tls: fix multi threaded medium scale test (VPP-1457)

- ensure session enqueue epoch does not wrap between two enqueues
- use 3 states for echo clients app, to distinguish between starting and
closing phases
- force tcp fin retransmit if out of buffers while sending a fin

Change-Id: I6f2cab46affd1148aba2a33fb6d58bcc54f32805
Signed-off-by: Florin Coras <fcoras@cisco.com>

src/vnet/session-apps/echo_client.c

@@ -208,7 +208,7 @@ echo_client_node_fn (vlib_main_t * vm, vlib_node_runtime_t * node,
   connections_this_batch =
     ecm->connections_this_batch_by_thread[my_thread_index];
 
-  if ((ecm->run_test == 0) ||
+  if ((ecm->run_test != ECHO_CLIENTS_RUNNING) ||
       ((vec_len (connection_indices) == 0)
        && vec_len (connections_this_batch) == 0))
     return 0;
@@ -352,6 +352,16 @@ echo_clients_init (vlib_main_t * vm)
   return 0;
 }
 
+static void
+echo_clients_session_disconnect (stream_session_t * s)
+{
+  echo_client_main_t *ecm = &echo_client_main;
+  vnet_disconnect_args_t _a, *a = &_a;
+  a->handle = session_handle (s);
+  a->app_index = ecm->app_index;
+  vnet_disconnect_session (a);
+}
+
 static int
 echo_clients_session_connected_callback (u32 app_index, u32 api_context,
 					 stream_session_t * s, u8 is_fail)
@@ -361,6 +371,9 @@ echo_clients_session_connected_callback (u32 app_index, u32 api_context,
   u32 session_index;
   u8 thread_index;
 
+  if (PREDICT_FALSE (ecm->run_test != ECHO_CLIENTS_STARTING))
+    return -1;
+
   if (is_fail)
     {
       clib_warning ("connection %d failed!", api_context);
@@ -407,7 +420,7 @@ echo_clients_session_connected_callback (u32 app_index, u32 api_context,
   __sync_fetch_and_add (&ecm->ready_connections, 1);
   if (ecm->ready_connections == ecm->expected_connections)
     {
-      ecm->run_test = 1;
+      ecm->run_test = ECHO_CLIENTS_RUNNING;
       /* Signal the CLI process that the action is starting... */
       signal_evt_to_cli (1);
     }
@@ -447,6 +460,12 @@ echo_clients_rx_callback (stream_session_t * s)
   echo_client_main_t *ecm = &echo_client_main;
   eclient_session_t *sp;
 
+  if (PREDICT_FALSE (ecm->run_test != ECHO_CLIENTS_RUNNING))
+    {
+      echo_clients_session_disconnect (s);
+      return -1;
+    }
+
   sp = pool_elt_at_index (ecm->sessions,
 			  s->server_rx_fifo->client_session_index);
   receive_data_chunk (ecm, sp);
@@ -624,6 +643,7 @@ echo_clients_command_fn (vlib_main_t * vm,
   ecm->vlib_main = vm;
   ecm->tls_engine = TLS_ENGINE_OPENSSL;
   ecm->no_copy = 0;
+  ecm->run_test = ECHO_CLIENTS_STARTING;
 
   if (thread_main->n_vlib_mains > 1)
     clib_spinlock_init (&ecm->sessions_lock);
@@ -825,7 +845,7 @@ echo_clients_command_fn (vlib_main_t * vm,
     error = clib_error_return (0, "failed: test bytes");
 
 cleanup:
-  ecm->run_test = 0;
+  ecm->run_test = ECHO_CLIENTS_EXITING;
   vlib_process_wait_for_event_or_clock (vm, 10e-3);
   for (i = 0; i < vec_len (ecm->connection_index_by_thread); i++)
     {

src/vnet/session-apps/echo_client.h

@@ -105,6 +105,12 @@ typedef struct
   vlib_main_t *vlib_main;
 } echo_client_main_t;
 
+enum
+{
+  ECHO_CLIENTS_STARTING,
+  ECHO_CLIENTS_RUNNING,
+  ECHO_CLIENTS_EXITING
+} echo_clients_test_state_e;
 extern echo_client_main_t echo_client_main;
 
 vlib_node_registration_t echo_clients_node;

src/vnet/session/session.c

@@ -153,7 +153,7 @@ session_free (stream_session_t * s)
     memset (s, 0xFA, sizeof (*s));
 }
 
-static void
+void
 session_free_w_fifos (stream_session_t * s)
 {
   segment_manager_dealloc_fifos (s->svm_segment_index, s->server_rx_fifo,
@@ -197,7 +197,7 @@ 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->session_state = SESSION_STATE_CONNECTING;
-  s->enqueue_epoch = ~0;
+  s->enqueue_epoch = (u64) ~ 0;
 
