FD.io VPP  v18.07-rc0-415-g6c78436
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->snd_una_adv = 0;
827  old_sacked_bytes = sb->sacked_bytes;
828  sb->last_bytes_delivered = 0;
829 
830  if (!tcp_opts_sack (&tc->rcv_opts)
831  && sb->head == TCP_INVALID_SACK_HOLE_INDEX)
832  return;
833 
834  /* Remove invalid blocks */
835  blk = tc->rcv_opts.sacks;
836  while (blk < vec_end (tc->rcv_opts.sacks))
837  {
838  if (seq_lt (blk->start, blk->end)
839  && seq_gt (blk->start, tc->snd_una)
840  && seq_gt (blk->start, ack) && seq_leq (blk->end, tc->snd_una_max))
841  {
842  blk++;
843  continue;
844  }
845  vec_del1 (tc->rcv_opts.sacks, blk - tc->rcv_opts.sacks);
846  }
847 
848  /* Add block for cumulative ack */
849  if (seq_gt (ack, tc->snd_una))
850  {
851  tmp.start = tc->snd_una;
852  tmp.end = ack;
853  vec_add1 (tc->rcv_opts.sacks, tmp);
854  }
855 
856  if (vec_len (tc->rcv_opts.sacks) == 0)
857  return;
858 
859  tcp_scoreboard_trace_add (tc, ack);
860 
861  /* Make sure blocks are ordered */
862  for (i = 0; i < vec_len (tc->rcv_opts.sacks); i++)
863  for (j = i + 1; j < vec_len (tc->rcv_opts.sacks); j++)
864  if (seq_lt (tc->rcv_opts.sacks[j].start, tc->rcv_opts.sacks[i].start))
865  {
866  tmp = tc->rcv_opts.sacks[i];
867  tc->rcv_opts.sacks[i] = tc->rcv_opts.sacks[j];
868  tc->rcv_opts.sacks[j] = tmp;
869  }
870 
871  if (sb->head == TCP_INVALID_SACK_HOLE_INDEX)
872  {
873  /* If no holes, insert the first that covers all outstanding bytes */
875  tc->snd_una, tc->snd_una_max);
876  sb->tail = scoreboard_hole_index (sb, last_hole);
877  tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1];
878  sb->high_sacked = tmp.end;
879  }
880  else
881  {
882  /* If we have holes but snd_una_max is beyond the last hole, update
883  * last hole end */
884  tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1];
885  last_hole = scoreboard_last_hole (sb);
886  if (seq_gt (tc->snd_una_max, last_hole->end))
887  {
888  if (seq_geq (last_hole->start, sb->high_sacked))
889  {
890  last_hole->end = tc->snd_una_max;
891  }
892  /* New hole after high sacked block */
893  else if (seq_lt (sb->high_sacked, tc->snd_una_max))
894  {
895  scoreboard_insert_hole (sb, sb->tail, sb->high_sacked,
896  tc->snd_una_max);
897  }
898  }
899  /* Keep track of max byte sacked for when the last hole
900  * is acked */
901  if (seq_gt (tmp.end, sb->high_sacked))
902  sb->high_sacked = tmp.end;
903  }
904 
905  /* Walk the holes with the SACK blocks */
906  hole = pool_elt_at_index (sb->holes, sb->head);
907  while (hole && blk_index < vec_len (tc->rcv_opts.sacks))
908  {
909  blk = &tc->rcv_opts.sacks[blk_index];
910  if (seq_leq (blk->start, hole->start))
911  {
912  /* Block covers hole. Remove hole */
913  if (seq_geq (blk->end, hole->end))
914  {
915  next_hole = scoreboard_next_hole (sb, hole);
916 
917  /* Byte accounting: snd_una needs to be advanced */
918  if (blk->end == ack)
919  {
920  if (next_hole)
921  {
922  if (seq_lt (ack, next_hole->start))
923  sb->snd_una_adv = next_hole->start - ack;
924  sb->last_bytes_delivered +=
925  next_hole->start - hole->end;
926  }
927  else
928  {
929  ASSERT (seq_geq (sb->high_sacked, ack));
930  sb->snd_una_adv = sb->high_sacked - ack;
931  sb->last_bytes_delivered += sb->high_sacked - hole->end;
932  }
933  }
934 
935  scoreboard_remove_hole (sb, hole);
936  hole = next_hole;
937  }
938  /* Partial 'head' overlap */
939  else
940  {
941  if (seq_gt (blk->end, hole->start))
942  {
943  hole->start = blk->end;
944  }
945  blk_index++;
946  }
947  }
948  else
949  {
950  /* Hole must be split */
951  if (seq_lt (blk->end, hole->end))
952  {
953  hole_index = scoreboard_hole_index (sb, hole);
954  next_hole = scoreboard_insert_hole (sb, hole_index, blk->end,
955  hole->end);
956 
957  /* Pool might've moved */
958  hole = scoreboard_get_hole (sb, hole_index);
959  hole->end = blk->start;
960  blk_index++;
961  ASSERT (hole->next == scoreboard_hole_index (sb, next_hole));
962  }
963  else if (seq_lt (blk->start, hole->end))
964  {
965  hole->end = blk->start;
966  }
967  hole = scoreboard_next_hole (sb, hole);
968  }
969  }
970 
971  scoreboard_update_bytes (tc, sb);
972  sb->last_sacked_bytes = sb->sacked_bytes
973  - (old_sacked_bytes - sb->last_bytes_delivered);
974  ASSERT (sb->last_sacked_bytes <= sb->sacked_bytes || tcp_in_recovery (tc));
975  ASSERT (sb->sacked_bytes == 0
976  || sb->sacked_bytes < tc->snd_una_max - seq_max (tc->snd_una, ack));
977  ASSERT (sb->last_sacked_bytes + sb->lost_bytes <= tc->snd_una_max
978  - seq_max (tc->snd_una, ack));
980  || sb->holes[sb->head].start == ack + sb->snd_una_adv);
981  TCP_EVT_DBG (TCP_EVT_CC_SCOREBOARD, tc);
982 }
983 
984 /**
985  * Try to update snd_wnd based on feedback received from peer.
986  *
987  * If successful, and new window is 'effectively' 0, activate persist
988  * timer.
989  */
990 static void
991 tcp_update_snd_wnd (tcp_connection_t * tc, u32 seq, u32 ack, u32 snd_wnd)
992 {
993  /* If (SND.WL1 < SEG.SEQ or (SND.WL1 = SEG.SEQ and SND.WL2 =< SEG.ACK)), set
994  * SND.WND <- SEG.WND, set SND.WL1 <- SEG.SEQ, and set SND.WL2 <- SEG.ACK */
995  if (seq_lt (tc->snd_wl1, seq)
996  || (tc->snd_wl1 == seq && seq_leq (tc->snd_wl2, ack)))
997  {
998  tc->snd_wnd = snd_wnd;
999  tc->snd_wl1 = seq;
1000  tc->snd_wl2 = ack;
1001  TCP_EVT_DBG (TCP_EVT_SND_WND, tc);
1002 
1003  if (tc->snd_wnd < tc->snd_mss)
1004  {
1005  /* Set persist timer if not set and we just got 0 wnd */
1006  if (!tcp_timer_is_active (tc, TCP_TIMER_PERSIST)
1007  && !tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT))
1008  tcp_persist_timer_set (tc);
1009  }
1010  else
1011  {
1013  if (!tcp_in_recovery (tc) && tc->rto_boff > 0)
1014  {
1015  tc->rto_boff = 0;
1016  tcp_update_rto (tc);
1017  }
1018  }
1019  }
1020 }
1021 
1022 /**
1023  * Init loss recovery/fast recovery.
1024  *
1025  * Triggered by dup acks as opposed to timer timeout. Note that cwnd is
1026  * updated in @ref tcp_cc_handle_event after fast retransmit
1027  */
1028 void
1030 {
1031  tcp_fastrecovery_on (tc);
1032  tc->snd_congestion = tc->snd_una_max;
1033  tc->cwnd_acc_bytes = 0;
1034  tc->cc_algo->congestion (tc);
1035  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 4);
1036 }
1037 
1038 static void
1040 {
1041  tc->rto_boff = 0;
1042  tcp_update_rto (tc);
1043  tc->snd_rxt_ts = 0;
1044  tc->snd_nxt = tc->snd_una_max;
1045  tcp_recovery_off (tc);
1046  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3);
1047 }
1048 
1049 void
1051 {
1052  tc->cc_algo->recovered (tc);
1053  tc->snd_rxt_bytes = 0;
1054  tc->rcv_dupacks = 0;
1055  tc->snd_nxt = tc->snd_una_max;
1056  tcp_fastrecovery_off (tc);
1058  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3);
1059 }
1060 
1061 static void
1063 {
1064  tc->cwnd = tc->prev_cwnd;
1065  tc->ssthresh = tc->prev_ssthresh;
1066  tc->snd_nxt = tc->snd_una_max;
1067  tc->rcv_dupacks = 0;
1068  if (tcp_in_recovery (tc))
1069  tcp_cc_recovery_exit (tc);
1070  ASSERT (tc->rto_boff == 0);
1071  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 5);
1072  /* TODO extend for fastrecovery */
1073 }
1074 
1075 static u8
1077 {
1078  return (tcp_in_recovery (tc) && tc->rto_boff == 1
1079  && tc->snd_rxt_ts
1080  && tcp_opts_tstamp (&tc->rcv_opts)
1081  && timestamp_lt (tc->rcv_opts.tsecr, tc->snd_rxt_ts));
1082 }
1083 
1084 static int
1086 {
1089  {
1091  return 1;
1092  }
1093 
1094  if (tcp_in_recovery (tc))
1095  tcp_cc_recovery_exit (tc);
1096  else if (tcp_in_fastrecovery (tc))
1098 
1099  ASSERT (tc->rto_boff == 0);
1100  ASSERT (!tcp_in_cong_recovery (tc));
1102  return 0;
1103 }
1104 
1105 static void
1107 {
1109 
1110  /* Congestion avoidance */
1111  tc->cc_algo->rcv_ack (tc);
1112  tc->tsecr_last_ack = tc->rcv_opts.tsecr;
1113 
1114  /* If a cumulative ack, make sure dupacks is 0 */
1115  tc->rcv_dupacks = 0;
1116 
1117  /* When dupacks hits the threshold we only enter fast retransmit if
1118  * cumulative ack covers more than snd_congestion. Should snd_una
1119  * wrap this test may fail under otherwise valid circumstances.
1120  * Therefore, proactively update snd_congestion when wrap detected. */
1121  if (PREDICT_FALSE
1122  (seq_leq (tc->snd_congestion, tc->snd_una - tc->bytes_acked)
1123  && seq_gt (tc->snd_congestion, tc->snd_una)))
1124  tc->snd_congestion = tc->snd_una - 1;
1125 }
1126 
1127 static u8
1129 {
1130  return (TCP_DUPACK_THRESHOLD - 1) * tc->snd_mss < tc->sack_sb.sacked_bytes;
1131 }
1132 
1133 static u8
1135 {
1136  return (tc->rcv_dupacks == TCP_DUPACK_THRESHOLD
1137  || tcp_should_fastrecover_sack (tc));
1138 }
1139 
1140 /**
1141  * One function to rule them all ... and in the darkness bind them
1142  */
1143 static void
1145 {
1146  u32 rxt_delivered;
1147 
1148  if (tcp_in_fastrecovery (tc) && tcp_opts_sack_permitted (&tc->rcv_opts))
1149  {
1150  if (tc->bytes_acked)
1151  goto partial_ack;
1152  tcp_fast_retransmit (tc);
1153  return;
1154  }
1155  /*
1156  * Duplicate ACK. Check if we should enter fast recovery, or if already in
1157  * it account for the bytes that left the network.
1158  */
1159  else if (is_dack && !tcp_in_recovery (tc))
1160  {
1161  TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc, 1);
1162  ASSERT (tc->snd_una != tc->snd_una_max
1163  || tc->sack_sb.last_sacked_bytes);
1164 
1165  tc->rcv_dupacks++;
1166 
1167  /* Pure duplicate ack. If some data got acked, it's handled lower */
1168  if (tc->rcv_dupacks > TCP_DUPACK_THRESHOLD && !tc->bytes_acked)
1169  {
1170  ASSERT (tcp_in_fastrecovery (tc));
1171  tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK);
1172  return;
1173  }
1174  else if (tcp_should_fastrecover (tc))
1175  {
1176  ASSERT (!tcp_in_fastrecovery (tc));
1177 
1178  /* If of of the two conditions lower hold, reset dupacks because
1179  * we're probably after timeout (RFC6582 heuristics).
