      /* Fix UDP length and set source port */
+	      udp0 = (udp_header_t *)(ip4_0+1);
+	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
+					     - sizeof (*ip4_0));
+	      udp0->length = new_l0;
+	      udp0->src_port = flow_hash0;
+	      udp1 = (udp_header_t *)(ip4_1+1);
+	      new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1)
+					     - sizeof (*ip4_1));
+	      udp1->length = new_l1;
+	      udp1->src_port = flow_hash1;
+
+	      /* Fix GTPU length */
+	      gtpu0 = (gtpu_header_t *)(udp0+1);
+	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
+					     - sizeof (*ip4_0) - sizeof(*udp0));
+	      gtpu0->length = new_l0;
+	      gtpu1 = (gtpu_header_t *)(udp1+1);
+	      new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1)
+					     - sizeof (*ip4_1) - sizeof(*udp1));
+	      gtpu1->length = new_l1;
+	    }
+	  else /* ipv6 */
+	    {
+              int bogus = 0;
+
+	      ip6_0 = vlib_buffer_get_current(b0);
+	      ip6_1 = vlib_buffer_get_current(b1);
+
+	      /* Copy the fixed header */
+	      copy_dst0 = (u64 *) ip6_0;
+	      copy_src0 = (u64 *) t0->rewrite;
+	      copy_dst1 = (u64 *) ip6_1;
+	      copy_src1 = (u64 *) t1->rewrite;
+	      /* Copy first 56 (ip6) octets 8-bytes at a time */
+#define _(offs) copy_dst0[offs] = copy_src0[offs];
+	      foreach_fixed_header6_offset;
+#undef _
+#define _(offs) copy_dst1[offs] = copy_src1[offs];
+	      foreach_fixed_header6_offset;
+#undef _
+	      /* Fix IP6 payload length */
+	      new_l0 =
+                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
+				      - sizeof(*ip6_0));
+	      ip6_0->payload_length = new_l0;
+	      new_l1 =
+                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1)
+				      - sizeof(*ip6_1));
+	      ip6_1->payload_length = new_l1;
+
+	      /* Fix UDP length  and set source port */
+	      udp0 = (udp_header_t *)(ip6_0+1);
+	      udp0->length = new_l0;
+	      udp0->src_port = flow_hash0;
+	      udp1 = (udp_header_t *)(ip6_1+1);
+	      udp1->length = new_l1;
+	      udp1->src_port = flow_hash1;
+
+	      /* IPv6 UDP checksum is mandatory */
+	      udp0->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b0,
+								 ip6_0, &bogus);
+	      if (udp0->checksum == 0)
+		udp0->checksum = 0xffff;
+	      udp1->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b1,
+                                                        ip6_1, &bogus);
+	      if (udp1->checksum == 0)
+		udp1->checksum = 0xffff;
+
+	      /* Fix GTPU length */
+	      gtpu0 = (gtpu_header_t *)(udp0+1);
+	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
+					     - sizeof (*ip4_0) - sizeof(*udp0));
+	      gtpu0->length = new_l0;
+	      gtpu1 = (gtpu_header_t *)(udp1+1);
+	      new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1)
+					     - sizeof (*ip4_1) - sizeof(*udp1));
+	      gtpu1->length = new_l1;
+	    }
+
+          pkts_encapsulated += 2;
+ 	  len0 = vlib_buffer_length_in_chain (vm, b0);
+ 	  len1 = vlib_buffer_length_in_chain (vm, b1);
+	  stats_n_packets += 2;
+	  stats_n_bytes += len0 + len1;
+
+	  /* Batch stats increment on the same gtpu tunnel so counter is not
+	     incremented per packet. Note stats are still incremented for deleted
+	     and admin-down tunnel where packets are dropped. It is not worthwhile
+	     to check for this rare case and affect normal path performance. */
+	  if (PREDICT_FALSE ((sw_if_index0 != stats_sw_if_index) ||
+			     (sw_if_index1 != stats_sw_if_index)))
+	    {
+	      stats_n_packets -= 2;
+	      stats_n_bytes -= len0 + len1;
+	      if (sw_if_index0 == sw_if_index1)
+	        {
+		  if (stats_n_packets)
+		    vlib_increment_combined_counter
+		      (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
+		       thread_index, stats_sw_if_index,
+		       stats_n_packets, stats_n_bytes);
+		  stats_sw_if_index = sw_if_index0;
+		  stats_n_packets = 2;
+		  stats_n_bytes = len0 + len1;
+	        }
+	      else
+	        {
+		  vlib_increment_combined_counter
+		      (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
+		       thread_index, sw_if_index0, 1, len0);
+		  vlib_increment_combined_counter
+		      (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
+		       thread_index, sw_if_index1, 1, len1);
+		}
+	    }
+
+	  if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
+            {
+              gtpu_encap_trace_t *tr =
+                vlib_add_trace (vm, node, b0, sizeof (*tr));
+              tr->tunnel_index = t0 - gtm->tunnels;
+              tr->teid = t0->teid;
+           }
+
+          if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED))
+            {
+              gtpu_encap_trace_t *tr =
+                vlib_add_trace (vm, node, b1, sizeof (*tr));
+              tr->tunnel_index = t1 - gtm->tunnels;
+              tr->teid = t1->teid;
+            }
+
+	  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)
+	{
+	  u32 bi0;
+	  vlib_buffer_t * b0;
+          u32 flow_hash0;
+	  u32 len0;
+          ip4_header_t * ip4_0;
+          ip6_header_t * ip6_0;
+          udp_header_t * udp0;
+          gtpu_header_t * gtpu0;
+          u64 * copy_src0, * copy_dst0;
+          u32 * copy_src_last0, * copy_dst_last0;
+          u16 new_l0;
+          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);
+
+          flow_hash0 = vnet_l2_compute_flow_hash(b0);
+
+	  /* Get next node index and adj index from tunnel next_dpo */
+	  if (sw_if_index0 != vnet_buffer(b0)->sw_if_index[VLIB_TX])
+	    {
+	      sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX];
+	      hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0);
+	      t0 = &gtm->tunnels[hi0->dev_instance];
+	      /* Note: change to always set next0 if it may be set to drop */
+	      next0 = t0->next_dpo.dpoi_next_node;
