FD.io VPP  v18.11-rc0-18-g2a3fb1a
Vector Packet Processing
tcp_input.c
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 #include <vppinfra/sparse_vec.h>
17 #include <vnet/tcp/tcp_packet.h>
18 #include <vnet/tcp/tcp.h>
19 #include <vnet/session/session.h>
20 #include <math.h>
21 
22 static char *tcp_error_strings[] = {
23 #define tcp_error(n,s) s,
24 #include <vnet/tcp/tcp_error.def>
25 #undef tcp_error
26 };
27 
28 /* All TCP nodes have the same outgoing arcs */
29 #define foreach_tcp_state_next \
30  _ (DROP4, "ip4-drop") \
31  _ (DROP6, "ip6-drop") \
32  _ (TCP4_OUTPUT, "tcp4-output") \
33  _ (TCP6_OUTPUT, "tcp6-output")
34 
35 typedef enum _tcp_established_next
36 {
37 #define _(s,n) TCP_ESTABLISHED_NEXT_##s,
39 #undef _
42 
43 typedef enum _tcp_rcv_process_next
44 {
45 #define _(s,n) TCP_RCV_PROCESS_NEXT_##s,
47 #undef _
50 
51 typedef enum _tcp_syn_sent_next
52 {
53 #define _(s,n) TCP_SYN_SENT_NEXT_##s,
55 #undef _
58 
59 typedef enum _tcp_listen_next
60 {
61 #define _(s,n) TCP_LISTEN_NEXT_##s,
63 #undef _
66 
67 /* Generic, state independent indices */
68 typedef enum _tcp_state_next
69 {
70 #define _(s,n) TCP_NEXT_##s,
72 #undef _
75 
76 #define tcp_next_output(is_ip4) (is_ip4 ? TCP_NEXT_TCP4_OUTPUT \
77  : TCP_NEXT_TCP6_OUTPUT)
78 
79 #define tcp_next_drop(is_ip4) (is_ip4 ? TCP_NEXT_DROP4 \
80  : TCP_NEXT_DROP6)
81 
84 
85 /**
86  * Validate segment sequence number. As per RFC793:
87  *
88  * Segment Receive Test
89  * Length Window
90  * ------- ------- -------------------------------------------
91  * 0 0 SEG.SEQ = RCV.NXT
92  * 0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND
93  * >0 0 not acceptable
94  * >0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND
95  * or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND
96  *
97  * This ultimately consists in checking if segment falls within the window.
98  * The one important difference compared to RFC793 is that we use rcv_las,
99  * or the rcv_nxt at last ack sent instead of rcv_nxt since that's the
100  * peer's reference when computing our receive window.
101  *
102  * This:
103  * seq_leq (end_seq, tc->rcv_las + tc->rcv_wnd) && seq_geq (seq, tc->rcv_las)
104  * however, is too strict when we have retransmits. Instead we just check that
105  * the seq is not beyond the right edge and that the end of the segment is not
106  * less than the left edge.
107  *
108  * N.B. rcv_nxt and rcv_wnd are both updated in this node if acks are sent, so
109  * use rcv_nxt in the right edge window test instead of rcv_las.
110  *
111  */
114 {
115  return (seq_geq (end_seq, tc->rcv_las)
116  && seq_leq (seq, tc->rcv_nxt + tc->rcv_wnd));
117 }
118 
119 /**
120  * Parse TCP header options.
121  *
122  * @param th TCP header
123  * @param to TCP options data structure to be populated
124  * @return -1 if parsing failed
125  */
126 static int
128 {
129  const u8 *data;
130  u8 opt_len, opts_len, kind;
131  int j;
132  sack_block_t b;
133 
134  opts_len = (tcp_doff (th) << 2) - sizeof (tcp_header_t);
135  data = (const u8 *) (th + 1);
136 
137  /* Zero out all flags but those set in SYN */
138  to->flags &= (TCP_OPTS_FLAG_SACK_PERMITTED | TCP_OPTS_FLAG_WSCALE);
139 
140  for (; opts_len > 0; opts_len -= opt_len, data += opt_len)
141  {
142  kind = data[0];
143 
144  /* Get options length */
145  if (kind == TCP_OPTION_EOL)
146  break;
147  else if (kind == TCP_OPTION_NOOP)
148  {
149  opt_len = 1;
150  continue;
151  }
152  else
153  {
154  /* broken options */
155  if (opts_len < 2)
156  return -1;
157  opt_len = data[1];
158 
159  /* weird option length */
160  if (opt_len < 2 || opt_len > opts_len)
161  return -1;
162  }
163 
164  /* Parse options */
165  switch (kind)
166  {
167  case TCP_OPTION_MSS:
168  if ((opt_len == TCP_OPTION_LEN_MSS) && tcp_syn (th))
169  {
170  to->flags |= TCP_OPTS_FLAG_MSS;
171  to->mss = clib_net_to_host_u16 (*(u16 *) (data + 2));
172  }
173  break;
175  if ((opt_len == TCP_OPTION_LEN_WINDOW_SCALE) && tcp_syn (th))
176  {
177  to->flags |= TCP_OPTS_FLAG_WSCALE;
178  to->wscale = data[2];
179  if (to->wscale > TCP_MAX_WND_SCALE)
180  {
181  clib_warning ("Illegal window scaling value: %d",
182  to->wscale);
184  }
185  }
186  break;
188  if (opt_len == TCP_OPTION_LEN_TIMESTAMP)
189  {
190  to->flags |= TCP_OPTS_FLAG_TSTAMP;
191  to->tsval = clib_net_to_host_u32 (*(u32 *) (data + 2));
192  to->tsecr = clib_net_to_host_u32 (*(u32 *) (data + 6));
193  }
194  break;
196  if (opt_len == TCP_OPTION_LEN_SACK_PERMITTED && tcp_syn (th))
197  to->flags |= TCP_OPTS_FLAG_SACK_PERMITTED;
198  break;
200  /* If SACK permitted was not advertised or a SYN, break */
201  if ((to->flags & TCP_OPTS_FLAG_SACK_PERMITTED) == 0 || tcp_syn (th))
202  break;
203 
204  /* If too short or not correctly formatted, break */
205  if (opt_len < 10 || ((opt_len - 2) % TCP_OPTION_LEN_SACK_BLOCK))
206  break;
207 
208  to->flags |= TCP_OPTS_FLAG_SACK;
209  to->n_sack_blocks = (opt_len - 2) / TCP_OPTION_LEN_SACK_BLOCK;
210  vec_reset_length (to->sacks);
211  for (j = 0; j < to->n_sack_blocks; j++)
212  {
213  b.start = clib_net_to_host_u32 (*(u32 *) (data + 2 + 8 * j));
214  b.end = clib_net_to_host_u32 (*(u32 *) (data + 6 + 8 * j));
215  vec_add1 (to->sacks, b);
216  }
217  break;
218  default:
219  /* Nothing to see here */
220  continue;
221  }
222  }
223  return 0;
224 }
225 
226 /**
227  * RFC1323: Check against wrapped sequence numbers (PAWS). If we have
228  * timestamp to echo and it's less than tsval_recent, drop segment
229  * but still send an ACK in order to retain TCP's mechanism for detecting
230  * and recovering from half-open connections
231  *
232  * Or at least that's what the theory says. It seems that this might not work
233  * very well with packet reordering and fast retransmit. XXX
234  */
235 always_inline int
237 {
238  return tcp_opts_tstamp (&tc->rcv_opts) && tc->tsval_recent
239  && timestamp_lt (tc->rcv_opts.tsval, tc->tsval_recent);
240 }
241 
242 /**
243  * Update tsval recent
244  */
245 always_inline void
247 {
248  /*
249  * RFC1323: If Last.ACK.sent falls within the range of sequence numbers
250  * of an incoming segment:
251  * SEG.SEQ <= Last.ACK.sent < SEG.SEQ + SEG.LEN
252  * then the TSval from the segment is copied to TS.Recent;
253  * otherwise, the TSval is ignored.
254  */
255  if (tcp_opts_tstamp (&tc->rcv_opts) && seq_leq (seq, tc->rcv_las)
256  && seq_leq (tc->rcv_las, seq_end))
257  {
258  ASSERT (timestamp_leq (tc->tsval_recent, tc->rcv_opts.tsval));
259  tc->tsval_recent = tc->rcv_opts.tsval;
260  tc->tsval_recent_age = tcp_time_now ();
261  }
262 }
263 
264 /**
265  * Validate incoming segment as per RFC793 p. 69 and RFC1323 p. 19
266  *
267  * It first verifies if segment has a wrapped sequence number (PAWS) and then
268  * does the processing associated to the first four steps (ignoring security
269  * and precedence): sequence number, rst bit and syn bit checks.
270  *
271  * @return 0 if segments passes validation.
272  */
273 static int
275  vlib_buffer_t * b0, tcp_header_t * th0,
276  u32 * next0, u32 * error0)
277 {
278  /* We could get a burst of RSTs interleaved with acks */
279  if (PREDICT_FALSE (tc0->state == TCP_STATE_CLOSED))
280  {
281  tcp_send_reset (tc0);
282  *error0 = TCP_ERROR_CONNECTION_CLOSED;
283  goto drop;
284  }
285 
286  if (PREDICT_FALSE (!tcp_ack (th0) && !tcp_rst (th0) && !tcp_syn (th0)))
287  {
288  *error0 = TCP_ERROR_SEGMENT_INVALID;
289  goto drop;
290  }
291 
292  if (PREDICT_FALSE (tcp_options_parse (th0, &tc0->rcv_opts)))
293  {
294  clib_warning ("options parse error");
295  *error0 = TCP_ERROR_OPTIONS;
296  goto drop;
297  }
298 
300  {
301  *error0 = TCP_ERROR_PAWS;
302  if (CLIB_DEBUG > 2)
303  clib_warning ("paws failed\n%U", format_tcp_connection, tc0, 2);
304  TCP_EVT_DBG (TCP_EVT_PAWS_FAIL, tc0, vnet_buffer (b0)->tcp.seq_number,
305  vnet_buffer (b0)->tcp.seq_end);
306 
307  /* If it just so happens that a segment updates tsval_recent for a
308  * segment over 24 days old, invalidate tsval_recent. */
309  if (timestamp_lt (tc0->tsval_recent_age + TCP_PAWS_IDLE,
310  tcp_time_now ()))
311  {
312  /* Age isn't reset until we get a valid tsval (bsd inspired) */
313  tc0->tsval_recent = 0;
314  clib_warning ("paws failed - really old segment. REALLY?");
315  }
316  else
317  {
318  /* Drop after ack if not rst */
319  if (!tcp_rst (th0))
320  {
321  tcp_make_ack (tc0, b0);
322  TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0, vnet_buffer (b0)->tcp);
323  goto error;
324  }
325  }
326  goto drop;
327  }
328 
329  /* 1st: check sequence number */
330  if (!tcp_segment_in_rcv_wnd (tc0, vnet_buffer (b0)->tcp.seq_number,
331  vnet_buffer (b0)->tcp.seq_end))
332  {
333  *error0 = TCP_ERROR_RCV_WND;
334  /* If our window is 0 and the packet is in sequence, let it pass
335  * through for ack processing. It should be dropped later. */
336  if (!(tc0->rcv_wnd == 0
337  && tc0->rcv_nxt == vnet_buffer (b0)->tcp.seq_number))
338  {
339  /* If not RST, send dup ack */
340  if (!tcp_rst (th0))
341  {
342  tcp_make_ack (tc0, b0);
343  TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0, vnet_buffer (b0)->tcp);
344  goto error;
345  }
346  goto drop;
347  }
348  }
349 
350  /* 2nd: check the RST bit */
351  if (PREDICT_FALSE (tcp_rst (th0)))
352  {
353  tcp_connection_reset (tc0);
354  *error0 = TCP_ERROR_RST_RCVD;
355  goto drop;
356  }
357 
358  /* 3rd: check security and precedence (skip) */
359 
360  /* 4th: check the SYN bit */
361  if (PREDICT_FALSE (tcp_syn (th0)))
362  {
363  /* TODO implement RFC 5961 */
364  if (tc0->state == TCP_STATE_SYN_RCVD)
365  {
366  tcp_make_synack (tc0, b0);
367  TCP_EVT_DBG (TCP_EVT_SYN_RCVD, tc0, 0);
368  }
369  else
370  {
371  tcp_make_ack (tc0, b0);
372  TCP_EVT_DBG (TCP_EVT_SYNACK_RCVD, tc0);
373  }
374  goto error;
375  }
376 
377  /* If segment in window, save timestamp */
378  tcp_update_timestamp (tc0, vnet_buffer (b0)->tcp.seq_number,
379  vnet_buffer (b0)->tcp.seq_end);
380  return 0;
381 
382 drop:
383  *next0 = tcp_next_drop (tc0->c_is_ip4);
384  return -1;
385 error:
386  *next0 = tcp_next_output (tc0->c_is_ip4);
387  return -1;
388 }
389 
390 always_inline int
392 {
393  /* SND.UNA =< SEG.ACK =< SND.NXT */
394  return (seq_leq (tc0->snd_una, vnet_buffer (tb0)->tcp.ack_number)
395  && seq_leq (vnet_buffer (tb0)->tcp.ack_number, tc0->snd_nxt));
396 }
397 
398 /**
399  * Compute smoothed RTT as per VJ's '88 SIGCOMM and RFC6298
400  *
401  * Note that although the original article, srtt and rttvar are scaled
402  * to minimize round-off errors, here we don't. Instead, we rely on
403  * better precision time measurements.
404  *
405  * TODO support us rtt resolution
406  */
407 static void
409 {
410  int err, diff;
411 
412  if (tc->srtt != 0)
413  {
414  err = mrtt - tc->srtt;
415 
416  /* XXX Drop in RTT results in RTTVAR increase and bigger RTO.
417  * The increase should be bound */
418  tc->srtt = clib_max ((int) tc->srtt + (err >> 3), 1);
419  diff = (clib_abs (err) - (int) tc->rttvar) >> 2;
420  tc->rttvar = clib_max ((int) tc->rttvar + diff, 1);
421  }
422  else
423  {
424  /* First measurement. */
425  tc->srtt = mrtt;
426  tc->rttvar = mrtt >> 1;
427  }
428 }
429 
430 void
432 {
433  tc->rto = clib_min (tc->srtt + (tc->rttvar << 2), TCP_RTO_MAX);
434  tc->rto = clib_max (tc->rto, TCP_RTO_MIN);
435 }
436 
437 /**
438  * Update RTT estimate and RTO timer
439  *
440  * Measure RTT: We have two sources of RTT measurements: TSOPT and ACK
441  * timing. Middle boxes are known to fiddle with TCP options so we
442  * should give higher priority to ACK timing.
443  *
444  * This should be called only if previously sent bytes have been acked.
445  *
446  * return 1 if valid rtt 0 otherwise
447  */
448 static int
450 {
451  u32 mrtt = 0;
452 
453  /* Karn's rule, part 1. Don't use retransmitted segments to estimate
454  * RTT because they're ambiguous. */
455  if (tcp_in_cong_recovery (tc) || tc->sack_sb.sacked_bytes)
456  goto done;
457 
458  if (tc->rtt_ts && seq_geq (ack, tc->rtt_seq))
459  {
460  mrtt = tcp_time_now () - tc->rtt_ts;
461  }
462  /* As per RFC7323 TSecr can be used for RTTM only if the segment advances
463  * snd_una, i.e., the left side of the send window:
464  * seq_lt (tc->snd_una, ack). This is a condition for calling update_rtt */
465  else if (tcp_opts_tstamp (&tc->rcv_opts) && tc->rcv_opts.tsecr)
466  {
467  mrtt = tcp_time_now () - tc->rcv_opts.tsecr;
468  }
469 
470  /* Ignore dubious measurements */
471  if (mrtt == 0 || mrtt > TCP_RTT_MAX)
472  goto done;
473 
474  tcp_estimate_rtt (tc, mrtt);
475 
476 done:
477 
478  /* Allow measuring of a new RTT */
479  tc->rtt_ts = 0;
480 
481  /* If we got here something must've been ACKed so make sure boff is 0,
482  * even if mrrt is not valid since we update the rto lower */
483  tc->rto_boff = 0;
484  tcp_update_rto (tc);
485 
486  return 0;
487 }
488 
489 /**
490  * Dequeue bytes that have been acked and while at it update RTT estimates.
491  */
492 static void
494 {
495  /* Dequeue the newly ACKed add SACKed bytes */
496  stream_session_dequeue_drop (&tc->connection,
497  tc->bytes_acked + tc->sack_sb.snd_una_adv);
498 
499  tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una);
500 
501  /* Update rtt and rto */
502  tcp_update_rtt (tc, ack);
503 
504  /* If everything has been acked, stop retransmit timer
505  * otherwise update. */
507 }
508 
509 /**
510  * Check if duplicate ack as per RFC5681 Sec. 2
511  */
512 static u8
514  u32 prev_snd_una)
515 {
516  return ((vnet_buffer (b)->tcp.ack_number == prev_snd_una)
517  && seq_gt (tc->snd_una_max, tc->snd_una)
518  && (vnet_buffer (b)->tcp.seq_end == vnet_buffer (b)->tcp.seq_number)
519  && (prev_snd_wnd == tc->snd_wnd));
520 }
521 
522 /**
523  * Checks if ack is a congestion control event.
