FD.io VPP  v19.08-27-gf4dcae4
Vector Packet Processing
tcp_input.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016-2019 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 
82 /**
83  * Validate segment sequence number. As per RFC793:
84  *
85  * Segment Receive Test
86  * Length Window
87  * ------- ------- -------------------------------------------
88  * 0 0 SEG.SEQ = RCV.NXT
89  * 0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND
90  * >0 0 not acceptable
91  * >0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND
92  * or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND
93  *
94  * This ultimately consists in checking if segment falls within the window.
95  * The one important difference compared to RFC793 is that we use rcv_las,
96  * or the rcv_nxt at last ack sent instead of rcv_nxt since that's the
97  * peer's reference when computing our receive window.
98  *
99  * This:
100  * seq_leq (end_seq, tc->rcv_las + tc->rcv_wnd) && seq_geq (seq, tc->rcv_las)
101  * however, is too strict when we have retransmits. Instead we just check that
102  * the seq is not beyond the right edge and that the end of the segment is not
103  * less than the left edge.
104  *
105  * N.B. rcv_nxt and rcv_wnd are both updated in this node if acks are sent, so
106  * use rcv_nxt in the right edge window test instead of rcv_las.
107  *
108  */
111 {
112  return (seq_geq (end_seq, tc->rcv_las)
113  && seq_leq (seq, tc->rcv_nxt + tc->rcv_wnd));
114 }
115 
116 /**
117  * Parse TCP header options.
118  *
119  * @param th TCP header
120  * @param to TCP options data structure to be populated
121  * @param is_syn set if packet is syn
122  * @return -1 if parsing failed
123  */
124 static inline int
126 {
127  const u8 *data;
128  u8 opt_len, opts_len, kind;
129  int j;
130  sack_block_t b;
131 
132  opts_len = (tcp_doff (th) << 2) - sizeof (tcp_header_t);
133  data = (const u8 *) (th + 1);
134 
135  /* Zero out all flags but those set in SYN */
136  to->flags &= (TCP_OPTS_FLAG_SACK_PERMITTED | TCP_OPTS_FLAG_WSCALE
137  | TCP_OPTS_FLAG_TSTAMP | TCP_OPTION_MSS);
138 
139  for (; opts_len > 0; opts_len -= opt_len, data += opt_len)
140  {
141  kind = data[0];
142 
143  /* Get options length */
144  if (kind == TCP_OPTION_EOL)
145  break;
146  else if (kind == TCP_OPTION_NOOP)
147  {
148  opt_len = 1;
149  continue;
150  }
151  else
152  {
153  /* broken options */
154  if (opts_len < 2)
155  return -1;
156  opt_len = data[1];
157 
158  /* weird option length */
159  if (opt_len < 2 || opt_len > opts_len)
160  return -1;
161  }
162 
163  /* Parse options */
164  switch (kind)
165  {
166  case TCP_OPTION_MSS:
167  if (!is_syn)
168  break;
169  if ((opt_len == TCP_OPTION_LEN_MSS) && tcp_syn (th))
170  {
171  to->flags |= TCP_OPTS_FLAG_MSS;
172  to->mss = clib_net_to_host_u16 (*(u16 *) (data + 2));
173  }
174  break;
176  if (!is_syn)
177  break;
178  if ((opt_len == TCP_OPTION_LEN_WINDOW_SCALE) && tcp_syn (th))
179  {
180  to->flags |= TCP_OPTS_FLAG_WSCALE;
181  to->wscale = data[2];
182  if (to->wscale > TCP_MAX_WND_SCALE)
184  }
185  break;
187  if (is_syn)
188  to->flags |= TCP_OPTS_FLAG_TSTAMP;
189  if ((to->flags & TCP_OPTS_FLAG_TSTAMP)
190  && opt_len == TCP_OPTION_LEN_TIMESTAMP)
191  {
192  to->tsval = clib_net_to_host_u32 (*(u32 *) (data + 2));
193  to->tsecr = clib_net_to_host_u32 (*(u32 *) (data + 6));
194  }
195  break;
197  if (!is_syn)
198  break;
199  if (opt_len == TCP_OPTION_LEN_SACK_PERMITTED && tcp_syn (th))
200  to->flags |= TCP_OPTS_FLAG_SACK_PERMITTED;
201  break;
203  /* If SACK permitted was not advertised or a SYN, break */
204  if ((to->flags & TCP_OPTS_FLAG_SACK_PERMITTED) == 0 || tcp_syn (th))
205  break;
206 
207  /* If too short or not correctly formatted, break */
208  if (opt_len < 10 || ((opt_len - 2) % TCP_OPTION_LEN_SACK_BLOCK))
209  break;
210 
211  to->flags |= TCP_OPTS_FLAG_SACK;
212  to->n_sack_blocks = (opt_len - 2) / TCP_OPTION_LEN_SACK_BLOCK;
213  vec_reset_length (to->sacks);
214  for (j = 0; j < to->n_sack_blocks; j++)
215  {
216  b.start = clib_net_to_host_u32 (*(u32 *) (data + 2 + 8 * j));
217  b.end = clib_net_to_host_u32 (*(u32 *) (data + 6 + 8 * j));
218  vec_add1 (to->sacks, b);
219  }
220  break;
221  default:
222  /* Nothing to see here */
223  continue;
224  }
225  }
226  return 0;
227 }
228 
229 /**
230  * RFC1323: Check against wrapped sequence numbers (PAWS). If we have
231  * timestamp to echo and it's less than tsval_recent, drop segment
232  * but still send an ACK in order to retain TCP's mechanism for detecting
233  * and recovering from half-open connections
234  *
235  * Or at least that's what the theory says. It seems that this might not work
236  * very well with packet reordering and fast retransmit. XXX
237  */
238 always_inline int
240 {
241  return tcp_opts_tstamp (&tc->rcv_opts)
242  && timestamp_lt (tc->rcv_opts.tsval, tc->tsval_recent);
243 }
244 
245 /**
246  * Update tsval recent
247  */
248 always_inline void
250 {
251  /*
252  * RFC1323: If Last.ACK.sent falls within the range of sequence numbers
253  * of an incoming segment:
254  * SEG.SEQ <= Last.ACK.sent < SEG.SEQ + SEG.LEN
255  * then the TSval from the segment is copied to TS.Recent;
256  * otherwise, the TSval is ignored.
257  */
258  if (tcp_opts_tstamp (&tc->rcv_opts) && seq_leq (seq, tc->rcv_las)
259  && seq_leq (tc->rcv_las, seq_end))
260  {
261  ASSERT (timestamp_leq (tc->tsval_recent, tc->rcv_opts.tsval));
262  tc->tsval_recent = tc->rcv_opts.tsval;
263  tc->tsval_recent_age = tcp_time_now_w_thread (tc->c_thread_index);
264  }
265 }
266 
267 /**
268  * Validate incoming segment as per RFC793 p. 69 and RFC1323 p. 19
269  *
270  * It first verifies if segment has a wrapped sequence number (PAWS) and then
271  * does the processing associated to the first four steps (ignoring security
272  * and precedence): sequence number, rst bit and syn bit checks.
273  *
274  * @return 0 if segments passes validation.
275  */
276 static int
278  vlib_buffer_t * b0, tcp_header_t * th0, u32 * error0)
279 {
280  /* We could get a burst of RSTs interleaved with acks */
281  if (PREDICT_FALSE (tc0->state == TCP_STATE_CLOSED))
282  {
283  tcp_send_reset (tc0);
284  *error0 = TCP_ERROR_CONNECTION_CLOSED;
285  goto error;
286  }
287 
288  if (PREDICT_FALSE (!tcp_ack (th0) && !tcp_rst (th0) && !tcp_syn (th0)))
289  {
290  *error0 = TCP_ERROR_SEGMENT_INVALID;
291  goto error;
292  }
293 
294  if (PREDICT_FALSE (tcp_options_parse (th0, &tc0->rcv_opts, 0)))
295  {
296  *error0 = TCP_ERROR_OPTIONS;
297  goto error;
298  }
299 
301  {
302  *error0 = TCP_ERROR_PAWS;
303  TCP_EVT_DBG (TCP_EVT_PAWS_FAIL, tc0, vnet_buffer (b0)->tcp.seq_number,
304  vnet_buffer (b0)->tcp.seq_end);
305 
306  /* If it just so happens that a segment updates tsval_recent for a
307  * segment over 24 days old, invalidate tsval_recent. */
308  if (timestamp_lt (tc0->tsval_recent_age + TCP_PAWS_IDLE,
309  tcp_time_now_w_thread (tc0->c_thread_index)))
310  {
311  tc0->tsval_recent = tc0->rcv_opts.tsval;
312  clib_warning ("paws failed: 24-day old segment");
313  }
314  /* Drop after ack if not rst. Resets can fail paws check as per
315  * RFC 7323 sec. 5.2: When an <RST> segment is received, it MUST NOT
316  * be subjected to the PAWS check by verifying an acceptable value in
317  * SEG.TSval */
318  else if (!tcp_rst (th0))
319  {
320  tcp_program_ack (tc0);
321  TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0, vnet_buffer (b0)->tcp);
322  goto error;
323  }
324  }
325 
326  /* 1st: check sequence number */
327  if (!tcp_segment_in_rcv_wnd (tc0, vnet_buffer (b0)->tcp.seq_number,
328  vnet_buffer (b0)->tcp.seq_end))
329  {
330  /* SYN/SYN-ACK retransmit */
331  if (tcp_syn (th0)
332  && vnet_buffer (b0)->tcp.seq_number == tc0->rcv_nxt - 1)
333  {
334  tcp_options_parse (th0, &tc0->rcv_opts, 1);
335  if (tc0->state == TCP_STATE_SYN_RCVD)
336  {
337  tcp_send_synack (tc0);
338  TCP_EVT_DBG (TCP_EVT_SYN_RCVD, tc0, 0);
339  *error0 = TCP_ERROR_SYNS_RCVD;
340  }
341  else
342  {
343  tcp_program_ack (tc0);
344  TCP_EVT_DBG (TCP_EVT_SYNACK_RCVD, tc0);
345  *error0 = TCP_ERROR_SYN_ACKS_RCVD;
346  }
347  goto error;
348  }
349 
350  /* If our window is 0 and the packet is in sequence, let it pass
351  * through for ack processing. It should be dropped later. */
352  if (tc0->rcv_wnd < tc0->snd_mss
353  && tc0->rcv_nxt == vnet_buffer (b0)->tcp.seq_number)
354  goto check_reset;
355 
356  /* If we entered recovery and peer did so as well, there's a chance that
357  * dup acks won't be acceptable on either end because seq_end may be less
358  * than rcv_las. This can happen if acks are lost in both directions. */
359  if (tcp_in_recovery (tc0)
360  && seq_geq (vnet_buffer (b0)->tcp.seq_number,
361  tc0->rcv_las - tc0->rcv_wnd)
362  && seq_leq (vnet_buffer (b0)->tcp.seq_end,
363  tc0->rcv_nxt + tc0->rcv_wnd))
364  goto check_reset;
365 
366  *error0 = TCP_ERROR_RCV_WND;
367 
368  tc0->errors.below_data_wnd += seq_lt (vnet_buffer (b0)->tcp.seq_end,
369  tc0->rcv_las);
370 
371  /* If not RST, send dup ack */
372  if (!tcp_rst (th0))
373  {
374  tcp_program_dupack (tc0);
375  TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0, vnet_buffer (b0)->tcp);
376  }
377  goto error;
378 
379  check_reset:
380  ;
381  }
382 
383  /* 2nd: check the RST bit */
384  if (PREDICT_FALSE (tcp_rst (th0)))
385  {
386  tcp_connection_reset (tc0);
387  *error0 = TCP_ERROR_RST_RCVD;
388  goto error;
389  }
390 
391  /* 3rd: check security and precedence (skip) */
392 
393  /* 4th: check the SYN bit (in window) */
394  if (PREDICT_FALSE (tcp_syn (th0)))
395  {
396  /* As per RFC5961 send challenge ack instead of reset */
397  tcp_program_ack (tc0);
398  *error0 = TCP_ERROR_SPURIOUS_SYN;
399  goto error;
400  }
401 
402  /* If segment in window, save timestamp */
403  tcp_update_timestamp (tc0, vnet_buffer (b0)->tcp.seq_number,
404  vnet_buffer (b0)->tcp.seq_end);
405  return 0;
406 
407 error:
408  return -1;
409 }
410 
411 always_inline int
413 {
414  /* SND.UNA =< SEG.ACK =< SND.NXT */
415  if (!(seq_leq (tc->snd_una, vnet_buffer (b)->tcp.ack_number)
416  && seq_leq (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt)))
417  {
418  if (seq_leq (vnet_buffer (b)->tcp.ack_number, tc->snd_una_max)
419  && seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_una))
420  {
421  tc->snd_nxt = vnet_buffer (b)->tcp.ack_number;
422  goto acceptable;
423  }
424  *error = TCP_ERROR_ACK_INVALID;
425  return -1;
426  }
427 
428 acceptable:
429  tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una;
430  tc->snd_una = vnet_buffer (b)->tcp.ack_number;
431  *error = TCP_ERROR_ACK_OK;
432  return 0;
433 }
434 
435 /**
436  * Compute smoothed RTT as per VJ's '88 SIGCOMM and RFC6298
437  *
438  * Note that although the original article, srtt and rttvar are scaled
439  * to minimize round-off errors, here we don't. Instead, we rely on
440  * better precision time measurements.
441  *
442  * TODO support us rtt resolution
443  */
444 static void
446 {
447  int err, diff;
448 
449  if (tc->srtt != 0)
450  {
451  err = mrtt - tc->srtt;
452 
453  /* XXX Drop in RTT results in RTTVAR increase and bigger RTO.
454  * The increase should be bound */
455  tc->srtt = clib_max ((int) tc->srtt + (err >> 3), 1);
456  diff = (clib_abs (err) - (int) tc->rttvar) >> 2;
457  tc->rttvar = clib_max ((int) tc->rttvar + diff, 1);
458  }
459  else
460  {
461  /* First measurement. */
462  tc->srtt = mrtt;
463  tc->rttvar = mrtt >> 1;
464  }
465 }
466 
467 #ifndef CLIB_MARCH_VARIANT
468 void
470 {
471  tc->rto = clib_min (tc->srtt + (tc->rttvar << 2), TCP_RTO_MAX);
472  tc->rto = clib_max (tc->rto, TCP_RTO_MIN);
473 }
474 #endif /* CLIB_MARCH_VARIANT */
475 
476 /**
477  * Update RTT estimate and RTO timer
478  *
479  * Measure RTT: We have two sources of RTT measurements: TSOPT and ACK
480  * timing. Middle boxes are known to fiddle with TCP options so we
481  * should give higher priority to ACK timing.
482  *
483  * This should be called only if previously sent bytes have been acked.
484  *
485  * return 1 if valid rtt 0 otherwise
486  */
487 static int
489 {
490  u32 mrtt = 0;
491 
492  /* Karn's rule, part 1. Don't use retransmitted segments to estimate
493  * RTT because they're ambiguous. */
494  if (tcp_in_cong_recovery (tc) || tc->sack_sb.sacked_bytes)
495  {
496  if (tcp_in_recovery (tc))
497  return 0;
498  goto done;
499  }
500 
501  if (tc->rtt_ts && seq_geq (ack, tc->rtt_seq))
502  {
503  f64 sample = tcp_time_now_us (tc->c_thread_index) - tc->rtt_ts;
504  tc->mrtt_us = tc->mrtt_us + (sample - tc->mrtt_us) * 0.125;
505  mrtt = clib_max ((u32) (sample * THZ), 1);
506  /* Allow measuring of a new RTT */
507  tc->rtt_ts = 0;
508  }
509  /* As per RFC7323 TSecr can be used for RTTM only if the segment advances
510  * snd_una, i.e., the left side of the send window:
511  * seq_lt (tc->snd_una, ack). This is a condition for calling update_rtt */
512  else if (tcp_opts_tstamp (&tc->rcv_opts) && tc->rcv_opts.tsecr)
513  {
514  u32 now = tcp_tstamp (tc);
515  mrtt = clib_max (now - tc->rcv_opts.tsecr, 1);
516  }
517 
518  /* Ignore dubious measurements */
519  if (mrtt == 0 || mrtt > TCP_RTT_MAX)
520  goto done;
521 
522  tcp_estimate_rtt (tc, mrtt);
523 
524 done:
525 
526  /* If we got here something must've been ACKed so make sure boff is 0,
527  * even if mrtt is not valid since we update the rto lower */
528  tc->rto_boff = 0;
529  tcp_update_rto (tc);
530 
531  return 0;
532 }
533 
534 static void
536 {
537  u8 thread_index = vlib_num_workers ()? 1 : 0;
538  int mrtt;
539 
540  if (tc->rtt_ts)
541  {
542  tc->mrtt_us = tcp_time_now_us (thread_index) - tc->rtt_ts;
543  tc->mrtt_us = clib_max (tc->mrtt_us, 0.0001);
544  mrtt = clib_max ((u32) (tc->mrtt_us * THZ), 1);
545  tc->rtt_ts = 0;
546  }
547  else
548  {
549  mrtt = tcp_time_now_w_thread (thread_index) - tc->rcv_opts.tsecr;
550  mrtt = clib_max (mrtt, 1);
551  /* Due to retransmits we don't know the initial mrtt */
552  if (tc->rto_boff && mrtt > 1 * THZ)
553  mrtt = 1 * THZ;
554  tc->mrtt_us = (f64) mrtt *TCP_TICK;
555  }
556 
557  if (mrtt > 0 && mrtt < TCP_RTT_MAX)
558  tcp_estimate_rtt (tc, mrtt);
559  tcp_update_rto (tc);
560 }
561 
562 /**
563  * Dequeue bytes for connections that have received acks in last burst
564  */
565 static void
567 {
568  u32 thread_index = wrk->vm->thread_index;
569  u32 *pending_deq_acked;
570  tcp_connection_t *tc;
571  int i;
572 
573  if (!vec_len (wrk->pending_deq_acked))
574  return;
575 
576  pending_deq_acked = wrk->pending_deq_acked;
577  for (i = 0; i < vec_len (pending_deq_acked); i++)
578  {
579  tc = tcp_connection_get (pending_deq_acked[i], thread_index);
580  tc->flags &= ~TCP_CONN_DEQ_PENDING;
581 
582  if (PREDICT_FALSE (!tc->burst_acked))
583  continue;
584 
585  /* Dequeue the newly ACKed bytes */
586  session_tx_fifo_dequeue_drop (&tc->connection, tc->burst_acked);
587  tc->burst_acked = 0;
588  tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una);
589 
590  if (PREDICT_FALSE (tc->flags & TCP_CONN_PSH_PENDING))
591  {
592  if (seq_leq (tc->psh_seq, tc->snd_una))
593  tc->flags &= ~TCP_CONN_PSH_PENDING;
594  }
595 
596  /* If everything has been acked, stop retransmit timer
597  * otherwise update. */
599 
600  /* If not congested, update pacer based on our new
601  * cwnd estimate */
602  if (!tcp_in_fastrecovery (tc))
604  }
605  _vec_len (wrk->pending_deq_acked) = 0;
606 }
607 
608 static void
610 {
611  if (!(tc->flags & TCP_CONN_DEQ_PENDING))
612  {
613  vec_add1 (wrk->pending_deq_acked, tc->c_c_index);
614  tc->flags |= TCP_CONN_DEQ_PENDING;
615  }
616  tc->burst_acked += tc->bytes_acked + tc->sack_sb.snd_una_adv;
617 }
618 
619 /**
620  * Check if duplicate ack as per RFC5681 Sec. 2
621  */
622 static u8
624  u32 prev_snd_una)
625 {
626  return ((vnet_buffer (b)->tcp.ack_number == prev_snd_una)
627  && seq_gt (tc->snd_nxt, tc->snd_una)
628  && (vnet_buffer (b)->tcp.seq_end == vnet_buffer (b)->tcp.seq_number)
629  && (prev_snd_wnd == tc->snd_wnd));
630 }
631 
632 /**
633  * Checks if ack is a congestion control event.
