FD.io VPP  v17.07-30-g839fa73
Vector Packet Processing
tcp_output.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14  */
15 
16 #include <vnet/tcp/tcp.h>
17 #include <vnet/lisp-cp/packets.h>
18 
21 
22 typedef enum _tcp_output_nect
23 {
27 
28 #define foreach_tcp4_output_next \
29  _ (DROP, "error-drop") \
30 
31 #define foreach_tcp6_output_next \
32  _ (DROP, "error-drop") \
33 
34 static char *tcp_error_strings[] = {
35 #define tcp_error(n,s) s,
36 #include <vnet/tcp/tcp_error.def>
37 #undef tcp_error
38 };
39 
40 typedef struct
41 {
45 
46 u16 dummy_mtu = 1460;
47 
48 u8 *
49 format_tcp_tx_trace (u8 * s, va_list * args)
50 {
51  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
52  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
53  tcp_tx_trace_t *t = va_arg (*args, tcp_tx_trace_t *);
54  uword indent = format_get_indent (s);
55 
56  s = format (s, "%U\n%U%U",
57  format_tcp_header, &t->tcp_header, 128,
58  format_white_space, indent,
60 
61  return s;
62 }
63 
64 static u8
65 tcp_window_compute_scale (u32 available_space)
66 {
67  u8 wnd_scale = 0;
68  while (wnd_scale < TCP_MAX_WND_SCALE
69  && (available_space >> wnd_scale) > TCP_WND_MAX)
70  wnd_scale++;
71  return wnd_scale;
72 }
73 
74 /**
75  * TCP's IW as recommended by RFC6928
76  */
79 {
80  return TCP_IW_N_SEGMENTS * tc->mss;
81 }
82 
83 /**
84  * Compute initial window and scale factor. As per RFC1323, window field in
85  * SYN and SYN-ACK segments is never scaled.
86  */
87 u32
89 {
90  u32 max_fifo;
91 
92  /* Initial wnd for SYN. Fifos are not allocated yet.
93  * Use some predefined value. For SYN-ACK we still want the
94  * scale to be computed in the same way */
95  max_fifo = TCP_MAX_RX_FIFO_SIZE;
96 
97  tc->rcv_wscale = tcp_window_compute_scale (max_fifo);
98  tc->rcv_wnd = tcp_initial_wnd_unscaled (tc);
99 
100  return clib_min (tc->rcv_wnd, TCP_WND_MAX);
101 }
102 
103 /**
104  * Compute and return window to advertise, scaled as per RFC1323
105  */
106 u32
108 {
109  if (state < TCP_STATE_ESTABLISHED)
111 
112  tcp_update_rcv_wnd (tc);
113 
114  if (tc->rcv_wnd == 0)
115  {
116  tc->flags |= TCP_CONN_SENT_RCV_WND0;
117  }
118  else
119  {
120  tc->flags &= ~TCP_CONN_SENT_RCV_WND0;
121  }
122 
123  return tc->rcv_wnd >> tc->rcv_wscale;
124 }
125 
126 void
128 {
129  i32 observed_wnd;
130  u32 available_space, max_fifo, wnd;
131 
132  /*
133  * Figure out how much space we have available
134  */
135  available_space = stream_session_max_rx_enqueue (&tc->connection);
136  max_fifo = stream_session_rx_fifo_size (&tc->connection);
137 
138  ASSERT (tc->rcv_opts.mss < max_fifo);
139  if (available_space < tc->rcv_opts.mss && available_space < max_fifo >> 3)
140  available_space = 0;
141 
142  /*
143  * Use the above and what we know about what we've previously advertised
144  * to compute the new window
145  */
146  observed_wnd = (i32) tc->rcv_wnd - (tc->rcv_nxt - tc->rcv_las);
147  if (observed_wnd < 0)
148  observed_wnd = 0;
149 
150  /* Bad. Thou shalt not shrink */
151  if (available_space < observed_wnd)
152  {
153  wnd = observed_wnd;
154  TCP_EVT_DBG (TCP_EVT_RCV_WND_SHRUNK, tc, observed_wnd, available_space);
155  }
156  else
157  {
158  wnd = available_space;
159  }
160 
161  /* Make sure we have a multiple of rcv_wscale */
162  if (wnd && tc->rcv_wscale)
163  {
164  wnd &= ~(1 << tc->rcv_wscale);
165  if (wnd == 0)
166  wnd = 1 << tc->rcv_wscale;
167  }
168 
169  tc->rcv_wnd = clib_min (wnd, TCP_WND_MAX << tc->rcv_wscale);
170 }
171 
172 /**
173  * Write TCP options to segment.
174  */
175 u32
177 {
178  u32 opts_len = 0;
179  u32 buf, seq_len = 4;
180 
181  if (tcp_opts_mss (opts))
182  {
183  *data++ = TCP_OPTION_MSS;
184  *data++ = TCP_OPTION_LEN_MSS;
185  buf = clib_host_to_net_u16 (opts->mss);
186  clib_memcpy (data, &buf, sizeof (opts->mss));
187  data += sizeof (opts->mss);
188  opts_len += TCP_OPTION_LEN_MSS;
189  }
190 
191  if (tcp_opts_wscale (opts))
192  {
193  *data++ = TCP_OPTION_WINDOW_SCALE;
194  *data++ = TCP_OPTION_LEN_WINDOW_SCALE;
195  *data++ = opts->wscale;
196  opts_len += TCP_OPTION_LEN_WINDOW_SCALE;
197  }
198 
199  if (tcp_opts_sack_permitted (opts))
200  {
201  *data++ = TCP_OPTION_SACK_PERMITTED;
203  opts_len += TCP_OPTION_LEN_SACK_PERMITTED;
204  }
205 
206  if (tcp_opts_tstamp (opts))
207  {
208  *data++ = TCP_OPTION_TIMESTAMP;
209  *data++ = TCP_OPTION_LEN_TIMESTAMP;
210  buf = clib_host_to_net_u32 (opts->tsval);
211  clib_memcpy (data, &buf, sizeof (opts->tsval));
212  data += sizeof (opts->tsval);
213  buf = clib_host_to_net_u32 (opts->tsecr);
214  clib_memcpy (data, &buf, sizeof (opts->tsecr));
215  data += sizeof (opts->tsecr);
216  opts_len += TCP_OPTION_LEN_TIMESTAMP;
217  }
218 
219  if (tcp_opts_sack (opts))
220  {
221  int i;
222  u32 n_sack_blocks = clib_min (vec_len (opts->sacks),
224 
225  if (n_sack_blocks != 0)
226  {
227  *data++ = TCP_OPTION_SACK_BLOCK;
228  *data++ = 2 + n_sack_blocks * TCP_OPTION_LEN_SACK_BLOCK;
229  for (i = 0; i < n_sack_blocks; i++)
230  {
231  buf = clib_host_to_net_u32 (opts->sacks[i].start);
232  clib_memcpy (data, &buf, seq_len);
233  data += seq_len;
234  buf = clib_host_to_net_u32 (opts->sacks[i].end);
235  clib_memcpy (data, &buf, seq_len);
236  data += seq_len;
237  }
238  opts_len += 2 + n_sack_blocks * TCP_OPTION_LEN_SACK_BLOCK;
239  }
240  }
241 
242  /* Terminate TCP options */
243  if (opts_len % 4)
244  {
245  *data++ = TCP_OPTION_EOL;
246  opts_len += TCP_OPTION_LEN_EOL;
247  }
248 
249  /* Pad with zeroes to a u32 boundary */
250  while (opts_len % 4)
251  {
252  *data++ = TCP_OPTION_NOOP;
253  opts_len += TCP_OPTION_LEN_NOOP;
254  }
255  return opts_len;
256 }
257 
258 always_inline int
260 {
261  u8 len = 0;
262 
263  opts->flags |= TCP_OPTS_FLAG_MSS;
264  opts->mss = dummy_mtu; /*XXX discover that */
265  len += TCP_OPTION_LEN_MSS;
266 
267  opts->flags |= TCP_OPTS_FLAG_WSCALE;
268  opts->wscale = wnd_scale;
270 
271  opts->flags |= TCP_OPTS_FLAG_TSTAMP;
272  opts->tsval = tcp_time_now ();
273  opts->tsecr = 0;
275 
276  if (TCP_USE_SACKS)
277  {
278  opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED;
280  }
281 
282  /* Align to needed boundary */
283  len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
284  return len;
285 }
286 
287 always_inline int
289 {
290  u8 len = 0;
291 
292  opts->flags |= TCP_OPTS_FLAG_MSS;
293  opts->mss = tc->mss;
294  len += TCP_OPTION_LEN_MSS;
295 
296  if (tcp_opts_wscale (&tc->rcv_opts))
297  {
298  opts->flags |= TCP_OPTS_FLAG_WSCALE;
299  opts->wscale = tc->rcv_wscale;
301  }
302 
303  if (tcp_opts_tstamp (&tc->rcv_opts))
304  {
305  opts->flags |= TCP_OPTS_FLAG_TSTAMP;
306  opts->tsval = tcp_time_now ();
307  opts->tsecr = tc->tsval_recent;
309  }
310 
311  if (tcp_opts_sack_permitted (&tc->rcv_opts))
312  {
313  opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED;
315  }
316 
317  /* Align to needed boundary */
318  len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
319  return len;
320 }
321 
322 always_inline int
324 {
325  u8 len = 0;
326 
327  opts->flags = 0;
328 
329  if (tcp_opts_tstamp (&tc->rcv_opts))
330  {
331  opts->flags |= TCP_OPTS_FLAG_TSTAMP;
332  opts->tsval = tcp_time_now ();
333  opts->tsecr = tc->tsval_recent;
335  }
336  if (tcp_opts_sack_permitted (&tc->rcv_opts))
337  {
338  if (vec_len (tc->snd_sacks))
339  {
340  opts->flags |= TCP_OPTS_FLAG_SACK;
341  opts->sacks = tc->snd_sacks;
342  opts->n_sack_blocks = clib_min (vec_len (tc->snd_sacks),
344  len += 2 + TCP_OPTION_LEN_SACK_BLOCK * opts->n_sack_blocks;
345  }
346  }
347 
348  /* Align to needed boundary */
349  len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
350  return len;
351 }
352 
353 always_inline int
356 {
357  switch (state)
358  {
359  case TCP_STATE_ESTABLISHED:
360  case TCP_STATE_FIN_WAIT_1:
361  return tcp_make_established_options (tc, opts);
362  case TCP_STATE_SYN_RCVD:
363  return tcp_make_synack_options (tc, opts);
364  case TCP_STATE_SYN_SENT:
365  return tcp_make_syn_options (opts, tc->rcv_wscale);
366  default:
367  clib_warning ("Not handled!");
368  return 0;
369  }
370 }
371 
372 /**
373  * Update max segment size we're able to process.
