tcp: add more connection stats

Type:feature

Change-Id: If02884d0f1f26bfe31ec609ea9611cb27b699868
Signed-off-by: Florin Coras <fcoras@cisco.com>

4 files changed, 81 insertions, 6 deletions

diff --git a/src/vnet/tcp/tcp.c b/src/vnet/tcp/tcp.c
index 5a215b658a0..adff77c14ad 100644
--- a/src/vnet/tcp/tcp.c
+++ b/src/vnet/tcp/tcp.c
@@ -670,6 +670,8 @@ tcp_connection_init_vars (tcp_connection_t * tc)
 
   if (tc->flags & TCP_CONN_RATE_SAMPLE)
     tcp_bt_init (tc);
+
+  tc->start_ts = tcp_time_now_us (tc->c_thread_index);
 }
 
 static int
@@ -864,6 +866,27 @@ format_tcp_congestion (u8 * s, va_list * args)
 }
 
 static u8 *
+format_tcp_stats (u8 * s, va_list * args)
+{
+  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
+  u32 indent = format_get_indent (s);
+  s = format (s, "in segs %lu dsegs %lu bytes %lu dupacks %u\n",
+	      tc->segs_in, tc->data_segs_in, tc->bytes_in, tc->dupacks_in);
+  s = format (s, "%Uout segs %lu dsegs %lu bytes %lu dupacks %u\n",
+	      format_white_space, indent, tc->segs_out,
+	      tc->data_segs_out, tc->bytes_out, tc->dupacks_out);
+  s = format (s, "%Ufr %u tr %u rxt segs %lu bytes %lu duration %.3f\n",
+	      format_white_space, indent, tc->fr_occurences,
+	      tc->tr_occurences, tc->segs_retrans, tc->bytes_retrans,
+	      tcp_time_now_us (tc->c_thread_index) - tc->start_ts);
+  s = format (s, "%Uerr wnd data below %u above %u ack below %u above %u",
+	      format_white_space, indent, tc->errors.below_data_wnd,
+	      tc->errors.above_data_wnd, tc->errors.below_ack_wnd,
+	      tc->errors.above_ack_wnd);
+  return s;
+}
+
+static u8 *
 format_tcp_vars (u8 * s, va_list * args)
 {
   tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
@@ -896,6 +919,7 @@ format_tcp_vars (u8 * s, va_list * args)
     {
       s = format (s, " sboard: %U\n", format_tcp_scoreboard, &tc->sack_sb,
 		  tc);
+      s = format (s, " stats: %U\n", format_tcp_stats, tc);
     }
   if (vec_len (tc->snd_sacks))
     s = format (s, " sacks tx: %U\n", format_tcp_sacks, tc);
diff --git a/src/vnet/tcp/tcp.h b/src/vnet/tcp/tcp.h
index 698173ee28a..fe02640f249 100644
--- a/src/vnet/tcp/tcp.h
+++ b/src/vnet/tcp/tcp.h
@@ -300,6 +300,17 @@ typedef enum tcp_cc_event_
   TCP_CC_EVT_START_TX,
 } tcp_cc_event_t;
 
+/*
+ * As per RFC4898 tcpEStatsStackSoftErrors
+ */
+typedef struct tcp_errors_
+{
+  u32 below_data_wnd;	/**< All data in seg is below snd_una */
+  u32 above_data_wnd;	/**< Some data in segment is above snd_wnd */
+  u32 below_ack_wnd;	/**< Acks for data below snd_una */
+  u32 above_ack_wnd;	/**< Acks for data not sent */
+} tcp_errors_t;
+
 typedef struct _tcp_connection
 {
   CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
@@ -309,7 +320,10 @@ typedef struct _tcp_connection
   u16 flags;			/**< Connection flags (see tcp_conn_flags_e) */
   u32 timers[TCP_N_TIMERS];	/**< Timer handles into timer wheel */
 
-  /* TODO RFC4898 */
+  u64 segs_in;		/** RFC4022/4898 tcpHCInSegs/tcpEStatsPerfSegsIn */
+  u64 bytes_in;		/** RFC4898 tcpEStatsPerfHCDataOctetsIn */
+  u64 segs_out;		/** RFC4898 tcpEStatsPerfSegsOut */
+  u64 bytes_out;	/** RFC4898 tcpEStatsPerfHCDataOctetsOut */
 
   /** Send sequence variables RFC793 */
   u32 snd_una;		/**< oldest unacknowledged sequence number */
@@ -320,6 +334,9 @@ typedef struct _tcp_connection
   u32 snd_nxt;		/**< next seq number to be sent */
   u16 snd_mss;		/**< Effective send max seg (data) size */
 
+  u64 data_segs_in;	/** RFC4898 tcpEStatsPerfDataSegsIn */
+  u64 data_segs_out;	/** RFC4898 tcpEStatsPerfDataSegsOut */
+
   /** Receive sequence variables RFC793 */
   u32 rcv_nxt;		/**< next sequence number expected */
   u32 rcv_wnd;		/**< receive window we expect */
@@ -342,8 +359,10 @@ typedef struct _tcp_connection
   sack_block_t *snd_sacks_fl;	/**< Vector for building new list */
   sack_scoreboard_t sack_sb;	/**< SACK "scoreboard" that tracks holes */
 
