FD.io VPP  v19.08.1-401-g8e4ed521a
Vector Packet Processing
tcp.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 /**
17  * @file
18  * @brief TCP host stack utilities
19  */
20 
21 #include <vnet/tcp/tcp.h>
22 #include <vnet/session/session.h>
23 #include <vnet/fib/fib.h>
24 #include <vnet/dpo/load_balance.h>
25 #include <vnet/dpo/receive_dpo.h>
26 #include <vnet/ip/ip6_neighbor.h>
27 #include <math.h>
28 #include <vnet/ethernet/arp.h>
29 
31 
32 typedef struct
33 {
36  ip46_address_t ip;
40 
41 static void
43 {
44  u32 ai;
45  if (args->is_add)
46  {
47  adj_nbr_add_or_lock (args->nh_proto, args->link_type, &args->ip,
48  args->sw_if_index);
49  }
50  else
51  {
53  args->sw_if_index);
54  if (ai != ADJ_INDEX_INVALID)
55  adj_unlock (ai);
56  }
57 }
58 
59 static void
61 {
62  tcp_add_del_adj_args_t args = {
64  .link_type = VNET_LINK_IP6,
65  .ip = tc->c_rmt_ip,
66  .sw_if_index = tc->sw_if_index,
67  .is_add = is_add
68  };
70  sizeof (args));
71 }
72 
73 static void
75 {
76  tc->cc_algo->init (tc);
77 }
78 
79 static void
81 {
82  if (tc->cc_algo->cleanup)
83  tc->cc_algo->cleanup (tc);
84 }
85 
86 void
88  const tcp_cc_algorithm_t * vft)
89 {
91  vec_validate (tm->cc_algos, type);
92 
93  tm->cc_algos[type] = *vft;
94  hash_set_mem (tm->cc_algo_by_name, vft->name, type);
95 }
96 
99 {
100  tcp_main_t *tm = vnet_get_tcp_main ();
101  return &tm->cc_algos[type];
102 }
103 
106 {
107  tcp_main_t *tm = vnet_get_tcp_main ();
108  tcp_cc_algo_register (++tm->cc_last_type, vft);
109  return tm->cc_last_type;
110 }
111 
112 static u32
114 {
115  tcp_main_t *tm = &tcp_main;
117  void *iface_ip;
118 
119  pool_get (tm->listener_pool, listener);
120  clib_memset (listener, 0, sizeof (*listener));
121 
122  listener->c_c_index = listener - tm->listener_pool;
123  listener->c_lcl_port = lcl->port;
124 
125  /* If we are provided a sw_if_index, bind using one of its ips */
126  if (ip_is_zero (&lcl->ip, 1) && lcl->sw_if_index != ENDPOINT_INVALID_INDEX)
127  {
128  if ((iface_ip = ip_interface_get_first_ip (lcl->sw_if_index,
129  lcl->is_ip4)))
130  ip_set (&lcl->ip, iface_ip, lcl->is_ip4);
131  }
132  ip_copy (&listener->c_lcl_ip, &lcl->ip, lcl->is_ip4);
133  listener->c_is_ip4 = lcl->is_ip4;
134  listener->c_proto = TRANSPORT_PROTO_TCP;
135  listener->c_s_index = session_index;
136  listener->c_fib_index = lcl->fib_index;
137  listener->state = TCP_STATE_LISTEN;
138  listener->cc_algo = tcp_cc_algo_get (tcp_cfg.cc_algo);
139 
140  tcp_connection_timers_init (listener);
141 
142  TCP_EVT (TCP_EVT_BIND, listener);
143 
144  return listener->c_c_index;
145 }
146 
147 static u32
149 {
150  return tcp_connection_bind (session_index, tep);
151 }
152 
153 static void
154 tcp_connection_unbind (u32 listener_index)
155 {
156  tcp_main_t *tm = vnet_get_tcp_main ();
157  tcp_connection_t *tc;
158 
159  tc = pool_elt_at_index (tm->listener_pool, listener_index);
160 
161  TCP_EVT (TCP_EVT_UNBIND, tc);
162 
163  /* Poison the entry */
164  if (CLIB_DEBUG > 0)
165  clib_memset (tc, 0xFA, sizeof (*tc));
166 
167  pool_put_index (tm->listener_pool, listener_index);
168 }
169 
170 static u32
171 tcp_session_unbind (u32 listener_index)
172 {
173  tcp_connection_unbind (listener_index);
174  return 0;
175 }
176 
177 static transport_connection_t *
179 {
180  tcp_main_t *tm = vnet_get_tcp_main ();
181  tcp_connection_t *tc;
182  tc = pool_elt_at_index (tm->listener_pool, listener_index);
183  return &tc->connection;
184 }
185 
186 /**
187  * Cleanup half-open connection
188  *
189  */
190 static void
192 {
193  tcp_main_t *tm = vnet_get_tcp_main ();
194  clib_spinlock_lock_if_init (&tm->half_open_lock);
195  if (CLIB_DEBUG)
196  clib_memset (tc, 0xFA, sizeof (*tc));
197  pool_put (tm->half_open_connections, tc);
198  clib_spinlock_unlock_if_init (&tm->half_open_lock);
199 }
200 
201 /**
202  * Try to cleanup half-open connection
203  *
204  * If called from a thread that doesn't own tc, the call won't have any
205  * effect.
206  *
207  * @param tc - connection to be cleaned up
208  * @return non-zero if cleanup failed.
209  */
210 int
212 {
213  /* Make sure this is the owning thread */
214  if (tc->c_thread_index != vlib_get_thread_index ())
215  return 1;
216  tcp_timer_reset (tc, TCP_TIMER_RETRANSMIT_SYN);
218  return 0;
219 }
220 
221 static tcp_connection_t *
223 {
224  tcp_main_t *tm = vnet_get_tcp_main ();
225  tcp_connection_t *tc = 0;
226  ASSERT (vlib_get_thread_index () == 0);
227  pool_get (tm->half_open_connections, tc);
228  clib_memset (tc, 0, sizeof (*tc));
229  tc->c_c_index = tc - tm->half_open_connections;
230  return tc;
231 }
232 
233 /**
234  * Cleans up connection state.
235  *
236  * No notifications.
237  */
238 void
240 {
241  tcp_main_t *tm = &tcp_main;
242 
243  TCP_EVT (TCP_EVT_DELETE, tc);
244 
245  /* Cleanup local endpoint if this was an active connect */
246  transport_endpoint_cleanup (TRANSPORT_PROTO_TCP, &tc->c_lcl_ip,
247  tc->c_lcl_port);
248 
249  /* Check if connection is not yet fully established */
250  if (tc->state == TCP_STATE_SYN_SENT)
251  {
252  /* Try to remove the half-open connection. If this is not the owning
253  * thread, tc won't be removed. Retransmit or establish timers will
254  * eventually expire and call again cleanup on the right thread. */
256  tc->flags |= TCP_CONN_HALF_OPEN_DONE;
257  }
258  else
259  {
260  int thread_index = tc->c_thread_index;
261 
262  /* Make sure all timers are cleared */
264 
265  if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6))
266  tcp_add_del_adjacency (tc, 0);
267 
268  tcp_cc_cleanup (tc);
269  vec_free (tc->snd_sacks);
270  vec_free (tc->snd_sacks_fl);
271  vec_free (tc->rcv_opts.sacks);
272  pool_free (tc->sack_sb.holes);
273 
274  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
275  tcp_bt_cleanup (tc);
276 
277  /* Poison the entry */
278  if (CLIB_DEBUG > 0)
279  clib_memset (tc, 0xFA, sizeof (*tc));
280  pool_put (tm->connections[thread_index], tc);
281  }
282 }
283 
284 /**
285  * Connection removal.
286  *
287  * This should be called only once connection enters CLOSED state. Note
288  * that it notifies the session of the removal event, so if the goal is to
289  * just remove the connection, call tcp_connection_cleanup instead.
290  */
291 void
293 {
294  session_transport_delete_notify (&tc->connection);
296 }
297 
299 tcp_connection_alloc (u8 thread_index)
300 {
301  tcp_main_t *tm = vnet_get_tcp_main ();
302  tcp_connection_t *tc;
303 
304  pool_get (tm->connections[thread_index], tc);
305  clib_memset (tc, 0, sizeof (*tc));
306  tc->c_c_index = tc - tm->connections[thread_index];
307  tc->c_thread_index = thread_index;
308  return tc;
309 }
310 
313 {
314  tcp_main_t *tm = vnet_get_tcp_main ();
315  tcp_connection_t *tc;
316 
317  pool_get (tm->connections[thread_index], tc);
318  clib_memcpy_fast (tc, base, sizeof (*tc));
319  tc->c_c_index = tc - tm->connections[thread_index];
320  tc->c_thread_index = thread_index;
321  return tc;
322 }
323 
324 void
326 {
327  tcp_main_t *tm = &tcp_main;
328  if (CLIB_DEBUG)
329  {
330  u8 thread_index = tc->c_thread_index;
331  clib_memset (tc, 0xFA, sizeof (*tc));
332  pool_put (tm->connections[thread_index], tc);
333  return;
334  }
335  pool_put (tm->connections[tc->c_thread_index], tc);
336 }
337 
338 /** Notify session that connection has been reset.
339  *
340  * Switch state to closed and wait for session to call cleanup.
341  */
342 void
344 {
345  TCP_EVT (TCP_EVT_RST_RCVD, tc);
346  switch (tc->state)
347  {
348  case TCP_STATE_SYN_RCVD:
349  /* Cleanup everything. App wasn't notified yet */
350  session_transport_delete_notify (&tc->connection);
352  break;
353  case TCP_STATE_SYN_SENT:
354  session_stream_connect_notify (&tc->connection, 1 /* fail */ );
356  break;
357  case TCP_STATE_ESTABLISHED:
359  /* Set the cleanup timer, in case the session layer/app don't
360  * cleanly close the connection */
361  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
362  session_transport_reset_notify (&tc->connection);
364  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
365  session_transport_closed_notify (&tc->connection);
366  break;
367  case TCP_STATE_CLOSE_WAIT:
368  case TCP_STATE_FIN_WAIT_1:
369  case TCP_STATE_FIN_WAIT_2:
370  case TCP_STATE_CLOSING:
371  case TCP_STATE_LAST_ACK:
373  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
375  /* Make sure we mark the session as closed. In some states we may
376  * be still trying to send data */
377  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
378  session_transport_closed_notify (&tc->connection);
379  break;
380  case TCP_STATE_CLOSED:
381  case TCP_STATE_TIME_WAIT:
382  break;
383  default:
384  TCP_DBG ("reset state: %u", tc->state);
385  }
386 }
387 
388 /**
389  * Begin connection closing procedure.
390  *
391  * If at the end the connection is not in CLOSED state, it is not removed.
392  * Instead, we rely on on TCP to advance through state machine to either
393  * 1) LAST_ACK (passive close) whereby when the last ACK is received
394  * tcp_connection_del is called. This notifies session of the delete and
395  * calls cleanup.
396  * 2) TIME_WAIT (active close) whereby after 2MSL the 2MSL timer triggers
397  * and cleanup is called.
398  *
399  * N.B. Half-close connections are not supported
400  */
401 void
403 {
404  TCP_EVT (TCP_EVT_CLOSE, tc);
405 
406  /* Send/Program FIN if needed and switch state */
407  switch (tc->state)
408  {
409  case TCP_STATE_SYN_SENT:
410  /* Try to cleanup. If not on the right thread, mark as half-open done.