634  */
635 static u8
637  u32 prev_snd_wnd, u32 prev_snd_una, u8 * is_dack)
638 {
639  /* Check if ack is duplicate. Per RFC 6675, ACKs that SACK new data are
640  * defined to be 'duplicate' */
641  *is_dack = tc->sack_sb.last_sacked_bytes
642  || tcp_ack_is_dupack (tc, b, prev_snd_wnd, prev_snd_una);
643 
644  return ((*is_dack || tcp_in_cong_recovery (tc)) && !tcp_is_lost_fin (tc));
645 }
646 
647 #ifndef CLIB_MARCH_VARIANT
648 static u32
650 {
651  ASSERT (!pool_is_free_index (sb->holes, hole - sb->holes));
652  return hole - sb->holes;
653 }
654 
655 static u32
657 {
658  return hole->end - hole->start;
659 }
660 
663 {
664  if (index != TCP_INVALID_SACK_HOLE_INDEX)
665  return pool_elt_at_index (sb->holes, index);
666  return 0;
667 }
668 
671 {
672  if (hole->next != TCP_INVALID_SACK_HOLE_INDEX)
673  return pool_elt_at_index (sb->holes, hole->next);
674  return 0;
675 }
676 
679 {
680  if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX)
681  return pool_elt_at_index (sb->holes, hole->prev);
682  return 0;
683 }
684 
687 {
688  if (sb->head != TCP_INVALID_SACK_HOLE_INDEX)
689  return pool_elt_at_index (sb->holes, sb->head);
690  return 0;
691 }
692 
695 {
696  if (sb->tail != TCP_INVALID_SACK_HOLE_INDEX)
697  return pool_elt_at_index (sb->holes, sb->tail);
698  return 0;
699 }
700 
701 static void
703 {
704  sack_scoreboard_hole_t *next, *prev;
705 
706  if (hole->next != TCP_INVALID_SACK_HOLE_INDEX)
707  {
708  next = pool_elt_at_index (sb->holes, hole->next);
709  next->prev = hole->prev;
710  }
711  else
712  {
713  sb->tail = hole->prev;
714  }
715 
716  if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX)
717  {
718  prev = pool_elt_at_index (sb->holes, hole->prev);
719  prev->next = hole->next;
720  }
721  else
722  {
723  sb->head = hole->next;
724  }
725 
726  if (scoreboard_hole_index (sb, hole) == sb->cur_rxt_hole)
727  sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
728 
729  /* Poison the entry */
730  if (CLIB_DEBUG > 0)
731  clib_memset (hole, 0xfe, sizeof (*hole));
732 
733  pool_put (sb->holes, hole);
734 }
735 
736 static sack_scoreboard_hole_t *
738  u32 start, u32 end)
739 {
740  sack_scoreboard_hole_t *hole, *next, *prev;
741  u32 hole_index;
742 
743  pool_get (sb->holes, hole);
744  clib_memset (hole, 0, sizeof (*hole));
745 
746  hole->start = start;
747  hole->end = end;
748  hole_index = scoreboard_hole_index (sb, hole);
749 
750  prev = scoreboard_get_hole (sb, prev_index);
751  if (prev)
752  {
753  hole->prev = prev_index;
754  hole->next = prev->next;
755 
756  if ((next = scoreboard_next_hole (sb, hole)))
757  next->prev = hole_index;
758  else
759  sb->tail = hole_index;
760 
761  prev->next = hole_index;
762  }
763  else
764  {
765  sb->head = hole_index;
766  hole->prev = TCP_INVALID_SACK_HOLE_INDEX;
767  hole->next = TCP_INVALID_SACK_HOLE_INDEX;
768  }
769 
770  return hole;
771 }
772 #endif /* CLIB_MARCH_VARIANT */
773 
774 #ifndef CLIB_MARCH_VARIANT
775 static void
777 {
779  u32 bytes = 0, blks = 0;
780 
781  sb->last_lost_bytes = 0;
782  sb->lost_bytes = 0;
783  sb->sacked_bytes = 0;
784  left = scoreboard_last_hole (sb);
785  if (!left)
786  return;
787 
788  if (seq_gt (sb->high_sacked, left->end))
789  {
790  bytes = sb->high_sacked - left->end;
791  blks = 1;
792  }
793 
794  while ((right = left)
795  && bytes < (TCP_DUPACK_THRESHOLD - 1) * tc->snd_mss
796  && blks < TCP_DUPACK_THRESHOLD
797  /* left not updated if above conditions fail */
798  && (left = scoreboard_prev_hole (sb, right)))
799  {
800  bytes += right->start - left->end;
801  blks++;
802  }
803 
804  /* left is first lost */
805  if (left)
806  {
807  do
808  {
809  sb->lost_bytes += scoreboard_hole_bytes (right);
810  sb->last_lost_bytes += left->is_lost ? 0 : left->end - left->start;
811  left->is_lost = 1;
812  left = scoreboard_prev_hole (sb, right);
813  if (left)
814  bytes += right->start - left->end;
815  }
816  while ((right = left));
817  }
818 
819  sb->sacked_bytes = bytes;
820 }
821 
822 /**
823  * Figure out the next hole to retransmit
824  *
825  * Follows logic proposed in RFC6675 Sec. 4, NextSeg()
826  */
829  sack_scoreboard_hole_t * start,
830  u8 have_unsent, u8 * can_rescue, u8 * snd_limited)
831 {
832  sack_scoreboard_hole_t *hole = 0;
833 
834  hole = start ? start : scoreboard_first_hole (sb);
835  while (hole && seq_leq (hole->end, sb->high_rxt) && hole->is_lost)
836  hole = scoreboard_next_hole (sb, hole);
837 
838  /* Nothing, return */
839  if (!hole)
840  {
841  sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
842  return 0;
843  }
844 
845  /* Rule (1): if higher than rxt, less than high_sacked and lost */
846  if (hole->is_lost && seq_lt (hole->start, sb->high_sacked))
847  {
848  sb->cur_rxt_hole = scoreboard_hole_index (sb, hole);
849  }
850  else
851  {
852  /* Rule (2): available unsent data */
853  if (have_unsent)
854  {
855  sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
856  return 0;
857  }
858  /* Rule (3): if hole not lost */
859  else if (seq_lt (hole->start, sb->high_sacked))
860  {
861  *snd_limited = 0;
862  sb->cur_rxt_hole = scoreboard_hole_index (sb, hole);
863  }
864  /* Rule (4): if hole beyond high_sacked */
865  else
866  {
867  ASSERT (seq_geq (hole->start, sb->high_sacked));
868  *snd_limited = 1;
869  *can_rescue = 1;
870  /* HighRxt MUST NOT be updated */
871  return 0;
872  }
873  }
874 
875  if (hole && seq_lt (sb->high_rxt, hole->start))
876  sb->high_rxt = hole->start;
877 
878  return hole;
879 }
880 #endif /* CLIB_MARCH_VARIANT */
881 
882 static void
884 {
886  hole = scoreboard_first_hole (sb);
887  if (hole)
888  {
889  snd_una = seq_gt (snd_una, hole->start) ? snd_una : hole->start;
890  sb->cur_rxt_hole = sb->head;
891  }
892  sb->high_rxt = snd_una;
893  sb->rescue_rxt = snd_una - 1;
894 }
895 
896 #ifndef CLIB_MARCH_VARIANT
897 void
899 {
900  sb->head = TCP_INVALID_SACK_HOLE_INDEX;
901  sb->tail = TCP_INVALID_SACK_HOLE_INDEX;
902  sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
903 }
904 
905 void
907 {
909  while ((hole = scoreboard_first_hole (sb)))
910  {
911  scoreboard_remove_hole (sb, hole);
912  }
913  ASSERT (sb->head == sb->tail && sb->head == TCP_INVALID_SACK_HOLE_INDEX);
914  ASSERT (pool_elts (sb->holes) == 0);
915  sb->sacked_bytes = 0;
916  sb->last_sacked_bytes = 0;
917  sb->last_bytes_delivered = 0;
918  sb->snd_una_adv = 0;
919  sb->high_sacked = 0;
920  sb->high_rxt = 0;
921  sb->lost_bytes = 0;
922  sb->last_lost_bytes = 0;
923  sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
924 }
925 #endif /* CLIB_MARCH_VARIANT */
926 
927 /**
928  * Test that scoreboard is sane after recovery
929  *
930  * Returns 1 if scoreboard is empty or if first hole beyond
931  * snd_una.
932  */
933 static u8
935 {
937  hole = scoreboard_first_hole (&tc->sack_sb);
938  return (!hole || (seq_geq (hole->start, tc->snd_una)
939  && seq_lt (hole->end, tc->snd_nxt)));
940 }
941 
942 #ifndef CLIB_MARCH_VARIANT
943 
944 void
946 {
947  sack_scoreboard_hole_t *hole, *next_hole, *last_hole;
948  u32 blk_index = 0, old_sacked_bytes, hole_index;
949  sack_scoreboard_t *sb = &tc->sack_sb;
950  sack_block_t *blk, tmp;
951  int i, j;
952 
953  sb->last_sacked_bytes = 0;
954  sb->last_bytes_delivered = 0;
955  sb->snd_una_adv = 0;
956 
957  if (!tcp_opts_sack (&tc->rcv_opts)
958  && sb->head == TCP_INVALID_SACK_HOLE_INDEX)
959  return;
960 
961  old_sacked_bytes = sb->sacked_bytes;
962 
963  /* Remove invalid blocks */
964  blk = tc->rcv_opts.sacks;
965  while (blk < vec_end (tc->rcv_opts.sacks))
966  {
967  if (seq_lt (blk->start, blk->end)
968  && seq_gt (blk->start, tc->snd_una)
969  && seq_gt (blk->start, ack)
970  && seq_lt (blk->start, tc->snd_nxt)
971  && seq_leq (blk->end, tc->snd_nxt))
972  {
973  blk++;
974  continue;
975  }
976  vec_del1 (tc->rcv_opts.sacks, blk - tc->rcv_opts.sacks);
977  }
978 
979  /* Add block for cumulative ack */
980  if (seq_gt (ack, tc->snd_una))
981  {
982  tmp.start = tc->snd_una;
983  tmp.end = ack;
984  vec_add1 (tc->rcv_opts.sacks, tmp);
985  }
986 
987  if (vec_len (tc->rcv_opts.sacks) == 0)
988  return;
989 
990  tcp_scoreboard_trace_add (tc, ack);
991 
992  /* Make sure blocks are ordered */
993  for (i = 0; i < vec_len (tc->rcv_opts.sacks); i++)
994  for (j = i + 1; j < vec_len (tc->rcv_opts.sacks); j++)
995  if (seq_lt (tc->rcv_opts.sacks[j].start, tc->rcv_opts.sacks[i].start))
996  {
997  tmp = tc->rcv_opts.sacks[i];
998  tc->rcv_opts.sacks[i] = tc->rcv_opts.sacks[j];
999  tc->rcv_opts.sacks[j] = tmp;
1000  }
1001 
1002  if (sb->head == TCP_INVALID_SACK_HOLE_INDEX)
1003  {
1004  /* If no holes, insert the first that covers all outstanding bytes */
1006  tc->snd_una, tc->snd_nxt);
1007  sb->tail = scoreboard_hole_index (sb, last_hole);
1008  tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1];
1009  sb->high_sacked = tmp.end;
1010  }
1011  else
1012  {
1013  /* If we have holes but snd_una_max is beyond the last hole, update
1014  * last hole end */
1015  tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1];
1016  last_hole = scoreboard_last_hole (sb);
1017  if (seq_gt (tc->snd_nxt, last_hole->end))
1018  {
1019  if (seq_geq (last_hole->start, sb->high_sacked))
1020  {
1021  last_hole->end = tc->snd_nxt;
1022  }
1023  /* New hole after high sacked block */
1024  else if (seq_lt (sb->high_sacked, tc->snd_nxt))
1025  {
1026  scoreboard_insert_hole (sb, sb->tail, sb->high_sacked,
1027  tc->snd_nxt);
1028  }
1029  }
1030  /* Keep track of max byte sacked for when the last hole
1031  * is acked */
1032  if (seq_gt (tmp.end, sb->high_sacked))
1033  sb->high_sacked = tmp.end;
1034  }
1035 
1036  /* Walk the holes with the SACK blocks */
1037  hole = pool_elt_at_index (sb->holes, sb->head);
1038  while (hole && blk_index < vec_len (tc->rcv_opts.sacks))
1039  {
1040  blk = &tc->rcv_opts.sacks[blk_index];
1041  if (seq_leq (blk->start, hole->start))
1042  {
1043  /* Block covers hole. Remove hole */
1044  if (seq_geq (blk->end, hole->end))
1045  {
1046  next_hole = scoreboard_next_hole (sb, hole);
1047 
1048  /* Byte accounting: snd_una needs to be advanced */
1049  if (blk->end == ack)
1050  {
1051  if (next_hole)
1052  {
1053  if (seq_lt (ack, next_hole->start))
1054  sb->snd_una_adv = next_hole->start - ack;
1055  sb->last_bytes_delivered +=
1056  next_hole->start - hole->end;
1057  }
1058  else
1059  {
1060  ASSERT (seq_geq (sb->high_sacked, ack));
1061  sb->snd_una_adv = sb->high_sacked - ack;
1062  sb->last_bytes_delivered += sb->high_sacked - hole->end;
1063  }
1064  }
1065  scoreboard_remove_hole (sb, hole);
1066  hole = next_hole;
1067  }
1068  /* Partial 'head' overlap */
1069  else
1070  {
1071  if (seq_gt (blk->end, hole->start))
1072  {
1073  hole->start = blk->end;
1074  }
1075  blk_index++;
1076  }
1077  }
1078  else
1079  {
1080  /* Hole must be split */
1081  if (seq_lt (blk->end, hole->end))
1082  {
1083  hole_index = scoreboard_hole_index (sb, hole);
1084  next_hole = scoreboard_insert_hole (sb, hole_index, blk->end,
1085  hole->end);
1086 
1087  /* Pool might've moved */
1088  hole = scoreboard_get_hole (sb, hole_index);
1089  hole->end = blk->start;
1090  blk_index++;
1091  ASSERT (hole->next == scoreboard_hole_index (sb, next_hole));
1092  }
1093  else if (seq_lt (blk->start, hole->end))
1094  {
1095  hole->end = blk->start;
1096  }
1097  hole = scoreboard_next_hole (sb, hole);
1098  }
1099  }
1100 
1101  if (pool_elts (sb->holes) == 1)
1102  {
1103  hole = scoreboard_first_hole (sb);
1104  if (hole->start == ack + sb->snd_una_adv && hole->end == tc->snd_nxt)
1105  scoreboard_remove_hole (sb, hole);
1106  }
1107 
1108  scoreboard_update_bytes (tc, sb);
1109  sb->last_sacked_bytes = sb->sacked_bytes
1110  - (old_sacked_bytes - sb->last_bytes_delivered);
1111 
1112  ASSERT (sb->last_sacked_bytes <= sb->sacked_bytes || tcp_in_recovery (tc));
1113  ASSERT (sb->sacked_bytes == 0 || tcp_in_recovery (tc)
1114  || sb->sacked_bytes < tc->snd_nxt - seq_max (tc->snd_una, ack));
1115  ASSERT (sb->last_sacked_bytes + sb->lost_bytes <= tc->snd_nxt
1116  - seq_max (tc->snd_una, ack) || tcp_in_recovery (tc));
1118  || sb->holes[sb->head].start == ack + sb->snd_una_adv);
1119  ASSERT (sb->last_lost_bytes <= sb->lost_bytes);
1120 
1121  TCP_EVT_DBG (TCP_EVT_CC_SCOREBOARD, tc);
1122 }
1123 #endif /* CLIB_MARCH_VARIANT */
1124 
1125 /**
1126  * Try to update snd_wnd based on feedback received from peer.
1127  *
1128  * If successful, and new window is 'effectively' 0, activate persist
1129  * timer.
1130  */
1131 static void
1132 tcp_update_snd_wnd (tcp_connection_t * tc, u32 seq, u32 ack, u32 snd_wnd)
1133 {
1134  /* If (SND.WL1 < SEG.SEQ or (SND.WL1 = SEG.SEQ and SND.WL2 =< SEG.ACK)), set
1135  * SND.WND <- SEG.WND, set SND.WL1 <- SEG.SEQ, and set SND.WL2 <- SEG.ACK */
1136  if (seq_lt (tc->snd_wl1, seq)
1137  || (tc->snd_wl1 == seq && seq_leq (tc->snd_wl2, ack)))
1138  {
1139  tc->snd_wnd = snd_wnd;
1140  tc->snd_wl1 = seq;
1141  tc->snd_wl2 = ack;
1142  TCP_EVT_DBG (TCP_EVT_SND_WND, tc);
1143 
1144  if (PREDICT_FALSE (tc->snd_wnd < tc->snd_mss))
1145  {
1146  /* Set persist timer if not set and we just got 0 wnd */
1147  if (!tcp_timer_is_active (tc, TCP_TIMER_PERSIST)
1148  && !tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT))
1149  tcp_persist_timer_set (tc);
1150  }
1151  else
1152  {
1154  if (PREDICT_FALSE (!tcp_in_recovery (tc) && tc->rto_boff > 0))
1155  {
1156  tc->rto_boff = 0;
1157  tcp_update_rto (tc);
1158  }
1159  }
1160  }
1161 }
1162 
1163 #ifndef CLIB_MARCH_VARIANT
1164 /**
1165  * Init loss recovery/fast recovery.
1166  *
1167  * Triggered by dup acks as opposed to timer timeout. Note that cwnd is
1168  * updated in @ref tcp_cc_handle_event after fast retransmit
1169  */
1170 void
1172 {
1173  tcp_fastrecovery_on (tc);
1174  tc->snd_congestion = tc->snd_nxt;
1175  tc->cwnd_acc_bytes = 0;
1176  tc->snd_rxt_bytes = 0;
1177  tc->prev_ssthresh = tc->ssthresh;
1178  tc->prev_cwnd = tc->cwnd;
1179  tc->cc_algo->congestion (tc);
1180  tc->fr_occurences += 1;
1181  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 4);
1182 }
1183 #endif /* CLIB_MARCH_VARIANT */
1184 
1185 static void
1187 {
1188  tc->rto_boff = 0;
1189  tcp_update_rto (tc);
1190  tc->snd_rxt_ts = 0;
1191  tc->rtt_ts = 0;
1192  tcp_recovery_off (tc);
1193  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3);
1194 }
1195 
1196 #ifndef CLIB_MARCH_VARIANT
1197 void
1199 {
1200  tc->snd_rxt_bytes = 0;
1201  tc->rcv_dupacks = 0;
1202  tc->rtt_ts = 0;
1203 
1204  tcp_fastrecovery_off (tc);
1206  tc->flags &= ~TCP_CONN_FRXT_PENDING;
1207 
1208  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3);
1209 }
1210 #endif /* CLIB_MARCH_VARIANT */
1211 
1212 static void
1214 {
1215  tc->cwnd = tc->prev_cwnd;
1216  tc->ssthresh = tc->prev_ssthresh;
1217  tc->rcv_dupacks = 0;
1218  if (tcp_in_recovery (tc))
1219  {
1220  tcp_cc_recovery_exit (tc);
1221  tc->snd_nxt = seq_max (tc->snd_nxt, tc->snd_congestion);
1222  }
1223  else if (tcp_in_fastrecovery (tc))
1224  {
1226  }
1227  tcp_cc_undo_recovery (tc);
1228  ASSERT (tc->rto_boff == 0);
1229  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 5);
1230 }
1231 
1232 static inline u8
1234 {
1235  return (tcp_in_recovery (tc) && tc->rto_boff == 1
1236  && tc->snd_rxt_ts
1237  && tcp_opts_tstamp (&tc->rcv_opts)
1238  && timestamp_lt (tc->rcv_opts.tsecr, tc->snd_rxt_ts));
1239 }
1240 
1241 static inline u8
1243 {
1244  return (tcp_in_fastrecovery (tc)
1245  && tc->cwnd > tc->ssthresh + 3 * tc->snd_mss);
1246 }
1247 
1248 static u8
1250 {
1251  return (tcp_cc_is_spurious_timeout_rxt (tc)
1252  || tcp_cc_is_spurious_fast_rxt (tc));
1253 }
1254 
1255 static int
1257 {
1260  {
1262  return 1;
1263  }
1264 
1265  if (tcp_in_recovery (tc))
1266  tcp_cc_recovery_exit (tc);
1267  else if (tcp_in_fastrecovery (tc))
1268  {
1269  tcp_cc_recovered (tc);
1271  }
1272 
1273  ASSERT (tc->rto_boff == 0);
1274  ASSERT (!tcp_in_cong_recovery (tc));
1276  return 0;
1277 }
1278 
1279 static void
1281 {
1283 
1284  /* Congestion avoidance */
1285  tcp_cc_rcv_ack (tc, rs);
1286 
1287  /* If a cumulative ack, make sure dupacks is 0 */
1288  tc->rcv_dupacks = 0;
1289 
1290  /* When dupacks hits the threshold we only enter fast retransmit if
1291  * cumulative ack covers more than snd_congestion. Should snd_una
1292  * wrap this test may fail under otherwise valid circumstances.