   /* Attach transport to session and vice versa */
   s->connection_index = tc->c_index;
@@ -393,7 +393,7 @@ session_enqueue_stream_connection (transport_connection_t * tc,
        * by calling stream_server_flush_enqueue_events () */
       session_manager_main_t *smm = vnet_get_session_manager_main ();
       u32 thread_index = s->thread_index;
-      u32 enqueue_epoch = smm->current_enqueue_epoch[tc->proto][thread_index];
+      u64 enqueue_epoch = smm->current_enqueue_epoch[tc->proto][thread_index];
 
       if (s->enqueue_epoch != enqueue_epoch)
 	{
@@ -434,7 +434,7 @@ session_enqueue_dgram_connection (stream_session_t * s,
        * by calling stream_server_flush_enqueue_events () */
       session_manager_main_t *smm = vnet_get_session_manager_main ();
       u32 thread_index = s->thread_index;
-      u32 enqueue_epoch = smm->current_enqueue_epoch[proto][thread_index];
+      u64 enqueue_epoch = smm->current_enqueue_epoch[proto][thread_index];
 
       if (s->enqueue_epoch != enqueue_epoch)
 	{

src/vnet/session/session.h

@@ -195,7 +195,7 @@ struct _session_manager_main
   clib_rwlock_t *peekers_rw_locks;
 
   /** Per-proto, per-worker enqueue epoch counters */
-  u32 *current_enqueue_epoch[TRANSPORT_N_PROTO];
+  u64 *current_enqueue_epoch[TRANSPORT_N_PROTO];
 
   /** Per-proto, per-worker thread vector of sessions to enqueue */
   u32 **session_to_enqueue[TRANSPORT_N_PROTO];
@@ -308,6 +308,7 @@ stream_session_is_valid (u32 si, u8 thread_index)
 stream_session_t *session_alloc (u32 thread_index);
 int session_alloc_fifos (segment_manager_t * sm, stream_session_t * s);
 void session_free (stream_session_t * s);
+void session_free_w_fifos (stream_session_t * s);
 
 always_inline stream_session_t *
 session_get (u32 si, u32 thread_index)

src/vnet/session/stream_session.h

@@ -67,7 +67,7 @@ typedef struct _stream_session_t
   u8 thread_index;
 
   /** To avoid n**2 "one event per frame" check */
-  u8 enqueue_epoch;
+  u64 enqueue_epoch;
 
   /** svm segment index where fifos were allocated */
   u32 svm_segment_index;
@@ -120,6 +120,9 @@ typedef struct local_session_
   /** Port for connection. Overlaps thread_index/enqueue_epoch */
   u16 port;
 
+  /** Partly overlaps enqueue_epoch */
+  u8 pad_epoch[7];
+
   /** Segment index where fifos were allocated */
   u32 svm_segment_index;
 

src/vnet/tcp/tcp_output.c

@@ -1078,7 +1078,15 @@ tcp_send_fin (tcp_connection_t * tc)
 
   tcp_retransmit_timer_force_update (tc);
   if (PREDICT_FALSE (tcp_get_free_buffer_index (tm, &bi)))
-    return;
+    {
+      /* Out of buffers so program fin retransmit ASAP */
+      tcp_timer_update (tc, TCP_TIMER_RETRANSMIT, 1);
+      tc->flags |= TCP_CONN_FINSNT;
+      tc->snd_una_max += 1;
+      tc->snd_nxt = tc->snd_una_max;
+      return;
+    }
+
   b = vlib_get_buffer (vm, bi);
   tcp_init_buffer (vm, b);
   fin_snt = tc->flags & TCP_CONN_FINSNT;

src/vnet/tls/tls.c

@@ -119,6 +119,7 @@ tls_ctx_half_open_alloc (void)
     {
       clib_rwlock_writer_lock (&tm->half_open_rwlock);
       pool_get (tm->half_open_ctx_pool, ctx);
+      ctx_index = ctx - tm->half_open_ctx_pool;
       clib_rwlock_writer_unlock (&tm->half_open_rwlock);
     }
   else
@@ -126,10 +127,10 @@ tls_ctx_half_open_alloc (void)
       /* reader lock assumption: only main thread will call pool_get */
       clib_rwlock_reader_lock (&tm->half_open_rwlock);
       pool_get (tm->half_open_ctx_pool, ctx);
+      ctx_index = ctx - tm->half_open_ctx_pool;
       clib_rwlock_reader_unlock (&tm->half_open_rwlock);
     }
   memset (ctx, 0, sizeof (*ctx));
-  ctx_index = ctx - tm->half_open_ctx_pool;
   return ctx_index;
 }
 
@@ -254,6 +255,8 @@ tls_notify_app_connected (tls_ctx_t * ctx, u8 is_failed)
     {
       TLS_DBG (1, "failed to notify app");
       tls_disconnect (ctx->tls_ctx_handle, vlib_get_thread_index ());
+      session_free_w_fifos (app_session);
+      return -1;
     }
 
   session_lookup_add_connection (&ctx->connection,