MTU: Setting of MTU on software interface (instead of hardware interface)

Change-Id: I98bd454a761a1032738a21edeb0fe847e801f901
Signed-off-by: Ole Troan <ot@cisco.com>

19 files changed, 232 insertions, 221 deletions

diff --git a/src/vnet/adj/adj.c b/src/vnet/adj/adj.c
index ed4bada6f24..0c9f7468b11 100644
--- a/src/vnet/adj/adj.c
+++ b/src/vnet/adj/adj.c
@@ -353,10 +353,8 @@ adj_mtu_update_walk_cb (adj_index_t ai,
     return (ADJ_WALK_RC_CONTINUE);
 }
 
-static void
-adj_sw_mtu_update (vnet_main_t * vnm,
-                   u32 sw_if_index,
-                   void *ctx)
+void
+adj_mtu_update (u32 sw_if_index)
 {
     /*
      * Walk all the adjacencies on the interface to update the cached MTU
@@ -364,18 +362,6 @@ adj_sw_mtu_update (vnet_main_t * vnm,
     adj_walk (sw_if_index, adj_mtu_update_walk_cb, NULL);
 }
 
-void
-adj_mtu_update (u32 hw_if_index)
-{
-    /*
-     * Walk all the SW interfaces on the HW interface to update the cached MTU
-     */
-    vnet_hw_interface_walk_sw(vnet_get_main(),
-                              hw_if_index,
-                              adj_sw_mtu_update,
-                              NULL);
-}
-
 /**
  * @brief Walk the Adjacencies on a given interface
  */
diff --git a/src/vnet/devices/virtio/vhost-user.c b/src/vnet/devices/virtio/vhost-user.c
index 5460f10b74e..34b131c556f 100644
--- a/src/vnet/devices/virtio/vhost-user.c
+++ b/src/vnet/devices/virtio/vhost-user.c
@@ -2869,8 +2869,8 @@ vhost_user_create_ethernet (vnet_main_t * vnm, vlib_main_t * vm,
   if (error)
     clib_error_report (error);
 
-  vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, vui->hw_if_index);
-  hi->max_l3_packet_bytes[VLIB_RX] = hi->max_l3_packet_bytes[VLIB_TX] = 9000;
+  vnet_sw_interface_t *si = vnet_get_hw_sw_interface (vnm, vui->hw_if_index);
+  vnet_sw_interface_set_mtu (vnm, si->sw_if_index, 9000);
 }
 
 /*
diff --git a/src/vnet/ethernet/interface.c b/src/vnet/ethernet/interface.c
index 2ed20e15c24..4e1d081635b 100644
--- a/src/vnet/ethernet/interface.c
+++ b/src/vnet/ethernet/interface.c
@@ -300,12 +300,9 @@ ethernet_register_interface (vnet_main_t * vnm,
     ETHERNET_MIN_PACKET_BYTES;
   hi->max_packet_bytes = hi->max_supported_packet_bytes =
     ETHERNET_MAX_PACKET_BYTES;
-  hi->per_packet_overhead_bytes =
-    /* preamble */ 8 + /* inter frame gap */ 12;
 
   /* Standard default ethernet MTU. */
-  hi->max_l3_packet_bytes[VLIB_RX] = hi->max_l3_packet_bytes[VLIB_TX] = 9000;
-
+  vnet_sw_interface_set_mtu (vnm, hi->sw_if_index, 9000);
   clib_memcpy (ei->address, address, sizeof (ei->address));
   vec_free (hi->hw_address);
   vec_add (hi->hw_address, address, sizeof (ei->address));
diff --git a/src/vnet/gre/interface.c b/src/vnet/gre/interface.c
index 5b165c858d3..013dde60af9 100644
--- a/src/vnet/gre/interface.c
+++ b/src/vnet/gre/interface.c
@@ -348,12 +348,8 @@ vnet_gre_tunnel_add (vnet_gre_add_del_tunnel_args_t * a,
 	64 + sizeof (gre_header_t) + sizeof (ip6_header_t);
     }
 
-  hi->per_packet_overhead_bytes =
-    /* preamble */ 8 + /* inter frame gap */ 12;
-
   /* Standard default gre MTU. */
-  hi->max_l3_packet_bytes[VLIB_RX] = hi->max_l3_packet_bytes[VLIB_TX] = 9000;
-
+  vnet_sw_interface_set_mtu (vnm, sw_if_index, 9000);
   /*
    * source the FIB entry for the tunnel's destination
    * and become a child thereof. The tunnel will then get poked
diff --git a/src/vnet/interface.api b/src/vnet/interface.api
index 25ba70342ee..0f88863dff3 100644
--- a/src/vnet/interface.api
+++ b/src/vnet/interface.api
@@ -75,7 +75,7 @@ autoreply define want_interface_events
     @param interface_name - name of the interface
     @param link_duplex - 1 if half duplex, 2 if full duplex
     @param link_speed - 1 = 10M, 2 = 100M, 4 = 1G, 8 = 10G, 16 = 40G, 32 = 100G
-    @param link_MTU - max. transmittion unit
+    @param MTU - max. transmittion unit
     @param sub_if_id - A number 0-N to uniquely identify this subif on super if
     @param sub_dot1ad - 0 = dot1q, 1 = dot1ad
     @param sub_dot1ah - 1 = dot1ah, 0 = otherwise
@@ -123,7 +123,7 @@ define sw_interface_details
   u8 link_speed;
 