1293  * Therefore, proactively update snd_congestion when wrap detected. */
1294  if (PREDICT_FALSE
1295  (seq_leq (tc->snd_congestion, tc->snd_una - tc->bytes_acked)
1296  && seq_gt (tc->snd_congestion, tc->snd_una)))
1297  tc->snd_congestion = tc->snd_una - 1;
1298 }
1299 
1300 static u8
1302 {
1303  return (TCP_DUPACK_THRESHOLD - 1) * tc->snd_mss < tc->sack_sb.sacked_bytes;
1304 }
1305 
1306 static u8
1308 {
1309  return (tc->rcv_dupacks == TCP_DUPACK_THRESHOLD
1310  || tcp_should_fastrecover_sack (tc));
1311 }
1312 
1313 /**
1314  * One function to rule them all ... and in the darkness bind them
1315  */
1316 static void
1318  u32 is_dack)
1319 {
1320  u32 rxt_delivered;
1321 
1322  if (tcp_in_fastrecovery (tc) && tcp_opts_sack_permitted (&tc->rcv_opts))
1323  {
1324  if (tc->bytes_acked)
1325  goto partial_ack;
1327  return;
1328  }
1329  /*
1330  * Duplicate ACK. Check if we should enter fast recovery, or if already in
1331  * it account for the bytes that left the network.
1332  */
1333  else if (is_dack && !tcp_in_recovery (tc))
1334  {
1335  TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc, 1);
1336  ASSERT (tc->snd_una != tc->snd_nxt || tc->sack_sb.last_sacked_bytes);
1337 
1338  tc->rcv_dupacks++;
1339 
1340  /* Pure duplicate ack. If some data got acked, it's handled lower */
1341  if (tc->rcv_dupacks > TCP_DUPACK_THRESHOLD && !tc->bytes_acked)
1342  {
1343  ASSERT (tcp_in_fastrecovery (tc));
1345  return;
1346  }
1347  else if (tcp_should_fastrecover (tc))
1348  {
1349  u32 pacer_wnd;
1350 
1351  ASSERT (!tcp_in_fastrecovery (tc));
1352 
1353  /* Heuristic to catch potential late dupacks
1354  * after fast retransmit exits */
1355  if (is_dack && tc->snd_una == tc->snd_congestion
1356  && timestamp_leq (tc->rcv_opts.tsecr, tc->tsecr_last_ack))
1357  {
1358  tc->rcv_dupacks = 0;
1359  return;
1360  }
1361 
1364 
1365  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1366  {
1367  tc->cwnd = tc->ssthresh;
1368  scoreboard_init_high_rxt (&tc->sack_sb, tc->snd_una);
1369  }
1370  else
1371  {
1372  /* Post retransmit update cwnd to ssthresh and account for the
1373  * three segments that have left the network and should've been
1374  * buffered at the receiver XXX */
1375  tc->cwnd = tc->ssthresh + 3 * tc->snd_mss;
1376  }
1377 
1378  /* Constrain rate until we get a partial ack */
1379  pacer_wnd = clib_max (0.1 * tc->cwnd, 2 * tc->snd_mss);
1380  tcp_connection_tx_pacer_reset (tc, pacer_wnd,
1381  0 /* start bucket */ );
1383  return;
1384  }
1385  else if (!tc->bytes_acked
1386  || (tc->bytes_acked && !tcp_in_cong_recovery (tc)))
1387  {
1389  return;
1390  }
1391  else
1392  goto partial_ack;
1393  }
1394  /* Don't allow entry in fast recovery if still in recovery, for now */
1395  else if (0 && is_dack && tcp_in_recovery (tc))
1396  {
1397  /* If of of the two conditions lower hold, reset dupacks because
1398  * we're probably after timeout (RFC6582 heuristics).
1399  * If Cumulative ack does not cover more than congestion threshold,
1400  * and:
1401  * 1) The following doesn't hold: The congestion window is greater
1402  * than SMSS bytes and the difference between highest_ack
1403  * and prev_highest_ack is at most 4*SMSS bytes
1404  * 2) Echoed timestamp in the last non-dup ack does not equal the
1405  * stored timestamp
1406  */
1407  if (seq_leq (tc->snd_una, tc->snd_congestion)
1408  && ((!(tc->cwnd > tc->snd_mss
1409  && tc->bytes_acked <= 4 * tc->snd_mss))
1410  || (tc->rcv_opts.tsecr != tc->tsecr_last_ack)))
1411  {
1412  tc->rcv_dupacks = 0;
1413  return;
1414  }
1415  }
1416 
1417  if (!tc->bytes_acked)
1418  return;
1419 
1420 partial_ack:
1421  TCP_EVT_DBG (TCP_EVT_CC_PACK, tc);
1422 
1423  /*
1424  * Legitimate ACK. 1) See if we can exit recovery
1425  */
1426 
1427  /* Update the pacing rate. For the first partial ack we move from
1428  * the artificially constrained rate to the one after congestion */
1430 
1431  if (seq_geq (tc->snd_una, tc->snd_congestion))
1432  {
1434 
1435  /* If spurious return, we've already updated everything */
1436  if (tcp_cc_recover (tc))
1437  {
1438  tc->tsecr_last_ack = tc->rcv_opts.tsecr;
1439  return;
1440  }
1441 
1442  /* Treat as congestion avoidance ack */
1443  tcp_cc_rcv_ack (tc, rs);
1444  return;
1445  }
1446 
1447  /*
1448  * Legitimate ACK. 2) If PARTIAL ACK try to retransmit
1449  */
1450 
1451  /* XXX limit this only to first partial ack? */
1453 
1454  /* RFC6675: If the incoming ACK is a cumulative acknowledgment,
1455  * reset dupacks to 0. Also needed if in congestion recovery */
1456  tc->rcv_dupacks = 0;
1457 
1458  /* Post RTO timeout don't try anything fancy */
1459  if (tcp_in_recovery (tc))
1460  {
1461  tcp_cc_rcv_ack (tc, rs);
1462  transport_add_tx_event (&tc->connection);
1463  return;
1464  }
1465 
1466  /* Remove retransmitted bytes that have been delivered */
1467  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1468  {
1469  ASSERT (tc->bytes_acked + tc->sack_sb.snd_una_adv
1470  >= tc->sack_sb.last_bytes_delivered
1471  || (tc->flags & TCP_CONN_FINSNT));
1472 
1473  /* If we have sacks and we haven't gotten an ack beyond high_rxt,
1474  * remove sacked bytes delivered */
1475  if (seq_lt (tc->snd_una, tc->sack_sb.high_rxt))
1476  {
1477  rxt_delivered = tc->bytes_acked + tc->sack_sb.snd_una_adv
1478  - tc->sack_sb.last_bytes_delivered;
1479  ASSERT (tc->snd_rxt_bytes >= rxt_delivered);
1480  tc->snd_rxt_bytes -= rxt_delivered;
1481  }
1482  else
1483  {
1484  /* Apparently all retransmitted holes have been acked */
1485  tc->snd_rxt_bytes = 0;
1486  tc->sack_sb.high_rxt = tc->snd_una;
1487  }
1488  }
1489  else
1490  {
1492  if (tc->snd_rxt_bytes > tc->bytes_acked)
1493  tc->snd_rxt_bytes -= tc->bytes_acked;
1494  else
1495  tc->snd_rxt_bytes = 0;
1496  }
1497 
1499 
1500  /*
1501  * Since this was a partial ack, try to retransmit some more data
1502  */
1504 }
1505 
1506 /**
1507  * Process incoming ACK
1508  */
1509 static int
1511  tcp_header_t * th, u32 * error)
1512 {
1513  u32 prev_snd_wnd, prev_snd_una;
1514  tcp_rate_sample_t rs = { 0 };
1515  u8 is_dack;
1516 
1517  TCP_EVT_DBG (TCP_EVT_CC_STAT, tc);
1518 
1519  /* If the ACK acks something not yet sent (SEG.ACK > SND.NXT) */
1520  if (PREDICT_FALSE (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt)))
1521  {
1522  /* We've probably entered recovery and the peer still has some
1523  * of the data we've sent. Update snd_nxt and accept the ack */
1524  if (seq_leq (vnet_buffer (b)->tcp.ack_number, tc->snd_una_max)
1525  && seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_una))
1526  {
1527  tc->snd_nxt = vnet_buffer (b)->tcp.ack_number;
1528  goto process_ack;
1529  }
1530 
1531  tc->errors.above_ack_wnd += 1;
1532  *error = TCP_ERROR_ACK_FUTURE;
1533  TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 0,
1534  vnet_buffer (b)->tcp.ack_number);
1535  return -1;
1536  }
1537 
1538  /* If old ACK, probably it's an old dupack */
1539  if (PREDICT_FALSE (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una)))
1540  {
1541  tc->errors.below_ack_wnd += 1;
1542  *error = TCP_ERROR_ACK_OLD;
1543  TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 1,
1544  vnet_buffer (b)->tcp.ack_number);
1545  if (tcp_in_fastrecovery (tc) && tc->rcv_dupacks == TCP_DUPACK_THRESHOLD)
1546  tcp_cc_handle_event (tc, 0, 1);
1547  /* Don't drop yet */
1548  return 0;
1549  }
1550 
1551 process_ack:
1552 
1553  /*
1554  * Looks okay, process feedback
1555  */
1556 
1557  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1558  tcp_rcv_sacks (tc, vnet_buffer (b)->tcp.ack_number);
1559 
1560  prev_snd_wnd = tc->snd_wnd;
1561  prev_snd_una = tc->snd_una;
1562  tcp_update_snd_wnd (tc, vnet_buffer (b)->tcp.seq_number,
1563  vnet_buffer (b)->tcp.ack_number,
1564  clib_net_to_host_u16 (th->window) << tc->snd_wscale);
1565  tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una;
1566  tc->snd_una = vnet_buffer (b)->tcp.ack_number + tc->sack_sb.snd_una_adv;
1567  tcp_validate_txf_size (tc, tc->bytes_acked);
1568 
1569  if (tc->bytes_acked)
1570  {
1571  tcp_program_dequeue (wrk, tc);
1572  tcp_update_rtt (tc, vnet_buffer (b)->tcp.ack_number);
1573  }
1574 
1575  if (tc->flags & TCP_CONN_RATE_SAMPLE)
1576  tcp_bt_sample_delivery_rate (tc, &rs);
1577 
1578  TCP_EVT_DBG (TCP_EVT_ACK_RCVD, tc);
1579 
1580  /*
1581  * Check if we have congestion event
1582  */
1583 
1584  if (tcp_ack_is_cc_event (tc, b, prev_snd_wnd, prev_snd_una, &is_dack))
1585  {
1586  tcp_cc_handle_event (tc, &rs, is_dack);
1587  tc->dupacks_in += is_dack;
1588  if (!tcp_in_cong_recovery (tc))
1589  {
1590  *error = TCP_ERROR_ACK_OK;
1591  return 0;
1592  }
1593  *error = TCP_ERROR_ACK_DUP;
1594  if (vnet_buffer (b)->tcp.data_len || tcp_is_fin (th))
1595  return 0;
1596  return -1;
1597  }
1598 
1599  /*
1600  * Update congestion control (slow start/congestion avoidance)
1601  */
1602  tcp_cc_update (tc, &rs);
1603  *error = TCP_ERROR_ACK_OK;
1604  return 0;
1605 }
1606 
1607 static void
1609 {
1610  if (!tcp_disconnect_pending (tc))
1611  {
1612  vec_add1 (wrk->pending_disconnects, tc->c_c_index);
1614  }
1615 }
1616 
1617 static void
1619 {
1620  u32 thread_index, *pending_disconnects;
1621  tcp_connection_t *tc;
1622  int i;
1623 
1624  if (!vec_len (wrk->pending_disconnects))
1625  return;
1626 
1627  thread_index = wrk->vm->thread_index;
1628  pending_disconnects = wrk->pending_disconnects;
1629  for (i = 0; i < vec_len (pending_disconnects); i++)
1630  {
1631  tc = tcp_connection_get (pending_disconnects[i], thread_index);
1633  session_transport_closing_notify (&tc->connection);
1634  }
1635  _vec_len (wrk->pending_disconnects) = 0;
1636 }
1637 
1638 static void
1640  u32 * error)
1641 {
1642  /* Reject out-of-order fins */
1643  if (vnet_buffer (b)->tcp.seq_end != tc->rcv_nxt)
1644  return;
1645 
1646  /* Account for the FIN and send ack */
1647  tc->rcv_nxt += 1;
1648  tcp_program_ack (tc);
1649  /* Enter CLOSE-WAIT and notify session. To avoid lingering
1650  * in CLOSE-WAIT, set timer (reuse WAITCLOSE). */
1651  tcp_connection_set_state (tc, TCP_STATE_CLOSE_WAIT);
1652  tcp_program_disconnect (wrk, tc);
1653  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
1654  TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc);
1655  *error = TCP_ERROR_FIN_RCVD;
1656 }
1657 
1658 #ifndef CLIB_MARCH_VARIANT
1659 static u8
1661 {
1662  int i;
1663  for (i = 1; i < vec_len (sacks); i++)
1664  {
1665  if (sacks[i - 1].end == sacks[i].start)
1666  return 0;
1667  }
1668  return 1;
1669 }
1670 
1671 /**
1672  * Build SACK list as per RFC2018.
1673  *
1674  * Makes sure the first block contains the segment that generated the current
1675  * ACK and the following ones are the ones most recently reported in SACK
1676  * blocks.
1677  *
1678  * @param tc TCP connection for which the SACK list is updated
1679  * @param start Start sequence number of the newest SACK block
1680  * @param end End sequence of the newest SACK block
1681  */
1682 void
1684 {
1685  sack_block_t *new_list = tc->snd_sacks_fl, *block = 0;
1686  int i;
1687 
1688  /* If the first segment is ooo add it to the list. Last write might've moved
1689  * rcv_nxt over the first segment. */
1690  if (seq_lt (tc->rcv_nxt, start))
1691  {
1692  vec_add2 (new_list, block, 1);
1693  block->start = start;
1694  block->end = end;
1695  }
1696 
1697  /* Find the blocks still worth keeping. */
1698  for (i = 0; i < vec_len (tc->snd_sacks); i++)
1699  {
1700  /* Discard if rcv_nxt advanced beyond current block */
1701  if (seq_leq (tc->snd_sacks[i].start, tc->rcv_nxt))
1702  continue;
1703 
1704  /* Merge or drop if segment overlapped by the new segment */
1705  if (block && (seq_geq (tc->snd_sacks[i].end, new_list[0].start)
1706  && seq_leq (tc->snd_sacks[i].start, new_list[0].end)))
1707  {
1708  if (seq_lt (tc->snd_sacks[i].start, new_list[0].start))
1709  new_list[0].start = tc->snd_sacks[i].start;
1710  if (seq_lt (new_list[0].end, tc->snd_sacks[i].end))
1711  new_list[0].end = tc->snd_sacks[i].end;
1712  continue;
1713  }
1714 
1715  /* Save to new SACK list if we have space. */
1716  if (vec_len (new_list) < TCP_MAX_SACK_BLOCKS)
1717  vec_add1 (new_list, tc->snd_sacks[i]);
1718  }
1719 
1720  ASSERT (vec_len (new_list) <= TCP_MAX_SACK_BLOCKS);
1721 
1722  /* Replace old vector with new one */
1723  vec_reset_length (tc->snd_sacks);
1724  tc->snd_sacks_fl = tc->snd_sacks;
1725  tc->snd_sacks = new_list;
1726 
1727  /* Segments should not 'touch' */
1728  ASSERT (tcp_sack_vector_is_sane (tc->snd_sacks));
1729 }
1730 
1731 u32
1733 {
1734  u32 bytes = 0, i;
1735  for (i = 0; i < vec_len (tc->snd_sacks); i++)
1736  bytes += tc->snd_sacks[i].end - tc->snd_sacks[i].start;
1737  return bytes;
1738 }
1739 #endif /* CLIB_MARCH_VARIANT */
1740 
1741 /** Enqueue data for delivery to application */
1742 static int
1744  u16 data_len)
1745 {
1746  int written, error = TCP_ERROR_ENQUEUED;
1747 
1748  ASSERT (seq_geq (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt));
1749  ASSERT (data_len);
1750  written = session_enqueue_stream_connection (&tc->connection, b, 0,
1751  1 /* queue event */ , 1);
1752  tc->bytes_in += written;
1753 
1754  TCP_EVT_DBG (TCP_EVT_INPUT, tc, 0, data_len, written);
1755 
1756  /* Update rcv_nxt */
1757  if (PREDICT_TRUE (written == data_len))
1758  {
1759  tc->rcv_nxt += written;
1760  }
1761  /* If more data written than expected, account for out-of-order bytes. */
1762  else if (written > data_len)
1763  {
1764  tc->rcv_nxt += written;
1765  TCP_EVT_DBG (TCP_EVT_CC_INPUT, tc, data_len, written);
1766  }
1767  else if (written > 0)
1768  {
1769  /* We've written something but FIFO is probably full now */
1770  tc->rcv_nxt += written;
1771  error = TCP_ERROR_PARTIALLY_ENQUEUED;
1772  }
1773  else
1774  {
1775  return TCP_ERROR_FIFO_FULL;
1776  }
1777 
1778  /* Update SACK list if need be */
1779  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1780  {
1781  /* Remove SACK blocks that have been delivered */
1782  tcp_update_sack_list (tc, tc->rcv_nxt, tc->rcv_nxt);
1783  }
1784 
1785  return error;
1786 }
1787 
1788 /** Enqueue out-of-order data */
1789 static int
1791  u16 data_len)
1792 {
1793  session_t *s0;
1794  int rv, offset;
1795 
1796  ASSERT (seq_gt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt));
1797  ASSERT (data_len);
1798 
1799  /* Enqueue out-of-order data with relative offset */
1800  rv = session_enqueue_stream_connection (&tc->connection, b,
1801  vnet_buffer (b)->tcp.seq_number -
1802  tc->rcv_nxt, 0 /* queue event */ ,
1803  0);
1804 
1805  /* Nothing written */
1806  if (rv)
1807  {
1808  TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, 0);
1809  return TCP_ERROR_FIFO_FULL;
1810  }
1811 
1812  TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, data_len);
1813  tc->bytes_in += data_len;
1814 
1815  /* Update SACK list if in use */
1816  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1817  {
1818  ooo_segment_t *newest;
1819  u32 start, end;
1820 
1821  s0 = session_get (tc->c_s_index, tc->c_thread_index);
1822 
1823  /* Get the newest segment from the fifo */
1824  newest = svm_fifo_newest_ooo_segment (s0->rx_fifo);
1825  if (newest)
1826  {
1827  offset = ooo_segment_offset_prod (s0->rx_fifo, newest);
1828  ASSERT (offset <= vnet_buffer (b)->tcp.seq_number - tc->rcv_nxt);
1829  start = tc->rcv_nxt + offset;
1830  end = start + ooo_segment_length (s0->rx_fifo, newest);
1831  tcp_update_sack_list (tc, start, end);
1833  TCP_EVT_DBG (TCP_EVT_CC_SACKS, tc);
1834  }
1835  }
1836 
1837  return TCP_ERROR_ENQUEUED_OOO;
1838 }
1839 
1840 /**
1841  * Check if ACK could be delayed. If ack can be delayed, it should return
1842  * true for a full frame. If we're always acking return 0.
