FD.io VPP  v20.05-21-gb1500e9ff
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  if (PREDICT_FALSE (available_space < tc->rcv_opts.mss))
124  {
125  tc->rcv_wnd = 0;
126  return;
127  }
128 
129  /*
130  * Use the above and what we know about what we've previously advertised
131  * to compute the new window
132  */
133  observed_wnd = (i32) tc->rcv_wnd - (tc->rcv_nxt - tc->rcv_las);
134 
135  /* Bad. Thou shalt not shrink */
136  if (PREDICT_FALSE ((i32) available_space < observed_wnd))
137  {
138  wnd = clib_max (observed_wnd, 0);
139  TCP_EVT (TCP_EVT_RCV_WND_SHRUNK, tc, observed_wnd, available_space);
140  }
141  else
142  {
143  wnd = available_space;
144  }
145 
146  /* Make sure we have a multiple of 1 << rcv_wscale. We round up to
147  * avoid advertising a window less than mss which could happen if
148  * 1 << rcv_wscale < mss */
149  if (wnd && tc->rcv_wscale)
150  wnd = round_pow2 (wnd, 1 << tc->rcv_wscale);
151 
152  tc->rcv_wnd = clib_min (wnd, TCP_WND_MAX << tc->rcv_wscale);
153 }
154 
155 /**
156  * Compute and return window to advertise, scaled as per RFC1323
157  */
158 static inline u32
160 {
161  if (state < TCP_STATE_ESTABLISHED)
163 
164  tcp_update_rcv_wnd (tc);
165  return tc->rcv_wnd >> tc->rcv_wscale;
166 }
167 
168 static int
170 {
171  u8 len = 0;
172 
173  opts->flags |= TCP_OPTS_FLAG_MSS;
174  opts->mss = tc->mss;
175  len += TCP_OPTION_LEN_MSS;
176 
177  opts->flags |= TCP_OPTS_FLAG_WSCALE;
178  opts->wscale = tc->rcv_wscale;
180 
181  opts->flags |= TCP_OPTS_FLAG_TSTAMP;
182  opts->tsval = tcp_time_now ();
183  opts->tsecr = 0;
185 
186  if (TCP_USE_SACKS)
187  {
188  opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED;
190  }
191 
192  /* Align to needed boundary */
193  len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
194  return len;
195 }
196 
197 static int
199 {
200  u8 len = 0;
201 
202  opts->flags |= TCP_OPTS_FLAG_MSS;
203  opts->mss = tc->mss;
204  len += TCP_OPTION_LEN_MSS;
205 
206  if (tcp_opts_wscale (&tc->rcv_opts))
207  {
208  opts->flags |= TCP_OPTS_FLAG_WSCALE;
209  opts->wscale = tc->rcv_wscale;
211  }
212 
213  if (tcp_opts_tstamp (&tc->rcv_opts))
214  {
215  opts->flags |= TCP_OPTS_FLAG_TSTAMP;
216  opts->tsval = tcp_time_now ();
217  opts->tsecr = tc->tsval_recent;
219  }
220 
221  if (tcp_opts_sack_permitted (&tc->rcv_opts))
222  {
223  opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED;
225  }
226 
227  /* Align to needed boundary */
228  len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
229  return len;
230 }
231 
232 static int
234 {
235  u8 len = 0;
236 
237  opts->flags = 0;
238 
239  if (tcp_opts_tstamp (&tc->rcv_opts))
240  {
241  opts->flags |= TCP_OPTS_FLAG_TSTAMP;
242  opts->tsval = tcp_tstamp (tc);
243  opts->tsecr = tc->tsval_recent;
245  }
246  if (tcp_opts_sack_permitted (&tc->rcv_opts))
247  {
248  if (vec_len (tc->snd_sacks))
249  {
250  opts->flags |= TCP_OPTS_FLAG_SACK;
251  if (tc->snd_sack_pos >= vec_len (tc->snd_sacks))
252  tc->snd_sack_pos = 0;
253  opts->sacks = &tc->snd_sacks[tc->snd_sack_pos];
254  opts->n_sack_blocks = vec_len (tc->snd_sacks) - tc->snd_sack_pos;
255  opts->n_sack_blocks = clib_min (opts->n_sack_blocks,
257  tc->snd_sack_pos += opts->n_sack_blocks;
258  len += 2 + TCP_OPTION_LEN_SACK_BLOCK * opts->n_sack_blocks;
259  }
260  }
261 
262  /* Align to needed boundary */
263  len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
264  return len;
265 }
266 
267 always_inline int
270 {
271  switch (state)
272  {
273  case TCP_STATE_ESTABLISHED:
274  case TCP_STATE_CLOSE_WAIT:
275  case TCP_STATE_FIN_WAIT_1:
276  case TCP_STATE_LAST_ACK:
277  case TCP_STATE_CLOSING:
278  case TCP_STATE_FIN_WAIT_2:
279  case TCP_STATE_TIME_WAIT:
280  case TCP_STATE_CLOSED:
281  return tcp_make_established_options (tc, opts);
282  case TCP_STATE_SYN_RCVD:
283  return tcp_make_synack_options (tc, opts);
284  case TCP_STATE_SYN_SENT:
285  return tcp_make_syn_options (tc, opts);
286  default:
287  clib_warning ("State not handled! %d", state);
288  return 0;
289  }
290 }
291 
292 /**
293  * Update burst send vars
294  *
295  * - Updates snd_mss to reflect the effective segment size that we can send
296  * by taking into account all TCP options, including SACKs.
297  * - Cache 'on the wire' options for reuse
298  * - Updates receive window which can be reused for a burst.
299  *
300  * This should *only* be called when doing bursts
301  */
302 void
304 {
305  tcp_main_t *tm = &tcp_main;
306 
307  /* Compute options to be used for connection. These may be reused when
308  * sending data or to compute the effective mss (snd_mss) */
309  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts,
310  TCP_STATE_ESTABLISHED);
311 
312  /* XXX check if MTU has been updated */
313  tc->snd_mss = clib_min (tc->mss, tc->rcv_opts.mss) - tc->snd_opts_len;
314  ASSERT (tc->snd_mss > 0);
315 
316  tcp_options_write (tm->wrk_ctx[tc->c_thread_index].cached_opts,
317  &tc->snd_opts);
318 
319  tcp_update_rcv_wnd (tc);
320 
321  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
323 
324  if (tc->snd_una == tc->snd_nxt)
325  {
328  }
329 }
330 
331 #endif /* CLIB_MARCH_VARIANT */
332 
333 static void *
335 {
336  if (b->flags & VLIB_BUFFER_NEXT_PRESENT)
338  /* Zero all flags but free list index and trace flag */
339  b->flags &= VLIB_BUFFER_NEXT_PRESENT - 1;
340  b->current_data = 0;
341  b->current_length = 0;
343  vnet_buffer (b)->tcp.flags = 0;
344 
345  /* Leave enough space for headers */
347 }
348 
349 #ifndef CLIB_MARCH_VARIANT
350 static void *
352 {
353  ASSERT ((b->flags & VLIB_BUFFER_NEXT_PRESENT) == 0);
354  b->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
356  b->current_data = 0;
357  vnet_buffer (b)->tcp.flags = 0;
359  /* Leave enough space for headers */
361 }
362 
363 
364 /* Compute TCP checksum in software when offloading is disabled for a connection */
365 u16
367  ip46_address_t * src, ip46_address_t * dst)
368 {
369  ip_csum_t sum0;
370  u16 payload_length_host_byte_order;
371  u32 i;
372 
373  /* Initialize checksum with ip header. */
374  sum0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, p0)) +
375  clib_host_to_net_u16 (IP_PROTOCOL_TCP);
376  payload_length_host_byte_order = vlib_buffer_length_in_chain (vm, p0);
377 
378  for (i = 0; i < ARRAY_LEN (src->ip6.as_uword); i++)
379  {
380  sum0 = ip_csum_with_carry
381  (sum0, clib_mem_unaligned (&src->ip6.as_uword[i], uword));
382  sum0 = ip_csum_with_carry
383  (sum0, clib_mem_unaligned (&dst->ip6.as_uword[i], uword));
384  }
385 
386  return ip_calculate_l4_checksum (vm, p0, sum0,
387  payload_length_host_byte_order, NULL, 0,
388  NULL);
389 }
390 
391 u16
393  ip46_address_t * src, ip46_address_t * dst)
394 {
395  ip_csum_t sum0;
396  u32 payload_length_host_byte_order;
397 
398  payload_length_host_byte_order = vlib_buffer_length_in_chain (vm, p0);
399  sum0 =
400  clib_host_to_net_u32 (payload_length_host_byte_order +
401  (IP_PROTOCOL_TCP << 16));
402 
403  sum0 = ip_csum_with_carry (sum0, clib_mem_unaligned (&src->ip4, u32));
404  sum0 = ip_csum_with_carry (sum0, clib_mem_unaligned (&dst->ip4, u32));
405 
406  return ip_calculate_l4_checksum (vm, p0, sum0,
407  payload_length_host_byte_order, NULL, 0,
408  NULL);
409 }
410 
411 static inline u16
413 {
414  u16 checksum = 0;
415  if (PREDICT_FALSE (tc->cfg_flags & TCP_CFG_F_NO_CSUM_OFFLOAD))
416  {
417  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
418  vlib_main_t *vm = wrk->vm;
419 
420  if (tc->c_is_ip4)
422  (vm, b, &tc->c_lcl_ip, &tc->c_rmt_ip);
423  else
425  (vm, b, &tc->c_lcl_ip, &tc->c_rmt_ip);
426  }
427  else
428  {
429  b->flags |= VNET_BUFFER_F_OFFLOAD_TCP_CKSUM;
430  }
431  return checksum;
432 }
433 
434 /**
435  * Prepare ACK
436  */
437 static inline void
439  u8 flags)
440 {
441  tcp_options_t _snd_opts, *snd_opts = &_snd_opts;
442  u8 tcp_opts_len, tcp_hdr_opts_len;
443  tcp_header_t *th;
444  u16 wnd;
445 
446  wnd = tcp_window_to_advertise (tc, state);
447 
448  /* Make and write options */
449  tcp_opts_len = tcp_make_established_options (tc, snd_opts);
450  tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
451 
452  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt,
453  tc->rcv_nxt, tcp_hdr_opts_len, flags, wnd);
454 
455  tcp_options_write ((u8 *) (th + 1), snd_opts);
456 
457  th->checksum = tcp_compute_checksum (tc, b);
458 
459  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
460 
461  if (wnd == 0)
463  else
465 }
466 
467 /**
468  * Convert buffer to ACK
469  */
470 static inline void
472 {
473  tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, TCP_FLAG_ACK);
474  TCP_EVT (TCP_EVT_ACK_SENT, tc);
475  tc->rcv_las = tc->rcv_nxt;
476 }
477 
478 /**
479  * Convert buffer to FIN-ACK