+	    }
+	  vnet_buffer(b0)->ip.adj_index[VLIB_TX] = t0->next_dpo.dpoi_index;
+
+          /* Apply the rewrite string. $$$$ vnet_rewrite? */
+          vlib_buffer_advance (b0, -(word)_vec_len(t0->rewrite));
+
+	  if (is_ip4)
+	    {
+	      ip4_0 = vlib_buffer_get_current(b0);
+
+	      /* Copy the fixed header */
+	      copy_dst0 = (u64 *) ip4_0;
+	      copy_src0 = (u64 *) t0->rewrite;
+	      /* Copy first 32 octets 8-bytes at a time */
+#define _(offs) copy_dst0[offs] = copy_src0[offs];
+	      foreach_fixed_header4_offset;
+#undef _
+	      /* Last 4 octets. Hopefully gcc will be our friend */
+              copy_dst_last0 = (u32 *)(&copy_dst0[4]);
+              copy_src_last0 = (u32 *)(&copy_src0[4]);
+              copy_dst_last0[0] = copy_src_last0[0];
+
+	      /* Fix the IP4 checksum and length */
+	      sum0 = ip4_0->checksum;
+	      new_l0 = /* old_l0 always 0, see the rewrite setup */
+                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 */);
+	      ip4_0->checksum = ip_csum_fold (sum0);
+	      ip4_0->length = new_l0;
+
+	      /* Fix UDP length and set source port */
+	      udp0 = (udp_header_t *)(ip4_0+1);
+	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
+					     - sizeof (*ip4_0));
+	      udp0->length = new_l0;
+	      udp0->src_port = flow_hash0;
+
+	      /* Fix GTPU length */
+	      gtpu0 = (gtpu_header_t *)(udp0+1);
+	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
+					     - sizeof (*ip4_0) - sizeof(*udp0));
+	      gtpu0->length = new_l0;
+	    }
+
+	  else /* ip6 path */
+	    {
+              int bogus = 0;
+
+	      ip6_0 = vlib_buffer_get_current(b0);
+	      /* Copy the fixed header */
+	      copy_dst0 = (u64 *) ip6_0;
+	      copy_src0 = (u64 *) t0->rewrite;
+	      /* Copy first 56 (ip6) octets 8-bytes at a time */
+#define _(offs) copy_dst0[offs] = copy_src0[offs];
+	      foreach_fixed_header6_offset;
+#undef _
+	      /* Fix IP6 payload length */
+	      new_l0 =
+                clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
+				      - sizeof(*ip6_0));
+	      ip6_0->payload_length = new_l0;
+
+	      /* Fix UDP length  and set source port */
+	      udp0 = (udp_header_t *)(ip6_0+1);
+	      udp0->length = new_l0;
+	      udp0->src_port = flow_hash0;
+
+	      /* IPv6 UDP checksum is mandatory */
+	      udp0->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b0,
+								 ip6_0, &bogus);
+	      if (udp0->checksum == 0)
+		udp0->checksum = 0xffff;
+
+	      /* Fix GTPU length */
+	      gtpu0 = (gtpu_header_t *)(udp0+1);
+	      new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
+					     - sizeof (*ip4_0) - sizeof(*udp0));
+	      gtpu0->length = new_l0;
+	    }
+
+          pkts_encapsulated ++;
+	  len0 = vlib_buffer_length_in_chain (vm, b0);
+	  stats_n_packets += 1;
+	  stats_n_bytes += len0;
+
+	  /* Batch stats increment on the same gtpu tunnel so counter is not
+	     incremented per packet. Note stats are still incremented for deleted
+	     and admin-down tunnel where packets are dropped. It is not worthwhile
+	     to check for this rare case and affect normal path performance. */
+	  if (PREDICT_FALSE (sw_if_index0 != stats_sw_if_index))
+	    {
+	      stats_n_packets -= 1;
+	      stats_n_bytes -= len0;
+	      if (stats_n_packets)
+		vlib_increment_combined_counter
+		  (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
+		   thread_index, stats_sw_if_index,
+		   stats_n_packets, stats_n_bytes);
+	      stats_n_packets = 1;
+	      stats_n_bytes = len0;
+	      stats_sw_if_index = sw_if_index0;
+	    }
+
+          if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
+            {
+              gtpu_encap_trace_t *tr =
+                vlib_add_trace (vm, node, b0, sizeof (*tr));
+              tr->tunnel_index = t0 - gtm->tunnels;
+              tr->teid = t0->teid;
+            }
+	  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);
+    }
+
+  /* Do we still need this now that tunnel tx stats is kept? */
+  vlib_node_increment_counter (vm, node->node_index,
+                               GTPU_ENCAP_ERROR_ENCAPSULATED,
+                               pkts_encapsulated);
+
+  /* Increment any remaining batch stats */
+  if (stats_n_packets)
+    {
+      vlib_increment_combined_counter
+	(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
+	 thread_index, stats_sw_if_index, stats_n_packets, stats_n_bytes);
+      node->runtime_data[0] = stats_sw_if_index;
+    }
+
+  return from_frame->n_vectors;
+}
+
+static uword
+gtpu4_encap (vlib_main_t * vm,
+	      vlib_node_runtime_t * node,
+	      vlib_frame_t * from_frame)
+{
+  return gtpu_encap_inline (vm, node, from_frame, /* is_ip4 */ 1);
+}
+
+static uword
+gtpu6_encap (vlib_main_t * vm,
+	      vlib_node_runtime_t * node,
+	      vlib_frame_t * from_frame)
+{
+  return gtpu_encap_inline (vm, node, from_frame, /* is_ip4 */ 0);
+}
+
+VLIB_REGISTER_NODE (gtpu4_encap_node) = {
+  .function = gtpu4_encap,
+  .name = "gtpu4-encap",
+  .vector_size = sizeof (u32),
+  .format_trace = format_gtpu_encap_trace,
+  .type = VLIB_NODE_TYPE_INTERNAL,
+  .n_errors = ARRAY_LEN(gtpu_encap_error_strings),
+  .error_strings = gtpu_encap_error_strings,
+  .n_next_nodes = GTPU_ENCAP_N_NEXT,
+  .next_nodes = {
+        [GTPU_ENCAP_NEXT_DROP] = "error-drop",
+  },
+};
+
+VLIB_NODE_FUNCTION_MULTIARCH (gtpu4_encap_node, gtpu4_encap)
+
+VLIB_REGISTER_NODE (gtpu6_encap_node) = {
+  .function = gtpu6_encap,
+  .name = "gtpu6-encap",
+  .vector_size = sizeof (u32),
+  .format_trace = format_gtpu_encap_trace,
+  .type = VLIB_NODE_TYPE_INTERNAL,
+  .n_errors = ARRAY_LEN(gtpu_encap_error_strings),
+  .error_strings = gtpu_encap_error_strings,
+  .n_next_nodes = GTPU_ENCAP_N_NEXT,
+  .next_nodes = {
+        [GTPU_ENCAP_NEXT_DROP] = "error-drop",
+  },
+};