1843  */
1844 always_inline int
1846 {
1847  /* Send ack if ... */
1848  if (TCP_ALWAYS_ACK
1849  /* just sent a rcv wnd 0
1850  || (tc->flags & TCP_CONN_SENT_RCV_WND0) != 0 */
1851  /* constrained to send ack */
1852  || (tc->flags & TCP_CONN_SNDACK) != 0
1853  /* we're almost out of tx wnd */
1854  || tcp_available_cc_snd_space (tc) < 4 * tc->snd_mss)
1855  return 0;
1856 
1857  return 1;
1858 }
1859 
1860 static int
1862 {
1863  u32 discard, first = b->current_length;
1864  vlib_main_t *vm = vlib_get_main ();
1865 
1866  /* Handle multi-buffer segments */
1867  if (n_bytes_to_drop > b->current_length)
1868  {
1869  if (!(b->flags & VLIB_BUFFER_NEXT_PRESENT))
1870  return -1;
1871  do
1872  {
1873  discard = clib_min (n_bytes_to_drop, b->current_length);
1874  vlib_buffer_advance (b, discard);
1875  b = vlib_get_buffer (vm, b->next_buffer);
1876  n_bytes_to_drop -= discard;
1877  }
1878  while (n_bytes_to_drop);
1879  if (n_bytes_to_drop > first)
1880  b->total_length_not_including_first_buffer -= n_bytes_to_drop - first;
1881  }
1882  else
1883  vlib_buffer_advance (b, n_bytes_to_drop);
1884  vnet_buffer (b)->tcp.data_len -= n_bytes_to_drop;
1885  return 0;
1886 }
1887 
1888 /**
1889  * Receive buffer for connection and handle acks
1890  *
1891  * It handles both in order or out-of-order data.
1892  */
1893 static int
1895  vlib_buffer_t * b)
1896 {
1897  u32 error, n_bytes_to_drop, n_data_bytes;
1898 
1899  vlib_buffer_advance (b, vnet_buffer (b)->tcp.data_offset);
1900  n_data_bytes = vnet_buffer (b)->tcp.data_len;
1901  ASSERT (n_data_bytes);
1902  tc->data_segs_in += 1;
1903 
1904  /* Handle out-of-order data */
1905  if (PREDICT_FALSE (vnet_buffer (b)->tcp.seq_number != tc->rcv_nxt))
1906  {
1907  /* Old sequence numbers allowed through because they overlapped
1908  * the rx window */
1909  if (seq_lt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt))
1910  {
1911  /* Completely in the past (possible retransmit). Ack
1912  * retransmissions since we may not have any data to send */
1913  if (seq_leq (vnet_buffer (b)->tcp.seq_end, tc->rcv_nxt))
1914  {
1915  tcp_program_ack (tc);
1916  error = TCP_ERROR_SEGMENT_OLD;
1917  goto done;
1918  }
1919 
1920  /* Chop off the bytes in the past and see if what is left
1921  * can be enqueued in order */
1922  n_bytes_to_drop = tc->rcv_nxt - vnet_buffer (b)->tcp.seq_number;
1923  n_data_bytes -= n_bytes_to_drop;
1924  vnet_buffer (b)->tcp.seq_number = tc->rcv_nxt;
1925  if (tcp_buffer_discard_bytes (b, n_bytes_to_drop))
1926  {
1927  error = TCP_ERROR_SEGMENT_OLD;
1928  goto done;
1929  }
1930  goto in_order;
1931  }
1932 
1933  /* RFC2581: Enqueue and send DUPACK for fast retransmit */
1934  error = tcp_session_enqueue_ooo (tc, b, n_data_bytes);
1935  tcp_program_dupack (tc);
1936  TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc, vnet_buffer (b)->tcp);
1937  tc->errors.above_data_wnd += seq_gt (vnet_buffer (b)->tcp.seq_end,
1938  tc->rcv_las + tc->rcv_wnd);
1939  goto done;
1940  }
1941 
1942 in_order:
1943 
1944  /* In order data, enqueue. Fifo figures out by itself if any out-of-order
1945  * segments can be enqueued after fifo tail offset changes. */
1946  error = tcp_session_enqueue_data (tc, b, n_data_bytes);
1947  if (tcp_can_delack (tc))
1948  {
1949  if (!tcp_timer_is_active (tc, TCP_TIMER_DELACK))
1950  tcp_timer_set (tc, TCP_TIMER_DELACK, tcp_cfg.delack_time);
1951  goto done;
1952  }
1953 
1954  tcp_program_ack (tc);
1955 
1956 done:
1957  return error;
1958 }
1959 
1960 typedef struct
1961 {
1964 } tcp_rx_trace_t;
1965 
1966 static u8 *
1967 format_tcp_rx_trace (u8 * s, va_list * args)
1968 {
1969  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1970  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1971  tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *);
1972  u32 indent = format_get_indent (s);
1973 
1974  s = format (s, "%U\n%U%U",
1975  format_tcp_header, &t->tcp_header, 128,
1976  format_white_space, indent,
1978 
1979  return s;
1980 }
1981 
1982 static u8 *
1983 format_tcp_rx_trace_short (u8 * s, va_list * args)
1984 {
1985  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1986  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1987  tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *);
1988 
1989  s = format (s, "%d -> %d (%U)",
1990  clib_net_to_host_u16 (t->tcp_header.dst_port),
1991  clib_net_to_host_u16 (t->tcp_header.src_port), format_tcp_state,
1992  t->tcp_connection.state);
1993 
1994  return s;
1995 }
1996 
1997 static void
1999  tcp_header_t * th0, vlib_buffer_t * b0, u8 is_ip4)
2000 {
2001  if (tc0)
2002  {
2003  clib_memcpy_fast (&t0->tcp_connection, tc0,
2004  sizeof (t0->tcp_connection));
2005  }
2006  else
2007  {
2008  th0 = tcp_buffer_hdr (b0);
2009  }
2010  clib_memcpy_fast (&t0->tcp_header, th0, sizeof (t0->tcp_header));
2011 }
2012 
2013 static void
2015  vlib_frame_t * frame, u8 is_ip4)
2016 {
2017  u32 *from, n_left;
2018 
2019  n_left = frame->n_vectors;
2020  from = vlib_frame_vector_args (frame);
2021 
2022  while (n_left >= 1)
2023  {
2024  tcp_connection_t *tc0;
2025  tcp_rx_trace_t *t0;
2026  tcp_header_t *th0;
2027  vlib_buffer_t *b0;
2028  u32 bi0;
2029 
2030  bi0 = from[0];
2031  b0 = vlib_get_buffer (vm, bi0);
2032 
2033  if (b0->flags & VLIB_BUFFER_IS_TRACED)
2034  {
2035  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
2036  tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
2037  vm->thread_index);
2038  th0 = tcp_buffer_hdr (b0);
2039  tcp_set_rx_trace_data (t0, tc0, th0, b0, is_ip4);
2040  }
2041 
2042  from += 1;
2043  n_left -= 1;
2044  }
2045 }
2046 
2047 always_inline void
2048 tcp_node_inc_counter_i (vlib_main_t * vm, u32 tcp4_node, u32 tcp6_node,
2049  u8 is_ip4, u32 evt, u32 val)
2050 {
2051  if (is_ip4)
2052  vlib_node_increment_counter (vm, tcp4_node, evt, val);
2053  else
2054  vlib_node_increment_counter (vm, tcp6_node, evt, val);
2055 }
2056 
2057 #define tcp_maybe_inc_counter(node_id, err, count) \
2058 { \
2059  if (next0 != tcp_next_drop (is_ip4)) \
2060  tcp_node_inc_counter_i (vm, tcp4_##node_id##_node.index, \
2061  tcp6_##node_id##_node.index, is_ip4, err, \
2062  1); \
2063 }
2064 #define tcp_inc_counter(node_id, err, count) \
2065  tcp_node_inc_counter_i (vm, tcp4_##node_id##_node.index, \
2066  tcp6_##node_id##_node.index, is_ip4, \
2067  err, count)
2068 #define tcp_maybe_inc_err_counter(cnts, err) \
2069 { \
2070  cnts[err] += (next0 != tcp_next_drop (is_ip4)); \
2071 }
2072 #define tcp_inc_err_counter(cnts, err, val) \
2073 { \
2074  cnts[err] += val; \
2075 }
2076 #define tcp_store_err_counters(node_id, cnts) \
2077 { \
2078  int i; \
2079  for (i = 0; i < TCP_N_ERROR; i++) \
2080  if (cnts[i]) \
2081  tcp_inc_counter(node_id, i, cnts[i]); \
2082 }
2083 
2084 
2087  vlib_frame_t * frame, int is_ip4)
2088 {
2089  u32 thread_index = vm->thread_index, errors = 0;
2090  tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index);
2091  u32 n_left_from, *from, *first_buffer;
2092  u16 err_counters[TCP_N_ERROR] = { 0 };
2093 
2094  if (node->flags & VLIB_NODE_FLAG_TRACE)
2095  tcp_established_trace_frame (vm, node, frame, is_ip4);
2096 
2097  first_buffer = from = vlib_frame_vector_args (frame);
2098  n_left_from = frame->n_vectors;
2099 
2100  while (n_left_from > 0)
2101  {
2102  u32 bi0, error0 = TCP_ERROR_ACK_OK;
2103  vlib_buffer_t *b0;
2104  tcp_header_t *th0;
2105  tcp_connection_t *tc0;
2106 
2107  if (n_left_from > 1)
2108  {
2109  vlib_buffer_t *pb;
2110  pb = vlib_get_buffer (vm, from[1]);
2111  vlib_prefetch_buffer_header (pb, LOAD);
2112  CLIB_PREFETCH (pb->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
2113  }
2114 
2115  bi0 = from[0];
2116  from += 1;
2117  n_left_from -= 1;
2118 
2119  b0 = vlib_get_buffer (vm, bi0);
2120  tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
2121  thread_index);
2122 
2123  if (PREDICT_FALSE (tc0 == 0))
2124  {
2125  error0 = TCP_ERROR_INVALID_CONNECTION;
2126  goto done;
2127  }
2128 
2129  th0 = tcp_buffer_hdr (b0);
2130 
2131  /* TODO header prediction fast path */
2132 
2133  /* 1-4: check SEQ, RST, SYN */
2134  if (PREDICT_FALSE (tcp_segment_validate (wrk, tc0, b0, th0, &error0)))
2135  {
2136  TCP_EVT_DBG (TCP_EVT_SEG_INVALID, tc0, vnet_buffer (b0)->tcp);
2137  goto done;
2138  }
2139 
2140  /* 5: check the ACK field */
2141  if (PREDICT_FALSE (tcp_rcv_ack (wrk, tc0, b0, th0, &error0)))
2142  goto done;
2143 
2144  /* 6: check the URG bit TODO */
2145 
2146  /* 7: process the segment text */
2147  if (vnet_buffer (b0)->tcp.data_len)
2148  error0 = tcp_segment_rcv (wrk, tc0, b0);
2149 
2150  /* 8: check the FIN bit */
2151  if (PREDICT_FALSE (tcp_is_fin (th0)))
2152  tcp_rcv_fin (wrk, tc0, b0, &error0);
2153 
2154  done:
2155  tcp_inc_err_counter (err_counters, error0, 1);
2156  }
2157 
2159  thread_index);
2160  err_counters[TCP_ERROR_MSG_QUEUE_FULL] = errors;
2161  tcp_store_err_counters (established, err_counters);
2163  tcp_handle_disconnects (wrk);
2164  vlib_buffer_free (vm, first_buffer, frame->n_vectors);
2165 
2166  return frame->n_vectors;
2167 }
2168 
2170  vlib_node_runtime_t * node,
2171  vlib_frame_t * from_frame)
2172 {
2173  return tcp46_established_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2174 }
2175 
2177  vlib_node_runtime_t * node,
2178  vlib_frame_t * from_frame)
2179 {
2180  return tcp46_established_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2181 }
2182 
2183 /* *INDENT-OFF* */
2185 {
2186  .name = "tcp4-established",
2187  /* Takes a vector of packets. */
2188  .vector_size = sizeof (u32),
2189  .n_errors = TCP_N_ERROR,
2190  .error_strings = tcp_error_strings,
2191  .n_next_nodes = TCP_ESTABLISHED_N_NEXT,
2192  .next_nodes =
2193  {
2194 #define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n,
2196 #undef _
2197  },
2198  .format_trace = format_tcp_rx_trace_short,
2199 };
2200 /* *INDENT-ON* */
2201 
2202 /* *INDENT-OFF* */
2204 {
2205  .name = "tcp6-established",
2206  /* Takes a vector of packets. */
2207  .vector_size = sizeof (u32),
2208  .n_errors = TCP_N_ERROR,
2209  .error_strings = tcp_error_strings,
2210  .n_next_nodes = TCP_ESTABLISHED_N_NEXT,
2211  .next_nodes =
2212  {
2213 #define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n,
2215 #undef _
2216  },
2217  .format_trace = format_tcp_rx_trace_short,
2218 };
2219 /* *INDENT-ON* */
2220 
2221 
2222 static u8
2224 {
2225  transport_connection_t *tmp = 0;
2226  u64 handle;
2227 
2228  if (!tc)
2229  return 1;
2230 
2231  /* Proxy case */
2232  if (tc->c_lcl_port == 0 && tc->state == TCP_STATE_LISTEN)
2233  return 1;
2234 
2235  u8 is_valid = (tc->c_lcl_port == hdr->dst_port
2236  && (tc->state == TCP_STATE_LISTEN
2237  || tc->c_rmt_port == hdr->src_port));
2238 
2239  if (!is_valid)
2240  {
2241  handle = session_lookup_half_open_handle (&tc->connection);
2242  tmp = session_lookup_half_open_connection (handle & 0xFFFFFFFF,
2243  tc->c_proto, tc->c_is_ip4);
2244 
2245  if (tmp)
2246  {
2247  if (tmp->lcl_port == hdr->dst_port
2248  && tmp->rmt_port == hdr->src_port)
2249  {
2250  TCP_DBG ("half-open is valid!");
2251  }
2252  }
2253  }
2254  return is_valid;
2255 }
2256 
2257 /**
2258  * Lookup transport connection
2259  */
2260 static tcp_connection_t *
2261 tcp_lookup_connection (u32 fib_index, vlib_buffer_t * b, u8 thread_index,
2262  u8 is_ip4)
2263 {
2264  tcp_header_t *tcp;
2265  transport_connection_t *tconn;
2266  tcp_connection_t *tc;
2267  u8 is_filtered = 0;
2268  if (is_ip4)
2269  {
2270  ip4_header_t *ip4;
2271  ip4 = vlib_buffer_get_current (b);
2272  tcp = ip4_next_header (ip4);
2273  tconn = session_lookup_connection_wt4 (fib_index,
2274  &ip4->dst_address,
2275  &ip4->src_address,
2276  tcp->dst_port,
2277  tcp->src_port,
2279  thread_index, &is_filtered);
2280  tc = tcp_get_connection_from_transport (tconn);
2281  ASSERT (tcp_lookup_is_valid (tc, tcp));
2282  }
2283  else
2284  {
2285  ip6_header_t *ip6;
2286  ip6 = vlib_buffer_get_current (b);
2287  tcp = ip6_next_header (ip6);
2288  tconn = session_lookup_connection_wt6 (fib_index,
2289  &ip6->dst_address,
2290  &ip6->src_address,
2291  tcp->dst_port,
2292  tcp->src_port,
2294  thread_index, &is_filtered);
2295  tc = tcp_get_connection_from_transport (tconn);
2296  ASSERT (tcp_lookup_is_valid (tc, tcp));
2297  }
2298  return tc;
2299 }
2300 
2303  vlib_frame_t * from_frame, int is_ip4)
2304 {
2305  tcp_main_t *tm = vnet_get_tcp_main ();
2306  u32 n_left_from, *from, *first_buffer, errors = 0;
2307  u32 my_thread_index = vm->thread_index;
2308  tcp_worker_ctx_t *wrk = tcp_get_worker (my_thread_index);
2309 
2310  from = first_buffer = vlib_frame_vector_args (from_frame);
2311  n_left_from = from_frame->n_vectors;
2312 
2313  while (n_left_from > 0)
2314  {
2315  u32 bi0, ack0, seq0, error0 = TCP_ERROR_NONE;
2316  tcp_connection_t *tc0, *new_tc0;
2317  tcp_header_t *tcp0 = 0;
2318  tcp_rx_trace_t *t0;
2319  vlib_buffer_t *b0;
2320 
2321  bi0 = from[0];
2322  from += 1;
2323  n_left_from -= 1;
2324 
2325  b0 = vlib_get_buffer (vm, bi0);
2326  tc0 =
2327  tcp_half_open_connection_get (vnet_buffer (b0)->tcp.connection_index);
2328  if (PREDICT_FALSE (tc0 == 0))
2329  {
2330  error0 = TCP_ERROR_INVALID_CONNECTION;
2331  goto drop;
2332  }
2333 
2334  /* Half-open completed recently but the connection was't removed
2335  * yet by the owning thread */
2336  if (PREDICT_FALSE (tc0->flags & TCP_CONN_HALF_OPEN_DONE))
2337  {
2338  /* Make sure the connection actually exists */
2339  ASSERT (tcp_lookup_connection (tc0->c_fib_index, b0,
2340  my_thread_index, is_ip4));
2341  error0 = TCP_ERROR_SPURIOUS_SYN_ACK;
2342  goto drop;
2343  }
2344 
2345  ack0 = vnet_buffer (b0)->tcp.ack_number;
2346  seq0 = vnet_buffer (b0)->tcp.seq_number;
2347  tcp0 = tcp_buffer_hdr (b0);
2348 
2349  /* Crude check to see if the connection handle does not match
2350  * the packet. Probably connection just switched to established */
2351  if (PREDICT_FALSE (tcp0->dst_port != tc0->c_lcl_port
2352  || tcp0->src_port != tc0->c_rmt_port))
2353  {
2354  error0 = TCP_ERROR_INVALID_CONNECTION;
2355  goto drop;
2356  }
2357 
2358  if (PREDICT_FALSE (!tcp_ack (tcp0) && !tcp_rst (tcp0)
2359  && !tcp_syn (tcp0)))
2360  {
2361  error0 = TCP_ERROR_SEGMENT_INVALID;
2362  goto drop;
2363  }
2364 
2365  /* SYNs consume sequence numbers */
2366  vnet_buffer (b0)->tcp.seq_end += tcp_is_syn (tcp0);
2367 
2368  /*
2369  * 1. check the ACK bit
2370  */
2371 
2372  /*
2373  * If the ACK bit is set
2374  * If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send a reset (unless
2375  * the RST bit is set, if so drop the segment and return)
2376  * <SEQ=SEG.ACK><CTL=RST>
2377  * and discard the segment. Return.
2378  * If SND.UNA =< SEG.ACK =< SND.NXT then the ACK is acceptable.
