FD.io VPP  v20.09-64-g4f7b92f0a
Vector Packet Processing
tcp_output.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 <vnet/tcp/tcp.h>
17 #include <vnet/tcp/tcp_inlines.h>
18 #include <math.h>
19 
20 typedef enum _tcp_output_next
21 {
28 
29 #define foreach_tcp4_output_next \
30  _ (DROP, "error-drop") \
31  _ (IP_LOOKUP, "ip4-lookup") \
32  _ (IP_REWRITE, "ip4-rewrite") \
33  _ (IP_ARP, "ip4-arp")
34 
35 #define foreach_tcp6_output_next \
36  _ (DROP, "error-drop") \
37  _ (IP_LOOKUP, "ip6-lookup") \
38  _ (IP_REWRITE, "ip6-rewrite") \
39  _ (IP_ARP, "ip6-discover-neighbor")
40 
41 static char *tcp_error_strings[] = {
42 #define tcp_error(n,s) s,
43 #include <vnet/tcp/tcp_error.def>
44 #undef tcp_error
45 };
46 
47 typedef struct
48 {
52 
53 static u8 *
54 format_tcp_tx_trace (u8 * s, va_list * args)
55 {
56  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
57  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
58  tcp_tx_trace_t *t = va_arg (*args, tcp_tx_trace_t *);
60  u32 indent = format_get_indent (s);
61 
62  s = format (s, "%U state %U\n%U%U", format_tcp_connection_id, tc,
63  format_tcp_state, tc->state, format_white_space, indent,
64  format_tcp_header, &t->tcp_header, 128);
65 
66  return s;
67 }
68 
69 #ifndef CLIB_MARCH_VARIANT
70 static u8
72 {
73  u8 wnd_scale = 0;
74  while (wnd_scale < TCP_MAX_WND_SCALE && (window >> wnd_scale) > TCP_WND_MAX)
75  wnd_scale++;
76  return wnd_scale;
77 }
78 
79 /**
80  * TCP's initial window
81  */
84 {
85  /* RFC 6928 recommends the value lower. However at the time our connections
86  * are initialized, fifos may not be allocated. Therefore, advertise the
87  * smallest possible unscaled window size and update once fifos are
88  * assigned to the session.
89  */
90  /*
91  tcp_update_rcv_mss (tc);
92  TCP_IW_N_SEGMENTS * tc->mss;
93  */
94  return tcp_cfg.min_rx_fifo;
95 }
96 
97 /**
98  * Compute initial window and scale factor. As per RFC1323, window field in
99  * SYN and SYN-ACK segments is never scaled.
100  */
101 u32
103 {
104  /* Compute rcv wscale only if peer advertised support for it */
105  if (tc->state != TCP_STATE_SYN_RCVD || tcp_opts_wscale (&tc->rcv_opts))
106  tc->rcv_wscale = tcp_window_compute_scale (tcp_cfg.max_rx_fifo);
107 
108  tc->rcv_wnd = tcp_initial_wnd_unscaled (tc);
109 
110  return clib_min (tc->rcv_wnd, TCP_WND_MAX);
111 }
112 
113 static inline void
115 {
116  u32 available_space, wnd;
117  i32 observed_wnd;
118 
119  /*
120  * Figure out how much space we have available
121  */
122  available_space = transport_max_rx_enqueue (&tc->connection);
123 
124  /*
125  * Use the above and what we know about what we've previously advertised
126  * to compute the new window
127  */
128  observed_wnd = (i32) tc->rcv_wnd - (tc->rcv_nxt - tc->rcv_las);
129 
130  /* Check if we are about to retract the window. Do the comparison before
131  * rounding to avoid errors. Per RFC7323 sec. 2.4 we could remove this */
132  if (PREDICT_FALSE ((i32) available_space < observed_wnd))
133  {
134  wnd = round_down_pow2 (clib_max (observed_wnd, 0), 1 << tc->rcv_wscale);
135  TCP_EVT (TCP_EVT_RCV_WND_SHRUNK, tc, observed_wnd, available_space);
136  }
137  else
138  {
139  /* Make sure we have a multiple of 1 << rcv_wscale. We round down to
140  * avoid advertising a window larger than what can be buffered */
141  wnd = round_down_pow2 (available_space, 1 << tc->rcv_wscale);
142  }
143 
144  if (PREDICT_FALSE (wnd < tc->rcv_opts.mss))
145  wnd = 0;
146 
147  tc->rcv_wnd = clib_min (wnd, TCP_WND_MAX << tc->rcv_wscale);
148 }
149 
150 /**
151  * Compute and return window to advertise, scaled as per RFC1323
152  */
153 static inline u32
155 {
156  if (state < TCP_STATE_ESTABLISHED)
158 
159  tcp_update_rcv_wnd (tc);
160  return tc->rcv_wnd >> tc->rcv_wscale;
161 }
162 
163 static int
165 {
166  u8 len = 0;
167 
168  opts->flags |= TCP_OPTS_FLAG_MSS;
169  opts->mss = tc->mss;
170  len += TCP_OPTION_LEN_MSS;
171 
172  opts->flags |= TCP_OPTS_FLAG_WSCALE;
173  opts->wscale = tc->rcv_wscale;
175 
176  opts->flags |= TCP_OPTS_FLAG_TSTAMP;
177  opts->tsval = tcp_time_now ();
178  opts->tsecr = 0;
180 
181  if (TCP_USE_SACKS)
182  {
183  opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED;
185  }
186 
187  /* Align to needed boundary */
188  len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
189  return len;
190 }
191 
192 static int
194 {
195  u8 len = 0;
196 
197  opts->flags |= TCP_OPTS_FLAG_MSS;
198  opts->mss = tc->mss;
199  len += TCP_OPTION_LEN_MSS;
200 
201  if (tcp_opts_wscale (&tc->rcv_opts))
202  {
203  opts->flags |= TCP_OPTS_FLAG_WSCALE;
204  opts->wscale = tc->rcv_wscale;
206  }
207 
208  if (tcp_opts_tstamp (&tc->rcv_opts))
209  {
210  opts->flags |= TCP_OPTS_FLAG_TSTAMP;
211  opts->tsval = tcp_time_now ();
212  opts->tsecr = tc->tsval_recent;
214  }
215 
216  if (tcp_opts_sack_permitted (&tc->rcv_opts))
217  {
218  opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED;
220  }
221 
222  /* Align to needed boundary */
223  len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
224  return len;
225 }
226 
227 static int
229 {
230  u8 len = 0;
231 
232  opts->flags = 0;
233 
234  if (tcp_opts_tstamp (&tc->rcv_opts))
235  {
236  opts->flags |= TCP_OPTS_FLAG_TSTAMP;
237  opts->tsval = tcp_tstamp (tc);
238  opts->tsecr = tc->tsval_recent;
240  }
241  if (tcp_opts_sack_permitted (&tc->rcv_opts))
242  {
243  if (vec_len (tc->snd_sacks))
244  {
245  opts->flags |= TCP_OPTS_FLAG_SACK;
246  if (tc->snd_sack_pos >= vec_len (tc->snd_sacks))
247  tc->snd_sack_pos = 0;
248  opts->sacks = &tc->snd_sacks[tc->snd_sack_pos];
249  opts->n_sack_blocks = vec_len (tc->snd_sacks) - tc->snd_sack_pos;
250  opts->n_sack_blocks = clib_min (opts->n_sack_blocks,
252  tc->snd_sack_pos += opts->n_sack_blocks;
253  len += 2 + TCP_OPTION_LEN_SACK_BLOCK * opts->n_sack_blocks;
254  }
255  }
256 
257  /* Align to needed boundary */
258  len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
259  return len;
260 }
261 
262 always_inline int
265 {
266  switch (state)
267  {
268  case TCP_STATE_ESTABLISHED:
269  case TCP_STATE_CLOSE_WAIT:
270  case TCP_STATE_FIN_WAIT_1:
271  case TCP_STATE_LAST_ACK:
272  case TCP_STATE_CLOSING:
273  case TCP_STATE_FIN_WAIT_2:
274  case TCP_STATE_TIME_WAIT:
275  case TCP_STATE_CLOSED:
276  return tcp_make_established_options (tc, opts);
277  case TCP_STATE_SYN_RCVD:
278  return tcp_make_synack_options (tc, opts);
279  case TCP_STATE_SYN_SENT:
280  return tcp_make_syn_options (tc, opts);
281  default:
282  clib_warning ("State not handled! %d", state);
283  return 0;
284  }
285 }
286 
287 /**
288  * Update burst send vars
289  *
290  * - Updates snd_mss to reflect the effective segment size that we can send
291  * by taking into account all TCP options, including SACKs.
292  * - Cache 'on the wire' options for reuse
293  * - Updates receive window which can be reused for a burst.
294  *
295  * This should *only* be called when doing bursts
296  */
297 void
299 {
300  tcp_main_t *tm = &tcp_main;
301 
302  /* Compute options to be used for connection. These may be reused when
303  * sending data or to compute the effective mss (snd_mss) */
304  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts,
305  TCP_STATE_ESTABLISHED);
306 
307  /* XXX check if MTU has been updated */
308  tc->snd_mss = clib_min (tc->mss, tc->rcv_opts.mss) - tc->snd_opts_len;
309  ASSERT (tc->snd_mss > 0);
310 
311  tcp_options_write (tm->wrk_ctx[tc->c_thread_index].cached_opts,
312  &tc->snd_opts);
313 
314  tcp_update_rcv_wnd (tc);
315 
316  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
318 
319  if (tc->snd_una == tc->snd_nxt)
320  {
323  }
324 }
325 
326 #endif /* CLIB_MARCH_VARIANT */
327 
328 static void *
330 {
331  if (b->flags & VLIB_BUFFER_NEXT_PRESENT)
333  /* Zero all flags but free list index and trace flag */
334  b->flags &= VLIB_BUFFER_NEXT_PRESENT - 1;
335  b->current_data = 0;
336  b->current_length = 0;
338  vnet_buffer (b)->tcp.flags = 0;
339 
340  /* Leave enough space for headers */
342 }
343 
344 #ifndef CLIB_MARCH_VARIANT
345 static void *
347 {
348  ASSERT ((b->flags & VLIB_BUFFER_NEXT_PRESENT) == 0);
349  b->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
351  b->current_data = 0;
352  vnet_buffer (b)->tcp.flags = 0;
354  /* Leave enough space for headers */
356 }
357 
358 
359 /* Compute TCP checksum in software when offloading is disabled for a connection */
360 u16
362  ip46_address_t * src, ip46_address_t * dst)
363 {
364  ip_csum_t sum0;
365  u16 payload_length_host_byte_order;
366  u32 i;
367 
368  /* Initialize checksum with ip header. */
369  sum0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, p0)) +
370  clib_host_to_net_u16 (IP_PROTOCOL_TCP);
371  payload_length_host_byte_order = vlib_buffer_length_in_chain (vm, p0);
372 
373  for (i = 0; i < ARRAY_LEN (src->ip6.as_uword); i++)
374  {
375  sum0 = ip_csum_with_carry
376  (sum0, clib_mem_unaligned (&src->ip6.as_uword[i], uword));
377  sum0 = ip_csum_with_carry
378  (sum0, clib_mem_unaligned (&dst->ip6.as_uword[i], uword));
379  }
380 
381  return ip_calculate_l4_checksum (vm, p0, sum0,
382  payload_length_host_byte_order, NULL, 0,
383  NULL);
384 }
385 
386 u16
388  ip46_address_t * src, ip46_address_t * dst)
389 {
390  ip_csum_t sum0;
391  u32 payload_length_host_byte_order;
392 
393  payload_length_host_byte_order = vlib_buffer_length_in_chain (vm, p0);
394  sum0 =
395  clib_host_to_net_u32 (payload_length_host_byte_order +
396  (IP_PROTOCOL_TCP << 16));
397 
398  sum0 = ip_csum_with_carry (sum0, clib_mem_unaligned (&src->ip4, u32));
399  sum0 = ip_csum_with_carry (sum0, clib_mem_unaligned (&dst->ip4, u32));
400 
401  return ip_calculate_l4_checksum (vm, p0, sum0,
402  payload_length_host_byte_order, NULL, 0,
403  NULL);
404 }
405 
406 static inline u16
408 {
409  u16 checksum = 0;
410  if (PREDICT_FALSE (tc->cfg_flags & TCP_CFG_F_NO_CSUM_OFFLOAD))
411  {
412  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
413  vlib_main_t *vm = wrk->vm;
414 
415  if (tc->c_is_ip4)
417  (vm, b, &tc->c_lcl_ip, &tc->c_rmt_ip);
418  else
420  (vm, b, &tc->c_lcl_ip, &tc->c_rmt_ip);
421  }
422  else
423  {
424  b->flags |= VNET_BUFFER_F_OFFLOAD_TCP_CKSUM;
425  }
426  return checksum;
427 }
428 
429 /**
430  * Prepare ACK
431  */
432 static inline void
434  u8 flags)
435 {
436  tcp_options_t _snd_opts, *snd_opts = &_snd_opts;
437  u8 tcp_opts_len, tcp_hdr_opts_len;
438  tcp_header_t *th;
439  u16 wnd;
440 
441  wnd = tcp_window_to_advertise (tc, state);
442 
443  /* Make and write options */
444  tcp_opts_len = tcp_make_established_options (tc, snd_opts);
445  tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
446 
447  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt,
448  tc->rcv_nxt, tcp_hdr_opts_len, flags, wnd);
449 
450  tcp_options_write ((u8 *) (th + 1), snd_opts);
451 
452  th->checksum = tcp_compute_checksum (tc, b);
453 
454  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
455 
456  if (wnd == 0)
458  else
460 }
461 
462 /**
463  * Convert buffer to ACK
464  */
465 static inline void
467 {
468  tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, TCP_FLAG_ACK);
469  TCP_EVT (TCP_EVT_ACK_SENT, tc);
470  tc->rcv_las = tc->rcv_nxt;
471 }
472 
473 /**
474  * Convert buffer to FIN-ACK
475  */
476 static void
478 {
479  tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_ACK);
480 }
481 
482 /**
483  * Convert buffer to SYN
484  */
485 void
487 {
488  u8 tcp_hdr_opts_len, tcp_opts_len;
489  tcp_header_t *th;
490  u16 initial_wnd;
491  tcp_options_t snd_opts;
492 