374  *
375  * The value is constrained by our interface's MTU and IP options. It is
376  * also what we advertise to our peer.
377  */
378 void
380 {
381  /* TODO find our iface MTU */
382  tc->mss = dummy_mtu;
383 }
384 
385 /**
386  * Update snd_mss to reflect the effective segment size that we can send
387  * by taking into account all TCP options, including SACKs
388  */
389 void
391 {
392  /* Compute options to be used for connection. These may be reused when
393  * sending data or to compute the effective mss (snd_mss) */
394  tc->snd_opts_len =
395  tcp_make_options (tc, &tc->snd_opts, TCP_STATE_ESTABLISHED);
396 
397  /* XXX check if MTU has been updated */
398  tc->snd_mss = clib_min (tc->mss, tc->rcv_opts.mss) - tc->snd_opts_len;
399  ASSERT (tc->snd_mss > 0);
400 }
401 
402 void
404 {
405  u16 default_min_mss = 536;
406  tcp_update_rcv_mss (tc);
407 
408  /* TODO cache mss and consider PMTU discovery */
409  tc->snd_mss = clib_min (tc->rcv_opts.mss, tc->mss);
410 
411  if (tc->snd_mss < 45)
412  {
413  clib_warning ("snd mss is 0");
414  /* Assume that at least the min default mss works */
415  tc->snd_mss = default_min_mss;
416  tc->rcv_opts.mss = default_min_mss;
417  }
418 
419  /* We should have enough space for 40 bytes of options */
420  ASSERT (tc->snd_mss > 45);
421 
422  /* If we use timestamp option, account for it */
423  if (tcp_opts_tstamp (&tc->rcv_opts))
424  tc->snd_mss -= TCP_OPTION_LEN_TIMESTAMP;
425 }
426 
427 #define tcp_get_free_buffer_index(tm, bidx) \
428 do { \
429  u32 *my_tx_buffers, n_free_buffers; \
430  u32 thread_index = vlib_get_thread_index(); \
431  my_tx_buffers = tm->tx_buffers[thread_index]; \
432  if (PREDICT_FALSE(vec_len (my_tx_buffers) == 0)) \
433  { \
434  n_free_buffers = 32; /* TODO config or macro */ \
435  vec_validate (my_tx_buffers, n_free_buffers - 1); \
436  _vec_len(my_tx_buffers) = vlib_buffer_alloc_from_free_list ( \
437  tm->vlib_main, my_tx_buffers, n_free_buffers, \
438  VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX); \
439  tm->tx_buffers[thread_index] = my_tx_buffers; \
440  } \
441  /* buffer shortage */ \
442  if (PREDICT_FALSE (vec_len (my_tx_buffers) == 0)) \
443  return; \
444  *bidx = my_tx_buffers[_vec_len (my_tx_buffers)-1]; \
445  _vec_len (my_tx_buffers) -= 1; \
446 } while (0)
447 
448 #define tcp_return_buffer(tm) \
449 do { \
450  u32 *my_tx_buffers; \
451  u32 thread_index = vlib_get_thread_index(); \
452  my_tx_buffers = tm->tx_buffers[thread_index]; \
453  _vec_len (my_tx_buffers) +=1; \
454 } while (0)
455 
456 always_inline void
458 {
459  vlib_buffer_t *it = b;
460  do
461  {
462  it->current_data = 0;
463  it->current_length = 0;
465  }
466  while ((it->flags & VLIB_BUFFER_NEXT_PRESENT)
467  && (it = vlib_get_buffer (vm, it->next_buffer)));
468 
469  /* Leave enough space for headers */
471  vnet_buffer (b)->tcp.flags = 0;
472 }
473 
474 /**
475  * Prepare ACK
476  */
477 void
479  u8 flags)
480 {
481  tcp_options_t _snd_opts, *snd_opts = &_snd_opts;
482  u8 tcp_opts_len, tcp_hdr_opts_len;
483  tcp_header_t *th;
484  u16 wnd;
485 
486  wnd = tcp_window_to_advertise (tc, state);
487 
488  /* Make and write options */
489  tcp_opts_len = tcp_make_established_options (tc, snd_opts);
490  tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
491 
492  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt,
493  tc->rcv_nxt, tcp_hdr_opts_len, flags, wnd);
494 
495  tcp_options_write ((u8 *) (th + 1), snd_opts);
496  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
497 }
498 
499 /**
500  * Convert buffer to ACK
501  */
502 void
504 {
505  vlib_main_t *vm = vlib_get_main ();
506 
507  tcp_reuse_buffer (vm, b);
508  tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, TCP_FLAG_ACK);
509  TCP_EVT_DBG (TCP_EVT_ACK_SENT, tc);
510  vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_ACK;
511  tc->rcv_las = tc->rcv_nxt;
512 }
513 
514 /**
515  * Convert buffer to FIN-ACK
516  */
517 void
519 {
520  vlib_main_t *vm = vlib_get_main ();
521  u8 flags = 0;
522 
523  tcp_reuse_buffer (vm, b);
524 
525  flags = TCP_FLAG_FIN | TCP_FLAG_ACK;
526  tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, flags);
527 
528  /* Reset flags, make sure ack is sent */
529  vnet_buffer (b)->tcp.flags &= ~TCP_BUF_FLAG_DUPACK;
530 
531  tc->snd_nxt += 1;
532 }
533 
534 /**
535  * Convert buffer to SYN-ACK
536  */
537 void
539 {
540  vlib_main_t *vm = vlib_get_main ();
541  tcp_options_t _snd_opts, *snd_opts = &_snd_opts;
542  u8 tcp_opts_len, tcp_hdr_opts_len;
543  tcp_header_t *th;
544  u16 initial_wnd;
545  u32 time_now;
546 
547  memset (snd_opts, 0, sizeof (*snd_opts));
548 
549  tcp_reuse_buffer (vm, b);
550 
551  /* Set random initial sequence */
552  time_now = tcp_time_now ();
553 
554  tc->iss = random_u32 (&time_now);
555  tc->snd_una = tc->iss;
556  tc->snd_nxt = tc->iss + 1;
557  tc->snd_una_max = tc->snd_nxt;
558 
559  initial_wnd = tcp_initial_window_to_advertise (tc);
560 
561  /* Make and write options */
562  tcp_opts_len = tcp_make_synack_options (tc, snd_opts);
563  tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
564 
565  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss,
566  tc->rcv_nxt, tcp_hdr_opts_len,
567  TCP_FLAG_SYN | TCP_FLAG_ACK, initial_wnd);
568 
569  tcp_options_write ((u8 *) (th + 1), snd_opts);
570 
571  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
572  vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_ACK;
573 
574  /* Init retransmit timer */
576 }
577 
578 always_inline void
580  u8 is_ip4)
581 {
582  u32 *to_next, next_index;
583  vlib_frame_t *f;
584 
586  b->error = 0;
587 
588  /* Default FIB for now */
589  vnet_buffer (b)->sw_if_index[VLIB_TX] = 0;
590 
591  /* Send to IP lookup */
592  next_index = is_ip4 ? ip4_lookup_node.index : ip6_lookup_node.index;
593  f = vlib_get_frame_to_node (vm, next_index);
594 
595  /* Enqueue the packet */
596  to_next = vlib_frame_vector_args (f);
597  to_next[0] = bi;
598  f->n_vectors = 1;
599  vlib_put_frame_to_node (vm, next_index, f);
600 }
601 
602 int
604  tcp_state_t state, u8 thread_index, u8 is_ip4)
605 {
606  ip4_header_t *ih4;
607  ip6_header_t *ih6;
608  tcp_header_t *th0;
609  ip4_address_t src_ip40, dst_ip40;
610  ip6_address_t src_ip60, dst_ip60;
611  u16 src_port, dst_port;
612  u32 tmp;
613  u32 seq, ack;
614  u8 flags;
615 
616  /* Find IP and TCP headers */
617  th0 = tcp_buffer_hdr (b0);
618 
619  /* Save src and dst ip */
620  if (is_ip4)
621  {
622  ih4 = vlib_buffer_get_current (b0);
623  ASSERT ((ih4->ip_version_and_header_length & 0xF0) == 0x40);
624  src_ip40.as_u32 = ih4->src_address.as_u32;
625  dst_ip40.as_u32 = ih4->dst_address.as_u32;
626  }
627  else
628  {
629  ih6 = vlib_buffer_get_current (b0);
630  ASSERT ((ih6->ip_version_traffic_class_and_flow_label & 0xF0) == 0x60);
631  clib_memcpy (&src_ip60, &ih6->src_address, sizeof (ip6_address_t));
632  clib_memcpy (&dst_ip60, &ih6->dst_address, sizeof (ip6_address_t));
633  }
634 
635  src_port = th0->src_port;
636  dst_port = th0->dst_port;
637 
638  /* Try to determine what/why we're actually resetting */
639  if (state == TCP_STATE_CLOSED)
640  {
641  if (!tcp_syn (th0))
642  return -1;
643 
644  tmp = clib_net_to_host_u32 (th0->seq_number);
645 
646  /* Got a SYN for no listener. */
647  flags = TCP_FLAG_RST | TCP_FLAG_ACK;
648  ack = clib_host_to_net_u32 (tmp + 1);
649  seq = 0;
650  }
651  else
652  {
653  flags = TCP_FLAG_RST;
654  seq = th0->ack_number;
655  ack = 0;
656  }
657 
658  tcp_reuse_buffer (vm, b0);
659  th0 = vlib_buffer_push_tcp_net_order (b0, dst_port, src_port, seq, ack,
660  sizeof (tcp_header_t), flags, 0);
661 
662  if (is_ip4)
663  {
664  ih4 = vlib_buffer_push_ip4 (vm, b0, &dst_ip40, &src_ip40,
665  IP_PROTOCOL_TCP);
666  th0->checksum = ip4_tcp_udp_compute_checksum (vm, b0, ih4);
667  }
668  else
669  {
670  int bogus = ~0;
671  ih6 = vlib_buffer_push_ip6 (vm, b0, &dst_ip60, &src_ip60,
672  IP_PROTOCOL_TCP);
673  th0->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b0, ih6, &bogus);
674  ASSERT (!bogus);
675  }
676 
677  return 0;
678 }
679 
680 /**
681  * Send reset without reusing existing buffer
682  */
683 void
685 {
686  vlib_buffer_t *b;
687  u32 bi;
688  tcp_main_t *tm = vnet_get_tcp_main ();
689  vlib_main_t *vm = vlib_get_main ();
690  u8 tcp_hdr_len, flags = 0;
691  tcp_header_t *th, *pkt_th;
692  u32 seq, ack;
693  ip4_header_t *ih4, *pkt_ih4;
694  ip6_header_t *ih6, *pkt_ih6;
695 
696  tcp_get_free_buffer_index (tm, &bi);
697  b = vlib_get_buffer (vm, bi);
698 
699  /* Leave enough space for headers */