2379  */
2380  if (tcp_ack (tcp0))
2381  {
2382  if (seq_leq (ack0, tc0->iss) || seq_gt (ack0, tc0->snd_nxt))
2383  {
2384  if (!tcp_rst (tcp0))
2385  tcp_send_reset_w_pkt (tc0, b0, my_thread_index, is_ip4);
2386  error0 = TCP_ERROR_RCV_WND;
2387  goto drop;
2388  }
2389 
2390  /* Make sure ACK is valid */
2391  if (seq_gt (tc0->snd_una, ack0))
2392  {
2393  error0 = TCP_ERROR_ACK_INVALID;
2394  goto drop;
2395  }
2396  }
2397 
2398  /*
2399  * 2. check the RST bit
2400  */
2401 
2402  if (tcp_rst (tcp0))
2403  {
2404  /* If ACK is acceptable, signal client that peer is not
2405  * willing to accept connection and drop connection*/
2406  if (tcp_ack (tcp0))
2407  tcp_connection_reset (tc0);
2408  error0 = TCP_ERROR_RST_RCVD;
2409  goto drop;
2410  }
2411 
2412  /*
2413  * 3. check the security and precedence (skipped)
2414  */
2415 
2416  /*
2417  * 4. check the SYN bit
2418  */
2419 
2420  /* No SYN flag. Drop. */
2421  if (!tcp_syn (tcp0))
2422  {
2423  error0 = TCP_ERROR_SEGMENT_INVALID;
2424  goto drop;
2425  }
2426 
2427  /* Parse options */
2428  if (tcp_options_parse (tcp0, &tc0->rcv_opts, 1))
2429  {
2430  error0 = TCP_ERROR_OPTIONS;
2431  goto drop;
2432  }
2433 
2434  /* Valid SYN or SYN-ACK. Move connection from half-open pool to
2435  * current thread pool. */
2436  pool_get (tm->connections[my_thread_index], new_tc0);
2437  clib_memcpy_fast (new_tc0, tc0, sizeof (*new_tc0));
2438  new_tc0->c_c_index = new_tc0 - tm->connections[my_thread_index];
2439  new_tc0->c_thread_index = my_thread_index;
2440  new_tc0->rcv_nxt = vnet_buffer (b0)->tcp.seq_end;
2441  new_tc0->irs = seq0;
2442  new_tc0->timers[TCP_TIMER_RETRANSMIT_SYN] = TCP_TIMER_HANDLE_INVALID;
2443  new_tc0->sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_RX];
2444 
2445  /* If this is not the owning thread, wait for syn retransmit to
2446  * expire and cleanup then */
2448  tc0->flags |= TCP_CONN_HALF_OPEN_DONE;
2449 
2450  if (tcp_opts_tstamp (&new_tc0->rcv_opts))
2451  {
2452  new_tc0->tsval_recent = new_tc0->rcv_opts.tsval;
2453  new_tc0->tsval_recent_age = tcp_time_now ();
2454  }
2455 
2456  if (tcp_opts_wscale (&new_tc0->rcv_opts))
2457  new_tc0->snd_wscale = new_tc0->rcv_opts.wscale;
2458  else
2459  new_tc0->rcv_wscale = 0;
2460 
2461  new_tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window)
2462  << new_tc0->snd_wscale;
2463  new_tc0->snd_wl1 = seq0;
2464  new_tc0->snd_wl2 = ack0;
2465 
2466  tcp_connection_init_vars (new_tc0);
2467 
2468  /* SYN-ACK: See if we can switch to ESTABLISHED state */
2469  if (PREDICT_TRUE (tcp_ack (tcp0)))
2470  {
2471  /* Our SYN is ACKed: we have iss < ack = snd_una */
2472 
2473  /* TODO Dequeue acknowledged segments if we support Fast Open */
2474  new_tc0->snd_una = ack0;
2475  new_tc0->state = TCP_STATE_ESTABLISHED;
2476 
2477  /* Make sure las is initialized for the wnd computation */
2478  new_tc0->rcv_las = new_tc0->rcv_nxt;
2479 
2480  /* Notify app that we have connection. If session layer can't
2481  * allocate session send reset */
2482  if (session_stream_connect_notify (&new_tc0->connection, 0))
2483  {
2484  tcp_send_reset_w_pkt (new_tc0, b0, my_thread_index, is_ip4);
2485  tcp_connection_cleanup (new_tc0);
2486  error0 = TCP_ERROR_CREATE_SESSION_FAIL;
2487  goto drop;
2488  }
2489 
2490  new_tc0->tx_fifo_size =
2491  transport_tx_fifo_size (&new_tc0->connection);
2492  /* Update rtt with the syn-ack sample */
2493  tcp_estimate_initial_rtt (new_tc0);
2494  TCP_EVT_DBG (TCP_EVT_SYNACK_RCVD, new_tc0);
2495  error0 = TCP_ERROR_SYN_ACKS_RCVD;
2496  }
2497  /* SYN: Simultaneous open. Change state to SYN-RCVD and send SYN-ACK */
2498  else
2499  {
2500  new_tc0->state = TCP_STATE_SYN_RCVD;
2501 
2502  /* Notify app that we have connection */
2503  if (session_stream_connect_notify (&new_tc0->connection, 0))
2504  {
2505  tcp_connection_cleanup (new_tc0);
2506  tcp_send_reset_w_pkt (tc0, b0, my_thread_index, is_ip4);
2507  TCP_EVT_DBG (TCP_EVT_RST_SENT, tc0);
2508  error0 = TCP_ERROR_CREATE_SESSION_FAIL;
2509  goto drop;
2510  }
2511 
2512  new_tc0->tx_fifo_size =
2513  transport_tx_fifo_size (&new_tc0->connection);
2514  new_tc0->rtt_ts = 0;
2515  tcp_init_snd_vars (new_tc0);
2516  tcp_send_synack (new_tc0);
2517  error0 = TCP_ERROR_SYNS_RCVD;
2518  goto drop;
2519  }
2520 
2521  /* Read data, if any */
2522  if (PREDICT_FALSE (vnet_buffer (b0)->tcp.data_len))
2523  {
2524  clib_warning ("rcvd data in syn-sent");
2525  error0 = tcp_segment_rcv (wrk, new_tc0, b0);
2526  if (error0 == TCP_ERROR_ACK_OK)
2527  error0 = TCP_ERROR_SYN_ACKS_RCVD;
2528  }
2529  else
2530  {
2531  tcp_program_ack (new_tc0);
2532  }
2533 
2534  drop:
2535 
2536  tcp_inc_counter (syn_sent, error0, 1);
2537  if (PREDICT_FALSE ((b0->flags & VLIB_BUFFER_IS_TRACED) && tcp0 != 0))
2538  {
2539  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
2540  clib_memcpy_fast (&t0->tcp_header, tcp0, sizeof (t0->tcp_header));
2541  clib_memcpy_fast (&t0->tcp_connection, tc0,
2542  sizeof (t0->tcp_connection));
2543  }
2544  }
2545 
2547  my_thread_index);
2548  tcp_inc_counter (syn_sent, TCP_ERROR_MSG_QUEUE_FULL, errors);
2549  vlib_buffer_free (vm, first_buffer, from_frame->n_vectors);
2550 
2551  return from_frame->n_vectors;
2552 }
2553 
2555  vlib_node_runtime_t * node,
2556  vlib_frame_t * from_frame)
2557 {
2558  return tcp46_syn_sent_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2559 }
2560 
2562  vlib_node_runtime_t * node,
2563  vlib_frame_t * from_frame)
2564 {
2565  return tcp46_syn_sent_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2566 }
2567 
2568 /* *INDENT-OFF* */
2570 {
2571  .name = "tcp4-syn-sent",
2572  /* Takes a vector of packets. */
2573  .vector_size = sizeof (u32),
2574  .n_errors = TCP_N_ERROR,
2575  .error_strings = tcp_error_strings,
2576  .n_next_nodes = TCP_SYN_SENT_N_NEXT,
2577  .next_nodes =
2578  {
2579 #define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n,
2581 #undef _
2582  },
2583  .format_trace = format_tcp_rx_trace_short,
2584 };
2585 /* *INDENT-ON* */
2586 
2587 /* *INDENT-OFF* */
2589 {
2590  .name = "tcp6-syn-sent",
2591  /* Takes a vector of packets. */
2592  .vector_size = sizeof (u32),
2593  .n_errors = TCP_N_ERROR,
2594  .error_strings = tcp_error_strings,
2595  .n_next_nodes = TCP_SYN_SENT_N_NEXT,
2596  .next_nodes =
2597  {
2598 #define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n,
2600 #undef _
2601  },
2602  .format_trace = format_tcp_rx_trace_short,
2603 };
2604 /* *INDENT-ON* */
2605 
2606 /**
2607  * Handles reception for all states except LISTEN, SYN-SENT and ESTABLISHED
2608  * as per RFC793 p. 64
2609  */
2612  vlib_frame_t * from_frame, int is_ip4)
2613 {
2614  u32 thread_index = vm->thread_index, errors = 0, *first_buffer;
2615  tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index);
2616  u32 n_left_from, *from, max_dequeue;
2617 
2618  from = first_buffer = vlib_frame_vector_args (from_frame);
2619  n_left_from = from_frame->n_vectors;
2620 
2621  while (n_left_from > 0)
2622  {
2623  u32 bi0, error0 = TCP_ERROR_NONE;
2624  tcp_header_t *tcp0 = 0;
2625  tcp_connection_t *tc0;
2626  vlib_buffer_t *b0;
2627  u8 is_fin0;
2628 
2629  bi0 = from[0];
2630  from += 1;
2631  n_left_from -= 1;
2632 
2633  b0 = vlib_get_buffer (vm, bi0);
2634  tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
2635  thread_index);
2636  if (PREDICT_FALSE (tc0 == 0))
2637  {
2638  error0 = TCP_ERROR_INVALID_CONNECTION;
2639  goto drop;
2640  }
2641 
2642  tcp0 = tcp_buffer_hdr (b0);
2643  is_fin0 = tcp_is_fin (tcp0);
2644 
2645  if (CLIB_DEBUG)
2646  {
2647  tcp_connection_t *tmp;
2648  tmp = tcp_lookup_connection (tc0->c_fib_index, b0, thread_index,
2649  is_ip4);
2650  if (tmp->state != tc0->state)
2651  {
2652  if (tc0->state != TCP_STATE_CLOSED)
2653  clib_warning ("state changed");
2654  goto drop;
2655  }
2656  }
2657 
2658  /*
2659  * Special treatment for CLOSED
2660  */
2661  if (PREDICT_FALSE (tc0->state == TCP_STATE_CLOSED))
2662  {
2663  error0 = TCP_ERROR_CONNECTION_CLOSED;
2664  goto drop;
2665  }
2666 
2667  /*
2668  * For all other states (except LISTEN)
2669  */
2670 
2671  /* 1-4: check SEQ, RST, SYN */
2672  if (PREDICT_FALSE (tcp_segment_validate (wrk, tc0, b0, tcp0, &error0)))
2673  goto drop;
2674 
2675  /* 5: check the ACK field */
2676  switch (tc0->state)
2677  {
2678  case TCP_STATE_SYN_RCVD:
2679 
2680  /* Make sure the segment is exactly right */
2681  if (tc0->rcv_nxt != vnet_buffer (b0)->tcp.seq_number || is_fin0)
2682  {
2683  tcp_connection_reset (tc0);
2684  error0 = TCP_ERROR_SEGMENT_INVALID;
2685  goto drop;
2686  }
2687 
2688  /*
2689  * If the segment acknowledgment is not acceptable, form a
2690  * reset segment,
2691  * <SEQ=SEG.ACK><CTL=RST>
2692  * and send it.
2693  */
2694  if (tcp_rcv_ack_no_cc (tc0, b0, &error0))
2695  {
2696  tcp_connection_reset (tc0);
2697  goto drop;
2698  }
2699 
2700  /* Update rtt and rto */
2702 
2703  /* Switch state to ESTABLISHED */
2704  tc0->state = TCP_STATE_ESTABLISHED;
2705  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2706 
2707  /* Initialize session variables */
2708  tc0->snd_una = vnet_buffer (b0)->tcp.ack_number;
2709  tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window)
2710  << tc0->rcv_opts.wscale;
2711  tc0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number;
2712  tc0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number;
2713 
2714  /* Reset SYN-ACK retransmit and SYN_RCV establish timers */
2716  if (session_stream_accept_notify (&tc0->connection))
2717  {
2718  error0 = TCP_ERROR_MSG_QUEUE_FULL;
2719  tcp_connection_reset (tc0);
2720  goto drop;
2721  }
2722  error0 = TCP_ERROR_ACK_OK;
2723  break;
2724  case TCP_STATE_ESTABLISHED:
2725  /* We can get packets in established state here because they
2726  * were enqueued before state change */
2727  if (tcp_rcv_ack (wrk, tc0, b0, tcp0, &error0))
2728  goto drop;
2729 
2730  break;
2731  case TCP_STATE_FIN_WAIT_1:
2732  /* In addition to the processing for the ESTABLISHED state, if
2733  * our FIN is now acknowledged then enter FIN-WAIT-2 and
2734  * continue processing in that state. */
2735  if (tcp_rcv_ack (wrk, tc0, b0, tcp0, &error0))
2736  goto drop;
2737 
2738  /* Still have to send the FIN */
2739  if (tc0->flags & TCP_CONN_FINPNDG)
2740  {
2741  /* TX fifo finally drained */
2742  max_dequeue = transport_max_tx_dequeue (&tc0->connection);
2743  if (max_dequeue <= tc0->burst_acked)
2744  tcp_send_fin (tc0);
2745  /* If a fin was received and data was acked extend wait */
2746  else if ((tc0->flags & TCP_CONN_FINRCVD) && tc0->bytes_acked)
2747  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE,
2748  tcp_cfg.closewait_time);
2749  }
2750  /* If FIN is ACKed */
2751  else if (tc0->snd_una == tc0->snd_nxt)
2752  {
2753  /* Stop all retransmit timers because we have nothing more
2754  * to send. */
2756 
2757  /* We already have a FIN but didn't transition to CLOSING
2758  * because of outstanding tx data. Close the connection. */
2759  if (tc0->flags & TCP_CONN_FINRCVD)
2760  {
2761  tcp_connection_set_state (tc0, TCP_STATE_CLOSED);
2762  tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE,
2763  tcp_cfg.cleanup_time);
2764  session_transport_closed_notify (&tc0->connection);
2765  goto drop;
2766  }
2767 
2768  tcp_connection_set_state (tc0, TCP_STATE_FIN_WAIT_2);
2769  /* Enable waitclose because we're willing to wait for peer's
2770  * FIN but not indefinitely. */
2771  tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, tcp_cfg.finwait2_time);
2772 
2773  /* Don't try to deq the FIN acked */
2774  if (tc0->burst_acked > 1)
2775  session_tx_fifo_dequeue_drop (&tc0->connection,
2776  tc0->burst_acked - 1);
2777  tc0->burst_acked = 0;
2778  }
2779  break;
2780  case TCP_STATE_FIN_WAIT_2:
2781  /* In addition to the processing for the ESTABLISHED state, if
2782  * the retransmission queue is empty, the user's CLOSE can be
2783  * acknowledged ("ok") but do not delete the TCB. */
2784  if (tcp_rcv_ack_no_cc (tc0, b0, &error0))
2785  goto drop;
2786  tc0->burst_acked = 0;
2787  break;
2788  case TCP_STATE_CLOSE_WAIT:
2789  /* Do the same processing as for the ESTABLISHED state. */
2790  if (tcp_rcv_ack (wrk, tc0, b0, tcp0, &error0))
2791  goto drop;
2792 
2793  if (!(tc0->flags & TCP_CONN_FINPNDG))
2794  break;
2795 
2796  /* Still have outstanding tx data */
2797  max_dequeue = transport_max_tx_dequeue (&tc0->connection);
2798  if (max_dequeue > tc0->burst_acked)
2799  break;
2800 
2801  tcp_send_fin (tc0);
2803  tcp_connection_set_state (tc0, TCP_STATE_LAST_ACK);
2804  tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, tcp_cfg.lastack_time);
2805  break;
2806  case TCP_STATE_CLOSING:
2807  /* In addition to the processing for the ESTABLISHED state, if
2808  * the ACK acknowledges our FIN then enter the TIME-WAIT state,
2809  * otherwise ignore the segment. */
2810  if (tcp_rcv_ack_no_cc (tc0, b0, &error0))
2811  goto drop;
2812 
2813  if (tc0->snd_una != tc0->snd_nxt)
2814  goto drop;
2815 
2817  tcp_connection_set_state (tc0, TCP_STATE_TIME_WAIT);
2818  tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, tcp_cfg.timewait_time);
2819  session_transport_closed_notify (&tc0->connection);
2820  goto drop;
2821 
2822  break;
2823  case TCP_STATE_LAST_ACK:
2824  /* The only thing that [should] arrive in this state is an
2825  * acknowledgment of our FIN. If our FIN is now acknowledged,
2826  * delete the TCB, enter the CLOSED state, and return. */
2827 
2828  if (tcp_rcv_ack_no_cc (tc0, b0, &error0))
2829  goto drop;
2830 
2831  /* Apparently our ACK for the peer's FIN was lost */
2832  if (is_fin0 && tc0->snd_una != tc0->snd_nxt)
2833  {
2834  tcp_send_fin (tc0);
2835  goto drop;
2836  }
2837 
2838  tcp_connection_set_state (tc0, TCP_STATE_CLOSED);
2839  session_transport_closed_notify (&tc0->connection);
2840 
2841  /* Don't free the connection from the data path since
2842  * we can't ensure that we have no packets already enqueued
2843  * to output. Rely instead on the waitclose timer */
2845  tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
2846 
2847  goto drop;
2848 
2849  break;
2850  case TCP_STATE_TIME_WAIT:
2851  /* The only thing that can arrive in this state is a
2852  * retransmission of the remote FIN. Acknowledge it, and restart
2853  * the 2 MSL timeout. */
2854 
2855  if (tcp_rcv_ack_no_cc (tc0, b0, &error0))
2856  goto drop;
2857 
2858  if (!is_fin0)
2859  goto drop;
2860 
2861  tcp_program_ack (tc0);
2862  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, tcp_cfg.timewait_time);
2863  goto drop;
2864 
2865  break;
2866  default:
2867  ASSERT (0);
2868  }
2869 
2870  /* 6: check the URG bit TODO */
2871 
2872  /* 7: process the segment text */
2873  switch (tc0->state)
2874  {
2875  case TCP_STATE_ESTABLISHED:
2876  case TCP_STATE_FIN_WAIT_1:
2877  case TCP_STATE_FIN_WAIT_2:
2878  if (vnet_buffer (b0)->tcp.data_len)
2879  error0 = tcp_segment_rcv (wrk, tc0, b0);
2880  break;
2881  case TCP_STATE_CLOSE_WAIT:
2882  case TCP_STATE_CLOSING:
2883  case TCP_STATE_LAST_ACK:
2884  case TCP_STATE_TIME_WAIT:
2885  /* This should not occur, since a FIN has been received from the
2886  * remote side. Ignore the segment text. */
2887  break;
2888  }
2889 
2890  /* 8: check the FIN bit */
2891  if (!is_fin0)
2892  goto drop;
2893 
2894  TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc0);
2895 
2896  switch (tc0->state)
2897  {
2898  case TCP_STATE_ESTABLISHED:
2899  /* Account for the FIN and send ack */
2900  tc0->rcv_nxt += 1;
2901  tcp_program_ack (tc0);
2902  tcp_connection_set_state (tc0, TCP_STATE_CLOSE_WAIT);
2903  tcp_program_disconnect (wrk, tc0);
2904  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
2905  break;
2906  case TCP_STATE_SYN_RCVD:
2907  /* Send FIN-ACK, enter LAST-ACK and because the app was not
2908  * notified yet, set a cleanup timer instead of relying on
2909  * disconnect notify and the implicit close call. */
2911  tc0->rcv_nxt += 1;
2912  tcp_send_fin (tc0);
2913  tcp_connection_set_state (tc0, TCP_STATE_LAST_ACK);
2914  tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, tcp_cfg.lastack_time);
2915  break;
2916  case TCP_STATE_CLOSE_WAIT:
2917  case TCP_STATE_CLOSING:
2918  case TCP_STATE_LAST_ACK:
2919  /* move along .. */
2920  break;
2921  case TCP_STATE_FIN_WAIT_1:
2922  tc0->rcv_nxt += 1;
2923 
2924  if (tc0->flags & TCP_CONN_FINPNDG)
2925  {
2926  /* If data is outstanding, stay in FIN_WAIT_1 and try to finish
2927  * sending it. Since we already received a fin, do not wait
2928  * for too long. */
2929  tc0->flags |= TCP_CONN_FINRCVD;
2930  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE,
2931  tcp_cfg.closewait_time);
2932  }
2933  else
2934  {
2935  tcp_connection_set_state (tc0, TCP_STATE_CLOSING);
2936  tcp_program_ack (tc0);
2937  /* Wait for ACK for our FIN but not forever */
2938  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE,
2939  tcp_cfg.closing_time);
2940  }
2941  break;
2942  case TCP_STATE_FIN_WAIT_2:
2943  /* Got FIN, send ACK! Be more aggressive with resource cleanup */
2944  tc0->rcv_nxt += 1;
2945  tcp_connection_set_state (tc0, TCP_STATE_TIME_WAIT);
2947  tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, tcp_cfg.timewait_time);
2948  tcp_program_ack (tc0);
2949  session_transport_closed_notify (&tc0->connection);
2950  break;
2951  case TCP_STATE_TIME_WAIT:
2952  /* Remain in the TIME-WAIT state. Restart the time-wait
2953  * timeout.