493  initial_wnd = tcp_initial_window_to_advertise (tc);
494 
495  /* Make and write options */
496  clib_memset (&snd_opts, 0, sizeof (snd_opts));
497  tcp_opts_len = tcp_make_syn_options (tc, &snd_opts);
498  tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
499 
500  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss,
501  tc->rcv_nxt, tcp_hdr_opts_len, TCP_FLAG_SYN,
502  initial_wnd);
503  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
504  tcp_options_write ((u8 *) (th + 1), &snd_opts);
505  th->checksum = tcp_compute_checksum (tc, b);
506 }
507 
508 /**
509  * Convert buffer to SYN-ACK
510  */
511 static void
513 {
514  tcp_options_t _snd_opts, *snd_opts = &_snd_opts;
515  u8 tcp_opts_len, tcp_hdr_opts_len;
516  tcp_header_t *th;
517  u16 initial_wnd;
518 
519  clib_memset (snd_opts, 0, sizeof (*snd_opts));
520  initial_wnd = tcp_initial_window_to_advertise (tc);
521  tcp_opts_len = tcp_make_synack_options (tc, snd_opts);
522  tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
523 
524  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss,
525  tc->rcv_nxt, tcp_hdr_opts_len,
526  TCP_FLAG_SYN | TCP_FLAG_ACK, initial_wnd);
527  tcp_options_write ((u8 *) (th + 1), snd_opts);
528 
529  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
530  th->checksum = tcp_compute_checksum (tc, b);
531 }
532 
533 static void
535  u8 is_ip4, u32 fib_index)
536 {
537  tcp_main_t *tm = &tcp_main;
538  vlib_main_t *vm = wrk->vm;
539 
540  b->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
541  b->error = 0;
542 
543  vnet_buffer (b)->sw_if_index[VLIB_TX] = fib_index;
544  vnet_buffer (b)->sw_if_index[VLIB_RX] = 0;
545 
547 
549  tm->ipl_next_node[!is_ip4]);
550 
551  if (vm->thread_index == 0 && vlib_num_workers ())
553 }
554 
555 static void
557  u8 is_ip4)
558 {
559  b->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
560  b->error = 0;
561 
563  wrk->tco_next_node[!is_ip4]);
564 }
565 
566 #endif /* CLIB_MARCH_VARIANT */
567 
568 static int
570 {
571  ip4_header_t *ih4;
572  ip6_header_t *ih6;
573  tcp_header_t *th;
574  ip4_address_t src_ip4, dst_ip4;
575  ip6_address_t src_ip6, dst_ip6;
577  u32 tmp, len, seq, ack;
578  u8 flags;
579 
580  /* Find IP and TCP headers */
581  th = tcp_buffer_hdr (b);
582 
583  /* Save src and dst ip */
584  if (is_ip4)
585  {
586  ih4 = vlib_buffer_get_current (b);
587  ASSERT ((ih4->ip_version_and_header_length & 0xF0) == 0x40);
588  src_ip4.as_u32 = ih4->src_address.as_u32;
589  dst_ip4.as_u32 = ih4->dst_address.as_u32;
590  }
591  else
592  {
593  ih6 = vlib_buffer_get_current (b);
594  ASSERT ((ih6->ip_version_traffic_class_and_flow_label & 0xF0) == 0x60);
595  clib_memcpy_fast (&src_ip6, &ih6->src_address, sizeof (ip6_address_t));
596  clib_memcpy_fast (&dst_ip6, &ih6->dst_address, sizeof (ip6_address_t));
597  }
598 
599  src_port = th->src_port;
600  dst_port = th->dst_port;
601  flags = TCP_FLAG_RST;
602 
603  /*
604  * RFC 793. If the ACK bit is off, sequence number zero is used,
605  * <SEQ=0><ACK=SEG.SEQ+SEG.LEN><CTL=RST,ACK>
606  * If the ACK bit is on,
607  * <SEQ=SEG.ACK><CTL=RST>
608  */
609  if (tcp_ack (th))
610  {
611  seq = th->ack_number;
612  ack = 0;
613  }
614  else
615  {
616  flags |= TCP_FLAG_ACK;
617  tmp = clib_net_to_host_u32 (th->seq_number);
618  len = vnet_buffer (b)->tcp.data_len + tcp_is_syn (th) + tcp_is_fin (th);
619  ack = clib_host_to_net_u32 (tmp + len);
620  seq = 0;
621  }
622 
623  tcp_reuse_buffer (vm, b);
625  th = vlib_buffer_push_tcp_net_order (b, dst_port, src_port, seq, ack,
626  sizeof (tcp_header_t), flags, 0);
627 
628  if (is_ip4)
629  {
630  ih4 = vlib_buffer_push_ip4 (vm, b, &dst_ip4, &src_ip4,
631  IP_PROTOCOL_TCP, 1);
632  th->checksum = ip4_tcp_udp_compute_checksum (vm, b, ih4);
633  }
634  else
635  {
636  int bogus = ~0;
637  ih6 = vlib_buffer_push_ip6 (vm, b, &dst_ip6, &src_ip6, IP_PROTOCOL_TCP);
638  th->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b, ih6, &bogus);
639  ASSERT (!bogus);
640  }
641 
642  return 0;
643 }
644 
645 #ifndef CLIB_MARCH_VARIANT
646 /**
647  * Send reset without reusing existing buffer
648  *
649  * It extracts connection info out of original packet
650  */
651 void
653  u32 thread_index, u8 is_ip4)
654 {
655  tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index);
656  vlib_main_t *vm = wrk->vm;
657  vlib_buffer_t *b;
658  u32 bi, sw_if_index, fib_index;
659  u8 tcp_hdr_len, flags = 0;
660  tcp_header_t *th, *pkt_th;
661  u32 seq, ack;
662  ip4_header_t *ih4, *pkt_ih4;
663  ip6_header_t *ih6, *pkt_ih6;
664  fib_protocol_t fib_proto;
665 
666  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
667  {
668  tcp_worker_stats_inc (wrk, no_buffer, 1);
669  return;
670  }
671 
672  b = vlib_get_buffer (vm, bi);
673  sw_if_index = vnet_buffer (pkt)->sw_if_index[VLIB_RX];
674  fib_proto = is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6;
675  fib_index = fib_table_get_index_for_sw_if_index (fib_proto, sw_if_index);
676  tcp_init_buffer (vm, b);
677 
678  /* Make and write options */
679  tcp_hdr_len = sizeof (tcp_header_t);
680 
681  if (is_ip4)
682  {
683  pkt_ih4 = vlib_buffer_get_current (pkt);
684  pkt_th = ip4_next_header (pkt_ih4);
685  }
686  else
687  {
688  pkt_ih6 = vlib_buffer_get_current (pkt);
689  pkt_th = ip6_next_header (pkt_ih6);
690  }
691 
692  if (tcp_ack (pkt_th))
693  {
694  flags = TCP_FLAG_RST;
695  seq = pkt_th->ack_number;
696  ack = (tc->state >= TCP_STATE_SYN_RCVD) ? tc->rcv_nxt : 0;
697  }
698  else
699  {
700  flags = TCP_FLAG_RST | TCP_FLAG_ACK;
701  seq = 0;
702  ack = clib_host_to_net_u32 (vnet_buffer (pkt)->tcp.seq_end);
703  }
704 
705  th = vlib_buffer_push_tcp_net_order (b, pkt_th->dst_port, pkt_th->src_port,
706  seq, ack, tcp_hdr_len, flags, 0);
707 
708  /* Swap src and dst ip */
709  if (is_ip4)
710  {
711  ASSERT ((pkt_ih4->ip_version_and_header_length & 0xF0) == 0x40);
712  ih4 = vlib_buffer_push_ip4 (vm, b, &pkt_ih4->dst_address,
713  &pkt_ih4->src_address, IP_PROTOCOL_TCP,
714  tcp_csum_offload (tc));
715  th->checksum = ip4_tcp_udp_compute_checksum (vm, b, ih4);
716  }
717  else
718  {
719  int bogus = ~0;
720  ASSERT ((pkt_ih6->ip_version_traffic_class_and_flow_label & 0xF0) ==
721  0x60);
722  ih6 = vlib_buffer_push_ip6_custom (vm, b, &pkt_ih6->dst_address,
723  &pkt_ih6->src_address,
724  IP_PROTOCOL_TCP,
725  tc->ipv6_flow_label);
726  th->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b, ih6, &bogus);
727  ASSERT (!bogus);
728  }
729 
730  tcp_enqueue_to_ip_lookup (wrk, b, bi, is_ip4, fib_index);
731  TCP_EVT (TCP_EVT_RST_SENT, tc);
732  vlib_node_increment_counter (vm, tcp_node_index (output, tc->c_is_ip4),
733  TCP_ERROR_RST_SENT, 1);
734 }
735 
736 /**
737  * Build and set reset packet for connection
738  */
739 void
741 {
742  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
743  vlib_main_t *vm = wrk->vm;
744  vlib_buffer_t *b;
745  u32 bi;
746  tcp_header_t *th;
747  u16 tcp_hdr_opts_len, advertise_wnd, opts_write_len;
748  u8 flags;
749 
750  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
751  {
752  tcp_worker_stats_inc (wrk, no_buffer, 1);
753  return;
754  }
755  b = vlib_get_buffer (vm, bi);
756  tcp_init_buffer (vm, b);
757 
758  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
759  tcp_hdr_opts_len = tc->snd_opts_len + sizeof (tcp_header_t);
760  advertise_wnd = tc->rcv_wnd >> tc->rcv_wscale;
761  flags = TCP_FLAG_RST;
762  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt,
763  tc->rcv_nxt, tcp_hdr_opts_len, flags,
764  advertise_wnd);
765  opts_write_len = tcp_options_write ((u8 *) (th + 1), &tc->snd_opts);
766  th->checksum = tcp_compute_checksum (tc, b);
767  ASSERT (opts_write_len == tc->snd_opts_len);
768  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
769  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
770  TCP_EVT (TCP_EVT_RST_SENT, tc);
771  vlib_node_increment_counter (vm, tcp_node_index (output, tc->c_is_ip4),
772  TCP_ERROR_RST_SENT, 1);
773 }
774 
775 static void
777  vlib_buffer_t * b)
778 {
779  if (tc->c_is_ip4)
780  {
781  vlib_buffer_push_ip4 (wrk->vm, b, &tc->c_lcl_ip4, &tc->c_rmt_ip4,
782  IP_PROTOCOL_TCP, tcp_csum_offload (tc));
783  }
784  else
785  {
786  vlib_buffer_push_ip6_custom (wrk->vm, b, &tc->c_lcl_ip6, &tc->c_rmt_ip6,
787  IP_PROTOCOL_TCP, tc->ipv6_flow_label);
788  }
789 }
790 
791 /**
792  * Send SYN
793  *
794  * Builds a SYN packet for a half-open connection and sends it to ipx_lookup.
795  * The packet is not forwarded through tcpx_output to avoid doing lookups
796  * in the half_open pool.
797  */
798 void
800 {
801  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
802  vlib_main_t *vm = wrk->vm;
803  vlib_buffer_t *b;
804  u32 bi;
805 
806  /*
807  * Setup retransmit and establish timers before requesting buffer
808  * such that we can return if we've ran out.
809  */
810  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN,
811  tc->rto * TCP_TO_TIMER_TICK);
812 
813  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
814  {
815  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN, 1);
816  tcp_worker_stats_inc (wrk, no_buffer, 1);
817  return;
818  }
819 
820  b = vlib_get_buffer (vm, bi);
821  tcp_init_buffer (vm, b);
822  tcp_make_syn (tc, b);
823 
824  /* Measure RTT with this */
825  tc->rtt_ts = tcp_time_now_us (vlib_num_workers ()? 1 : 0);
826  tc->rtt_seq = tc->snd_nxt;
827  tc->rto_boff = 0;
828 
829  tcp_push_ip_hdr (wrk, tc, b);
830  tcp_enqueue_to_ip_lookup (wrk, b, bi, tc->c_is_ip4, tc->c_fib_index);
831  TCP_EVT (TCP_EVT_SYN_SENT, tc);
832 }
833 
834 void
836 {
837  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
838  vlib_main_t *vm = wrk->vm;
839  vlib_buffer_t *b;
840  u32 bi;
841 
843 
844  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
845  {
846  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT, 1);
847  tcp_worker_stats_inc (wrk, no_buffer, 1);
848  return;
849  }
850 
851  tc->rtt_ts = tcp_time_now_us (tc->c_thread_index);
852  b = vlib_get_buffer (vm, bi);
853  tcp_init_buffer (vm, b);
854  tcp_make_synack (tc, b);
855  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
856  TCP_EVT (TCP_EVT_SYNACK_SENT, tc);
857 }
858 
859 /**
860  * Send FIN
861  */
862 void
864 {
865  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
866  vlib_main_t *vm = wrk->vm;
867  vlib_buffer_t *b;
868  u32 bi;
869  u8 fin_snt = 0;
870 
871  fin_snt = tc->flags & TCP_CONN_FINSNT;
872  if (fin_snt)
873  tc->snd_nxt -= 1;
874 
875  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
876  {
877  /* Out of buffers so program fin retransmit ASAP */
878  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT, 1);
879  if (fin_snt)
880  tc->snd_nxt += 1;
881  else
882  /* Make sure retransmit retries a fin not data */
883  tc->flags |= TCP_CONN_FINSNT;
884  tcp_worker_stats_inc (wrk, no_buffer, 1);
885  return;
886  }
887 
888  /* If we have non-dupacks programmed, no need to send them */
889  if ((tc->flags & TCP_CONN_SNDACK) && !tc->pending_dupacks)
890  tc->flags &= ~TCP_CONN_SNDACK;
891 
893  b = vlib_get_buffer (vm, bi);
894  tcp_init_buffer (vm, b);
895  tcp_make_fin (tc, b);
896  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
897  TCP_EVT (TCP_EVT_FIN_SENT, tc);
898  /* Account for the FIN */
899  tc->snd_nxt += 1;
900  if (!fin_snt)
901  {
902  tc->flags |= TCP_CONN_FINSNT;
903  tc->flags &= ~TCP_CONN_FINPNDG;
904  tc->snd_una_max = seq_max (tc->snd_una_max, tc->snd_nxt);
905  }
906 }
907 
908 /**
909  * Push TCP header and update connection variables. Should only be called
910  * for segments with data, not for 'control' packets.