701 
702  /* Make and write options */
703  tcp_hdr_len = sizeof (tcp_header_t);
704 
705  if (is_ip4)
706  {
707  pkt_ih4 = vlib_buffer_get_current (pkt);
708  pkt_th = ip4_next_header (pkt_ih4);
709  }
710  else
711  {
712  pkt_ih6 = vlib_buffer_get_current (pkt);
713  pkt_th = ip6_next_header (pkt_ih6);
714  }
715 
716  if (tcp_ack (pkt_th))
717  {
718  flags = TCP_FLAG_RST;
719  seq = pkt_th->ack_number;
720  ack = 0;
721  }
722  else
723  {
724  flags = TCP_FLAG_RST | TCP_FLAG_ACK;
725  seq = 0;
726  ack = clib_host_to_net_u32 (vnet_buffer (pkt)->tcp.seq_end);
727  }
728 
729  th = vlib_buffer_push_tcp_net_order (b, pkt_th->dst_port, pkt_th->src_port,
730  seq, ack, tcp_hdr_len, flags, 0);
731 
732  /* Swap src and dst ip */
733  if (is_ip4)
734  {
735  ASSERT ((pkt_ih4->ip_version_and_header_length & 0xF0) == 0x40);
736  ih4 = vlib_buffer_push_ip4 (vm, b, &pkt_ih4->dst_address,
737  &pkt_ih4->src_address, IP_PROTOCOL_TCP);
738  th->checksum = ip4_tcp_udp_compute_checksum (vm, b, ih4);
739  }
740  else
741  {
742  int bogus = ~0;
743  pkt_ih6 = (ip6_header_t *) (pkt_th - 1);
744  ASSERT ((pkt_ih6->ip_version_traffic_class_and_flow_label & 0xF0) ==
745  0x60);
746  ih6 =
747  vlib_buffer_push_ip6 (vm, b, &pkt_ih6->dst_address,
748  &pkt_ih6->src_address, IP_PROTOCOL_TCP);
749  th->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b, ih6, &bogus);
750  ASSERT (!bogus);
751  }
752 
753  tcp_enqueue_to_ip_lookup (vm, b, bi, is_ip4);
754 }
755 
756 void
758 {
760 
761  if (tc->c_is_ip4)
762  {
763  ip4_header_t *ih;
764  ih = vlib_buffer_push_ip4 (tm->vlib_main, b, &tc->c_lcl_ip4,
765  &tc->c_rmt_ip4, IP_PROTOCOL_TCP);
766  th->checksum = ip4_tcp_udp_compute_checksum (tm->vlib_main, b, ih);
767  }
768  else
769  {
770  ip6_header_t *ih;
771  int bogus = ~0;
772 
773  ih = vlib_buffer_push_ip6 (tm->vlib_main, b, &tc->c_lcl_ip6,
774  &tc->c_rmt_ip6, IP_PROTOCOL_TCP);
775  th->checksum = ip6_tcp_udp_icmp_compute_checksum (tm->vlib_main, b, ih,
776  &bogus);
777  ASSERT (!bogus);
778  }
779 }
780 
781 /**
782  * Send SYN
783  *
784  * Builds a SYN packet for a half-open connection and sends it to ipx_lookup.
785  * The packet is not forwarded through tcpx_output to avoid doing lookups
786  * in the half_open pool.
787  */
788 void
790 {
791  vlib_buffer_t *b;
792  u32 bi;
793  tcp_main_t *tm = vnet_get_tcp_main ();
794  vlib_main_t *vm = vlib_get_main ();
795  u8 tcp_hdr_opts_len, tcp_opts_len;
796  tcp_header_t *th;
797  u32 time_now;
798  u16 initial_wnd;
799  tcp_options_t snd_opts;
800 
801  tcp_get_free_buffer_index (tm, &bi);
802  b = vlib_get_buffer (vm, bi);
803 
804  /* Leave enough space for headers */
806 
807  /* Set random initial sequence */
808  time_now = tcp_time_now ();
809 
810  tc->iss = random_u32 (&time_now);
811  tc->snd_una = tc->iss;
812  tc->snd_una_max = tc->snd_nxt = tc->iss + 1;
813 
814  initial_wnd = tcp_initial_window_to_advertise (tc);
815 
816  /* Make and write options */
817  memset (&snd_opts, 0, sizeof (snd_opts));
818  tcp_opts_len = tcp_make_syn_options (&snd_opts, tc->rcv_wscale);
819  tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
820 
821  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss,
822  tc->rcv_nxt, tcp_hdr_opts_len, TCP_FLAG_SYN,
823  initial_wnd);
824 
825  tcp_options_write ((u8 *) (th + 1), &snd_opts);
826 
827  /* Measure RTT with this */
828  tc->rtt_ts = tcp_time_now ();
829  tc->rtt_seq = tc->snd_nxt;
830 
831  /* Start retransmit trimer */
832  tcp_timer_set (tc, TCP_TIMER_RETRANSMIT_SYN, tc->rto * TCP_TO_TIMER_TICK);
833  tc->rto_boff = 0;
834 
835  /* Set the connection establishment timer */
836  tcp_timer_set (tc, TCP_TIMER_ESTABLISH, TCP_ESTABLISH_TIME);
837 
838  tcp_push_ip_hdr (tm, tc, b);
839  tcp_enqueue_to_ip_lookup (vm, b, bi, tc->c_is_ip4);
840 }
841 
842 always_inline void
844 {
845  u32 *to_next, next_index;
846  vlib_frame_t *f;
847 
849  b->error = 0;
850 
851  /* Decide where to send the packet */
852  next_index = is_ip4 ? tcp4_output_node.index : tcp6_output_node.index;
853 
854  /* Enqueue the packet */
855  f = vlib_get_frame_to_node (vm, next_index);
856  to_next = vlib_frame_vector_args (f);
857  to_next[0] = bi;
858  f->n_vectors = 1;
859  vlib_put_frame_to_node (vm, next_index, f);
860 }
861 
862 /**
863  * Send FIN
864  */
865 void
867 {
868  vlib_buffer_t *b;
869  u32 bi;
870  tcp_main_t *tm = vnet_get_tcp_main ();
871  vlib_main_t *vm = vlib_get_main ();
872 
873  tcp_get_free_buffer_index (tm, &bi);
874  b = vlib_get_buffer (vm, bi);
875 
876  /* Leave enough space for headers */
878 
879  tcp_make_fin (tc, b);
880  tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4);
881  tc->flags |= TCP_CONN_FINSNT;
883  TCP_EVT_DBG (TCP_EVT_FIN_SENT, tc);
884 }
885 
888 {
889  switch (next_state)
890  {
891  case TCP_STATE_ESTABLISHED:
892  return TCP_FLAG_ACK;
893  case TCP_STATE_SYN_RCVD:
894  return TCP_FLAG_SYN | TCP_FLAG_ACK;
895  case TCP_STATE_SYN_SENT:
896  return TCP_FLAG_SYN;
897  case TCP_STATE_LAST_ACK:
898  case TCP_STATE_FIN_WAIT_1:
899  return TCP_FLAG_FIN;
900  default:
901  clib_warning ("Shouldn't be here!");
902  }
903  return 0;
904 }
905 
906 /**
907  * Push TCP header and update connection variables
908  */
909 static void
911  tcp_state_t next_state, u8 compute_opts)
912 {
913  u32 advertise_wnd, data_len;
914  u8 tcp_hdr_opts_len, opts_write_len, flags;
915  tcp_header_t *th;
916 
918  vnet_buffer (b)->tcp.flags = 0;
919 
920  if (compute_opts)
921  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
922 
923  tcp_hdr_opts_len = tc->snd_opts_len + sizeof (tcp_header_t);
924  advertise_wnd = tcp_window_to_advertise (tc, next_state);
925  flags = tcp_make_state_flags (next_state);
926 
927  /* Push header and options */
928  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt,
929  tc->rcv_nxt, tcp_hdr_opts_len, flags,
930  advertise_wnd);
931  opts_write_len = tcp_options_write ((u8 *) (th + 1), &tc->snd_opts);
932 
933  ASSERT (opts_write_len == tc->snd_opts_len);
934  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
935 
936  /*
937  * Update connection variables
938  */
939 
940  tc->snd_nxt += data_len;
941  tc->rcv_las = tc->rcv_nxt;
942 
943  /* TODO this is updated in output as well ... */
944  if (seq_gt (tc->snd_nxt, tc->snd_una_max))
945  {
946  tc->snd_una_max = tc->snd_nxt;
947  tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una);
948  }
949 
950  TCP_EVT_DBG (TCP_EVT_PKTIZE, tc);
951 }
952 
953 void
955 {
956  tcp_main_t *tm = vnet_get_tcp_main ();
957  vlib_main_t *vm = vlib_get_main ();
958 
959  vlib_buffer_t *b;
960  u32 bi;
961 
962  /* Get buffer */
963  tcp_get_free_buffer_index (tm, &bi);
964  b = vlib_get_buffer (vm, bi);
965 
966  /* Fill in the ACK */
967  tcp_make_ack (tc, b);
968  tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4);
969 }
970 
971 /* Send delayed ACK when timer expires */
972 void
974 {
975  u32 thread_index = vlib_get_thread_index ();
976  tcp_connection_t *tc;
977 
978  tc = tcp_connection_get (index, thread_index);
979  tc->timers[TCP_TIMER_DELACK] = TCP_TIMER_HANDLE_INVALID;
980  tcp_send_ack (tc);
981 }
982 
983 /** Build a retransmit segment
984  *
985  * @return the number of bytes in the segment or 0 if there's nothing to
986  * retransmit
987  */
988 u32
990  u32 offset, u32 max_bytes)
991 {
992  vlib_main_t *vm = vlib_get_main ();
993  int n_bytes = 0;
994  u32 start;
995 
996  tcp_reuse_buffer (vm, b);
997 
998  ASSERT (tc->state >= TCP_STATE_ESTABLISHED);
999  ASSERT (max_bytes != 0);
1000 
1001  max_bytes = clib_min (tc->snd_mss, max_bytes);
1002  start = tc->snd_una + offset;
1003 
1004  /* Start is beyond snd_congestion */
1005  if (seq_geq (start, tc->snd_congestion))
1006  goto done;
1007 
1008  /* Don't overshoot snd_congestion */
1009  if (seq_gt (start + max_bytes, tc->snd_congestion))
1010  {
1011  max_bytes = tc->snd_congestion - start;
1012  if (max_bytes == 0)
1013  goto done;
1014  }
1015 
1016  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
1017 
1018  ASSERT (max_bytes <= tc->snd_mss);
1019 
1020  n_bytes = stream_session_peek_bytes (&tc->connection,
1021  vlib_buffer_get_current (b), offset,
1022  max_bytes);
1023  ASSERT (n_bytes > 0);
1024  b->current_length = n_bytes;
1025  tcp_push_hdr_i (tc, b, tc->state, 0);
1026 
1027  if (tcp_in_fastrecovery (tc))
1028  tc->snd_rxt_bytes += n_bytes;
1029 
1030 done:
1031  TCP_EVT_DBG (TCP_EVT_CC_RTX, tc, offset, n_bytes);
1032  return n_bytes;
1033 }
1034 
1035 /**
1036  * Reset congestion control, switch cwnd to loss window and try again.