524  */
525 static u8
527  u32 prev_snd_wnd, u32 prev_snd_una, u8 * is_dack)
528 {
529  /* Check if ack is duplicate. Per RFC 6675, ACKs that SACK new data are
530  * defined to be 'duplicate' */
531  *is_dack = tc->sack_sb.last_sacked_bytes
532  || tcp_ack_is_dupack (tc, b, prev_snd_wnd, prev_snd_una);
533 
534  return ((*is_dack || tcp_in_cong_recovery (tc)) && !tcp_is_lost_fin (tc));
535 }
536 
537 static u32
539 {
540  ASSERT (!pool_is_free_index (sb->holes, hole - sb->holes));
541  return hole - sb->holes;
542 }
543 
544 static u32
546 {
547  return hole->end - hole->start;
548 }
549 
552 {
553  if (index != TCP_INVALID_SACK_HOLE_INDEX)
554  return pool_elt_at_index (sb->holes, index);
555  return 0;
556 }
557 
560 {
561  if (hole->next != TCP_INVALID_SACK_HOLE_INDEX)
562  return pool_elt_at_index (sb->holes, hole->next);
563  return 0;
564 }
565 
568 {
569  if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX)
570  return pool_elt_at_index (sb->holes, hole->prev);
571  return 0;
572 }
573 
576 {
577  if (sb->head != TCP_INVALID_SACK_HOLE_INDEX)
578  return pool_elt_at_index (sb->holes, sb->head);
579  return 0;
580 }
581 
584 {
585  if (sb->tail != TCP_INVALID_SACK_HOLE_INDEX)
586  return pool_elt_at_index (sb->holes, sb->tail);
587  return 0;
588 }
589 
590 static void
592 {
593  sack_scoreboard_hole_t *next, *prev;
594 
595  if (hole->next != TCP_INVALID_SACK_HOLE_INDEX)
596  {
597  next = pool_elt_at_index (sb->holes, hole->next);
598  next->prev = hole->prev;
599  }
600  else
601  {
602  sb->tail = hole->prev;
603  }
604 
605  if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX)
606  {
607  prev = pool_elt_at_index (sb->holes, hole->prev);
608  prev->next = hole->next;
609  }
610  else
611  {
612  sb->head = hole->next;
613  }
614 
615  if (scoreboard_hole_index (sb, hole) == sb->cur_rxt_hole)
616  sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
617 
618  /* Poison the entry */
619  if (CLIB_DEBUG > 0)
620  memset (hole, 0xfe, sizeof (*hole));
621 
622  pool_put (sb->holes, hole);
623 }
624 
625 static sack_scoreboard_hole_t *
627  u32 start, u32 end)
628 {
629  sack_scoreboard_hole_t *hole, *next, *prev;
630  u32 hole_index;
631 
632  pool_get (sb->holes, hole);
633  memset (hole, 0, sizeof (*hole));
634 
635  hole->start = start;
636  hole->end = end;
637  hole_index = scoreboard_hole_index (sb, hole);
638 
639  prev = scoreboard_get_hole (sb, prev_index);
640  if (prev)
641  {
642  hole->prev = prev_index;
643  hole->next = prev->next;
644 
645  if ((next = scoreboard_next_hole (sb, hole)))
646  next->prev = hole_index;
647  else
648  sb->tail = hole_index;
649 
650  prev->next = hole_index;
651  }
652  else
653  {
654  sb->head = hole_index;
655  hole->prev = TCP_INVALID_SACK_HOLE_INDEX;
656  hole->next = TCP_INVALID_SACK_HOLE_INDEX;
657  }
658 
659  return hole;
660 }
661 
662 static void
664 {
665  sack_scoreboard_hole_t *hole, *prev;
666  u32 bytes = 0, blks = 0;
667 
668  sb->lost_bytes = 0;
669  sb->sacked_bytes = 0;
670  hole = scoreboard_last_hole (sb);
671  if (!hole)
672  return;
673 
674  if (seq_gt (sb->high_sacked, hole->end))
675  {
676  bytes = sb->high_sacked - hole->end;
677  blks = 1;
678  }
679 
680  while ((prev = scoreboard_prev_hole (sb, hole))
681  && (bytes < (TCP_DUPACK_THRESHOLD - 1) * tc->snd_mss
682  && blks < TCP_DUPACK_THRESHOLD))
683  {
684  bytes += hole->start - prev->end;
685  blks++;
686  hole = prev;
687  }
688 
689  while (hole)
690  {
691  sb->lost_bytes += scoreboard_hole_bytes (hole);
692  hole->is_lost = 1;
693  prev = hole;
694  hole = scoreboard_prev_hole (sb, hole);
695  if (hole)
696  bytes += prev->start - hole->end;
697  }
698  sb->sacked_bytes = bytes;
699 }
700 
701 /**
702  * Figure out the next hole to retransmit
703  *
704  * Follows logic proposed in RFC6675 Sec. 4, NextSeg()
705  */
708  sack_scoreboard_hole_t * start,
709  u8 have_sent_1_smss,
710  u8 * can_rescue, u8 * snd_limited)
711 {
712  sack_scoreboard_hole_t *hole = 0;
713 
714  hole = start ? start : scoreboard_first_hole (sb);
715  while (hole && seq_leq (hole->end, sb->high_rxt) && hole->is_lost)
716  hole = scoreboard_next_hole (sb, hole);
717 
718  /* Nothing, return */
719  if (!hole)
720  {
721  sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
722  return 0;
723  }
724 
725  /* Rule (1): if higher than rxt, less than high_sacked and lost */
726  if (hole->is_lost && seq_lt (hole->start, sb->high_sacked))
727  {
728  sb->cur_rxt_hole = scoreboard_hole_index (sb, hole);
729  }
730  else
731  {
732  /* Rule (2): output takes care of transmitting new data */
733  if (!have_sent_1_smss)
734  {
735  hole = 0;
736  sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
737  }
738  /* Rule (3): if hole not lost */
739  else if (seq_lt (hole->start, sb->high_sacked))
740  {
741  *snd_limited = 1;
742  sb->cur_rxt_hole = scoreboard_hole_index (sb, hole);
743  }
744  /* Rule (4): if hole beyond high_sacked */
745  else
746  {
747  ASSERT (seq_geq (hole->start, sb->high_sacked));
748  *snd_limited = 1;
749  *can_rescue = 1;
750  /* HighRxt MUST NOT be updated */
751  return 0;
752  }
753  }
754 
755  if (hole && seq_lt (sb->high_rxt, hole->start))
756  sb->high_rxt = hole->start;
757 
758  return hole;
759 }
760 
761 static void
763 {
765  hole = scoreboard_first_hole (sb);
766  if (hole)
767  {
768  seq = seq_gt (seq, hole->start) ? seq : hole->start;
769  sb->cur_rxt_hole = sb->head;
770  }
771  sb->high_rxt = seq;
772 }
773 
774 void
776 {
777  sb->head = TCP_INVALID_SACK_HOLE_INDEX;
778  sb->tail = TCP_INVALID_SACK_HOLE_INDEX;
779  sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
780 }
781 
782 void
784 {
786  while ((hole = scoreboard_first_hole (sb)))
787  {
788  scoreboard_remove_hole (sb, hole);
789  }
790  ASSERT (sb->head == sb->tail && sb->head == TCP_INVALID_SACK_HOLE_INDEX);
791  ASSERT (pool_elts (sb->holes) == 0);
792  sb->sacked_bytes = 0;
793  sb->last_sacked_bytes = 0;
794  sb->last_bytes_delivered = 0;
795  sb->snd_una_adv = 0;
796  sb->high_sacked = 0;
797  sb->high_rxt = 0;
798  sb->lost_bytes = 0;
799  sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
800 }
801 
802 /**
803  * Test that scoreboard is sane after recovery
804  *
805  * Returns 1 if scoreboard is empty or if first hole beyond
806  * snd_una.
807  */
808 static u8
810 {
812  hole = scoreboard_first_hole (&tc->sack_sb);
813  return (!hole || seq_geq (hole->start, tc->snd_una));
814 }
815 
816 void
818 {
819  sack_scoreboard_t *sb = &tc->sack_sb;
820  sack_block_t *blk, tmp;
821  sack_scoreboard_hole_t *hole, *next_hole, *last_hole;
822  u32 blk_index = 0, old_sacked_bytes, hole_index;
823  int i, j;
824 
825  sb->last_sacked_bytes = 0;
826  sb->last_bytes_delivered = 0;
827  sb->snd_una_adv = 0;
828 
829  if (!tcp_opts_sack (&tc->rcv_opts)
830  && sb->head == TCP_INVALID_SACK_HOLE_INDEX)
831  return;
832 
833  old_sacked_bytes = sb->sacked_bytes;
834 
835  /* Remove invalid blocks */
836  blk = tc->rcv_opts.sacks;
837  while (blk < vec_end (tc->rcv_opts.sacks))
838  {
839  if (seq_lt (blk->start, blk->end)
840  && seq_gt (blk->start, tc->snd_una)
841  && seq_gt (blk->start, ack) && seq_leq (blk->end, tc->snd_una_max))
842  {
843  blk++;
844  continue;
845  }
846  vec_del1 (tc->rcv_opts.sacks, blk - tc->rcv_opts.sacks);
847  }
848 
849  /* Add block for cumulative ack */
850  if (seq_gt (ack, tc->snd_una))
851  {
852  tmp.start = tc->snd_una;
853  tmp.end = ack;
854  vec_add1 (tc->rcv_opts.sacks, tmp);
855  }
856 
857  if (vec_len (tc->rcv_opts.sacks) == 0)
858  return;
859 
860  tcp_scoreboard_trace_add (tc, ack);
861 
862  /* Make sure blocks are ordered */
863  for (i = 0; i < vec_len (tc->rcv_opts.sacks); i++)
864  for (j = i + 1; j < vec_len (tc->rcv_opts.sacks); j++)
865  if (seq_lt (tc->rcv_opts.sacks[j].start, tc->rcv_opts.sacks[i].start))
866  {
867  tmp = tc->rcv_opts.sacks[i];
868  tc->rcv_opts.sacks[i] = tc->rcv_opts.sacks[j];
869  tc->rcv_opts.sacks[j] = tmp;
870  }
871 
872  if (sb->head == TCP_INVALID_SACK_HOLE_INDEX)
873  {
874  /* If no holes, insert the first that covers all outstanding bytes */
876  tc->snd_una, tc->snd_una_max);
877  sb->tail = scoreboard_hole_index (sb, last_hole);
878  tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1];
879  sb->high_sacked = tmp.end;
880  }
881  else
882  {
883  /* If we have holes but snd_una_max is beyond the last hole, update
884  * last hole end */
885  tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1];
886  last_hole = scoreboard_last_hole (sb);
887  if (seq_gt (tc->snd_una_max, last_hole->end))
888  {
889  if (seq_geq (last_hole->start, sb->high_sacked))
890  {
891  last_hole->end = tc->snd_una_max;
892  }
893  /* New hole after high sacked block */
894  else if (seq_lt (sb->high_sacked, tc->snd_una_max))
895  {
896  scoreboard_insert_hole (sb, sb->tail, sb->high_sacked,
897  tc->snd_una_max);
898  }
899  }
900  /* Keep track of max byte sacked for when the last hole
901  * is acked */
902  if (seq_gt (tmp.end, sb->high_sacked))
903  sb->high_sacked = tmp.end;
904  }
905 
906  /* Walk the holes with the SACK blocks */
907  hole = pool_elt_at_index (sb->holes, sb->head);
908  while (hole && blk_index < vec_len (tc->rcv_opts.sacks))
909  {
910  blk = &tc->rcv_opts.sacks[blk_index];
911  if (seq_leq (blk->start, hole->start))
912  {
913  /* Block covers hole. Remove hole */
914  if (seq_geq (blk->end, hole->end))
915  {
916  next_hole = scoreboard_next_hole (sb, hole);
917 
918  /* Byte accounting: snd_una needs to be advanced */
919  if (blk->end == ack)
920  {
921  if (next_hole)
922  {
923  if (seq_lt (ack, next_hole->start))
924  sb->snd_una_adv = next_hole->start - ack;
925  sb->last_bytes_delivered +=
926  next_hole->start - hole->end;
927  }
928  else
929  {
930  ASSERT (seq_geq (sb->high_sacked, ack));
931  sb->snd_una_adv = sb->high_sacked - ack;
932  sb->last_bytes_delivered += sb->high_sacked - hole->end;
933  }
934  }
935 
936  scoreboard_remove_hole (sb, hole);
937  hole = next_hole;
938  }
939  /* Partial 'head' overlap */
940  else
941  {
942  if (seq_gt (blk->end, hole->start))
943  {
944  hole->start = blk->end;
945  }
946  blk_index++;
947  }
948  }
949  else
950  {
951  /* Hole must be split */
952  if (seq_lt (blk->end, hole->end))
953  {
954  hole_index = scoreboard_hole_index (sb, hole);
955  next_hole = scoreboard_insert_hole (sb, hole_index, blk->end,
956  hole->end);
957 
958  /* Pool might've moved */
959  hole = scoreboard_get_hole (sb, hole_index);
960  hole->end = blk->start;
961  blk_index++;
962  ASSERT (hole->next == scoreboard_hole_index (sb, next_hole));
963  }
964  else if (seq_lt (blk->start, hole->end))
965  {
966  hole->end = blk->start;
967  }
968  hole = scoreboard_next_hole (sb, hole);
969  }
970  }
971 
972  scoreboard_update_bytes (tc, sb);
973  sb->last_sacked_bytes = sb->sacked_bytes
974  - (old_sacked_bytes - sb->last_bytes_delivered);
975  ASSERT (sb->last_sacked_bytes <= sb->sacked_bytes || tcp_in_recovery (tc));
976  ASSERT (sb->sacked_bytes == 0
977  || sb->sacked_bytes < tc->snd_una_max - seq_max (tc->snd_una, ack));
978  ASSERT (sb->last_sacked_bytes + sb->lost_bytes <= tc->snd_una_max
979  - seq_max (tc->snd_una, ack));
981  || sb->holes[sb->head].start == ack + sb->snd_una_adv);
982  TCP_EVT_DBG (TCP_EVT_CC_SCOREBOARD, tc);
983 }
984 
985 /**
986  * Try to update snd_wnd based on feedback received from peer.
987  *
988  * If successful, and new window is 'effectively' 0, activate persist
989  * timer.
990  */
991 static void
992 tcp_update_snd_wnd (tcp_connection_t * tc, u32 seq, u32 ack, u32 snd_wnd)
993 {
994  /* If (SND.WL1 < SEG.SEQ or (SND.WL1 = SEG.SEQ and SND.WL2 =< SEG.ACK)), set
995  * SND.WND <- SEG.WND, set SND.WL1 <- SEG.SEQ, and set SND.WL2 <- SEG.ACK */
996  if (seq_lt (tc->snd_wl1, seq)
997  || (tc->snd_wl1 == seq && seq_leq (tc->snd_wl2, ack)))
998  {
999  tc->snd_wnd = snd_wnd;
1000  tc->snd_wl1 = seq;
1001  tc->snd_wl2 = ack;
1002  TCP_EVT_DBG (TCP_EVT_SND_WND, tc);
1003 
1004  if (tc->snd_wnd < tc->snd_mss)
1005  {
1006  /* Set persist timer if not set and we just got 0 wnd */
1007  if (!tcp_timer_is_active (tc, TCP_TIMER_PERSIST)
1008  && !tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT))
1009  tcp_persist_timer_set (tc);
1010  }
1011  else
1012  {
1014  if (!tcp_in_recovery (tc) && tc->rto_boff > 0)
1015  {
1016  tc->rto_boff = 0;
1017  tcp_update_rto (tc);
1018  }
1019  }
1020  }
1021 }
1022 
1023 /**
1024  * Init loss recovery/fast recovery.
1025  *
1026  * Triggered by dup acks as opposed to timer timeout. Note that cwnd is
1027  * updated in @ref tcp_cc_handle_event after fast retransmit
1028  */
1029 void
1031 {
1032  tcp_fastrecovery_on (tc);
1033  tc->snd_congestion = tc->snd_una_max;
1034  tc->cwnd_acc_bytes = 0;
1035  tc->cc_algo->congestion (tc);
1036  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 4);
1037 }
1038 
1039 static void
1041 {
1042  tc->rto_boff = 0;
1043  tcp_update_rto (tc);
1044  tc->snd_rxt_ts = 0;
1045  tc->snd_nxt = tc->snd_una_max;
1046  tcp_recovery_off (tc);
1047  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3);
1048 }
1049 
1050 void
1052 {
1053  tc->cc_algo->recovered (tc);
1054  tc->snd_rxt_bytes = 0;
1055  tc->rcv_dupacks = 0;
1056  tc->snd_nxt = tc->snd_una_max;
1057  tcp_fastrecovery_off (tc);
1059  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3);
1060 }
1061 
1062 static void
1064 {
1065  tc->cwnd = tc->prev_cwnd;
1066  tc->ssthresh = tc->prev_ssthresh;
1067  tc->snd_nxt = tc->snd_una_max;
1068  tc->rcv_dupacks = 0;
1069  if (tcp_in_recovery (tc))
1070  tcp_cc_recovery_exit (tc);
1071  ASSERT (tc->rto_boff == 0);
1072  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 5);
1073  /* TODO extend for fastrecovery */
1074 }
1075 
1076 static u8
1078 {
1079  return (tcp_in_recovery (tc) && tc->rto_boff == 1
1080  && tc->snd_rxt_ts
1081  && tcp_opts_tstamp (&tc->rcv_opts)
1082  && timestamp_lt (tc->rcv_opts.tsecr, tc->snd_rxt_ts));
1083 }
1084 
1085 static int
1087 {
1090  {
1092  return 1;
1093  }
1094 
1095  if (tcp_in_recovery (tc))
1096  tcp_cc_recovery_exit (tc);
1097  else if (tcp_in_fastrecovery (tc))
1099 
1100  ASSERT (tc->rto_boff == 0);
1101  ASSERT (!tcp_in_cong_recovery (tc));
1103  return 0;
1104 }
1105 
1106 static void
1108 {
1110 
1111  /* Congestion avoidance */
1112  tc->cc_algo->rcv_ack (tc);
1113  tc->tsecr_last_ack = tc->rcv_opts.tsecr;
1114 
1115  /* If a cumulative ack, make sure dupacks is 0 */
1116  tc->rcv_dupacks = 0;
1117 
1118  /* When dupacks hits the threshold we only enter fast retransmit if
1119  * cumulative ack covers more than snd_congestion. Should snd_una
1120  * wrap this test may fail under otherwise valid circumstances.
1121  * Therefore, proactively update snd_congestion when wrap detected. */
1122  if (PREDICT_FALSE
1123  (seq_leq (tc->snd_congestion, tc->snd_una - tc->bytes_acked)
1124  && seq_gt (tc->snd_congestion, tc->snd_una)))
1125  tc->snd_congestion = tc->snd_una - 1;
1126 }
1127 
1128 static u8
1130 {
1131  return (TCP_DUPACK_THRESHOLD - 1) * tc->snd_mss < tc->sack_sb.sacked_bytes;
1132 }
1133 
1134 static u8
1136 {
1137  return (tc->rcv_dupacks == TCP_DUPACK_THRESHOLD
1138  || tcp_should_fastrecover_sack (tc));
1139 }
1140 
1141 /**
1142  * One function to rule them all ... and in the darkness bind them
1143  */
1144 static void
1146 {
1147  u32 rxt_delivered;
1148 
1149  if (tcp_in_fastrecovery (tc) && tcp_opts_sack_permitted (&tc->rcv_opts))
1150  {
1151  if (tc->bytes_acked)
1152  goto partial_ack;
1153  tcp_fast_retransmit (tc);
1154  return;
1155  }
1156  /*
1157  * Duplicate ACK. Check if we should enter fast recovery, or if already in
1158  * it account for the bytes that left the network.
1159  */
1160  else if (is_dack && !tcp_in_recovery (tc))
1161  {
1162  TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc, 1);
1163  ASSERT (tc->snd_una != tc->snd_una_max
1164  || tc->sack_sb.last_sacked_bytes);
1165 
1166  tc->rcv_dupacks++;
1167 
1168  /* Pure duplicate ack. If some data got acked, it's handled lower */
1169  if (tc->rcv_dupacks > TCP_DUPACK_THRESHOLD && !tc->bytes_acked)
1170  {
1171  ASSERT (tcp_in_fastrecovery (tc));
1172  tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK);
1173  return;
1174  }
1175  else if (tcp_should_fastrecover (tc))
1176  {
1177  ASSERT (!tcp_in_fastrecovery (tc));
1178 
1179  /* If of of the two conditions lower hold, reset dupacks because
1180  * we're probably after timeout (RFC6582 heuristics).
