vcl/ldp: add write msg function and fine tuning

Allows app to push data. Additionally, ensure reset/close replies are
not sent unless vcl closes the session.

Change-Id: Icbbf933cf57b55cfbcc7b802af0f83919a066f65
Signed-off-by: Florin Coras <fcoras@cisco.com>

5 files changed, 61 insertions, 26 deletions

diff --git a/src/vnet/sctp/sctp.c b/src/vnet/sctp/sctp.c
index 482f81a1e2a..10ec7700a2c 100644
--- a/src/vnet/sctp/sctp.c
+++ b/src/vnet/sctp/sctp.c
@@ -23,6 +23,7 @@ sctp_connection_bind (u32 session_index, transport_endpoint_t * tep)
   sctp_main_t *tm = &sctp_main;
   sctp_connection_t *listener;
   void *iface_ip;
+  u32 mtu = 1460;
 
   pool_get (tm->listener_pool, listener);
   clib_memset (listener, 0, sizeof (*listener));
@@ -43,11 +44,13 @@ sctp_connection_bind (u32 session_index, transport_endpoint_t * tep)
   ip_copy (&listener->sub_conn[SCTP_PRIMARY_PATH_IDX].connection.lcl_ip,
 	   &tep->ip, tep->is_ip4);
 
-  u32 mtu = tep->is_ip4 ? vnet_sw_interface_get_mtu (vnet_get_main (),
-						     tep->sw_if_index,
-						     VNET_MTU_IP4) :
-    vnet_sw_interface_get_mtu (vnet_get_main (), tep->sw_if_index,
-			       VNET_MTU_IP6);
+  if (tep->sw_if_index != ENDPOINT_INVALID_INDEX)
+    mtu = tep->is_ip4 ? vnet_sw_interface_get_mtu (vnet_get_main (),
+						   tep->sw_if_index,
+						   VNET_MTU_IP4) :
+      vnet_sw_interface_get_mtu (vnet_get_main (), tep->sw_if_index,
+				 VNET_MTU_IP6);
+
   listener->sub_conn[SCTP_PRIMARY_PATH_IDX].PMTU = mtu;
   listener->sub_conn[SCTP_PRIMARY_PATH_IDX].connection.is_ip4 = tep->is_ip4;
   listener->sub_conn[SCTP_PRIMARY_PATH_IDX].connection.proto =
@@ -192,12 +195,13 @@ format_sctp_connection_id (u8 * s, va_list * args)
   u8 i;
   for (i = 0; i < MAX_SCTP_CONNECTIONS; i++)
     {
+      if (i > 0 && sctp_conn->sub_conn[i].state == SCTP_SUBCONN_STATE_DOWN)
+	continue;
       if (sctp_conn->sub_conn[i].connection.is_ip4)
 	{
-	  s = format (s, "%U[#%d][%s] %U:%d->%U:%d",
-		      s,
+	  s = format (s, "[#%d][%s] %U:%d->%U:%d",
 		      sctp_conn->sub_conn[i].connection.thread_index,
-		      "T",
+		      "S",
 		      format_ip4_address,
 		      &sctp_conn->sub_conn[i].connection.lcl_ip.ip4,
 		      clib_net_to_host_u16 (sctp_conn->sub_conn[i].
@@ -209,10 +213,9 @@ format_sctp_connection_id (u8 * s, va_list * args)
 	}
       else
 	{
-	  s = format (s, "%U[#%d][%s] %U:%d->%U:%d",
-		      s,
+	  s = format (s, "[#%d][%s] %U:%d->%U:%d",
 		      sctp_conn->sub_conn[i].connection.thread_index,
-		      "T",
+		      "S",
 		      format_ip6_address,
 		      &sctp_conn->sub_conn[i].connection.lcl_ip.ip6,
 		      clib_net_to_host_u16 (sctp_conn->sub_conn[i].
@@ -238,6 +241,8 @@ format_sctp_connection (u8 * s, va_list * args)
   if (verbose)
     {
       s = format (s, "%-15U", format_sctp_state, sctp_conn->state);
+      if (verbose > 1)
+	s = format (s, "\n");
     }
 
   return s;
@@ -458,6 +463,7 @@ sctp_connection_open (transport_endpoint_cfg_t * rmt)
   ip46_address_t lcl_addr;
   u16 lcl_port;
   uword thread_id;
+  u32 mtu = 1460;
   int rv;
 
