reassembly: feature/concurrency

This change makes ip reassembly an interface feature, while adding
concurrency support. Due to this, punt is no longer needed to test
reassembly.

Change-Id: I467669514ec33283ce935be0f1dd08f07684f0c7
Signed-off-by: Klement Sekera <ksekera@cisco.com>
diff --git a/src/vnet/buffer.h b/src/vnet/buffer.h
index e785145..02688e2 100644
--- a/src/vnet/buffer.h
+++ b/src/vnet/buffer.h
@@ -161,15 +161,24 @@
 	} icmp;
 
 	/* reassembly */
-	struct
+	union
 	{
-	  u16 fragment_first;
-	  u16 fragment_last;
-	  u16 range_first;
-	  u16 range_last;
-	  u32 next_range_bi;
-	  u16 ip6_frag_hdr_offset;
-	  u16 estimated_mtu;
+	  /* in/out variables */
+	  struct
+	  {
+	    u32 next_index;	/* index of next node - ignored if "feature" node */
+	    u16 estimated_mtu;	/* estimated MTU calculated during reassembly */
+	  };
+	  /* internal variables used during reassembly */
+	  struct
+	  {
+	    u16 fragment_first;
+	    u16 fragment_last;
+	    u16 range_first;
+	    u16 range_last;
+	    u32 next_range_bi;
+	    u16 ip6_frag_hdr_offset;
+	  };
 	} reass;
       };
 
diff --git a/src/vnet/ip/ip.api b/src/vnet/ip/ip.api
index d6c5f6b..2ca2f56 100644
--- a/src/vnet/ip/ip.api
+++ b/src/vnet/ip/ip.api
@@ -910,6 +910,22 @@
   u8 is_ip6;
 };
 
+/** \brief Enable/disable reassembly feature
+    @param client_index - opaque cookie to identify the sender
+    @param context - sender context, to match reply w/ request
+    @param sw_if_index - interface to enable/disable feature on
+    @param enable_ip4 - enable ip4 reassembly if non-zero, disable if 0
+    @param enable_ip6 - enable ip6 reassembly if non-zero, disable if 0
+*/
+autoreply define ip_reassembly_enable_disable
+{
+  u32 client_index;
+  u32 context;
+  u32 sw_if_index;
+  u8 enable_ip4;
+  u8 enable_ip6;
+};
+
 /*
  * Local Variables:
  * eval: (c-set-style "gnu")
diff --git a/src/vnet/ip/ip46_cli.c b/src/vnet/ip/ip46_cli.c
index 668c650..2508f0e 100644
--- a/src/vnet/ip/ip46_cli.c
+++ b/src/vnet/ip/ip46_cli.c
@@ -38,6 +38,8 @@
  */
 
 #include <vnet/ip/ip.h>
+#include <vnet/ip/ip4_reassembly.h>
+#include <vnet/ip/ip6_reassembly.h>
 
 /**
  * @file
@@ -218,6 +220,87 @@
 };
 /* *INDENT-ON* */
 
+static clib_error_t *
+set_reassembly_command_fn (vlib_main_t * vm,
+			   unformat_input_t * input, vlib_cli_command_t * cmd)
+{
+  vnet_main_t *vnm = vnet_get_main ();
+  unformat_input_t _line_input, *line_input = &_line_input;
+  u32 sw_if_index = ~0;
+  u8 ip4_on = 0;
+  u8 ip6_on = 0;
+
+  /* Get a line of input. */
+  if (!unformat_user (input, unformat_line_input, line_input))
+    {
+      return NULL;
+    }
+
+  if (!unformat (input, "%U", unformat_vnet_sw_interface, vnm, &sw_if_index))
+    {
+      return clib_error_return (0, "Invalid interface name");
+    }
+
+  if (unformat (input, "on"))
+    {
+      ip4_on = 1;
+      ip6_on = 1;
+    }
+  else if (unformat (input, "off"))
+    {
+      ip4_on = 0;
+      ip6_on = 0;
+    }
+  else if (unformat (input, "ip4"))
+    {
+      ip4_on = 1;
+      ip6_on = 0;
+    }
+  else if (unformat (input, "ip6"))
+    {
+      ip4_on = 0;
+      ip6_on = 1;
+    }
+  else
+    {
+      return clib_error_return (0, "Unknown input `%U'",
+				format_unformat_error, line_input);
+    }
+
+
+  vnet_api_error_t rv4 = ip4_reass_enable_disable (sw_if_index, ip4_on);
+  vnet_api_error_t rv6 = ip6_reass_enable_disable (sw_if_index, ip6_on);
+  if (rv4 && rv6)
+    {
+      return clib_error_return (0,
+				"`ip4_reass_enable_disable' API call failed, rv=%d:%U, "
+				"`ip6_reass_enable_disable' API call failed, rv=%d:%U",
+				(int) rv4, format_vnet_api_errno, rv4,
+				(int) rv6, format_vnet_api_errno, rv6);
+    }
+  else if (rv4)
+    {
+      return clib_error_return (0,
+				"`ip4_reass_enable_disable' API call failed, rv=%d:%U",
+				(int) rv4, format_vnet_api_errno, rv4);
+    }
+  else if (rv6)
+    {
+      return clib_error_return (0,
+				"`ip6_reass_enable_disable' API call failed, rv=%d:%U",
+				(int) rv6, format_vnet_api_errno, rv6);
+    }
+  return NULL;
+}
+
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (set_reassembly_command, static) = {
+    .path = "set interface reassembly",
+    .short_help = "set interface reassembly <interface-name> [on|off|ip4|ip6]",
+    .function = set_reassembly_command_fn,
+};
+/* *INDENT-ON* */
+
 /* Dummy init function to get us linked in. */
 static clib_error_t *
 ip4_cli_init (vlib_main_t * vm)
diff --git a/src/vnet/ip/ip4_forward.c b/src/vnet/ip/ip4_forward.c
index 19babce..7c56a29 100644
--- a/src/vnet/ip/ip4_forward.c
+++ b/src/vnet/ip/ip4_forward.c
@@ -1217,14 +1217,8 @@
 	  sw_if_index0 = vnet_buffer (p0)->sw_if_index[VLIB_RX];
 	  sw_if_index1 = vnet_buffer (p1)->sw_if_index[VLIB_RX];
 
-	  /* Treat IP frag packets as "experimental" protocol for now
-	     until support of IP frag reassembly is implemented */
-	  proto0 =
-	    ip4_is_fragment (ip0) ? IP_PROTOCOL_VPP_FRAGMENTATION :
-	    ip0->protocol;
-	  proto1 =
-	    ip4_is_fragment (ip1) ? IP_PROTOCOL_VPP_FRAGMENTATION :
-	    ip1->protocol;
+	  proto0 = ip0->protocol;
+	  proto1 = ip1->protocol;
 
 	  if (head_of_feature_arc == 0)
 	    goto skip_checks;
@@ -1387,11 +1381,7 @@
 	  vnet_buffer (p0)->l3_hdr_offset = p0->current_data;
 	  sw_if_index0 = vnet_buffer (p0)->sw_if_index[VLIB_RX];
 
-	  /* Treat IP frag packets as "experimental" protocol for now
-	     until support of IP frag reassembly is implemented */
-	  proto0 =
-	    ip4_is_fragment (ip0) ? IP_PROTOCOL_VPP_FRAGMENTATION :
-	    ip0->protocol;
+	  proto0 = ip0->protocol;
 
 	  if (head_of_feature_arc == 0 || p0->flags & VNET_BUFFER_F_IS_NATED)
 	    goto skip_check;
@@ -1482,7 +1472,6 @@
     [IP_LOCAL_NEXT_PUNT] = "ip4-punt",
     [IP_LOCAL_NEXT_UDP_LOOKUP] = "ip4-udp-lookup",
     [IP_LOCAL_NEXT_ICMP] = "ip4-icmp-input",
-    [IP_LOCAL_NEXT_REASSEMBLY] = "ip4-reassembly",
   },
 };
 /* *INDENT-ON* */
diff --git a/src/vnet/ip/ip4_reassembly.c b/src/vnet/ip/ip4_reassembly.c
index ac3db16..114d8e0 100644
--- a/src/vnet/ip/ip4_reassembly.c
+++ b/src/vnet/ip/ip4_reassembly.c
@@ -29,7 +29,7 @@
 #define MSEC_PER_SEC 1000
 #define IP4_REASS_TIMEOUT_DEFAULT_MS 100
 #define IP4_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS 10000	// 10 seconds default
-#define IP4_REASS_MAX_REASSEMBLIES_DEAFULT 1024
+#define IP4_REASS_MAX_REASSEMBLIES_DEFAULT 1024
 #define IP4_REASS_HT_LOAD_FACTOR (0.75)
 
 #define IP4_REASS_DEBUG_BUFFERS 0
@@ -111,11 +111,10 @@
 {
   // hash table key
   ip4_reass_key_t key;
-  f64 first_heard;
   // time when last packet was received
   f64 last_heard;
   // internal id of this reassembly
-  u32 id;
+  u64 id;
   // buffer index of first buffer in this reassembly context
   u32 first_bi;
   // last octet of packet, ~0 until fragment without more_fragments arrives
@@ -124,10 +123,23 @@
   u32 data_len;
   // trace operation counter
   u32 trace_op_counter;
+  // next index - used by non-feature node
+  u8 next_index;
+  // minimum fragment length for this reassembly - used to estimate MTU
+  u16 min_fragment_length;
 } ip4_reass_t;
 
 typedef struct
 {
+  ip4_reass_t *pool;
+  u32 reass_n;
+  u32 buffers_n;
+  u32 id_counter;
+  clib_spinlock_t lock;
+} ip4_reass_per_thread_t;
+
+typedef struct
+{
   // IPv4 config
   u32 timeout_ms;
   f64 timeout;
@@ -135,11 +147,9 @@
   u32 max_reass_n;
 
   // IPv4 runtime
-  ip4_reass_t *pool;
   clib_bihash_24_8_t hash;
-  u32 reass_n;
-  u32 id_counter;
-  u32 buffers_n;
+  // per-thread data
+  ip4_reass_per_thread_t *per_thread_data;
 
   // convenience
   vlib_main_t *vlib_main;
@@ -148,7 +158,6 @@
   // node index of ip4-drop node
   u32 ip4_drop_idx;
   u32 ip4_reass_expire_node_idx;
-
 } ip4_reass_main_t;
 
 ip4_reass_main_t ip4_reass_main;
@@ -182,7 +191,6 @@
 typedef struct
 {
   ip4_reass_trace_operation_e action;
-  u32 pool_index;
   u32 reass_id;
   ip4_reass_range_trace_t trace_range;
   u32 size_diff;
@@ -192,7 +200,7 @@
   u32 total_data_len;
 } ip4_reass_trace_t;
 
-void
+static void
 ip4_reass_trace_details (vlib_main_t * vm, u32 bi,
 			 ip4_reass_range_trace_t * trace)
 {
@@ -205,7 +213,7 @@
   trace->range_bi = bi;
 }
 
-u8 *
+static u8 *
 format_ip4_reass_range_trace (u8 * s, va_list * args)
 {
   ip4_reass_range_trace_t *trace = va_arg (*args, ip4_reass_range_trace_t *);
@@ -252,15 +260,20 @@
   return s;
 }
 
-void
+static void
 ip4_reass_add_trace (vlib_main_t * vm, vlib_node_runtime_t * node,
 		     ip4_reass_main_t * rm, ip4_reass_t * reass, u32 bi,
 		     ip4_reass_trace_operation_e action, u32 size_diff)
 {
   vlib_buffer_t *b = vlib_get_buffer (vm, bi);
   vnet_buffer_opaque_t *vnb = vnet_buffer (b);
+  if (pool_is_free_index (vm->trace_main.trace_buffer_pool, b->trace_index))
+    {
+      // this buffer's trace is gone
+      b->flags &= ~VLIB_BUFFER_IS_TRACED;
+      return;
+    }
   ip4_reass_trace_t *t = vlib_add_trace (vm, node, b, sizeof (t[0]));
-  t->pool_index = reass - rm->pool;
   t->reass_id = reass->id;
   t->action = action;
   ip4_reass_trace_details (vm, bi, &t->trace_range);
@@ -280,19 +293,20 @@
 #endif
 }
 
-void
-ip4_reass_free (ip4_reass_main_t * rm, ip4_reass_t * reass)
+always_inline void
+ip4_reass_free (ip4_reass_main_t * rm, ip4_reass_per_thread_t * rt,
+		ip4_reass_t * reass)
 {
   clib_bihash_kv_24_8_t kv;
   kv.key[0] = reass->key.as_u64[0];
   kv.key[1] = reass->key.as_u64[1];
   kv.key[2] = reass->key.as_u64[2];
   clib_bihash_add_del_24_8 (&rm->hash, &kv, 0);
-  pool_put (rm->pool, reass);
-  --rm->reass_n;
+  pool_put (rt->pool, reass);
+  --rt->reass_n;
 }
 
-static void
+always_inline void
 ip4_reass_on_timeout (vlib_main_t * vm, ip4_reass_main_t * rm,
 		      ip4_reass_t * reass, u32 ** vec_drop_timeout)
 {
@@ -324,6 +338,7 @@
 
 ip4_reass_t *
 ip4_reass_find_or_create (vlib_main_t * vm, ip4_reass_main_t * rm,
+			  ip4_reass_per_thread_t * rt,
 			  ip4_reass_key_t * k, u32 ** vec_drop_timeout)
 {
   ip4_reass_t *reass = NULL;
@@ -335,11 +350,11 @@
 
   if (!clib_bihash_search_24_8 (&rm->hash, &kv, &value))
     {
-      reass = pool_elt_at_index (rm->pool, value.value);
+      reass = pool_elt_at_index (rt->pool, value.value);
       if (now > reass->last_heard + rm->timeout)
 	{
 	  ip4_reass_on_timeout (vm, rm, reass, vec_drop_timeout);
-	  ip4_reass_free (rm, reass);
+	  ip4_reass_free (rm, rt, reass);
 	  reass = NULL;
 	}
     }
@@ -350,43 +365,45 @@
       return reass;
     }
 
-  if (rm->reass_n >= rm->max_reass_n)
+  if (rt->reass_n >= rm->max_reass_n)
     {
       reass = NULL;
       return reass;
     }
   else
     {
-      pool_get (rm->pool, reass);
+      pool_get (rt->pool, reass);
       memset (reass, 0, sizeof (*reass));
-      reass->id = rm->id_counter;
-      ++rm->id_counter;
+      reass->id =
+	((u64) os_get_thread_index () * 1000000000) + rt->id_counter;
+      ++rt->id_counter;
       reass->first_bi = ~0;
       reass->last_packet_octet = ~0;
       reass->data_len = 0;
-      ++rm->reass_n;
+      ++rt->reass_n;
     }
 
   reass->key.as_u64[0] = kv.key[0] = k->as_u64[0];
   reass->key.as_u64[1] = kv.key[1] = k->as_u64[1];
   reass->key.as_u64[2] = kv.key[2] = k->as_u64[2];
-  kv.value = reass - rm->pool;
+  kv.value = reass - rt->pool;
   reass->last_heard = now;
 
   if (clib_bihash_add_del_24_8 (&rm->hash, &kv, 1))
     {
-      ip4_reass_free (rm, reass);
+      ip4_reass_free (rm, rt, reass);
       reass = NULL;
     }
 
   return reass;
 }
 
-void
+always_inline void
 ip4_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node,
-		    ip4_reass_main_t * rm, ip4_reass_t * reass, u32 * bi0,
-		    u32 * next0, vlib_error_t * error0, u32 next_input,
-		    u32 ** vec_drop_compress, u32 ** vec_drop_overlap)
+		    ip4_reass_main_t * rm, ip4_reass_per_thread_t * rt,
+		    ip4_reass_t * reass, u32 * bi0, u32 * next0,
+		    vlib_error_t * error0, u32 ** vec_drop_compress,
+		    u32 ** vec_drop_overlap, bool is_feature)
 {
   ASSERT (~0 != reass->first_bi);
   vlib_buffer_t *first_b = vlib_get_buffer (vm, reass->first_bi);
@@ -481,8 +498,8 @@
     }
   while (~0 != sub_chain_bi);
   last_b->flags &= ~VLIB_BUFFER_NEXT_PRESENT;
-  ASSERT (rm->buffers_n >= (buf_cnt - dropped_cnt));
-  rm->buffers_n -= buf_cnt - dropped_cnt;
+  ASSERT (rt->buffers_n >= (buf_cnt - dropped_cnt));
+  rt->buffers_n -= buf_cnt - dropped_cnt;
   ASSERT (total_length >= first_b->current_length);
   total_length -= first_b->current_length;
   first_b->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID;
@@ -523,13 +540,21 @@
 #endif
     }
   *bi0 = reass->first_bi;
-  *next0 = next_input;
+  if (is_feature)
+    {
+      *next0 = IP4_REASSEMBLY_NEXT_INPUT;
+    }
+  else
+    {
+      *next0 = reass->next_index;
+    }
+  vnet_buffer (first_b)->ip.reass.estimated_mtu = reass->min_fragment_length;
   *error0 = IP4_ERROR_NONE;
-  ip4_reass_free (rm, reass);
+  ip4_reass_free (rm, rt, reass);
   reass = NULL;
 }
 