1181  * If Cumulative ack does not cover more than congestion threshold,
1182  * and:
1183  * 1) The following doesn't hold: The congestion window is greater
1184  * than SMSS bytes and the difference between highest_ack
1185  * and prev_highest_ack is at most 4*SMSS bytes
1186  * 2) Echoed timestamp in the last non-dup ack does not equal the
1187  * stored timestamp
1188  */
1189  if (seq_leq (tc->snd_una, tc->snd_congestion)
1190  && ((!(tc->cwnd > tc->snd_mss
1191  && tc->bytes_acked <= 4 * tc->snd_mss))
1192  || (tc->rcv_opts.tsecr != tc->tsecr_last_ack)))
1193  {
1194  tc->rcv_dupacks = 0;
1195  return;
1196  }
1197 
1199  tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK);
1200 
1201  /* The first segment MUST be retransmitted */
1203 
1204  /* Post retransmit update cwnd to ssthresh and account for the
1205  * three segments that have left the network and should've been
1206  * buffered at the receiver XXX */
1207  tc->cwnd = tc->ssthresh + tc->rcv_dupacks * tc->snd_mss;
1208  ASSERT (tc->cwnd >= tc->snd_mss);
1209 
1210  /* If cwnd allows, send more data */
1211  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1212  {
1213  scoreboard_init_high_rxt (&tc->sack_sb,
1214  tc->snd_una + tc->snd_mss);
1216  }
1217  else
1218  {
1220  }
1221  return;
1222  }
1223  else if (!tc->bytes_acked
1224  || (tc->bytes_acked && !tcp_in_cong_recovery (tc)))
1225  {
1226  tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK);
1227  return;
1228  }
1229  else
1230  goto partial_ack;
1231  }
1232 
1233  if (!tc->bytes_acked)
1234  return;
1235 
1236 partial_ack:
1237  TCP_EVT_DBG (TCP_EVT_CC_PACK, tc);
1238 
1239  /*
1240  * Legitimate ACK. 1) See if we can exit recovery
1241  */
1242  /* XXX limit this only to first partial ack? */
1244 
1245  if (seq_geq (tc->snd_una, tc->snd_congestion))
1246  {
1247  /* If spurious return, we've already updated everything */
1248  if (tcp_cc_recover (tc))
1249  {
1250  tc->tsecr_last_ack = tc->rcv_opts.tsecr;
1251  return;
1252  }
1253 
1254  tc->snd_nxt = tc->snd_una_max;
1255 
1256  /* Treat as congestion avoidance ack */
1257  tc->cc_algo->rcv_ack (tc);
1258  tc->tsecr_last_ack = tc->rcv_opts.tsecr;
1259  return;
1260  }
1261 
1262  /*
1263  * Legitimate ACK. 2) If PARTIAL ACK try to retransmit
1264  */
1265 
1266  /* RFC6675: If the incoming ACK is a cumulative acknowledgment,
1267  * reset dupacks to 0. Also needed if in congestion recovery */
1268  tc->rcv_dupacks = 0;
1269 
1270  /* Post RTO timeout don't try anything fancy */
1271  if (tcp_in_recovery (tc))
1272  {
1273  tc->cc_algo->rcv_ack (tc);
1274  tc->tsecr_last_ack = tc->rcv_opts.tsecr;
1275  return;
1276  }
1277 
1278  /* Remove retransmitted bytes that have been delivered */
1279  ASSERT (tc->bytes_acked + tc->sack_sb.snd_una_adv
1280  >= tc->sack_sb.last_bytes_delivered
1281  || (tc->flags & TCP_CONN_FINSNT));
1282 
1283  if (seq_lt (tc->snd_una, tc->sack_sb.high_rxt))
1284  {
1285  /* If we have sacks and we haven't gotten an ack beyond high_rxt,
1286  * remove sacked bytes delivered */
1287  rxt_delivered = tc->bytes_acked + tc->sack_sb.snd_una_adv
1288  - tc->sack_sb.last_bytes_delivered;
1289  ASSERT (tc->snd_rxt_bytes >= rxt_delivered);
1290  tc->snd_rxt_bytes -= rxt_delivered;
1291  }
1292  else
1293  {
1294  /* Either all retransmitted holes have been acked, or we're
1295  * "in the blind" and retransmitting segment by segment */
1296  tc->snd_rxt_bytes = 0;
1297  }
1298 
1299  tc->cc_algo->rcv_cong_ack (tc, TCP_CC_PARTIALACK);
1300 
1301  /*
1302  * Since this was a partial ack, try to retransmit some more data
1303  */
1304  tcp_fast_retransmit (tc);
1305 }
1306 
1307 /**
1308  * Process incoming ACK
1309  */
1310 static int
1312  tcp_header_t * th, u32 * next, u32 * error)
1313 {
1314  u32 prev_snd_wnd, prev_snd_una;
1315  u8 is_dack;
1316 
1317  TCP_EVT_DBG (TCP_EVT_CC_STAT, tc);
1318 
1319  /* If the ACK acks something not yet sent (SEG.ACK > SND.NXT) */
1320  if (PREDICT_FALSE (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt)))
1321  {
1322  /* When we entered recovery, we reset snd_nxt to snd_una. Seems peer
1323  * still has the data so accept the ack */
1324  if (tcp_in_recovery (tc)
1325  && seq_leq (vnet_buffer (b)->tcp.ack_number, tc->snd_congestion)
1326  && seq_geq (vnet_buffer (b)->tcp.ack_number, tc->snd_una))
1327  {
1328  tc->snd_una_max = tc->snd_nxt = vnet_buffer (b)->tcp.ack_number;
1329  goto process_ack;
1330  }
1331 
1332  /* If we have outstanding data and this is within the window, accept it,
1333  * probably retransmit has timed out. Otherwise ACK segment and then
1334  * drop it */
1335  if (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_una_max))
1336  {
1337  tcp_make_ack (tc, b);
1338  *next = tcp_next_output (tc->c_is_ip4);
1339  *error = TCP_ERROR_ACK_INVALID;
1340  TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 0,
1341  vnet_buffer (b)->tcp.ack_number);
1342  return -1;
1343  }
1344 
1345  TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 2,
1346  vnet_buffer (b)->tcp.ack_number);
1347 
1348  tc->snd_nxt = vnet_buffer (b)->tcp.ack_number;
1349  *error = TCP_ERROR_ACK_FUTURE;
1350  }
1351 
1352  /* If old ACK, probably it's an old dupack */
1353  if (PREDICT_FALSE (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una)))
1354  {
1355  *error = TCP_ERROR_ACK_OLD;
1356  TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 1,
1357  vnet_buffer (b)->tcp.ack_number);
1358  if (tcp_in_fastrecovery (tc) && tc->rcv_dupacks == TCP_DUPACK_THRESHOLD)
1359  tcp_cc_handle_event (tc, 1);
1360  /* Don't drop yet */
1361  return 0;
1362  }
1363 
1364  /*
1365  * Looks okay, process feedback
1366  */
1367 process_ack:
1368  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1369  tcp_rcv_sacks (tc, vnet_buffer (b)->tcp.ack_number);
1370 
1371  prev_snd_wnd = tc->snd_wnd;
1372  prev_snd_una = tc->snd_una;
1373  tcp_update_snd_wnd (tc, vnet_buffer (b)->tcp.seq_number,
1374  vnet_buffer (b)->tcp.ack_number,
1375  clib_net_to_host_u16 (th->window) << tc->snd_wscale);
1376  tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una;
1377  tc->snd_una = vnet_buffer (b)->tcp.ack_number + tc->sack_sb.snd_una_adv;
1378  tcp_validate_txf_size (tc, tc->bytes_acked);
1379 
1380  if (tc->bytes_acked)
1381  tcp_dequeue_acked (tc, vnet_buffer (b)->tcp.ack_number);
1382 
1383  TCP_EVT_DBG (TCP_EVT_ACK_RCVD, tc);
1384 
1385  /*
1386  * Check if we have congestion event
1387  */
1388 
1389  if (tcp_ack_is_cc_event (tc, b, prev_snd_wnd, prev_snd_una, &is_dack))
1390  {
1391  tcp_cc_handle_event (tc, is_dack);
1392  if (!tcp_in_cong_recovery (tc))
1393  return 0;
1394  *error = TCP_ERROR_ACK_DUP;
1395  return vnet_buffer (b)->tcp.data_len ? 0 : -1;
1396  }
1397 
1398  /*
1399  * Update congestion control (slow start/congestion avoidance)
1400  */
1401  tcp_cc_update (tc, b);
1402  *error = TCP_ERROR_ACK_OK;
1403  return 0;
1404 }
1405 
1406 static u8
1408 {
1409  int i;
1410  for (i = 1; i < vec_len (sacks); i++)
1411  {
1412  if (sacks[i - 1].end == sacks[i].start)
1413  return 0;
1414  }
1415  return 1;
1416 }
1417 
1418 /**
1419  * Build SACK list as per RFC2018.
1420  *
1421  * Makes sure the first block contains the segment that generated the current
1422  * ACK and the following ones are the ones most recently reported in SACK
1423  * blocks.
1424  *
1425  * @param tc TCP connection for which the SACK list is updated
1426  * @param start Start sequence number of the newest SACK block
1427  * @param end End sequence of the newest SACK block
1428  */
1429 void
1431 {
1432  sack_block_t *new_list = 0, *block = 0;
1433  int i;
1434 
1435  /* If the first segment is ooo add it to the list. Last write might've moved
1436  * rcv_nxt over the first segment. */
1437  if (seq_lt (tc->rcv_nxt, start))
1438  {
1439  vec_add2 (new_list, block, 1);
1440  block->start = start;
1441  block->end = end;
1442  }
1443 
1444  /* Find the blocks still worth keeping. */
1445  for (i = 0; i < vec_len (tc->snd_sacks); i++)
1446  {
1447  /* Discard if rcv_nxt advanced beyond current block */
1448  if (seq_leq (tc->snd_sacks[i].start, tc->rcv_nxt))
1449  continue;
1450 
1451  /* Merge or drop if segment overlapped by the new segment */
1452  if (block && (seq_geq (tc->snd_sacks[i].end, new_list[0].start)
1453  && seq_leq (tc->snd_sacks[i].start, new_list[0].end)))
1454  {
1455  if (seq_lt (tc->snd_sacks[i].start, new_list[0].start))
1456  new_list[0].start = tc->snd_sacks[i].start;
1457  if (seq_lt (new_list[0].end, tc->snd_sacks[i].end))
1458  new_list[0].end = tc->snd_sacks[i].end;
1459  continue;
1460  }
1461 
1462  /* Save to new SACK list if we have space. */
1463  if (vec_len (new_list) < TCP_MAX_SACK_BLOCKS)
1464  {
1465  vec_add1 (new_list, tc->snd_sacks[i]);
1466  }
1467  else
1468  {
1469  clib_warning ("sack discarded");
1470  }
1471  }
1472 
1473  ASSERT (vec_len (new_list) <= TCP_MAX_SACK_BLOCKS);
1474 
1475  /* Replace old vector with new one */
1476  vec_free (tc->snd_sacks);
1477  tc->snd_sacks = new_list;
1478 
1479  /* Segments should not 'touch' */
1480  ASSERT (tcp_sack_vector_is_sane (tc->snd_sacks));
1481 }
1482 
1483 u32
1485 {
1486  u32 bytes = 0, i;
1487  for (i = 0; i < vec_len (tc->snd_sacks); i++)
1488  bytes += tc->snd_sacks[i].end - tc->snd_sacks[i].start;
1489  return bytes;
1490 }
1491 
1492 /** Enqueue data for delivery to application */
1493 static int
1495  u16 data_len)
1496 {
1497  int written, error = TCP_ERROR_ENQUEUED;
1498 
1499  ASSERT (seq_geq (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt));
1500  ASSERT (data_len);
1501  written = session_enqueue_stream_connection (&tc->connection, b, 0,
1502  1 /* queue event */ , 1);
1503 
1504  TCP_EVT_DBG (TCP_EVT_INPUT, tc, 0, data_len, written);
1505 
1506  /* Update rcv_nxt */
1507  if (PREDICT_TRUE (written == data_len))
1508  {
1509  tc->rcv_nxt += written;
1510  }
1511  /* If more data written than expected, account for out-of-order bytes. */
1512  else if (written > data_len)
1513  {
1514  tc->rcv_nxt += written;
1515 
1516  /* Send ACK confirming the update */
1517  tc->flags |= TCP_CONN_SNDACK;
1518  TCP_EVT_DBG (TCP_EVT_CC_INPUT, tc, data_len, written);
1519  }
1520  else if (written > 0)
1521  {
1522  /* We've written something but FIFO is probably full now */
1523  tc->rcv_nxt += written;
1524 
1525  /* Depending on how fast the app is, all remaining buffers in burst will
1526  * not be enqueued. Inform peer */
1527  tc->flags |= TCP_CONN_SNDACK;
1528 
1529  error = TCP_ERROR_PARTIALLY_ENQUEUED;
1530  }
1531  else
1532  {
1533  tc->flags |= TCP_CONN_SNDACK;
1534  return TCP_ERROR_FIFO_FULL;
1535  }
1536 
1537  /* Update SACK list if need be */
1538  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1539  {
1540  /* Remove SACK blocks that have been delivered */
1541  tcp_update_sack_list (tc, tc->rcv_nxt, tc->rcv_nxt);
1542  }
1543 
1544  return error;
1545 }
1546 
1547 /** Enqueue out-of-order data */
1548 static int
1550  u16 data_len)
1551 {
1552  stream_session_t *s0;
1553  int rv, offset;
1554 
1555  ASSERT (seq_gt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt));
1556  ASSERT (data_len);
1557 
1558  /* Enqueue out-of-order data with relative offset */
1559  rv = session_enqueue_stream_connection (&tc->connection, b,
1560  vnet_buffer (b)->tcp.seq_number -
1561  tc->rcv_nxt, 0 /* queue event */ ,
1562  0);
1563 
1564  /* Nothing written */
1565  if (rv)
1566  {
1567  TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, 0);
1568  return TCP_ERROR_FIFO_FULL;
1569  }
1570 
1571  TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, data_len);
1572 
1573  /* Update SACK list if in use */
1574  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1575  {
1576  ooo_segment_t *newest;
1577  u32 start, end;
1578 
1579  s0 = session_get (tc->c_s_index, tc->c_thread_index);
1580 
1581  /* Get the newest segment from the fifo */
1582  newest = svm_fifo_newest_ooo_segment (s0->server_rx_fifo);
1583  if (newest)
1584  {
1585  offset = ooo_segment_offset (s0->server_rx_fifo, newest);
1586  ASSERT (offset <= vnet_buffer (b)->tcp.seq_number - tc->rcv_nxt);
1587  start = tc->rcv_nxt + offset;
1588  end = start + ooo_segment_length (s0->server_rx_fifo, newest);
1589  tcp_update_sack_list (tc, start, end);
1590  svm_fifo_newest_ooo_segment_reset (s0->server_rx_fifo);
1591  TCP_EVT_DBG (TCP_EVT_CC_SACKS, tc);
1592  }
1593  }
1594 
1595  return TCP_ERROR_ENQUEUED_OOO;
1596 }
1597 
1598 /**
1599  * Check if ACK could be delayed. If ack can be delayed, it should return
1600  * true for a full frame. If we're always acking return 0.
1601  */
1602 always_inline int
1604 {
1605  /* Send ack if ... */
1606  if (TCP_ALWAYS_ACK
1607  /* just sent a rcv wnd 0 */
1608  || (tc->flags & TCP_CONN_SENT_RCV_WND0) != 0
1609  /* constrained to send ack */
1610  || (tc->flags & TCP_CONN_SNDACK) != 0
1611  /* we're almost out of tx wnd */
1612  || tcp_available_cc_snd_space (tc) < 4 * tc->snd_mss)
1613  return 0;
1614 
1615  return 1;
1616 }
1617 
1618 static int
1620 {
1621  u32 discard, first = b->current_length;
1623 
1624  /* Handle multi-buffer segments */
1625  if (n_bytes_to_drop > b->current_length)
1626  {
1627  if (!(b->flags & VLIB_BUFFER_NEXT_PRESENT))
1628  return -1;
1629  do
1630  {
1631  discard = clib_min (n_bytes_to_drop, b->current_length);
1632  vlib_buffer_advance (b, discard);
1633  b = vlib_get_buffer (vm, b->next_buffer);
1634  n_bytes_to_drop -= discard;
1635  }
1636  while (n_bytes_to_drop);
1637  if (n_bytes_to_drop > first)
1638  b->total_length_not_including_first_buffer -= n_bytes_to_drop - first;
1639  }
1640  else
1641  vlib_buffer_advance (b, n_bytes_to_drop);
1642  vnet_buffer (b)->tcp.data_len -= n_bytes_to_drop;
1643  return 0;
1644 }
1645 
1646 /**
1647  * Receive buffer for connection and handle acks
1648  *
1649  * It handles both in order or out-of-order data.
