FD.io VPP  v18.07-34-g55fbdb9
Vector Packet Processing
tcp.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14  */
15 
16 #ifndef _vnet_tcp_h_
17 #define _vnet_tcp_h_
18 
19 #include <vnet/vnet.h>
20 #include <vnet/ip/ip.h>
21 #include <vnet/tcp/tcp_packet.h>
22 #include <vnet/tcp/tcp_timer.h>
23 #include <vnet/session/transport.h>
24 #include <vnet/session/session.h>
25 #include <vnet/tcp/tcp_debug.h>
26 
27 #define TCP_TICK 0.001 /**< TCP tick period (s) */
28 #define THZ (u32) (1/TCP_TICK) /**< TCP tick frequency */
29 #define TCP_TSTAMP_RESOLUTION TCP_TICK /**< Time stamp resolution */
30 #define TCP_PAWS_IDLE 24 * 24 * 60 * 60 * THZ /**< 24 days */
31 #define TCP_FIB_RECHECK_PERIOD 1 * THZ /**< Recheck every 1s */
32 #define TCP_MAX_OPTION_SPACE 40
33 
34 #define TCP_DUPACK_THRESHOLD 3
35 #define TCP_MAX_RX_FIFO_SIZE 4 << 20
36 #define TCP_MIN_RX_FIFO_SIZE 4 << 10
37 #define TCP_IW_N_SEGMENTS 10
38 #define TCP_ALWAYS_ACK 1 /**< On/off delayed acks */
39 #define TCP_USE_SACKS 1 /**< Disable only for testing */
40 
41 /** TCP FSM state definitions as per RFC793. */
42 #define foreach_tcp_fsm_state \
43  _(CLOSED, "CLOSED") \
44  _(LISTEN, "LISTEN") \
45  _(SYN_SENT, "SYN_SENT") \
46  _(SYN_RCVD, "SYN_RCVD") \
47  _(ESTABLISHED, "ESTABLISHED") \
48  _(CLOSE_WAIT, "CLOSE_WAIT") \
49  _(FIN_WAIT_1, "FIN_WAIT_1") \
50  _(LAST_ACK, "LAST_ACK") \
51  _(CLOSING, "CLOSING") \
52  _(FIN_WAIT_2, "FIN_WAIT_2") \
53  _(TIME_WAIT, "TIME_WAIT")
54 
55 typedef enum _tcp_state
56 {
57 #define _(sym, str) TCP_STATE_##sym,
59 #undef _
61 } tcp_state_t;
62 
67 
68 /** TCP timers */
69 #define foreach_tcp_timer \
70  _(RETRANSMIT, "RETRANSMIT") \
71  _(DELACK, "DELAYED ACK") \
72  _(PERSIST, "PERSIST") \
73  _(KEEP, "KEEP") \
74  _(WAITCLOSE, "WAIT CLOSE") \
75  _(RETRANSMIT_SYN, "RETRANSMIT SYN") \
76  _(ESTABLISH, "ESTABLISH")
77 
78 typedef enum _tcp_timers
79 {
80 #define _(sym, str) TCP_TIMER_##sym,
82 #undef _
84 } tcp_timers_e;
85 
86 typedef void (timer_expiration_handler) (u32 index);
87 
92 
93 #define TCP_TIMER_HANDLE_INVALID ((u32) ~0)
94 
95 /* Timer delays as multiples of 100ms */
96 #define TCP_TO_TIMER_TICK TCP_TICK*10 /* Period for converting from TCP
97  * ticks to timer units */
98 #define TCP_DELACK_TIME 1 /* 0.1s */
99 #define TCP_ESTABLISH_TIME 750 /* 75s */
100 #define TCP_SYN_RCVD_TIME 600 /* 60s */
101 #define TCP_2MSL_TIME 300 /* 30s */
102 #define TCP_CLOSEWAIT_TIME 20 /* 2s */
103 #define TCP_TIMEWAIT_TIME 100 /* 10s */
104 #define TCP_CLEANUP_TIME 10 /* 1s Time to wait before cleanup */
105 #define TCP_TIMER_PERSIST_MIN 2 /* 0.2s */
106 
107 #define TCP_RTO_MAX 60 * THZ /* Min max RTO (60s) as per RFC6298 */
108 #define TCP_RTO_MIN 0.2 * THZ /* Min RTO (200ms) - lower than standard */
109 #define TCP_RTT_MAX 30 * THZ /* 30s (probably too much) */
110 #define TCP_RTO_SYN_RETRIES 3 /* SYN retries without doubling RTO */
111 #define TCP_RTO_INIT 1 * THZ /* Initial retransmit timer */
112 
113 /** TCP connection flags */
114 #define foreach_tcp_connection_flag \
115  _(SNDACK, "Send ACK") \
116  _(FINSNT, "FIN sent") \
117  _(SENT_RCV_WND0, "Sent 0 rcv_wnd") \
118  _(RECOVERY, "Recovery") \
119  _(FAST_RECOVERY, "Fast Recovery") \
120  _(FR_1_SMSS, "Sent 1 SMSS") \
121  _(HALF_OPEN_DONE, "Half-open completed") \
122  _(FINPNDG, "FIN pending") \
123 
124 typedef enum _tcp_connection_flag_bits
125 {
126 #define _(sym, str) TCP_CONN_##sym##_BIT,
128 #undef _
131 
132 typedef enum _tcp_connection_flag
133 {
134 #define _(sym, str) TCP_CONN_##sym = 1 << TCP_CONN_##sym##_BIT,
136 #undef _
139 
140 /** TCP buffer flags */
141 #define foreach_tcp_buf_flag \
142  _ (ACK) /**< Sending ACK. */ \
143  _ (DUPACK) /**< Sending DUPACK. */ \
144 
145 enum
146 {
147 #define _(f) TCP_BUF_BIT_##f,
149 #undef _
151 };
152 
153 enum
154 {
155 #define _(f) TCP_BUF_FLAG_##f = 1 << TCP_BUF_BIT_##f,
157 #undef _
158 };
159 
160 #define TCP_SCOREBOARD_TRACE (0)
161 #define TCP_MAX_SACK_BLOCKS 15 /**< Max number of SACK blocks stored */
162 #define TCP_INVALID_SACK_HOLE_INDEX ((u32)~0)
164 typedef struct _scoreboard_trace_elt
165 {
166  u32 start;
167  u32 end;
168  u32 ack;
169  u32 snd_una_max;
170  u32 group;
172 
173 typedef struct _sack_scoreboard_hole
174 {
175  u32 next; /**< Index for next entry in linked list */
176  u32 prev; /**< Index for previous entry in linked list */
177  u32 start; /**< Start sequence number */
178  u32 end; /**< End sequence number */
179  u8 is_lost; /**< Mark hole as lost */
181 
182 typedef struct _sack_scoreboard
183 {
184  sack_scoreboard_hole_t *holes; /**< Pool of holes */
185  u32 head; /**< Index of first entry */
186  u32 tail; /**< Index of last entry */
