FD.io VPP  v20.01-48-g3e0dafb74
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>
27 #include <math.h>
28 
30 
31 typedef struct
32 {
35  ip46_address_t ip;
39 
40 static void
42 {
43  u32 ai;
44  if (args->is_add)
45  {
46  adj_nbr_add_or_lock (args->nh_proto, args->link_type, &args->ip,
47  args->sw_if_index);
48  }
49  else
50  {
52  args->sw_if_index);
53  if (ai != ADJ_INDEX_INVALID)
54  adj_unlock (ai);
55  }
56 }
57 
58 static void
60 {
61  tcp_add_del_adj_args_t args = {
63  .link_type = VNET_LINK_IP6,
64  .ip = tc->c_rmt_ip,
65  .sw_if_index = tc->sw_if_index,
66  .is_add = is_add
67  };
69  sizeof (args));
70 }
71 
72 static void
74 {
75  tc->cc_algo->init (tc);
76 }
77 
78 static void
80 {
81  if (tc->cc_algo->cleanup)
82  tc->cc_algo->cleanup (tc);
83 }
84 
85 void
87  const tcp_cc_algorithm_t * vft)
88 {
90  vec_validate (tm->cc_algos, type);
91 
92  tm->cc_algos[type] = *vft;
93  hash_set_mem (tm->cc_algo_by_name, vft->name, type);
94 }
95 
98 {
100  return &tm->cc_algos[type];
101 }
102 
105 {
106  tcp_main_t *tm = vnet_get_tcp_main ();
107  tcp_cc_algo_register (++tm->cc_last_type, vft);
108  return tm->cc_last_type;
109 }
110 
111 static u32
113 {
114  tcp_main_t *tm = &tcp_main;
116  void *iface_ip;
117 
118  pool_get (tm->listener_pool, listener);
119  clib_memset (listener, 0, sizeof (*listener));
120 
121  listener->c_c_index = listener - tm->listener_pool;
122  listener->c_lcl_port = lcl->port;
123 
124  /* If we are provided a sw_if_index, bind using one of its ips */
125  if (ip_is_zero (&lcl->ip, 1) && lcl->sw_if_index != ENDPOINT_INVALID_INDEX)
126  {
127  if ((iface_ip = ip_interface_get_first_ip (lcl->sw_if_index,
128  lcl->is_ip4)))
129  ip_set (&lcl->ip, iface_ip, lcl->is_ip4);
130  }
131  ip_copy (&listener->c_lcl_ip, &lcl->ip, lcl->is_ip4);
132  listener->c_is_ip4 = lcl->is_ip4;
133  listener->c_proto = TRANSPORT_PROTO_TCP;
134  listener->c_s_index = session_index;
135  listener->c_fib_index = lcl->fib_index;
136  listener->state = TCP_STATE_LISTEN;
137  listener->cc_algo = tcp_cc_algo_get (tcp_cfg.cc_algo);
138 
139  tcp_connection_timers_init (listener);
140 
141  TCP_EVT (TCP_EVT_BIND, listener);
142 
143  return listener->c_c_index;
144 }
145 
146 static u32
148 {
149  return tcp_connection_bind (session_index, tep);
150 }
151 
152 static void
153 tcp_connection_unbind (u32 listener_index)
154 {
155  tcp_main_t *tm = vnet_get_tcp_main ();
156  tcp_connection_t *tc;
157 
158  tc = pool_elt_at_index (tm->listener_pool, listener_index);
159 
160  TCP_EVT (TCP_EVT_UNBIND, tc);
161 
162  /* Poison the entry */
163  if (CLIB_DEBUG > 0)
164  clib_memset (tc, 0xFA, sizeof (*tc));
165 
166  pool_put_index (tm->listener_pool, listener_index);
167 }
168 
169 static u32
170 tcp_session_unbind (u32 listener_index)
171 {
172  tcp_connection_unbind (listener_index);
173  return 0;
174 }
175 
176 static transport_connection_t *
178 {
179  tcp_main_t *tm = vnet_get_tcp_main ();
180  tcp_connection_t *tc;
181  tc = pool_elt_at_index (tm->listener_pool, listener_index);
182  return &tc->connection;
183 }
184 
185 /**
186  * Cleanup half-open connection
187  *
188  */
189 static void
191 {
192  tcp_main_t *tm = vnet_get_tcp_main ();
193  clib_spinlock_lock_if_init (&tm->half_open_lock);
194  if (CLIB_DEBUG)
195  clib_memset (tc, 0xFA, sizeof (*tc));
196  pool_put (tm->half_open_connections, tc);
197  clib_spinlock_unlock_if_init (&tm->half_open_lock);
198 }
199 
200 /**
201  * Try to cleanup half-open connection
202  *
203  * If called from a thread that doesn't own tc, the call won't have any
204  * effect.
205  *
206  * @param tc - connection to be cleaned up
207  * @return non-zero if cleanup failed.
208  */
209 int
211 {
212  /* Make sure this is the owning thread */
213  if (tc->c_thread_index != vlib_get_thread_index ())
214  return 1;
215  tcp_timer_reset (tc, TCP_TIMER_RETRANSMIT_SYN);
217  return 0;
218 }
219 
220 static tcp_connection_t *
222 {
223  tcp_main_t *tm = vnet_get_tcp_main ();
224  tcp_connection_t *tc = 0;
225  ASSERT (vlib_get_thread_index () == 0);
226  pool_get (tm->half_open_connections, tc);
227  clib_memset (tc, 0, sizeof (*tc));
228  tc->c_c_index = tc - tm->half_open_connections;
229  return tc;
230 }
231 
232 /**
233  * Cleans up connection state.
234  *
235  * No notifications.
236  */
237 void
239 {
240  tcp_main_t *tm = &tcp_main;
241 
242  TCP_EVT (TCP_EVT_DELETE, tc);
243 
244  /* Cleanup local endpoint if this was an active connect */
245  transport_endpoint_cleanup (TRANSPORT_PROTO_TCP, &tc->c_lcl_ip,
246  tc->c_lcl_port);
247 
248  /* Check if connection is not yet fully established */
249  if (tc->state == TCP_STATE_SYN_SENT)
250  {
251  /* Try to remove the half-open connection. If this is not the owning
252  * thread, tc won't be removed. Retransmit or establish timers will
253  * eventually expire and call again cleanup on the right thread. */
255  tc->flags |= TCP_CONN_HALF_OPEN_DONE;
256  }
257  else
258  {
259  int thread_index = tc->c_thread_index;
260 
261  /* Make sure all timers are cleared */
263 
264  if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6))
265  tcp_add_del_adjacency (tc, 0);
266 
267  tcp_cc_cleanup (tc);
268  vec_free (tc->snd_sacks);
269  vec_free (tc->snd_sacks_fl);
270  vec_free (tc->rcv_opts.sacks);
271  pool_free (tc->sack_sb.holes);
272 
273  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
274  tcp_bt_cleanup (tc);
275 
276  /* Poison the entry */
277  if (CLIB_DEBUG > 0)
278  clib_memset (tc, 0xFA, sizeof (*tc));
279  pool_put (tm->connections[thread_index], tc);
280  }
281 }
282 
283 /**
284  * Connection removal.
285  *
286  * This should be called only once connection enters CLOSED state. Note
287  * that it notifies the session of the removal event, so if the goal is to
288  * just remove the connection, call tcp_connection_cleanup instead.
289  */
290 void
292 {
293  session_transport_delete_notify (&tc->connection);
295 }
296 
298 tcp_connection_alloc (u8 thread_index)
299 {
300  tcp_main_t *tm = vnet_get_tcp_main ();
301  tcp_connection_t *tc;
302 
303  pool_get (tm->connections[thread_index], tc);
304  clib_memset (tc, 0, sizeof (*tc));
305  tc->c_c_index = tc - tm->connections[thread_index];
306  tc->c_thread_index = thread_index;
307  return tc;
308 }
309 
312 {
313  tcp_main_t *tm = vnet_get_tcp_main ();
314  tcp_connection_t *tc;
315 
316  pool_get (tm->connections[thread_index], tc);
317  clib_memcpy_fast (tc, base, sizeof (*tc));
318  tc->c_c_index = tc - tm->connections[thread_index];
319  tc->c_thread_index = thread_index;
320  return tc;
321 }
322 
323 void
325 {
326  tcp_main_t *tm = &tcp_main;
327  if (CLIB_DEBUG)
328  {
329  u8 thread_index = tc->c_thread_index;
330  clib_memset (tc, 0xFA, sizeof (*tc));
331  pool_put (tm->connections[thread_index], tc);
332  return;
333  }
334  pool_put (tm->connections[tc->c_thread_index], tc);
335 }
336 
337 /** Notify session that connection has been reset.
338  *
339  * Switch state to closed and wait for session to call cleanup.
340  */
341 void
343 {
344  TCP_EVT (TCP_EVT_RST_RCVD, tc);
345  switch (tc->state)
346  {
347  case TCP_STATE_SYN_RCVD:
348  /* Cleanup everything. App wasn't notified yet */
349  session_transport_delete_notify (&tc->connection);
351  break;
352  case TCP_STATE_SYN_SENT:
353  session_stream_connect_notify (&tc->connection, 1 /* fail */ );
355  break;
356  case TCP_STATE_ESTABLISHED:
358  /* Set the cleanup timer, in case the session layer/app don't
359  * cleanly close the connection */
360  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
361  session_transport_reset_notify (&tc->connection);
363  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
364  session_transport_closed_notify (&tc->connection);
365  break;
366  case TCP_STATE_CLOSE_WAIT:
367  case TCP_STATE_FIN_WAIT_1:
368  case TCP_STATE_FIN_WAIT_2:
369  case TCP_STATE_CLOSING:
370  case TCP_STATE_LAST_ACK:
372  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
374  /* Make sure we mark the session as closed. In some states we may
375  * be still trying to send data */
376  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
377  session_transport_closed_notify (&tc->connection);
378  break;
379  case TCP_STATE_CLOSED:
380  case TCP_STATE_TIME_WAIT:
381  break;
382  default:
383  TCP_DBG ("reset state: %u", tc->state);
384  }
385 }
386 
387 /**
388  * Begin connection closing procedure.
389  *
390  * If at the end the connection is not in CLOSED state, it is not removed.
391  * Instead, we rely on on TCP to advance through state machine to either
392  * 1) LAST_ACK (passive close) whereby when the last ACK is received
393  * tcp_connection_del is called. This notifies session of the delete and
394  * calls cleanup.
395  * 2) TIME_WAIT (active close) whereby after 2MSL the 2MSL timer triggers
396  * and cleanup is called.
397  *
398  * N.B. Half-close connections are not supported
399  */
400 void
402 {
403  TCP_EVT (TCP_EVT_CLOSE, tc);
404 
405  /* Send/Program FIN if needed and switch state */
406  switch (tc->state)
407  {
408  case TCP_STATE_SYN_SENT:
409  /* Try to cleanup. If not on the right thread, mark as half-open done.