1650  */
1651 static int
1653 {
1654  u32 error, n_bytes_to_drop, n_data_bytes;
1655 
1656  vlib_buffer_advance (b, vnet_buffer (b)->tcp.data_offset);
1657  n_data_bytes = vnet_buffer (b)->tcp.data_len;
1658  ASSERT (n_data_bytes);
1659 
1660  /* Handle out-of-order data */
1661  if (PREDICT_FALSE (vnet_buffer (b)->tcp.seq_number != tc->rcv_nxt))
1662  {
1663  /* Old sequence numbers allowed through because they overlapped
1664  * the rx window */
1665  if (seq_lt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt))
1666  {
1667  /* Completely in the past (possible retransmit). Ack
1668  * retransmissions since we may not have any data to send */
1669  if (seq_leq (vnet_buffer (b)->tcp.seq_end, tc->rcv_nxt))
1670  {
1671  tcp_make_ack (tc, b);
1672  error = TCP_ERROR_SEGMENT_OLD;
1673  *next0 = tcp_next_output (tc->c_is_ip4);
1674  goto done;
1675  }
1676 
1677  /* Chop off the bytes in the past and see if what is left
1678  * can be enqueued in order */
1679  n_bytes_to_drop = tc->rcv_nxt - vnet_buffer (b)->tcp.seq_number;
1680  n_data_bytes -= n_bytes_to_drop;
1681  vnet_buffer (b)->tcp.seq_number = tc->rcv_nxt;
1682  if (tcp_buffer_discard_bytes (b, n_bytes_to_drop))
1683  {
1684  error = TCP_ERROR_SEGMENT_OLD;
1685  *next0 = tcp_next_drop (tc->c_is_ip4);
1686  goto done;
1687  }
1688  goto in_order;
1689  }
1690 
1691  /* RFC2581: Enqueue and send DUPACK for fast retransmit */
1692  error = tcp_session_enqueue_ooo (tc, b, n_data_bytes);
1693  *next0 = tcp_next_output (tc->c_is_ip4);
1694  tcp_make_ack (tc, b);
1695  vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_DUPACK;
1696  TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc, vnet_buffer (b)->tcp);
1697  goto done;
1698  }
1699 
1700 in_order:
1701 
1702  /* In order data, enqueue. Fifo figures out by itself if any out-of-order
1703  * segments can be enqueued after fifo tail offset changes. */
1704  error = tcp_session_enqueue_data (tc, b, n_data_bytes);
1705  if (tcp_can_delack (tc))
1706  {
1707  *next0 = tcp_next_drop (tc->c_is_ip4);
1708  if (!tcp_timer_is_active (tc, TCP_TIMER_DELACK))
1709  tcp_timer_set (tc, TCP_TIMER_DELACK, TCP_DELACK_TIME);
1710  goto done;
1711  }
1712 
1713  *next0 = tcp_next_output (tc->c_is_ip4);
1714  tcp_make_ack (tc, b);
1715 
1716 done:
1717  return error;
1718 }
1719 
1720 typedef struct
1721 {
1724 } tcp_rx_trace_t;
1725 
1726 static u8 *
1727 format_tcp_rx_trace (u8 * s, va_list * args)
1728 {
1729  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1730  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1731  tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *);
1732  u32 indent = format_get_indent (s);
1733 
1734  s = format (s, "%U\n%U%U",
1735  format_tcp_header, &t->tcp_header, 128,
1736  format_white_space, indent,
1738 
1739  return s;
1740 }
1741 
1742 static u8 *
1743 format_tcp_rx_trace_short (u8 * s, va_list * args)
1744 {
1745  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1746  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1747  tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *);
1748 
1749  s = format (s, "%d -> %d (%U)",
1750  clib_net_to_host_u16 (t->tcp_header.dst_port),
1751  clib_net_to_host_u16 (t->tcp_header.src_port), format_tcp_state,
1752  t->tcp_connection.state);
1753 
1754  return s;
1755 }
1756 
1757 static void
1759  tcp_header_t * th0, vlib_buffer_t * b0, u8 is_ip4)
1760 {
1761  if (tc0)
1762  {
1763  clib_memcpy (&t0->tcp_connection, tc0, sizeof (t0->tcp_connection));
1764  }
1765  else
1766  {
1767  th0 = tcp_buffer_hdr (b0);
1768  }
1769  clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header));
1770 }
1771 
1772 static void
1774  vlib_frame_t * frame, u8 is_ip4)
1775 {
1776  u32 *from, n_left;
1777 
1778  n_left = frame->n_vectors;
1779  from = vlib_frame_vector_args (frame);
1780 
1781  while (n_left >= 1)
1782  {
1783  tcp_connection_t *tc0;
1784  tcp_rx_trace_t *t0;
1785  tcp_header_t *th0;
1786  vlib_buffer_t *b0;
1787  u32 bi0;
1788 
1789  bi0 = from[0];
1790  b0 = vlib_get_buffer (vm, bi0);
1791 
1792  if (b0->flags & VLIB_BUFFER_IS_TRACED)
1793  {
1794  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
1795  tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
1796  vm->thread_index);
1797  th0 = tcp_buffer_hdr (b0);
1798  tcp_set_rx_trace_data (t0, tc0, th0, b0, is_ip4);
1799  }
1800 
1801  from += 1;
1802  n_left -= 1;
1803  }
1804 }
1805 
1806 always_inline void
1807 tcp_node_inc_counter_i (vlib_main_t * vm, u32 tcp4_node, u32 tcp6_node,
1808  u8 is_ip4, u32 evt, u32 val)
1809 {
1810  if (is_ip4)
1811  vlib_node_increment_counter (vm, tcp4_node, evt, val);
1812  else
1813  vlib_node_increment_counter (vm, tcp6_node, evt, val);
1814 }
1815 
1816 #define tcp_maybe_inc_counter(node_id, err, count) \
1817 { \
1818  if (next0 != tcp_next_drop (is_ip4)) \
1819  tcp_node_inc_counter_i (vm, tcp4_##node_id##_node.index, \
1820  tcp6_##node_id##_node.index, is_ip4, err, \
1821  1); \
1822 }
1823 #define tcp_inc_counter(node_id, err, count) \
1824  tcp_node_inc_counter_i (vm, tcp4_##node_id##_node.index, \
1825  tcp6_##node_id##_node.index, is_ip4, \
1826  err, count)
1827 #define tcp_maybe_inc_err_counter(cnts, err) \
1828 { \
1829  cnts[err] += (next0 != tcp_next_drop (is_ip4)); \
1830 }
1831 #define tcp_inc_err_counter(cnts, err, val) \
1832 { \
1833  cnts[err] += val; \
1834 }
1835 #define tcp_store_err_counters(node_id, cnts) \
1836 { \
1837  int i; \
1838  for (i = 0; i < TCP_N_ERROR; i++) \
1839  if (cnts[i]) \
1840  tcp_inc_counter(node_id, i, cnts[i]); \
1841 }
1842 
1843 
1846  vlib_frame_t * frame, int is_ip4)
1847 {
1848  u32 thread_index = vm->thread_index, errors = 0;
1849  u32 n_left_from, next_index, *from, *to_next;
1850  u16 err_counters[TCP_N_ERROR] = { 0 };
1851  u8 is_fin = 0;
1852 
1853  if (node->flags & VLIB_NODE_FLAG_TRACE)
1854  tcp_established_trace_frame (vm, node, frame, is_ip4);
1855 
1856  from = vlib_frame_vector_args (frame);
1857  n_left_from = frame->n_vectors;
1858  next_index = node->cached_next_index;
1859 
1860  while (n_left_from > 0)
1861  {
1862  u32 n_left_to_next;
1863 
1864  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
1865  while (n_left_from > 0 && n_left_to_next > 0)
1866  {
1867  u32 bi0;
1868  vlib_buffer_t *b0;
1869  tcp_header_t *th0 = 0;
1870  tcp_connection_t *tc0;
1871  u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_ACK_OK;
1872 
1873  if (n_left_from > 1)
1874  {
1875  vlib_buffer_t *pb;
1876  pb = vlib_get_buffer (vm, from[1]);
1877  vlib_prefetch_buffer_header (pb, LOAD);
1878  CLIB_PREFETCH (pb->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
1879  }
1880 
1881  bi0 = from[0];
1882  to_next[0] = bi0;
1883  from += 1;
1884  to_next += 1;
1885  n_left_from -= 1;
1886  n_left_to_next -= 1;
1887 
1888  b0 = vlib_get_buffer (vm, bi0);
1889  tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
1890  thread_index);
1891 
1892  if (PREDICT_FALSE (tc0 == 0))
1893  {
1894  error0 = TCP_ERROR_INVALID_CONNECTION;
1895  goto done;
1896  }
1897 
1898  th0 = tcp_buffer_hdr (b0);
1899  /* N.B. buffer is rewritten if segment is ooo. Thus, th0 becomes a
1900  * dangling reference. */
1901  is_fin = tcp_is_fin (th0);
1902 
1903  /* SYNs, FINs and data consume sequence numbers */
1904  vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number
1905  + tcp_is_syn (th0) + is_fin + vnet_buffer (b0)->tcp.data_len;
1906 
1907  /* TODO header prediction fast path */
1908 
1909  /* 1-4: check SEQ, RST, SYN */
1910  if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, th0, &next0,
1911  &error0)))
1912  {
1913  tcp_maybe_inc_err_counter (err_counters, error0);
1914  TCP_EVT_DBG (TCP_EVT_SEG_INVALID, tc0, vnet_buffer (b0)->tcp);
1915  goto done;
1916  }
1917 
1918  /* 5: check the ACK field */
1919  if (PREDICT_FALSE (tcp_rcv_ack (tc0, b0, th0, &next0, &error0)))
1920  {
1921  tcp_maybe_inc_err_counter (err_counters, error0);
1922  goto done;
1923  }
1924 
1925  /* 6: check the URG bit TODO */
1926 
1927  /* 7: process the segment text */
1928  if (vnet_buffer (b0)->tcp.data_len)
1929  {
1930  error0 = tcp_segment_rcv (tc0, b0, &next0);
1931  tcp_maybe_inc_err_counter (err_counters, error0);
1932  }
1933 
1934  /* 8: check the FIN bit */
1935  if (PREDICT_FALSE (is_fin))
1936  {
1937  /* Enter CLOSE-WAIT and notify session. To avoid lingering
1938  * in CLOSE-WAIT, set timer (reuse WAITCLOSE). */
1939  /* Account for the FIN if nothing else was received */
1940  if (vnet_buffer (b0)->tcp.data_len == 0)
1941  tc0->rcv_nxt += 1;
1942  tcp_make_ack (tc0, b0);
1943  next0 = tcp_next_output (tc0->c_is_ip4);
1944  tc0->state = TCP_STATE_CLOSE_WAIT;
1945  stream_session_disconnect_notify (&tc0->connection);
1946  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME);
1947  TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc0);
1948  tcp_inc_err_counter (err_counters, TCP_ERROR_FIN_RCVD, 1);
1949  }
1950 
1951  done:
1952  b0->error = node->errors[error0];
1953  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
1954  n_left_to_next, bi0, next0);
1955  }
1956 
1957  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
1958  }
1959 
1961  thread_index);
1962  err_counters[TCP_ERROR_EVENT_FIFO_FULL] = errors;
1963  tcp_store_err_counters (established, err_counters);
1964  tcp_flush_frame_to_output (vm, thread_index, is_ip4);
1965 
1966  return frame->n_vectors;
1967 }
1968 
1969 static uword
1971  vlib_frame_t * from_frame)
1972 {
1973  return tcp46_established_inline (vm, node, from_frame, 1 /* is_ip4 */ );
1974 }
1975 
1976 static uword
1978  vlib_frame_t * from_frame)
1979 {
1980  return tcp46_established_inline (vm, node, from_frame, 0 /* is_ip4 */ );
1981 }
1982 
1983 /* *INDENT-OFF* */
1985 {
1986  .function = tcp4_established,
1987  .name = "tcp4-established",
1988  /* Takes a vector of packets. */
1989  .vector_size = sizeof (u32),
1990  .n_errors = TCP_N_ERROR,
1991  .error_strings = tcp_error_strings,
1992  .n_next_nodes = TCP_ESTABLISHED_N_NEXT,
1993  .next_nodes =
1994  {
1995 #define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n,
1997 #undef _
1998  },
1999  .format_trace = format_tcp_rx_trace_short,
2000 };
2001 /* *INDENT-ON* */
2002 
2004 
2005 /* *INDENT-OFF* */
2007 {
2008  .function = tcp6_established,
2009  .name = "tcp6-established",
2010  /* Takes a vector of packets. */
2011  .vector_size = sizeof (u32),
2012  .n_errors = TCP_N_ERROR,
2013  .error_strings = tcp_error_strings,
2014  .n_next_nodes = TCP_ESTABLISHED_N_NEXT,
2015  .next_nodes =
2016  {
2017 #define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n,
2019 #undef _
2020  },
2021  .format_trace = format_tcp_rx_trace_short,
2022 };
2023 /* *INDENT-ON* */
2024 
2025 
2027 
2030 
2031 static u8
2033 {
2034  transport_connection_t *tmp = 0;
2035  u64 handle;
2036 
2037  if (!tc)
2038  return 1;
2039 
2040  /* Proxy case */
2041  if (tc->c_lcl_port == 0 && tc->state == TCP_STATE_LISTEN)
2042  return 1;
2043 
2044  u8 is_valid = (tc->c_lcl_port == hdr->dst_port
2045  && (tc->state == TCP_STATE_LISTEN
2046  || tc->c_rmt_port == hdr->src_port));
2047 
2048  if (!is_valid)
2049  {
2050  handle = session_lookup_half_open_handle (&tc->connection);
2051  tmp = session_lookup_half_open_connection (handle & 0xFFFFFFFF,
2052  tc->c_proto, tc->c_is_ip4);
2053 
2054  if (tmp)
2055  {
2056  if (tmp->lcl_port == hdr->dst_port
2057  && tmp->rmt_port == hdr->src_port)
2058  {
2059  TCP_DBG ("half-open is valid!");
2060  }
2061  }
2062  }
2063  return is_valid;
2064 }
2065 
2066 /**
2067  * Lookup transport connection
2068  */
2069 static tcp_connection_t *
2070 tcp_lookup_connection (u32 fib_index, vlib_buffer_t * b, u8 thread_index,
2071  u8 is_ip4)
2072 {
2073  tcp_header_t *tcp;
2074  transport_connection_t *tconn;
2075  tcp_connection_t *tc;
2076  u8 is_filtered = 0;
2077  if (is_ip4)
2078  {
2079  ip4_header_t *ip4;
2080  ip4 = vlib_buffer_get_current (b);
2081  tcp = ip4_next_header (ip4);
2082  tconn = session_lookup_connection_wt4 (fib_index,
2083  &ip4->dst_address,
2084  &ip4->src_address,
2085  tcp->dst_port,
2086  tcp->src_port,
2088  thread_index, &is_filtered);
2089  tc = tcp_get_connection_from_transport (tconn);
2090  ASSERT (tcp_lookup_is_valid (tc, tcp));
2091  }
2092  else
2093  {
2094  ip6_header_t *ip6;
2095  ip6 = vlib_buffer_get_current (b);
2096  tcp = ip6_next_header (ip6);
2097  tconn = session_lookup_connection_wt6 (fib_index,
2098  &ip6->dst_address,
2099  &ip6->src_address,
2100  tcp->dst_port,
2101  tcp->src_port,
2103  thread_index, &is_filtered);
2104  tc = tcp_get_connection_from_transport (tconn);
2105  ASSERT (tcp_lookup_is_valid (tc, tcp));
2106  }
2107  return tc;
2108 }
2109 
2112  vlib_frame_t * from_frame, int is_ip4)
2113 {
2114  tcp_main_t *tm = vnet_get_tcp_main ();
2115  u32 n_left_from, next_index, *from, *to_next;
2116  u32 my_thread_index = vm->thread_index, errors = 0;
2117 
2118  from = vlib_frame_vector_args (from_frame);
2119  n_left_from = from_frame->n_vectors;
2120 
2121  next_index = node->cached_next_index;
2122 
2123  while (n_left_from > 0)
2124  {
2125  u32 n_left_to_next;
2126 
2127  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
2128 
2129  while (n_left_from > 0 && n_left_to_next > 0)
2130  {
2131  u32 bi0, ack0, seq0;
2132  vlib_buffer_t *b0;
2133  tcp_rx_trace_t *t0;
2134  tcp_header_t *tcp0 = 0;
2135  tcp_connection_t *tc0;
2136  tcp_connection_t *new_tc0;
2137  u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_ENQUEUED;
2138 
2139  bi0 = from[0];
2140  to_next[0] = bi0;
2141  from += 1;
2142  to_next += 1;
2143  n_left_from -= 1;
2144  n_left_to_next -= 1;
2145 
2146  b0 = vlib_get_buffer (vm, bi0);
2147  tc0 =
2149  tcp.connection_index);
2150  if (PREDICT_FALSE (tc0 == 0))
2151  {
2152  error0 = TCP_ERROR_INVALID_CONNECTION;
2153  goto drop;
2154  }
2155 
2156  /* Half-open completed recently but the connection was't removed
2157  * yet by the owning thread */
2158  if (PREDICT_FALSE (tc0->flags & TCP_CONN_HALF_OPEN_DONE))
2159  {
2160  /* Make sure the connection actually exists */
2161  ASSERT (tcp_lookup_connection (tc0->c_fib_index, b0,
2162  my_thread_index, is_ip4));
2163  goto drop;
2164  }
2165 
2166  ack0 = vnet_buffer (b0)->tcp.ack_number;
2167  seq0 = vnet_buffer (b0)->tcp.seq_number;
2168  tcp0 = tcp_buffer_hdr (b0);
2169 
2170  /* Crude check to see if the connection handle does not match
2171  * the packet. Probably connection just switched to established */
2172  if (PREDICT_FALSE (tcp0->dst_port != tc0->c_lcl_port
2173  || tcp0->src_port != tc0->c_rmt_port))
2174  goto drop;
2175 
2176  if (PREDICT_FALSE
2177  (!tcp_ack (tcp0) && !tcp_rst (tcp0) && !tcp_syn (tcp0)))
2178  goto drop;
2179 
2180  /* SYNs, FINs and data consume sequence numbers */
2181  vnet_buffer (b0)->tcp.seq_end = seq0 + tcp_is_syn (tcp0)
2182  + tcp_is_fin (tcp0) + vnet_buffer (b0)->tcp.data_len;
2183 
2184  /*
2185  * 1. check the ACK bit
2186  */
2187 
2188  /*
2189  * If the ACK bit is set
2190  * If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send a reset (unless
2191  * the RST bit is set, if so drop the segment and return)
2192  * <SEQ=SEG.ACK><CTL=RST>
2193  * and discard the segment. Return.
2194  * If SND.UNA =< SEG.ACK =< SND.NXT then the ACK is acceptable.