187  u32 sacked_bytes; /**< Number of bytes sacked in sb */
188  u32 last_sacked_bytes; /**< Number of bytes last sacked */
189  u32 last_bytes_delivered; /**< Number of sack bytes delivered */
190  u32 snd_una_adv; /**< Bytes to add to snd_una */
191  u32 high_sacked; /**< Highest byte sacked (fack) */
192  u32 high_rxt; /**< Highest retransmitted sequence */
193  u32 rescue_rxt; /**< Rescue sequence number */
194  u32 lost_bytes; /**< Bytes lost as per RFC6675 */
195  u32 cur_rxt_hole; /**< Retransmitting from this hole */
196 
197 #if TCP_SCOREBOARD_TRACE
199 #endif
200 
202 
203 #if TCP_SCOREBOARD_TRACE
204 #define tcp_scoreboard_trace_add(_tc, _ack) \
205 { \
206  static u64 _group = 0; \
207  sack_scoreboard_t *_sb = &_tc->sack_sb; \
208  sack_block_t *_sack, *_sacks; \
209  scoreboard_trace_elt_t *_elt; \
210  int i; \
211  _group++; \
212  _sacks = _tc->rcv_opts.sacks; \
213  for (i = 0; i < vec_len (_sacks); i++) \
214  { \
215  _sack = &_sacks[i]; \
216  vec_add2 (_sb->trace, _elt, 1); \
217  _elt->start = _sack->start; \
218  _elt->end = _sack->end; \
219  _elt->ack = _elt->end == _ack ? _ack : 0; \
220  _elt->snd_una_max = _elt->end == _ack ? _tc->snd_una_max : 0; \
221  _elt->group = _group; \
222  } \
223 }
224 #else
225 #define tcp_scoreboard_trace_add(_tc, _ack)
226 #endif
227 
230  start, u8 have_sent_1_smss,
231  u8 * can_rescue,
232  u8 * snd_limited);
234  u32 index);
235 
237  sack_scoreboard_hole_t * hole);
239  sack_scoreboard_hole_t * hole);
244 u8 *format_tcp_scoreboard (u8 * s, va_list * args);
245 
246 typedef enum _tcp_cc_algorithm_type
247 {
250 
251 typedef struct _tcp_cc_algorithm tcp_cc_algorithm_t;
253 typedef enum _tcp_cc_ack_t
254 {
255  TCP_CC_ACK,
259 
260 typedef struct _tcp_connection
261 {
262  transport_connection_t connection; /**< Common transport data. First! */
263 
264  u8 state; /**< TCP state as per tcp_state_t */
265  u16 flags; /**< Connection flags (see tcp_conn_flags_e) */
266  u32 timers[TCP_N_TIMERS]; /**< Timer handles into timer wheel */
267 
268  /* TODO RFC4898 */
269 
270  /** Send sequence variables RFC793 */
271  u32 snd_una; /**< oldest unacknowledged sequence number */
272  u32 snd_una_max; /**< newest unacknowledged sequence number + 1*/
273  u32 snd_wnd; /**< send window */
274  u32 snd_wl1; /**< seq number used for last snd.wnd update */
275  u32 snd_wl2; /**< ack number used for last snd.wnd update */
276  u32 snd_nxt; /**< next seq number to be sent */
277  u16 snd_mss; /**< Effective send max seg (data) size */
278 
279  /** Receive sequence variables RFC793 */
280  u32 rcv_nxt; /**< next sequence number expected */
281  u32 rcv_wnd; /**< receive window we expect */
282 
283  u32 rcv_las; /**< rcv_nxt at last ack sent/rcv_wnd update */
284  u32 iss; /**< initial sent sequence */
285  u32 irs; /**< initial remote sequence */
286 
287  /* Options */
288  u8 snd_opts_len; /**< Tx options len */
289  u8 rcv_wscale; /**< Window scale to advertise to peer */
290  u8 snd_wscale; /**< Window scale to use when sending */
291  u32 tsval_recent; /**< Last timestamp received */
292  u32 tsval_recent_age; /**< When last updated tstamp_recent*/
293  tcp_options_t snd_opts; /**< Tx options for connection */
294  tcp_options_t rcv_opts; /**< Rx options for connection */
295 
296  sack_block_t *snd_sacks; /**< Vector of SACKs to send. XXX Fixed size? */
297  sack_scoreboard_t sack_sb; /**< SACK "scoreboard" that tracks holes */
298 
299  u16 rcv_dupacks; /**< Number of DUPACKs received */
300  u8 snt_dupacks; /**< Number of DUPACKs sent in a burst */
301 
302  /* Congestion control */
303  u32 cwnd; /**< Congestion window */
304  u32 cwnd_acc_bytes; /**< Bytes accumulated for cwnd increment */
305  u32 ssthresh; /**< Slow-start threshold */
306  u32 prev_ssthresh; /**< ssthresh before congestion */
307  u32 prev_cwnd; /**< ssthresh before congestion */
308  u32 bytes_acked; /**< Bytes acknowledged by current segment */
309  u32 snd_rxt_bytes; /**< Retransmitted bytes */
310  u32 snd_rxt_ts; /**< Timestamp when first packet is retransmitted */
311  u32 tsecr_last_ack; /**< Timestamp echoed to us in last healthy ACK */
312  u32 snd_congestion; /**< snd_una_max when congestion is detected */
313  tcp_cc_algorithm_t *cc_algo; /**< Congestion control algorithm */
314 
315  /* RTT and RTO */
316  u32 rto; /**< Retransmission timeout */
317  u32 rto_boff; /**< Index for RTO backoff */
318  u32 srtt; /**< Smoothed RTT */
319  u32 rttvar; /**< Smoothed mean RTT difference. Approximates variance */
320  u32 rtt_ts; /**< Timestamp for tracked ACK */
321  u32 rtt_seq; /**< Sequence number for tracked ACK */
322 
323  u16 mss; /**< Our max seg size that includes options */
