Add Vxlan-Gpe over IPv6

PatchSet4: consolidate code as per comments.

PatchSet3: simplify the code using ip_udp_encap_one/two

PatchSet2: consolidate comments and indent style

Change-Id: Ia8b43f854a46d77e838e198566200ad28fd72472
Signed-off-by: Hongjun Ni <hongjun.ni@intel.com>

9 files changed, 1021 insertions, 715 deletions

diff --git a/vnet/vnet/ip/udp.h b/vnet/vnet/ip/udp.h
index a6991487507..1cf525c6093 100644
--- a/vnet/vnet/ip/udp.h
+++ b/vnet/vnet/ip/udp.h
@@ -51,6 +51,7 @@ _ (547, dhcpv6_to_server)                       \
 _ (546, dhcpv6_to_client)			\
 _ (4341, lisp_gpe6)                             \
 _ (4342, lisp_cp6)                          	\
+_ (4790, vxlan6_gpe)      \
 _ (6633, vpath6_3)
 
 typedef enum {
diff --git a/vnet/vnet/vxlan-gpe/decap.c b/vnet/vnet/vxlan-gpe/decap.c
index 870871e7de6..34bcccd9333 100644
--- a/vnet/vnet/vxlan-gpe/decap.c
+++ b/vnet/vnet/vxlan-gpe/decap.c
@@ -19,6 +19,8 @@
 #include <vnet/pg/pg.h>
 #include <vnet/vxlan-gpe/vxlan_gpe.h>
 
+vlib_node_registration_t vxlan_gpe_input_node;
+
 typedef struct {
   u32 next_index;
   u32 tunnel_index;
@@ -32,15 +34,15 @@ static u8 * format_vxlan_gpe_rx_trace (u8 * s, va_list * args)
   vxlan_gpe_rx_trace_t * t = va_arg (*args, vxlan_gpe_rx_trace_t *);
 
   if (t->tunnel_index != ~0)
-    {
-      s = format (s, "VXLAN-GPE: tunnel %d next %d error %d", t->tunnel_index, 
-                  t->next_index, t->error);
-    }
+  {
+    s = format (s, "VXLAN-GPE: tunnel %d next %d error %d", t->tunnel_index,
+        t->next_index, t->error);
+  }
   else
-    {
-      s = format (s, "VXLAN-GPE: no tunnel next %d error %d\n", t->next_index, 
-                  t->error);
-    }
+  {
+    s = format (s, "VXLAN-GPE: no tunnel next %d error %d\n", t->next_index,
+        t->error);
+  }
   return s;
 }
 
@@ -54,22 +56,24 @@ static u8 * format_vxlan_gpe_with_length (u8 * s, va_list * args)
   return s;
 }
 
-static uword
+always_inline uword
 vxlan_gpe_input (vlib_main_t * vm,
                      vlib_node_runtime_t * node,
-                     vlib_frame_t * from_frame)
+                     vlib_frame_t * from_frame,
+					 u8 is_ip4)
 {
-  u32 n_left_from, next_index, * from, * to_next;
+  u32 n_left_from, next_index, *from, *to_next;
   vxlan_gpe_main_t * ngm = &vxlan_gpe_main;
   vnet_main_t * vnm = ngm->vnet_main;
   vnet_interface_main_t * im = &vnm->interface_main;
   u32 last_tunnel_index = ~0;
-  vxlan_gpe_tunnel_key_t last_key;
+  vxlan4_gpe_tunnel_key_t last_key4;
+  vxlan6_gpe_tunnel_key_t last_key6;
   u32 pkts_decapsulated = 0;
-  u32 cpu_index = os_get_cpu_number();
+  u32 cpu_index = os_get_cpu_number ();
   u32 stats_sw_if_index, stats_n_packets, stats_n_bytes;
 
-  memset (&last_key, 0xff, sizeof (last_key));
+  memset (&last_key4, 0xff, sizeof(last_key4));
 
   from = vlib_frame_vector_args (from_frame);
   n_left_from = from_frame->n_vectors;
@@ -79,339 +83,458 @@ vxlan_gpe_input (vlib_main_t * vm,
   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_vxlan_gpe_header_t * iuvn0, * iuvn1;
-	  uword * p0, * p1;
-          u32 tunnel_index0, tunnel_index1;
-          vxlan_gpe_tunnel_t * t0, * t1;
-          vxlan_gpe_tunnel_key_t key0, key1;
-          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 vxlan header */
-          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)));
-
-          iuvn0 = vlib_buffer_get_current (b0);
-          iuvn1 = vlib_buffer_get_current (b1);
-
-          /* pop (ip, udp, vxlan) */
-          vlib_buffer_advance (b0, sizeof (*iuvn0));
-          vlib_buffer_advance (b1, sizeof (*iuvn1));
-
-          tunnel_index0 = ~0;
-          tunnel_index1 = ~0;
-          error0 = 0;
-          error1 = 0;
-
-	  next0 = (iuvn0->vxlan.protocol < node->n_next_nodes) ? iuvn0->vxlan.protocol : VXLAN_GPE_INPUT_NEXT_DROP;
-	  next1 = (iuvn1->vxlan.protocol < node->n_next_nodes) ? iuvn1->vxlan.protocol : VXLAN_GPE_INPUT_NEXT_DROP;
-
-
-
-
-          key0.local = iuvn0->ip4.dst_address.as_u32;
-          key1.local = iuvn1->ip4.dst_address.as_u32;
-
-	  key0.remote = iuvn0->ip4.src_address.as_u32;
-	  key1.remote = iuvn1->ip4.src_address.as_u32;
-
-          key0.vni = iuvn0->vxlan.vni_res;
-          key1.vni = iuvn1->vxlan.vni_res;
-
-          key0.pad = 0;
-          key1.pad = 0;
-
-	  /* Processing for key0 */
-          if (PREDICT_FALSE ((key0.as_u64[0] != last_key.as_u64[0])
-                             || (key0.as_u64[1] != last_key.as_u64[1])))
-            {
-              p0 = hash_get_mem (ngm->vxlan_gpe_tunnel_by_key, &key0);
-
-              if (p0 == 0)
-                {
-                  error0 = VXLAN_GPE_ERROR_NO_SUCH_TUNNEL;
-                  goto trace0;
-                }
+  {
+    u32 n_left_to_next;
 
-              last_key.as_u64[0] = key0.as_u64[0];
-              last_key.as_u64[1] = key0.as_u64[1];
-              tunnel_index0 = last_tunnel_index = p0[0];
-            }
-          else
-            tunnel_index0 = last_tunnel_index;
+    vlib_get_next_frame(vm, node, next_index, to_next, n_left_to_next);
 
-          t0 = pool_elt_at_index (ngm->tunnels, tunnel_index0);
+    while (n_left_from >= 4 && n_left_to_next >= 2)
+    {
+      u32 bi0, bi1;
+      vlib_buffer_t * b0, *b1;
+      u32 next0, next1;
+      ip4_vxlan_gpe_header_t * iuvn4_0, *iuvn4_1;
+      ip6_vxlan_gpe_header_t * iuvn6_0, *iuvn6_1;
+      uword * p0, *p1;
+      u32 tunnel_index0, tunnel_index1;
+      vxlan_gpe_tunnel_t * t0, *t1;
+      vxlan4_gpe_tunnel_key_t key4_0, key4_1;
+      vxlan6_gpe_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);
+
+      if (is_ip4)
+      {
+        /* udp leaves current_data pointing at the vxlan-gpe header */
+        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)));
+
+        iuvn4_0 = vlib_buffer_get_current (b0);
+        iuvn4_1 = vlib_buffer_get_current (b1);
+
+        /* pop (ip, udp, vxlan) */
+        vlib_buffer_advance (b0, sizeof(*iuvn4_0));
+        vlib_buffer_advance (b1, sizeof(*iuvn4_1));
+      }
+      else
+      {
+        /* udp leaves current_data pointing at the vxlan-gpe header */
+        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)));
+
+        iuvn6_0 = vlib_buffer_get_current (b0);
+        iuvn6_1 = vlib_buffer_get_current (b1);
+
+        /* pop (ip, udp, vxlan) */
+        vlib_buffer_advance (b0, sizeof(*iuvn6_0));
+        vlib_buffer_advance (b1, sizeof(*iuvn6_1));
+      }
+
+      tunnel_index0 = ~0;
+      tunnel_index1 = ~0;
+      error0 = 0;
+      error1 = 0;
+
+      if (is_ip4)
+      {
+        next0 = (iuvn4_0->vxlan.protocol < node->n_next_nodes) ?
+                iuvn4_0->vxlan.protocol : VXLAN_GPE_INPUT_NEXT_DROP;
+        next1 = (iuvn4_1->vxlan.protocol < node->n_next_nodes) ?
+                iuvn4_1->vxlan.protocol : VXLAN_GPE_INPUT_NEXT_DROP;
+
+        key4_0.local = iuvn4_0->ip4.dst_address.as_u32;
+        key4_1.local = iuvn4_1->ip4.dst_address.as_u32;
+
+        key4_0.remote = iuvn4_0->ip4.src_address.as_u32;
+        key4_1.remote = iuvn4_1->ip4.src_address.as_u32;
+
+        key4_0.vni = iuvn4_0->vxlan.vni_res;
+        key4_1.vni = iuvn4_1->vxlan.vni_res;
+
+        key4_0.pad = 0;
+        key4_1.pad = 0;
+
+        /* Processing for key4_0 */
+        if (PREDICT_FALSE((key4_0.as_u64[0] != last_key4.as_u64[0])
+                || (key4_0.as_u64[1] != last_key4.as_u64[1])))
+        {
+          p0 = hash_get_mem(ngm->vxlan4_gpe_tunnel_by_key, &key4_0);
+
+          if (p0 == 0)
+          {
+            error0 = VXLAN_GPE_ERROR_NO_SUCH_TUNNEL;
+            goto trace0;
+          }
 
