IPIP and IPv6 fragmentation

- Error where ICMPv6 error code doesn't reset VLIB_TX = -1
  Leading to crash for ICMP generated on tunnelled packets
- Missed setting VNET_BUFFER_F_LOCALLY_ORIGINATED, so
  IP in IPv6 packets never got fragmented.
- Add support for fragmentation of buffer chains.
- Remove support for inner fragmentation in frag code itself.

Change-Id: If9a97301b7e35ca97ffa5c0fada2b9e7e7dbfb27
Signed-off-by: Ole Troan <ot@cisco.com>

5 files changed, 206 insertions, 286 deletions

diff --git a/src/vnet/ip/icmp6.c b/src/vnet/ip/icmp6.c
index 6ebdef4745e..4252ac7ad59 100644
--- a/src/vnet/ip/icmp6.c
+++ b/src/vnet/ip/icmp6.c
@@ -524,14 +524,24 @@ ip6_icmp_error (vlib_main_t * vm,
 		{
 		  b = vlib_get_buffer (vm, b->next_buffer);
 		  b->current_length = 0;
+		  // XXX: Buffer leak???
 		}
 	    }
 
 	  /* Add IP header and ICMPv6 header including a 4 byte data field */
-	  vlib_buffer_advance (p0,
-			       -sizeof (ip6_header_t) -
-			       sizeof (icmp46_header_t) - 4);
+	  int headroom = sizeof (ip6_header_t) + sizeof (icmp46_header_t) + 4;
 
+	  /* Verify that we're not falling off the edge */
+	  if (p0->current_data - headroom < -VLIB_BUFFER_PRE_DATA_SIZE)
+	    {
+	      next0 = IP6_ICMP_ERROR_NEXT_DROP;
+	      error0 = ICMP6_ERROR_DROP;
+	      goto error;
+	    }
+
+	  vlib_buffer_advance (p0, -headroom);
+	  vnet_buffer (p0)->sw_if_index[VLIB_TX] = ~0;
+	  p0->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
 	  p0->current_length =
 	    p0->current_length > 1280 ? 1280 : p0->current_length;
 
@@ -561,6 +571,7 @@ ip6_icmp_error (vlib_main_t * vm,
 	    {
 	      next0 = IP6_ICMP_ERROR_NEXT_DROP;
 	      error0 = ICMP6_ERROR_DROP;
+	      goto error;
 	    }
 
 	  /* Fill icmp header fields */
@@ -573,11 +584,11 @@ ip6_icmp_error (vlib_main_t * vm,
 	    ip6_tcp_udp_icmp_compute_checksum (vm, p0, out_ip0,
 					       &bogus_length);
 
-
-
 	  /* Update error status */
 	  if (error0 == ICMP6_ERROR_NONE)
 	    error0 = icmp6_icmp_type_to_error (icmp0->type);
+
+	error:
 	  vlib_error_count (vm, node->node_index, error0, 1);
 
 	  /* Verify speculative enqueue, maybe switch current next frame */
@@ -602,7 +613,7 @@ VLIB_REGISTER_NODE (ip6_icmp_error_node) = {
 
   .n_next_nodes = IP6_ICMP_ERROR_N_NEXT,
   .next_nodes = {
-    [IP6_ICMP_ERROR_NEXT_DROP] = "ip6-drop",
+    [IP6_ICMP_ERROR_NEXT_DROP] = "error-drop",
     [IP6_ICMP_ERROR_NEXT_LOOKUP] = "ip6-lookup",
   },
 