324  u32 limited_transmit; /**< snd_nxt when limited transmit starts */
325  u32 last_fib_check; /**< Last time we checked fib route for peer */
326  u32 sw_if_index; /**< Interface for the connection */
328 
329 struct _tcp_cc_algorithm
330 {
331  void (*rcv_ack) (tcp_connection_t * tc);
332  void (*rcv_cong_ack) (tcp_connection_t * tc, tcp_cc_ack_t ack);
333  void (*congestion) (tcp_connection_t * tc);
334  void (*recovered) (tcp_connection_t * tc);
335  void (*init) (tcp_connection_t * tc);
336 };
337 
338 #define tcp_fastrecovery_on(tc) (tc)->flags |= TCP_CONN_FAST_RECOVERY
339 #define tcp_fastrecovery_off(tc) (tc)->flags &= ~TCP_CONN_FAST_RECOVERY
340 #define tcp_recovery_on(tc) (tc)->flags |= TCP_CONN_RECOVERY
341 #define tcp_recovery_off(tc) (tc)->flags &= ~TCP_CONN_RECOVERY
342 #define tcp_in_fastrecovery(tc) ((tc)->flags & TCP_CONN_FAST_RECOVERY)
343 #define tcp_in_recovery(tc) ((tc)->flags & (TCP_CONN_RECOVERY))
344 #define tcp_in_slowstart(tc) (tc->cwnd < tc->ssthresh)
345 #define tcp_fastrecovery_sent_1_smss(tc) ((tc)->flags & TCP_CONN_FR_1_SMSS)
346 #define tcp_fastrecovery_1_smss_on(tc) ((tc)->flags |= TCP_CONN_FR_1_SMSS)
347 #define tcp_fastrecovery_1_smss_off(tc) ((tc)->flags &= ~TCP_CONN_FR_1_SMSS)
349 #define tcp_in_cong_recovery(tc) ((tc)->flags & \
350  (TCP_CONN_FAST_RECOVERY | TCP_CONN_RECOVERY))
351 
352 always_inline void
354 {
355  tc->flags &= ~(TCP_CONN_FAST_RECOVERY | TCP_CONN_RECOVERY);
357 }
358 
359 typedef enum _tcp_error
360 {
361 #define tcp_error(n,s) TCP_ERROR_##n,
363 #undef tcp_error
364  TCP_N_ERROR,
366 
367 typedef struct _tcp_lookup_dispatch
368 {
369  u8 next, error;
371 
372 typedef struct tcp_worker_ctx_
373 {
374  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
375  u32 time_now; /**< worker time */
376  tw_timer_wheel_16t_2w_512sl_t timer_wheel; /**< worker timer wheel */
377  u32 *tx_buffers; /**< tx buffer free list */
378  vlib_frame_t *tx_frames[2]; /**< tx frames for tcp 4/6
379  output nodes */
380  vlib_frame_t *ip_lookup_tx_frames[2]; /**< tx frames for ip 4/6
381  lookup nodes */
382  CLIB_CACHE_LINE_ALIGN_MARK (cacheline1);
383  u8 cached_opts[40]; /**< cached 'on the wire'
384  options for bursts */
386 
387 typedef struct _tcp_main
388 {
389  /* Per-worker thread tcp connection pools */
390  tcp_connection_t **connections;
391 
392  /* Pool of listeners. */
393  tcp_connection_t *listener_pool;
394 
395  /** Dispatch table by state and flags */
396  tcp_lookup_dispatch_t dispatch_table[TCP_N_STATES][64];
397 
398  u8 log2_tstamp_clocks_per_tick;
399  f64 tstamp_ticks_per_clock;
400 
401  /** per-worker context */
402  tcp_worker_ctx_t *wrk_ctx;
403 
404  /* Pool of half-open connections on which we've sent a SYN */
405  tcp_connection_t *half_open_connections;
406  clib_spinlock_t half_open_lock;
407 
408  /* Congestion control algorithms registered */
409  tcp_cc_algorithm_t *cc_algos;
410 
411  /* Flag that indicates if stack is on or off */
412  u8 is_enabled;
413 
414  /** Number of preallocated connections */
415  u32 preallocated_connections;
416  u32 preallocated_half_open_connections;
417 
418  /** Vectors of src addresses. Optional unless one needs > 63K active-opens */
419  ip4_address_t *ip4_src_addresses;
420  u32 last_v4_address_rotor;
421  u32 last_v6_address_rotor;
422  ip6_address_t *ip6_src_addresses;
423 
424  /** vlib buffer size */
425  u32 bytes_per_buffer;
426 
427  u8 punt_unknown4;
428  u8 punt_unknown6;
429 
430  /** fault-injection */
431  f64 buffer_fail_fraction;
432 } tcp_main_t;
433 
434 extern tcp_main_t tcp_main;
439 
442 {
443  return &tcp_main;
444 }
445 
448 {
449  ASSERT ((signed) b->current_data >= (signed) -VLIB_BUFFER_PRE_DATA_SIZE);
450  return (tcp_header_t *) (b->data + b->current_data
451  + vnet_buffer (b)->tcp.hdr_offset);
452 }
453 
454 #if (VLIB_BUFFER_TRACE_TRAJECTORY)
455 #define tcp_trajectory_add_start(b, start) \
456 { \
457  (*vlib_buffer_trace_trajectory_cb) (b, start); \
458 }
459 #else
460 #define tcp_trajectory_add_start(b, start)
461 #endif
462 
464 
465 void tcp_punt_unknown (vlib_main_t * vm, u8 is_ip4, u8 is_add);
466 
468 tcp_connection_get (u32 conn_index, u32 thread_index)
469 {
470  if (PREDICT_FALSE
471  (pool_is_free_index (tcp_main.connections[thread_index], conn_index)))
472  return 0;
473  return pool_elt_at_index (tcp_main.connections[thread_index], conn_index);
474 }
475 
477 tcp_connection_get_if_valid (u32 conn_index, u32 thread_index)
478 {
479  if (tcp_main.connections[thread_index] == 0)
480  return 0;
481  if (pool_is_free_index (tcp_main.connections[thread_index], conn_index))
482  return 0;
483  return pool_elt_at_index (tcp_main.connections[thread_index], conn_index);