410  * Connection will be cleaned up when establish timer pops */
412  break;
413  case TCP_STATE_SYN_RCVD:
415  tcp_send_fin (tc);
416  tcp_connection_set_state (tc, TCP_STATE_FIN_WAIT_1);
417  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.finwait1_time);
418  break;
419  case TCP_STATE_ESTABLISHED:
420  /* If closing with unread data, reset the connection */
421  if (transport_max_rx_dequeue (&tc->connection))
422  {
423  tcp_send_reset (tc);
425  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
426  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.closewait_time);
427  session_transport_closed_notify (&tc->connection);
428  break;
429  }
430  if (!transport_max_tx_dequeue (&tc->connection))
431  tcp_send_fin (tc);
432  else
433  tc->flags |= TCP_CONN_FINPNDG;
434  tcp_connection_set_state (tc, TCP_STATE_FIN_WAIT_1);
435  /* Set a timer in case the peer stops responding. Otherwise the
436  * connection will be stuck here forever. */
437  ASSERT (tc->timers[TCP_TIMER_WAITCLOSE] == TCP_TIMER_HANDLE_INVALID);
438  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.finwait1_time);
439  break;
440  case TCP_STATE_CLOSE_WAIT:
441  if (!transport_max_tx_dequeue (&tc->connection))
442  {
443  tcp_send_fin (tc);
445  tcp_connection_set_state (tc, TCP_STATE_LAST_ACK);
446  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.lastack_time);
447  }
448  else
449  tc->flags |= TCP_CONN_FINPNDG;
450  break;
451  case TCP_STATE_FIN_WAIT_1:
452  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.finwait1_time);
453  break;
454  case TCP_STATE_CLOSED:
456  /* Delete connection but instead of doing it now wait until next
457  * dispatch cycle to give the session layer a chance to clear
458  * unhandled events */
459  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
460  break;
461  default:
462  TCP_DBG ("state: %u", tc->state);
463  }
464 }
465 
466 static void
467 tcp_session_close (u32 conn_index, u32 thread_index)
468 {
469  tcp_connection_t *tc;
470  tc = tcp_connection_get (conn_index, thread_index);
472 }
473 
474 static void
475 tcp_session_cleanup (u32 conn_index, u32 thread_index)
476 {
477  tcp_connection_t *tc;
478  tc = tcp_connection_get (conn_index, thread_index);
479  if (!tc)
480  return;
481  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
483 }
484 
485 static void
486 tcp_session_reset (u32 conn_index, u32 thread_index)
487 {
488  tcp_connection_t *tc;
489  tc = tcp_connection_get (conn_index, thread_index);
490  session_transport_closed_notify (&tc->connection);
491  tcp_send_reset (tc);
494  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
495  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
496 }
497 
498 /**
499  * Initialize all connection timers as invalid
500  */
501 void
503 {
504  int i;
505 
506  /* Set all to invalid */
507  for (i = 0; i < TCP_N_TIMERS; i++)
508  {
509  tc->timers[i] = TCP_TIMER_HANDLE_INVALID;
510  }
511 
512  tc->rto = TCP_RTO_INIT;
513 }
514 
515 /**
516  * Stop all connection timers
517  */
518 void
520 {
521  int i;
522  for (i = 0; i < TCP_N_TIMERS; i++)
523  {
524  tcp_timer_reset (tc, i);
525  }
526 }
527 
528 #if 0
529 typedef struct ip4_tcp_hdr
530 {
532  tcp_header_t tcp;
533 } ip4_tcp_hdr_t;
534 
535 typedef struct ip6_tcp_hdr
536 {
538  tcp_header_t tcp;
539 } ip6_tcp_hdr_t;
540 
541 static void
542 tcp_connection_select_lb_bucket (tcp_connection_t * tc, const dpo_id_t * dpo,
543  dpo_id_t * result)
544 {
545  const dpo_id_t *choice;
546  load_balance_t *lb;
547  int hash;
548 
549  lb = load_balance_get (dpo->dpoi_index);
550  if (tc->c_is_ip4)
551  {
552  ip4_tcp_hdr_t hdr;
553  clib_memset (&hdr, 0, sizeof (hdr));
554  hdr.ip.protocol = IP_PROTOCOL_TCP;
555  hdr.ip.address_pair.src.as_u32 = tc->c_lcl_ip.ip4.as_u32;
556  hdr.ip.address_pair.dst.as_u32 = tc->c_rmt_ip.ip4.as_u32;
557  hdr.tcp.src_port = tc->c_lcl_port;
558  hdr.tcp.dst_port = tc->c_rmt_port;
559  hash = ip4_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
560  }
561  else
562  {
563  ip6_tcp_hdr_t hdr;
564  clib_memset (&hdr, 0, sizeof (hdr));
565  hdr.ip.protocol = IP_PROTOCOL_TCP;
566  clib_memcpy_fast (&hdr.ip.src_address, &tc->c_lcl_ip.ip6,
567  sizeof (ip6_address_t));
568  clib_memcpy_fast (&hdr.ip.dst_address, &tc->c_rmt_ip.ip6,
569  sizeof (ip6_address_t));
570  hdr.tcp.src_port = tc->c_lcl_port;
571  hdr.tcp.dst_port = tc->c_rmt_port;
572  hash = ip6_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
573  }
574  choice = load_balance_get_bucket_i (lb, hash & lb->lb_n_buckets_minus_1);
575  dpo_copy (result, choice);
576 }
577 
580 {
582  u32 fib_index;
583 
584  clib_memcpy_fast (&prefix.fp_addr, &tc->c_rmt_ip, sizeof (prefix.fp_addr));
585  prefix.fp_proto = tc->c_is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6;
586  prefix.fp_len = tc->c_is_ip4 ? 32 : 128;
587  fib_index = fib_table_find (prefix.fp_proto, tc->c_fib_index);
588  return fib_table_lookup (fib_index, &prefix);
589 }
590 
591 static int
592 tcp_connection_stack_on_fib_entry (tcp_connection_t * tc)
593 {
594  dpo_id_t choice = DPO_INVALID;
595  u32 output_node_index;
596  fib_entry_t *fe;
597 
598  fe = fib_entry_get (tc->c_rmt_fei);
599  if (fe->fe_lb.dpoi_type != DPO_LOAD_BALANCE)
600  return -1;
601 
602  tcp_connection_select_lb_bucket (tc, &fe->fe_lb, &choice);
603 
604  output_node_index =
605  tc->c_is_ip4 ? tcp4_output_node.index : tcp6_output_node.index;
606  dpo_stack_from_node (output_node_index, &tc->c_rmt_dpo, &choice);
607  return 0;
608 }
609 
610 /** Stack tcp connection on peer's fib entry.
611  *
612  * This ultimately populates the dpo the connection will use to send packets.
613  */
614 static void
615 tcp_connection_fib_attach (tcp_connection_t * tc)
616 {
617  tc->c_rmt_fei = tcp_lookup_rmt_in_fib (tc);
618 
619  ASSERT (tc->c_rmt_fei != FIB_NODE_INDEX_INVALID);
620 
621  tcp_connection_stack_on_fib_entry (tc);
622 }
623 #endif /* 0 */
624 
625 /**
626  * Generate random iss as per rfc6528
627  */
628 static u32
630 {
631  tcp_main_t *tm = &tcp_main;
632  u64 tmp;
633 
634  if (tc->c_is_ip4)
635  tmp = (u64) tc->c_lcl_ip.ip4.as_u32 << 32 | (u64) tc->c_rmt_ip.ip4.as_u32;
636  else
637  tmp = tc->c_lcl_ip.ip6.as_u64[0] ^ tc->c_lcl_ip.ip6.as_u64[1]
638  ^ tc->c_rmt_ip.ip6.as_u64[0] ^ tc->c_rmt_ip.ip6.as_u64[1];
639 
640  tmp ^= tm->iss_seed.first | ((u64) tc->c_lcl_port << 16 | tc->c_rmt_port);
641  tmp ^= tm->iss_seed.second;
642  tmp = clib_xxhash (tmp) + clib_cpu_time_now ();
643  return ((tmp >> 32) ^ (tmp & 0xffffffff));
644 }
645 
646 /**
647  * Initialize max segment size we're able to process.
648  *
649  * The value is constrained by the output interface's MTU and by the size
650  * of the IP and TCP headers (see RFC6691). It is also what we advertise
651  * to our peer.
652  */
653 static void
655 {
656  u8 ip_hdr_len;
657 
658  ip_hdr_len = tc->c_is_ip4 ? sizeof (ip4_header_t) : sizeof (ip6_header_t);
659  tc->mss = tcp_cfg.default_mtu - sizeof (tcp_header_t) - ip_hdr_len;
660 }
661 
662 static void
664 {
665  u16 default_min_mss = 536;
666 
667  tcp_init_rcv_mss (tc);
668 
669  /* TODO consider PMTU discovery */
670  tc->snd_mss = clib_min (tc->rcv_opts.mss, tc->mss);
671 
672  if (tc->snd_mss < 45)
673  {
674  /* Assume that at least the min default mss works */
675  tc->snd_mss = default_min_mss;
676  tc->rcv_opts.mss = default_min_mss;
677  }
678 
679  /* We should have enough space for 40 bytes of options */
680  ASSERT (tc->snd_mss > 45);
681 
682  /* If we use timestamp option, account for it */
683  if (tcp_opts_tstamp (&tc->rcv_opts))
684  tc->snd_mss -= TCP_OPTION_LEN_TIMESTAMP;
685 }
686 
687 /**
688  * Initialize connection send variables.
689  */
690 void
692 {
693  /*
694  * We use the time to randomize iss and for setting up the initial
695  * timestamp. Make sure it's updated otherwise syn and ack in the
696  * handshake may make it look as if time has flown in the opposite
697  * direction for us.