diff --git a/src/vnet/ip/ip4_forward.c b/src/vnet/ip/ip4_forward.c
index d5cf011b677..d5e2cd56e77 100644
--- a/src/vnet/ip/ip4_forward.c
+++ b/src/vnet/ip/ip4_forward.c
@@ -2097,7 +2097,7 @@ ip4_mtu_check (vlib_buffer_t * b, u16 packet_len,
       else
 	{
 	  /* IP fragmentation */
-	  ip_frag_set_vnet_buffer (b, 0, adj_packet_bytes,
+	  ip_frag_set_vnet_buffer (b, adj_packet_bytes,
 				   IP4_FRAG_NEXT_IP4_REWRITE, 0);
 	  *next = IP4_REWRITE_NEXT_FRAGMENT;
 	}
diff --git a/src/vnet/ip/ip6_forward.c b/src/vnet/ip/ip6_forward.c
index 5abbba59617..e05792fe28f 100644
--- a/src/vnet/ip/ip6_forward.c
+++ b/src/vnet/ip/ip6_forward.c
@@ -1566,9 +1566,10 @@ ip6_mtu_check (vlib_buffer_t * b, u16 packet_bytes,
       if (is_locally_generated)
 	{
 	  /* IP fragmentation */
-	  ip_frag_set_vnet_buffer (b, 0, adj_packet_bytes,
+	  ip_frag_set_vnet_buffer (b, adj_packet_bytes,
 				   IP6_FRAG_NEXT_IP6_REWRITE, 0);
 	  *next = IP6_REWRITE_NEXT_FRAGMENT;
+	  *error = IP6_ERROR_MTU_EXCEEDED;
 	}
       else
 	{
diff --git a/src/vnet/ip/ip_frag.c b/src/vnet/ip/ip_frag.c
index eb9bb4a5ebc..628d9d66474 100644
--- a/src/vnet/ip/ip_frag.c
+++ b/src/vnet/ip/ip_frag.c
@@ -27,7 +27,6 @@
 typedef struct
 {
   u8 ipv6;
-  u16 header_offset;
   u16 mtu;
   u8 next;
   u16 n_fragments;
@@ -39,13 +38,48 @@ format_ip_frag_trace (u8 * s, va_list * args)
   CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
   CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
   ip_frag_trace_t *t = va_arg (*args, ip_frag_trace_t *);
-  s = format (s, "IPv%s offset: %u mtu: %u fragments: %u",
-	      t->ipv6 ? "6" : "4", t->header_offset, t->mtu, t->n_fragments);
+  s = format (s, "IPv%s mtu: %u fragments: %u",
+	      t->ipv6 ? "6" : "4", t->mtu, t->n_fragments);
   return s;
 }
 
 static u32 running_fragment_id;
 
+static void
+frag_set_sw_if_index (vlib_buffer_t * to, vlib_buffer_t * from)
+{
+  vnet_buffer (to)->sw_if_index[VLIB_RX] =
+    vnet_buffer (from)->sw_if_index[VLIB_RX];
+  vnet_buffer (to)->sw_if_index[VLIB_TX] =
+    vnet_buffer (from)->sw_if_index[VLIB_TX];
+
+  /* Copy adj_index in case DPO based node is sending for the
+   * fragmentation, the packet would be sent back to the proper
+   * DPO next node and Index
+   */
+  vnet_buffer (to)->ip.adj_index[VLIB_RX] =
+    vnet_buffer (from)->ip.adj_index[VLIB_RX];
+  vnet_buffer (to)->ip.adj_index[VLIB_TX] =
+    vnet_buffer (from)->ip.adj_index[VLIB_TX];
+}
+
+static vlib_buffer_t *
+frag_buffer_alloc (vlib_buffer_t * org_b, u32 * bi)
+{
+  vlib_main_t *vm = vlib_get_main ();
+  if (vlib_buffer_alloc (vm, bi, 1) != 1)
+    return 0;
+
+  vlib_buffer_t *b = vlib_get_buffer (vm, *bi);
+  vlib_buffer_free_list_t *fl =
+    vlib_buffer_get_free_list (vm, VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX);
+  vlib_buffer_init_for_free_list (b, fl);
+  VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b);
+  vlib_buffer_copy_trace_flag (vm, org_b, *bi);
+
+  return b;
+}
+
 /*
  * Limitation: Does follow buffer chains in the packet to fragment,
  * but does not generate buffer chains. I.e. a fragment is always
@@ -58,23 +92,19 @@ ip4_frag_do_fragment (vlib_main_t * vm, u32 from_bi, u32 ** buffer,
 {
   vlib_buffer_t *from_b;
   ip4_header_t *ip4;
-  u16 mtu, len, max, rem, offset, ip_frag_id, ip_frag_offset;
+  u16 mtu, len, max, rem, ip_frag_id, ip_frag_offset;
   u8 *org_from_packet, more;
 
   from_b = vlib_get_buffer (vm, from_bi);
-  offset = vnet_buffer (from_b)->ip_frag.header_offset;
   mtu = vnet_buffer (from_b)->ip_frag.mtu;
   org_from_packet = vlib_buffer_get_current (from_b);
-  ip4 = (ip4_header_t *) vlib_buffer_get_current (from_b) + offset;
+  ip4 = (ip4_header_t *) vlib_buffer_get_current (from_b);
 