484 }
485 
488 {
489  return (tcp_connection_t *) tconn;
490 }
491 
496 tcp_connection_t *tcp_connection_new (u8 thread_index);
499  ip4_address_t * start,
500  ip4_address_t * end, u32 table_id);
502  ip6_address_t * start,
503  ip6_address_t * end, u32 table_id);
504 void tcp_api_reference (void);
505 u8 *format_tcp_connection (u8 * s, va_list * args);
506 
508 tcp_listener_get (u32 tli)
509 {
510  return pool_elt_at_index (tcp_main.listener_pool, tli);
511 }
512 
515 {
516  tcp_connection_t *tc = 0;
517  clib_spinlock_lock_if_init (&tcp_main.half_open_lock);
518  if (!pool_is_free_index (tcp_main.half_open_connections, conn_index))
519  tc = pool_elt_at_index (tcp_main.half_open_connections, conn_index);
520  clib_spinlock_unlock_if_init (&tcp_main.half_open_lock);
521  return tc;
522 }
523 
528  u8 is_ip4);
529 void tcp_send_reset (tcp_connection_t * tc);
530 void tcp_send_syn (tcp_connection_t * tc);
531 void tcp_send_fin (tcp_connection_t * tc);
532 void tcp_init_mss (tcp_connection_t * tc);
534 void tcp_update_rto (tcp_connection_t * tc);
535 void tcp_flush_frame_to_output (vlib_main_t * vm, u8 thread_index, u8 is_ip4);
536 void tcp_flush_frames_to_output (u8 thread_index);
537 
539 tcp_end_seq (tcp_header_t * th, u32 len)
540 {
541  return th->seq_number + tcp_is_syn (th) + tcp_is_fin (th) + len;
542 }
543 
544 /* Modulo arithmetic for TCP sequence numbers */
545 #define seq_lt(_s1, _s2) ((i32)((_s1)-(_s2)) < 0)
546 #define seq_leq(_s1, _s2) ((i32)((_s1)-(_s2)) <= 0)
547 #define seq_gt(_s1, _s2) ((i32)((_s1)-(_s2)) > 0)
548 #define seq_geq(_s1, _s2) ((i32)((_s1)-(_s2)) >= 0)
549 #define seq_max(_s1, _s2) (seq_gt((_s1), (_s2)) ? (_s1) : (_s2))
551 /* Modulo arithmetic for timestamps */
552 #define timestamp_lt(_t1, _t2) ((i32)((_t1)-(_t2)) < 0)
553 #define timestamp_leq(_t1, _t2) ((i32)((_t1)-(_t2)) <= 0)
555 /**
556  * Our estimate of the number of bytes that have left the network
557  */
559 tcp_bytes_out (const tcp_connection_t * tc)
560 {
561  if (tcp_opts_sack_permitted (&tc->rcv_opts))
562  return tc->sack_sb.sacked_bytes + tc->sack_sb.lost_bytes;
563  else
564  return tc->rcv_dupacks * tc->snd_mss;
565 }
566 
567 /**
568  * Our estimate of the number of bytes in flight (pipe size)
569  */
572 {
573  int flight_size;
574 
575  flight_size = (int) (tc->snd_una_max - tc->snd_una) - tcp_bytes_out (tc)
576  + tc->snd_rxt_bytes;
577 
578  if (flight_size < 0)
579  {
580  if (0)
582  ("Negative: %u %u %u dupacks %u sacked bytes %u flags %d",
583  tc->snd_una_max - tc->snd_una, tcp_bytes_out (tc),
584  tc->snd_rxt_bytes, tc->rcv_dupacks, tc->sack_sb.sacked_bytes,
585  tc->rcv_opts.flags);
586  return 0;
587  }
588 
589  return flight_size;
590 }
591 
592 /**
593  * Initial cwnd as per RFC5681
594  */
597 {
598  if (tc->snd_mss > 2190)
599  return 2 * tc->snd_mss;
600  else if (tc->snd_mss > 1095)
601  return 3 * tc->snd_mss;
602  else
603  return 4 * tc->snd_mss;
604 }
605 
607 tcp_loss_wnd (const tcp_connection_t * tc)
608 {
609  return tc->snd_mss;
610 }
611 
614 {
615  return clib_min (tc->cwnd, tc->snd_wnd);
616 }
617 
620 {
621  u32 available_wnd = tcp_available_snd_wnd (tc);
622  int flight_size = (int) (tc->snd_nxt - tc->snd_una);
623 
624  if (available_wnd <= flight_size)
625  return 0;
626 
627  return available_wnd - flight_size;
628 }
629 
630 /**
631  * Estimate of how many bytes we can still push into the network
632  */
635 {
636  u32 available_wnd = tcp_available_snd_wnd (tc);
637  u32 flight_size = tcp_flight_size (tc);
638 
639  if (available_wnd <= flight_size)
640  return 0;
641 
642  return available_wnd - flight_size;
643 }
644 
647 {
648  if ((tc->flags & TCP_CONN_FINSNT) && tc->snd_una_max - tc->snd_una == 1)
649  return 1;
650  return 0;
651 }
652 
659 
661 
662 /* Made public for unit testing only */
663 void tcp_update_sack_list (tcp_connection_t * tc, u32 start, u32 end);
665 
667 tcp_time_now (void)
668 {
669  return tcp_main.wrk_ctx[vlib_get_thread_index ()].time_now;
670 }
671 
673 tcp_set_time_now (u32 thread_index)
674 {
675  tcp_main.wrk_ctx[thread_index].time_now = clib_cpu_time_now ()
676  * tcp_main.tstamp_ticks_per_clock;
677  return tcp_main.wrk_ctx[thread_index].time_now;
678 }
679 
681 
686 
687 always_inline void
689 {
690  /* Reset flags, make sure ack is sent */
691  tc->flags = TCP_CONN_SNDACK;
692  vnet_buffer (b)->tcp.flags &= ~TCP_BUF_FLAG_DUPACK;
693 }
694 
695 always_inline void
696 tcp_timer_set (tcp_connection_t * tc, u8 timer_id, u32 interval)
697 {
698  ASSERT (tc->c_thread_index == vlib_get_thread_index ());
699  ASSERT (tc->timers[timer_id] == TCP_TIMER_HANDLE_INVALID);
700  tc->timers[timer_id] =
701  tw_timer_start_16t_2w_512sl (&tcp_main.
