FD.io VPP  v18.10-34-gcce845e
Vector Packet Processing
tcp.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14  */
15 
16 /**
17  * @file
18  * @brief TCP host stack utilities
19  */
20 
21 #include <vnet/tcp/tcp.h>
22 #include <vnet/session/session.h>
23 #include <vnet/fib/fib.h>
24 #include <vnet/dpo/load_balance.h>
25 #include <vnet/dpo/receive_dpo.h>
26 #include <vnet/ip/ip6_neighbor.h>
27 #include <math.h>
28 
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 u32
74 {
75  tcp_main_t *tm = &tcp_main;
76  tcp_connection_t *listener;
77  void *iface_ip;
78 
79  pool_get (tm->listener_pool, listener);
80  memset (listener, 0, sizeof (*listener));
81 
82  listener->c_c_index = listener - tm->listener_pool;
83  listener->c_lcl_port = lcl->port;
84 
85  /* If we are provided a sw_if_index, bind using one of its ips */
86  if (ip_is_zero (&lcl->ip, 1) && lcl->sw_if_index != ENDPOINT_INVALID_INDEX)
87  {
88  if ((iface_ip = ip_interface_get_first_ip (lcl->sw_if_index,
89  lcl->is_ip4)))
90  ip_set (&lcl->ip, iface_ip, lcl->is_ip4);
91  }
92  ip_copy (&listener->c_lcl_ip, &lcl->ip, lcl->is_ip4);
93  listener->c_is_ip4 = lcl->is_ip4;
94  listener->c_proto = TRANSPORT_PROTO_TCP;
95  listener->c_s_index = session_index;
96  listener->c_fib_index = lcl->fib_index;
97  listener->state = TCP_STATE_LISTEN;
98 
99  tcp_connection_timers_init (listener);
100 
101  TCP_EVT_DBG (TCP_EVT_BIND, listener);
102 
103  return listener->c_c_index;
104 }
105 
106 static u32
108 {
109  return tcp_connection_bind (session_index, tep);
110 }
111 
112 static void
113 tcp_connection_unbind (u32 listener_index)
114 {
115  tcp_main_t *tm = vnet_get_tcp_main ();
116  tcp_connection_t *tc;
117 
118  tc = pool_elt_at_index (tm->listener_pool, listener_index);
119 
120  TCP_EVT_DBG (TCP_EVT_UNBIND, tc);
121 
122  /* Poison the entry */
123  if (CLIB_DEBUG > 0)
124  memset (tc, 0xFA, sizeof (*tc));
125 
126  pool_put_index (tm->listener_pool, listener_index);
127 }
128 
129 static u32
130 tcp_session_unbind (u32 listener_index)
131 {
132  tcp_connection_unbind (listener_index);
133  return 0;
134 }
135 
136 static transport_connection_t *
138 {
139  tcp_main_t *tm = vnet_get_tcp_main ();
140  tcp_connection_t *tc;
141  tc = pool_elt_at_index (tm->listener_pool, listener_index);
142  return &tc->connection;
143 }
144 
145 /**
146  * Cleanup half-open connection
147  *
148  */
149 static void
151 {
152  tcp_main_t *tm = vnet_get_tcp_main ();
153  clib_spinlock_lock_if_init (&tm->half_open_lock);
154  pool_put_index (tm->half_open_connections, tc->c_c_index);
155  if (CLIB_DEBUG)
156  memset (tc, 0xFA, sizeof (*tc));
157  clib_spinlock_unlock_if_init (&tm->half_open_lock);
158 }
159 
160 /**
161  * Try to cleanup half-open connection
162  *
163  * If called from a thread that doesn't own tc, the call won't have any
164  * effect.
165  *
166  * @param tc - connection to be cleaned up
167  * @return non-zero if cleanup failed.
168  */
169 int
171 {
172  /* Make sure this is the owning thread */
173  if (tc->c_thread_index != vlib_get_thread_index ())
174  return 1;
175  tcp_timer_reset (tc, TCP_TIMER_ESTABLISH);
176  tcp_timer_reset (tc, TCP_TIMER_RETRANSMIT_SYN);
178  return 0;
179 }
180 
181 static tcp_connection_t *
183 {
184  tcp_main_t *tm = vnet_get_tcp_main ();
185  tcp_connection_t *tc = 0;
186  ASSERT (vlib_get_thread_index () == 0);
187  pool_get (tm->half_open_connections, tc);
188  memset (tc, 0, sizeof (*tc));
189  tc->c_c_index = tc - tm->half_open_connections;
190  return tc;
191 }
192 
193 /**
194  * Cleans up connection state.
