tcp: postpone rst handling

Type: refactor

Signed-off-by: Florin Coras <fcoras@cisco.com>
Change-Id: Id09fa54cc5657b3b5616ea6a6180c1344b0141d4

3 files changed, 133 insertions, 71 deletions

diff --git a/src/vnet/tcp/tcp.c b/src/vnet/tcp/tcp.c
index 0c2523c5bff..c3fa6d49da5 100644
--- a/src/vnet/tcp/tcp.c
+++ b/src/vnet/tcp/tcp.c
@@ -335,56 +335,6 @@ tcp_connection_free (tcp_connection_t * tc)
   pool_put (tm->connections[tc->c_thread_index], tc);
 }
 
-/** Notify session that connection has been reset.
- *
- * Switch state to closed and wait for session to call cleanup.
- */
-void
-tcp_connection_reset (tcp_connection_t * tc)
-{
-  TCP_EVT (TCP_EVT_RST_RCVD, tc);
-  switch (tc->state)
-    {
-    case TCP_STATE_SYN_RCVD:
-      /* Cleanup everything. App wasn't notified yet */
-      session_transport_delete_notify (&tc->connection);
-      tcp_connection_cleanup (tc);
-      break;
-    case TCP_STATE_SYN_SENT:
-      session_stream_connect_notify (&tc->connection, 1 /* fail */ );
-      tcp_connection_cleanup (tc);
-      break;
-    case TCP_STATE_ESTABLISHED:
-      tcp_connection_timers_reset (tc);
-      /* Set the cleanup timer, in case the session layer/app don't
-       * cleanly close the connection */
-      tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
-      session_transport_reset_notify (&tc->connection);
-      tcp_cong_recovery_off (tc);
-      tcp_connection_set_state (tc, TCP_STATE_CLOSED);
-      session_transport_closed_notify (&tc->connection);
-      break;
-    case TCP_STATE_CLOSE_WAIT:
-    case TCP_STATE_FIN_WAIT_1:
-    case TCP_STATE_FIN_WAIT_2:
-    case TCP_STATE_CLOSING:
-    case TCP_STATE_LAST_ACK:
-      tcp_connection_timers_reset (tc);
-      tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
-      tcp_cong_recovery_off (tc);
-      /* Make sure we mark the session as closed. In some states we may
-       * be still trying to send data */
-      tcp_connection_set_state (tc, TCP_STATE_CLOSED);
-      session_transport_closed_notify (&tc->connection);
-      break;
-    case TCP_STATE_CLOSED:
-    case TCP_STATE_TIME_WAIT:
-      break;
-    default:
-      TCP_DBG ("reset state: %u", tc->state);
-    }
-}
-
 /**
  * Begin connection closing procedure.
  *
diff --git a/src/vnet/tcp/tcp.h b/src/vnet/tcp/tcp.h
index 73d59706da5..36e98a98dd6 100644
--- a/src/vnet/tcp/tcp.h
+++ b/src/vnet/tcp/tcp.h
@@ -436,6 +436,8 @@ typedef struct _tcp_connection
   u16 mss;		/**< Our max seg size that includes options */
   u32 timestamp_delta;	/**< Offset for timestamp */
   u32 ipv6_flow_label;	/**< flow label for ipv6 header */
+
+#define rst_state snd_wl1
 } tcp_connection_t;
 
 /* *INDENT-OFF* */
@@ -503,11 +505,6 @@ typedef struct _tcp_lookup_dispatch
 typedef struct tcp_worker_ctx_
 {
   CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
-  /** worker time */
-  u32 time_now;
-
-  /** worker timer wheel */
-  tw_timer_wheel_16t_2w_512sl_t timer_wheel;
 
   /** tx buffer free list */
   u32 *tx_buffers;
@@ -521,14 +518,22 @@ typedef struct tcp_worker_ctx_
   /** vector of pending disconnect notifications */
   u32 *pending_disconnects;
 
+  u32 *pending_resets;
+
   /** convenience pointer to this thread's vlib main */
   vlib_main_t *vm;
 
+  /** worker time */
+  u32 time_now;
+
     CLIB_CACHE_LINE_ALIGN_MARK (cacheline1);
 
   /** cached 'on the wire' options for bursts */
   u8 cached_opts[40];
 