-static u32
+always_inline u32
 ip4_reass_get_buffer_chain_length (vlib_main_t * vm, vlib_buffer_t * b)
 {
   u32 len = 0;
@@ -548,9 +573,10 @@
   return len;
 }
 
-static void
+always_inline void
 ip4_reass_insert_range_in_chain (vlib_main_t * vm,
 				 ip4_reass_main_t * rm,
+				 ip4_reass_per_thread_t * rt,
 				 ip4_reass_t * reass,
 				 u32 prev_range_bi, u32 new_next_bi)
 {
@@ -573,10 +599,10 @@
       reass->first_bi = new_next_bi;
     }
   reass->data_len += ip4_reass_buffer_get_data_len (new_next_b);
-  rm->buffers_n += ip4_reass_get_buffer_chain_length (vm, new_next_b);
+  rt->buffers_n += ip4_reass_get_buffer_chain_length (vm, new_next_b);
 }
 
-static void
+always_inline void
 ip4_reass_remove_range_from_chain (vlib_main_t * vm,
 				   vlib_node_runtime_t * node,
 				   ip4_reass_main_t * rm,
@@ -619,12 +645,12 @@
     }
 }
 
-void
+always_inline void
 ip4_reass_update (vlib_main_t * vm, vlib_node_runtime_t * node,
-		  ip4_reass_main_t * rm, ip4_reass_t * reass, u32 * bi0,
-		  u32 * next0, vlib_error_t * error0,
-		  u32 ** vec_drop_overlap, u32 ** vec_drop_compress,
-		  u32 next_input, u32 next_drop)
+		  ip4_reass_main_t * rm, ip4_reass_per_thread_t * rt,
+		  ip4_reass_t * reass, u32 * bi0, u32 * next0,
+		  vlib_error_t * error0, u32 ** vec_drop_overlap,
+		  u32 ** vec_drop_compress, bool is_feature)
 {
   int consumed = 0;
   vlib_buffer_t *fb = vlib_get_buffer (vm, *bi0);
@@ -640,6 +666,7 @@
   int more_fragments = ip4_get_fragment_more (fip);
   u32 candidate_range_bi = reass->first_bi;
   u32 prev_range_bi = ~0;
+  reass->next_index = fvnb->ip.reass.next_index;	// store next_index before it's overwritten
   fvnb->ip.reass.range_first = fragment_first;
   fvnb->ip.reass.range_last = fragment_last;
   fvnb->ip.reass.next_range_bi = ~0;
@@ -650,17 +677,18 @@
   if (~0 == reass->first_bi)
     {
       // starting a new reassembly
-      ip4_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi, *bi0);
+      ip4_reass_insert_range_in_chain (vm, rm, rt, reass, prev_range_bi,
+				       *bi0);
       if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED))
 	{
 	  ip4_reass_add_trace (vm, node, rm, reass, *bi0, RANGE_NEW, 0);
 	}
       *bi0 = ~0;
-      fvnb->ip.reass.estimated_mtu = clib_net_to_host_u16 (fip->length);
+      reass->min_fragment_length = clib_net_to_host_u16 (fip->length);
       return;
     }
-  fvnb->ip.reass.estimated_mtu = clib_min (clib_net_to_host_u16 (fip->length),
-					   fvnb->ip.reass.estimated_mtu);
+  reass->min_fragment_length = clib_min (clib_net_to_host_u16 (fip->length),
+					 fvnb->ip.reass.estimated_mtu);
   while (~0 != candidate_range_bi)
     {
       vlib_buffer_t *candidate_b = vlib_get_buffer (vm, candidate_range_bi);
@@ -674,8 +702,8 @@
 	      ~0 == candidate_range_bi)
 	    {
 	      // special case - this fragment falls beyond all known ranges
-	      ip4_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi,
-					       *bi0);
+	      ip4_reass_insert_range_in_chain (vm, rm, rt, reass,
+					       prev_range_bi, *bi0);
 	      consumed = 1;
 	      break;
 	    }
@@ -684,7 +712,7 @@
       if (fragment_last < candidate_vnb->ip.reass.range_first)
 	{
 	  // this fragment ends before candidate range without any overlap
-	  ip4_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi,
+	  ip4_reass_insert_range_in_chain (vm, rm, rt, reass, prev_range_bi,
 					   *bi0);
 	  consumed = 1;
 	}
@@ -717,7 +745,7 @@
 					   candidate_range_bi, RANGE_SHRINK,
 					   overlap);
 		    }
-		  ip4_reass_insert_range_in_chain (vm, rm, reass,
+		  ip4_reass_insert_range_in_chain (vm, rm, rt, reass,
 						   prev_range_bi, *bi0);
 		  consumed = 1;
 		}
@@ -743,7 +771,7 @@
 		  else
 		    {
 		      // special case - last range discarded
-		      ip4_reass_insert_range_in_chain (vm, rm, reass,
+		      ip4_reass_insert_range_in_chain (vm, rm, rt, reass,
 						       candidate_range_bi,
 						       *bi0);
 		      consumed = 1;
@@ -774,7 +802,7 @@
 	      else
 		{
 		  // special case - last range discarded
-		  ip4_reass_insert_range_in_chain (vm, rm, reass,
+		  ip4_reass_insert_range_in_chain (vm, rm, rt, reass,
 						   prev_range_bi, *bi0);
 		  consumed = 1;
 		}
@@ -792,8 +820,8 @@
   if (~0 != reass->last_packet_octet &&
       reass->data_len == reass->last_packet_octet + 1)
     {
-      ip4_reass_finalize (vm, node, rm, reass, bi0, next0, error0, next_input,
-			  vec_drop_compress, vec_drop_overlap);
+      ip4_reass_finalize (vm, node, rm, rt, reass, bi0, next0, error0,
+			  vec_drop_compress, vec_drop_overlap, is_feature);
     }
   else
     {
@@ -803,19 +831,22 @@
 	}
       else
 	{
-	  *next0 = next_drop;
+	  *next0 = IP4_REASSEMBLY_NEXT_DROP;
 	  *error0 = IP4_ERROR_REASS_DUPLICATE_FRAGMENT;
 	}
     }
 }
 
 always_inline uword
-ip4_reassembly (vlib_main_t * vm, vlib_node_runtime_t * node,
-		vlib_frame_t * frame)
+ip4_reassembly_inline (vlib_main_t * vm,
+		       vlib_node_runtime_t * node,
+		       vlib_frame_t * frame, bool is_feature)
 {
   u32 *from = vlib_frame_vector_args (frame);
   u32 n_left_from, n_left_to_next, *to_next, next_index;
   ip4_reass_main_t *rm = &ip4_reass_main;
+  ip4_reass_per_thread_t *rt = &rm->per_thread_data[os_get_thread_index ()];
+  clib_spinlock_lock (&rt->lock);
 
   n_left_from = frame->n_vectors;
   next_index = node->cached_next_index;
@@ -823,7 +854,7 @@
   static u32 *vec_drop_overlap = NULL;	// indexes of buffers which were discarded due to overlap
   static u32 *vec_drop_compress = NULL;	// indexes of buffers dicarded due to buffer compression
   while (n_left_from > 0 || vec_len (vec_drop_timeout) > 0 ||
-	 vec_len (vec_drop_overlap) > 0)
+	 vec_len (vec_drop_overlap) > 0 || vec_len (vec_drop_compress) > 0)
     {
       vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
 
@@ -839,8 +870,8 @@
 					   n_left_to_next, bi,
 					   IP4_REASSEMBLY_NEXT_DROP);
 	  IP4_REASS_DEBUG_BUFFER (bi, enqueue_drop_timeout);
-	  ASSERT (rm->buffers_n > 0);
-	  --rm->buffers_n;
+	  ASSERT (rt->buffers_n > 0);
+	  --rt->buffers_n;
 	}
 
       while (vec_len (vec_drop_overlap) > 0 && n_left_to_next > 0)
@@ -855,8 +886,8 @@
 					   n_left_to_next, bi,
 					   IP4_REASSEMBLY_NEXT_DROP);
 	  IP4_REASS_DEBUG_BUFFER (bi, enqueue_drop_duplicate_fragment);
-	  ASSERT (rm->buffers_n > 0);
-	  --rm->buffers_n;
+	  ASSERT (rt->buffers_n > 0);
+	  --rt->buffers_n;
 	}
 
       while (vec_len (vec_drop_compress) > 0 && n_left_to_next > 0)
@@ -871,51 +902,70 @@
 					   n_left_to_next, bi,
 					   IP4_REASSEMBLY_NEXT_DROP);
 	  IP4_REASS_DEBUG_BUFFER (bi, enqueue_drop_compress);
-	  ASSERT (rm->buffers_n > 0);
-	  --rm->buffers_n;
+	  ASSERT (rt->buffers_n > 0);
+	  --rt->buffers_n;
 	}
 
       while (n_left_from > 0 && n_left_to_next > 0)
 	{
 	  u32 bi0;
 	  vlib_buffer_t *b0;
-	  u32 next0;		//, error0;
+	  u32 next0;
+	  u32 error0 = IP4_ERROR_NONE;
 
 	  bi0 = from[0];
 	  b0 = vlib_get_buffer (vm, bi0);
 
 	  ip4_header_t *ip0 = vlib_buffer_get_current (b0);
-	  ip4_reass_key_t k;
-	  k.src.as_u32 = ip0->src_address.as_u32;
-	  k.dst.as_u32 = ip0->dst_address.as_u32;
-	  k.xx_id = vnet_buffer (b0)->sw_if_index[VLIB_RX];
-	  k.frag_id = ip0->fragment_id;
-	  k.proto = ip0->protocol;
-	  k.unused = 0;
-	  ip4_reass_t *reass =
-	    ip4_reass_find_or_create (vm, rm, &k, &vec_drop_timeout);
-
-	  u32 error0 = IP4_ERROR_NONE;
-	  if (reass)
+	  if (!ip4_get_fragment_more (ip0) && !ip4_get_fragment_offset (ip0))
 	    {
-	      ip4_reass_update (vm, node, rm, reass, &bi0, &next0, &error0,
-				&vec_drop_overlap, &vec_drop_compress,
-				IP4_REASSEMBLY_NEXT_INPUT,
-				IP4_REASSEMBLY_NEXT_DROP);
+	      // this is a whole packet - no fragmentation
+	      if (is_feature)
+		{
+		  next0 = IP4_REASSEMBLY_NEXT_INPUT;
+		}
+	      else
+		{
+		  next0 = vnet_buffer (b0)->ip.reass.next_index;
+		}
 	    }
 	  else
 	    {
-	      next0 = IP4_REASSEMBLY_NEXT_DROP;
-	      error0 = IP4_ERROR_REASS_LIMIT_REACHED;
-	    }
+	      ip4_reass_key_t k;
+	      k.src.as_u32 = ip0->src_address.as_u32;
+	      k.dst.as_u32 = ip0->dst_address.as_u32;
+	      k.xx_id = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+	      k.frag_id = ip0->fragment_id;
+	      k.proto = ip0->protocol;
+	      k.unused = 0;
+	      ip4_reass_t *reass =
+		ip4_reass_find_or_create (vm, rm, rt, &k, &vec_drop_timeout);
 
-	  b0->error = node->errors[error0];
+	      if (reass)
+		{
+		  ip4_reass_update (vm, node, rm, rt, reass, &bi0, &next0,
+				    &error0, &vec_drop_overlap,
+				    &vec_drop_compress, is_feature);
+		}
+	      else
+		{
+		  next0 = IP4_REASSEMBLY_NEXT_DROP;
+		  error0 = IP4_ERROR_REASS_LIMIT_REACHED;
+		}
+
+	      b0->error = node->errors[error0];
+	    }
 
 	  if (bi0 != ~0)
 	    {
 	      to_next[0] = bi0;
 	      to_next += 1;
 	      n_left_to_next -= 1;
+	      if (is_feature && IP4_ERROR_NONE == error0)
+		{
+		  vnet_feature_next (vnet_buffer (b0)->sw_if_index[VLIB_RX],
+				     &next0, b0);
+		}
 	      vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
 					       n_left_to_next, bi0, next0);
 	      IP4_REASS_DEBUG_BUFFER (bi0, enqueue_next);
@@ -928,6 +978,7 @@
       vlib_put_next_frame (vm, node, next_index, n_left_to_next);
     }
 
+  clib_spinlock_unlock (&rt->lock);
   return frame->n_vectors;
 }
 
@@ -937,6 +988,13 @@
 #undef _
 };
 
+always_inline uword
+ip4_reassembly (vlib_main_t * vm, vlib_node_runtime_t * node,
+		vlib_frame_t * frame)
+{
+  return ip4_reassembly_inline (vm, node, frame, false /* is_feature */ );
+}
+
 /* *INDENT-OFF* */
 VLIB_REGISTER_NODE (ip4_reass_node, static) = {
     .function = ip4_reassembly,
@@ -954,8 +1012,45 @@
 };
 /* *INDENT-ON* */
 
-VLIB_NODE_FUNCTION_MULTIARCH (ip4_reass_node, ip4_reassembly)
-     static u32 ip4_reass_get_nbuckets ()
+VLIB_NODE_FUNCTION_MULTIARCH (ip4_reass_node, ip4_reassembly);
+
+always_inline uword
+ip4_reassembly_feature (vlib_main_t * vm,
+			vlib_node_runtime_t * node, vlib_frame_t * frame)
+{
+  return ip4_reassembly_inline (vm, node, frame, true /* is_feature */ );
+}
+
+/* *INDENT-OFF* */
+VLIB_REGISTER_NODE (ip4_reass_node_feature, static) = {
+    .function = ip4_reassembly_feature,
+    .name = "ip4-reassembly-feature",
+    .vector_size = sizeof (u32),
+    .format_trace = format_ip4_reass_trace,
+    .n_errors = ARRAY_LEN (ip4_reassembly_error_strings),
+    .error_strings = ip4_reassembly_error_strings,
+    .n_next_nodes = IP4_REASSEMBLY_N_NEXT,
+    .next_nodes =
+        {
+                [IP4_REASSEMBLY_NEXT_INPUT] = "ip4-input",
+                [IP4_REASSEMBLY_NEXT_DROP] = "ip4-drop",
+        },
+};
+/* *INDENT-ON* */
+
+VLIB_NODE_FUNCTION_MULTIARCH (ip4_reass_node_feature, ip4_reassembly_feature);
+
+/* *INDENT-OFF* */
+VNET_FEATURE_INIT (ip4_reassembly_feature, static) = {
+    .arc_name = "ip4-unicast",
+    .node_name = "ip4-reassembly-feature",
+    .runs_before = VNET_FEATURES ("ip4-lookup"),
+    .runs_after = 0,
+};
+/* *INDENT-ON* */
+
+always_inline u32
+ip4_reass_get_nbuckets ()
 {
   ip4_reass_main_t *rm = &ip4_reass_main;
   u32 nbuckets;
@@ -982,7 +1077,7 @@
   clib_bihash_24_8_t *new_hash;
 } ip4_rehash_cb_ctx;
 