411  * Connection will be cleaned up when establish timer pops */
413  break;
414  case TCP_STATE_SYN_RCVD:
416  tcp_send_fin (tc);
417  tcp_connection_set_state (tc, TCP_STATE_FIN_WAIT_1);
418  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.finwait1_time);
419  break;
420  case TCP_STATE_ESTABLISHED:
421  /* If closing with unread data, reset the connection */
422  if (transport_max_rx_dequeue (&tc->connection))
423  {
424  tcp_send_reset (tc);
426  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
427  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
428  session_transport_closed_notify (&tc->connection);
429  break;
430  }
431  if (!transport_max_tx_dequeue (&tc->connection))
432  tcp_send_fin (tc);
433  else
434  tc->flags |= TCP_CONN_FINPNDG;
435  tcp_connection_set_state (tc, TCP_STATE_FIN_WAIT_1);
436  /* Set a timer in case the peer stops responding. Otherwise the
437  * connection will be stuck here forever. */
438  ASSERT (tc->timers[TCP_TIMER_WAITCLOSE] == TCP_TIMER_HANDLE_INVALID);
439  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.finwait1_time);
440  break;
441  case TCP_STATE_CLOSE_WAIT:
442  if (!transport_max_tx_dequeue (&tc->connection))
443  {
444  tcp_send_fin (tc);
446  tcp_connection_set_state (tc, TCP_STATE_LAST_ACK);
447  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.lastack_time);
448  }
449  else
450  tc->flags |= TCP_CONN_FINPNDG;
451  break;
452  case TCP_STATE_FIN_WAIT_1:
453  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.finwait1_time);
454  break;
455  case TCP_STATE_CLOSED:
457  /* Delete connection but instead of doing it now wait until next
458  * dispatch cycle to give the session layer a chance to clear
459  * unhandled events */
460  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
461  break;
462  default:
463  TCP_DBG ("state: %u", tc->state);
464  }
465 }
466 
467 static void
468 tcp_session_close (u32 conn_index, u32 thread_index)
469 {
470  tcp_connection_t *tc;
471  tc = tcp_connection_get (conn_index, thread_index);
473 }
474 
475 static void
476 tcp_session_cleanup (u32 conn_index, u32 thread_index)
477 {
478  tcp_connection_t *tc;
479  tc = tcp_connection_get (conn_index, thread_index);
480  if (!tc)
481  return;
482  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
484 }
485 
486 static void
487 tcp_session_reset (u32 conn_index, u32 thread_index)
488 {
489  tcp_connection_t *tc;
490  tc = tcp_connection_get (conn_index, thread_index);
491  session_transport_closed_notify (&tc->connection);
492  tcp_send_reset (tc);
495  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
496  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
497 }
498 
499 /**
500  * Initialize all connection timers as invalid
501  */
502 void
504 {
505  int i;
506 
507  /* Set all to invalid */
508  for (i = 0; i < TCP_N_TIMERS; i++)
509  {
510  tc->timers[i] = TCP_TIMER_HANDLE_INVALID;
511  }
512 
513  tc->rto = TCP_RTO_INIT;
514 }
515 
516 /**
517  * Stop all connection timers
518  */
519 void
521 {
522  int i;
523  for (i = 0; i < TCP_N_TIMERS; i++)
524  {
525  tcp_timer_reset (tc, i);
526  }
527 }
528 
529 #if 0
530 typedef struct ip4_tcp_hdr
531 {
533  tcp_header_t tcp;
534 } ip4_tcp_hdr_t;
535 
536 typedef struct ip6_tcp_hdr
537 {
539  tcp_header_t tcp;
540 } ip6_tcp_hdr_t;
541 
542 static void
543 tcp_connection_select_lb_bucket (tcp_connection_t * tc, const dpo_id_t * dpo,
544  dpo_id_t * result)
545 {
546  const dpo_id_t *choice;
547  load_balance_t *lb;
548  int hash;
549 
550  lb = load_balance_get (dpo->dpoi_index);
551  if (tc->c_is_ip4)
552  {
553  ip4_tcp_hdr_t hdr;
554  clib_memset (&hdr, 0, sizeof (hdr));
555  hdr.ip.protocol = IP_PROTOCOL_TCP;
556  hdr.ip.address_pair.src.as_u32 = tc->c_lcl_ip.ip4.as_u32;
557  hdr.ip.address_pair.dst.as_u32 = tc->c_rmt_ip.ip4.as_u32;
558  hdr.tcp.src_port = tc->c_lcl_port;
559  hdr.tcp.dst_port = tc->c_rmt_port;
560  hash = ip4_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
561  }
562  else
563  {
564  ip6_tcp_hdr_t hdr;
565  clib_memset (&hdr, 0, sizeof (hdr));
566  hdr.ip.protocol = IP_PROTOCOL_TCP;
567  clib_memcpy_fast (&hdr.ip.src_address, &tc->c_lcl_ip.ip6,
568  sizeof (ip6_address_t));
569  clib_memcpy_fast (&hdr.ip.dst_address, &tc->c_rmt_ip.ip6,
570  sizeof (ip6_address_t));
571  hdr.tcp.src_port = tc->c_lcl_port;
572  hdr.tcp.dst_port = tc->c_rmt_port;
573  hash = ip6_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
574  }
575  choice = load_balance_get_bucket_i (lb, hash & lb->lb_n_buckets_minus_1);
576  dpo_copy (result, choice);
577 }
578 
581 {
583  u32 fib_index;
584 
585  clib_memcpy_fast (&prefix.fp_addr, &tc->c_rmt_ip, sizeof (prefix.fp_addr));
586  prefix.fp_proto = tc->c_is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6;
587  prefix.fp_len = tc->c_is_ip4 ? 32 : 128;
588  fib_index = fib_table_find (prefix.fp_proto, tc->c_fib_index);
589  return fib_table_lookup (fib_index, &prefix);
590 }
591 
592 static int
593 tcp_connection_stack_on_fib_entry (tcp_connection_t * tc)
594 {
595  dpo_id_t choice = DPO_INVALID;
596  u32 output_node_index;
597  fib_entry_t *fe;
598 
599  fe = fib_entry_get (tc->c_rmt_fei);
600  if (fe->fe_lb.dpoi_type != DPO_LOAD_BALANCE)
601  return -1;
602 
603  tcp_connection_select_lb_bucket (tc, &fe->fe_lb, &choice);
604 
605  output_node_index =
606  tc->c_is_ip4 ? tcp4_output_node.index : tcp6_output_node.index;
607  dpo_stack_from_node (output_node_index, &tc->c_rmt_dpo, &choice);
608  return 0;
609 }
610 
611 /** Stack tcp connection on peer's fib entry.
612  *
613  * This ultimately populates the dpo the connection will use to send packets.
614  */
615 static void
616 tcp_connection_fib_attach (tcp_connection_t * tc)
617 {
618  tc->c_rmt_fei = tcp_lookup_rmt_in_fib (tc);
619 
620  ASSERT (tc->c_rmt_fei != FIB_NODE_INDEX_INVALID);
621 
622  tcp_connection_stack_on_fib_entry (tc);
623 }
624 #endif /* 0 */
625 
626 /**
627  * Generate random iss as per rfc6528
628  */
629 static u32
631 {
632  tcp_main_t *tm = &tcp_main;
633  u64 tmp;
634 
635  if (tc->c_is_ip4)
636  tmp = (u64) tc->c_lcl_ip.ip4.as_u32 << 32 | (u64) tc->c_rmt_ip.ip4.as_u32;
637  else
638  tmp = tc->c_lcl_ip.ip6.as_u64[0] ^ tc->c_lcl_ip.ip6.as_u64[1]
639  ^ tc->c_rmt_ip.ip6.as_u64[0] ^ tc->c_rmt_ip.ip6.as_u64[1];
640 
641  tmp ^= tm->iss_seed.first | ((u64) tc->c_lcl_port << 16 | tc->c_rmt_port);
642  tmp ^= tm->iss_seed.second;
643  tmp = clib_xxhash (tmp) + clib_cpu_time_now ();
644  return ((tmp >> 32) ^ (tmp & 0xffffffff));
645 }
646 
647 /**
648  * Initialize max segment size we're able to process.
649  *
650  * The value is constrained by the output interface's MTU and by the size
651  * of the IP and TCP headers (see RFC6691). It is also what we advertise
652  * to our peer.
653  */
654 static void
656 {
657  u8 ip_hdr_len;
658 
659  ip_hdr_len = tc->c_is_ip4 ? sizeof (ip4_header_t) : sizeof (ip6_header_t);
660  tc->mss = tcp_cfg.default_mtu - sizeof (tcp_header_t) - ip_hdr_len;
661 }
662 
663 static void
665 {
666  u16 default_min_mss = 536;
667 
668  tcp_init_rcv_mss (tc);
669 
670  /* TODO consider PMTU discovery */
671  tc->snd_mss = clib_min (tc->rcv_opts.mss, tc->mss);
672 
673  if (tc->snd_mss < 45)
674  {
675  /* Assume that at least the min default mss works */
676  tc->snd_mss = default_min_mss;
677  tc->rcv_opts.mss = default_min_mss;
678  }
679 
680  /* We should have enough space for 40 bytes of options */
681  ASSERT (tc->snd_mss > 45);
682 
683  /* If we use timestamp option, account for it */
684  if (tcp_opts_tstamp (&tc->rcv_opts))
685  tc->snd_mss -= TCP_OPTION_LEN_TIMESTAMP;
686 }
687 
688 /**
689  * Initialize connection send variables.
690  */
691 void
693 {
694  /*
695  * We use the time to randomize iss and for setting up the initial
696  * timestamp. Make sure it's updated otherwise syn and ack in the
697  * handshake may make it look as if time has flown in the opposite
698  * direction for us.
699  */
701 
702  tcp_init_rcv_mss (tc);
703  tc->iss = tcp_generate_random_iss (tc);
704  tc->snd_una = tc->iss;
705  tc->snd_nxt = tc->iss + 1;
706  tc->snd_una_max = tc->snd_nxt;
707  tc->srtt = 100; /* 100 ms */
708 }
709 
710 void
712 {
713  u32 byte_rate;
714  byte_rate = tc->cwnd / (tc->srtt * TCP_TICK);
715  transport_connection_tx_pacer_init (&tc->connection, byte_rate, tc->cwnd);
716  tc->mrtt_us = (u32) ~ 0;
717 }
718 
719 /** Initialize tcp connection variables
720  *
721  * Should be called after having received a msg from the peer, i.e., a SYN or
722  * a SYNACK, such that connection options have already been exchanged. */
723 void
725 {
727  tcp_init_mss (tc);
728  scoreboard_init (&tc->sack_sb);
729  if (tc->state == TCP_STATE_SYN_RCVD)
730  tcp_init_snd_vars (tc);
731 
732  tcp_cc_init (tc);
733 
734  if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6))
735  tcp_add_del_adjacency (tc, 1);
736 
737  /* tcp_connection_fib_attach (tc); */
738 
739  if (transport_connection_is_tx_paced (&tc->connection)
740  || tcp_cfg.enable_tx_pacing)
741  tcp_enable_pacing (tc);
742 
743  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
744  tcp_bt_init (tc);
745 
746  if (!tcp_cfg.allow_tso)
747  tc->cfg_flags |= TCP_CFG_F_NO_TSO;
748 
749  tc->start_ts = tcp_time_now_us (tc->c_thread_index);
750 }
751 
752 static int
753 tcp_alloc_custom_local_endpoint (tcp_main_t * tm, ip46_address_t * lcl_addr,
754  u16 * lcl_port, u8 is_ip4)
755 {
756  int index, port;
757  if (is_ip4)
758  {
759  index = tm->last_v4_addr_rotor++;
760  if (tm->last_v4_addr_rotor >= vec_len (tcp_cfg.ip4_src_addrs))
761  tm->last_v4_addr_rotor = 0;
762  lcl_addr->ip4.as_u32 = tcp_cfg.ip4_src_addrs[index].as_u32;
763  }
764  else
765  {
766  index = tm->last_v6_addr_rotor++;
767  if (tm->last_v6_addr_rotor >= vec_len (tcp_cfg.ip6_src_addrs))
768  tm->last_v6_addr_rotor = 0;
769  clib_memcpy_fast (&lcl_addr->ip6, &tcp_cfg.ip6_src_addrs[index],
770  sizeof (ip6_address_t));
771  }
772  port = transport_alloc_local_port (TRANSPORT_PROTO_TCP, lcl_addr);
773  if (port < 1)
774  {
775  clib_warning ("Failed to allocate src port");
776  return -1;
777  }
778  *lcl_port = port;
779  return 0;
780 }
781 
782 static int
784 {
785  tcp_main_t *tm = vnet_get_tcp_main ();
786  tcp_connection_t *tc;
787  ip46_address_t lcl_addr;
788  u16 lcl_port;
789  int rv;
790 
791  /*
792  * Allocate local endpoint
793  */
794  if ((rmt->is_ip4 && vec_len (tcp_cfg.ip4_src_addrs))
795  || (!rmt->is_ip4 && vec_len (tcp_cfg.ip6_src_addrs)))
796  rv = tcp_alloc_custom_local_endpoint (tm, &lcl_addr, &lcl_port,
797  rmt->is_ip4);
798  else
799  rv = transport_alloc_local_endpoint (TRANSPORT_PROTO_TCP,
800  rmt, &lcl_addr, &lcl_port);
801 
802  if (rv)
803  return -1;
804 
805  /*
806  * Create connection and send SYN
807  */
808  clib_spinlock_lock_if_init (&tm->half_open_lock);
810  ip_copy (&tc->c_rmt_ip, &rmt->ip, rmt->is_ip4);
811  ip_copy (&tc->c_lcl_ip, &lcl_addr, rmt->is_ip4);
812  tc->c_rmt_port = rmt->port;
813  tc->c_lcl_port = clib_host_to_net_u16 (lcl_port);
814  tc->c_is_ip4 = rmt->is_ip4;
815  tc->c_proto = TRANSPORT_PROTO_TCP;
816  tc->c_fib_index = rmt->fib_index;
817  tc->cc_algo = tcp_cc_algo_get (tcp_cfg.cc_algo);
818  /* The other connection vars will be initialized after SYN ACK */
820 
821  TCP_EVT (TCP_EVT_OPEN, tc);
822  tc->state = TCP_STATE_SYN_SENT;
823  tcp_init_snd_vars (tc);
824  tcp_send_syn (tc);
825  clib_spinlock_unlock_if_init (&tm->half_open_lock);
826 
827  return tc->c_c_index;
828 }
829 
830 const char *tcp_fsm_states[] = {
831 #define _(sym, str) str,
833 #undef _
834 };
835 
836 u8 *
837 format_tcp_state (u8 * s, va_list * args)
838 {
839  u32 state = va_arg (*args, u32);
840 
841  if (state < TCP_N_STATES)
842  s = format (s, "%s", tcp_fsm_states[state]);
843  else
844  s = format (s, "UNKNOWN (%d (0x%x))", state, state);
845  return s;
846 }
847 
848 const char *tcp_cfg_flags_str[] = {
849 #define _(sym, str) str,
851 #undef _
852 };
853 
854 static u8 *
855 format_tcp_cfg_flags (u8 * s, va_list * args)