1037  */
1038 static void
1040 {
1041  tc->prev_ssthresh = tc->ssthresh;
1042  tc->prev_cwnd = tc->cwnd;
1043 
1044  /* Cleanly recover cc (also clears up fast retransmit) */
1045  if (tcp_in_fastrecovery (tc))
1047 
1048  /* Start again from the beginning */
1049  tc->ssthresh = clib_max (tcp_flight_size (tc) / 2, 2 * tc->snd_mss);
1050  tc->cwnd = tcp_loss_wnd (tc);
1051  tc->snd_congestion = tc->snd_una_max;
1052 
1053  tcp_recovery_on (tc);
1054 }
1055 
1056 static void
1058 {
1059  tcp_main_t *tm = vnet_get_tcp_main ();
1060  vlib_main_t *vm = vlib_get_main ();
1061  u32 thread_index = vlib_get_thread_index ();
1062  tcp_connection_t *tc;
1063  vlib_buffer_t *b;
1064  u32 bi, n_bytes;
1065 
1066  if (is_syn)
1067  {
1068  tc = tcp_half_open_connection_get (index);
1069  }
1070  else
1071  {
1072  tc = tcp_connection_get (index, thread_index);
1073  }
1074 
1075  /* Make sure timer handle is set to invalid */
1076  tc->timers[TCP_TIMER_RETRANSMIT] = TCP_TIMER_HANDLE_INVALID;
1077 
1078  if (!tcp_in_recovery (tc) && tc->rto_boff > 0
1079  && tc->state >= TCP_STATE_ESTABLISHED)
1080  {
1081  tc->rto_boff = 0;
1082  tcp_update_rto (tc);
1083  }
1084 
1085  /* Increment RTO backoff (also equal to number of retries) */
1086  tc->rto_boff += 1;
1087 
1088  /* Go back to first un-acked byte */
1089  tc->snd_nxt = tc->snd_una;
1090 
1091  tcp_get_free_buffer_index (tm, &bi);
1092  b = vlib_get_buffer (vm, bi);
1093 
1094  if (tc->state >= TCP_STATE_ESTABLISHED)
1095  {
1096  /* Lost FIN, retransmit and return */
1097  if (tc->flags & TCP_CONN_FINSNT)
1098  {
1099  tcp_send_fin (tc);
1100  return;
1101  }
1102 
1103  /* First retransmit timeout */
1104  if (tc->rto_boff == 1)
1105  tcp_rtx_timeout_cc (tc);
1106 
1107  /* Exponential backoff */
1108  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1109 
1110  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 1);
1111 
1112  /* Send one segment */
1113  n_bytes = tcp_prepare_retransmit_segment (tc, b, 0, tc->snd_mss);
1114  /* TODO be less aggressive about this */
1115  scoreboard_clear (&tc->sack_sb);
1116 
1117  if (n_bytes == 0)
1118  {
1119  clib_warning ("could not retransmit anything");
1120  clib_warning ("%U", format_tcp_connection, tc, 2);
1121 
1122  /* Try again eventually */
1124  ASSERT (0 || (tc->rto_boff > 1
1125  && tc->snd_una == tc->snd_congestion));
1126  return;
1127  }
1128 
1129  /* For first retransmit, record timestamp (Eifel detection RFC3522) */
1130  if (tc->rto_boff == 1)
1131  tc->snd_rxt_ts = tcp_time_now ();
1132  }
1133  /* Retransmit for SYN/SYNACK */
1134  else if (tc->state == TCP_STATE_SYN_RCVD || tc->state == TCP_STATE_SYN_SENT)
1135  {
1136  /* Try without increasing RTO a number of times. If this fails,
1137  * start growing RTO exponentially */
1138  if (tc->rto_boff > TCP_RTO_SYN_RETRIES)
1139  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1140 
1142 
1143  tcp_push_hdr_i (tc, b, tc->state, 1);
1144 
1145  /* Account for the SYN */
1146  tc->snd_nxt += 1;
1147  }
1148  else
1149  {
1150  ASSERT (tc->state == TCP_STATE_CLOSED);
1151  clib_warning ("connection closed ...");
1152  return;
1153  }
1154 
1155  if (!is_syn)
1156  {
1157  tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4);
1158 
1159  /* Re-enable retransmit timer */
1161  }
1162  else
1163  {
1164  ASSERT (tc->state == TCP_STATE_SYN_SENT);
1165 
1166  TCP_EVT_DBG (TCP_EVT_SYN_RTX, tc);
1167 
1168  /* This goes straight to ipx_lookup */
1169  tcp_push_ip_hdr (tm, tc, b);
1170  tcp_enqueue_to_ip_lookup (vm, b, bi, tc->c_is_ip4);
1171 
1172  /* Re-enable retransmit timer */
1173  tcp_timer_set (tc, TCP_TIMER_RETRANSMIT_SYN,
1174  tc->rto * TCP_TO_TIMER_TICK);
1175  }
1176 }
1177 
1178 void
1180 {
1181  tcp_timer_retransmit_handler_i (index, 0);
1182 }
1183 
1184 void
1186 {
1187  tcp_timer_retransmit_handler_i (index, 1);
1188 }
1189 
1190 /**
1191  * Got 0 snd_wnd from peer, try to do something about it.