   /* MTU */
-  u16 link_mtu;
+  u16 mtu;
 
   /* Subinterface ID. A number 0-N to uniquely identify this subinterface under the super interface */
   u32 sub_id;
diff --git a/src/vnet/interface.c b/src/vnet/interface.c
index b07a9ba7553..28c46110851 100644
--- a/src/vnet/interface.c
+++ b/src/vnet/interface.c
@@ -122,20 +122,55 @@ unserialize_vnet_sw_interface_set_flags (serialize_main_t * m, va_list * va)
      /* helper_flags no redistribution */ 0);
 }
 
-void
-vnet_hw_interface_set_mtu (vnet_main_t * vnm, u32 hw_if_index, u32 mtu)
+static void
+vnet_sw_interface_set_mtu_cb (vnet_main_t * vnm, u32 sw_if_index, void *ctx)
 {
-  vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, hw_if_index);
+  u32 *mtu = ctx;
+  vnet_sw_interface_t *si = vnet_get_sw_interface (vnm, sw_if_index);
+  ASSERT (si);
+
+  si->max_l3_packet_bytes[VLIB_TX] = si->max_l3_packet_bytes[VLIB_RX] = *mtu;
+  adj_mtu_update (sw_if_index);
+}
+
+/*
+ * MTU is set per software interface. Setting MTU on a parent
+ * interface will override the MTU setting on sub-interfaces.
+ * TODO: If sub-interface MTU is ~0 inherit from parent?
+ */
+int
+vnet_sw_interface_set_mtu (vnet_main_t * vnm, u32 sw_if_index, u32 mtu)
+{
+  vnet_sw_interface_t *si = vnet_get_sw_interface (vnm, sw_if_index);
+  vnet_hw_interface_t *hi = vnet_get_sw_hw_interface (vnm, sw_if_index);
+
+  if (mtu < hi->min_packet_bytes)
+    return VNET_API_ERROR_INVALID_VALUE;
+  if (mtu > hi->max_packet_bytes)
+    return VNET_API_ERROR_INVALID_VALUE;
 
-  if (hi->max_packet_bytes != mtu)
+  /* If done on a parent interface */
+  if (si->sw_if_index == si->sup_sw_if_index)
     {
-      u16 l3_pad = hi->max_packet_bytes - hi->max_l3_packet_bytes[VLIB_TX];
-      hi->max_packet_bytes = mtu;
-      hi->max_l3_packet_bytes[VLIB_TX] =
-	hi->max_l3_packet_bytes[VLIB_RX] = mtu - l3_pad;
-      ethernet_set_flags (vnm, hw_if_index, ETHERNET_INTERFACE_FLAG_MTU);
-      adj_mtu_update (hw_if_index);
+      if (hi->hw_class_index == ethernet_hw_interface_class.index)
+	{
+	  ethernet_set_flags (vnm, hi->hw_if_index,
+			      ETHERNET_INTERFACE_FLAG_MTU);
+	}
+
+      /* Override MTU on any sub-interface */
+      vnet_hw_interface_walk_sw (vnm,
+				 hi->hw_if_index,
+				 vnet_sw_interface_set_mtu_cb, &mtu);
     }
+  else
+    {
+      si->max_l3_packet_bytes[VLIB_TX] = si->max_l3_packet_bytes[VLIB_RX] =
+	mtu;
+      adj_mtu_update (sw_if_index);
+    }
+
+  return 0;
 }
 
 static void
@@ -584,6 +619,9 @@ vnet_create_sw_interface_no_callbacks (vnet_main_t * vnm,
   if (sw->type == VNET_SW_INTERFACE_TYPE_HARDWARE)
     sw->sup_sw_if_index = sw->sw_if_index;
 
+  sw->max_l3_packet_bytes[VLIB_RX] = ~0;
+  sw->max_l3_packet_bytes[VLIB_TX] = ~0;
+
   /* Allocate counters for this interface. */
   {
     u32 i;
@@ -758,9 +796,7 @@ vnet_register_interface (vnet_main_t * vnm,
 
   hw->max_rate_bits_per_sec = 0;
   hw->min_packet_bytes = 0;
-  hw->per_packet_overhead_bytes = 0;
-  hw->max_l3_packet_bytes[VLIB_RX] = ~0;
-  hw->max_l3_packet_bytes[VLIB_TX] = ~0;
+  hw->max_packet_bytes = 9000;	/* default */
 
   if (dev_class->tx_function == 0)
     goto no_output_nodes;	/* No output/tx nodes to create */
diff --git a/src/vnet/interface.h b/src/vnet/interface.h
index 7556bc5544e..d462e1e8448 100644
--- a/src/vnet/interface.h
+++ b/src/vnet/interface.h
@@ -512,14 +512,6 @@ typedef struct vnet_hw_interface_t
   /* Largest packet size for this interface. */
   u32 max_packet_bytes;
 
-  /* Number of extra bytes that go on the wire.
-     Packet length on wire
-     = max (length + per_packet_overhead_bytes, min_packet_bytes). */
-  u32 per_packet_overhead_bytes;
-
-  /* Receive and transmit layer 3 packet size limits (MRU/MTU). */
-  u32 max_l3_packet_bytes[VLIB_N_RX_TX];
-
   /* Hash table mapping sub interface id to sw_if_index. */
   uword *sub_interface_sw_if_index_by_id;
 