-          next0 = t0->protocol;
-
-          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 */
-          vnet_update_l2_len (b0);
-	  
-	  /*
-	   * ip[46] lookup in the configured FIB
-	   */
-	  vnet_buffer(b0)->sw_if_index[VLIB_TX] = t0->decap_fib_index;
+          last_key4.as_u64[0] = key4_0.as_u64[0];
+          last_key4.as_u64[1] = key4_0.as_u64[1];
+          tunnel_index0 = last_tunnel_index = p0[0];
+        }
+        else
+          tunnel_index0 = last_tunnel_index;
+      }
+      else /* is_ip6 */
+      {
+        next0 = (iuvn6_0->vxlan.protocol < node->n_next_nodes) ?
+                iuvn6_0->vxlan.protocol : VXLAN_GPE_INPUT_NEXT_DROP;
+        next1 = (iuvn6_1->vxlan.protocol < node->n_next_nodes) ?
+                iuvn6_1->vxlan.protocol : VXLAN_GPE_INPUT_NEXT_DROP;
+
+        key6_0.local.as_u64[0] = iuvn6_0->ip6.dst_address.as_u64[0];
+        key6_0.local.as_u64[1] = iuvn6_0->ip6.dst_address.as_u64[1];
+        key6_1.local.as_u64[0] = iuvn6_1->ip6.dst_address.as_u64[0];
+        key6_1.local.as_u64[1] = iuvn6_1->ip6.dst_address.as_u64[1];
+
+        key6_0.remote.as_u64[0] = iuvn6_0->ip6.src_address.as_u64[0];
+        key6_0.remote.as_u64[1] = iuvn6_0->ip6.src_address.as_u64[1];
+        key6_1.remote.as_u64[0] = iuvn6_1->ip6.src_address.as_u64[0];
+        key6_1.remote.as_u64[1] = iuvn6_1->ip6.src_address.as_u64[1];
+
+        key6_0.vni = iuvn6_0->vxlan.vni_res;
+        key6_1.vni = iuvn6_1->vxlan.vni_res;
+
+        /* Processing for key6_0 */
+        if (PREDICT_FALSE(memcmp (&key6_0, &last_key6, sizeof(last_key6)) != 0))
+        {
+          p0 = hash_get_mem(ngm->vxlan6_gpe_tunnel_by_key, &key6_0);
+
+          if (p0 == 0)
+          {
+            error0 = VXLAN_GPE_ERROR_NO_SUCH_TUNNEL;
+            goto trace0;
+          }
 
-          pkts_decapsulated++;
-          stats_n_packets += 1;
-          stats_n_bytes += len0;
+          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(ngm->tunnels, tunnel_index0);
+
+      next0 = t0->protocol;
+
+      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 */
+      vnet_update_l2_len (b0);
+
+      /*
+       * ip[46] lookup in the configured FIB
+       */
+      vnet_buffer(b0)->sw_if_index[VLIB_TX] = t0->decap_fib_index;
+
+      pkts_decapsulated++;
+      stats_n_packets += 1;
+      stats_n_bytes += len0;
+
+      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,
+              cpu_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))
+      {
+        vxlan_gpe_rx_trace_t *tr = vlib_add_trace (vm, node, b0, sizeof(*tr));
+        tr->next_index = next0;
+        tr->error = error0;
+        tr->tunnel_index = tunnel_index0;
+      }
+
+      /* Process packet 1 */
+      if (is_ip4)
+      {
+        /* Processing for key4_1 */
+        if (PREDICT_FALSE(
+            (key4_1.as_u64[0] != last_key4.as_u64[0])
+                || (key4_1.as_u64[1] != last_key4.as_u64[1])))
+        {
+          p1 = hash_get_mem(ngm->vxlan4_gpe_tunnel_by_key, &key4_1);
+
+          if (p1 == 0)
+          {
+            error1 = VXLAN_GPE_ERROR_NO_SUCH_TUNNEL;
+            goto trace1;
+          }
 
-          if (PREDICT_FALSE(sw_if_index0 != stats_sw_if_index))
+          last_key4.as_u64[0] = key4_1.as_u64[0];
+          last_key4.as_u64[1] = key4_1.as_u64[1];
+          tunnel_index1 = last_tunnel_index = p1[0];
+        }
+        else
+          tunnel_index1 = last_tunnel_index;
+      }
+      else /* is_ip6 */
+      {
+        /* Processing for key6_1 */
+        if (PREDICT_FALSE(memcmp (&key6_1, &last_key6, sizeof(last_key6)) != 0))
+        {
+          p1 = hash_get_mem(ngm->vxlan6_gpe_tunnel_by_key, &key6_1);
+
+          if (p1 == 0)
           {
-            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,
-                  cpu_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;
+            error1 = VXLAN_GPE_ERROR_NO_SUCH_TUNNEL;
+            goto trace1;
           }
 
