tcp: handle link-local addresses

Change-Id: I9ede6bc861350c7d9e78fa4d96cd584c2816d06f
Signed-off-by: Florin Coras <fcoras@cisco.com>

4 files changed, 66 insertions, 7 deletions

diff --git a/src/vnet/tcp/tcp.h b/src/vnet/tcp/tcp.h
index 8a6f74ab13a..837b5b4d0d2 100644
--- a/src/vnet/tcp/tcp.h
+++ b/src/vnet/tcp/tcp.h
@@ -305,6 +305,7 @@ typedef struct _tcp_connection
   u16 mss;		/**< Our max seg size that includes options */
   u32 limited_transmit;	/**< snd_nxt when limited transmit starts */
   u32 last_fib_check;	/**< Last time we checked fib route for peer */
+  u32 sw_if_index;	/**< Interface for the connection */
 } tcp_connection_t;
 
 struct _tcp_cc_algorithm
diff --git a/src/vnet/tcp/tcp_error.def b/src/vnet/tcp/tcp_error.def
index 8f006dfc287..44bf0c04168 100644
--- a/src/vnet/tcp/tcp_error.def
+++ b/src/vnet/tcp/tcp_error.def
@@ -44,4 +44,5 @@ tcp_error (FILTERED, "Packets filtered")
 tcp_error (OPTIONS, "Could not parse options")
 tcp_error (PAWS, "PAWS check failed")
 tcp_error (RCV_WND, "Segment not in receive window")
-tcp_error (FIN_RCVD, "FINs received")
\ No newline at end of file
+tcp_error (FIN_RCVD, "FINs received")
+tcp_error (LINK_LOCAL_RW, "No rewrite for link local connection")
\ No newline at end of file
diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c
index 497b709bec0..42db82e1214 100644
--- a/src/vnet/tcp/tcp_input.c
+++ b/src/vnet/tcp/tcp_input.c
@@ -2126,6 +2126,7 @@ tcp46_syn_sent_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	  new_tc0->timers[TCP_TIMER_ESTABLISH] = TCP_TIMER_HANDLE_INVALID;
 	  new_tc0->timers[TCP_TIMER_RETRANSMIT_SYN] =
 	    TCP_TIMER_HANDLE_INVALID;
+	  new_tc0->sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_RX];
 
 	  /* If this is not the owning thread, wait for syn retransmit to
 	   * expire and cleanup then */
@@ -2818,6 +2819,7 @@ tcp46_listen_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	  child0->irs = vnet_buffer (b0)->tcp.seq_number;
 	  child0->rcv_nxt = vnet_buffer (b0)->tcp.seq_number + 1;
 	  child0->rcv_las = child0->rcv_nxt;
+	  child0->sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_RX];
 
 	  /* RFC1323: TSval timestamps sent on {SYN} and {SYN,ACK}
 	   * segments are used to initialize PAWS. */
diff --git a/src/vnet/tcp/tcp_output.c b/src/vnet/tcp/tcp_output.c
index df882bc5ff1..fe6d09c2c0e 100644
--- a/src/vnet/tcp/tcp_output.c
+++ b/src/vnet/tcp/tcp_output.c
@@ -24,16 +24,22 @@ typedef enum _tcp_output_next
 {
   TCP_OUTPUT_NEXT_DROP,
   TCP_OUTPUT_NEXT_IP_LOOKUP,
+  TCP_OUTPUT_NEXT_IP_REWRITE,
+  TCP_OUTPUT_NEXT_IP_ARP,
   TCP_OUTPUT_N_NEXT
 } tcp_output_next_t;
 
 #define foreach_tcp4_output_next              	\
   _ (DROP, "error-drop")                        \
-  _ (IP_LOOKUP, "ip4-lookup")
+  _ (IP_LOOKUP, "ip4-lookup")			\
+  _ (IP_REWRITE, "ip4-rewrite")			\
+  _ (IP_ARP, "ip4-arp")
 