911  */
912 always_inline void
914  u8 compute_opts, u8 maybe_burst, u8 update_snd_nxt)
915 {
916  u8 tcp_hdr_opts_len, flags = TCP_FLAG_ACK;
917  u32 advertise_wnd, data_len;
918  tcp_main_t *tm = &tcp_main;
919  tcp_header_t *th;
920 
921  data_len = b->current_length;
922  if (PREDICT_FALSE (b->flags & VLIB_BUFFER_NEXT_PRESENT))
924 
925  vnet_buffer (b)->tcp.flags = 0;
926  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
927 
928  if (compute_opts)
929  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
930 
931  tcp_hdr_opts_len = tc->snd_opts_len + sizeof (tcp_header_t);
932 
933  if (maybe_burst)
934  advertise_wnd = tc->rcv_wnd >> tc->rcv_wscale;
935  else
936  advertise_wnd = tcp_window_to_advertise (tc, TCP_STATE_ESTABLISHED);
937 
938  if (PREDICT_FALSE (tc->flags & TCP_CONN_PSH_PENDING))
939  {
940  if (seq_geq (tc->psh_seq, snd_nxt)
941  && seq_lt (tc->psh_seq, snd_nxt + data_len))
942  flags |= TCP_FLAG_PSH;
943  }
944  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, snd_nxt,
945  tc->rcv_nxt, tcp_hdr_opts_len, flags,
946  advertise_wnd);
947 
948  if (maybe_burst)
949  {
950  clib_memcpy_fast ((u8 *) (th + 1),
951  tm->wrk_ctx[tc->c_thread_index].cached_opts,
952  tc->snd_opts_len);
953  }
954  else
955  {
956  u8 len = tcp_options_write ((u8 *) (th + 1), &tc->snd_opts);
957  ASSERT (len == tc->snd_opts_len);
958  }
959 
960  /*
961  * Update connection variables
962  */
963 
964  if (update_snd_nxt)
965  tc->snd_nxt += data_len;
966  tc->rcv_las = tc->rcv_nxt;
967 
968  tc->bytes_out += data_len;
969  tc->data_segs_out += 1;
970 
971  th->checksum = tcp_compute_checksum (tc, b);
972 
973  TCP_EVT (TCP_EVT_PKTIZE, tc);
974 }
975 
978 {
980  if (PREDICT_FALSE (b->flags & VLIB_BUFFER_NEXT_PRESENT))
982  return data_len;
983 }
984 
985 u32
987 {
988  tcp_connection_t *tc = (tcp_connection_t *) tconn;
989 
990  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
992 
993  tcp_push_hdr_i (tc, b, tc->snd_nxt, /* compute opts */ 0, /* burst */ 1,
994  /* update_snd_nxt */ 1);
995 
996  tc->snd_una_max = seq_max (tc->snd_nxt, tc->snd_una_max);
997  tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una);
998  /* If not tracking an ACK, start tracking */
999  if (tc->rtt_ts == 0 && !tcp_in_cong_recovery (tc))
1000  {
1001  tc->rtt_ts = tcp_time_now_us (tc->c_thread_index);
1002  tc->rtt_seq = tc->snd_nxt;
1003  }
1004  if (PREDICT_FALSE (!tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT)))
1005  {
1006  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1008  tc->rto_boff = 0;
1009  }
1010  tcp_trajectory_add_start (b, 3);
1011  return 0;
1012 }
1013 
1014 void
1016 {
1017  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1018  vlib_main_t *vm = wrk->vm;
1019  vlib_buffer_t *b;
1020  u32 bi;
1021 
1022  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1023  {
1024  tcp_update_rcv_wnd (tc);
1025  tcp_worker_stats_inc (wrk, no_buffer, 1);
1026  return;
1027  }
1028  b = vlib_get_buffer (vm, bi);
1029  tcp_init_buffer (vm, b);
1030  tcp_make_ack (tc, b);
1031  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1032 }
1033 
1034 void
1036 {
1037  if (!(tc->flags & TCP_CONN_SNDACK))
1038  {
1039  session_add_self_custom_tx_evt (&tc->connection, 1);
1040  tc->flags |= TCP_CONN_SNDACK;
1041  }
1042 }
1043 
1044 void
1046 {
1047  if (!(tc->flags & TCP_CONN_SNDACK))
1048  {
1049  session_add_self_custom_tx_evt (&tc->connection, 1);
1050  tc->flags |= TCP_CONN_SNDACK;
1051  }
1052  if (tc->pending_dupacks < 255)
1053  tc->pending_dupacks += 1;
1054 }
1055 
1056 void
1058 {
1059  if (!(tc->flags & TCP_CONN_RXT_PENDING))
1060  {
1061  session_add_self_custom_tx_evt (&tc->connection, 0);
1062  tc->flags |= TCP_CONN_RXT_PENDING;
1063  }
1064 }
1065 
1066 /**
1067  * Send window update ack
1068  *
1069  * Ensures that it will be sent only once, after a zero rwnd has been
1070  * advertised in a previous ack, and only if rwnd has grown beyond a
1071  * configurable value.
1072  */
1073 void
1075 {
1076  if (tcp_zero_rwnd_sent (tc))
1077  {
1078  tcp_update_rcv_wnd (tc);
1079  if (tc->rcv_wnd >= tcp_cfg.rwnd_min_update_ack * tc->snd_mss)
1080  {
1082  tcp_program_ack (tc);
1083  }
1084  }
1085 }
1086 
1087 /**
1088  * Allocate a new buffer and build a new tcp segment
1089  *
1090  * @param wrk tcp worker
1091  * @param tc connection for which the segment will be allocated
1092  * @param offset offset of the first byte in the tx fifo
1093  * @param max_deq_byte segment size
1094  * @param[out] b pointer to buffer allocated
1095  *
1096  * @return the number of bytes in the segment or 0 if buffer cannot be
1097  * allocated or no data available
1098  */
1099 static int
1101  u32 offset, u32 max_deq_bytes, vlib_buffer_t ** b)
1102 {
1103  u32 bytes_per_buffer = vnet_get_tcp_main ()->bytes_per_buffer;
1104  vlib_main_t *vm = wrk->vm;
1105  u32 bi, seg_size;
1106  int n_bytes = 0;
1107  u8 *data;
1108 
1109  seg_size = max_deq_bytes + TRANSPORT_MAX_HDRS_LEN;
1110 
1111  /*
1112  * Prepare options
1113  */
1114  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
1115 
1116  /*
1117  * Allocate and fill in buffer(s)
1118  */
1119 
1120  /* Easy case, buffer size greater than mss */
1121  if (PREDICT_TRUE (seg_size <= bytes_per_buffer))
1122  {
1123  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1124  {
1125  tcp_worker_stats_inc (wrk, no_buffer, 1);
1126  return 0;
1127  }
1128  *b = vlib_get_buffer (vm, bi);
1129  data = tcp_init_buffer (vm, *b);
1130  n_bytes = session_tx_fifo_peek_bytes (&tc->connection, data, offset,
1131  max_deq_bytes);
1132  ASSERT (n_bytes == max_deq_bytes);
1133  b[0]->current_length = n_bytes;
1134  tcp_push_hdr_i (tc, *b, tc->snd_una + offset, /* compute opts */ 0,
1135  /* burst */ 0, /* update_snd_nxt */ 0);
1136  }
1137  /* Split mss into multiple buffers */
1138  else
1139  {
1140  u32 chain_bi = ~0, n_bufs_per_seg, n_bufs;
1141  u16 n_peeked, len_to_deq;
1142  vlib_buffer_t *chain_b, *prev_b;
1143  int i;
1144 
1145  /* Make sure we have enough buffers */
1146  n_bufs_per_seg = ceil ((double) seg_size / bytes_per_buffer);
1147  vec_validate_aligned (wrk->tx_buffers, n_bufs_per_seg - 1,
1149  n_bufs = vlib_buffer_alloc (vm, wrk->tx_buffers, n_bufs_per_seg);
1150  if (PREDICT_FALSE (n_bufs != n_bufs_per_seg))
1151  {
1152  if (n_bufs)
1153  vlib_buffer_free (vm, wrk->tx_buffers, n_bufs);
1154  tcp_worker_stats_inc (wrk, no_buffer, 1);
1155  return 0;
1156  }
1157 
1158  *b = vlib_get_buffer (vm, wrk->tx_buffers[--n_bufs]);
1159  data = tcp_init_buffer (vm, *b);
1160  n_bytes = session_tx_fifo_peek_bytes (&tc->connection, data, offset,
1161  bytes_per_buffer -
1162  TRANSPORT_MAX_HDRS_LEN);
1163  b[0]->current_length = n_bytes;
1164  b[0]->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID;
1166  max_deq_bytes -= n_bytes;
1167 
1168  chain_b = *b;
1169  for (i = 1; i < n_bufs_per_seg; i++)
1170  {
1171  prev_b = chain_b;
1172  len_to_deq = clib_min (max_deq_bytes, bytes_per_buffer);
1173  chain_bi = wrk->tx_buffers[--n_bufs];
1174  chain_b = vlib_get_buffer (vm, chain_bi);
1175  chain_b->current_data = 0;
1176  data = vlib_buffer_get_current (chain_b);
1177  n_peeked = session_tx_fifo_peek_bytes (&tc->connection, data,
1178  offset + n_bytes,
1179  len_to_deq);
1180  ASSERT (n_peeked == len_to_deq);
1181  n_bytes += n_peeked;
1182  chain_b->current_length = n_peeked;
1183  chain_b->next_buffer = 0;
1184 
1185  /* update previous buffer */
1186  prev_b->next_buffer = chain_bi;
1187  prev_b->flags |= VLIB_BUFFER_NEXT_PRESENT;
1188 
1189  max_deq_bytes -= n_peeked;
1190  b[0]->total_length_not_including_first_buffer += n_peeked;
1191  }
1192 
1193  tcp_push_hdr_i (tc, *b, tc->snd_una + offset, /* compute opts */ 0,
1194  /* burst */ 0, /* update_snd_nxt */ 0);
1195 
1196  if (PREDICT_FALSE (n_bufs))
1197  {
1198  clib_warning ("not all buffers consumed");
1199  vlib_buffer_free (vm, wrk->tx_buffers, n_bufs);
1200  }
1201  }
1202 
1203  ASSERT (n_bytes > 0);
1204  ASSERT (((*b)->current_data + (*b)->current_length) <= bytes_per_buffer);
1205 
1206  return n_bytes;
1207 }
1208 
1209 /**
1210  * Build a retransmit segment
1211  *
1212  * @return the number of bytes in the segment or 0 if there's nothing to
1213  * retransmit
1214  */
1215 static u32
1217  tcp_connection_t * tc, u32 offset,
1218  u32 max_deq_bytes, vlib_buffer_t ** b)
1219 {
1220  u32 start, available_bytes;
1221  int n_bytes = 0;
1222 
1223  ASSERT (tc->state >= TCP_STATE_ESTABLISHED);
1224  ASSERT (max_deq_bytes != 0);
1225 
1226  /*
1227  * Make sure we can retransmit something
1228  */
1229  available_bytes = transport_max_tx_dequeue (&tc->connection);
1230  ASSERT (available_bytes >= offset);
1231  available_bytes -= offset;
1232  if (!available_bytes)
1233  return 0;
1234 
1235  max_deq_bytes = clib_min (tc->snd_mss, max_deq_bytes);
1236  max_deq_bytes = clib_min (available_bytes, max_deq_bytes);
1237 
1238  start = tc->snd_una + offset;
1239  ASSERT (seq_leq (start + max_deq_bytes, tc->snd_nxt));
1240 
1241  n_bytes = tcp_prepare_segment (wrk, tc, offset, max_deq_bytes, b);
1242  if (!n_bytes)
1243  return 0;
1244 
1245  tc->snd_rxt_bytes += n_bytes;
1246 
1247  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1248  tcp_bt_track_rxt (tc, start, start + n_bytes);
1249 
1250  tc->bytes_retrans += n_bytes;
1251  tc->segs_retrans += 1;
1252  tcp_worker_stats_inc (wrk, rxt_segs, 1);
1253  TCP_EVT (TCP_EVT_CC_RTX, tc, offset, n_bytes);
1254 
1255  return n_bytes;
1256 }
1257 
1258 static void
1260 {
1261  sack_scoreboard_t *sb = &tc->sack_sb;
1262  sack_scoreboard_hole_t *hole;
1263 
1264  hole = scoreboard_first_hole (sb);
1265  if (!sb->is_reneging && (!hole || hole->start == tc->snd_una))
1266  return;
1267 
1268  scoreboard_clear_reneging (sb, tc->snd_una, tc->snd_nxt);
1269 }
1270 
1271 /**
1272  * Reset congestion control, switch cwnd to loss window and try again.