195  *
196  * No notifications.
197  */
198 void
200 {
201  tcp_main_t *tm = &tcp_main;
202 
203  /* Cleanup local endpoint if this was an active connect */
205  tc->c_lcl_port);
206 
207  /* Check if connection is not yet fully established */
208  if (tc->state == TCP_STATE_SYN_SENT)
209  {
210  /* Try to remove the half-open connection. If this is not the owning
211  * thread, tc won't be removed. Retransmit or establish timers will
212  * eventually expire and call again cleanup on the right thread. */
214  tc->flags |= TCP_CONN_HALF_OPEN_DONE;
215  }
216  else
217  {
218  int thread_index = tc->c_thread_index;
219 
220  /* Make sure all timers are cleared */
222 
223  if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6))
224  tcp_add_del_adjacency (tc, 0);
225 
226  /* Poison the entry */
227  if (CLIB_DEBUG > 0)
228  memset (tc, 0xFA, sizeof (*tc));
229  pool_put (tm->connections[thread_index], tc);
230  }
231 }
232 
233 /**
234  * Connection removal.
235  *
236  * This should be called only once connection enters CLOSED state. Note
237  * that it notifies the session of the removal event, so if the goal is to
238  * just remove the connection, call tcp_connection_cleanup instead.
239  */
240 void
242 {
243  TCP_EVT_DBG (TCP_EVT_DELETE, tc);
244  stream_session_delete_notify (&tc->connection);
246 }
247 
249 tcp_connection_new (u8 thread_index)
250 {
251  tcp_main_t *tm = vnet_get_tcp_main ();
252  tcp_connection_t *tc;
253 
254  pool_get (tm->connections[thread_index], tc);
255  memset (tc, 0, sizeof (*tc));
256  tc->c_c_index = tc - tm->connections[thread_index];
257  tc->c_thread_index = thread_index;
258  return tc;
259 }
260 
261 /** Notify session that connection has been reset.
262  *
263  * Switch state to closed and wait for session to call cleanup.
264  */
265 void
267 {
268  TCP_EVT_DBG (TCP_EVT_RST_RCVD, tc);
269  switch (tc->state)
270  {
271  case TCP_STATE_SYN_RCVD:
272  /* Cleanup everything. App wasn't notified yet */
273  stream_session_delete_notify (&tc->connection);
275  break;
276  case TCP_STATE_SYN_SENT:
277  session_stream_connect_notify (&tc->connection, 1 /* fail */ );
279  break;
280  case TCP_STATE_ESTABLISHED:
282  /* Set the cleanup timer, in case the session layer/app don't
283  * cleanly close the connection */
284  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
285  stream_session_reset_notify (&tc->connection);
286  break;
287  case TCP_STATE_CLOSE_WAIT:
288  case TCP_STATE_FIN_WAIT_1:
289  case TCP_STATE_FIN_WAIT_2:
290  case TCP_STATE_CLOSING:
291  tc->state = TCP_STATE_CLOSED;
292  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc);
294  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
295  break;
296  case TCP_STATE_CLOSED:
297  return;
298  }
299 }
300 
301 /**
302  * Begin connection closing procedure.
303  *
304  * If at the end the connection is not in CLOSED state, it is not removed.
305  * Instead, we rely on on TCP to advance through state machine to either
306  * 1) LAST_ACK (passive close) whereby when the last ACK is received
307  * tcp_connection_del is called. This notifies session of the delete and
308  * calls cleanup.
309  * 2) TIME_WAIT (active close) whereby after 2MSL the 2MSL timer triggers
310  * and cleanup is called.