-void
+static void
 ip4_rehash_cb (clib_bihash_kv_24_8_t * kv, void *_ctx)
 {
   ip4_rehash_cb_ctx *ctx = _ctx;
@@ -992,16 +1087,23 @@
     }
 }
 
+static void
+ip4_reass_set_params (u32 timeout_ms, u32 max_reassemblies,
+		      u32 expire_walk_interval_ms)
+{
+  ip4_reass_main.timeout_ms = timeout_ms;
+  ip4_reass_main.timeout = (f64) timeout_ms / (f64) MSEC_PER_SEC;
+  ip4_reass_main.max_reass_n = max_reassemblies;
+  ip4_reass_main.expire_walk_interval_ms = expire_walk_interval_ms;
+}
+
 vnet_api_error_t
 ip4_reass_set (u32 timeout_ms, u32 max_reassemblies,
 	       u32 expire_walk_interval_ms)
 {
   u32 old_nbuckets = ip4_reass_get_nbuckets ();
-  ip4_reass_main.timeout_ms = timeout_ms;
-  ip4_reass_main.timeout = (f64) timeout_ms / (f64) MSEC_PER_SEC;
-  ip4_reass_main.max_reass_n = max_reassemblies;
-  ip4_reass_main.expire_walk_interval_ms = expire_walk_interval_ms;
-
+  ip4_reass_set_params (timeout_ms, max_reassemblies,
+			expire_walk_interval_ms);
   vlib_process_signal_event (ip4_reass_main.vlib_main,
 			     ip4_reass_main.ip4_reass_expire_node_idx,
 			     IP4_EVENT_CONFIG_CHANGED, 0);
@@ -1043,7 +1145,7 @@
   return 0;
 }
 
-clib_error_t *
+static clib_error_t *
 ip4_reass_init_function (vlib_main_t * vm)
 {
   ip4_reass_main_t *rm = &ip4_reass_main;
@@ -1054,23 +1156,29 @@
   rm->vlib_main = vm;
   rm->vnet_main = vnet_get_main ();
 
-  rm->reass_n = 0;
-  pool_alloc (rm->pool, rm->max_reass_n);
+  vec_validate (rm->per_thread_data, vlib_num_workers () + 1);
+  ip4_reass_per_thread_t *rt;
+  vec_foreach (rt, rm->per_thread_data)
+  {
+    clib_spinlock_init (&rt->lock);
+    pool_alloc (rt->pool, rm->max_reass_n);
+  }
 
   node = vlib_get_node_by_name (vm, (u8 *) "ip4-reassembly-expire-walk");
   ASSERT (node);
   rm->ip4_reass_expire_node_idx = node->index;
 
-  ip4_reass_set (IP4_REASS_TIMEOUT_DEFAULT_MS,
-		 IP4_REASS_MAX_REASSEMBLIES_DEAFULT,
-		 IP4_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS);
-
   nbuckets = ip4_reass_get_nbuckets ();
   clib_bihash_init_24_8 (&rm->hash, "ip4-reass", nbuckets, nbuckets * 1024);
 
   node = vlib_get_node_by_name (vm, (u8 *) "ip4-drop");
   ASSERT (node);
   rm->ip4_drop_idx = node->index;
+
+  ip4_reass_set_params (IP4_REASS_TIMEOUT_DEFAULT_MS,
+			IP4_REASS_MAX_REASSEMBLIES_DEFAULT,
+			IP4_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS);
+
   return error;
 }
 
@@ -1107,32 +1215,57 @@
       u32 *vec_drop_timeout = NULL;
       int *pool_indexes_to_free = NULL;
 
+      uword thread_index = 0;
       int index;
-      /* *INDENT-OFF* */
-      pool_foreach_index (index, rm->pool, ({
-                            reass = pool_elt_at_index (rm->pool, index);
-                            if (now > reass->last_heard + rm->timeout)
-                              {
-                                vec_add1 (pool_indexes_to_free, index);
-                              }
-                          }));
-      /* *INDENT-ON* */
-      int *i;
-      /* *INDENT-OFF* */
-      vec_foreach (i, pool_indexes_to_free)
-      {
-        ip4_reass_t *reass = pool_elt_at_index (rm->pool, i[0]);
-        ip4_reass_on_timeout (vm, rm, reass, &vec_drop_timeout);
-        ip4_reass_free (rm, reass);
-      }
-      /* *INDENT-ON* */
+      const uword nthreads = os_get_nthreads ();
+      for (thread_index = 0; thread_index < nthreads; ++thread_index)
+	{
+	  ip4_reass_per_thread_t *rt = &rm->per_thread_data[thread_index];
+	  clib_spinlock_lock (&rt->lock);
+
+	  vec_reset_length (pool_indexes_to_free);
+          /* *INDENT-OFF* */
+          pool_foreach_index (index, rt->pool, ({
+                                reass = pool_elt_at_index (rt->pool, index);
+                                if (now > reass->last_heard + rm->timeout)
+                                  {
+                                    vec_add1 (pool_indexes_to_free, index);
+                                  }
+                              }));
+          /* *INDENT-ON* */
+	  int *i;
+          /* *INDENT-OFF* */
+          vec_foreach (i, pool_indexes_to_free)
+          {
+            ip4_reass_t *reass = pool_elt_at_index (rt->pool, i[0]);
+	    u32 before = vec_len (vec_drop_timeout);
+	    vlib_buffer_t *b = vlib_get_buffer (vm, reass->first_bi);
+	    if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED))
+	      {
+		if (pool_is_free_index (vm->trace_main.trace_buffer_pool,
+					b->trace_index))
+		  {
+		    /* the trace is gone, don't trace this buffer anymore */
+		    b->flags &= ~VLIB_BUFFER_IS_TRACED;
+		  }
+	      }
+            ip4_reass_on_timeout (vm, rm, reass, &vec_drop_timeout);
+            u32 after = vec_len (vec_drop_timeout);
+            ASSERT (rt->buffers_n >= (after - before));
+            rt->buffers_n -= (after - before);
+            ip4_reass_free (rm, rt, reass);
+          }
+          /* *INDENT-ON* */
+
+	  clib_spinlock_unlock (&rt->lock);
+	}
 
       while (vec_len (vec_drop_timeout) > 0)
 	{
 	  vlib_frame_t *f = vlib_get_frame_to_node (vm, rm->ip4_drop_idx);
 	  u32 *to_next = vlib_frame_vector_args (f);
 	  u32 n_left_to_next = VLIB_FRAME_SIZE - f->n_vectors;
-	  u32 n_trace = 0;
+	  int trace_frame = 0;
 	  while (vec_len (vec_drop_timeout) > 0 && n_left_to_next > 0)
 	    {
 	      u32 bi = vec_pop (vec_drop_timeout);
@@ -1147,7 +1280,7 @@
 		    }
 		  else
 		    {
-		      ++n_trace;
+		      trace_frame = 1;
 		    }
 		}
 	      b->error = node->errors[IP4_ERROR_REASS_TIMEOUT];
@@ -1156,13 +1289,8 @@
 	      to_next += 1;
 	      n_left_to_next -= 1;
 	      IP4_REASS_DEBUG_BUFFER (bi, enqueue_drop_timeout_walk);
-	      ASSERT (rm->buffers_n > 0);
-	      --rm->buffers_n;
 	    }
-	  if (PREDICT_FALSE (n_trace > 0))
-	    {
-	      f->flags |= VLIB_FRAME_TRACE;
-	    }
+	  f->flags |= (trace_frame * VLIB_FRAME_TRACE);
 	  vlib_put_frame_to_node (vm, rm->ip4_drop_idx, f);
 	}
 
@@ -1207,7 +1335,7 @@
   vlib_main_t *vm = va_arg (*args, vlib_main_t *);
   ip4_reass_t *reass = va_arg (*args, ip4_reass_t *);
 
-  s = format (s, "ID: %u, key: %U\n  first_bi: %u, data_len: %u, "
+  s = format (s, "ID: %lu, key: %U\n  first_bi: %u, data_len: %u, "
 	      "last_packet_octet: %u, trace_op_counter: %u\n",
 	      reass->id, format_ip4_reass_key, &reass->key, reass->first_bi,
 	      reass->data_len, reass->last_packet_octet,
@@ -1246,22 +1374,41 @@
   vlib_cli_output (vm, "---------------------");
   vlib_cli_output (vm, "IP4 reassembly status");
   vlib_cli_output (vm, "---------------------");
+  bool details = false;
   if (unformat (input, "details"))
     {
-      ip4_reass_t *reass;
-      /* *INDENT-OFF* */
-      pool_foreach (reass, rm->pool, {
-        vlib_cli_output (vm, "%U", format_ip4_reass, vm, reass);
-      });
-      /* *INDENT-ON* */
+      details = true;
+    }
+
+  u32 sum_reass_n = 0;
+  u64 sum_buffers_n = 0;
+  ip4_reass_t *reass;
+  uword thread_index;
+  const uword nthreads = os_get_nthreads ();
+  for (thread_index = 0; thread_index < nthreads; ++thread_index)
+    {
+      ip4_reass_per_thread_t *rt = &rm->per_thread_data[thread_index];
+      clib_spinlock_lock (&rt->lock);
+      if (details)
+	{
+          /* *INDENT-OFF* */
+          pool_foreach (reass, rt->pool, {
+            vlib_cli_output (vm, "%U", format_ip4_reass, vm, reass);
+          });
+          /* *INDENT-ON* */
+	}
+      sum_reass_n += rt->reass_n;
+      sum_buffers_n += rt->buffers_n;
+      clib_spinlock_unlock (&rt->lock);
     }
   vlib_cli_output (vm, "---------------------");
-  vlib_cli_output (vm, "Current IP4 reassemblies count: %lu\n", rm->reass_n);
+  vlib_cli_output (vm, "Current IP4 reassemblies count: %lu\n",
+		   (long unsigned) sum_reass_n);
   vlib_cli_output (vm,
-		   "Maximum configured concurrent IP4 reassemblies: %lu\n",
+		   "Maximum configured concurrent IP4 reassemblies per worker-thread: %lu\n",
 		   (long unsigned) rm->max_reass_n);
   vlib_cli_output (vm, "Buffers in use: %lu\n",
-		   (long unsigned) rm->buffers_n);
+		   (long unsigned) sum_buffers_n);
   return 0;
 }
 
@@ -1273,6 +1420,13 @@
 };
 /* *INDENT-ON* */
 
+vnet_api_error_t
+ip4_reass_enable_disable (u32 sw_if_index, u8 enable_disable)
+{
+  return vnet_feature_enable_disable ("ip4-unicast", "ip4-reassembly-feature",
+				      sw_if_index, enable_disable, 0, 0);
+}
+
 /*
  * fd.io coding-style-patch-verification: ON
  *
diff --git a/src/vnet/ip/ip4_reassembly.h b/src/vnet/ip/ip4_reassembly.h
index a48c07e..521ca0f 100644
--- a/src/vnet/ip/ip4_reassembly.h
+++ b/src/vnet/ip/ip4_reassembly.h
@@ -38,6 +38,9 @@
 vnet_api_error_t ip4_reass_get (u32 * timeout_ms, u32 * max_reassemblies,
 				u32 * expire_walk_interval_ms);
 
+vnet_api_error_t ip4_reass_enable_disable (u32 sw_if_index,
+					   u8 enable_disable);
+
 #endif /* __included_ip4_reassembly_h */
 
 /*
diff --git a/src/vnet/ip/ip6_forward.c b/src/vnet/ip/ip6_forward.c
index 3eb84e8..2c00111 100644
--- a/src/vnet/ip/ip6_forward.c
+++ b/src/vnet/ip/ip6_forward.c
@@ -1352,7 +1352,6 @@
     [IP_LOCAL_NEXT_PUNT] = "ip6-punt",
     [IP_LOCAL_NEXT_UDP_LOOKUP] = "ip6-udp-lookup",
     [IP_LOCAL_NEXT_ICMP] = "ip6-icmp-input",
-    [IP_LOCAL_NEXT_REASSEMBLY] = "ip6-reassembly",
   },
 };
 /* *INDENT-ON* */
diff --git a/src/vnet/ip/ip6_reassembly.c b/src/vnet/ip/ip6_reassembly.c
index df9c4fb..c6424a7 100644
--- a/src/vnet/ip/ip6_reassembly.c
+++ b/src/vnet/ip/ip6_reassembly.c
@@ -29,7 +29,7 @@
 #define MSEC_PER_SEC 1000
 #define IP6_REASS_TIMEOUT_DEFAULT_MS 100
 #define IP6_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS 10000	// 10 seconds default
-#define IP6_REASS_MAX_REASSEMBLIES_DEAFULT 1024
+#define IP6_REASS_MAX_REASSEMBLIES_DEFAULT 1024
 #define IP6_REASS_HT_LOAD_FACTOR (0.75)
 
 static vlib_node_registration_t ip6_reass_node;
@@ -89,12 +89,10 @@
 {
   // hash table key
   ip6_reass_key_t key;
-  // time when first packet was received
-  f64 first_heard;
   // time when last packet was received
   f64 last_heard;
   // internal id of this reassembly
-  u32 id;
+  u64 id;
   // buffer index of first buffer in this reassembly context
   u32 first_bi;
   // last octet of packet, ~0 until fragment without more_fragments arrives
@@ -103,10 +101,23 @@
   u32 data_len;
   // trace operation counter
   u32 trace_op_counter;
+  // next index - used by non-feature node
+  u8 next_index;
+  // minimum fragment length for this reassembly - used to estimate MTU
+  u16 min_fragment_length;
 } ip6_reass_t;
 
 typedef struct
 {
+  ip6_reass_t *pool;
+  u32 reass_n;
+  u32 buffers_n;
+  u32 id_counter;
+  clib_spinlock_t lock;
+} ip6_reass_per_thread_t;
+
+typedef struct
+{
   // IPv6 config
   u32 timeout_ms;
   f64 timeout;
@@ -114,11 +125,10 @@
   u32 max_reass_n;
 
   // IPv6 runtime
-  ip6_reass_t *pool;
   clib_bihash_48_8_t hash;
-  u32 reass_n;
-  u32 id_counter;
-  u32 buffers_n;
+
+  // per-thread data
+  ip6_reass_per_thread_t *per_thread_data;
 
   // convenience
   vlib_main_t *vlib_main;
@@ -164,7 +174,6 @@
 typedef struct
 {
   ip6_reass_trace_operation_e action;
-  u32 pool_index;
   u32 reass_id;
   ip6_reass_range_trace_t trace_range;
   u32 size_diff;
@@ -187,7 +196,7 @@
   trace->range_bi = bi;
 }
 
-u8 *
+static u8 *
 format_ip6_reass_range_trace (u8 * s, va_list * args)
 {
   ip6_reass_range_trace_t *trace = va_arg (*args, ip6_reass_range_trace_t *);
@@ -197,7 +206,7 @@
   return s;
 }
 
-u8 *
+static u8 *
 format_ip6_reass_trace (u8 * s, va_list * args)
 {
   CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
@@ -247,8 +256,13 @@
 {
   vlib_buffer_t *b = vlib_get_buffer (vm, bi);
   vnet_buffer_opaque_t *vnb = vnet_buffer (b);
+  if (pool_is_free_index (vm->trace_main.trace_buffer_pool, b->trace_index))
+    {
+      // this buffer's trace is gone
+      b->flags &= ~VLIB_BUFFER_IS_TRACED;
+      return;
+    }
   ip6_reass_trace_t *t = vlib_add_trace (vm, node, b, sizeof (t[0]));
-  t->pool_index = reass - rm->pool;
   t->reass_id = reass->id;
   t->action = action;
   ip6_reass_trace_details (vm, bi, &t->trace_range);
@@ -268,8 +282,9 @@
 #endif
 }
 