1192  *
1193  */
1194 void
1196 {
1197  tcp_main_t *tm = vnet_get_tcp_main ();
1198  vlib_main_t *vm = vlib_get_main ();
1199  u32 thread_index = vlib_get_thread_index ();
1200  tcp_connection_t *tc;
1201  vlib_buffer_t *b;
1202  u32 bi, old_snd_nxt;
1203  int n_bytes = 0;
1204 
1205  tc = tcp_connection_get_if_valid (index, thread_index);
1206 
1207  if (!tc)
1208  return;
1209 
1210  /* Make sure timer handle is set to invalid */
1211  tc->timers[TCP_TIMER_PERSIST] = TCP_TIMER_HANDLE_INVALID;
1212 
1213  /* Problem already solved or worse */
1214  if (tc->state == TCP_STATE_CLOSED || tc->state > TCP_STATE_ESTABLISHED
1215  || tc->snd_wnd > tc->snd_mss || tcp_in_recovery (tc))
1216  return;
1217 
1218  /* Increment RTO backoff */
1219  tc->rto_boff += 1;
1220  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1221 
1222  /* Try to force the first unsent segment */
1223  tcp_get_free_buffer_index (tm, &bi);
1224  b = vlib_get_buffer (vm, bi);
1225 
1226  tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una);
1227  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
1228  n_bytes = stream_session_peek_bytes (&tc->connection,
1230  tc->snd_una_max - tc->snd_una,
1231  tc->snd_mss);
1232  /* Nothing to send */
1233  if (n_bytes <= 0)
1234  {
1235  clib_warning ("persist found nothing to send");
1236  tcp_return_buffer (tm);
1237  return;
1238  }
1239 
1240  b->current_length = n_bytes;
1241  ASSERT (tc->snd_nxt == tc->snd_una_max || tc->rto_boff > 1
1242  || tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT));
1243 
1244  /* Allow updating of snd_una_max but don't update snd_nxt */
1245  old_snd_nxt = tc->snd_nxt;
1246  tcp_push_hdr_i (tc, b, tc->state, 0);
1247  tc->snd_nxt = old_snd_nxt;
1248  tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4);
1249 
1250  /* Re-enable persist timer */
1251  tcp_persist_timer_set (tc);
1252 }
1253 
1254 /**
1255  * Retransmit first unacked segment
1256  */
1257 void
1259 {
1260  tcp_main_t *tm = vnet_get_tcp_main ();
1261  vlib_main_t *vm = vlib_get_main ();
1262  vlib_buffer_t *b;
1263  u32 bi, n_bytes, old_snd_nxt;
1264 
1265  old_snd_nxt = tc->snd_nxt;
1266  tc->snd_nxt = tc->snd_una;
1267 
1268  /* Get buffer */
1269  tcp_get_free_buffer_index (tm, &bi);
1270  b = vlib_get_buffer (vm, bi);
1271 
1272  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 2);
1273 
1274  n_bytes = tcp_prepare_retransmit_segment (tc, b, 0, tc->snd_mss);
1275  if (n_bytes == 0)
1276  {
1277  tcp_return_buffer (tm);
1278  goto done;
1279  }
1280 
1281  tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4);
1282 
1283 done:
1284  tc->snd_nxt = old_snd_nxt;
1285 }
1286 
1287 /**
1288  * Do fast retransmit with SACKs
1289  */
1290 void
1292 {
1293  tcp_main_t *tm = vnet_get_tcp_main ();
1294  vlib_main_t *vm = vlib_get_main ();
1295  u32 n_written = 0, offset = 0, max_bytes;
1296  vlib_buffer_t *b;
1297  sack_scoreboard_hole_t *hole;
1298  sack_scoreboard_t *sb;
1299  u32 bi, old_snd_nxt;
1300  int snd_space;
1301  u8 snd_limited = 0, can_rescue = 0;
1302 
1303  ASSERT (tcp_in_fastrecovery (tc));
1304  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 0);
1305 
1306  old_snd_nxt = tc->snd_nxt;
1307  sb = &tc->sack_sb;
1308  snd_space = tcp_available_snd_space (tc);
1309 
1310  hole = scoreboard_get_hole (sb, sb->cur_rxt_hole);
1311  while (hole && snd_space > 0)
1312  {
1313  tcp_get_free_buffer_index (tm, &bi);
1314  b = vlib_get_buffer (vm, bi);
1315 
1316  hole = scoreboard_next_rxt_hole (sb, hole,
1318  &can_rescue, &snd_limited);
1319  if (!hole)
1320  {
1321  if (!can_rescue || !(seq_lt (sb->rescue_rxt, tc->snd_una)
1322  || seq_gt (sb->rescue_rxt,
1323  tc->snd_congestion)))
1324  break;
1325 
1326  /* If rescue rxt undefined or less than snd_una then one segment of
1327  * up to SMSS octets that MUST include the highest outstanding
1328  * unSACKed sequence number SHOULD be returned, and RescueRxt set to
1329  * RecoveryPoint. HighRxt MUST NOT be updated.
1330  */
1331  max_bytes = clib_min (tc->snd_mss, snd_space);
1332  offset = tc->snd_congestion - tc->snd_una - max_bytes;
1333  sb->rescue_rxt = tc->snd_congestion;
1334  tc->snd_nxt = tc->snd_una + offset;
1335  tcp_prepare_retransmit_segment (tc, b, offset, max_bytes);
1336  tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4);
1337  break;
1338  }
1339 
1340  max_bytes = snd_limited ? tc->snd_mss : hole->end - sb->high_rxt;
1341  offset = sb->high_rxt - tc->snd_una;
1342  tc->snd_nxt = tc->snd_una + offset;
1343  n_written = tcp_prepare_retransmit_segment (tc, b, offset, max_bytes);
1344 
1345  /* Nothing left to retransmit */
1346  if (n_written == 0)
1347  {
1348  tcp_return_buffer (tm);
1349  break;
1350  }
1351 
1352  sb->high_rxt += n_written;
1353  tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4);
1354  snd_space -= n_written;
1355  }
1356 
1357  /* If window allows, send 1 SMSS of new data */
1358  tc->snd_nxt = old_snd_nxt;
1359 }
1360 
1361 /**
1362  * Fast retransmit without SACK info
1363  */
1364 void
1366 {
1367  tcp_main_t *tm = vnet_get_tcp_main ();
1368  vlib_main_t *vm = vlib_get_main ();
1369  u32 n_written = 0, offset = 0, bi, old_snd_nxt;
1370  int snd_space;
1371  vlib_buffer_t *b;
1372 
1373  ASSERT (tcp_in_fastrecovery (tc));
1374  TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 0);
1375 
1376  /* Start resending from first un-acked segment */
1377  old_snd_nxt = tc->snd_nxt;
1378  tc->snd_nxt = tc->snd_una;
1379  snd_space = tcp_available_snd_space (tc);
1380 
1381  while (snd_space > 0)
1382  {
1383  tcp_get_free_buffer_index (tm, &bi);
1384  b = vlib_get_buffer (vm, bi);
1385 
1386  offset += n_written;
1387  n_written = tcp_prepare_retransmit_segment (tc, b, offset, snd_space);
1388 
1389  /* Nothing left to retransmit */
1390  if (n_written == 0)
1391  {
1392  tcp_return_buffer (tm);
1393  break;
1394  }
1395 
1396  tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4);
1397  snd_space -= n_written;
1398  }
1399 
1400  /* Restore snd_nxt. If window allows, send 1 SMSS of new data */
1401  tc->snd_nxt = old_snd_nxt;
1402 }
1403 
1404 /**
1405  * Do fast retransmit
1406  */
1407 void
1409 {
1410  if (tcp_opts_sack_permitted (&tc->rcv_opts)
1411  && scoreboard_first_hole (&tc->sack_sb))
1413  else
1415 }
1416 
1419 {
1420  stream_session_t *s =
1421  stream_session_get (tc->c_s_index, tc->c_thread_index);
1422  return svm_fifo_has_ooo_data (s->server_rx_fifo);
1423 }
1424 
1427  vlib_node_runtime_t * node,
1428  vlib_frame_t * from_frame, int is_ip4)
1429 {
1430  u32 n_left_from, next_index, *from, *to_next;
1431  u32 my_thread_index = vm->thread_index;
1432 
1433  from = vlib_frame_vector_args (from_frame);
1434  n_left_from = from_frame->n_vectors;
1435 
1436  next_index = node->cached_next_index;
1437 
1438  while (n_left_from > 0)
1439  {
1440  u32 n_left_to_next;
1441 
1442  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
1443 
1444  while (n_left_from > 0 && n_left_to_next > 0)
1445  {
1446  u32 bi0;
1447  vlib_buffer_t *b0;
1448  tcp_connection_t *tc0;
1449  tcp_tx_trace_t *t0;
1450  tcp_header_t *th0 = 0;
1451  u32 error0 = TCP_ERROR_PKTS_SENT, next0 = TCP_OUTPUT_NEXT_DROP;
1452 
1453  bi0 = from[0];
1454  to_next[0] = bi0;
1455  from += 1;
1456  to_next += 1;
1457  n_left_from -= 1;
1458  n_left_to_next -= 1;
1459 
1460  b0 = vlib_get_buffer (vm, bi0);
1461  tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
1462  my_thread_index);
1463  if (PREDICT_FALSE (tc0 == 0 || tc0->state == TCP_STATE_CLOSED))
1464  {
1465  error0 = TCP_ERROR_INVALID_CONNECTION;
1466  next0 = TCP_OUTPUT_NEXT_DROP;
1467  goto done;
1468  }
1469 
1470  th0 = vlib_buffer_get_current (b0);
1471  TCP_EVT_DBG (TCP_EVT_OUTPUT, tc0, th0->flags, b0->current_length);
1472 
1473  if (is_ip4)
1474  {
1475  ip4_header_t *ih0;
1476  ih0 = vlib_buffer_push_ip4 (vm, b0, &tc0->c_lcl_ip4,
1477  &tc0->c_rmt_ip4, IP_PROTOCOL_TCP);
1478  th0->checksum = ip4_tcp_udp_compute_checksum (vm, b0, ih0);
1479  }
1480  else
1481  {
1482  ip6_header_t *ih0;
1483  int bogus = ~0;
1484 
1485  ih0 = vlib_buffer_push_ip6 (vm, b0, &tc0->c_lcl_ip6,
1486  &tc0->c_rmt_ip6, IP_PROTOCOL_TCP);
1487  th0->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b0, ih0,
1488  &bogus);
1489  ASSERT (!bogus);
1490  }
1491 
1492  /* Filter out DUPACKs if there are no OOO segments left */
1493  if (PREDICT_FALSE
1494  (vnet_buffer (b0)->tcp.flags & TCP_BUF_FLAG_DUPACK))
1495  {
1496  if (!tcp_session_has_ooo_data (tc0))
1497  {
1498  error0 = TCP_ERROR_FILTERED_DUPACKS;
1499  next0 = TCP_OUTPUT_NEXT_DROP;
1500  goto done;
1501  }
1502  }
1503 
