tcp: fix fin_wait_1 condition to send fin

Also add the closed-waiting session state wherein the session still
allows the transport to send oustanding data.

Change-Id: Ic47807379906ef2010934381ff0b9e53c7e631d8
Signed-off-by: Florin Coras <fcoras@cisco.com>

4 files changed, 20 insertions, 3 deletions

diff --git a/src/vnet/session/session.c b/src/vnet/session/session.c
index b48459d5081..069818ef1be 100644
--- a/src/vnet/session/session.c
+++ b/src/vnet/session/session.c
@@ -819,6 +819,7 @@ stream_session_delete_notify (transport_connection_t * tc)
       break;
     case SESSION_STATE_CLOSED:
     case SESSION_STATE_ACCEPTING:
+    case SESSION_STATE_CLOSED_WAITING:
       stream_session_delete (s);
       break;
     default:
@@ -1112,7 +1113,18 @@ stream_session_disconnect_transport (stream_session_t * s)
       session_free_w_fifos (s);
       return;
     }
-  s->session_state = SESSION_STATE_CLOSED;
+
+  /* If tx queue wasn't drained, change state to closed waiting for transport.
+   * This way, the transport, if it so wishes, can continue to try sending the
+   * outstanding data (in closed state it cannot). It MUST however at one
+   * point, either after sending everything or after a timeout, call delete
+   * notify. This will finally lead to the complete cleanup of the session.
+   */
+  if (svm_fifo_max_dequeue (s->server_tx_fifo))
+    s->session_state = SESSION_STATE_CLOSED_WAITING;
+  else
+    s->session_state = SESSION_STATE_CLOSED;
+
   tp_vfts[session_get_transport_proto (s)].close (s->connection_index,
 						  s->thread_index);
 }
diff --git a/src/vnet/session/stream_session.h b/src/vnet/session/stream_session.h
index c335c5b833e..79a6839ad21 100644
--- a/src/vnet/session/stream_session.h
+++ b/src/vnet/session/stream_session.h
@@ -33,6 +33,7 @@ typedef enum
   SESSION_STATE_OPENED,
   SESSION_STATE_TRANSPORT_CLOSING,
   SESSION_STATE_CLOSING,
+  SESSION_STATE_CLOSED_WAITING,
   SESSION_STATE_CLOSED,
   SESSION_STATE_N_STATES,
 } stream_session_state_t;
diff --git a/src/vnet/tcp/tcp.c b/src/vnet/tcp/tcp.c
index beadf9e1561..cbf570f2dc7 100644
--- a/src/vnet/tcp/tcp.c
+++ b/src/vnet/tcp/tcp.c
@@ -1271,6 +1271,9 @@ tcp_timer_waitclose_handler (u32 conn_index)
   else if (tc->state == TCP_STATE_FIN_WAIT_1)
     {
       tcp_connection_timers_reset (tc);
+      /* If FIN pending send it before closing */
+      if (tc->flags & TCP_CONN_FINPNDG)
+	tcp_send_fin (tc);
       tc->state = TCP_STATE_CLOSED;
       /* Wait for session layer to clean up tx events */
       tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c
index dc8fffe2291..32de369b4cc 100644
--- a/src/vnet/tcp/tcp_input.c
+++ b/src/vnet/tcp/tcp_input.c
@@ -2645,7 +2645,7 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 {
   u32 thread_index = vm->thread_index, errors = 0, *first_buffer;
   tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index);
-  u32 n_left_from, *from;
+  u32 n_left_from, *from, max_dequeue;
 
   from = first_buffer = vlib_frame_vector_args (from_frame);
   n_left_from = from_frame->n_vectors;
@@ -2776,7 +2776,8 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	  if (tc0->flags & TCP_CONN_FINPNDG)
 	    {
 	      /* TX fifo finally drained */
-	      if (!session_tx_fifo_max_dequeue (&tc0->connection))
+	      max_dequeue = session_tx_fifo_max_dequeue (&tc0->connection);
+	      if (max_dequeue <= tc0->burst_acked)
 		tcp_send_fin (tc0);
 	    }
 	  /* If FIN is ACKed */