-static void
-ip6_reass_free (ip6_reass_main_t * rm, ip6_reass_t * reass)
+always_inline void
+ip6_reass_free (ip6_reass_main_t * rm, ip6_reass_per_thread_t * rt,
+		ip6_reass_t * reass)
 {
   clib_bihash_kv_48_8_t kv;
   kv.key[0] = reass->key.as_u64[0];
@@ -279,11 +294,11 @@
   kv.key[4] = reass->key.as_u64[4];
   kv.key[5] = reass->key.as_u64[5];
   clib_bihash_add_del_48_8 (&rm->hash, &kv, 0);
-  pool_put (rm->pool, reass);
-  --rm->reass_n;
+  pool_put (rt->pool, reass);
+  --rt->reass_n;
 }
 
-static void
+always_inline void
 ip6_reass_drop_all (vlib_main_t * vm, ip6_reass_main_t * rm,
 		    ip6_reass_t * reass, u32 ** vec_drop_bi)
 {
@@ -313,7 +328,7 @@
     }
 }
 
-static void
+always_inline void
 ip6_reass_on_timeout (vlib_main_t * vm, vlib_node_runtime_t * node,
 		      ip6_reass_main_t * rm, ip6_reass_t * reass,
 		      u32 * icmp_bi, u32 ** vec_timeout)
@@ -349,10 +364,9 @@
   ip6_reass_drop_all (vm, rm, reass, vec_timeout);
 }
 
-static ip6_reass_t *
-ip6_reass_find_or_create (vlib_main_t * vm,
-			  vlib_node_runtime_t * node,
-			  ip6_reass_main_t * rm,
+always_inline ip6_reass_t *
+ip6_reass_find_or_create (vlib_main_t * vm, vlib_node_runtime_t * node,
+			  ip6_reass_main_t * rm, ip6_reass_per_thread_t * rt,
 			  ip6_reass_key_t * k, u32 * icmp_bi,
 			  u32 ** vec_timeout)
 {
@@ -368,11 +382,11 @@
 
   if (!clib_bihash_search_48_8 (&rm->hash, &kv, &value))
     {
-      reass = pool_elt_at_index (rm->pool, value.value);
+      reass = pool_elt_at_index (rt->pool, value.value);
       if (now > reass->last_heard + rm->timeout)
 	{
 	  ip6_reass_on_timeout (vm, node, rm, reass, icmp_bi, vec_timeout);
-	  ip6_reass_free (rm, reass);
+	  ip6_reass_free (rm, rt, reass);
 	  reass = NULL;
 	}
     }
@@ -383,21 +397,22 @@
       return reass;
     }
 
-  if (rm->reass_n >= rm->max_reass_n)
+  if (rt->reass_n >= rm->max_reass_n)
     {
       reass = NULL;
       return reass;
     }
   else
     {
-      pool_get (rm->pool, reass);
+      pool_get (rt->pool, reass);
       memset (reass, 0, sizeof (*reass));
-      reass->id = rm->id_counter;
-      ++rm->id_counter;
+      reass->id =
+	((u64) os_get_thread_index () * 1000000000) + rt->id_counter;
+      ++rt->id_counter;
       reass->first_bi = ~0;
       reass->last_packet_octet = ~0;
       reass->data_len = 0;
-      ++rm->reass_n;
+      ++rt->reass_n;
     }
 
   reass->key.as_u64[0] = kv.key[0] = k->as_u64[0];
@@ -406,27 +421,27 @@
   reass->key.as_u64[3] = kv.key[3] = k->as_u64[3];
   reass->key.as_u64[4] = kv.key[4] = k->as_u64[4];
   reass->key.as_u64[5] = kv.key[5] = k->as_u64[5];
-  kv.value = reass - rm->pool;
+  kv.value = reass - rt->pool;
   reass->last_heard = now;
 
   if (clib_bihash_add_del_48_8 (&rm->hash, &kv, 1))
     {
-      ip6_reass_free (rm, reass);
+      ip6_reass_free (rm, rt, reass);
       reass = NULL;
     }
 
   return reass;
 }
 
-void
+always_inline void
 ip6_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node,
-		    ip6_reass_main_t * rm, ip6_reass_t * reass, u32 * bi0,
-		    u32 * next0, vlib_error_t * error0, u32 next_input,
-		    u32 ** vec_drop_compress)
+		    ip6_reass_main_t * rm, ip6_reass_per_thread_t * rt,
+		    ip6_reass_t * reass, u32 * bi0, u32 * next0,
+		    vlib_error_t * error0, u32 ** vec_drop_compress,
+		    bool is_feature)
 {
   ASSERT (~0 != reass->first_bi);
   *bi0 = reass->first_bi;
-  *next0 = next_input;
   *error0 = IP6_ERROR_NONE;
   ip6_frag_hdr_t *frag_hdr;
   vlib_buffer_t *last_b = NULL;
@@ -548,7 +563,6 @@
   ip->payload_length =
     clib_host_to_net_u16 (total_length + first_b->current_length -
 			  sizeof (*ip));
-  ip6_reass_free (rm, reass);
   vlib_buffer_chain_compress (vm, first_b, vec_drop_compress);
   if (PREDICT_FALSE (first_b->flags & VLIB_BUFFER_IS_TRACED))
     {
@@ -580,10 +594,20 @@
       while (0);
 #endif
     }
+  if (is_feature)
+    {
+      *next0 = IP6_REASSEMBLY_NEXT_INPUT;
+    }
+  else
+    {
+      *next0 = reass->next_index;
+    }
+  vnet_buffer (first_b)->ip.reass.estimated_mtu = reass->min_fragment_length;
+  ip6_reass_free (rm, rt, reass);
   reass = NULL;
 }
 
-static u32
+always_inline u32
 ip6_reass_get_buffer_chain_length (vlib_main_t * vm, vlib_buffer_t * b)
 {
   u32 len = 0;
@@ -602,11 +626,11 @@
   return len;
 }
 
-static void
-ip6_reass_insert_range_in_chain (vlib_main_t * vm,
-				 ip6_reass_main_t * rm,
-				 ip6_reass_t * reass,
-				 u32 prev_range_bi, u32 new_next_bi)
+always_inline void
+ip6_reass_insert_range_in_chain (vlib_main_t * vm, ip6_reass_main_t * rm,
+				 ip6_reass_per_thread_t * rt,
+				 ip6_reass_t * reass, u32 prev_range_bi,
+				 u32 new_next_bi)
 {
 
   vlib_buffer_t *new_next_b = vlib_get_buffer (vm, new_next_bi);
@@ -627,16 +651,16 @@
       reass->first_bi = new_next_bi;
     }
   reass->data_len += ip6_reass_buffer_get_data_len (new_next_b);
-  rm->buffers_n += ip6_reass_get_buffer_chain_length (vm, new_next_b);
+  rt->buffers_n += ip6_reass_get_buffer_chain_length (vm, new_next_b);
 }
 
-void
+always_inline void
 ip6_reass_update (vlib_main_t * vm, vlib_node_runtime_t * node,
-		  ip6_reass_main_t * rm, ip6_reass_t * reass, u32 * bi0,
-		  u32 * next0, vlib_error_t * error0,
-		  ip6_frag_hdr_t * frag_hdr, u32 ** vec_drop_overlap,
-		  u32 ** vec_drop_compress, u32 next_input, u32 next_drop,
-		  u32 next_icmp_error)
+		  ip6_reass_main_t * rm, ip6_reass_per_thread_t * rt,
+		  ip6_reass_t * reass, u32 * bi0, u32 * next0,
+		  vlib_error_t * error0, ip6_frag_hdr_t * frag_hdr,
+		  u32 ** vec_drop_overlap, u32 ** vec_drop_compress,
+		  bool is_feature)
 {
   int consumed = 0;
   vlib_buffer_t *fb = vlib_get_buffer (vm, *bi0);
@@ -660,34 +684,6 @@
   fvnb->ip.reass.range_first = fragment_first;
   fvnb->ip.reass.range_last = fragment_last;
   fvnb->ip.reass.next_range_bi = ~0;
-  if (more_fragments && 0 != fragment_length % 8)
-    {
-      if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED))
-	{
-	  ip6_reass_add_trace (vm, node, rm, reass, *bi0,
-			       ICMP_ERROR_FL_NOT_MULT_8, 0);
-	}
-      *next0 = next_icmp_error;
-      icmp6_error_set_vnet_buffer (fb, ICMP6_parameter_problem,
-				   ICMP6_parameter_problem_erroneous_header_field,
-				   (u8 *) & fip->payload_length - (u8 *) fip);
-      return;
-    }
-  if (fragment_first + fragment_length > 65535)
-    {
-      if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED))
-	{
-	  ip6_reass_add_trace (vm, node, rm, reass, *bi0,
-			       ICMP_ERROR_FL_TOO_BIG, 0);
-	}
-      *next0 = next_icmp_error;
-      ip6_header_t *ip0 = vlib_buffer_get_current (fb);
-      icmp6_error_set_vnet_buffer (fb, ICMP6_parameter_problem,
-				   ICMP6_parameter_problem_erroneous_header_field,
-				   (u8 *) & frag_hdr->fragment_offset_and_more
-				   - (u8 *) ip0);
-      return;
-    }
   if (!more_fragments)
     {
       reass->last_packet_octet = fragment_last;
@@ -695,15 +691,17 @@
   if (~0 == reass->first_bi)
     {
       // starting a new reassembly
-      ip6_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi, *bi0);
+      ip6_reass_insert_range_in_chain (vm, rm, rt, reass, prev_range_bi,
+				       *bi0);
       if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED))
 	{
 	  ip6_reass_add_trace (vm, node, rm, reass, *bi0, RANGE_NEW, 0);
 	}
+      reass->min_fragment_length = clib_net_to_host_u16 (fip->payload_length);
       *bi0 = ~0;
       return;
     }
-  fvnb->ip.reass.estimated_mtu =
+  reass->min_fragment_length =
     clib_min (clib_net_to_host_u16 (fip->payload_length),
 	      fvnb->ip.reass.estimated_mtu);
   while (~0 != candidate_range_bi)
@@ -719,8 +717,8 @@
 	      ~0 == candidate_range_bi)
 	    {
 	      // special case - this fragment falls beyond all known ranges
-	      ip6_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi,
-					       *bi0);
+	      ip6_reass_insert_range_in_chain (vm, rm, rt, reass,
+					       prev_range_bi, *bi0);
 	      consumed = 1;
 	      break;
 	    }
@@ -729,7 +727,7 @@
       if (fragment_last < candidate_vnb->ip.reass.range_first)
 	{
 	  // this fragment ends before candidate range without any overlap
-	  ip6_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi,
+	  ip6_reass_insert_range_in_chain (vm, rm, rt, reass, prev_range_bi,
 					   *bi0);
 	  consumed = 1;
 	}
@@ -742,13 +740,13 @@
 	{
 	  // overlapping fragment - not allowed by RFC 8200
 	  ip6_reass_drop_all (vm, rm, reass, vec_drop_overlap);
-	  ip6_reass_free (rm, reass);
+	  ip6_reass_free (rm, rt, reass);
 	  if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED))
 	    {
 	      ip6_reass_add_trace (vm, node, rm, reass, *bi0, RANGE_OVERLAP,
 				   0);
 	    }
-	  *next0 = next_drop;
+	  *next0 = IP6_REASSEMBLY_NEXT_DROP;
 	  *error0 = IP6_ERROR_REASS_OVERLAPPING_FRAGMENT;
 	}
       break;
@@ -763,8 +761,8 @@
   if (~0 != reass->last_packet_octet &&
       reass->data_len == reass->last_packet_octet + 1)
     {
-      ip6_reass_finalize (vm, node, rm, reass, bi0, next0, error0, next_input,
-			  vec_drop_compress);
+      ip6_reass_finalize (vm, node, rm, rt, reass, bi0, next0, error0,
+			  vec_drop_compress, is_feature);
     }
   else
     {
@@ -774,19 +772,90 @@
 	}
       else
 	{
-	  *next0 = next_drop;
+	  *next0 = IP6_REASSEMBLY_NEXT_DROP;
+	  ;
 	  *error0 = IP6_ERROR_REASS_DUPLICATE_FRAGMENT;
 	}
     }
 }
 
+always_inline bool
+ip6_reass_verify_upper_layer_present (vlib_node_runtime_t * node,
+				      vlib_buffer_t * b,
+				      ip6_frag_hdr_t * frag_hdr)
+{
+  ip6_ext_header_t *tmp = (ip6_ext_header_t *) frag_hdr;
+  while (ip6_ext_hdr (tmp->next_hdr))
+    {
+      tmp = ip6_ext_next_header (tmp);
+    }
+  if (IP_PROTOCOL_IP6_NONXT == tmp->next_hdr)
+    {
+      icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem,
+				   ICMP6_parameter_problem_first_fragment_has_incomplete_header_chain,
+				   0);
+      b->error = node->errors[IP6_ERROR_REASS_MISSING_UPPER];
+
+      return false;
+    }
+  return true;
+}
+
+always_inline bool
+ip6_reass_verify_fragment_multiple_8 (vlib_main_t * vm,
+				      vlib_node_runtime_t * node,
+				      vlib_buffer_t * b,
+				      ip6_frag_hdr_t * frag_hdr)
+{
+  vnet_buffer_opaque_t *vnb = vnet_buffer (b);
+  ip6_header_t *ip = vlib_buffer_get_current (b);
+  int more_fragments = ip6_frag_hdr_more (frag_hdr);
+  u32 fragment_length =
+    vlib_buffer_length_in_chain (vm, b) -
+    (vnb->ip.reass.ip6_frag_hdr_offset + sizeof (*frag_hdr));
+  if (more_fragments && 0 != fragment_length % 8)
+    {
+      icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem,
+				   ICMP6_parameter_problem_erroneous_header_field,
+				   (u8 *) & ip->payload_length - (u8 *) ip);
+      return false;
+    }
+  return true;
+}
+
+always_inline bool
+ip6_reass_verify_packet_size_lt_64k (vlib_main_t * vm,
+				     vlib_node_runtime_t * node,
+				     vlib_buffer_t * b,
+				     ip6_frag_hdr_t * frag_hdr)
+{
+  vnet_buffer_opaque_t *vnb = vnet_buffer (b);
+  u32 fragment_first = ip6_frag_hdr_offset_bytes (frag_hdr);
+  u32 fragment_length =
+    vlib_buffer_length_in_chain (vm, b) -
+    (vnb->ip.reass.ip6_frag_hdr_offset + sizeof (*frag_hdr));
+  if (fragment_first + fragment_length > 65535)
+    {
+      ip6_header_t *ip0 = vlib_buffer_get_current (b);
+      icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem,
+				   ICMP6_parameter_problem_erroneous_header_field,
+				   (u8 *) & frag_hdr->fragment_offset_and_more
+				   - (u8 *) ip0);
+      return false;
+    }
+  return true;
+}
+
 always_inline uword
-ip6_reassembly (vlib_main_t * vm, vlib_node_runtime_t * node,
-		vlib_frame_t * frame)
+ip6_reassembly_inline (vlib_main_t * vm,
+		       vlib_node_runtime_t * node,
+		       vlib_frame_t * frame, bool is_feature)
 {
   u32 *from = vlib_frame_vector_args (frame);
   u32 n_left_from, n_left_to_next, *to_next, next_index;
   ip6_reass_main_t *rm = &ip6_reass_main;
+  ip6_reass_per_thread_t *rt = &rm->per_thread_data[os_get_thread_index ()];
+  clib_spinlock_lock (&rt->lock);
 
   n_left_from = frame->n_vectors;
   next_index = node->cached_next_index;
@@ -794,7 +863,7 @@
   static u32 *vec_drop_overlap = NULL;	// indexes of buffers dropped due to overlap
   static u32 *vec_drop_compress = NULL;	// indexes of buffers dropped due to buffer compression
   while (n_left_from > 0 || vec_len (vec_timeout) > 0 ||
-	 vec_len (vec_drop_overlap) > 0)
+	 vec_len (vec_drop_overlap) > 0 || vec_len (vec_drop_compress) > 0)
     {
       vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
 
