session/tcp: support tx flush mark

For tcp this means that the last enqueued data goes out with a psh bit
set.

Change-Id: I29d357ecae6f02e748b59a7b799150ec73d14ba2
Signed-off-by: Florin Coras <fcoras@cisco.com>

4 files changed, 26 insertions, 1 deletions

diff --git a/src/vnet/tcp/tcp.c b/src/vnet/tcp/tcp.c
index f703d634b54..6d6a880eda8 100644
--- a/src/vnet/tcp/tcp.c
+++ b/src/vnet/tcp/tcp.c
@@ -1123,6 +1123,16 @@ tcp_session_push_header (transport_connection_t * tconn, vlib_buffer_t * b)
   return tcp_push_header (tc, b);
 }
 
+static void
+tcp_session_flush_data (transport_connection_t * tconn)
+{
+  tcp_connection_t *tc = (tcp_connection_t *) tconn;
+  if (tc->flags & TCP_CONN_PSH_PENDING)
+    return;
+  tc->flags |= TCP_CONN_PSH_PENDING;
+  tc->psh_seq = tc->snd_una_max + transport_max_tx_dequeue (tconn) - 1;
+}
+
 /* *INDENT-OFF* */
 const static transport_proto_vft_t tcp_proto = {
   .enable = vnet_tcp_enable_disable,
@@ -1139,6 +1149,7 @@ const static transport_proto_vft_t tcp_proto = {
   .send_space = tcp_session_send_space,
   .update_time = tcp_update_time,
   .tx_fifo_offset = tcp_session_tx_fifo_offset,
+  .flush_data = tcp_session_flush_data,
   .format_connection = format_tcp_session,
   .format_listener = format_tcp_listener_session,
   .format_half_open = format_tcp_half_open_session,
diff --git a/src/vnet/tcp/tcp.h b/src/vnet/tcp/tcp.h
index 5b235b65d74..46b03ac14b4 100644
--- a/src/vnet/tcp/tcp.h
+++ b/src/vnet/tcp/tcp.h
@@ -124,6 +124,7 @@ extern timer_expiration_handler tcp_timer_retransmit_syn_handler;
   _(FRXT_PENDING, "Fast-retransmit pending")	\
   _(FRXT_FIRST, "Fast-retransmit first again")	\
   _(DEQ_PENDING, "Pending dequeue acked")	\
+  _(PSH_PENDING, "Pending psh packet")		\
 
 typedef enum _tcp_connection_flag_bits
 {
@@ -334,6 +335,8 @@ typedef struct _tcp_connection
   u32 last_fib_check;	/**< Last time we checked fib route for peer */
   u32 sw_if_index;	/**< Interface for the connection */
   u32 tx_fifo_size;	/**< Tx fifo size. Used to constrain cwnd */
+
+  u32 psh_seq;		/**< Add psh header for seg that includes this */
 } tcp_connection_t;
 
 /* *INDENT-OFF* */
diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c
index dff18029155..f04fa5d8901 100644
--- a/src/vnet/tcp/tcp_input.c
+++ b/src/vnet/tcp/tcp_input.c
@@ -543,6 +543,12 @@ tcp_handle_postponed_dequeues (tcp_worker_ctx_t * wrk)
       tc->burst_acked = 0;
       tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una);
 
+      if (PREDICT_FALSE (tc->flags & TCP_CONN_PSH_PENDING))
+	{
+	  if (seq_leq (tc->psh_seq, tc->snd_una))
+	    tc->flags &= ~TCP_CONN_PSH_PENDING;
+	}
+
       /* If everything has been acked, stop retransmit timer
        * otherwise update. */
       tcp_retransmit_timer_update (tc);
diff --git a/src/vnet/tcp/tcp_output.c b/src/vnet/tcp/tcp_output.c
index 74fc15fe6cc..7cee34995e3 100644
--- a/src/vnet/tcp/tcp_output.c
+++ b/src/vnet/tcp/tcp_output.c
@@ -1169,7 +1169,12 @@ tcp_push_hdr_i (tcp_connection_t * tc, vlib_buffer_t * b,
     advertise_wnd = tcp_window_to_advertise (tc, next_state);
 
   flags = tcp_make_state_flags (tc, next_state);
-
+  if (PREDICT_FALSE (tc->flags & TCP_CONN_PSH_PENDING))
+    {
+      if (seq_geq (tc->psh_seq, tc->snd_nxt)
+	  && seq_lt (tc->psh_seq, tc->snd_nxt + data_len))
+	flags |= TCP_FLAG_PSH;
+    }
   th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt,
 			     tc->rcv_nxt, tcp_hdr_opts_len, flags,
 			     advertise_wnd);