1180  * If Cumulative ack does not cover more than congestion threshold,
1181  * and:
1182  * 1) The following doesn't hold: The congestion window is greater
1183  * than SMSS bytes and the difference between highest_ack
1184  * and prev_highest_ack is at most 4*SMSS bytes
1185  * 2) Echoed timestamp in the last non-dup ack does not equal the
1186  * stored timestamp
1187  */
1188  if (seq_leq (tc->snd_una, tc->snd_congestion)
1189  && ((!(tc->cwnd > tc->snd_mss
1190  && tc->bytes_acked <= 4 * tc->snd_mss))
1191  || (tc->rcv_opts.tsecr != tc->tsecr_last_ack)))
1192  {
1193  tc->rcv_dupacks = 0;
1194  return;
1195  }
1196 
1198  tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK);
1199 
1200  /* The first segment MUST be retransmitted */
1202 
1203  /* Post retransmit update cwnd to ssthresh and account for the
1204  * three segments that have left the network and should've been
1205  * buffered at the receiver XXX */
1206  tc->cwnd = tc->ssthresh + tc->rcv_dupacks * tc->snd_mss;
1207  ASSERT (tc->cwnd >= tc->snd_mss);
1208 
1209  /* If cwnd allows, send more data */
1210  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1211  {
1212  scoreboard_init_high_rxt (&tc->sack_sb,
1213  tc->snd_una + tc->snd_mss);
1215  }
1216  else
1217  {
1219  }
1220  return;
1221  }
1222  else if (!tc->bytes_acked
1223  || (tc->bytes_acked && !tcp_in_cong_recovery (tc)))
1224  {
1225  tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK);
1226  return;
1227  }
1228  else
1229  goto partial_ack;
1230  }
1231 
1232  if (!tc->bytes_acked)
1233  return;
1234 
1235 partial_ack:
1236  TCP_EVT_DBG (TCP_EVT_CC_PACK, tc);
1237 
1238  /*
1239  * Legitimate ACK. 1) See if we can exit recovery
1240  */
1241  /* XXX limit this only to first partial ack? */
1243 
1244  if (seq_geq (tc->snd_una, tc->snd_congestion))
1245  {
1246  /* If spurious return, we've already updated everything */
1247  if (tcp_cc_recover (tc))
1248  {
1249  tc->tsecr_last_ack = tc->rcv_opts.tsecr;
1250  return;
1251  }
1252 
1253  tc->snd_nxt = tc->snd_una_max;
1254 
1255  /* Treat as congestion avoidance ack */
1256  tc->cc_algo->rcv_ack (tc);
1257  tc->tsecr_last_ack = tc->rcv_opts.tsecr;
1258  return;
1259  }
1260 
1261  /*
1262  * Legitimate ACK. 2) If PARTIAL ACK try to retransmit
1263  */
1264 
1265  /* RFC6675: If the incoming ACK is a cumulative acknowledgment,
1266  * reset dupacks to 0. Also needed if in congestion recovery */
1267  tc->rcv_dupacks = 0;
1268 
1269  /* Post RTO timeout don't try anything fancy */
1270  if (tcp_in_recovery (tc))
1271  {
1272  tc->cc_algo->rcv_ack (tc);
1273  tc->tsecr_last_ack = tc->rcv_opts.tsecr;
1274  return;
1275  }
1276 
1277  /* Remove retransmitted bytes that have been delivered */
1278  ASSERT (tc->bytes_acked + tc->sack_sb.snd_una_adv
1279  >= tc->sack_sb.last_bytes_delivered
1280  || (tc->flags & TCP_CONN_FINSNT));
1281 
1282  if (seq_lt (tc->snd_una, tc->sack_sb.high_rxt))
1283  {
1284  /* If we have sacks and we haven't gotten an ack beyond high_rxt,
1285  * remove sacked bytes delivered */
1286  rxt_delivered = tc->bytes_acked + tc->sack_sb.snd_una_adv
1287  - tc->sack_sb.last_bytes_delivered;
1288  ASSERT (tc->snd_rxt_bytes >= rxt_delivered);
1289  tc->snd_rxt_bytes -= rxt_delivered;
1290  }
1291  else
1292  {
1293  /* Either all retransmitted holes have been acked, or we're
1294  * "in the blind" and retransmitting segment by segment */
1295  tc->snd_rxt_bytes = 0;
1296  }
1297 
1298  tc->cc_algo->rcv_cong_ack (tc, TCP_CC_PARTIALACK);
1299 
1300  /*
1301  * Since this was a partial ack, try to retransmit some more data
1302  */
1303  tcp_fast_retransmit (tc);
1304 }
1305 
1306 /**
1307  * Process incoming ACK
1308  */
1309 static int
1311  tcp_header_t * th, u32 * next, u32 * error)
1312 {
1313  u32 prev_snd_wnd, prev_snd_una;
1314  u8 is_dack;
1315 
1316  TCP_EVT_DBG (TCP_EVT_CC_STAT, tc);
1317 
1318  /* If the ACK acks something not yet sent (SEG.ACK > SND.NXT) */
1319  if (PREDICT_FALSE (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt)))
1320  {
1321  /* When we entered recovery, we reset snd_nxt to snd_una. Seems peer
1322  * still has the data so accept the ack */
1323  if (tcp_in_recovery (tc)
1324  && seq_leq (vnet_buffer (b)->tcp.ack_number, tc->snd_congestion)
1325  && seq_geq (vnet_buffer (b)->tcp.ack_number, tc->snd_una))
1326  {
1327  tc->snd_una_max = tc->snd_nxt = vnet_buffer (b)->tcp.ack_number;
1328  goto process_ack;
1329  }
1330 
1331  /* If we have outstanding data and this is within the window, accept it,
1332  * probably retransmit has timed out. Otherwise ACK segment and then
1333  * drop it */
1334  if (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_una_max))
1335  {
1336  tcp_make_ack (tc, b);
1337  *next = tcp_next_output (tc->c_is_ip4);
1338  *error = TCP_ERROR_ACK_INVALID;
1339  TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 0,
1340  vnet_buffer (b)->tcp.ack_number);
1341  return -1;
1342  }
1343 
1344  TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 2,
1345  vnet_buffer (b)->tcp.ack_number);
1346 
1347  tc->snd_nxt = vnet_buffer (b)->tcp.ack_number;
1348  *error = TCP_ERROR_ACK_FUTURE;
1349  }
1350 
1351  /* If old ACK, probably it's an old dupack */
1352  if (PREDICT_FALSE (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una)))
1353  {
1354  *error = TCP_ERROR_ACK_OLD;
1355  TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 1,
1356  vnet_buffer (b)->tcp.ack_number);
1357  if (tcp_in_fastrecovery (tc) && tc->rcv_dupacks == TCP_DUPACK_THRESHOLD)
1358  tcp_cc_handle_event (tc, 1);
1359  /* Don't drop yet */
1360  return 0;
1361  }
1362 
1363  /*
1364  * Looks okay, process feedback
1365  */
1366 process_ack:
1367  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1368  tcp_rcv_sacks (tc, vnet_buffer (b)->tcp.ack_number);
1369 
1370  prev_snd_wnd = tc->snd_wnd;
1371  prev_snd_una = tc->snd_una;
1372  tcp_update_snd_wnd (tc, vnet_buffer (b)->tcp.seq_number,
1373  vnet_buffer (b)->tcp.ack_number,
1374  clib_net_to_host_u16 (th->window) << tc->snd_wscale);
1375  tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una;
1376  tc->snd_una = vnet_buffer (b)->tcp.ack_number + tc->sack_sb.snd_una_adv;
1377  tcp_validate_txf_size (tc, tc->bytes_acked);
1378 
1379  if (tc->bytes_acked)
1380  tcp_dequeue_acked (tc, vnet_buffer (b)->tcp.ack_number);
1381 
1382  TCP_EVT_DBG (TCP_EVT_ACK_RCVD, tc);
1383 
1384  /*
1385  * Check if we have congestion event
1386  */
1387 
1388  if (tcp_ack_is_cc_event (tc, b, prev_snd_wnd, prev_snd_una, &is_dack))
1389  {
1390  tcp_cc_handle_event (tc, is_dack);
1391  if (!tcp_in_cong_recovery (tc))
1392  return 0;
1393  *error = TCP_ERROR_ACK_DUP;
1394  return vnet_buffer (b)->tcp.data_len ? 0 : -1;
1395  }
1396 
1397  /*
1398  * Update congestion control (slow start/congestion avoidance)
1399  */
1400  tcp_cc_update (tc, b);
1401  *error = TCP_ERROR_ACK_OK;
1402  return 0;
1403 }
1404 
1405 static u8
1407 {
1408  int i;
1409  for (i = 1; i < vec_len (sacks); i++)
1410  {
1411  if (sacks[i - 1].end == sacks[i].start)
1412  return 0;
1413  }
1414  return 1;
1415 }
1416 
1417 /**
1418  * Build SACK list as per RFC2018.
1419  *
1420  * Makes sure the first block contains the segment that generated the current
1421  * ACK and the following ones are the ones most recently reported in SACK
1422  * blocks.
1423  *
1424  * @param tc TCP connection for which the SACK list is updated
1425  * @param start Start sequence number of the newest SACK block
1426  * @param end End sequence of the newest SACK block
1427  */
1428 void
1430 {
1431  sack_block_t *new_list = 0, *block = 0;
1432  int i;
1433 
1434  /* If the first segment is ooo add it to the list. Last write might've moved
1435  * rcv_nxt over the first segment. */
1436  if (seq_lt (tc->rcv_nxt, start))
1437  {
1438  vec_add2 (new_list, block, 1);
1439  block->start = start;
1440  block->end = end;
1441  }
1442 
1443  /* Find the blocks still worth keeping. */
1444  for (i = 0; i < vec_len (tc->snd_sacks); i++)
1445  {
1446  /* Discard if rcv_nxt advanced beyond current block */
1447  if (seq_leq (tc->snd_sacks[i].start, tc->rcv_nxt))
1448  continue;
1449 
1450  /* Merge or drop if segment overlapped by the new segment */
1451  if (block && (seq_geq (tc->snd_sacks[i].end, new_list[0].start)
1452  && seq_leq (tc->snd_sacks[i].start, new_list[0].end)))
1453  {
1454  if (seq_lt (tc->snd_sacks[i].start, new_list[0].start))
1455  new_list[0].start = tc->snd_sacks[i].start;
1456  if (seq_lt (new_list[0].end, tc->snd_sacks[i].end))
1457  new_list[0].end = tc->snd_sacks[i].end;
1458  continue;
1459  }
1460 
1461  /* Save to new SACK list if we have space. */
1462  if (vec_len (new_list) < TCP_MAX_SACK_BLOCKS)
1463  {
1464  vec_add1 (new_list, tc->snd_sacks[i]);
1465  }
1466  else
1467  {
1468  clib_warning ("sack discarded");
1469  }
1470  }
1471 
1472  ASSERT (vec_len (new_list) <= TCP_MAX_SACK_BLOCKS);
1473 
1474  /* Replace old vector with new one */
1475  vec_free (tc->snd_sacks);
1476  tc->snd_sacks = new_list;
1477 
1478  /* Segments should not 'touch' */
1479  ASSERT (tcp_sack_vector_is_sane (tc->snd_sacks));
1480 }
1481 
1482 u32
1484 {
1485  u32 bytes = 0, i;
1486  for (i = 0; i < vec_len (tc->snd_sacks); i++)
1487  bytes += tc->snd_sacks[i].end - tc->snd_sacks[i].start;
1488  return bytes;
1489 }
1490 
1491 /** Enqueue data for delivery to application */
1492 static int
1494  u16 data_len)
1495 {
1496  int written, error = TCP_ERROR_ENQUEUED;
1497 
1498  ASSERT (seq_geq (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt));
1499  ASSERT (data_len);
1500  written = session_enqueue_stream_connection (&tc->connection, b, 0,
1501  1 /* queue event */ , 1);
1502 
1503  TCP_EVT_DBG (TCP_EVT_INPUT, tc, 0, data_len, written);
1504 
1505  /* Update rcv_nxt */
1506  if (PREDICT_TRUE (written == data_len))
1507  {
1508  tc->rcv_nxt += written;
1509  }
1510  /* If more data written than expected, account for out-of-order bytes. */
1511  else if (written > data_len)
1512  {
1513  tc->rcv_nxt += written;
1514 
1515  /* Send ACK confirming the update */
1516  tc->flags |= TCP_CONN_SNDACK;
1517  TCP_EVT_DBG (TCP_EVT_CC_INPUT, tc, data_len, written);
1518  }
1519  else if (written > 0)
1520  {
1521  /* We've written something but FIFO is probably full now */
1522  tc->rcv_nxt += written;
1523 
1524  /* Depending on how fast the app is, all remaining buffers in burst will
1525  * not be enqueued. Inform peer */
1526  tc->flags |= TCP_CONN_SNDACK;
1527 
1528  error = TCP_ERROR_PARTIALLY_ENQUEUED;
1529  }
1530  else
1531  {
1532  tc->flags |= TCP_CONN_SNDACK;
1533  return TCP_ERROR_FIFO_FULL;
1534  }
1535 
1536  /* Update SACK list if need be */
1537  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1538  {
1539  /* Remove SACK blocks that have been delivered */
1540  tcp_update_sack_list (tc, tc->rcv_nxt, tc->rcv_nxt);
1541  }
1542 
1543  return error;
1544 }
1545 
1546 /** Enqueue out-of-order data */
1547 static int
1549  u16 data_len)
1550 {
1551  stream_session_t *s0;
1552  int rv, offset;
1553 
1554  ASSERT (seq_gt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt));
1555  ASSERT (data_len);
1556 
1557  /* Enqueue out-of-order data with relative offset */
1558  rv = session_enqueue_stream_connection (&tc->connection, b,
1559  vnet_buffer (b)->tcp.seq_number -
1560  tc->rcv_nxt, 0 /* queue event */ ,
1561  0);
1562 
1563  /* Nothing written */
1564  if (rv)
1565  {
1566  TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, 0);
1567  return TCP_ERROR_FIFO_FULL;
1568  }
1569 
1570  TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, data_len);
1571 
1572  /* Update SACK list if in use */
1573  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1574  {
1575  ooo_segment_t *newest;
1576  u32 start, end;
1577 
1578  s0 = session_get (tc->c_s_index, tc->c_thread_index);
1579 
1580  /* Get the newest segment from the fifo */
1581  newest = svm_fifo_newest_ooo_segment (s0->server_rx_fifo);
1582  if (newest)
1583  {
1584  offset = ooo_segment_offset (s0->server_rx_fifo, newest);
1585  ASSERT (offset <= vnet_buffer (b)->tcp.seq_number - tc->rcv_nxt);
1586  start = tc->rcv_nxt + offset;
1587  end = start + ooo_segment_length (s0->server_rx_fifo, newest);
1588  tcp_update_sack_list (tc, start, end);
1589  svm_fifo_newest_ooo_segment_reset (s0->server_rx_fifo);
1590  TCP_EVT_DBG (TCP_EVT_CC_SACKS, tc);
1591  }
1592  }
1593 
1594  return TCP_ERROR_ENQUEUED_OOO;
1595 }
1596 
1597 /**
1598  * Check if ACK could be delayed. If ack can be delayed, it should return
1599  * true for a full frame. If we're always acking return 0.