+
+VLIB_NODE_FUNCTION_MULTIARCH (gtpu6_encap_node, gtpu6_encap)
+
diff --git a/src/plugins/gtpu/gtpu_error.def b/src/plugins/gtpu/gtpu_error.def
new file mode 100644
index 00000000..a55b2026
--- /dev/null
+++ b/src/plugins/gtpu/gtpu_error.def
@@ -0,0 +1,18 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+gtpu_error (DECAPSULATED, "good packets decapsulated")
+gtpu_error (NO_SUCH_TUNNEL, "no such tunnel packets")
+gtpu_error (BAD_VER, "packets with bad version in gtpu header")
+gtpu_error (BAD_FLAGS, "packets with bad flags field in gtpu header")
diff --git a/src/plugins/gtpu/gtpu_msg_enum.h b/src/plugins/gtpu/gtpu_msg_enum.h
new file mode 100644
index 00000000..6a2ea448
--- /dev/null
+++ b/src/plugins/gtpu/gtpu_msg_enum.h
@@ -0,0 +1,31 @@
+/*
+ * gtpu_msg_enum.h - vpp engine plug-in message enumeration
+ *
+ * Copyright (c) <current-year> <your-organization>
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef included_gtpu_msg_enum_h
+#define included_gtpu_msg_enum_h
+
+#include <vppinfra/byte_order.h>
+
+#define vl_msg_id(n,h) n,
+typedef enum
+{
+#include <gtpu/gtpu_all_api_h.h>
+  /* We'll want to know how many messages IDs we need... */
+  VL_MSG_FIRST_AVAILABLE,
+} vl_msg_id_t;
+#undef vl_msg_id
+
+#endif /* included_gtpu_msg_enum_h */
diff --git a/src/plugins/gtpu/gtpu_test.c b/src/plugins/gtpu/gtpu_test.c
new file mode 100644
index 00000000..91c5c616
--- /dev/null
+++ b/src/plugins/gtpu/gtpu_test.c
@@ -0,0 +1,498 @@
+/*
+ * Copyright (c) 2016 Intel and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <vat/vat.h>
+#include <vlibapi/api.h>
+#include <vlibmemory/api.h>
+#include <vlibsocket/api.h>
+#include <vppinfra/error.h>
+#include <gtpu/gtpu.h>
+
+#define __plugin_msg_base gtpu_test_main.msg_id_base
+#include <vlibapi/vat_helper_macros.h>
+
+
+uword unformat_ip46_address (unformat_input_t * input, va_list * args)
+{
+  ip46_address_t *ip46 = va_arg (*args, ip46_address_t *);
+  ip46_type_t type = va_arg (*args, ip46_type_t);
+  if ((type != IP46_TYPE_IP6) &&
+      unformat(input, "%U", unformat_ip4_address, &ip46->ip4)) {
+    ip46_address_mask_ip4(ip46);
+    return 1;
+  } else if ((type != IP46_TYPE_IP4) &&
+      unformat(input, "%U", unformat_ip6_address, &ip46->ip6)) {
+    return 1;
+  }
+  return 0;
+}
+uword unformat_ip46_prefix (unformat_input_t * input, va_list * args)
+{
+  ip46_address_t *ip46 = va_arg (*args, ip46_address_t *);
+  u8 *len = va_arg (*args, u8 *);
+  ip46_type_t type = va_arg (*args, ip46_type_t);
+
+  u32 l;
+  if ((type != IP46_TYPE_IP6) && unformat(input, "%U/%u", unformat_ip4_address, &ip46->ip4, &l)) {
+    if (l > 32)
+      return 0;
+    *len = l + 96;
+    ip46->pad[0] = ip46->pad[1] = ip46->pad[2] = 0;
+  } else if ((type != IP46_TYPE_IP4) && unformat(input, "%U/%u", unformat_ip6_address, &ip46->ip6, &l)) {
+    if (l > 128)
+      return 0;
+    *len = l;
+  } else {
+    return 0;
+  }
+  return 1;
+}
+/////////////////////////
+
+#define vl_msg_id(n,h) n,
+typedef enum {
+#include <gtpu/gtpu.api.h>
+    /* We'll want to know how many messages IDs we need... */
+    VL_MSG_FIRST_AVAILABLE,
+} vl_msg_id_t;
+#undef vl_msg_id
+
+/* define message structures */
+#define vl_typedefs
+#include <gtpu/gtpu.api.h>
+#undef vl_typedefs
+
+/* declare message handlers for each api */
+
+#define vl_endianfun             /* define message structures */
+#include <gtpu/gtpu.api.h>
+#undef vl_endianfun
+
+/* instantiate all the print functions we know about */
+#define vl_print(handle, ...)
+#define vl_printfun
+#include <gtpu/gtpu.api.h>
+#undef vl_printfun
+
+/* Get the API version number. */
+#define vl_api_version(n,v) static u32 api_version=(v);
+#include <gtpu/gtpu.api.h>
+#undef vl_api_version
+
+typedef struct {
+    /* API message ID base */
+    u16 msg_id_base;
+    vat_main_t *vat_main;
+} gtpu_test_main_t;
+
+gtpu_test_main_t gtpu_test_main;
+
+static void vl_api_gtpu_add_del_tunnel_reply_t_handler
+  (vl_api_gtpu_add_del_tunnel_reply_t * mp)
+{
+  vat_main_t *vam = &vat_main;
+  i32 retval = ntohl (mp->retval);
+  if (vam->async_mode)
+    {
+      vam->async_errors += (retval < 0);
+    }
+  else
+    {
+      vam->retval = retval;
+      vam->sw_if_index = ntohl (mp->sw_if_index);
+      vam->result_ready = 1;
+    }
+}
+
+
+#define foreach_standard_reply_retval_handler   \
+    _(sw_interface_set_gtpu_bypass_reply)
+
+#define _(n)                                            \
+    static void vl_api_##n##_t_handler                  \
+    (vl_api_##n##_t * mp)                               \
+    {                                                   \
+        vat_main_t * vam = gtpu_test_main.vat_main;   \
+        i32 retval = ntohl(mp->retval);                 \
+        if (vam->async_mode) {                          \
+            vam->async_errors += (retval < 0);          \
+        } else {                                        \
+            vam->retval = retval;                       \
+            vam->result_ready = 1;                      \
+        }                                               \
+    }
+  foreach_standard_reply_retval_handler;
+#undef _
+
+/*
+ * Table of message reply handlers, must include boilerplate handlers
+ * we just generated
+ */
+#define foreach_vpe_api_reply_msg                               \
+  _(SW_INTERFACE_SET_GTPU_BYPASS_REPLY, sw_interface_set_gtpu_bypass_reply) \
+  _(GTPU_ADD_DEL_TUNNEL_REPLY, gtpu_add_del_tunnel_reply)               \
+  _(GTPU_TUNNEL_DETAILS, gtpu_tunnel_details)
+
+
+static uword
+api_unformat_sw_if_index (unformat_input_t * input, va_list * args)
+{
+  vat_main_t *vam = va_arg (*args, vat_main_t *);
+  u32 *result = va_arg (*args, u32 *);
+  u8 *if_name;
+  uword *p;
+
+  if (!unformat (input, "%s", &if_name))
+    return 0;
+
+  p = hash_get_mem (vam->sw_if_index_by_interface_name, if_name);