311  *
312  * N.B. Half-close connections are not supported
313  */
314 void
316 {
317  TCP_EVT_DBG (TCP_EVT_CLOSE, tc);
318 
319  /* Send/Program FIN if needed and switch state */
320  switch (tc->state)
321  {
322  case TCP_STATE_SYN_SENT:
323  tc->state = TCP_STATE_CLOSED;
324  break;
325  case TCP_STATE_SYN_RCVD:
327  tcp_send_fin (tc);
328  tc->state = TCP_STATE_FIN_WAIT_1;
329  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
330  break;
331  case TCP_STATE_ESTABLISHED:
332  if (!session_tx_fifo_max_dequeue (&tc->connection))
333  tcp_send_fin (tc);
334  else
335  tc->flags |= TCP_CONN_FINPNDG;
336  tc->state = TCP_STATE_FIN_WAIT_1;
337  break;
338  case TCP_STATE_CLOSE_WAIT:
339  if (!session_tx_fifo_max_dequeue (&tc->connection))
340  {
341  tcp_send_fin (tc);
343  tc->state = TCP_STATE_LAST_ACK;
344  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
345  }
346  else
347  tc->flags |= TCP_CONN_FINPNDG;
348  break;
349  case TCP_STATE_FIN_WAIT_1:
350  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
351  break;
352  default:
353  TCP_DBG ("state: %u", tc->state);
354  }
355 
356  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc);
357 
358  /* If in CLOSED and WAITCLOSE timer is not set, delete connection now */
359  if (!tcp_timer_is_active (tc, TCP_TIMER_WAITCLOSE)
360  && tc->state == TCP_STATE_CLOSED)
361  tcp_connection_del (tc);
362 }
363 
364 static void
365 tcp_session_close (u32 conn_index, u32 thread_index)
366 {
367  tcp_connection_t *tc;
368  tc = tcp_connection_get (conn_index, thread_index);
370 }
371 
372 static void
373 tcp_session_cleanup (u32 conn_index, u32 thread_index)
374 {
375  tcp_connection_t *tc;
376  tc = tcp_connection_get (conn_index, thread_index);
378  tc->state = TCP_STATE_CLOSED;
379  TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc);
381 }
382 
383 /**
384  * Initialize all connection timers as invalid
385  */
386 void
388 {
389  int i;
390 
391  /* Set all to invalid */
392  for (i = 0; i < TCP_N_TIMERS; i++)
393  {
394  tc->timers[i] = TCP_TIMER_HANDLE_INVALID;
395  }
396 
397  tc->rto = TCP_RTO_INIT;
398 }
399 
400 /**
401  * Stop all connection timers
402  */
403 void
405 {
406  int i;
407  for (i = 0; i < TCP_N_TIMERS; i++)
408  {
409  tcp_timer_reset (tc, i);
410  }
411 }
412 
413 #if 0
414 typedef struct ip4_tcp_hdr
415 {
416  ip4_header_t ip;
417  tcp_header_t tcp;
418 } ip4_tcp_hdr_t;
419 
420 typedef struct ip6_tcp_hdr
421 {
422  ip6_header_t ip;
423  tcp_header_t tcp;
424 } ip6_tcp_hdr_t;
425 
426 static void
427 tcp_connection_select_lb_bucket (tcp_connection_t * tc, const dpo_id_t * dpo,
428  dpo_id_t * result)
429 {
430  const dpo_id_t *choice;
431  load_balance_t *lb;
432  int hash;
433 
434  lb = load_balance_get (dpo->dpoi_index);
435  if (tc->c_is_ip4)
436  {
437  ip4_tcp_hdr_t hdr;
438  memset (&hdr, 0, sizeof (hdr));
439  hdr.ip.protocol = IP_PROTOCOL_TCP;
440  hdr.ip.address_pair.src.as_u32 = tc->c_lcl_ip.ip4.as_u32;
441  hdr.ip.address_pair.dst.as_u32 = tc->c_rmt_ip.ip4.as_u32;
442  hdr.tcp.src_port = tc->c_lcl_port;
443  hdr.tcp.dst_port = tc->c_rmt_port;
444  hash = ip4_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
445  }
446  else
447  {
448  ip6_tcp_hdr_t hdr;
449  memset (&hdr, 0, sizeof (hdr));
450  hdr.ip.protocol = IP_PROTOCOL_TCP;
451  clib_memcpy (&hdr.ip.src_address, &tc->c_lcl_ip.ip6,
452  sizeof (ip6_address_t));
453  clib_memcpy (&hdr.ip.dst_address, &tc->c_rmt_ip.ip6,
454  sizeof (ip6_address_t));
455  hdr.tcp.src_port = tc->c_lcl_port;
456  hdr.tcp.dst_port = tc->c_rmt_port;
457  hash = ip6_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
458  }
459  choice = load_balance_get_bucket_i (lb, hash & lb->lb_n_buckets_minus_1);
460  dpo_copy (result, choice);
461 }
462 
465 {
467  u32 fib_index;
468 
469  clib_memcpy (&prefix.fp_addr, &tc->c_rmt_ip, sizeof (prefix.fp_addr));
470  prefix.fp_proto = tc->c_is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6;
471  prefix.fp_len = tc->c_is_ip4 ? 32 : 128;
472  fib_index = fib_table_find (prefix.fp_proto, tc->c_fib_index);
473  return fib_table_lookup (fib_index, &prefix);
474 }
475 
476 static int
477 tcp_connection_stack_on_fib_entry (tcp_connection_t * tc)
478 {
479  dpo_id_t choice = DPO_INVALID;
480  u32 output_node_index;
481  fib_entry_t *fe;
482 
483  fe = fib_entry_get (tc->c_rmt_fei);
484  if (fe->fe_lb.dpoi_type != DPO_LOAD_BALANCE)
485  return -1;
486 
487  tcp_connection_select_lb_bucket (tc, &fe->fe_lb, &choice);
488 
489  output_node_index =
490  tc->c_is_ip4 ? tcp4_output_node.index : tcp6_output_node.index;
491  dpo_stack_from_node (output_node_index, &tc->c_rmt_dpo, &choice);
492  return 0;
493 }
494 
495 /** Stack tcp connection on peer's fib entry.