1600  */
1601 always_inline int
1603 {
1604  /* Send ack if ... */
1605  if (TCP_ALWAYS_ACK
1606  /* just sent a rcv wnd 0 */
1607  || (tc->flags & TCP_CONN_SENT_RCV_WND0) != 0
1608  /* constrained to send ack */
1609  || (tc->flags & TCP_CONN_SNDACK) != 0
1610  /* we're almost out of tx wnd */
1611  || tcp_available_cc_snd_space (tc) < 4 * tc->snd_mss)
1612  return 0;
1613 
1614  return 1;
1615 }
1616 
1617 static int
1619 {
1620  u32 discard, first = b->current_length;
1622 
1623  /* Handle multi-buffer segments */
1624  if (n_bytes_to_drop > b->current_length)
1625  {
1626  if (!(b->flags & VLIB_BUFFER_NEXT_PRESENT))
1627  return -1;
1628  do
1629  {
1630  discard = clib_min (n_bytes_to_drop, b->current_length);
1631  vlib_buffer_advance (b, discard);
1632  b = vlib_get_buffer (vm, b->next_buffer);
1633  n_bytes_to_drop -= discard;
1634  }
1635  while (n_bytes_to_drop);
1636  if (n_bytes_to_drop > first)
1637  b->total_length_not_including_first_buffer -= n_bytes_to_drop - first;
1638  }
1639  else
1640  vlib_buffer_advance (b, n_bytes_to_drop);
1641  vnet_buffer (b)->tcp.data_len -= n_bytes_to_drop;
1642  return 0;
1643 }
1644 
1645 /**
1646  * Receive buffer for connection and handle acks
1647  *
1648  * It handles both in order or out-of-order data.
1649  */
1650 static int
1652 {
1653  u32 error, n_bytes_to_drop, n_data_bytes;
1654 
1655  vlib_buffer_advance (b, vnet_buffer (b)->tcp.data_offset);
1656  n_data_bytes = vnet_buffer (b)->tcp.data_len;
1657  ASSERT (n_data_bytes);
1658 
1659  /* Handle out-of-order data */
1660  if (PREDICT_FALSE (vnet_buffer (b)->tcp.seq_number != tc->rcv_nxt))
1661  {
1662  /* Old sequence numbers allowed through because they overlapped
1663  * the rx window */
1664  if (seq_lt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt))
1665  {
1666  /* Completely in the past (possible retransmit). Ack
1667  * retransmissions since we may not have any data to send */
1668  if (seq_leq (vnet_buffer (b)->tcp.seq_end, tc->rcv_nxt))
1669  {
1670  tcp_make_ack (tc, b);
1671  error = TCP_ERROR_SEGMENT_OLD;
1672  *next0 = tcp_next_output (tc->c_is_ip4);
1673  goto done;
1674  }
1675 
1676  /* Chop off the bytes in the past and see if what is left
1677  * can be enqueued in order */
1678  n_bytes_to_drop = tc->rcv_nxt - vnet_buffer (b)->tcp.seq_number;
1679  n_data_bytes -= n_bytes_to_drop;
1680  vnet_buffer (b)->tcp.seq_number = tc->rcv_nxt;
1681  if (tcp_buffer_discard_bytes (b, n_bytes_to_drop))
1682  {
1683  error = TCP_ERROR_SEGMENT_OLD;
1684  *next0 = tcp_next_drop (tc->c_is_ip4);
1685  goto done;
1686  }
1687  goto in_order;
1688  }
1689 
1690  /* RFC2581: Enqueue and send DUPACK for fast retransmit */
1691  error = tcp_session_enqueue_ooo (tc, b, n_data_bytes);
1692  *next0 = tcp_next_output (tc->c_is_ip4);
1693  tcp_make_ack (tc, b);
1694  vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_DUPACK;
1695  TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc, vnet_buffer (b)->tcp);
1696  goto done;
1697  }
1698 
1699 in_order:
1700 
1701  /* In order data, enqueue. Fifo figures out by itself if any out-of-order
1702  * segments can be enqueued after fifo tail offset changes. */
1703  error = tcp_session_enqueue_data (tc, b, n_data_bytes);
1704  if (tcp_can_delack (tc))
1705  {
1706  *next0 = tcp_next_drop (tc->c_is_ip4);
1707  if (!tcp_timer_is_active (tc, TCP_TIMER_DELACK))
1708  tcp_timer_set (tc, TCP_TIMER_DELACK, TCP_DELACK_TIME);
1709  goto done;
1710  }
1711 
1712  *next0 = tcp_next_output (tc->c_is_ip4);
1713  tcp_make_ack (tc, b);
1714 
1715 done:
1716  return error;
1717 }
1718 
1719 typedef struct
1720 {
1723 } tcp_rx_trace_t;
1724 
1725 static u8 *
1726 format_tcp_rx_trace (u8 * s, va_list * args)
1727 {
1728  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1729  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1730  tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *);
1731  u32 indent = format_get_indent (s);
1732 
1733  s = format (s, "%U\n%U%U",
1734  format_tcp_header, &t->tcp_header, 128,
1735  format_white_space, indent,
1737 
1738  return s;
1739 }
1740 
1741 static u8 *
1742 format_tcp_rx_trace_short (u8 * s, va_list * args)
1743 {
1744  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1745  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1746  tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *);
1747 
1748  s = format (s, "%d -> %d (%U)",
1749  clib_net_to_host_u16 (t->tcp_header.dst_port),
1750  clib_net_to_host_u16 (t->tcp_header.src_port), format_tcp_state,
1751  t->tcp_connection.state);
1752 
1753  return s;
1754 }
1755 
1756 static void
1758  tcp_header_t * th0, vlib_buffer_t * b0, u8 is_ip4)
1759 {
1760  if (tc0)
1761  {
1762  clib_memcpy (&t0->tcp_connection, tc0, sizeof (t0->tcp_connection));
1763  }
1764  else
1765  {
1766  th0 = tcp_buffer_hdr (b0);
1767  }
1768  clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header));
1769 }
1770 
1771 static void
1773  vlib_frame_t * frame, u8 is_ip4)
1774 {
1775  u32 *from, n_left;
1776 
1777  n_left = frame->n_vectors;
1778  from = vlib_frame_vector_args (frame);
1779 
1780  while (n_left >= 1)
1781  {
1782  tcp_connection_t *tc0;
1783  tcp_rx_trace_t *t0;
1784  tcp_header_t *th0;
1785  vlib_buffer_t *b0;
1786  u32 bi0;
1787 
1788  bi0 = from[0];
1789  b0 = vlib_get_buffer (vm, bi0);
1790 
1791  if (b0->flags & VLIB_BUFFER_IS_TRACED)
1792  {
1793  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
1794  tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
1795  vm->thread_index);
1796  th0 = tcp_buffer_hdr (b0);
1797  tcp_set_rx_trace_data (t0, tc0, th0, b0, is_ip4);
1798  }
1799 
1800  from += 1;
1801  n_left -= 1;
1802  }
1803 }
1804 
1805 always_inline void
1806 tcp_node_inc_counter_i (vlib_main_t * vm, u32 tcp4_node, u32 tcp6_node,
1807  u8 is_ip4, u32 evt, u32 val)
1808 {
1809  if (is_ip4)
1810  vlib_node_increment_counter (vm, tcp4_node, evt, val);
1811  else
1812  vlib_node_increment_counter (vm, tcp6_node, evt, val);
1813 }
1814 
1815 #define tcp_maybe_inc_counter(node_id, err, count) \
1816 { \
1817  if (next0 != tcp_next_drop (is_ip4)) \
1818  tcp_node_inc_counter_i (vm, tcp4_##node_id##_node.index, \
1819  tcp6_##node_id##_node.index, is_ip4, err, \
1820  1); \
1821 }
1822 #define tcp_inc_counter(node_id, err, count) \
1823  tcp_node_inc_counter_i (vm, tcp4_##node_id##_node.index, \
1824  tcp6_##node_id##_node.index, is_ip4, \
1825  err, count)
1826 #define tcp_maybe_inc_err_counter(cnts, err) \
1827 { \
1828  cnts[err] += (next0 != tcp_next_drop (is_ip4)); \
1829 }
1830 #define tcp_inc_err_counter(cnts, err, val) \
1831 { \
1832  cnts[err] += val; \
1833 }
1834 #define tcp_store_err_counters(node_id, cnts) \
1835 { \
1836  int i; \
1837  for (i = 0; i < TCP_N_ERROR; i++) \
1838  if (cnts[i]) \
1839  tcp_inc_counter(node_id, i, cnts[i]); \
1840 }
1841 
1842 
1845  vlib_frame_t * frame, int is_ip4)
1846 {
1847  u32 thread_index = vm->thread_index, errors = 0;
1848  u32 n_left_from, next_index, *from, *to_next;
1849  u16 err_counters[TCP_N_ERROR] = { 0 };
1850  u8 is_fin = 0;
1851 
1852  if (node->flags & VLIB_NODE_FLAG_TRACE)
1853  tcp_established_trace_frame (vm, node, frame, is_ip4);
1854 
1855  from = vlib_frame_vector_args (frame);
1856  n_left_from = frame->n_vectors;
1857  next_index = node->cached_next_index;
1858 
1859  while (n_left_from > 0)
1860  {
1861  u32 n_left_to_next;
1862 
1863  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
1864  while (n_left_from > 0 && n_left_to_next > 0)
1865  {
1866  u32 bi0;
1867  vlib_buffer_t *b0;
1868  tcp_header_t *th0 = 0;
1869  tcp_connection_t *tc0;
1870  u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_ACK_OK;
1871 
1872  if (n_left_from > 1)
1873  {
1874  vlib_buffer_t *pb;
1875  pb = vlib_get_buffer (vm, from[1]);
1876  vlib_prefetch_buffer_header (pb, LOAD);
1877  CLIB_PREFETCH (pb->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
1878  }
1879 
1880  bi0 = from[0];
1881  to_next[0] = bi0;
1882  from += 1;
1883  to_next += 1;
1884  n_left_from -= 1;
1885  n_left_to_next -= 1;
1886 
1887  b0 = vlib_get_buffer (vm, bi0);
1888  tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
1889  thread_index);
1890 
1891  if (PREDICT_FALSE (tc0 == 0))
1892  {
1893  error0 = TCP_ERROR_INVALID_CONNECTION;