698  */
700 
701  tcp_init_rcv_mss (tc);
702  tc->iss = tcp_generate_random_iss (tc);
703  tc->snd_una = tc->iss;
704  tc->snd_nxt = tc->iss + 1;
705  tc->snd_una_max = tc->snd_nxt;
706  tc->srtt = 100; /* 100 ms */
707 
708  if (!tcp_cfg.csum_offload)
709  tc->cfg_flags |= TCP_CFG_F_NO_CSUM_OFFLOAD;
710 }
711 
712 void
714 {
715  u32 byte_rate;
716  byte_rate = tc->cwnd / (tc->srtt * TCP_TICK);
717  transport_connection_tx_pacer_init (&tc->connection, byte_rate, tc->cwnd);
718  tc->mrtt_us = (u32) ~ 0;
719 }
720 
721 /** Initialize tcp connection variables
722  *
723  * Should be called after having received a msg from the peer, i.e., a SYN or
724  * a SYNACK, such that connection options have already been exchanged. */
725 void
727 {
729  tcp_init_mss (tc);
730  scoreboard_init (&tc->sack_sb);
731  if (tc->state == TCP_STATE_SYN_RCVD)
732  tcp_init_snd_vars (tc);
733 
734  tcp_cc_init (tc);
735 
736  if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6))
737  tcp_add_del_adjacency (tc, 1);
738 
739  /* tcp_connection_fib_attach (tc); */
740 
741  if (transport_connection_is_tx_paced (&tc->connection)
742  || tcp_cfg.enable_tx_pacing)
743  tcp_enable_pacing (tc);
744 
745  if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
746  tcp_bt_init (tc);
747 
748  if (!tcp_cfg.allow_tso)
749  tc->cfg_flags |= TCP_CFG_F_NO_TSO;
750 
751  tc->start_ts = tcp_time_now_us (tc->c_thread_index);
752 }
753 
754 static int
755 tcp_alloc_custom_local_endpoint (tcp_main_t * tm, ip46_address_t * lcl_addr,
756  u16 * lcl_port, u8 is_ip4)
757 {
758  int index, port;
759  if (is_ip4)
760  {
761  index = tm->last_v4_addr_rotor++;
762  if (tm->last_v4_addr_rotor >= vec_len (tcp_cfg.ip4_src_addrs))
763  tm->last_v4_addr_rotor = 0;
764  lcl_addr->ip4.as_u32 = tcp_cfg.ip4_src_addrs[index].as_u32;
765  }
766  else
767  {
768  index = tm->last_v6_addr_rotor++;
769  if (tm->last_v6_addr_rotor >= vec_len (tcp_cfg.ip6_src_addrs))
770  tm->last_v6_addr_rotor = 0;
771  clib_memcpy_fast (&lcl_addr->ip6, &tcp_cfg.ip6_src_addrs[index],
772  sizeof (ip6_address_t));
773  }
774  port = transport_alloc_local_port (TRANSPORT_PROTO_TCP, lcl_addr);
775  if (port < 1)
776  {
777  clib_warning ("Failed to allocate src port");
778  return -1;
779  }
780  *lcl_port = port;
781  return 0;
782 }
783 
784 static int
786 {
787  tcp_main_t *tm = vnet_get_tcp_main ();
788  tcp_connection_t *tc;
789  ip46_address_t lcl_addr;
790  u16 lcl_port;
791  int rv;
792 
793  /*
794  * Allocate local endpoint
795  */
796  if ((rmt->is_ip4 && vec_len (tcp_cfg.ip4_src_addrs))
797  || (!rmt->is_ip4 && vec_len (tcp_cfg.ip6_src_addrs)))
798  rv = tcp_alloc_custom_local_endpoint (tm, &lcl_addr, &lcl_port,
799  rmt->is_ip4);
800  else
801  rv = transport_alloc_local_endpoint (TRANSPORT_PROTO_TCP,
802  rmt, &lcl_addr, &lcl_port);
803 
804  if (rv)
805  return -1;
806 
807  /*
808  * Create connection and send SYN
809  */
810  clib_spinlock_lock_if_init (&tm->half_open_lock);
812  ip_copy (&tc->c_rmt_ip, &rmt->ip, rmt->is_ip4);
813  ip_copy (&tc->c_lcl_ip, &lcl_addr, rmt->is_ip4);
814  tc->c_rmt_port = rmt->port;
815  tc->c_lcl_port = clib_host_to_net_u16 (lcl_port);
816  tc->c_is_ip4 = rmt->is_ip4;
817  tc->c_proto = TRANSPORT_PROTO_TCP;
818  tc->c_fib_index = rmt->fib_index;
819  tc->cc_algo = tcp_cc_algo_get (tcp_cfg.cc_algo);
820  /* The other connection vars will be initialized after SYN ACK */
822 
823  TCP_EVT (TCP_EVT_OPEN, tc);
824  tc->state = TCP_STATE_SYN_SENT;
825  tcp_init_snd_vars (tc);
826  tcp_send_syn (tc);
827  clib_spinlock_unlock_if_init (&tm->half_open_lock);
828 
829  return tc->c_c_index;
830 }
831 
832 const char *tcp_fsm_states[] = {
833 #define _(sym, str) str,
835 #undef _
836 };
837 
838 u8 *
839 format_tcp_state (u8 * s, va_list * args)
840 {
841  u32 state = va_arg (*args, u32);
842 
843  if (state < TCP_N_STATES)
844  s = format (s, "%s", tcp_fsm_states[state]);
845  else
846  s = format (s, "UNKNOWN (%d (0x%x))", state, state);
847  return s;
848 }
849 
850 const char *tcp_cfg_flags_str[] = {
851 #define _(sym, str) str,
853 #undef _
854 };
855 
856 static u8 *
857 format_tcp_cfg_flags (u8 * s, va_list * args)
858 {
859  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
860  int i, last = -1;
861 
862  for (i = 0; i < TCP_CFG_N_FLAG_BITS; i++)
863  if (tc->cfg_flags & (1 << i))
864  last = i;
865  for (i = 0; i < last; i++)
866  {
867  if (tc->cfg_flags & (1 << i))
868  s = format (s, "%s, ", tcp_cfg_flags_str[i]);
869  }
870  if (last >= 0)
871  s = format (s, "%s", tcp_cfg_flags_str[last]);
872  return s;
873 }
874 
875 const char *tcp_connection_flags_str[] = {
876 #define _(sym, str) str,
878 #undef _
879 };
880 
881 static u8 *
882 format_tcp_connection_flags (u8 * s, va_list * args)
883 {
884  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
885  int i, last = -1;
886 
887  for (i = 0; i < TCP_CONN_N_FLAG_BITS; i++)
888  if (tc->flags & (1 << i))
889  last = i;
890  for (i = 0; i < last; i++)
891  {
892  if (tc->flags & (1 << i))
893  s = format (s, "%s, ", tcp_connection_flags_str[i]);
894  }
895  if (last >= 0)
896  s = format (s, "%s", tcp_connection_flags_str[last]);
897  return s;
898 }
899 
900 const char *tcp_conn_timers[] = {
901 #define _(sym, str) str,
903 #undef _
904 };
905 
906 static u8 *
907 format_tcp_timers (u8 * s, va_list * args)
908 {
909  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
910  int i, last = -1;
911 
912  for (i = 0; i < TCP_N_TIMERS; i++)
913  if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
914  last = i;
915 
916  for (i = 0; i < last; i++)
917  {
918  if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
919  s = format (s, "%s,", tcp_conn_timers[i]);
920  }
921 
922  if (last >= 0)
923  s = format (s, "%s", tcp_conn_timers[i]);
924 
925  return s;
926 }
927 
928 static u8 *
929 format_tcp_congestion_status (u8 * s, va_list * args)
930 {
931  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
932  if (tcp_in_recovery (tc))
933  s = format (s, "recovery");
934  else if (tcp_in_fastrecovery (tc))
935  s = format (s, "fastrecovery");
936  else
937  s = format (s, "none");
938  return s;
939 }
940 
941 static i32
943 {
944  return (i32) tc->rcv_wnd - (tc->rcv_nxt - tc->rcv_las);
945 }
946 
947 static u8 *
948 format_tcp_congestion (u8 * s, va_list * args)
949 {
950  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
951  u32 indent = format_get_indent (s), prr_space = 0;
952 
953  s = format (s, "%U ", format_tcp_congestion_status, tc);
954  s = format (s, "algo %s cwnd %u ssthresh %u bytes_acked %u\n",
955  tc->cc_algo->name, tc->cwnd, tc->ssthresh, tc->bytes_acked);
956  s = format (s, "%Ucc space %u prev_cwnd %u prev_ssthresh %u\n",
958  tc->prev_cwnd, tc->prev_ssthresh);
959  s = format (s, "%Usnd_cong %u dupack %u limited_tx %u\n",
960  format_white_space, indent, tc->snd_congestion - tc->iss,
961  tc->rcv_dupacks, tc->limited_transmit - tc->iss);
962  s = format (s, "%Urxt_bytes %u rxt_delivered %u rxt_head %u rxt_ts %u\n",
963  format_white_space, indent, tc->snd_rxt_bytes,
964  tc->rxt_delivered, tc->rxt_head - tc->iss,
965  tcp_time_now_w_thread (tc->c_thread_index) - tc->snd_rxt_ts);
966  if (tcp_in_fastrecovery (tc))
967  prr_space = tcp_fastrecovery_prr_snd_space (tc);
968  s = format (s, "%Uprr_start %u prr_delivered %u prr space %u\n",
969  format_white_space, indent, tc->prr_start - tc->iss,
970  tc->prr_delivered, prr_space);
971  return s;
972 }
973 
974 static u8 *
975 format_tcp_stats (u8 * s, va_list * args)
976 {
977  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
978  u32 indent = format_get_indent (s);
979  s = format (s, "in segs %lu dsegs %lu bytes %lu dupacks %u\n",
980  tc->segs_in, tc->data_segs_in, tc->bytes_in, tc->dupacks_in);
981  s = format (s, "%Uout segs %lu dsegs %lu bytes %lu dupacks %u\n",
982  format_white_space, indent, tc->segs_out,
983  tc->data_segs_out, tc->bytes_out, tc->dupacks_out);
984  s = format (s, "%Ufr %u tr %u rxt segs %lu bytes %lu duration %.3f\n",
985  format_white_space, indent, tc->fr_occurences,
986  tc->tr_occurences, tc->segs_retrans, tc->bytes_retrans,
987  tcp_time_now_us (tc->c_thread_index) - tc->start_ts);
988  s = format (s, "%Uerr wnd data below %u above %u ack below %u above %u",
989  format_white_space, indent, tc->errors.below_data_wnd,
990  tc->errors.above_data_wnd, tc->errors.below_ack_wnd,
991  tc->errors.above_ack_wnd);
992  return s;
993 }
994 
995 static u8 *
996 format_tcp_vars (u8 * s, va_list * args)
997 {
998  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
999  s = format (s, " index: %u cfg: %U flags: %U timers: %U\n", tc->c_c_index,
1001  format_tcp_timers, tc);
1002  s = format (s, " snd_una %u snd_nxt %u snd_una_max %u",
1003  tc->snd_una - tc->iss, tc->snd_nxt - tc->iss,
1004  tc->snd_una_max - tc->iss);
1005  s = format (s, " rcv_nxt %u rcv_las %u\n",
1006  tc->rcv_nxt - tc->irs, tc->rcv_las - tc->irs);
1007  s = format (s, " snd_wnd %u rcv_wnd %u rcv_wscale %u ",
1008  tc->snd_wnd, tc->rcv_wnd, tc->rcv_wscale);
1009  s = format (s, "snd_wl1 %u snd_wl2 %u\n", tc->snd_wl1 - tc->irs,
1010  tc->snd_wl2 - tc->iss);
1011  s = format (s, " flight size %u out space %u rcv_wnd_av %u",
1013  tcp_rcv_wnd_available (tc));
1014  s = format (s, " tsval_recent %u\n", tc->tsval_recent);
1015  s = format (s, " tsecr %u tsecr_last_ack %u tsval_recent_age %u",
1016  tc->rcv_opts.tsecr, tc->tsecr_last_ack,
1017  tcp_time_now () - tc->tsval_recent_age);
1018  s = format (s, " snd_mss %u\n", tc->snd_mss);
1019  s = format (s, " rto %u rto_boff %u srtt %u us %.3f rttvar %u rtt_ts %.4f",
1020  tc->rto, tc->rto_boff, tc->srtt, tc->mrtt_us * 1000, tc->rttvar,
1021  tc->rtt_ts);
1022  s = format (s, " rtt_seq %u\n", tc->rtt_seq - tc->iss);
1023  s = format (s, " next_node %u opaque 0x%x\n", tc->next_node_index,
1024  tc->next_node_opaque);
1025  s = format (s, " cong: %U", format_tcp_congestion, tc);
1026 
1027  if (tc->state >= TCP_STATE_ESTABLISHED)
1028  {
1029  s = format (s, " sboard: %U\n", format_tcp_scoreboard, &tc->sack_sb,
1030  tc);
1031  s = format (s, " stats: %U\n", format_tcp_stats, tc);
1032  }
1033  if (vec_len (tc->snd_sacks))
1034  s = format (s, " sacks tx: %U\n", format_tcp_sacks, tc);
1035 
1036  return s;
1037 }
1038 
1039 u8 *
1040 format_tcp_connection_id (u8 * s, va_list * args)
1041 {
1042  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1043  if (!tc)
1044  return s;
1045  if (tc->c_is_ip4)
1046  {
1047  s = format (s, "[%d:%d][%s] %U:%d->%U:%d", tc->c_thread_index,
1048  tc->c_s_index, "T", format_ip4_address, &tc->c_lcl_ip4,
1049  clib_net_to_host_u16 (tc->c_lcl_port), format_ip4_address,
1050  &tc->c_rmt_ip4, clib_net_to_host_u16 (tc->c_rmt_port));
1051  }
1052  else
1053  {
1054  s = format (s, "[%d:%d][%s] %U:%d->%U:%d", tc->c_thread_index,
1055  tc->c_s_index, "T", format_ip6_address, &tc->c_lcl_ip6,
1056  clib_net_to_host_u16 (tc->c_lcl_port), format_ip6_address,
1057  &tc->c_rmt_ip6, clib_net_to_host_u16 (tc->c_rmt_port));
1058  }
1059 
1060  return s;
1061 }
1062 
1063 u8 *
1064 format_tcp_connection (u8 * s, va_list * args)
1065 {
1066  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1067  u32 verbose = va_arg (*args, u32);
1068 
1069  if (!tc)
1070  return s;