-        trace0:
-          b0->error = error0 ? node->errors[error0] : 0;
-
-          if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
-            {
-              vxlan_gpe_rx_trace_t *tr
-                = vlib_add_trace (vm, node, b0, sizeof (*tr));
-              tr->next_index = next0;
-              tr->error = error0;
-              tr->tunnel_index = tunnel_index0;
-            }
-
-
-	  /* Processing for key1 */
-          if (PREDICT_FALSE ((key1.as_u64[0] != last_key.as_u64[0])
-                             || (key1.as_u64[1] != last_key.as_u64[1])))
-            {
-              p1 = hash_get_mem (ngm->vxlan_gpe_tunnel_by_key, &key1);
-
-              if (p1 == 0)
-                {
-                  error1 = VXLAN_GPE_ERROR_NO_SUCH_TUNNEL;
-                  goto trace1;
-                }
-
-              last_key.as_u64[0] = key1.as_u64[0];
-              last_key.as_u64[1] = key1.as_u64[1];
-              tunnel_index1 = last_tunnel_index = p1[0];
-            }
-          else
-            tunnel_index1 = last_tunnel_index;
-
-          t1 = pool_elt_at_index (ngm->tunnels, tunnel_index1);
-
-          next1 = t1->protocol;
-          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 */
-          vnet_update_l2_len (b1);
-
-	  /*
-	   * ip[46] lookup in the configured FIB
-	   */
-	  vnet_buffer(b1)->sw_if_index[VLIB_TX] = t1->decap_fib_index;
-
-
-          pkts_decapsulated++;
-          stats_n_packets += 1;
-          stats_n_bytes += len1;
-          
-	  /* Batch stats increment on the same vxlan tunnel so counter
-           is not incremented per packet */
-          if (PREDICT_FALSE(sw_if_index1 != stats_sw_if_index))
+          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(ngm->tunnels, tunnel_index1);
+
+      next1 = t1->protocol;
+      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 */
+      vnet_update_l2_len (b1);
+
+      /*
+       * ip[46] lookup in the configured FIB
+       */
+      vnet_buffer(b1)->sw_if_index[VLIB_TX] = t1->decap_fib_index;
+
+      pkts_decapsulated++;
+      stats_n_packets += 1;
+      stats_n_bytes += len1;
+
+      /* Batch stats increment on the same vxlan 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,
+              cpu_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;
+      }
+      vnet_buffer(b1)->sw_if_index[VLIB_TX] = t1->decap_fib_index;
+
+      trace1: b1->error = error1 ? node->errors[error1] : 0;
+
+      if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED))
+      {
+        vxlan_gpe_rx_trace_t *tr = vlib_add_trace (vm, node, b1, sizeof(*tr));
+        tr->next_index = next1;
+        tr->error = error1;
+        tr->tunnel_index = tunnel_index1;
+      }
+
+      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_vxlan_gpe_header_t * iuvn4_0;
+      ip6_vxlan_gpe_header_t * iuvn6_0;
+      uword * p0;
+      u32 tunnel_index0;
+      vxlan_gpe_tunnel_t * t0;
+      vxlan4_gpe_tunnel_key_t key4_0;
+      vxlan6_gpe_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);
+
+      if (is_ip4)
+      {
+        /* udp leaves current_data pointing at the vxlan-gpe header */
+        vlib_buffer_advance (
+            b0, -(word) (sizeof(udp_header_t) + sizeof(ip4_header_t)));
+
+        iuvn4_0 = vlib_buffer_get_current (b0);
+
+        /* pop (ip, udp, vxlan) */
+        vlib_buffer_advance (b0, sizeof(*iuvn4_0));
+      }
+      else
+      {
+        /* udp leaves current_data pointing at the vxlan-gpe header */
+        vlib_buffer_advance (
+            b0, -(word) (sizeof(udp_header_t) + sizeof(ip6_header_t)));
+
+        iuvn6_0 = vlib_buffer_get_current (b0);
+
+        /* pop (ip, udp, vxlan) */
+        vlib_buffer_advance (b0, sizeof(*iuvn6_0));
+      }
+
+      tunnel_index0 = ~0;
+      error0 = 0;
+
+      if (is_ip4)
+      {
+        next0 =
+            (iuvn4_0->vxlan.protocol < node->n_next_nodes) ?
+                iuvn4_0->vxlan.protocol : VXLAN_GPE_INPUT_NEXT_DROP;
+
+        key4_0.local = iuvn4_0->ip4.dst_address.as_u32;
+        key4_0.remote = iuvn4_0->ip4.src_address.as_u32;
+        key4_0.vni = iuvn4_0->vxlan.vni_res;
+        key4_0.pad = 0;
+
+        /* Processing for key4_0 */
+        if (PREDICT_FALSE(
+            (key4_0.as_u64[0] != last_key4.as_u64[0])
+                || (key4_0.as_u64[1] != last_key4.as_u64[1])))
+        {
+          p0 = hash_get_mem(ngm->vxlan4_gpe_tunnel_by_key, &key4_0);
+
+          if (p0 == 0)
           {
-            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,
-                  cpu_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;
+            error0 = VXLAN_GPE_ERROR_NO_SUCH_TUNNEL;
+            goto trace00;
           }
-          vnet_buffer(b1)->sw_if_index[VLIB_TX] = t1->decap_fib_index;
-
-        trace1:
-          b1->error = error1 ? node->errors[error1] : 0;
-
-          if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED))
-            {
-              vxlan_gpe_rx_trace_t *tr
-                = vlib_add_trace (vm, node, b1, sizeof (*tr));
-              tr->next_index = next1;
-              tr->error = error1;
-              tr->tunnel_index = tunnel_index1;
-            }
-
-	  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_vxlan_gpe_header_t * iuvn0;
-	  uword * p0;
-          u32 tunnel_index0;
-          vxlan_gpe_tunnel_t * t0;
-          vxlan_gpe_tunnel_key_t key0;
-          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 vxlan header */
-          vlib_buffer_advance 
-            (b0, -(word)(sizeof(udp_header_t)+sizeof(ip4_header_t)));
-
-          iuvn0 = vlib_buffer_get_current (b0);
-
-          /* pop (ip, udp, vxlan) */
-          vlib_buffer_advance (b0, sizeof (*iuvn0));
-
-          tunnel_index0 = ~0;
-          error0 = 0;
-          next0 = (iuvn0->vxlan.protocol < node->n_next_nodes) ? iuvn0->vxlan.protocol : VXLAN_GPE_INPUT_NEXT_DROP;
-
-          key0.local = iuvn0->ip4.dst_address.as_u32;
-	  key0.remote = iuvn0->ip4.src_address.as_u32;
-          key0.vni = iuvn0->vxlan.vni_res;
-          key0.pad = 0;
-
-          if (PREDICT_FALSE ((key0.as_u64[0] != last_key.as_u64[0])
-                             || (key0.as_u64[1] != last_key.as_u64[1])))
-            {
-              p0 = hash_get_mem (ngm->vxlan_gpe_tunnel_by_key, &key0);
-          
-              if (p0 == 0)
-                {
-                  error0 = VXLAN_GPE_ERROR_NO_SUCH_TUNNEL;
-                  goto trace00;
-                }
-
-              last_key.as_u64[0] = key0.as_u64[0];
-              last_key.as_u64[1] = key0.as_u64[1];
-              tunnel_index0 = last_tunnel_index = p0[0];
-            }
-          else
-            tunnel_index0 = last_tunnel_index;
-
-          t0 = pool_elt_at_index (ngm->tunnels, tunnel_index0);
-
-          next0 = t0->protocol;
-	  
-	  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 */
-          vnet_update_l2_len (b0);
-
-	  /*
-	   * ip[46] lookup in the configured FIB
-	   */
-	  vnet_buffer(b0)->sw_if_index[VLIB_TX] = t0->decap_fib_index;
-
-          pkts_decapsulated ++;
-          stats_n_packets += 1;
-          stats_n_bytes += len0;
-
-          /* Batch stats increment on the same vxlan-gpe tunnel so counter
-           is not incremented per packet */
-          if (PREDICT_FALSE(sw_if_index0 != stats_sw_if_index))
+
+          last_key4.as_u64[0] = key4_0.as_u64[0];
+          last_key4.as_u64[1] = key4_0.as_u64[1];
+          tunnel_index0 = last_tunnel_index = p0[0];
+        }
+        else
+          tunnel_index0 = last_tunnel_index;
+      }
+      else /* is_ip6 */
+      {
+        next0 = (iuvn6_0->vxlan.protocol < node->n_next_nodes) ?
+                iuvn6_0->vxlan.protocol : VXLAN_GPE_INPUT_NEXT_DROP;
+
+        key6_0.local.as_u64[0] = iuvn6_0->ip6.dst_address.as_u64[0];
+        key6_0.local.as_u64[1] = iuvn6_0->ip6.dst_address.as_u64[1];
+        key6_0.remote.as_u64[0] = iuvn6_0->ip6.src_address.as_u64[0];
+        key6_0.remote.as_u64[1] = iuvn6_0->ip6.src_address.as_u64[1];
+        key6_0.vni = iuvn6_0->vxlan.vni_res;
+
+        /* Processing for key6_0 */
+        if (PREDICT_FALSE(memcmp (&key6_0, &last_key6, sizeof(last_key6)) != 0))
+        {
+          p0 = hash_get_mem(ngm->vxlan6_gpe_tunnel_by_key, &key6_0);
+
+          if (p0 == 0)
           {
-            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,
-                  cpu_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;
+            error0 = VXLAN_GPE_ERROR_NO_SUCH_TUNNEL;
+            goto trace00;
           }
 
-        trace00:
-          b0->error = error0 ? node->errors[error0] : 0;
-
-          if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
-            {
-              vxlan_gpe_rx_trace_t *tr
-                = vlib_add_trace (vm, node, b0, sizeof (*tr));
-              tr->next_index = next0;
-              tr->error = error0;
-              tr->tunnel_index = tunnel_index0;
-            }
-	  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);
+          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(ngm->tunnels, tunnel_index0);
+
+      next0 = t0->protocol;
+
+      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 */
+      vnet_update_l2_len (b0);
+
+      /*
+       * ip[46] lookup in the configured FIB
+       */
+      vnet_buffer(b0)->sw_if_index[VLIB_TX] = t0->decap_fib_index;
+
+      pkts_decapsulated++;
+      stats_n_packets += 1;
+      stats_n_bytes += len0;
+
+      /* Batch stats increment on the same vxlan-gpe 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,
+              cpu_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))
+      {
+        vxlan_gpe_rx_trace_t *tr = vlib_add_trace (vm, node, b0, sizeof(*tr));
+        tr->next_index = next0;
+        tr->error = error0;
+        tr->tunnel_index = tunnel_index0;
+      }
+      vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next,
+                                      n_left_to_next, bi0, next0);
     }
+
+    vlib_put_next_frame (vm, node, next_index, n_left_to_next);
+  }
   vlib_node_increment_counter (vm, vxlan_gpe_input_node.index,
-                               VXLAN_GPE_ERROR_DECAPSULATED,
-                               pkts_decapsulated);
+                               VXLAN_GPE_ERROR_DECAPSULATED, pkts_decapsulated);
   /* Increment any remaining batch stats */
   if (stats_n_packets)
   {
-    vlib_increment_combined_counter(
+    vlib_increment_combined_counter (
         im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_RX, cpu_index,
         stats_sw_if_index, stats_n_packets, stats_n_bytes);
     node->runtime_data[0] = stats_sw_if_index;
@@ -419,6 +542,20 @@ vxlan_gpe_input (vlib_main_t * vm,
   return from_frame->n_vectors;
 }
 
+static uword
+vxlan4_gpe_input (vlib_main_t * vm, vlib_node_runtime_t * node,
+                  vlib_frame_t * from_frame)
+{
+  return vxlan_gpe_input (vm, node, from_frame, /* is_ip4 */1);
+}
+
+static uword
+vxlan6_gpe_input (vlib_main_t * vm, vlib_node_runtime_t * node,
+                  vlib_frame_t * from_frame)
+{
+  return vxlan_gpe_input (vm, node, from_frame, /* is_ip4 */0);
+}
+
 static char * vxlan_gpe_error_strings[] = {
 #define vxlan_gpe_error(n,s) s,
 #include <vnet/vxlan-gpe/vxlan_gpe_error.def>
@@ -426,9 +563,9 @@ static char * vxlan_gpe_error_strings[] = {
 #undef _
 };
 
-VLIB_REGISTER_NODE (vxlan_gpe_input_node) = {
-  .function = vxlan_gpe_input,
-  .name = "vxlan-gpe-input",
+VLIB_REGISTER_NODE (vxlan4_gpe_input_node) = {
+  .function = vxlan4_gpe_input,
+  .name = "vxlan4-gpe-input",
   /* Takes a vector of packets. */
   .vector_size = sizeof (u32),
   .type = VLIB_NODE_TYPE_INTERNAL,
@@ -447,4 +584,27 @@ VLIB_REGISTER_NODE (vxlan_gpe_input_node) = {
   // $$$$ .unformat_buffer = unformat_vxlan_gpe_header,
 };
 
+VLIB_NODE_FUNCTION_MULTIARCH (vxlan4_gpe_input_node, vxlan4_gpe_input)
+
+VLIB_REGISTER_NODE (vxlan6_gpe_input_node) = {
+  .function = vxlan6_gpe_input,
+  .name = "vxlan6-gpe-input",
+  /* Takes a vector of packets. */
+  .vector_size = sizeof (u32),
+  .type = VLIB_NODE_TYPE_INTERNAL,
+  .n_errors = ARRAY_LEN(vxlan_gpe_error_strings),
+  .error_strings = vxlan_gpe_error_strings,
+
+  .n_next_nodes = VXLAN_GPE_INPUT_N_NEXT,
+  .next_nodes = {
+#define _(s,n) [VXLAN_GPE_INPUT_NEXT_##s] = n,
+    foreach_vxlan_gpe_input_next
+#undef _
+  },
+
+  .format_buffer = format_vxlan_gpe_with_length,
+  .format_trace = format_vxlan_gpe_rx_trace,
+  // $$$$ .unformat_buffer = unformat_vxlan_gpe_header,
+};
 