+  if (p == 0)
+    return 0;
+  *result = p[0];
+  return 1;
+}
+
+static int
+api_sw_interface_set_gtpu_bypass (vat_main_t * vam)
+{
+  unformat_input_t *i = vam->input;
+  vl_api_sw_interface_set_gtpu_bypass_t *mp;
+  u32 sw_if_index = 0;
+  u8 sw_if_index_set = 0;
+  u8 is_enable = 1;
+  u8 is_ipv6 = 0;
+  int ret;
+
+  /* Parse args required to build the message */
+  while (unformat_check_input (i) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (i, "%U", api_unformat_sw_if_index, vam, &sw_if_index))
+      sw_if_index_set = 1;
+      else if (unformat (i, "sw_if_index %d", &sw_if_index))
+      sw_if_index_set = 1;
+      else if (unformat (i, "enable"))
+      is_enable = 1;
+      else if (unformat (i, "disable"))
+      is_enable = 0;
+      else if (unformat (i, "ip4"))
+      is_ipv6 = 0;
+      else if (unformat (i, "ip6"))
+      is_ipv6 = 1;
+      else
+      break;
+    }
+
+  if (sw_if_index_set == 0)
+    {
+      errmsg ("missing interface name or sw_if_index");
+      return -99;
+    }
+
+  /* Construct the API message */
+  M (SW_INTERFACE_SET_GTPU_BYPASS, mp);
+
+  mp->sw_if_index = ntohl (sw_if_index);
+  mp->enable = is_enable;
+  mp->is_ipv6 = is_ipv6;
+
+  /* send it... */
+  S (mp);
+
+  /* Wait for a reply... */
+  W (ret);
+  return ret;
+}
+
+static uword unformat_gtpu_decap_next
+  (unformat_input_t * input, va_list * args)
+{
+  u32 *result = va_arg (*args, u32 *);
+  u32 tmp;
+
+  if (unformat (input, "l2"))
+    *result = GTPU_INPUT_NEXT_L2_INPUT;
+  else if (unformat (input, "%d", &tmp))
+    *result = tmp;
+  else
+    return 0;
+  return 1;
+}
+
+static int
+api_gtpu_add_del_tunnel (vat_main_t * vam)
+{
+  unformat_input_t *line_input = vam->input;
+  vl_api_gtpu_add_del_tunnel_t *mp;
+  ip46_address_t src, dst;
+  u8 is_add = 1;
+  u8 ipv4_set = 0, ipv6_set = 0;
+  u8 src_set = 0;
+  u8 dst_set = 0;
+  u8 grp_set = 0;
+  u32 mcast_sw_if_index = ~0;
+  u32 encap_vrf_id = 0;
+  u32 decap_next_index = ~0;
+  u32 teid = 0;
+  int ret;
+
+  /* Can't "universally zero init" (={0}) due to GCC bug 53119 */
+  memset (&src, 0, sizeof src);
+  memset (&dst, 0, sizeof dst);
+
+  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (line_input, "del"))
+      is_add = 0;
+      else
+      if (unformat (line_input, "src %U", unformat_ip4_address, &src.ip4))
+      {
+	ipv4_set = 1;
+	src_set = 1;
+      }
+      else
+      if (unformat (line_input, "dst %U", unformat_ip4_address, &dst.ip4))
+      {
+	ipv4_set = 1;
+	dst_set = 1;
+      }
+      else
+      if (unformat (line_input, "src %U", unformat_ip6_address, &src.ip6))
+      {
+	ipv6_set = 1;
+	src_set = 1;
+      }
+      else
+      if (unformat (line_input, "dst %U", unformat_ip6_address, &dst.ip6))
+      {
+	ipv6_set = 1;
+	dst_set = 1;
+      }
+      else if (unformat (line_input, "group %U %U",
+		       unformat_ip4_address, &dst.ip4,
+		       api_unformat_sw_if_index, vam, &mcast_sw_if_index))
+      {
+	grp_set = dst_set = 1;
+	ipv4_set = 1;
+      }
+      else if (unformat (line_input, "group %U",
+		       unformat_ip4_address, &dst.ip4))
+      {
+	grp_set = dst_set = 1;
+	ipv4_set = 1;
+      }
+      else if (unformat (line_input, "group %U %U",
+		       unformat_ip6_address, &dst.ip6,
+		       api_unformat_sw_if_index, vam, &mcast_sw_if_index))
+      {
+	grp_set = dst_set = 1;
+	ipv6_set = 1;
+      }
+      else if (unformat (line_input, "group %U",
+		       unformat_ip6_address, &dst.ip6))
+      {
+	grp_set = dst_set = 1;
+	ipv6_set = 1;
+      }
+      else
+      if (unformat (line_input, "mcast_sw_if_index %u", &mcast_sw_if_index))
+      ;
+      else if (unformat (line_input, "encap-vrf-id %d", &encap_vrf_id))
+      ;
+      else if (unformat (line_input, "decap-next %U",
+		       unformat_gtpu_decap_next, &decap_next_index))
+      ;
+      else if (unformat (line_input, "teid %d", &teid))
+      ;
+      else
+      {
+	errmsg ("parse error '%U'", format_unformat_error, line_input);
+	return -99;
+      }
+    }
+
+  if (src_set == 0)
+    {
+      errmsg ("tunnel src address not specified");
+      return -99;
+    }
+  if (dst_set == 0)
+    {
+      errmsg ("tunnel dst address not specified");
+      return -99;
+    }
+
+  if (grp_set && !ip46_address_is_multicast (&dst))
+    {
+      errmsg ("tunnel group address not multicast");
+      return -99;
+    }
+  if (grp_set && mcast_sw_if_index == ~0)
+    {
+      errmsg ("tunnel nonexistent multicast device");
+      return -99;
+    }
+  if (grp_set == 0 && ip46_address_is_multicast (&dst))
+    {
+      errmsg ("tunnel dst address must be unicast");
+      return -99;
+    }
+
+
+  if (ipv4_set && ipv6_set)
+    {
+      errmsg ("both IPv4 and IPv6 addresses specified");
+      return -99;
+    }
+
+  M (GTPU_ADD_DEL_TUNNEL, mp);
+
+  if (ipv6_set)
+    {
+      clib_memcpy (mp->src_address, &src.ip6, sizeof (src.ip6));
+      clib_memcpy (mp->dst_address, &dst.ip6, sizeof (dst.ip6));
+    }
+  else
+    {
+      clib_memcpy (mp->src_address, &src.ip4, sizeof (src.ip4));
+      clib_memcpy (mp->dst_address, &dst.ip4, sizeof (dst.ip4));
+    }
+  mp->encap_vrf_id = ntohl (encap_vrf_id);
+  mp->decap_next_index = ntohl (decap_next_index);
+  mp->mcast_sw_if_index = ntohl (mcast_sw_if_index);
+  mp->teid = ntohl (teid);
+  mp->is_add = is_add;
+  mp->is_ipv6 = ipv6_set;
+
+  S (mp);
+  W (ret);
+  return ret;
+}
+
+static void vl_api_gtpu_tunnel_details_t_handler
+  (vl_api_gtpu_tunnel_details_t * mp)
+{
+  vat_main_t *vam = &vat_main;
+  ip46_address_t src = to_ip46 (mp->is_ipv6, mp->dst_address);
+  ip46_address_t dst = to_ip46 (mp->is_ipv6, mp->src_address);
+
+  print (vam->ofp, "%11d%24U%24U%14d%18d%13d%19d",
+       ntohl (mp->sw_if_index),
+       format_ip46_address, &src, IP46_TYPE_ANY,