1071  s = format (s, "%-50U", format_tcp_connection_id, tc);
1072  if (verbose)
1073  {
1074  s = format (s, "%-15U", format_tcp_state, tc->state);
1075  if (verbose > 1)
1076  s = format (s, "\n%U", format_tcp_vars, tc);
1077  }
1078 
1079  return s;
1080 }
1081 
1082 static u8 *
1083 format_tcp_session (u8 * s, va_list * args)
1084 {
1085  u32 tci = va_arg (*args, u32);
1086  u32 thread_index = va_arg (*args, u32);
1087  u32 verbose = va_arg (*args, u32);
1088  tcp_connection_t *tc;
1089 
1090  tc = tcp_connection_get (tci, thread_index);
1091  if (tc)
1092  s = format (s, "%U", format_tcp_connection, tc, verbose);
1093  else
1094  s = format (s, "empty\n");
1095  return s;
1096 }
1097 
1098 static u8 *
1099 format_tcp_listener_session (u8 * s, va_list * args)
1100 {
1101  u32 tci = va_arg (*args, u32);
1102  u32 __clib_unused thread_index = va_arg (*args, u32);
1103  u32 verbose = va_arg (*args, u32);
1104  tcp_connection_t *tc = tcp_listener_get (tci);
1105  s = format (s, "%-50U", format_tcp_connection_id, tc);
1106  if (verbose)
1107  s = format (s, "%-15U", format_tcp_state, tc->state);
1108  return s;
1109 }
1110 
1111 static u8 *
1112 format_tcp_half_open_session (u8 * s, va_list * args)
1113 {
1114  u32 tci = va_arg (*args, u32);
1115  u32 __clib_unused thread_index = va_arg (*args, u32);
1117  return format (s, "%U", format_tcp_connection_id, tc);
1118 }
1119 
1120 u8 *
1121 format_tcp_sacks (u8 * s, va_list * args)
1122 {
1123  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1124  sack_block_t *sacks = tc->snd_sacks;
1125  sack_block_t *block;
1126  int i, len = 0;
1127 
1128  len = vec_len (sacks);
1129  for (i = 0; i < len - 1; i++)
1130  {
1131  block = &sacks[i];
1132  s = format (s, " start %u end %u\n", block->start - tc->irs,
1133  block->end - tc->irs);
1134  }
1135  if (len)
1136  {
1137  block = &sacks[len - 1];
1138  s = format (s, " start %u end %u", block->start - tc->irs,
1139  block->end - tc->irs);
1140  }
1141  return s;
1142 }
1143 
1144 u8 *
1145 format_tcp_rcv_sacks (u8 * s, va_list * args)
1146 {
1147  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1148  sack_block_t *sacks = tc->rcv_opts.sacks;
1149  sack_block_t *block;
1150  int i, len = 0;
1151 
1152  len = vec_len (sacks);
1153  for (i = 0; i < len - 1; i++)
1154  {
1155  block = &sacks[i];
1156  s = format (s, " start %u end %u\n", block->start - tc->iss,
1157  block->end - tc->iss);
1158  }
1159  if (len)
1160  {
1161  block = &sacks[len - 1];
1162  s = format (s, " start %u end %u", block->start - tc->iss,
1163  block->end - tc->iss);
1164  }
1165  return s;
1166 }
1167 
1168 static u8 *
1169 format_tcp_sack_hole (u8 * s, va_list * args)
1170 {
1171  sack_scoreboard_hole_t *hole = va_arg (*args, sack_scoreboard_hole_t *);
1172  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1173  if (tc)
1174  s = format (s, " [%u, %u]", hole->start - tc->iss, hole->end - tc->iss);
1175  else
1176  s = format (s, " [%u, %u]", hole->start, hole->end);
1177  return s;
1178 }
1179 
1180 u8 *
1181 format_tcp_scoreboard (u8 * s, va_list * args)
1182 {
1183  sack_scoreboard_t *sb = va_arg (*args, sack_scoreboard_t *);
1184  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1185  sack_scoreboard_hole_t *hole;
1186  u32 indent = format_get_indent (s);
1187 
1188  s = format (s, "sacked %u last_sacked %u lost %u last_lost %u"
1189  " rxt_sacked %u\n",
1190  sb->sacked_bytes, sb->last_sacked_bytes, sb->lost_bytes,
1191  sb->last_lost_bytes, sb->rxt_sacked);
1192  s = format (s, "%Ulast_delivered %u high_sacked %u is_reneging %u\n",
1193  format_white_space, indent, sb->last_bytes_delivered,
1194  sb->high_sacked - tc->iss, sb->is_reneging);
1195  s = format (s, "%Ucur_rxt_hole %u high_rxt %u rescue_rxt %u",
1196  format_white_space, indent, sb->cur_rxt_hole,
1197  sb->high_rxt - tc->iss, sb->rescue_rxt - tc->iss);
1198 
1199  hole = scoreboard_first_hole (sb);
1200  if (hole)
1201  s = format (s, "\n%Uhead %u tail %u %u holes:\n%U", format_white_space,
1202  indent, sb->head, sb->tail, pool_elts (sb->holes),
1203  format_white_space, indent);
1204 
1205  while (hole)
1206  {
1207  s = format (s, "%U", format_tcp_sack_hole, hole, tc);
1208  hole = scoreboard_next_hole (sb, hole);
1209  }
1210 
1211  return s;
1212 }
1213 
1214 static transport_connection_t *
1215 tcp_session_get_transport (u32 conn_index, u32 thread_index)
1216 {
1217  tcp_connection_t *tc = tcp_connection_get (conn_index, thread_index);
1218  if (PREDICT_FALSE (!tc))
1219  return 0;
1220  return &tc->connection;
1221 }
1222 
1223 static transport_connection_t *
1225 {
1227  return &tc->connection;
1228 }
1229 
1230 static u16
1232 {
1233  u16 goal_size = tc->snd_mss;
1234 
1235  goal_size = TCP_MAX_GSO_SZ - tc->snd_mss % TCP_MAX_GSO_SZ;
1236  goal_size = clib_min (goal_size, tc->snd_wnd / 2);
1237 
1238  return goal_size > tc->snd_mss ? goal_size : tc->snd_mss;
1239 }
1240 
1241 /**
1242  * Compute maximum segment size for session layer.
1243  *
1244  * Since the result needs to be the actual data length, it first computes
1245  * the tcp options to be used in the next burst and subtracts their
1246  * length from the connection's snd_mss.
1247  */
1248 static u16
1250 {
1251  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1252 
1253  /* Ensure snd_mss does accurately reflect the amount of data we can push
1254  * in a segment. This also makes sure that options are updated according to
1255  * the current state of the connection. */
1257 
1258  if (PREDICT_FALSE (tc->cfg_flags & TCP_CFG_F_TSO))
1259  return tcp_session_cal_goal_size (tc);
1260 
1261  return tc->snd_mss;
1262 }
1263 
1266 {
1267  if (PREDICT_FALSE (tc->snd_wnd < tc->snd_mss))
1268  {
1269  return tc->snd_wnd <= snd_space ? tc->snd_wnd : 0;
1270  }
1271 
1272  /* If not snd_wnd constrained and we can't write at least a segment,
1273  * don't try at all */
1274  if (PREDICT_FALSE (snd_space < tc->snd_mss))
1275  return snd_space < tc->cwnd ? 0 : snd_space;
1276 
1277  /* round down to mss multiple */
1278  return snd_space - (snd_space % tc->snd_mss);
1279 }
1280 
1281 /**
1282  * Compute tx window session is allowed to fill.
1283  *
1284  * Takes into account available send space, snd_mss and the congestion
1285  * state of the connection. If possible, the value returned is a multiple
1286  * of snd_mss.
1287  *
1288  * @param tc tcp connection
1289  * @return number of bytes session is allowed to write
1290  */
1291 static inline u32
1293 {
1294  int snd_space;
1295 
1297  || tc->state == TCP_STATE_CLOSED))
1298  return 0;
1299 
1300  snd_space = tcp_available_output_snd_space (tc);
1301 
1302  /* If we got dupacks or sacked bytes but we're not yet in recovery, try
1303  * to force the peer to send enough dupacks to start retransmitting as
1304  * per Limited Transmit (RFC3042)
1305  */
1306  if (PREDICT_FALSE (tc->rcv_dupacks != 0 || tc->sack_sb.sacked_bytes))
1307  {
1308  if (tc->limited_transmit != tc->snd_nxt
1309  && (seq_lt (tc->limited_transmit, tc->snd_nxt - 2 * tc->snd_mss)
1310  || seq_gt (tc->limited_transmit, tc->snd_nxt)))
1311  tc->limited_transmit = tc->snd_nxt;
1312 
1313  ASSERT (seq_leq (tc->limited_transmit, tc->snd_nxt));
1314 
1315  int snt_limited = tc->snd_nxt - tc->limited_transmit;
1316  snd_space = clib_max ((int) 2 * tc->snd_mss - snt_limited, 0);
1317  }
1318  return tcp_round_snd_space (tc, snd_space);
1319 }
1320 
1321 u32
1323 {
1324  return tcp_snd_space_inline (tc);
1325 }
1326 
1327 static u32
1329 {
1330  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1331  return clib_min (tcp_snd_space_inline (tc),
1332  tc->snd_wnd - (tc->snd_nxt - tc->snd_una));
1333 }
1334 
1335 static u32
1337 {
1338  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1339 
1340  ASSERT (seq_geq (tc->snd_nxt, tc->snd_una));
1341 
1342  /* This still works if fast retransmit is on */
1343  return (tc->snd_nxt - tc->snd_una);
1344 }
1345 
1346 static void
1347 tcp_update_time (f64 now, u8 thread_index)
1348 {
1349  tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index);
1350 
1351  tcp_set_time_now (wrk);
1352  tw_timer_expire_timers_16t_2w_512sl (&wrk->timer_wheel, now);
1354 }
1355 
1356 static void
1358 {
1359  tcp_connection_t *tc = (tcp_connection_t *) tconn;
1360  if (tc->flags & TCP_CONN_PSH_PENDING)
1361  return;
1362  tc->flags |= TCP_CONN_PSH_PENDING;
1363  tc->psh_seq = tc->snd_una + transport_max_tx_dequeue (tconn) - 1;
1364 }
1365 
1366 /* *INDENT-OFF* */
1368  .enable = vnet_tcp_enable_disable,
1369  .start_listen = tcp_session_bind,
1370  .stop_listen = tcp_session_unbind,
1371  .push_header = tcp_session_push_header,
1372  .get_connection = tcp_session_get_transport,
1373  .get_listener = tcp_session_get_listener,
1374  .get_half_open = tcp_half_open_session_get_transport,
1375  .connect = tcp_session_open,
1376  .close = tcp_session_close,
1377  .cleanup = tcp_session_cleanup,
1378  .reset = tcp_session_reset,
1379  .send_mss = tcp_session_send_mss,
1380  .send_space = tcp_session_send_space,
1381  .update_time = tcp_update_time,
1382  .tx_fifo_offset = tcp_session_tx_fifo_offset,
1383  .flush_data = tcp_session_flush_data,
1384  .custom_tx = tcp_session_custom_tx,
1385  .format_connection = format_tcp_session,
1386  .format_listener = format_tcp_listener_session,
1387  .format_half_open = format_tcp_half_open_session,
1388  .transport_options = {
1389  .tx_type = TRANSPORT_TX_PEEK,
1390  .service_type = TRANSPORT_SERVICE_VC,
1391  },
1392 };
1393 /* *INDENT-ON* */
1394 
1395 void
1397 {
1398  if (!transport_connection_is_tx_paced (&tc->connection))
1399  return;
1400 
1401  f64 srtt = clib_min ((f64) tc->srtt * TCP_TICK, tc->mrtt_us);
1402 
1403  transport_connection_tx_pacer_update (&tc->connection,
1405  srtt * CLIB_US_TIME_FREQ);
1406 }
1407 
1408 void
1410  u32 start_bucket)
1411 {
1412  f64 srtt = clib_min ((f64) tc->srtt * TCP_TICK, tc->mrtt_us);
1413  transport_connection_tx_pacer_reset (&tc->connection,
1415  start_bucket,
1416  srtt * CLIB_US_TIME_FREQ);
1417 }
1418 
1419 static void
1420 tcp_timer_waitclose_handler (u32 conn_index, u32 thread_index)
1421 {
1422  tcp_connection_t *tc;
1423 
1424  tc = tcp_connection_get (conn_index, thread_index);
1425  if (!tc)
1426  return;
1427 
1428  switch (tc->state)
1429  {
1430  case TCP_STATE_CLOSE_WAIT:
1432  session_transport_closed_notify (&tc->connection);
1433 
1434  if (!(tc->flags & TCP_CONN_FINPNDG))
1435  {
1436  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1437  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
1438  break;
1439  }
1440 
1441  /* Session didn't come back with a close. Send FIN either way
1442  * and switch to LAST_ACK. */
1443  tcp_cong_recovery_off (tc);
1444  /* Make sure we don't try to send unsent data */
1445  tc->snd_nxt = tc->snd_una;
1446  tcp_send_fin (tc);
1447  tcp_connection_set_state (tc, TCP_STATE_LAST_ACK);
1448 
1449  /* Make sure we don't wait in LAST ACK forever */
1450  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.lastack_time);
1451 
1452  /* Don't delete the connection yet */
1453  break;
1454  case TCP_STATE_FIN_WAIT_1:
1456  session_transport_closed_notify (&tc->connection);
1457  if (tc->flags & TCP_CONN_FINPNDG)
1458  {
1459  /* If FIN pending, we haven't sent everything, but we did try.