480  */
481 static void
483 {
484  tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_ACK);
485 }
486 
487 /**
488  * Convert buffer to SYN
489  */
490 void
492 {
493  u8 tcp_hdr_opts_len, tcp_opts_len;
494  tcp_header_t *th;
495  u16 initial_wnd;
496  tcp_options_t snd_opts;
497 
498  initial_wnd = tcp_initial_window_to_advertise (tc);
499 
500  /* Make and write options */
501  clib_memset (&snd_opts, 0, sizeof (snd_opts));
502  tcp_opts_len = tcp_make_syn_options (tc, &snd_opts);
503  tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
504 
505  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss,
506  tc->rcv_nxt, tcp_hdr_opts_len, TCP_FLAG_SYN,
507  initial_wnd);
508  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
509  tcp_options_write ((u8 *) (th + 1), &snd_opts);
510  th->checksum = tcp_compute_checksum (tc, b);
511 }
512 
513 /**
514  * Convert buffer to SYN-ACK
515  */
516 static void
518 {
519  tcp_options_t _snd_opts, *snd_opts = &_snd_opts;
520  u8 tcp_opts_len, tcp_hdr_opts_len;
521  tcp_header_t *th;
522  u16 initial_wnd;
523 
524  clib_memset (snd_opts, 0, sizeof (*snd_opts));
525  initial_wnd = tcp_initial_window_to_advertise (tc);
526  tcp_opts_len = tcp_make_synack_options (tc, snd_opts);
527  tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
528 
529  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss,
530  tc->rcv_nxt, tcp_hdr_opts_len,
531  TCP_FLAG_SYN | TCP_FLAG_ACK, initial_wnd);
532  tcp_options_write ((u8 *) (th + 1), snd_opts);
533 
534  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
535  th->checksum = tcp_compute_checksum (tc, b);
536 }
537 
538 static void
540  u8 is_ip4, u32 fib_index)
541 {
542  tcp_main_t *tm = &tcp_main;
543  vlib_main_t *vm = wrk->vm;
544 
545  b->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
546  b->error = 0;
547 
548  vnet_buffer (b)->sw_if_index[VLIB_TX] = fib_index;
549  vnet_buffer (b)->sw_if_index[VLIB_RX] = 0;
550 
552 
554  tm->ipl_next_node[!is_ip4]);
555 
556  if (vm->thread_index == 0 && vlib_num_workers ())
558 }
559 
560 static void
562  u8 is_ip4)
563 {
564  b->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
565  b->error = 0;
566 
568  wrk->tco_next_node[!is_ip4]);
569 }
570 
571 #endif /* CLIB_MARCH_VARIANT */
572 
573 static int
575 {
576  ip4_header_t *ih4;
577  ip6_header_t *ih6;
578  tcp_header_t *th;
579  ip4_address_t src_ip4, dst_ip4;
580  ip6_address_t src_ip6, dst_ip6;
582  u32 tmp, len, seq, ack;
583  u8 flags;
584 
585  /* Find IP and TCP headers */
586  th = tcp_buffer_hdr (b);
587 
588  /* Save src and dst ip */
589  if (is_ip4)
590  {
591  ih4 = vlib_buffer_get_current (b);
592  ASSERT ((ih4->ip_version_and_header_length & 0xF0) == 0x40);
593  src_ip4.as_u32 = ih4->src_address.as_u32;
594  dst_ip4.as_u32 = ih4->dst_address.as_u32;
595  }
596  else
597  {
598  ih6 = vlib_buffer_get_current (b);
599  ASSERT ((ih6->ip_version_traffic_class_and_flow_label & 0xF0) == 0x60);
600  clib_memcpy_fast (&src_ip6, &ih6->src_address, sizeof (ip6_address_t));
601  clib_memcpy_fast (&dst_ip6, &ih6->dst_address, sizeof (ip6_address_t));
602  }
603 
604  src_port = th->src_port;
605  dst_port = th->dst_port;
606  flags = TCP_FLAG_RST;
607 
608  /*
609  * RFC 793. If the ACK bit is off, sequence number zero is used,
610  * <SEQ=0><ACK=SEG.SEQ+SEG.LEN><CTL=RST,ACK>
611  * If the ACK bit is on,
612  * <SEQ=SEG.ACK><CTL=RST>
613  */
614  if (tcp_ack (th))
615  {
616  seq = th->ack_number;
617  ack = 0;
618  }
619  else
620  {
621  flags |= TCP_FLAG_ACK;
622  tmp = clib_net_to_host_u32 (th->seq_number);
623  len = vnet_buffer (b)->tcp.data_len + tcp_is_syn (th) + tcp_is_fin (th);
624  ack = clib_host_to_net_u32 (tmp + len);
625  seq = 0;
626  }
627 
628  tcp_reuse_buffer (vm, b);
630  th = vlib_buffer_push_tcp_net_order (b, dst_port, src_port, seq, ack,
631  sizeof (tcp_header_t), flags, 0);
632 
633  if (is_ip4)
634  {
635  ih4 = vlib_buffer_push_ip4 (vm, b, &dst_ip4, &src_ip4,
636  IP_PROTOCOL_TCP, 1);
637  th->checksum = ip4_tcp_udp_compute_checksum (vm, b, ih4);
638  }
639  else
640  {
641  int bogus = ~0;
642  ih6 = vlib_buffer_push_ip6 (vm, b, &dst_ip6, &src_ip6, IP_PROTOCOL_TCP);
643  th->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b, ih6, &bogus);
644  ASSERT (!bogus);
645  }
646 
647  return 0;
648 }
649 
650 #ifndef CLIB_MARCH_VARIANT
651 /**
652  * Send reset without reusing existing buffer
653  *
654  * It extracts connection info out of original packet
655  */
656 void
658  u32 thread_index, u8 is_ip4)
659 {
660  tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index);
661  vlib_main_t *vm = wrk->vm;
662  vlib_buffer_t *b;
663  u32 bi, sw_if_index, fib_index;
664  u8 tcp_hdr_len, flags = 0;
665  tcp_header_t *th, *pkt_th;
666  u32 seq, ack;
667  ip4_header_t *ih4, *pkt_ih4;
668  ip6_header_t *ih6, *pkt_ih6;
669  fib_protocol_t fib_proto;
670 
671  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
672  return;
673 
674  b = vlib_get_buffer (vm, bi);
675  sw_if_index = vnet_buffer (pkt)->sw_if_index[VLIB_RX];
676  fib_proto = is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6;
677  fib_index = fib_table_get_index_for_sw_if_index (fib_proto, sw_if_index);
678  tcp_init_buffer (vm, b);
679 
680  /* Make and write options */
681  tcp_hdr_len = sizeof (tcp_header_t);
682 
683  if (is_ip4)
684  {
685  pkt_ih4 = vlib_buffer_get_current (pkt);
686  pkt_th = ip4_next_header (pkt_ih4);
687  }
688  else
689  {
690  pkt_ih6 = vlib_buffer_get_current (pkt);
691  pkt_th = ip6_next_header (pkt_ih6);
692  }
693 
694  if (tcp_ack (pkt_th))
695  {
696  flags = TCP_FLAG_RST;
697  seq = pkt_th->ack_number;
698  ack = (tc->state >= TCP_STATE_SYN_RCVD) ? tc->rcv_nxt : 0;
699  }
700  else
701  {
702  flags = TCP_FLAG_RST | TCP_FLAG_ACK;
703  seq = 0;
704  ack = clib_host_to_net_u32 (vnet_buffer (pkt)->tcp.seq_end);
705  }
706 
707  th = vlib_buffer_push_tcp_net_order (b, pkt_th->dst_port, pkt_th->src_port,
708  seq, ack, tcp_hdr_len, flags, 0);
709 
710  /* Swap src and dst ip */
711  if (is_ip4)
712  {
713  ASSERT ((pkt_ih4->ip_version_and_header_length & 0xF0) == 0x40);
714  ih4 = vlib_buffer_push_ip4 (vm, b, &pkt_ih4->dst_address,
715  &pkt_ih4->src_address, IP_PROTOCOL_TCP,
716  tcp_csum_offload (tc));
717  th->checksum = ip4_tcp_udp_compute_checksum (vm, b, ih4);
718  }
719  else
720  {
721  int bogus = ~0;
722  ASSERT ((pkt_ih6->ip_version_traffic_class_and_flow_label & 0xF0) ==
723  0x60);
724  ih6 = vlib_buffer_push_ip6_custom (vm, b, &pkt_ih6->dst_address,
725  &pkt_ih6->src_address,
726  IP_PROTOCOL_TCP,
727  tc->ipv6_flow_label);
728  th->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b, ih6, &bogus);
729  ASSERT (!bogus);
730  }
731 
732  tcp_enqueue_to_ip_lookup (wrk, b, bi, is_ip4, fib_index);
733  TCP_EVT (TCP_EVT_RST_SENT, tc);
734  vlib_node_increment_counter (vm, tcp_node_index (output, tc->c_is_ip4),
735  TCP_ERROR_RST_SENT, 1);
736 }
737 
738 /**
739  * Build and set reset packet for connection
740  */
741 void
743 {
744  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
745  vlib_main_t *vm = wrk->vm;
746  vlib_buffer_t *b;
747  u32 bi;
748  tcp_header_t *th;
749  u16 tcp_hdr_opts_len, advertise_wnd, opts_write_len;
750  u8 flags;
751 
752  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
753  return;
754  b = vlib_get_buffer (vm, bi);
755  tcp_init_buffer (vm, b);
756 
757  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
758  tcp_hdr_opts_len = tc->snd_opts_len + sizeof (tcp_header_t);
759  advertise_wnd = tc->rcv_wnd >> tc->rcv_wscale;
760  flags = TCP_FLAG_RST;
761  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt,
762  tc->rcv_nxt, tcp_hdr_opts_len, flags,
763  advertise_wnd);
764  opts_write_len = tcp_options_write ((u8 *) (th + 1), &tc->snd_opts);
765  th->checksum = tcp_compute_checksum (tc, b);
766  ASSERT (opts_write_len == tc->snd_opts_len);
767  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
768  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
769  TCP_EVT (TCP_EVT_RST_SENT, tc);
770  vlib_node_increment_counter (vm, tcp_node_index (output, tc->c_is_ip4),
771  TCP_ERROR_RST_SENT, 1);
772 }
773 
774 static void
776  vlib_buffer_t * b)
777 {
778  if (tc->c_is_ip4)
779  {
780  vlib_buffer_push_ip4 (wrk->vm, b, &tc->c_lcl_ip4, &tc->c_rmt_ip4,
781  IP_PROTOCOL_TCP, tcp_csum_offload (tc));
782  }
783  else
784  {
785  vlib_buffer_push_ip6_custom (wrk->vm, b, &tc->c_lcl_ip6, &tc->c_rmt_ip6,
786  IP_PROTOCOL_TCP, tc->ipv6_flow_label);
787  }
788 }
789 
790 /**
791  * Send SYN
792  *
793  * Builds a SYN packet for a half-open connection and sends it to ipx_lookup.
794  * The packet is not forwarded through tcpx_output to avoid doing lookups
795  * in the half_open pool.