1894  goto done;
1895  }
1896 
1897  th0 = tcp_buffer_hdr (b0);
1898  /* N.B. buffer is rewritten if segment is ooo. Thus, th0 becomes a
1899  * dangling reference. */
1900  is_fin = tcp_is_fin (th0);
1901 
1902  /* SYNs, FINs and data consume sequence numbers */
1903  vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number
1904  + tcp_is_syn (th0) + is_fin + vnet_buffer (b0)->tcp.data_len;
1905 
1906  /* TODO header prediction fast path */
1907 
1908  /* 1-4: check SEQ, RST, SYN */
1909  if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, th0, &next0,
1910  &error0)))
1911  {
1912  tcp_maybe_inc_err_counter (err_counters, error0);
1913  TCP_EVT_DBG (TCP_EVT_SEG_INVALID, tc0, vnet_buffer (b0)->tcp);
1914  goto done;
1915  }
1916 
1917  /* 5: check the ACK field */
1918  if (PREDICT_FALSE (tcp_rcv_ack (tc0, b0, th0, &next0, &error0)))
1919  {
1920  tcp_maybe_inc_err_counter (err_counters, error0);
1921  goto done;
1922  }
1923 
1924  /* 6: check the URG bit TODO */
1925 
1926  /* 7: process the segment text */
1927  if (vnet_buffer (b0)->tcp.data_len)
1928  {
1929  error0 = tcp_segment_rcv (tc0, b0, &next0);
1930  tcp_maybe_inc_err_counter (err_counters, error0);
1931  }
1932 
1933  /* 8: check the FIN bit */
1934  if (PREDICT_FALSE (is_fin))
1935  {
1936  /* Enter CLOSE-WAIT and notify session. To avoid lingering
1937  * in CLOSE-WAIT, set timer (reuse WAITCLOSE). */
1938  /* Account for the FIN if nothing else was received */
1939  if (vnet_buffer (b0)->tcp.data_len == 0)
1940  tc0->rcv_nxt += 1;
1941  tcp_make_ack (tc0, b0);
1942  next0 = tcp_next_output (tc0->c_is_ip4);
1943  tc0->state = TCP_STATE_CLOSE_WAIT;
1944  stream_session_disconnect_notify (&tc0->connection);
1945  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME);
1946  TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc0);
1947  tcp_inc_err_counter (err_counters, TCP_ERROR_FIN_RCVD, 1);
1948  }
1949 
1950  done:
1951  b0->error = node->errors[error0];
1952  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
1953  n_left_to_next, bi0, next0);
1954  }
1955 
1956  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
1957  }
1958 
1960  thread_index);
1961  err_counters[TCP_ERROR_EVENT_FIFO_FULL] = errors;
1962  tcp_store_err_counters (established, err_counters);
1963  tcp_flush_frame_to_output (vm, thread_index, is_ip4);
1964 
1965  return frame->n_vectors;
1966 }
1967 
1968 static uword
1970  vlib_frame_t * from_frame)
1971 {
1972  return tcp46_established_inline (vm, node, from_frame, 1 /* is_ip4 */ );
1973 }
1974 
1975 static uword
1977  vlib_frame_t * from_frame)
1978 {
1979  return tcp46_established_inline (vm, node, from_frame, 0 /* is_ip4 */ );
1980 }
1981 
1982 /* *INDENT-OFF* */
1984 {
1985  .function = tcp4_established,
1986  .name = "tcp4-established",
1987  /* Takes a vector of packets. */
1988  .vector_size = sizeof (u32),
1989  .n_errors = TCP_N_ERROR,
1990  .error_strings = tcp_error_strings,
1991  .n_next_nodes = TCP_ESTABLISHED_N_NEXT,
1992  .next_nodes =
1993  {
1994 #define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n,
1996 #undef _
1997  },
1998  .format_trace = format_tcp_rx_trace_short,
1999 };
2000 /* *INDENT-ON* */
2001 
2003 
2004 /* *INDENT-OFF* */
2006 {
2007  .function = tcp6_established,
2008  .name = "tcp6-established",
2009  /* Takes a vector of packets. */
2010  .vector_size = sizeof (u32),
2011  .n_errors = TCP_N_ERROR,
2012  .error_strings = tcp_error_strings,
2013  .n_next_nodes = TCP_ESTABLISHED_N_NEXT,
2014  .next_nodes =
2015  {
2016 #define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n,
2018 #undef _
2019  },
2020  .format_trace = format_tcp_rx_trace_short,
2021 };
2022 /* *INDENT-ON* */
2023 
2024 
2026 
2029 
2030 static u8
2032 {
2033  transport_connection_t *tmp = 0;
2034  u64 handle;
2035 
2036  if (!tc)
2037  return 1;
2038 
2039  /* Proxy case */
2040  if (tc->c_lcl_port == 0 && tc->state == TCP_STATE_LISTEN)
2041  return 1;
2042 
2043  u8 is_valid = (tc->c_lcl_port == hdr->dst_port
2044  && (tc->state == TCP_STATE_LISTEN
2045  || tc->c_rmt_port == hdr->src_port));
2046 
2047  if (!is_valid)
2048  {
2049  handle = session_lookup_half_open_handle (&tc->connection);
2050  tmp = session_lookup_half_open_connection (handle & 0xFFFFFFFF,
2051  tc->c_proto, tc->c_is_ip4);
2052 
2053  if (tmp)
2054  {
2055  if (tmp->lcl_port == hdr->dst_port
2056  && tmp->rmt_port == hdr->src_port)
2057  {
2058  TCP_DBG ("half-open is valid!");
2059  }
2060  }
2061  }
2062  return is_valid;
2063 }
2064 
2065 /**
2066  * Lookup transport connection
2067  */
2068 static tcp_connection_t *
2069 tcp_lookup_connection (u32 fib_index, vlib_buffer_t * b, u8 thread_index,
2070  u8 is_ip4)
2071 {
2072  tcp_header_t *tcp;
2073  transport_connection_t *tconn;
2074  tcp_connection_t *tc;
2075  u8 is_filtered = 0;
2076  if (is_ip4)
2077  {
2078  ip4_header_t *ip4;
2079  ip4 = vlib_buffer_get_current (b);
2080  tcp = ip4_next_header (ip4);
2081  tconn = session_lookup_connection_wt4 (fib_index,
2082  &ip4->dst_address,
2083  &ip4->src_address,
2084  tcp->dst_port,
2085  tcp->src_port,
2087  thread_index, &is_filtered);
2088  tc = tcp_get_connection_from_transport (tconn);
2089  ASSERT (tcp_lookup_is_valid (tc, tcp));
2090  }
2091  else
2092  {
2093  ip6_header_t *ip6;
2094  ip6 = vlib_buffer_get_current (b);
2095  tcp = ip6_next_header (ip6);
2096  tconn = session_lookup_connection_wt6 (fib_index,
2097  &ip6->dst_address,
2098  &ip6->src_address,
2099  tcp->dst_port,
2100  tcp->src_port,
2102  thread_index, &is_filtered);
2103  tc = tcp_get_connection_from_transport (tconn);
2104  ASSERT (tcp_lookup_is_valid (tc, tcp));
2105  }
2106  return tc;
2107 }
2108 
2111  vlib_frame_t * from_frame, int is_ip4)
2112 {
2113  tcp_main_t *tm = vnet_get_tcp_main ();
2114  u32 n_left_from, next_index, *from, *to_next;
2115  u32 my_thread_index = vm->thread_index, errors = 0;
2116 
2117  from = vlib_frame_vector_args (from_frame);
2118  n_left_from = from_frame->n_vectors;
2119 
2120  next_index = node->cached_next_index;
2121 
2122  while (n_left_from > 0)
2123  {
2124  u32 n_left_to_next;
2125 
2126  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
2127 
2128  while (n_left_from > 0 && n_left_to_next > 0)
2129  {
2130  u32 bi0, ack0, seq0;
2131  vlib_buffer_t *b0;
2132  tcp_rx_trace_t *t0;
2133  tcp_header_t *tcp0 = 0;
2134  tcp_connection_t *tc0;
2135  tcp_connection_t *new_tc0;
2136  u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_ENQUEUED;
2137 
2138  bi0 = from[0];
2139  to_next[0] = bi0;
2140  from += 1;
2141  to_next += 1;
2142  n_left_from -= 1;
2143  n_left_to_next -= 1;
2144 
2145  b0 = vlib_get_buffer (vm, bi0);
2146  tc0 =
2148  tcp.connection_index);
2149  if (PREDICT_FALSE (tc0 == 0))
2150  {
2151  error0 = TCP_ERROR_INVALID_CONNECTION;
2152  goto drop;
2153  }
2154 
2155  /* Half-open completed recently but the connection was't removed
2156  * yet by the owning thread */
2157  if (PREDICT_FALSE (tc0->flags & TCP_CONN_HALF_OPEN_DONE))
2158  {
2159  /* Make sure the connection actually exists */
2160  ASSERT (tcp_lookup_connection (tc0->c_fib_index, b0,
2161  my_thread_index, is_ip4));
2162  goto drop;
2163  }
2164 
2165  ack0 = vnet_buffer (b0)->tcp.ack_number;
2166  seq0 = vnet_buffer (b0)->tcp.seq_number;
2167  tcp0 = tcp_buffer_hdr (b0);
2168 
2169  /* Crude check to see if the connection handle does not match
2170  * the packet. Probably connection just switched to established */
2171  if (PREDICT_FALSE (tcp0->dst_port != tc0->c_lcl_port
2172  || tcp0->src_port != tc0->c_rmt_port))
2173  goto drop;
2174 
2175  if (PREDICT_FALSE
2176  (!tcp_ack (tcp0) && !tcp_rst (tcp0) && !tcp_syn (tcp0)))
2177  goto drop;
2178 
2179  /* SYNs, FINs and data consume sequence numbers */
2180  vnet_buffer (b0)->tcp.seq_end = seq0 + tcp_is_syn (tcp0)
2181  + tcp_is_fin (tcp0) + vnet_buffer (b0)->tcp.data_len;
2182 
2183  /*
2184  * 1. check the ACK bit
2185  */
2186 
2187  /*
2188  * If the ACK bit is set
2189  * If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send a reset (unless
2190  * the RST bit is set, if so drop the segment and return)
2191  * <SEQ=SEG.ACK><CTL=RST>
2192  * and discard the segment. Return.
2193  * If SND.UNA =< SEG.ACK =< SND.NXT then the ACK is acceptable.