1460  * Notify session layer that transport is closed. */
1461  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1462  tcp_send_reset (tc);
1463  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
1464  }
1465  else
1466  {
1467  /* We've sent the fin but no progress. Close the connection and
1468  * to make sure everything is flushed, setup a cleanup timer */
1469  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1470  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
1471  }
1472  break;
1473  case TCP_STATE_LAST_ACK:
1474  case TCP_STATE_CLOSING:
1476  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1477  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, tcp_cfg.cleanup_time);
1478  session_transport_closed_notify (&tc->connection);
1479  break;
1480  default:
1481  tcp_connection_del (tc);
1482  break;
1483  }
1484 }
1485 
1486 /* *INDENT-OFF* */
1488 {
1494 };
1495 /* *INDENT-ON* */
1496 
1497 static void
1499 {
1500  u32 thread_index = vlib_get_thread_index ();
1501  u32 connection_index, timer_id;
1502  tcp_connection_t *tc;
1503  int i;
1504 
1505  /*
1506  * Invalidate all timer handles before dispatching. This avoids dangling
1507  * index references to timer wheel pool entries that have been freed.
1508  */
1509  for (i = 0; i < vec_len (expired_timers); i++)
1510  {
1511  connection_index = expired_timers[i] & 0x0FFFFFFF;
1512  timer_id = expired_timers[i] >> 28;
1513 
1514  if (timer_id != TCP_TIMER_RETRANSMIT_SYN)
1515  tc = tcp_connection_get (connection_index, thread_index);
1516  else
1517  tc = tcp_half_open_connection_get (connection_index);
1518 
1519  TCP_EVT (TCP_EVT_TIMER_POP, connection_index, timer_id);
1520 
1521  tc->timers[timer_id] = TCP_TIMER_HANDLE_INVALID;
1522  }
1523 
1524  /*
1525  * Dispatch expired timers
1526  */
1527  for (i = 0; i < vec_len (expired_timers); i++)
1528  {
1529  connection_index = expired_timers[i] & 0x0FFFFFFF;
1530  timer_id = expired_timers[i] >> 28;
1531  (*timer_expiration_handlers[timer_id]) (connection_index, thread_index);
1532  }
1533 }
1534 
1535 static void
1537 {
1539  tw_timer_wheel_16t_2w_512sl_t *tw;
1540  /* *INDENT-OFF* */
1541  foreach_vlib_main (({
1542  tw = &tm->wrk_ctx[ii].timer_wheel;
1543  tw_timer_wheel_init_16t_2w_512sl (tw, tcp_expired_timers_dispatch,
1544  TCP_TIMER_TICK, ~0);
1545  tw->last_run_time = vlib_time_now (vm);
1546  }));
1547  /* *INDENT-ON* */
1548 }
1549 
1550 static void
1552 {
1553  u32 default_seed = random_default_seed ();
1554  u64 time_now = clib_cpu_time_now ();
1555 
1556  tm->iss_seed.first = (u64) random_u32 (&default_seed) << 32;
1557  tm->iss_seed.second = random_u64 (&time_now);
1558 }
1559 
1560 static clib_error_t *
1562 {
1564  u32 num_threads, n_workers, prealloc_conn_per_wrk;
1565  tcp_connection_t *tc __attribute__ ((unused));
1566  tcp_main_t *tm = vnet_get_tcp_main ();
1567  clib_error_t *error = 0;
1568  int thread;
1569 
1570  if ((error = vlib_call_init_function (vm, ip_main_init)))
1571  return error;
1572  if ((error = vlib_call_init_function (vm, ip4_lookup_init)))
1573  return error;
1574  if ((error = vlib_call_init_function (vm, ip6_lookup_init)))
1575  return error;
1576 
1577  /*
1578  * Registrations
1579  */
1580 
1581  ip4_register_protocol (IP_PROTOCOL_TCP, tcp4_input_node.index);
1582  ip6_register_protocol (IP_PROTOCOL_TCP, tcp6_input_node.index);
1583 
1584  /*
1585  * Initialize data structures
1586  */
1587 
1588  num_threads = 1 /* main thread */ + vtm->n_threads;
1589  vec_validate (tm->connections, num_threads - 1);
1590  vec_validate (tm->wrk_ctx, num_threads - 1);
1591  n_workers = num_threads == 1 ? 1 : vtm->n_threads;
1592  prealloc_conn_per_wrk = tcp_cfg.preallocated_connections / n_workers;
1593 
1594  for (thread = 0; thread < num_threads; thread++)
1595  {
1596  vec_validate (tm->wrk_ctx[thread].pending_deq_acked, 255);
1597  vec_validate (tm->wrk_ctx[thread].pending_disconnects, 255);
1598  vec_reset_length (tm->wrk_ctx[thread].pending_deq_acked);
1599  vec_reset_length (tm->wrk_ctx[thread].pending_disconnects);
1600  tm->wrk_ctx[thread].vm = vlib_mains[thread];
1601 
1602  /*
1603  * Preallocate connections. Assume that thread 0 won't
1604  * use preallocated threads when running multi-core
1605  */
1606  if ((thread > 0 || num_threads == 1) && prealloc_conn_per_wrk)
1607  pool_init_fixed (tm->connections[thread], prealloc_conn_per_wrk);
1608  }
1609 
1610  /*
1611  * Use a preallocated half-open connection pool?
1612  */
1613  if (tcp_cfg.preallocated_half_open_connections)
1614  pool_init_fixed (tm->half_open_connections,
1615  tcp_cfg.preallocated_half_open_connections);
1616 
1617  /* Initialize clocks per tick for TCP timestamp. Used to compute
1618  * monotonically increasing timestamps. */
1619  tm->tstamp_ticks_per_clock = vm->clib_time.seconds_per_clock
1621 
1622  if (num_threads > 1)
1623  {
1624  clib_spinlock_init (&tm->half_open_lock);
1625  }
1626 
1629 
1630  tm->bytes_per_buffer = vlib_buffer_get_default_data_size (vm);
1631  tm->cc_last_type = TCP_CC_LAST;
1632  return error;
1633 }
1634 
1635 clib_error_t *
1637 {
1638  if (is_en)
1639  {
1640  if (tcp_main.is_enabled)
1641  return 0;
1642 
1643  return tcp_main_enable (vm);
1644  }
1645  else
1646  {
1647  tcp_main.is_enabled = 0;
1648  }
1649 
1650  return 0;
1651 }
1652 
1653 void
1655 {
1656  tcp_main_t *tm = &tcp_main;
1657  if (is_ip4)
1658  tm->punt_unknown4 = is_add;
1659  else
1660  tm->punt_unknown6 = is_add;
1661 }
1662 
1663 /**
1664  * Initialize default values for tcp parameters
1665  */
1666 static void
1668 {
1669  /* Initial wnd for SYN. Fifos are not allocated at that point so use some
1670  * predefined value. For SYN-ACK we still want the scale to be computed in
1671  * the same way */
1672  tcp_cfg.max_rx_fifo = 32 << 20;
1673  tcp_cfg.min_rx_fifo = 4 << 10;
1674 
1675  tcp_cfg.default_mtu = 1500;
1676  tcp_cfg.initial_cwnd_multiplier = 0;
1677  tcp_cfg.enable_tx_pacing = 1;
1678  tcp_cfg.allow_tso = 0;
1679  tcp_cfg.csum_offload = 1;
1680  tcp_cfg.cc_algo = TCP_CC_NEWRENO;
1681  tcp_cfg.rwnd_min_update_ack = 1;
1682 
1683  /* Time constants defined as timer tick (100ms) multiples */
1684  tcp_cfg.delack_time = 1; /* 0.1s */
1685  tcp_cfg.closewait_time = 20; /* 2s */
1686  tcp_cfg.timewait_time = 100; /* 10s */
1687  tcp_cfg.finwait1_time = 600; /* 60s */
1688  tcp_cfg.lastack_time = 300; /* 30s */
1689  tcp_cfg.finwait2_time = 300; /* 30s */
1690  tcp_cfg.closing_time = 300; /* 30s */
1691  tcp_cfg.cleanup_time = 1; /* 0.1s */
1692 }
1693 
1694 static clib_error_t *
1696 {
1697  tcp_main_t *tm = vnet_get_tcp_main ();
1698  ip_main_t *im = &ip_main;
1699  ip_protocol_info_t *pi;
1700 
1701  /* Session layer, and by implication tcp, are disabled by default */
1702  tm->is_enabled = 0;
1703 
1704  /* Register with IP for header parsing */
1705  pi = ip_get_protocol_info (im, IP_PROTOCOL_TCP);
1706  if (pi == 0)
1707  return clib_error_return (0, "TCP protocol info AWOL");
1710 
1711  /* Register as transport with session layer */
1712  transport_register_protocol (TRANSPORT_PROTO_TCP, &tcp_proto,
1714  transport_register_protocol (TRANSPORT_PROTO_TCP, &tcp_proto,
1716 
1717  tcp_api_reference ();
1719 
1720  tm->cc_algo_by_name = hash_create_string (0, sizeof (uword));
1721 
1722  return 0;
1723 }
1724 
1726 
1727 uword
1729 {
1730  tcp_cc_algorithm_type_e *result = va_arg (*va, tcp_cc_algorithm_type_e *);
1731  tcp_main_t *tm = &tcp_main;
1732  char *cc_algo_name;
1733  u8 found = 0;
1734  uword *p;
1735 
1736  if (unformat (input, "%s", &cc_algo_name)
1737  && ((p = hash_get_mem (tm->cc_algo_by_name, cc_algo_name))))
1738  {
1739  *result = *p;
1740  found = 1;
1741  }
1742 
1743  vec_free (cc_algo_name);
1744  return found;
1745 }
1746 
1747 uword
1749 {
1750  tcp_main_t *tm = vnet_get_tcp_main ();
1751  tcp_cc_algorithm_t *cc_alg;
1752  unformat_input_t sub_input;
1753  int found = 0;
1754 
1755  vec_foreach (cc_alg, tm->cc_algos)
1756  {
1757  if (!unformat (input, cc_alg->name))
1758  continue;
1759 
1760  if (cc_alg->unformat_cfg
1761  && unformat (input, "%U", unformat_vlib_cli_sub_input, &sub_input))
1762  {
1763  if (cc_alg->unformat_cfg (&sub_input))
1764  found = 1;
1765  }
1766  }
1767  return found;
1768 }
1769 
1770 static clib_error_t *
1772 {
1773  u32 cwnd_multiplier, tmp_time;
1775 
1776  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1777  {
1778  if (unformat (input, "preallocated-connections %d",
1779  &tcp_cfg.preallocated_connections))
1780  ;
1781  else if (unformat (input, "preallocated-half-open-connections %d",
1782  &tcp_cfg.preallocated_half_open_connections))
1783  ;
1784  else if (unformat (input, "buffer-fail-fraction %f",
1785  &tcp_cfg.buffer_fail_fraction))
1786  ;
1787  else if (unformat (input, "max-rx-fifo %U", unformat_memory_size,
1788  &memory_size))
1789  {
1790  if (memory_size >= 0x100000000)
1791  {
1792  return clib_error_return
1793  (0, "max-rx-fifo %llu (0x%llx) too large", memory_size,
1794  memory_size);
1795  }
1796  tcp_cfg.max_rx_fifo = memory_size;
1797  }
1798  else if (unformat (input, "min-rx-fifo %U", unformat_memory_size,
1799  &memory_size))
1800  {
1801  if (memory_size >= 0x100000000)
1802  {
1803  return clib_error_return
1804  (0, "min-rx-fifo %llu (0x%llx) too large", memory_size,
1805  memory_size);
1806  }
1807  tcp_cfg.min_rx_fifo = memory_size;
1808  }
1809  else if (unformat (input, "mtu %u", &tcp_cfg.default_mtu))
1810  ;
1811  else if (unformat (input, "rwnd-min-update-ack %d",
1812  &tcp_cfg.rwnd_min_update_ack))
1813  ;
1814  else if (unformat (input, "initial-cwnd-multiplier %u",
1815  &cwnd_multiplier))
1816  tcp_cfg.initial_cwnd_multiplier = cwnd_multiplier;
1817  else if (unformat (input, "no-tx-pacing"))
1818  tcp_cfg.enable_tx_pacing = 0;
1819  else if (unformat (input, "tso"))
1820  tcp_cfg.allow_tso = 1;
1821  else if (unformat (input, "no-csum-offload"))
1822  tcp_cfg.csum_offload = 0;
1823  else if (unformat (input, "cc-algo %U", unformat_tcp_cc_algo,
1824  &tcp_cfg.cc_algo))
1825  ;
1826  else if (unformat (input, "%U", unformat_tcp_cc_algo_cfg))
1827  ;
1828  else if (unformat (input, "closewait-time %u", &tmp_time))
1829  tcp_cfg.closewait_time = tmp_time / TCP_TIMER_TICK;