856 {
857  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
858  int i, last = -1;
859 
860  for (i = 0; i < TCP_CFG_N_FLAG_BITS; i++)
861  if (tc->cfg_flags & (1 << i))
862  last = i;
863  for (i = 0; i < last; i++)
864  {
865  if (tc->cfg_flags & (1 << i))
866  s = format (s, "%s, ", tcp_cfg_flags_str[i]);
867  }
868  if (last >= 0)
869  s = format (s, "%s", tcp_cfg_flags_str[last]);
870  return s;
871 }
872 
873 const char *tcp_connection_flags_str[] = {
874 #define _(sym, str) str,
876 #undef _
877 };
878 
879 static u8 *
880 format_tcp_connection_flags (u8 * s, va_list * args)
881 {
882  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
883  int i, last = -1;
884 
885  for (i = 0; i < TCP_CONN_N_FLAG_BITS; i++)
886  if (tc->flags & (1 << i))
887  last = i;
888  for (i = 0; i < last; i++)
889  {
890  if (tc->flags & (1 << i))
891  s = format (s, "%s, ", tcp_connection_flags_str[i]);
892  }
893  if (last >= 0)
894  s = format (s, "%s", tcp_connection_flags_str[last]);
895  return s;
896 }
897 
898 const char *tcp_conn_timers[] = {
899 #define _(sym, str) str,
901 #undef _
902 };
903 
904 static u8 *
905 format_tcp_timers (u8 * s, va_list * args)
906 {
907  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
908  int i, last = -1;
909 
910  for (i = 0; i < TCP_N_TIMERS; i++)
911  if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
912  last = i;
913 
914  for (i = 0; i < last; i++)
915  {
916  if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
917  s = format (s, "%s,", tcp_conn_timers[i]);
918  }
919 
920  if (last >= 0)
921  s = format (s, "%s", tcp_conn_timers[i]);
922 
923  return s;
924 }
925 
926 static u8 *
927 format_tcp_congestion_status (u8 * s, va_list * args)
928 {
929  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
930  if (tcp_in_recovery (tc))
931  s = format (s, "recovery");
932  else if (tcp_in_fastrecovery (tc))
933  s = format (s, "fastrecovery");
934  else
935  s = format (s, "none");
936  return s;
937 }
938 
939 static i32
941 {
942  return (i32) tc->rcv_wnd - (tc->rcv_nxt - tc->rcv_las);
943 }
944 
945 static u8 *
946 format_tcp_congestion (u8 * s, va_list * args)
947 {
948  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
949  u32 indent = format_get_indent (s), prr_space = 0;
950 
951  s = format (s, "%U ", format_tcp_congestion_status, tc);
952  s = format (s, "algo %s cwnd %u ssthresh %u bytes_acked %u\n",
953  tc->cc_algo->name, tc->cwnd, tc->ssthresh, tc->bytes_acked);
954  s = format (s, "%Ucc space %u prev_cwnd %u prev_ssthresh %u\n",
956  tc->prev_cwnd, tc->prev_ssthresh);
957  s = format (s, "%Usnd_cong %u dupack %u limited_tx %u\n",
958  format_white_space, indent, tc->snd_congestion - tc->iss,
959  tc->rcv_dupacks, tc->limited_transmit - tc->iss);
960  s = format (s, "%Urxt_bytes %u rxt_delivered %u rxt_head %u rxt_ts %u\n",
961  format_white_space, indent, tc->snd_rxt_bytes,
962  tc->rxt_delivered, tc->rxt_head - tc->iss,
963  tcp_time_now_w_thread (tc->c_thread_index) - tc->snd_rxt_ts);
964  if (tcp_in_fastrecovery (tc))
965  prr_space = tcp_fastrecovery_prr_snd_space (tc);
966  s = format (s, "%Uprr_start %u prr_delivered %u prr space %u\n",
967  format_white_space, indent, tc->prr_start - tc->iss,
968  tc->prr_delivered, prr_space);
969  return s;
970 }
971 
972 static u8 *
973 format_tcp_stats (u8 * s, va_list * args)
974 {
975  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
976  u32 indent = format_get_indent (s);
977  s = format (s, "in segs %lu dsegs %lu bytes %lu dupacks %u\n",
978  tc->segs_in, tc->data_segs_in, tc->bytes_in, tc->dupacks_in);
979  s = format (s, "%Uout segs %lu dsegs %lu bytes %lu dupacks %u\n",
980  format_white_space, indent, tc->segs_out,
981  tc->data_segs_out, tc->bytes_out, tc->dupacks_out);
982  s = format (s, "%Ufr %u tr %u rxt segs %lu bytes %lu duration %.3f\n",
983  format_white_space, indent, tc->fr_occurences,
984  tc->tr_occurences, tc->segs_retrans, tc->bytes_retrans,
985  tcp_time_now_us (tc->c_thread_index) - tc->start_ts);
986  s = format (s, "%Uerr wnd data below %u above %u ack below %u above %u",
987  format_white_space, indent, tc->errors.below_data_wnd,
988  tc->errors.above_data_wnd, tc->errors.below_ack_wnd,
989  tc->errors.above_ack_wnd);
990  return s;
991 }
992 
993 static u8 *
994 format_tcp_vars (u8 * s, va_list * args)
995 {
996  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
997  s = format (s, " index: %u cfg: %U flags: %U timers: %U\n", tc->c_c_index,
999  format_tcp_timers, tc);
1000  s = format (s, " snd_una %u snd_nxt %u snd_una_max %u",
1001  tc->snd_una - tc->iss, tc->snd_nxt - tc->iss,
1002  tc->snd_una_max - tc->iss);
1003  s = format (s, " rcv_nxt %u rcv_las %u\n",
1004  tc->rcv_nxt - tc->irs, tc->rcv_las - tc->irs);
1005  s = format (s, " snd_wnd %u rcv_wnd %u rcv_wscale %u ",
1006  tc->snd_wnd, tc->rcv_wnd, tc->rcv_wscale);
1007  s = format (s, "snd_wl1 %u snd_wl2 %u\n", tc->snd_wl1 - tc->irs,
1008  tc->snd_wl2 - tc->iss);
1009  s = format (s, " flight size %u out space %u rcv_wnd_av %u",
1011  tcp_rcv_wnd_available (tc));
1012  s = format (s, " tsval_recent %u\n", tc->tsval_recent);
1013  s = format (s, " tsecr %u tsecr_last_ack %u tsval_recent_age %u",
1014  tc->rcv_opts.tsecr, tc->tsecr_last_ack,
1015  tcp_time_now () - tc->tsval_recent_age);
1016  s = format (s, " snd_mss %u\n", tc->snd_mss);
1017  s = format (s, " rto %u rto_boff %u srtt %u us %.3f rttvar %u rtt_ts %.4f",
1018  tc->rto, tc->rto_boff, tc->srtt, tc->mrtt_us * 1000, tc->rttvar,
1019  tc->rtt_ts);
1020  s = format (s, " rtt_seq %u\n", tc->rtt_seq - tc->iss);
1021  s = format (s, " cong: %U", format_tcp_congestion, tc);
1022 
1023  if (tc->state >= TCP_STATE_ESTABLISHED)
1024  {
1025  s = format (s, " sboard: %U\n", format_tcp_scoreboard, &tc->sack_sb,
1026  tc);
1027  s = format (s, " stats: %U\n", format_tcp_stats, tc);
1028  }
1029  if (vec_len (tc->snd_sacks))
1030  s = format (s, " sacks tx: %U\n", format_tcp_sacks, tc);
1031 
1032  return s;
1033 }
1034 
1035 static u8 *
1036 format_tcp_connection_id (u8 * s, va_list * args)
1037 {
1038  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1039  if (!tc)
1040  return s;
1041  if (tc->c_is_ip4)
1042  {
1043  s = format (s, "[%d:%d][%s] %U:%d->%U:%d", tc->c_thread_index,
1044  tc->c_s_index, "T", format_ip4_address, &tc->c_lcl_ip4,
1045  clib_net_to_host_u16 (tc->c_lcl_port), format_ip4_address,
1046  &tc->c_rmt_ip4, clib_net_to_host_u16 (tc->c_rmt_port));
1047  }
1048  else
1049  {
1050  s = format (s, "[%d:%d][%s] %U:%d->%U:%d", tc->c_thread_index,
1051  tc->c_s_index, "T", format_ip6_address, &tc->c_lcl_ip6,
1052  clib_net_to_host_u16 (tc->c_lcl_port), format_ip6_address,
1053  &tc->c_rmt_ip6, clib_net_to_host_u16 (tc->c_rmt_port));
1054  }
1055 
1056  return s;
1057 }
1058 
1059 u8 *
1060 format_tcp_connection (u8 * s, va_list * args)
1061 {
1062  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1063  u32 verbose = va_arg (*args, u32);
1064 
1065  if (!tc)
1066  return s;
1067  s = format (s, "%-50U", format_tcp_connection_id, tc);
1068  if (verbose)
1069  {
1070  s = format (s, "%-15U", format_tcp_state, tc->state);
1071  if (verbose > 1)
1072  s = format (s, "\n%U", format_tcp_vars, tc);
1073  }
1074 
1075  return s;
1076 }
1077 
1078 static u8 *
1079 format_tcp_session (u8 * s, va_list * args)
1080 {
1081  u32 tci = va_arg (*args, u32);
1082  u32 thread_index = va_arg (*args, u32);
1083  u32 verbose = va_arg (*args, u32);
1084  tcp_connection_t *tc;
1085 
1086  tc = tcp_connection_get (tci, thread_index);
1087  if (tc)
1088  s = format (s, "%U", format_tcp_connection, tc, verbose);
1089  else
1090  s = format (s, "empty\n");
1091  return s;
1092 }
1093 
1094 static u8 *
1095 format_tcp_listener_session (u8 * s, va_list * args)
1096 {
1097  u32 tci = va_arg (*args, u32);
1098  u32 __clib_unused thread_index = va_arg (*args, u32);
1099  u32 verbose = va_arg (*args, u32);
1100  tcp_connection_t *tc = tcp_listener_get (tci);
1101  s = format (s, "%-50U", format_tcp_connection_id, tc);
1102  if (verbose)
1103  s = format (s, "%-15U", format_tcp_state, tc->state);
1104  return s;
1105 }
1106 
1107 static u8 *
1108 format_tcp_half_open_session (u8 * s, va_list * args)
1109 {
1110  u32 tci = va_arg (*args, u32);
1111  u32 __clib_unused thread_index = va_arg (*args, u32);
1113  return format (s, "%U", format_tcp_connection_id, tc);
1114 }
1115 
1116 u8 *
1117 format_tcp_sacks (u8 * s, va_list * args)
1118 {
1119  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1120  sack_block_t *sacks = tc->snd_sacks;
1121  sack_block_t *block;
1122  int i, len = 0;
1123 
1124  len = vec_len (sacks);
1125  for (i = 0; i < len - 1; i++)
1126  {
1127  block = &sacks[i];
1128  s = format (s, " start %u end %u\n", block->start - tc->irs,
1129  block->end - tc->irs);
1130  }
1131  if (len)
1132  {
1133  block = &sacks[len - 1];
1134  s = format (s, " start %u end %u", block->start - tc->irs,
1135  block->end - tc->irs);
1136  }
1137  return s;
1138 }
1139 
1140 u8 *
1141 format_tcp_rcv_sacks (u8 * s, va_list * args)
1142 {
1143  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1144  sack_block_t *sacks = tc->rcv_opts.sacks;
1145  sack_block_t *block;
1146  int i, len = 0;
1147 
1148  len = vec_len (sacks);
1149  for (i = 0; i < len - 1; i++)
1150  {
1151  block = &sacks[i];
1152  s = format (s, " start %u end %u\n", block->start - tc->iss,
1153  block->end - tc->iss);
1154  }
1155  if (len)
1156  {
1157  block = &sacks[len - 1];
1158  s = format (s, " start %u end %u", block->start - tc->iss,
1159  block->end - tc->iss);
1160  }
1161  return s;
1162 }
1163 
1164 static u8 *
1165 format_tcp_sack_hole (u8 * s, va_list * args)
1166 {
1167  sack_scoreboard_hole_t *hole = va_arg (*args, sack_scoreboard_hole_t *);
1168  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1169  if (tc)
1170  s = format (s, " [%u, %u]", hole->start - tc->iss, hole->end - tc->iss);
1171  else
1172  s = format (s, " [%u, %u]", hole->start, hole->end);
1173  return s;
1174 }
1175 
1176 u8 *
1177 format_tcp_scoreboard (u8 * s, va_list * args)
1178 {
1179  sack_scoreboard_t *sb = va_arg (*args, sack_scoreboard_t *);
1180  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1181  sack_scoreboard_hole_t *hole;
1182  u32 indent = format_get_indent (s);
1183 
1184  s = format (s, "sacked %u last_sacked %u lost %u last_lost %u"
1185  " rxt_sacked %u\n",
1186  sb->sacked_bytes, sb->last_sacked_bytes, sb->lost_bytes,
1187  sb->last_lost_bytes, sb->rxt_sacked);
1188  s = format (s, "%Ulast_delivered %u high_sacked %u is_reneging %u\n",
1189  format_white_space, indent, sb->last_bytes_delivered,
1190  sb->high_sacked - tc->iss, sb->is_reneging);
1191  s = format (s, "%Ucur_rxt_hole %u high_rxt %u rescue_rxt %u",
1192  format_white_space, indent, sb->cur_rxt_hole,
1193  sb->high_rxt - tc->iss, sb->rescue_rxt - tc->iss);
1194 
1195  hole = scoreboard_first_hole (sb);
1196  if (hole)
1197  s = format (s, "\n%Uhead %u tail %u %u holes:\n%U", format_white_space,
1198  indent, sb->head, sb->tail, pool_elts (sb->holes),
1199  format_white_space, indent);
1200 
1201  while (hole)
1202  {
1203  s = format (s, "%U", format_tcp_sack_hole, hole, tc);
1204  hole = scoreboard_next_hole (sb, hole);
1205  }
1206 
1207  return s;
1208 }
1209 
1210 static transport_connection_t *
1211 tcp_session_get_transport (u32 conn_index, u32 thread_index)
1212 {
1213  tcp_connection_t *tc = tcp_connection_get (conn_index, thread_index);
1214  if (PREDICT_FALSE (!tc))
1215  return 0;
1216  return &tc->connection;
1217 }
1218 
1219 static transport_connection_t *
1221 {
1223  return &tc->connection;
1224 }
1225 
1226 static u16
1228 {
1229  u16 goal_size = tc->snd_mss;
1230 
1231  goal_size = TCP_MAX_GSO_SZ - tc->snd_mss % TCP_MAX_GSO_SZ;
1232  goal_size = clib_min (goal_size, tc->snd_wnd / 2);
1233 
1234  return goal_size > tc->snd_mss ? goal_size : tc->snd_mss;
1235 }
1236 
1237 /**
1238  * Compute maximum segment size for session layer.