702  wrk_ctx[tc->c_thread_index].timer_wheel,
703  tc->c_c_index, timer_id, interval);
704 }
705 
706 always_inline void
707 tcp_timer_reset (tcp_connection_t * tc, u8 timer_id)
708 {
709  ASSERT (tc->c_thread_index == vlib_get_thread_index ());
710  if (tc->timers[timer_id] == TCP_TIMER_HANDLE_INVALID)
711  return;
712 
713  tw_timer_stop_16t_2w_512sl (&tcp_main.
714  wrk_ctx[tc->c_thread_index].timer_wheel,
715  tc->timers[timer_id]);
716  tc->timers[timer_id] = TCP_TIMER_HANDLE_INVALID;
717 }
718 
719 always_inline void
720 tcp_timer_update (tcp_connection_t * tc, u8 timer_id, u32 interval)
721 {
722  ASSERT (tc->c_thread_index == vlib_get_thread_index ());
723  if (tc->timers[timer_id] != TCP_TIMER_HANDLE_INVALID)
724  tw_timer_update_16t_2w_512sl (&tcp_main.
725  wrk_ctx[tc->c_thread_index].timer_wheel,
726  tc->timers[timer_id], interval);
727  else
728  tc->timers[timer_id] =
729  tw_timer_start_16t_2w_512sl (&tcp_main.
730  wrk_ctx[tc->c_thread_index].timer_wheel,
731  tc->c_c_index, timer_id, interval);
732 }
733 
734 always_inline void
736 {
737  ASSERT (tc->snd_una != tc->snd_una_max);
738  tcp_timer_set (tc, TCP_TIMER_RETRANSMIT,
739  clib_max (tc->rto * TCP_TO_TIMER_TICK, 1));
740 }
741 
742 always_inline void
744 {
745  tcp_timer_reset (tc, TCP_TIMER_RETRANSMIT);
746 }
747 
748 always_inline void
750 {
751  tcp_timer_update (tc, TCP_TIMER_RETRANSMIT,
752  clib_max (tc->rto * TCP_TO_TIMER_TICK, 1));
753 }
754 
755 always_inline void
757 {
758  /* Reuse RTO. It's backed off in handler */
759  tcp_timer_set (tc, TCP_TIMER_PERSIST,
760  clib_max (tc->rto * TCP_TO_TIMER_TICK,
762 }
763 
764 always_inline void
766 {
767  tcp_timer_update (tc, TCP_TIMER_PERSIST,
768  clib_max (tc->rto * TCP_TO_TIMER_TICK,
770 }
771 
772 always_inline void
774 {
775  tcp_timer_reset (tc, TCP_TIMER_PERSIST);
776 }
777 
778 always_inline void
780 {
781  if (tc->snd_una == tc->snd_una_max)
782  {
784  if (tc->snd_wnd < tc->snd_mss)
786  }
787  else
788  tcp_timer_update (tc, TCP_TIMER_RETRANSMIT,
789  clib_max (tc->rto * TCP_TO_TIMER_TICK, 1));
790 }
791 
794 {
795  return tc->timers[timer] != TCP_TIMER_HANDLE_INVALID;
796 }
797 
798 #define tcp_validate_txf_size(_tc, _a) \
799  ASSERT(_tc->state != TCP_STATE_ESTABLISHED \
800  || session_tx_fifo_max_dequeue (&_tc->connection) >= _a)
801 
802 void tcp_rcv_sacks (tcp_connection_t * tc, u32 ack);
803 u8 *tcp_scoreboard_replay (u8 * s, tcp_connection_t * tc, u8 verbose);
804 
806  const tcp_cc_algorithm_t * vft);
807 
809 
810 /**
811  * Push TCP header to buffer
812  *
813  * @param vm - vlib_main
814  * @param b - buffer to write the header to
815  * @param sp_net - source port net order
816  * @param dp_net - destination port net order
817  * @param seq - sequence number net order
818  * @param ack - ack number net order
819  * @param tcp_hdr_opts_len - header and options length in bytes
820  * @param flags - header flags
821  * @param wnd - window size
822  *
823  * @return - pointer to start of TCP header
824  */
825 always_inline void *
827  u32 ack, u8 tcp_hdr_opts_len, u8 flags,
828  u16 wnd)
829 {
830  tcp_header_t *th;
831 
832  th = vlib_buffer_push_uninit (b, tcp_hdr_opts_len);
833 
834  th->src_port = sp;
835  th->dst_port = dp;
836  th->seq_number = seq;
837  th->ack_number = ack;
838  th->data_offset_and_reserved = (tcp_hdr_opts_len >> 2) << 4;
839  th->flags = flags;
840  th->window = wnd;
841  th->checksum = 0;
842  th->urgent_pointer = 0;
843  return th;
844 }
845 
846 /**
847  * Push TCP header to buffer
848  *
849  * @param b - buffer to write the header to
850  * @param sp_net - source port net order
851  * @param dp_net - destination port net order
852  * @param seq - sequence number host order
853  * @param ack - ack number host order
854  * @param tcp_hdr_opts_len - header and options length in bytes
855  * @param flags - header flags
856  * @param wnd - window size
857  *
858  * @return - pointer to start of TCP header