@@ -656,6 +648,9 @@ typedef struct
 
   u32 link_speed;
 
+  /* Receive and transmit layer 3 packet size limits (MRU/MTU). */
+  u32 max_l3_packet_bytes[VLIB_N_RX_TX];
+
   union
   {
     /* VNET_SW_INTERFACE_TYPE_HARDWARE. */
diff --git a/src/vnet/interface_api.c b/src/vnet/interface_api.c
index 8d982e36b73..116ee63bdba 100644
--- a/src/vnet/interface_api.c
+++ b/src/vnet/interface_api.c
@@ -102,40 +102,11 @@ vl_api_sw_interface_set_mtu_t_handler (vl_api_sw_interface_set_mtu_t * mp)
   vnet_main_t *vnm = vnet_get_main ();
   u32 sw_if_index = ntohl (mp->sw_if_index);
   u16 mtu = ntohs (mp->mtu);
-  ethernet_main_t *em = &ethernet_main;
   int rv = 0;
 
   VALIDATE_SW_IF_INDEX (mp);
 
-  vnet_sw_interface_t *si = vnet_get_sw_interface (vnm, sw_if_index);
-  if (si->type != VNET_SW_INTERFACE_TYPE_HARDWARE)
-    {
-      rv = VNET_API_ERROR_INVALID_VALUE;
-      goto bad_sw_if_index;
-    }
-
-  vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, si->hw_if_index);
-  ethernet_interface_t *eif = ethernet_get_interface (em, si->hw_if_index);
-
-  if (!eif)
-    {
-      rv = VNET_API_ERROR_FEATURE_DISABLED;
-      goto bad_sw_if_index;
-    }
-
-  if (mtu < hi->min_supported_packet_bytes)
-    {
-      rv = VNET_API_ERROR_INVALID_VALUE;
-      goto bad_sw_if_index;
-    }
-
-  if (mtu > hi->max_supported_packet_bytes)
-    {
-      rv = VNET_API_ERROR_INVALID_VALUE;
-      goto bad_sw_if_index;
-    }
-
-  vnet_hw_interface_set_mtu (vnm, si->hw_if_index, mtu);
+  rv = vnet_sw_interface_set_mtu (vnm, sw_if_index, mtu);
 
   BAD_SW_IF_INDEX_LABEL;
   REPLY_MACRO (VL_API_SW_INTERFACE_SET_MTU_REPLY);
@@ -161,7 +132,7 @@ send_sw_interface_details (vpe_api_main_t * am,
 		     VNET_HW_INTERFACE_FLAG_DUPLEX_SHIFT);
   mp->link_speed = ((hi->flags & VNET_HW_INTERFACE_FLAG_SPEED_MASK) >>
 		    VNET_HW_INTERFACE_FLAG_SPEED_SHIFT);
-  mp->link_mtu = ntohs (hi->max_packet_bytes);
+  mp->mtu = ntohs (swif->max_l3_packet_bytes[VLIB_TX]);
   mp->context = context;
 
   strncpy ((char *) mp->interface_name,
diff --git a/src/vnet/interface_cli.c b/src/vnet/interface_cli.c
index d151335aa1f..fe17c823b36 100644
--- a/src/vnet/interface_cli.c
+++ b/src/vnet/interface_cli.c
@@ -1103,32 +1103,29 @@ static clib_error_t *
 mtu_cmd (vlib_main_t * vm, unformat_input_t * input, vlib_cli_command_t * cmd)
 {
   vnet_main_t *vnm = vnet_get_main ();
-  u32 hw_if_index, mtu;
-  ethernet_main_t *em = &ethernet_main;
+  u32 sw_if_index, mtu;
 
-  if (unformat (input, "%d %U", &mtu,
-		unformat_vnet_hw_interface, vnm, &hw_if_index))
+  if (unformat (input, "%d %U", &mtu, unformat_vnet_sw_interface, vnm,
+		&sw_if_index))
     {
-      vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, hw_if_index);
-      ethernet_interface_t *eif = ethernet_get_interface (em, hw_if_index);
-
-      if (!eif)
-	return clib_error_return (0, "not supported");
-
-      if (mtu < hi->min_supported_packet_bytes)
-	return clib_error_return (0, "Invalid mtu (%d): "
-				  "must be >= min pkt bytes (%d)", mtu,
-				  hi->min_supported_packet_bytes);
-
-      if (mtu > hi->max_supported_packet_bytes)
-	return clib_error_return (0, "Invalid mtu (%d): must be <= (%d)", mtu,
-				  hi->max_supported_packet_bytes);
-
-      vnet_hw_interface_set_mtu (vnm, hw_if_index, mtu);
+      ;
     }
   else
-    return clib_error_return (0, "unknown input `%U'",
-			      format_unformat_error, input);
+    {
+      return clib_error_return (0, "unknown input `%U'",
+				format_unformat_error, input);
+    }
+
+  int rv = vnet_sw_interface_set_mtu (vnm, sw_if_index, mtu);
+  if (rv < 0)
+    {
+      vnet_hw_interface_t *hi = vnet_get_sw_hw_interface (vnm, sw_if_index);
+      ASSERT (hi);
+      return clib_error_return (0, "Invalid mtu (%d): "
+				"must be  between min pkt bytes (%d) and max pkt bytes (%d)",
+				mtu, hi->min_packet_bytes,
+				hi->max_packet_bytes);
+    }
   return 0;
 }
 