496  *
497  * This ultimately populates the dpo the connection will use to send packets.
498  */
499 static void
500 tcp_connection_fib_attach (tcp_connection_t * tc)
501 {
502  tc->c_rmt_fei = tcp_lookup_rmt_in_fib (tc);
503 
504  ASSERT (tc->c_rmt_fei != FIB_NODE_INDEX_INVALID);
505 
506  tcp_connection_stack_on_fib_entry (tc);
507 }
508 #endif /* 0 */
509 
510 static void
512 {
513  tc->cc_algo = tcp_cc_algo_get (TCP_CC_NEWRENO);
514  tc->cc_algo->init (tc);
515 }
516 
517 void
519  const tcp_cc_algorithm_t * vft)
520 {
521  tcp_main_t *tm = vnet_get_tcp_main ();
522  vec_validate (tm->cc_algos, type);
523 
524  tm->cc_algos[type] = *vft;
525 }
526 
529 {
530  tcp_main_t *tm = vnet_get_tcp_main ();
531  return &tm->cc_algos[type];
532 }
533 
534 
535 /**
536  * Initialize connection send variables.
537  */
538 void
540 {
541  u32 time_now;
542 
543  /*
544  * We use the time to randomize iss and for setting up the initial
545  * timestamp. Make sure it's updated otherwise syn and ack in the
546  * handshake may make it look as if time has flown in the opposite
547  * direction for us.
548  */
550  time_now = tcp_time_now ();
551 
552  tc->iss = random_u32 (&time_now);
553  tc->snd_una = tc->iss;
554  tc->snd_nxt = tc->iss + 1;
555  tc->snd_una_max = tc->snd_nxt;
556 }
557 
558 /** Initialize tcp connection variables
559  *
560  * Should be called after having received a msg from the peer, i.e., a SYN or
561  * a SYNACK, such that connection options have already been exchanged. */
562 void
564 {
566  tcp_init_mss (tc);
567  scoreboard_init (&tc->sack_sb);
568  tcp_cc_init (tc);
569  if (tc->state == TCP_STATE_SYN_RCVD)
570  tcp_init_snd_vars (tc);
571 
572  if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6))
573  tcp_add_del_adjacency (tc, 1);
574 
575  // tcp_connection_fib_attach (tc);
576 }
577 
578 static int
579 tcp_alloc_custom_local_endpoint (tcp_main_t * tm, ip46_address_t * lcl_addr,
580  u16 * lcl_port, u8 is_ip4)
581 {
582  int index, port;
583  if (is_ip4)
584  {
585  index = tm->last_v4_address_rotor++;
586  if (tm->last_v4_address_rotor >= vec_len (tm->ip4_src_addresses))
587  tm->last_v4_address_rotor = 0;
588  lcl_addr->ip4.as_u32 = tm->ip4_src_addresses[index].as_u32;
589  }
590  else
591  {
592  index = tm->last_v6_address_rotor++;
593  if (tm->last_v6_address_rotor >= vec_len (tm->ip6_src_addresses))
594  tm->last_v6_address_rotor = 0;
595  clib_memcpy (&lcl_addr->ip6, &tm->ip6_src_addresses[index],
596  sizeof (ip6_address_t));
597  }
599  if (port < 1)
600  {
601  clib_warning ("Failed to allocate src port");
602  return -1;
603  }
604  *lcl_port = port;
605  return 0;
606 }
607 
608 static int
610 {
611  tcp_main_t *tm = vnet_get_tcp_main ();
612  tcp_connection_t *tc;
613  ip46_address_t lcl_addr;
614  u16 lcl_port;
615  int rv;
616 
617  /*
618  * Allocate local endpoint
619  */
620  if ((rmt->is_ip4 && vec_len (tm->ip4_src_addresses))
621  || (!rmt->is_ip4 && vec_len (tm->ip6_src_addresses)))
622  rv = tcp_alloc_custom_local_endpoint (tm, &lcl_addr, &lcl_port,
623  rmt->is_ip4);
624  else
626  rmt, &lcl_addr, &lcl_port);
627 
628  if (rv)
629  return -1;
630 
631  /*
632  * Create connection and send SYN
633  */
634  clib_spinlock_lock_if_init (&tm->half_open_lock);
636  ip_copy (&tc->c_rmt_ip, &rmt->ip, rmt->is_ip4);
637  ip_copy (&tc->c_lcl_ip, &lcl_addr, rmt->is_ip4);
638  tc->c_rmt_port = rmt->port;
639  tc->c_lcl_port = clib_host_to_net_u16 (lcl_port);
640  tc->c_is_ip4 = rmt->is_ip4;