+       format_ip46_address, &dst, IP46_TYPE_ANY,
+       ntohl (mp->encap_vrf_id),
+       ntohl (mp->decap_next_index), ntohl (mp->teid),
+       ntohl (mp->mcast_sw_if_index));
+}
+
+static int
+api_gtpu_tunnel_dump (vat_main_t * vam)
+{
+  unformat_input_t *i = vam->input;
+  vl_api_gtpu_tunnel_dump_t *mp;
+  u32 sw_if_index;
+  u8 sw_if_index_set = 0;
+  int ret;
+
+  /* Parse args required to build the message */
+  while (unformat_check_input (i) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (i, "sw_if_index %d", &sw_if_index))
+      sw_if_index_set = 1;
+      else
+      break;
+    }
+
+  if (sw_if_index_set == 0)
+    {
+      sw_if_index = ~0;
+    }
+
+  if (!vam->json_output)
+    {
+      print (vam->ofp, "%11s%24s%24s%14s%18s%13s%19s",
+	   "sw_if_index", "src_address", "dst_address",
+	   "encap_vrf_id", "decap_next_index", "teid", "mcast_sw_if_index");
+    }
+
+  /* Get list of gtpu-tunnel interfaces */
+  M (GTPU_TUNNEL_DUMP, mp);
+
+  mp->sw_if_index = htonl (sw_if_index);
+
+  S (mp);
+
+  W (ret);
+  return ret;
+}
+
+/*
+ * List of messages that the api test plugin sends,
+ * and that the data plane plugin processes
+ */
+#define foreach_vpe_api_msg                                            \
+_(sw_interface_set_gtpu_bypass,                                        \
+      "<intfc> | sw_if_index <id> [ip4 | ip6] [enable | disable]")     \
+_(gtpu_add_del_tunnel,                                                 \
+        "src <ip-addr> { dst <ip-addr> | group <mcast-ip-addr>\n"      \
+        "{ <intfc> | mcast_sw_if_index <nn> } }\n"                     \
+        "teid <teid> [encap-vrf-id <nn>] [decap-next <l2|nn>] [del]")  \
+_(gtpu_tunnel_dump, "[<intfc> | sw_if_index <nn>]")                    \
+
+static void
+gtpu_vat_api_hookup (vat_main_t *vam)
+{
+  gtpu_test_main_t * gtm = &gtpu_test_main;
+  /* Hook up handlers for replies from the data plane plug-in */
+#define _(N,n)                                                  \
+  vl_msg_api_set_handlers((VL_API_##N + gtm->msg_id_base),       \
+                          #n,                                   \
+                          vl_api_##n##_t_handler,               \
+                          vl_noop_handler,                      \
+                          vl_api_##n##_t_endian,                \
+                          vl_api_##n##_t_print,                 \
+                          sizeof(vl_api_##n##_t), 1);
+  foreach_vpe_api_reply_msg;
+#undef _
+
+  /* API messages we can send */
+#define _(n,h) hash_set_mem (vam->function_by_name, #n, api_##n);
+  foreach_vpe_api_msg;
+#undef _
+
+  /* Help strings */
+#define _(n,h) hash_set_mem (vam->help_by_name, #n, h);
+  foreach_vpe_api_msg;
+#undef _
+}
+
+clib_error_t * vat_plugin_register (vat_main_t *vam)
+{
+  gtpu_test_main_t * gtm = &gtpu_test_main;
+
+  u8 * name;
+
+  gtm->vat_main = vam;
+
+  /* Ask the vpp engine for the first assigned message-id */
+  name = format (0, "gtpu_%08x%c", api_version, 0);
+  gtm->msg_id_base = vl_client_get_first_plugin_msg_id ((char *) name);
+
+  if (gtm->msg_id_base != (u16) ~0)
+    gtpu_vat_api_hookup (vam);
+
+  vec_free(name);
+
+  return 0;
+}
diff --git a/src/vnet/mfib/mfib_types.h b/src/vnet/mfib/mfib_types.h
index fe53aa68..e278af7f 100644
--- a/src/vnet/mfib/mfib_types.h
+++ b/src/vnet/mfib/mfib_types.h
@@ -167,6 +167,7 @@ typedef enum mfib_source_t_
     MFIB_SOURCE_DHCP,
     MFIB_SOURCE_SRv6,
     MFIB_SOURCE_DEFAULT_ROUTE,
+    MFIB_SOURCE_GTPU,
 } mfib_source_t;
 
 #define MFIB_SOURCE_NAMES {                        \
@@ -177,6 +178,7 @@ typedef enum mfib_source_t_
     [MFIB_SOURCE_VXLAN] = "VXLAN",                 \
     [MFIB_SOURCE_SRv6] = "SRv6",                   \
     [MFIB_SOURCE_DEFAULT_ROUTE] = "Default Route", \
+    [MFIB_SOURCE_GTPU] = "GTPU",                   \
 }
 
 /**
diff --git a/src/vnet/udp/udp.h b/src/vnet/udp/udp.h
index bd163e27..5b63f44d 100644
--- a/src/vnet/udp/udp.h
+++ b/src/vnet/udp/udp.h
@@ -81,6 +81,7 @@ typedef enum
 _ (67, dhcp_to_server)                          \
 _ (68, dhcp_to_client)                          \
 _ (500, ikev2)                                  \
+_ (2152, GTPU)					\
 _ (3784, bfd4)                                  \
 _ (3785, bfd_echo4)                             \
 _ (4341, lisp_gpe)                              \
@@ -95,6 +96,7 @@ _ (6633, vpath_3)
 #define foreach_udp6_dst_port                   \
 _ (547, dhcpv6_to_server)                       \
 _ (546, dhcpv6_to_client)			\
+_ (2152, GTPU6)					\
 _ (3784, bfd6)                                  \
 _ (3785, bfd_echo6)                             \
 _ (4341, lisp_gpe6)                             \
diff --git a/src/vpp-api/java/Makefile.am b/src/vpp-api/java/Makefile.am
index f18e0c24..4f1f7ed2 100644
--- a/src/vpp-api/java/Makefile.am
+++ b/src/vpp-api/java/Makefile.am
@@ -129,6 +129,26 @@ jvpp-acl/io_fd_vpp_jvpp_acl_JVppAclImpl.h: $(jvpp_registry_ok) $(jvpp_acl_json_f
 endif
 
 #
+# GTPU Plugin
+#
+if ENABLE_GTPU_PLUGIN
+noinst_LTLIBRARIES += libjvpp_gtpu.la
+libjvpp_gtpu_la_SOURCES = jvpp-gtpu/jvpp_gtpu.c
+libjvpp_gtpu_la_CPPFLAGS = -Ijvpp-gtpu
+libjvpp_gtpu_la_LIBADD = $(JVPP_LIBS)
+libjvpp_gtpu_la_DEPENDENCIES = libjvpp_common.la
+
+BUILT_SOURCES += jvpp-gtpu/io_fd_vpp_jvpp_gtpu_JVppGtpuImpl.h
+JAR_FILES += jvpp-gtpu-$(PACKAGE_VERSION).jar
+CLEANDIRS += jvpp-gtpu/target
+
+jvpp_gtpu_json_files = @top_builddir@/plugins/gtpu/gtpu.api.json
+
+jvpp-gtpu/io_fd_vpp_jvpp_gtpu_JVppGtpuImpl.h: $(jvpp_registry_ok) $(jvpp_gtpu_json_files)
+	$(call japigen,gtpu,JVppGtpuImpl)
+endif
+
+#
 # SNAT Plugin
 #
 if ENABLE_SNAT_PLUGIN
diff --git a/src/vpp-api/java/jvpp-gtpu/jvpp_gtpu.c b/src/vpp-api/java/jvpp-gtpu/jvpp_gtpu.c
new file mode 100644
index 00000000..12b3090e
--- /dev/null
+++ b/src/vpp-api/java/jvpp-gtpu/jvpp_gtpu.c