796  */
797 void
799 {
800  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
801  vlib_main_t *vm = wrk->vm;
802  vlib_buffer_t *b;
803  u32 bi;
804 
805  /*
806  * Setup retransmit and establish timers before requesting buffer
807  * such that we can return if we've ran out.
808  */
809  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN,
810  tc->rto * TCP_TO_TIMER_TICK);
811 
812  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
813  {
814  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN, 1);
815  return;
816  }
817 
818  b = vlib_get_buffer (vm, bi);
819  tcp_init_buffer (vm, b);
820  tcp_make_syn (tc, b);
821 
822  /* Measure RTT with this */
823  tc->rtt_ts = tcp_time_now_us (vlib_num_workers ()? 1 : 0);
824  tc->rtt_seq = tc->snd_nxt;
825  tc->rto_boff = 0;
826 
827  tcp_push_ip_hdr (wrk, tc, b);
828  tcp_enqueue_to_ip_lookup (wrk, b, bi, tc->c_is_ip4, tc->c_fib_index);
829  TCP_EVT (TCP_EVT_SYN_SENT, tc);
830 }
831 
832 void
834 {
835  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
836  vlib_main_t *vm = wrk->vm;
837  vlib_buffer_t *b;
838  u32 bi;
839 
841 
842  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
843  {
844  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT, 1);
845  return;
846  }
847 
848  tc->rtt_ts = tcp_time_now_us (tc->c_thread_index);
849  b = vlib_get_buffer (vm, bi);
850  tcp_init_buffer (vm, b);
851  tcp_make_synack (tc, b);
852  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
853  TCP_EVT (TCP_EVT_SYNACK_SENT, tc);
854 }
855 
856 /**
857  * Send FIN
858  */
859 void
861 {
862  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
863  vlib_main_t *vm = wrk->vm;
864  vlib_buffer_t *b;
865  u32 bi;
866  u8 fin_snt = 0;
867 
868  fin_snt = tc->flags & TCP_CONN_FINSNT;
869  if (fin_snt)
870  tc->snd_nxt -= 1;
871 
872  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
873  {
874  /* Out of buffers so program fin retransmit ASAP */
875  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT, 1);
876  if (fin_snt)
877  tc->snd_nxt += 1;
878  else
879  /* Make sure retransmit retries a fin not data */
880  tc->flags |= TCP_CONN_FINSNT;
881  return;
882  }
883 
884  /* If we have non-dupacks programmed, no need to send them */
885  if ((tc->flags & TCP_CONN_SNDACK) && !tc->pending_dupacks)
886  tc->flags &= ~TCP_CONN_SNDACK;
887 
889  b = vlib_get_buffer (vm, bi);
890  tcp_init_buffer (vm, b);
891  tcp_make_fin (tc, b);
892  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
893  TCP_EVT (TCP_EVT_FIN_SENT, tc);
894  /* Account for the FIN */
895  tc->snd_nxt += 1;
896  if (!fin_snt)
897  {
898  tc->flags |= TCP_CONN_FINSNT;
899  tc->flags &= ~TCP_CONN_FINPNDG;
900  tc->snd_una_max = seq_max (tc->snd_una_max, tc->snd_nxt);
901  }
902 }
903 
904 /**
905  * Push TCP header and update connection variables. Should only be called
906  * for segments with data, not for 'control' packets.
907  */
908 always_inline void
910  u8 compute_opts, u8 maybe_burst, u8 update_snd_nxt)
911 {
912  u8 tcp_hdr_opts_len, flags = TCP_FLAG_ACK;
913  u32 advertise_wnd, data_len;
914  tcp_main_t *tm = &tcp_main;
915  tcp_header_t *th;
916 
917  data_len = b->current_length;
918  if (PREDICT_FALSE (b->flags & VLIB_BUFFER_NEXT_PRESENT))
920 
921  vnet_buffer (b)->tcp.flags = 0;
922  vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
923 
924  if (compute_opts)
925  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
926 
927  tcp_hdr_opts_len = tc->snd_opts_len + sizeof (tcp_header_t);
928 
929  if (maybe_burst)
930  advertise_wnd = tc->rcv_wnd >> tc->rcv_wscale;
931  else
932  advertise_wnd = tcp_window_to_advertise (tc, TCP_STATE_ESTABLISHED);
933 
934  if (PREDICT_FALSE (tc->flags & TCP_CONN_PSH_PENDING))
935  {
936  if (seq_geq (tc->psh_seq, snd_nxt)
937  && seq_lt (tc->psh_seq, snd_nxt + data_len))
938  flags |= TCP_FLAG_PSH;
939  }
940  th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, snd_nxt,
941  tc->rcv_nxt, tcp_hdr_opts_len, flags,
942  advertise_wnd);
943 
944  if (maybe_burst)
945  {
946  clib_memcpy_fast ((u8 *) (th + 1),
947  tm->wrk_ctx[tc->c_thread_index].cached_opts,
948  tc->snd_opts_len);
949  }
950  else
951  {
952  u8 len = tcp_options_write ((u8 *) (th + 1), &tc->snd_opts);
953  ASSERT (len == tc->snd_opts_len);
954  }
955 
956  /*
957  * Update connection variables
958  */
959 
960  if (update_snd_nxt)
961  tc->snd_nxt += data_len;
962  tc->rcv_las = tc->rcv_nxt;
963 
964  tc->bytes_out += data_len;
965  tc->data_segs_out += 1;
966 
967  th->checksum = tcp_compute_checksum (tc, b);
968 
969  TCP_EVT (TCP_EVT_PKTIZE, tc);
970 }
971 
974 {
975  u32 data_len = b->current_length;
976  if (PREDICT_FALSE (b->flags & VLIB_BUFFER_NEXT_PRESENT))
978  return data_len;
979 }
980 
981 u32
983 {
984  tcp_connection_t *tc = (tcp_connection_t *) tconn;
985 
986  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
988 
989  tcp_push_hdr_i (tc, b, tc->snd_nxt, /* compute opts */ 0, /* burst */ 1,
990  /* update_snd_nxt */ 1);
991 
992  tc->snd_una_max = seq_max (tc->snd_nxt, tc->snd_una_max);
993  tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una);
994  /* If not tracking an ACK, start tracking */
995  if (tc->rtt_ts == 0 && !tcp_in_cong_recovery (tc))
996  {
997  tc->rtt_ts = tcp_time_now_us (tc->c_thread_index);
998  tc->rtt_seq = tc->snd_nxt;
999  }
1000  if (PREDICT_FALSE (!tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT)))
1001  {
1002  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1004  tc->rto_boff = 0;
1005  }
1006  tcp_trajectory_add_start (b, 3);
1007  return 0;
1008 }
1009 
1010 void
1012 {
1013  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1014  vlib_main_t *vm = wrk->vm;
1015  vlib_buffer_t *b;
1016  u32 bi;
1017 
1018  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1019  {
1020  tcp_update_rcv_wnd (tc);
1021  return;
1022  }
1023  b = vlib_get_buffer (vm, bi);
1024  tcp_init_buffer (vm, b);
1025  tcp_make_ack (tc, b);
1026  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1027 }
1028 
1029 void
1031 {
1032  if (!(tc->flags & TCP_CONN_SNDACK))
1033  {
1034  session_add_self_custom_tx_evt (&tc->connection, 1);
1035  tc->flags |= TCP_CONN_SNDACK;
1036  }
1037 }
1038 
1039 void
1041 {
1042  if (!(tc->flags & TCP_CONN_SNDACK))
1043  {
1044  session_add_self_custom_tx_evt (&tc->connection, 1);
1045  tc->flags |= TCP_CONN_SNDACK;
1046  }
1047  if (tc->pending_dupacks < 255)
1048  tc->pending_dupacks += 1;
1049 }
1050 
1051 void
1053 {
1054  if (!(tc->flags & TCP_CONN_RXT_PENDING))
1055  {
1056  session_add_self_custom_tx_evt (&tc->connection, 0);
1057  tc->flags |= TCP_CONN_RXT_PENDING;
1058  }
1059 }
1060 
1061 /**
1062  * Delayed ack timer handler
1063  *
1064  * Sends delayed ACK when timer expires
1065  */
1066 void
1068 {
1069  tcp_send_ack (tc);
1070 }
1071 
1072 /**
1073  * Send window update ack
1074  *
1075  * Ensures that it will be sent only once, after a zero rwnd has been
1076  * advertised in a previous ack, and only if rwnd has grown beyond a
1077  * configurable value.