+  /** worker timer wheel */
+  tw_timer_wheel_16t_2w_512sl_t timer_wheel;
+
 } tcp_worker_ctx_t;
 
 typedef struct tcp_iss_seed_
@@ -753,7 +758,6 @@ tcp_connection_t *tcp_connection_alloc (u8 thread_index);
 tcp_connection_t *tcp_connection_alloc_w_base (u8 thread_index,
 					       tcp_connection_t * base);
 void tcp_connection_free (tcp_connection_t * tc);
-void tcp_connection_reset (tcp_connection_t * tc);
 int tcp_configure_v4_source_address_range (vlib_main_t * vm,
 					   ip4_address_t * start,
 					   ip4_address_t * end, u32 table_id);
diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c
index f981351c4b9..164a1b3431f 100755
--- a/src/vnet/tcp/tcp_input.c
+++ b/src/vnet/tcp/tcp_input.c
@@ -266,6 +266,100 @@ tcp_update_timestamp (tcp_connection_t * tc, u32 seq, u32 seq_end)
     }
 }
 
+static void
+tcp_handle_rst (tcp_connection_t * tc)
+{
+  switch (tc->rst_state)
+    {
+    case TCP_STATE_SYN_RCVD:
+      /* Cleanup everything. App wasn't notified yet */
+      session_transport_delete_notify (&tc->connection);
+      tcp_connection_cleanup (tc);
+      break;
+    case TCP_STATE_SYN_SENT:
+      session_stream_connect_notify (&tc->connection, 1 /* fail */ );
+      tcp_connection_cleanup (tc);
+      break;
+    case TCP_STATE_ESTABLISHED:
+      session_transport_reset_notify (&tc->connection);
+      session_transport_closed_notify (&tc->connection);
+      break;
+    case TCP_STATE_CLOSE_WAIT:
+    case TCP_STATE_FIN_WAIT_1:
+    case TCP_STATE_FIN_WAIT_2:
+    case TCP_STATE_CLOSING:
+    case TCP_STATE_LAST_ACK:
+      session_transport_closed_notify (&tc->connection);
+      break;
+    case TCP_STATE_CLOSED:
+    case TCP_STATE_TIME_WAIT:
+      break;
+    default:
+      TCP_DBG ("reset state: %u", tc->state);
+    }
+}
+
+static void
+tcp_program_reset_ntf (tcp_worker_ctx_t * wrk, tcp_connection_t * tc)
+{
+  if (!tcp_disconnect_pending (tc))
+    {
+      tc->rst_state = tc->state;
+      vec_add1 (wrk->pending_resets, tc->c_c_index);
+      tcp_disconnect_pending_on (tc);
+    }
+}
+
+/**
+ * Handle reset packet
+ *
+ * Programs disconnect/reset notification that should be sent
+ * later by calling @ref tcp_handle_disconnects
+ */
+static void
+tcp_rcv_rst (tcp_worker_ctx_t * wrk, tcp_connection_t * tc)
+{
+  TCP_EVT (TCP_EVT_RST_RCVD, tc);
+  switch (tc->state)
+    {
+    case TCP_STATE_SYN_RCVD:
+      tcp_program_reset_ntf (wrk, tc);
+      tcp_connection_set_state (tc, TCP_STATE_CLOSED);
+      break;
+    case TCP_STATE_SYN_SENT:
+      tcp_program_reset_ntf (wrk, tc);
+      tcp_connection_set_state (tc, TCP_STATE_CLOSED);
+      break;
+    case TCP_STATE_ESTABLISHED:
+      tcp_connection_timers_reset (tc);
+      /* Set the cleanup timer, in case the session layer/app don't
+       * cleanly close the connection */
+      tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
+      tcp_cong_recovery_off (tc);
+      tcp_program_reset_ntf (wrk, tc);
+      tcp_connection_set_state (tc, TCP_STATE_CLOSED);
+      break;
+    case TCP_STATE_CLOSE_WAIT:
+    case TCP_STATE_FIN_WAIT_1:
+    case TCP_STATE_FIN_WAIT_2:
+    case TCP_STATE_CLOSING:
+    case TCP_STATE_LAST_ACK:
+      tcp_connection_timers_reset (tc);
+      tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
+      tcp_cong_recovery_off (tc);
+      tcp_program_reset_ntf (wrk, tc);
+      /* Make sure we mark the session as closed. In some states we may
+       * be still trying to send data */
+      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);
+    }
+}
+
 /**
  * Validate incoming segment as per RFC793 p. 69 and RFC1323 p. 19
  *
@@ -392,7 +486,7 @@ tcp_segment_validate (tcp_worker_ctx_t * wrk, tcp_connection_t * tc0,
   /* 2nd: check the RST bit */
   if (PREDICT_FALSE (tcp_rst (th0)))
     {
-      tcp_connection_reset (tc0);
+      tcp_rcv_rst (wrk, tc0);
       *error0 = TCP_ERROR_RST_RCVD;
       goto error;
     }
@@ -1676,22 +1770,35 @@ tcp_program_disconnect (tcp_worker_ctx_t * wrk, tcp_connection_t * tc)
 static void
 tcp_handle_disconnects (tcp_worker_ctx_t * wrk)
 {
-  u32 thread_index, *pending_disconnects;
+  u32 thread_index, *pending_disconnects, *pending_resets;
   tcp_connection_t *tc;
   int i;
 