2195  */
2196  if (tcp_ack (tcp0))
2197  {
2198  if (seq_leq (ack0, tc0->iss) || seq_gt (ack0, tc0->snd_nxt))
2199  {
2200  clib_warning ("ack not in rcv wnd");
2201  if (!tcp_rst (tcp0))
2202  tcp_send_reset_w_pkt (tc0, b0, is_ip4);
2203  goto drop;
2204  }
2205 
2206  /* Make sure ACK is valid */
2207  if (seq_gt (tc0->snd_una, ack0))
2208  {
2209  clib_warning ("ack invalid");
2210  goto drop;
2211  }
2212  }
2213 
2214  /*
2215  * 2. check the RST bit
2216  */
2217 
2218  if (tcp_rst (tcp0))
2219  {
2220  /* If ACK is acceptable, signal client that peer is not
2221  * willing to accept connection and drop connection*/
2222  if (tcp_ack (tcp0))
2223  tcp_connection_reset (tc0);
2224  goto drop;
2225  }
2226 
2227  /*
2228  * 3. check the security and precedence (skipped)
2229  */
2230 
2231  /*
2232  * 4. check the SYN bit
2233  */
2234 
2235  /* No SYN flag. Drop. */
2236  if (!tcp_syn (tcp0))
2237  {
2238  clib_warning ("not synack");
2239  goto drop;
2240  }
2241 
2242  /* Parse options */
2243  if (tcp_options_parse (tcp0, &tc0->rcv_opts))
2244  {
2245  clib_warning ("options parse fail");
2246  goto drop;
2247  }
2248 
2249  /* Valid SYN or SYN-ACK. Move connection from half-open pool to
2250  * current thread pool. */
2251  pool_get (tm->connections[my_thread_index], new_tc0);
2252  clib_memcpy (new_tc0, tc0, sizeof (*new_tc0));
2253  new_tc0->c_c_index = new_tc0 - tm->connections[my_thread_index];
2254  new_tc0->c_thread_index = my_thread_index;
2255  new_tc0->rcv_nxt = vnet_buffer (b0)->tcp.seq_end;
2256  new_tc0->irs = seq0;
2257  new_tc0->timers[TCP_TIMER_ESTABLISH] = TCP_TIMER_HANDLE_INVALID;
2258  new_tc0->timers[TCP_TIMER_RETRANSMIT_SYN] =
2260  new_tc0->sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_RX];
2261 
2262  /* If this is not the owning thread, wait for syn retransmit to
2263  * expire and cleanup then */
2265  tc0->flags |= TCP_CONN_HALF_OPEN_DONE;
2266 
2267  if (tcp_opts_tstamp (&new_tc0->rcv_opts))
2268  {
2269  new_tc0->tsval_recent = new_tc0->rcv_opts.tsval;
2270  new_tc0->tsval_recent_age = tcp_time_now ();
2271  }
2272 
2273  if (tcp_opts_wscale (&new_tc0->rcv_opts))
2274  new_tc0->snd_wscale = new_tc0->rcv_opts.wscale;
2275 
2276  /* RFC1323: SYN and SYN-ACK wnd not scaled */
2277  new_tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window);
2278  new_tc0->snd_wl1 = seq0;
2279  new_tc0->snd_wl2 = ack0;
2280 
2281  tcp_connection_init_vars (new_tc0);
2282 
2283  /* SYN-ACK: See if we can switch to ESTABLISHED state */
2284  if (PREDICT_TRUE (tcp_ack (tcp0)))
2285  {
2286  /* Our SYN is ACKed: we have iss < ack = snd_una */
2287 
2288  /* TODO Dequeue acknowledged segments if we support Fast Open */
2289  new_tc0->snd_una = ack0;
2290  new_tc0->state = TCP_STATE_ESTABLISHED;
2291 
2292  /* Make sure las is initialized for the wnd computation */
2293  new_tc0->rcv_las = new_tc0->rcv_nxt;
2294 
2295  /* Notify app that we have connection. If session layer can't
2296  * allocate session send reset */
2297  if (session_stream_connect_notify (&new_tc0->connection, 0))
2298  {
2299  clib_warning ("connect notify fail");
2300  tcp_send_reset_w_pkt (new_tc0, b0, is_ip4);
2301  tcp_connection_cleanup (new_tc0);
2302  goto drop;
2303  }
2304 
2305  /* Make sure after data segment processing ACK is sent */
2306  new_tc0->flags |= TCP_CONN_SNDACK;
2307 
2308  /* Update rtt with the syn-ack sample */
2309  tcp_update_rtt (new_tc0, vnet_buffer (b0)->tcp.ack_number);
2310  TCP_EVT_DBG (TCP_EVT_SYNACK_RCVD, new_tc0);
2311  }
2312  /* SYN: Simultaneous open. Change state to SYN-RCVD and send SYN-ACK */
2313  else
2314  {
2315  new_tc0->state = TCP_STATE_SYN_RCVD;
2316 
2317  /* Notify app that we have connection */
2318  if (session_stream_connect_notify (&new_tc0->connection, 0))
2319  {
2320  tcp_connection_cleanup (new_tc0);
2321  tcp_send_reset_w_pkt (tc0, b0, is_ip4);
2322  TCP_EVT_DBG (TCP_EVT_RST_SENT, tc0);
2323  goto drop;
2324  }
2325 
2326  tc0->rtt_ts = 0;
2327  tcp_init_snd_vars (tc0);
2328  tcp_make_synack (new_tc0, b0);
2329  next0 = tcp_next_output (is_ip4);
2330 
2331  goto drop;
2332  }
2333 
2334  /* Read data, if any */
2335  if (PREDICT_FALSE (vnet_buffer (b0)->tcp.data_len))
2336  {
2337  clib_warning ("rcvd data in syn-sent");
2338  error0 = tcp_segment_rcv (new_tc0, b0, &next0);
2339  if (error0 == TCP_ERROR_ACK_OK)
2340  error0 = TCP_ERROR_SYN_ACKS_RCVD;
2341  tcp_maybe_inc_counter (syn_sent, error0, 1);
2342  }
2343  else
2344  {
2345  tcp_make_ack (new_tc0, b0);
2346  next0 = tcp_next_output (new_tc0->c_is_ip4);
2347  }
2348 
2349  drop:
2350 
2351  b0->error = error0 ? node->errors[error0] : 0;
2352  if (PREDICT_FALSE
2353  ((b0->flags & VLIB_BUFFER_IS_TRACED) && tcp0 != 0))
2354  {
2355  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
2356  clib_memcpy (&t0->tcp_header, tcp0, sizeof (t0->tcp_header));
2357  clib_memcpy (&t0->tcp_connection, tc0,
2358  sizeof (t0->tcp_connection));
2359  }
2360 
2361  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
2362  n_left_to_next, bi0, next0);
2363  }
2364 
2365  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
2366  }
2367 
2369  my_thread_index);
2370  tcp_inc_counter (syn_sent, TCP_ERROR_EVENT_FIFO_FULL, errors);
2371  return from_frame->n_vectors;
2372 }
2373 
2374 static uword
2376  vlib_frame_t * from_frame)
2377 {
2378  return tcp46_syn_sent_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2379 }
2380 
2381 static uword
2383  vlib_frame_t * from_frame)
2384 {
2385  return tcp46_syn_sent_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2386 }
2387 
2388 /* *INDENT-OFF* */
2390 {
2391  .function = tcp4_syn_sent,
2392  .name = "tcp4-syn-sent",
2393  /* Takes a vector of packets. */
2394  .vector_size = sizeof (u32),
2395  .n_errors = TCP_N_ERROR,
2396  .error_strings = tcp_error_strings,
2397  .n_next_nodes = TCP_SYN_SENT_N_NEXT,
2398  .next_nodes =
2399  {
2400 #define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n,
2402 #undef _
2403  },
2404  .format_trace = format_tcp_rx_trace_short,
2405 };
2406 /* *INDENT-ON* */
2407 
2409 
2410 /* *INDENT-OFF* */
2412 {
2413  .function = tcp6_syn_sent_rcv,
2414  .name = "tcp6-syn-sent",
2415  /* Takes a vector of packets. */
2416  .vector_size = sizeof (u32),
2417  .n_errors = TCP_N_ERROR,
2418  .error_strings = tcp_error_strings,
2419  .n_next_nodes = TCP_SYN_SENT_N_NEXT,
2420  .next_nodes =
2421  {
2422 #define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n,
2424 #undef _
2425  },
2426  .format_trace = format_tcp_rx_trace_short,
2427 };
2428 /* *INDENT-ON* */
2429 
2431 
2434 
2435 /**
2436  * Handles reception for all states except LISTEN, SYN-SENT and ESTABLISHED
2437  * as per RFC793 p. 64
2438  */
2441  vlib_frame_t * from_frame, int is_ip4)
2442 {
2443  u32 n_left_from, next_index, *from, *to_next, n_fins = 0;
2444  u32 my_thread_index = vm->thread_index, errors = 0;
2445 
2446  from = vlib_frame_vector_args (from_frame);
2447  n_left_from = from_frame->n_vectors;
2448  next_index = node->cached_next_index;
2449 
2450  while (n_left_from > 0)
2451  {
2452  u32 n_left_to_next;
2453 
2454  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
2455 
2456  while (n_left_from > 0 && n_left_to_next > 0)
2457  {
2458  u32 bi0;
2459  vlib_buffer_t *b0;
2460  tcp_header_t *tcp0 = 0;
2461  tcp_connection_t *tc0;
2462  u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_NONE;
2463  u8 is_fin0;
2464 
2465  bi0 = from[0];
2466  to_next[0] = bi0;
2467  from += 1;
2468  to_next += 1;
2469  n_left_from -= 1;
2470  n_left_to_next -= 1;
2471 
2472  b0 = vlib_get_buffer (vm, bi0);
2473  tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
2474  my_thread_index);
2475  if (PREDICT_FALSE (tc0 == 0))
2476  {
2477  error0 = TCP_ERROR_INVALID_CONNECTION;
2478  goto drop;
2479  }
2480 
2481  tcp0 = tcp_buffer_hdr (b0);
2482  is_fin0 = tcp_is_fin (tcp0);
2483 
2484  /* SYNs, FINs and data consume sequence numbers */
2485  vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number
2486  + tcp_is_syn (tcp0) + is_fin0 + vnet_buffer (b0)->tcp.data_len;
2487 
2488  if (CLIB_DEBUG)
2489  {
2490  tcp_connection_t *tmp;
2491  tmp = tcp_lookup_connection (tc0->c_fib_index, b0,
2492  my_thread_index, is_ip4);
2493  if (tmp->state != tc0->state)
2494  {
2495  clib_warning ("state changed");
2496  goto drop;
2497  }
2498  }
2499 
2500  /*
2501  * Special treatment for CLOSED
2502  */
2503  if (PREDICT_FALSE (tc0->state == TCP_STATE_CLOSED))
2504  {
2505  error0 = TCP_ERROR_CONNECTION_CLOSED;
2506  goto drop;
2507  }
2508 
2509  /*
2510  * For all other states (except LISTEN)
2511  */
2512 
2513  /* 1-4: check SEQ, RST, SYN */
2514  if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, tcp0,
2515  &next0, &error0)))
2516  {
2517  tcp_maybe_inc_counter (rcv_process, error0, 1);
2518  goto drop;
2519  }
2520 
2521  /* 5: check the ACK field */
2522  switch (tc0->state)
2523  {
2524  case TCP_STATE_SYN_RCVD:
2525  /*
2526  * If the segment acknowledgment is not acceptable, form a
2527  * reset segment,
2528  * <SEQ=SEG.ACK><CTL=RST>
2529  * and send it.
2530  */
2531  if (!tcp_rcv_ack_is_acceptable (tc0, b0))
2532  {
2533  TCP_DBG ("connection not accepted");
2534  tcp_send_reset_w_pkt (tc0, b0, is_ip4);
2535  error0 = TCP_ERROR_ACK_INVALID;
2536  goto drop;
2537  }
2538 
2539  /* Update rtt and rto */
2540  tcp_update_rtt (tc0, vnet_buffer (b0)->tcp.ack_number);
2541 
2542  /* Switch state to ESTABLISHED */
2543  tc0->state = TCP_STATE_ESTABLISHED;
2544  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2545 
2546  /* Initialize session variables */
2547  tc0->snd_una = vnet_buffer (b0)->tcp.ack_number;
2548  tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window)
2549  << tc0->rcv_opts.wscale;
2550  tc0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number;
2551  tc0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number;
2552 
2553  /* Reset SYN-ACK retransmit and SYN_RCV establish timers */
2555  tcp_timer_reset (tc0, TCP_TIMER_ESTABLISH);
2556  stream_session_accept_notify (&tc0->connection);
2557  error0 = TCP_ERROR_ACK_OK;
2558  break;
2559  case TCP_STATE_ESTABLISHED:
2560  /* We can get packets in established state here because they
2561  * were enqueued before state change */
2562  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2563  {
2564  tcp_maybe_inc_counter (rcv_process, error0, 1);
2565  goto drop;
2566  }
2567 
2568  break;
2569  case TCP_STATE_FIN_WAIT_1:
2570  /* In addition to the processing for the ESTABLISHED state, if
2571  * our FIN is now acknowledged then enter FIN-WAIT-2 and
2572  * continue processing in that state. */
2573  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2574  {
2575  tcp_maybe_inc_counter (rcv_process, error0, 1);
2576  goto drop;
2577  }
2578 
2579  /* Still have to send the FIN */
2580  if (tc0->flags & TCP_CONN_FINPNDG)
2581  {
2582  /* TX fifo finally drained */
2583  if (!session_tx_fifo_max_dequeue (&tc0->connection))
2584  tcp_send_fin (tc0);
2585  }
2586  /* If FIN is ACKed */
2587  else if (tc0->snd_una == tc0->snd_una_max)
2588  {
2589  tc0->state = TCP_STATE_FIN_WAIT_2;
2590  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2591 
2592  /* Stop all retransmit timers because we have nothing more
2593  * to send. Enable waitclose though because we're willing to
2594  * wait for peer's FIN but not indefinitely. */
2596  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
2597  }
2598  break;
2599  case TCP_STATE_FIN_WAIT_2:
2600  /* In addition to the processing for the ESTABLISHED state, if
2601  * the retransmission queue is empty, the user's CLOSE can be
2602  * acknowledged ("ok") but do not delete the TCB. */
2603  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2604  {
2605  tcp_maybe_inc_counter (rcv_process, error0, 1);
2606  goto drop;
2607  }
2608  break;
2609  case TCP_STATE_CLOSE_WAIT:
2610  /* Do the same processing as for the ESTABLISHED state. */
2611  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2612  {
2613  tcp_maybe_inc_counter (rcv_process, error0, 1);
2614  goto drop;
2615  }
2616  if (tc0->flags & TCP_CONN_FINPNDG)
2617  {
2618  /* TX fifo finally drained */
2619  if (!session_tx_fifo_max_dequeue (&tc0->connection))
2620  {
2621  tcp_send_fin (tc0);
2623  tc0->state = TCP_STATE_LAST_ACK;
2624  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE,
2625  TCP_2MSL_TIME);
2626  }
2627  }
2628  break;
2629  case TCP_STATE_CLOSING:
2630  /* In addition to the processing for the ESTABLISHED state, if
2631  * the ACK acknowledges our FIN then enter the TIME-WAIT state,
2632  * otherwise ignore the segment. */
2633  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2634  {
2635  tcp_maybe_inc_counter (rcv_process, error0, 1);
2636  goto drop;
2637  }
2638 
2639  tc0->state = TCP_STATE_TIME_WAIT;
2640  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2641  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
2642  goto drop;
2643 
2644  break;
2645  case TCP_STATE_LAST_ACK:
2646  /* The only thing that [should] arrive in this state is an
2647  * acknowledgment of our FIN. If our FIN is now acknowledged,
2648  * delete the TCB, enter the CLOSED state, and return. */
2649 
2650  if (!tcp_rcv_ack_is_acceptable (tc0, b0))
2651  {
2652  error0 = TCP_ERROR_ACK_INVALID;
2653  goto drop;
2654  }
2655  error0 = TCP_ERROR_ACK_OK;
2656  tc0->snd_una = vnet_buffer (b0)->tcp.ack_number;
2657  /* Apparently our ACK for the peer's FIN was lost */
2658  if (is_fin0 && tc0->snd_una != tc0->snd_una_max)
2659  {
2660  tcp_send_fin (tc0);
2661  goto drop;
2662  }
2663 
2664  tc0->state = TCP_STATE_CLOSED;
2665  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2666  /* Delete the connection/session since the pipes should be
2667  * clear by now */
2668  tcp_connection_del (tc0);
2669 
2670  goto drop;
2671 
2672  break;
2673  case TCP_STATE_TIME_WAIT:
2674  /* The only thing that can arrive in this state is a
2675  * retransmission of the remote FIN. Acknowledge it, and restart
2676  * the 2 MSL timeout. */
2677 
2678  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2679  {
2680  tcp_maybe_inc_counter (rcv_process, error0, 1);
2681  goto drop;
2682  }
2683 
2684  tcp_make_ack (tc0, b0);
2685  next0 = tcp_next_output (is_ip4);
2686  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
2687  goto drop;
2688 
2689  break;
2690  default:
2691  ASSERT (0);
2692  }
2693 
2694  /* 6: check the URG bit TODO */
2695 
2696  /* 7: process the segment text */
2697  switch (tc0->state)
2698  {
2699  case TCP_STATE_ESTABLISHED:
2700  case TCP_STATE_FIN_WAIT_1:
2701  case TCP_STATE_FIN_WAIT_2:
2702  if (vnet_buffer (b0)->tcp.data_len)
2703  {
2704  error0 = tcp_segment_rcv (tc0, b0, &next0);
2705  tcp_maybe_inc_counter (rcv_process, error0, 1);
2706  }
2707  else if (is_fin0)
2708  tc0->rcv_nxt += 1;
2709  break;
2710  case TCP_STATE_CLOSE_WAIT:
2711  case TCP_STATE_CLOSING:
2712  case TCP_STATE_LAST_ACK:
2713  case TCP_STATE_TIME_WAIT:
2714  /* This should not occur, since a FIN has been received from the
2715  * remote side. Ignore the segment text. */
2716  break;
2717  }
2718 
2719  /* 8: check the FIN bit */
2720  if (!is_fin0)
2721  goto drop;
2722 
2723  switch (tc0->state)
2724  {
2725  case TCP_STATE_ESTABLISHED:
2726  case TCP_STATE_SYN_RCVD:
2727  /* Send FIN-ACK notify app and enter CLOSE-WAIT */
2730  tcp_make_fin (tc0, b0);
2731  tc0->snd_nxt += 1;
2732  tc0->snd_una_max = tc0->snd_nxt;
2733  next0 = tcp_next_output (tc0->c_is_ip4);
2734  stream_session_disconnect_notify (&tc0->connection);
2735  tc0->state = TCP_STATE_CLOSE_WAIT;
2736  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2737  break;
2738  case TCP_STATE_CLOSE_WAIT:
2739  case TCP_STATE_CLOSING:
2740  case TCP_STATE_LAST_ACK:
2741  /* move along .. */
2742  break;
2743  case TCP_STATE_FIN_WAIT_1:
2744  tc0->state = TCP_STATE_CLOSING;
2745  tcp_make_ack (tc0, b0);
2746  next0 = tcp_next_output (is_ip4);
2747  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2748  /* Wait for ACK but not forever */
2749  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
2750  break;
2751  case TCP_STATE_FIN_WAIT_2:
2752  /* Got FIN, send ACK! Be more aggressive with resource cleanup */
2753  tc0->state = TCP_STATE_TIME_WAIT;
2755  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
2756  tcp_make_ack (tc0, b0);
2757  next0 = tcp_next_output (is_ip4);
2758  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2759  break;
2760  case TCP_STATE_TIME_WAIT:
2761  /* Remain in the TIME-WAIT state. Restart the time-wait
2762  * timeout.