2954  */
2955  tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, tcp_cfg.timewait_time);
2956  break;
2957  }
2958  error0 = TCP_ERROR_FIN_RCVD;
2959 
2960  drop:
2961 
2962  tcp_inc_counter (rcv_process, error0, 1);
2963  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
2964  {
2965  tcp_rx_trace_t *t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
2966  tcp_set_rx_trace_data (t0, tc0, tcp0, b0, is_ip4);
2967  }
2968  }
2969 
2971  thread_index);
2972  tcp_inc_counter (rcv_process, TCP_ERROR_MSG_QUEUE_FULL, errors);
2974  tcp_handle_disconnects (wrk);
2975  vlib_buffer_free (vm, first_buffer, from_frame->n_vectors);
2976 
2977  return from_frame->n_vectors;
2978 }
2979 
2981  vlib_node_runtime_t * node,
2982  vlib_frame_t * from_frame)
2983 {
2984  return tcp46_rcv_process_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2985 }
2986 
2988  vlib_node_runtime_t * node,
2989  vlib_frame_t * from_frame)
2990 {
2991  return tcp46_rcv_process_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2992 }
2993 
2994 /* *INDENT-OFF* */
2996 {
2997  .name = "tcp4-rcv-process",
2998  /* Takes a vector of packets. */
2999  .vector_size = sizeof (u32),
3000  .n_errors = TCP_N_ERROR,
3001  .error_strings = tcp_error_strings,
3002  .n_next_nodes = TCP_RCV_PROCESS_N_NEXT,
3003  .next_nodes =
3004  {
3005 #define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n,
3007 #undef _
3008  },
3009  .format_trace = format_tcp_rx_trace_short,
3010 };
3011 /* *INDENT-ON* */
3012 
3013 /* *INDENT-OFF* */
3015 {
3016  .name = "tcp6-rcv-process",
3017  /* Takes a vector of packets. */
3018  .vector_size = sizeof (u32),
3019  .n_errors = TCP_N_ERROR,
3020  .error_strings = tcp_error_strings,
3021  .n_next_nodes = TCP_RCV_PROCESS_N_NEXT,
3022  .next_nodes =
3023  {
3024 #define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n,
3026 #undef _
3027  },
3028  .format_trace = format_tcp_rx_trace_short,
3029 };
3030 /* *INDENT-ON* */
3031 
3032 /**
3033  * LISTEN state processing as per RFC 793 p. 65
3034  */
3037  vlib_frame_t * from_frame, int is_ip4)
3038 {
3039  u32 n_left_from, *from, n_syns = 0, *first_buffer;
3040  u32 my_thread_index = vm->thread_index;
3041 
3042  from = first_buffer = vlib_frame_vector_args (from_frame);
3043  n_left_from = from_frame->n_vectors;
3044 
3045  while (n_left_from > 0)
3046  {
3047  u32 bi0;
3048  vlib_buffer_t *b0;
3049  tcp_rx_trace_t *t0;
3050  tcp_header_t *th0 = 0;
3051  tcp_connection_t *lc0;
3052  ip4_header_t *ip40;
3053  ip6_header_t *ip60;
3054  tcp_connection_t *child0;
3055  u32 error0 = TCP_ERROR_NONE;
3056 
3057  bi0 = from[0];
3058  from += 1;
3059  n_left_from -= 1;
3060 
3061  b0 = vlib_get_buffer (vm, bi0);
3062  lc0 = tcp_listener_get (vnet_buffer (b0)->tcp.connection_index);
3063 
3064  if (is_ip4)
3065  {
3066  ip40 = vlib_buffer_get_current (b0);
3067  th0 = ip4_next_header (ip40);
3068  }
3069  else
3070  {
3071  ip60 = vlib_buffer_get_current (b0);
3072  th0 = ip6_next_header (ip60);
3073  }
3074 
3075  /* Create child session. For syn-flood protection use filter */
3076 
3077  /* 1. first check for an RST: handled in dispatch */
3078  /* if (tcp_rst (th0))
3079  goto drop;
3080  */
3081 
3082  /* 2. second check for an ACK: handled in dispatch */
3083  /* if (tcp_ack (th0))
3084  {
3085  tcp_send_reset (b0, is_ip4);
3086  goto drop;
3087  }
3088  */
3089 
3090  /* 3. check for a SYN (did that already) */
3091 
3092  /* Make sure connection wasn't just created */
3093  child0 = tcp_lookup_connection (lc0->c_fib_index, b0, my_thread_index,
3094  is_ip4);
3095  if (PREDICT_FALSE (child0->state != TCP_STATE_LISTEN))
3096  {
3097  error0 = TCP_ERROR_CREATE_EXISTS;
3098  goto drop;
3099  }
3100 
3101  /* Create child session and send SYN-ACK */
3102  child0 = tcp_connection_alloc (my_thread_index);
3103  child0->c_lcl_port = th0->dst_port;
3104  child0->c_rmt_port = th0->src_port;
3105  child0->c_is_ip4 = is_ip4;
3106  child0->state = TCP_STATE_SYN_RCVD;
3107  child0->c_fib_index = lc0->c_fib_index;
3108 
3109  if (is_ip4)
3110  {
3111  child0->c_lcl_ip4.as_u32 = ip40->dst_address.as_u32;
3112  child0->c_rmt_ip4.as_u32 = ip40->src_address.as_u32;
3113  }
3114  else
3115  {
3116  clib_memcpy_fast (&child0->c_lcl_ip6, &ip60->dst_address,
3117  sizeof (ip6_address_t));
3118  clib_memcpy_fast (&child0->c_rmt_ip6, &ip60->src_address,
3119  sizeof (ip6_address_t));
3120  }
3121 
3122  if (tcp_options_parse (th0, &child0->rcv_opts, 1))
3123  {
3124  error0 = TCP_ERROR_OPTIONS;
3125  tcp_connection_free (child0);
3126  goto drop;
3127  }
3128 
3129  child0->irs = vnet_buffer (b0)->tcp.seq_number;
3130  child0->rcv_nxt = vnet_buffer (b0)->tcp.seq_number + 1;
3131  child0->rcv_las = child0->rcv_nxt;
3132  child0->sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_RX];
3133 
3134  /* RFC1323: TSval timestamps sent on {SYN} and {SYN,ACK}
3135  * segments are used to initialize PAWS. */
3136  if (tcp_opts_tstamp (&child0->rcv_opts))
3137  {
3138  child0->tsval_recent = child0->rcv_opts.tsval;
3139  child0->tsval_recent_age = tcp_time_now ();
3140  }
3141 
3142  if (tcp_opts_wscale (&child0->rcv_opts))
3143  child0->snd_wscale = child0->rcv_opts.wscale;
3144 
3145  child0->snd_wnd = clib_net_to_host_u16 (th0->window)
3146  << child0->snd_wscale;
3147  child0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number;
3148  child0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number;
3149 
3150  tcp_connection_init_vars (child0);
3151  child0->rto = TCP_RTO_MIN;
3152 
3153  if (session_stream_accept (&child0->connection, lc0->c_s_index,
3154  lc0->c_thread_index, 0 /* notify */ ))
3155  {
3156  tcp_connection_cleanup (child0);
3157  error0 = TCP_ERROR_CREATE_SESSION_FAIL;
3158  goto drop;
3159  }
3160 
3161  TCP_EVT_DBG (TCP_EVT_SYN_RCVD, child0, 1);
3162  child0->tx_fifo_size = transport_tx_fifo_size (&child0->connection);
3163  tcp_send_synack (child0);
3164 
3165  drop:
3166 
3167  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
3168  {
3169  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
3170  clib_memcpy_fast (&t0->tcp_header, th0, sizeof (t0->tcp_header));
3171  clib_memcpy_fast (&t0->tcp_connection, lc0,
3172  sizeof (t0->tcp_connection));
3173  }
3174 
3175  n_syns += (error0 == TCP_ERROR_NONE);
3176  }
3177 
3178  tcp_inc_counter (listen, TCP_ERROR_SYNS_RCVD, n_syns);
3179  vlib_buffer_free (vm, first_buffer, from_frame->n_vectors);
3180 
3181  return from_frame->n_vectors;
3182 }
3183 
3185  vlib_frame_t * from_frame)
3186 {
3187  return tcp46_listen_inline (vm, node, from_frame, 1 /* is_ip4 */ );
3188 }
3189 
3191  vlib_frame_t * from_frame)
3192 {
3193  return tcp46_listen_inline (vm, node, from_frame, 0 /* is_ip4 */ );
3194 }
3195 
3196 /* *INDENT-OFF* */
3198 {
3199  .name = "tcp4-listen",
3200  /* Takes a vector of packets. */
3201  .vector_size = sizeof (u32),
3202  .n_errors = TCP_N_ERROR,
3203  .error_strings = tcp_error_strings,
3204  .n_next_nodes = TCP_LISTEN_N_NEXT,
3205  .next_nodes =
3206  {
3207 #define _(s,n) [TCP_LISTEN_NEXT_##s] = n,
3209 #undef _
3210  },
3211  .format_trace = format_tcp_rx_trace_short,
3212 };
3213 /* *INDENT-ON* */
3214 
3215 /* *INDENT-OFF* */
3217 {
3218  .name = "tcp6-listen",
3219  /* Takes a vector of packets. */
3220  .vector_size = sizeof (u32),
3221  .n_errors = TCP_N_ERROR,
3222  .error_strings = tcp_error_strings,
3223  .n_next_nodes = TCP_LISTEN_N_NEXT,
3224  .next_nodes =
3225  {
3226 #define _(s,n) [TCP_LISTEN_NEXT_##s] = n,
3228 #undef _
3229  },
3230  .format_trace = format_tcp_rx_trace_short,
3231 };
3232 /* *INDENT-ON* */
3233 
3234 typedef enum _tcp_input_next
3235 {
3245 
3246 #define foreach_tcp4_input_next \
3247  _ (DROP, "ip4-drop") \
3248  _ (LISTEN, "tcp4-listen") \
3249  _ (RCV_PROCESS, "tcp4-rcv-process") \
3250  _ (SYN_SENT, "tcp4-syn-sent") \
3251  _ (ESTABLISHED, "tcp4-established") \
3252  _ (RESET, "tcp4-reset") \
3253  _ (PUNT, "ip4-punt")
3254 
3255 #define foreach_tcp6_input_next \
3256  _ (DROP, "ip6-drop") \
3257  _ (LISTEN, "tcp6-listen") \
3258  _ (RCV_PROCESS, "tcp6-rcv-process") \
3259  _ (SYN_SENT, "tcp6-syn-sent") \
3260  _ (ESTABLISHED, "tcp6-established") \
3261  _ (RESET, "tcp6-reset") \
3262  _ (PUNT, "ip6-punt")
3263 
3264 #define filter_flags (TCP_FLAG_SYN|TCP_FLAG_ACK|TCP_FLAG_RST|TCP_FLAG_FIN)
3265 
3266 static void
3268  vlib_buffer_t ** bs, u32 n_bufs, u8 is_ip4)
3269 {
3270  tcp_connection_t *tc;
3271  tcp_header_t *tcp;
3272  tcp_rx_trace_t *t;
3273  int i;
3274 
3275  for (i = 0; i < n_bufs; i++)
3276  {
3277  if (bs[i]->flags & VLIB_BUFFER_IS_TRACED)
3278  {
3279  t = vlib_add_trace (vm, node, bs[i], sizeof (*t));
3280  tc = tcp_connection_get (vnet_buffer (bs[i])->tcp.connection_index,
3281  vm->thread_index);
3282  tcp = vlib_buffer_get_current (bs[i]);
3283  tcp_set_rx_trace_data (t, tc, tcp, bs[i], is_ip4);
3284  }
3285  }
3286 }
3287 
3288 static void
3289 tcp_input_set_error_next (tcp_main_t * tm, u16 * next, u32 * error, u8 is_ip4)
3290 {
3291  if (*error == TCP_ERROR_FILTERED || *error == TCP_ERROR_WRONG_THREAD)
3292  {
3293  *next = TCP_INPUT_NEXT_DROP;
3294  }
3295  else if ((is_ip4 && tm->punt_unknown4) || (!is_ip4 && tm->punt_unknown6))
3296  {
3297  *next = TCP_INPUT_NEXT_PUNT;
3298  *error = TCP_ERROR_PUNT;
3299  }
3300  else
3301  {
3302  *next = TCP_INPUT_NEXT_RESET;
3303  *error = TCP_ERROR_NO_LISTENER;
3304  }
3305 }
3306 
3308 tcp_input_lookup_buffer (vlib_buffer_t * b, u8 thread_index, u32 * error,
3309  u8 is_ip4, u8 is_nolookup)
3310 {
3311  u32 fib_index = vnet_buffer (b)->ip.fib_index;
3312  int n_advance_bytes, n_data_bytes;
3314  tcp_header_t *tcp;
3315  u8 result = 0;
3316 
3317  if (is_ip4)
3318  {
3320  int ip_hdr_bytes = ip4_header_bytes (ip4);
3321  if (PREDICT_FALSE (b->current_length < ip_hdr_bytes + sizeof (*tcp)))
3322  {
3323  *error = TCP_ERROR_LENGTH;
3324  return 0;
3325  }
3326  tcp = ip4_next_header (ip4);
3327  vnet_buffer (b)->tcp.hdr_offset = (u8 *) tcp - (u8 *) ip4;
3328  n_advance_bytes = (ip_hdr_bytes + tcp_header_bytes (tcp));
3329  n_data_bytes = clib_net_to_host_u16 (ip4->length) - n_advance_bytes;
3330 
3331  /* Length check. Checksum computed by ipx_local no need to compute again */
3332  if (PREDICT_FALSE (n_data_bytes < 0))
3333  {
3334  *error = TCP_ERROR_LENGTH;
3335  return 0;
3336  }
3337 
3338  if (!is_nolookup)
3339  tc = session_lookup_connection_wt4 (fib_index, &ip4->dst_address,
3340  &ip4->src_address, tcp->dst_port,
3341  tcp->src_port,
3342  TRANSPORT_PROTO_TCP, thread_index,
3343  &result);
3344  }
3345  else
3346  {
3348  if (PREDICT_FALSE (b->current_length < sizeof (*ip6) + sizeof (*tcp)))
3349  {
3350  *error = TCP_ERROR_LENGTH;
3351  return 0;
3352  }
3353  tcp = ip6_next_header (ip6);
3354  vnet_buffer (b)->tcp.hdr_offset = (u8 *) tcp - (u8 *) ip6;
3355  n_advance_bytes = tcp_header_bytes (tcp);
3356  n_data_bytes = clib_net_to_host_u16 (ip6->payload_length)
3357  - n_advance_bytes;
3358  n_advance_bytes += sizeof (ip6[0]);
3359 
3360  if (PREDICT_FALSE (n_data_bytes < 0))
3361  {
3362  *error = TCP_ERROR_LENGTH;
3363  return 0;
3364  }
3365 
3366  if (!is_nolookup)
3367  {
3368  if (PREDICT_FALSE
3370  {
3371  ip4_main_t *im = &ip4_main;
3372  fib_index = vec_elt (im->fib_index_by_sw_if_index,
3374  }
3375 
3376  tc = session_lookup_connection_wt6 (fib_index, &ip6->dst_address,
3377  &ip6->src_address,
3378  tcp->dst_port, tcp->src_port,
3380  thread_index, &result);
3381  }
3382  }
3383 
3384  if (is_nolookup)
3385  tc =
3387  tcp.connection_index,
3388  thread_index);
3389 
3390  vnet_buffer (b)->tcp.seq_number = clib_net_to_host_u32 (tcp->seq_number);
3391  vnet_buffer (b)->tcp.ack_number = clib_net_to_host_u32 (tcp->ack_number);
3392  vnet_buffer (b)->tcp.data_offset = n_advance_bytes;
3393  vnet_buffer (b)->tcp.data_len = n_data_bytes;
3394  vnet_buffer (b)->tcp.seq_end = vnet_buffer (b)->tcp.seq_number
3395  + n_data_bytes;
3396  vnet_buffer (b)->tcp.flags = 0;
3397 
3398  *error = result ? TCP_ERROR_NONE + result : *error;
3399 
3401 }
3402 
3403 static inline void
3405  vlib_buffer_t * b, u16 * next, u32 * error)
3406 {
3407  tcp_header_t *tcp;
3408  u8 flags;
3409 
3410  tcp = tcp_buffer_hdr (b);
3411  flags = tcp->flags & filter_flags;
3412  *next = tm->dispatch_table[tc->state][flags].next;
3413  *error = tm->dispatch_table[tc->state][flags].error;
3414  tc->segs_in += 1;
3415 
3416  if (PREDICT_FALSE (*error == TCP_ERROR_DISPATCH
3417  || *next == TCP_INPUT_NEXT_RESET))
3418  {
3419  /* Overload tcp flags to store state */
3420  tcp_state_t state = tc->state;
3421  vnet_buffer (b)->tcp.flags = tc->state;
3422 
3423  if (*error == TCP_ERROR_DISPATCH)
3424  clib_warning ("tcp conn %u disp error state %U flags %U",
3425  tc->c_c_index, format_tcp_state, state,
3426  format_tcp_flags, (int) flags);
3427  }
3428 }
3429 
3432  vlib_frame_t * frame, int is_ip4, u8 is_nolookup)
3433 {
3434  u32 n_left_from, *from, thread_index = vm->thread_index;
3435  tcp_main_t *tm = vnet_get_tcp_main ();
3436  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
3437  u16 nexts[VLIB_FRAME_SIZE], *next;
3438 
3439  tcp_set_time_now (tcp_get_worker (thread_index));
3440 
3441  from = vlib_frame_vector_args (frame);
3442  n_left_from = frame->n_vectors;
3443  vlib_get_buffers (vm, from, bufs, n_left_from);
3444 
3445  b = bufs;
3446  next = nexts;
3447 
3448  while (n_left_from >= 4)
3449  {
3450  u32 error0 = TCP_ERROR_NO_LISTENER, error1 = TCP_ERROR_NO_LISTENER;
3451  tcp_connection_t *tc0, *tc1;
3452 
3453  {
3454  vlib_prefetch_buffer_header (b[2], STORE);
3455  CLIB_PREFETCH (b[2]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3456 
3457  vlib_prefetch_buffer_header (b[3], STORE);
3458  CLIB_PREFETCH (b[3]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3459  }
3460 
3461  next[0] = next[1] = TCP_INPUT_NEXT_DROP;
3462 
3463  tc0 = tcp_input_lookup_buffer (b[0], thread_index, &error0, is_ip4,
3464  is_nolookup);
3465  tc1 = tcp_input_lookup_buffer (b[1], thread_index, &error1, is_ip4,
3466  is_nolookup);
3467 
3468  if (PREDICT_TRUE (!tc0 + !tc1 == 0))
3469  {
3470  ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3471  ASSERT (tcp_lookup_is_valid (tc1, tcp_buffer_hdr (b[1])));
3472 
3473  vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3474  vnet_buffer (b[1])->tcp.connection_index = tc1->c_c_index;
3475 
3476  tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3477  tcp_input_dispatch_buffer (tm, tc1, b[1], &next[1], &error1);
3478  }
3479  else
3480  {
3481  if (PREDICT_TRUE (tc0 != 0))
3482  {
3483  ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3484  vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3485  tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3486  }
3487  else
3488  tcp_input_set_error_next (tm, &next[0], &error0, is_ip4);
3489 
3490  if (PREDICT_TRUE (tc1 != 0))
3491  {
3492  ASSERT (tcp_lookup_is_valid (tc1, tcp_buffer_hdr (b[1])));
3493  vnet_buffer (b[1])->tcp.connection_index = tc1->c_c_index;
3494  tcp_input_dispatch_buffer (tm, tc1, b[1], &next[1], &error1);
3495  }
3496  else
3497  tcp_input_set_error_next (tm, &next[1], &error1, is_ip4);
3498  }
3499 
3500  b += 2;
3501  next += 2;
3502  n_left_from -= 2;
3503  }
3504  while (n_left_from > 0)
3505  {
3506  tcp_connection_t *tc0;
3507  u32 error0 = TCP_ERROR_NO_LISTENER;
3508 
3509  if (n_left_from > 1)
3510  {
3511  vlib_prefetch_buffer_header (b[1], STORE);
3512  CLIB_PREFETCH (b[1]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3513  }
3514 
3515  next[0] = TCP_INPUT_NEXT_DROP;
3516  tc0 = tcp_input_lookup_buffer (b[0], thread_index, &error0, is_ip4,
3517  is_nolookup);
3518  if (PREDICT_TRUE (tc0 != 0))
3519  {
3520  ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3521  vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3522  tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3523  }
3524  else
3525  tcp_input_set_error_next (tm, &next[0], &error0, is_ip4);
3526 
3527  b += 1;
3528  next += 1;
3529  n_left_from -= 1;
3530  }
3531 
3533  tcp_input_trace_frame (vm, node, bufs, frame->n_vectors, is_ip4);
3534 
3535  vlib_buffer_enqueue_to_next (vm, node, from, nexts, frame->n_vectors);
3536  return frame->n_vectors;
3537 }
3538 
3540  vlib_node_runtime_t * node,
3541  vlib_frame_t * from_frame)
3542 {
3543  return tcp46_input_inline (vm, node, from_frame, 1 /* is_ip4 */ ,
3544  1 /* is_nolookup */ );
3545 }
3546 
3548  vlib_node_runtime_t * node,
3549  vlib_frame_t * from_frame)
3550 {
3551  return tcp46_input_inline (vm, node, from_frame, 0 /* is_ip4 */ ,
3552  1 /* is_nolookup */ );
3553 }
3554 
3555 /* *INDENT-OFF* */
3557 {
3558  .name = "tcp4-input-nolookup",
3559  /* Takes a vector of packets. */
3560  .vector_size = sizeof (u32),