@@ -1406,7 +1403,7 @@ set_interface_rx_mode (vlib_main_t * vm, unformat_input_t * input,
  * @cliexend
 ?*/
 /* *INDENT-OFF* */
-VLIB_CLI_COMMAND (cmd_set_if_rx_mode,static) = {
+VLIB_CLI_COMMAND (cmd_set_if_rx_mode, static) = {
     .path = "set interface rx-mode",
     .short_help = "set interface rx-mode <interface> [queue <n>] [polling | interrupt | adaptive]",
     .function = set_interface_rx_mode,
diff --git a/src/vnet/interface_funcs.h b/src/vnet/interface_funcs.h
index 6e188f7f543..42b69bdf789 100644
--- a/src/vnet/interface_funcs.h
+++ b/src/vnet/interface_funcs.h
@@ -87,6 +87,15 @@ vnet_get_sup_hw_interface (vnet_main_t * vnm, u32 sw_if_index)
   return vnet_get_hw_interface (vnm, sw->hw_if_index);
 }
 
+always_inline vnet_hw_interface_t *
+vnet_get_sw_hw_interface (vnet_main_t * vnm, u32 sw_if_index)
+{
+  vnet_sw_interface_t *sw = vnet_get_sw_interface (vnm, sw_if_index);
+  if (sw->type == VNET_SW_INTERFACE_TYPE_HARDWARE)
+    return vnet_get_hw_interface (vnm, sw->hw_if_index);
+  return vnet_get_sup_hw_interface (vnm, sw_if_index);
+}
+
 always_inline vnet_hw_interface_class_t *
 vnet_get_hw_interface_class (vnet_main_t * vnm, u32 hw_class_index)
 {
@@ -219,19 +228,11 @@ vnet_hw_interface_get_flags (vnet_main_t * vnm, u32 hw_if_index)
 }
 
 always_inline uword
-vnet_hw_interface_get_mtu (vnet_main_t * vnm, u32 hw_if_index,
-			   vlib_rx_or_tx_t dir)
-{
-  vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index);
-  return hw->max_l3_packet_bytes[dir];
-}
-
-always_inline uword
 vnet_sw_interface_get_mtu (vnet_main_t * vnm, u32 sw_if_index,
 			   vlib_rx_or_tx_t dir)
 {
-  vnet_hw_interface_t *hw = vnet_get_sup_hw_interface (vnm, sw_if_index);
-  return (hw->max_l3_packet_bytes[dir]);
+  vnet_sw_interface_t *sw = vnet_get_sw_interface (vnm, sw_if_index);
+  return (sw->max_l3_packet_bytes[dir]);
 }
 
 always_inline uword
@@ -292,8 +293,8 @@ clib_error_t *set_hw_interface_change_rx_mode (vnet_main_t * vnm,
 					       vnet_hw_interface_rx_mode
 					       mode);
 
-/* Set the MTU on the HW interface */
-void vnet_hw_interface_set_mtu (vnet_main_t * vnm, u32 hw_if_index, u32 mtu);
+/* Set the MTU on the SW interface */
+int vnet_sw_interface_set_mtu (vnet_main_t * vnm, u32 sw_if_index, u32 mtu);
 
 /* update the unnumbered state of an interface */
 void vnet_sw_interface_update_unnumbered (u32 sw_if_index,
diff --git a/src/vnet/ip/icmp4.c b/src/vnet/ip/icmp4.c
index 1fe01e48500..a4808f23271 100644
--- a/src/vnet/ip/icmp4.c
+++ b/src/vnet/ip/icmp4.c
@@ -513,13 +513,15 @@ ip4_icmp_error (vlib_main_t * vm,
 		  b->current_length = 0;
 		}
 	    }
-	  p0->current_length =
-	    p0->current_length > 576 ? 576 : p0->current_length;
 
 	  /* Add IP header and ICMPv4 header including a 4 byte data field */
 	  vlib_buffer_advance (p0,
 			       -sizeof (ip4_header_t) -
 			       sizeof (icmp46_header_t) - 4);
+
+	  p0->current_length =
+	    p0->current_length > 576 ? 576 : p0->current_length;
+
 	  out_ip0 = vlib_buffer_get_current (p0);
 	  icmp0 = (icmp46_header_t *) & out_ip0[1];
 
diff --git a/src/vnet/ip/icmp6.c b/src/vnet/ip/icmp6.c
index fd5d0ecba38..6beec28c2cf 100644
--- a/src/vnet/ip/icmp6.c
+++ b/src/vnet/ip/icmp6.c
@@ -526,13 +526,14 @@ ip6_icmp_error (vlib_main_t * vm,
 		  b->current_length = 0;
 		}
 	    }
-	  p0->current_length =
-	    p0->current_length > 1280 ? 1280 : p0->current_length;
-
 	  /* 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);
+
+	  p0->current_length =
+	    p0->current_length > 1280 ? 1280 : p0->current_length;
+
 	  out_ip0 = vlib_buffer_get_current (p0);
 	  icmp0 = (icmp46_header_t *) & out_ip0[1];
 
diff --git a/src/vnet/ip/ip4_forward.c b/src/vnet/ip/ip4_forward.c
index 7c56a294436..3dce590c807 100644
--- a/src/vnet/ip/ip4_forward.c
+++ b/src/vnet/ip/ip4_forward.c
@@ -1940,6 +1940,29 @@ typedef enum
   IP4_REWRITE_NEXT_ICMP_ERROR,
 } ip4_rewrite_next_t;
 