   rem = clib_net_to_host_u16 (ip4->length) - sizeof (ip4_header_t);
-  max =
-    (mtu - sizeof (ip4_header_t) -
-     vnet_buffer (from_b)->ip_frag.header_offset) & ~0x7;
+  max = (mtu - sizeof (ip4_header_t)) & ~0x7;
 
   if (rem >
-      (vlib_buffer_length_in_chain (vm, from_b) - offset -
-       sizeof (ip4_header_t)))
+      (vlib_buffer_length_in_chain (vm, from_b) - sizeof (ip4_header_t)))
     {
       *error = IP_FRAG_ERROR_MALFORMED;
       return;
@@ -110,9 +140,9 @@ ip4_frag_do_fragment (vlib_main_t * vm, u32 from_bi, u32 ** buffer,
 
   u8 *from_data = (void *) (ip4 + 1);
   vlib_buffer_t *org_from_b = from_b;
-  u16 ptr = 0, fo = 0;
-  u16 left_in_from_buffer =
-    from_b->current_length - offset - sizeof (ip4_header_t);
+  u16 fo = 0;
+  u16 left_in_from_buffer = from_b->current_length - sizeof (ip4_header_t);
+  u16 ptr = 0;
 
   /* Do the actual fragmentation */
   while (rem)
@@ -122,40 +152,28 @@ ip4_frag_do_fragment (vlib_main_t * vm, u32 from_bi, u32 ** buffer,
       ip4_header_t *to_ip4;
       u8 *to_data;
 
-      len = (rem > (mtu - sizeof (ip4_header_t) - offset) ? max : rem);
+      len = (rem > (mtu - sizeof (ip4_header_t)) ? max : rem);
       if (len != rem)		/* Last fragment does not need to divisible by 8 */
 	len &= ~0x7;
-      if (!vlib_buffer_alloc (vm, &to_bi, 1))
+      if ((to_b = frag_buffer_alloc (org_from_b, &to_bi)) == 0)
 	{
 	  *error = IP_FRAG_ERROR_MEMORY;
 	  return;
 	}
       vec_add1 (*buffer, to_bi);
-      to_b = vlib_get_buffer (vm, to_bi);
-      vnet_buffer (to_b)->sw_if_index[VLIB_RX] =
-	vnet_buffer (org_from_b)->sw_if_index[VLIB_RX];
-      vnet_buffer (to_b)->sw_if_index[VLIB_TX] =
-	vnet_buffer (org_from_b)->sw_if_index[VLIB_TX];
-      /* Copy adj_index in case DPO based node is sending for the
-       * fragmentation, the packet would be sent back to the proper
-       * DPO next node and Index
-       */
-      vnet_buffer (to_b)->ip.adj_index[VLIB_RX] =
-	vnet_buffer (org_from_b)->ip.adj_index[VLIB_RX];
-      vnet_buffer (to_b)->ip.adj_index[VLIB_TX] =
-	vnet_buffer (org_from_b)->ip.adj_index[VLIB_TX];
-
-      /* Copy offset and ip4 header */
-      clib_memcpy (to_b->data, org_from_packet,
-		   offset + sizeof (ip4_header_t));
-      to_ip4 = vlib_buffer_get_current (to_b) + offset;
+      frag_set_sw_if_index (to_b, org_from_b);
+
+      /* Copy ip4 header */
+      clib_memcpy (to_b->data, org_from_packet, sizeof (ip4_header_t));
+      to_ip4 = vlib_buffer_get_current (to_b);
       to_data = (void *) (to_ip4 + 1);
 
       /* Spin through from buffers filling up the to buffer */
-      u16 to_ptr = 0;
-      u16 bytes_to_copy, left_in_to_buffer = len;
+      u16 left_in_to_buffer = len, to_ptr = 0;
       while (1)
 	{
+	  u16 bytes_to_copy;
+
 	  /* Figure out how many bytes we can safely copy */
 	  bytes_to_copy = left_in_to_buffer <= left_in_from_buffer ?
 	    left_in_to_buffer : left_in_from_buffer;
@@ -166,7 +184,7 @@ ip4_frag_do_fragment (vlib_main_t * vm, u32 from_bi, u32 ** buffer,
 	  if (left_in_to_buffer == 0)
 	    break;
 