641  tc->c_proto = TRANSPORT_PROTO_TCP;
642  tc->c_fib_index = rmt->fib_index;
643  /* The other connection vars will be initialized after SYN ACK */
645 
646  TCP_EVT_DBG (TCP_EVT_OPEN, tc);
647  tc->state = TCP_STATE_SYN_SENT;
648  tcp_init_snd_vars (tc);
649  tcp_send_syn (tc);
650  clib_spinlock_unlock_if_init (&tm->half_open_lock);
651 
652  return tc->c_c_index;
653 }
654 
655 static int
657 {
658  return tcp_connection_open (tep);
659 }
660 
661 const char *tcp_dbg_evt_str[] = {
662 #define _(sym, str) str,
664 #undef _
665 };
666 
667 const char *tcp_fsm_states[] = {
668 #define _(sym, str) str,
670 #undef _
671 };
672 
673 u8 *
674 format_tcp_state (u8 * s, va_list * args)
675 {
676  u32 state = va_arg (*args, u32);
677 
678  if (state < TCP_N_STATES)
679  s = format (s, "%s", tcp_fsm_states[state]);
680  else
681  s = format (s, "UNKNOWN (%d (0x%x))", state, state);
682  return s;
683 }
684 
685 const char *tcp_connection_flags_str[] = {
686 #define _(sym, str) str,
688 #undef _
689 };
690 
691 static u8 *
692 format_tcp_connection_flags (u8 * s, va_list * args)
693 {
694  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
695  int i, last = -1;
696 
697  for (i = 0; i < TCP_CONN_N_FLAG_BITS; i++)
698  if (tc->flags & (1 << i))
699  last = i;
700  for (i = 0; i < last; i++)
701  {
702  if (tc->flags & (1 << i))
703  s = format (s, "%s, ", tcp_connection_flags_str[i]);
704  }
705  if (last >= 0)
706  s = format (s, "%s", tcp_connection_flags_str[last]);
707  return s;
708 }
709 
710 const char *tcp_conn_timers[] = {
711 #define _(sym, str) str,
713 #undef _
714 };
715 
716 static u8 *
717 format_tcp_timers (u8 * s, va_list * args)
718 {
719  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
720  int i, last = -1;
721 
722  for (i = 0; i < TCP_N_TIMERS; i++)
723  if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
724  last = i;
725 
726  s = format (s, "[");
727  for (i = 0; i < last; i++)
728  {
729  if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
730  s = format (s, "%s,", tcp_conn_timers[i]);
731  }
732 
733  if (last >= 0)
734  s = format (s, "%s]", tcp_conn_timers[i]);
735  else
736  s = format (s, "]");
737 
738  return s;
739 }
740 
741 static u8 *
742 format_tcp_congestion_status (u8 * s, va_list * args)
743 {
744  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
745  if (tcp_in_recovery (tc))
746  s = format (s, "recovery");
747  else if (tcp_in_fastrecovery (tc))
748  s = format (s, "fastrecovery");
749  else
750  s = format (s, "none");
751  return s;
752 }
753 
754 static i32
756 {
757  return (i32) tc->rcv_wnd - (tc->rcv_nxt - tc->rcv_las);
758 }
759 
760 static u8 *
761 format_tcp_vars (u8 * s, va_list * args)
762 {
763  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
764  s = format (s, " flags: %U timers: %U\n", format_tcp_connection_flags, tc,
765  format_tcp_timers, tc);
766  s = format (s, " snd_una %u snd_nxt %u snd_una_max %u",
767  tc->snd_una - tc->iss, tc->snd_nxt - tc->iss,
768  tc->snd_una_max - tc->iss);
769  s = format (s, " rcv_nxt %u rcv_las %u\n",
770  tc->rcv_nxt - tc->irs, tc->rcv_las - tc->irs);
771  s = format (s, " snd_wnd %u rcv_wnd %u rcv_wscale %u ",
772  tc->snd_wnd, tc->rcv_wnd, tc->rcv_wscale);
773  s = format (s, "snd_wl1 %u snd_wl2 %u\n", tc->snd_wl1 - tc->irs,
774  tc->snd_wl2 - tc->iss);
775  s = format (s, " flight size %u out space %u cc space %u rcv_wnd_av %u\n",
778  s = format (s, " cong %U ", format_tcp_congestion_status, tc);