859  */
860 always_inline void *
861 vlib_buffer_push_tcp (vlib_buffer_t * b, u16 sp_net, u16 dp_net, u32 seq,
862  u32 ack, u8 tcp_hdr_opts_len, u8 flags, u16 wnd)
863 {
864  return vlib_buffer_push_tcp_net_order (b, sp_net, dp_net,
865  clib_host_to_net_u32 (seq),
866  clib_host_to_net_u32 (ack),
867  tcp_hdr_opts_len, flags,
868  clib_host_to_net_u16 (wnd));
869 }
870 
871 #endif /* _vnet_tcp_h_ */
872 
873 /*
874  * fd.io coding-style-patch-verification: ON
875  *
876  * Local Variables:
877  * eval: (c-set-style "gnu")
878  * End:
879  */
tcp_main_t tcp_main
Definition: tcp.c:29
static vlib_cli_command_t trace
(constructor) VLIB_CLI_COMMAND (trace)
Definition: vlib_api_cli.c:862
#define CLIB_CACHE_LINE_ALIGN_MARK(mark)
Definition: cache.h:63
#define tcp_fastrecovery_1_smss_off(tc)
Definition: tcp.h:348
#define clib_min(x, y)
Definition: clib.h:289
void scoreboard_clear(sack_scoreboard_t *sb)
Definition: tcp_input.c:783
static void tcp_retransmit_timer_set(tcp_connection_t *tc)
Definition: tcp.h:736
void tcp_make_fin(tcp_connection_t *tc, vlib_buffer_t *b)
Convert buffer to FIN-ACK.
Definition: tcp_output.c:573
struct _sack_block sack_block_t
void tcp_cc_init_congestion(tcp_connection_t *tc)
Init loss recovery/fast recovery.
Definition: tcp_input.c:1030
struct _scoreboard_trace_elt scoreboard_trace_elt_t
void tcp_connection_timers_reset(tcp_connection_t *tc)
Stop all connection timers.
Definition: tcp.c:401
struct _transport_connection transport_connection_t
#define TCP_TO_TIMER_TICK
Definition: tcp.h:96
vlib_node_registration_t tcp4_output_node
(constructor) VLIB_REGISTER_NODE (tcp4_output_node)
Definition: tcp_output.c:2038
void scoreboard_init(sack_scoreboard_t *sb)
Definition: tcp_input.c:775
tcp_connection_t * tcp_connection_new(u8 thread_index)
Definition: tcp.c:248
void tcp_fast_retransmit(tcp_connection_t *tc)
Do fast retransmit.
Definition: tcp_output.c:1805
static u32 tcp_bytes_out(const tcp_connection_t *tc)
Our estimate of the number of bytes that have left the network.
Definition: tcp.h:560
static tcp_connection_t * tcp_connection_get_if_valid(u32 conn_index, u32 thread_index)
Definition: tcp.h:478
void tcp_connection_del(tcp_connection_t *tc)
Connection removal.
Definition: tcp.c:240
struct _sack_scoreboard sack_scoreboard_t
void tcp_update_sack_list(tcp_connection_t *tc, u32 start, u32 end)
Build SACK list as per RFC2018.
Definition: tcp_input.c:1430
static tcp_connection_t * tcp_half_open_connection_get(u32 conn_index)
Definition: tcp.h:515
sack_scoreboard_hole_t * scoreboard_last_hole(sack_scoreboard_t *sb)
Definition: tcp_input.c:583
sack_scoreboard_hole_t * scoreboard_next_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:559
#define VLIB_BUFFER_PRE_DATA_SIZE
Definition: buffer.h:52
static_always_inline void clib_spinlock_unlock_if_init(clib_spinlock_t *p)
Definition: lock.h:98
struct _tcp_main tcp_main_t
vlib_node_registration_t tcp6_output_node
(constructor) VLIB_REGISTER_NODE (tcp6_output_node)
Definition: tcp_output.c:2059
static u64 clib_cpu_time_now(void)
Definition: time.h:73
timer_expiration_handler tcp_timer_retransmit_handler
u8 * format_tcp_scoreboard(u8 *s, va_list *args)
Definition: tcp.c:933
struct _tcp_lookup_dispatch tcp_lookup_dispatch_t
void tcp_update_burst_snd_vars(tcp_connection_t *tc)
Update burst send vars.
Definition: tcp_output.c:416
struct _tcp_connection tcp_connection_t
static u32 tcp_available_cc_snd_space(const tcp_connection_t *tc)
Estimate of how many bytes we can still push into the network.
Definition: tcp.h:635
static u32 tcp_available_snd_wnd(const tcp_connection_t *tc)
Definition: tcp.h:614
void tcp_fast_retransmit_sack(tcp_connection_t *tc)
Do fast retransmit with SACKs.
Definition: tcp_output.c:1685
static tcp_connection_t * tcp_get_connection_from_transport(transport_connection_t *tconn)
Definition: tcp.h:488
void tcp_connection_cleanup(tcp_connection_t *tc)
Cleans up connection state.