+VLIB_NODE_FUNCTION_MULTIARCH (vxlan6_gpe_input_node, vxlan6_gpe_input)
diff --git a/vnet/vnet/vxlan-gpe/encap.c b/vnet/vnet/vxlan-gpe/encap.c
index 3ffe2a62db2..ba0eca2a0a1 100644
--- a/vnet/vnet/vxlan-gpe/encap.c
+++ b/vnet/vnet/vxlan-gpe/encap.c
@@ -38,6 +38,7 @@ typedef enum {
 
 typedef enum {
   VXLAN_GPE_ENCAP_NEXT_IP4_LOOKUP,
+  VXLAN_GPE_ENCAP_NEXT_IP6_LOOKUP,
   VXLAN_GPE_ENCAP_NEXT_DROP,
   VXLAN_GPE_ENCAP_N_NEXT
 } vxlan_gpe_encap_next_t;
@@ -58,21 +59,59 @@ u8 * format_vxlan_gpe_encap_trace (u8 * s, va_list * args)
   return s;
 }
 
-#define foreach_fixed_header_offset             \
-_(0) _(1) _(2) _(3) _(4) _(5) _(6)
+always_inline void
+vxlan_gpe_encap_one_inline (vxlan_gpe_main_t * ngm, vlib_buffer_t * b0,
+                            vxlan_gpe_tunnel_t * t0, u32 * next0, u8 is_v4)
+{
+  ASSERT(sizeof(ip4_vxlan_gpe_header_t) == 36);
+  ASSERT(sizeof(ip6_vxlan_gpe_header_t) == 56);
+
+  if (is_v4)
+    {
+      ip_udp_encap_one (ngm->vlib_main, b0, t0->rewrite, 36, 1);
+      next0[0] = VXLAN_GPE_ENCAP_NEXT_IP4_LOOKUP;
+
+    }
+  else
+    {
+      ip_udp_encap_one (ngm->vlib_main, b0, t0->rewrite, 56, 0);
+      next0[0] = VXLAN_GPE_ENCAP_NEXT_IP6_LOOKUP;
+    }
+}
+
+always_inline void
+vxlan_gpe_encap_two_inline (vxlan_gpe_main_t * ngm, vlib_buffer_t * b0, vlib_buffer_t * b1,
+                            vxlan_gpe_tunnel_t * t0, vxlan_gpe_tunnel_t * t1, u32 * next0,
+                            u32 * next1, u8 is_v4)
+{
+  ASSERT(sizeof(ip4_vxlan_gpe_header_t) == 36);
+  ASSERT(sizeof(ip6_vxlan_gpe_header_t) == 56);
+
+  if (is_v4)
+    {
+      ip_udp_encap_one (ngm->vlib_main, b0, t0->rewrite, 36, 1);
+      ip_udp_encap_one (ngm->vlib_main, b1, t1->rewrite, 36, 1);
+      next0[0] = next1[0] = VXLAN_GPE_ENCAP_NEXT_IP4_LOOKUP;
+    }
+  else
+    {
+      ip_udp_encap_one (ngm->vlib_main, b0, t0->rewrite, 56, 0);
+      ip_udp_encap_one (ngm->vlib_main, b1, t1->rewrite, 56, 0);
+      next0[0] = next1[0] = VXLAN_GPE_ENCAP_NEXT_IP6_LOOKUP;
+    }
+}
 
 static uword
 vxlan_gpe_encap (vlib_main_t * vm,
                vlib_node_runtime_t * node,
                vlib_frame_t * from_frame)
 {
-  u32 n_left_from, next_index, * from, * to_next;
+  u32 n_left_from, next_index, *from, *to_next;
   vxlan_gpe_main_t * ngm = &vxlan_gpe_main;
   vnet_main_t * vnm = ngm->vnet_main;
   vnet_interface_main_t * im = &vnm->interface_main;
   u32 pkts_encapsulated = 0;
-  u16 old_l0 = 0, old_l1 = 0;
-  u32 cpu_index = os_get_cpu_number();
+  u32 cpu_index = os_get_cpu_number ();
   u32 stats_sw_if_index, stats_n_packets, stats_n_bytes;
 
   from = vlib_frame_vector_args (from_frame);
@@ -83,324 +122,207 @@ vxlan_gpe_encap (vlib_main_t * vm,
   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 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 = VXLAN_GPE_ENCAP_NEXT_IP4_LOOKUP;
-          u32 next1 = VXLAN_GPE_ENCAP_NEXT_IP4_LOOKUP;
-          u32 sw_if_index0, sw_if_index1, len0, len1;
-          vnet_hw_interface_t * hi0, * hi1;
-          ip4_header_t * ip0, * ip1;
-          udp_header_t * udp0, * udp1;
-          u64 * copy_src0, * copy_dst0;
-          u64 * copy_src1, * copy_dst1;
-          u32 * copy_src_last0, * copy_dst_last0;
-          u32 * copy_src_last1, * copy_dst_last1;
-          vxlan_gpe_tunnel_t * t0, * t1;
-          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);
-
-          /* 1-wide cache? */
-          sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX];
-          sw_if_index1 = vnet_buffer(b1)->sw_if_index[VLIB_TX];
-          hi0 = vnet_get_sup_hw_interface
-            (vnm, vnet_buffer(b0)->sw_if_index[VLIB_TX]);
-          hi1 = vnet_get_sup_hw_interface
-            (vnm, vnet_buffer(b1)->sw_if_index[VLIB_TX]);
-
-          t0 = pool_elt_at_index (ngm->tunnels, hi0->dev_instance);
-          t1 = pool_elt_at_index (ngm->tunnels, hi1->dev_instance);
-
-          ASSERT(vec_len(t0->rewrite) >= 24);
-          ASSERT(vec_len(t1->rewrite) >= 24);
-
-          /* Apply the rewrite string. $$$$ vnet_rewrite? */
-          vlib_buffer_advance (b0, -(word)_vec_len(t0->rewrite));
-          vlib_buffer_advance (b1, -(word)_vec_len(t1->rewrite));
-
-          ip0 = vlib_buffer_get_current(b0);
-          ip1 = vlib_buffer_get_current(b1);
-          /* Copy the fixed header */
-          copy_dst0 = (u64 *) ip0;
-          copy_src0 = (u64 *) t0->rewrite;
-          copy_dst1 = (u64 *) ip1;
-          copy_src1 = (u64 *) t1->rewrite;
-
-          ASSERT (sizeof (ip4_vxlan_gpe_header_t) == 36);
-
-          /* Copy first 36 octets 8-bytes at a time */
-#define _(offs) copy_dst0[offs] = copy_src0[offs];
-          foreach_fixed_header_offset;
-#undef _
-#define _(offs) copy_dst1[offs] = copy_src1[offs];
-          foreach_fixed_header_offset;
-#undef _
+      u32 bi0, bi1;
+      vlib_buffer_t * b0, *b1;
+      u32 next0, next1;
+      u32 sw_if_index0, sw_if_index1, len0, len1;
+      vnet_hw_interface_t * hi0, *hi1;
+      vxlan_gpe_tunnel_t * t0, *t1;
+      u8 is_ip4_0, is_ip4_1;
+
+      next0 = next1 = VXLAN_GPE_ENCAP_NEXT_IP4_LOOKUP;
+
+      /* 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);
+
+      /* 1-wide cache? */
+      sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX];
+      sw_if_index1 = vnet_buffer(b1)->sw_if_index[VLIB_TX];
+      hi0 = vnet_get_sup_hw_interface (vnm, vnet_buffer(b0)->sw_if_index[VLIB_TX]);
+      hi1 = vnet_get_sup_hw_interface (vnm, vnet_buffer(b1)->sw_if_index[VLIB_TX]);
+
+      t0 = pool_elt_at_index(ngm->tunnels, hi0->dev_instance);
+      t1 = pool_elt_at_index(ngm->tunnels, hi1->dev_instance);
+
+      is_ip4_0 = (t0->flags & VXLAN_GPE_TUNNEL_IS_IPV4);
+      is_ip4_1 = (t1->flags & VXLAN_GPE_TUNNEL_IS_IPV4);
+
+      if (PREDICT_TRUE(is_ip4_0 == is_ip4_1))
+      {
+        vxlan_gpe_encap_two_inline (ngm, b0, b1, t0, t1, &next0, &next1,is_ip4_0);
+      }
+      else
+      {
+        vxlan_gpe_encap_one_inline (ngm, b0, t0, &next0, is_ip4_0);
+        vxlan_gpe_encap_one_inline (ngm, b1, t1, &next1, is_ip4_1);
+      }
+
+      /* Reset to look up tunnel partner in the configured FIB */
+      vnet_buffer(b0)->sw_if_index[VLIB_TX] = t0->encap_fib_index;
+      vnet_buffer(b1)->sw_if_index[VLIB_TX] = t1->encap_fib_index;
+      vnet_buffer(b0)->sw_if_index[VLIB_RX] = sw_if_index0;
+      vnet_buffer(b1)->sw_if_index[VLIB_RX] = sw_if_index1;
+      pkts_encapsulated += 2;
+
+      len0 = vlib_buffer_length_in_chain (vm, b0);
+      len1 = vlib_buffer_length_in_chain (vm, b0);
+      stats_n_packets += 2;
+      stats_n_bytes += len0 + len1;
+
+      /* Batch stats increment on the same vxlan 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,
+                cpu_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,
+              cpu_index, sw_if_index0, 1, len0);
+          vlib_increment_combined_counter (
+              im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
+              cpu_index, sw_if_index1, 1, len1);
+        }
+      }
+
+      if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
+      {
+        vxlan_gpe_encap_trace_t *tr = vlib_add_trace (vm, node, b0, sizeof(*tr));
+        tr->tunnel_index = t0 - ngm->tunnels;
+      }
+
+      if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED))
+      {
+        vxlan_gpe_encap_trace_t *tr = vlib_add_trace (vm, node, b1,
+                                                      sizeof(*tr));
+        tr->tunnel_index = t1 - ngm->tunnels;
+      }
+
+      vlib_validate_buffer_enqueue_x2(vm, node, next_index, to_next,
+                                      n_left_to_next, bi0, bi1, next0, next1);
+    }
 
