tcp: avoid sending acks when data available

Type: fix

Change-Id: I02e2cb2d349f9ddb2bfed040a08ca448f379d565
Signed-off-by: Florin Coras <fcoras@cisco.com>
(cherry picked from commit cb711a4ec908b89edc9da29d3c2e093dab85fbbf)

3 files changed, 24 insertions, 5 deletions

diff --git a/src/vnet/tcp/tcp.h b/src/vnet/tcp/tcp.h
index 8636a7c85db..657f1bea449 100644
--- a/src/vnet/tcp/tcp.h
+++ b/src/vnet/tcp/tcp.h
@@ -786,6 +786,7 @@ void tcp_send_reset (tcp_connection_t * tc);
 void tcp_send_syn (tcp_connection_t * tc);
 void tcp_send_synack (tcp_connection_t * tc);
 void tcp_send_fin (tcp_connection_t * tc);
+void tcp_send_ack (tcp_connection_t * tc);
 void tcp_update_burst_snd_vars (tcp_connection_t * tc);
 void tcp_update_rto (tcp_connection_t * tc);
 void tcp_flush_frame_to_output (tcp_worker_ctx_t * wrk, u8 is_ip4);
diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c
index de59a4da0af..83769465525 100755
--- a/src/vnet/tcp/tcp_input.c
+++ b/src/vnet/tcp/tcp_input.c
@@ -1687,6 +1687,7 @@ tcp_rcv_fin (tcp_worker_ctx_t * wrk, tcp_connection_t * tc, vlib_buffer_t * b,
 
   /* Account for the FIN and send ack */
   tc->rcv_nxt += 1;
+  tc->flags |= TCP_CONN_FINRCVD;
   tcp_program_ack (tc);
   /* Enter CLOSE-WAIT and notify session. To avoid lingering
    * in CLOSE-WAIT, set timer (reuse WAITCLOSE). */
@@ -2599,7 +2600,9 @@ tcp46_syn_sent_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	}
       else
 	{
-	  tcp_program_ack (new_tc0);
+	  /* Send ack now instead of programming it because connection was
+	   * just established and it's not optional. */
+	  tcp_send_ack (new_tc0);
 	}
 
     drop:
diff --git a/src/vnet/tcp/tcp_output.c b/src/vnet/tcp/tcp_output.c
index 33bfe100a10..2437b7bf28f 100644
--- a/src/vnet/tcp/tcp_output.c
+++ b/src/vnet/tcp/tcp_output.c
@@ -518,7 +518,6 @@ tcp_compute_checksum (tcp_connection_t * tc, vlib_buffer_t * b)
   return checksum;
 }
 
-
 /**
  * Prepare ACK
  */
@@ -1086,6 +1085,10 @@ tcp_send_fin (tcp_connection_t * tc)
       return;
     }
 
+  /* If we have non-dupacks programmed, no need to send them */
+  if ((tc->flags & TCP_CONN_SNDACK) && !tc->pending_dupacks)
+    tc->flags &= ~TCP_CONN_SNDACK;
+
   tcp_retransmit_timer_force_update (tc);
   b = vlib_get_buffer (vm, bi);
   tcp_init_buffer (vm, b);
@@ -1165,7 +1168,6 @@ tcp_push_hdr_i (tcp_connection_t * tc, vlib_buffer_t * b, u32 snd_nxt,
   tc->bytes_out += data_len;
   tc->data_segs_out += 1;
 
-
   th->checksum = tcp_compute_checksum (tc, b);
 
   TCP_EVT (TCP_EVT_PKTIZE, tc);
@@ -1897,6 +1899,13 @@ tcp_retransmit_should_retry_head (tcp_connection_t * tc,
   return (tx_adv_sack > (tc->snd_una - tc->prr_start) * rr);
 }
 
+static inline u8
+tcp_max_tx_deq (tcp_connection_t * tc)
+{
+  return (transport_max_tx_dequeue (&tc->connection)
+	  - (tc->snd_nxt - tc->snd_una));
+}
+
 #define scoreboard_rescue_rxt_valid(_sb, _tc)			\
     (seq_geq (_sb->rescue_rxt, _tc->snd_una) 			\
 	&& seq_leq (_sb->rescue_rxt, _tc->snd_congestion))
@@ -2161,8 +2170,13 @@ tcp_send_acks (tcp_connection_t * tc, u32 max_burst_size)
 
   if (!tc->pending_dupacks)
     {
-      tcp_send_ack (tc);
-      return 1;
+      if (tcp_in_cong_recovery (tc) || !tcp_max_tx_deq (tc)
+	  || tc->state != TCP_STATE_ESTABLISHED)
+	{
+	  tcp_send_ack (tc);
+	  return 1;
+	}
+      return 0;
     }
 
   /* If we're supposed to send dupacks but have no ooo data
@@ -2170,6 +2184,7 @@ tcp_send_acks (tcp_connection_t * tc, u32 max_burst_size)
   if (!vec_len (tc->snd_sacks))
     {
       tcp_send_ack (tc);
+      tc->pending_dupacks = 0;
       return 1;
     }