   u8 idx = SCTP_PRIMARY_PATH_IDX;
@@ -484,11 +490,12 @@ sctp_connection_open (transport_endpoint_cfg_t * rmt)
 
   clib_spinlock_lock_if_init (&tm->half_open_lock);
   sctp_conn = sctp_half_open_connection_new (thread_id);
-  u32 mtu = rmt->is_ip4 ? vnet_sw_interface_get_mtu (vnet_get_main (),
-						     rmt->peer.sw_if_index,
-						     VNET_MTU_IP4) :
-    vnet_sw_interface_get_mtu (vnet_get_main (), rmt->peer.sw_if_index,
-			       VNET_MTU_IP6);
+  if (rmt->peer.sw_if_index != ENDPOINT_INVALID_INDEX)
+    mtu = rmt->is_ip4 ? vnet_sw_interface_get_mtu (vnet_get_main (),
+						   rmt->peer.sw_if_index,
+						   VNET_MTU_IP4) :
+      vnet_sw_interface_get_mtu (vnet_get_main (), rmt->peer.sw_if_index,
+				 VNET_MTU_IP6);
   sctp_conn->sub_conn[idx].PMTU = mtu;
 
   transport_connection_t *trans_conn = &sctp_conn->sub_conn[idx].connection;
diff --git a/src/vnet/session/session.c b/src/vnet/session/session.c
index cbe0dd76430..d30254e5fc0 100644
--- a/src/vnet/session/session.c
+++ b/src/vnet/session/session.c
@@ -234,6 +234,7 @@ 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->enqueue_epoch = (u64) ~ 0;
+  s->session_state = SESSION_STATE_CLOSED;
 
   /* Attach transport to session and vice versa */
   s->connection_index = tc->c_index;
@@ -773,6 +774,7 @@ stream_session_accept_notify (transport_connection_t * tc)
   app_wrk = app_worker_get_if_valid (s->app_wrk_index);
   if (!app_wrk)
     return -1;
+  s->session_state = SESSION_STATE_ACCEPTING;
   app = application_get (app_wrk->app_index);
   return app->cb_fns.session_accept_callback (s);
 }
@@ -824,6 +826,7 @@ session_transport_delete_notify (transport_connection_t * tc)
 
   switch (s->session_state)
     {
+    case SESSION_STATE_ACCEPTING:
     case SESSION_STATE_TRANSPORT_CLOSING:
       /* If transport finishes or times out before we get a reply
        * from the app, mark transport as closed and wait for reply
@@ -844,11 +847,13 @@ session_transport_delete_notify (transport_connection_t * tc)
       s->session_state = SESSION_STATE_TRANSPORT_CLOSED;
       session_program_transport_close (s);
       break;
+    case SESSION_STATE_TRANSPORT_CLOSED:
+      break;
     case SESSION_STATE_CLOSED:
-    case SESSION_STATE_ACCEPTING:
       session_delete (s);
       break;
     default:
+      clib_warning ("session state %u", s->session_state);
       session_delete (s);
       break;
     }
@@ -869,7 +874,16 @@ session_transport_closed_notify (transport_connection_t * tc)
 
   if (!(s = session_get_if_valid (tc->s_index, tc->thread_index)))
     return;
-  s->session_state = SESSION_STATE_CLOSED;
+
+  /* If app close has not been received or has not yet resulted in
+   * a transport close, only mark the session transport as closed */
+  if (s->session_state <= SESSION_STATE_CLOSING)
+    {
+      session_lookup_del_session (s);
+      s->session_state = SESSION_STATE_TRANSPORT_CLOSED;
+    }
+  else
+    s->session_state = SESSION_STATE_CLOSED;
 }
 
 /**
@@ -913,7 +927,6 @@ stream_session_accept (transport_connection_t * tc, u32 listener_index,
 
   s->app_wrk_index = app_wrk->wrk_index;
   s->listener_index = listener_index;
-  s->session_state = SESSION_STATE_ACCEPTING;
 