+always_inline void
+ip4_mtu_check (vlib_buffer_t * b, u16 buffer_packet_bytes,
+	       u16 adj_packet_bytes, bool df, u32 * next, u32 * error)
+{
+  if (buffer_packet_bytes > adj_packet_bytes)
+    {
+      *error = IP4_ERROR_MTU_EXCEEDED;
+      if (df)
+	{
+	  icmp4_error_set_vnet_buffer
+	    (b, ICMP4_destination_unreachable,
+	     ICMP4_destination_unreachable_fragmentation_needed_and_dont_fragment_set,
+	     adj_packet_bytes);
+	  *next = IP4_REWRITE_NEXT_ICMP_ERROR;
+	}
+      else
+	{
+	  /* Add support for fragmentation here */
+	  *next = IP4_REWRITE_NEXT_DROP;
+	}
+    }
+}
+
 always_inline uword
 ip4_rewrite_inline (vlib_main_t * vm,
 		    vlib_node_runtime_t * node,
@@ -2100,26 +2123,20 @@ ip4_rewrite_inline (vlib_main_t * vm,
 	  vnet_buffer (p1)->ip.save_rewrite_length = rw_len1;
 
 	  /* Check MTU of outgoing interface. */
-	  if (vlib_buffer_length_in_chain (vm, p0) >
-	      adj0[0].rewrite_header.max_l3_packet_bytes)
-	    {
-	      error0 = IP4_ERROR_MTU_EXCEEDED;
-	      next0 = IP4_REWRITE_NEXT_ICMP_ERROR;
-	      icmp4_error_set_vnet_buffer
-		(p0, ICMP4_destination_unreachable,
-		 ICMP4_destination_unreachable_fragmentation_needed_and_dont_fragment_set,
-		 0);
-	    }
-	  if (vlib_buffer_length_in_chain (vm, p1) >
-	      adj1[0].rewrite_header.max_l3_packet_bytes)
-	    {
-	      error1 = IP4_ERROR_MTU_EXCEEDED;
-	      next1 = IP4_REWRITE_NEXT_ICMP_ERROR;
-	      icmp4_error_set_vnet_buffer
-		(p1, ICMP4_destination_unreachable,
-		 ICMP4_destination_unreachable_fragmentation_needed_and_dont_fragment_set,
-		 0);
-	    }
+	  ip4_mtu_check (p0, vlib_buffer_length_in_chain (vm, p0),
+			 adj0[0].rewrite_header.max_l3_packet_bytes,
+			 ip0->flags_and_fragment_offset &
+			 clib_host_to_net_u16 (IP4_HEADER_FLAG_DONT_FRAGMENT),
+			 &next0, &error0);
+	  ip4_mtu_check (p1, vlib_buffer_length_in_chain (vm, p1),
+			 adj1[0].rewrite_header.max_l3_packet_bytes,
+			 ip1->flags_and_fragment_offset &
+			 clib_host_to_net_u16 (IP4_HEADER_FLAG_DONT_FRAGMENT),
+			 &next1, &error1);
+
+	  /* Guess we are only writing on simple Ethernet header. */
+	  vnet_rewrite_two_headers (adj0[0], adj1[0],
+				    ip0, ip1, sizeof (ethernet_header_t));
 
 	  if (is_mcast)
 	    {
@@ -2143,10 +2160,17 @@ ip4_rewrite_inline (vlib_main_t * vm,
 	      tx_sw_if_index0 = adj0[0].rewrite_header.sw_if_index;
 	      vnet_buffer (p0)->sw_if_index[VLIB_TX] = tx_sw_if_index0;
 
+	      if (is_midchain)
+		{
+		  adj0->sub_type.midchain.fixup_func
+		    (vm, adj0, p0, adj0->sub_type.midchain.fixup_data);
+		}
+
 	      if (PREDICT_FALSE
 		  (adj0[0].rewrite_header.flags & VNET_REWRITE_HAS_FEATURES))
 		vnet_feature_arc_start (lm->output_feature_arc_index,
 					tx_sw_if_index0, &next0, p0);
+
 	    }
 	  if (PREDICT_TRUE (error1 == IP4_ERROR_NONE))
 	    {
@@ -2157,16 +2181,18 @@ ip4_rewrite_inline (vlib_main_t * vm,
 	      tx_sw_if_index1 = adj1[0].rewrite_header.sw_if_index;
 	      vnet_buffer (p1)->sw_if_index[VLIB_TX] = tx_sw_if_index1;
 
+	      if (is_midchain)
+		{
+		  adj1->sub_type.midchain.fixup_func
+		    (vm, adj1, p1, adj0->sub_type.midchain.fixup_data);
+		}
+
 	      if (PREDICT_FALSE
 		  (adj1[0].rewrite_header.flags & VNET_REWRITE_HAS_FEATURES))
 		vnet_feature_arc_start (lm->output_feature_arc_index,
 					tx_sw_if_index1, &next1, p1);
 	    }
 