-  if (!vec_len (wrk->pending_disconnects))
-    return;
+  if (vec_len (wrk->pending_disconnects))
+    {
+      thread_index = wrk->vm->thread_index;
+      pending_disconnects = wrk->pending_disconnects;
+      for (i = 0; i < vec_len (pending_disconnects); i++)
+	{
+	  tc = tcp_connection_get (pending_disconnects[i], thread_index);
+	  tcp_disconnect_pending_off (tc);
+	  session_transport_closing_notify (&tc->connection);
+	}
+      _vec_len (wrk->pending_disconnects) = 0;
+    }
 
-  thread_index = wrk->vm->thread_index;
-  pending_disconnects = wrk->pending_disconnects;
-  for (i = 0; i < vec_len (pending_disconnects); i++)
+  if (vec_len (wrk->pending_resets))
     {
-      tc = tcp_connection_get (pending_disconnects[i], thread_index);
-      tcp_disconnect_pending_off (tc);
-      session_transport_closing_notify (&tc->connection);
+      thread_index = wrk->vm->thread_index;
+      pending_resets = wrk->pending_resets;
+      for (i = 0; i < vec_len (pending_resets); i++)
+	{
+	  tc = tcp_connection_get (pending_resets[i], thread_index);
+	  tcp_disconnect_pending_off (tc);
+	  tcp_handle_rst (tc);
+	}
+      _vec_len (wrk->pending_resets) = 0;
     }
-  _vec_len (wrk->pending_disconnects) = 0;
 }
 
 static void
@@ -2556,7 +2663,7 @@ tcp46_syn_sent_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	  /* If ACK is acceptable, signal client that peer is not
 	   * willing to accept connection and drop connection*/
 	  if (tcp_ack (tcp0))
-	    tcp_connection_reset (tc0);
+	    tcp_rcv_rst (wrk, tc0);
 	  error0 = TCP_ERROR_RST_RCVD;
 	  goto drop;
 	}
@@ -2701,6 +2808,7 @@ tcp46_syn_sent_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 					      my_thread_index);
   tcp_inc_counter (syn_sent, TCP_ERROR_MSG_QUEUE_FULL, errors);
   vlib_buffer_free (vm, first_buffer, from_frame->n_vectors);
+  tcp_handle_disconnects (wrk);
 
   return from_frame->n_vectors;
 }
@@ -2837,7 +2945,7 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	  /* Make sure the segment is exactly right */
 	  if (tc0->rcv_nxt != vnet_buffer (b0)->tcp.seq_number || is_fin0)
 	    {
-	      tcp_connection_reset (tc0);
+	      tcp_rcv_rst (wrk, tc0);
 	      error0 = TCP_ERROR_SEGMENT_INVALID;
 	      goto drop;
 	    }
@@ -2850,7 +2958,7 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	   */
 	  if (tcp_rcv_ack_no_cc (tc0, b0, &error0))
 	    {
-	      tcp_connection_reset (tc0);
+	      tcp_rcv_rst (wrk, tc0);
 	      goto drop;
 	    }
 
@@ -2877,7 +2985,7 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	  if (session_stream_accept_notify (&tc0->connection))
 	    {
 	      error0 = TCP_ERROR_MSG_QUEUE_FULL;
-	      tcp_connection_reset (tc0);
+	      tcp_rcv_rst (wrk, tc0);
 	      goto drop;
 	    }
 	  error0 = TCP_ERROR_ACK_OK;