779  s = format (s, "cwnd %u ssthresh %u rtx_bytes %u bytes_acked %u\n",
780  tc->cwnd, tc->ssthresh, tc->snd_rxt_bytes, tc->bytes_acked);
781  s = format (s, " prev_ssthresh %u snd_congestion %u dupack %u",
782  tc->prev_ssthresh, tc->snd_congestion - tc->iss,
783  tc->rcv_dupacks);
784  s = format (s, " limited_transmit %u\n", tc->limited_transmit - tc->iss);
785  s = format (s, " tsecr %u tsecr_last_ack %u\n", tc->rcv_opts.tsecr,
786  tc->tsecr_last_ack);
787  s = format (s, " rto %u rto_boff %u srtt %u rttvar %u rtt_ts %u ", tc->rto,
788  tc->rto_boff, tc->srtt, tc->rttvar, tc->rtt_ts);
789  s = format (s, "rtt_seq %u\n", tc->rtt_seq);
790  s = format (s, " tsval_recent %u tsval_recent_age %u\n", tc->tsval_recent,
791  tcp_time_now () - tc->tsval_recent_age);
792  if (tc->state >= TCP_STATE_ESTABLISHED)
793  s = format (s, " scoreboard: %U\n", format_tcp_scoreboard, &tc->sack_sb,
794  tc);
795  if (vec_len (tc->snd_sacks))
796  s = format (s, " sacks tx: %U\n", format_tcp_sacks, tc);
797 
798  return s;
799 }
800 
801 static u8 *
802 format_tcp_connection_id (u8 * s, va_list * args)
803 {
804  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
805  if (!tc)
806  return s;
807  if (tc->c_is_ip4)
808  {
809  s = format (s, "[#%d][%s] %U:%d->%U:%d", tc->c_thread_index, "T",
810  format_ip4_address, &tc->c_lcl_ip4,
811  clib_net_to_host_u16 (tc->c_lcl_port), format_ip4_address,
812  &tc->c_rmt_ip4, clib_net_to_host_u16 (tc->c_rmt_port));
813  }
814  else
815  {
816  s = format (s, "[#%d][%s] %U:%d->%U:%d", tc->c_thread_index, "T",
817  format_ip6_address, &tc->c_lcl_ip6,
818  clib_net_to_host_u16 (tc->c_lcl_port), format_ip6_address,
819  &tc->c_rmt_ip6, clib_net_to_host_u16 (tc->c_rmt_port));
820  }
821 
822  return s;
823 }
824 
825 u8 *
826 format_tcp_connection (u8 * s, va_list * args)
827 {
828  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
829  u32 verbose = va_arg (*args, u32);
830 
831  if (!tc)
832  return s;
833  s = format (s, "%-50U", format_tcp_connection_id, tc);
834  if (verbose)
835  {
836  s = format (s, "%-15U", format_tcp_state, tc->state);
837  if (verbose > 1)
838  s = format (s, "\n%U", format_tcp_vars, tc);
839  }
840 
841  return s;
842 }
843 
844 static u8 *
845 format_tcp_session (u8 * s, va_list * args)
846 {
847  u32 tci = va_arg (*args, u32);
848  u32 thread_index = va_arg (*args, u32);
849  u32 verbose = va_arg (*args, u32);
850  tcp_connection_t *tc;
851 
852  tc = tcp_connection_get (tci, thread_index);
853  if (tc)
854  s = format (s, "%U", format_tcp_connection, tc, verbose);
855  else
856  s = format (s, "empty\n");
857  return s;
858 }
859 
860 static u8 *
861 format_tcp_listener_session (u8 * s, va_list * args)
862 {
863  u32 tci = va_arg (*args, u32);
865  return format (s, "%U", format_tcp_connection_id, tc);
866 }
867 
868 static u8 *
869 format_tcp_half_open_session (u8 * s, va_list * args)
870 {
871  u32 tci = va_arg (*args, u32);
873  return format (s, "%U", format_tcp_connection_id, tc);
874 }
875 
876 u8 *
877 format_tcp_sacks (u8 * s, va_list * args)
878 {
879  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
880  sack_block_t *sacks = tc->snd_sacks;
881  sack_block_t *block;
882  int i, len = 0;
883 
884  len = vec_len (sacks);
885  for (i = 0; i < len - 1; i++)
886  {
887  block = &sacks[i];
888  s = format (s, " start %u end %u\n", block->start - tc->irs,
889  block->end - tc->irs);
890  }
891  if (len)
892  {
893  block = &sacks[len - 1];