-	  /* Guess we are only writing on simple Ethernet header. */
-	  vnet_rewrite_two_headers (adj0[0], adj1[0],
-				    ip0, ip1, sizeof (ethernet_header_t));
-
 	  /*
 	   * Bump the per-adjacency counters
 	   */
@@ -2185,13 +2211,6 @@ ip4_rewrite_inline (vlib_main_t * vm,
 		 vlib_buffer_length_in_chain (vm, p1) + rw_len1);
 	    }
 
-	  if (is_midchain)
-	    {
-	      adj0->sub_type.midchain.fixup_func
-		(vm, adj0, p0, adj0->sub_type.midchain.fixup_data);
-	      adj1->sub_type.midchain.fixup_func
-		(vm, adj1, p1, adj0->sub_type.midchain.fixup_data);
-	    }
 	  if (is_mcast)
 	    {
 	      /*
@@ -2272,6 +2291,7 @@ ip4_rewrite_inline (vlib_main_t * vm,
 
 	  /* Guess we are only writing on simple Ethernet header. */
 	  vnet_rewrite_one_header (adj0[0], ip0, sizeof (ethernet_header_t));
+
 	  if (is_mcast)
 	    {
 	      /*
@@ -2291,16 +2311,12 @@ ip4_rewrite_inline (vlib_main_t * vm,
 	       vlib_buffer_length_in_chain (vm, p0) + rw_len0);
 
 	  /* Check MTU of outgoing interface. */
-	  if (vlib_buffer_length_in_chain (vm, p0) >
-	      adj0[0].rewrite_header.max_l3_packet_bytes)
-	    {
-	      error0 = IP4_ERROR_MTU_EXCEEDED;
-	      next0 = IP4_REWRITE_NEXT_ICMP_ERROR;
-	      icmp4_error_set_vnet_buffer
-		(p0, ICMP4_destination_unreachable,
-		 ICMP4_destination_unreachable_fragmentation_needed_and_dont_fragment_set,
-		 0);
-	    }
+	  ip4_mtu_check (p0, vlib_buffer_length_in_chain (vm, p0),
+			 adj0[0].rewrite_header.max_l3_packet_bytes,
+			 ip0->flags_and_fragment_offset &
+			 clib_host_to_net_u16 (IP4_HEADER_FLAG_DONT_FRAGMENT),
+			 &next0, &error0);
+
 	  if (is_mcast)
 	    {
 	      error0 = ((adj0[0].rewrite_header.sw_if_index ==
diff --git a/src/vnet/ip/ip6_forward.c b/src/vnet/ip/ip6_forward.c
index 588cd0675a4..7599733fcb5 100644
--- a/src/vnet/ip/ip6_forward.c
+++ b/src/vnet/ip/ip6_forward.c
@@ -1774,6 +1774,19 @@ typedef enum
   IP6_REWRITE_NEXT_ICMP_ERROR,
 } ip6_rewrite_next_t;
 
+always_inline void
+ip6_mtu_check (vlib_buffer_t * b, u16 buffer_packet_bytes,
+	       u16 adj_packet_bytes, u32 * next, u32 * error)
+{
+  if (adj_packet_bytes >= 1280 && buffer_packet_bytes > adj_packet_bytes)
+    {
+      *error = IP6_ERROR_MTU_EXCEEDED;
+      icmp6_error_set_vnet_buffer (b, ICMP6_packet_too_big, 0,
+				   adj_packet_bytes);
+      *next = IP6_REWRITE_NEXT_ICMP_ERROR;
+    }
+}
+
 always_inline uword
 ip6_rewrite_inline (vlib_main_t * vm,
 		    vlib_node_runtime_t * node,
@@ -1898,9 +1911,14 @@ ip6_rewrite_inline (vlib_main_t * vm,
 	    {
 	      p1->flags &= ~VNET_BUFFER_F_LOCALLY_ORIGINATED;
 	    }
+
 	  adj0 = adj_get (adj_index0);
 	  adj1 = adj_get (adj_index1);
 
+	  /* Guess we are only writing on simple Ethernet header. */
+	  vnet_rewrite_two_headers (adj0[0], adj1[0],
+				    ip0, ip1, sizeof (ethernet_header_t));
+
 	  rw_len0 = adj0[0].rewrite_header.data_bytes;
 	  rw_len1 = adj1[0].rewrite_header.data_bytes;
 	  vnet_buffer (p0)->ip.save_rewrite_length = rw_len0;
@@ -1919,16 +1937,12 @@ ip6_rewrite_inline (vlib_main_t * vm,
 	    }
 
 	  /* Check MTU of outgoing interface. */
-	  error0 =
-	    (vlib_buffer_length_in_chain (vm, p0) >
-	     adj0[0].
-	     rewrite_header.max_l3_packet_bytes ? IP6_ERROR_MTU_EXCEEDED :
-	     error0);
-	  error1 =
-	    (vlib_buffer_length_in_chain (vm, p1) >
-	     adj1[0].
-	     rewrite_header.max_l3_packet_bytes ? IP6_ERROR_MTU_EXCEEDED :
-	     error1);
+	  ip6_mtu_check (p0, vlib_buffer_length_in_chain (vm, p0),
+			 adj0[0].rewrite_header.max_l3_packet_bytes, &next0,
+			 &error0);
+	  ip6_mtu_check (p1, vlib_buffer_length_in_chain (vm, p1),
+			 adj1[0].rewrite_header.max_l3_packet_bytes, &next1,
+			 &error1);
 