1078  */
1079 void
1081 {
1082  if (tcp_zero_rwnd_sent (tc))
1083  {
1084  tcp_update_rcv_wnd (tc);
1085  if (tc->rcv_wnd >= tcp_cfg.rwnd_min_update_ack * tc->snd_mss)
1086  {
1088  tcp_program_ack (tc);
1089  }
1090  }
1091 }
1092 
1093 /**
1094  * Allocate a new buffer and build a new tcp segment
1095  *
1096  * @param wrk tcp worker
1097  * @param tc connection for which the segment will be allocated
1098  * @param offset offset of the first byte in the tx fifo
1099  * @param max_deq_byte segment size
1100  * @param[out] b pointer to buffer allocated
1101  *
1102  * @return the number of bytes in the segment or 0 if buffer cannot be
1103  * allocated or no data available
1104  */
1105 static int
1107  u32 offset, u32 max_deq_bytes, vlib_buffer_t ** b)
1108 {
1109  u32 bytes_per_buffer = vnet_get_tcp_main ()->bytes_per_buffer;
1110  vlib_main_t *vm = wrk->vm;
1111  u32 bi, seg_size;
1112  int n_bytes = 0;
1113  u8 *data;
1114 
1115  seg_size = max_deq_bytes + TRANSPORT_MAX_HDRS_LEN;
1116 
1117  /*
1118  * Prepare options
1119  */
1120  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
1121 
1122  /*
1123  * Allocate and fill in buffer(s)
1124  */
1125 
1126  /* Easy case, buffer size greater than mss */
1127  if (PREDICT_TRUE (seg_size <= bytes_per_buffer))
1128  {
1129  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1130  return 0;
1131  *b = vlib_get_buffer (vm, bi);
1132  data = tcp_init_buffer (vm, *b);
1133  n_bytes = session_tx_fifo_peek_bytes (&tc->connection, data, offset,
1134  max_deq_bytes);
1135  ASSERT (n_bytes == max_deq_bytes);
1136  b[0]->current_length = n_bytes;
1137  tcp_push_hdr_i (tc, *b, tc->snd_una + offset, /* compute opts */ 0,
1138  /* burst */ 0, /* update_snd_nxt */ 0);
1139  }
1140  /* Split mss into multiple buffers */
1141  else
1142  {
1143  u32 chain_bi = ~0, n_bufs_per_seg, n_bufs;
1144  u16 n_peeked, len_to_deq;
1145  vlib_buffer_t *chain_b, *prev_b;
1146  int i;
1147 
1148  /* Make sure we have enough buffers */
1149  n_bufs_per_seg = ceil ((double) seg_size / bytes_per_buffer);
1150  vec_validate_aligned (wrk->tx_buffers, n_bufs_per_seg - 1,
1152  n_bufs = vlib_buffer_alloc (vm, wrk->tx_buffers, n_bufs_per_seg);
1153  if (PREDICT_FALSE (n_bufs != n_bufs_per_seg))
1154  {
1155  if (n_bufs)
1156  vlib_buffer_free (vm, wrk->tx_buffers, n_bufs);
1157  return 0;
1158  }
1159 
1160  *b = vlib_get_buffer (vm, wrk->tx_buffers[--n_bufs]);
1161  data = tcp_init_buffer (vm, *b);
1162  n_bytes = session_tx_fifo_peek_bytes (&tc->connection, data, offset,
1163  bytes_per_buffer -
1164  TRANSPORT_MAX_HDRS_LEN);
1165  b[0]->current_length = n_bytes;
1166  b[0]->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID;
1168  max_deq_bytes -= n_bytes;
1169 
1170  chain_b = *b;
1171  for (i = 1; i < n_bufs_per_seg; i++)
1172  {
1173  prev_b = chain_b;
1174  len_to_deq = clib_min (max_deq_bytes, bytes_per_buffer);
1175  chain_bi = wrk->tx_buffers[--n_bufs];
1176  chain_b = vlib_get_buffer (vm, chain_bi);
1177  chain_b->current_data = 0;
1178  data = vlib_buffer_get_current (chain_b);
1179  n_peeked = session_tx_fifo_peek_bytes (&tc->connection, data,
1180  offset + n_bytes,
1181  len_to_deq);
1182  ASSERT (n_peeked == len_to_deq);
1183  n_bytes += n_peeked;
1184  chain_b->current_length = n_peeked;
1185  chain_b->next_buffer = 0;
1186 
1187  /* update previous buffer */
1188  prev_b->next_buffer = chain_bi;
1189  prev_b->flags |= VLIB_BUFFER_NEXT_PRESENT;
1190 
1191  max_deq_bytes -= n_peeked;
1192  b[0]->total_length_not_including_first_buffer += n_peeked;
1193  }
1194 
1195  tcp_push_hdr_i (tc, *b, tc->snd_una + offset, /* compute opts */ 0,
1196  /* burst */ 0, /* update_snd_nxt */ 0);
1197 
1198  if (PREDICT_FALSE (n_bufs))
1199  {
1200  clib_warning ("not all buffers consumed");
1201  vlib_buffer_free (vm, wrk->tx_buffers, n_bufs);
1202  }
1203  }
1204 
1205  ASSERT (n_bytes > 0);
1206  ASSERT (((*b)->current_data + (*b)->current_length) <= bytes_per_buffer);
1207 
1208  return n_bytes;
1209 }
1210 
1211 /**
1212  * Build a retransmit segment
1213  *
1214  * @return the number of bytes in the segment or 0 if there's nothing to
1215  * retransmit
1216  */
1217 static u32
1219  tcp_connection_t * tc, u32 offset,
1220  u32 max_deq_bytes, vlib_buffer_t ** b)
1221 {
1222  u32 start, available_bytes;
1223  int n_bytes = 0;
1224 
1225  ASSERT (tc->state >= TCP_STATE_ESTABLISHED);
1226  ASSERT (max_deq_bytes != 0);
1227 
1228  /*
1229  * Make sure we can retransmit something
1230  */
1231  available_bytes = transport_max_tx_dequeue (&tc->connection);
1232  ASSERT (available_bytes >= offset);
1233  available_bytes -= offset;
1234  if (!available_bytes)
1235  return 0;
1236 
1237  max_deq_bytes = clib_min (tc->snd_mss, max_deq_bytes);
1238  max_deq_bytes = clib_min (available_bytes, max_deq_bytes);
1239 
1240  start = tc->snd_una + offset;
1241  ASSERT (seq_leq (start + max_deq_bytes, tc->snd_nxt));
1242 
1243  n_bytes = tcp_prepare_segment (wrk, tc, offset, max_deq_bytes, b);
1244  if (!n_bytes)
1245  return 0;
1246 
1247  tc->snd_rxt_bytes += n_bytes;
1248 
1249  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1250  tcp_bt_track_rxt (tc, start, start + n_bytes);
1251 
1252  tc->bytes_retrans += n_bytes;
1253  tc->segs_retrans += 1;
1254  tcp_worker_stats_inc (wrk, rxt_segs, 1);
1255  TCP_EVT (TCP_EVT_CC_RTX, tc, offset, n_bytes);
1256 
1257  return n_bytes;
1258 }
1259 
1260 static void
1262 {
1263  sack_scoreboard_t *sb = &tc->sack_sb;
1264  sack_scoreboard_hole_t *hole;
1265 
1266  hole = scoreboard_first_hole (sb);
1267  if (!sb->is_reneging && (!hole || hole->start == tc->snd_una))
1268  return;
1269 
1270  scoreboard_clear_reneging (sb, tc->snd_una, tc->snd_nxt);
1271 }
1272 
1273 /**
1274  * Reset congestion control, switch cwnd to loss window and try again.
1275  */
1276 static void
1278 {
1279  TCP_EVT (TCP_EVT_CC_EVT, tc, 6);
1280 
1281  tc->prev_ssthresh = tc->ssthresh;
1282  tc->prev_cwnd = tc->cwnd;
1283 
1284  /* If we entrered loss without fast recovery, notify cc algo of the
1285  * congestion event such that it can update ssthresh and its state */
1286  if (!tcp_in_fastrecovery (tc))
1287  tcp_cc_congestion (tc);
1288 
1289  /* Let cc algo decide loss cwnd and ssthresh post unrecovered loss */
1290  tcp_cc_loss (tc);
1291 
1292  tc->rtt_ts = 0;
1293  tc->cwnd_acc_bytes = 0;
1294  tc->tr_occurences += 1;
1295  tcp_recovery_on (tc);
1296 }
1297 
1298 void
1300 {
1301  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1302  vlib_main_t *vm = wrk->vm;
1303  vlib_buffer_t *b = 0;
1304  u32 bi, n_bytes;
1305 
1306  tcp_worker_stats_inc (wrk, tr_events, 1);
1307 
1308  /* Should be handled by a different handler */
1309  if (PREDICT_FALSE (tc->state == TCP_STATE_SYN_SENT))
1310  return;
1311 
1312  /* Wait-close and retransmit could pop at the same time */
1313  if (tc->state == TCP_STATE_CLOSED)
1314  return;
1315 
1316  if (tc->state >= TCP_STATE_ESTABLISHED)
1317  {
1318  TCP_EVT (TCP_EVT_CC_EVT, tc, 2);
1319 
1320  /* Lost FIN, retransmit and return */
1321  if (tc->flags & TCP_CONN_FINSNT)
1322  {
1323  tcp_send_fin (tc);
1324  tc->rto_boff += 1;
1325  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1326  return;
1327  }
1328 
1329  /* Shouldn't be here. This condition is tricky because it has to take
1330  * into account boff > 0 due to persist timeout. */
1331  if ((tc->rto_boff == 0 && tc->snd_una == tc->snd_nxt)
1332  || (tc->rto_boff > 0 && seq_geq (tc->snd_una, tc->snd_congestion)
1333  && !tcp_flight_size (tc)))
1334  {
1335  ASSERT (!tcp_in_recovery (tc));
1336  tc->rto_boff = 0;
1337  return;
1338  }
1339 
1340  /* We're not in recovery so make sure rto_boff is 0. Can be non 0 due
1341  * to persist timer timeout */
1342  if (!tcp_in_recovery (tc) && tc->rto_boff > 0)
1343  {
1344  tc->rto_boff = 0;
1345  tcp_update_rto (tc);
1346  }
1347 
1348  /* Peer is dead or network connectivity is lost. Close connection.
1349  * RFC 1122 section 4.2.3.5 recommends a value of at least 100s. For
1350  * a min rto of 0.2s we need to retry about 8 times. */
1351  if (tc->rto_boff >= TCP_RTO_BOFF_MAX)
1352  {
1353  tcp_send_reset (tc);
1354  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1355  session_transport_closing_notify (&tc->connection);
1356  session_transport_closed_notify (&tc->connection);
1358  tcp_program_cleanup (wrk, tc);
1359  tcp_worker_stats_inc (wrk, tr_abort, 1);
1360  return;
1361  }
1362 
1363  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1365 
1366  /* Update send congestion to make sure that rxt has data to send */
1367  tc->snd_congestion = tc->snd_nxt;
1368 
1369  /* Send the first unacked segment. If we're short on buffers, return
1370  * as soon as possible */
1371  n_bytes = clib_min (tc->snd_mss, tc->snd_nxt - tc->snd_una);
1372  n_bytes = tcp_prepare_retransmit_segment (wrk, tc, 0, n_bytes, &b);
1373  if (!n_bytes)
1374  {
1375  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT, 1);
1376  return;
1377  }
1378 
1379  bi = vlib_get_buffer_index (vm, b);
1380  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1381 
1382  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1384 
1385  tc->rto_boff += 1;
1386  if (tc->rto_boff == 1)
1387  {
1389  /* Record timestamp. Eifel detection algorithm RFC3522 */
1390  tc->snd_rxt_ts = tcp_tstamp (tc);
1391  }
1392 
1393  if (tcp_opts_sack_permitted (&tc->rcv_opts))
1394  scoreboard_init_rxt (&tc->sack_sb, tc->snd_una + n_bytes);
1395 
1397  }
1398  /* Retransmit SYN-ACK */
1399  else if (tc->state == TCP_STATE_SYN_RCVD)
1400  {
1401  TCP_EVT (TCP_EVT_CC_EVT, tc, 2);
1402 
1403  tc->rtt_ts = 0;
1404 
1405  /* Passive open establish timeout */
1406  if (tc->rto > TCP_ESTABLISH_TIME >> 1)
1407  {
1408  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1410  tcp_program_cleanup (wrk, tc);
1411  tcp_worker_stats_inc (wrk, tr_abort, 1);
1412  return;
1413  }
1414 
1415  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1416  {
1417  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT, 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  return;
1482  }
1483 
1484  /* Try without increasing RTO a number of times. If this fails,
1485  * start growing RTO exponentially */
1486  tc->rto_boff += 1;
1487  if (tc->rto_boff > TCP_RTO_SYN_RETRIES)
1488  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1489 
1490  b = vlib_get_buffer (vm, bi);
1491  tcp_init_buffer (vm, b);
1492  tcp_make_syn (tc, b);
1493 
1494  TCP_EVT (TCP_EVT_SYN_RXT, tc, 0);
1495 
1496  /* This goes straight to ipx_lookup */
1497  tcp_push_ip_hdr (wrk, tc, b);
1498  tcp_enqueue_to_ip_lookup (wrk, b, bi, tc->c_is_ip4, tc->c_fib_index);
1499 
1500  tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN,
1501  tc->rto * TCP_TO_TIMER_TICK);
1502 }
1503 
1504 /**
1505  * Got 0 snd_wnd from peer, try to do something about it.