-  u16 rcv_dupacks;	/**< Number of DUPACKs received */
+  u16 rcv_dupacks;	/**< Number of recent DUPACKs received */
+  u32 dupacks_in;	/**< RFC4898 tcpEStatsStackDupAcksIn*/
   u8 pending_dupacks;	/**< Number of DUPACKs to be sent */
+  u32 dupacks_out;	/**< RFC4898 tcpEStatsPathDupAcksOut */
 
   /* Congestion control */
   u32 cwnd;		/**< Congestion window */
@@ -361,6 +380,12 @@ typedef struct _tcp_connection
   tcp_cc_algorithm_t *cc_algo;	/**< Congestion control algorithm */
   u8 cc_data[TCP_CC_DATA_SZ];	/**< Congestion control algo private data */
 
+  u32 fr_occurences;	/**< fast-retransmit occurrences RFC4898
+			     tcpEStatsStackFastRetran */
+  u32 tr_occurences;	/**< timer-retransmit occurrences */
+  u64 bytes_retrans;	/**< RFC4898 tcpEStatsPerfOctetsRetrans */
+  u64 segs_retrans;	/**< RFC4898 tcpEStatsPerfSegsRetrans*/
+
   /* RTT and RTO */
   u32 rto;		/**< Retransmission timeout */
   u32 rto_boff;		/**< Index for RTO backoff */
@@ -382,9 +407,12 @@ typedef struct _tcp_connection
   f64 delivered_time;		/**< Time last bytes were acked */
   tcp_byte_tracker_t *bt;	/**< Tx byte tracker */
 
+  tcp_errors_t errors;	/**< Soft connection errors */
+
+  f64 start_ts;		/**< Timestamp when connection initialized */
   u32 last_fib_check;	/**< Last time we checked fib route for peer */
   u16 mss;		/**< Our max seg size that includes options */
-  u32 timestamp_delta;
+  u32 timestamp_delta;	/**< Offset for timestamp */
 } tcp_connection_t;
 
 /* *INDENT-OFF* */
diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c
index 4695fbb1161..2fadab59eb8 100755
--- a/src/vnet/tcp/tcp_input.c
+++ b/src/vnet/tcp/tcp_input.c
@@ -365,6 +365,9 @@ tcp_segment_validate (tcp_worker_ctx_t * wrk, tcp_connection_t * tc0,
 
       *error0 = TCP_ERROR_RCV_WND;
 
+      tc0->errors.below_data_wnd += seq_lt (vnet_buffer (b0)->tcp.seq_end,
+					    tc0->rcv_las);
+
       /* If not RST, send dup ack */
       if (!tcp_rst (th0))
 	{
@@ -1174,6 +1177,7 @@ tcp_cc_init_congestion (tcp_connection_t * tc)
   tc->prev_ssthresh = tc->ssthresh;
   tc->prev_cwnd = tc->cwnd;
   tc->cc_algo->congestion (tc);
+  tc->fr_occurences += 1;
   TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 4);
 }
 #endif /* CLIB_MARCH_VARIANT */
@@ -1524,6 +1528,7 @@ tcp_rcv_ack (tcp_worker_ctx_t * wrk, tcp_connection_t * tc, vlib_buffer_t * b,
 	  goto process_ack;
 	}
 
+      tc->errors.above_ack_wnd += 1;
       *error = TCP_ERROR_ACK_FUTURE;
       TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 0,
 		   vnet_buffer (b)->tcp.ack_number);
@@ -1533,6 +1538,7 @@ tcp_rcv_ack (tcp_worker_ctx_t * wrk, tcp_connection_t * tc, vlib_buffer_t * b,
   /* If old ACK, probably it's an old dupack */
   if (PREDICT_FALSE (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una)))
     {
+      tc->errors.below_ack_wnd += 1;
       *error = TCP_ERROR_ACK_OLD;
       TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 1,
 		   vnet_buffer (b)->tcp.ack_number);
@@ -1547,6 +1553,7 @@ process_ack:
   /*
    * Looks okay, process feedback
    */
+
   if (tcp_opts_sack_permitted (&tc->rcv_opts))
     tcp_rcv_sacks (tc, vnet_buffer (b)->tcp.ack_number);
 
@@ -1577,6 +1584,7 @@ process_ack:
   if (tcp_ack_is_cc_event (tc, b, prev_snd_wnd, prev_snd_una, &is_dack))
     {
       tcp_cc_handle_event (tc, &rs, is_dack);
+      tc->dupacks_in += is_dack;
       if (!tcp_in_cong_recovery (tc))
 	{
 	  *error = TCP_ERROR_ACK_OK;
@@ -1741,6 +1749,7 @@ tcp_session_enqueue_data (tcp_connection_t * tc, vlib_buffer_t * b,
   ASSERT (data_len);
   written = session_enqueue_stream_connection (&tc->connection, b, 0,
 					       1 /* queue event */ , 1);
+  tc->bytes_in += written;
 