2194  */
2195  if (tcp_ack (tcp0))
2196  {
2197  if (seq_leq (ack0, tc0->iss) || seq_gt (ack0, tc0->snd_nxt))
2198  {
2199  clib_warning ("ack not in rcv wnd");
2200  if (!tcp_rst (tcp0))
2201  tcp_send_reset_w_pkt (tc0, b0, is_ip4);
2202  goto drop;
2203  }
2204 
2205  /* Make sure ACK is valid */
2206  if (seq_gt (tc0->snd_una, ack0))
2207  {
2208  clib_warning ("ack invalid");
2209  goto drop;
2210  }
2211  }
2212 
2213  /*
2214  * 2. check the RST bit
2215  */
2216 
2217  if (tcp_rst (tcp0))
2218  {
2219  /* If ACK is acceptable, signal client that peer is not
2220  * willing to accept connection and drop connection*/
2221  if (tcp_ack (tcp0))
2222  tcp_connection_reset (tc0);
2223  goto drop;
2224  }
2225 
2226  /*
2227  * 3. check the security and precedence (skipped)
2228  */
2229 
2230  /*
2231  * 4. check the SYN bit
2232  */
2233 
2234  /* No SYN flag. Drop. */
2235  if (!tcp_syn (tcp0))
2236  {
2237  clib_warning ("not synack");
2238  goto drop;
2239  }
2240 
2241  /* Parse options */
2242  if (tcp_options_parse (tcp0, &tc0->rcv_opts))
2243  {
2244  clib_warning ("options parse fail");
2245  goto drop;
2246  }
2247 
2248  /* Valid SYN or SYN-ACK. Move connection from half-open pool to
2249  * current thread pool. */
2250  pool_get (tm->connections[my_thread_index], new_tc0);
2251  clib_memcpy (new_tc0, tc0, sizeof (*new_tc0));
2252  new_tc0->c_c_index = new_tc0 - tm->connections[my_thread_index];
2253  new_tc0->c_thread_index = my_thread_index;
2254  new_tc0->rcv_nxt = vnet_buffer (b0)->tcp.seq_end;
2255  new_tc0->irs = seq0;
2256  new_tc0->timers[TCP_TIMER_ESTABLISH] = TCP_TIMER_HANDLE_INVALID;
2257  new_tc0->timers[TCP_TIMER_RETRANSMIT_SYN] =
2259  new_tc0->sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_RX];
2260 
2261  /* If this is not the owning thread, wait for syn retransmit to
2262  * expire and cleanup then */
2264  tc0->flags |= TCP_CONN_HALF_OPEN_DONE;
2265 
2266  if (tcp_opts_tstamp (&new_tc0->rcv_opts))
2267  {
2268  new_tc0->tsval_recent = new_tc0->rcv_opts.tsval;
2269  new_tc0->tsval_recent_age = tcp_time_now ();
2270  }
2271 
2272  if (tcp_opts_wscale (&new_tc0->rcv_opts))
2273  new_tc0->snd_wscale = new_tc0->rcv_opts.wscale;
2274 
2275  /* RFC1323: SYN and SYN-ACK wnd not scaled */
2276  new_tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window);
2277  new_tc0->snd_wl1 = seq0;
2278  new_tc0->snd_wl2 = ack0;
2279 
2280  tcp_connection_init_vars (new_tc0);
2281 
2282  /* SYN-ACK: See if we can switch to ESTABLISHED state */
2283  if (PREDICT_TRUE (tcp_ack (tcp0)))
2284  {
2285  /* Our SYN is ACKed: we have iss < ack = snd_una */
2286 
2287  /* TODO Dequeue acknowledged segments if we support Fast Open */
2288  new_tc0->snd_una = ack0;
2289  new_tc0->state = TCP_STATE_ESTABLISHED;
2290 
2291  /* Make sure las is initialized for the wnd computation */
2292  new_tc0->rcv_las = new_tc0->rcv_nxt;
2293 
2294  /* Notify app that we have connection. If session layer can't
2295  * allocate session send reset */
2296  if (session_stream_connect_notify (&new_tc0->connection, 0))
2297  {
2298  clib_warning ("connect notify fail");
2299  tcp_send_reset_w_pkt (new_tc0, b0, is_ip4);
2300  tcp_connection_cleanup (new_tc0);
2301  goto drop;
2302  }
2303 
2304  /* Make sure after data segment processing ACK is sent */
2305  new_tc0->flags |= TCP_CONN_SNDACK;
2306 
2307  /* Update rtt with the syn-ack sample */
2308  tcp_update_rtt (new_tc0, vnet_buffer (b0)->tcp.ack_number);
2309  TCP_EVT_DBG (TCP_EVT_SYNACK_RCVD, new_tc0);
2310  }
2311  /* SYN: Simultaneous open. Change state to SYN-RCVD and send SYN-ACK */
2312  else
2313  {
2314  new_tc0->state = TCP_STATE_SYN_RCVD;
2315 
2316  /* Notify app that we have connection */
2317  if (session_stream_connect_notify (&new_tc0->connection, 0))
2318  {
2319  tcp_connection_cleanup (new_tc0);
2320  tcp_send_reset_w_pkt (tc0, b0, is_ip4);
2321  TCP_EVT_DBG (TCP_EVT_RST_SENT, tc0);
2322  goto drop;
2323  }
2324 
2325  tc0->rtt_ts = 0;
2326  tcp_init_snd_vars (tc0);
2327  tcp_make_synack (new_tc0, b0);
2328  next0 = tcp_next_output (is_ip4);
2329 
2330  goto drop;
2331  }
2332 
2333  /* Read data, if any */
2334  if (PREDICT_FALSE (vnet_buffer (b0)->tcp.data_len))
2335  {
2336  clib_warning ("rcvd data in syn-sent");
2337  error0 = tcp_segment_rcv (new_tc0, b0, &next0);
2338  if (error0 == TCP_ERROR_ACK_OK)
2339  error0 = TCP_ERROR_SYN_ACKS_RCVD;
2340  tcp_maybe_inc_counter (syn_sent, error0, 1);
2341  }
2342  else
2343  {
2344  tcp_make_ack (new_tc0, b0);
2345  next0 = tcp_next_output (new_tc0->c_is_ip4);
2346  }
2347 
2348  drop:
2349 
2350  b0->error = error0 ? node->errors[error0] : 0;
2351  if (PREDICT_FALSE
2352  ((b0->flags & VLIB_BUFFER_IS_TRACED) && tcp0 != 0))
2353  {
2354  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
2355  clib_memcpy (&t0->tcp_header, tcp0, sizeof (t0->tcp_header));
2356  clib_memcpy (&t0->tcp_connection, tc0,
2357  sizeof (t0->tcp_connection));
2358  }
2359 
2360  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
2361  n_left_to_next, bi0, next0);
2362  }
2363 
2364  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
2365  }
2366 
2368  my_thread_index);
2369  tcp_inc_counter (syn_sent, TCP_ERROR_EVENT_FIFO_FULL, errors);
2370  return from_frame->n_vectors;
2371 }
2372 
2373 static uword
2375  vlib_frame_t * from_frame)
2376 {
2377  return tcp46_syn_sent_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2378 }
2379 
2380 static uword
2382  vlib_frame_t * from_frame)
2383 {
2384  return tcp46_syn_sent_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2385 }
2386 
2387 /* *INDENT-OFF* */
2389 {
2390  .function = tcp4_syn_sent,
2391  .name = "tcp4-syn-sent",
2392  /* Takes a vector of packets. */
2393  .vector_size = sizeof (u32),
2394  .n_errors = TCP_N_ERROR,
2395  .error_strings = tcp_error_strings,
2396  .n_next_nodes = TCP_SYN_SENT_N_NEXT,
2397  .next_nodes =
2398  {
2399 #define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n,
2401 #undef _
2402  },
2403  .format_trace = format_tcp_rx_trace_short,
2404 };
2405 /* *INDENT-ON* */
2406 
2408 
2409 /* *INDENT-OFF* */
2411 {
2412  .function = tcp6_syn_sent_rcv,
2413  .name = "tcp6-syn-sent",
2414  /* Takes a vector of packets. */
2415  .vector_size = sizeof (u32),
2416  .n_errors = TCP_N_ERROR,
2417  .error_strings = tcp_error_strings,
2418  .n_next_nodes = TCP_SYN_SENT_N_NEXT,
2419  .next_nodes =
2420  {
2421 #define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n,
2423 #undef _
2424  },
2425  .format_trace = format_tcp_rx_trace_short,
2426 };
2427 /* *INDENT-ON* */
2428 
2430 
2433 
2434 /**
2435  * Handles reception for all states except LISTEN, SYN-SENT and ESTABLISHED
2436  * as per RFC793 p. 64
2437  */
2440  vlib_frame_t * from_frame, int is_ip4)
2441 {
2442  u32 n_left_from, next_index, *from, *to_next, n_fins = 0;
2443  u32 my_thread_index = vm->thread_index, errors = 0;
2444 
2445  from = vlib_frame_vector_args (from_frame);
2446  n_left_from = from_frame->n_vectors;
2447  next_index = node->cached_next_index;
2448 
2449  while (n_left_from > 0)
2450  {
2451  u32 n_left_to_next;
2452 
2453  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
2454 
2455  while (n_left_from > 0 && n_left_to_next > 0)
2456  {
2457  u32 bi0;
2458  vlib_buffer_t *b0;
2459  tcp_header_t *tcp0 = 0;
2460  tcp_connection_t *tc0;
2461  u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_NONE;
2462  u8 is_fin0;
2463 
2464  bi0 = from[0];
2465  to_next[0] = bi0;
2466  from += 1;
2467  to_next += 1;
2468  n_left_from -= 1;
2469  n_left_to_next -= 1;
2470 
2471  b0 = vlib_get_buffer (vm, bi0);
2472  tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
2473  my_thread_index);
2474  if (PREDICT_FALSE (tc0 == 0))
2475  {
2476  error0 = TCP_ERROR_INVALID_CONNECTION;
2477  goto drop;
2478  }
2479 
2480  tcp0 = tcp_buffer_hdr (b0);
2481  is_fin0 = tcp_is_fin (tcp0);
2482 
2483  /* SYNs, FINs and data consume sequence numbers */
2484  vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number
2485  + tcp_is_syn (tcp0) + is_fin0 + vnet_buffer (b0)->tcp.data_len;
2486 
2487  if (CLIB_DEBUG)
2488  {
2489  tcp_connection_t *tmp;
2490  tmp = tcp_lookup_connection (tc0->c_fib_index, b0,
2491  my_thread_index, is_ip4);
2492  if (tmp->state != tc0->state)
2493  {
2494  clib_warning ("state changed");
2495  goto drop;
2496  }
2497  }
2498 
2499  /*
2500  * Special treatment for CLOSED
2501  */
2502  if (PREDICT_FALSE (tc0->state == TCP_STATE_CLOSED))
2503  {
2504  error0 = TCP_ERROR_CONNECTION_CLOSED;
2505  goto drop;
2506  }
2507 
2508  /*
2509  * For all other states (except LISTEN)
2510  */
2511 
2512  /* 1-4: check SEQ, RST, SYN */
2513  if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, tcp0,
2514  &next0, &error0)))
2515  {
2516  tcp_maybe_inc_counter (rcv_process, error0, 1);
2517  goto drop;
2518  }
2519 
2520  /* 5: check the ACK field */
2521  switch (tc0->state)
2522  {
2523  case TCP_STATE_SYN_RCVD:
2524  /*
2525  * If the segment acknowledgment is not acceptable, form a
2526  * reset segment,
2527  * <SEQ=SEG.ACK><CTL=RST>
2528  * and send it.