1830  else if (unformat (input, "timewait-time %u", &tmp_time))
1831  tcp_cfg.timewait_time = tmp_time / TCP_TIMER_TICK;
1832  else if (unformat (input, "finwait1-time %u", &tmp_time))
1833  tcp_cfg.finwait1_time = tmp_time / TCP_TIMER_TICK;
1834  else if (unformat (input, "finwait2-time %u", &tmp_time))
1835  tcp_cfg.finwait2_time = tmp_time / TCP_TIMER_TICK;
1836  else if (unformat (input, "lastack-time %u", &tmp_time))
1837  tcp_cfg.lastack_time = tmp_time / TCP_TIMER_TICK;
1838  else if (unformat (input, "closing-time %u", &tmp_time))
1839  tcp_cfg.closing_time = tmp_time / TCP_TIMER_TICK;
1840  else if (unformat (input, "cleanup-time %u", &tmp_time))
1841  tcp_cfg.cleanup_time = tmp_time / TCP_TIMER_TICK;
1842  else
1843  return clib_error_return (0, "unknown input `%U'",
1844  format_unformat_error, input);
1845  }
1846  return 0;
1847 }
1848 
1850 
1851 
1852 /**
1853  * \brief Configure an ipv4 source address range
1854  * @param vm vlib_main_t pointer
1855  * @param start first ipv4 address in the source address range
1856  * @param end last ipv4 address in the source address range
1857  * @param table_id VRF / table ID, 0 for the default FIB
1858  * @return 0 if all OK, else an error indication from api_errno.h
1859  */
1860 
1861 int
1863  ip4_address_t * start,
1865 {
1866  u32 start_host_byte_order, end_host_byte_order;
1868  fib_node_index_t fei;
1869  u32 fib_index = 0;
1870  u32 sw_if_index;
1871  int rv;
1872 
1873  clib_memset (&prefix, 0, sizeof (prefix));
1874 
1875  fib_index = fib_table_find (FIB_PROTOCOL_IP4, table_id);
1876 
1877  if (fib_index == ~0)
1878  return VNET_API_ERROR_NO_SUCH_FIB;
1879 
1880  start_host_byte_order = clib_net_to_host_u32 (start->as_u32);
1881  end_host_byte_order = clib_net_to_host_u32 (end->as_u32);
1882 
1883  /* sanity check for reversed args or some such */
1884  if ((end_host_byte_order - start_host_byte_order) > (10 << 10))
1885  return VNET_API_ERROR_INVALID_ARGUMENT;
1886 
1887  /* Lookup the last address, to identify the interface involved */
1888  prefix.fp_len = 32;
1889  prefix.fp_proto = FIB_PROTOCOL_IP4;
1890  memcpy (&prefix.fp_addr.ip4, end, sizeof (ip4_address_t));
1891 
1892  fei = fib_table_lookup (fib_index, &prefix);
1893 
1894  /* Couldn't find route to destination. Bail out. */
1895  if (fei == FIB_NODE_INDEX_INVALID)
1896  return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1897 
1898  sw_if_index = fib_entry_get_resolving_interface (fei);
1899 
1900  /* Configure proxy arp across the range */
1901  rv = ip4_neighbor_proxy_add (fib_index, start, end);
1902 
1903  if (rv)
1904  return rv;
1905 
1906  rv = ip4_neighbor_proxy_enable (sw_if_index);
1907 
1908  if (rv)
1909  return rv;
1910 
1911  do
1912  {
1913  dpo_id_t dpo = DPO_INVALID;
1914 
1915  vec_add1 (tcp_cfg.ip4_src_addrs, start[0]);
1916 
1917  /* Add local adjacencies for the range */
1918 
1919  receive_dpo_add_or_lock (DPO_PROTO_IP4, ~0 /* sw_if_index */ ,
1920  NULL, &dpo);
1921  prefix.fp_len = 32;
1922  prefix.fp_proto = FIB_PROTOCOL_IP4;
1923  prefix.fp_addr.ip4.as_u32 = start->as_u32;
1924 
1926  &prefix,
1928  FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
1929  dpo_reset (&dpo);
1930 
1931  start_host_byte_order++;
1932  start->as_u32 = clib_host_to_net_u32 (start_host_byte_order);
1933  }
1934  while (start_host_byte_order <= end_host_byte_order);
1935 
1936  return 0;
1937 }
1938 
1939 /**
1940  * \brief Configure an ipv6 source address range
1941  * @param vm vlib_main_t pointer
1942  * @param start first ipv6 address in the source address range
1943  * @param end last ipv6 address in the source address range
1944  * @param table_id VRF / table ID, 0 for the default FIB
1945  * @return 0 if all OK, else an error indication from api_errno.h
1946  */
1947 
1948 int
1950  ip6_address_t * start,
1952 {
1954  u32 fib_index = 0;
1955  fib_node_index_t fei;
1956  u32 sw_if_index;
1957 
1958  clib_memset (&prefix, 0, sizeof (prefix));
1959 
1960  fib_index = fib_table_find (FIB_PROTOCOL_IP6, table_id);
1961 
1962  if (fib_index == ~0)
1963  return VNET_API_ERROR_NO_SUCH_FIB;
1964 
1965  while (1)
1966  {
1967  int i;
1968  ip6_address_t tmp;
1969  dpo_id_t dpo = DPO_INVALID;
1970 
1971  /* Remember this address */
1972  vec_add1 (tcp_cfg.ip6_src_addrs, start[0]);
1973 
1974  /* Lookup the prefix, to identify the interface involved */
1975  prefix.fp_len = 128;
1976  prefix.fp_proto = FIB_PROTOCOL_IP6;
1977  memcpy (&prefix.fp_addr.ip6, start, sizeof (ip6_address_t));
1978 
1979  fei = fib_table_lookup (fib_index, &prefix);
1980 
1981  /* Couldn't find route to destination. Bail out. */
1982  if (fei == FIB_NODE_INDEX_INVALID)
1983  return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1984 
1985  sw_if_index = fib_entry_get_resolving_interface (fei);
1986 
1987  if (sw_if_index == (u32) ~ 0)
1988  return VNET_API_ERROR_NO_MATCHING_INTERFACE;
1989 
1990  /* Add a proxy neighbor discovery entry for this address */
1991  ip6_neighbor_proxy_add (sw_if_index, start);
1992 
1993  /* Add a receive adjacency for this address */
1994  receive_dpo_add_or_lock (DPO_PROTO_IP6, ~0 /* sw_if_index */ ,
1995  NULL, &dpo);
1996 
1998  &prefix,
2000  FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
2001  dpo_reset (&dpo);
2002 
2003  /* Done with the entire range? */
2004  if (!memcmp (start, end, sizeof (start[0])))
2005  break;
2006 
2007  /* Increment the address. DGMS. */
2008  tmp = start[0];
2009  for (i = 15; i >= 0; i--)
2010  {
2011  tmp.as_u8[i] += 1;
2012  if (tmp.as_u8[i] != 0)
2013  break;
2014  }
2015  start[0] = tmp;
2016  }
2017  return 0;
2018 }
2019 
2020 static clib_error_t *
2022  unformat_input_t * input, vlib_cli_command_t * cmd_arg)
2023 {
2024  ip4_address_t v4start, v4end;
2025  ip6_address_t v6start, v6end;
2026  u32 table_id = 0;
2027  int v4set = 0;
2028  int v6set = 0;
2029  int rv;
2030 
2031  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
2032  {
2033  if (unformat (input, "%U - %U", unformat_ip4_address, &v4start,
2034  unformat_ip4_address, &v4end))
2035  v4set = 1;
2036  else if (unformat (input, "%U", unformat_ip4_address, &v4start))
2037  {
2038  memcpy (&v4end, &v4start, sizeof (v4start));
2039  v4set = 1;
2040  }
2041  else if (unformat (input, "%U - %U", unformat_ip6_address, &v6start,
2042  unformat_ip6_address, &v6end))
2043  v6set = 1;
2044  else if (unformat (input, "%U", unformat_ip6_address, &v6start))
2045  {
2046  memcpy (&v6end, &v6start, sizeof (v6start));
2047  v6set = 1;
2048  }
2049  else if (unformat (input, "fib-table %d", &table_id))
2050  ;
2051  else
2052  break;
2053  }
2054 
2055  if (!v4set && !v6set)
2056  return clib_error_return (0, "at least one v4 or v6 address required");
2057 
2058  if (v4set)
2059  {
2060  rv = tcp_configure_v4_source_address_range (vm, &v4start, &v4end,
2061  table_id);
2062  switch (rv)
2063  {
2064  case 0:
2065  break;
2066 
2067  case VNET_API_ERROR_NO_SUCH_FIB:
2068  return clib_error_return (0, "Invalid table-id %d", table_id);
2069 
2070  case VNET_API_ERROR_INVALID_ARGUMENT:
2071  return clib_error_return (0, "Invalid address range %U - %U",
2072  format_ip4_address, &v4start,
2073  format_ip4_address, &v4end);
2074  default:
2075  return clib_error_return (0, "error %d", rv);
2076  break;
2077  }
2078  }
2079  if (v6set)
2080  {
2081  rv = tcp_configure_v6_source_address_range (vm, &v6start, &v6end,
2082  table_id);
2083  switch (rv)
2084  {
2085  case 0:
2086  break;
2087 
2088  case VNET_API_ERROR_NO_SUCH_FIB:
2089  return clib_error_return (0, "Invalid table-id %d", table_id);
2090 
2091  default:
2092  return clib_error_return (0, "error %d", rv);
2093  break;
2094  }
2095  }
2096  return 0;
2097 }
2098 
2099 /* *INDENT-OFF* */
2100 VLIB_CLI_COMMAND (tcp_src_address_command, static) =
2101 {
2102  .path = "tcp src-address",
2103  .short_help = "tcp src-address <ip-addr> [- <ip-addr>] add src address range",
2104  .function = tcp_src_address,
2105 };
2106 /* *INDENT-ON* */
2107 
2108 static u8 *
2110 {
2111 #if TCP_SCOREBOARD_TRACE
2112 
2113  scoreboard_trace_elt_t *block;
2114  int i = 0;
2115 
2116  if (!sb->trace)
2117  return s;
2118 
2119  s = format (s, "scoreboard trace:");
2120  vec_foreach (block, sb->trace)
2121  {
2122  s = format (s, "{%u, %u, %u, %u, %u}, ", block->start, block->end,
2123  block->ack, block->snd_una_max, block->group);
2124  if ((++i % 3) == 0)
2125  s = format (s, "\n");
2126  }
2127  return s;
2128 #else
2129  return 0;
2130 #endif
2131 }
2132 
2133 static clib_error_t *
2135  vlib_cli_command_t * cmd_arg)
2136 {
2137  transport_connection_t *tconn = 0;
2138  tcp_connection_t *tc;
2139  u8 *s = 0;
2140  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
2141  {
2142  if (unformat (input, "%U", unformat_transport_connection, &tconn,
2143  TRANSPORT_PROTO_TCP))
2144  ;
2145  else
2146  return clib_error_return (0, "unknown input `%U'",
2147  format_unformat_error, input);
2148  }
2149 
2150  if (!TCP_SCOREBOARD_TRACE)
2151  {
2152  vlib_cli_output (vm, "scoreboard tracing not enabled");
2153  return 0;
2154  }
2155 
2156  tc = tcp_get_connection_from_transport (tconn);
2157  s = tcp_scoreboard_dump_trace (s, &tc->sack_sb);
2158  vlib_cli_output (vm, "%v", s);
2159  return 0;
2160 }
2161 
2162 /* *INDENT-OFF* */
2163 VLIB_CLI_COMMAND (tcp_show_scoreboard_trace_command, static) =
2164 {
2165  .path = "show tcp scoreboard trace",
2166  .short_help = "show tcp scoreboard trace <connection>",
2167  .function = tcp_show_scoreboard_trace_fn,
2168 };
2169 /* *INDENT-ON* */
2170 
2171 u8 *
2173 {
2174  int i, trace_len;
2176  u32 next_ack, left, group, has_new_ack = 0;
2177  tcp_connection_t _dummy_tc, *dummy_tc = &_dummy_tc;
2178  sack_block_t *block;
2179 
2180  if (!TCP_SCOREBOARD_TRACE)
2181  {
2182  s = format (s, "scoreboard tracing not enabled");
2183  return s;
2184  }
2185 
2186  if (!tc)
2187  return s;
2188 
2189  clib_memset (dummy_tc, 0, sizeof (*dummy_tc));
2190  tcp_connection_timers_init (dummy_tc);
2191  scoreboard_init (&dummy_tc->sack_sb);
2192  dummy_tc->rcv_opts.flags |= TCP_OPTS_FLAG_SACK;
2193 
2194 #if TCP_SCOREBOARD_TRACE
2195  trace = tc->sack_sb.trace;
2196  trace_len = vec_len (tc->sack_sb.trace);
2197 #endif
2198 
2199  for (i = 0; i < trace_len; i++)