894  s = format (s, " start %u end %u", block->start - tc->irs,
895  block->end - tc->irs);
896  }
897  return s;
898 }
899 
900 u8 *
901 format_tcp_rcv_sacks (u8 * s, va_list * args)
902 {
903  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
904  sack_block_t *sacks = tc->rcv_opts.sacks;
905  sack_block_t *block;
906  int i, len = 0;
907 
908  len = vec_len (sacks);
909  for (i = 0; i < len - 1; i++)
910  {
911  block = &sacks[i];
912  s = format (s, " start %u end %u\n", block->start - tc->iss,
913  block->end - tc->iss);
914  }
915  if (len)
916  {
917  block = &sacks[len - 1];
918  s = format (s, " start %u end %u", block->start - tc->iss,
919  block->end - tc->iss);
920  }
921  return s;
922 }
923 
924 static u8 *
925 format_tcp_sack_hole (u8 * s, va_list * args)
926 {
927  sack_scoreboard_hole_t *hole = va_arg (*args, sack_scoreboard_hole_t *);
928  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
929  if (tc)
930  s = format (s, " [%u, %u]", hole->start - tc->iss, hole->end - tc->iss);
931  else
932  s = format (s, " [%u, %u]", hole->start, hole->end);
933  return s;
934 }
935 
936 u8 *
937 format_tcp_scoreboard (u8 * s, va_list * args)
938 {
939  sack_scoreboard_t *sb = va_arg (*args, sack_scoreboard_t *);
940  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
942  s = format (s, "sacked_bytes %u last_sacked_bytes %u lost_bytes %u\n",
943  sb->sacked_bytes, sb->last_sacked_bytes, sb->lost_bytes);
944  s = format (s, " last_bytes_delivered %u high_sacked %u snd_una_adv %u\n",
945  sb->last_bytes_delivered, sb->high_sacked - tc->iss,
946  sb->snd_una_adv);
947  s = format (s, " cur_rxt_hole %u high_rxt %u rescue_rxt %u",
948  sb->cur_rxt_hole, sb->high_rxt - tc->iss,
949  sb->rescue_rxt - tc->iss);
950 
951  hole = scoreboard_first_hole (sb);
952  if (hole)
953  s = format (s, "\n head %u tail %u holes:\n", sb->head, sb->tail);
954 
955  while (hole)
956  {
957  s = format (s, "%U", format_tcp_sack_hole, hole, tc);
958  hole = scoreboard_next_hole (sb, hole);
959  }
960 
961  return s;
962 }
963 
964 static transport_connection_t *
965 tcp_session_get_transport (u32 conn_index, u32 thread_index)
966 {
967  tcp_connection_t *tc = tcp_connection_get (conn_index, thread_index);
968  return &tc->connection;
969 }
970 
971 static transport_connection_t *
973 {
975  return &tc->connection;
976 }
977 
978 /**
979  * Compute maximum segment size for session layer.
980  *
981  * Since the result needs to be the actual data length, it first computes
982  * the tcp options to be used in the next burst and subtracts their
983  * length from the connection's snd_mss.
984  */
985 static u16
987 {
988  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
989 
990  /* Ensure snd_mss does accurately reflect the amount of data we can push
991  * in a segment. This also makes sure that options are updated according to
992  * the current state of the connection. */
994 
995  return tc->snd_mss;
996 }
997 
1000 {
1001  if (PREDICT_FALSE (tc->snd_wnd < tc->snd_mss))
1002  {
1003  return tc->snd_wnd <= snd_space ? tc->snd_wnd : 0;
1004  }
1005 
1006  /* If not snd_wnd constrained and we can't write at least a segment,
1007  * don't try at all */
1008  if (PREDICT_FALSE (snd_space < tc->snd_mss))
1009  return snd_space < tc->cwnd ? 0 : snd_space;
1010 
1011  /* round down to mss multiple */
1012  return snd_space - (snd_space % tc->snd_mss);
1013 }
1014 
1015 /**
1016  * Compute tx window session is allowed to fill.