2529  */
2530  if (!tcp_rcv_ack_is_acceptable (tc0, b0))
2531  {
2532  TCP_DBG ("connection not accepted");
2533  tcp_send_reset_w_pkt (tc0, b0, is_ip4);
2534  error0 = TCP_ERROR_ACK_INVALID;
2535  goto drop;
2536  }
2537 
2538  /* Update rtt and rto */
2539  tcp_update_rtt (tc0, vnet_buffer (b0)->tcp.ack_number);
2540 
2541  /* Switch state to ESTABLISHED */
2542  tc0->state = TCP_STATE_ESTABLISHED;
2543  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2544 
2545  /* Initialize session variables */
2546  tc0->snd_una = vnet_buffer (b0)->tcp.ack_number;
2547  tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window)
2548  << tc0->rcv_opts.wscale;
2549  tc0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number;
2550  tc0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number;
2551 
2552  /* Reset SYN-ACK retransmit and SYN_RCV establish timers */
2554  tcp_timer_reset (tc0, TCP_TIMER_ESTABLISH);
2555  stream_session_accept_notify (&tc0->connection);
2556  error0 = TCP_ERROR_ACK_OK;
2557  break;
2558  case TCP_STATE_ESTABLISHED:
2559  /* We can get packets in established state here because they
2560  * were enqueued before state change */
2561  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2562  {
2563  tcp_maybe_inc_counter (rcv_process, error0, 1);
2564  goto drop;
2565  }
2566 
2567  break;
2568  case TCP_STATE_FIN_WAIT_1:
2569  /* In addition to the processing for the ESTABLISHED state, if
2570  * our FIN is now acknowledged then enter FIN-WAIT-2 and
2571  * continue processing in that state. */
2572  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2573  {
2574  tcp_maybe_inc_counter (rcv_process, error0, 1);
2575  goto drop;
2576  }
2577 
2578  /* Still have to send the FIN */
2579  if (tc0->flags & TCP_CONN_FINPNDG)
2580  {
2581  /* TX fifo finally drained */
2582  if (!session_tx_fifo_max_dequeue (&tc0->connection))
2583  tcp_send_fin (tc0);
2584  }
2585  /* If FIN is ACKed */
2586  else if (tc0->snd_una == tc0->snd_una_max)
2587  {
2588  tc0->state = TCP_STATE_FIN_WAIT_2;
2589  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2590 
2591  /* Stop all retransmit timers because we have nothing more
2592  * to send. Enable waitclose though because we're willing to
2593  * wait for peer's FIN but not indefinitely. */
2595  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
2596  }
2597  break;
2598  case TCP_STATE_FIN_WAIT_2:
2599  /* In addition to the processing for the ESTABLISHED state, if
2600  * the retransmission queue is empty, the user's CLOSE can be
2601  * acknowledged ("ok") but do not delete the TCB. */
2602  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2603  {
2604  tcp_maybe_inc_counter (rcv_process, error0, 1);
2605  goto drop;
2606  }
2607  break;
2608  case TCP_STATE_CLOSE_WAIT:
2609  /* Do the same processing as for the ESTABLISHED state. */
2610  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2611  {
2612  tcp_maybe_inc_counter (rcv_process, error0, 1);
2613  goto drop;
2614  }
2615  if (tc0->flags & TCP_CONN_FINPNDG)
2616  {
2617  /* TX fifo finally drained */
2618  if (!session_tx_fifo_max_dequeue (&tc0->connection))
2619  {
2620  tcp_send_fin (tc0);
2622  tc0->state = TCP_STATE_LAST_ACK;
2623  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE,
2624  TCP_2MSL_TIME);
2625  }
2626  }
2627  break;
2628  case TCP_STATE_CLOSING:
2629  /* In addition to the processing for the ESTABLISHED state, if
2630  * the ACK acknowledges our FIN then enter the TIME-WAIT state,
2631  * otherwise ignore the segment. */
2632  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2633  {
2634  tcp_maybe_inc_counter (rcv_process, error0, 1);
2635  goto drop;
2636  }
2637 
2638  tc0->state = TCP_STATE_TIME_WAIT;
2639  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2640  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
2641  goto drop;
2642 
2643  break;
2644  case TCP_STATE_LAST_ACK:
2645  /* The only thing that [should] arrive in this state is an
2646  * acknowledgment of our FIN. If our FIN is now acknowledged,
2647  * delete the TCB, enter the CLOSED state, and return. */
2648 
2649  if (!tcp_rcv_ack_is_acceptable (tc0, b0))
2650  {
2651  error0 = TCP_ERROR_ACK_INVALID;
2652  goto drop;
2653  }
2654  error0 = TCP_ERROR_ACK_OK;
2655  tc0->snd_una = vnet_buffer (b0)->tcp.ack_number;
2656  /* Apparently our ACK for the peer's FIN was lost */
2657  if (is_fin0 && tc0->snd_una != tc0->snd_una_max)
2658  {
2659  tcp_send_fin (tc0);
2660  goto drop;
2661  }
2662 
2663  tc0->state = TCP_STATE_CLOSED;
2664  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2665  /* Delete the connection/session since the pipes should be
2666  * clear by now */
2667  tcp_connection_del (tc0);
2668 
2669  goto drop;
2670 
2671  break;
2672  case TCP_STATE_TIME_WAIT:
2673  /* The only thing that can arrive in this state is a
2674  * retransmission of the remote FIN. Acknowledge it, and restart
2675  * the 2 MSL timeout. */
2676 
2677  if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
2678  {
2679  tcp_maybe_inc_counter (rcv_process, error0, 1);
2680  goto drop;
2681  }
2682 
2683  tcp_make_ack (tc0, b0);
2684  next0 = tcp_next_output (is_ip4);
2685  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
2686  goto drop;
2687 
2688  break;
2689  default:
2690  ASSERT (0);
2691  }
2692 
2693  /* 6: check the URG bit TODO */
2694 
2695  /* 7: process the segment text */
2696  switch (tc0->state)
2697  {
2698  case TCP_STATE_ESTABLISHED:
2699  case TCP_STATE_FIN_WAIT_1:
2700  case TCP_STATE_FIN_WAIT_2:
2701  if (vnet_buffer (b0)->tcp.data_len)
2702  {
2703  error0 = tcp_segment_rcv (tc0, b0, &next0);
2704  tcp_maybe_inc_counter (rcv_process, error0, 1);
2705  }
2706  else if (is_fin0)
2707  tc0->rcv_nxt += 1;
2708  break;
2709  case TCP_STATE_CLOSE_WAIT:
2710  case TCP_STATE_CLOSING:
2711  case TCP_STATE_LAST_ACK:
2712  case TCP_STATE_TIME_WAIT:
2713  /* This should not occur, since a FIN has been received from the
2714  * remote side. Ignore the segment text. */
2715  break;
2716  }
2717 
2718  /* 8: check the FIN bit */
2719  if (!is_fin0)
2720  goto drop;
2721 
2722  switch (tc0->state)
2723  {
2724  case TCP_STATE_ESTABLISHED:
2725  case TCP_STATE_SYN_RCVD:
2726  /* Send FIN-ACK notify app and enter CLOSE-WAIT */
2729  tcp_make_fin (tc0, b0);
2730  tc0->snd_nxt += 1;
2731  tc0->snd_una_max = tc0->snd_nxt;
2732  next0 = tcp_next_output (tc0->c_is_ip4);
2733  stream_session_disconnect_notify (&tc0->connection);
2734  tc0->state = TCP_STATE_CLOSE_WAIT;
2735  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2736  break;
2737  case TCP_STATE_CLOSE_WAIT:
2738  case TCP_STATE_CLOSING:
2739  case TCP_STATE_LAST_ACK:
2740  /* move along .. */
2741  break;
2742  case TCP_STATE_FIN_WAIT_1:
2743  tc0->state = TCP_STATE_CLOSING;
2744  tcp_make_ack (tc0, b0);
2745  next0 = tcp_next_output (is_ip4);
2746  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2747  /* Wait for ACK but not forever */
2748  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
2749  break;
2750  case TCP_STATE_FIN_WAIT_2:
2751  /* Got FIN, send ACK! Be more aggressive with resource cleanup */
2752  tc0->state = TCP_STATE_TIME_WAIT;
2754  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
2755  tcp_make_ack (tc0, b0);
2756  next0 = tcp_next_output (is_ip4);
2757  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2758  break;
2759  case TCP_STATE_TIME_WAIT:
2760  /* Remain in the TIME-WAIT state. Restart the time-wait
2761  * timeout.
2762  */
2763  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
2764  break;
2765  }
2766  TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc0);
2767  n_fins += 1;
2768 
2769  drop:
2770  b0->error = error0 ? node->errors[error0] : 0;
2771 
2772  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
2773  {
2774  tcp_rx_trace_t *t0 =
2775  vlib_add_trace (vm, node, b0, sizeof (*t0));
2776  tcp_set_rx_trace_data (t0, tc0, tcp0, b0, is_ip4);
2777  }
2778 
2779  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
2780  n_left_to_next, bi0, next0);
2781  }
2782 
2783  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
2784  }
2785 
2787  my_thread_index);
2788  tcp_inc_counter (rcv_process, TCP_ERROR_EVENT_FIFO_FULL, errors);
2789  tcp_inc_counter (rcv_process, TCP_ERROR_FIN_RCVD, n_fins);
2790  return from_frame->n_vectors;
2791 }
2792 
2793 static uword
2795  vlib_frame_t * from_frame)
2796 {
2797  return tcp46_rcv_process_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2798 }
2799 
2800 static uword
2802  vlib_frame_t * from_frame)
2803 {
2804  return tcp46_rcv_process_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2805 }
2806 
2807 /* *INDENT-OFF* */
2809 {
2810  .function = tcp4_rcv_process,
2811  .name = "tcp4-rcv-process",
2812  /* Takes a vector of packets. */
2813  .vector_size = sizeof (u32),
2814  .n_errors = TCP_N_ERROR,
2815  .error_strings = tcp_error_strings,
2816  .n_next_nodes = TCP_RCV_PROCESS_N_NEXT,
2817  .next_nodes =
2818  {
2819 #define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n,
2821 #undef _
2822  },
2823  .format_trace = format_tcp_rx_trace_short,
2824 };
2825 /* *INDENT-ON* */
2826 
2828 
2829 /* *INDENT-OFF* */
2831 {
2832  .function = tcp6_rcv_process,
2833  .name = "tcp6-rcv-process",
2834  /* Takes a vector of packets. */
2835  .vector_size = sizeof (u32),
2836  .n_errors = TCP_N_ERROR,
2837  .error_strings = tcp_error_strings,
2838  .n_next_nodes = TCP_RCV_PROCESS_N_NEXT,
2839  .next_nodes =
2840  {
2841 #define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n,
2843 #undef _
2844  },
2845  .format_trace = format_tcp_rx_trace_short,
2846 };
2847 /* *INDENT-ON* */
2848 
2850 
2853 
2854 /**
2855  * LISTEN state processing as per RFC 793 p. 65
2856  */
2859  vlib_frame_t * from_frame, int is_ip4)
2860 {
2861  u32 n_left_from, next_index, *from, *to_next, n_syns = 0;
2862  u32 my_thread_index = vm->thread_index;
2863 
2864  from = vlib_frame_vector_args (from_frame);
2865  n_left_from = from_frame->n_vectors;
2866 
2867  next_index = node->cached_next_index;
2868 
2869  while (n_left_from > 0)
2870  {
2871  u32 n_left_to_next;
2872 
2873  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
2874 
2875  while (n_left_from > 0 && n_left_to_next > 0)
2876  {
2877  u32 bi0;
2878  vlib_buffer_t *b0;
2879  tcp_rx_trace_t *t0;
2880  tcp_header_t *th0 = 0;
2881  tcp_connection_t *lc0;
2882  ip4_header_t *ip40;
2883  ip6_header_t *ip60;
2884  tcp_connection_t *child0;
2885  u32 error0 = TCP_ERROR_NONE, next0 = tcp_next_drop (is_ip4);
2886 
2887  bi0 = from[0];
2888  to_next[0] = bi0;
2889  from += 1;
2890  to_next += 1;
2891  n_left_from -= 1;
2892  n_left_to_next -= 1;
2893 
2894  b0 = vlib_get_buffer (vm, bi0);
2895  lc0 = tcp_listener_get (vnet_buffer (b0)->tcp.connection_index);
2896 
2897  if (is_ip4)
2898  {
2899  ip40 = vlib_buffer_get_current (b0);
2900  th0 = ip4_next_header (ip40);
2901  }
2902  else
2903  {
2904  ip60 = vlib_buffer_get_current (b0);
2905  th0 = ip6_next_header (ip60);
2906  }
2907 
2908  /* Create child session. For syn-flood protection use filter */
2909 
2910  /* 1. first check for an RST: handled in dispatch */
2911  /* if (tcp_rst (th0))
2912  goto drop; */
2913 
2914  /* 2. second check for an ACK: handled in dispatch */
2915  /* if (tcp_ack (th0))
2916  {
2917  tcp_send_reset (b0, is_ip4);
2918  goto drop;
2919  } */
2920 
2921  /* 3. check for a SYN (did that already) */
2922 
2923  /* Make sure connection wasn't just created */
2924  child0 = tcp_lookup_connection (lc0->c_fib_index, b0,
2925  my_thread_index, is_ip4);
2926  if (PREDICT_FALSE (child0->state != TCP_STATE_LISTEN))
2927  {
2928  error0 = TCP_ERROR_CREATE_EXISTS;
2929  goto drop;
2930  }
2931 
2932  /* Create child session and send SYN-ACK */
2933  child0 = tcp_connection_new (my_thread_index);
2934  child0->c_lcl_port = th0->dst_port;
2935  child0->c_rmt_port = th0->src_port;
2936  child0->c_is_ip4 = is_ip4;
2937  child0->state = TCP_STATE_SYN_RCVD;
2938  child0->c_fib_index = lc0->c_fib_index;
2939 
2940  if (is_ip4)
2941  {
2942  child0->c_lcl_ip4.as_u32 = ip40->dst_address.as_u32;
2943  child0->c_rmt_ip4.as_u32 = ip40->src_address.as_u32;
2944  }
2945  else
2946  {
2947  clib_memcpy (&child0->c_lcl_ip6, &ip60->dst_address,
2948  sizeof (ip6_address_t));
2949  clib_memcpy (&child0->c_rmt_ip6, &ip60->src_address,
2950  sizeof (ip6_address_t));
2951  }
2952 
2953  if (tcp_options_parse (th0, &child0->rcv_opts))
2954  {
2955  clib_warning ("options parse fail");
2956  goto drop;
2957  }
2958 
2959  child0->irs = vnet_buffer (b0)->tcp.seq_number;