@@ -0,0 +1,107 @@
+/*
+ * Copyright (c) 2016 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <vnet/vnet.h>
+
+#include <gtpu/gtpu_msg_enum.h>
+#define vl_typedefs             /* define message structures */
+#include <gtpu/gtpu_all_api_h.h>
+#undef vl_typedefs
+
+#include <vnet/api_errno.h>
+#include <vlibapi/api.h>
+#include <vlibmemory/api.h>
+
+#if VPPJNI_DEBUG == 1
+  #define DEBUG_LOG(...) clib_warning(__VA_ARGS__)
+#else
+  #define DEBUG_LOG(...)
+#endif
+
+#include <jvpp-common/jvpp_common.h>
+
+#include "jvpp-gtpu/io_fd_vpp_jvpp_gtpu_JVppGtpuImpl.h"
+#include "jvpp_gtpu.h"
+#include "jvpp-gtpu/jvpp_gtpu_gen.h"
+
+/*
+ * Class:     io_fd_vpp_jvpp_gtpu_JVppgtpuImpl
+ * Method:    init0
+ * Signature: (JI)V
+ */
+JNIEXPORT void JNICALL Java_io_fd_vpp_jvpp_gtpu_JVppGtpuImpl_init0
+  (JNIEnv *env, jclass clazz, jobject callback, jlong queue_address, jint my_client_index) {
+  gtpu_main_t * plugin_main = &gtpu_main;
+  clib_warning ("Java_io_fd_vpp_jvpp_gtpu_JVppGtpuImpl_init0");
+
+  plugin_main->my_client_index = my_client_index;
+  plugin_main->vl_input_queue = (unix_shared_memory_queue_t *)queue_address;
+
+  plugin_main->callbackObject = (*env)->NewGlobalRef(env, callback);
+  plugin_main->callbackClass = (jclass)(*env)->NewGlobalRef(env, (*env)->GetObjectClass(env, callback));
+
+  // verify API has not changed since jar generation
+  #define _(N)             \
+      get_message_id(env, #N);  \
+      foreach_supported_api_message;
+  #undef _
+
+  #define _(N,n)                                  \
+      vl_msg_api_set_handlers(get_message_id(env, #N), #n,     \
+              vl_api_##n##_t_handler,             \
+              vl_noop_handler,                    \
+              vl_noop_handler,                    \
+              vl_noop_handler,                    \
+              sizeof(vl_api_##n##_t), 1);
+      foreach_api_reply_handler;
+  #undef _
+}
+
+JNIEXPORT void JNICALL Java_io_fd_vpp_jvpp_gtpu_JVppGtpuImpl_close0
+(JNIEnv *env, jclass clazz) {
+  gtpu_main_t * plugin_main = &gtpu_main;
+
+    // cleanup:
+    (*env)->DeleteGlobalRef(env, plugin_main->callbackClass);
+    (*env)->DeleteGlobalRef(env, plugin_main->callbackObject);
+
+    plugin_main->callbackClass = NULL;
+    plugin_main->callbackObject = NULL;
+}
+
+/* Attach thread to JVM and cache class references when initiating JVPP ACL */
+jint JNI_OnLoad(JavaVM *vm, void *reserved) {
+    JNIEnv* env;
+
+    if ((*vm)->GetEnv(vm, (void**) &env, JNI_VERSION_1_8) != JNI_OK) {
+        return JNI_EVERSION;
+    }
+
+    if (cache_class_references(env) != 0) {
+        clib_warning ("Failed to cache class references\n");
+        return JNI_ERR;
+    }
+
+    return JNI_VERSION_1_8;
+}
+
+/* Clean up cached references when disposing JVPP ACL */
+void JNI_OnUnload(JavaVM *vm, void *reserved) {
+    JNIEnv* env;
+    if ((*vm)->GetEnv(vm, (void**) &env, JNI_VERSION_1_8) != JNI_OK) {
+        return;
+    }
+    delete_class_references(env);
+}
diff --git a/src/vpp-api/java/jvpp-gtpu/jvpp_gtpu.h b/src/vpp-api/java/jvpp-gtpu/jvpp_gtpu.h
new file mode 100644
index 00000000..447776ce
--- /dev/null
+++ b/src/vpp-api/java/jvpp-gtpu/jvpp_gtpu.h
@@ -0,0 +1,42 @@
+/*
+ * Copyright (c) 2016 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef __included_jvpp_gtpu_h__
+#define __included_jvpp_gtpu_h__
+
+#include <vnet/vnet.h>
+#include <vnet/ip/ip.h>
+#include <vnet/api_errno.h>
+#include <vlibapi/api.h>
+#include <vlibmemory/api.h>
+#include <jni.h>
+
+/* Global state for JVPP-gtpu */
+typedef struct {
+    /* Pointer to shared memory queue */
+    unix_shared_memory_queue_t * vl_input_queue;
+
+    /* VPP api client index */
+    u32 my_client_index;
+
+    /* Callback object and class references enabling asynchronous Java calls */
+    jobject callbackObject;
+    jclass callbackClass;
+
+} gtpu_main_t;
+
+gtpu_main_t gtpu_main __attribute__((aligned (64)));
+
+
+#endif /* __included_jvpp_gtpu_h__ */
diff --git a/test/test_gtpu.py b/test/test_gtpu.py
new file mode 100644
index 00000000..9411fffc
--- /dev/null
+++ b/test/test_gtpu.py
@@ -0,0 +1,289 @@
+#!/usr/bin/env python
+
+import socket
+from util import ip4n_range
+import unittest
+from framework import VppTestCase, VppTestRunner
+from template_bd import BridgeDomain
+
+from scapy.layers.l2 import Ether, Raw
+from scapy.layers.inet import IP, UDP
+from scapy.contrib.gtp import GTP_U_Header
+from scapy.utils import atol
+
+
+class TestGtpu(BridgeDomain, VppTestCase):
+    """ GTPU Test Case """
+
+    def __init__(self, *args):
+        BridgeDomain.__init__(self)
+        VppTestCase.__init__(self, *args)
+
+    def encapsulate(self, pkt, vni):
+        """
+        Encapsulate the original payload frame by adding GTPU header with its
+        UDP, IP and Ethernet fields
+        """
+        return (Ether(src=self.pg0.remote_mac, dst=self.pg0.local_mac) /
+                IP(src=self.pg0.remote_ip4, dst=self.pg0.local_ip4) /
+                UDP(sport=self.dport, dport=self.dport, chksum=0) /
+                GTP_U_Header(TEID=vni, gtp_type=self.gtp_type, length=150) /
+                pkt)
+
+    def encap_mcast(self, pkt, src_ip, src_mac, vni):
+        """
+        Encapsulate the original payload frame by adding GTPU header with its
+        UDP, IP and Ethernet fields
+        """
+        return (Ether(src=src_mac, dst=self.mcast_mac) /
+                IP(src=src_ip, dst=self.mcast_ip4) /
+                UDP(sport=self.dport, dport=self.dport, chksum=0) /
+                GTP_U_Header(TEID=vni, gtp_type=self.gtp_type, length=150) /
+                pkt)
+
+    def decapsulate(self, pkt):
+        """
+        Decapsulate the original payload frame by removing GTPU header
+        """
+        return pkt[GTP_U_Header].payload
+
+    # Method for checking GTPU encapsulation.