2763  */
2764  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
2765  break;
2766  }
2767  TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc0);
2768  n_fins += 1;
2769 
2770  drop:
2771  b0->error = error0 ? node->errors[error0] : 0;
2772 
2773  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
2774  {
2775  tcp_rx_trace_t *t0 =
2776  vlib_add_trace (vm, node, b0, sizeof (*t0));
2777  tcp_set_rx_trace_data (t0, tc0, tcp0, b0, is_ip4);
2778  }
2779 
2780  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
2781  n_left_to_next, bi0, next0);
2782  }
2783 
2784  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
2785  }
2786 
2788  my_thread_index);
2789  tcp_inc_counter (rcv_process, TCP_ERROR_EVENT_FIFO_FULL, errors);
2790  tcp_inc_counter (rcv_process, TCP_ERROR_FIN_RCVD, n_fins);
2791  return from_frame->n_vectors;
2792 }
2793 
2794 static uword
2796  vlib_frame_t * from_frame)
2797 {
2798  return tcp46_rcv_process_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2799 }
2800 
2801 static uword
2803  vlib_frame_t * from_frame)
2804 {
2805  return tcp46_rcv_process_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2806 }
2807 
2808 /* *INDENT-OFF* */
2810 {
2811  .function = tcp4_rcv_process,
2812  .name = "tcp4-rcv-process",
2813  /* Takes a vector of packets. */
2814  .vector_size = sizeof (u32),
2815  .n_errors = TCP_N_ERROR,
2816  .error_strings = tcp_error_strings,
2817  .n_next_nodes = TCP_RCV_PROCESS_N_NEXT,
2818  .next_nodes =
2819  {
2820 #define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n,
2822 #undef _
2823  },
2824  .format_trace = format_tcp_rx_trace_short,
2825 };
2826 /* *INDENT-ON* */
2827 
2829 
2830 /* *INDENT-OFF* */
2832 {
2833  .function = tcp6_rcv_process,
2834  .name = "tcp6-rcv-process",
2835  /* Takes a vector of packets. */
2836  .vector_size = sizeof (u32),
2837  .n_errors = TCP_N_ERROR,
2838  .error_strings = tcp_error_strings,
2839  .n_next_nodes = TCP_RCV_PROCESS_N_NEXT,
2840  .next_nodes =
2841  {
2842 #define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n,
2844 #undef _
2845  },
2846  .format_trace = format_tcp_rx_trace_short,
2847 };
2848 /* *INDENT-ON* */
2849 
2851 
2854 
2855 /**
2856  * LISTEN state processing as per RFC 793 p. 65
2857  */
2860  vlib_frame_t * from_frame, int is_ip4)
2861 {
2862  u32 n_left_from, next_index, *from, *to_next, n_syns = 0;
2863  u32 my_thread_index = vm->thread_index;
2864 
2865  from = vlib_frame_vector_args (from_frame);
2866  n_left_from = from_frame->n_vectors;
2867 
2868  next_index = node->cached_next_index;
2869 
2870  while (n_left_from > 0)
2871  {
2872  u32 n_left_to_next;
2873 
2874  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
2875 
2876  while (n_left_from > 0 && n_left_to_next > 0)
2877  {
2878  u32 bi0;
2879  vlib_buffer_t *b0;
2880  tcp_rx_trace_t *t0;
2881  tcp_header_t *th0 = 0;
2882  tcp_connection_t *lc0;
2883  ip4_header_t *ip40;
2884  ip6_header_t *ip60;
2885  tcp_connection_t *child0;
2886  u32 error0 = TCP_ERROR_NONE, next0 = tcp_next_drop (is_ip4);
2887 
2888  bi0 = from[0];
2889  to_next[0] = bi0;
2890  from += 1;
2891  to_next += 1;
2892  n_left_from -= 1;
2893  n_left_to_next -= 1;
2894 
2895  b0 = vlib_get_buffer (vm, bi0);
2896  lc0 = tcp_listener_get (vnet_buffer (b0)->tcp.connection_index);
2897 
2898  if (is_ip4)
2899  {
2900  ip40 = vlib_buffer_get_current (b0);
2901  th0 = ip4_next_header (ip40);
2902  }
2903  else
2904  {
2905  ip60 = vlib_buffer_get_current (b0);
2906  th0 = ip6_next_header (ip60);
2907  }
2908 
2909  /* Create child session. For syn-flood protection use filter */
2910 
2911  /* 1. first check for an RST: handled in dispatch */
2912  /* if (tcp_rst (th0))
2913  goto drop; */
2914 
2915  /* 2. second check for an ACK: handled in dispatch */
2916  /* if (tcp_ack (th0))
2917  {
2918  tcp_send_reset (b0, is_ip4);
2919  goto drop;
2920  } */
2921 
2922  /* 3. check for a SYN (did that already) */
2923 
2924  /* Make sure connection wasn't just created */
2925  child0 = tcp_lookup_connection (lc0->c_fib_index, b0,
2926  my_thread_index, is_ip4);
2927  if (PREDICT_FALSE (child0->state != TCP_STATE_LISTEN))
2928  {
2929  error0 = TCP_ERROR_CREATE_EXISTS;
2930  goto drop;
2931  }
2932 
2933  /* Create child session and send SYN-ACK */
2934  child0 = tcp_connection_new (my_thread_index);
2935  child0->c_lcl_port = th0->dst_port;
2936  child0->c_rmt_port = th0->src_port;
2937  child0->c_is_ip4 = is_ip4;
2938  child0->state = TCP_STATE_SYN_RCVD;
2939  child0->c_fib_index = lc0->c_fib_index;
2940 
2941  if (is_ip4)
2942  {
2943  child0->c_lcl_ip4.as_u32 = ip40->dst_address.as_u32;
2944  child0->c_rmt_ip4.as_u32 = ip40->src_address.as_u32;
2945  }
2946  else
2947  {
2948  clib_memcpy (&child0->c_lcl_ip6, &ip60->dst_address,
2949  sizeof (ip6_address_t));
2950  clib_memcpy (&child0->c_rmt_ip6, &ip60->src_address,
2951  sizeof (ip6_address_t));
2952  }
2953 
2954  if (tcp_options_parse (th0, &child0->rcv_opts))
2955  {
2956  clib_warning ("options parse fail");
2957  goto drop;
2958  }
2959 
2960  child0->irs = vnet_buffer (b0)->tcp.seq_number;
2961  child0->rcv_nxt = vnet_buffer (b0)->tcp.seq_number + 1;
2962  child0->rcv_las = child0->rcv_nxt;
2963  child0->sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_RX];
2964 
2965  /* RFC1323: TSval timestamps sent on {SYN} and {SYN,ACK}
2966  * segments are used to initialize PAWS. */
2967  if (tcp_opts_tstamp (&child0->rcv_opts))
2968  {
2969  child0->tsval_recent = child0->rcv_opts.tsval;
2970  child0->tsval_recent_age = tcp_time_now ();
2971  }
2972 
2973  if (tcp_opts_wscale (&child0->rcv_opts))
2974  child0->snd_wscale = child0->rcv_opts.wscale;
2975 
2976  child0->snd_wnd = clib_net_to_host_u16 (th0->window)
2977  << child0->snd_wscale;
2978  child0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number;
2979  child0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number;
2980 
2981  tcp_connection_init_vars (child0);
2982  TCP_EVT_DBG (TCP_EVT_SYN_RCVD, child0, 1);
2983 
2984  if (stream_session_accept (&child0->connection, lc0->c_s_index,
2985  0 /* notify */ ))
2986  {
2987  clib_warning ("session accept fail");
2988  tcp_connection_cleanup (child0);
2989  error0 = TCP_ERROR_CREATE_SESSION_FAIL;
2990  goto drop;
2991  }
2992 
2993  /* Reuse buffer to make syn-ack and send */
2994  tcp_make_synack (child0, b0);
2995  next0 = tcp_next_output (is_ip4);
2996  tcp_timer_set (child0, TCP_TIMER_ESTABLISH, TCP_SYN_RCVD_TIME);
2997 
2998  drop:
2999  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
3000  {
3001  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
3002  clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header));
3003  clib_memcpy (&t0->tcp_connection, lc0,
3004  sizeof (t0->tcp_connection));
3005  }
3006 
3007  n_syns += (error0 == TCP_ERROR_NONE);
3008  b0->error = node->errors[error0];
3009 
3010  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
3011  n_left_to_next, bi0, next0);
3012  }
3013 
3014  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
3015  }
3016 
3017  tcp_inc_counter (listen, TCP_ERROR_SYNS_RCVD, n_syns);
3018  return from_frame->n_vectors;
3019 }
3020 
3021 static uword
3023  vlib_frame_t * from_frame)
3024 {
3025  return tcp46_listen_inline (vm, node, from_frame, 1 /* is_ip4 */ );
3026 }
3027 
3028 static uword
3030  vlib_frame_t * from_frame)
3031 {
3032  return tcp46_listen_inline (vm, node, from_frame, 0 /* is_ip4 */ );
3033 }
3034 
3035 /* *INDENT-OFF* */
3037 {
3038  .function = tcp4_listen,
3039  .name = "tcp4-listen",
3040  /* Takes a vector of packets. */
3041  .vector_size = sizeof (u32),
3042  .n_errors = TCP_N_ERROR,
3043  .error_strings = tcp_error_strings,
3044  .n_next_nodes = TCP_LISTEN_N_NEXT,
3045  .next_nodes =
3046  {
3047 #define _(s,n) [TCP_LISTEN_NEXT_##s] = n,
3049 #undef _
3050  },
3051  .format_trace = format_tcp_rx_trace_short,
3052 };
3053 /* *INDENT-ON* */
3054 
3056 
3057 /* *INDENT-OFF* */
3059 {
3060  .function = tcp6_listen,
3061  .name = "tcp6-listen",
3062  /* Takes a vector of packets. */
3063  .vector_size = sizeof (u32),
3064  .n_errors = TCP_N_ERROR,
3065  .error_strings = tcp_error_strings,
3066  .n_next_nodes = TCP_LISTEN_N_NEXT,
3067  .next_nodes =
3068  {
3069 #define _(s,n) [TCP_LISTEN_NEXT_##s] = n,
3071 #undef _
3072  },
3073  .format_trace = format_tcp_rx_trace_short,
3074 };
3075 /* *INDENT-ON* */
3076 
3078 
3081 
3082 typedef enum _tcp_input_next
3083 {
3093 
3094 #define foreach_tcp4_input_next \
3095  _ (DROP, "ip4-drop") \
3096  _ (LISTEN, "tcp4-listen") \
3097  _ (RCV_PROCESS, "tcp4-rcv-process") \
3098  _ (SYN_SENT, "tcp4-syn-sent") \
3099  _ (ESTABLISHED, "tcp4-established") \
3100  _ (RESET, "tcp4-reset") \
3101  _ (PUNT, "ip4-punt")
3102 
3103 #define foreach_tcp6_input_next \
3104  _ (DROP, "ip6-drop") \
3105  _ (LISTEN, "tcp6-listen") \
3106  _ (RCV_PROCESS, "tcp6-rcv-process") \
3107  _ (SYN_SENT, "tcp6-syn-sent") \
3108  _ (ESTABLISHED, "tcp6-established") \
3109  _ (RESET, "tcp6-reset") \
3110  _ (PUNT, "ip6-punt")
3111 
3112 #define filter_flags (TCP_FLAG_SYN|TCP_FLAG_ACK|TCP_FLAG_RST|TCP_FLAG_FIN)
3113 
3114 static void
3116  vlib_buffer_t ** bs, u32 n_bufs, u8 is_ip4)
3117 {
3118  tcp_connection_t *tc;
3119  tcp_header_t *tcp;
3120  tcp_rx_trace_t *t;
3121  int i;
3122 
3123  for (i = 0; i < n_bufs; i++)
3124  {
3125  if (bs[i]->flags & VLIB_BUFFER_IS_TRACED)
3126  {
3127  t = vlib_add_trace (vm, node, bs[i], sizeof (*t));
3128  tc = tcp_connection_get (vnet_buffer (bs[i])->tcp.connection_index,
3129  vm->thread_index);
3130  tcp = vlib_buffer_get_current (bs[i]);
3131  tcp_set_rx_trace_data (t, tc, tcp, bs[i], is_ip4);
3132  }
3133  }
3134 }
3135 
3136 static void
3137 tcp_input_set_error_next (tcp_main_t * tm, u16 * next, u32 * error, u8 is_ip4)
3138 {
3139  if (*error == TCP_ERROR_FILTERED)
3140  {
3141  *next = TCP_INPUT_NEXT_DROP;
3142  }
3143  else if ((is_ip4 && tm->punt_unknown4) || (!is_ip4 && tm->punt_unknown6))
3144  {
3145  *next = TCP_INPUT_NEXT_PUNT;
3146  *error = TCP_ERROR_PUNT;
3147  }
3148  else
3149  {
3150  *next = TCP_INPUT_NEXT_RESET;
3151  *error = TCP_ERROR_NO_LISTENER;
3152  }
3153 }
3154 
3155 static inline tcp_connection_t *
3156 tcp_input_lookup_buffer (vlib_buffer_t * b, u8 thread_index, u32 * error,
3157  u8 is_ip4)
3158 {
3159  u32 fib_index = vnet_buffer (b)->ip.fib_index;
3160  int n_advance_bytes, n_data_bytes;
3162  tcp_header_t *tcp;
3163  u8 is_filtered = 0;
3164 
3165  if (is_ip4)
3166  {
3168  tcp = ip4_next_header (ip4);
3169  vnet_buffer (b)->tcp.hdr_offset = (u8 *) tcp - (u8 *) ip4;
3170  n_advance_bytes = (ip4_header_bytes (ip4) + tcp_header_bytes (tcp));
3171  n_data_bytes = clib_net_to_host_u16 (ip4->length) - n_advance_bytes;
3172 
3173  /* Length check. Checksum computed by ipx_local no need to compute again */
3174  if (PREDICT_FALSE (n_advance_bytes < 0))
3175  {
3176  *error = TCP_ERROR_LENGTH;
3177  return 0;
3178  }
3179 
3180  tc = session_lookup_connection_wt4 (fib_index, &ip4->dst_address,
3181  &ip4->src_address, tcp->dst_port,
3182  tcp->src_port, TRANSPORT_PROTO_TCP,
3183  thread_index, &is_filtered);
3184  }
3185  else
3186  {
3188  tcp = ip6_next_header (ip6);
3189  vnet_buffer (b)->tcp.hdr_offset = (u8 *) tcp - (u8 *) ip6;
3190  n_advance_bytes = tcp_header_bytes (tcp);
3191  n_data_bytes = clib_net_to_host_u16 (ip6->payload_length)
3192  - n_advance_bytes;
3193  n_advance_bytes += sizeof (ip6[0]);
3194 
3195  if (PREDICT_FALSE (n_advance_bytes < 0))
3196  {
3197  *error = TCP_ERROR_LENGTH;
3198  return 0;
3199  }
3200 
3201  tc = session_lookup_connection_wt6 (fib_index, &ip6->dst_address,
3202  &ip6->src_address, tcp->dst_port,
3203  tcp->src_port, TRANSPORT_PROTO_TCP,
3204  thread_index, &is_filtered);
3205  }
3206 
3207  vnet_buffer (b)->tcp.seq_number = clib_net_to_host_u32 (tcp->seq_number);
3208  vnet_buffer (b)->tcp.ack_number = clib_net_to_host_u32 (tcp->ack_number);
3209  vnet_buffer (b)->tcp.data_offset = n_advance_bytes;
3210  vnet_buffer (b)->tcp.data_len = n_data_bytes;
3211  vnet_buffer (b)->tcp.flags = 0;
3212 
3213  *error = is_filtered ? TCP_ERROR_FILTERED : *error;
3214 
3216 }
3217 
3218 static inline void
3220  vlib_buffer_t * b, u16 * next, u32 * error)
3221 {
3222  tcp_header_t *tcp;
3223  u8 flags;
3224 
3225  tcp = tcp_buffer_hdr (b);
3226  flags = tcp->flags & filter_flags;
3227  *next = tm->dispatch_table[tc->state][flags].next;
3228  *error = tm->dispatch_table[tc->state][flags].error;
3229 
3230  if (PREDICT_FALSE (*error == TCP_ERROR_DISPATCH
3231  || *next == TCP_INPUT_NEXT_RESET))
3232  {
3233  /* Overload tcp flags to store state */
3234  tcp_state_t state = tc->state;
3235  vnet_buffer (b)->tcp.flags = tc->state;
3236 
3237  if (*error == TCP_ERROR_DISPATCH)
3238  clib_warning ("disp error state %U flags %U", format_tcp_state,
3239  state, format_tcp_flags, (int) flags);
3240  }
3241 }
3242 
3245  vlib_frame_t * frame, int is_ip4)
3246 {
3247  u32 n_left_from, *from, thread_index = vm->thread_index;
3248  tcp_main_t *tm = vnet_get_tcp_main ();
3249  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
3250  u16 nexts[VLIB_FRAME_SIZE], *next;
3251 
3252  tcp_set_time_now (thread_index);
3253 
3254  from = vlib_frame_vector_args (frame);
3255  n_left_from = frame->n_vectors;
3256  vlib_get_buffers (vm, from, bufs, n_left_from);
3257 
3258  b = bufs;
3259  next = nexts;
3260 
3261  while (n_left_from >= 4)
3262  {
3263  u32 error0 = TCP_ERROR_NO_LISTENER, error1 = TCP_ERROR_NO_LISTENER;
3264  tcp_connection_t *tc0, *tc1;
3265 
3266  {
3267  vlib_prefetch_buffer_header (b[2], STORE);
3268  CLIB_PREFETCH (b[2]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3269 
3270  vlib_prefetch_buffer_header (b[3], STORE);
3271  CLIB_PREFETCH (b[3]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3272  }
3273 
3274  next[0] = next[1] = TCP_INPUT_NEXT_DROP;
3275 
3276  tc0 = tcp_input_lookup_buffer (b[0], thread_index, &error0, is_ip4);
3277  tc1 = tcp_input_lookup_buffer (b[1], thread_index, &error1, is_ip4);
3278 
3279  if (PREDICT_TRUE (!tc0 + !tc1 == 0))
3280  {
3281  ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3282  ASSERT (tcp_lookup_is_valid (tc1, tcp_buffer_hdr (b[1])));
3283 
3284  vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3285  vnet_buffer (b[1])->tcp.connection_index = tc1->c_c_index;
3286 
3287  tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3288  tcp_input_dispatch_buffer (tm, tc1, b[1], &next[1], &error1);
3289  }
3290  else
3291  {
3292  if (PREDICT_TRUE (tc0 != 0))
3293  {
3294  ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3295  vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3296  tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3297  }
3298  else
3299  tcp_input_set_error_next (tm, &next[0], &error0, is_ip4);
3300 
3301  if (PREDICT_TRUE (tc1 != 0))
3302  {
3303  ASSERT (tcp_lookup_is_valid (tc1, tcp_buffer_hdr (b[1])));
3304  vnet_buffer (b[1])->tcp.connection_index = tc1->c_c_index;
3305  tcp_input_dispatch_buffer (tm, tc1, b[1], &next[1], &error1);
3306  }
3307  else
3308  tcp_input_set_error_next (tm, &next[1], &error1, is_ip4);
3309  }
3310 
3311  b += 2;
3312  next += 2;
3313  n_left_from -= 2;
3314  }
3315  while (n_left_from > 0)
3316  {
3317  tcp_connection_t *tc0;
3318  u32 error0 = TCP_ERROR_NO_LISTENER;
3319 
3320  if (n_left_from > 1)
3321  {
3322  vlib_prefetch_buffer_header (b[1], STORE);
3323  CLIB_PREFETCH (b[1]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3324  }
3325 
3326  next[0] = TCP_INPUT_NEXT_DROP;
3327  tc0 = tcp_input_lookup_buffer (b[0], thread_index, &error0, is_ip4);