@@ -809,8 +878,8 @@
 	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
 					   n_left_to_next, bi,
 					   IP6_REASSEMBLY_NEXT_DROP);
-	  ASSERT (rm->buffers_n > 0);
-	  --rm->buffers_n;
+	  ASSERT (rt->buffers_n > 0);
+	  --rt->buffers_n;
 	}
 
       while (vec_len (vec_drop_overlap) > 0 && n_left_to_next > 0)
@@ -824,8 +893,8 @@
 	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
 					   n_left_to_next, bi,
 					   IP6_REASSEMBLY_NEXT_DROP);
-	  ASSERT (rm->buffers_n > 0);
-	  --rm->buffers_n;
+	  ASSERT (rt->buffers_n > 0);
+	  --rt->buffers_n;
 	}
 
       while (vec_len (vec_drop_compress) > 0 && n_left_to_next > 0)
@@ -839,15 +908,17 @@
 	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
 					   n_left_to_next, bi,
 					   IP6_REASSEMBLY_NEXT_DROP);
-	  ASSERT (rm->buffers_n > 0);
-	  --rm->buffers_n;
+	  ASSERT (rt->buffers_n > 0);
+	  --rt->buffers_n;
 	}
 
       while (n_left_from > 0 && n_left_to_next > 0)
 	{
 	  u32 bi0;
 	  vlib_buffer_t *b0;
-	  u32 next0;		//, error0;
+	  u32 next0;
+	  u32 error0 = IP6_ERROR_NONE;
+	  u32 icmp_bi = ~0;
 
 	  bi0 = from[0];
 	  b0 = vlib_get_buffer (vm, bi0);
@@ -857,30 +928,27 @@
 	  ip6_ext_header_t *prev_hdr;
 	  ip6_ext_header_find_t (ip0, prev_hdr, frag_hdr,
 				 IP_PROTOCOL_IPV6_FRAGMENTATION);
+	  if (!frag_hdr)
+	    {
+	      // this is a regular packet - no fragmentation
+	      next0 = IP6_REASSEMBLY_NEXT_INPUT;
+	      goto skip_reass;
+	    }
 	  if (0 == ip6_frag_hdr_offset (frag_hdr))
 	    {
 	      // first fragment - verify upper-layer is present
-	      ip6_ext_header_t *tmp = (ip6_ext_header_t *) frag_hdr;
-	      while (ip6_ext_hdr (tmp->next_hdr))
+	      if (!ip6_reass_verify_upper_layer_present (node, b0, frag_hdr))
 		{
-		  tmp = ip6_ext_next_header (tmp);
+		  next0 = IP6_REASSEMBLY_NEXT_ICMP_ERROR;
+		  goto skip_reass;
 		}
-	      if (IP_PROTOCOL_IP6_NONXT == tmp->next_hdr)
-		{
-		  icmp6_error_set_vnet_buffer (b0, ICMP6_parameter_problem,
-					       ICMP6_parameter_problem_first_fragment_has_incomplete_header_chain,
-					       0);
-		  b0->error = node->errors[IP6_ERROR_REASS_MISSING_UPPER];
-
-		  to_next[0] = bi0;
-		  to_next += 1;
-		  n_left_to_next -= 1;
-		  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
-						   to_next, n_left_to_next,
-						   bi0,
-						   IP6_REASSEMBLY_NEXT_ICMP_ERROR);
-		  goto next;
-		}
+	    }
+	  if (!ip6_reass_verify_fragment_multiple_8 (vm, node, b0, frag_hdr)
+	      || !ip6_reass_verify_packet_size_lt_64k (vm, node, b0,
+						       frag_hdr))
+	    {
+	      next0 = IP6_REASSEMBLY_NEXT_ICMP_ERROR;
+	      goto skip_reass;
 	    }
 	  vnet_buffer (b0)->ip.reass.ip6_frag_hdr_offset =
 	    (u8 *) frag_hdr - (u8 *) ip0;
@@ -894,19 +962,15 @@
 	  k.frag_id = frag_hdr->identification;
 	  k.proto = ip0->protocol;
 	  k.unused = 0;
-	  u32 icmp_bi = ~0;
 	  ip6_reass_t *reass =
-	    ip6_reass_find_or_create (vm, node, rm, &k, &icmp_bi,
+	    ip6_reass_find_or_create (vm, node, rm, rt, &k, &icmp_bi,
 				      &vec_timeout);
 
-	  u32 error0 = IP6_ERROR_NONE;
 	  if (reass)
 	    {
-	      ip6_reass_update (vm, node, rm, reass, &bi0, &next0, &error0,
-				frag_hdr, &vec_drop_overlap,
-				&vec_drop_compress, IP6_REASSEMBLY_NEXT_INPUT,
-				IP6_REASSEMBLY_NEXT_DROP,
-				IP6_REASSEMBLY_NEXT_ICMP_ERROR);
+	      ip6_reass_update (vm, node, rm, rt, reass, &bi0, &next0,
+				&error0, frag_hdr, &vec_drop_overlap,
+				&vec_drop_compress, is_feature);
 	    }
 	  else
 	    {
@@ -918,9 +982,15 @@
 
 	  if (~0 != bi0)
 	    {
+	    skip_reass:
 	      to_next[0] = bi0;
 	      to_next += 1;
 	      n_left_to_next -= 1;
+	      if (is_feature && IP6_ERROR_NONE == error0)
+		{
+		  vnet_feature_next (vnet_buffer (b0)->sw_if_index[VLIB_RX],
+				     &next0, b0);
+		}
 	      vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
 					       n_left_to_next, bi0, next0);
 	    }
@@ -935,7 +1005,6 @@
 					       n_left_to_next, icmp_bi,
 					       next0);
 	    }
-	next:
 	  from += 1;
 	  n_left_from -= 1;
 	}
@@ -943,6 +1012,7 @@
       vlib_put_next_frame (vm, node, next_index, n_left_to_next);
     }
 
+  clib_spinlock_unlock (&rt->lock);
   return frame->n_vectors;
 }
 
@@ -952,6 +1022,13 @@
 #undef _
 };
 
+always_inline uword
+ip6_reassembly (vlib_main_t * vm, vlib_node_runtime_t * node,
+		vlib_frame_t * frame)
+{
+  return ip6_reassembly_inline (vm, node, frame, false /* is_feature */ );
+}
+
 /* *INDENT-OFF* */
 VLIB_REGISTER_NODE (ip6_reass_node, static) = {
     .function = ip6_reassembly,
@@ -970,8 +1047,46 @@
 };
 /* *INDENT-ON* */
 
-VLIB_NODE_FUNCTION_MULTIARCH (ip6_reass_node, ip6_reassembly)
-     static u32 ip6_reass_get_nbuckets ()
+VLIB_NODE_FUNCTION_MULTIARCH (ip6_reass_node, ip6_reassembly);
+
+always_inline uword
+ip6_reassembly_feature (vlib_main_t * vm,
+			vlib_node_runtime_t * node, vlib_frame_t * frame)
+{
+  return ip6_reassembly_inline (vm, node, frame, true /* is_feature */ );
+}
+
+/* *INDENT-OFF* */
+VLIB_REGISTER_NODE (ip6_reass_node_feature, static) = {
+    .function = ip6_reassembly_feature,
+    .name = "ip6-reassembly-feature",
+    .vector_size = sizeof (u32),
+    .format_trace = format_ip6_reass_trace,
+    .n_errors = ARRAY_LEN (ip6_reassembly_error_strings),
+    .error_strings = ip6_reassembly_error_strings,
+    .n_next_nodes = IP6_REASSEMBLY_N_NEXT,
+    .next_nodes =
+        {
+                [IP6_REASSEMBLY_NEXT_INPUT] = "ip6-input",
+                [IP6_REASSEMBLY_NEXT_DROP] = "ip6-drop",
+                [IP6_REASSEMBLY_NEXT_ICMP_ERROR] = "ip6-icmp-error",
+        },
+};
+/* *INDENT-ON* */
+
+VLIB_NODE_FUNCTION_MULTIARCH (ip6_reass_node_feature, ip6_reassembly_feature);
+
+/* *INDENT-OFF* */
+VNET_FEATURE_INIT (ip6_reassembly_feature, static) = {
+    .arc_name = "ip6-unicast",
+    .node_name = "ip6-reassembly-feature",
+    .runs_before = VNET_FEATURES ("ip6-lookup"),
+    .runs_after = 0,
+};
+/* *INDENT-ON* */
+
+static u32
+ip6_reass_get_nbuckets ()
 {
   ip6_reass_main_t *rm = &ip6_reass_main;
   u32 nbuckets;
@@ -998,7 +1113,7 @@
   clib_bihash_48_8_t *new_hash;
 } ip6_rehash_cb_ctx;
 
-void
+static void
 ip6_rehash_cb (clib_bihash_kv_48_8_t * kv, void *_ctx)
 {
   ip6_rehash_cb_ctx *ctx = _ctx;
@@ -1008,15 +1123,23 @@
     }
 }
 
+static void
+ip6_reass_set_params (u32 timeout_ms, u32 max_reassemblies,
+		      u32 expire_walk_interval_ms)
+{
+  ip6_reass_main.timeout_ms = timeout_ms;
+  ip6_reass_main.timeout = (f64) timeout_ms / (f64) MSEC_PER_SEC;
+  ip6_reass_main.max_reass_n = max_reassemblies;
+  ip6_reass_main.expire_walk_interval_ms = expire_walk_interval_ms;
+}
+
 vnet_api_error_t
 ip6_reass_set (u32 timeout_ms, u32 max_reassemblies,
 	       u32 expire_walk_interval_ms)
 {
   u32 old_nbuckets = ip6_reass_get_nbuckets ();
-  ip6_reass_main.timeout_ms = timeout_ms;
-  ip6_reass_main.timeout = (f64) timeout_ms / (f64) MSEC_PER_SEC;
-  ip6_reass_main.max_reass_n = max_reassemblies;
-  ip6_reass_main.expire_walk_interval_ms = expire_walk_interval_ms;
+  ip6_reass_set_params (timeout_ms, max_reassemblies,
+			expire_walk_interval_ms);
   vlib_process_signal_event (ip6_reass_main.vlib_main,
 			     ip6_reass_main.ip6_reass_expire_node_idx,
 			     IP6_EVENT_CONFIG_CHANGED, 0);
@@ -1058,7 +1181,7 @@
   return 0;
 }
 
-clib_error_t *
+static clib_error_t *
 ip6_reass_init_function (vlib_main_t * vm)
 {
   ip6_reass_main_t *rm = &ip6_reass_main;
@@ -1069,17 +1192,18 @@
   rm->vlib_main = vm;
   rm->vnet_main = vnet_get_main ();
 
-  rm->reass_n = 0;
-  pool_alloc (rm->pool, rm->max_reass_n);
+  vec_validate (rm->per_thread_data, vlib_num_workers () + 1);
+  ip6_reass_per_thread_t *rt;
+  vec_foreach (rt, rm->per_thread_data)
+  {
+    clib_spinlock_init (&rt->lock);
+    pool_alloc (rt->pool, rm->max_reass_n);
+  }
 
   node = vlib_get_node_by_name (vm, (u8 *) "ip6-reassembly-expire-walk");
   ASSERT (node);
   rm->ip6_reass_expire_node_idx = node->index;
 
-  ip6_reass_set (IP6_REASS_TIMEOUT_DEFAULT_MS,
-		 IP6_REASS_MAX_REASSEMBLIES_DEAFULT,
-		 IP6_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS);
-
   nbuckets = ip6_reass_get_nbuckets ();
   clib_bihash_init_48_8 (&rm->hash, "ip6-reass", nbuckets, nbuckets * 1024);
 
@@ -1094,6 +1218,11 @@
     return error;
   ip6_register_protocol (IP_PROTOCOL_IPV6_FRAGMENTATION,
 			 ip6_reass_node.index);
+
+  ip6_reass_set_params (IP6_REASS_TIMEOUT_DEFAULT_MS,
+			IP6_REASS_MAX_REASSEMBLIES_DEFAULT,
+			IP6_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS);
+
   return error;
 }
 
@@ -1130,38 +1259,65 @@
       u32 *vec_timeout = NULL;
       int *pool_indexes_to_free = NULL;
 
+      uword thread_index = 0;
       int index;
-      /* *INDENT-OFF* */
-      pool_foreach_index (index, rm->pool, ({
-                            reass = pool_elt_at_index (rm->pool, index);
-                            if (now > reass->last_heard + rm->timeout)
-                              {
-                                vec_add1 (pool_indexes_to_free, index);
-                              }
-                          }));
-      /* *INDENT-ON* */
-      int *i;
+      const uword nthreads = os_get_nthreads ();
       u32 *vec_icmp_bi = NULL;
-      /* *INDENT-OFF* */
-      vec_foreach (i, pool_indexes_to_free)
-      {
-        ip6_reass_t *reass = pool_elt_at_index (rm->pool, i[0]);
-        u32 icmp_bi = ~0;
-        ip6_reass_on_timeout (vm, node, rm, reass, &icmp_bi, &vec_timeout);
-        if (~0 != icmp_bi)
+      for (thread_index = 0; thread_index < nthreads; ++thread_index)
+	{
+	  ip6_reass_per_thread_t *rt = &rm->per_thread_data[thread_index];
+	  clib_spinlock_lock (&rt->lock);
+
+	  vec_reset_length (pool_indexes_to_free);
+          /* *INDENT-OFF* */
+          pool_foreach_index (index, rt->pool, ({
+                                reass = pool_elt_at_index (rt->pool, index);
+                                if (now > reass->last_heard + rm->timeout)
+                                  {
+                                    vec_add1 (pool_indexes_to_free, index);
+                                  }
+                              }));
+          /* *INDENT-ON* */
+	  int *i;
+          /* *INDENT-OFF* */
+          vec_foreach (i, pool_indexes_to_free)
           {
-            vec_add1 (vec_icmp_bi, icmp_bi);
+            ip6_reass_t *reass = pool_elt_at_index (rt->pool, i[0]);
+            u32 icmp_bi = ~0;
+            u32 before = vec_len (vec_timeout);
+            vlib_buffer_t *b = vlib_get_buffer (vm, reass->first_bi);
+            if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED))
+              {
+                if (pool_is_free_index (vm->trace_main.trace_buffer_pool,
+                                        b->trace_index))
+                  {
+                    /* the trace is gone, don't trace this buffer anymore */
+                    b->flags &= ~VLIB_BUFFER_IS_TRACED;
+                  }
+              }
+            ip6_reass_on_timeout (vm, node, rm, reass, &icmp_bi, &vec_timeout);
+            u32 after = vec_len (vec_timeout);
+            ASSERT (rt->buffers_n >= (after - before));
+            rt->buffers_n -= (after - before);
+            if (~0 != icmp_bi)
+              {
+                vec_add1 (vec_icmp_bi, icmp_bi);
+                ASSERT (rt->buffers_n > 0);
+                --rt->buffers_n;
+              }
+            ip6_reass_free (rm, rt, reass);
           }
-        ip6_reass_free (rm, reass);
-      }
-      /* *INDENT-ON* */
+          /* *INDENT-ON* */
+
+	  clib_spinlock_unlock (&rt->lock);
+	}
 
       while (vec_len (vec_timeout) > 0)
 	{
 	  vlib_frame_t *f = vlib_get_frame_to_node (vm, rm->ip6_drop_idx);
 	  u32 *to_next = vlib_frame_vector_args (f);
 	  u32 n_left_to_next = VLIB_FRAME_SIZE - f->n_vectors;
-	  u32 n_trace = 0;
+	  int trace_frame = 0;
 	  while (vec_len (vec_timeout) > 0 && n_left_to_next > 0)
 	    {
 	      u32 bi = vec_pop (vec_timeout);
@@ -1176,7 +1332,7 @@
 		    }
 		  else
 		    {
-		      ++n_trace;
+		      trace_frame = 1;
 		    }
 		}
 	      b->error = node->errors[IP6_ERROR_REASS_TIMEOUT];
@@ -1184,13 +1340,8 @@
 	      ++f->n_vectors;
 	      to_next += 1;
 	      n_left_to_next -= 1;
-	      ASSERT (rm->buffers_n > 0);
-	      --rm->buffers_n;
 	    }
-	  if (PREDICT_FALSE (n_trace > 0))
-	    {
-	      f->flags |= VLIB_FRAME_TRACE;
-	    }
+	  f->flags |= (trace_frame * VLIB_FRAME_TRACE);
 	  vlib_put_frame_to_node (vm, rm->ip6_drop_idx, f);
 	}
 