+    #
+    def check_encapsulation(self, pkt, vni, local_only=False, mcast_pkt=False):
+        # Verify source MAC is VPP_MAC and destination MAC is MY_MAC resolved
+        #  by VPP using ARP.
+        self.assertEqual(pkt[Ether].src, self.pg0.local_mac)
+        if not local_only:
+            if not mcast_pkt:
+                self.assertEqual(pkt[Ether].dst, self.pg0.remote_mac)
+            else:
+                self.assertEqual(pkt[Ether].dst, type(self).mcast_mac)
+        # Verify GTPU tunnel source IP is VPP_IP and destination IP is MY_IP.
+        self.assertEqual(pkt[IP].src, self.pg0.local_ip4)
+        if not local_only:
+            if not mcast_pkt:
+                self.assertEqual(pkt[IP].dst, self.pg0.remote_ip4)
+            else:
+                self.assertEqual(pkt[IP].dst, type(self).mcast_ip4)
+        # Verify UDP destination port is GTPU 2152, source UDP port could be
+        #  arbitrary.
+        self.assertEqual(pkt[UDP].dport, type(self).dport)
+        # Verify TEID
+        self.assertEqual(pkt[GTP_U_Header].TEID, vni)
+
+    def test_encap(self):
+        """ Encapsulation test
+        Send frames from pg1
+        Verify receipt of encapsulated frames on pg0
+        """
+        self.pg1.add_stream([self.frame_reply])
+
+        self.pg0.enable_capture()
+
+        self.pg_start()
+
+        # Pick first received frame and check if it's corectly encapsulated.
+        out = self.pg0.get_capture(1)
+        pkt = out[0]
+        self.check_encapsulation(pkt, self.single_tunnel_bd)
+
+        # payload = self.decapsulate(pkt)
+        # self.assert_eq_pkts(payload, self.frame_reply)
+
+    def test_ucast_flood(self):
+        """ Unicast flood test
+        Send frames from pg3
+        Verify receipt of encapsulated frames on pg0
+        """
+        self.pg3.add_stream([self.frame_reply])
+
+        self.pg0.enable_capture()
+
+        self.pg_start()
+
+        # Get packet from each tunnel and assert it's corectly encapsulated.
+        out = self.pg0.get_capture(self.n_ucast_tunnels)
+        for pkt in out:
+            self.check_encapsulation(pkt, self.ucast_flood_bd, True)
+            # payload = self.decapsulate(pkt)
+            # self.assert_eq_pkts(payload, self.frame_reply)
+
+    def test_mcast_flood(self):
+        """ Multicast flood test
+        Send frames from pg2
+        Verify receipt of encapsulated frames on pg0
+        """
+        self.pg2.add_stream([self.frame_reply])
+
+        self.pg0.enable_capture()
+
+        self.pg_start()
+
+        # Pick first received frame and check if it's corectly encapsulated.
+        out = self.pg0.get_capture(1)
+        pkt = out[0]
+        self.check_encapsulation(pkt, self.mcast_flood_bd,
+                                 local_only=False, mcast_pkt=True)
+
+        # payload = self.decapsulate(pkt)
+        # self.assert_eq_pkts(payload, self.frame_reply)
+
+    @classmethod
+    def create_gtpu_flood_test_bd(cls, teid, n_ucast_tunnels):
+        # Create 10 ucast gtpu tunnels under bd
+        ip_range_start = 10
+        ip_range_end = ip_range_start + n_ucast_tunnels
+        next_hop_address = cls.pg0.remote_ip4n
+        for dest_ip4n in ip4n_range(next_hop_address, ip_range_start,
+                                    ip_range_end):
+            # add host route so dest_ip4n will not be resolved
+            cls.vapi.ip_add_del_route(dest_ip4n, 32, next_hop_address)
+            r = cls.vapi.gtpu_add_del_tunnel(
+                src_addr=cls.pg0.local_ip4n,
+                dst_addr=dest_ip4n,
+                teid=teid)
+            cls.vapi.sw_interface_set_l2_bridge(r.sw_if_index, bd_id=teid)
+
+    @classmethod
+    def add_del_shared_mcast_dst_load(cls, is_add):
+        """
+        add or del tunnels sharing the same mcast dst
+        to test gtpu ref_count mechanism
+        """
+        n_shared_dst_tunnels = 20
+        teid_start = 1000
+        teid_end = teid_start + n_shared_dst_tunnels
+        for teid in range(teid_start, teid_end):
+            r = cls.vapi.gtpu_add_del_tunnel(
+                src_addr=cls.pg0.local_ip4n,
+                dst_addr=cls.mcast_ip4n,
+                mcast_sw_if_index=1,
+                teid=teid,
+                is_add=is_add)
+            if r.sw_if_index == 0xffffffff:
+                raise "bad sw_if_index"
+
+    @classmethod
+    def add_shared_mcast_dst_load(cls):
+        cls.add_del_shared_mcast_dst_load(is_add=1)
+
+    @classmethod
+    def del_shared_mcast_dst_load(cls):
+        cls.add_del_shared_mcast_dst_load(is_add=0)
+
+    @classmethod
+    def add_del_mcast_tunnels_load(cls, is_add):
+        """
+        add or del tunnels to test gtpu stability
+        """
+        n_distinct_dst_tunnels = 20
+        ip_range_start = 10
+        ip_range_end = ip_range_start + n_distinct_dst_tunnels
+        for dest_ip4n in ip4n_range(cls.mcast_ip4n, ip_range_start,
+                                    ip_range_end):
+            teid = bytearray(dest_ip4n)[3]
+            cls.vapi.gtpu_add_del_tunnel(
+                src_addr=cls.pg0.local_ip4n,
+                dst_addr=dest_ip4n,
+                mcast_sw_if_index=1,
+                teid=teid,
+                is_add=is_add)
+
+    @classmethod
+    def add_mcast_tunnels_load(cls):
+        cls.add_del_mcast_tunnels_load(is_add=1)
+
+    @classmethod
+    def del_mcast_tunnels_load(cls):
+        cls.add_del_mcast_tunnels_load(is_add=0)
+
+    # Class method to start the GTPU test case.