   TCP_EVT_DBG (TCP_EVT_INPUT, tc, 0, data_len, written);
 
@@ -1801,6 +1810,7 @@ tcp_session_enqueue_ooo (tcp_connection_t * tc, vlib_buffer_t * b,
     }
 
   TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, data_len);
+  tc->bytes_in += data_len;
 
   /* Update SACK list if in use */
   if (tcp_opts_sack_permitted (&tc->rcv_opts))
@@ -1889,6 +1899,7 @@ tcp_segment_rcv (tcp_worker_ctx_t * wrk, tcp_connection_t * tc,
   vlib_buffer_advance (b, vnet_buffer (b)->tcp.data_offset);
   n_data_bytes = vnet_buffer (b)->tcp.data_len;
   ASSERT (n_data_bytes);
+  tc->data_segs_in += 1;
 
   /* Handle out-of-order data */
   if (PREDICT_FALSE (vnet_buffer (b)->tcp.seq_number != tc->rcv_nxt))
@@ -1923,6 +1934,8 @@ tcp_segment_rcv (tcp_worker_ctx_t * wrk, tcp_connection_t * tc,
       error = tcp_session_enqueue_ooo (tc, b, n_data_bytes);
       tcp_program_dupack (tc);
       TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc, vnet_buffer (b)->tcp);
+      tc->errors.above_data_wnd += seq_gt (vnet_buffer (b)->tcp.seq_end,
+					   tc->rcv_las + tc->rcv_wnd);
       goto done;
     }
 
@@ -3398,6 +3411,7 @@ tcp_input_dispatch_buffer (tcp_main_t * tm, tcp_connection_t * tc,
   flags = tcp->flags & filter_flags;
   *next = tm->dispatch_table[tc->state][flags].next;
   *error = tm->dispatch_table[tc->state][flags].error;
+  tc->segs_in += 1;
 
   if (PREDICT_FALSE (*error == TCP_ERROR_DISPATCH
 		     || *next == TCP_INPUT_NEXT_RESET))
diff --git a/src/vnet/tcp/tcp_output.c b/src/vnet/tcp/tcp_output.c
index 010397b9e04..751d8b31382 100644
--- a/src/vnet/tcp/tcp_output.c
+++ b/src/vnet/tcp/tcp_output.c
@@ -1137,6 +1137,9 @@ tcp_push_hdr_i (tcp_connection_t * tc, vlib_buffer_t * b, u32 snd_nxt,
     tc->snd_nxt += data_len;
   tc->rcv_las = tc->rcv_nxt;
 
+  tc->bytes_out += data_len;
+  tc->data_segs_out += 1;
+
   TCP_EVT_DBG (TCP_EVT_PKTIZE, tc);
 }
 
@@ -1410,14 +1413,14 @@ tcp_prepare_retransmit_segment (tcp_worker_ctx_t * wrk,
   /* Start is beyond snd_congestion */
   start = tc->snd_una + offset;
   if (seq_geq (start, tc->snd_congestion))
-    goto done;
+    return 0;
 
   /* Don't overshoot snd_congestion */
   if (seq_gt (start + max_deq_bytes, tc->snd_congestion))
     {
       max_deq_bytes = tc->snd_congestion - start;
       if (max_deq_bytes == 0)
-	goto done;
+	return 0;
     }
 
   n_bytes = tcp_prepare_segment (wrk, tc, offset, max_deq_bytes, b);
@@ -1431,7 +1434,8 @@ tcp_prepare_retransmit_segment (tcp_worker_ctx_t * wrk,
 	tcp_bt_track_rxt (tc, start, start + n_bytes);
     }
 
-done:
+  tc->bytes_retrans += n_bytes;
+  tc->segs_retrans += 1;
   TCP_EVT_DBG (TCP_EVT_CC_RTX, tc, offset, n_bytes);
   return n_bytes;
 }
@@ -1458,6 +1462,7 @@ tcp_cc_init_rxt_timeout (tcp_connection_t * tc)
   tc->rcv_dupacks = 0;
   tc->rtt_ts = 0;
   tc->cwnd_acc_bytes = 0;
+  tc->tr_occurences += 1;
   tcp_connection_tx_pacer_reset (tc, tc->cwnd, 2 * tc->snd_mss);
   tcp_recovery_on (tc);
 }
@@ -2015,12 +2020,14 @@ tcp_send_acks (tcp_connection_t * tc, u32 max_burst_size)
     {
       tc->pending_dupacks = 0;
       tc->snd_sack_pos = 0;
+      tc->dupacks_out += n_acks;
       return n_acks;
     }
   else
     {
       TCP_DBG ("constrained by burst size");
       tc->pending_dupacks = n_acks - max_burst_size;
+      tc->dupacks_out += max_burst_size;
       tcp_program_dupack (tc);
       return max_burst_size;
     }
@@ -2196,6 +2203,8 @@ tcp_output_handle_packet (tcp_connection_t * tc0, vlib_buffer_t * b0,
 
   if (!TCP_ALWAYS_ACK)
     tcp_timer_reset (tc0, TCP_TIMER_DELACK);
+
+  tc0->segs_out += 1;
 }
 
 always_inline uword