1273  */
1274 static void
1276 {
1277  TCP_EVT (TCP_EVT_CC_EVT, tc, 6);
1278 
1279  tc->prev_ssthresh = tc->ssthresh;
1280  tc->prev_cwnd = tc->cwnd;
1281 
1282  /* If we entrered loss without fast recovery, notify cc algo of the
1283  * congestion event such that it can update ssthresh and its state */
1284  if (!tcp_in_fastrecovery (tc))
1285  tcp_cc_congestion (tc);
1286 
1287  /* Let cc algo decide loss cwnd and ssthresh post unrecovered loss */
1288  tcp_cc_loss (tc);
1289 
1290  tc->rtt_ts = 0;
1291  tc->cwnd_acc_bytes = 0;
1292  tc->tr_occurences += 1;
1293  tc->sack_sb.reorder = TCP_DUPACK_THRESHOLD;
1294  tcp_recovery_on (tc);
1295 }
1296 
1297 void
1299 {
1300  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1301  vlib_main_t *vm = wrk->vm;
1302  vlib_buffer_t *b = 0;
1303  u32 bi, n_bytes;
1304 
1305  tcp_worker_stats_inc (wrk, tr_events, 1);
1306 
1307  /* Should be handled by a different handler */
1308  if (PREDICT_FALSE (tc->state == TCP_STATE_SYN_SENT))
1309  return;
1310 
1311  /* Wait-close and retransmit could pop at the same time */
1312  if (tc->state == TCP_STATE_CLOSED)
1313  return;
1314 
1315  if (tc->state >= TCP_STATE_ESTABLISHED)
1316  {
1317  TCP_EVT (TCP_EVT_CC_EVT, tc, 2);
1318 
1319  /* Lost FIN, retransmit and return */
1320  if (tc->flags & TCP_CONN_FINSNT)
1321  {
1322  tcp_send_fin (tc);
1323  tc->rto_boff += 1;
1324  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1325  return;
1326  }
1327 
1328  /* Shouldn't be here. This condition is tricky because it has to take
1329  * into account boff > 0 due to persist timeout. */
1330  if ((tc->rto_boff == 0 && tc->snd_una == tc->snd_nxt)
1331  || (tc->rto_boff > 0 && seq_geq (tc->snd_una, tc->snd_congestion)
1332  && !tcp_flight_size (tc)))
1333  {
1334  ASSERT (!tcp_in_recovery (tc));
1335  tc->rto_boff = 0;
1336  return;
1337  }
1338 
1339  /* We're not in recovery so make sure rto_boff is 0. Can be non 0 due
1340  * to persist timer timeout */
1341  if (!tcp_in_recovery (tc) && tc->rto_boff > 0)
1342  {
1343  tc->rto_boff = 0;
1344  tcp_update_rto (tc);
1345  }
1346 
1347  /* Peer is dead or network connectivity is lost. Close connection.
1348  * RFC 1122 section 4.2.3.5 recommends a value of at least 100s. For
1349  * a min rto of 0.2s we need to retry about 8 times. */
1350  if (tc->rto_boff >= TCP_RTO_BOFF_MAX)
1351  {
1352  tcp_send_reset (tc);
1353  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1354  session_transport_closing_notify (&tc->connection);
1355  session_transport_closed_notify (&tc->connection);
1357  tcp_program_cleanup (wrk, tc);
1358  tcp_worker_stats_inc (wrk, tr_abort, 1);
1359  return;
1360  }
1361 
1362  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1364 
1365  /* Update send congestion to make sure that rxt has data to send */
1366  tc->snd_congestion = tc->snd_nxt;
1367 
1368  /* Send the first unacked segment. If we're short on buffers, return
1369  * as soon as possible */
1370  n_bytes = clib_min (tc->snd_mss, tc->snd_nxt - tc->snd_una);
1371  n_bytes = tcp_prepare_retransmit_segment (wrk, tc, 0, n_bytes, &b);
1372  if (!n_bytes)
1373  {
1374  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT, 1);
1375  return;
1376  }
1377 
1378  bi = vlib_get_buffer_index (vm, b);
1379  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1380 
1381  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1383 
1384  tc->rto_boff += 1;
1385  if (tc->rto_boff == 1)
1386  {
1388  /* Record timestamp. Eifel detection algorithm RFC3522 */
1389  tc->snd_rxt_ts = tcp_tstamp (tc);
1390  }
1391 
1392  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1393  scoreboard_init_rxt (&tc->sack_sb, tc->snd_una + n_bytes);
1394 
1396  }
1397  /* Retransmit SYN-ACK */
1398  else if (tc->state == TCP_STATE_SYN_RCVD)
1399  {
1400  TCP_EVT (TCP_EVT_CC_EVT, tc, 2);
1401 
1402  tc->rtt_ts = 0;
1403 
1404  /* Passive open establish timeout */
1405  if (tc->rto > TCP_ESTABLISH_TIME >> 1)
1406  {
1407  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1409  tcp_program_cleanup (wrk, tc);
1410  tcp_worker_stats_inc (wrk, tr_abort, 1);
1411  return;
1412  }
1413 
1414  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1415  {
1416  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT, 1);
1417  tcp_worker_stats_inc (wrk, no_buffer, 1);
1418  return;
1419  }
1420 
1421  tc->rto_boff += 1;
1422  if (tc->rto_boff > TCP_RTO_SYN_RETRIES)
1423  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1424 
1426 
1427  b = vlib_get_buffer (vm, bi);
1428  tcp_init_buffer (vm, b);
1429  tcp_make_synack (tc, b);
1430  TCP_EVT (TCP_EVT_SYN_RXT, tc, 1);
1431 
1432  /* Retransmit timer already updated, just enqueue to output */
1433  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1434  }
1435  else
1436  {
1437  ASSERT (tc->state == TCP_STATE_CLOSED);
1438  return;
1439  }
1440 }
1441 
1442 /**
1443  * SYN retransmit timer handler. Active open only.
1444  */
1445 void
1447 {
1448  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1449  vlib_main_t *vm = wrk->vm;
1450  vlib_buffer_t *b = 0;
1451  u32 bi;
1452 
1453  /* Note: the connection may have transitioned to ESTABLISHED... */
1454  if (PREDICT_FALSE (tc->state != TCP_STATE_SYN_SENT))
1455  return;
1456 
1457  /* Half-open connection actually moved to established but we were
1458  * waiting for syn retransmit to pop to call cleanup from the right
1459  * thread. */
1460  if (tc->flags & TCP_CONN_HALF_OPEN_DONE)
1461  {
1463  TCP_DBG ("could not remove half-open connection");
1464  return;
1465  }
1466 
1467  TCP_EVT (TCP_EVT_CC_EVT, tc, 2);
1468  tc->rtt_ts = 0;
1469 
1470  /* Active open establish timeout */
1471  if (tc->rto >= TCP_ESTABLISH_TIME >> 1)
1472  {
1473  session_stream_connect_notify (&tc->connection, SESSION_E_TIMEDOUT);
1475  return;
1476  }
1477 
1478  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1479  {
1480  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN, 1);
1481  tcp_worker_stats_inc (wrk, no_buffer, 1);
1482  return;
1483  }
1484 
1485  /* Try without increasing RTO a number of times. If this fails,
1486  * start growing RTO exponentially */
1487  tc->rto_boff += 1;
1488  if (tc->rto_boff > TCP_RTO_SYN_RETRIES)
1489  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1490 
1491  b = vlib_get_buffer (vm, bi);
1492  tcp_init_buffer (vm, b);
1493  tcp_make_syn (tc, b);
1494 
1495  TCP_EVT (TCP_EVT_SYN_RXT, tc, 0);
1496 
1497  /* This goes straight to ipx_lookup */
1498  tcp_push_ip_hdr (wrk, tc, b);
1499  tcp_enqueue_to_ip_lookup (wrk, b, bi, tc->c_is_ip4, tc->c_fib_index);
1500 
1501  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN,
1502  tc->rto * TCP_TO_TIMER_TICK);
1503 }
1504 
1505 /**
1506  * Got 0 snd_wnd from peer, try to do something about it.
1507  *
1508  */
1509 void
1511 {
1512  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1513  u32 bi, max_snd_bytes, available_bytes, offset;
1514  tcp_main_t *tm = vnet_get_tcp_main ();
1515  vlib_main_t *vm = wrk->vm;
1516  vlib_buffer_t *b;
1517  int n_bytes = 0;
1518  u8 *data;
1519 
1520  /* Problem already solved or worse */
1521  if (tc->state == TCP_STATE_CLOSED || tc->snd_wnd > tc->snd_mss
1522  || (tc->flags & TCP_CONN_FINSNT))
1523  goto update_scheduler;
1524 
1525  available_bytes = transport_max_tx_dequeue (&tc->connection);
1526  offset = tc->snd_nxt - tc->snd_una;
1527 
1528  /* Reprogram persist if no new bytes available to send. We may have data
1529  * next time */
1530  if (!available_bytes)
1531  {
1532  tcp_persist_timer_set (&wrk->timer_wheel, tc);
1533  return;
1534  }
1535 
1536  if (available_bytes <= offset)
1537  goto update_scheduler;
1538 
1539  /* Increment RTO backoff */
1540  tc->rto_boff += 1;
1541  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1542 
1543  /*
1544  * Try to force the first unsent segment (or buffer)
1545  */
1546  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1547  {
1548  tcp_persist_timer_set (&wrk->timer_wheel, tc);
1549  tcp_worker_stats_inc (wrk, no_buffer, 1);
1550  return;
1551  }
1552 
1553  b = vlib_get_buffer (vm, bi);
1554  data = tcp_init_buffer (vm, b);
1555 
1556  tcp_validate_txf_size (tc, offset);
1557  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
1558  max_snd_bytes = clib_min (tc->snd_mss,
1559  tm->bytes_per_buffer - TRANSPORT_MAX_HDRS_LEN);
1560  n_bytes = session_tx_fifo_peek_bytes (&tc->connection, data, offset,
1561  max_snd_bytes);
1562  b->current_length = n_bytes;
1563  ASSERT (n_bytes != 0 && (tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT)
1564  || tc->snd_nxt == tc->snd_una_max
1565  || tc->rto_boff > 1));
1566 
1567  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1568  {
1570  tcp_bt_track_tx (tc, n_bytes);
1571  }
1572 
1573  tcp_push_hdr_i (tc, b, tc->snd_nxt, /* compute opts */ 0,
1574  /* burst */ 0, /* update_snd_nxt */ 1);
1575  tc->snd_una_max = seq_max (tc->snd_nxt, tc->snd_una_max);
1576  tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una);
1577  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1578 
1579  /* Just sent new data, enable retransmit */
1581 
1582  return;
1583 
1584 update_scheduler:
1585 
1586  if (tcp_is_descheduled (tc))
1587  transport_connection_reschedule (&tc->connection);
1588 }
1589 
1590 /**
1591  * Retransmit first unacked segment
1592  */
1593 int
1595 {
1596  vlib_main_t *vm = wrk->vm;
1597  vlib_buffer_t *b;
1598  u32 bi, n_bytes;
1599 
1600  TCP_EVT (TCP_EVT_CC_EVT, tc, 1);
1601 
1602  n_bytes = tcp_prepare_retransmit_segment (wrk, tc, 0, tc->snd_mss, &b);
1603  if (!n_bytes)
1604  return -1;
1605 
1606  bi = vlib_get_buffer_index (vm, b);
1607  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1608 
1609  return 0;
1610 }
1611 
1612 static int
1614  u32 burst_size)
1615 {
1616  u32 offset, n_segs = 0, n_written, bi, available_wnd;
1617  vlib_main_t *vm = wrk->vm;
1618  vlib_buffer_t *b = 0;
1619 
1620  offset = tc->snd_nxt - tc->snd_una;
1621  available_wnd = tc->snd_wnd - offset;
1622  burst_size = clib_min (burst_size, available_wnd / tc->snd_mss);
1623 
1624  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1626 
1627  while (n_segs < burst_size)
1628  {
1629  n_written = tcp_prepare_segment (wrk, tc, offset, tc->snd_mss, &b);
1630  if (!n_written)
1631  goto done;
1632 
1633  bi = vlib_get_buffer_index (vm, b);
1634  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1635  offset += n_written;
1636  n_segs += 1;
1637 
1638  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1639  tcp_bt_track_tx (tc, n_written);
1640 
1641  tc->snd_nxt += n_written;
1642  tc->snd_una_max = seq_max (tc->snd_nxt, tc->snd_una_max);
1643  }
1644 
1645 done:
1646  return n_segs;
1647 }
1648 
1649 /**
1650  * Estimate send space using proportional rate reduction (RFC6937)
1651  */
1652 int
1654 {
1655  u32 pipe, prr_out;
1656  int space;
1657 
1658  pipe = tcp_flight_size (tc);
1659  prr_out = tc->snd_rxt_bytes + (tc->snd_nxt - tc->snd_congestion);
1660 
1661  if (pipe > tc->ssthresh)
1662  {
1663  space = ((int) tc->prr_delivered * ((f64) tc->ssthresh / tc->prev_cwnd))
1664  - prr_out;
1665  }
1666  else
1667  {
1668  int limit;
1669  limit = clib_max ((int) (tc->prr_delivered - prr_out), 0) + tc->snd_mss;
1670  space = clib_min (tc->ssthresh - pipe, limit);
1671  }
1672  space = clib_max (space, prr_out ? 0 : tc->snd_mss);
1673  return space;
1674 }
1675 
1676 static inline u8
1678  sack_scoreboard_t * sb)
1679 {
1680  u32 tx_adv_sack = sb->high_sacked - tc->snd_congestion;
1681  f64 rr = (f64) tc->ssthresh / tc->prev_cwnd;
1682 
1683  if (tcp_fastrecovery_first (tc))
1684  return 1;
1685 
1686  return (tx_adv_sack > (tc->snd_una - tc->prr_start) * rr);
1687 }
1688 
1689 static inline u8
1691 {
1692  return (transport_max_tx_dequeue (&tc->connection)
1693  - (tc->snd_nxt - tc->snd_una));
1694 }
1695 
1696 #define scoreboard_rescue_rxt_valid(_sb, _tc) \
1697  (seq_geq (_sb->rescue_rxt, _tc->snd_una) \
1698  && seq_leq (_sb->rescue_rxt, _tc->snd_congestion))
1699 
1700 /**
1701  * Do retransmit with SACKs
1702  */
1703 static int
1705  u32 burst_size)
1706 {
1707  u32 n_written = 0, offset, max_bytes, n_segs = 0;
1708  u8 snd_limited = 0, can_rescue = 0;
1709  u32 bi, max_deq, burst_bytes;
1710  sack_scoreboard_hole_t *hole;
1711  vlib_main_t *vm = wrk->vm;
1712  vlib_buffer_t *b = 0;
1713  sack_scoreboard_t *sb;
1714  int snd_space;
1715 
1717 
1718  burst_bytes = transport_connection_tx_pacer_burst (&tc->connection);
1719  burst_size = clib_min (burst_size, burst_bytes / tc->snd_mss);
1720  if (!burst_size)
1721  {
1723  return 0;
1724  }
1725 
1726  if (tcp_in_recovery (tc))
1727  snd_space = tcp_available_cc_snd_space (tc);
1728  else
1729  snd_space = tcp_fastrecovery_prr_snd_space (tc);
1730 
1731  if (snd_space < tc->snd_mss)
1732  goto done;
1733 
1734  sb = &tc->sack_sb;
1735 
1736  /* Check if snd_una is a lost retransmit */
1737  if (pool_elts (sb->holes)
1738  && seq_gt (sb->high_sacked, tc->snd_congestion)
1739  && tc->rxt_head != tc->snd_una
1741  {
1742  max_bytes = clib_min (tc->snd_mss, tc->snd_congestion - tc->snd_una);
1743  n_written = tcp_prepare_retransmit_segment (wrk, tc, 0, max_bytes, &b);
1744  if (!n_written)
1745  {
1747  goto done;
1748  }
1749  bi = vlib_get_buffer_index (vm, b);
1750  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1751  n_segs = 1;
1752 
1753  tc->rxt_head = tc->snd_una;
1754  tc->rxt_delivered += n_written;
1755  tc->prr_delivered += n_written;
1756  ASSERT (tc->rxt_delivered <= tc->snd_rxt_bytes);
1757  }
1758 
1760 
1761  TCP_EVT (TCP_EVT_CC_EVT, tc, 0);
1762  hole = scoreboard_get_hole (sb, sb->cur_rxt_hole);
1763 
1764  max_deq = transport_max_tx_dequeue (&tc->connection);
1765  max_deq -= tc->snd_nxt - tc->snd_una;
1766 
1767  while (snd_space > 0 && n_segs < burst_size)
1768  {
1769  hole = scoreboard_next_rxt_hole (sb, hole, max_deq != 0, &can_rescue,
1770  &snd_limited);
1771  if (!hole)
1772  {
1773  /* We are out of lost holes to retransmit so send some new data. */
1774  if (max_deq > tc->snd_mss)
1775  {
1776  u32 n_segs_new;
1777  int av_wnd;
1778 
1779  /* Make sure we don't exceed available window and leave space
1780  * for one more packet, to avoid zero window acks */
1781  av_wnd = (int) tc->snd_wnd - (tc->snd_nxt - tc->snd_una);
1782  av_wnd = clib_max (av_wnd - tc->snd_mss, 0);
1783  snd_space = clib_min (snd_space, av_wnd);
1784  snd_space = clib_min (max_deq, snd_space);
1785  burst_size = clib_min (burst_size - n_segs,
1786  snd_space / tc->snd_mss);
1787  burst_size = clib_min (burst_size, TCP_RXT_MAX_BURST);
1788  n_segs_new = tcp_transmit_unsent (wrk, tc, burst_size);
1789  if (max_deq > n_segs_new * tc->snd_mss)
1791 
1792  n_segs += n_segs_new;
1793  goto done;
1794  }
1795 
1796  if (tcp_in_recovery (tc) || !can_rescue
1797  || scoreboard_rescue_rxt_valid (sb, tc))
1798  break;
1799 
1800  /* If rescue rxt undefined or less than snd_una then one segment of
1801  * up to SMSS octets that MUST include the highest outstanding
1802  * unSACKed sequence number SHOULD be returned, and RescueRxt set to
1803  * RecoveryPoint. HighRxt MUST NOT be updated.