-	  ASSERT (left_in_from_buffer == 0);
+	  ASSERT (left_in_from_buffer <= 0);
 	  /* Move buffer */
 	  if (!(from_b->flags & VLIB_BUFFER_NEXT_PRESENT))
 	    {
@@ -180,7 +198,7 @@ ip4_frag_do_fragment (vlib_main_t * vm, u32 from_bi, u32 ** buffer,
 	  to_ptr += bytes_to_copy;
 	}
 
-      to_b->current_length = offset + len + sizeof (ip4_header_t);
+      to_b->current_length = len + sizeof (ip4_header_t);
 
       to_ip4->fragment_id = ip_frag_id;
       to_ip4->flags_and_fragment_offset =
@@ -214,21 +232,20 @@ ip4_frag_do_fragment (vlib_main_t * vm, u32 from_bi, u32 ** buffer,
 }
 
 void
-ip_frag_set_vnet_buffer (vlib_buffer_t * b, u16 offset, u16 mtu,
-			 u8 next_index, u8 flags)
+ip_frag_set_vnet_buffer (vlib_buffer_t * b, u16 mtu, u8 next_index, u8 flags)
 {
-  vnet_buffer (b)->ip_frag.header_offset = offset;
   vnet_buffer (b)->ip_frag.mtu = mtu;
   vnet_buffer (b)->ip_frag.next_index = next_index;
   vnet_buffer (b)->ip_frag.flags = flags;
 }
 
-static uword
-ip4_frag (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame)
+
+static inline uword
+frag_node_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
+		  vlib_frame_t * frame, u32 node_index, bool is_ip6)
 {
   u32 n_left_from, *from, next_index, *to_next, n_left_to_next;
-  vlib_node_runtime_t *error_node =
-    vlib_node_get_runtime (vm, ip4_frag_node.index);
+  vlib_node_runtime_t *error_node = vlib_node_get_runtime (vm, node_index);
   from = vlib_frame_vector_args (frame);
   n_left_from = frame->n_vectors;
   next_index = node->cached_next_index;
@@ -244,58 +261,68 @@ ip4_frag (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame)
 	  u32 pi0, *frag_from, frag_left;
 	  vlib_buffer_t *p0;
 	  ip_frag_error_t error0;
-	  ip4_frag_next_t next0;
+	  int next0;
 
-	  //Note: The packet is not enqueued now.
-	  //It is instead put in a vector where other fragments
-	  //will be put as well.
+	  /*
+	   * Note: The packet is not enqueued now. It is instead put
+	   * in a vector where other fragments will be put as well.
+	   */
 	  pi0 = from[0];
 	  from += 1;
 	  n_left_from -= 1;
 	  error0 = IP_FRAG_ERROR_NONE;
 
 	  p0 = vlib_get_buffer (vm, pi0);
-	  ip4_frag_do_fragment (vm, pi0, &buffer, &error0);
+	  if (is_ip6)
+	    ip6_frag_do_fragment (vm, pi0, &buffer, &error0);
+	  else
+	    ip4_frag_do_fragment (vm, pi0, &buffer, &error0);
 
 	  if (PREDICT_FALSE (p0->flags & VLIB_BUFFER_IS_TRACED))
 	    {
 	      ip_frag_trace_t *tr =
 		vlib_add_trace (vm, node, p0, sizeof (*tr));
-	      tr->header_offset = vnet_buffer (p0)->ip_frag.header_offset;
 	      tr->mtu = vnet_buffer (p0)->ip_frag.mtu;
-	      tr->ipv6 = 0;
+	      tr->ipv6 = is_ip6 ? 1 : 0;
 	      tr->n_fragments = vec_len (buffer);
 	      tr->next = vnet_buffer (p0)->ip_frag.next_index;
 	    }
 
-	  if (error0 == IP_FRAG_ERROR_DONT_FRAGMENT_SET)
+	  if (!is_ip6 && error0 == IP_FRAG_ERROR_DONT_FRAGMENT_SET)
 	    {
 	      icmp4_error_set_vnet_buffer (p0, ICMP4_destination_unreachable,
 					   ICMP4_destination_unreachable_fragmentation_needed_and_dont_fragment_set,
 					   vnet_buffer (p0)->ip_frag.mtu);
-	      vlib_buffer_advance (p0,
-				   vnet_buffer (p0)->ip_frag.header_offset);
 	      next0 = IP4_FRAG_NEXT_ICMP_ERROR;
 	    }
 	  else
 	    {
-              /* *INDENT-OFF* */
-              next0 = (error0 == IP_FRAG_ERROR_NONE) ? vnet_buffer (p0)->
-                ip_frag.next_index : IP4_FRAG_NEXT_DROP;
-              /* *INDENT-ON* */
+	      if (is_ip6)
+		next0 =
+		  (error0 ==
+		   IP_FRAG_ERROR_NONE) ? vnet_buffer (p0)->
+		  ip_frag.next_index : IP6_FRAG_NEXT_DROP;
+	      else
+		next0 =
+		  (error0 ==
+		   IP_FRAG_ERROR_NONE) ? vnet_buffer (p0)->
+		  ip_frag.next_index : IP4_FRAG_NEXT_DROP;
 	    }
 