-          /* Last 4 octets. Hopefully gcc will be our friend */
-          copy_dst_last0 = (u32 *)(&copy_dst0[7]);
-          copy_src_last0 = (u32 *)(&copy_src0[7]);
-          copy_dst_last1 = (u32 *)(&copy_dst1[7]);
-          copy_src_last1 = (u32 *)(&copy_src1[7]);
-          
-          copy_dst_last0[0] = copy_src_last0[0];
-          copy_dst_last1[0] = copy_src_last1[0];
-
-          /* If there are TLVs to copy, do so */
-          if (PREDICT_FALSE (_vec_len(t0->rewrite) > 64))
-            clib_memcpy (&copy_dst0[3], t0->rewrite + 64 , 
-                    _vec_len (t0->rewrite)-64);
-
-          if (PREDICT_FALSE (_vec_len(t1->rewrite) > 64))
-            clib_memcpy (&copy_dst0[3], t1->rewrite + 64 , 
-                    _vec_len (t1->rewrite)-64);
-
-          /* fix the <bleep>ing outer-IP checksum */
-          sum0 = ip0->checksum;
-          /* old_l0 always 0, see the rewrite setup */
-          new_l0 = 
-            clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
-          
-          sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t,
-                                 length /* changed member */);
-          ip0->checksum = ip_csum_fold (sum0);
-          ip0->length = new_l0;
-
-          sum1 = ip1->checksum;
-          /* old_l1 always 0, see the rewrite setup */
-          new_l1 = 
-            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 */);
-          ip1->checksum = ip_csum_fold (sum1);
-          ip1->length = new_l1;
-          
-          /* Fix UDP length */
-          udp0 = (udp_header_t *)(ip0+1);
-          new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
-                                         - sizeof (*ip0));
-          udp1 = (udp_header_t *)(ip1+1);
-          new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1)
-                                         - sizeof (*ip1));
-
-          udp0->length = new_l0;
-          udp1->length = new_l1;
-
-          /* Reset to look up tunnel partner in the configured FIB */
-          vnet_buffer(b0)->sw_if_index[VLIB_TX] = t0->encap_fib_index;
-          vnet_buffer(b1)->sw_if_index[VLIB_TX] = t1->encap_fib_index;
-          vnet_buffer(b0)->sw_if_index[VLIB_RX] = sw_if_index0;
-          vnet_buffer(b1)->sw_if_index[VLIB_RX] = sw_if_index1;
-          pkts_encapsulated += 2;
-
-          len0 = vlib_buffer_length_in_chain(vm, b0);
-          len1 = vlib_buffer_length_in_chain(vm, b0);
-          stats_n_packets += 2;
-          stats_n_bytes += len0 + len1;
-
-          /* Batch stats increment on the same vxlan 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,
-                    cpu_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,
-                  cpu_index, sw_if_index0, 1, len0);
-              vlib_increment_combined_counter(
-                  im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
-                  cpu_index, sw_if_index1, 1, len1);
-            }
-          }
-
-          if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
-            {
-              vxlan_gpe_encap_trace_t *tr =
-                vlib_add_trace (vm, node, b0, sizeof (*tr));
-              tr->tunnel_index = t0 - ngm->tunnels;
-            }
-
-          if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED))
-            {
-              vxlan_gpe_encap_trace_t *tr =
-                vlib_add_trace (vm, node, b1, sizeof (*tr));
-              tr->tunnel_index = t1 - ngm->tunnels;
-            }
-
-	  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 = VXLAN_GPE_ENCAP_NEXT_IP4_LOOKUP;
-	  u32 sw_if_index0, len0;
-          vnet_hw_interface_t * hi0;
-          ip4_header_t * ip0;
-          udp_header_t * udp0;
-          u64 * copy_src0, * copy_dst0;
-          u32 * copy_src_last0, * copy_dst_last0;
-          vxlan_gpe_tunnel_t * t0;
-          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);
-
-          /* 1-wide cache? */
-	  sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX];
-          hi0 = vnet_get_sup_hw_interface
-            (vnm, vnet_buffer(b0)->sw_if_index[VLIB_TX]);
-
-          t0 = pool_elt_at_index (ngm->tunnels, hi0->dev_instance);
-
-          ASSERT(vec_len(t0->rewrite) >= 24);
-
-          /* Apply the rewrite string. $$$$ vnet_rewrite? */
-          vlib_buffer_advance (b0, -(word)_vec_len(t0->rewrite));
-
-          ip0 = vlib_buffer_get_current(b0);
-          /* Copy the fixed header */
-          copy_dst0 = (u64 *) ip0;
-          copy_src0 = (u64 *) t0->rewrite;
-
-          ASSERT (sizeof (ip4_vxlan_gpe_header_t) == 36);
-
-          /* Copy first 36 octets 8-bytes at a time */
-#define _(offs) copy_dst0[offs] = copy_src0[offs];
-          foreach_fixed_header_offset;
-#undef _
-          /* Last 4 octets. Hopefully gcc will be our friend */
-          copy_dst_last0 = (u32 *)(&copy_dst0[7]);
-          copy_src_last0 = (u32 *)(&copy_src0[7]);
-          
-          copy_dst_last0[0] = copy_src_last0[0];
-
-          /* If there are TLVs to copy, do so */
-          if (PREDICT_FALSE (_vec_len(t0->rewrite) > 64))
-            clib_memcpy (&copy_dst0[3], t0->rewrite + 64 , 
-                    _vec_len (t0->rewrite)-64);
-
-          /* fix the <bleep>ing outer-IP checksum */
-          sum0 = ip0->checksum;
-          /* old_l0 always 0, see the rewrite setup */
-          new_l0 = 
-            clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
-          
-          sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t,
-                                 length /* changed member */);
-          ip0->checksum = ip_csum_fold (sum0);
-          ip0->length = new_l0;
-
-          /* Fix UDP length */
-          udp0 = (udp_header_t *)(ip0+1);
-          new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
-                                         - sizeof (*ip0));
-
-          udp0->length = new_l0;
-
-          /* Reset to look up tunnel partner in the configured FIB */
-          vnet_buffer(b0)->sw_if_index[VLIB_TX] = t0->encap_fib_index;
-	  vnet_buffer(b0)->sw_if_index[VLIB_RX] = sw_if_index0;
-          pkts_encapsulated ++;
-
-	  len0 = vlib_buffer_length_in_chain(vm, b0); 
-	  stats_n_packets += 1; 
-	  stats_n_bytes += len0; 
-	  
-          /* Batch stats increment on the same vxlan 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, 
-						cpu_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))
-	    {
-	      vxlan_gpe_encap_trace_t *tr =
-		vlib_add_trace (vm, node, b0, sizeof (*tr));
-              tr->tunnel_index = t0 - ngm->tunnels;
-            }
-	  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);
+    while (n_left_from > 0 && n_left_to_next > 0)
+    {
+      u32 bi0;
+      vlib_buffer_t * b0;
+      u32 next0 = VXLAN_GPE_ENCAP_NEXT_IP4_LOOKUP;
+      u32 sw_if_index0, len0;
+      vnet_hw_interface_t * hi0;
+      vxlan_gpe_tunnel_t * t0;
+      u8 is_ip4_0;
+
+      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);
+
+      /* 1-wide cache? */
+      sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX];
+      hi0 = vnet_get_sup_hw_interface (vnm, vnet_buffer(b0)->sw_if_index[VLIB_TX]);
+
+      t0 = pool_elt_at_index(ngm->tunnels, hi0->dev_instance);
+
+      is_ip4_0 = (t0->flags & VXLAN_GPE_TUNNEL_IS_IPV4);
+
+      vxlan_gpe_encap_one_inline (ngm, b0, t0, &next0, is_ip4_0);
+
+      /* Reset to look up tunnel partner in the configured FIB */
+      vnet_buffer(b0)->sw_if_index[VLIB_TX] = t0->encap_fib_index;
+      vnet_buffer(b0)->sw_if_index[VLIB_RX] = sw_if_index0;
+      pkts_encapsulated++;
+
+      len0 = vlib_buffer_length_in_chain (vm, b0);
+      stats_n_packets += 1;
+      stats_n_bytes += len0;
+
+      /* Batch stats increment on the same vxlan 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,
+              cpu_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))
+      {
+        vxlan_gpe_encap_trace_t *tr = vlib_add_trace (vm, node, b0,
+                                                      sizeof(*tr));
+        tr->tunnel_index = t0 - ngm->tunnels;
+      }
+      vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next,
+                                      n_left_to_next, bi0, next0);
     }
+
+    vlib_put_next_frame (vm, node, next_index, n_left_to_next);
+  }
   vlib_node_increment_counter (vm, node->node_index,
                                VXLAN_GPE_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, cpu_index, 
-				    stats_sw_if_index, stats_n_packets, stats_n_bytes); 
-    node->runtime_data[0] = stats_sw_if_index; 
-  } 
+  if (stats_n_packets)
+  {
+    vlib_increment_combined_counter (
+        im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX, cpu_index,
+        stats_sw_if_index, stats_n_packets, stats_n_bytes);
+    node->runtime_data[0] = stats_sw_if_index;
+  }
 
   return from_frame->n_vectors;
 }
@@ -418,8 +340,9 @@ VLIB_REGISTER_NODE (vxlan_gpe_encap_node) = {
   .n_next_nodes = VXLAN_GPE_ENCAP_N_NEXT,
 