Definition: tcp.c:199
u8 *( format_function_t)(u8 *s, va_list *args)
Definition: format.h:48
void tcp_send_reset_w_pkt(tcp_connection_t *tc, vlib_buffer_t *pkt, u8 is_ip4)
Send reset without reusing existing buffer.
Definition: tcp_output.c:830
format_function_t format_tcp_flags
Definition: tcp.h:64
struct _tcp_header tcp_header_t
void tcp_connection_reset(tcp_connection_t *tc)
Notify session that connection has been reset.
Definition: tcp.c:265
unsigned char u8
Definition: types.h:56
struct _sack_scoreboard_hole sack_scoreboard_hole_t
double f64
Definition: types.h:142
void tcp_flush_frame_to_output(vlib_main_t *vm, u8 thread_index, u8 is_ip4)
Flush tx frame populated by retransmits and timer pops.
Definition: tcp_output.c:1022
u8 * tcp_scoreboard_replay(u8 *s, tcp_connection_t *tc, u8 verbose)
Definition: tcp.c:1737
i16 current_data
signed offset in data[], pre_data[] that we are currently processing.
Definition: buffer.h:104
#define tcp_is_fin(_th)
Definition: tcp_packet.h:90
sack_scoreboard_hole_t * scoreboard_get_hole(sack_scoreboard_t *sb, u32 index)
Definition: tcp_input.c:551
#define always_inline
Definition: clib.h:92
static u32 tcp_available_output_snd_space(const tcp_connection_t *tc)
Definition: tcp.h:620
static tcp_header_t * tcp_buffer_hdr(vlib_buffer_t *b)
Definition: tcp.h:448
static timer_callback_t * timers
Definition: timer.c:57
enum _tcp_state tcp_state_t
vhost_vring_state_t state
Definition: vhost_user.h:115
timer_expiration_handler tcp_timer_retransmit_syn_handler
static u32 tcp_time_now(void)
Definition: tcp.h:668
unsigned int u32
Definition: types.h:88
sack_scoreboard_hole_t * scoreboard_next_rxt_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *start, u8 have_sent_1_smss, u8 *can_rescue, u8 *snd_limited)
Figure out the next hole to retransmit.
Definition: tcp_input.c:707
void tcp_api_reference(void)
Definition: tcp_api.c:109
static void tcp_timer_set(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:697
struct tcp_worker_ctx_ tcp_worker_ctx_t
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:464
timer_expiration_handler tcp_timer_persist_handler
u32 tcp_push_header(tcp_connection_t *tconn, vlib_buffer_t *b)
Definition: tcp_output.c:1185
u32 tcp_sack_list_bytes(tcp_connection_t *tc)
Definition: tcp_input.c:1484
static void * vlib_buffer_push_tcp_net_order(vlib_buffer_t *b, u16 sp, u16 dp, u32 seq, u32 ack, u8 tcp_hdr_opts_len, u8 flags, u16 wnd)
Push TCP header to buffer.
Definition: tcp.h:827
void tcp_retransmit_first_unacked(tcp_connection_t *tc)
Retransmit first unacked segment.
Definition: tcp_output.c:1662
void tcp_rcv_sacks(tcp_connection_t *tc, u32 ack)
Definition: tcp_input.c:817
clib_error_t * vnet_tcp_enable_disable(vlib_main_t *vm, u8 is_en)
Definition: tcp.c:1328
void tcp_send_syn(tcp_connection_t *tc)
Send SYN.
Definition: tcp_output.c:986
unsigned short u16
Definition: types.h:57
sack_scoreboard_hole_t * scoreboard_prev_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:567
#define TCP_TIMER_HANDLE_INVALID
Definition: tcp.h:93
void tcp_fast_retransmit_no_sack(tcp_connection_t *tc)
Fast retransmit without SACK info.
Definition: tcp_output.c:1768
static u32 tcp_flight_size(const tcp_connection_t *tc)
Our estimate of the number of bytes in flight (pipe size)
Definition: tcp.h:572
#define PREDICT_FALSE(x)
Definition: clib.h:105
void tcp_cc_algo_register(tcp_cc_algorithm_type_e type, const tcp_cc_algorithm_t *vft)
Definition: tcp.c:515
void( timer_expiration_handler)(u32 index)
Definition: tcp.h:86
enum _tcp_cc_ack_t tcp_cc_ack_t
u32 flags
Definition: vhost_user.h:110
static void tcp_timer_reset(tcp_connection_t *tc, u8 timer_id)
Definition: tcp.h:708
enum _tcp_error tcp_error_t
static void * vlib_buffer_push_tcp(vlib_buffer_t *b, u16 sp_net, u16 dp_net, u32 seq, u32 ack, u8 tcp_hdr_opts_len, u8 flags, u16 wnd)
Push TCP header to buffer.
Definition: tcp.h:862
int tcp_configure_v4_source_address_range(vlib_main_t *vm, ip4_address_t *start, ip4_address_t *end, u32 table_id)
Configure an ipv4 source address range.
Definition: tcp.c:1421
static_always_inline uword vlib_get_thread_index(void)
Definition: threads.h:221
void tcp_send_reset(tcp_connection_t *tc)
Build and set reset packet for connection.
Definition: tcp_output.c:909
void tcp_punt_unknown(vlib_main_t *vm, u8 is_ip4, u8 is_add)
Definition: tcp.c:1346
format_function_t format_tcp_state
Definition: tcp.h:63
#define clib_warning(format, args...)
Definition: error.h:59
enum _tcp_timers tcp_timers_e
static void tcp_connection_force_ack(tcp_connection_t *tc, vlib_buffer_t *b)
Definition: tcp.h:689
int tcp_half_open_connection_cleanup(tcp_connection_t *tc)
Try to cleanup half-open connection.
Definition: tcp.c:170
u32 fib_node_index_t
A typedef of a node index.