 	  if (error0 == IP_FRAG_ERROR_NONE)
 	    {
 	      /* Free original buffer chain */
-	      vlib_buffer_free_one (vm, pi0);
 	      frag_sent += vec_len (buffer);
 	      small_packets += (vec_len (buffer) == 1);
+	      vlib_buffer_free_one (vm, pi0);	/* Free original packet */
 	    }
 	  else
-	    vlib_error_count (vm, ip4_frag_node.index, error0, 1);
+	    {
+	      vlib_error_count (vm, node_index, error0, 1);
+	      vec_add1 (buffer, pi0);	/* Get rid of the original buffer */
+	    }
 
-	  //Send fragments that were added in the frame
+	  /* Send fragments that were added in the frame */
 	  frag_from = buffer;
 	  frag_left = vec_len (buffer);
 
@@ -325,262 +352,143 @@ ip4_frag (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame)
     }
   vec_free (buffer);
 
-  vlib_node_increment_counter (vm, ip4_frag_node.index,
+  vlib_node_increment_counter (vm, node_index,
 			       IP_FRAG_ERROR_FRAGMENT_SENT, frag_sent);
-  vlib_node_increment_counter (vm, ip4_frag_node.index,
+  vlib_node_increment_counter (vm, node_index,
 			       IP_FRAG_ERROR_SMALL_PACKET, small_packets);
 
   return frame->n_vectors;
 }
 
 
+
+static uword
+ip4_frag (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame)
+{
+  return frag_node_inline (vm, node, frame, ip4_frag_node.index,
+			   0 /* is_ip6 */ );
+}
+
+static uword
+ip6_frag (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame)
+{
+  return frag_node_inline (vm, node, frame, ip6_frag_node.index,
+			   1 /* is_ip6 */ );
+}
+
+/*
+ * Fragments the packet given in from_bi. Fragments are returned in the buffer vector.
+ * Caller must ensure the original packet is freed.
+ */
 void
-ip6_frag_do_fragment (vlib_main_t * vm, u32 pi, u32 ** buffer,
+ip6_frag_do_fragment (vlib_main_t * vm, u32 from_bi, u32 ** buffer,
 		      ip_frag_error_t * error)
 {
-  vlib_buffer_t *p;
-  ip6_header_t *ip6_hdr;
-  ip6_frag_hdr_t *frag_hdr;
-  u8 *payload, *next_header;
-
-  p = vlib_get_buffer (vm, pi);
-
-  //Parsing the IPv6 headers
-  ip6_hdr =
-    vlib_buffer_get_current (p) + vnet_buffer (p)->ip_frag.header_offset;
-  payload = (u8 *) (ip6_hdr + 1);
-  next_header = &ip6_hdr->protocol;
-  if (*next_header == IP_PROTOCOL_IP6_HOP_BY_HOP_OPTIONS)
-    {
-      next_header = payload;
-      payload += payload[1] * 8;
-    }
+  vlib_buffer_t *from_b;
+  ip6_header_t *ip6;
+  u16 mtu, len, max, rem, ip_frag_id;
 
-  if (*next_header == IP_PROTOCOL_IP6_DESTINATION_OPTIONS)
-    {
-      next_header = payload;
-      payload += payload[1] * 8;
-    }
+  from_b = vlib_get_buffer (vm, from_bi);
+  mtu = vnet_buffer (from_b)->ip_frag.mtu;
+  ip6 = (ip6_header_t *) vlib_buffer_get_current (from_b);
 
-  if (*next_header == IP_PROTOCOL_IPV6_ROUTE)
-    {
-      next_header = payload;
-      payload += payload[1] * 8;
-    }
+  rem = clib_net_to_host_u16 (ip6->payload_length);
+  max = (mtu - sizeof (ip6_header_t) - sizeof (ip6_frag_hdr_t)) & ~0x7;	// TODO: Is max correct??
 