1504  /* Stop DELACK timer and fix flags */
1505  tc0->flags &= ~(TCP_CONN_SNDACK);
1506  tcp_timer_reset (tc0, TCP_TIMER_DELACK);
1507 
1508  /* If not retransmitting
1509  * 1) update snd_una_max (SYN, SYNACK, FIN)
1510  * 2) If we're not tracking an ACK, start tracking */
1511  if (seq_lt (tc0->snd_una_max, tc0->snd_nxt))
1512  {
1513  tc0->snd_una_max = tc0->snd_nxt;
1514  if (tc0->rtt_ts == 0)
1515  {
1516  tc0->rtt_ts = tcp_time_now ();
1517  tc0->rtt_seq = tc0->snd_nxt;
1518  }
1519  }
1520 
1521  /* Set the retransmit timer if not set already and not
1522  * doing a pure ACK */
1523  if (!tcp_timer_is_active (tc0, TCP_TIMER_RETRANSMIT)
1524  && tc0->snd_nxt != tc0->snd_una)
1525  {
1527  tc0->rto_boff = 0;
1528  }
1529 
1530  /* Make sure we haven't lost route to our peer */
1531  if (PREDICT_FALSE (tc0->last_fib_check
1532  < tc0->snd_opts.tsval + TCP_FIB_RECHECK_PERIOD))
1533  {
1534  if (PREDICT_TRUE
1535  (tc0->c_rmt_fei == tcp_lookup_rmt_in_fib (tc0)))
1536  {
1537  tc0->last_fib_check = tc0->snd_opts.tsval;
1538  }
1539  else
1540  {
1541  clib_warning ("lost connection to peer");
1542  tcp_connection_reset (tc0);
1543  goto done;
1544  }
1545  }
1546 
1547  /* Use pre-computed dpo to set next node */
1548  next0 = tc0->c_rmt_dpo.dpoi_next_node;
1549  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = tc0->c_rmt_dpo.dpoi_index;
1550 
1552  done:
1553  b0->error = node->errors[error0];
1555  {
1556  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
1557  if (th0)
1558  {
1559  clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header));
1560  }
1561  else
1562  {
1563  memset (&t0->tcp_header, 0, sizeof (t0->tcp_header));
1564  }
1565  clib_memcpy (&t0->tcp_connection, tc0,
1566  sizeof (t0->tcp_connection));
1567  }
1568 
1569  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
1570  n_left_to_next, bi0, next0);
1571  }
1572 
1573  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
1574  }
1575 
1576  return from_frame->n_vectors;
1577 }
1578 
1579 static uword
1581  vlib_frame_t * from_frame)
1582 {
1583  return tcp46_output_inline (vm, node, from_frame, 1 /* is_ip4 */ );
1584 }
1585 
1586 static uword
1588  vlib_frame_t * from_frame)
1589 {
1590  return tcp46_output_inline (vm, node, from_frame, 0 /* is_ip4 */ );
1591 }
1592 
1593 /* *INDENT-OFF* */
1595 {
1596  .function = tcp4_output,.name = "tcp4-output",
1597  /* Takes a vector of packets. */
1598  .vector_size = sizeof (u32),
1599  .n_errors = TCP_N_ERROR,
1600  .error_strings = tcp_error_strings,
1601  .n_next_nodes = TCP_OUTPUT_N_NEXT,
1602  .next_nodes = {
1603 #define _(s,n) [TCP_OUTPUT_NEXT_##s] = n,
1605 #undef _
1606  },
1607  .format_buffer = format_tcp_header,
1608  .format_trace = format_tcp_tx_trace,
1609 };
1610 /* *INDENT-ON* */
1611 
1613 
1614 /* *INDENT-OFF* */
1616 {
1617  .function = tcp6_output,
1618  .name = "tcp6-output",
1619  /* Takes a vector of packets. */
1620  .vector_size = sizeof (u32),
1621  .n_errors = TCP_N_ERROR,
1622  .error_strings = tcp_error_strings,
1623  .n_next_nodes = TCP_OUTPUT_N_NEXT,
1624  .next_nodes = {
1625 #define _(s,n) [TCP_OUTPUT_NEXT_##s] = n,
1627 #undef _
1628  },
1629  .format_buffer = format_tcp_header,
1630  .format_trace = format_tcp_tx_trace,
1631 };
1632 /* *INDENT-ON* */
1633 
1635 
1636 u32
1638 {
1639  tcp_connection_t *tc;
1640 
1641  tc = (tcp_connection_t *) tconn;
1642  tcp_push_hdr_i (tc, b, TCP_STATE_ESTABLISHED, 0);
1643 
1644  if (tc->rtt_ts == 0 && !tcp_in_cong_recovery (tc))
1645  {
1646  tc->rtt_ts = tcp_time_now ();
1647  tc->rtt_seq = tc->snd_nxt;
1648  }
1649  return 0;
1650 }
1651 
1652 typedef enum _tcp_reset_next
1653 {
1658 
1659 #define foreach_tcp4_reset_next \
1660  _(DROP, "error-drop") \
1661  _(IP_LOOKUP, "ip4-lookup")
1662 
1663 #define foreach_tcp6_reset_next \
1664  _(DROP, "error-drop") \
1665  _(IP_LOOKUP, "ip6-lookup")
1666 
1667 static uword
1669  vlib_frame_t * from_frame, u8 is_ip4)
1670 {
1671  u32 n_left_from, next_index, *from, *to_next;
1672  u32 my_thread_index = vm->thread_index;
1673 
1674  from = vlib_frame_vector_args (from_frame);
1675  n_left_from = from_frame->n_vectors;
1676 
1677  next_index = node->cached_next_index;
1678 
1679  while (n_left_from > 0)
1680  {
1681  u32 n_left_to_next;
1682 
1683  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
1684 
1685  while (n_left_from > 0 && n_left_to_next > 0)
1686  {
1687  u32 bi0;
1688  vlib_buffer_t *b0;
1689  tcp_tx_trace_t *t0;
1690  tcp_header_t *th0;
1691  u32 error0 = TCP_ERROR_RST_SENT, next0 = TCP_RESET_NEXT_IP_LOOKUP;
1692 
1693  bi0 = from[0];
1694  to_next[0] = bi0;
1695  from += 1;
1696  to_next += 1;
1697  n_left_from -= 1;
1698  n_left_to_next -= 1;
1699 
1700  b0 = vlib_get_buffer (vm, bi0);
1701 
1702  if (tcp_make_reset_in_place (vm, b0, vnet_buffer (b0)->tcp.flags,
1703  my_thread_index, is_ip4))
1704  {
1705  error0 = TCP_ERROR_LOOKUP_DROPS;
1706  next0 = TCP_RESET_NEXT_DROP;
1707  goto done;
1708  }
1709 
1710  /* Prepare to send to IP lookup */
1711  vnet_buffer (b0)->sw_if_index[VLIB_TX] = 0;
1712  next0 = TCP_RESET_NEXT_IP_LOOKUP;
1713 
1714  done:
1715  b0->error = node->errors[error0];
1718  {
1719  th0 = vlib_buffer_get_current (b0);
1720  if (is_ip4)
1721  th0 = ip4_next_header ((ip4_header_t *) th0);
1722  else
1723  th0 = ip6_next_header ((ip6_header_t *) th0);
1724  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
1725  clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header));
1726  }
1727 
1728  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
1729  n_left_to_next, bi0, next0);
1730  }
1731  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
1732  }
1733  return from_frame->n_vectors;
1734 }
1735 
1736 static uword
1738  vlib_frame_t * from_frame)
1739 {
1740  return tcp46_send_reset_inline (vm, node, from_frame, 1);
1741 }
1742 
1743 static uword
1745  vlib_frame_t * from_frame)
1746 {
1747  return tcp46_send_reset_inline (vm, node, from_frame, 0);
1748 }
1749 
1750 /* *INDENT-OFF* */
1752  .function = tcp4_send_reset,
1753  .name = "tcp4-reset",
1754  .vector_size = sizeof (u32),
1755  .n_errors = TCP_N_ERROR,
1756  .error_strings = tcp_error_strings,
1757  .n_next_nodes = TCP_RESET_N_NEXT,
1758  .next_nodes = {
1759 #define _(s,n) [TCP_RESET_NEXT_##s] = n,
1761 #undef _
1762  },
1763  .format_trace = format_tcp_tx_trace,
1764 };
1765 /* *INDENT-ON* */
1766 
1768 
1769 /* *INDENT-OFF* */
1771  .function = tcp6_send_reset,
1772  .name = "tcp6-reset",
1773  .vector_size = sizeof (u32),
1774  .n_errors = TCP_N_ERROR,
1775  .error_strings = tcp_error_strings,
1776  .n_next_nodes = TCP_RESET_N_NEXT,
1777  .next_nodes = {
1778 #define _(s,n) [TCP_RESET_NEXT_##s] = n,
1780 #undef _
1781  },
1782  .format_trace = format_tcp_tx_trace,
1783 };
1784 /* *INDENT-ON* */
1785 
1787 
1788 /*
1789  * fd.io coding-style-patch-verification: ON
1790  *
1791  * Local Variables:
1792  * eval: (c-set-style "gnu")
1793  * End:
1794  */
void tcp_make_fin(tcp_connection_t *tc, vlib_buffer_t *b)
Convert buffer to FIN-ACK.
Definition: tcp_output.c:518
#define TCP_IW_N_SEGMENTS
Definition: tcp.h:36
#define tcp_in_cong_recovery(tc)
Definition: tcp.h:284
#define TCP_FIB_RECHECK_PERIOD
Recheck every 1s.
Definition: tcp.h:31
vlib_frame_t * vlib_get_frame_to_node(vlib_main_t *vm, u32 to_node_index)
Definition: main.c:187
End of options.
Definition: tcp_packet.h:104
sll srl srl sll sra u16x4 i
Definition: vector_sse2.h:337
#define clib_min(x, y)
Definition: clib.h:332
#define TCP_OPTION_LEN_EOL
Definition: tcp_packet.h:163
#define CLIB_UNUSED(x)
Definition: clib.h:79
static void tcp_enqueue_to_ip_lookup(vlib_main_t *vm, vlib_buffer_t *b, u32 bi, u8 is_ip4)
Definition: tcp_output.c:579
#define tcp_in_recovery(tc)
Definition: tcp.h:278
static void tcp_retransmit_timer_set(tcp_connection_t *tc)
Definition: tcp.h:602
#define TCP_OPTION_LEN_SACK_PERMITTED
Definition: tcp_packet.h:167
void tcp_timer_retransmit_handler(u32 index)
Definition: tcp_output.c:1179
ip4_address_t src_address
Definition: ip4_packet.h:164
struct _transport_connection transport_connection_t
#define TCP_TO_TIMER_TICK
Definition: tcp.h:94
fib_node_index_t tcp_lookup_rmt_in_fib(tcp_connection_t *tc)
Definition: tcp.c:396
Selective Ack permitted.
Definition: tcp_packet.h:108
#define TCP_FLAG_SYN
Definition: fa_node.h:8
static u8 svm_fifo_has_ooo_data(svm_fifo_t *f)
Definition: svm_fifo.h:83
#define tcp_opts_tstamp(_to)
Definition: tcp_packet.h:157
static uword tcp4_send_reset(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_output.c:1737
void tcp_make_synack(tcp_connection_t *tc, vlib_buffer_t *b)
Convert buffer to SYN-ACK.