3328  if (PREDICT_TRUE (tc0 != 0))
3329  {
3330  ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3331  vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3332  tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3333  }
3334  else
3335  tcp_input_set_error_next (tm, &next[0], &error0, is_ip4);
3336 
3337  b += 1;
3338  next += 1;
3339  n_left_from -= 1;
3340  }
3341 
3343  tcp_input_trace_frame (vm, node, bufs, frame->n_vectors, is_ip4);
3344 
3345  vlib_buffer_enqueue_to_next (vm, node, from, nexts, frame->n_vectors);
3346  return frame->n_vectors;
3347 }
3348 
3349 static uword
3351  vlib_frame_t * from_frame)
3352 {
3353  return tcp46_input_inline (vm, node, from_frame, 1 /* is_ip4 */ );
3354 }
3355 
3356 static uword
3358  vlib_frame_t * from_frame)
3359 {
3360  return tcp46_input_inline (vm, node, from_frame, 0 /* is_ip4 */ );
3361 }
3362 
3363 /* *INDENT-OFF* */
3365 {
3366  .function = tcp4_input,
3367  .name = "tcp4-input",
3368  /* Takes a vector of packets. */
3369  .vector_size = sizeof (u32),
3370  .n_errors = TCP_N_ERROR,
3371  .error_strings = tcp_error_strings,
3372  .n_next_nodes = TCP_INPUT_N_NEXT,
3373  .next_nodes =
3374  {
3375 #define _(s,n) [TCP_INPUT_NEXT_##s] = n,
3377 #undef _
3378  },
3379  .format_buffer = format_tcp_header,
3380  .format_trace = format_tcp_rx_trace,
3381 };
3382 /* *INDENT-ON* */
3383 
3385 
3386 /* *INDENT-OFF* */
3388 {
3389  .function = tcp6_input,
3390  .name = "tcp6-input",
3391  /* Takes a vector of packets. */
3392  .vector_size = sizeof (u32),
3393  .n_errors = TCP_N_ERROR,
3394  .error_strings = tcp_error_strings,
3395  .n_next_nodes = TCP_INPUT_N_NEXT,
3396  .next_nodes =
3397  {
3398 #define _(s,n) [TCP_INPUT_NEXT_##s] = n,
3400 #undef _
3401  },
3402  .format_buffer = format_tcp_header,
3403  .format_trace = format_tcp_rx_trace,
3404 };
3405 /* *INDENT-ON* */
3406 
3408 
3409 static void
3411 {
3412  int i, j;
3413  for (i = 0; i < ARRAY_LEN (tm->dispatch_table); i++)
3414  for (j = 0; j < ARRAY_LEN (tm->dispatch_table[i]); j++)
3415  {
3416  tm->dispatch_table[i][j].next = TCP_INPUT_NEXT_DROP;
3417  tm->dispatch_table[i][j].error = TCP_ERROR_DISPATCH;
3418  }
3419 
3420 #define _(t,f,n,e) \
3421 do { \
3422  tm->dispatch_table[TCP_STATE_##t][f].next = (n); \
3423  tm->dispatch_table[TCP_STATE_##t][f].error = (e); \
3424 } while (0)
3425 
3426  /* SYNs for new connections -> tcp-listen. */
3427  _(LISTEN, TCP_FLAG_SYN, TCP_INPUT_NEXT_LISTEN, TCP_ERROR_NONE);
3428  _(LISTEN, TCP_FLAG_ACK, TCP_INPUT_NEXT_RESET, TCP_ERROR_NONE);
3429  _(LISTEN, TCP_FLAG_RST, TCP_INPUT_NEXT_DROP, TCP_ERROR_RST_RCVD);
3431  TCP_ERROR_NONE);
3432  /* ACK for for a SYN-ACK -> tcp-rcv-process. */
3433  _(SYN_RCVD, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3434  _(SYN_RCVD, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3435  _(SYN_RCVD, TCP_FLAG_SYN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3436  /* SYN-ACK for a SYN */
3438  TCP_ERROR_NONE);
3439  _(SYN_SENT, TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE);
3440  _(SYN_SENT, TCP_FLAG_RST, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE);
3442  TCP_ERROR_NONE);
3443  /* ACK for for established connection -> tcp-established. */
3444  _(ESTABLISHED, TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3445  /* FIN for for established connection -> tcp-established. */
3446  _(ESTABLISHED, TCP_FLAG_FIN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3448  TCP_ERROR_NONE);
3449  _(ESTABLISHED, TCP_FLAG_RST, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3451  TCP_ERROR_NONE);
3452  _(ESTABLISHED, TCP_FLAG_SYN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3454  TCP_ERROR_NONE);
3455  /* ACK or FIN-ACK to our FIN */
3456  _(FIN_WAIT_1, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3458  TCP_ERROR_NONE);
3459  /* FIN in reply to our FIN from the other side */
3460  _(FIN_WAIT_1, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3461  _(FIN_WAIT_1, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3462  _(CLOSING, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3463  /* FIN confirming that the peer (app) has closed */
3464  _(FIN_WAIT_2, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3465  _(FIN_WAIT_2, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3467  TCP_ERROR_NONE);
3468  _(CLOSE_WAIT, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3470  TCP_ERROR_NONE);
3471  _(LAST_ACK, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3472  _(LAST_ACK, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3474  TCP_ERROR_NONE);
3475  _(LAST_ACK, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3476  _(TIME_WAIT, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3478  TCP_ERROR_NONE);
3479  _(TIME_WAIT, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3480  _(TIME_WAIT, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3481  _(CLOSED, TCP_FLAG_ACK, TCP_INPUT_NEXT_DROP, TCP_ERROR_CONNECTION_CLOSED);
3482  _(CLOSED, TCP_FLAG_RST, TCP_INPUT_NEXT_DROP, TCP_ERROR_CONNECTION_CLOSED);
3484  TCP_ERROR_CONNECTION_CLOSED);
3485 #undef _
3486 }
3487 
3488 static clib_error_t *
3490 {
3491  clib_error_t *error = 0;
3492  tcp_main_t *tm = vnet_get_tcp_main ();
3493 
3494  if ((error = vlib_call_init_function (vm, tcp_init)))
3495  return error;
3496 
3497  /* Initialize dispatch table. */
3499 
3500  return error;
3501 }
3502 
3504 
3505 /*
3506  * fd.io coding-style-patch-verification: ON
3507  *
3508  * Local Variables:
3509  * eval: (c-set-style "gnu")
3510  * End:
3511  */
#define tcp_in_cong_recovery(tc)
Definition: tcp.h:350
static int tcp_session_enqueue_ooo(tcp_connection_t *tc, vlib_buffer_t *b, u16 data_len)
Enqueue out-of-order data.
Definition: tcp_input.c:1549
static void tcp_update_timestamp(tcp_connection_t *tc, u32 seq, u32 seq_end)
Update tsval recent.
Definition: tcp_input.c:246
static sack_scoreboard_hole_t * scoreboard_insert_hole(sack_scoreboard_t *sb, u32 prev_index, u32 start, u32 end)
Definition: tcp_input.c:626
static u8 tcp_scoreboard_is_sane_post_recovery(tcp_connection_t *tc)
Test that scoreboard is sane after recovery.
Definition: tcp_input.c:809
void scoreboard_clear(sack_scoreboard_t *sb)
Definition: tcp_input.c:783
static u8 tcp_should_fastrecover(tcp_connection_t *tc)
Definition: tcp_input.c:1135
#define TCP_2MSL_TIME
Definition: tcp.h:101
End of options.
Definition: tcp_packet.h:104
#define tcp_next_drop(is_ip4)
Definition: tcp_input.c:79
#define tcp_fastrecovery_1_smss_off(tc)
Definition: tcp.h:348
#define clib_min(x, y)
Definition: clib.h:289
#define CLIB_UNUSED(x)
Definition: clib.h:79
static void tcp_cc_update(tcp_connection_t *tc, vlib_buffer_t *b)
Definition: tcp_input.c:1107
vlib_node_registration_t tcp6_rcv_process_node
(constructor) VLIB_REGISTER_NODE (tcp6_rcv_process_node)
Definition: tcp_input.c:2433
#define tcp_in_recovery(tc)
Definition: tcp.h:344
static void tcp_retransmit_timer_set(tcp_connection_t *tc)
Definition: tcp.h:736
#define TCP_OPTION_LEN_SACK_PERMITTED
Definition: tcp_packet.h:167
static int tcp_rcv_ack_is_acceptable(tcp_connection_t *tc0, vlib_buffer_t *tb0)
Definition: tcp_input.c:391
#define seq_leq(_s1, _s2)
Definition: tcp.h:547
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_rcv_sacks(tcp_connection_t *tc, u32 ack)
Definition: tcp_input.c:817
#define timestamp_leq(_t1, _t2)
Definition: tcp.h:554
ip4_address_t src_address
Definition: ip4_packet.h:169
static u8 tcp_cc_is_spurious_retransmit(tcp_connection_t *tc)
Definition: tcp_input.c:1077
enum _tcp_state_next tcp_state_next_t
struct _transport_connection transport_connection_t
#define tcp_rst(_th)
Definition: tcp_packet.h:81
#define TCP_TIMEWAIT_TIME
Definition: tcp.h:103
static uword tcp6_input(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:3357
Selective Ack permitted.
Definition: tcp_packet.h:108
#define TCP_FLAG_SYN
Definition: fa_node.h:13
#define tcp_opts_tstamp(_to)
Definition: tcp_packet.h:157
#define PREDICT_TRUE(x)
Definition: clib.h:106
void tcp_fast_retransmit(tcp_connection_t *tc)
Do fast retransmit.
Definition: tcp_output.c:1805
tcp_connection_t * tcp_connection_new(u8 thread_index)
Definition: tcp.c:248
#define tcp_inc_err_counter(cnts, err, val)
Definition: tcp_input.c:1831
unsigned long u64
Definition: types.h:89
#define tcp_store_err_counters(node_id, cnts)
Definition: tcp_input.c:1835
static void tcp_dispatch_table_init(tcp_main_t *tm)
Definition: tcp_input.c:3410
static int ip4_header_bytes(ip4_header_t *i)
Definition: ip4_packet.h:234
static u8 * format_tcp_rx_trace_short(u8 *s, va_list *args)
Definition: tcp_input.c:1743
struct _sack_scoreboard sack_scoreboard_t
static uword tcp46_established_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame, int is_ip4)
Definition: tcp_input.c:1845
static tcp_connection_t * tcp_half_open_connection_get(u32 conn_index)
Definition: tcp.h:515
void tcp_update_rto(tcp_connection_t *tc)
Definition: tcp_input.c:431
static int tcp_segment_rcv(tcp_connection_t *tc, vlib_buffer_t *b, u32 *next0)
Receive buffer for connection and handle acks.
Definition: tcp_input.c:1652
#define tcp_doff(_th)
Definition: tcp_packet.h:78
struct _tcp_main tcp_main_t
u32 thread_index
Definition: main.h:176
void tcp_connection_timers_reset(tcp_connection_t *tc)
Stop all connection timers.
Definition: tcp.c:401
static int tcp_segment_validate(vlib_main_t *vm, tcp_connection_t *tc0, vlib_buffer_t *b0, tcp_header_t *th0, u32 *next0, u32 *error0)
Validate incoming segment as per RFC793 p.
Definition: tcp_input.c:274
#define vec_add1(V, E)
Add 1 element to end of vector (unspecified alignment).
Definition: vec.h:523
static int tcp_options_parse(tcp_header_t *th, tcp_options_t *to)
Parse TCP header options.
Definition: tcp_input.c:127
#define tcp_recovery_off(tc)
Definition: tcp.h:342
#define clib_abs(x)
Definition: clib.h:296
static int tcp_update_rtt(tcp_connection_t *tc, u32 ack)
Update RTT estimate and RTO timer.
Definition: tcp_input.c:449
#define vec_add2(V, P, N)
Add N elements to end of vector V, return pointer to new elements in P.
Definition: vec.h:562
int i
static u32 format_get_indent(u8 *s)
Definition: format.h:72
vlib_node_registration_t tcp4_rcv_process_node
(constructor) VLIB_REGISTER_NODE (tcp4_rcv_process_node)
Definition: tcp_input.c:2432
struct _tcp_connection tcp_connection_t
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:419
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
#define tcp_opts_sack(_to)
Definition: tcp_packet.h:159
void tcp_fast_retransmit_sack(tcp_connection_t *tc)
Do fast retransmit with SACKs.
Definition: tcp_output.c:1685
tcp_connection_t tcp_connection
Definition: tcp_input.c:1723
static u8 tcp_sack_vector_is_sane(sack_block_t *sacks)
Definition: tcp_input.c:1407
static tcp_connection_t * tcp_get_connection_from_transport(transport_connection_t *tconn)
Definition: tcp.h:488
static void tcp_cc_congestion_undo(tcp_connection_t *tc)
Definition: tcp_input.c:1063
int session_enqueue_stream_connection(transport_connection_t *tc, vlib_buffer_t *b, u32 offset, u8 queue_event, u8 is_in_order)
Definition: session.c:346
u64 session_lookup_half_open_handle(transport_connection_t *tc)
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
vlib_error_t * errors
Vector of errors for this node.
Definition: node.h:451
No operation.
Definition: tcp_packet.h:105
format_function_t format_tcp_flags
Definition: tcp.h:64
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:228
u8 n_sack_blocks
Number of SACKs blocks.
Definition: tcp_packet.h:152
struct _tcp_header tcp_header_t
int tcp_half_open_connection_cleanup(tcp_connection_t *tc)
Try to cleanup half-open connection.
Definition: tcp.c:170
static uword tcp6_listen(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:3029
ip6_address_t src_address
Definition: ip6_packet.h:347
unsigned char u8
Definition: types.h:56
#define tcp_inc_counter(node_id, err, count)
Definition: tcp_input.c:1823
vlib_node_registration_t tcp6_syn_sent_node
(constructor) VLIB_REGISTER_NODE (tcp6_syn_sent_node)
Definition: tcp_input.c:2029
struct _sack_scoreboard_hole sack_scoreboard_hole_t
u8 wscale
Window scale advertised.
Definition: tcp_packet.h:148
#define vec_reset_length(v)
Reset vector length to zero NULL-pointer tolerant.
static tcp_connection_t * tcp_lookup_connection(u32 fib_index, vlib_buffer_t *b, u8 thread_index, u8 is_ip4)
Lookup transport connection.
Definition: tcp_input.c:2070
static void tcp_dequeue_acked(tcp_connection_t *tc, u32 ack)
Dequeue bytes that have been acked and while at it update RTT estimates.
Definition: tcp_input.c:493
#define tcp_fastrecovery_on(tc)
Definition: tcp.h:339
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
Limit MSS.
Definition: tcp_packet.h:106
static uword tcp4_listen(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:3022
static void scoreboard_init_high_rxt(sack_scoreboard_t *sb, u32 seq)
Definition: tcp_input.c:762
sack_scoreboard_hole_t * scoreboard_get_hole(sack_scoreboard_t *sb, u32 index)
Definition: tcp_input.c:551
#define tcp_is_fin(_th)
Definition: tcp_packet.h:90
static uword tcp6_rcv_process(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:2802
static uword tcp4_syn_sent(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:2375
#define seq_gt(_s1, _s2)
Definition: tcp.h:548
static u8 * format_tcp_rx_trace(u8 *s, va_list *args)
Definition: tcp_input.c:1727
void tcp_init_snd_vars(tcp_connection_t *tc)
Initialize connection send variables.
Definition: tcp.c:536
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:156
vlib_node_registration_t tcp4_established_node
(constructor) VLIB_REGISTER_NODE (tcp4_established_node)
Definition: tcp_input.c:82
#define TCP_CLOSEWAIT_TIME
Definition: tcp.h:102
void stream_session_accept_notify(transport_connection_t *tc)
Definition: session.c:732
#define always_inline
Definition: clib.h:92
#define TCP_OPTION_LEN_SACK_BLOCK
Definition: tcp_packet.h:169
ip4_address_t dst_address
Definition: ip4_packet.h:169
#define TCP_FLAG_ACK
Definition: fa_node.h:16
u8 * format_white_space(u8 *s, va_list *va)
Definition: std-formats.c:113
#define TCP_DELACK_TIME
Definition: tcp.h:98
static tcp_header_t * tcp_buffer_hdr(vlib_buffer_t *b)
Definition: tcp.h:448
static void tcp_cc_recovery_exit(tcp_connection_t *tc)
Definition: tcp_input.c:1040
#define vlib_prefetch_buffer_header(b, type)
Prefetch buffer metadata.
Definition: buffer.h:184
enum _tcp_state tcp_state_t
#define TCP_ALWAYS_ACK
On/off delayed acks.
Definition: tcp.h:38
vlib_node_registration_t tcp6_input_node
(constructor) VLIB_REGISTER_NODE (tcp6_input_node)
Definition: tcp_input.c:3080
static u8 tcp_ack_is_dupack(tcp_connection_t *tc, vlib_buffer_t *b, u32 prev_snd_wnd, u32 prev_snd_una)
Check if duplicate ack as per RFC5681 Sec.
Definition: tcp_input.c:513
vhost_vring_state_t state
Definition: vhost_user.h:115
#define TCP_RTO_MAX
Definition: tcp.h:107
static u32 ooo_segment_length(svm_fifo_t *f, ooo_segment_t *s)
Definition: svm_fifo.h:261
static void * ip4_next_header(ip4_header_t *i)
Definition: ip4_packet.h:240
static u32 tcp_time_now(void)
Definition: tcp.h:668
sack_block_t * sacks
SACK blocks.
Definition: tcp_packet.h:151
unsigned int u32
Definition: types.h:88
#define vec_end(v)
End (last data address) of vector.
struct _stream_session_t stream_session_t
#define vlib_call_init_function(vm, x)
Definition: init.h:227
static void tcp_node_inc_counter_i(vlib_main_t *vm, u32 tcp4_node, u32 tcp6_node, u8 is_ip4, u32 evt, u32 val)
Definition: tcp_input.c:1807
#define TCP_MAX_SACK_BLOCKS
Max number of SACK blocks stored.