-  if (PREDICT_FALSE
-      (payload >= (u8 *) vlib_buffer_get_current (p) + p->current_length))
+  if (rem >
+      (vlib_buffer_length_in_chain (vm, from_b) - sizeof (ip6_header_t)))
     {
-      //A malicious packet could set an extension header with a too big size
-      //and make us modify another vlib_buffer
       *error = IP_FRAG_ERROR_MALFORMED;
       return;
     }
 
-  if (p->flags & VLIB_BUFFER_NEXT_PRESENT)
+  /* TODO: Look through header chain for fragmentation header */
+  if (ip6->protocol == IP_PROTOCOL_IPV6_FRAGMENTATION)
     {
       *error = IP_FRAG_ERROR_MALFORMED;
       return;
     }
 
-  u8 has_more;
-  u16 initial_offset;
-  if (*next_header == IP_PROTOCOL_IPV6_FRAGMENTATION)
-    {
-      //The fragmentation header is already there
-      frag_hdr = (ip6_frag_hdr_t *) payload;
-      has_more = ip6_frag_hdr_more (frag_hdr);
-      initial_offset = ip6_frag_hdr_offset (frag_hdr);
-    }
-  else
-    {
-      //Insert a fragmentation header in the packet
-      u8 nh = *next_header;
-      *next_header = IP_PROTOCOL_IPV6_FRAGMENTATION;
-      vlib_buffer_advance (p, -sizeof (*frag_hdr));
-      u8 *start = vlib_buffer_get_current (p);
-      memmove (start, start + sizeof (*frag_hdr),
-	       payload - (start + sizeof (*frag_hdr)));
-      frag_hdr = (ip6_frag_hdr_t *) (payload - sizeof (*frag_hdr));
-      frag_hdr->identification = ++running_fragment_id;
-      frag_hdr->next_hdr = nh;
-      frag_hdr->rsv = 0;
-      has_more = 0;
-      initial_offset = 0;
-    }
-  payload = (u8 *) (frag_hdr + 1);
-
-  u16 headers_len = payload - (u8 *) vlib_buffer_get_current (p);
-  u16 max_payload = vnet_buffer (p)->ip_frag.mtu - headers_len;
-  u16 rem = p->current_length - headers_len;
+  u8 *from_data = (void *) (ip6 + 1);
+  vlib_buffer_t *org_from_b = from_b;
+  u16 fo = 0;
+  u16 left_in_from_buffer = from_b->current_length - sizeof (ip6_header_t);
   u16 ptr = 0;
 
-  if (max_payload < 8)
-    {
-      *error = IP_FRAG_ERROR_CANT_FRAGMENT_HEADER;
-      return;
-    }
+  ip_frag_id = ++running_fragment_id;	// Fix
 
+  /* Do the actual fragmentation */
   while (rem)
     {
-      u32 bi;
-      vlib_buffer_t *b;
-      u16 len = (rem > max_payload) ? (max_payload & ~0x7) : rem;
-      rem -= len;
-
-      if (ptr != 0)
-	{
-	  if (!vlib_buffer_alloc (vm, &bi, 1))
-	    {
-	      *error = IP_FRAG_ERROR_MEMORY;
-	      return;
-	    }
-	  b = vlib_get_buffer (vm, bi);
-	  vnet_buffer (b)->sw_if_index[VLIB_RX] =
-	    vnet_buffer (p)->sw_if_index[VLIB_RX];
-	  vnet_buffer (b)->sw_if_index[VLIB_TX] =
-	    vnet_buffer (p)->sw_if_index[VLIB_TX];
-
-	  /* Copy Adj_index in case DPO based node is sending for the fragmentation,
-	     the packet would be sent back to the proper DPO next node and Index */
-	  vnet_buffer (b)->ip.adj_index[VLIB_RX] =
-	    vnet_buffer (p)->ip.adj_index[VLIB_RX];
-	  vnet_buffer (b)->ip.adj_index[VLIB_TX] =
-	    vnet_buffer (p)->ip.adj_index[VLIB_TX];
-
-	  clib_memcpy (vlib_buffer_get_current (b),
-		       vlib_buffer_get_current (p), headers_len);
-	  clib_memcpy (vlib_buffer_get_current (b) + headers_len,
-		       payload + ptr, len);
-	  frag_hdr =
-	    vlib_buffer_get_current (b) + headers_len - sizeof (*frag_hdr);
-	}
-      else
-	{
-	  bi = pi;
-	  b = vlib_get_buffer (vm, bi);
-	  //frag_hdr already set here
-	}
+      u32 to_bi;
+      vlib_buffer_t *to_b;
+      ip6_header_t *to_ip6;
+      ip6_frag_hdr_t *to_frag_hdr;
+      u8 *to_data;
 