 #define foreach_tcp6_output_next              	\
   _ (DROP, "error-drop")                        \
-  _ (IP_LOOKUP, "ip6-lookup")
+  _ (IP_LOOKUP, "ip6-lookup")			\
+  _ (IP_REWRITE, "ip6-rewrite")			\
+  _ (IP_ARP, "ip6-discover-neighbor")
 
 static char *tcp_error_strings[] = {
 #define tcp_error(n,s) s,
@@ -1747,6 +1753,53 @@ tcp_session_has_ooo_data (tcp_connection_t * tc)
   return svm_fifo_has_ooo_data (s->server_rx_fifo);
 }
 
+typedef struct
+{
+  fib_protocol_t nh_proto;
+  vnet_link_t link_type;
+  ip46_address_t ip;
+  u32 sw_if_index;
+} tcp_adj_add_args_t;
+
+void
+tcp_output_add_adj (tcp_adj_add_args_t * args)
+{
+  adj_nbr_add_or_lock (args->nh_proto, args->link_type, &args->ip,
+		       args->sw_if_index);
+}
+
+static void
+tcp_output_handle_link_local (tcp_connection_t * tc0, vlib_buffer_t * b0,
+			      u32 * next0, u32 * error0)
+{
+  ip_adjacency_t *adj;
+  adj_index_t ai;
+
+  ai = adj_nbr_find (FIB_PROTOCOL_IP6, VNET_LINK_IP6, &tc0->c_rmt_ip,
+		     tc0->sw_if_index);
+  if (ai == ADJ_INDEX_INVALID)
+    {
+      tcp_adj_add_args_t args = {
+	.nh_proto = FIB_PROTOCOL_IP6,
+	.link_type = VNET_LINK_IP6,
+	.ip = tc0->c_rmt_ip,
+	.sw_if_index = tc0->sw_if_index
+      };
+      vlib_rpc_call_main_thread (tcp_output_add_adj, (u8 *) & args,
+				 sizeof (args));
+      vnet_buffer (b0)->sw_if_index[VLIB_TX] = ~0;
+      *next0 = TCP_OUTPUT_NEXT_DROP;
+      *error0 = TCP_ERROR_LINK_LOCAL_RW;
+      return;
+    }
+
+  adj = adj_get (ai);
+  if (adj->lookup_next_index == IP_LOOKUP_NEXT_REWRITE)
+    *next0 = TCP_OUTPUT_NEXT_IP_REWRITE;
+  else if (adj->lookup_next_index == IP_LOOKUP_NEXT_ARP)
+    *next0 = TCP_OUTPUT_NEXT_IP_ARP;
+}
+
 always_inline uword
 tcp46_output_inline (vlib_main_t * vm,
 		     vlib_node_runtime_t * node,
@@ -1802,6 +1855,8 @@ tcp46_output_inline (vlib_main_t * vm,
 
 	  th0 = vlib_buffer_get_current (b0);
 	  TCP_EVT_DBG (TCP_EVT_OUTPUT, tc0, th0->flags, b0->current_length);
+	  vnet_buffer (b0)->sw_if_index[VLIB_TX] = tc0->c_fib_index;
+	  vnet_buffer (b0)->sw_if_index[VLIB_RX] = 0;
 
 	  if (is_ip4)
 	    {
@@ -1820,6 +1875,10 @@ tcp46_output_inline (vlib_main_t * vm,
 	      vnet_buffer (b0)->l3_hdr_offset = (u8 *) ih0 - b0->data;
 	      vnet_buffer (b0)->l4_hdr_offset = (u8 *) th0 - b0->data;
 	      th0->checksum = 0;
+
+	      if (PREDICT_FALSE
+		  (ip6_address_is_link_local_unicast (&tc0->c_rmt_ip6)))
+		tcp_output_handle_link_local (tc0, b0, &next0, &error0);
 	    }
 
 	  /* Filter out DUPACKs if there are no OOO segments left */
@@ -1889,10 +1948,6 @@ tcp46_output_inline (vlib_main_t * vm,
 	  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = tc0->c_rmt_dpo.dpoi_index;
 #endif
 
-	  vnet_buffer (b0)->sw_if_index[VLIB_RX] = 0;
-	  vnet_buffer (b0)->sw_if_index[VLIB_TX] = tc0->c_fib_index;
-
-	  b0->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
 	done:
 	  b0->error = node->errors[error0];
 	  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))