1804  */
1805  hole = scoreboard_last_hole (sb);
1806  max_bytes = clib_min (tc->snd_mss, hole->end - hole->start);
1807  max_bytes = clib_min (max_bytes, snd_space);
1808  offset = hole->end - tc->snd_una - max_bytes;
1809  n_written = tcp_prepare_retransmit_segment (wrk, tc, offset,
1810  max_bytes, &b);
1811  if (!n_written)
1812  goto done;
1813 
1814  sb->rescue_rxt = tc->snd_congestion;
1815  bi = vlib_get_buffer_index (vm, b);
1816  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1817  n_segs += 1;
1818  break;
1819  }
1820 
1821  max_bytes = clib_min (hole->end - sb->high_rxt, snd_space);
1822  max_bytes = snd_limited ? clib_min (max_bytes, tc->snd_mss) : max_bytes;
1823  if (max_bytes == 0)
1824  break;
1825 
1826  offset = sb->high_rxt - tc->snd_una;
1827  n_written = tcp_prepare_retransmit_segment (wrk, tc, offset, max_bytes,
1828  &b);
1829  ASSERT (n_written <= snd_space);
1830 
1831  /* Nothing left to retransmit */
1832  if (n_written == 0)
1833  break;
1834 
1835  bi = vlib_get_buffer_index (vm, b);
1836  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1837 
1838  sb->high_rxt += n_written;
1839  ASSERT (seq_leq (sb->high_rxt, tc->snd_nxt));
1840 
1841  snd_space -= n_written;
1842  n_segs += 1;
1843  }
1844 
1845  if (hole)
1847 
1848 done:
1849 
1850  transport_connection_tx_pacer_reset_bucket (&tc->connection, 0);
1851  return n_segs;
1852 }
1853 
1854 /**
1855  * Fast retransmit without SACK info
1856  */
1857 static int
1859  u32 burst_size)
1860 {
1861  u32 n_written = 0, offset = 0, bi, max_deq, n_segs_now, max_bytes;
1862  u32 burst_bytes, sent_bytes;
1863  vlib_main_t *vm = wrk->vm;
1864  int snd_space, n_segs = 0;
1865  u8 cc_limited = 0;
1866  vlib_buffer_t *b;
1867 
1869  TCP_EVT (TCP_EVT_CC_EVT, tc, 0);
1870 
1871  burst_bytes = transport_connection_tx_pacer_burst (&tc->connection);
1872  burst_size = clib_min (burst_size, burst_bytes / tc->snd_mss);
1873  if (!burst_size)
1874  {
1876  return 0;
1877  }
1878 
1879  snd_space = tcp_available_cc_snd_space (tc);
1880  cc_limited = snd_space < burst_bytes;
1881 
1882  if (!tcp_fastrecovery_first (tc))
1883  goto send_unsent;
1884 
1885  /* RFC 6582: [If a partial ack], retransmit the first unacknowledged
1886  * segment. */
1887  while (snd_space > 0 && n_segs < burst_size)
1888  {
1889  max_bytes = clib_min (tc->snd_mss,
1890  tc->snd_congestion - tc->snd_una - offset);
1891  if (!max_bytes)
1892  break;
1893  n_written = tcp_prepare_retransmit_segment (wrk, tc, offset, max_bytes,
1894  &b);
1895 
1896  /* Nothing left to retransmit */
1897  if (n_written == 0)
1898  break;
1899 
1900  bi = vlib_get_buffer_index (vm, b);
1901  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1902  snd_space -= n_written;
1903  offset += n_written;
1904  n_segs += 1;
1905  }
1906 
1907  if (n_segs == burst_size)
1908  goto done;
1909 
1910 send_unsent:
1911 
1912  /* RFC 6582: Send a new segment if permitted by the new value of cwnd. */
1913  if (snd_space < tc->snd_mss || tc->snd_mss == 0)
1914  goto done;
1915 
1916  max_deq = transport_max_tx_dequeue (&tc->connection);
1917  max_deq -= tc->snd_nxt - tc->snd_una;
1918  if (max_deq)
1919  {
1920  snd_space = clib_min (max_deq, snd_space);
1921  burst_size = clib_min (burst_size - n_segs, snd_space / tc->snd_mss);
1922  n_segs_now = tcp_transmit_unsent (wrk, tc, burst_size);
1923  if (n_segs_now && max_deq > n_segs_now * tc->snd_mss)
1925  n_segs += n_segs_now;
1926  }
1927 
1928 done:
1930 
1931  sent_bytes = clib_min (n_segs * tc->snd_mss, burst_bytes);
1932  sent_bytes = cc_limited ? burst_bytes : sent_bytes;
1933  transport_connection_tx_pacer_update_bytes (&tc->connection, sent_bytes);
1934 
1935  return n_segs;
1936 }
1937 
1938 static int
1939 tcp_send_acks (tcp_connection_t * tc, u32 max_burst_size)
1940 {
1941  int j, n_acks;
1942 
1943  if (!tc->pending_dupacks)
1944  {
1945  if (tcp_in_cong_recovery (tc) || !tcp_max_tx_deq (tc)
1946  || tc->state != TCP_STATE_ESTABLISHED)
1947  {
1948  tcp_send_ack (tc);
1949  return 1;
1950  }
1951  return 0;
1952  }
1953 
1954  /* If we're supposed to send dupacks but have no ooo data
1955  * send only one ack */
1956  if (!vec_len (tc->snd_sacks))
1957  {
1958  tcp_send_ack (tc);
1959  tc->dupacks_out += 1;
1960  tc->pending_dupacks = 0;
1961  return 1;
1962  }
1963 
1964  /* Start with first sack block */
1965  tc->snd_sack_pos = 0;
1966 
1967  /* Generate enough dupacks to cover all sack blocks. Do not generate
1968  * more sacks than the number of packets received. But do generate at
1969  * least 3, i.e., the number needed to signal congestion, if needed. */
1970  n_acks = vec_len (tc->snd_sacks) / TCP_OPTS_MAX_SACK_BLOCKS;
1971  n_acks = clib_min (n_acks, tc->pending_dupacks);
1972  n_acks = clib_max (n_acks, clib_min (tc->pending_dupacks, 3));
1973  for (j = 0; j < clib_min (n_acks, max_burst_size); j++)
1974  tcp_send_ack (tc);
1975 
1976  if (n_acks < max_burst_size)
1977  {
1978  tc->pending_dupacks = 0;
1979  tc->snd_sack_pos = 0;
1980  tc->dupacks_out += n_acks;
1981  return n_acks;
1982  }
1983  else
1984  {
1985  TCP_DBG ("constrained by burst size");
1986  tc->pending_dupacks = n_acks - max_burst_size;
1987  tc->dupacks_out += max_burst_size;
1988  tcp_program_dupack (tc);
1989  return max_burst_size;
1990  }
1991 }
1992 
1993 static int
1995 {
1996  tcp_worker_ctx_t *wrk;
1997  u32 n_segs;
1998 
1999  if (PREDICT_FALSE (tc->state == TCP_STATE_CLOSED))
2000  return 0;
2001 
2002  wrk = tcp_get_worker (tc->c_thread_index);
2003 
2004  if (tcp_opts_sack_permitted (&tc->rcv_opts))
2005  n_segs = tcp_retransmit_sack (wrk, tc, max_burst_size);
2006  else
2007  n_segs = tcp_retransmit_no_sack (wrk, tc, max_burst_size);
2008 
2009  return n_segs;
2010 }
2011 
2012 int
2014 {
2015  tcp_connection_t *tc = (tcp_connection_t *) conn;
2016  u32 n_segs = 0;
2017 
2018  if (tcp_in_cong_recovery (tc) && (tc->flags & TCP_CONN_RXT_PENDING))
2019  {
2020  tc->flags &= ~TCP_CONN_RXT_PENDING;
2021  n_segs = tcp_do_retransmit (tc, sp->max_burst_size);
2022  }
2023 
2024  if (!(tc->flags & TCP_CONN_SNDACK))
2025  return n_segs;
2026 
2027  tc->flags &= ~TCP_CONN_SNDACK;
2028 
2029  /* We have retransmitted packets and no dupack */
2030  if (n_segs && !tc->pending_dupacks)
2031  return n_segs;
2032 
2033  if (sp->max_burst_size <= n_segs)
2034  {
2035  tcp_program_ack (tc);
2036  return n_segs;
2037  }
2038 
2039  n_segs += tcp_send_acks (tc, sp->max_burst_size - n_segs);
2040 
2041  return n_segs;
2042 }
2043 #endif /* CLIB_MARCH_VARIANT */
2044 
2045 static void
2047  u16 * next0, u32 * error0)
2048 {
2049  ip_adjacency_t *adj;
2050  adj_index_t ai;
2051 
2052  /* Not thread safe but as long as the connection exists the adj should
2053  * not be removed */
2054  ai = adj_nbr_find (FIB_PROTOCOL_IP6, VNET_LINK_IP6, &tc0->c_rmt_ip,
2055  tc0->sw_if_index);
2056  if (ai == ADJ_INDEX_INVALID)
2057  {
2058  vnet_buffer (b0)->sw_if_index[VLIB_TX] = ~0;
2059  *next0 = TCP_OUTPUT_NEXT_DROP;
2060  *error0 = TCP_ERROR_LINK_LOCAL_RW;
2061  return;
2062  }
2063 
2064  adj = adj_get (ai);
2066  *next0 = TCP_OUTPUT_NEXT_IP_REWRITE;
2067  else if (adj->lookup_next_index == IP_LOOKUP_NEXT_ARP)
2068  *next0 = TCP_OUTPUT_NEXT_IP_ARP;
2069  else
2070  {
2071  *next0 = TCP_OUTPUT_NEXT_DROP;
2072  *error0 = TCP_ERROR_LINK_LOCAL_RW;
2073  }
2074  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = ai;
2075 }
2076 
2077 static void
2079  u32 * to_next, u32 n_bufs)
2080 {
2081  tcp_connection_t *tc;
2082  tcp_tx_trace_t *t;
2083  vlib_buffer_t *b;
2084  tcp_header_t *th;
2085  int i;
2086 
2087  for (i = 0; i < n_bufs; i++)
2088  {
2089  b = vlib_get_buffer (vm, to_next[i]);
2090  if (!(b->flags & VLIB_BUFFER_IS_TRACED))
2091  continue;
2092  th = vlib_buffer_get_current (b);
2093  tc = tcp_connection_get (vnet_buffer (b)->tcp.connection_index,
2094  vm->thread_index);
2095  t = vlib_add_trace (vm, node, b, sizeof (*t));
2096  clib_memcpy_fast (&t->tcp_header, th, sizeof (t->tcp_header));
2097  clib_memcpy_fast (&t->tcp_connection, tc, sizeof (t->tcp_connection));
2098  }
2099 }
2100 
2101 always_inline void
2103  tcp_connection_t * tc0, u8 is_ip4)
2104 {
2105  TCP_EVT (TCP_EVT_OUTPUT, tc0,
2107  b0->current_length);
2108 
2109  if (is_ip4)
2110  vlib_buffer_push_ip4 (vm, b0, &tc0->c_lcl_ip4, &tc0->c_rmt_ip4,
2111  IP_PROTOCOL_TCP, tcp_csum_offload (tc0));
2112  else
2113  vlib_buffer_push_ip6_custom (vm, b0, &tc0->c_lcl_ip6, &tc0->c_rmt_ip6,
2114  IP_PROTOCOL_TCP, tc0->ipv6_flow_label);
2115 }
2116 
2117 always_inline void
2119 {
2120  if (PREDICT_TRUE (!(tc->cfg_flags & TCP_CFG_F_TSO)))
2121  return;
2122 
2123  u16 data_len = b->current_length - sizeof (tcp_header_t) - tc->snd_opts_len;
2124 
2125  if (PREDICT_FALSE (b->flags & VLIB_BUFFER_TOTAL_LENGTH_VALID))
2127 
2128  if (PREDICT_TRUE (data_len <= tc->snd_mss))
2129  return;
2130  else
2131  {
2132  ASSERT ((b->flags & VNET_BUFFER_F_L3_HDR_OFFSET_VALID) != 0);
2133  ASSERT ((b->flags & VNET_BUFFER_F_L4_HDR_OFFSET_VALID) != 0);
2134  b->flags |= VNET_BUFFER_F_GSO;
2135  vnet_buffer2 (b)->gso_l4_hdr_sz =
2136  sizeof (tcp_header_t) + tc->snd_opts_len;
2137  vnet_buffer2 (b)->gso_size = tc->snd_mss;
2138  }
2139 }
2140 
2141 always_inline void
2143  vlib_node_runtime_t * error_node, u16 * next0,
2144  u8 is_ip4)
2145 {
2146  /* If next_index is not drop use it */
2147  if (tc0->next_node_index)
2148  {
2149  *next0 = tc0->next_node_index;
2150  vnet_buffer (b0)->tcp.next_node_opaque = tc0->next_node_opaque;
2151  }
2152  else
2153  {
2154  *next0 = TCP_OUTPUT_NEXT_IP_LOOKUP;
2155  }
2156 
2157  vnet_buffer (b0)->sw_if_index[VLIB_TX] = tc0->c_fib_index;
2158  vnet_buffer (b0)->sw_if_index[VLIB_RX] = 0;