-      ip6_hdr =
-	vlib_buffer_get_current (b) + vnet_buffer (p)->ip_frag.header_offset;
-      frag_hdr->fragment_offset_and_more =
-	ip6_frag_hdr_offset_and_more (initial_offset + (ptr >> 3),
-				      (rem || has_more));
-      b->current_length = headers_len + len;
-      ip6_hdr->payload_length =
-	clib_host_to_net_u16 (b->current_length -
-			      vnet_buffer (p)->ip_frag.header_offset -
-			      sizeof (*ip6_hdr));
-
-      if (vnet_buffer (p)->ip_frag.flags & IP_FRAG_FLAG_IP4_HEADER)
-	{
-	  //Encapsulating ipv4 header
-	  ip4_header_t *encap_header4 =
-	    (ip4_header_t *) vlib_buffer_get_current (b);
-	  encap_header4->length = clib_host_to_net_u16 (b->current_length);
-	  encap_header4->checksum = ip4_header_checksum (encap_header4);
-	}
-      else if (vnet_buffer (p)->ip_frag.flags & IP_FRAG_FLAG_IP6_HEADER)
+      len =
+	(rem >
+	 (mtu - sizeof (ip6_header_t) - sizeof (ip6_frag_hdr_t)) ? max : rem);
+      if (len != rem)		/* Last fragment does not need to divisible by 8 */
+	len &= ~0x7;
+      if ((to_b = frag_buffer_alloc (org_from_b, &to_bi)) == 0)
 	{
-	  //Encapsulating ipv6 header
-	  ip6_header_t *encap_header6 =
-	    (ip6_header_t *) vlib_buffer_get_current (b);
-	  encap_header6->payload_length =
-	    clib_host_to_net_u16 (b->current_length -
-				  sizeof (*encap_header6));
+	  *error = IP_FRAG_ERROR_MEMORY;
+	  return;
 	}
+      vec_add1 (*buffer, to_bi);
+      frag_set_sw_if_index (to_b, org_from_b);
 
-      vec_add1 (*buffer, bi);
-
-      ptr += len;
-    }
-}
-
-static uword
-ip6_frag (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame)
-{
-  u32 n_left_from, *from, next_index, *to_next, n_left_to_next;
-  vlib_node_runtime_t *error_node =
-    vlib_node_get_runtime (vm, ip6_frag_node.index);
-  from = vlib_frame_vector_args (frame);
-  n_left_from = frame->n_vectors;
-  next_index = node->cached_next_index;
-  u32 frag_sent = 0, small_packets = 0;
-  u32 *buffer = 0;
-
-  while (n_left_from > 0)
-    {
-      vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
+      /* Copy ip6 header */
+      clib_memcpy (to_b->data, ip6, sizeof (ip6_header_t));
+      to_ip6 = vlib_buffer_get_current (to_b);
+      to_frag_hdr = (ip6_frag_hdr_t *) (to_ip6 + 1);
+      to_data = (void *) (to_frag_hdr + 1);
 
-      while (n_left_from > 0 && n_left_to_next > 0)
+      /* Spin through from buffers filling up the to buffer */
+      u16 left_in_to_buffer = len, to_ptr = 0;
+      while (1)
 	{
-	  u32 pi0, *frag_from, frag_left;
-	  vlib_buffer_t *p0;
-	  ip_frag_error_t error0;
-	  ip6_frag_next_t next0;
-
-	  pi0 = from[0];
-	  from += 1;
-	  n_left_from -= 1;
-	  error0 = IP_FRAG_ERROR_NONE;
+	  u16 bytes_to_copy;
 
-	  p0 = vlib_get_buffer (vm, pi0);
-	  ip6_frag_do_fragment (vm, pi0, &buffer, &error0);
+	  /* Figure out how many bytes we can safely copy */
+	  bytes_to_copy = left_in_to_buffer <= left_in_from_buffer ?
+	    left_in_to_buffer : left_in_from_buffer;
+	  clib_memcpy (to_data + to_ptr, from_data + ptr, bytes_to_copy);
+	  left_in_to_buffer -= bytes_to_copy;
+	  ptr += bytes_to_copy;
+	  left_in_from_buffer -= bytes_to_copy;
+	  if (left_in_to_buffer == 0)
+	    break;
 