1239  *
1240  * Since the result needs to be the actual data length, it first computes
1241  * the tcp options to be used in the next burst and subtracts their
1242  * length from the connection's snd_mss.
1243  */
1244 static u16
1246 {
1247  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1248 
1249  /* Ensure snd_mss does accurately reflect the amount of data we can push
1250  * in a segment. This also makes sure that options are updated according to
1251  * the current state of the connection. */
1253 
1254  if (PREDICT_FALSE (tc->cfg_flags & TCP_CFG_F_TSO))
1255  return tcp_session_cal_goal_size (tc);
1256 
1257  return tc->snd_mss;
1258 }
1259 
1262 {
1263  if (PREDICT_FALSE (tc->snd_wnd < tc->snd_mss))
1264  {
1265  return tc->snd_wnd <= snd_space ? tc->snd_wnd : 0;
1266  }
1267 
1268  /* If not snd_wnd constrained and we can't write at least a segment,
1269  * don't try at all */
1270  if (PREDICT_FALSE (snd_space < tc->snd_mss))
1271  return snd_space < tc->cwnd ? 0 : snd_space;
1272 
1273  /* round down to mss multiple */
1274  return snd_space - (snd_space % tc->snd_mss);
1275 }
1276 
1277 /**
1278  * Compute tx window session is allowed to fill.
1279  *
1280  * Takes into account available send space, snd_mss and the congestion
1281  * state of the connection. If possible, the value returned is a multiple
1282  * of snd_mss.
1283  *
1284  * @param tc tcp connection
1285  * @return number of bytes session is allowed to write
1286  */
1287 static inline u32
1289 {
1290  int snd_space;
1291 
1293  || tc->state == TCP_STATE_CLOSED))
1294  return 0;
1295 
1296  snd_space = tcp_available_output_snd_space (tc);
1297 
1298  /* If we got dupacks or sacked bytes but we're not yet in recovery, try
1299  * to force the peer to send enough dupacks to start retransmitting as
1300  * per Limited Transmit (RFC3042)
1301  */
1302  if (PREDICT_FALSE (tc->rcv_dupacks != 0 || tc->sack_sb.sacked_bytes))
1303  {
1304  if (tc->limited_transmit != tc->snd_nxt
1305  && (seq_lt (tc->limited_transmit, tc->snd_nxt - 2 * tc->snd_mss)
1306  || seq_gt (tc->limited_transmit, tc->snd_nxt)))
1307  tc->limited_transmit = tc->snd_nxt;
1308 
1309  ASSERT (seq_leq (tc->limited_transmit, tc->snd_nxt));
1310 
1311  int snt_limited = tc->snd_nxt - tc->limited_transmit;
1312  snd_space = clib_max ((int) 2 * tc->snd_mss - snt_limited, 0);
1313  }
1314  return tcp_round_snd_space (tc, snd_space);
1315 }
1316 
1317 u32
1319 {
1320  return tcp_snd_space_inline (tc);
1321 }
1322 
1323 static u32
1325 {
1326  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1327  return clib_min (tcp_snd_space_inline (tc),
1328  tc->snd_wnd - (tc->snd_nxt - tc->snd_una));
1329 }
1330 
1331 static u32
1333 {
1334  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1335 
1336  ASSERT (seq_geq (tc->snd_nxt, tc->snd_una));
1337 
1338  /* This still works if fast retransmit is on */
1339  return (tc->snd_nxt - tc->snd_una);
1340 }
1341 
1342 static void
1343 tcp_update_time (f64 now, u8 thread_index)
1344 {
1345  tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index);
1346 
1347  tcp_set_time_now (wrk);
1348  tw_timer_expire_timers_16t_2w_512sl (&wrk->timer_wheel, now);
1350 }
1351 
1352 static void
1354 {
1355  tcp_connection_t *tc = (tcp_connection_t *) tconn;
1356  if (tc->flags & TCP_CONN_PSH_PENDING)
1357  return;
1358  tc->flags |= TCP_CONN_PSH_PENDING;
1359  tc->psh_seq = tc->snd_una + transport_max_tx_dequeue (tconn) - 1;
1360 }
1361 
1362 /* *INDENT-OFF* */
1364  .enable = vnet_tcp_enable_disable,
1365  .start_listen = tcp_session_bind,
1366  .stop_listen = tcp_session_unbind,
1367  .push_header = tcp_session_push_header,
1368  .get_connection = tcp_session_get_transport,
1369  .get_listener = tcp_session_get_listener,
1370  .get_half_open = tcp_half_open_session_get_transport,
1371  .connect = tcp_session_open,
1372  .close = tcp_session_close,
1373  .cleanup = tcp_session_cleanup,
1374  .reset = tcp_session_reset,
1375  .send_mss = tcp_session_send_mss,
1376  .send_space = tcp_session_send_space,
1377  .update_time = tcp_update_time,
1378  .tx_fifo_offset = tcp_session_tx_fifo_offset,
1379  .flush_data = tcp_session_flush_data,
1380  .custom_tx = tcp_session_custom_tx,
1381  .format_connection = format_tcp_session,
1382  .format_listener = format_tcp_listener_session,
1383  .format_half_open = format_tcp_half_open_session,
1384  .transport_options = {
1385  .tx_type = TRANSPORT_TX_PEEK,
1386  .service_type = TRANSPORT_SERVICE_VC,
1387  },
1388 };
1389 /* *INDENT-ON* */
1390 
1391 void
1393 {
1394  if (!transport_connection_is_tx_paced (&tc->connection))
1395  return;
1396 
1397  transport_connection_tx_pacer_update (&tc->connection,
1398  tcp_cc_get_pacing_rate (tc));
1399 }
1400 
1401 void
1403  u32 start_bucket)
1404 {
1405  f64 srtt = clib_min ((f64) tc->srtt * TCP_TICK, tc->mrtt_us);
1406  u64 rate = (u64) window / srtt;
1407  transport_connection_tx_pacer_reset (&tc->connection, rate, start_bucket);
1408 }
1409 
1410 static void
1412 {
1413  u32 thread_index = vlib_get_thread_index ();
1414  tcp_connection_t *tc;
1415 
1416  tc = tcp_connection_get (conn_index, thread_index);
1417  if (!tc)
1418  return;
1419 
1420  tc->timers[TCP_TIMER_WAITCLOSE] = TCP_TIMER_HANDLE_INVALID;
1421 
1422  switch (tc->state)
1423  {
1424  case TCP_STATE_CLOSE_WAIT:
1426  session_transport_closed_notify (&tc->connection);
1427 
1428  if (!(tc->flags & TCP_CONN_FINPNDG))
1429  {
1430  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1431  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
1432  break;
1433  }
1434 
1435  /* Session didn't come back with a close. Send FIN either way
1436  * and switch to LAST_ACK. */
1437  tcp_cong_recovery_off (tc);
1438  /* Make sure we don't try to send unsent data */
1439  tc->snd_nxt = tc->snd_una;
1440  tcp_send_fin (tc);
1441  tcp_connection_set_state (tc, TCP_STATE_LAST_ACK);
1442 
1443  /* Make sure we don't wait in LAST ACK forever */
1444  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.lastack_time);
1445 
1446  /* Don't delete the connection yet */
1447  break;
1448  case TCP_STATE_FIN_WAIT_1:
1450  session_transport_closed_notify (&tc->connection);
1451  if (tc->flags & TCP_CONN_FINPNDG)
1452  {
1453  /* If FIN pending, we haven't sent everything, but we did try.
1454  * Notify session layer that transport is closed. */
1455  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1456  tcp_send_reset (tc);
1457  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
1458  }
1459  else
1460  {
1461  /* We've sent the fin but no progress. Close the connection and
1462  * to make sure everything is flushed, setup a cleanup timer */
1463  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1464  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
1465  }
1466  break;
1467  case TCP_STATE_LAST_ACK:
1468  case TCP_STATE_CLOSING:
1470  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1471  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
1472  session_transport_closed_notify (&tc->connection);
1473  break;
1474  default:
1475  tcp_connection_del (tc);
1476  break;
1477  }
1478 }
1479 
1480 /* *INDENT-OFF* */
1482 {
1488 };
1489 /* *INDENT-ON* */
1490 
1491 static void
1493 {
1494  int i;
1495  u32 connection_index, timer_id;
1496 
1497  for (i = 0; i < vec_len (expired_timers); i++)
1498  {
1499  /* Get session index and timer id */
1500  connection_index = expired_timers[i] & 0x0FFFFFFF;
1501  timer_id = expired_timers[i] >> 28;
1502 
1503  TCP_EVT (TCP_EVT_TIMER_POP, connection_index, timer_id);
1504 
1505  /* Handle expiration */
1506  (*timer_expiration_handlers[timer_id]) (connection_index);
1507  }
1508 }
1509 
1510 static void
1512 {
1514  tw_timer_wheel_16t_2w_512sl_t *tw;
1515  /* *INDENT-OFF* */
1516  foreach_vlib_main (({
1517  tw = &tm->wrk_ctx[ii].timer_wheel;
1518  tw_timer_wheel_init_16t_2w_512sl (tw, tcp_expired_timers_dispatch,
1519  TCP_TIMER_TICK, ~0);
1520  tw->last_run_time = vlib_time_now (vm);
1521  }));
1522  /* *INDENT-ON* */
1523 }
1524 
1525 static void
1527 {
1528  u32 default_seed = random_default_seed ();
1529  u64 time_now = clib_cpu_time_now ();
1530 
1531  tm->iss_seed.first = (u64) random_u32 (&default_seed) << 32;
1532  tm->iss_seed.second = random_u64 (&time_now);
1533 }
1534 
1535 static clib_error_t *
1537 {
1539  u32 num_threads, n_workers, prealloc_conn_per_wrk;
1540  tcp_connection_t *tc __attribute__ ((unused));
1541  tcp_main_t *tm = vnet_get_tcp_main ();
1542  clib_error_t *error = 0;
1543  int thread;
1544 
1545  if ((error = vlib_call_init_function (vm, ip_main_init)))
1546  return error;
1547  if ((error = vlib_call_init_function (vm, ip4_lookup_init)))
1548  return error;
1549  if ((error = vlib_call_init_function (vm, ip6_lookup_init)))
1550  return error;
1551 
1552  /*
1553  * Registrations
1554  */
1555 
1556  ip4_register_protocol (IP_PROTOCOL_TCP, tcp4_input_node.index);
1557  ip6_register_protocol (IP_PROTOCOL_TCP, tcp6_input_node.index);
1558 
1559  /*
1560  * Initialize data structures
1561  */
1562 
1563  num_threads = 1 /* main thread */ + vtm->n_threads;
1564  vec_validate (tm->connections, num_threads - 1);
1565  vec_validate (tm->wrk_ctx, num_threads - 1);
1566  n_workers = num_threads == 1 ? 1 : vtm->n_threads;
1567  prealloc_conn_per_wrk = tcp_cfg.preallocated_connections / n_workers;
1568 
1569  for (thread = 0; thread < num_threads; thread++)
1570  {
1571  vec_validate (tm->wrk_ctx[thread].pending_deq_acked, 255);
1572  vec_validate (tm->wrk_ctx[thread].pending_disconnects, 255);
1573  vec_reset_length (tm->wrk_ctx[thread].pending_deq_acked);
1574  vec_reset_length (tm->wrk_ctx[thread].pending_disconnects);
1575  tm->wrk_ctx[thread].vm = vlib_mains[thread];
1576 
1577  /*
1578  * Preallocate connections. Assume that thread 0 won't
1579  * use preallocated threads when running multi-core
1580  */
1581  if ((thread > 0 || num_threads == 1) && prealloc_conn_per_wrk)
1582  pool_init_fixed (tm->connections[thread], prealloc_conn_per_wrk);
1583  }
1584 
1585  /*
1586  * Use a preallocated half-open connection pool?