Definition: fib_types.h:30
void tcp_connection_timers_init(tcp_connection_t *tc)
Initialize all connection timers as invalid.
Definition: tcp.c:384
void tcp_make_synack(tcp_connection_t *ts, vlib_buffer_t *b)
Convert buffer to SYN-ACK.
Definition: tcp_output.c:616
void tcp_flush_frames_to_output(u8 thread_index)
Flush v4 and v6 tcp and ip-lookup tx frames for thread index.
Definition: tcp_output.c:1057
#define pool_is_free_index(P, I)
Use free bitmap to query whether given index is free.
Definition: pool.h:271
format_function_t format_tcp_rcv_sacks
Definition: tcp.h:66
vlib_node_registration_t tcp6_input_node
(constructor) VLIB_REGISTER_NODE (tcp6_input_node)
Definition: tcp_input.c:3387
fib_node_index_t tcp_lookup_rmt_in_fib(tcp_connection_t *tc)
void tcp_make_ack(tcp_connection_t *ts, vlib_buffer_t *b)
Convert buffer to ACK.
Definition: tcp_output.c:558
static void tcp_timer_update(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:721
#define ASSERT(truth)
enum _tcp_cc_algorithm_type tcp_cc_algorithm_type_e
enum _tcp_connection_flag_bits tcp_connection_flag_bits_e
void tcp_connection_init_vars(tcp_connection_t *tc)
Initialize tcp connection variables.
Definition: tcp.c:560
u8 * format_tcp_connection(u8 *s, va_list *args)
Definition: tcp.c:822
static u32 tcp_end_seq(tcp_header_t *th, u32 len)
Definition: tcp.h:540
static void init(void)
Definition: client.c:105
struct _tcp_cc_algorithm tcp_cc_algorithm_t
Definition: tcp.h:252
#define clib_max(x, y)
Definition: clib.h:282
struct _vlib_node_registration vlib_node_registration_t
#define tcp_is_syn(_th)
Definition: tcp_packet.h:89
static void tcp_persist_timer_update(tcp_connection_t *tc)
Definition: tcp.h:766
static void * vlib_buffer_push_uninit(vlib_buffer_t *b, u8 size)
Prepend uninitialized data to buffer.
Definition: buffer.h:299
vlib_node_registration_t tcp4_input_node
(constructor) VLIB_REGISTER_NODE (tcp4_input_node)
Definition: tcp_input.c:3364
tcp_cc_algorithm_t * tcp_cc_algo_get(tcp_cc_algorithm_type_e type)
Definition: tcp.c:525
static u32 tcp_initial_cwnd(const tcp_connection_t *tc)
Initial cwnd as per RFC5681.
Definition: tcp.h:597
#define foreach_tcp_fsm_state
TCP FSM state definitions as per RFC793.
Definition: tcp.h:42
void tcp_send_fin(tcp_connection_t *tc)
Send FIN.
Definition: tcp_output.c:1070
#define foreach_tcp_connection_flag
TCP connection flags.
Definition: tcp.h:114
#define foreach_tcp_timer
TCP timers.
Definition: tcp.h:69
static u32 tcp_set_time_now(u32 thread_index)
Definition: tcp.h:674
static u8 tcp_is_lost_fin(tcp_connection_t *tc)
Definition: tcp.h:647
sack_scoreboard_hole_t * scoreboard_first_hole(sack_scoreboard_t *sb)
Definition: tcp_input.c:575
static void tcp_retransmit_timer_update(tcp_connection_t *tc)
Definition: tcp.h:780
void tcp_init_snd_vars(tcp_connection_t *tc)
Initialize connection send variables.
Definition: tcp.c:536
void tcp_connection_close(tcp_connection_t *tc)
Begin connection closing procedure.
Definition: tcp.c:314
static void tcp_retransmit_timer_force_update(tcp_connection_t *tc)
Definition: tcp.h:750
#define vnet_buffer(b)
Definition: buffer.h:360
static tcp_connection_t * tcp_connection_get(u32 conn_index, u32 thread_index)
Definition: tcp.h:469
void tcp_update_rto(tcp_connection_t *tc)
Definition: tcp_input.c:431
void tcp_init_mss(tcp_connection_t *tc)
Definition: tcp_output.c:436
u8 data[0]
Packet data.
Definition: buffer.h:172
#define foreach_tcp_buf_flag
TCP buffer flags.
Definition: tcp.h:141
enum _tcp_connection_flag tcp_connection_flags_e
format_function_t format_tcp_sacks
Definition: tcp.h:65
#define TCP_TIMER_PERSIST_MIN
Definition: tcp.h:105
#define tcp_opts_sack_permitted(_to)
Definition: tcp_packet.h:160
static u32 tcp_loss_wnd(const tcp_connection_t *tc)
Definition: tcp.h:608
int tcp_configure_v6_source_address_range(vlib_main_t *vm, ip6_address_t *start, ip6_address_t *end, u32 table_id)
Configure an ipv6 source address range.
Definition: tcp.c:1513
static void tcp_persist_timer_set(tcp_connection_t *tc)
Definition: tcp.h:757
static tcp_main_t * vnet_get_tcp_main()
Definition: tcp.h:442
static void tcp_cong_recovery_off(tcp_connection_t *tc)
Definition: tcp.h:354
timer_expiration_handler tcp_timer_delack_handler
static_always_inline void clib_spinlock_lock_if_init(clib_spinlock_t *p)
Definition: lock.h:82
static void tcp_retransmit_timer_reset(tcp_connection_t *tc)
Definition: tcp.h:744
void tcp_cc_fastrecovery_exit(tcp_connection_t *tc)
Definition: tcp_input.c:1051
static u8 tcp_timer_is_active(tcp_connection_t *tc, tcp_timers_e timer)
Definition: tcp.h:794
static tcp_connection_t * tcp_listener_get(u32 tli)
Definition: tcp.h:509
static void tcp_persist_timer_reset(tcp_connection_t *tc)
Definition: tcp.h:774