1506  *
1507  */
1508 void
1510 {
1511  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1512  u32 bi, max_snd_bytes, available_bytes, offset;
1513  tcp_main_t *tm = vnet_get_tcp_main ();
1514  vlib_main_t *vm = wrk->vm;
1515  vlib_buffer_t *b;
1516  int n_bytes = 0;
1517  u8 *data;
1518 
1519  /* Problem already solved or worse */
1520  if (tc->state == TCP_STATE_CLOSED || tc->snd_wnd > tc->snd_mss
1521  || (tc->flags & TCP_CONN_FINSNT))
1522  goto update_scheduler;
1523 
1524  available_bytes = transport_max_tx_dequeue (&tc->connection);
1525  offset = tc->snd_nxt - tc->snd_una;
1526 
1527  /* Reprogram persist if no new bytes available to send. We may have data
1528  * next time */
1529  if (!available_bytes)
1530  {
1531  tcp_persist_timer_set (&wrk->timer_wheel, tc);
1532  return;
1533  }
1534 
1535  if (available_bytes <= offset)
1536  goto update_scheduler;
1537 
1538  /* Increment RTO backoff */
1539  tc->rto_boff += 1;
1540  tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1541 
1542  /*
1543  * Try to force the first unsent segment (or buffer)
1544  */
1545  if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1546  {
1547  tcp_persist_timer_set (&wrk->timer_wheel, tc);
1548  return;
1549  }
1550 
1551  b = vlib_get_buffer (vm, bi);
1552  data = tcp_init_buffer (vm, b);
1553 
1554  tcp_validate_txf_size (tc, offset);
1555  tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
1556  max_snd_bytes = clib_min (tc->snd_mss,
1557  tm->bytes_per_buffer - TRANSPORT_MAX_HDRS_LEN);
1558  n_bytes = session_tx_fifo_peek_bytes (&tc->connection, data, offset,
1559  max_snd_bytes);
1560  b->current_length = n_bytes;
1561  ASSERT (n_bytes != 0 && (tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT)
1562  || tc->snd_nxt == tc->snd_una_max
1563  || tc->rto_boff > 1));
1564 
1565  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1566  {
1568  tcp_bt_track_tx (tc, n_bytes);
1569  }
1570 
1571  tcp_push_hdr_i (tc, b, tc->snd_nxt, /* compute opts */ 0,
1572  /* burst */ 0, /* update_snd_nxt */ 1);
1573  tc->snd_una_max = seq_max (tc->snd_nxt, tc->snd_una_max);
1574  tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una);
1575  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1576 
1577  /* Just sent new data, enable retransmit */
1579 
1580  return;
1581 
1582 update_scheduler:
1583 
1584  if (tcp_is_descheduled (tc))
1585  transport_connection_reschedule (&tc->connection);
1586 }
1587 
1588 /**
1589  * Retransmit first unacked segment
1590  */
1591 int
1593 {
1594  vlib_main_t *vm = wrk->vm;
1595  vlib_buffer_t *b;
1596  u32 bi, n_bytes;
1597 
1598  TCP_EVT (TCP_EVT_CC_EVT, tc, 1);
1599 
1600  n_bytes = tcp_prepare_retransmit_segment (wrk, tc, 0, tc->snd_mss, &b);
1601  if (!n_bytes)
1602  return -1;
1603 
1604  bi = vlib_get_buffer_index (vm, b);
1605  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1606 
1607  return 0;
1608 }
1609 
1610 static int
1612  u32 burst_size)
1613 {
1614  u32 offset, n_segs = 0, n_written, bi, available_wnd;
1615  vlib_main_t *vm = wrk->vm;
1616  vlib_buffer_t *b = 0;
1617 
1618  offset = tc->snd_nxt - tc->snd_una;
1619  available_wnd = tc->snd_wnd - offset;
1620  burst_size = clib_min (burst_size, available_wnd / tc->snd_mss);
1621 
1622  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1624 
1625  while (n_segs < burst_size)
1626  {
1627  n_written = tcp_prepare_segment (wrk, tc, offset, tc->snd_mss, &b);
1628  if (!n_written)
1629  goto done;
1630 
1631  bi = vlib_get_buffer_index (vm, b);
1632  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1633  offset += n_written;
1634  n_segs += 1;
1635 
1636  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1637  tcp_bt_track_tx (tc, n_written);
1638 
1639  tc->snd_nxt += n_written;
1640  tc->snd_una_max = seq_max (tc->snd_nxt, tc->snd_una_max);
1641  }
1642 
1643 done:
1644  return n_segs;
1645 }
1646 
1647 /**
1648  * Estimate send space using proportional rate reduction (RFC6937)
1649  */
1650 int
1652 {
1653  u32 pipe, prr_out;
1654  int space;
1655 
1656  pipe = tcp_flight_size (tc);
1657  prr_out = tc->snd_rxt_bytes + (tc->snd_nxt - tc->snd_congestion);
1658 
1659  if (pipe > tc->ssthresh)
1660  {
1661  space = ((int) tc->prr_delivered * ((f64) tc->ssthresh / tc->prev_cwnd))
1662  - prr_out;
1663  }
1664  else
1665  {
1666  int limit;
1667  limit = clib_max ((int) (tc->prr_delivered - prr_out), 0) + tc->snd_mss;
1668  space = clib_min (tc->ssthresh - pipe, limit);
1669  }
1670  space = clib_max (space, prr_out ? 0 : tc->snd_mss);
1671  return space;
1672 }
1673 
1674 static inline u8
1676  sack_scoreboard_t * sb)
1677 {
1678  u32 tx_adv_sack = sb->high_sacked - tc->snd_congestion;
1679  f64 rr = (f64) tc->ssthresh / tc->prev_cwnd;
1680 
1681  if (tcp_fastrecovery_first (tc))
1682  return 1;
1683 
1684  return (tx_adv_sack > (tc->snd_una - tc->prr_start) * rr);
1685 }
1686 
1687 static inline u8
1689 {
1690  return (transport_max_tx_dequeue (&tc->connection)
1691  - (tc->snd_nxt - tc->snd_una));
1692 }
1693 
1694 #define scoreboard_rescue_rxt_valid(_sb, _tc) \
1695  (seq_geq (_sb->rescue_rxt, _tc->snd_una) \
1696  && seq_leq (_sb->rescue_rxt, _tc->snd_congestion))
1697 
1698 /**
1699  * Do retransmit with SACKs
1700  */
1701 static int
1703  u32 burst_size)
1704 {
1705  u32 n_written = 0, offset, max_bytes, n_segs = 0;
1706  u8 snd_limited = 0, can_rescue = 0;
1707  u32 bi, max_deq, burst_bytes;
1708  sack_scoreboard_hole_t *hole;
1709  vlib_main_t *vm = wrk->vm;
1710  vlib_buffer_t *b = 0;
1711  sack_scoreboard_t *sb;
1712  int snd_space;
1713 
1715 
1716  burst_bytes = transport_connection_tx_pacer_burst (&tc->connection);
1717  burst_size = clib_min (burst_size, burst_bytes / tc->snd_mss);
1718  if (!burst_size)
1719  {
1721  return 0;
1722  }
1723 
1724  if (tcp_in_recovery (tc))
1725  snd_space = tcp_available_cc_snd_space (tc);
1726  else
1727  snd_space = tcp_fastrecovery_prr_snd_space (tc);
1728 
1729  if (snd_space < tc->snd_mss)
1730  goto done;
1731 
1732  sb = &tc->sack_sb;
1733 
1734  /* Check if snd_una is a lost retransmit */
1735  if (pool_elts (sb->holes)
1736  && seq_gt (sb->high_sacked, tc->snd_congestion)
1737  && tc->rxt_head != tc->snd_una
1739  {
1740  max_bytes = clib_min (tc->snd_mss, tc->snd_congestion - tc->snd_una);
1741  n_written = tcp_prepare_retransmit_segment (wrk, tc, 0, max_bytes, &b);
1742  if (!n_written)
1743  {
1745  goto done;
1746  }
1747  bi = vlib_get_buffer_index (vm, b);
1748  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1749  n_segs = 1;
1750 
1751  tc->rxt_head = tc->snd_una;
1752  tc->rxt_delivered += n_written;
1753  tc->prr_delivered += n_written;
1754  ASSERT (tc->rxt_delivered <= tc->snd_rxt_bytes);
1755  }
1756 
1758 
1759  TCP_EVT (TCP_EVT_CC_EVT, tc, 0);
1760  hole = scoreboard_get_hole (sb, sb->cur_rxt_hole);
1761 
1762  max_deq = transport_max_tx_dequeue (&tc->connection);
1763  max_deq -= tc->snd_nxt - tc->snd_una;
1764 
1765  while (snd_space > 0 && n_segs < burst_size)
1766  {
1767  hole = scoreboard_next_rxt_hole (sb, hole, max_deq != 0, &can_rescue,
1768  &snd_limited);
1769  if (!hole)
1770  {
1771  /* We are out of lost holes to retransmit so send some new data. */
1772  if (max_deq > tc->snd_mss)
1773  {
1774  u32 n_segs_new;
1775  int av_wnd;
1776 
1777  /* Make sure we don't exceed available window and leave space
1778  * for one more packet, to avoid zero window acks */
1779  av_wnd = (int) tc->snd_wnd - (tc->snd_nxt - tc->snd_una);
1780  av_wnd = clib_max (av_wnd - tc->snd_mss, 0);
1781  snd_space = clib_min (snd_space, av_wnd);
1782  snd_space = clib_min (max_deq, snd_space);
1783  burst_size = clib_min (burst_size - n_segs,
1784  snd_space / tc->snd_mss);
1785  burst_size = clib_min (burst_size, TCP_RXT_MAX_BURST);
1786  n_segs_new = tcp_transmit_unsent (wrk, tc, burst_size);
1787  if (max_deq > n_segs_new * tc->snd_mss)
1789 
1790  n_segs += n_segs_new;
1791  goto done;
1792  }
1793 
1794  if (tcp_in_recovery (tc) || !can_rescue
1795  || scoreboard_rescue_rxt_valid (sb, tc))
1796  break;
1797 
1798  /* If rescue rxt undefined or less than snd_una then one segment of
1799  * up to SMSS octets that MUST include the highest outstanding
1800  * unSACKed sequence number SHOULD be returned, and RescueRxt set to
1801  * RecoveryPoint. HighRxt MUST NOT be updated.