@@ -1200,7 +1351,7 @@
 	    vlib_get_frame_to_node (vm, rm->ip6_icmp_error_idx);
 	  u32 *to_next = vlib_frame_vector_args (f);
 	  u32 n_left_to_next = VLIB_FRAME_SIZE - f->n_vectors;
-	  u32 n_trace = 0;
+	  int trace_frame = 0;
 	  while (vec_len (vec_icmp_bi) > 0 && n_left_to_next > 0)
 	    {
 	      u32 bi = vec_pop (vec_icmp_bi);
@@ -1215,7 +1366,7 @@
 		    }
 		  else
 		    {
-		      ++n_trace;
+		      trace_frame = 1;
 		    }
 		}
 	      b->error = node->errors[IP6_ERROR_REASS_TIMEOUT];
@@ -1223,13 +1374,8 @@
 	      ++f->n_vectors;
 	      to_next += 1;
 	      n_left_to_next -= 1;
-	      ASSERT (rm->buffers_n > 0);
-	      --rm->buffers_n;
 	    }
-	  if (PREDICT_FALSE (n_trace > 0))
-	    {
-	      f->flags |= VLIB_FRAME_TRACE;
-	    }
+	  f->flags |= (trace_frame * VLIB_FRAME_TRACE);
 	  vlib_put_frame_to_node (vm, rm->ip6_icmp_error_idx, f);
 	}
 
@@ -1276,7 +1422,7 @@
   vlib_main_t *vm = va_arg (*args, vlib_main_t *);
   ip6_reass_t *reass = va_arg (*args, ip6_reass_t *);
 
-  s = format (s, "ID: %u, key: %U\n  first_bi: %u, data_len: %u, "
+  s = format (s, "ID: %lu, key: %U\n  first_bi: %u, data_len: %u, "
 	      "last_packet_octet: %u, trace_op_counter: %u\n",
 	      reass->id, format_ip6_reass_key, &reass->key, reass->first_bi,
 	      reass->data_len, reass->last_packet_octet,
@@ -1315,22 +1461,40 @@
   vlib_cli_output (vm, "---------------------");
   vlib_cli_output (vm, "IP6 reassembly status");
   vlib_cli_output (vm, "---------------------");
+  bool details = false;
   if (unformat (input, "details"))
     {
-      ip6_reass_t *reass;
-      /* *INDENT-OFF* */
-      pool_foreach (reass, rm->pool, {
-        vlib_cli_output (vm, "%U", format_ip6_reass, vm, reass);
-      });
-      /* *INDENT-ON* */
+      details = true;
+    }
+
+  u32 sum_reass_n = 0;
+  u64 sum_buffers_n = 0;
+  ip6_reass_t *reass;
+  uword thread_index;
+  const uword nthreads = os_get_nthreads ();
+  for (thread_index = 0; thread_index < nthreads; ++thread_index)
+    {
+      ip6_reass_per_thread_t *rt = &rm->per_thread_data[thread_index];
+      clib_spinlock_lock (&rt->lock);
+      if (details)
+	{
+          /* *INDENT-OFF* */
+          pool_foreach (reass, rt->pool, {
+            vlib_cli_output (vm, "%U", format_ip6_reass, vm, reass);
+          });
+          /* *INDENT-ON* */
+	}
+      sum_reass_n += rt->reass_n;
+      sum_buffers_n += rt->buffers_n;
+      clib_spinlock_unlock (&rt->lock);
     }
   vlib_cli_output (vm, "---------------------");
-  vlib_cli_output (vm, "Current IP6 reassemblies count: %lu\n", rm->reass_n);
-  vlib_cli_output (vm,
-		   "Maximum configured concurrent IP6 reassemblies: %lu\n",
-		   (long unsigned) rm->max_reass_n);
+  vlib_cli_output (vm, "Current IP6 reassemblies count: %lu\n",
+		   (long unsigned) sum_reass_n);
+  vlib_cli_output (vm, "Maximum configured concurrent IP6 reassemblies per "
+		   "worker-thread: %lu\n", (long unsigned) rm->max_reass_n);
   vlib_cli_output (vm, "Buffers in use: %lu\n",
-		   (long unsigned) rm->buffers_n);
+		   (long unsigned) sum_buffers_n);
   return 0;
 }
 
@@ -1342,6 +1506,13 @@
 };
 /* *INDENT-ON* */
 
+vnet_api_error_t
+ip6_reass_enable_disable (u32 sw_if_index, u8 enable_disable)
+{
+  return vnet_feature_enable_disable ("ip6-unicast", "ip6-reassembly-feature",
+				      sw_if_index, enable_disable, 0, 0);
+}
+
 /*
  * fd.io coding-style-patch-verification: ON
  *
diff --git a/src/vnet/ip/ip6_reassembly.h b/src/vnet/ip/ip6_reassembly.h
index 0de4e04..5084eda 100644
--- a/src/vnet/ip/ip6_reassembly.h
+++ b/src/vnet/ip/ip6_reassembly.h
@@ -38,6 +38,9 @@
 vnet_api_error_t ip6_reass_get (u32 * timeout_ms, u32 * max_reassemblies,
 				u32 * expire_walk_interval_ms);
 
+vnet_api_error_t ip6_reass_enable_disable (u32 sw_if_index,
+					   u8 enable_disable);
+
 #endif /* __included_ip6_reassembly_h */
 
 /*
diff --git a/src/vnet/ip/ip_api.c b/src/vnet/ip/ip_api.c
index 0a05511..a8dbb3a 100644
--- a/src/vnet/ip/ip_api.c
+++ b/src/vnet/ip/ip_api.c
@@ -103,7 +103,8 @@
 _(IP_SOURCE_AND_PORT_RANGE_CHECK_INTERFACE_ADD_DEL,                     \
   ip_source_and_port_range_check_interface_add_del)                     \
 _(IP_REASSEMBLY_SET, ip_reassembly_set)                                 \
-_(IP_REASSEMBLY_GET, ip_reassembly_get)
+_(IP_REASSEMBLY_GET, ip_reassembly_get)                                 \
+_(IP_REASSEMBLY_ENABLE_DISABLE, ip_reassembly_enable_disable)
 
 extern void stats_dslock_with_hint (int hint, int tag);
 extern void stats_dsunlock (void);
@@ -2904,6 +2905,23 @@
   vl_msg_api_send_shmem (q, (u8 *) & rmp);
 }
 
+void
+  vl_api_ip_reassembly_enable_disable_t_handler
+  (vl_api_ip_reassembly_enable_disable_t * mp)
+{
+  vl_api_ip_reassembly_enable_disable_reply_t *rmp;
+  int rv = 0;
+  rv = ip4_reass_enable_disable (clib_net_to_host_u32 (mp->sw_if_index),
+				 mp->enable_ip4);
+  if (0 == rv)
+    {
+      rv = ip6_reass_enable_disable (clib_net_to_host_u32 (mp->sw_if_index),
+				     mp->enable_ip6);
+    }
+
+  REPLY_MACRO (VL_API_IP_REASSEMBLY_SET_REPLY);
+}
+
 #define vl_msg_name_crc_list
 #include <vnet/ip/ip.api.h>
 #undef vl_msg_name_crc_list
diff --git a/src/vnet/ip/lookup.c b/src/vnet/ip/lookup.c
index 3bbf2ab..216af4c 100644
--- a/src/vnet/ip/lookup.c
+++ b/src/vnet/ip/lookup.c
@@ -230,8 +230,6 @@
       }
 
     lm->local_next_by_ip_protocol[IP_PROTOCOL_UDP] = IP_LOCAL_NEXT_UDP_LOOKUP;
-    lm->local_next_by_ip_protocol[IP_PROTOCOL_VPP_FRAGMENTATION] =
-      IP_LOCAL_NEXT_REASSEMBLY;
     lm->local_next_by_ip_protocol[is_ip6 ? IP_PROTOCOL_ICMP6 :
 				  IP_PROTOCOL_ICMP] = IP_LOCAL_NEXT_ICMP;
     lm->builtin_protocol_by_ip_protocol[IP_PROTOCOL_UDP] =
diff --git a/src/vnet/ip/lookup.h b/src/vnet/ip/lookup.h
index 9fe7a1b..a89546f 100644
--- a/src/vnet/ip/lookup.h
+++ b/src/vnet/ip/lookup.h
@@ -111,7 +111,6 @@
   IP_LOCAL_NEXT_PUNT,
   IP_LOCAL_NEXT_UDP_LOOKUP,
   IP_LOCAL_NEXT_ICMP,
-  IP_LOCAL_NEXT_REASSEMBLY,
   IP_LOCAL_N_NEXT,
 } ip_local_next_t;
 
diff --git a/test/framework.py b/test/framework.py
index 423feaf..2167399 100644
--- a/test/framework.py
+++ b/test/framework.py
@@ -377,6 +377,10 @@
             try:
                 cls.vapi.connect()
             except:
+                try:
+                    cls.vapi.disconnect()
+                except:
+                    pass
                 if cls.debug_gdbserver:
                     print(colorize("You're running VPP inside gdbserver but "
                                    "VPP-API connection failed, did you forget "
diff --git a/test/sys_req/set_system_parameters.sh b/test/sys_req/set_system_parameters.sh
deleted file mode 100755
index aa23634..0000000
--- a/test/sys_req/set_system_parameters.sh
+++ /dev/null
@@ -1,68 +0,0 @@
-#!/bin/bash
-
-file="$1"
-
-usage(){
-	echo "Usage: $0 <requirements file>"
-}
-
-if [ "$file" == "" ]
-then
-	echo "Invalid parameters specified."
-	usage
-	exit 1
-fi
-
-if [ ! -f $file ]
-then
-	echo "File '$file' does not exist."
-	usage
-	exit 1
-fi
-
-if test "$DOCKER_TEST" = "True"
-then
-	echo "=============================================================================="
-	echo "DOCKER_TEST is set to '$DOCKER_TEST'."
-	echo "Skipping verification of some system parameters."
-	echo "Make sure these are set properly, otherwise tests might fail."
-	echo "Required values/criteria are in '`readlink -e $file`'."
-	echo "=============================================================================="
-	exit 0
-fi
-
-cat $file | grep -v -e '^#.*$' | grep -v -e '^ *$' | while read line
-do
-	value_file=`echo $line | awk '{print $1}'`
-	operator=`echo $line | awk '{print $2}'`
-	value=`echo $line | awk '{print $3}'`
-	set_value=`echo $line | awk '{print $4}'`
-	if [[ "$value_file" == "" || "$operator" == "" || "$value" == "" || "$set_value" == "" ]]
-	then
-		echo "Syntax error in requirements file."
-		exit 1
-	fi
-	current_value=`cat $value_file`
-	if test "$current_value" $operator "$value"
-	then
-		if test "$V" = "2"
-		then
-			echo "Requirement '$value_file $operator $value' satisfied."
-		fi
-	else
-		echo "Requirement '$value_file $operator $value' not satisfied."
-		echo "Writing '$set_value' to '$value_file'."
-		echo "$set_value" | tee "$value_file" > /dev/null
-		if ! test "`cat $value_file`" = "$set_value"
-		then
-			echo "Repeating the write using sudo..."
-			echo "$set_value" | sudo -n tee "$value_file" > /dev/null
-			if ! test "`cat $value_file`" = "$set_value"
-			then
-				echo "Couldn't set the required value. Is that value allowed? Is sudo working?"
-				exit 1
-			fi
-		fi
-		echo "Succesfully wrote '$set_value' to '$value_file'."
-	fi
-done
diff --git a/test/sys_req/system_parameters b/test/sys_req/system_parameters
deleted file mode 100644
index 6373774..0000000
--- a/test/sys_req/system_parameters
+++ /dev/null
@@ -1,18 +0,0 @@
-# test framework system requirements
-# format of this file is
-# <path> <operator> <comparison-value> <set-value>
-#
-# path - path to value e.g. in /proc which needs to be checked
-# operator - test operator (e.g. -gt)
-# comparison-value - value, against which the value read from <path> is compared
-# set-value - value, to which the path is set if the test fails
-#
-# the comparison is done using `test' command
-
-
-# test_reassembly.py
-# needed by test_reassembly which uses udp punt via unix domain sockets
-# to ensure that all data which vpp might produce in a burst fits into
-# the socket send buffer
-/proc/sys/net/core/wmem_max -ge 4636252 4636252
-/proc/sys/net/core/wmem_default -ge 4636252 4636252
diff --git a/test/test_reassembly.py.disabled b/test/test_reassembly.py
similarity index 73%
rename from test/test_reassembly.py.disabled
rename to test/test_reassembly.py
index a2d77d4..531caa4 100644
--- a/test/test_reassembly.py.disabled
+++ b/test/test_reassembly.py
@@ -8,7 +8,6 @@
 from scapy.layers.l2 import Ether, GRE
 from scapy.layers.inet import IP, UDP
 from util import ppp, fragment_rfc791, fragment_rfc8200
-from vpp_punt_socket import VppUDSPuntSocket
 from scapy.layers.inet6 import IPv6, IPv6ExtHdrFragment, ICMPv6ParamProblem,\
     ICMPv6TimeExceeded
 from vpp_gre_interface import VppGreInterface, VppGre6Interface
@@ -24,8 +23,9 @@
     def setUpClass(cls):
         super(TestIPv4Reassembly, cls).setUpClass()
 
-        cls.create_pg_interfaces([0])
-        cls.pg_if = cls.pg0
+        cls.create_pg_interfaces([0, 1])
+        cls.src_if = cls.pg0
+        cls.dst_if = cls.pg1
 
         # setup all interfaces
         for i in cls.pg_interfaces:
@@ -33,9 +33,6 @@
             i.config_ip4()
             i.resolve_arp()
 