 	  /* Don't adjust the buffer for hop count issue; icmp-error node
 	   * wants to see the IP headerr */
@@ -1945,6 +1959,19 @@ ip6_rewrite_inline (vlib_main_t * vm,
 		  (adj0[0].rewrite_header.flags & VNET_REWRITE_HAS_FEATURES))
 		vnet_feature_arc_start (lm->output_feature_arc_index,
 					tx_sw_if_index0, &next0, p0);
+
+	      if (is_midchain)
+		{
+		  adj0->sub_type.midchain.fixup_func
+		    (vm, adj0, p0, adj0->sub_type.midchain.fixup_data);
+		}
+	      if (is_mcast)
+		{
+		  /*
+		   * copy bytes from the IP address into the MAC rewrite
+		   */
+		  vnet_fixup_one_header (adj0[0], &ip0->dst_address, ip0);
+		}
 	    }
 	  if (PREDICT_TRUE (error1 == IP6_ERROR_NONE))
 	    {
@@ -1959,26 +1986,19 @@ ip6_rewrite_inline (vlib_main_t * vm,
 		  (adj1[0].rewrite_header.flags & VNET_REWRITE_HAS_FEATURES))
 		vnet_feature_arc_start (lm->output_feature_arc_index,
 					tx_sw_if_index1, &next1, p1);
-	    }
 
-	  /* Guess we are only writing on simple Ethernet header. */
-	  vnet_rewrite_two_headers (adj0[0], adj1[0],
-				    ip0, ip1, sizeof (ethernet_header_t));
-
-	  if (is_midchain)
-	    {
-	      adj0->sub_type.midchain.fixup_func
-		(vm, adj0, p0, adj0->sub_type.midchain.fixup_data);
-	      adj1->sub_type.midchain.fixup_func
-		(vm, adj1, p1, adj1->sub_type.midchain.fixup_data);
-	    }
-	  if (is_mcast)
-	    {
-	      /*
-	       * copy bytes from the IP address into the MAC rewrite
-	       */
-	      vnet_fixup_one_header (adj0[0], &ip0->dst_address, ip0);
-	      vnet_fixup_one_header (adj1[0], &ip1->dst_address, ip1);
+	      if (is_midchain)
+		{
+		  adj1->sub_type.midchain.fixup_func
+		    (vm, adj1, p1, adj1->sub_type.midchain.fixup_data);
+		}
+	      if (is_mcast)
+		{
+		  /*
+		   * copy bytes from the IP address into the MAC rewrite
+		   */
+		  vnet_fixup_one_header (adj1[0], &ip1->dst_address, ip1);
+		}
 	    }
 
 	  vlib_validate_buffer_enqueue_x2 (vm, node, next_index,
@@ -2054,11 +2074,9 @@ ip6_rewrite_inline (vlib_main_t * vm,
 	    }
 
 	  /* Check MTU of outgoing interface. */
-	  error0 =
-	    (vlib_buffer_length_in_chain (vm, p0) >
-	     adj0[0].
-	     rewrite_header.max_l3_packet_bytes ? IP6_ERROR_MTU_EXCEEDED :
-	     error0);
+	  ip6_mtu_check (p0, vlib_buffer_length_in_chain (vm, p0),
+			 adj0[0].rewrite_header.max_l3_packet_bytes, &next0,
+			 &error0);
 
 	  /* Don't adjust the buffer for hop count issue; icmp-error node
 	   * wants to see the IP headerr */
@@ -2076,16 +2094,16 @@ ip6_rewrite_inline (vlib_main_t * vm,
 		  (adj0[0].rewrite_header.flags & VNET_REWRITE_HAS_FEATURES))
 		vnet_feature_arc_start (lm->output_feature_arc_index,
 					tx_sw_if_index0, &next0, p0);
-	    }
 
-	  if (is_midchain)
-	    {
-	      adj0->sub_type.midchain.fixup_func
-		(vm, adj0, p0, adj0->sub_type.midchain.fixup_data);
-	    }
-	  if (is_mcast)
-	    {
-	      vnet_fixup_one_header (adj0[0], &ip0->dst_address, ip0);
+	      if (is_midchain)
+		{
+		  adj0->sub_type.midchain.fixup_func
+		    (vm, adj0, p0, adj0->sub_type.midchain.fixup_data);
+		}
+	      if (is_mcast)
+		{
+		  vnet_fixup_one_header (adj0[0], &ip0->dst_address, ip0);
+		}
 	    }
 
 	  p0->error = error_node->errors[error0];
diff --git a/src/vnet/ip/ip6_neighbor.c b/src/vnet/ip/ip6_neighbor.c
index fee4356f5e0..a1439faa154 100644
--- a/src/vnet/ip/ip6_neighbor.c
+++ b/src/vnet/ip/ip6_neighbor.c
@@ -2575,10 +2575,6 @@ ip6_neighbor_sw_interface_add_del (vnet_main_t * vnm,
     {
       if (is_add)
 	{
-	  vnet_hw_interface_t *hw_if0;
-
-	  hw_if0 = vnet_get_sup_hw_interface (vnm, sw_if_index);
-
 	  pool_get (nm->if_radv_pool, a);
 
 	  ri = a - nm->if_radv_pool;
@@ -2612,7 +2608,8 @@ ip6_neighbor_sw_interface_add_del (vnet_main_t * vnm,
 	  a->send_radv = 1;
 