2200  {
2201  if (trace[i].ack != 0)
2202  {
2203  dummy_tc->snd_una = trace[i].ack - 1448;
2204  dummy_tc->snd_una_max = trace[i].ack;
2205  }
2206  }
2207 
2208  left = 0;
2209  while (left < trace_len)
2210  {
2211  group = trace[left].group;
2212  vec_reset_length (dummy_tc->rcv_opts.sacks);
2213  has_new_ack = 0;
2214  while (trace[left].group == group)
2215  {
2216  if (trace[left].ack != 0)
2217  {
2218  if (verbose)
2219  s = format (s, "Adding ack %u, snd_una_max %u, segs: ",
2220  trace[left].ack, trace[left].snd_una_max);
2221  dummy_tc->snd_una_max = trace[left].snd_una_max;
2222  next_ack = trace[left].ack;
2223  has_new_ack = 1;
2224  }
2225  else
2226  {
2227  if (verbose)
2228  s = format (s, "[%u, %u], ", trace[left].start,
2229  trace[left].end);
2230  vec_add2 (dummy_tc->rcv_opts.sacks, block, 1);
2231  block->start = trace[left].start;
2232  block->end = trace[left].end;
2233  }
2234  left++;
2235  }
2236 
2237  /* Push segments */
2238  tcp_rcv_sacks (dummy_tc, next_ack);
2239  if (has_new_ack)
2240  dummy_tc->snd_una = next_ack;
2241 
2242  if (verbose)
2243  s = format (s, "result: %U", format_tcp_scoreboard,
2244  &dummy_tc->sack_sb);
2245 
2246  }
2247  s = format (s, "result: %U", format_tcp_scoreboard, &dummy_tc->sack_sb);
2248 
2249  return s;
2250 }
2251 
2252 static clib_error_t *
2254  vlib_cli_command_t * cmd_arg)
2255 {
2256  transport_connection_t *tconn = 0;
2257  tcp_connection_t *tc = 0;
2258  u8 *str = 0;
2259  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
2260  {
2261  if (unformat (input, "%U", unformat_transport_connection, &tconn,
2262  TRANSPORT_PROTO_TCP))
2263  ;
2264  else
2265  return clib_error_return (0, "unknown input `%U'",
2266  format_unformat_error, input);
2267  }
2268 
2269  if (!TCP_SCOREBOARD_TRACE)
2270  {
2271  vlib_cli_output (vm, "scoreboard tracing not enabled");
2272  return 0;
2273  }
2274 
2275  tc = tcp_get_connection_from_transport (tconn);
2276  if (!tc)
2277  {
2278  vlib_cli_output (vm, "connection not found");
2279  return 0;
2280  }
2281  str = tcp_scoreboard_replay (str, tc, 1);
2282  vlib_cli_output (vm, "%v", str);
2283  return 0;
2284 }
2285 
2286 /* *INDENT-OFF* */
2287 VLIB_CLI_COMMAND (tcp_replay_scoreboard_command, static) =
2288 {
2289  .path = "tcp replay scoreboard",
2290  .short_help = "tcp replay scoreboard <connection>",
2291  .function = tcp_scoreboard_trace_fn,
2292 };
2293 /* *INDENT-ON* */
2294 
2295 static clib_error_t *
2297  vlib_cli_command_t * cmd_arg)
2298 {
2299  tcp_main_t *tm = vnet_get_tcp_main ();
2301  return clib_error_return (0, "unknown input `%U'", format_unformat_error,
2302  input);
2303  vlib_cli_output (vm, "IPv4 TCP punt: %s",
2304  tm->punt_unknown4 ? "enabled" : "disabled");
2305  vlib_cli_output (vm, "IPv6 TCP punt: %s",
2306  tm->punt_unknown6 ? "enabled" : "disabled");
2307  return 0;
2308 }
2309 /* *INDENT-OFF* */
2310 VLIB_CLI_COMMAND (show_tcp_punt_command, static) =
2311 {
2312  .path = "show tcp punt",
2313  .short_help = "show tcp punt",
2314  .function = show_tcp_punt_fn,
2315 };
2316 /* *INDENT-ON* */
2317 
2318 /*
2319  * fd.io coding-style-patch-verification: ON
2320  *
2321  * Local Variables:
2322  * eval: (c-set-style "gnu")
2323  * End:
2324  */
static void tcp_session_close(u32 conn_index, u32 thread_index)
Definition: tcp.c:467
#define vec_validate(V, I)
Make sure vector is long enough for given index (no header, unspecified alignment) ...
Definition: vec.h:440
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
void() timer_expiration_handler(u32 index, u32 thread_index)
Definition: tcp.h:85
#define ENDPOINT_INVALID_INDEX
static vlib_cli_command_t trace
(constructor) VLIB_CLI_COMMAND (trace)
Definition: vlib_api_cli.c:898
static void tcp_add_del_adjacency(tcp_connection_t *tc, u8 is_add)
Definition: tcp.c:59
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:305
#define clib_min(x, y)
Definition: clib.h:295
static u8 * format_tcp_listener_session(u8 *s, va_list *args)
Definition: tcp.c:1099
const char * tcp_fsm_states[]
Definition: tcp.c:832
static clib_error_t * tcp_config_fn(vlib_main_t *vm, unformat_input_t *input)
Definition: tcp.c:1771
#define tcp_in_recovery(tc)
Definition: tcp.h:465
static f64 tcp_time_now_us(u32 thread_index)
Definition: tcp.h:1028
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:874
struct _sack_block sack_block_t
u8 * format_tcp_rcv_sacks(u8 *s, va_list *args)
Definition: tcp.c:1145
void ip6_register_protocol(u32 protocol, u32 node_index)
Definition: ip6_forward.c:1590
struct _scoreboard_trace_elt scoreboard_trace_elt_t
u8 * format_tcp_sacks(u8 *s, va_list *args)
Definition: tcp.c:1121
static clib_error_t * ip4_lookup_init(vlib_main_t *vm)
Definition: ip4_forward.c:1083
const char * tcp_connection_flags_str[]
Definition: tcp.c:875
vlib_node_registration_t tcp4_output_node
(constructor) VLIB_REGISTER_NODE (tcp4_output_node)
Definition: tcp_output.c:2456
void scoreboard_init(sack_scoreboard_t *sb)
Definition: tcp_input.c:919
uword unformat_transport_connection(unformat_input_t *input, va_list *args)
Definition: session_cli.c:268
#define tcp_opts_tstamp(_to)
Definition: tcp_packet.h:156
static u32 tcp_session_send_space(transport_connection_t *trans_conn)
Definition: tcp.c:1328
static transport_connection_t * tcp_half_open_session_get_transport(u32 conn_index)
Definition: tcp.c:1224
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:1336
#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:304
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:880
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:755
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:221
static u8 * format_tcp_timers(u8 *s, va_list *args)
Definition: tcp.c:907
static tcp_connection_t * tcp_half_open_connection_get(u32 conn_index)
Definition: tcp.h:777
sack_scoreboard_hole_t * scoreboard_next_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:657
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:519
vlib_node_registration_t tcp6_output_node
(constructor) VLIB_REGISTER_NODE (tcp6_output_node)
Definition: tcp_output.c:2476
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:523
static u64 clib_cpu_time_now(void)
Definition: time.h:75
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:1129
void session_transport_reset_notify(transport_connection_t *tc)
Notify application that connection has been reset.
Definition: session.c:981
#define vec_add2(V, P, N)
Add N elements to end of vector V, return pointer to new elements in P.
Definition: vec.h:561
int i
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:977
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:1249
Definition: ip.h:106
static tcp_connection_t * tcp_get_connection_from_transport(transport_connection_t *tconn)
Definition: tcp.h:736
vl_api_prefix_t prefix
Definition: ip.api:144
static void tcp_initialize_iss_seed(tcp_main_t *tm)
Definition: tcp.c:1551
void ip4_register_protocol(u32 protocol, u32 node_index)
Definition: ip4_forward.c:1866
static void tcp_initialize_timer_wheels(tcp_main_t *tm)
Definition: tcp.c:1536
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:237
struct _tcp_header tcp_header_t
fib_protocol_t nh_proto
Definition: tcp.c:33
int tcp_half_open_connection_cleanup(tcp_connection_t *tc)
Try to cleanup half-open connection.
Definition: tcp.c:210
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:190
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:97
int ip4_neighbor_proxy_add(u32 fib_index, const ip4_address_t *start, const ip4_address_t *end)
The set of function that vnet requires from the IP neighbour module.
Definition: ip_neighbor.c:1030
#define TCP_RTO_INIT
Definition: tcp.h:103
format_function_t format_ip4_address
Definition: format.h:73
#define TCP_TICK
TCP tick period (s)
Definition: tcp.h:27
unformat_function_t * unformat_pg_edit
Definition: ip.h:87
#define seq_gt(_s1, _s2)
Definition: tcp.h:875
static void tcp_connection_set_state(tcp_connection_t *tc, tcp_state_t state)
Definition: tcp.h:742
unformat_function_t unformat_ip4_address
Definition: format.h:68
void tcp_init_snd_vars(tcp_connection_t *tc)
Initialize connection send variables.
Definition: tcp.c:691
#define tcp_cfg
Definition: tcp.h:679
vl_api_interface_index_t sw_if_index
Definition: gre.api:59
u8 * format_tcp_connection_id(u8 *s, va_list *args)
Definition: tcp.c:1040
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:173
static u8 * format_tcp_vars(u8 *s, va_list *args)
Definition: tcp.c:996
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:502
static u32 tcp_available_output_snd_space(const tcp_connection_t *tc)
Definition: tcp.h:962
u8 * format_white_space(u8 *s, va_list *va)
Definition: std-formats.c:129
tcp_main_t tcp_main
Definition: tcp.c:29
static void tcp_connection_unbind(u32 listener_index)
Definition: tcp.c:153
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:324
static u8 * tcp_scoreboard_dump_trace(u8 *s, sack_scoreboard_t *sb)
Definition: tcp.c:2109
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:526
static u32 tcp_time_now(void)
Definition: tcp.h:1006
u32 tcp_snd_space(tcp_connection_t *tc)
Definition: tcp.c:1322
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:1097
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:119
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:68
static u32 tcp_connection_bind(u32 session_index, transport_endpoint_t *lcl)
Definition: tcp.c:112
static i32 tcp_rcv_wnd_available(tcp_connection_t *tc)
Definition: tcp.c:942
#define hash_create_string(elts, value_bytes)
Definition: hash.h:690
static void tcp_cc_init(tcp_connection_t *tc)
Definition: tcp.c:73
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:1112
uword unformat_tcp_cc_algo_cfg(unformat_input_t *input, va_list *va)
Definition: tcp.c:1748
vl_api_fib_path_type_t type
Definition: fib_types.api:123
static u32 tcp_session_unbind(u32 listener_index)
Definition: tcp.c:170
static timer_expiration_handler * timer_expiration_handlers[TCP_N_TIMERS]
Definition: tcp.c:1487
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:1561
Definition: fib_entry.h:116
#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:519
void tcp_bt_init(tcp_connection_t *tc)
Byte tracker initialize.