2159 
2160  if (!is_ip4)
2161  {
2162  u32 error0 = 0;
2163 
2164  if (PREDICT_FALSE (ip6_address_is_link_local_unicast (&tc0->c_rmt_ip6)))
2165  tcp_output_handle_link_local (tc0, b0, next0, &error0);
2166 
2167  if (PREDICT_FALSE (error0))
2168  {
2169  b0->error = error_node->errors[error0];
2170  return;
2171  }
2172  }
2173 
2174  tc0->segs_out += 1;
2175 }
2176 
2179  vlib_frame_t * frame, int is_ip4)
2180 {
2181  u32 n_left_from, *from, thread_index = vm->thread_index;
2182  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
2183  u16 nexts[VLIB_FRAME_SIZE], *next;
2184 
2185  from = vlib_frame_vector_args (frame);
2186  n_left_from = frame->n_vectors;
2187  tcp_set_time_now (tcp_get_worker (thread_index));
2188 
2190  tcp46_output_trace_frame (vm, node, from, n_left_from);
2191 
2192  vlib_get_buffers (vm, from, bufs, n_left_from);
2193  b = bufs;
2194  next = nexts;
2195 
2196  while (n_left_from >= 4)
2197  {
2198  tcp_connection_t *tc0, *tc1;
2199 
2200  {
2201  vlib_prefetch_buffer_header (b[2], STORE);
2202  CLIB_PREFETCH (b[2]->data, 2 * CLIB_CACHE_LINE_BYTES, STORE);
2203 
2204  vlib_prefetch_buffer_header (b[3], STORE);
2205  CLIB_PREFETCH (b[3]->data, 2 * CLIB_CACHE_LINE_BYTES, STORE);
2206  }
2207 
2208  tc0 = tcp_connection_get (vnet_buffer (b[0])->tcp.connection_index,
2209  thread_index);
2210  tc1 = tcp_connection_get (vnet_buffer (b[1])->tcp.connection_index,
2211  thread_index);
2212 
2213  if (PREDICT_TRUE (!tc0 + !tc1 == 0))
2214  {
2215  tcp_output_push_ip (vm, b[0], tc0, is_ip4);
2216  tcp_output_push_ip (vm, b[1], tc1, is_ip4);
2217 
2218  tcp_check_if_gso (tc0, b[0]);
2219  tcp_check_if_gso (tc1, b[1]);
2220 
2221  tcp_output_handle_packet (tc0, b[0], node, &next[0], is_ip4);
2222  tcp_output_handle_packet (tc1, b[1], node, &next[1], is_ip4);
2223  }
2224  else
2225  {
2226  if (tc0 != 0)
2227  {
2228  tcp_output_push_ip (vm, b[0], tc0, is_ip4);
2229  tcp_check_if_gso (tc0, b[0]);
2230  tcp_output_handle_packet (tc0, b[0], node, &next[0], is_ip4);
2231  }
2232  else
2233  {
2234  b[0]->error = node->errors[TCP_ERROR_INVALID_CONNECTION];
2235  next[0] = TCP_OUTPUT_NEXT_DROP;
2236  }
2237  if (tc1 != 0)
2238  {
2239  tcp_output_push_ip (vm, b[1], tc1, is_ip4);
2240  tcp_check_if_gso (tc1, b[1]);
2241  tcp_output_handle_packet (tc1, b[1], node, &next[1], is_ip4);
2242  }
2243  else
2244  {
2245  b[1]->error = node->errors[TCP_ERROR_INVALID_CONNECTION];
2246  next[1] = TCP_OUTPUT_NEXT_DROP;
2247  }
2248  }
2249 
2250  b += 2;
2251  next += 2;
2252  n_left_from -= 2;
2253  }
2254  while (n_left_from > 0)
2255  {
2256  tcp_connection_t *tc0;
2257 
2258  if (n_left_from > 1)
2259  {
2260  vlib_prefetch_buffer_header (b[1], STORE);
2261  CLIB_PREFETCH (b[1]->data, 2 * CLIB_CACHE_LINE_BYTES, STORE);
2262  }
2263 
2264  tc0 = tcp_connection_get (vnet_buffer (b[0])->tcp.connection_index,
2265  thread_index);
2266 
2267  if (PREDICT_TRUE (tc0 != 0))
2268  {
2269  tcp_output_push_ip (vm, b[0], tc0, is_ip4);
2270  tcp_check_if_gso (tc0, b[0]);
2271  tcp_output_handle_packet (tc0, b[0], node, &next[0], is_ip4);
2272  }
2273  else
2274  {
2275  b[0]->error = node->errors[TCP_ERROR_INVALID_CONNECTION];
2276  next[0] = TCP_OUTPUT_NEXT_DROP;
2277  }
2278 
2279  b += 1;
2280  next += 1;
2281  n_left_from -= 1;
2282  }
2283 
2284  vlib_buffer_enqueue_to_next (vm, node, from, nexts, frame->n_vectors);
2285  vlib_node_increment_counter (vm, tcp_node_index (output, is_ip4),
2286  TCP_ERROR_PKTS_SENT, frame->n_vectors);
2287  return frame->n_vectors;
2288 }
2289 
2291  vlib_frame_t * from_frame)
2292 {
2293  return tcp46_output_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2294 }
2295 
2297  vlib_frame_t * from_frame)
2298 {
2299  return tcp46_output_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2300 }
2301 
2302 /* *INDENT-OFF* */
2304 {
2305  .name = "tcp4-output",
2306  /* Takes a vector of packets. */
2307  .vector_size = sizeof (u32),
2308  .n_errors = TCP_N_ERROR,
2309  .protocol_hint = VLIB_NODE_PROTO_HINT_TCP,
2310  .error_strings = tcp_error_strings,
2311  .n_next_nodes = TCP_OUTPUT_N_NEXT,
2312  .next_nodes = {
2313 #define _(s,n) [TCP_OUTPUT_NEXT_##s] = n,
2315 #undef _
2316  },
2317  .format_buffer = format_tcp_header,
2318  .format_trace = format_tcp_tx_trace,
2319 };
2320 /* *INDENT-ON* */
2321 
2322 /* *INDENT-OFF* */
2324 {
2325  .name = "tcp6-output",
2326  /* Takes a vector of packets. */
2327  .vector_size = sizeof (u32),
2328  .n_errors = TCP_N_ERROR,
2329  .protocol_hint = VLIB_NODE_PROTO_HINT_TCP,
2330  .error_strings = tcp_error_strings,
2331  .n_next_nodes = TCP_OUTPUT_N_NEXT,
2332  .next_nodes = {
2333 #define _(s,n) [TCP_OUTPUT_NEXT_##s] = n,
2335 #undef _
2336  },
2337  .format_buffer = format_tcp_header,
2338  .format_trace = format_tcp_tx_trace,
2339 };
2340 /* *INDENT-ON* */
2341 
2342 typedef enum _tcp_reset_next
2343 {
2348 
2349 #define foreach_tcp4_reset_next \
2350  _(DROP, "error-drop") \
2351  _(IP_LOOKUP, "ip4-lookup")
2352 
2353 #define foreach_tcp6_reset_next \
2354  _(DROP, "error-drop") \
2355  _(IP_LOOKUP, "ip6-lookup")
2356 
2357 static uword
2359  vlib_frame_t * from_frame, u8 is_ip4)
2360 {
2361  u32 error0 = TCP_ERROR_RST_SENT, next0 = TCP_RESET_NEXT_IP_LOOKUP;
2362  u32 n_left_from, next_index, *from, *to_next;
2363 
2364  from = vlib_frame_vector_args (from_frame);
2365  n_left_from = from_frame->n_vectors;
2366 
2367  next_index = node->cached_next_index;
2368 
2369  while (n_left_from > 0)
2370  {
2371  u32 n_left_to_next;
2372 
2373  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
2374 
2375  while (n_left_from > 0 && n_left_to_next > 0)
2376  {
2377  vlib_buffer_t *b0;
2378  tcp_tx_trace_t *t0;
2379  tcp_header_t *th0;
2380  u32 bi0;
2381 
2382  bi0 = from[0];
2383  to_next[0] = bi0;
2384  from += 1;
2385  to_next += 1;
2386  n_left_from -= 1;
2387  n_left_to_next -= 1;
2388 
2389  b0 = vlib_get_buffer (vm, bi0);
2390  tcp_make_reset_in_place (vm, b0, is_ip4);
2391 
2392  /* Prepare to send to IP lookup */
2393  vnet_buffer (b0)->sw_if_index[VLIB_TX] = ~0;
2394 
2395  b0->error = node->errors[error0];
2396  b0->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
2397  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
2398  {
2399  th0 = vlib_buffer_get_current (b0);
2400  if (is_ip4)
2401  th0 = ip4_next_header ((ip4_header_t *) th0);
2402  else
2403  th0 = ip6_next_header ((ip6_header_t *) th0);
2404  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
2405  clib_memcpy_fast (&t0->tcp_header, th0,
2406  sizeof (t0->tcp_header));
2407  }
2408 
2409  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
2410  n_left_to_next, bi0, next0);
2411  }
2412  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
2413  }
2414  return from_frame->n_vectors;
2415 }
2416 
2418  vlib_frame_t * from_frame)
2419 {
2420  return tcp46_send_reset_inline (vm, node, from_frame, 1);
2421 }
2422 
2424  vlib_frame_t * from_frame)
2425 {
2426  return tcp46_send_reset_inline (vm, node, from_frame, 0);
2427 }
2428 
2429 /* *INDENT-OFF* */
2431  .name = "tcp4-reset",
2432  .vector_size = sizeof (u32),
2433  .n_errors = TCP_N_ERROR,
2434  .error_strings = tcp_error_strings,
2435  .n_next_nodes = TCP_RESET_N_NEXT,
2436  .next_nodes = {
2437 #define _(s,n) [TCP_RESET_NEXT_##s] = n,
2439 #undef _
2440  },
2441  .format_trace = format_tcp_tx_trace,
2442 };
2443 /* *INDENT-ON* */
2444 
2445 /* *INDENT-OFF* */
2447  .name = "tcp6-reset",
2448  .vector_size = sizeof (u32),
2449  .n_errors = TCP_N_ERROR,
2450  .error_strings = tcp_error_strings,
2451  .n_next_nodes = TCP_RESET_N_NEXT,
2452  .next_nodes = {
2453 #define _(s,n) [TCP_RESET_NEXT_##s] = n,
2455 #undef _
2456  },
2457  .format_trace = format_tcp_tx_trace,
2458 };
2459 /* *INDENT-ON* */
2460 
2461 /*
2462  * fd.io coding-style-patch-verification: ON
2463  *
2464  * Local Variables:
2465  * eval: (c-set-style "gnu")
2466  * End:
2467  */
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
static void tcp_check_if_gso(tcp_connection_t *tc, vlib_buffer_t *b)
Definition: tcp_output.c:2118
static void tcp_check_sack_reneging(tcp_connection_t *tc)
Definition: tcp_output.c:1259
#define TCP_RXT_MAX_BURST
Definition: tcp_types.h:37
#define clib_min(x, y)
Definition: clib.h:327
static int tcp_send_acks(tcp_connection_t *tc, u32 max_burst_size)
Definition: tcp_output.c:1939
u16 ip4_tcp_compute_checksum_custom(vlib_main_t *vm, vlib_buffer_t *p0, ip46_address_t *src, ip46_address_t *dst)
Definition: tcp_output.c:387
#define CLIB_UNUSED(x)
Definition: clib.h:87
static void tcp_persist_timer_set(tcp_timer_wheel_t *tw, tcp_connection_t *tc)
Definition: tcp_timer.h:77
#define TCP_OPTION_LEN_SACK_PERMITTED
Definition: tcp_packet.h:166
static u32 tcp_time_now(void)
Definition: tcp_inlines.h:191
void tcp_timer_persist_handler(tcp_connection_t *tc)
Got 0 snd_wnd from peer, try to do something about it.
Definition: tcp_output.c:1510
vl_api_wireguard_peer_flags_t flags
Definition: wireguard.api:103
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:937
static tcp_connection_t * tcp_connection_get(u32 conn_index, u32 thread_index)
Definition: tcp_inlines.h:30
ip4_address_t src_address
Definition: ip4_packet.h:125
#define tcp_node_index(node_id, is_ip4)
Definition: tcp.h:272
int session_tx_fifo_peek_bytes(transport_connection_t *tc, u8 *buffer, u32 offset, u32 max_bytes)
Definition: session.c:553
void session_queue_run_on_main_thread(vlib_main_t *vm)
Definition: session.c:1650
#define vnet_buffer2(b)
Definition: buffer.h:482
#define TCP_FLAG_SYN
Definition: fa_node.h:13
static void tcp_make_synack(tcp_connection_t *tc, vlib_buffer_t *b)
Convert buffer to SYN-ACK.
