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

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

5 files changed, 128 insertions, 94 deletions

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);