1802  */
1803  hole = scoreboard_last_hole (sb);
1804  max_bytes = clib_min (tc->snd_mss, hole->end - hole->start);
1805  max_bytes = clib_min (max_bytes, snd_space);
1806  offset = hole->end - tc->snd_una - max_bytes;
1807  n_written = tcp_prepare_retransmit_segment (wrk, tc, offset,
1808  max_bytes, &b);
1809  if (!n_written)
1810  goto done;
1811 
1812  sb->rescue_rxt = tc->snd_congestion;
1813  bi = vlib_get_buffer_index (vm, b);
1814  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1815  n_segs += 1;
1816  break;
1817  }
1818 
1819  max_bytes = clib_min (hole->end - sb->high_rxt, snd_space);
1820  max_bytes = snd_limited ? clib_min (max_bytes, tc->snd_mss) : max_bytes;
1821  if (max_bytes == 0)
1822  break;
1823 
1824  offset = sb->high_rxt - tc->snd_una;
1825  n_written = tcp_prepare_retransmit_segment (wrk, tc, offset, max_bytes,
1826  &b);
1827  ASSERT (n_written <= snd_space);
1828 
1829  /* Nothing left to retransmit */
1830  if (n_written == 0)
1831  break;
1832 
1833  bi = vlib_get_buffer_index (vm, b);
1834  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1835 
1836  sb->high_rxt += n_written;
1837  ASSERT (seq_leq (sb->high_rxt, tc->snd_nxt));
1838 
1839  snd_space -= n_written;
1840  n_segs += 1;
1841  }
1842 
1843  if (hole)
1845 
1846 done:
1847 
1848  transport_connection_tx_pacer_reset_bucket (&tc->connection, 0);
1849  return n_segs;
1850 }
1851 
1852 /**
1853  * Fast retransmit without SACK info
1854  */
1855 static int
1857  u32 burst_size)
1858 {
1859  u32 n_written = 0, offset = 0, bi, max_deq, n_segs_now, max_bytes;
1860  u32 burst_bytes, sent_bytes;
1861  vlib_main_t *vm = wrk->vm;
1862  int snd_space, n_segs = 0;
1863  u8 cc_limited = 0;
1864  vlib_buffer_t *b;
1865 
1867  TCP_EVT (TCP_EVT_CC_EVT, tc, 0);
1868 
1869  burst_bytes = transport_connection_tx_pacer_burst (&tc->connection);
1870  burst_size = clib_min (burst_size, burst_bytes / tc->snd_mss);
1871  if (!burst_size)
1872  {
1874  return 0;
1875  }
1876 
1877  snd_space = tcp_available_cc_snd_space (tc);
1878  cc_limited = snd_space < burst_bytes;
1879 
1880  if (!tcp_fastrecovery_first (tc))
1881  goto send_unsent;
1882 
1883  /* RFC 6582: [If a partial ack], retransmit the first unacknowledged
1884  * segment. */
1885  while (snd_space > 0 && n_segs < burst_size)
1886  {
1887  max_bytes = clib_min (tc->snd_mss,
1888  tc->snd_congestion - tc->snd_una - offset);
1889  if (!max_bytes)
1890  break;
1891  n_written = tcp_prepare_retransmit_segment (wrk, tc, offset, max_bytes,
1892  &b);
1893 
1894  /* Nothing left to retransmit */
1895  if (n_written == 0)
1896  break;
1897 
1898  bi = vlib_get_buffer_index (vm, b);
1899  tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1900  snd_space -= n_written;
1901  offset += n_written;
1902  n_segs += 1;
1903  }
1904 
1905  if (n_segs == burst_size)
1906  goto done;
1907 
1908 send_unsent:
1909 
1910  /* RFC 6582: Send a new segment if permitted by the new value of cwnd. */
1911  if (snd_space < tc->snd_mss || tc->snd_mss == 0)
1912  goto done;
1913 
1914  max_deq = transport_max_tx_dequeue (&tc->connection);
1915  max_deq -= tc->snd_nxt - tc->snd_una;
1916  if (max_deq)
1917  {
1918  snd_space = clib_min (max_deq, snd_space);
1919  burst_size = clib_min (burst_size - n_segs, snd_space / tc->snd_mss);
1920  n_segs_now = tcp_transmit_unsent (wrk, tc, burst_size);
1921  if (n_segs_now && max_deq > n_segs_now * tc->snd_mss)
1923  n_segs += n_segs_now;
1924  }
1925 
1926 done:
1928 
1929  sent_bytes = clib_min (n_segs * tc->snd_mss, burst_bytes);
1930  sent_bytes = cc_limited ? burst_bytes : sent_bytes;
1931  transport_connection_tx_pacer_update_bytes (&tc->connection, sent_bytes);
1932 
1933  return n_segs;
1934 }
1935 
1936 static int
1937 tcp_send_acks (tcp_connection_t * tc, u32 max_burst_size)
1938 {
1939  int j, n_acks;
1940 
1941  if (!tc->pending_dupacks)
1942  {
1943  if (tcp_in_cong_recovery (tc) || !tcp_max_tx_deq (tc)
1944  || tc->state != TCP_STATE_ESTABLISHED)
1945  {
1946  tcp_send_ack (tc);
1947  return 1;
1948  }
1949  return 0;
1950  }
1951 
1952  /* If we're supposed to send dupacks but have no ooo data
1953  * send only one ack */
1954  if (!vec_len (tc->snd_sacks))
1955  {
1956  tcp_send_ack (tc);
1957  tc->dupacks_out += 1;
1958  tc->pending_dupacks = 0;
1959  return 1;
1960  }
1961 
1962  /* Start with first sack block */
1963  tc->snd_sack_pos = 0;
1964 
1965  /* Generate enough dupacks to cover all sack blocks. Do not generate
1966  * more sacks than the number of packets received. But do generate at
1967  * least 3, i.e., the number needed to signal congestion, if needed. */
1968  n_acks = vec_len (tc->snd_sacks) / TCP_OPTS_MAX_SACK_BLOCKS;
1969  n_acks = clib_min (n_acks, tc->pending_dupacks);
1970  n_acks = clib_max (n_acks, clib_min (tc->pending_dupacks, 3));
1971  for (j = 0; j < clib_min (n_acks, max_burst_size); j++)
1972  tcp_send_ack (tc);
1973 
1974  if (n_acks < max_burst_size)
1975  {
1976  tc->pending_dupacks = 0;
1977  tc->snd_sack_pos = 0;
1978  tc->dupacks_out += n_acks;
1979  return n_acks;
1980  }
1981  else
1982  {
1983  TCP_DBG ("constrained by burst size");
1984  tc->pending_dupacks = n_acks - max_burst_size;
1985  tc->dupacks_out += max_burst_size;
1986  tcp_program_dupack (tc);
1987  return max_burst_size;
1988  }
1989 }
1990 
1991 static int
1993 {
1994  tcp_worker_ctx_t *wrk;
1995  u32 n_segs;
1996 
1997  if (PREDICT_FALSE (tc->state == TCP_STATE_CLOSED))
1998  return 0;
1999 
2000  wrk = tcp_get_worker (tc->c_thread_index);
2001 
2002  if (tcp_opts_sack_permitted (&tc->rcv_opts))
2003  n_segs = tcp_retransmit_sack (wrk, tc, max_burst_size);
2004  else
2005  n_segs = tcp_retransmit_no_sack (wrk, tc, max_burst_size);
2006 
2007  return n_segs;
2008 }
2009 
2010 int
2012 {
2013  tcp_connection_t *tc = (tcp_connection_t *) conn;
2014  u32 n_segs = 0;
2015 
2016  if (tcp_in_cong_recovery (tc) && (tc->flags & TCP_CONN_RXT_PENDING))
2017  {
2018  tc->flags &= ~TCP_CONN_RXT_PENDING;
2019  n_segs = tcp_do_retransmit (tc, sp->max_burst_size);
2020  }
2021 
2022  if (!(tc->flags & TCP_CONN_SNDACK))
2023  return n_segs;
2024 
2025  tc->flags &= ~TCP_CONN_SNDACK;
2026 
2027  /* We have retransmitted packets and no dupack */
2028  if (n_segs && !tc->pending_dupacks)
2029  return n_segs;
2030 
2031  if (sp->max_burst_size <= n_segs)
2032  {
2033  tcp_program_ack (tc);
2034  return n_segs;
2035  }
2036 
2037  n_segs += tcp_send_acks (tc, sp->max_burst_size - n_segs);
2038 
2039  return n_segs;
2040 }
2041 #endif /* CLIB_MARCH_VARIANT */
2042 
2043 static void
2045  u16 * next0, u32 * error0)
2046 {
2047  ip_adjacency_t *adj;
2048  adj_index_t ai;
2049 
2050  /* Not thread safe but as long as the connection exists the adj should
2051  * not be removed */
2052  ai = adj_nbr_find (FIB_PROTOCOL_IP6, VNET_LINK_IP6, &tc0->c_rmt_ip,
2053  tc0->sw_if_index);
2054  if (ai == ADJ_INDEX_INVALID)
2055  {
2056  vnet_buffer (b0)->sw_if_index[VLIB_TX] = ~0;
2057  *next0 = TCP_OUTPUT_NEXT_DROP;
2058  *error0 = TCP_ERROR_LINK_LOCAL_RW;
2059  return;
2060  }
2061 
2062  adj = adj_get (ai);
2064  *next0 = TCP_OUTPUT_NEXT_IP_REWRITE;
2065  else if (adj->lookup_next_index == IP_LOOKUP_NEXT_ARP)
2066  *next0 = TCP_OUTPUT_NEXT_IP_ARP;
2067  else
2068  {
2069  *next0 = TCP_OUTPUT_NEXT_DROP;
2070  *error0 = TCP_ERROR_LINK_LOCAL_RW;
2071  }
2072  vnet_buffer (b0)->ip.adj_index[VLIB_TX] = ai;
2073 }
2074 
2075 static void
2077  u32 * to_next, u32 n_bufs)
2078 {
2079  tcp_connection_t *tc;
2080  tcp_tx_trace_t *t;
2081  vlib_buffer_t *b;
2082  tcp_header_t *th;
2083  int i;
2084 
2085  for (i = 0; i < n_bufs; i++)
2086  {
2087  b = vlib_get_buffer (vm, to_next[i]);
2088  if (!(b->flags & VLIB_BUFFER_IS_TRACED))
2089  continue;
2090  th = vlib_buffer_get_current (b);
2091  tc = tcp_connection_get (vnet_buffer (b)->tcp.connection_index,
2092  vm->thread_index);
2093  t = vlib_add_trace (vm, node, b, sizeof (*t));
2094  clib_memcpy_fast (&t->tcp_header, th, sizeof (t->tcp_header));
2095  clib_memcpy_fast (&t->tcp_connection, tc, sizeof (t->tcp_connection));
2096  }
2097 }
2098 
2099 always_inline void
2101  tcp_connection_t * tc0, u8 is_ip4)
2102 {
2103  TCP_EVT (TCP_EVT_OUTPUT, tc0,
2105  b0->current_length);
2106 
2107  if (is_ip4)
2108  vlib_buffer_push_ip4 (vm, b0, &tc0->c_lcl_ip4, &tc0->c_rmt_ip4,
2109  IP_PROTOCOL_TCP, tcp_csum_offload (tc0));
2110  else
2111  vlib_buffer_push_ip6_custom (vm, b0, &tc0->c_lcl_ip6, &tc0->c_rmt_ip6,
2112  IP_PROTOCOL_TCP, tc0->ipv6_flow_label);
2113 }
2114 
2115 always_inline void
2117 {
2118  if (PREDICT_TRUE (!(tc->cfg_flags & TCP_CFG_F_TSO)))
2119  return;
2120 
2121  u16 data_len = b->current_length - sizeof (tcp_header_t) - tc->snd_opts_len;
2122 
2123  if (PREDICT_FALSE (b->flags & VLIB_BUFFER_TOTAL_LENGTH_VALID))
2125 
2126  if (PREDICT_TRUE (data_len <= tc->snd_mss))
2127  return;
2128  else
2129  {
2130  ASSERT ((b->flags & VNET_BUFFER_F_L3_HDR_OFFSET_VALID) != 0);
2131  ASSERT ((b->flags & VNET_BUFFER_F_L4_HDR_OFFSET_VALID) != 0);
2132  b->flags |= VNET_BUFFER_F_GSO;
2133  vnet_buffer2 (b)->gso_l4_hdr_sz =
2134  sizeof (tcp_header_t) + tc->snd_opts_len;
2135  vnet_buffer2 (b)->gso_size = tc->snd_mss;
2136  }
2137 }
2138 
2139 always_inline void
2141  vlib_node_runtime_t * error_node, u16 * next0,
2142  u8 is_ip4)
2143 {
2144  /* If next_index is not drop use it */