3561  .n_errors = TCP_N_ERROR,
3562  .error_strings = tcp_error_strings,
3563  .n_next_nodes = TCP_INPUT_N_NEXT,
3564  .next_nodes =
3565  {
3566 #define _(s,n) [TCP_INPUT_NEXT_##s] = n,
3568 #undef _
3569  },
3570  .format_buffer = format_tcp_header,
3571  .format_trace = format_tcp_rx_trace,
3572 };
3573 /* *INDENT-ON* */
3574 
3575 /* *INDENT-OFF* */
3577 {
3578  .name = "tcp6-input-nolookup",
3579  /* Takes a vector of packets. */
3580  .vector_size = sizeof (u32),
3581  .n_errors = TCP_N_ERROR,
3582  .error_strings = tcp_error_strings,
3583  .n_next_nodes = TCP_INPUT_N_NEXT,
3584  .next_nodes =
3585  {
3586 #define _(s,n) [TCP_INPUT_NEXT_##s] = n,
3588 #undef _
3589  },
3590  .format_buffer = format_tcp_header,
3591  .format_trace = format_tcp_rx_trace,
3592 };
3593 /* *INDENT-ON* */
3594 
3596  vlib_frame_t * from_frame)
3597 {
3598  return tcp46_input_inline (vm, node, from_frame, 1 /* is_ip4 */ ,
3599  0 /* is_nolookup */ );
3600 }
3601 
3603  vlib_frame_t * from_frame)
3604 {
3605  return tcp46_input_inline (vm, node, from_frame, 0 /* is_ip4 */ ,
3606  0 /* is_nolookup */ );
3607 }
3608 
3609 /* *INDENT-OFF* */
3611 {
3612  .name = "tcp4-input",
3613  /* Takes a vector of packets. */
3614  .vector_size = sizeof (u32),
3615  .n_errors = TCP_N_ERROR,
3616  .error_strings = tcp_error_strings,
3617  .n_next_nodes = TCP_INPUT_N_NEXT,
3618  .next_nodes =
3619  {
3620 #define _(s,n) [TCP_INPUT_NEXT_##s] = n,
3622 #undef _
3623  },
3624  .format_buffer = format_tcp_header,
3625  .format_trace = format_tcp_rx_trace,
3626 };
3627 /* *INDENT-ON* */
3628 
3629 /* *INDENT-OFF* */
3631 {
3632  .name = "tcp6-input",
3633  /* Takes a vector of packets. */
3634  .vector_size = sizeof (u32),
3635  .n_errors = TCP_N_ERROR,
3636  .error_strings = tcp_error_strings,
3637  .n_next_nodes = TCP_INPUT_N_NEXT,
3638  .next_nodes =
3639  {
3640 #define _(s,n) [TCP_INPUT_NEXT_##s] = n,
3642 #undef _
3643  },
3644  .format_buffer = format_tcp_header,
3645  .format_trace = format_tcp_rx_trace,
3646 };
3647 /* *INDENT-ON* */
3648 
3649 #ifndef CLIB_MARCH_VARIANT
3650 static void
3652 {
3653  int i, j;
3654  for (i = 0; i < ARRAY_LEN (tm->dispatch_table); i++)
3655  for (j = 0; j < ARRAY_LEN (tm->dispatch_table[i]); j++)
3656  {
3657  tm->dispatch_table[i][j].next = TCP_INPUT_NEXT_DROP;
3658  tm->dispatch_table[i][j].error = TCP_ERROR_DISPATCH;
3659  }
3660 
3661 #define _(t,f,n,e) \
3662 do { \
3663  tm->dispatch_table[TCP_STATE_##t][f].next = (n); \
3664  tm->dispatch_table[TCP_STATE_##t][f].error = (e); \
3665 } while (0)
3666 
3667  /* RFC 793: In LISTEN if RST drop and if ACK return RST */
3668  _(LISTEN, 0, TCP_INPUT_NEXT_DROP, TCP_ERROR_SEGMENT_INVALID);
3669  _(LISTEN, TCP_FLAG_ACK, TCP_INPUT_NEXT_RESET, TCP_ERROR_ACK_INVALID);
3670  _(LISTEN, TCP_FLAG_RST, TCP_INPUT_NEXT_DROP, TCP_ERROR_INVALID_CONNECTION);
3671  _(LISTEN, TCP_FLAG_SYN, TCP_INPUT_NEXT_LISTEN, TCP_ERROR_NONE);
3673  TCP_ERROR_ACK_INVALID);
3675  TCP_ERROR_SEGMENT_INVALID);
3677  TCP_ERROR_SEGMENT_INVALID);
3679  TCP_ERROR_INVALID_CONNECTION);
3680  _(LISTEN, TCP_FLAG_FIN, TCP_INPUT_NEXT_RESET, TCP_ERROR_SEGMENT_INVALID);
3682  TCP_ERROR_SEGMENT_INVALID);
3684  TCP_ERROR_SEGMENT_INVALID);
3686  TCP_ERROR_NONE);
3688  TCP_ERROR_SEGMENT_INVALID);
3690  TCP_ERROR_SEGMENT_INVALID);
3692  TCP_ERROR_SEGMENT_INVALID);
3694  TCP_INPUT_NEXT_DROP, TCP_ERROR_SEGMENT_INVALID);
3695  /* ACK for for a SYN-ACK -> tcp-rcv-process. */
3696  _(SYN_RCVD, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3697  _(SYN_RCVD, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3699  TCP_ERROR_NONE);
3700  _(SYN_RCVD, TCP_FLAG_SYN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3702  TCP_ERROR_NONE);
3704  TCP_ERROR_NONE);
3705  _(SYN_RCVD, TCP_FLAG_SYN | TCP_FLAG_RST | TCP_FLAG_ACK,
3706  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3707  _(SYN_RCVD, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3709  TCP_ERROR_NONE);
3711  TCP_ERROR_NONE);
3712  _(SYN_RCVD, TCP_FLAG_FIN | TCP_FLAG_RST | TCP_FLAG_ACK,
3713  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3715  TCP_ERROR_NONE);
3716  _(SYN_RCVD, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_RST,
3717  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3718  _(SYN_RCVD, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_ACK,
3719  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3721  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3722  _(SYN_RCVD, 0, TCP_INPUT_NEXT_DROP, TCP_ERROR_SEGMENT_INVALID);
3723  /* SYN-ACK for a SYN */
3725  TCP_ERROR_NONE);
3726  _(SYN_SENT, TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE);
3727  _(SYN_SENT, TCP_FLAG_RST, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE);
3729  TCP_ERROR_NONE);
3730  _(SYN_SENT, TCP_FLAG_FIN, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE);
3732  TCP_ERROR_NONE);
3733  /* ACK for for established connection -> tcp-established. */
3734  _(ESTABLISHED, TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3735  /* FIN for for established connection -> tcp-established. */
3736  _(ESTABLISHED, TCP_FLAG_FIN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3738  TCP_ERROR_NONE);
3740  TCP_ERROR_NONE);
3741  _(ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_RST | TCP_FLAG_ACK,
3742  TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3744  TCP_ERROR_NONE);
3745  _(ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_ACK,
3746  TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3747  _(ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_RST,
3748  TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3749  _(ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_RST | TCP_FLAG_ACK,
3750  TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3751  _(ESTABLISHED, TCP_FLAG_RST, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3753  TCP_ERROR_NONE);
3754  _(ESTABLISHED, TCP_FLAG_SYN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3756  TCP_ERROR_NONE);
3758  TCP_ERROR_NONE);
3759  _(ESTABLISHED, TCP_FLAG_SYN | TCP_FLAG_RST | TCP_FLAG_ACK,
3760  TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3761  _(ESTABLISHED, 0, TCP_INPUT_NEXT_DROP, TCP_ERROR_SEGMENT_INVALID);
3762  /* ACK or FIN-ACK to our FIN */
3763  _(FIN_WAIT_1, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3765  TCP_ERROR_NONE);
3766  /* FIN in reply to our FIN from the other side */
3767  _(FIN_WAIT_1, 0, TCP_INPUT_NEXT_DROP, TCP_ERROR_SEGMENT_INVALID);
3768  _(FIN_WAIT_1, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3770  TCP_ERROR_NONE);
3771  _(FIN_WAIT_1, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_ACK,
3772  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3773  _(FIN_WAIT_1, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_RST,
3774  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3775  _(FIN_WAIT_1, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_RST | TCP_FLAG_ACK,
3776  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3778  TCP_ERROR_NONE);
3779  _(FIN_WAIT_1, TCP_FLAG_FIN | TCP_FLAG_RST | TCP_FLAG_ACK,
3780  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3781  _(FIN_WAIT_1, TCP_FLAG_SYN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3783  TCP_ERROR_NONE);
3785  TCP_ERROR_NONE);
3786  _(FIN_WAIT_1, TCP_FLAG_SYN | TCP_FLAG_RST | TCP_FLAG_ACK,
3787  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3788  _(FIN_WAIT_1, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3790  TCP_ERROR_NONE);
3791  _(CLOSING, 0, TCP_INPUT_NEXT_DROP, TCP_ERROR_SEGMENT_INVALID);
3792  _(CLOSING, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3793  _(CLOSING, TCP_FLAG_SYN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3795  TCP_ERROR_NONE);
3797  TCP_ERROR_NONE);
3798  _(CLOSING, TCP_FLAG_SYN | TCP_FLAG_RST | TCP_FLAG_ACK,
3799  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3800  _(CLOSING, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3802  TCP_ERROR_NONE);
3803  _(CLOSING, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3805  TCP_ERROR_NONE);
3807  TCP_ERROR_NONE);
3808  _(CLOSING, TCP_FLAG_FIN | TCP_FLAG_RST | TCP_FLAG_ACK,
3809  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3811  TCP_ERROR_NONE);
3812  _(CLOSING, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_ACK,
3813  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3814  _(CLOSING, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_RST,
3815  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3817  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3818  /* FIN confirming that the peer (app) has closed */
3819  _(FIN_WAIT_2, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3820  _(FIN_WAIT_2, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3822  TCP_ERROR_NONE);
3823  _(FIN_WAIT_2, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3825  TCP_ERROR_NONE);
3826  _(CLOSE_WAIT, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3828  TCP_ERROR_NONE);
3829  _(CLOSE_WAIT, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3831  TCP_ERROR_NONE);
3832  _(LAST_ACK, 0, TCP_INPUT_NEXT_DROP, TCP_ERROR_SEGMENT_INVALID);
3833  _(LAST_ACK, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3834  _(LAST_ACK, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3836  TCP_ERROR_NONE);
3838  TCP_ERROR_NONE);
3839  _(LAST_ACK, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_ACK,
3840  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3842  TCP_ERROR_NONE);
3843  _(LAST_ACK, TCP_FLAG_FIN | TCP_FLAG_RST | TCP_FLAG_ACK,
3844  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3845  _(LAST_ACK, TCP_FLAG_FIN | TCP_FLAG_SYN | TCP_FLAG_RST,
3846  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3848  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3849  _(LAST_ACK, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3851  TCP_ERROR_NONE);
3852  _(LAST_ACK, TCP_FLAG_SYN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3854  TCP_ERROR_NONE);
3856  TCP_ERROR_NONE);
3857  _(LAST_ACK, TCP_FLAG_SYN | TCP_FLAG_RST | TCP_FLAG_ACK,
3858  TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3859  _(TIME_WAIT, TCP_FLAG_SYN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3860  _(TIME_WAIT, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3862  TCP_ERROR_NONE);
3863  _(TIME_WAIT, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3865  TCP_ERROR_NONE);
3866  _(TIME_WAIT, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3867  /* RFC793 CLOSED: An incoming segment containing a RST is discarded. An
3868  * incoming segment not containing a RST causes a RST to be sent in
3869  * response.*/
3870  _(CLOSED, TCP_FLAG_RST, TCP_INPUT_NEXT_DROP, TCP_ERROR_CONNECTION_CLOSED);
3872  TCP_ERROR_CONNECTION_CLOSED);
3873  _(CLOSED, TCP_FLAG_ACK, TCP_INPUT_NEXT_RESET, TCP_ERROR_NONE);
3874  _(CLOSED, TCP_FLAG_SYN, TCP_INPUT_NEXT_RESET, TCP_ERROR_NONE);
3876  TCP_ERROR_NONE);
3877 #undef _
3878 }
3879 
3880 static clib_error_t *
3882 {
3883  clib_error_t *error = 0;
3884  tcp_main_t *tm = vnet_get_tcp_main ();
3885 
3886  if ((error = vlib_call_init_function (vm, tcp_init)))
3887  return error;
3888 
3889  /* Initialize dispatch table. */
3891 
3892  return error;
3893 }
3894 
3896 
3897 #endif /* CLIB_MARCH_VARIANT */
3898 
3899 /*
3900  * fd.io coding-style-patch-verification: ON
3901  *
3902  * Local Variables:
3903  * eval: (c-set-style "gnu")
3904  * End:
3905  */
static void tcp_program_disconnect(tcp_worker_ctx_t *wrk, tcp_connection_t *tc)
Definition: tcp_input.c:1608
#define tcp_in_cong_recovery(tc)
Definition: tcp.h:438
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:1790
static void tcp_update_timestamp(tcp_connection_t *tc, u32 seq, u32 seq_end)
Update tsval recent.
Definition: tcp_input.c:249
static sack_scoreboard_hole_t * scoreboard_insert_hole(sack_scoreboard_t *sb, u32 prev_index, u32 start, u32 end)
Definition: tcp_input.c:737
static u8 tcp_scoreboard_is_sane_post_recovery(tcp_connection_t *tc)
Test that scoreboard is sane after recovery.
Definition: tcp_input.c:934
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:124
void scoreboard_clear(sack_scoreboard_t *sb)
Definition: tcp_input.c:906
static u8 tcp_should_fastrecover(tcp_connection_t *tc)
Definition: tcp_input.c:1307
End of options.
Definition: tcp_packet.h:104
u32 flags
Definition: vhost_user.h:141
#define clib_min(x, y)
Definition: clib.h:295
#define CLIB_UNUSED(x)
Definition: clib.h:82
u32 * pending_disconnects
vector of pending disconnect notifications
Definition: tcp.h:487
vlib_node_registration_t tcp6_rcv_process_node
(constructor) VLIB_REGISTER_NODE (tcp6_rcv_process_node)
Definition: tcp_input.c:3014
#define tcp_in_recovery(tc)
Definition: tcp.h:429
static f64 tcp_time_now_us(u32 thread_index)
Definition: tcp.h:993
static void tcp_rcv_fin(tcp_worker_ctx_t *wrk, tcp_connection_t *tc, vlib_buffer_t *b, u32 *error)
Definition: tcp_input.c:1639
#define TCP_OPTION_LEN_SACK_PERMITTED
Definition: tcp_packet.h:167
#define seq_leq(_s1, _s2)
Definition: tcp.h:823
struct _sack_block sack_block_t
void tcp_rcv_sacks(tcp_connection_t *tc, u32 ack)
Definition: tcp_input.c:945
static void vlib_buffer_free(vlib_main_t *vm, u32 *buffers, u32 n_buffers)
Free buffers Frees the entire buffer chain for each buffer.
Definition: buffer_funcs.h:865
#define timestamp_leq(_t1, _t2)
Definition: tcp.h:830
ip4_address_t src_address
Definition: ip4_packet.h:170
static u8 tcp_cc_is_spurious_retransmit(tcp_connection_t *tc)
Definition: tcp_input.c:1249
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 *result)
Lookup connection with ip6 and transport layer information.
enum _tcp_state_next tcp_state_next_t
#define tcp_rst(_th)
Definition: tcp_packet.h:81
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:112
#define tcp_inc_err_counter(cnts, err, val)
Definition: tcp_input.c:2072
unsigned long u64
Definition: types.h:89
#define tcp_store_err_counters(node_id, cnts)
Definition: tcp_input.c:2076
static void tcp_dispatch_table_init(tcp_main_t *tm)
Definition: tcp_input.c:3651
#define clib_memcpy_fast(a, b, c)
Definition: string.h:81
static u8 * format_tcp_rx_trace_short(u8 *s, va_list *args)
Definition: tcp_input.c:1983
static int tcp_segment_rcv(tcp_worker_ctx_t *wrk, tcp_connection_t *tc, vlib_buffer_t *b)
Receive buffer for connection and handle acks.
Definition: tcp_input.c:1894
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:2086
static tcp_connection_t * tcp_half_open_connection_get(u32 conn_index)
Definition: tcp.h:726
void tcp_update_rto(tcp_connection_t *tc)
Definition: tcp_input.c:469
svm_fifo_t * rx_fifo
Pointers to rx/tx buffers.
#define tcp_doff(_th)
Definition: tcp_packet.h:78
struct _tcp_main tcp_main_t
u32 thread_index
Definition: main.h:197
void tcp_connection_timers_reset(tcp_connection_t *tc)
Stop all connection timers.
Definition: tcp.c:488
u16 current_length
Nbytes between current data and the end of this buffer.
Definition: buffer.h:113
int session_main_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:631
u8 data[0]
Packet data.
Definition: buffer.h:181
#define vec_add1(V, E)
Add 1 element to end of vector (unspecified alignment).
Definition: vec.h:522
#define tcp_recovery_off(tc)
Definition: tcp.h:427
#define clib_abs(x)
Definition: clib.h:302
static int tcp_update_rtt(tcp_connection_t *tc, u32 ack)
Update RTT estimate and RTO timer.
Definition: tcp_input.c:488
#define vec_add2(V, P, N)
Add N elements to end of vector V, return pointer to new elements in P.
Definition: vec.h:560
int i
#define THZ
TCP tick frequency.
Definition: tcp.h:28
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:2995
clib_memset(h->entries, 0, sizeof(h->entries[0])*entries)
u32 * fib_index_by_sw_if_index
Table index indexed by software interface.
Definition: ip4.h:121
struct _tcp_connection tcp_connection_t
static session_t * session_get(u32 si, u32 thread_index)
Definition: session.h:258
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:424
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:933
#define tcp_opts_sack(_to)
Definition: tcp_packet.h:159
u8 data[128]
Definition: ipsec.api:249
tcp_connection_t tcp_connection
Definition: tcp_input.c:1963
static u8 tcp_sack_vector_is_sane(sack_block_t *sacks)
Definition: tcp_input.c:1660
static tcp_connection_t * tcp_get_connection_from_transport(transport_connection_t *tconn)
Definition: tcp.h:691
#define VLIB_NODE_FN(node)
Definition: node.h:201
static void tcp_cc_congestion_undo(tcp_connection_t *tc)
Definition: tcp_input.c:1213
#define tcp_disconnect_pending_on(tc)
Definition: tcp.h:432
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:386
u64 session_lookup_half_open_handle(transport_connection_t *tc)
No operation.