1587  */
1588  if (tcp_cfg.preallocated_half_open_connections)
1589  pool_init_fixed (tm->half_open_connections,
1590  tcp_cfg.preallocated_half_open_connections);
1591 
1592  /* Initialize clocks per tick for TCP timestamp. Used to compute
1593  * monotonically increasing timestamps. */
1594  tm->tstamp_ticks_per_clock = vm->clib_time.seconds_per_clock
1596 
1597  if (num_threads > 1)
1598  {
1599  clib_spinlock_init (&tm->half_open_lock);
1600  }
1601 
1604 
1605  tm->bytes_per_buffer = vlib_buffer_get_default_data_size (vm);
1606  tm->cc_last_type = TCP_CC_LAST;
1607  return error;
1608 }
1609 
1610 clib_error_t *
1612 {
1613  if (is_en)
1614  {
1615  if (tcp_main.is_enabled)
1616  return 0;
1617 
1618  return tcp_main_enable (vm);
1619  }
1620  else
1621  {
1622  tcp_main.is_enabled = 0;
1623  }
1624 
1625  return 0;
1626 }
1627 
1628 void
1629 tcp_punt_unknown (vlib_main_t * vm, u8 is_ip4, u8 is_add)
1630 {
1631  tcp_main_t *tm = &tcp_main;
1632  if (is_ip4)
1633  tm->punt_unknown4 = is_add;
1634  else
1635  tm->punt_unknown6 = is_add;
1636 }
1637 
1638 /**
1639  * Initialize default values for tcp parameters
1640  */
1641 static void
1643 {
1644  /* Initial wnd for SYN. Fifos are not allocated at that point so use some
1645  * predefined value. For SYN-ACK we still want the scale to be computed in
1646  * the same way */
1647  tcp_cfg.max_rx_fifo = 32 << 20;
1648  tcp_cfg.min_rx_fifo = 4 << 10;
1649 
1650  tcp_cfg.default_mtu = 1500;
1651  tcp_cfg.initial_cwnd_multiplier = 0;
1652  tcp_cfg.enable_tx_pacing = 1;
1653  tcp_cfg.allow_tso = 0;
1654  tcp_cfg.cc_algo = TCP_CC_NEWRENO;
1655  tcp_cfg.rwnd_min_update_ack = 1;
1656 
1657  /* Time constants defined as timer tick (100ms) multiples */
1658  tcp_cfg.delack_time = 1; /* 0.1s */
1659  tcp_cfg.closewait_time = 20; /* 2s */
1660  tcp_cfg.timewait_time = 100; /* 10s */
1661  tcp_cfg.finwait1_time = 600; /* 60s */
1662  tcp_cfg.lastack_time = 300; /* 30s */
1663  tcp_cfg.finwait2_time = 300; /* 30s */
1664  tcp_cfg.closing_time = 300; /* 30s */
1665  tcp_cfg.cleanup_time = 1; /* 0.1s */
1666 }
1667 
1668 static clib_error_t *
1670 {
1671  tcp_main_t *tm = vnet_get_tcp_main ();
1672  ip_main_t *im = &ip_main;
1673  ip_protocol_info_t *pi;
1674 
1675  /* Session layer, and by implication tcp, are disabled by default */
1676  tm->is_enabled = 0;
1677 
1678  /* Register with IP for header parsing */
1679  pi = ip_get_protocol_info (im, IP_PROTOCOL_TCP);
1680  if (pi == 0)
1681  return clib_error_return (0, "TCP protocol info AWOL");
1684 
1685  /* Register as transport with session layer */
1686  transport_register_protocol (TRANSPORT_PROTO_TCP, &tcp_proto,
1688  transport_register_protocol (TRANSPORT_PROTO_TCP, &tcp_proto,
1690 
1691  tcp_api_reference ();
1693 
1694  tm->cc_algo_by_name = hash_create_string (0, sizeof (uword));
1695 
1696  return 0;
1697 }
1698 
1700 
1701 uword
1703 {
1704  tcp_cc_algorithm_type_e *result = va_arg (*va, tcp_cc_algorithm_type_e *);
1705  tcp_main_t *tm = &tcp_main;
1706  char *cc_algo_name;
1707  u8 found = 0;
1708  uword *p;
1709 
1710  if (unformat (input, "%s", &cc_algo_name)
1711  && ((p = hash_get_mem (tm->cc_algo_by_name, cc_algo_name))))
1712  {
1713  *result = *p;
1714  found = 1;
1715  }
1716 
1717  vec_free (cc_algo_name);
1718  return found;
1719 }
1720 
1721 uword
1723 {
1724  tcp_main_t *tm = vnet_get_tcp_main ();
1725  tcp_cc_algorithm_t *cc_alg;
1726  unformat_input_t sub_input;
1727  int found = 0;
1728 
1729  vec_foreach (cc_alg, tm->cc_algos)
1730  {
1731  if (!unformat (input, cc_alg->name))
1732  continue;
1733 
1734  if (cc_alg->unformat_cfg
1735  && unformat (input, "%U", unformat_vlib_cli_sub_input, &sub_input))
1736  {
1737  if (cc_alg->unformat_cfg (&sub_input))
1738  found = 1;
1739  }
1740  }
1741  return found;
1742 }
1743 
1744 static clib_error_t *
1746 {
1747  u32 cwnd_multiplier, tmp_time;
1749 
1750  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1751  {
1752  if (unformat (input, "preallocated-connections %d",
1753  &tcp_cfg.preallocated_connections))
1754  ;
1755  else if (unformat (input, "preallocated-half-open-connections %d",
1756  &tcp_cfg.preallocated_half_open_connections))
1757  ;
1758  else if (unformat (input, "buffer-fail-fraction %f",
1759  &tcp_cfg.buffer_fail_fraction))
1760  ;
1761  else if (unformat (input, "max-rx-fifo %U", unformat_memory_size,
1762  &memory_size))
1763  {
1764  if (memory_size >= 0x100000000)
1765  {
1766  return clib_error_return
1767  (0, "max-rx-fifo %llu (0x%llx) too large", memory_size,
1768  memory_size);
1769  }
1770  tcp_cfg.max_rx_fifo = memory_size;
1771  }
1772  else if (unformat (input, "min-rx-fifo %U", unformat_memory_size,
1773  &memory_size))
1774  {
1775  if (memory_size >= 0x100000000)
1776  {
1777  return clib_error_return
1778  (0, "min-rx-fifo %llu (0x%llx) too large", memory_size,
1779  memory_size);
1780  }
1781  tcp_cfg.min_rx_fifo = memory_size;
1782  }
1783  else if (unformat (input, "mtu %u", &tcp_cfg.default_mtu))
1784  ;
1785  else if (unformat (input, "rwnd-min-update-ack %d",
1786  &tcp_cfg.rwnd_min_update_ack))
1787  ;
1788  else if (unformat (input, "initial-cwnd-multiplier %u",
1789  &cwnd_multiplier))
1790  tcp_cfg.initial_cwnd_multiplier = cwnd_multiplier;
1791  else if (unformat (input, "no-tx-pacing"))
1792  tcp_cfg.enable_tx_pacing = 0;
1793  else if (unformat (input, "tso"))
1794  tcp_cfg.allow_tso = 1;
1795  else if (unformat (input, "cc-algo %U", unformat_tcp_cc_algo,
1796  &tcp_cfg.cc_algo))
1797  ;
1798  else if (unformat (input, "%U", unformat_tcp_cc_algo_cfg))
1799  ;
1800  else if (unformat (input, "closewait-time %u", &tmp_time))
1801  tcp_cfg.closewait_time = tmp_time / TCP_TIMER_TICK;
1802  else if (unformat (input, "timewait-time %u", &tmp_time))
1803  tcp_cfg.timewait_time = tmp_time / TCP_TIMER_TICK;
1804  else if (unformat (input, "finwait1-time %u", &tmp_time))
1805  tcp_cfg.finwait1_time = tmp_time / TCP_TIMER_TICK;
1806  else if (unformat (input, "finwait2-time %u", &tmp_time))
1807  tcp_cfg.finwait2_time = tmp_time / TCP_TIMER_TICK;
1808  else if (unformat (input, "lastack-time %u", &tmp_time))
1809  tcp_cfg.lastack_time = tmp_time / TCP_TIMER_TICK;
1810  else if (unformat (input, "closing-time %u", &tmp_time))
1811  tcp_cfg.closing_time = tmp_time / TCP_TIMER_TICK;
1812  else if (unformat (input, "cleanup-time %u", &tmp_time))
1813  tcp_cfg.cleanup_time = tmp_time / TCP_TIMER_TICK;
1814  else
1815  return clib_error_return (0, "unknown input `%U'",
1816  format_unformat_error, input);
1817  }
1818  return 0;
1819 }
1820 
1822 
1823 
1824 /**
1825  * \brief Configure an ipv4 source address range
1826  * @param vm vlib_main_t pointer
1827  * @param start first ipv4 address in the source address range
1828  * @param end last ipv4 address in the source address range
1829  * @param table_id VRF / table ID, 0 for the default FIB
1830  * @return 0 if all OK, else an error indication from api_errno.h
1831  */
1832 
1833 int
1835  ip4_address_t * start,
1836  ip4_address_t * end, u32 table_id)
1837 {
1838  vnet_main_t *vnm = vnet_get_main ();
1839  u32 start_host_byte_order, end_host_byte_order;
1841  fib_node_index_t fei;
1842  u32 fib_index = 0;
1843  u32 sw_if_index;
1844  int rv;
1845 
1846  clib_memset (&prefix, 0, sizeof (prefix));
1847 
1848  fib_index = fib_table_find (FIB_PROTOCOL_IP4, table_id);
1849 
1850  if (fib_index == ~0)
1851  return VNET_API_ERROR_NO_SUCH_FIB;
1852 
1853  start_host_byte_order = clib_net_to_host_u32 (start->as_u32);
1854  end_host_byte_order = clib_net_to_host_u32 (end->as_u32);
1855 
1856  /* sanity check for reversed args or some such */
1857  if ((end_host_byte_order - start_host_byte_order) > (10 << 10))
1858  return VNET_API_ERROR_INVALID_ARGUMENT;
1859 
1860  /* Lookup the last address, to identify the interface involved */
1861  prefix.fp_len = 32;
1862  prefix.fp_proto = FIB_PROTOCOL_IP4;
1863  memcpy (&prefix.fp_addr.ip4, end, sizeof (ip4_address_t));
1864 
1865  fei = fib_table_lookup (fib_index, &prefix);
1866 
1867  /* Couldn't find route to destination. Bail out. */
1868  if (fei == FIB_NODE_INDEX_INVALID)
1869  return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1870 
1871  sw_if_index = fib_entry_get_resolving_interface (fei);
1872 
1873  /* Configure proxy arp across the range */
1874  rv = vnet_proxy_arp_add_del (start, end, fib_index, 0 /* is_del */ );
1875 
1876  if (rv)
1877  return rv;
1878 
1879  rv = vnet_proxy_arp_enable_disable (vnm, sw_if_index, 1);
1880 
1881  if (rv)
1882  return rv;
1883 
1884  do
1885  {
1886  dpo_id_t dpo = DPO_INVALID;
1887 
1888  vec_add1 (tcp_cfg.ip4_src_addrs, start[0]);
1889 
1890  /* Add local adjacencies for the range */
1891 
1892  receive_dpo_add_or_lock (DPO_PROTO_IP4, ~0 /* sw_if_index */ ,
1893  NULL, &dpo);
1894  prefix.fp_len = 32;
1895  prefix.fp_proto = FIB_PROTOCOL_IP4;
1896  prefix.fp_addr.ip4.as_u32 = start->as_u32;
1897 
1899  &prefix,
1901  FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
1902  dpo_reset (&dpo);
1903 
1904  start_host_byte_order++;
1905  start->as_u32 = clib_host_to_net_u32 (start_host_byte_order);
1906  }
1907  while (start_host_byte_order <= end_host_byte_order);
1908 
1909  return 0;
1910 }
1911 
1912 /**
1913  * \brief Configure an ipv6 source address range
1914  * @param vm vlib_main_t pointer
1915  * @param start first ipv6 address in the source address range
1916  * @param end last ipv6 address in the source address range
1917  * @param table_id VRF / table ID, 0 for the default FIB
1918  * @return 0 if all OK, else an error indication from api_errno.h
1919  */
1920 
1921 int
1923  ip6_address_t * start,
1924  ip6_address_t * end, u32 table_id)
1925 {
1927  u32 fib_index = 0;
1928  fib_node_index_t fei;
1929  u32 sw_if_index;
1930 
1931  clib_memset (&prefix, 0, sizeof (prefix));
1932 
1933  fib_index = fib_table_find (FIB_PROTOCOL_IP6, table_id);
1934 
1935  if (fib_index == ~0)
1936  return VNET_API_ERROR_NO_SUCH_FIB;
1937 
1938  while (1)
1939  {
1940  int i;
1941  ip6_address_t tmp;
1942  dpo_id_t dpo = DPO_INVALID;
1943 
1944  /* Remember this address */
1945  vec_add1 (tcp_cfg.ip6_src_addrs, start[0]);
1946 
1947  /* Lookup the prefix, to identify the interface involved */
1948  prefix.fp_len = 128;
1949  prefix.fp_proto = FIB_PROTOCOL_IP6;
1950  memcpy (&prefix.fp_addr.ip6, start, sizeof (ip6_address_t));
1951 
1952  fei = fib_table_lookup (fib_index, &prefix);
1953 
1954  /* Couldn't find route to destination. Bail out. */
1955  if (fei == FIB_NODE_INDEX_INVALID)
1956  return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1957 
1958  sw_if_index = fib_entry_get_resolving_interface (fei);
1959 
1960  if (sw_if_index == (u32) ~ 0)
1961  return VNET_API_ERROR_NO_MATCHING_INTERFACE;
1962 
1963  /* Add a proxy neighbor discovery entry for this address */
1964  ip6_neighbor_proxy_add_del (sw_if_index, start, 0 /* is_del */ );
1965 
1966  /* Add a receive adjacency for this address */
1967  receive_dpo_add_or_lock (DPO_PROTO_IP6, ~0 /* sw_if_index */ ,
1968  NULL, &dpo);
1969 
1971  &prefix,
1973  FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
1974  dpo_reset (&dpo);
1975 
1976  /* Done with the entire range? */
1977  if (!memcmp (start, end, sizeof (start[0])))
1978  break;
1979 
1980  /* Increment the address. DGMS. */
1981  tmp = start[0];
1982  for (i = 15; i >= 0; i--)
1983  {
1984  tmp.as_u8[i] += 1;
1985  if (tmp.as_u8[i] != 0)
1986  break;
1987  }
1988  start[0] = tmp;
1989  }
1990  return 0;
1991 }
1992 
1993 static clib_error_t *
1995  unformat_input_t * input, vlib_cli_command_t * cmd_arg)
1996 {
1997  ip4_address_t v4start, v4end;
1998  ip6_address_t v6start, v6end;
1999  u32 table_id = 0;
2000  int v4set = 0;