2145  if (tc0->next_node_index)
2146  {
2147  *next0 = tc0->next_node_index;
2148  vnet_buffer (b0)->tcp.next_node_opaque = tc0->next_node_opaque;
2149  }
2150  else
2151  {
2152  *next0 = TCP_OUTPUT_NEXT_IP_LOOKUP;
2153  }
2154 
2155  vnet_buffer (b0)->sw_if_index[VLIB_TX] = tc0->c_fib_index;
2156  vnet_buffer (b0)->sw_if_index[VLIB_RX] = 0;
2157 
2158  if (!is_ip4)
2159  {
2160  u32 error0 = 0;
2161 
2162  if (PREDICT_FALSE (ip6_address_is_link_local_unicast (&tc0->c_rmt_ip6)))
2163  tcp_output_handle_link_local (tc0, b0, next0, &error0);
2164 
2165  if (PREDICT_FALSE (error0))
2166  {
2167  b0->error = error_node->errors[error0];
2168  return;
2169  }
2170  }
2171 
2172  tc0->segs_out += 1;
2173 }
2174 
2177  vlib_frame_t * frame, int is_ip4)
2178 {
2179  u32 n_left_from, *from, thread_index = vm->thread_index;
2180  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
2181  u16 nexts[VLIB_FRAME_SIZE], *next;
2182 
2183  from = vlib_frame_vector_args (frame);
2184  n_left_from = frame->n_vectors;
2185  tcp_set_time_now (tcp_get_worker (thread_index));
2186 
2188  tcp46_output_trace_frame (vm, node, from, n_left_from);
2189 
2190  vlib_get_buffers (vm, from, bufs, n_left_from);
2191  b = bufs;
2192  next = nexts;
2193 
2194  while (n_left_from >= 4)
2195  {
2196  tcp_connection_t *tc0, *tc1;
2197 
2198  {
2199  vlib_prefetch_buffer_header (b[2], STORE);
2200  CLIB_PREFETCH (b[2]->data, 2 * CLIB_CACHE_LINE_BYTES, STORE);
2201 
2202  vlib_prefetch_buffer_header (b[3], STORE);
2203  CLIB_PREFETCH (b[3]->data, 2 * CLIB_CACHE_LINE_BYTES, STORE);
2204  }
2205 
2206  tc0 = tcp_connection_get (vnet_buffer (b[0])->tcp.connection_index,
2207  thread_index);
2208  tc1 = tcp_connection_get (vnet_buffer (b[1])->tcp.connection_index,
2209  thread_index);
2210 
2211  if (PREDICT_TRUE (!tc0 + !tc1 == 0))
2212  {
2213  tcp_output_push_ip (vm, b[0], tc0, is_ip4);
2214  tcp_output_push_ip (vm, b[1], tc1, is_ip4);
2215 
2216  tcp_check_if_gso (tc0, b[0]);
2217  tcp_check_if_gso (tc1, b[1]);
2218 
2219  tcp_output_handle_packet (tc0, b[0], node, &next[0], is_ip4);
2220  tcp_output_handle_packet (tc1, b[1], node, &next[1], is_ip4);
2221  }
2222  else
2223  {
2224  if (tc0 != 0)
2225  {
2226  tcp_output_push_ip (vm, b[0], tc0, is_ip4);
2227  tcp_check_if_gso (tc0, b[0]);
2228  tcp_output_handle_packet (tc0, b[0], node, &next[0], is_ip4);
2229  }
2230  else
2231  {
2232  b[0]->error = node->errors[TCP_ERROR_INVALID_CONNECTION];
2233  next[0] = TCP_OUTPUT_NEXT_DROP;
2234  }
2235  if (tc1 != 0)
2236  {
2237  tcp_output_push_ip (vm, b[1], tc1, is_ip4);
2238  tcp_check_if_gso (tc1, b[1]);
2239  tcp_output_handle_packet (tc1, b[1], node, &next[1], is_ip4);
2240  }
2241  else
2242  {
2243  b[1]->error = node->errors[TCP_ERROR_INVALID_CONNECTION];
2244  next[1] = TCP_OUTPUT_NEXT_DROP;
2245  }
2246  }
2247 
2248  b += 2;
2249  next += 2;
2250  n_left_from -= 2;
2251  }
2252  while (n_left_from > 0)
2253  {
2254  tcp_connection_t *tc0;
2255 
2256  if (n_left_from > 1)
2257  {
2258  vlib_prefetch_buffer_header (b[1], STORE);
2259  CLIB_PREFETCH (b[1]->data, 2 * CLIB_CACHE_LINE_BYTES, STORE);
2260  }
2261 
2262  tc0 = tcp_connection_get (vnet_buffer (b[0])->tcp.connection_index,
2263  thread_index);
2264 
2265  if (PREDICT_TRUE (tc0 != 0))
2266  {
2267  tcp_output_push_ip (vm, b[0], tc0, is_ip4);
2268  tcp_check_if_gso (tc0, b[0]);
2269  tcp_output_handle_packet (tc0, b[0], node, &next[0], is_ip4);
2270  }
2271  else
2272  {
2273  b[0]->error = node->errors[TCP_ERROR_INVALID_CONNECTION];
2274  next[0] = TCP_OUTPUT_NEXT_DROP;
2275  }
2276 
2277  b += 1;
2278  next += 1;
2279  n_left_from -= 1;
2280  }
2281 
2282  vlib_buffer_enqueue_to_next (vm, node, from, nexts, frame->n_vectors);
2283  vlib_node_increment_counter (vm, tcp_node_index (output, is_ip4),
2284  TCP_ERROR_PKTS_SENT, frame->n_vectors);
2285  return frame->n_vectors;
2286 }
2287 
2289  vlib_frame_t * from_frame)
2290 {
2291  return tcp46_output_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2292 }
2293 
2295  vlib_frame_t * from_frame)
2296 {
2297  return tcp46_output_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2298 }
2299 
2300 /* *INDENT-OFF* */
2302 {
2303  .name = "tcp4-output",
2304  /* Takes a vector of packets. */
2305  .vector_size = sizeof (u32),
2306  .n_errors = TCP_N_ERROR,
2307  .protocol_hint = VLIB_NODE_PROTO_HINT_TCP,
2308  .error_strings = tcp_error_strings,
2309  .n_next_nodes = TCP_OUTPUT_N_NEXT,
2310  .next_nodes = {
2311 #define _(s,n) [TCP_OUTPUT_NEXT_##s] = n,
2313 #undef _
2314  },
2315  .format_buffer = format_tcp_header,
2316  .format_trace = format_tcp_tx_trace,
2317 };
2318 /* *INDENT-ON* */
2319 
2320 /* *INDENT-OFF* */
2322 {
2323  .name = "tcp6-output",
2324  /* Takes a vector of packets. */
2325  .vector_size = sizeof (u32),
2326  .n_errors = TCP_N_ERROR,
2327  .protocol_hint = VLIB_NODE_PROTO_HINT_TCP,
2328  .error_strings = tcp_error_strings,
2329  .n_next_nodes = TCP_OUTPUT_N_NEXT,
2330  .next_nodes = {
2331 #define _(s,n) [TCP_OUTPUT_NEXT_##s] = n,
2333 #undef _
2334  },
2335  .format_buffer = format_tcp_header,
2336  .format_trace = format_tcp_tx_trace,
2337 };
2338 /* *INDENT-ON* */
2339 
2340 typedef enum _tcp_reset_next
2341 {
2346 
2347 #define foreach_tcp4_reset_next \
2348  _(DROP, "error-drop") \
2349  _(IP_LOOKUP, "ip4-lookup")
2350 
2351 #define foreach_tcp6_reset_next \
2352  _(DROP, "error-drop") \
2353  _(IP_LOOKUP, "ip6-lookup")
2354 
2355 static uword
2357  vlib_frame_t * from_frame, u8 is_ip4)
2358 {
2359  u32 error0 = TCP_ERROR_RST_SENT, next0 = TCP_RESET_NEXT_IP_LOOKUP;
2360  u32 n_left_from, next_index, *from, *to_next;
2361 
2362  from = vlib_frame_vector_args (from_frame);
2363  n_left_from = from_frame->n_vectors;
2364 
2365  next_index = node->cached_next_index;
2366 
2367  while (n_left_from > 0)
2368  {
2369  u32 n_left_to_next;
2370 
2371  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
2372 
2373  while (n_left_from > 0 && n_left_to_next > 0)
2374  {
2375  vlib_buffer_t *b0;
2376  tcp_tx_trace_t *t0;
2377  tcp_header_t *th0;
2378  u32 bi0;
2379 
2380  bi0 = from[0];
2381  to_next[0] = bi0;
2382  from += 1;
2383  to_next += 1;
2384  n_left_from -= 1;
2385  n_left_to_next -= 1;
2386 
2387  b0 = vlib_get_buffer (vm, bi0);
2388  tcp_make_reset_in_place (vm, b0, is_ip4);
2389 
2390  /* Prepare to send to IP lookup */
2391  vnet_buffer (b0)->sw_if_index[VLIB_TX] = ~0;
2392 
2393  b0->error = node->errors[error0];
2394  b0->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
2395  if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
2396  {
2397  th0 = vlib_buffer_get_current (b0);
2398  if (is_ip4)
2399  th0 = ip4_next_header ((ip4_header_t *) th0);
2400  else
2401  th0 = ip6_next_header ((ip6_header_t *) th0);
2402  t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
2403  clib_memcpy_fast (&t0->tcp_header, th0,
2404  sizeof (t0->tcp_header));
2405  }
2406 
2407  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
2408  n_left_to_next, bi0, next0);
2409  }
2410  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
2411  }
2412  return from_frame->n_vectors;
2413 }
2414 
2416  vlib_frame_t * from_frame)
2417 {
2418  return tcp46_send_reset_inline (vm, node, from_frame, 1);
2419 }
2420 
2422  vlib_frame_t * from_frame)
2423 {
2424  return tcp46_send_reset_inline (vm, node, from_frame, 0);
2425 }
2426 
2427 /* *INDENT-OFF* */
2429  .name = "tcp4-reset",
2430  .vector_size = sizeof (u32),
2431  .n_errors = TCP_N_ERROR,
2432  .error_strings = tcp_error_strings,
2433  .n_next_nodes = TCP_RESET_N_NEXT,
2434  .next_nodes = {
2435 #define _(s,n) [TCP_RESET_NEXT_##s] = n,
2437 #undef _
2438  },
2439  .format_trace = format_tcp_tx_trace,
2440 };
2441 /* *INDENT-ON* */
2442 
2443 /* *INDENT-OFF* */
2445  .name = "tcp6-reset",
2446  .vector_size = sizeof (u32),
2447  .n_errors = TCP_N_ERROR,
2448  .error_strings = tcp_error_strings,
2449  .n_next_nodes = TCP_RESET_N_NEXT,
2450  .next_nodes = {
2451 #define _(s,n) [TCP_RESET_NEXT_##s] = n,
2453 #undef _
2454  },
2455  .format_trace = format_tcp_tx_trace,
2456 };
2457 /* *INDENT-ON* */
2458 
2459 /*
2460  * fd.io coding-style-patch-verification: ON
2461  *
2462  * Local Variables:
2463  * eval: (c-set-style "gnu")
2464  * End:
2465  */
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:2116
static void tcp_check_sack_reneging(tcp_connection_t *tc)
Definition: tcp_output.c:1261
#define TCP_RXT_MAX_BURST
Definition: tcp_types.h:33
#define clib_min(x, y)
Definition: clib.h:319
static int tcp_send_acks(tcp_connection_t *tc, u32 max_burst_size)
Definition: tcp_output.c:1937
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:392
#define CLIB_UNUSED(x)
Definition: clib.h:86
static void tcp_persist_timer_set(tcp_timer_wheel_t *tw, tcp_connection_t *tc)
Definition: tcp_timer.h:76
#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:1509
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:170
#define tcp_node_index(node_id, is_ip4)
Definition: tcp.h:274
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:1647
#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:517
#define tcp_opts_tstamp(_to)
Definition: tcp_packet.h:156
#define PREDICT_TRUE(x)
Definition: clib.h:119
i16 current_data
signed offset in data[], pre_data[] that we are currently processing.