Definition: tcp_output.c:538
#define PREDICT_TRUE(x)
Definition: clib.h:98
static void tcp_enqueue_to_output(vlib_main_t *vm, vlib_buffer_t *b, u32 bi, u8 is_ip4)
Definition: tcp_output.c:843
#define tcp_return_buffer(tm)
Definition: tcp_output.c:448
static tcp_connection_t * tcp_connection_get_if_valid(u32 conn_index, u32 thread_index)
Definition: tcp.h:390
static int tcp_make_syn_options(tcp_options_t *opts, u8 wnd_scale)
Definition: tcp_output.c:259
struct _sack_scoreboard sack_scoreboard_t
static tcp_connection_t * tcp_half_open_connection_get(u32 conn_index)
Definition: tcp.h:415
void vlib_put_next_frame(vlib_main_t *vm, vlib_node_runtime_t *r, u32 next_index, u32 n_vectors_left)
Release pointer to next frame vector data.
Definition: main.c:459
void tcp_update_rcv_mss(tcp_connection_t *tc)
Update max segment size we&#39;re able to process.
Definition: tcp_output.c:379
struct _tcp_main tcp_main_t
u32 thread_index
Definition: main.h:159
#define TCP_OPTS_ALIGN
Definition: tcp_packet.h:173
static u32 tcp_initial_wnd_unscaled(tcp_connection_t *tc)
TCP&#39;s IW as recommended by RFC6928.
Definition: tcp_output.c:78
struct _vlib_node_registration vlib_node_registration_t
static u32 tcp_session_has_ooo_data(tcp_connection_t *tc)
Definition: tcp_output.c:1418
struct _tcp_connection tcp_connection_t
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:419
#define tcp_get_free_buffer_index(tm, bidx)
Definition: tcp_output.c:427
#define tcp_opts_sack(_to)
Definition: tcp_packet.h:159
void tcp_make_ack_i(tcp_connection_t *tc, vlib_buffer_t *b, tcp_state_t state, u8 flags)
Prepare ACK.
Definition: tcp_output.c:478
static void scoreboard_clear(sack_scoreboard_t *sb)
Definition: tcp.h:723
static uword tcp46_send_reset_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame, u8 is_ip4)
Definition: tcp_output.c:1668
vlib_error_t * errors
Vector of errors for this node.
Definition: node.h:419
No operation.
Definition: tcp_packet.h:105
int tcp_make_reset_in_place(vlib_main_t *vm, vlib_buffer_t *b0, tcp_state_t state, u8 thread_index, u8 is_ip4)
Definition: tcp_output.c:603
u8 n_sack_blocks
Number of SACKs blocks.
Definition: tcp_packet.h:152
struct _tcp_header tcp_header_t
static u32 tcp_available_snd_space(const tcp_connection_t *tc)
Definition: tcp.h:511
static u32 stream_session_max_rx_enqueue(transport_connection_t *tc)
Definition: session.h:348
ip6_address_t src_address
Definition: ip6_packet.h:341
struct _sack_scoreboard_hole sack_scoreboard_hole_t
u8 wscale
Window scale advertised.
Definition: tcp_packet.h:148
static void tcp_reuse_buffer(vlib_main_t *vm, vlib_buffer_t *b)
Definition: tcp_output.c:457
#define TCP_OPTS_MAX_SACK_BLOCKS
Definition: tcp_packet.h:174
#define TCP_MAX_RX_FIFO_SIZE
Definition: tcp.h:35
u16 dummy_mtu
Definition: tcp_output.c:46
vlib_node_registration_t ip4_lookup_node
(constructor) VLIB_REGISTER_NODE (ip4_lookup_node)
Definition: ip4_forward.c:479
struct _stream_session_t stream_session_t
static void tcp_timer_retransmit_handler_i(u32 index, u8 is_syn)
Definition: tcp_output.c:1057
#define foreach_tcp4_reset_next
Definition: tcp_output.c:1659
Limit MSS.
Definition: tcp_packet.h:106
#define VLIB_BUFFER_NEXT_PRESENT
Definition: buffer.h:87
i16 current_data
signed offset in data[], pre_data[] that we are currently processing.
Definition: buffer.h:67
#define seq_gt(_s1, _s2)
Definition: tcp.h:439
void tcp_push_ip_hdr(tcp_main_t *tm, tcp_connection_t *tc, vlib_buffer_t *b)
Definition: tcp_output.c:757
#define always_inline
Definition: clib.h:84
static uword format_get_indent(u8 *s)
Definition: format.h:72
#define TCP_OPTION_LEN_SACK_BLOCK
Definition: tcp_packet.h:169
ip4_address_t dst_address
Definition: ip4_packet.h:164
#define TCP_FLAG_ACK
Definition: fa_node.h:11
u8 * format_white_space(u8 *s, va_list *va)
Definition: std-formats.c:113
static tcp_header_t * tcp_buffer_hdr(vlib_buffer_t *b)
Definition: tcp.h:372
int i32
Definition: types.h:81
u8 * format_tcp_tx_trace(u8 *s, va_list *args)
Definition: tcp_output.c:49
enum _tcp_state tcp_state_t
void tcp_fast_retransmit_no_sack(tcp_connection_t *tc)
Fast retransmit without SACK info.
Definition: tcp_output.c:1365
#define TCP_RTO_MAX
Definition: tcp.h:103
static void * ip4_next_header(ip4_header_t *i)
Definition: ip4_packet.h:233
static u32 tcp_time_now(void)
Definition: tcp.h:540
sack_block_t * sacks
SACK blocks.
Definition: tcp_packet.h:151
#define TCP_ESTABLISH_TIME
Definition: tcp.h:97
sack_scoreboard_hole_t * scoreboard_next_rxt_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *start, u8 have_sent_1_smss, u8 *can_rescue, u8 *snd_limited)
Figure out the next hole to retransmit.
Definition: tcp_input.c:622
#define tcp_validate_txf_size(_tc, _a)
Definition: tcp.h:664
#define TCP_EVT_DBG(_evt, _args...)
Definition: tcp_debug.h:183
static void tcp_timer_set(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:571
#define TCP_OPTION_LEN_WINDOW_SCALE
Definition: tcp_packet.h:166
vlib_node_registration_t tcp6_reset_node
(constructor) VLIB_REGISTER_NODE (tcp6_reset_node)
Definition: tcp_output.c:1770
#define TCP_RTO_SYN_RETRIES
Definition: tcp.h:106
static int tcp_make_synack_options(tcp_connection_t *tc, tcp_options_t *opts)
Definition: tcp_output.c:288
u16 current_length
Nbytes between current data and the end of this buffer.
Definition: buffer.h:71
static void tcp_push_hdr_i(tcp_connection_t *tc, vlib_buffer_t *b, tcp_state_t next_state, u8 compute_opts)
Push TCP header and update connection variables.
Definition: tcp_output.c:910
static void * vlib_buffer_make_headroom(vlib_buffer_t *b, u8 size)
Make head room, typically for packet headers.
Definition: buffer.h:300
#define MAX_HDRS_LEN
Definition: session.h:27
#define tcp_in_fastrecovery(tc)
Definition: tcp.h:277
static void * vlib_buffer_push_tcp_net_order(vlib_buffer_t *b, u16 sp, u16 dp, u32 seq, u32 ack, u8 tcp_hdr_opts_len, u8 flags, u16 wnd)
Push TCP header to buffer.
Definition: tcp.h:798
#define tcp_opts_mss(_to)
Definition: tcp_packet.h:156
static void * vlib_buffer_get_current(vlib_buffer_t *b)
Get pointer to current data to process.
Definition: buffer.h:188
#define TCP_TIMER_HANDLE_INVALID
Definition: tcp.h:91
#define foreach_tcp6_output_next
Definition: tcp_output.c:31
static u32 tcp_flight_size(const tcp_connection_t *tc)
Our estimate of the number of bytes in flight (pipe size)
Definition: tcp.h:463
#define PREDICT_FALSE(x)
Definition: clib.h:97
static void * vlib_buffer_push_ip4(vlib_main_t *vm, vlib_buffer_t *b, ip4_address_t *src, ip4_address_t *dst, int proto)
Push IPv4 header to buffer.
Definition: ip4.h:330
void tcp_fast_retransmit(tcp_connection_t *tc)
Do fast retransmit.
Definition: tcp_output.c:1408
#define TCP_FLAG_FIN
Definition: fa_node.h:7
static void tcp_rtx_timeout_cc(tcp_connection_t *tc)
Reset congestion control, switch cwnd to loss window and try again.
Definition: tcp_output.c:1039
void tcp_timer_retransmit_syn_handler(u32 index)
Definition: tcp_output.c:1185
#define vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next, n_left_to_next, bi0, next0)
Finish enqueueing one buffer forward in the graph.
Definition: buffer_node.h:216
#define vlib_get_next_frame(vm, node, next_index, vectors, n_vectors_left)
Get pointer to next frame vector data by (vlib_node_runtime_t, next_index).
Definition: node_funcs.h:366
#define TCP_OPTION_LEN_TIMESTAMP
Definition: tcp_packet.h:168
#define foreach_tcp4_output_next
Definition: tcp_output.c:28
vlib_error_t error
Error code for buffers to be enqueued to error handler.
Definition: buffer.h:113
#define TCP_WND_MAX
Definition: tcp_packet.h:171
Selective Ack block.
Definition: tcp_packet.h:109
void tcp_make_ack(tcp_connection_t *tc, vlib_buffer_t *b)
Convert buffer to ACK.
Definition: tcp_output.c:503
#define TCP_FLAG_RST
Definition: fa_node.h:9
#define TCP_MAX_WND_SCALE
Definition: tcp_packet.h:172
static void tcp_timer_reset(tcp_connection_t *tc, u8 timer_id)
Definition: tcp.h:579
static sack_scoreboard_hole_t * scoreboard_first_hole(sack_scoreboard_t *sb)
Definition: tcp.h:707
#define tcp_fastrecovery_sent_1_smss(tc)
Definition: tcp.h:280
u32 tcp_prepare_retransmit_segment(tcp_connection_t *tc, vlib_buffer_t *b, u32 offset, u32 max_bytes)
Build a retransmit segment.