-        cls.punt_port = 9999
-        cls.punt_socket = VppUDSPuntSocket(cls, cls.punt_port)
-
         # packet sizes
         cls.packet_sizes = [64, 512, 1518, 9018]
         cls.padding = " abcdefghijklmn"
@@ -45,6 +42,8 @@
     def setUp(self):
         """ Test setup - force timeout on existing reassemblies """
         super(TestIPv4Reassembly, self).setUp()
+        self.vapi.ip_reassembly_enable_disable(
+            sw_if_index=self.src_if.sw_if_index, enable_ip4=True)
         self.vapi.ip_reassembly_set(timeout_ms=0, max_reassemblies=1000,
                                     expire_walk_interval_ms=10)
         self.sleep(.25)
@@ -62,12 +61,12 @@
         :param list packet_sizes: Required packet sizes.
         """
         for i in range(0, packet_count):
-            info = cls.create_packet_info(cls.pg_if, cls.pg_if)
+            info = cls.create_packet_info(cls.src_if, cls.src_if)
             payload = cls.info_to_payload(info)
-            p = (Ether(dst=cls.pg_if.local_mac, src=cls.pg_if.remote_mac) /
-                 IP(id=info.index, src=cls.pg_if.remote_ip4,
-                    dst=cls.pg_if.local_ip4) /
-                 UDP(sport=1234, dport=cls.punt_port) /
+            p = (Ether(dst=cls.src_if.local_mac, src=cls.src_if.remote_mac) /
+                 IP(id=info.index, src=cls.src_if.remote_ip4,
+                    dst=cls.dst_if.remote_ip4) /
+                 UDP(sport=1234, dport=5678) /
                  Raw(payload))
             size = packet_sizes[(i // 2) % len(packet_sizes)]
             cls.extend_packet(p, size, cls.padding)
@@ -79,7 +78,7 @@
         cls.pkt_infos = []
         for index, info in infos.iteritems():
             p = info.data
-            # self.logger.debug(ppp("Packet:", p.__class__(str(p))))
+            # cls.logger.debug(ppp("Packet:", p.__class__(str(p))))
             fragments_400 = fragment_rfc791(p, 400)
             fragments_300 = fragment_rfc791(p, 300)
             fragments_200 = [
@@ -106,11 +105,7 @@
         seen = set()
         for packet in capture:
             try:
-                sw_if_index = packet['sw_if_index']
-                punt_action = packet['punt_action']
-                packet = Ether(packet['packet'])
-                self.logger.debug(ppp("Got packet from %s, action %s" %
-                                      (sw_if_index, punt_action), packet))
+                self.logger.debug(ppp("Got packet:", packet))
                 ip = packet[IP]
                 udp = packet[UDP]
                 payload_info = self.payload_to_info(str(packet[Raw]))
@@ -121,7 +116,7 @@
                 if packet_index in seen:
                     raise Exception(ppp("Duplicate packet received", packet))
                 seen.add(packet_index)
-                self.assertEqual(payload_info.dst, self.pg_if.sw_if_index)
+                self.assertEqual(payload_info.dst, self.src_if.sw_if_index)
                 info = self._packet_infos[packet_index]
                 self.assertTrue(info is not None)
                 self.assertEqual(packet_index, info.index)
@@ -129,7 +124,7 @@
                 self.assertEqual(ip.src, saved_packet[IP].src)
                 self.assertEqual(ip.dst, saved_packet[IP].dst)
                 self.assertEqual(udp.payload, saved_packet[UDP].payload)
-            except:
+            except Exception:
                 self.logger.error(ppp("Unexpected or invalid packet:", packet))
                 raise
         for index in self._packet_infos:
@@ -140,21 +135,21 @@
         """ basic reassembly """
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(self.fragments_200)
+        self.src_if.add_stream(self.fragments_200)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
         # run it all again to verify correctness
         self.pg_enable_capture()
-        self.pg_if.add_stream(self.fragments_200)
+        self.src_if.add_stream(self.fragments_200)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_reversed(self):
         """ reverse order reassembly """
@@ -163,21 +158,21 @@
         fragments.reverse()
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.packet_infos))
+        packets = self.dst_if.get_capture(len(self.packet_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
         # run it all again to verify correctness
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.packet_infos))
+        packets = self.dst_if.get_capture(len(self.packet_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_random(self):
         """ random order reassembly """
@@ -186,21 +181,21 @@
         shuffle(fragments)
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.packet_infos))
+        packets = self.dst_if.get_capture(len(self.packet_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
         # run it all again to verify correctness
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.packet_infos))
+        packets = self.dst_if.get_capture(len(self.packet_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_duplicates(self):
         """ duplicate fragments """
@@ -212,12 +207,12 @@
         ]
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_overlap1(self):
         """ overlapping fragments case #1 """
@@ -232,21 +227,21 @@
                     fragments.extend(j)
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
         # run it all to verify correctness
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_overlap2(self):
         """ overlapping fragments case #2 """
@@ -267,21 +262,21 @@
                 fragments.append(zipped[-1][0])
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
         # run it all to verify correctness
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_timeout_inline(self):
         """ timeout (inline) """
@@ -294,13 +289,13 @@
                                     expire_walk_interval_ms=10000)
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(self.fragments_400)
+        self.src_if.add_stream(self.fragments_400)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(
+        packets = self.dst_if.get_capture(
             len(self.pkt_infos) - len(dropped_packet_indexes))
         self.verify_capture(packets, dropped_packet_indexes)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_timeout_cleanup(self):
         """ timeout (cleanup) """
@@ -324,19 +319,18 @@
                                     expire_walk_interval_ms=50)
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
         self.sleep(.25, "wait before sending rest of fragments")
 
-        self.pg_if.add_stream(fragments2)
+        self.src_if.add_stream(fragments2)
         self.pg_start()
-        self.sleep(.25, "wait for vpp to process packets")
 
-        packets = self.punt_socket.wait_for_packets(
+        packets = self.dst_if.get_capture(
             len(self.pkt_infos) - len(dropped_packet_indexes))
         self.verify_capture(packets, dropped_packet_indexes)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_disabled(self):
         """ reassembly disabled """
@@ -349,13 +343,13 @@
                                     expire_walk_interval_ms=10000)
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(self.fragments_400)
+        self.src_if.add_stream(self.fragments_400)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(
+        packets = self.dst_if.get_capture(
             len(self.pkt_infos) - len(dropped_packet_indexes))
         self.verify_capture(packets, dropped_packet_indexes)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
 
 class TestIPv6Reassembly(VppTestCase):
@@ -365,8 +359,9 @@
     def setUpClass(cls):
         super(TestIPv6Reassembly, cls).setUpClass()
 
-        cls.create_pg_interfaces([0])
-        cls.pg_if = cls.pg0
+        cls.create_pg_interfaces([0, 1])
+        cls.src_if = cls.pg0
+        cls.dst_if = cls.pg1
 
         # setup all interfaces
         for i in cls.pg_interfaces:
@@ -374,9 +369,6 @@
             i.config_ip6()
             i.resolve_ndp()
 
-        cls.punt_port = 9999
-        cls.punt_socket = VppUDSPuntSocket(cls, cls.punt_port, is_ip4=0)
-
         # packet sizes
         cls.packet_sizes = [64, 512, 1518, 9018]
         cls.padding = " abcdefghijklmn"
@@ -386,11 +378,14 @@
     def setUp(self):
         """ Test setup - force timeout on existing reassemblies """
         super(TestIPv6Reassembly, self).setUp()
+        self.vapi.ip_reassembly_enable_disable(
+            sw_if_index=self.src_if.sw_if_index, enable_ip6=True)
         self.vapi.ip_reassembly_set(timeout_ms=0, max_reassemblies=1000,
                                     expire_walk_interval_ms=10, is_ip6=1)
         self.sleep(.25)
         self.vapi.ip_reassembly_set(timeout_ms=1000000, max_reassemblies=1000,
                                     expire_walk_interval_ms=10000, is_ip6=1)
+        self.logger.debug(self.vapi.ppcli("show ip6-reassembly details"))
 
     def tearDown(self):
         super(TestIPv6Reassembly, self).tearDown()
@@ -403,12 +398,12 @@
         :param list packet_sizes: Required packet sizes.
         """
         for i in range(0, packet_count):
-            info = cls.create_packet_info(cls.pg_if, cls.pg_if)
+            info = cls.create_packet_info(cls.src_if, cls.src_if)
             payload = cls.info_to_payload(info)
-            p = (Ether(dst=cls.pg_if.local_mac, src=cls.pg_if.remote_mac) /
-                 IPv6(src=cls.pg_if.remote_ip6,
-                      dst=cls.pg_if.local_ip6) /
-                 UDP(sport=1234, dport=cls.punt_port) /
+            p = (Ether(dst=cls.src_if.local_mac, src=cls.src_if.remote_mac) /
+                 IPv6(src=cls.src_if.remote_ip6,
+                      dst=cls.dst_if.remote_ip6) /
+                 UDP(sport=1234, dport=5678) /
                  Raw(payload))
             size = packet_sizes[(i // 2) % len(packet_sizes)]
             cls.extend_packet(p, size, cls.padding)
@@ -420,7 +415,7 @@
         cls.pkt_infos = []
         for index, info in infos.iteritems():
             p = info.data
-            # self.logger.debug(ppp("Packet:", p.__class__(str(p))))
+            # cls.logger.debug(ppp("Packet:", p.__class__(str(p))))
             fragments_400 = fragment_rfc8200(p, info.index, 400)
             fragments_300 = fragment_rfc8200(p, info.index, 300)
             cls.pkt_infos.append((index, fragments_400, fragments_300))
@@ -442,11 +437,7 @@
         seen = set()
         for packet in capture:
             try:
-                sw_if_index = packet['sw_if_index']
-                punt_action = packet['punt_action']
-                packet = Ether(packet['packet'])
-                self.logger.debug(ppp("Got packet from %s, action %s" %
-                                      (sw_if_index, punt_action), packet))
+                self.logger.debug(ppp("Got packet:", packet))
                 ip = packet[IPv6]
                 udp = packet[UDP]
                 payload_info = self.payload_to_info(str(packet[Raw]))
@@ -457,7 +448,7 @@
                 if packet_index in seen:
                     raise Exception(ppp("Duplicate packet received", packet))
                 seen.add(packet_index)
-                self.assertEqual(payload_info.dst, self.pg_if.sw_if_index)
+                self.assertEqual(payload_info.dst, self.src_if.sw_if_index)
                 info = self._packet_infos[packet_index]
                 self.assertTrue(info is not None)
                 self.assertEqual(packet_index, info.index)
@@ -465,7 +456,7 @@
                 self.assertEqual(ip.src, saved_packet[IPv6].src)
                 self.assertEqual(ip.dst, saved_packet[IPv6].dst)
                 self.assertEqual(udp.payload, saved_packet[UDP].payload)
-            except:
+            except Exception:
                 self.logger.error(ppp("Unexpected or invalid packet:", packet))
                 raise
         for index in self._packet_infos:
@@ -476,21 +467,21 @@
         """ basic reassembly """
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(self.fragments_400)
+        self.src_if.add_stream(self.fragments_400)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
         # run it all again to verify correctness
         self.pg_enable_capture()
-        self.pg_if.add_stream(self.fragments_400)
+        self.src_if.add_stream(self.fragments_400)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_reversed(self):
         """ reverse order reassembly """
@@ -499,21 +490,21 @@
         fragments.reverse()
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
         # run it all again to verify correctness
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_random(self):
         """ random order reassembly """
@@ -522,21 +513,21 @@
         shuffle(fragments)
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
         # run it all again to verify correctness
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_duplicates(self):
         """ duplicate fragments """
@@ -548,12 +539,12 @@
         ]
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(len(self.pkt_infos))
+        packets = self.dst_if.get_capture(len(self.pkt_infos))
         self.verify_capture(packets)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_overlap1(self):
         """ overlapping fragments case #1 """
@@ -572,20 +563,19 @@
         )
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        self.sleep(.1, "wait for vpp to process packets")
-        packets = self.punt_socket.wait_for_packets(
+        packets = self.dst_if.get_capture(
             len(self.pkt_infos) - len(dropped_packet_indexes))
         self.verify_capture(packets, dropped_packet_indexes)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_overlap2(self):
         """ overlapping fragments case #2 """
 
         fragments = []
-        for _, frags_400, frags_30 in self.pkt_infos:
+        for _, frags_400, frags_300 in self.pkt_infos:
             if len(frags_400) == 1:
                 fragments.extend(frags_400)
             else:
@@ -593,7 +583,7 @@
                 # received by vpp after reassembly is finished, otherwise
                 # new reassemblies will be started and packet generator will
                 # freak out when it detects unfreed buffers
-                zipped = zip(frags_400, frags_30)
+                zipped = zip(frags_400, frags_300)
                 for i, j in zipped[:-1]:
                     fragments.extend(i)
                     fragments.extend(j)
@@ -604,14 +594,13 @@
         )
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
-        self.sleep(.1, "wait for vpp to process packets")
-        packets = self.punt_socket.wait_for_packets(
+        packets = self.dst_if.get_capture(
             len(self.pkt_infos) - len(dropped_packet_indexes))
         self.verify_capture(packets, dropped_packet_indexes)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_timeout_inline(self):
         """ timeout (inline) """
@@ -624,13 +613,13 @@
                                     expire_walk_interval_ms=10000, is_ip6=1)
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(self.fragments_400)
+        self.src_if.add_stream(self.fragments_400)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(
+        packets = self.dst_if.get_capture(
             len(self.pkt_infos) - len(dropped_packet_indexes))
         self.verify_capture(packets, dropped_packet_indexes)
-        pkts = self.pg_if.get_capture(
+        pkts = self.src_if.get_capture(
             expected_count=len(dropped_packet_indexes))
         for icmp in pkts:
             self.assertIn(ICMPv6TimeExceeded, icmp)
@@ -663,19 +652,18 @@
                                     expire_walk_interval_ms=50, is_ip6=1)
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(fragments)
+        self.src_if.add_stream(fragments)
         self.pg_start()
 
         self.sleep(.25, "wait before sending rest of fragments")
 
-        self.pg_if.add_stream(fragments2)
+        self.src_if.add_stream(fragments2)
         self.pg_start()
-        self.sleep(.25, "wait for vpp to process packets")
 
-        packets = self.punt_socket.wait_for_packets(
+        packets = self.dst_if.get_capture(
             len(self.pkt_infos) - len(dropped_packet_indexes))
         self.verify_capture(packets, dropped_packet_indexes)
-        pkts = self.pg_if.get_capture(
+        pkts = self.src_if.get_capture(
             expected_count=len(dropped_packet_indexes))
         for icmp in pkts:
             self.assertIn(ICMPv6TimeExceeded, icmp)
@@ -694,20 +682,20 @@
                                     expire_walk_interval_ms=10000, is_ip6=1)
 
         self.pg_enable_capture()
-        self.pg_if.add_stream(self.fragments_400)
+        self.src_if.add_stream(self.fragments_400)
         self.pg_start()
 
-        packets = self.punt_socket.wait_for_packets(
+        packets = self.dst_if.get_capture(
             len(self.pkt_infos) - len(dropped_packet_indexes))
         self.verify_capture(packets, dropped_packet_indexes)
-        self.pg_if.assert_nothing_captured()
+        self.src_if.assert_nothing_captured()
 
     def test_missing_upper(self):
         """ missing upper layer """
-        p = (Ether(dst=self.pg_if.local_mac, src=self.pg_if.remote_mac) /
-             IPv6(src=self.pg_if.remote_ip6,
-                  dst=self.pg_if.local_ip6) /
-             UDP(sport=1234, dport=self.punt_port) /
+        p = (Ether(dst=self.src_if.local_mac, src=self.src_if.remote_mac) /
+             IPv6(src=self.src_if.remote_ip6,
+                  dst=self.src_if.local_ip6) /
+             UDP(sport=1234, dport=5678) /
              Raw())
         self.extend_packet(p, 1000, self.padding)
         fragments = fragment_rfc8200(p, 1, 500)
@@ -715,52 +703,53 @@
         bad_fragment[IPv6ExtHdrFragment].nh = 59
         bad_fragment[IPv6ExtHdrFragment].offset = 0
         self.pg_enable_capture()
-        self.pg_if.add_stream([bad_fragment])
+        self.src_if.add_stream([bad_fragment])
         self.pg_start()
-        pkts = self.pg_if.get_capture(expected_count=1)
+        pkts = self.src_if.get_capture(expected_count=1)
         icmp = pkts[0]
         self.assertIn(ICMPv6ParamProblem, icmp)
         self.assert_equal(icmp[ICMPv6ParamProblem].code, 3, "ICMP code")
 
     def test_invalid_frag_size(self):
         """ fragment size not a multiple of 8 """
-        p = (Ether(dst=self.pg_if.local_mac, src=self.pg_if.remote_mac) /
-             IPv6(src=self.pg_if.remote_ip6,
-                  dst=self.pg_if.local_ip6) /
-             UDP(sport=1234, dport=self.punt_port) /
+        p = (Ether(dst=self.src_if.local_mac, src=self.src_if.remote_mac) /
+             IPv6(src=self.src_if.remote_ip6,
+                  dst=self.src_if.local_ip6) /
+             UDP(sport=1234, dport=5678) /
              Raw())
         self.extend_packet(p, 1000, self.padding)
         fragments = fragment_rfc8200(p, 1, 500)
         bad_fragment = fragments[0]
         self.extend_packet(bad_fragment, len(bad_fragment) + 5)
         self.pg_enable_capture()
-        self.pg_if.add_stream([bad_fragment])
+        self.src_if.add_stream([bad_fragment])
         self.pg_start()
-        pkts = self.pg_if.get_capture(expected_count=1)
+        pkts = self.src_if.get_capture(expected_count=1)
         icmp = pkts[0]
         self.assertIn(ICMPv6ParamProblem, icmp)
         self.assert_equal(icmp[ICMPv6ParamProblem].code, 0, "ICMP code")
 
     def test_invalid_packet_size(self):
         """ total packet size > 65535 """
-        p = (Ether(dst=self.pg_if.local_mac, src=self.pg_if.remote_mac) /
-             IPv6(src=self.pg_if.remote_ip6,
-                  dst=self.pg_if.local_ip6) /
-             UDP(sport=1234, dport=self.punt_port) /
+        p = (Ether(dst=self.src_if.local_mac, src=self.src_if.remote_mac) /
+             IPv6(src=self.src_if.remote_ip6,
+                  dst=self.src_if.local_ip6) /
+             UDP(sport=1234, dport=5678) /
              Raw())
         self.extend_packet(p, 1000, self.padding)
         fragments = fragment_rfc8200(p, 1, 500)
         bad_fragment = fragments[1]
         bad_fragment[IPv6ExtHdrFragment].offset = 65500
         self.pg_enable_capture()
-        self.pg_if.add_stream([bad_fragment])
+        self.src_if.add_stream([bad_fragment])
         self.pg_start()
-        pkts = self.pg_if.get_capture(expected_count=1)
+        pkts = self.src_if.get_capture(expected_count=1)
         icmp = pkts[0]
         self.assertIn(ICMPv6ParamProblem, icmp)
         self.assert_equal(icmp[ICMPv6ParamProblem].code, 0, "ICMP code")
 