Definition: buffer.h:110
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:1992
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:421
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:1299
struct _tcp_main tcp_main_t
u32 thread_index
Definition: main.h:218
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:403
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:1611
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
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:88
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:775
#define vec_validate_aligned(V, I, A)
Make sure vector is long enough for given index (no header, specified alignment)
Definition: vec.h:518
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:2356
vlib_error_t * errors
Vector of errors for this node.
Definition: node.h:472
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:382
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:1694
ip6_address_t src_address
Definition: ip6_packet.h:310
#define tcp_in_cong_recovery(tc)
Definition: tcp_types.h:423
unsigned char u8
Definition: types.h:56
u8 wscale
Option flags, see above.
Definition: tcp_packet.h:146
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:2347
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:1218
u16 src_port
Definition: udp.api:41
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:2011
void session_transport_closing_notify(transport_connection_t *tc)
Notification from transport that connection is being closed.
Definition: session.c:945
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:2176
static void * tcp_init_buffer(vlib_main_t *vm, vlib_buffer_t *b)
Definition: tcp_output.c:351
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:491
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:1106
#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:482
#define tcp_cfg
Definition: tcp.h:273
vl_api_interface_index_t sw_if_index
Definition: gre.api:53
#define TCP_OPTION_LEN_SACK_BLOCK
Definition: tcp_packet.h:168
ip4_address_t dst_address
Definition: ip4_packet.h:170
#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:85
static void tcp_retransmit_timer_force_update(tcp_timer_wheel_t *tw, tcp_connection_t *tc)
Definition: tcp_timer.h:68
static void * ip4_next_header(ip4_header_t *i)
Definition: ip4_packet.h:241
#define tcp_zero_rwnd_sent(tc)
Definition: tcp_types.h:435
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:2076
#define VLIB_FRAME_SIZE
Definition: node.h:380
#define tcp_validate_txf_size(_tc, _a)
Definition: tcp.h:356
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:909
#define tcp_in_fastrecovery(tc)
Definition: tcp_types.h:413
#define tcp_csum_offload(tc)
Definition: tcp_types.h:433
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:103
u32 tcp_session_push_header(transport_connection_t *tconn, vlib_buffer_t *b)
Definition: tcp_output.c:982
#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:2444
void scoreboard_clear_reneging(sack_scoreboard_t *sb, u32 start, u32 end)
Definition: tcp_sack.c:276
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:296
#define TRANSPORT_MAX_HDRS_LEN
static void tcp_retransmit_timer_set(tcp_timer_wheel_t *tw, tcp_connection_t *tc)
Definition: tcp_timer.h:54
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:742
void tcp_send_synack(tcp_connection_t *tc)
Definition: tcp_output.c:833
#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:198
static int tcp_make_syn_options(tcp_connection_t *tc, tcp_options_t *opts)
Definition: tcp_output.c:169
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:1702
void tcp_connection_tx_pacer_reset(tcp_connection_t *tc, u32 window, u32 start_bucket)
Definition: tcp.c:1196
static void tcp_cc_loss(tcp_connection_t *tc)
Definition: tcp_cc.h:42
format_function_t format_tcp_connection_id
Definition: tcp.h:354
unsigned short u16
Definition: types.h:57
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:91
static void tcp_output_handle_link_local(tcp_connection_t *tc0, vlib_buffer_t *b0, u16 *next0, u32 *error0)
Definition: tcp_output.c:2044
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:118
#define always_inline
Definition: ipsec.h:28
void tcp_program_dupack(tcp_connection_t *tc)
Definition: tcp_output.c:1040
#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:1651
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:338
#define TCP_OPTION_LEN_TIMESTAMP
Definition: tcp_packet.h:167
vlib_main_t * vm
Definition: in2out_ed.c:1599
#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:539
static void vlib_node_increment_counter(vlib_main_t *vm, u32 node_index, u32 counter_index, u64 increment)
Definition: node_funcs.h:1150
#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:2140
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
u32 flags
Definition: vhost_user.h:248
u16 n_vectors
Definition: node.h:399
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:1080
sll srl srl sll sra u16x4 i
Definition: vector_sse42.h:317
void tcp_program_retransmit(tcp_connection_t *tc)
Definition: tcp_output.c:1052
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:657
format_function_t format_tcp_state
Definition: tcp.h:349
static void tcp_update_rto(tcp_connection_t *tc)
Definition: tcp_inlines.h:375
#define clib_warning(format, args...)
Definition: error.h:59
#define tcp_in_recovery(tc)
Definition: tcp_types.h:414
format_function_t format_tcp_header
Definition: format.h:100
void tcp_timer_delack_handler(tcp_connection_t *tc)
Delayed ack timer handler.
Definition: tcp_output.c:1067
struct _transport_connection transport_connection_t
#define TCP_TO_TIMER_TICK
Factor for converting ticks to timer ticks.
Definition: tcp_types.h:81
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:159
u32 adj_index_t
An index for adjacencies.
Definition: adj_types.h:30
#define ARRAY_LEN(x)
Definition: clib.h:66
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:436
static uword round_pow2(uword x, uword pow2)
Definition: clib.h:256
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
vlib_main_t vlib_node_runtime_t * node
Definition: in2out_ed.c:1599
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:1856
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:440
static void tcp_make_ack(tcp_connection_t *tc, vlib_buffer_t *b)
Convert buffer to ACK.
Definition: tcp_output.c:471
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:233
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:517
#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:1277
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:2100
#define tcp_recovery_on(tc)
Definition: tcp_types.h:411
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
u8 data[128]
Definition: ipsec_types.api:89
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:303
#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:90
void tcp_send_fin(tcp_connection_t *tc)
Send FIN.
Definition: tcp_output.c:860
#define clib_max(x, y)
Definition: clib.h:312
void tcp_send_ack(tcp_connection_t *tc)
Definition: tcp_output.c:1011
static void * vlib_add_trace(vlib_main_t *vm, vlib_node_runtime_t *r, vlib_buffer_t *b, u32 n_data_bytes)
Definition: trace_funcs.h:55
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:1592
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:42
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:419
#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:1688
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
#define foreach_tcp6_reset_next
Definition: tcp_output.c:2351
#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:284
void session_transport_closed_notify(transport_connection_t *tc)
Notification from transport that it is closed.
Definition: session.c:1033
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:244
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:438
static int tcp_make_reset_in_place(vlib_main_t *vm, vlib_buffer_t *b, u8 is_ip4)
Definition: tcp_output.c:574
#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:175
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:366
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:235
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:38
vl_api_dhcp_client_state_t state
Definition: dhcp.api:201
static u32 vlib_num_workers()
Definition: threads.h:376
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:973
static u8 tcp_retransmit_should_retry_head(tcp_connection_t *tc, sack_scoreboard_t *sb)
Definition: tcp_output.c:1675
void tcp_send_syn(tcp_connection_t *tc)
Send SYN.
Definition: tcp_output.c:798
vlib_node_registration_t tcp6_output_node
(constructor) VLIB_REGISTER_NODE (tcp6_output_node)
Definition: tcp_output.c:2321
vlib_main_t vlib_node_runtime_t vlib_frame_t * frame
Definition: in2out_ed.c:1600
u16 flags
Copy of main node flags.
Definition: node.h:511
static u8 tcp_timer_is_active(tcp_connection_t *tc, tcp_timers_e timer)
Definition: tcp_timer.h:117
static u16 tcp_compute_checksum(tcp_connection_t *tc, vlib_buffer_t *b)
Definition: tcp_output.c:412
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:1030
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
u16 dst_port
Definition: udp.api:42
static void * tcp_reuse_buffer(vlib_main_t *vm, vlib_buffer_t *b)
Definition: tcp_output.c:334
u8 ip_version_and_header_length
Definition: ip4_packet.h:138
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:2428
#define VLIB_NODE_FLAG_TRACE
Definition: node.h:304
vlib_node_registration_t tcp4_output_node
(constructor) VLIB_REGISTER_NODE (tcp4_output_node)
Definition: tcp_output.c:2301
#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:561
static u32 vlib_buffer_alloc(vlib_main_t *vm, u32 *buffers, u32 n_buffers)
Allocate buffers into supplied array.
Definition: buffer_funcs.h:677
static tcp_main_t * vnet_get_tcp_main()
Definition: tcp.h:278
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:417
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
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:437
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:92
#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:268
static uword pool_elts(void *v)
Number of active elements in a pool.
Definition: pool.h:128