+    #  Overrides setUpClass method in VppTestCase class.
+    #  Python try..except statement is used to ensure that the tear down of
+    #  the class will be executed even if exception is raised.
+    #  @param cls The class pointer.
+    @classmethod
+    def setUpClass(cls):
+        super(TestGtpu, cls).setUpClass()
+
+        try:
+            cls.dport = 2152
+            cls.gtp_type = 0xff
+
+            # Create 2 pg interfaces.
+            cls.create_pg_interfaces(range(4))
+            for pg in cls.pg_interfaces:
+                pg.admin_up()
+
+            # Configure IPv4 addresses on VPP pg0.
+            cls.pg0.config_ip4()
+
+            # Resolve MAC address for VPP's IP address on pg0.
+            cls.pg0.resolve_arp()
+
+            # Our Multicast address
+            cls.mcast_ip4 = '239.1.1.1'
+            cls.mcast_ip4n = socket.inet_pton(socket.AF_INET, cls.mcast_ip4)
+            iplong = atol(cls.mcast_ip4)
+            cls.mcast_mac = "01:00:5e:%02x:%02x:%02x" % (
+                (iplong >> 16) & 0x7F, (iplong >> 8) & 0xFF, iplong & 0xFF)
+
+            # Create GTPU VTEP on VPP pg0, and put gtpu_tunnel0 and pg1
+            #  into BD.
+            cls.single_tunnel_bd = 11
+            r = cls.vapi.gtpu_add_del_tunnel(
+                src_addr=cls.pg0.local_ip4n,
+                dst_addr=cls.pg0.remote_ip4n,
+                teid=cls.single_tunnel_bd)
+            cls.vapi.sw_interface_set_l2_bridge(r.sw_if_index,
+                                                bd_id=cls.single_tunnel_bd)
+            cls.vapi.sw_interface_set_l2_bridge(cls.pg1.sw_if_index,
+                                                bd_id=cls.single_tunnel_bd)
+
+            # Setup teid 2 to test multicast flooding
+            cls.n_ucast_tunnels = 10
+            cls.mcast_flood_bd = 12
+            cls.create_gtpu_flood_test_bd(cls.mcast_flood_bd,
+                                          cls.n_ucast_tunnels)
+            r = cls.vapi.gtpu_add_del_tunnel(
+                src_addr=cls.pg0.local_ip4n,
+                dst_addr=cls.mcast_ip4n,
+                mcast_sw_if_index=1,
+                teid=cls.mcast_flood_bd)
+            cls.vapi.sw_interface_set_l2_bridge(r.sw_if_index,
+                                                bd_id=cls.mcast_flood_bd)
+            cls.vapi.sw_interface_set_l2_bridge(cls.pg2.sw_if_index,
+                                                bd_id=cls.mcast_flood_bd)
+
+            # Add and delete mcast tunnels to check stability
+            cls.add_shared_mcast_dst_load()
+            cls.add_mcast_tunnels_load()
+            cls.del_shared_mcast_dst_load()
+            cls.del_mcast_tunnels_load()
+
+            # Setup teid 3 to test unicast flooding
+            cls.ucast_flood_bd = 13
+            cls.create_gtpu_flood_test_bd(cls.ucast_flood_bd,
+                                          cls.n_ucast_tunnels)
+            cls.vapi.sw_interface_set_l2_bridge(cls.pg3.sw_if_index,
+                                                bd_id=cls.ucast_flood_bd)
+        except Exception:
+            super(TestGtpu, cls).tearDownClass()
+            raise
+
+    # Method to define VPP actions before tear down of the test case.
+    #  Overrides tearDown method in VppTestCase class.
+    #  @param self The object pointer.
+    def tearDown(self):
+        super(TestGtpu, self).tearDown()
+        if not self.vpp_dead:
+            self.logger.info(self.vapi.cli("show bridge-domain 11 detail"))
+            self.logger.info(self.vapi.cli("show bridge-domain 12 detail"))
+            self.logger.info(self.vapi.cli("show bridge-domain 13 detail"))
+            self.logger.info(self.vapi.cli("show int"))
+            self.logger.info(self.vapi.cli("show gtpu tunnel"))
+            self.logger.info(self.vapi.cli("show trace"))
+
+
+if __name__ == '__main__':
+    unittest.main(testRunner=VppTestRunner)
diff --git a/test/vpp_papi_provider.py b/test/vpp_papi_provider.py
index 83c4a83b..ff8e2cfd 100644
--- a/test/vpp_papi_provider.py
+++ b/test/vpp_papi_provider.py
@@ -1775,3 +1775,35 @@ class VppPapiProvider(object):
                 'is_translation': is_translation,
                 'mtu': mtu
             })
+
+    def gtpu_add_del_tunnel(
+            self,
+            src_addr,
+            dst_addr,
+            is_add=1,
+            is_ipv6=0,
+            mcast_sw_if_index=0xFFFFFFFF,
+            encap_vrf_id=0,
+            decap_next_index=0xFFFFFFFF,
+            teid=0):
+        """
+
+        :param is_add:  (Default value = 1)
+        :param is_ipv6:  (Default value = 0)
+        :param src_addr:
+        :param dst_addr:
+        :param mcast_sw_if_index:  (Default value = 0xFFFFFFFF)
+        :param encap_vrf_id:  (Default value = 0)
+        :param decap_next_index:  (Default value = 0xFFFFFFFF)
+        :param teid:  (Default value = 0)
+
+        """
+        return self.api(self.papi.gtpu_add_del_tunnel,
+                        {'is_add': is_add,
+                         'is_ipv6': is_ipv6,
+                         'src_address': src_addr,
+                         'dst_address': dst_addr,
+                         'mcast_sw_if_index': mcast_sw_if_index,
+                         'encap_vrf_id': encap_vrf_id,
+                         'decap_next_index': decap_next_index,
+                         'teid': teid})