Definition: tcp.h:162
#define VLIB_FRAME_SIZE
Definition: node.h:364
#define tcp_validate_txf_size(_tc, _a)
Definition: tcp.h:799
#define TCP_EVT_DBG(_evt, _args...)
Definition: tcp_debug.h:238
#define timestamp_lt(_t1, _t2)
Definition: tcp.h:553
static void tcp_timer_set(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:697
#define TCP_OPTION_LEN_WINDOW_SCALE
Definition: tcp_packet.h:166
static void svm_fifo_newest_ooo_segment_reset(svm_fifo_t *f)
Definition: svm_fifo.h:184
static heap_elt_t * first(heap_header_t *h)
Definition: heap.c:59
void scoreboard_init(sack_scoreboard_t *sb)
Definition: tcp_input.c:775
u32 stream_session_dequeue_drop(transport_connection_t *tc, u32 max_bytes)
Definition: session.c:474
#define TCP_INVALID_SACK_HOLE_INDEX
Definition: tcp.h:163
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:464
u16 current_length
Nbytes between current data and the end of this buffer.
Definition: buffer.h:108
void tcp_cc_fastrecovery_exit(tcp_connection_t *tc)
Definition: tcp_input.c:1051
static uword tcp46_listen_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame, int is_ip4)
LISTEN state processing as per RFC 793 p.
Definition: tcp_input.c:2859
#define tcp_in_fastrecovery(tc)
Definition: tcp.h:343
static void tcp_input_set_error_next(tcp_main_t *tm, u16 *next, u32 *error, u8 is_ip4)
Definition: tcp_input.c:3137
void tcp_retransmit_first_unacked(tcp_connection_t *tc)
Retransmit first unacked segment.
Definition: tcp_output.c:1662
unsigned short u16
Definition: types.h:57
#define foreach_tcp4_input_next
Definition: tcp_input.c:3094
static u32 ooo_segment_offset(svm_fifo_t *f, ooo_segment_t *s)
Definition: svm_fifo.h:249
static void * vlib_buffer_get_current(vlib_buffer_t *b)
Get pointer to current data to process.
Definition: buffer.h:202
#define filter_flags
Definition: tcp_input.c:3112
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:274
static int tcp_buffer_discard_bytes(vlib_buffer_t *b, u32 n_bytes_to_drop)
Definition: tcp_input.c:1619
#define foreach_tcp6_input_next
Definition: tcp_input.c:3103
#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 void tcp_input_trace_frame(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_buffer_t **bs, u32 n_bufs, u8 is_ip4)
Definition: tcp_input.c:3115
#define PREDICT_FALSE(x)
Definition: clib.h:105
#define vec_del1(v, i)
Delete the element at index I.
Definition: vec.h:806
#define TCP_FLAG_FIN
Definition: fa_node.h:12
#define vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next, n_left_to_next, bi0, next0)
Finish enqueueing one buffer forward in the graph.
Definition: buffer_node.h:218
int stream_session_accept(transport_connection_t *tc, u32 listener_index, u8 notify)
Accept a stream session.
Definition: session.c:819
#define vlib_get_next_frame(vm, node, next_index, vectors, n_vectors_left)
Get pointer to next frame vector data by (vlib_node_runtime_t, next_index).
Definition: node_funcs.h:364
vlib_node_registration_t tcp4_listen_node
(constructor) VLIB_REGISTER_NODE (tcp4_listen_node)
Definition: tcp_input.c:2852
#define TCP_OPTION_LEN_TIMESTAMP
Definition: tcp_packet.h:168
static u8 tcp_lookup_is_valid(tcp_connection_t *tc, tcp_header_t *hdr)
Definition: tcp_input.c:2032
static ooo_segment_t * svm_fifo_newest_ooo_segment(svm_fifo_t *f)
Definition: svm_fifo.h:176
vlib_error_t error
Error code for buffers to be enqueued to error handler.
Definition: buffer.h:135
u32 tcp_sack_list_bytes(tcp_connection_t *tc)
Definition: tcp_input.c:1484
Selective Ack block.
Definition: tcp_packet.h:109
vlib_node_registration_t tcp6_established_node
(constructor) VLIB_REGISTER_NODE (tcp6_established_node)
Definition: tcp_input.c:83
sack_scoreboard_hole_t * scoreboard_first_hole(sack_scoreboard_t *sb)
Definition: tcp_input.c:575
static int tcp_can_delack(tcp_connection_t *tc)
Check if ACK could be delayed.
Definition: tcp_input.c:1603
static void vlib_node_increment_counter(vlib_main_t *vm, u32 node_index, u32 counter_index, u64 increment)
Definition: node_funcs.h:1168
static int tcp_cc_recover(tcp_connection_t *tc)
Definition: tcp_input.c:1086
#define TCP_FLAG_RST
Definition: fa_node.h:14
static stream_session_t * session_get(u32 si, u32 thread_index)
Definition: session.h:309
#define TCP_DBG(_fmt, _args...)
Definition: tcp_debug.h:89
u32 flags
Definition: vhost_user.h:110
#define TCP_MAX_WND_SCALE
Definition: tcp_packet.h:173
static void tcp_timer_reset(tcp_connection_t *tc, u8 timer_id)
Definition: tcp.h:708
static uword tcp6_syn_sent_rcv(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:2382
#define VLIB_REGISTER_NODE(x,...)
Definition: node.h:153
vlib_node_registration_t tcp4_syn_sent_node
(constructor) VLIB_REGISTER_NODE (tcp4_syn_sent_node)
Definition: tcp_input.c:2028
u16 n_vectors
Definition: node.h:380
#define CLIB_PREFETCH(addr, size, type)
Definition: cache.h:77
vlib_main_t * vm
Definition: buffer.c:294
static_always_inline void vlib_buffer_enqueue_to_next(vlib_main_t *vm, vlib_node_runtime_t *node, u32 *buffers, u16 *nexts, uword count)
Definition: buffer_node.h:332
static void tcp_set_rx_trace_data(tcp_rx_trace_t *t0, tcp_connection_t *tc0, tcp_header_t *th0, vlib_buffer_t *b0, u8 is_ip4)
Definition: tcp_input.c:1758
void tcp_send_reset(tcp_connection_t *tc)
Build and set reset packet for connection.
Definition: tcp_output.c:909
#define vec_free(V)
Free vector&#39;s memory (no header).
Definition: vec.h:339
#define TCP_DUPACK_THRESHOLD
Definition: tcp.h:34
format_function_t format_tcp_state
Definition: tcp.h:63
#define clib_warning(format, args...)
Definition: error.h:59
#define clib_memcpy(a, b, c)
Definition: string.h:75
static int tcp_rcv_ack(tcp_connection_t *tc, vlib_buffer_t *b, tcp_header_t *th, u32 *next, u32 *error)
Process incoming ACK.
Definition: tcp_input.c:1311
tcp_header_t tcp_header
Definition: tcp_input.c:1722
format_function_t format_tcp_header
Definition: format.h:107
void tcp_make_synack(tcp_connection_t *ts, vlib_buffer_t *b)
Convert buffer to SYN-ACK.
Definition: tcp_output.c:616
#define pool_is_free_index(P, I)
Use free bitmap to query whether given index is free.
Definition: pool.h:271
#define ARRAY_LEN(x)
Definition: clib.h:59
void vlib_put_next_frame(vlib_main_t *vm, vlib_node_runtime_t *r, u32 next_index, u32 n_vectors_left)
Release pointer to next frame vector data.
Definition: main.c:454
#define TCP_RTT_MAX
Definition: tcp.h:109
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
u16 mss
Option flags, see above.
Definition: tcp_packet.h:147
static void * ip6_next_header(ip6_header_t *i)
Definition: ip6_packet.h:374
void tcp_make_ack(tcp_connection_t *ts, vlib_buffer_t *b)
Convert buffer to ACK.
Definition: tcp_output.c:558
void stream_session_disconnect_notify(transport_connection_t *tc)
Notification from transport that connection is being closed.
Definition: session.c:750
transport_connection_t * session_lookup_connection_wt4(u32 fib_index, ip4_address_t *lcl, ip4_address_t *rmt, u16 lcl_port, u16 rmt_port, u8 proto, u32 thread_index, u8 *is_filtered)
Lookup connection with ip4 and transport layer information.
static void tcp_timer_update(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:721
#define TCP_PAWS_IDLE
24 days
Definition: tcp.h:30
u16 cached_next_index
Next frame index that vector arguments were last enqueued to last time this node ran.
Definition: node.h:492
#define ASSERT(truth)
#define tcp_syn(_th)
Definition: tcp_packet.h:80
static clib_error_t * tcp_input_init(vlib_main_t *vm)
Definition: tcp_input.c:3489
u32 session_tx_fifo_max_dequeue(transport_connection_t *tc)
Definition: session.c:457
static void tcp_estimate_rtt(tcp_connection_t *tc, u32 mrtt)
Compute smoothed RTT as per VJ&#39;s &#39;88 SIGCOMM and RFC6298.
Definition: tcp_input.c:408
enum _tcp_rcv_process_next tcp_rcv_process_next_t
static uword tcp4_established(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:1970
#define seq_geq(_s1, _s2)
Definition: tcp.h:549
u32 next_buffer
Next buffer for this linked-list of buffers.
Definition: buffer.h:126
static void vlib_buffer_advance(vlib_buffer_t *b, word l)
Advance current data pointer by the supplied (signed!) amount.
Definition: buffer.h:215
static int tcp_segment_check_paws(tcp_connection_t *tc)
RFC1323: Check against wrapped sequence numbers (PAWS).
Definition: tcp_input.c:236
static void tcp_cc_handle_event(tcp_connection_t *tc, u32 is_dack)
One function to rule them all ...
Definition: tcp_input.c:1145
static void tcp_established_trace_frame(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame, u8 is_ip4)
Definition: tcp_input.c:1773
enum _tcp_input_next tcp_input_next_t
transport_connection_t * session_lookup_connection_wt6(u32 fib_index, ip6_address_t *lcl, ip6_address_t *rmt, u16 lcl_port, u16 rmt_port, u8 proto, u32 thread_index, u8 *is_filtered)
Lookup connection with ip6 and transport layer information.
void tcp_update_sack_list(tcp_connection_t *tc, u32 start, u32 end)
Build SACK list as per RFC2018.
Definition: tcp_input.c:1430
Out-of-order segment.
Definition: svm_fifo.h:27
static u8 tcp_segment_in_rcv_wnd(tcp_connection_t *tc, u32 seq, u32 end_seq)
Validate segment sequence number.
Definition: tcp_input.c:113
#define clib_max(x, y)
Definition: clib.h:282
static vlib_main_t * vlib_get_main(void)
Definition: global_funcs.h:23
static clib_error_t * tcp_init(vlib_main_t *vm)
Definition: tcp.c:1356
static void * vlib_add_trace(vlib_main_t *vm, vlib_node_runtime_t *r, vlib_buffer_t *b, u32 n_data_bytes)
Definition: trace_funcs.h:55
VLIB_NODE_FUNCTION_MULTIARCH(tcp4_established_node, tcp4_established)
static uword tcp6_established(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:1977
u32 total_length_not_including_first_buffer
Only valid for first buffer in chain.
Definition: buffer.h:152
#define seq_lt(_s1, _s2)
Definition: tcp.h:546
struct _vlib_node_registration vlib_node_registration_t
#define tcp_is_syn(_th)
Definition: tcp_packet.h:89
#define tcp_opts_wscale(_to)
Definition: tcp_packet.h:158
enum _tcp_syn_sent_next tcp_syn_sent_next_t
static void tcp_update_snd_wnd(tcp_connection_t *tc, u32 seq, u32 ack, u32 snd_wnd)
Try to update snd_wnd based on feedback received from peer.
Definition: tcp_input.c:992
void tcp_connection_reset(tcp_connection_t *tc)
Notify session that connection has been reset.
Definition: tcp.c:265
u32 tsval
Timestamp value.
Definition: tcp_packet.h:149
enum _tcp_established_next tcp_established_next_t
u16 payload_length
Definition: ip6_packet.h:338
u32 tsecr
Echoed/reflected time stamp.
Definition: tcp_packet.h:150
vlib_node_registration_t tcp4_input_node
(constructor) VLIB_REGISTER_NODE (tcp4_input_node)
Definition: tcp_input.c:3079
void tcp_send_fin(tcp_connection_t *tc)
Send FIN.
Definition: tcp_output.c:1070
#define vec_len(v)
Number of elements in vector (rvalue-only, NULL tolerant)
enum _tcp_listen_next tcp_listen_next_t
#define foreach_tcp_state_next
Definition: tcp_input.c:29
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
static uword tcp4_rcv_process(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:2795
static u32 scoreboard_hole_bytes(sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:545
static void tcp_retransmit_timer_update(tcp_connection_t *tc)
Definition: tcp.h:780
static int tcp_session_enqueue_data(tcp_connection_t *tc, vlib_buffer_t *b, u16 data_len)
Enqueue data for delivery to application.
Definition: tcp_input.c:1494
static u8 tcp_should_fastrecover_sack(tcp_connection_t *tc)
Definition: tcp_input.c:1129
u64 uword
Definition: types.h:112
#define seq_max(_s1, _s2)
Definition: tcp.h:550
sack_scoreboard_hole_t * scoreboard_next_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:559
sack_scoreboard_hole_t * scoreboard_prev_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:567
static void * vlib_frame_vector_args(vlib_frame_t *f)
Get pointer to frame vector data.
Definition: node_funcs.h:267
void tcp_connection_init_vars(tcp_connection_t *tc)
Initialize tcp connection variables.
Definition: tcp.c:560
static void scoreboard_remove_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:591
#define tcp_maybe_inc_counter(node_id, err, count)
Definition: tcp_input.c:1816
#define TCP_OPTION_LEN_MSS
Definition: tcp_packet.h:165
sack_scoreboard_hole_t * scoreboard_last_hole(sack_scoreboard_t *sb)
Definition: tcp_input.c:583
static void scoreboard_update_bytes(tcp_connection_t *tc, sack_scoreboard_t *sb)
Definition: tcp_input.c:663
#define tcp_next_output(is_ip4)
Definition: tcp_input.c:76
#define TCP_RTO_MIN
Definition: tcp.h:108
struct clib_bihash_value offset
template key/value backing page structure
#define tcp_scoreboard_trace_add(_tc, _ack)
Definition: tcp.h:226
u8 * format_tcp_connection(u8 *s, va_list *args)
Definition: tcp.c:822
#define vnet_buffer(b)
Definition: buffer.h:359
int session_manager_flush_enqueue_events(u8 transport_proto, u32 thread_index)
Flushes queue of sessions that are to be notified of new data enqueued events.
Definition: session.c:537
static tcp_connection_t * tcp_connection_get(u32 conn_index, u32 thread_index)
Definition: tcp.h:469
#define tcp_maybe_inc_err_counter(cnts, err)
Definition: tcp_input.c:1827
static u32 scoreboard_hole_index(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:538
static int tcp_header_bytes(tcp_header_t *t)
Definition: tcp_packet.h:93
int session_stream_connect_notify(transport_connection_t *tc, u8 is_fail)
Definition: session.c:594
void tcp_connection_cleanup(tcp_connection_t *tc)
Cleans up connection state.
Definition: tcp.c:199
void tcp_connection_del(tcp_connection_t *tc)
Connection removal.
Definition: tcp.c:240
static uword tcp4_input(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:3350
u8 data[0]
Packet data.
Definition: buffer.h:172
static tcp_connection_t * tcp_input_lookup_buffer(vlib_buffer_t *b, u8 thread_index, u32 *error, u8 is_ip4)
Definition: tcp_input.c:3156
u16 flags
Copy of main node flags.
Definition: node.h:486
Window scale.
Definition: tcp_packet.h:107
vlib_node_registration_t tcp6_listen_node
(constructor) VLIB_REGISTER_NODE (tcp6_listen_node)
Definition: tcp_input.c:2853
#define tcp_opts_sack_permitted(_to)
Definition: tcp_packet.h:160
Timestamps.
Definition: tcp_packet.h:110
static_always_inline void vlib_get_buffers(vlib_main_t *vm, u32 *bi, vlib_buffer_t **b, int count)
Translate array of buffer indices into buffer pointers.
Definition: buffer_funcs.h:128
#define VLIB_NODE_FLAG_TRACE
Definition: node.h:295
#define CLIB_CACHE_LINE_BYTES
Definition: cache.h:62
#define TCP_SYN_RCVD_TIME
Definition: tcp.h:100
u32 flags
buffer flags: VLIB_BUFFER_FREE_LIST_INDEX_MASK: bits used to store free list index, VLIB_BUFFER_IS_TRACED: trace this buffer.
Definition: buffer.h:111
static uword tcp46_input_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame, int is_ip4)
Definition: tcp_input.c:3244
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 uword tcp46_syn_sent_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame, int is_ip4)
Definition: tcp_input.c:2111
#define tcp_fastrecovery_off(tc)
Definition: tcp.h:340
static uword tcp46_rcv_process_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame, int is_ip4)
Handles reception for all states except LISTEN, SYN-SENT and ESTABLISHED as per RFC793 p...
Definition: tcp_input.c:2440
static void tcp_retransmit_timer_reset(tcp_connection_t *tc)
Definition: tcp.h:744
static vlib_buffer_t * vlib_get_buffer(vlib_main_t *vm, u32 buffer_index)
Translate buffer index into buffer pointer.
Definition: buffer_funcs.h:57
static void tcp_input_dispatch_buffer(tcp_main_t *tm, tcp_connection_t *tc, vlib_buffer_t *b, u16 *next, u32 *error)
Definition: tcp_input.c:3219
#define tcp_ack(_th)
Definition: tcp_packet.h:83
static u8 tcp_timer_is_active(tcp_connection_t *tc, tcp_timers_e timer)
Definition: tcp.h:794
transport_connection_t * session_lookup_half_open_connection(u64 handle, u8 proto, u8 is_ip4)
Definition: defs.h:46
static tcp_connection_t * tcp_listener_get(u32 tli)
Definition: tcp.h:509
ip6_address_t dst_address
Definition: ip6_packet.h:347
static u8 tcp_ack_is_cc_event(tcp_connection_t *tc, vlib_buffer_t *b, u32 prev_snd_wnd, u32 prev_snd_una, u8 *is_dack)
Checks if ack is a congestion control event.
Definition: tcp_input.c:526
void tcp_cc_init_congestion(tcp_connection_t *tc)
Init loss recovery/fast recovery.
Definition: tcp_input.c:1030
static void tcp_persist_timer_reset(tcp_connection_t *tc)
Definition: tcp.h:774
static char * tcp_error_strings[]
Definition: tcp_input.c:22
static uword pool_elts(void *v)
Number of active elements in a pool.
Definition: pool.h:128