Definition: tcp_output.c:989
static void * vlib_buffer_push_tcp(vlib_buffer_t *b, u16 sp_net, u16 dp_net, u32 seq, u32 ack, u8 tcp_hdr_opts_len, u8 flags, u16 wnd)
Push TCP header to buffer.
Definition: tcp.h:833
tcp_header_t tcp_header
Definition: tcp_output.c:42
u16 n_vectors
Definition: node.h:345
#define VNET_BUFFER_LOCALLY_ORIGINATED
Definition: buffer.h:68
static_always_inline uword vlib_get_thread_index(void)
Definition: threads.h:185
enum _tcp_output_nect tcp_output_next_t
#define clib_warning(format, args...)
Definition: error.h:59
#define VLIB_BUFFER_IS_TRACED
Definition: buffer.h:85
#define clib_memcpy(a, b, c)
Definition: string.h:69
format_function_t format_tcp_header
Definition: format.h:102
#define TCP_USE_SACKS
Disable only for testing.
Definition: tcp.h:38
#define tcp_recovery_on(tc)
Definition: tcp.h:275
u16 mss
Option flags, see above.
Definition: tcp_packet.h:147
static void * ip6_next_header(ip6_header_t *i)
Definition: ip6_packet.h:351
u16 ip6_tcp_udp_icmp_compute_checksum(vlib_main_t *vm, vlib_buffer_t *p0, ip6_header_t *ip0, int *bogus_lengthp)
Definition: ip6_forward.c:1195
vlib_node_registration_t ip6_lookup_node
(constructor) VLIB_REGISTER_NODE (ip6_lookup_node)
Definition: ip6_forward.c:733
static int tcp_make_established_options(tcp_connection_t *tc, tcp_options_t *opts)
Definition: tcp_output.c:323
u16 cached_next_index
Next frame index that vector arguments were last enqueued to last time this node ran.
Definition: node.h:460
#define ASSERT(truth)
#define tcp_syn(_th)
Definition: tcp_packet.h:80
unsigned int u32
Definition: types.h:88
static uword tcp6_output(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_output.c:1587
u16 ip4_tcp_udp_compute_checksum(vlib_main_t *vm, vlib_buffer_t *p0, ip4_header_t *ip0)
Definition: ip4_forward.c:1445
u32 tcp_push_header(transport_connection_t *tconn, vlib_buffer_t *b)
Definition: tcp_output.c:1637
#define seq_geq(_s1, _s2)
Definition: tcp.h:440
u32 tcp_window_to_advertise(tcp_connection_t *tc, tcp_state_t state)
Compute and return window to advertise, scaled as per RFC1323.
Definition: tcp_output.c:107
vhost_vring_state_t state
Definition: vhost-user.h:81
void tcp_retransmit_first_unacked(tcp_connection_t *tc)
Retransmit first unacked segment.
Definition: tcp_output.c:1258
u32 next_buffer
Next buffer for this linked-list of buffers.
Definition: buffer.h:109
void tcp_init_mss(tcp_connection_t *tc)
Definition: tcp_output.c:403
static uword tcp6_send_reset(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_output.c:1744
static uword tcp46_output_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame, int is_ip4)
Definition: tcp_output.c:1426
static u8 tcp_make_state_flags(tcp_state_t next_state)
Definition: tcp_output.c:887
void tcp_send_fin(tcp_connection_t *tc)
Send FIN.
Definition: tcp_output.c:866
static u8 tcp_window_compute_scale(u32 available_space)
Definition: tcp_output.c:65
#define clib_max(x, y)
Definition: clib.h:325
static vlib_main_t * vlib_get_main(void)
Definition: global_funcs.h:23
VLIB_NODE_FUNCTION_MULTIARCH(tcp4_output_node, tcp4_output)
void tcp_send_ack(tcp_connection_t *tc)
Definition: tcp_output.c:954
u64 uword
Definition: types.h:112
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
u32 total_length_not_including_first_buffer
Only valid for first buffer in chain.
Definition: buffer.h:103
#define seq_lt(_s1, _s2)
Definition: tcp.h:437
template key/value backing page structure
Definition: bihash_doc.h:44
u32 ip_version_traffic_class_and_flow_label
Definition: ip6_packet.h:328
#define tcp_opts_wscale(_to)
Definition: tcp_packet.h:158
static sack_scoreboard_hole_t * scoreboard_get_hole(sack_scoreboard_t *sb, u32 index)
Definition: tcp.h:683
Definition: defs.h:47
unsigned short u16
Definition: types.h:57
void tcp_connection_reset(tcp_connection_t *tc)
Notify session that connection has been reset.
Definition: tcp.c:152
u32 tsval
Timestamp value.
Definition: tcp_packet.h:149
u32 tsecr
Echoed/reflected time stamp.
Definition: tcp_packet.h:150
static void * vlib_buffer_push_ip6(vlib_main_t *vm, vlib_buffer_t *b, ip6_address_t *src, ip6_address_t *dst, int proto)
Push IPv6 header to buffer.
Definition: ip6.h:555
#define vec_len(v)
Number of elements in vector (rvalue-only, NULL tolerant)
unsigned char u8
Definition: types.h:56
#define foreach_tcp6_reset_next
Definition: tcp_output.c:1663
static void * vlib_frame_vector_args(vlib_frame_t *f)
Get pointer to frame vector data.
Definition: node_funcs.h:269
void tcp_timer_delack_handler(u32 index)
Definition: tcp_output.c:973
#define TCP_OPTION_LEN_MSS
Definition: tcp_packet.h:165
void tcp_update_snd_mss(tcp_connection_t *tc)
Update snd_mss to reflect the effective segment size that we can send by taking into account all TCP ...
Definition: tcp_output.c:390
struct clib_bihash_value offset
template key/value backing page structure
static void tcp_retransmit_timer_force_update(tcp_connection_t *tc)
Definition: tcp.h:615
u8 * format_tcp_connection(u8 *s, va_list *args)
Definition: tcp.c:669
u32 tcp_initial_window_to_advertise(tcp_connection_t *tc)
Compute initial window and scale factor.
Definition: tcp_output.c:88
static u32 stream_session_rx_fifo_size(transport_connection_t *tc)
Definition: session.h:355
#define vnet_buffer(b)
Definition: buffer.h:303
static tcp_connection_t * tcp_connection_get(u32 conn_index, u32 thread_index)
Definition: tcp.h:382
static u32 random_u32(u32 *seed)
32-bit random number generator
Definition: random.h:69
void tcp_update_rcv_wnd(tcp_connection_t *tc)
Definition: tcp_output.c:127
#define VLIB_REGISTER_NODE(x,...)
Definition: node.h:144
void tcp_update_rto(tcp_connection_t *tc)
Definition: tcp_input.c:405
void tcp_fast_retransmit_sack(tcp_connection_t *tc)
Do fast retransmit with SACKs.
Definition: tcp_output.c:1291
#define TCP_OPTION_LEN_NOOP
Definition: tcp_packet.h:164
void tcp_send_syn(tcp_connection_t *tc)
Send SYN.
Definition: tcp_output.c:789
vlib_node_registration_t tcp6_output_node
(constructor) VLIB_REGISTER_NODE (tcp6_output_node)
Definition: tcp_output.c:20
Window scale.
Definition: tcp_packet.h:107
enum _tcp_reset_next tcp_reset_next_t
#define tcp_opts_sack_permitted(_to)
Definition: tcp_packet.h:160
tcp_connection_t tcp_connection
Definition: tcp_output.c:43
void vlib_put_frame_to_node(vlib_main_t *vm, u32 to_node_index, vlib_frame_t *f)
Definition: main.c:196
static u32 tcp_loss_wnd(const tcp_connection_t *tc)
Definition: tcp.h:499
u8 ip_version_and_header_length
Definition: ip4_packet.h:132
Timestamps.
Definition: tcp_packet.h:110
vlib_node_registration_t tcp4_reset_node
(constructor) VLIB_REGISTER_NODE (tcp4_reset_node)
Definition: tcp_output.c:1751
vlib_node_registration_t tcp4_output_node
(constructor) VLIB_REGISTER_NODE (tcp4_output_node)
Definition: tcp_output.c:19
u32 flags
Definition: vhost-user.h:76
u32 flags
buffer flags: VLIB_BUFFER_IS_TRACED: trace this buffer.
Definition: buffer.h:74
static void tcp_persist_timer_set(tcp_connection_t *tc)
Definition: tcp.h:622
static tcp_main_t * vnet_get_tcp_main()
Definition: tcp.h:366
int stream_session_peek_bytes(transport_connection_t *tc, u8 *buffer, u32 offset, u32 max_bytes)
Definition: session.c:575
static char * tcp_error_strings[]
Definition: tcp_output.c:34
static uword tcp4_output(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *from_frame)
Definition: tcp_output.c:1580
static vlib_buffer_t * vlib_get_buffer(vlib_main_t *vm, u32 buffer_index)
Translate buffer index into buffer pointer.
Definition: buffer_funcs.h:57
void tcp_cc_fastrecovery_exit(tcp_connection_t *tc)
Definition: tcp_input.c:894
void tcp_timer_persist_handler(u32 index)
Got 0 snd_wnd from peer, try to do something about it.
Definition: tcp_output.c:1195
#define tcp_ack(_th)
Definition: tcp_packet.h:83
static u8 tcp_timer_is_active(tcp_connection_t *tc, tcp_timers_e timer)
Definition: tcp.h:659
ip6_address_t dst_address
Definition: ip6_packet.h:341
static stream_session_t * stream_session_get(u32 si, u32 thread_index)
Definition: session.h:281
u32 tcp_options_write(u8 *data, tcp_options_t *opts)
Write TCP options to segment.
Definition: tcp_output.c:176
void tcp_send_reset(vlib_buffer_t *pkt, u8 is_ip4)
Send reset without reusing existing buffer.
Definition: tcp_output.c:684
static int tcp_make_options(tcp_connection_t *tc, tcp_options_t *opts, tcp_state_t state)
Definition: tcp_output.c:354