2001  int v6set = 0;
2002  int rv;
2003 
2004  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
2005  {
2006  if (unformat (input, "%U - %U", unformat_ip4_address, &v4start,
2007  unformat_ip4_address, &v4end))
2008  v4set = 1;
2009  else if (unformat (input, "%U", unformat_ip4_address, &v4start))
2010  {
2011  memcpy (&v4end, &v4start, sizeof (v4start));
2012  v4set = 1;
2013  }
2014  else if (unformat (input, "%U - %U", unformat_ip6_address, &v6start,
2015  unformat_ip6_address, &v6end))
2016  v6set = 1;
2017  else if (unformat (input, "%U", unformat_ip6_address, &v6start))
2018  {
2019  memcpy (&v6end, &v6start, sizeof (v6start));
2020  v6set = 1;
2021  }
2022  else if (unformat (input, "fib-table %d", &table_id))
2023  ;
2024  else
2025  break;
2026  }
2027 
2028  if (!v4set && !v6set)
2029  return clib_error_return (0, "at least one v4 or v6 address required");
2030 
2031  if (v4set)
2032  {
2033  rv = tcp_configure_v4_source_address_range (vm, &v4start, &v4end,
2034  table_id);
2035  switch (rv)
2036  {
2037  case 0:
2038  break;
2039 
2040  case VNET_API_ERROR_NO_SUCH_FIB:
2041  return clib_error_return (0, "Invalid table-id %d", table_id);
2042 
2043  case VNET_API_ERROR_INVALID_ARGUMENT:
2044  return clib_error_return (0, "Invalid address range %U - %U",
2045  format_ip4_address, &v4start,
2046  format_ip4_address, &v4end);
2047  default:
2048  return clib_error_return (0, "error %d", rv);
2049  break;
2050  }
2051  }
2052  if (v6set)
2053  {
2054  rv = tcp_configure_v6_source_address_range (vm, &v6start, &v6end,
2055  table_id);
2056  switch (rv)
2057  {
2058  case 0:
2059  break;
2060 
2061  case VNET_API_ERROR_NO_SUCH_FIB:
2062  return clib_error_return (0, "Invalid table-id %d", table_id);
2063 
2064  default:
2065  return clib_error_return (0, "error %d", rv);
2066  break;
2067  }
2068  }
2069  return 0;
2070 }
2071 
2072 /* *INDENT-OFF* */
2073 VLIB_CLI_COMMAND (tcp_src_address_command, static) =
2074 {
2075  .path = "tcp src-address",
2076  .short_help = "tcp src-address <ip-addr> [- <ip-addr>] add src address range",
2077  .function = tcp_src_address,
2078 };
2079 /* *INDENT-ON* */
2080 
2081 static u8 *
2083 {
2084 #if TCP_SCOREBOARD_TRACE
2085 
2086  scoreboard_trace_elt_t *block;
2087  int i = 0;
2088 
2089  if (!sb->trace)
2090  return s;
2091 
2092  s = format (s, "scoreboard trace:");
2093  vec_foreach (block, sb->trace)
2094  {
2095  s = format (s, "{%u, %u, %u, %u, %u}, ", block->start, block->end,
2096  block->ack, block->snd_una_max, block->group);
2097  if ((++i % 3) == 0)
2098  s = format (s, "\n");
2099  }
2100  return s;
2101 #else
2102  return 0;
2103 #endif
2104 }
2105 
2106 static clib_error_t *
2108  vlib_cli_command_t * cmd_arg)
2109 {
2110  transport_connection_t *tconn = 0;
2111  tcp_connection_t *tc;
2112  u8 *s = 0;
2113  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
2114  {
2115  if (unformat (input, "%U", unformat_transport_connection, &tconn,
2116  TRANSPORT_PROTO_TCP))
2117  ;
2118  else
2119  return clib_error_return (0, "unknown input `%U'",
2120  format_unformat_error, input);
2121  }
2122 
2123  if (!TCP_SCOREBOARD_TRACE)
2124  {
2125  vlib_cli_output (vm, "scoreboard tracing not enabled");
2126  return 0;
2127  }
2128 
2129  tc = tcp_get_connection_from_transport (tconn);
2130  s = tcp_scoreboard_dump_trace (s, &tc->sack_sb);
2131  vlib_cli_output (vm, "%v", s);
2132  return 0;
2133 }
2134 
2135 /* *INDENT-OFF* */
2136 VLIB_CLI_COMMAND (tcp_show_scoreboard_trace_command, static) =
2137 {
2138  .path = "show tcp scoreboard trace",
2139  .short_help = "show tcp scoreboard trace <connection>",
2140  .function = tcp_show_scoreboard_trace_fn,
2141 };
2142 /* *INDENT-ON* */
2143 
2144 u8 *
2146 {
2147  int i, trace_len;
2149  u32 next_ack, left, group, has_new_ack = 0;
2150  tcp_connection_t _dummy_tc, *dummy_tc = &_dummy_tc;
2151  sack_block_t *block;
2152 
2153  if (!TCP_SCOREBOARD_TRACE)
2154  {
2155  s = format (s, "scoreboard tracing not enabled");
2156  return s;
2157  }
2158 
2159  if (!tc)
2160  return s;
2161 
2162  clib_memset (dummy_tc, 0, sizeof (*dummy_tc));
2163  tcp_connection_timers_init (dummy_tc);
2164  scoreboard_init (&dummy_tc->sack_sb);
2165  dummy_tc->rcv_opts.flags |= TCP_OPTS_FLAG_SACK;
2166 
2167 #if TCP_SCOREBOARD_TRACE
2168  trace = tc->sack_sb.trace;
2169  trace_len = vec_len (tc->sack_sb.trace);
2170 #endif
2171 
2172  for (i = 0; i < trace_len; i++)
2173  {
2174  if (trace[i].ack != 0)
2175  {
2176  dummy_tc->snd_una = trace[i].ack - 1448;
2177  dummy_tc->snd_una_max = trace[i].ack;
2178  }
2179  }
2180 
2181  left = 0;
2182  while (left < trace_len)
2183  {
2184  group = trace[left].group;
2185  vec_reset_length (dummy_tc->rcv_opts.sacks);
2186  has_new_ack = 0;
2187  while (trace[left].group == group)
2188  {
2189  if (trace[left].ack != 0)
2190  {
2191  if (verbose)
2192  s = format (s, "Adding ack %u, snd_una_max %u, segs: ",
2193  trace[left].ack, trace[left].snd_una_max);
2194  dummy_tc->snd_una_max = trace[left].snd_una_max;
2195  next_ack = trace[left].ack;
2196  has_new_ack = 1;
2197  }
2198  else
2199  {
2200  if (verbose)
2201  s = format (s, "[%u, %u], ", trace[left].start,
2202  trace[left].end);
2203  vec_add2 (dummy_tc->rcv_opts.sacks, block, 1);
2204  block->start = trace[left].start;
2205  block->end = trace[left].end;
2206  }
2207  left++;
2208  }
2209 
2210  /* Push segments */
2211  tcp_rcv_sacks (dummy_tc, next_ack);
2212  if (has_new_ack)
2213  dummy_tc->snd_una = next_ack;
2214 
2215  if (verbose)
2216  s = format (s, "result: %U", format_tcp_scoreboard,
2217  &dummy_tc->sack_sb);
2218 
2219  }
2220  s = format (s, "result: %U", format_tcp_scoreboard, &dummy_tc->sack_sb);
2221 
2222  return s;
2223 }
2224 
2225 static clib_error_t *
2227  vlib_cli_command_t * cmd_arg)
2228 {
2229  transport_connection_t *tconn = 0;
2230  tcp_connection_t *tc = 0;
2231  u8 *str = 0;
2232  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
2233  {
2234  if (unformat (input, "%U", unformat_transport_connection, &tconn,
2235  TRANSPORT_PROTO_TCP))
2236  ;
2237  else
2238  return clib_error_return (0, "unknown input `%U'",
2239  format_unformat_error, input);
2240  }
2241 
2242  if (!TCP_SCOREBOARD_TRACE)
2243  {
2244  vlib_cli_output (vm, "scoreboard tracing not enabled");
2245  return 0;
2246  }
2247 
2248  tc = tcp_get_connection_from_transport (tconn);
2249  if (!tc)
2250  {
2251  vlib_cli_output (vm, "connection not found");
2252  return 0;
2253  }
2254  str = tcp_scoreboard_replay (str, tc, 1);
2255  vlib_cli_output (vm, "%v", str);
2256  return 0;
2257 }
2258 
2259 /* *INDENT-OFF* */
2260 VLIB_CLI_COMMAND (tcp_replay_scoreboard_command, static) =
2261 {
2262  .path = "tcp replay scoreboard",
2263  .short_help = "tcp replay scoreboard <connection>",
2264  .function = tcp_scoreboard_trace_fn,
2265 };
2266 /* *INDENT-ON* */
2267 
2268 static clib_error_t *
2270  vlib_cli_command_t * cmd_arg)
2271 {
2272  tcp_main_t *tm = vnet_get_tcp_main ();
2274  return clib_error_return (0, "unknown input `%U'", format_unformat_error,
2275  input);
2276  vlib_cli_output (vm, "IPv4 TCP punt: %s",
2277  tm->punt_unknown4 ? "enabled" : "disabled");
2278  vlib_cli_output (vm, "IPv6 TCP punt: %s",
2279  tm->punt_unknown6 ? "enabled" : "disabled");
2280  return 0;
2281 }
2282 /* *INDENT-OFF* */
2283 VLIB_CLI_COMMAND (show_tcp_punt_command, static) =
2284 {
2285  .path = "show tcp punt",
2286  .short_help = "show tcp punt",
2287  .function = show_tcp_punt_fn,
2288 };
2289 /* *INDENT-ON* */
2290 
2291 /*
2292  * fd.io coding-style-patch-verification: ON
2293  *
2294  * Local Variables:
2295  * eval: (c-set-style "gnu")
2296  * End:
2297  */
static void tcp_session_close(u32 conn_index, u32 thread_index)
Definition: tcp.c:468
#define vec_validate(V, I)
Make sure vector is long enough for given index (no header, unspecified alignment) ...
Definition: vec.h:439
void dpo_stack_from_node(u32 child_node_index, dpo_id_t *dpo, const dpo_id_t *parent)
Stack one DPO object on another, and thus establish a child parent relationship.
Definition: dpo.c:531
fib_protocol_t fp_proto
protocol type
Definition: fib_types.h:212
#define foreach_tcp_cfg_flag
Connection configuration flags.
Definition: tcp.h:108
#define ENDPOINT_INVALID_INDEX
static vlib_cli_command_t trace
(constructor) VLIB_CLI_COMMAND (trace)
Definition: vlib_api_cli.c:889
static void tcp_add_del_adjacency(tcp_connection_t *tc, u8 is_add)
Definition: tcp.c:60
void receive_dpo_add_or_lock(dpo_proto_t proto, u32 sw_if_index, const ip46_address_t *nh_addr, dpo_id_t *dpo)
Definition: receive_dpo.c:62
An entry in a FIB table.
Definition: fib_entry.h:468
#define clib_min(x, y)
Definition: clib.h:295
static u8 * format_tcp_listener_session(u8 *s, va_list *args)
Definition: tcp.c:1095
const char * tcp_fsm_states[]
Definition: tcp.c:830
static clib_error_t * tcp_config_fn(vlib_main_t *vm, unformat_input_t *input)
Definition: tcp.c:1745
#define tcp_in_recovery(tc)
Definition: tcp.h:465
static f64 tcp_time_now_us(u32 thread_index)
Definition: tcp.h:1021
void ip_copy(ip46_address_t *dst, ip46_address_t *src, u8 is_ip4)
Definition: ip.c:81
static uword random_default_seed(void)
Default random seed (unix/linux user-mode)
Definition: random.h:91
#define seq_leq(_s1, _s2)
Definition: tcp.h:867
struct _sack_block sack_block_t
u8 * format_tcp_rcv_sacks(u8 *s, va_list *args)
Definition: tcp.c:1141
void ip6_register_protocol(u32 protocol, u32 node_index)
Definition: ip6_forward.c:1508
struct _scoreboard_trace_elt scoreboard_trace_elt_t
u8 * format_tcp_sacks(u8 *s, va_list *args)
Definition: tcp.c:1117
vnet_main_t * vnet_get_main(void)
Definition: misc.c:46
static clib_error_t * ip4_lookup_init(vlib_main_t *vm)
Definition: ip4_forward.c:1080
const char * tcp_connection_flags_str[]
Definition: tcp.c:873
vlib_node_registration_t tcp4_output_node
(constructor) VLIB_REGISTER_NODE (tcp4_output_node)
Definition: tcp_output.c:2493
void scoreboard_init(sack_scoreboard_t *sb)
Definition: tcp_input.c:939
uword unformat_transport_connection(unformat_input_t *input, va_list *args)
Definition: session_cli.c:217
#define tcp_opts_tstamp(_to)
Definition: tcp_packet.h:156
static u32 tcp_session_send_space(transport_connection_t *trans_conn)
Definition: tcp.c:1324
static transport_connection_t * tcp_half_open_session_get_transport(u32 conn_index)
Definition: tcp.c:1220
u8 as_u8[16]
Definition: ip6_packet.h:48
unsigned long u64
Definition: types.h:89
void ip_set(ip46_address_t *dst, void *src, u8 is_ip4)
Definition: ip.c:90
#define clib_memcpy_fast(a, b, c)
Definition: string.h:81
static u32 tcp_session_tx_fifo_offset(transport_connection_t *trans_conn)
Definition: tcp.c:1332
#define NULL
Definition: clib.h:58
static u32 ip4_compute_flow_hash(const ip4_header_t *ip, flow_hash_config_t flow_hash_config)
Definition: ip4.h:310
clib_memset(h->entries, 0, sizeof(h->entries[0]) *entries)
struct _sack_scoreboard sack_scoreboard_t
void session_transport_delete_notify(transport_connection_t *tc)
Notification from transport that connection is being deleted.
Definition: session.c:878
static int tcp_alloc_custom_local_endpoint(tcp_main_t *tm, ip46_address_t *lcl_addr, u16 *lcl_port, u8 is_ip4)
Definition: tcp.c:753
static f64 vlib_time_now(vlib_main_t *vm)
Definition: main.h:279
static tcp_connection_t * tcp_half_open_connection_new(void)
Definition: tcp.c:222
static u8 * format_tcp_timers(u8 *s, va_list *args)
Definition: tcp.c:905
static tcp_connection_t * tcp_half_open_connection_get(u32 conn_index)
Definition: tcp.h:770
sack_scoreboard_hole_t * scoreboard_next_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:679
flow_hash_config_t lb_hash_config
the hash config to use when selecting a bucket.