   .next_nodes = {
-        [VXLAN_GPE_ENCAP_NEXT_IP4_LOOKUP] = "ip4-lookup",
-        [VXLAN_GPE_ENCAP_NEXT_DROP] = "error-drop",
+    [VXLAN_GPE_ENCAP_NEXT_IP4_LOOKUP] = "ip4-lookup",
+		[VXLAN_GPE_ENCAP_NEXT_IP6_LOOKUP] = "ip6-lookup",
+    [VXLAN_GPE_ENCAP_NEXT_DROP] = "error-drop",
   },
 };
 
diff --git a/vnet/vnet/vxlan-gpe/vxlan_gpe.c b/vnet/vnet/vxlan-gpe/vxlan_gpe.c
index 13eda440ed1..bf1623af372 100644
--- a/vnet/vnet/vxlan-gpe/vxlan_gpe.c
+++ b/vnet/vnet/vxlan-gpe/vxlan_gpe.c
@@ -13,6 +13,7 @@
  * limitations under the License.
  */
 #include <vnet/vxlan-gpe/vxlan_gpe.h>
+#include <vnet/ip/format.h>
 
 vxlan_gpe_main_t vxlan_gpe_main;
 
@@ -23,8 +24,10 @@ u8 * format_vxlan_gpe_tunnel (u8 * s, va_list * args)
 
   s = format (s, "[%d] local: %U remote: %U ",
               t - gm->tunnels,
-              format_ip4_address, &t->local,
-              format_ip4_address, &t->remote);    
+              format_ip46_address, &t->local,
+              format_ip46_address, &t->remote);
+
+  s = format (s, "  vxlan VNI %d ", t->vni);
 
   switch (t->protocol)
     {
@@ -48,8 +51,6 @@ u8 * format_vxlan_gpe_tunnel (u8 * s, va_list * args)
               t->encap_fib_index,
               t->decap_fib_index);
 
-  s = format (s, "  vxlan VNI %d ", t->vni);
-
   return s;
 }
 
@@ -121,24 +122,26 @@ VNET_HW_INTERFACE_CLASS (vxlan_gpe_hw_class) = {
 
 
 #define foreach_gpe_copy_field                  \
-_(local.as_u32)                                 \
-_(remote.as_u32)				\
 _(vni)                                          \
 _(protocol)                                \
 _(encap_fib_index)                              \
-_(decap_fib_index)                              
-
-#define foreach_copy_field			\
-_(src.as_u32)                                   \
-_(dst.as_u32)                                   \
-_(vni)                                          \
-_(encap_fib_index)                              \
 _(decap_fib_index)                              \
 _(decap_next_index)
 
+#define foreach_copy_ipv4 {                     \
+  _(local.ip4.as_u32)                           \
+  _(remote.ip4.as_u32)                          \
+}
 
+#define foreach_copy_ipv6 {                     \
+  _(local.ip6.as_u64[0])                        \
+  _(local.ip6.as_u64[1])                        \
+  _(remote.ip6.as_u64[0])                       \
+  _(remote.ip6.as_u64[1])                       \
+}
 
-static int vxlan_gpe_rewrite (vxlan_gpe_tunnel_t * t)
+
+static int vxlan4_gpe_rewrite (vxlan_gpe_tunnel_t * t)
 {
   u8 *rw = 0;
   ip4_header_t * ip0;
@@ -158,8 +161,8 @@ static int vxlan_gpe_rewrite (vxlan_gpe_tunnel_t * t)
   ip0->protocol = IP_PROTOCOL_UDP;
 
   /* we fix up the ip4 header length and checksum after-the-fact */
-  ip0->src_address.as_u32 = t->local.as_u32; 
-  ip0->dst_address.as_u32 = t->remote.as_u32; 
+  ip0->src_address.as_u32 = t->local.ip4.as_u32;
+  ip0->dst_address.as_u32 = t->remote.ip4.as_u32;
   ip0->checksum = ip4_header_checksum (ip0);
 
   /* UDP header, randomize src port on something, maybe? */
@@ -176,6 +179,44 @@ static int vxlan_gpe_rewrite (vxlan_gpe_tunnel_t * t)
   return (0);
 }
 
+static int vxlan6_gpe_rewrite (vxlan_gpe_tunnel_t * t)
+{
+  u8 *rw = 0;
+  ip6_header_t * ip0;
+  ip6_vxlan_gpe_header_t * h0;
+  int len;
+
+  len = sizeof (*h0);
+
+  vec_validate_aligned (rw, len-1, CLIB_CACHE_LINE_BYTES);
+
+  h0 = (ip6_vxlan_gpe_header_t *) rw;
+
+  /* Fixed portion of the (outer) ip4 header */
+  ip0 = &h0->ip6;
+  ip0->ip_version_traffic_class_and_flow_label = clib_host_to_net_u32(6 << 28);
+  ip0->hop_limit = 255;
+  ip0->protocol = IP_PROTOCOL_UDP;
+
+  ip0->src_address.as_u64[0] = t->local.ip6.as_u64[0];
+  ip0->src_address.as_u64[1] = t->local.ip6.as_u64[1];
+  ip0->dst_address.as_u64[0] = t->remote.ip6.as_u64[0];
+  ip0->dst_address.as_u64[1] = t->remote.ip6.as_u64[1];
+
+  /* UDP header, randomize src port on something, maybe? */
+  h0->udp.src_port = clib_host_to_net_u16 (4790);
+  h0->udp.dst_port = clib_host_to_net_u16 (UDP_DST_PORT_vxlan_gpe);
+
+  /* VXLAN header. Are we having fun yet? */
+  h0->vxlan.flags = VXLAN_GPE_FLAGS_I | VXLAN_GPE_FLAGS_P;
+  h0->vxlan.ver_res = VXLAN_GPE_VERSION;
+  h0->vxlan.protocol = VXLAN_GPE_PROTOCOL_IP4;
+  h0->vxlan.vni_res = clib_host_to_net_u32 (t->vni<<8);
+
+  t->rewrite = rw;
+  return (0);
+}
+
 int vnet_vxlan_gpe_add_del_tunnel 
 (vnet_vxlan_gpe_add_del_tunnel_args_t *a, u32 * sw_if_indexp)
 {
@@ -187,15 +228,29 @@ int vnet_vxlan_gpe_add_del_tunnel
   u32 hw_if_index = ~0;
   u32 sw_if_index = ~0;
   int rv;
-  vxlan_gpe_tunnel_key_t key, *key_copy;
+  vxlan4_gpe_tunnel_key_t key4, *key4_copy;
+  vxlan6_gpe_tunnel_key_t key6, *key6_copy;
   hash_pair_t *hp;
   
-  key.local = a->local.as_u32; 
-  key.remote = a->remote.as_u32; 
-  key.vni = clib_host_to_net_u32 (a->vni << 8);
-  key.pad = 0;
-
-  p = hash_get_mem (gm->vxlan_gpe_tunnel_by_key, &key);
+  if (!a->is_ip6)
+  {
+    key4.local = a->local.ip4.as_u32;
+    key4.remote = a->remote.ip4.as_u32;
+    key4.vni = clib_host_to_net_u32 (a->vni << 8);
+    key4.pad = 0;
+
+    p = hash_get_mem(gm->vxlan4_gpe_tunnel_by_key, &key4);
+  }
+  else
+  {
+    key6.local.as_u64[0] = a->local.ip6.as_u64[0];
+    key6.local.as_u64[1] = a->local.ip6.as_u64[1];
+    key6.remote.as_u64[0] = a->remote.ip6.as_u64[0];
+    key6.remote.as_u64[1] = a->remote.ip6.as_u64[1];
+    key6.vni = clib_host_to_net_u32 (a->vni << 8);
+
+    p = hash_get(gm->vxlan6_gpe_tunnel_by_key, &key6);
+  }
   
   if (a->is_add)
     {
@@ -212,21 +267,43 @@ int vnet_vxlan_gpe_add_del_tunnel
       /* copy from arg structure */
 #define _(x) t->x = a->x;
       foreach_gpe_copy_field;
+      if (!a->is_ip6) foreach_copy_ipv4
+      else            foreach_copy_ipv6
 #undef _
 
-      rv = vxlan_gpe_rewrite (t);
+      if (a->is_ip6) {
+        /* copy the key */
+        t->key6 = key6;
+      }
+
+      if (!a->is_ip6) t->flags |= VXLAN_GPE_TUNNEL_IS_IPV4;
+
+      if (!a->is_ip6) {
+        rv = vxlan4_gpe_rewrite (t);
+      } else {
+        rv = vxlan6_gpe_rewrite (t);
+      }
 
       if (rv)
-        {
+      {
           pool_put (gm->tunnels, t);
           return rv;
-        }
-
-      key_copy = clib_mem_alloc (sizeof (*key_copy));
-      clib_memcpy (key_copy, &key, sizeof (*key_copy));
+      }
 