2960  child0->rcv_nxt = vnet_buffer (b0)->tcp.seq_number + 1;
2961  child0->rcv_las = child0->rcv_nxt;
2962  child0->sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_RX];
2963 
2964  /* RFC1323: TSval timestamps sent on {SYN} and {SYN,ACK}
2965  * segments are used to initialize PAWS. */
2966  if (tcp_opts_tstamp (&child0->rcv_opts))
2967  {
2968  child0->tsval_recent = child0->rcv_opts.tsval;
2969  child0->tsval_recent_age = tcp_time_now ();
2970  }
2971 
2972  if (tcp_opts_wscale (&child0->rcv_opts))
2973  child0->snd_wscale = child0->rcv_opts.wscale;
2974 
2975  child0->snd_wnd = clib_net_to_host_u16 (th0->window)
2976  << child0->snd_wscale;
2977  child0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number;
2978  child0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number;
2979 
2980  tcp_connection_init_vars (child0);
2981  TCP_EVT_DBG (TCP_EVT_SYN_RCVD, child0, 1);
2982 
2983  if (stream_session_accept (&child0->connection, lc0->c_s_index,
2984  0 /* notify */ ))
2985  {
2986  clib_warning ("session accept fail");
2987  tcp_connection_cleanup (child0);
2988  error0 = TCP_ERROR_CREATE_SESSION_FAIL;
2989  goto drop;
2990  }
2991 
2992  /* Reuse buffer to make syn-ack and send */
2993  tcp_make_synack (child0, b0);
2994  next0 = tcp_next_output (is_ip4);
2995  tcp_timer_set (child0, TCP_TIMER_ESTABLISH, TCP_SYN_RCVD_TIME);
2996 
2997  drop:
2998  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
2999  {
3000  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
3001  clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header));
3002  clib_memcpy (&t0->tcp_connection, lc0,
3003  sizeof (t0->tcp_connection));
3004  }
3005 
3006  n_syns += (error0 == TCP_ERROR_NONE);
3007  b0->error = node->errors[error0];
3008 
3009  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
3010  n_left_to_next, bi0, next0);
3011  }
3012 
3013  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
3014  }
3015 
3016  tcp_inc_counter (listen, TCP_ERROR_SYNS_RCVD, n_syns);
3017  return from_frame->n_vectors;
3018 }
3019 
3020 static uword
3022  vlib_frame_t * from_frame)
3023 {
3024  return tcp46_listen_inline (vm, node, from_frame, 1 /* is_ip4 */ );
3025 }
3026 
3027 static uword
3029  vlib_frame_t * from_frame)
3030 {
3031  return tcp46_listen_inline (vm, node, from_frame, 0 /* is_ip4 */ );
3032 }
3033 
3034 /* *INDENT-OFF* */
3036 {
3037  .function = tcp4_listen,
3038  .name = "tcp4-listen",
3039  /* Takes a vector of packets. */
3040  .vector_size = sizeof (u32),
3041  .n_errors = TCP_N_ERROR,
3042  .error_strings = tcp_error_strings,
3043  .n_next_nodes = TCP_LISTEN_N_NEXT,
3044  .next_nodes =
3045  {
3046 #define _(s,n) [TCP_LISTEN_NEXT_##s] = n,
3048 #undef _
3049  },
3050  .format_trace = format_tcp_rx_trace_short,
3051 };
3052 /* *INDENT-ON* */
3053 
3055 
3056 /* *INDENT-OFF* */
3058 {
3059  .function = tcp6_listen,
3060  .name = "tcp6-listen",
3061  /* Takes a vector of packets. */
3062  .vector_size = sizeof (u32),
3063  .n_errors = TCP_N_ERROR,
3064  .error_strings = tcp_error_strings,
3065  .n_next_nodes = TCP_LISTEN_N_NEXT,
3066  .next_nodes =
3067  {
3068 #define _(s,n) [TCP_LISTEN_NEXT_##s] = n,
3070 #undef _
3071  },
3072  .format_trace = format_tcp_rx_trace_short,
3073 };
3074 /* *INDENT-ON* */
3075 
3077 
3080 
3081 typedef enum _tcp_input_next
3082 {
3092 
3093 #define foreach_tcp4_input_next \
3094  _ (DROP, "ip4-drop") \
3095  _ (LISTEN, "tcp4-listen") \
3096  _ (RCV_PROCESS, "tcp4-rcv-process") \
3097  _ (SYN_SENT, "tcp4-syn-sent") \
3098  _ (ESTABLISHED, "tcp4-established") \
3099  _ (RESET, "tcp4-reset") \
3100  _ (PUNT, "ip4-punt")
3101 
3102 #define foreach_tcp6_input_next \
3103  _ (DROP, "ip6-drop") \
3104  _ (LISTEN, "tcp6-listen") \
3105  _ (RCV_PROCESS, "tcp6-rcv-process") \
3106  _ (SYN_SENT, "tcp6-syn-sent") \
3107  _ (ESTABLISHED, "tcp6-established") \
3108  _ (RESET, "tcp6-reset") \
3109  _ (PUNT, "ip6-punt")
3110 
3111 #define filter_flags (TCP_FLAG_SYN|TCP_FLAG_ACK|TCP_FLAG_RST|TCP_FLAG_FIN)
3112 
3113 static void
3115  vlib_buffer_t ** bs, u32 n_bufs, u8 is_ip4)
3116 {
3117  tcp_connection_t *tc;
3118  tcp_header_t *tcp;
3119  tcp_rx_trace_t *t;
3120  int i;
3121 
3122  for (i = 0; i < n_bufs; i++)
3123  {
3124  if (bs[i]->flags & VLIB_BUFFER_IS_TRACED)
3125  {
3126  t = vlib_add_trace (vm, node, bs[i], sizeof (*t));
3127  tc = tcp_connection_get (vnet_buffer (bs[i])->tcp.connection_index,
3128  vm->thread_index);
3129  tcp = vlib_buffer_get_current (bs[i]);
3130  tcp_set_rx_trace_data (t, tc, tcp, bs[i], is_ip4);
3131  }
3132  }
3133 }
3134 
3135 static void
3136 tcp_input_set_error_next (tcp_main_t * tm, u16 * next, u32 * error, u8 is_ip4)
3137 {
3138  if (*error == TCP_ERROR_FILTERED)
3139  {
3140  *next = TCP_INPUT_NEXT_DROP;
3141  }
3142  else if ((is_ip4 && tm->punt_unknown4) || (!is_ip4 && tm->punt_unknown6))
3143  {
3144  *next = TCP_INPUT_NEXT_PUNT;
3145  *error = TCP_ERROR_PUNT;
3146  }
3147  else
3148  {
3149  *next = TCP_INPUT_NEXT_RESET;
3150  *error = TCP_ERROR_NO_LISTENER;
3151  }
3152 }
3153 
3154 static inline tcp_connection_t *
3155 tcp_input_lookup_buffer (vlib_buffer_t * b, u8 thread_index, u32 * error,
3156  u8 is_ip4)
3157 {
3158  u32 fib_index = vnet_buffer (b)->ip.fib_index;
3159  int n_advance_bytes, n_data_bytes;
3161  tcp_header_t *tcp;
3162  u8 is_filtered = 0;
3163 
3164  if (is_ip4)
3165  {
3167  tcp = ip4_next_header (ip4);
3168  vnet_buffer (b)->tcp.hdr_offset = (u8 *) tcp - (u8 *) ip4;
3169  n_advance_bytes = (ip4_header_bytes (ip4) + tcp_header_bytes (tcp));
3170  n_data_bytes = clib_net_to_host_u16 (ip4->length) - n_advance_bytes;
3171 
3172  /* Length check. Checksum computed by ipx_local no need to compute again */
3173  if (PREDICT_FALSE (n_advance_bytes < 0))
3174  {
3175  *error = TCP_ERROR_LENGTH;
3176  return 0;
3177  }
3178 
3179  tc = session_lookup_connection_wt4 (fib_index, &ip4->dst_address,
3180  &ip4->src_address, tcp->dst_port,
3181  tcp->src_port, TRANSPORT_PROTO_TCP,
3182  thread_index, &is_filtered);
3183  }
3184  else
3185  {
3187  tcp = ip6_next_header (ip6);
3188  vnet_buffer (b)->tcp.hdr_offset = (u8 *) tcp - (u8 *) ip6;
3189  n_advance_bytes = tcp_header_bytes (tcp);
3190  n_data_bytes = clib_net_to_host_u16 (ip6->payload_length)
3191  - n_advance_bytes;
3192  n_advance_bytes += sizeof (ip6[0]);
3193 
3194  if (PREDICT_FALSE (n_advance_bytes < 0))
3195  {
3196  *error = TCP_ERROR_LENGTH;
3197  return 0;
3198  }
3199 
3200  tc = session_lookup_connection_wt6 (fib_index, &ip6->dst_address,
3201  &ip6->src_address, tcp->dst_port,
3202  tcp->src_port, TRANSPORT_PROTO_TCP,
3203  thread_index, &is_filtered);
3204  }
3205 
3206  vnet_buffer (b)->tcp.seq_number = clib_net_to_host_u32 (tcp->seq_number);
3207  vnet_buffer (b)->tcp.ack_number = clib_net_to_host_u32 (tcp->ack_number);
3208  vnet_buffer (b)->tcp.data_offset = n_advance_bytes;
3209  vnet_buffer (b)->tcp.data_len = n_data_bytes;
3210  vnet_buffer (b)->tcp.flags = 0;
3211 
3212  *error = is_filtered ? TCP_ERROR_FILTERED : *error;
3213 
3215 }
3216 
3217 static inline void
3219  vlib_buffer_t * b, u16 * next, u32 * error)
3220 {
3221  tcp_header_t *tcp;
3222  u8 flags;
3223 
3224  tcp = tcp_buffer_hdr (b);
3225  flags = tcp->flags & filter_flags;
3226  *next = tm->dispatch_table[tc->state][flags].next;
3227  *error = tm->dispatch_table[tc->state][flags].error;
3228 
3229  if (PREDICT_FALSE (*error == TCP_ERROR_DISPATCH
3230  || *next == TCP_INPUT_NEXT_RESET))
3231  {
3232  /* Overload tcp flags to store state */
3233  tcp_state_t state = tc->state;
3234  vnet_buffer (b)->tcp.flags = tc->state;
3235 
3236  if (*error == TCP_ERROR_DISPATCH)
3237  clib_warning ("disp error state %U flags %U", format_tcp_state,
3238  state, format_tcp_flags, (int) flags);
3239  }
3240 }
3241 
3244  vlib_frame_t * frame, int is_ip4)
3245 {
3246  u32 n_left_from, *from, thread_index = vm->thread_index;
3247  tcp_main_t *tm = vnet_get_tcp_main ();
3248  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
3249  u16 nexts[VLIB_FRAME_SIZE], *next;
3250 
3251  tcp_set_time_now (thread_index);
3252 
3253  from = vlib_frame_vector_args (frame);
3254  n_left_from = frame->n_vectors;
3255  vlib_get_buffers (vm, from, bufs, n_left_from);
3256 
3257  b = bufs;
3258  next = nexts;
3259 
3260  while (n_left_from >= 4)
3261  {
3262  u32 error0 = TCP_ERROR_NO_LISTENER, error1 = TCP_ERROR_NO_LISTENER;
3263  tcp_connection_t *tc0, *tc1;
3264 
3265  {
3266  vlib_prefetch_buffer_header (b[2], STORE);
3267  CLIB_PREFETCH (b[2]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3268 
3269  vlib_prefetch_buffer_header (b[3], STORE);
3270  CLIB_PREFETCH (b[3]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3271  }
3272 
3273  next[0] = next[1] = TCP_INPUT_NEXT_DROP;
3274 
3275  tc0 = tcp_input_lookup_buffer (b[0], thread_index, &error0, is_ip4);
3276  tc1 = tcp_input_lookup_buffer (b[1], thread_index, &error1, is_ip4);
3277 
3278  if (PREDICT_TRUE (!tc0 + !tc1 == 0))
3279  {
3280  ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3281  ASSERT (tcp_lookup_is_valid (tc1, tcp_buffer_hdr (b[1])));
3282 
3283  vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3284  vnet_buffer (b[1])->tcp.connection_index = tc1->c_c_index;
3285 
3286  tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3287  tcp_input_dispatch_buffer (tm, tc1, b[1], &next[1], &error1);
3288  }
3289  else
3290  {
3291  if (PREDICT_TRUE (tc0 != 0))
3292  {
3293  ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3294  vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3295  tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3296  }
3297  else
3298  tcp_input_set_error_next (tm, &next[0], &error0, is_ip4);
3299 
3300  if (PREDICT_TRUE (tc1 != 0))
3301  {
3302  ASSERT (tcp_lookup_is_valid (tc1, tcp_buffer_hdr (b[1])));
3303  vnet_buffer (b[1])->tcp.connection_index = tc1->c_c_index;
3304  tcp_input_dispatch_buffer (tm, tc1, b[1], &next[1], &error1);
3305  }
3306  else
3307  tcp_input_set_error_next (tm, &next[1], &error1, is_ip4);
3308  }
3309 
3310  b += 2;
3311  next += 2;
3312  n_left_from -= 2;
3313  }
3314  while (n_left_from > 0)
3315  {
3316  tcp_connection_t *tc0;
3317  u32 error0 = TCP_ERROR_NO_LISTENER;
3318 
3319  if (n_left_from > 1)
3320  {
3321  vlib_prefetch_buffer_header (b[1], STORE);
3322  CLIB_PREFETCH (b[1]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3323  }
3324 
3325  next[0] = TCP_INPUT_NEXT_DROP;
3326  tc0 = tcp_input_lookup_buffer (b[0], thread_index, &error0, is_ip4);
3327  if (PREDICT_TRUE (tc0 != 0))
3328  {
3329  ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3330  vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3331  tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3332  }
3333  else
3334  tcp_input_set_error_next (tm, &next[0], &error0, is_ip4);
3335 
3336  b += 1;
3337  next += 1;
3338  n_left_from -= 1;
3339  }
3340 
3342  tcp_input_trace_frame (vm, node, bufs, frame->n_vectors, is_ip4);
3343 
3344  vlib_buffer_enqueue_to_next (vm, node, from, nexts, frame->n_vectors);
3345  return frame->n_vectors;
3346 }
3347 
3348 static uword
3350  vlib_frame_t * from_frame)
3351 {
3352  return tcp46_input_inline (vm, node, from_frame, 1 /* is_ip4 */ );
3353 }
3354 
3355 static uword
3357  vlib_frame_t * from_frame)
3358 {
3359  return tcp46_input_inline (vm, node, from_frame, 0 /* is_ip4 */ );
3360 }
3361 
3362 /* *INDENT-OFF* */
3364 {
3365  .function = tcp4_input,
3366  .name = "tcp4-input",
3367  /* Takes a vector of packets. */
3368  .vector_size = sizeof (u32),
3369  .n_errors = TCP_N_ERROR,
3370  .error_strings = tcp_error_strings,
3371  .n_next_nodes = TCP_INPUT_N_NEXT,
3372  .next_nodes =
3373  {
3374 #define _(s,n) [TCP_INPUT_NEXT_##s] = n,
3376 #undef _
3377  },
3378  .format_buffer = format_tcp_header,
3379  .format_trace = format_tcp_rx_trace,
3380 };
3381 /* *INDENT-ON* */
3382 
3384 
3385 /* *INDENT-OFF* */
3387 {
3388  .function = tcp6_input,
3389  .name = "tcp6-input",
3390  /* Takes a vector of packets. */
3391  .vector_size = sizeof (u32),
3392  .n_errors = TCP_N_ERROR,
3393  .error_strings = tcp_error_strings,
3394  .n_next_nodes = TCP_INPUT_N_NEXT,
3395  .next_nodes =
3396  {
3397 #define _(s,n) [TCP_INPUT_NEXT_##s] = n,
3399 #undef _
3400  },
3401  .format_buffer = format_tcp_header,
3402  .format_trace = format_tcp_rx_trace,
3403 };
3404 /* *INDENT-ON* */
3405 
3407 
3408 static void
3410 {
3411  int i, j;
3412  for (i = 0; i < ARRAY_LEN (tm->dispatch_table); i++)
3413  for (j = 0; j < ARRAY_LEN (tm->dispatch_table[i]); j++)
3414  {