-	  if (PREDICT_FALSE (p0->flags & VLIB_BUFFER_IS_TRACED))
+	  ASSERT (left_in_from_buffer <= 0);
+	  /* Move buffer */
+	  if (!(from_b->flags & VLIB_BUFFER_NEXT_PRESENT))
 	    {
-	      ip_frag_trace_t *tr =
-		vlib_add_trace (vm, node, p0, sizeof (*tr));
-	      tr->header_offset = vnet_buffer (p0)->ip_frag.header_offset;
-	      tr->mtu = vnet_buffer (p0)->ip_frag.mtu;
-	      tr->ipv6 = 1;
-	      tr->n_fragments = vec_len (buffer);
-	      tr->next = vnet_buffer (p0)->ip_frag.next_index;
+	      *error = IP_FRAG_ERROR_MALFORMED;
+	      return;
 	    }
+	  from_b = vlib_get_buffer (vm, from_b->next_buffer);
+	  from_data = (u8 *) vlib_buffer_get_current (from_b);
+	  ptr = 0;
+	  left_in_from_buffer = from_b->current_length;
+	  to_ptr += bytes_to_copy;
+	}
 
-          /* *INDENT-OFF* */
-	  next0 = (error0 == IP_FRAG_ERROR_NONE) ? vnet_buffer (p0)->
-	    ip_frag.next_index : IP6_FRAG_NEXT_DROP;
-          /* *INDENT-ON* */
-
-	  frag_sent += vec_len (buffer);
-	  small_packets += (vec_len (buffer) == 1);
-
-	  //Send fragments that were added in the frame
-	  frag_from = buffer;
-	  frag_left = vec_len (buffer);
-	  while (frag_left > 0)
-	    {
-	      while (frag_left > 0 && n_left_to_next > 0)
-		{
-		  u32 i;
-		  i = to_next[0] = frag_from[0];
-		  frag_from += 1;
-		  frag_left -= 1;
-		  to_next += 1;
-		  n_left_to_next -= 1;
+      to_b->current_length =
+	len + sizeof (ip6_header_t) + sizeof (ip6_frag_hdr_t);
+      to_ip6->payload_length =
+	clib_host_to_net_u16 (len + sizeof (ip6_frag_hdr_t));
+      to_ip6->protocol = IP_PROTOCOL_IPV6_FRAGMENTATION;
+      to_frag_hdr->fragment_offset_and_more =
+	ip6_frag_hdr_offset_and_more ((fo >> 3), len != rem);
+      to_frag_hdr->identification = ip_frag_id;
+      to_frag_hdr->next_hdr = ip6->protocol;
+      to_frag_hdr->rsv = 0;
 
-		  vlib_get_buffer (vm, i)->error = error_node->errors[error0];
-		  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
-						   to_next, n_left_to_next, i,
-						   next0);
-		}
-	      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
-	      vlib_get_next_frame (vm, node, next_index, to_next,
-				   n_left_to_next);
-	    }
-	  vec_reset_length (buffer);
-	}
-      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
+      rem -= len;
+      fo += len;
     }
-  vec_free (buffer);
-  vlib_node_increment_counter (vm, ip6_frag_node.index,
-			       IP_FRAG_ERROR_FRAGMENT_SENT, frag_sent);
-  vlib_node_increment_counter (vm, ip6_frag_node.index,
-			       IP_FRAG_ERROR_SMALL_PACKET, small_packets);
-
-  return frame->n_vectors;
 }
 
 static char *ip4_frag_error_strings[] = {
diff --git a/src/vnet/ip/ip_frag.h b/src/vnet/ip/ip_frag.h
index ef5eb4c0543..06eeee82bc6 100644
--- a/src/vnet/ip/ip_frag.h
+++ b/src/vnet/ip/ip_frag.h
@@ -84,7 +84,7 @@ typedef enum
     IP_FRAG_N_ERROR,
 } ip_frag_error_t;
 
-void ip_frag_set_vnet_buffer (vlib_buffer_t * b, u16 offset, u16 mtu,
+void ip_frag_set_vnet_buffer (vlib_buffer_t * b, u16 mtu,
 			      u8 next_index, u8 flags);
 void
 ip4_frag_do_fragment (vlib_main_t * vm, u32 pi, u32 ** buffer,