Definition: tcp_output.c:512
#define tcp_opts_tstamp(_to)
Definition: tcp_packet.h:156
#define PREDICT_TRUE(x)
Definition: clib.h:121
i16 current_data
signed offset in data[], pre_data[] that we are currently processing.
Definition: buffer.h:110
vl_api_ip_port_and_mask_t dst_port
Definition: flow_types.api:92
void tcp_timer_retransmit_syn_handler(tcp_connection_t *tc)
SYN retransmit timer handler.
Definition: tcp_output.c:1446
#define clib_memcpy_fast(a, b, c)
Definition: string.h:81
clib_memset(h->entries, 0, sizeof(h->entries[0]) *entries)
static int tcp_do_retransmit(tcp_connection_t *tc, u32 max_burst_size)
Definition: tcp_output.c:1994
IP unicast adjacency.
Definition: adj.h:227
u32 fib_table_get_index_for_sw_if_index(fib_protocol_t proto, u32 sw_if_index)
Get the index of the FIB bound to the interface.
Definition: fib_table.c:989
#define tcp_fastrecovery_first_off(tc)
Definition: tcp_types.h:427
static u32 tcp_options_write(u8 *data, tcp_options_t *opts)
Write TCP options to segment.
Definition: tcp_packet.h:308
void session_add_self_custom_tx_evt(transport_connection_t *tc, u8 has_prio)
Definition: session.c:122
void tcp_timer_retransmit_handler(tcp_connection_t *tc)
Definition: tcp_output.c:1298
struct _tcp_main tcp_main_t
u32 thread_index
Definition: main.h:249
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_inlines.h:401
static sack_scoreboard_hole_t * scoreboard_get_hole(sack_scoreboard_t *sb, u32 index)
Definition: tcp_sack.h:35
void tcp_connection_timers_reset(tcp_connection_t *tc)
Stop all connection timers.
Definition: tcp.c:493
This packet is to be rewritten and forwarded to the next processing node.
Definition: adj.h:73
u16 current_length
Nbytes between current data and the end of this buffer.
Definition: buffer.h:113
static int tcp_transmit_unsent(tcp_worker_ctx_t *wrk, tcp_connection_t *tc, u32 burst_size)
Definition: tcp_output.c:1613
struct _tcp_connection tcp_connection_t
#define TCP_OPTS_ALIGN
Definition: tcp_packet.h:173
static u32 tcp_set_time_now(tcp_worker_ctx_t *wrk)
Definition: tcp_inlines.h:219
static u32 tcp_initial_wnd_unscaled(tcp_connection_t *tc)
TCP&#39;s initial window.
Definition: tcp_output.c:83
enum _tcp_output_next tcp_output_next_t
vl_api_address_t src
Definition: gre.api:54
static u32 format_get_indent(u8 *s)
Definition: format.h:72
static void tcp_cc_congestion(tcp_connection_t *tc)
Definition: tcp_cc.h:36
uword ip_csum_t
Definition: ip_packet.h:244
vlib_main_t * vm
Definition: in2out_ed.c:1582
static ip_csum_t ip_csum_with_carry(ip_csum_t sum, ip_csum_t x)
Definition: ip_packet.h:247
#define TCP_RTO_SYN_RETRIES
Definition: tcp_types.h:91
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:424
#define VLIB_NODE_FN(node)
Definition: node.h:202
static void tcp_push_ip_hdr(tcp_worker_ctx_t *wrk, tcp_connection_t *tc, vlib_buffer_t *b)
Definition: tcp_output.c:776
#define vec_validate_aligned(V, I, A)
Make sure vector is long enough for given index (no header, specified alignment)
Definition: vec.h:520
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:2358
vlib_error_t * errors
Vector of errors for this node.
Definition: node.h:469
static uword vlib_buffer_length_in_chain(vlib_main_t *vm, vlib_buffer_t *b)
Get length in bytes of the buffer chain.
Definition: buffer_funcs.h:402
static u8 tcp_is_descheduled(tcp_connection_t *tc)
Definition: tcp_inlines.h:380
u8 n_sack_blocks
Number of SACKs blocks.
Definition: tcp_packet.h:151
struct _sack_scoreboard sack_scoreboard_t
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:209
#define scoreboard_rescue_rxt_valid(_sb, _tc)
Definition: tcp_output.c:1696
ip6_address_t src_address
Definition: ip6_packet.h:310
#define tcp_in_cong_recovery(tc)
Definition: tcp_types.h:429
unsigned char u8
Definition: types.h:56
u8 wscale
Option flags, see above.
Definition: tcp_packet.h:146
u8 data[128]
Definition: ipsec_types.api:89
enum fib_protocol_t_ fib_protocol_t
Protocol Type.
#define TCP_OPTS_MAX_SACK_BLOCKS
Definition: tcp_packet.h:174
double f64
Definition: types.h:142
#define foreach_tcp4_reset_next
Definition: tcp_output.c:2349
static u32 tcp_prepare_retransmit_segment(tcp_worker_ctx_t *wrk, tcp_connection_t *tc, u32 offset, u32 max_deq_bytes, vlib_buffer_t **b)
Build a retransmit segment.
Definition: tcp_output.c:1216
static u16 ip_calculate_l4_checksum(vlib_main_t *vm, vlib_buffer_t *p0, ip_csum_t sum0, u32 payload_length, u8 *iph, u32 ip_header_size, u8 *l4h)
Definition: ip.h:184
int tcp_session_custom_tx(void *conn, transport_send_params_t *sp)
Definition: tcp_output.c:2013
void session_transport_closing_notify(transport_connection_t *tc)
Notification from transport that connection is being closed.
Definition: session.c:948
static uword tcp46_output_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame, int is_ip4)
Definition: tcp_output.c:2178
static void * tcp_init_buffer(vlib_main_t *vm, vlib_buffer_t *b)
Definition: tcp_output.c:346
static ip_adjacency_t * adj_get(adj_index_t adj_index)
Get a pointer to an adjacency object from its index.
Definition: adj.h:459
void tcp_make_syn(tcp_connection_t *tc, vlib_buffer_t *b)
Convert buffer to SYN.
Definition: tcp_output.c:486
static int tcp_prepare_segment(tcp_worker_ctx_t *wrk, tcp_connection_t *tc, u32 offset, u32 max_deq_bytes, vlib_buffer_t **b)
Allocate a new buffer and build a new tcp segment.
Definition: tcp_output.c:1100
#define tcp_is_fin(_th)
Definition: tcp_packet.h:90
static void tcp_make_fin(tcp_connection_t *tc, vlib_buffer_t *b)
Convert buffer to FIN-ACK.
Definition: tcp_output.c:477
#define tcp_cfg
Definition: tcp.h:271
#define TCP_OPTION_LEN_SACK_BLOCK
Definition: tcp_packet.h:168
ip4_address_t dst_address
Definition: ip4_packet.h:125
#define seq_leq(_s1, _s2)
Definition: tcp_packet.h:178
#define TCP_FLAG_ACK
Definition: fa_node.h:16
u8 * format_white_space(u8 *s, va_list *va)
Definition: std-formats.c:129
tcp_main_t tcp_main
Definition: tcp.c:28
#define vlib_prefetch_buffer_header(b, type)
Prefetch buffer metadata.
Definition: buffer.h:203
#define TCP_RTO_MAX
Definition: tcp_types.h:88
static void tcp_retransmit_timer_force_update(tcp_timer_wheel_t *tw, tcp_connection_t *tc)
Definition: tcp_timer.h:69
static void * ip4_next_header(ip4_header_t *i)
Definition: ip4_packet.h:196
#define tcp_zero_rwnd_sent(tc)
Definition: tcp_types.h:441
sack_block_t * sacks
SACK blocks.
Definition: tcp_packet.h:150
unsigned int u32
Definition: types.h:88
static sack_scoreboard_hole_t * scoreboard_first_hole(sack_scoreboard_t *sb)
Definition: tcp_sack.h:59
static tcp_header_t * tcp_buffer_hdr(vlib_buffer_t *b)
Definition: tcp_inlines.h:22
static void tcp46_output_trace_frame(vlib_main_t *vm, vlib_node_runtime_t *node, u32 *to_next, u32 n_bufs)
Definition: tcp_output.c:2078
#define VLIB_FRAME_SIZE
Definition: node.h:377
#define tcp_validate_txf_size(_tc, _a)
Definition: tcp.h:354
static void tcp_push_hdr_i(tcp_connection_t *tc, vlib_buffer_t *b, u32 snd_nxt, u8 compute_opts, u8 maybe_burst, u8 update_snd_nxt)
Push TCP header and update connection variables.
Definition: tcp_output.c:913
#define tcp_in_fastrecovery(tc)
Definition: tcp_types.h:419
#define tcp_csum_offload(tc)
Definition: tcp_types.h:439
static u32 vlib_get_buffer_index(vlib_main_t *vm, void *p)
Translate buffer pointer into buffer index.
Definition: buffer_funcs.h:293
static void tcp_retransmit_timer_update(tcp_timer_wheel_t *tw, tcp_connection_t *tc)
Definition: tcp_timer.h:104
u32 tcp_session_push_header(transport_connection_t *tconn, vlib_buffer_t *b)
Definition: tcp_output.c:986
#define TCP_OPTION_LEN_WINDOW_SCALE
Definition: tcp_packet.h:165
vlib_node_registration_t tcp6_reset_node
(constructor) VLIB_REGISTER_NODE (tcp6_reset_node)
Definition: tcp_output.c:2446
void scoreboard_clear_reneging(sack_scoreboard_t *sb, u32 start, u32 end)
Definition: tcp_sack.c:297
vlib_error_t error
Error code for buffers to be enqueued to error handler.
Definition: buffer.h:136
#define tcp_trajectory_add_start(b, start)
Definition: tcp.h:294
#define TRANSPORT_MAX_HDRS_LEN
static void tcp_retransmit_timer_set(tcp_timer_wheel_t *tw, tcp_connection_t *tc)
Definition: tcp_timer.h:55
vlib_main_t * vm
convenience pointer to this thread&#39;s vlib main
Definition: tcp.h:92
void tcp_send_reset(tcp_connection_t *tc)
Build and set reset packet for connection.
Definition: tcp_output.c:740
void tcp_send_synack(tcp_connection_t *tc)
Definition: tcp_output.c:835
#define ADJ_INDEX_INVALID
Invalid ADJ index - used when no adj is known likewise blazoned capitals INVALID speak volumes where ...
Definition: adj_types.h:36
static int tcp_make_synack_options(tcp_connection_t *tc, tcp_options_t *opts)
Definition: tcp_output.c:193
static int tcp_make_syn_options(tcp_connection_t *tc, tcp_options_t *opts)
Definition: tcp_output.c:164
static void * vlib_buffer_make_headroom(vlib_buffer_t *b, u8 size)
Make head room, typically for packet headers.
Definition: buffer.h:350
static int tcp_retransmit_sack(tcp_worker_ctx_t *wrk, tcp_connection_t *tc, u32 burst_size)
Do retransmit with SACKs.
Definition: tcp_output.c:1704
void tcp_connection_tx_pacer_reset(tcp_connection_t *tc, u32 window, u32 start_bucket)
Definition: tcp.c:1206
static void tcp_cc_loss(tcp_connection_t *tc)
Definition: tcp_cc.h:42
format_function_t format_tcp_connection_id
Definition: tcp.h:352
static __clib_warn_unused_result u32 vlib_buffer_alloc(vlib_main_t *vm, u32 *buffers, u32 n_buffers)
Allocate buffers into supplied array.
Definition: buffer_funcs.h:677
unsigned short u16
Definition: types.h:57
#define TCP_DUPACK_THRESHOLD
Definition: tcp_types.h:39
u8 data_len
Definition: ikev2_types.api:24
static void * vlib_buffer_get_current(vlib_buffer_t *b)
Get pointer to current data to process.
Definition: buffer.h:229
#define TCP_ESTABLISH_TIME
Definition: tcp_types.h:94
static void tcp_output_handle_link_local(tcp_connection_t *tc0, vlib_buffer_t *b0, u16 *next0, u32 *error0)
Definition: tcp_output.c:2046
tcp_timer_wheel_t timer_wheel
worker timer wheel
Definition: tcp.h:118
#define foreach_tcp6_output_next
Definition: tcp_output.c:35
#define PREDICT_FALSE(x)
Definition: clib.h:120
#define always_inline
Definition: ipsec.h:28
void tcp_program_dupack(tcp_connection_t *tc)
Definition: tcp_output.c:1045
#define TCP_FLAG_FIN
Definition: fa_node.h:12
int tcp_fastrecovery_prr_snd_space(tcp_connection_t *tc)
Estimate send space using proportional rate reduction (RFC6937)
Definition: tcp_output.c:1653
static u8 tcp_window_compute_scale(u32 window)
Definition: tcp_output.c:71
#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:224
vl_api_address_t dst
Definition: gre.api:55
#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:391
#define TCP_OPTION_LEN_TIMESTAMP
Definition: tcp_packet.h:167
#define foreach_tcp4_output_next
Definition: tcp_output.c:29
#define TCP_WND_MAX
Definition: tcp_packet.h:171
static void tcp_enqueue_to_ip_lookup(tcp_worker_ctx_t *wrk, vlib_buffer_t *b, u32 bi, u8 is_ip4, u32 fib_index)
Definition: tcp_output.c:534
static void vlib_node_increment_counter(vlib_main_t *vm, u32 node_index, u32 counter_index, u64 increment)
Definition: node_funcs.h:1231
#define TCP_FLAG_RST
Definition: fa_node.h:14
#define TCP_DBG(_fmt, _args...)