Definition: load_balance.h:161
static_always_inline void clib_spinlock_unlock_if_init(clib_spinlock_t *p)
Definition: lock.h:110
struct _tcp_main tcp_main_t
void tcp_connection_timers_reset(tcp_connection_t *tc)
Stop all connection timers.
Definition: tcp.c:520
vlib_node_registration_t tcp6_output_node
(constructor) VLIB_REGISTER_NODE (tcp6_output_node)
Definition: tcp_output.c:2513
void dpo_copy(dpo_id_t *dst, const dpo_id_t *src)
atomic copy a data-plane object.
Definition: dpo.c:262
#define vec_add1(V, E)
Add 1 element to end of vector (unspecified alignment).
Definition: vec.h:522
static u64 clib_cpu_time_now(void)
Definition: time.h:81
timer_expiration_handler tcp_timer_retransmit_handler
static u64 clib_xxhash(u64 key)
Definition: xxhash.h:58
static heap_elt_t * last(heap_header_t *h)
Definition: heap.c:53
u32 tcp_session_push_header(transport_connection_t *tconn, vlib_buffer_t *b)
Definition: tcp_output.c:1137
void session_transport_reset_notify(transport_connection_t *tc)
Notify application that connection has been reset.
Definition: session.c:979
#define vec_add2(V, P, N)
Add N elements to end of vector V, return pointer to new elements in P.
Definition: vec.h:560
int i
int ip6_neighbor_proxy_add_del(u32 sw_if_index, ip6_address_t *addr, u8 is_del)
void tcp_update_burst_snd_vars(tcp_connection_t *tc)
Update burst send vars.
Definition: tcp_output.c:390
static u32 format_get_indent(u8 *s)
Definition: format.h:72
#define hash_set_mem(h, key, value)
Definition: hash.h:275
void * ip_interface_get_first_ip(u32 sw_if_index, u8 is_ip4)
Definition: ip.c:134
struct _tcp_connection tcp_connection_t
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:424
static u32 tcp_available_cc_snd_space(const tcp_connection_t *tc)
Estimate of how many bytes we can still push into the network.
Definition: tcp.h:970
clib_time_t clib_time
Definition: main.h:87
static u64 random_u64(u64 *seed)
64-bit random number generator Again, constants courtesy of Donald Knuth.
Definition: random.h:126
static u16 tcp_session_send_mss(transport_connection_t *trans_conn)
Compute maximum segment size for session layer.
Definition: tcp.c:1245
Definition: ip.h:107
static tcp_connection_t * tcp_get_connection_from_transport(transport_connection_t *tconn)
Definition: tcp.h:733
vl_api_mprefix_t prefix
Definition: ip.api:456
static void tcp_initialize_iss_seed(tcp_main_t *tm)
Definition: tcp.c:1526
void ip4_register_protocol(u32 protocol, u32 node_index)
Definition: ip4_forward.c:1863
static void tcp_initialize_timer_wheels(tcp_main_t *tm)
Definition: tcp.c:1511
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:236
struct _tcp_header tcp_header_t
fib_protocol_t nh_proto
Definition: tcp.c:34
int tcp_half_open_connection_cleanup(tcp_connection_t *tc)
Try to cleanup half-open connection.
Definition: tcp.c:211
vlib_main_t ** vlib_mains
Definition: buffer.c:332
static void tcp_half_open_connection_del(tcp_connection_t *tc)
Cleanup half-open connection.
Definition: tcp.c:191
unsigned char u8
Definition: types.h:56
struct _sack_scoreboard_hole sack_scoreboard_hole_t
enum fib_protocol_t_ fib_protocol_t
Protocol Type.
#define vec_reset_length(v)
Reset vector length to zero NULL-pointer tolerant.
double f64
Definition: types.h:142
tcp_cc_algorithm_t * tcp_cc_algo_get(tcp_cc_algorithm_type_e type)
Definition: tcp.c:98
#define TCP_RTO_INIT
Definition: tcp.h:103
format_function_t format_ip4_address
Definition: format.h:75
#define TCP_TICK
TCP tick period (s)
Definition: tcp.h:27
unformat_function_t * unformat_pg_edit
Definition: ip.h:88
#define seq_gt(_s1, _s2)
Definition: tcp.h:868
static void tcp_connection_set_state(tcp_connection_t *tc, tcp_state_t state)
Definition: tcp.h:739
unformat_function_t unformat_ip4_address
Definition: format.h:70
void tcp_init_snd_vars(tcp_connection_t *tc)
Initialize connection send variables.
Definition: tcp.c:692
#define tcp_cfg
Definition: tcp.h:676
vl_api_interface_index_t sw_if_index
Definition: gre.api:50
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:173
#define always_inline
Definition: clib.h:98
static u8 * format_tcp_vars(u8 *s, va_list *args)
Definition: tcp.c:994
u16 lb_n_buckets_minus_1
number of buckets in the load-balance - 1.
Definition: load_balance.h:121
void tcp_connection_timers_init(tcp_connection_t *tc)
Initialize all connection timers as invalid.
Definition: tcp.c:503
static u32 tcp_available_output_snd_space(const tcp_connection_t *tc)
Definition: tcp.h:955
u8 * format_white_space(u8 *s, va_list *va)
Definition: std-formats.c:129
tcp_main_t tcp_main
Definition: tcp.c:30
static void tcp_connection_unbind(u32 listener_index)
Definition: tcp.c:154
Aggregate type for a prefix.
Definition: fib_types.h:203
tw_timer_wheel_16t_2w_512sl_t timer_wheel
worker timer wheel
Definition: tcp.h:510
#define clib_error_return(e, args...)
Definition: error.h:99
void adj_unlock(adj_index_t adj_index)
Release a reference counting lock on the adjacency.
Definition: adj.c:322
vhost_vring_state_t state
Definition: vhost_user.h:146
static u8 * tcp_scoreboard_dump_trace(u8 *s, sack_scoreboard_t *sb)
Definition: tcp.c:2082
timer_expiration_handler tcp_timer_retransmit_syn_handler
pthread_t thread[MAX_CONNS]
Definition: main.c:142
int transport_alloc_local_endpoint(u8 proto, transport_endpoint_cfg_t *rmt_cfg, ip46_address_t *lcl_addr, u16 *lcl_port)
Definition: transport.c:497
static u32 tcp_time_now(void)
Definition: tcp.h:999
u32 tcp_snd_space(tcp_connection_t *tc)
Definition: tcp.c:1318
unsigned int u32
Definition: types.h:88
u32 fib_table_find(fib_protocol_t proto, u32 table_id)
Get the index of the FIB for a Table-ID.
Definition: fib_table.c:1080
u16 fp_len
The mask length.
Definition: fib_types.h:207
#define vlib_call_init_function(vm, x)
Definition: init.h:270
void tcp_api_reference(void)
Definition: tcp_api.c:109
fib_node_index_t fib_table_lookup(u32 fib_index, const fib_prefix_t *prefix)
Perfom a longest prefix match in the non-forwarding table.
Definition: fib_table.c:66
static u32 tcp_connection_bind(u32 session_index, transport_endpoint_t *lcl)
Definition: tcp.c:113
static i32 tcp_rcv_wnd_available(tcp_connection_t *tc)
Definition: tcp.c:940
#define hash_create_string(elts, value_bytes)
Definition: hash.h:690
static void tcp_cc_init(tcp_connection_t *tc)
Definition: tcp.c:74
struct _transport_proto_vft transport_proto_vft_t
static void clib_spinlock_init(clib_spinlock_t *p)
Definition: lock.h:63
static void tcp_timer_set(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:1105
uword unformat_tcp_cc_algo_cfg(unformat_input_t *input, va_list *va)
Definition: tcp.c:1722
vl_api_fib_path_type_t type
Definition: fib_types.api:123
static u32 tcp_session_unbind(u32 listener_index)
Definition: tcp.c:171
static timer_expiration_handler * timer_expiration_handlers[TCP_N_TIMERS]
Definition: tcp.c:1481
The identity of a DPO is a combination of its type and its instance number/index of objects of that t...
Definition: dpo.h:170
static clib_error_t * tcp_main_enable(vlib_main_t *vm)
Definition: tcp.c:1536
Definition: fib_entry.h:285
#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
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:514
void tcp_bt_init(tcp_connection_t *tc)
Byte tracker initialize.
Definition: tcp_bt.c:650
timer_expiration_handler tcp_timer_persist_handler
static transport_connection_t * tcp_session_get_transport(u32 conn_index, u32 thread_index)
Definition: tcp.c:1211
static ip_protocol_info_t * ip_get_protocol_info(ip_main_t *im, u32 protocol)
Definition: ip.h:134
int vnet_proxy_arp_enable_disable(vnet_main_t *vnm, u32 sw_if_index, u8 enable)
Definition: arp.c:2412
#define tcp_in_fastrecovery(tc)
Definition: tcp.h:464
ip46_address_t fp_addr
The address type is not deriveable from the fp_addr member.
Definition: fib_types.h:226
void tcp_connection_tx_pacer_reset(tcp_connection_t *tc, u32 window, u32 start_bucket)
Definition: tcp.c:1402
int tcp_fastrecovery_prr_snd_space(tcp_connection_t *tc)
Estimate send space using proportional rate reduction (RFC6937)
Definition: tcp_output.c:1821
dpo_type_t dpoi_type
the type
Definition: dpo.h:174
format_function_t * format_header
Definition: ip.h:79
tcp_connection_t * tcp_connection_alloc_w_base(u8 thread_index, tcp_connection_t *base)
Definition: tcp.c:312
static const dpo_id_t * load_balance_get_bucket_i(const load_balance_t *lb, u32 bucket)
Definition: load_balance.h:229
void tcp_rcv_sacks(tcp_connection_t *tc, u32 ack)
Definition: tcp_input.c:1001
static void tcp_init_rcv_mss(tcp_connection_t *tc)
Initialize max segment size we&#39;re able to process.
Definition: tcp.c:655
void tcp_send_syn(tcp_connection_t *tc)
Send SYN.
Definition: tcp_output.c:927
struct _unformat_input_t unformat_input_t
unsigned short u16
Definition: types.h:57
u64 memory_size
Definition: vhost_user.h:141
tcp_connection_t * tcp_connection_alloc(u8 thread_index)
Definition: tcp.c:299
load-balancing over a choice of [un]equal cost paths
Definition: dpo.h:102
static transport_connection_t * tcp_session_get_listener(u32 listener_index)
Definition: tcp.c:178
static u32 transport_max_rx_dequeue(transport_connection_t *tc)
Definition: session.h:483
void tcp_connection_tx_pacer_update(tcp_connection_t *tc)
Definition: tcp.c:1392
static u32 ip6_compute_flow_hash(const ip6_header_t *ip, flow_hash_config_t flow_hash_config)
Definition: ip6.h:482
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:286
static void tcp_expired_timers_dispatch(u32 *expired_timers)
Definition: tcp.c:1492
void transport_connection_tx_pacer_init(transport_connection_t *tc, u64 rate_bytes_per_sec, u32 initial_bucket)
Initialize tx pacer for connection.
Definition: transport.c:646
#define TCP_TIMER_HANDLE_INVALID
Definition: tcp.h:92
static void tcp_session_cleanup(u32 conn_index, u32 thread_index)
Definition: tcp.c:476
The FIB DPO provieds;.
Definition: load_balance.h:106
static u32 tcp_flight_size(const tcp_connection_t *tc)
Our estimate of the number of bytes in flight (pipe size)
Definition: tcp.h:893
#define PREDICT_FALSE(x)
Definition: clib.h:111
#define VLIB_CONFIG_FUNCTION(x, n,...)
Definition: init.h:182
static u8 * format_tcp_stats(u8 *s, va_list *args)
Definition: tcp.c:973
u8 * format_tcp_scoreboard(u8 *s, va_list *args)
Definition: tcp.c:1177
clib_error_t * vnet_tcp_enable_disable(vlib_main_t *vm, u8 is_en)
Definition: tcp.c:1611
u16 port
Definition: punt.api:40
f64 seconds_per_clock
Definition: time.h:58
static u8 transport_connection_is_tx_paced(transport_connection_t *tc)
Check if transport connection is paced.
Definition: transport.h:236
static void tcp_timer_waitclose_handler(u32 conn_index)
Definition: tcp.c:1411
static u8 * format_tcp_cfg_flags(u8 *s, va_list *args)
Definition: tcp.c:855
#define foreach_vlib_main(body)
Definition: threads.h:236
#define TCP_MAX_GSO_SZ
Definition: tcp.h:34
#define TCP_OPTION_LEN_TIMESTAMP
Definition: tcp_packet.h:167
const char * tcp_cfg_flags_str[]
Definition: tcp.c:848
void transport_connection_tx_pacer_update(transport_connection_t *tc, u64 bytes_per_sec)
Update tx pacer pacing rate.
Definition: transport.c:656
#define TCP_DBG(_fmt, _args...)
Definition: tcp_debug.h:146
u8 len
Definition: ip_types.api:90
u32 fib_entry_get_resolving_interface(fib_node_index_t entry_index)
Definition: fib_entry.c:1458
static void tcp_timer_reset(tcp_connection_t *tc, u8 timer_id)
Definition: tcp.h:1116
void tcp_connection_free(tcp_connection_t *tc)
Definition: tcp.c:325
unformat_function_t unformat_ip6_address
Definition: format.h:91
static_always_inline u32 vlib_buffer_get_default_data_size(vlib_main_t *vm)
Definition: buffer_funcs.h:96
#define pool_free(p)
Free a pool.