-      hash_set_mem (gm->vxlan_gpe_tunnel_by_key, key_copy, 
-                    t - gm->tunnels);
+      if (!a->is_ip6)
+      {
+        key4_copy = clib_mem_alloc (sizeof (*key4_copy));
+        clib_memcpy (key4_copy, &key4, sizeof (*key4_copy));
+        hash_set_mem (gm->vxlan4_gpe_tunnel_by_key, key4_copy,
+                      t - gm->tunnels);
+      }
+      else
+      {
+          key6_copy = clib_mem_alloc (sizeof (*key6_copy));
+          clib_memcpy (key6_copy, &key4, sizeof (*key6_copy));
+          hash_set_mem (gm->vxlan4_gpe_tunnel_by_key, key6_copy,
+                        t - gm->tunnels);
+      }
       
       if (vec_len (gm->free_vxlan_gpe_tunnel_hw_if_indices) > 0)
         {
@@ -264,10 +341,20 @@ int vnet_vxlan_gpe_add_del_tunnel
       vnet_sw_interface_set_flags (vnm, t->sw_if_index, 0 /* down */);
       vec_add1 (gm->free_vxlan_gpe_tunnel_hw_if_indices, t->hw_if_index);
 
-      hp = hash_get_pair (gm->vxlan_gpe_tunnel_by_key, &key);
-      key_copy = (void *)(hp->key);
-      hash_unset_mem (gm->vxlan_gpe_tunnel_by_key, &key);
-      clib_mem_free (key_copy);
+      if (!a->is_ip6)
+      {
+        hp = hash_get_pair (gm->vxlan4_gpe_tunnel_by_key, &key4);
+        key4_copy = (void *)(hp->key);
+        hash_unset_mem (gm->vxlan4_gpe_tunnel_by_key, &key4);
+        clib_mem_free (key4_copy);
+      }
+      else
+      {
+        hp = hash_get_pair (gm->vxlan6_gpe_tunnel_by_key, &key6);
+        key6_copy = (void *)(hp->key);
+        hash_unset_mem (gm->vxlan4_gpe_tunnel_by_key, &key6);
+        clib_mem_free (key6_copy);
+      }
 
       vec_free (t->rewrite);
       pool_put (gm->tunnels, t);
@@ -279,7 +366,7 @@ int vnet_vxlan_gpe_add_del_tunnel
   return 0;
 }
 
-static u32 fib_index_from_fib_id (u32 fib_id)
+static u32 fib4_index_from_fib_id (u32 fib_id)
 {
   ip4_main_t * im = &ip4_main;
   uword * p;
@@ -291,6 +378,18 @@ static u32 fib_index_from_fib_id (u32 fib_id)
   return p[0];
 }
 
+static u32 fib6_index_from_fib_id (u32 fib_id)
+{
+  ip6_main_t * im = &ip6_main;
+  uword * p;
+
+  p = hash_get (im->fib_index_by_table_id, fib_id);
+  if (!p)
+    return ~0;
+
+  return p[0];
+}
+
 static uword unformat_gpe_decap_next (unformat_input_t * input, va_list * args)
 {
   u32 * result = va_arg (*args, u32 *);
@@ -318,9 +417,11 @@ vxlan_gpe_add_del_tunnel_command_fn (vlib_main_t * vm,
 {
   unformat_input_t _line_input, * line_input = &_line_input;
   u8 is_add = 1;
-  ip4_address_t local, remote;
+  ip46_address_t local, remote;
   u8 local_set = 0;
   u8 remote_set = 0;
+  u8 ipv4_set = 0;
+  u8 ipv6_set = 0;
   u32 encap_fib_index = 0;
   u32 decap_fib_index = 0;
   u8 protocol = VXLAN_GPE_PROTOCOL_IP4;
@@ -340,20 +441,47 @@ vxlan_gpe_add_del_tunnel_command_fn (vlib_main_t * vm,
     if (unformat (line_input, "del"))
       is_add = 0;
     else if (unformat (line_input, "local %U", 
-                       unformat_ip4_address, &local))
+                       unformat_ip4_address, &local.ip4))
+    {
       local_set = 1;
+      ipv4_set = 1;
+    }
+    else if (unformat (line_input, "remote %U",
+                       unformat_ip4_address, &remote.ip4))
+    {
+      remote_set = 1;
+      ipv4_set = 1;
+    }
+    else if (unformat (line_input, "local %U",
+                       unformat_ip6_address, &local.ip6))
+    {
+      local_set = 1;
+      ipv6_set = 1;
+    }
     else if (unformat (line_input, "remote %U",
-                       unformat_ip4_address, &remote))
+                       unformat_ip6_address, &remote.ip6))
+    {
       remote_set = 1;
+      ipv6_set = 1;
+    }
     else if (unformat (line_input, "encap-vrf-id %d", &tmp))
       {
-        encap_fib_index = fib_index_from_fib_id (tmp);
+        if (ipv6_set)
+          encap_fib_index = fib6_index_from_fib_id (tmp);
+        else
+          encap_fib_index = fib4_index_from_fib_id (tmp);
+
         if (encap_fib_index == ~0)
           return clib_error_return (0, "nonexistent encap fib id %d", tmp);
       }
     else if (unformat (line_input, "decap-vrf-id %d", &tmp))
       {
-        decap_fib_index = fib_index_from_fib_id (tmp);
+
+        if (ipv6_set)
+          decap_fib_index = fib6_index_from_fib_id (tmp);
+        else
+          decap_fib_index = fib4_index_from_fib_id (tmp);
+
         if (decap_fib_index == ~0)
           return clib_error_return (0, "nonexistent decap fib id %d", tmp);
       }
@@ -383,15 +511,25 @@ vxlan_gpe_add_del_tunnel_command_fn (vlib_main_t * vm,
   if (remote_set == 0)
     return clib_error_return (0, "tunnel remote address not specified");
 
+  if (ipv4_set && ipv6_set)
+    return clib_error_return (0, "both IPv4 and IPv6 addresses specified");
+
+  if ((ipv4_set && memcmp(&local.ip4, &remote.ip4, sizeof(local.ip4)) == 0) ||
+      (ipv6_set && memcmp(&local.ip6, &remote.ip6, sizeof(local.ip6)) == 0))
+    return clib_error_return (0, "src and dst addresses are identical");
+
   if (vni_set == 0)
     return clib_error_return (0, "vni not specified");
 
   memset (a, 0, sizeof (*a));
 
   a->is_add = is_add;
+  a->is_ip6 = ipv6_set;
 
 #define _(x) a->x = x;
   foreach_gpe_copy_field;
+  if (ipv4_set) foreach_copy_ipv4
+  else          foreach_copy_ipv6
 #undef _
 
   rv = vnet_vxlan_gpe_add_del_tunnel (a, &sw_if_index);
@@ -421,7 +559,7 @@ vxlan_gpe_add_del_tunnel_command_fn (vlib_main_t * vm,
 VLIB_CLI_COMMAND (create_vxlan_gpe_tunnel_command, static) = {
   .path = "create vxlan-gpe tunnel",
   .short_help = 
-  "create vxlan-gpe tunnel local <ip4-addr> remote <ip4-addr>"
+  "create vxlan-gpe tunnel local <local-addr> remote <remote-addr>"
   " vni <nn> [next-ip4][next-ip6][next-ethernet][next-nsh]"
   " [encap-vrf-id <nn>] [decap-vrf-id <nn>]"
   " [del]\n",
@@ -458,12 +596,18 @@ clib_error_t *vxlan_gpe_init (vlib_main_t *vm)
   
   gm->vnet_main = vnet_get_main();
   gm->vlib_main = vm;
-  
-  gm->vxlan_gpe_tunnel_by_key 
-    = hash_create_mem (0, sizeof(vxlan_gpe_tunnel_key_t), sizeof (uword));
 
-  udp_register_dst_port (vm, UDP_DST_PORT_vxlan_gpe, 
-                         vxlan_gpe_input_node.index, 1 /* is_ip4 */);
+  gm->vxlan4_gpe_tunnel_by_key
+    = hash_create_mem (0, sizeof(vxlan4_gpe_tunnel_key_t), sizeof (uword));
+
+  gm->vxlan6_gpe_tunnel_by_key
+    = hash_create_mem (0, sizeof(vxlan6_gpe_tunnel_key_t), sizeof (uword));
+
+
+  udp_register_dst_port (vm, UDP_DST_PORT_vxlan_gpe,
+                         vxlan4_gpe_input_node.index, 1 /* is_ip4 */);
+  udp_register_dst_port (vm, UDP_DST_PORT_vxlan6_gpe,
+                         vxlan6_gpe_input_node.index, 0 /* is_ip4 */);
   return 0;
 }
 
diff --git a/vnet/vnet/vxlan-gpe/vxlan_gpe.h b/vnet/vnet/vxlan-gpe/vxlan_gpe.h
index a51e0b31e3e..730abe70253 100644
--- a/vnet/vnet/vxlan-gpe/vxlan_gpe.h
+++ b/vnet/vnet/vxlan-gpe/vxlan_gpe.h
@@ -23,6 +23,7 @@
 #include <vnet/ethernet/ethernet.h>
 #include <vnet/vxlan-gpe/vxlan_gpe_packet.h>
 #include <vnet/ip/ip4_packet.h>
+#include <vnet/ip/ip6_packet.h>
 #include <vnet/ip/udp.h>
 
 
@@ -32,6 +33,12 @@ typedef CLIB_PACKED (struct {
   vxlan_gpe_header_t vxlan;     /* 8 bytes */
 }) ip4_vxlan_gpe_header_t;
 