Definition: tcp_bt.c:650
timer_expiration_handler tcp_timer_persist_handler
#define CLIB_US_TIME_FREQ
Definition: time.h:201
static transport_connection_t * tcp_session_get_transport(u32 conn_index, u32 thread_index)
Definition: tcp.c:1215
static ip_protocol_info_t * ip_get_protocol_info(ip_main_t *im, u32 protocol)
Definition: ip.h:133
#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:1409
int tcp_fastrecovery_prr_snd_space(tcp_connection_t *tc)
Estimate send space using proportional rate reduction (RFC6937)
Definition: tcp_output.c:1798
dpo_type_t dpoi_type
the type
Definition: dpo.h:174
format_function_t * format_header
Definition: ip.h:78
tcp_connection_t * tcp_connection_alloc_w_base(u8 thread_index, tcp_connection_t *base)
Definition: tcp.c:311
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:981
static void tcp_init_rcv_mss(tcp_connection_t *tc)
Initialize max segment size we&#39;re able to process.
Definition: tcp.c:654
void tcp_send_syn(tcp_connection_t *tc)
Send SYN.
Definition: tcp_output.c:919
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:298
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:177
static u32 transport_max_rx_dequeue(transport_connection_t *tc)
Definition: session.h:485
void tcp_connection_tx_pacer_update(tcp_connection_t *tc)
Definition: tcp.c:1396
static u32 ip6_compute_flow_hash(const ip6_header_t *ip, flow_hash_config_t flow_hash_config)
Definition: ip6.h:380
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:287
static void tcp_expired_timers_dispatch(u32 *expired_timers)
Definition: tcp.c:1498
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:679
#define TCP_TIMER_HANDLE_INVALID
Definition: tcp.h:92
static void tcp_session_cleanup(u32 conn_index, u32 thread_index)
Definition: tcp.c:475
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:900
#define PREDICT_FALSE(x)
Definition: clib.h:111
#define VLIB_CONFIG_FUNCTION(x, n,...)
Definition: init.h:182
#define always_inline
Definition: ipsec.h:28
static u8 * format_tcp_stats(u8 *s, va_list *args)
Definition: tcp.c:975
u8 * format_tcp_scoreboard(u8 *s, va_list *args)
Definition: tcp.c:1181
clib_error_t * vnet_tcp_enable_disable(vlib_main_t *vm, u8 is_en)
Definition: tcp.c:1636
f64 seconds_per_clock
Definition: time.h:57
static u8 transport_connection_is_tx_paced(transport_connection_t *tc)
Check if transport connection is paced.
Definition: transport.h:246
void transport_connection_tx_pacer_reset(transport_connection_t *tc, u64 rate_bytes_per_sec, u32 start_bucket, clib_us_time_t rtt)
Definition: transport.c:662
static u8 * format_tcp_cfg_flags(u8 *s, va_list *args)
Definition: tcp.c:857
#define foreach_vlib_main(body)
Definition: threads.h:241
static void tcp_timer_waitclose_handler(u32 conn_index, u32 thread_index)
Definition: tcp.c:1420
#define TCP_MAX_GSO_SZ
Definition: tcp.h:34
#define TCP_OPTION_LEN_TIMESTAMP
Definition: tcp_packet.h:167
vlib_main_t * vm
Definition: in2out_ed.c:1810
const char * tcp_cfg_flags_str[]
Definition: tcp.c:850
#define TCP_DBG(_fmt, _args...)
Definition: tcp_debug.h:146
u8 len
Definition: ip_types.api:91
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:1123
void tcp_connection_free(tcp_connection_t *tc)
Definition: tcp.c:324
unformat_function_t unformat_ip6_address
Definition: format.h:89
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:412
u8 is_ip4
Definition: lisp_gpe.api:232
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:363
#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:1862
static_always_inline uword vlib_get_thread_index(void)
Definition: threads.h:218
format_function_t format_ip6_address
Definition: format.h:91
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:41
static clib_error_t * tcp_src_address(vlib_main_t *vm, unformat_input_t *input, vlib_cli_command_t *cmd_arg)
Definition: tcp.c:2021
void tcp_connection_close(tcp_connection_t *tc)
Begin connection closing procedure.
Definition: tcp.c:401
void tcp_send_reset(tcp_connection_t *tc)
Build and set reset packet for connection.
Definition: tcp_output.c:863
#define vec_free(V)
Free vector&#39;s memory (no header).
Definition: vec.h:342
virtual circuit service
static void tcp_session_flush_data(transport_connection_t *tconn)
Definition: tcp.c:1357
#define clib_warning(format, args...)
Definition: error.h:59
static u8 * format_tcp_congestion(u8 *s, va_list *args)
Definition: tcp.c:948
format_function_t format_tcp_header
Definition: format.h:100
struct _transport_connection transport_connection_t
static u8 * format_tcp_connection_flags(u8 *s, va_list *args)
Definition: tcp.c:882
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:1728
vlib_node_registration_t tcp6_input_node
(constructor) VLIB_REGISTER_NODE (tcp6_input_node)
Definition: tcp_input.c:3806
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:51
#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:395
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:239
#define VLIB_CLI_COMMAND(x,...)
Definition: cli.h:152
static u32 transport_max_tx_dequeue(transport_connection_t *tc)
Definition: session.h:478
static void tcp_timer_update(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:1136
signed int i32
Definition: types.h:77
#define pool_put_index(p, i)
Free pool element with given index.
Definition: pool.h:316
#define ASSERT(truth)
static int tcp_session_open(transport_endpoint_cfg_t *rmt)
Definition: tcp.c:785
tcp_cc_algorithm_type_e tcp_cc_algo_new_type(const tcp_cc_algorithm_t *vft)
Register new cc algo type.
Definition: tcp.c:104
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:2253
unformat_function_t unformat_pg_tcp_header
Definition: format.h:102
static void tcp_configuration_init(void)
Initialize default values for tcp parameters.
Definition: tcp.c:1667
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:997
static load_balance_t * load_balance_get(index_t lbi)
Definition: load_balance.h:220
#define seq_geq(_s1, _s2)
Definition: tcp.h:876
static void tcp_init_mss(tcp_connection_t *tc)
Definition: tcp.c:663
int ip6_neighbor_proxy_add(u32 sw_if_index, const ip6_address_t *addr)
Definition: ip_neighbor.c:1077
From the control plane API.
Definition: fib_source.h:72
static u8 * format_tcp_congestion_status(u8 *s, va_list *args)
Definition: tcp.c:929
static u32 tcp_generate_random_iss(tcp_connection_t *tc)
Generate random iss as per rfc6528.
Definition: tcp.c:629
reliable transport protos
static uword ip6_address_is_link_local_unicast(const ip6_address_t *a)
Definition: ip6_packet.h:250
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:86
u8 * tcp_scoreboard_replay(u8 *s, tcp_connection_t *tc, u8 verbose)
Definition: tcp.c:2172
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:1949
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:1012
static clib_error_t * tcp_init(vlib_main_t *vm)
Definition: tcp.c:1695
void transport_connection_tx_pacer_update(transport_connection_t *tc, u64 bytes_per_sec, clib_us_time_t rtt)
Update tx pacer pacing rate.
Definition: transport.c:689
dpo_id_t fe_lb
The load-balance used for forwarding.
Definition: fib_entry.h:331
int ip4_neighbor_proxy_enable(u32 sw_if_index)
Definition: ip_neighbor.c:1057
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:1367
#define seq_lt(_s1, _s2)
Definition: tcp.h:873
void tcp_punt_unknown(vlib_main_t *vm, u8 is_ip4, u8 is_add)
Definition: tcp.c:1654
void tcp_connection_reset(tcp_connection_t *tc)
Notify session that connection has been reset.
Definition: tcp.c:342
vlib_node_registration_t tcp4_input_node
(constructor) VLIB_REGISTER_NODE (tcp4_input_node)
Definition: tcp_input.c:3786
static u32 tcp_session_bind(u32 session_index, transport_endpoint_t *tep)
Definition: tcp.c:147
index_t dpoi_index
the index of objects of that type
Definition: dpo.h:186
vl_api_address_t ip
Definition: l2.api:490
ip46_address_t ip
Definition: tcp.c:35
#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:1007
#define foreach_tcp_connection_flag
TCP connection flags.
Definition: tcp.h:131
#define foreach_tcp_timer
TCP timers.
Definition: tcp.h:70
#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:1873
static void tcp_cc_cleanup(tcp_connection_t *tc)
Definition: tcp.c:79
uword unformat_vlib_cli_sub_input(unformat_input_t *i, va_list *args)
Definition: cli.c:162
sack_scoreboard_hole_t * scoreboard_first_hole(sack_scoreboard_t *sb)
Definition: tcp_input.c:673
static u32 tcp_round_snd_space(tcp_connection_t *tc, u32 snd_space)
Definition: tcp.c:1265
static tcp_worker_ctx_t * tcp_get_worker(u32 thread_index)
Definition: tcp.h:690
void session_transport_closed_notify(transport_connection_t *tc)
Notification from transport that it is closed.
Definition: session.c:946
u64 uword
Definition: types.h:112
void tcp_connection_init_vars(tcp_connection_t *tc)
Initialize tcp connection variables.
Definition: tcp.c:726
#define DPO_INVALID
An initialiser for DPOs declared on the stack.
Definition: dpo.h:197
const char * tcp_conn_timers[]
Definition: tcp.c:900
static u8 * format_tcp_session(u8 *s, va_list *args)
Definition: tcp.c:1083
u16 port
Definition: lb_types.api:72
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:1064
static u8 * format_tcp_sack_hole(u8 *s, va_list *args)
Definition: tcp.c:1169
vnet_link_t link_type
Definition: tcp.c:34
static tcp_connection_t * tcp_connection_get(u32 conn_index, u32 thread_index)
Definition: tcp.h:717
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:757
static u16 tcp_session_cal_goal_size(tcp_connection_t *tc)
Definition: tcp.c:1231
static vlib_thread_main_t * vlib_get_thread_main()
Definition: global_funcs.h:32
vl_api_dhcp_client_state_t state
Definition: dhcp.api:201
void tcp_connection_cleanup(tcp_connection_t *tc)
Cleans up connection state.
Definition: tcp.c:238
void tcp_connection_del(tcp_connection_t *tc)
Connection removal.
Definition: tcp.c:291
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.
f64 end
end of the time range
Definition: mactime.api:44
#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:1112
static u32 tcp_snd_space_inline(tcp_connection_t *tc)
Compute tx window session is allowed to fill.
Definition: tcp.c:1292
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:429
u8 * format_tcp_state(u8 *s, va_list *args)
Definition: tcp.c:839
static clib_error_t * ip6_lookup_init(vlib_main_t *vm)
Definition: ip6_forward.c:2707
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:2296
u32 table_id
Definition: fib_types.api:118
static void tcp_update_time(f64 now, u8 thread_index)
Definition: tcp.c:1347
void tcp_enable_pacing(tcp_connection_t *tc)
Definition: tcp.c:713
static u64 tcp_cc_get_pacing_rate(tcp_connection_t *tc)
Definition: tcp.h:1099
static tcp_main_t * vnet_get_tcp_main()
Definition: tcp.h:684
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:486
void vlib_cli_output(vlib_main_t *vm, char *fmt,...)
Definition: cli.c:689
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:1034
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:2134
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:768
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:2157
static uword pool_elts(void *v)
Number of active elements in a pool.
Definition: pool.h:128