   /* Shoulder-tap the server */
   if (notify)
@@ -1141,7 +1154,7 @@ void
 session_transport_close (stream_session_t * s)
 {
   /* If transport is already closed, just free the session */
-  if (s->session_state == SESSION_STATE_TRANSPORT_CLOSED)
+  if (s->session_state >= SESSION_STATE_TRANSPORT_CLOSED)
     {
       session_free_w_fifos (s);
       return;
diff --git a/src/vnet/session/session_node.c b/src/vnet/session/session_node.c
index 4323ed83cf8..45018daf45a 100644
--- a/src/vnet/session/session_node.c
+++ b/src/vnet/session/session_node.c
@@ -29,6 +29,7 @@ session_mq_accepted_reply_handler (void *data)
 {
   session_accepted_reply_msg_t *mp = (session_accepted_reply_msg_t *) data;
   vnet_disconnect_args_t _a = { 0 }, *a = &_a;
+  stream_session_state_t old_state;
   app_worker_t *app_wrk;
   local_session_t *ls;
   stream_session_t *s;
@@ -64,18 +65,29 @@ session_mq_accepted_reply_handler (void *data)
   else
     {
       s = session_get_from_handle_if_valid (mp->handle);
-      /* Closed while waiting for app to reply */
-      if (!s || s->session_state > SESSION_STATE_READY)
+      if (!s)
 	return;
+
       app_wrk = app_worker_get (s->app_wrk_index);
       if (app_wrk->app_index != mp->context)
 	{
 	  clib_warning ("app doesn't own session");
 	  return;
 	}
+
+      old_state = s->session_state;
       s->session_state = SESSION_STATE_READY;
       if (!svm_fifo_is_empty (s->server_rx_fifo))
 	app_worker_lock_and_send_event (app_wrk, s, FIFO_EVENT_APP_RX);
+
+      /* Closed while waiting for app to reply. Resend disconnect */
+      if (old_state >= SESSION_STATE_TRANSPORT_CLOSING)
+	{
+	  application_t *app = application_get (app_wrk->app_index);
+	  app->cb_fns.session_disconnect_callback (s);
+	  s->session_state = old_state;
+	  return;
+	}
     }
 }
 
diff --git a/src/vnet/tcp/tcp.c b/src/vnet/tcp/tcp.c
index 564f200bc00..7f6a087b2f3 100644
--- a/src/vnet/tcp/tcp.c
+++ b/src/vnet/tcp/tcp.c
@@ -284,7 +284,8 @@ tcp_connection_reset (tcp_connection_t * tc)
       break;
     case TCP_STATE_SYN_SENT:
       session_stream_connect_notify (&tc->connection, 1 /* fail */ );
-      tcp_connection_cleanup (tc);
+      tcp_connection_set_state (tc, TCP_STATE_CLOSED);
+      tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
       break;
     case TCP_STATE_ESTABLISHED:
       tcp_connection_timers_reset (tc);
@@ -307,6 +308,7 @@ tcp_connection_reset (tcp_connection_t * tc)
       tcp_connection_set_state (tc, TCP_STATE_CLOSED);
       break;
     case TCP_STATE_CLOSED:
+    case TCP_STATE_TIME_WAIT:
       break;
     default:
       TCP_DBG ("reset state: %u", tc->state);
@@ -1293,9 +1295,9 @@ tcp_timer_waitclose_handler (u32 conn_index)
 	   * is closed. We haven't sent everything but we did try. */
 	  tcp_cong_recovery_off (tc);
 	  tcp_send_fin (tc);
-	  rto = clib_max (tc->rto >> tc->rto_boff, 1);
+	  rto = clib_max ((tc->rto >> tc->rto_boff) * TCP_TO_TIMER_TICK, 1);
 	  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE,
-			 clib_min (rto * TCP_TO_TIMER_TICK, TCP_2MSL_TIME));
+			 clib_min (rto, TCP_2MSL_TIME));
 	  session_transport_closed_notify (&tc->connection);
 	}
       else
diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c
index c3ce2eb1ae1..fb80d7f5759 100644
--- a/src/vnet/tcp/tcp_input.c
+++ b/src/vnet/tcp/tcp_input.c
@@ -2671,7 +2671,8 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 				       is_ip4);
 	  if (tmp->state != tc0->state)
 	    {
-	      clib_warning ("state changed");
+	      if (tc0->state != TCP_STATE_CLOSED)
+		clib_warning ("state changed");
 	      goto drop;
 	    }
 	}