Definition: tcp_debug.h:146
u8 len
Definition: ip_types.api:92
#define TCP_MAX_WND_SCALE
Definition: tcp_packet.h:172
void tcp_program_cleanup(tcp_worker_ctx_t *wrk, tcp_connection_t *tc)
Definition: tcp.c:335
static void tcp_output_handle_packet(tcp_connection_t *tc0, vlib_buffer_t *b0, vlib_node_runtime_t *error_node, u16 *next0, u8 is_ip4)
Definition: tcp_output.c:2142
vl_api_ip_port_and_mask_t src_port
Definition: flow_types.api:91
static void tcp_cc_event(tcp_connection_t *tc, tcp_cc_event_t evt)
Definition: tcp_cc.h:61
void transport_connection_reschedule(transport_connection_t *tc)
Definition: transport.c:756
static u32 tcp_flight_size(const tcp_connection_t *tc)
Our estimate of the number of bytes in flight (pipe size)
Definition: tcp_inlines.h:94
This packet matches an "incomplete adjacency" and packets need to be passed to ARP to find rewrite st...
Definition: adj.h:63
void tcp_bt_track_tx(tcp_connection_t *tc, u32 len)
Track a tcp tx burst.
Definition: tcp_bt.c:301
#define VLIB_REGISTER_NODE(x,...)
Definition: node.h:169
tcp_header_t tcp_header
Definition: tcp_output.c:49
u16 n_vectors
Definition: node.h:396
static sack_scoreboard_hole_t * scoreboard_last_hole(sack_scoreboard_t *sb)
Definition: tcp_sack.h:67
#define seq_max(_s1, _s2)
Definition: tcp_packet.h:181
#define CLIB_PREFETCH(addr, size, type)
Definition: cache.h:80
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:339
void tcp_send_window_update_ack(tcp_connection_t *tc)
Send window update ack.
Definition: tcp_output.c:1074
sll srl srl sll sra u16x4 i
Definition: vector_sse42.h:317
void tcp_program_retransmit(tcp_connection_t *tc)
Definition: tcp_output.c:1057
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:652
format_function_t format_tcp_state
Definition: tcp.h:347
static void tcp_update_rto(tcp_connection_t *tc)
Definition: tcp_inlines.h:373
#define clib_warning(format, args...)
Definition: error.h:59
#define tcp_in_recovery(tc)
Definition: tcp_types.h:420
format_function_t format_tcp_header
Definition: format.h:100
struct _transport_connection transport_connection_t
#define TCP_TO_TIMER_TICK
Factor for converting ticks to timer ticks.
Definition: tcp_types.h:84
static 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:154
u32 adj_index_t
An index for adjacencies.
Definition: adj_types.h:30
#define ARRAY_LEN(x)
Definition: clib.h:67
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:483
#define tcp_zero_rwnd_sent_on(tc)
Definition: tcp_types.h:442
u16 mss
Maximum segment size advertised.
Definition: tcp_packet.h:147
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_inlines.h:171
static void * ip6_next_header(ip6_header_t *i)
Definition: ip6_packet.h:371
static int tcp_retransmit_no_sack(tcp_worker_ctx_t *wrk, tcp_connection_t *tc, u32 burst_size)
Fast retransmit without SACK info.
Definition: tcp_output.c:1858
vlib_main_t vlib_node_runtime_t * node
Definition: in2out_ed.c:1582
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_inlines.h:438
static void tcp_make_ack(tcp_connection_t *tc, vlib_buffer_t *b)
Convert buffer to ACK.
Definition: tcp_output.c:466
static u32 transport_max_tx_dequeue(transport_connection_t *tc)
Definition: session.h:503
#define seq_geq(_s1, _s2)
Definition: tcp_packet.h:180
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:1095
signed int i32
Definition: types.h:77
static int tcp_make_established_options(tcp_connection_t *tc, tcp_options_t *opts)
Definition: tcp_output.c:228
void tcp_bt_check_app_limited(tcp_connection_t *tc)
Check if sample to be generated is app limited.
Definition: tcp_bt.c:286
u16 cached_next_index
Next frame index that vector arguments were last enqueued to last time this node ran.
Definition: node.h:510
#define ASSERT(truth)
static void tcp_cc_init_rxt_timeout(tcp_connection_t *tc)
Reset congestion control, switch cwnd to loss window and try again.
Definition: tcp_output.c:1275
static void tcp_output_push_ip(vlib_main_t *vm, vlib_buffer_t *b0, tcp_connection_t *tc0, u8 is_ip4)
Definition: tcp_output.c:2102
#define tcp_recovery_on(tc)
Definition: tcp_types.h:417
static u8 * format_tcp_tx_trace(u8 *s, va_list *args)
Definition: tcp_output.c:54
u16 ip4_tcp_udp_compute_checksum(vlib_main_t *vm, vlib_buffer_t *p0, ip4_header_t *ip0)
Definition: ip4_forward.c:1367
void transport_connection_tx_pacer_reset_bucket(transport_connection_t *tc, u32 bucket)
Reset tx pacer bucket.
Definition: transport.c:705
void tcp_bt_track_rxt(tcp_connection_t *tc, u32 start, u32 end)
Track a tcp retransmission.
Definition: tcp_bt.c:337
void tcp_update_burst_snd_vars(tcp_connection_t *tc)
Update burst send vars.
Definition: tcp_output.c:298
#define TRANSPORT_PACER_MIN_BURST
Definition: transport.h:23
#define seq_gt(_s1, _s2)
Definition: tcp_packet.h:179
static uword ip6_address_is_link_local_unicast(const ip6_address_t *a)
Definition: ip6_packet.h:253
#define clib_mem_unaligned(pointer, type)
Definition: types.h:155
static void tcp_update_rcv_wnd(tcp_connection_t *tc)
Definition: tcp_output.c:114
struct _sack_scoreboard_hole sack_scoreboard_hole_t
#define TCP_RTO_BOFF_MAX
Definition: tcp_types.h:93
void tcp_send_fin(tcp_connection_t *tc)
Send FIN.
Definition: tcp_output.c:863
#define clib_max(x, y)
Definition: clib.h:320
void tcp_send_ack(tcp_connection_t *tc)
Definition: tcp_output.c:1015
void transport_connection_tx_pacer_update_bytes(transport_connection_t *tc, u32 bytes)
Definition: transport.c:749
int tcp_retransmit_first_unacked(tcp_worker_ctx_t *wrk, tcp_connection_t *tc)
Retransmit first unacked segment.
Definition: tcp_output.c:1594
template key/value backing page structure
Definition: bihash_doc.h:44
#define tcp_is_syn(_th)
Definition: tcp_packet.h:89
u32 ip_version_traffic_class_and_flow_label
Definition: ip6_packet.h:297
#define tcp_opts_wscale(_to)
Definition: tcp_packet.h:157
Definition: defs.h:47
u32 tsval
Timestamp value.
Definition: tcp_packet.h:148
static void tcp_timer_update(tcp_timer_wheel_t *tw, tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp_timer.h:43
u32 tsecr
Echoed/reflected time stamp.
Definition: tcp_packet.h:149
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:606
#define tcp_fastrecovery_first(tc)
Definition: tcp_types.h:425
#define vec_len(v)
Number of elements in vector (rvalue-only, NULL tolerant)
static u8 tcp_max_tx_deq(tcp_connection_t *tc)
Definition: tcp_output.c:1690
ip_lookup_next_t lookup_next_index
Next hop after ip4-lookup.
Definition: adj.h:329
u32 next_buffer
Next buffer for this linked-list of buffers.
Definition: buffer.h:140
vlib_main_t vlib_node_runtime_t vlib_frame_t * frame
Definition: in2out_ed.c:1583
#define foreach_tcp6_reset_next
Definition: tcp_output.c:2353
#define VLIB_BUFFER_TRACE_TRAJECTORY_INIT(b)
Definition: buffer.h:492
static tcp_worker_ctx_t * tcp_get_worker(u32 thread_index)
Definition: tcp.h:282
void session_transport_closed_notify(transport_connection_t *tc)
Notification from transport that it is closed.
Definition: session.c:1036
VLIB buffer representation.
Definition: buffer.h:102
u64 uword
Definition: types.h:112
int session_stream_connect_notify(transport_connection_t *tc, session_error_t err)
Definition: session.c:757
static void * vlib_frame_vector_args(vlib_frame_t *f)
Get pointer to frame vector data.
Definition: node_funcs.h:297
static 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:433
static int tcp_make_reset_in_place(vlib_main_t *vm, vlib_buffer_t *b, u8 is_ip4)
Definition: tcp_output.c:569
#define TCP_OPTION_LEN_MSS
Definition: tcp_packet.h:164
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_sack.c:194
u16 ip6_tcp_compute_checksum_custom(vlib_main_t *vm, vlib_buffer_t *p0, ip46_address_t *src, ip46_address_t *dst)
Definition: tcp_output.c:361
static f64 tcp_time_now_us(u32 thread_index)
Definition: tcp_inlines.h:213
void scoreboard_init_rxt(sack_scoreboard_t *sb, u32 snd_una)
Definition: tcp_sack.c:254
static void tcp_connection_set_state(tcp_connection_t *tc, tcp_state_t state)
Definition: tcp_inlines.h:51
struct clib_bihash_value offset
template key/value backing page structure
u32 tcp_initial_window_to_advertise(tcp_connection_t *tc)
Compute initial window and scale factor.
Definition: tcp_output.c:102
#define vnet_buffer(b)
Definition: buffer.h:417
#define TCP_USE_SACKS
Disable only for testing.
Definition: tcp_types.h:42
vl_api_dhcp_client_state_t state
Definition: dhcp.api:201
static u32 vlib_num_workers()
Definition: threads.h:377
void tcp_connection_cleanup(tcp_connection_t *tc)
Cleans up connection state.
Definition: tcp.c:242
static u32 tcp_buffer_len(vlib_buffer_t *b)
Definition: tcp_output.c:977
static u8 tcp_retransmit_should_retry_head(tcp_connection_t *tc, sack_scoreboard_t *sb)
Definition: tcp_output.c:1677
void tcp_send_syn(tcp_connection_t *tc)
Send SYN.
Definition: tcp_output.c:799
vlib_node_registration_t tcp6_output_node
(constructor) VLIB_REGISTER_NODE (tcp6_output_node)
Definition: tcp_output.c:2323
u16 flags
Copy of main node flags.
Definition: node.h:500
void * vlib_add_trace(vlib_main_t *vm, vlib_node_runtime_t *r, vlib_buffer_t *b, u32 n_data_bytes)
Definition: trace.c:577
static u8 tcp_timer_is_active(tcp_connection_t *tc, tcp_timers_e timer)
Definition: tcp_timer.h:118
static u16 tcp_compute_checksum(tcp_connection_t *tc, vlib_buffer_t *b)
Definition: tcp_output.c:407
enum _tcp_reset_next tcp_reset_next_t
static u32 transport_max_rx_enqueue(transport_connection_t *tc)
Definition: session.h:496
#define tcp_opts_sack_permitted(_to)
Definition: tcp_packet.h:159
static void vlib_buffer_free_one(vlib_main_t *vm, u32 buffer_index)
Free one buffer Shorthand to free a single buffer chain.
Definition: buffer_funcs.h:970
tcp_connection_t tcp_connection
Definition: tcp_output.c:50
static u32 tcp_tstamp(tcp_connection_t *tc)
Generate timestamp for tcp connection.
Definition: tcp_inlines.h:206
void tcp_program_ack(tcp_connection_t *tc)
Definition: tcp_output.c:1035
static void * vlib_buffer_push_ip6_custom(vlib_main_t *vm, vlib_buffer_t *b, ip6_address_t *src, ip6_address_t *dst, int proto, u32 flow_label)
Push IPv6 header to buffer.
Definition: ip6.h:563
static void * tcp_reuse_buffer(vlib_main_t *vm, vlib_buffer_t *b)
Definition: tcp_output.c:329
u8 ip_version_and_header_length
Definition: ip4_packet.h:93
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:280
vlib_node_registration_t tcp4_reset_node
(constructor) VLIB_REGISTER_NODE (tcp4_reset_node)
Definition: tcp_output.c:2430
#define VLIB_NODE_FLAG_TRACE
Definition: node.h:301
static uword round_down_pow2(uword x, uword pow2)
Definition: clib.h:258
vlib_node_registration_t tcp4_output_node
(constructor) VLIB_REGISTER_NODE (tcp4_output_node)
Definition: tcp_output.c:2303
#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 void tcp_enqueue_to_output(tcp_worker_ctx_t *wrk, vlib_buffer_t *b, u32 bi, u8 is_ip4)
Definition: tcp_output.c:556
static tcp_main_t * vnet_get_tcp_main()
Definition: tcp.h:276
u32 tco_next_node[2]
Session layer edge indices to tcp output.
Definition: tcp.h:101
static char * tcp_error_strings[]
Definition: tcp_output.c:41
static void * vlib_buffer_push_ip4(vlib_main_t *vm, vlib_buffer_t *b, ip4_address_t *src, ip4_address_t *dst, int proto, u8 csum_offload)
Push IPv4 header to buffer.
Definition: ip4.h:420
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
#define tcp_ack(_th)
Definition: tcp_packet.h:83
#define seq_lt(_s1, _s2)
Definition: tcp_packet.h:177
enum _tcp_state tcp_state_t
u32 transport_connection_tx_pacer_burst(transport_connection_t *tc)
Get tx pacer max burst.
Definition: transport.c:729
vl_api_interface_index_t sw_if_index
Definition: wireguard.api:33
Definition: defs.h:46
#define tcp_worker_stats_inc(_wrk, _stat, _val)
Definition: tcp.h:125
#define tcp_zero_rwnd_sent_off(tc)
Definition: tcp_types.h:443
ip6_address_t dst_address
Definition: ip6_packet.h:310
u32 * tx_buffers
tx buffer free list
Definition: tcp.h:112
static void session_add_pending_tx_buffer(u32 thread_index, u32 bi, u32 next_node)
Add session node pending buffer with custom node.
Definition: session.h:663
adj_index_t adj_nbr_find(fib_protocol_t nh_proto, vnet_link_t link_type, const ip46_address_t *nh_addr, u32 sw_if_index)
Lookup neighbor adjancency.
Definition: adj_nbr.c:109
#define TCP_EVT(_evt, _args...)
Definition: tcp_debug.h:145
static int tcp_make_options(tcp_connection_t *tc, tcp_options_t *opts, tcp_state_t state)
Definition: tcp_output.c:263
static uword pool_elts(void *v)
Number of active elements in a pool.
Definition: pool.h:128