 
+@unittest.skip("removing GRE tunnels broken, need fix")
 class TestFIFReassembly(VppTestCase):
     """ Fragments in fragments reassembly """
 
@@ -768,23 +757,28 @@
     def setUpClass(cls):
         super(TestFIFReassembly, cls).setUpClass()
 
-        cls.create_pg_interfaces([0])
-        cls.pg_if = cls.pg0
-        cls.pg_if.admin_up()
-        cls.pg_if.config_ip4()
-        cls.pg_if.resolve_arp()
-        cls.pg_if.config_ip6()
-        cls.pg_if.resolve_ndp()
+        cls.create_pg_interfaces([0, 1])
+        cls.src_if = cls.pg0
+        cls.dst_if = cls.pg1
+        for i in cls.pg_interfaces:
+            i.admin_up()
+            i.config_ip4()
+            i.resolve_arp()
+            i.config_ip6()
+            i.resolve_ndp()
 
-        cls.punt_port = 9999
-        cls.punt4_socket = VppUDSPuntSocket(cls, cls.punt_port)
-        cls.punt6_socket = VppUDSPuntSocket(cls, cls.punt_port, is_ip4=0)
         cls.packet_sizes = [64, 512, 1518, 9018]
         cls.padding = " abcdefghijklmn"
 
     def setUp(self):
         """ Test setup - force timeout on existing reassemblies """
         super(TestFIFReassembly, self).setUp()
+        self.vapi.ip_reassembly_enable_disable(
+            sw_if_index=self.src_if.sw_if_index, enable_ip4=True,
+            enable_ip6=True)
+        self.vapi.ip_reassembly_enable_disable(
+            sw_if_index=self.dst_if.sw_if_index, enable_ip4=True,
+            enable_ip6=True)
         self.vapi.ip_reassembly_set(timeout_ms=0, max_reassemblies=1000,
                                     expire_walk_interval_ms=10)
         self.vapi.ip_reassembly_set(timeout_ms=0, max_reassemblies=1000,
@@ -809,11 +803,7 @@
         seen = set()
         for packet in capture:
             try:
-                sw_if_index = packet['sw_if_index']
-                punt_action = packet['punt_action']
-                packet = Ether(packet['packet'])
-                self.logger.debug(ppp("Got packet from %s, action %s" %
-                                      (sw_if_index, punt_action), packet))
+                self.logger.debug(ppp("Got packet:", packet))
                 ip = packet[ip_class]
                 udp = packet[UDP]
                 payload_info = self.payload_to_info(str(packet[Raw]))
@@ -824,7 +814,7 @@
                 if packet_index in seen:
                     raise Exception(ppp("Duplicate packet received", packet))
                 seen.add(packet_index)
-                self.assertEqual(payload_info.dst, self.pg_if.sw_if_index)
+                self.assertEqual(payload_info.dst, self.dst_if.sw_if_index)
                 info = self._packet_infos[packet_index]
                 self.assertTrue(info is not None)
                 self.assertEqual(packet_index, info.index)
@@ -832,7 +822,7 @@
                 self.assertEqual(ip.src, saved_packet[ip_class].src)
                 self.assertEqual(ip.dst, saved_packet[ip_class].dst)
                 self.assertEqual(udp.payload, saved_packet[UDP].payload)
-            except:
+            except Exception:
                 self.logger.error(ppp("Unexpected or invalid packet:", packet))
                 raise
         for index in self._packet_infos:
@@ -848,33 +838,41 @@
         # it shared for multiple test cases
         self.tun_ip4 = "1.1.1.2"
 
-        self.gre4 = VppGreInterface(self, self.pg0.local_ip4, self.tun_ip4)
+        self.gre4 = VppGreInterface(self, self.src_if.local_ip4, self.tun_ip4)
         self.gre4.add_vpp_config()
         self.gre4.admin_up()
         self.gre4.config_ip4()
 
+        self.vapi.ip_reassembly_enable_disable(
+            sw_if_index=self.gre4.sw_if_index, enable_ip4=True)
+
         self.route4 = VppIpRoute(self, self.tun_ip4, 32,
-                                 [VppRoutePath(self.pg0.remote_ip4,
-                                               self.pg0.sw_if_index)])
+                                 [VppRoutePath(self.src_if.remote_ip4,
+                                               self.src_if.sw_if_index)])
         self.route4.add_vpp_config()
 
         self.reset_packet_infos()
         for i in range(test_packet_count):
-            info = self.create_packet_info(self.pg0, self.pg0)
+            info = self.create_packet_info(self.src_if, self.dst_if)
             payload = self.info_to_payload(info)
-            p = (IP(id=i, src=self.pg0.remote_ip4, dst=self.pg0.local_ip4) /
-                 UDP(sport=1234, dport=self.punt_port) /
+            # Ethernet header here is only for size calculation, thus it
+            # doesn't matter how it's initialized. This is to ensure that
+            # reassembled packet is not > 9000 bytes, so that it's not dropped
+            p = (Ether() /
+                 IP(id=i, src=self.src_if.remote_ip4,
+                    dst=self.dst_if.remote_ip4) /
+                 UDP(sport=1234, dport=5678) /
                  Raw(payload))
             size = self.packet_sizes[(i // 2) % len(self.packet_sizes)]
             self.extend_packet(p, size, self.padding)
-            info.data = p
+            info.data = p[IP]  # use only IP part, without ethernet header
 
         fragments = [x for _, p in self._packet_infos.iteritems()
                      for x in fragment_rfc791(p.data, 400)]
 
         encapped_fragments = \
-            [Ether(dst=self.pg0.local_mac, src=self.pg0.remote_mac) /
-             IP(src=self.tun_ip4, dst=self.pg0.local_ip4) /
+            [Ether(dst=self.src_if.local_mac, src=self.src_if.remote_mac) /
+             IP(src=self.tun_ip4, dst=self.src_if.local_ip4) /
                 GRE() /
                 p
                 for p in fragments]
@@ -883,18 +881,19 @@
             [x for p in encapped_fragments
              for x in fragment_rfc791(p, 200)]
 
-        self.pg0.add_stream(fragmented_encapped_fragments)
+        self.src_if.add_stream(fragmented_encapped_fragments)
 
         self.pg_enable_capture(self.pg_interfaces)
         self.pg_start()
 
-        self.pg0.assert_nothing_captured()
-        packets = self.punt4_socket.wait_for_packets(len(self._packet_infos))
+        self.src_if.assert_nothing_captured()
+        packets = self.dst_if.get_capture(len(self._packet_infos))
         self.verify_capture(packets, IP)
 
         # TODO remove gre vpp config by hand until VppIpRoute gets fixed
         # so that it's query_vpp_config() works as it should
         self.gre4.remove_vpp_config()
+        self.logger.debug(self.vapi.ppcli("show interface"))
 
     def test_fif6(self):
         """ Fragments in fragments (6o6) """
@@ -904,36 +903,43 @@
         # it shared for multiple test cases
         self.tun_ip6 = "1002::1"
 
-        self.gre6 = VppGre6Interface(self, self.pg0.local_ip6, self.tun_ip6)
+        self.gre6 = VppGre6Interface(self, self.src_if.local_ip6, self.tun_ip6)
         self.gre6.add_vpp_config()
         self.gre6.admin_up()
         self.gre6.config_ip6()
 
+        self.vapi.ip_reassembly_enable_disable(
+            sw_if_index=self.gre6.sw_if_index, enable_ip6=True)
+
         self.route6 = VppIpRoute(self, self.tun_ip6, 128,
-                                 [VppRoutePath(self.pg0.remote_ip6,
-                                               self.pg0.sw_if_index,
+                                 [VppRoutePath(self.src_if.remote_ip6,
+                                               self.src_if.sw_if_index,
                                                proto=DpoProto.DPO_PROTO_IP6)],
                                  is_ip6=1)
         self.route6.add_vpp_config()
 
         self.reset_packet_infos()
         for i in range(test_packet_count):
-            info = self.create_packet_info(self.pg0, self.pg0)
+            info = self.create_packet_info(self.src_if, self.dst_if)
             payload = self.info_to_payload(info)
-            p = (IPv6(src=self.pg0.remote_ip6, dst=self.pg0.local_ip6) /
-                 UDP(sport=1234, dport=self.punt_port) /
+            # Ethernet header here is only for size calculation, thus it
+            # doesn't matter how it's initialized. This is to ensure that
+            # reassembled packet is not > 9000 bytes, so that it's not dropped
+            p = (Ether() /
+                 IPv6(src=self.src_if.remote_ip6, dst=self.dst_if.remote_ip6) /
+                 UDP(sport=1234, dport=5678) /
                  Raw(payload))
             size = self.packet_sizes[(i // 2) % len(self.packet_sizes)]
             self.extend_packet(p, size, self.padding)
-            info.data = p
+            info.data = p[IPv6]  # use only IPv6 part, without ethernet header
 
         fragments = [x for _, i in self._packet_infos.iteritems()
                      for x in fragment_rfc8200(
                          i.data, i.index, 400)]
 
         encapped_fragments = \
-            [Ether(dst=self.pg0.local_mac, src=self.pg0.remote_mac) /
-             IPv6(src=self.tun_ip6, dst=self.pg0.local_ip6) /
+            [Ether(dst=self.src_if.local_mac, src=self.src_if.remote_mac) /
+             IPv6(src=self.tun_ip6, dst=self.src_if.local_ip6) /
                 GRE() /
                 p
                 for p in fragments]
@@ -948,18 +954,18 @@
             )
             ]
 
-        self.pg0.add_stream(fragmented_encapped_fragments)
+        self.src_if.add_stream(fragmented_encapped_fragments)
 
         self.pg_enable_capture(self.pg_interfaces)
         self.pg_start()
 
-        self.pg0.assert_nothing_captured()
-        packets = self.punt6_socket.wait_for_packets(len(self._packet_infos))
+        self.src_if.assert_nothing_captured()
+        packets = self.dst_if.get_capture(len(self._packet_infos))
         self.verify_capture(packets, IPv6)
 
         # TODO remove gre vpp config by hand until VppIpRoute gets fixed
         # so that it's query_vpp_config() works as it should
-        self.gre6.remove_vpp_config()
+        # self.gre6.remove_vpp_config()
 
 
 if __name__ == '__main__':
diff --git a/test/vpp_papi_provider.py b/test/vpp_papi_provider.py
index dd553cb..52fc364 100644
--- a/test/vpp_papi_provider.py
+++ b/test/vpp_papi_provider.py
@@ -3203,6 +3203,15 @@
         """ Get IP reassembly parameters """
         return self.api(self.papi.ip_reassembly_get, {'is_ip6': is_ip6})
 
+    def ip_reassembly_enable_disable(self, sw_if_index, enable_ip4=False,
+                                     enable_ip6=False):
+        """ Enable/disable IP reassembly """
+        return self.api(self.papi.ip_reassembly_enable_disable,
+                        {'sw_if_index': sw_if_index,
+                         'enable_ip4': 1 if enable_ip4 else 0,
+                         'enable_ip6': 1 if enable_ip6 else 0,
+                         })
+
     def gbp_endpoint_add_del(self, is_add, sw_if_index, addr, is_ip6, epg):
         """ GBP endpoint Add/Del """
         return self.api(self.papi.gbp_endpoint_add_del,
diff --git a/test/vpp_punt_socket.py b/test/vpp_punt_socket.py
deleted file mode 100644
index 5004b23..0000000
--- a/test/vpp_punt_socket.py
+++ /dev/null
@@ -1,75 +0,0 @@
-from socket import socket, AF_UNIX, SOCK_DGRAM
-from select import select
-from time import time
-from struct import unpack, calcsize
-from util import ppc
-from scapy.layers.l2 import Ether
-
-client_uds_socket_name = "client-uds-socket"
-vpp_uds_socket_name = "vpp-uds-socket"
-
-VPP_PUNT_HEADER_FMT = '=Ii'
-VPP_PUNT_HEADER_SIZE = calcsize(VPP_PUNT_HEADER_FMT)
-
-
-class VppPuntAction:
-    PUNT_L2 = 0
-    PUNT_IP4_ROUTED = 1
-    PUNT_IP6_ROUTED = 2
-
-
-class VppUDSPuntSocket(object):
-    def __init__(self, testcase, port, is_ip4=1, l4_protocol=0x11):
-        client_path = '%s/%s-%s-%s' % (testcase.tempdir,
-                                       client_uds_socket_name,
-                                       "4" if is_ip4 else "6", port)
-        testcase.vapi.punt_socket_register(
-            port, client_path, is_ip4=is_ip4, l4_protocol=l4_protocol)
-        self.testcase = testcase
-        self.uds = socket(AF_UNIX, SOCK_DGRAM)
-        self.uds.bind(client_path)
-        self.uds.connect(testcase.punt_socket_path)
-
-    def wait_for_packets(self, count, timeout=1):
-        packets = []
-        now = time()
-        deadline = now + timeout
-        while len(packets) < count and now < deadline:
-            r, w, e = select([self.uds], [], [self.uds], deadline - now)
-            if self.uds in r:
-                x = self.uds.recv(1024 * 1024)
-                sw_if_index, punt_action = unpack(
-                    VPP_PUNT_HEADER_FMT, x[:VPP_PUNT_HEADER_SIZE])
-                packets.append({'sw_if_index': sw_if_index,
-                                'punt_action': punt_action,
-                                'packet': x[VPP_PUNT_HEADER_SIZE:]})
-
-            if self.uds in e:
-                raise Exception("select() indicates error on UDS socket")
-            now = time()
-
-        if len(packets) != count:
-            raise Exception("Unexpected packet count received, got %s packets,"
-                            " expected %s packets" % (len(packets), count))
-        self.testcase.logger.debug(
-            "Got %s packets via punt socket" % len(packets))
-        return packets
-
-    def assert_nothing_captured(self, timeout=.25):
-        packets = []
-        now = time()
-        deadline = now + timeout
-        while now < deadline:
-            r, w, e = select([self.uds], [], [self.uds], deadline - now)
-            if self.uds in r:
-                x = self.uds.recv(1024 * 1024)
-                packets.append(Ether(x[VPP_PUNT_HEADER_SIZE:]))
-            if self.uds in e:
-                raise Exception("select() indicates error on UDS socket")
-            now = time()
-
-        if len(packets) > 0:
-            self.testcase.logger.error(
-                ppc("Unexpected packets captured:", packets))
-            raise Exception("Unexpected packet count received, got %s packets,"
-                            " expected no packets" % len(packets))