Definition: tcp_packet.h:105
format_function_t format_tcp_flags
Definition: tcp.h:63
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:236
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:211
ip6_address_t src_address
Definition: ip6_packet.h:383
u32 * pending_deq_acked
vector of pending ack dequeues
Definition: tcp.h:484
unsigned char u8
Definition: types.h:56
#define tcp_inc_counter(node_id, err, count)
Definition: tcp_input.c:2064
vlib_node_registration_t tcp6_syn_sent_node
(constructor) VLIB_REGISTER_NODE (tcp6_syn_sent_node)
Definition: tcp_input.c:2588
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:2261
double f64
Definition: types.h:142
#define tcp_fastrecovery_on(tc)
Definition: tcp.h:424
Limit MSS.
Definition: tcp_packet.h:106
void session_transport_closing_notify(transport_connection_t *tc)
Notification from transport that connection is being closed.
Definition: session.c:829
sack_scoreboard_hole_t * scoreboard_get_hole(sack_scoreboard_t *sb, u32 index)
Definition: tcp_input.c:662
#define TCP_TICK
TCP tick period (s)
Definition: tcp.h:27
#define tcp_is_fin(_th)
Definition: tcp_packet.h:90
#define seq_gt(_s1, _s2)
Definition: tcp.h:824
static u8 * format_tcp_rx_trace(u8 *s, va_list *args)
Definition: tcp_input.c:1967
static void tcp_connection_set_state(tcp_connection_t *tc, tcp_state_t state)
Definition: tcp.h:697
void tcp_init_snd_vars(tcp_connection_t *tc)
Initialize connection send variables.
Definition: tcp.c:619
#define tcp_cfg
Definition: tcp.h:634
vl_api_interface_index_t sw_if_index
Definition: gre.api:50
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:173
vlib_node_registration_t tcp4_established_node
(constructor) VLIB_REGISTER_NODE (tcp4_established_node)
Definition: tcp_input.c:2184
#define always_inline
Definition: clib.h:98
#define TCP_OPTION_LEN_SACK_BLOCK
Definition: tcp_packet.h:169
ip4_address_t dst_address
Definition: ip4_packet.h:170
#define TCP_FLAG_ACK
Definition: fa_node.h:16
u8 * format_white_space(u8 *s, va_list *va)
Definition: std-formats.c:129
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 *result)
Lookup connection with ip4 and transport layer information.
static tcp_header_t * tcp_buffer_hdr(vlib_buffer_t *b)
Definition: tcp.h:651
static void tcp_cc_recovery_exit(tcp_connection_t *tc)
Definition: tcp_input.c:1186
#define vlib_prefetch_buffer_header(b, type)
Prefetch buffer metadata.
Definition: buffer.h:203
static int tcp_segment_validate(tcp_worker_ctx_t *wrk, tcp_connection_t *tc0, vlib_buffer_t *b0, tcp_header_t *th0, u32 *error0)
Validate incoming segment as per RFC793 p.
Definition: tcp_input.c:277
enum _tcp_state tcp_state_t
#define TCP_ALWAYS_ACK
On/off delayed acks.
Definition: tcp.h:37
vlib_node_registration_t tcp6_input_node
(constructor) VLIB_REGISTER_NODE (tcp6_input_node)
Definition: tcp_input.c:3630
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:623
vhost_vring_state_t state
Definition: vhost_user.h:146
#define TCP_RTO_MAX
Definition: tcp.h:97
static u32 ooo_segment_length(svm_fifo_t *f, ooo_segment_t *s)
Definition: svm_fifo.h:713
static void * ip4_next_header(ip4_header_t *i)
Definition: ip4_packet.h:241
static u32 tcp_time_now(void)
Definition: tcp.h:971
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.
#define vlib_call_init_function(vm, x)
Definition: init.h:270
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:2048
#define TCP_MAX_SACK_BLOCKS
Max number of SACK blocks stored.
Definition: tcp.h:140
#define VLIB_FRAME_SIZE
Definition: node.h:376
#define tcp_validate_txf_size(_tc, _a)
Definition: tcp.h:1158
#define TCP_EVT_DBG(_evt, _args...)
Definition: tcp_debug.h:243
static int tcp_options_parse(tcp_header_t *th, tcp_options_t *to, u8 is_syn)
Parse TCP header options.
Definition: tcp_input.c:125
#define timestamp_lt(_t1, _t2)
Definition: tcp.h:829
static void tcp_timer_set(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:1058
#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:697
static heap_elt_t * first(heap_header_t *h)
Definition: heap.c:59
void scoreboard_init(sack_scoreboard_t *sb)
Definition: tcp_input.c:898
vlib_main_t * vm
convenience pointer to this thread&#39;s vlib main
Definition: tcp.h:490
#define TCP_INVALID_SACK_HOLE_INDEX
Definition: tcp.h:141
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:514
static void tcp_program_dequeue(tcp_worker_ctx_t *wrk, tcp_connection_t *tc)
Definition: tcp_input.c:609
static void tcp_handle_disconnects(tcp_worker_ctx_t *wrk)
Definition: tcp_input.c:1618
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:3036
#define tcp_in_fastrecovery(tc)
Definition: tcp.h:428
void tcp_connection_tx_pacer_reset(tcp_connection_t *tc, u32 window, u32 start_bucket)
Definition: tcp.c:1280
static void tcp_input_set_error_next(tcp_main_t *tm, u16 *next, u32 *error, u8 is_ip4)
Definition: tcp_input.c:3289
vlib_node_registration_t tcp4_input_nolookup_node
(constructor) VLIB_REGISTER_NODE (tcp4_input_nolookup_node)
Definition: tcp_input.c:3556
unsigned short u16
Definition: types.h:57
#define foreach_tcp4_input_next
Definition: tcp_input.c:3246
tcp_connection_t * tcp_connection_alloc(u8 thread_index)
Definition: tcp.c:297
static void * vlib_buffer_get_current(vlib_buffer_t *b)
Get pointer to current data to process.
Definition: buffer.h:229
#define filter_flags
Definition: tcp_input.c:3264
void tcp_connection_tx_pacer_update(tcp_connection_t *tc)
Definition: tcp.c:1264
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:286
static int tcp_buffer_discard_bytes(vlib_buffer_t *b, u32 n_bytes_to_drop)
Definition: tcp_input.c:1861
#define foreach_tcp6_input_next
Definition: tcp_input.c:3255
#define TCP_TIMER_HANDLE_INVALID
Definition: tcp.h:90
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:3267
static void tcp_cc_rcv_cong_ack(tcp_connection_t *tc, tcp_cc_ack_t ack_type, tcp_rate_sample_t *rs)
Definition: tcp.h:1025
#define PREDICT_FALSE(x)
Definition: clib.h:111
static int tcp_rcv_ack_no_cc(tcp_connection_t *tc, vlib_buffer_t *b, u32 *error)
Definition: tcp_input.c:412
#define vec_del1(v, i)
Delete the element at index I.
Definition: vec.h:804
#define TCP_FLAG_FIN
Definition: fa_node.h:12
static u8 tcp_cc_is_spurious_fast_rxt(tcp_connection_t *tc)
Definition: tcp_input.c:1242
static void tcp_cc_handle_event(tcp_connection_t *tc, tcp_rate_sample_t *rs, u32 is_dack)
One function to rule them all ...
Definition: tcp_input.c:1317
vlib_node_registration_t tcp4_listen_node
(constructor) VLIB_REGISTER_NODE (tcp4_listen_node)
Definition: tcp_input.c:3197
static void scoreboard_init_high_rxt(sack_scoreboard_t *sb, u32 snd_una)
Definition: tcp_input.c:883
#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:2223
static ooo_segment_t * svm_fifo_newest_ooo_segment(svm_fifo_t *f)
Definition: svm_fifo.h:689
u32 tcp_sack_list_bytes(tcp_connection_t *tc)
Definition: tcp_input.c:1732
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:2203
sack_scoreboard_hole_t * scoreboard_first_hole(sack_scoreboard_t *sb)
Definition: tcp_input.c:686
static int tcp_can_delack(tcp_connection_t *tc)
Check if ACK could be delayed.
Definition: tcp_input.c:1845
static void vlib_node_increment_counter(vlib_main_t *vm, u32 node_index, u32 counter_index, u64 increment)
Definition: node_funcs.h:1150
static int tcp_cc_recover(tcp_connection_t *tc)
Definition: tcp_input.c:1256
#define TCP_FLAG_RST
Definition: fa_node.h:14
#define TCP_DBG(_fmt, _args...)
Definition: tcp_debug.h:244
static int tcp_rcv_ack(tcp_worker_ctx_t *wrk, tcp_connection_t *tc, vlib_buffer_t *b, tcp_header_t *th, u32 *error)
Process incoming ACK.
Definition: tcp_input.c:1510
#define TCP_MAX_WND_SCALE
Definition: tcp_packet.h:173
void tcp_connection_free(tcp_connection_t *tc)
Definition: tcp.c:310
#define VLIB_REGISTER_NODE(x,...)
Definition: node.h:169
vlib_node_registration_t tcp4_syn_sent_node
(constructor) VLIB_REGISTER_NODE (tcp4_syn_sent_node)
Definition: tcp_input.c:2569
u16 n_vectors
Definition: node.h:395
#define CLIB_PREFETCH(addr, size, type)
Definition: cache.h:80
vlib_main_t * vm
Definition: buffer.c:312
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:1998
void tcp_program_dupack(tcp_connection_t *tc)
Definition: tcp_output.c:1199
void tcp_send_reset(tcp_connection_t *tc)
Build and set reset packet for connection.
Definition: tcp_output.c:846
#define TCP_DUPACK_THRESHOLD
Definition: tcp.h:35
static u32 tcp_tstamp(tcp_connection_t *tc)
Generate timestamp for tcp connection.
Definition: tcp.h:986
static tcp_connection_t * tcp_input_lookup_buffer(vlib_buffer_t *b, u8 thread_index, u32 *error, u8 is_ip4, u8 is_nolookup)
Definition: tcp_input.c:3308
format_function_t format_tcp_state
Definition: tcp.h:62
static void tcp_cc_undo_recovery(tcp_connection_t *tc)
Definition: tcp.h:1044
#define clib_warning(format, args...)
Definition: error.h:59
tcp_header_t tcp_header
Definition: tcp_input.c:1962
format_function_t format_tcp_header
Definition: format.h:101
struct _transport_connection transport_connection_t
#define pool_is_free_index(P, I)
Use free bitmap to query whether given index is free.
Definition: pool.h:283
#define ARRAY_LEN(x)
Definition: clib.h:62
#define TCP_RTT_MAX
Definition: tcp.h:99
u16 mss
Option flags, see above.
Definition: tcp_packet.h:147
static void * ip6_next_header(ip6_header_t *i)
Definition: ip6_packet.h:410
static u32 transport_max_tx_dequeue(transport_connection_t *tc)
Definition: session.h:439
void tcp_send_synack(tcp_connection_t *tc)
Definition: tcp_output.c:961
static void tcp_timer_update(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:1082
#define TCP_PAWS_IDLE
24 days
Definition: tcp.h:30
vslo right
#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:3881
#define tcp_fastrecovery_first_on(tc)
Definition: tcp.h:435
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:445
enum _tcp_rcv_process_next tcp_rcv_process_next_t
#define seq_geq(_s1, _s2)
Definition: tcp.h:825
IPv4 main type.
Definition: ip4.h:105
static void transport_add_tx_event(transport_connection_t *tc)
Definition: session.h:480
static void tcp_cc_update(tcp_connection_t *tc, tcp_rate_sample_t *rs)
Definition: tcp_input.c:1280
static void tcp_handle_postponed_dequeues(tcp_worker_ctx_t *wrk)
Dequeue bytes for connections that have received acks in last burst.
Definition: tcp_input.c:566
void tcp_bt_sample_delivery_rate(tcp_connection_t *tc, tcp_rate_sample_t *rs)
Generate a delivery rate sample from recently acked bytes.
Definition: tcp_bt.c:508
static void tcp_estimate_initial_rtt(tcp_connection_t *tc)
Definition: tcp_input.c:535
static void vlib_buffer_advance(vlib_buffer_t *b, word l)
Advance current data pointer by the supplied (signed!) amount.
Definition: buffer.h:248
static int tcp_segment_check_paws(tcp_connection_t *tc)
RFC1323: Check against wrapped sequence numbers (PAWS).
Definition: tcp_input.c:239
static uword ip6_address_is_link_local_unicast(const ip6_address_t *a)
Definition: ip6_packet.h:326
static u8 tcp_cc_is_spurious_timeout_rxt(tcp_connection_t *tc)
Definition: tcp_input.c:1233
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:2014
enum _tcp_input_next tcp_input_next_t
void tcp_update_sack_list(tcp_connection_t *tc, u32 start, u32 end)
Build SACK list as per RFC2018.
Definition: tcp_input.c:1683
#define tcp_fastrecovery_first_off(tc)
Definition: tcp.h:436
int session_stream_accept_notify(transport_connection_t *tc)
Definition: session.c:966
Out-of-order segment.
Definition: svm_fifo.h:29
static u8 tcp_segment_in_rcv_wnd(tcp_connection_t *tc, u32 seq, u32 end_seq)
Validate segment sequence number.
Definition: tcp_input.c:110
#define clib_max(x, y)
Definition: clib.h:288
static vlib_main_t * vlib_get_main(void)
Definition: global_funcs.h:23
static u32 tcp_time_now_w_thread(u32 thread_index)
Definition: tcp.h:977
static clib_error_t * tcp_init(vlib_main_t *vm)
Definition: tcp.c:1546
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
#define vec_elt(v, i)
Get vector value at index i.
#define seq_lt(_s1, _s2)
Definition: tcp.h:822
#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
void tcp_send_reset_w_pkt(tcp_connection_t *tc, vlib_buffer_t *pkt, u32 thread_index, u8 is_ip4)
Send reset without reusing existing buffer.
Definition: tcp_output.c:764
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:1132
void tcp_connection_reset(tcp_connection_t *tc)
Notify session that connection has been reset.
Definition: tcp.c:328
u32 tsval
Timestamp value.
Definition: tcp_packet.h:149
enum _tcp_established_next tcp_established_next_t
u16 payload_length
Definition: ip6_packet.h:374
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:3610
void tcp_send_fin(tcp_connection_t *tc)
Send FIN.
Definition: tcp_output.c:1031
#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
u32 next_buffer
Next buffer for this linked-list of buffers.
Definition: buffer.h:140
static u8 tcp_is_lost_fin(tcp_connection_t *tc)
Definition: tcp.h:945
static u32 scoreboard_hole_bytes(sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:656
static void tcp_cc_rcv_ack(tcp_connection_t *tc, tcp_rate_sample_t *rs)
Definition: tcp.h:1018
static tcp_worker_ctx_t * tcp_get_worker(u32 thread_index)
Definition: tcp.h:645
void session_transport_closed_notify(transport_connection_t *tc)
Notification from transport that it is closed.
Definition: session.c:916
static void tcp_retransmit_timer_update(tcp_connection_t *tc)
Definition: tcp.h:1139
VLIB buffer representation.
Definition: buffer.h:102
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:1743
static u8 tcp_should_fastrecover_sack(tcp_connection_t *tc)
Definition: tcp_input.c:1301
u64 uword
Definition: types.h:112
#define seq_max(_s1, _s2)
Definition: tcp.h:826
sack_scoreboard_hole_t * scoreboard_next_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:670
sack_scoreboard_hole_t * scoreboard_prev_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:678
static void * vlib_frame_vector_args(vlib_frame_t *f)
Get pointer to frame vector data.
Definition: node_funcs.h:244
void tcp_connection_init_vars(tcp_connection_t *tc)
Initialize tcp connection variables.
Definition: tcp.c:652
void tcp_cc_fastrecovery_clear(tcp_connection_t *tc)
Definition: tcp_input.c:1198
static void tcp_cc_recovered(tcp_connection_t *tc)
Definition: tcp.h:1038
static void scoreboard_remove_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:702
sack_scoreboard_hole_t * scoreboard_next_rxt_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *start, u8 have_unsent, u8 *can_rescue, u8 *snd_limited)
Figure out the next hole to retransmit.
Definition: tcp_input.c:828
#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:694
static void scoreboard_update_bytes(tcp_connection_t *tc, sack_scoreboard_t *sb)
Definition: tcp_input.c:776
#define tcp_disconnect_pending(tc)
Definition: tcp.h:431
left
void tcp_program_fastretransmit(tcp_connection_t *tc)
Definition: tcp_output.c:1211
#define TCP_RTO_MIN
Definition: tcp.h:98
static u32 ooo_segment_offset_prod(svm_fifo_t *f, ooo_segment_t *s)
Definition: svm_fifo.h:703
struct clib_bihash_value offset
template key/value backing page structure
#define tcp_scoreboard_trace_add(_tc, _ack)
Definition: tcp.h:205
u8 * format_tcp_connection(u8 *s, va_list *args)
Definition: tcp.c:954
#define vnet_buffer(b)
Definition: buffer.h:361
static tcp_connection_t * tcp_connection_get(u32 conn_index, u32 thread_index)
Definition: tcp.h:672
static u32 scoreboard_hole_index(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:649
ip4_main_t ip4_main
Global ip4 main structure.
Definition: ip4_forward.c:921
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:729
#define tcp_disconnect_pending_off(tc)
Definition: tcp.h:433
static u32 vlib_num_workers()
Definition: threads.h:367
void tcp_connection_cleanup(tcp_connection_t *tc)
Cleans up connection state.
Definition: tcp.c:239
u16 flags
Copy of main node flags.
Definition: node.h:507
Window scale.
Definition: tcp_packet.h:107
u32 session_tx_fifo_dequeue_drop(transport_connection_t *tc, u32 max_bytes)
Definition: session.c:483
void tcp_program_ack(tcp_connection_t *tc)
Definition: tcp_output.c:1189
vlib_node_registration_t tcp6_listen_node
(constructor) VLIB_REGISTER_NODE (tcp6_listen_node)
Definition: tcp_input.c:3216
#define tcp_opts_sack_permitted(_to)
Definition: tcp_packet.h:160
static int ip4_header_bytes(const ip4_header_t *i)
Definition: ip4_packet.h:235
Timestamps.
Definition: tcp_packet.h:110
int session_stream_accept(transport_connection_t *tc, u32 listener_index, u32 thread_index, u8 notify)
Accept a stream session.
Definition: session.c:983
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:244
#define VLIB_NODE_FLAG_TRACE
Definition: node.h:301
#define CLIB_CACHE_LINE_BYTES
Definition: cache.h:59
u32 total_length_not_including_first_buffer
Only valid for first buffer in chain.
Definition: buffer.h:167
static uword tcp46_input_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame, int is_ip4, u8 is_nolookup)
Definition: tcp_input.c:3431
static void tcp_persist_timer_set(tcp_connection_t *tc)
Definition: tcp.h:1118
static tcp_main_t * vnet_get_tcp_main()
Definition: tcp.h:639
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:2302
#define tcp_fastrecovery_off(tc)
Definition: tcp.h:425
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:2611
static void tcp_retransmit_timer_reset(tcp_connection_t *tc)
Definition: tcp.h:1105
static vlib_buffer_t * vlib_get_buffer(vlib_main_t *vm, u32 buffer_index)
Translate buffer index into buffer pointer.
Definition: buffer_funcs.h:85
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:3404
vlib_node_registration_t tcp6_input_nolookup_node
(constructor) VLIB_REGISTER_NODE (tcp6_input_nolookup_node)
Definition: tcp_input.c:3576
static u32 tcp_set_time_now(tcp_worker_ctx_t *wrk)
Definition: tcp.h:999
#define tcp_ack(_th)
Definition: tcp_packet.h:83
static u32 transport_tx_fifo_size(transport_connection_t *tc)
Definition: session.h:460
static u8 tcp_timer_is_active(tcp_connection_t *tc, tcp_timers_e timer)
Definition: tcp.h:1153
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:720
ip6_address_t dst_address
Definition: ip6_packet.h:383
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:636
void tcp_cc_init_congestion(tcp_connection_t *tc)
Init loss recovery/fast recovery.
Definition: tcp_input.c:1171
static void tcp_persist_timer_reset(tcp_connection_t *tc)
Definition: tcp.h:1133
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