 	  /* fill in radv_info for this interface that will be needed later */
-	  a->adv_link_mtu = hw_if0->max_l3_packet_bytes[VLIB_RX];
+	  a->adv_link_mtu =
+	    vnet_sw_interface_get_mtu (vnm, sw_if_index, VLIB_TX);
 
 	  clib_memcpy (a->link_layer_address, eth_if0->address, 6);
 
diff --git a/src/vnet/ipip/ipip.c b/src/vnet/ipip/ipip.c
index 82c961cdddd..9b808d4ac0f 100644
--- a/src/vnet/ipip/ipip.c
+++ b/src/vnet/ipip/ipip.c
@@ -476,17 +476,16 @@ ipip_add_tunnel (ipip_transport_t transport,
     {
       vec_validate (im4->fib_index_by_sw_if_index, sw_if_index);
       hi->min_packet_bytes = 64 + sizeof (ip4_header_t);
+      hi->max_packet_bytes = 65536 - sizeof (ip4_header_t);
     }
   else
     {
       vec_validate (im6->fib_index_by_sw_if_index, sw_if_index);
       hi->min_packet_bytes = 64 + sizeof (ip6_header_t);
+      hi->max_packet_bytes = 65536 - sizeof (ip6_header_t);
     }
 
-  hi->per_packet_overhead_bytes = /* preamble */ 8 + /* inter frame gap */ 12;
-
-  /* Standard default ipip MTU. */
-  hi->max_l3_packet_bytes[VLIB_RX] = hi->max_l3_packet_bytes[VLIB_TX] = 9000;
+  vnet_sw_interface_set_mtu (vnm, sw_if_index, hi->max_packet_bytes);
 
   t->tunnel_src = *src;
   t->tunnel_dst = *dst;
diff --git a/src/vnet/ipip/sixrd.c b/src/vnet/ipip/sixrd.c
index cfdd0f87e3a..998025782fb 100644
--- a/src/vnet/ipip/sixrd.c
+++ b/src/vnet/ipip/sixrd.c
@@ -337,7 +337,8 @@ sixrd_add_tunnel (ip6_address_t * ip6_prefix, u8 ip6_prefix_len,
   t->dev_instance = t_idx;
   t->user_instance = t_idx;
 
-  hi->max_l3_packet_bytes[VLIB_RX] = hi->max_l3_packet_bytes[VLIB_TX] = 1480;
+  /* Set default MTU to 1480 */
+  vnet_sw_interface_set_mtu (vnet_get_main (), t->sw_if_index, 1480);
 
   ipip_tunnel_db_add (t, &key);
 
diff --git a/src/vnet/ipsec-gre/interface.c b/src/vnet/ipsec-gre/interface.c
index 0772ce73df2..fa33684c50d 100644
--- a/src/vnet/ipsec-gre/interface.c
+++ b/src/vnet/ipsec-gre/interface.c
@@ -174,12 +174,9 @@ vnet_ipsec_gre_add_del_tunnel (vnet_ipsec_gre_add_del_tunnel_args_t * a,
 
       hi->min_packet_bytes = 64 + sizeof (gre_header_t) +
 	sizeof (ip4_header_t) + sizeof (esp_header_t) + sizeof (esp_footer_t);
-      hi->per_packet_overhead_bytes =
-	/* preamble */ 8 + /* inter frame gap */ 12;
 
       /* Standard default gre MTU. */
-      hi->max_l3_packet_bytes[VLIB_RX] = hi->max_l3_packet_bytes[VLIB_TX] =
-	9000;
+      vnet_sw_interface_set_mtu (vnm, hi->sw_if_index, 9000);
 
       clib_memcpy (&t->tunnel_src, &a->src, sizeof (t->tunnel_src));
       clib_memcpy (&t->tunnel_dst, &a->dst, sizeof (t->tunnel_dst));
diff --git a/src/vnet/unix/tapcli.c b/src/vnet/unix/tapcli.c
index 10a86a457e8..5c550bc3320 100644
--- a/src/vnet/unix/tapcli.c
+++ b/src/vnet/unix/tapcli.c
@@ -1068,8 +1068,9 @@ vnet_tap_connect (vlib_main_t * vm, vnet_tap_connect_args_t * ap)
     hw = vnet_get_hw_interface (tm->vnet_main, ti->hw_if_index);
     hw->min_supported_packet_bytes = TAP_MTU_MIN;
     hw->max_supported_packet_bytes = TAP_MTU_MAX;
-    hw->max_l3_packet_bytes[VLIB_RX] = hw->max_l3_packet_bytes[VLIB_TX] =
-      hw->max_supported_packet_bytes - sizeof (ethernet_header_t);
+    vnet_sw_interface_set_mtu (tm->vnet_main, hw->sw_if_index,
+			       hw->max_supported_packet_bytes -
+			       sizeof (ethernet_header_t));
     ti->sw_if_index = hw->sw_if_index;
     if (ap->sw_if_indexp)
       *(ap->sw_if_indexp) = hw->sw_if_index;