Definition: pool.h:407
clib_error_t * ip_main_init(vlib_main_t *vm)
Definition: ip_init.c:45
fib_node_index_t fib_table_entry_special_dpo_update(u32 fib_index, const fib_prefix_t *prefix, fib_source_t source, fib_entry_flag_t flags, const dpo_id_t *dpo)
Update a &#39;special&#39; entry to the FIB that links to the DPO passed A special entry is an entry that the...
Definition: fib_table.c:346
#define UNFORMAT_END_OF_INPUT
Definition: format.h:145
int tcp_configure_v4_source_address_range(vlib_main_t *vm, ip4_address_t *start, ip4_address_t *end, u32 table_id)
Configure an ipv4 source address range.
Definition: tcp.c:1834
static_always_inline uword vlib_get_thread_index(void)
Definition: threads.h:213
format_function_t format_ip6_address
Definition: format.h:93
vlib_main_t * vm
Definition: buffer.c:323
ip_main_t ip_main
Definition: ip_init.c:42
static void tcp_add_del_adj_cb(tcp_add_del_adj_args_t *args)
Definition: tcp.c:42
static clib_error_t * tcp_src_address(vlib_main_t *vm, unformat_input_t *input, vlib_cli_command_t *cmd_arg)
Definition: tcp.c:1994
void tcp_connection_close(tcp_connection_t *tc)
Begin connection closing procedure.
Definition: tcp.c:402
void tcp_send_reset(tcp_connection_t *tc)
Build and set reset packet for connection.
Definition: tcp_output.c:861
#define vec_free(V)
Free vector&#39;s memory (no header).
Definition: vec.h:341
virtual circuit service
static void tcp_session_flush_data(transport_connection_t *tconn)
Definition: tcp.c:1353
#define clib_warning(format, args...)
Definition: error.h:59
static u8 * format_tcp_congestion(u8 *s, va_list *args)
Definition: tcp.c:946
format_function_t format_tcp_header
Definition: format.h:101
struct _transport_connection transport_connection_t
static u8 * format_tcp_connection_flags(u8 *s, va_list *args)
Definition: tcp.c:880
u32 fib_node_index_t
A typedef of a node index.
Definition: fib_types.h:30
uword unformat_tcp_cc_algo(unformat_input_t *input, va_list *va)
Definition: tcp.c:1702
vlib_node_registration_t tcp6_input_node
(constructor) VLIB_REGISTER_NODE (tcp6_input_node)
Definition: tcp_input.c:3752
fib_node_index_t tcp_lookup_rmt_in_fib(tcp_connection_t *tc)
fib_entry_t * fib_entry_get(fib_node_index_t index)
Definition: fib_entry.c:52
int vnet_proxy_arp_add_del(ip4_address_t *lo_addr, ip4_address_t *hi_addr, u32 fib_index, int is_del)
Definition: arp.c:2444
#define pool_init_fixed(pool, max_elts)
initialize a fixed-size, preallocated pool
Definition: pool.h:86
void transport_endpoint_cleanup(u8 proto, ip46_address_t *lcl_ip, u16 port)
Definition: transport.c:366
void transport_register_protocol(transport_proto_t transport_proto, const transport_proto_vft_t *vft, fib_protocol_t fib_proto, u32 output_node)
Register transport virtual function table.
Definition: transport.c:210
#define VLIB_CLI_COMMAND(x,...)
Definition: cli.h:161
static u32 transport_max_tx_dequeue(transport_connection_t *tc)
Definition: session.h:476
static void tcp_timer_update(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:1129
signed int i32
Definition: types.h:77
#define pool_put_index(p, i)
Free pool element with given index.
Definition: pool.h:311
#define ASSERT(truth)
static int tcp_session_open(transport_endpoint_cfg_t *rmt)
Definition: tcp.c:783
tcp_cc_algorithm_type_e tcp_cc_algo_new_type(const tcp_cc_algorithm_t *vft)
Register new cc algo type.
Definition: tcp.c:105
enum _tcp_cc_algorithm_type tcp_cc_algorithm_type_e
static clib_error_t * tcp_scoreboard_trace_fn(vlib_main_t *vm, unformat_input_t *input, vlib_cli_command_t *cmd_arg)
Definition: tcp.c:2226
unformat_function_t unformat_pg_tcp_header
Definition: format.h:103
static void tcp_configuration_init(void)
Initialize default values for tcp parameters.
Definition: tcp.c:1642
enum vnet_link_t_ vnet_link_t
Link Type: A description of the protocol of packets on the link.
void tcp_flush_frames_to_output(tcp_worker_ctx_t *wrk)
Flush v4 and v6 tcp and ip-lookup tx frames for thread index.
Definition: tcp_output.c:1005
static load_balance_t * load_balance_get(index_t lbi)
Definition: load_balance.h:220
#define seq_geq(_s1, _s2)
Definition: tcp.h:869
static void tcp_init_mss(tcp_connection_t *tc)
Definition: tcp.c:664
static u8 * format_tcp_congestion_status(u8 *s, va_list *args)
Definition: tcp.c:927
static u32 tcp_generate_random_iss(tcp_connection_t *tc)
Generate random iss as per rfc6528.
Definition: tcp.c:630
reliable transport protos
static uword ip6_address_is_link_local_unicast(const ip6_address_t *a)
Definition: ip6_packet.h:326
From the control plane API.
Definition: fib_entry.h:79
void tcp_cc_algo_register(tcp_cc_algorithm_type_e type, const tcp_cc_algorithm_t *vft)
Register exiting cc algo type.
Definition: tcp.c:87
u8 * tcp_scoreboard_replay(u8 *s, tcp_connection_t *tc, u8 verbose)
Definition: tcp.c:2145
int tcp_configure_v6_source_address_range(vlib_main_t *vm, ip6_address_t *start, ip6_address_t *end, u32 table_id)
Configure an ipv6 source address range.
Definition: tcp.c:1922
struct _tcp_cc_algorithm tcp_cc_algorithm_t
Definition: tcp.h:305
#define clib_max(x, y)
Definition: clib.h:288
static vlib_main_t * vlib_get_main(void)
Definition: global_funcs.h:23
static u32 tcp_time_now_w_thread(u32 thread_index)
Definition: tcp.h:1005
static clib_error_t * tcp_init(vlib_main_t *vm)
Definition: tcp.c:1669
dpo_id_t fe_lb
The load-balance used for forwarding.
Definition: fib_entry.h:494
u8 ip_is_zero(ip46_address_t *ip46_address, u8 is_ip4)
Definition: ip.c:20
static const transport_proto_vft_t tcp_proto
Definition: tcp.c:1363
#define seq_lt(_s1, _s2)
Definition: tcp.h:866
void tcp_punt_unknown(vlib_main_t *vm, u8 is_ip4, u8 is_add)
Definition: tcp.c:1629
void tcp_connection_reset(tcp_connection_t *tc)
Notify session that connection has been reset.
Definition: tcp.c:343
vlib_node_registration_t tcp4_input_node
(constructor) VLIB_REGISTER_NODE (tcp4_input_node)
Definition: tcp_input.c:3732
static u32 tcp_session_bind(u32 session_index, transport_endpoint_t *tep)
Definition: tcp.c:148
index_t dpoi_index
the index of objects of that type
Definition: dpo.h:186
vl_api_address_t ip
Definition: l2.api:489
ip46_address_t ip
Definition: tcp.c:36
#define FIB_NODE_INDEX_INVALID
Definition: fib_types.h:31
#define foreach_tcp_fsm_state
TCP FSM state definitions as per RFC793.
Definition: tcp.h:43
void tcp_send_fin(tcp_connection_t *tc)
Send FIN.
Definition: tcp_output.c:1015
#define foreach_tcp_connection_flag
TCP connection flags.
Definition: tcp.h:131
#define foreach_tcp_timer
TCP timers.
Definition: tcp.h:70
static u8 * format_tcp_connection_id(u8 *s, va_list *args)
Definition: tcp.c:1036
#define vec_len(v)
Number of elements in vector (rvalue-only, NULL tolerant)
void vlib_rpc_call_main_thread(void *callback, u8 *args, u32 arg_size)
Definition: threads.c:1826
static void tcp_cc_cleanup(tcp_connection_t *tc)
Definition: tcp.c:80
uword unformat_vlib_cli_sub_input(unformat_input_t *i, va_list *args)
Definition: cli.c:162
void() timer_expiration_handler(u32 index)
Definition: tcp.h:85
sack_scoreboard_hole_t * scoreboard_first_hole(sack_scoreboard_t *sb)
Definition: tcp_input.c:695
static u32 tcp_round_snd_space(tcp_connection_t *tc, u32 snd_space)
Definition: tcp.c:1261
static tcp_worker_ctx_t * tcp_get_worker(u32 thread_index)
Definition: tcp.h:687
void session_transport_closed_notify(transport_connection_t *tc)
Notification from transport that it is closed.
Definition: session.c:944
u64 uword
Definition: types.h:112
void tcp_connection_init_vars(tcp_connection_t *tc)
Initialize tcp connection variables.
Definition: tcp.c:724
#define DPO_INVALID
An initialiser for DPOs declared on the stack.
Definition: dpo.h:197
const char * tcp_conn_timers[]
Definition: tcp.c:898
static u8 * format_tcp_session(u8 *s, va_list *args)
Definition: tcp.c:1079
left
#define hash_get_mem(h, key)
Definition: hash.h:269
unformat_function_t unformat_memory_size
Definition: format.h:296
u8 * format_tcp_connection(u8 *s, va_list *args)
Definition: tcp.c:1060
static u8 * format_tcp_sack_hole(u8 *s, va_list *args)
Definition: tcp.c:1165
vnet_link_t link_type
Definition: tcp.c:35
static tcp_connection_t * tcp_connection_get(u32 conn_index, u32 thread_index)
Definition: tcp.h:714
static u32 random_u32(u32 *seed)
32-bit random number generator
Definition: random.h:69
u8 * format_unformat_error(u8 *s, va_list *va)
Definition: unformat.c:91
#define TCP_SCOREBOARD_TRACE
Definition: tcp.h:162
#define TCP_TIMER_TICK
Timer tick in seconds.
Definition: tcp.h:94
int session_stream_connect_notify(transport_connection_t *tc, u8 is_fail)
Definition: session.c:756
static u16 tcp_session_cal_goal_size(tcp_connection_t *tc)
Definition: tcp.c:1227
static vlib_thread_main_t * vlib_get_thread_main()
Definition: global_funcs.h:32
void tcp_connection_cleanup(tcp_connection_t *tc)
Cleans up connection state.
Definition: tcp.c:239
void tcp_connection_del(tcp_connection_t *tc)
Connection removal.
Definition: tcp.c:292
adj_index_t adj_nbr_add_or_lock(fib_protocol_t nh_proto, vnet_link_t link_type, const ip46_address_t *nh_addr, u32 sw_if_index)
Neighbour Adjacency sub-type.
Definition: adj_nbr.c:218
void dpo_reset(dpo_id_t *dpo)
reset a DPO ID The DPO will be unlocked.
Definition: dpo.c:232
#define vec_foreach(var, vec)
Vector iterator.
#define TCP_TSTAMP_RESOLUTION
Time stamp resolution.
Definition: tcp.h:29
static u8 * format_tcp_half_open_session(u8 *s, va_list *args)
Definition: tcp.c:1108
static u32 tcp_snd_space_inline(tcp_connection_t *tc)
Compute tx window session is allowed to fill.
Definition: tcp.c:1288
int transport_alloc_local_port(u8 proto, ip46_address_t *ip)
Allocate local port and add if successful add entry to local endpoint table to mark the pair as used...
Definition: transport.c:400
u8 * format_tcp_state(u8 *s, va_list *args)
Definition: tcp.c:837
static clib_error_t * ip6_lookup_init(vlib_main_t *vm)
Definition: ip6_forward.c:2736
static clib_error_t * show_tcp_punt_fn(vlib_main_t *vm, unformat_input_t *input, vlib_cli_command_t *cmd_arg)
Definition: tcp.c:2269
u32 table_id
Definition: fib_types.api:118
static void tcp_update_time(f64 now, u8 thread_index)
Definition: tcp.c:1343
void tcp_enable_pacing(tcp_connection_t *tc)
Definition: tcp.c:711
static u64 tcp_cc_get_pacing_rate(tcp_connection_t *tc)
Definition: tcp.h:1092
static tcp_main_t * vnet_get_tcp_main()
Definition: tcp.h:681
static void tcp_cong_recovery_off(tcp_connection_t *tc)
Definition: tcp.h:480
timer_expiration_handler tcp_timer_delack_handler
static void tcp_session_reset(u32 conn_index, u32 thread_index)
Definition: tcp.c:487
void vlib_cli_output(vlib_main_t *vm, char *fmt,...)
Definition: cli.c:772
static_always_inline void clib_spinlock_lock_if_init(clib_spinlock_t *p)
Definition: lock.h:95
static u32 tcp_set_time_now(tcp_worker_ctx_t *wrk)
Definition: tcp.h:1027
void transport_connection_tx_pacer_reset(transport_connection_t *tc, u64 rate_bytes_per_sec, u32 start_bucket)
Definition: transport.c:631
static clib_error_t * tcp_show_scoreboard_trace_fn(vlib_main_t *vm, unformat_input_t *input, vlib_cli_command_t *cmd_arg)
Definition: tcp.c:2107
uword unformat(unformat_input_t *i, const char *fmt,...)
Definition: unformat.c:978
void tcp_bt_cleanup(tcp_connection_t *tc)
Byte tracker cleanup.
Definition: tcp_bt.c:639
static tcp_connection_t * tcp_listener_get(u32 tli)
Definition: tcp.h:764
static uword unformat_check_input(unformat_input_t *i)
Definition: format.h:171
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:99
#define TCP_EVT(_evt, _args...)
Definition: tcp_debug.h:145
int tcp_session_custom_tx(void *conn, u32 max_burst_size)
Definition: tcp_output.c:2193
static uword pool_elts(void *v)
Number of active elements in a pool.
Definition: pool.h:128