3415  tm->dispatch_table[i][j].next = TCP_INPUT_NEXT_DROP;
3416  tm->dispatch_table[i][j].error = TCP_ERROR_DISPATCH;
3417  }
3418 
3419 #define _(t,f,n,e) \
3420 do { \
3421  tm->dispatch_table[TCP_STATE_##t][f].next = (n); \
3422  tm->dispatch_table[TCP_STATE_##t][f].error = (e); \
3423 } while (0)
3424 
3425  /* SYNs for new connections -> tcp-listen. */
3426  _(LISTEN, TCP_FLAG_SYN, TCP_INPUT_NEXT_LISTEN, TCP_ERROR_NONE);
3427  _(LISTEN, TCP_FLAG_ACK, TCP_INPUT_NEXT_RESET, TCP_ERROR_NONE);
3428  _(LISTEN, TCP_FLAG_RST, TCP_INPUT_NEXT_DROP, TCP_ERROR_RST_RCVD);
3430  TCP_ERROR_NONE);
3431  /* ACK for for a SYN-ACK -> tcp-rcv-process. */
3432  _(SYN_RCVD, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3433  _(SYN_RCVD, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3434  _(SYN_RCVD, TCP_FLAG_SYN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3435  /* SYN-ACK for a SYN */
3437  TCP_ERROR_NONE);
3438  _(SYN_SENT, TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE);
3439  _(SYN_SENT, TCP_FLAG_RST, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE);
3441  TCP_ERROR_NONE);
3442  /* ACK for for established connection -> tcp-established. */
3443  _(ESTABLISHED, TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3444  /* FIN for for established connection -> tcp-established. */
3445  _(ESTABLISHED, TCP_FLAG_FIN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3447  TCP_ERROR_NONE);
3448  _(ESTABLISHED, TCP_FLAG_RST, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3450  TCP_ERROR_NONE);
3451  _(ESTABLISHED, TCP_FLAG_SYN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3453  TCP_ERROR_NONE);
3454  /* ACK or FIN-ACK to our FIN */
3455  _(FIN_WAIT_1, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3457  TCP_ERROR_NONE);
3458  /* FIN in reply to our FIN from the other side */
3459  _(FIN_WAIT_1, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3460  _(FIN_WAIT_1, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3461  _(CLOSING, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3462  /* FIN confirming that the peer (app) has closed */
3463  _(FIN_WAIT_2, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3464  _(FIN_WAIT_2, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3466  TCP_ERROR_NONE);
3467  _(CLOSE_WAIT, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3469  TCP_ERROR_NONE);
3470  _(LAST_ACK, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3471  _(LAST_ACK, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3473  TCP_ERROR_NONE);
3474  _(LAST_ACK, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3475  _(TIME_WAIT, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3477  TCP_ERROR_NONE);
3478  _(TIME_WAIT, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3479  _(TIME_WAIT, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3480  _(CLOSED, TCP_FLAG_ACK, TCP_INPUT_NEXT_DROP, TCP_ERROR_CONNECTION_CLOSED);
3481  _(CLOSED, TCP_FLAG_RST, TCP_INPUT_NEXT_DROP, TCP_ERROR_CONNECTION_CLOSED);
3483  TCP_ERROR_CONNECTION_CLOSED);
3484 #undef _
3485 }
3486 
3487 static clib_error_t *
3489 {
3490  clib_error_t *error = 0;
3491  tcp_main_t *tm = vnet_get_tcp_main ();
3492 
3493  if ((error = vlib_call_init_function (vm, tcp_init)))
3494  return error;
3495 
3496  /* Initialize dispatch table. */
3498 
3499  return error;
3500 }
3501 
3503 
3504 /*
3505  * fd.io coding-style-patch-verification: ON
3506  *
3507  * Local Variables:
3508  * eval: (c-set-style "gnu")
3509  * End:
3510  */
#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:1548
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:1134
#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:1106
vlib_node_registration_t tcp6_rcv_process_node
(constructor) VLIB_REGISTER_NODE (tcp6_rcv_process_node)
Definition: tcp_input.c:2432
#define tcp_in_recovery(tc)
Definition: tcp.h:344
static void tcp_retransmit_timer_set(tcp_connection_t *tc)
Definition: tcp.h:733
#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:544
void tcp_make_fin(tcp_connection_t *tc, vlib_buffer_t *b)
Convert buffer to FIN-ACK.
Definition: tcp_output.c:560
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:551
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:1076
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:3356
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:1772
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:1830
unsigned long u64
Definition: types.h:89
#define tcp_store_err_counters(node_id, cnts)
Definition: tcp_input.c:1834
static void tcp_dispatch_table_init(tcp_main_t *tm)
Definition: tcp_input.c:3409
static int ip4_header_bytes(ip4_header_t *i)
Definition: ip4_packet.h:232
static u8 * format_tcp_rx_trace_short(u8 *s, va_list *args)
Definition: tcp_input.c:1742
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:1844
static tcp_connection_t * tcp_half_open_connection_get(u32 conn_index)
Definition: tcp.h:512
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:1651
#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:2431
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:632
#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:1652
tcp_connection_t tcp_connection
Definition: tcp_input.c:1722
static u8 tcp_sack_vector_is_sane(sack_block_t *sacks)
Definition: tcp_input.c:1406
static tcp_connection_t * tcp_get_connection_from_transport(transport_connection_t *tconn)
Definition: tcp.h:485
static void tcp_cc_congestion_undo(tcp_connection_t *tc)
Definition: tcp_input.c:1062
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:306
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:817
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:227
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:3028
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:1822
vlib_node_registration_t tcp6_syn_sent_node
(constructor) VLIB_REGISTER_NODE (tcp6_syn_sent_node)
Definition: tcp_input.c:2028
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:2069
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:1009
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:3021
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:2801
static uword tcp4_syn_sent(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:2374
#define seq_gt(_s1, _s2)
Definition: tcp.h:545
static u8 * format_tcp_rx_trace(u8 *s, va_list *args)
Definition: tcp_input.c:1726
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:702
#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:445
static void tcp_cc_recovery_exit(tcp_connection_t *tc)
Definition: tcp_input.c:1039
#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:3079
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
#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:249
static void * ip4_next_header(ip4_header_t *i)
Definition: ip4_packet.h:238
static u32 tcp_time_now(void)
Definition: tcp.h:665
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:1806
#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:796
#define TCP_EVT_DBG(_evt, _args...)
Definition: tcp_debug.h:238
#define timestamp_lt(_t1, _t2)
Definition: tcp.h:550
static void tcp_timer_set(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:694
#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:172
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:434
#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:461
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:1050
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:2858
#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:3136
void tcp_retransmit_first_unacked(tcp_connection_t *tc)
Retransmit first unacked segment.
Definition: tcp_output.c:1629
unsigned short u16
Definition: types.h:57
#define foreach_tcp4_input_next
Definition: tcp_input.c:3093
static u32 ooo_segment_offset(svm_fifo_t *f, ooo_segment_t *s)
Definition: svm_fifo.h:237
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:3111
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:273
static int tcp_buffer_discard_bytes(vlib_buffer_t *b, u32 n_bytes_to_drop)
Definition: tcp_input.c:1618
#define foreach_tcp6_input_next
Definition: tcp_input.c:3102
#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:1735
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:3114
#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:789
#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:2851
#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:2031
static ooo_segment_t * svm_fifo_newest_ooo_segment(svm_fifo_t *f)
Definition: svm_fifo.h:164
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:1483
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:1602
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:1085
#define TCP_FLAG_RST
Definition: fa_node.h:14
static stream_session_t * session_get(u32 si, u32 thread_index)
Definition: session.h:290
#define TCP_DBG(_fmt, _args...)
Definition: tcp_debug.h:89
#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:705
static uword tcp6_syn_sent_rcv(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:2381
#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:2027
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:1757
void tcp_send_reset(tcp_connection_t *tc)
Build and set reset packet for connection.
Definition: tcp_output.c:896
#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:1310
tcp_header_t tcp_header
Definition: tcp_input.c:1721
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:603
#define pool_is_free_index(P, I)
Use free bitmap to query whether given index is free.
Definition: pool.h:270
#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:545
void stream_session_disconnect_notify(transport_connection_t *tc)
Notification from transport that connection is being closed.
Definition: session.c:720
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:718
#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:3488
u32 session_tx_fifo_max_dequeue(transport_connection_t *tc)
Definition: session.c:417
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:1969
#define seq_geq(_s1, _s2)
Definition: tcp.h:546
vhost_vring_state_t state
Definition: vhost-user.h:82
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:1144
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:1772
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:1429
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:1976
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:543
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:991
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:3078
void tcp_send_fin(tcp_connection_t *tc)
Send FIN.
Definition: tcp_output.c:1057
#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:671
static u8 tcp_is_lost_fin(tcp_connection_t *tc)
Definition: tcp.h:644
static uword tcp4_rcv_process(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_input.c:2794
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:777
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:1493
static u8 tcp_should_fastrecover_sack(tcp_connection_t *tc)
Definition: tcp_input.c:1128
u64 uword
Definition: types.h:112
#define seq_max(_s1, _s2)
Definition: tcp.h:547
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:1815
#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:360
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:515
static tcp_connection_t * tcp_connection_get(u32 conn_index, u32 thread_index)
Definition: tcp.h:466
#define tcp_maybe_inc_err_counter(cnts, err)
Definition: tcp_input.c:1826
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:564
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:3349
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:3155
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:2852
#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
u32 flags
Definition: vhost-user.h:77
#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:3243
static void tcp_persist_timer_set(tcp_connection_t *tc)
Definition: tcp.h:754
static tcp_main_t * vnet_get_tcp_main()
Definition: tcp.h:439
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:2110
#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:2439
static void tcp_retransmit_timer_reset(tcp_connection_t *tc)
Definition: tcp.h:741
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:3218
#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:791
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:506
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:1029
static void tcp_persist_timer_reset(tcp_connection_t *tc)
Definition: tcp.h:771
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