+typedef CLIB_PACKED (struct {
+  ip6_header_t ip6;             /* 40 bytes */
+  udp_header_t udp;             /* 8 bytes */
+  vxlan_gpe_header_t vxlan;     /* 8 bytes */
+}) ip6_vxlan_gpe_header_t;
+
 typedef CLIB_PACKED(struct {
   /* 
    * Key fields: local remote, vni 
@@ -46,7 +53,17 @@ typedef CLIB_PACKED(struct {
     };
     u64 as_u64[2];
   };
-}) vxlan_gpe_tunnel_key_t;
+}) vxlan4_gpe_tunnel_key_t;
+
+typedef CLIB_PACKED(struct {
+  /*
+   * Key fields: local remote, vni
+   * all fields in NET byte order
+   */
+  ip6_address_t local;
+  ip6_address_t remote;
+  u32 vni;                      /* shifted 8 bits */
+}) vxlan6_gpe_tunnel_key_t;
 
 typedef struct {
   /* Rewrite string. $$$$ embed vnet_rewrite header */
@@ -56,8 +73,8 @@ typedef struct {
   u8 protocol;
 
   /* tunnel src and dst addresses */
-  ip4_address_t local;
-  ip4_address_t remote;
+  ip46_address_t local;
+  ip46_address_t remote;
 
   /* FIB indices */
   u32 encap_fib_index;          /* tunnel partner lookup here */
@@ -66,12 +83,25 @@ typedef struct {
   /* vxlan VNI in HOST byte order, shifted left 8 bits */
   u32 vni;
 
+  /*decap next index*/
+  u32 decap_next_index;
+
   /* vnet intfc hw/sw_if_index */
   u32 hw_if_index;
   u32 sw_if_index;
 
+  union { /* storage for the hash key */
+	vxlan4_gpe_tunnel_key_t key4;
+	vxlan6_gpe_tunnel_key_t key6;
+  };
+
+  /* flags */
+  u32 flags;
 } vxlan_gpe_tunnel_t;
 
+/* Flags for vxlan_gpe_tunnel_t.flags */
+#define VXLAN_GPE_TUNNEL_IS_IPV4	1
+
 #define foreach_vxlan_gpe_input_next        \
 _(DROP, "error-drop")                           \
 _(IP4_INPUT, "ip4-input")                       \
@@ -97,7 +127,8 @@ typedef struct {
   vxlan_gpe_tunnel_t *tunnels;
 
   /* lookup tunnel by key */
-  uword * vxlan_gpe_tunnel_by_key;
+  uword * vxlan4_gpe_tunnel_by_key;
+  uword * vxlan6_gpe_tunnel_by_key;
 
   /* Free vlib hw_if_indices */
   u32 * free_vxlan_gpe_tunnel_hw_if_indices;
@@ -113,13 +144,15 @@ typedef struct {
 vxlan_gpe_main_t vxlan_gpe_main;
 
 extern vlib_node_registration_t vxlan_gpe_encap_node;
-extern vlib_node_registration_t vxlan_gpe_input_node;
+extern vlib_node_registration_t vxlan4_gpe_input_node;
+extern vlib_node_registration_t vxlan6_gpe_input_node;
 
 u8 * format_vxlan_gpe_encap_trace (u8 * s, va_list * args);
 
 typedef struct {
   u8 is_add;
-  ip4_address_t local, remote;
+  u8 is_ip6;
+  ip46_address_t local, remote;
   u8 protocol;
   u32 encap_fib_index;
   u32 decap_fib_index;
diff --git a/vpp-api-test/vat/api_format.c b/vpp-api-test/vat/api_format.c
index 3fc56d700e6..59da077e053 100644
--- a/vpp-api-test/vat/api_format.c
+++ b/vpp-api-test/vat/api_format.c
@@ -8062,8 +8062,10 @@ static int api_vxlan_gpe_add_del_tunnel (vat_main_t * vam)
     unformat_input_t * line_input = vam->input;
     vl_api_vxlan_gpe_add_del_tunnel_t *mp;
     f64 timeout;
-    ip4_address_t local, remote;
+    ip4_address_t local4, remote4;
+    ip6_address_t local6, remote6;
     u8 is_add = 1;
+    u8 ipv4_set = 0, ipv6_set = 0;
     u8 local_set = 0;
     u8 remote_set = 0;
     u32 encap_vrf_id = 0;
@@ -8076,11 +8078,29 @@ static int api_vxlan_gpe_add_del_tunnel (vat_main_t * vam)
         if (unformat (line_input, "del"))
             is_add = 0;
         else if (unformat (line_input, "local %U", 
-                           unformat_ip4_address, &local))
+                           unformat_ip4_address, &local4))
+        {
+            local_set = 1;
+            ipv4_set = 1;
+        }
+        else if (unformat (line_input, "remote %U",
+                           unformat_ip4_address, &remote4))
+        {
+            remote_set = 1;
+            ipv4_set = 1;
+        }
+        else if (unformat (line_input, "local %U",
+                           unformat_ip6_address, &local6))
+        {
             local_set = 1;
+            ipv6_set = 1;
+        }
         else if (unformat (line_input, "remote %U",
-                           unformat_ip4_address, &remote))
+                           unformat_ip6_address, &remote6))
+        {
             remote_set = 1;
+            ipv6_set = 1;
+        }
         else if (unformat (line_input, "encap-vrf-id %d", &encap_vrf_id))
             ;
         else if (unformat (line_input, "decap-vrf-id %d", &decap_vrf_id))
@@ -8109,6 +8129,10 @@ static int api_vxlan_gpe_add_del_tunnel (vat_main_t * vam)
         errmsg ("tunnel remote address not specified\n");
         return -99;
     }
+    if (ipv4_set && ipv6_set) {
+        errmsg ("both IPv4 and IPv6 addresses specified");
+        return -99;
+    }
 
     if (vni_set == 0) {
         errmsg ("vni not specified\n");
@@ -8117,14 +8141,21 @@ static int api_vxlan_gpe_add_del_tunnel (vat_main_t * vam)
 
     M(VXLAN_GPE_ADD_DEL_TUNNEL, vxlan_gpe_add_del_tunnel);
     
-    mp->local = local.as_u32;
-    mp->remote = remote.as_u32;
+
+    if (ipv6_set) {
+        clib_memcpy(&mp->local, &local6, sizeof(local6));
+        clib_memcpy(&mp->remote, &remote6, sizeof(remote6));
+    } else {
+        clib_memcpy(&mp->local, &local4, sizeof(local4));
+        clib_memcpy(&mp->remote, &remote4, sizeof(remote4));
+    }
+
     mp->encap_vrf_id = ntohl(encap_vrf_id);
     mp->decap_vrf_id = ntohl(decap_vrf_id);
     mp->protocol = ntohl(protocol);
     mp->vni = ntohl(vni);
     mp->is_add = is_add;
-
+    mp->is_ipv6 = ipv6_set;
 
     S; W;
     /* NOTREACHED */
diff --git a/vpp/api/api.c b/vpp/api/api.c
index 3bf2ed47088..8b801e5a5c0 100644
--- a/vpp/api/api.c
+++ b/vpp/api/api.c
@@ -4568,12 +4568,24 @@ vl_api_vxlan_gpe_add_del_tunnel_t_handler
         decap_fib_index = ntohl(mp->decap_vrf_id);
     }
 
+    /* Check src & dst are different */
+    if ((a->is_ip6 && memcmp(mp->local, mp->remote, 16) == 0) ||
+       (!a->is_ip6 && memcmp(mp->local, mp->remote, 4) == 0)) {
+        rv = VNET_API_ERROR_SAME_SRC_DST;
+        goto out;
+    }
     memset (a, 0, sizeof (*a));
 
     a->is_add = mp->is_add;
+    a->is_ip6 = mp->is_ipv6;
     /* ip addresses sent in network byte order */
-    a->local.as_u32 = ntohl(mp->local);
-    a->remote.as_u32 = ntohl(mp->remote);
+    if (a->is_ip6) {
+      clib_memcpy(&(a->local.ip6), mp->local, 16);
+      clib_memcpy(&(a->remote.ip6), mp->remote, 16);
+    } else {
+      clib_memcpy(&(a->local.ip4), mp->local, 4);
+      clib_memcpy(&(a->remote.ip4), mp->remote, 4);
+    }
     a->encap_fib_index = encap_fib_index;
     a->decap_fib_index = decap_fib_index;
     a->protocol = protocol;
diff --git a/vpp/api/custom_dump.c b/vpp/api/custom_dump.c
index 47c9f662517..ba4cf538d2f 100644
--- a/vpp/api/custom_dump.c
+++ b/vpp/api/custom_dump.c
@@ -1580,8 +1580,9 @@ static void *vl_api_vxlan_gpe_add_del_tunnel_t_print
 
     s = format (0, "SCRIPT: vxlan_gpe_add_del_tunnel ");
 
-    s = format (s, "local %U remote %U ", format_ip4_address, &mp->local,
-                format_ip4_address, &mp->remote);
+    s = format (s, "local %U ", format_ip46_address, &mp->local, mp->is_ipv6);
+
+    s = format (s, "remote %U ", format_ip46_address, &mp->remote, mp->is_ipv6);
 
     s = format (s, "protocol %d ", ntohl(mp->protocol));
 
diff --git a/vpp/api/vpe.api b/vpp/api/vpe.api
index 2a827d12188..9c90c338e73 100644
--- a/vpp/api/vpe.api
+++ b/vpp/api/vpe.api
@@ -2113,8 +2113,9 @@ manual_java define l2_fib_table_dump {
 define vxlan_gpe_add_del_tunnel {
     u32 client_index;
     u32 context;
-    u32 local;
-    u32 remote;
+    u8 is_ipv6;
+    u8 local[16];
+    u8 remote[16];
     u32 encap_vrf_id;
     u32 decap_vrf_id;
     u8 protocol;