tcp/session: drop connections if message queue is full

Also cleanup session table when transport closes and app didn't reply to
the close notification.

Change-Id: Ie3d518e3afff73437561561b46dbf695c24632ad
Signed-off-by: Florin Coras <fcoras@cisco.com>

4 files changed, 28 insertions, 21 deletions

diff --git a/src/vnet/session/session.c b/src/vnet/session/session.c
index 7f9a32f46de..a03b53dd3bd 100644
--- a/src/vnet/session/session.c
+++ b/src/vnet/session/session.c
@@ -194,7 +194,6 @@ session_alloc_for_connection (transport_connection_t * tc)
 
   s = session_alloc (thread_index);
   s->session_type = session_type_from_proto_and_ip (tc->proto, tc->is_ip4);
-  s->session_state = SESSION_STATE_CONNECTING;
   s->enqueue_epoch = (u64) ~ 0;
 
   /* Attach transport to session and vice versa */
@@ -629,6 +628,7 @@ session_stream_connect_notify (transport_connection_t * tc, u8 is_fail)
 	}
       else
 	{
+	  new_s->session_state = SESSION_STATE_CONNECTING;
 	  new_s->app_wrk_index = app_wrk->wrk_index;
 	  new_si = new_s->session_index;
 	  new_ti = new_s->thread_index;
@@ -723,7 +723,7 @@ session_dgram_connect_notify (transport_connection_t * tc,
   return 0;
 }
 
-void
+int
 stream_session_accept_notify (transport_connection_t * tc)
 {
   app_worker_t *app_wrk;
@@ -733,9 +733,9 @@ stream_session_accept_notify (transport_connection_t * tc)
   s = session_get (tc->s_index, tc->thread_index);
   app_wrk = app_worker_get_if_valid (s->app_wrk_index);
   if (!app_wrk)
-    return;
+    return -1;
   app = application_get (app_wrk->app_index);
-  app->cb_fns.session_accept_callback (s);
+  return app->cb_fns.session_accept_callback (s);
 }
 
 /**
@@ -805,7 +805,10 @@ stream_session_delete_notify (transport_connection_t * tc)
     case SESSION_STATE_TRANSPORT_CLOSING:
       /* If transport finishes or times out before we get a reply
        * from the app, do the whole disconnect since we might still
-       * have lingering events */
+       * have lingering events. Cleanup session table in advance
+       * because transport will soon be closed and closed sessions
+       * are assumed to have been removed from the lookup table */
+      session_lookup_del_session (s);
       stream_session_disconnect (s);
       s->session_state = SESSION_STATE_CLOSED;
       break;
@@ -872,7 +875,7 @@ stream_session_accept (transport_connection_t * tc, u32 listener_index,
   if (notify)
     {
       application_t *app = application_get (app_wrk->app_index);
-      app->cb_fns.session_accept_callback (s);
+      return app->cb_fns.session_accept_callback (s);
     }
 
   return 0;
diff --git a/src/vnet/session/session.h b/src/vnet/session/session.h
index be2490fbe95..6c1bdb639df 100644
--- a/src/vnet/session/session.h
+++ b/src/vnet/session/session.h
@@ -610,7 +610,7 @@ int session_dequeue_notify (stream_session_t * s);
 void stream_session_init_fifos_pointers (transport_connection_t * tc,
 					 u32 rx_pointer, u32 tx_pointer);
 
-void stream_session_accept_notify (transport_connection_t * tc);
+int stream_session_accept_notify (transport_connection_t * tc);
 void stream_session_disconnect_notify (transport_connection_t * tc);
 void stream_session_delete_notify (transport_connection_t * tc);
 void stream_session_reset_notify (transport_connection_t * tc);
diff --git a/src/vnet/tcp/tcp_error.def b/src/vnet/tcp/tcp_error.def
index 44bf0c04168..141ca515995 100644
--- a/src/vnet/tcp/tcp_error.def
+++ b/src/vnet/tcp/tcp_error.def
@@ -25,7 +25,7 @@ tcp_error (SEGMENT_OLD, "Old segment")
 tcp_error (SEGMENT_INVALID, "Invalid segments")
 tcp_error (SYNS_RCVD, "SYNs received")
 tcp_error (SYN_ACKS_RCVD, "SYN-ACKs received")
-tcp_error (EVENT_FIFO_FULL, "Events not sent for lack of event fifo space") 
+tcp_error (MSG_QUEUE_FULL, "Events not sent for lack of msg queue space") 
 tcp_error (CREATE_SESSION_FAIL, "Sessions couldn't be allocated")
 tcp_error (ACK_OK, "Pure ACKs received")
 tcp_error (ACK_INVALID, "Invalid ACK")
diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c
index 91876feaf42..8b16275359c 100644
--- a/src/vnet/tcp/tcp_input.c
+++ b/src/vnet/tcp/tcp_input.c
@@ -183,11 +183,7 @@ tcp_options_parse (tcp_header_t * th, tcp_options_t * to, u8 is_syn)
 	      to->flags |= TCP_OPTS_FLAG_WSCALE;
 	      to->wscale = data[2];
 	      if (to->wscale > TCP_MAX_WND_SCALE)
-		{
-		  clib_warning ("Illegal window scaling value: %d",
-				to->wscale);
-		  to->wscale = TCP_MAX_WND_SCALE;
-		}
+		to->wscale = TCP_MAX_WND_SCALE;
 	    }
 	  break;
 	case TCP_OPTION_TIMESTAMP:
@@ -1598,7 +1594,10 @@ process_ack:
     {
       tcp_cc_handle_event (tc, is_dack);
       if (!tcp_in_cong_recovery (tc))
-	return 0;
+	{
+	  *error = TCP_ERROR_ACK_OK;
+	  return 0;
+	}
       *error = TCP_ERROR_ACK_DUP;
       if (vnet_buffer (b)->tcp.data_len || tcp_is_fin (th))
 	return 0;
@@ -2169,7 +2168,7 @@ tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 
   errors = session_manager_flush_enqueue_events (TRANSPORT_PROTO_TCP,
 						 thread_index);
-  err_counters[TCP_ERROR_EVENT_FIFO_FULL] = errors;
+  err_counters[TCP_ERROR_MSG_QUEUE_FULL] = errors;
   tcp_store_err_counters (established, err_counters);
   tcp_handle_postponed_dequeues (wrk);
   tcp_handle_disconnects (wrk);
@@ -2569,7 +2568,7 @@ tcp46_syn_sent_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 
   errors = session_manager_flush_enqueue_events (TRANSPORT_PROTO_TCP,
 						 my_thread_index);
-  tcp_inc_counter (syn_sent, TCP_ERROR_EVENT_FIFO_FULL, errors);
+  tcp_inc_counter (syn_sent, TCP_ERROR_MSG_QUEUE_FULL, errors);
   vlib_buffer_free (vm, first_buffer, from_frame->n_vectors);
 
   return from_frame->n_vectors;
@@ -2750,7 +2749,12 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	  /* Reset SYN-ACK retransmit and SYN_RCV establish timers */
 	  tcp_retransmit_timer_reset (tc0);
 	  tcp_timer_reset (tc0, TCP_TIMER_ESTABLISH);
-	  stream_session_accept_notify (&tc0->connection);
+	  if (stream_session_accept_notify (&tc0->connection))
+	    {
+	      error0 = TCP_ERROR_MSG_QUEUE_FULL;
+	      tcp_connection_reset (tc0);
+	      goto drop;
+	    }
 	  error0 = TCP_ERROR_ACK_OK;
 	  break;
 	case TCP_STATE_ESTABLISHED:
@@ -2958,7 +2962,7 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 
   errors = session_manager_flush_enqueue_events (TRANSPORT_PROTO_TCP,
 						 thread_index);
-  tcp_inc_counter (rcv_process, TCP_ERROR_EVENT_FIFO_FULL, errors);
+  tcp_inc_counter (rcv_process, TCP_ERROR_MSG_QUEUE_FULL, errors);
   tcp_handle_postponed_dequeues (wrk);
   vlib_buffer_free (vm, first_buffer, from_frame->n_vectors);
 
@@ -3150,7 +3154,6 @@ tcp46_listen_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
       if (stream_session_accept (&child0->connection, lc0->c_s_index,
 				 0 /* notify */ ))
 	{
-	  clib_warning ("session accept fail");
 	  tcp_connection_cleanup (child0);
 	  error0 = TCP_ERROR_CREATE_SESSION_FAIL;
 	  goto drop;
@@ -3407,8 +3410,9 @@ tcp_input_dispatch_buffer (tcp_main_t * tm, tcp_connection_t * tc,
       vnet_buffer (b)->tcp.flags = tc->state;
 
       if (*error == TCP_ERROR_DISPATCH)
-	clib_warning ("disp error state %U flags %U", format_tcp_state,
-		      state, format_tcp_flags, (int) flags);
+	clib_warning ("tcp conn %u disp error state %U flags %U",
+		      tc->c_c_index, format_tcp_state, state,
+		      format_tcp_flags, (int) flags);
     }
 }