1017  *
1018  * Takes into account available send space, snd_mss and the congestion
1019  * state of the connection. If possible, the value returned is a multiple
1020  * of snd_mss.
1021  *
1022  * @param tc tcp connection
1023  * @return number of bytes session is allowed to write
1024  */
1025 static inline u32
1027 {
1028  int snd_space, snt_limited;
1029 
1030  if (PREDICT_TRUE (tcp_in_cong_recovery (tc) == 0))
1031  {
1032  snd_space = tcp_available_output_snd_space (tc);
1033 
1034  /* If we haven't gotten dupacks or if we did and have gotten sacked
1035  * bytes then we can still send as per Limited Transmit (RFC3042) */
1036  if (PREDICT_FALSE (tc->rcv_dupacks != 0
1037  && (tcp_opts_sack_permitted (tc)
1038  && tc->sack_sb.last_sacked_bytes == 0)))
1039  {
1040  if (tc->rcv_dupacks == 1 && tc->limited_transmit != tc->snd_nxt)
1041  tc->limited_transmit = tc->snd_nxt;
1042  ASSERT (seq_leq (tc->limited_transmit, tc->snd_nxt));
1043 
1044  snt_limited = tc->snd_nxt - tc->limited_transmit;
1045  snd_space = clib_max (2 * tc->snd_mss - snt_limited, 0);
1046  }
1047  return tcp_round_snd_space (tc, snd_space);
1048  }
1049 
1050  if (tcp_in_recovery (tc))
1051  {
1052  tc->snd_nxt = tc->snd_una_max;
1053  snd_space = tcp_available_snd_wnd (tc) - tc->snd_rxt_bytes
1054  - (tc->snd_una_max - tc->snd_congestion);
1055  if (snd_space <= 0 || (tc->snd_una_max - tc->snd_una) >= tc->snd_wnd)
1056  return 0;
1057  return tcp_round_snd_space (tc, snd_space);
1058  }
1059 
1060  /* RFC 5681: When previously unsent data is available and the new value of
1061  * cwnd and the receiver's advertised window allow, a TCP SHOULD send 1*SMSS
1062  * bytes of previously unsent data. */
1064  {
1065  if (tcp_available_cc_snd_space (tc) < tc->snd_mss)
1066  return 0;
1068  return tc->snd_mss;
1069  }
1070 
1071  return 0;
1072 }
1073 
1074 u32
1076 {
1077  return tcp_snd_space_inline (tc);
1078 }
1079 
1080 static u32
1082 {
1083  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1084  return clib_min (tcp_snd_space_inline (tc),
1085  tc->snd_wnd - (tc->snd_nxt - tc->snd_una));
1086 }
1087 
1088 static u32
1090 {
1091  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1092 
1093  ASSERT (seq_geq (tc->snd_nxt, tc->snd_una));
1094 
1095  /* This still works if fast retransmit is on */
1096  return (tc->snd_nxt - tc->snd_una);
1097 }
1098 
1099 static void
1100 tcp_update_time (f64 now, u8 thread_index)
1101 {
1102  tcp_set_time_now (thread_index);
1103  tw_timer_expire_timers_16t_2w_512sl (&tcp_main.
1104  wrk_ctx[thread_index].timer_wheel,
1105  now);
1106  tcp_flush_frames_to_output (thread_index);
1107 }
1108 
1109 static u32
1111 {
1112  tcp_connection_t *tc = (tcp_connection_t *) tconn;
1113  return tcp_push_header (tc, b);
1114 }
1115 
1116 /* *INDENT-OFF* */
1118  .enable = vnet_tcp_enable_disable,
1119  .bind = tcp_session_bind,
1120  .unbind = tcp_session_unbind,
1121  .push_header = tcp_session_push_header,
1122  .get_connection = tcp_session_get_transport,
1123  .get_listener = tcp_session_get_listener,
1124  .get_half_open = tcp_half_open_session_get_transport,
1125  .open = tcp_session_open,
1126  .close = tcp_session_close,
1127  .cleanup = tcp_session_cleanup,
1128  .send_mss = tcp_session_send_mss,
1129  .send_space = tcp_session_send_space,
1130  .update_time = tcp_update_time,
1131  .tx_fifo_offset = tcp_session_tx_fifo_offset,
1132  .format_connection = format_tcp_session,
1133  .format_listener = format_tcp_listener_session,
1134  .format_half_open = format_tcp_half_open_session,
1135  .tx_type = TRANSPORT_TX_PEEK,
1136  .service_type = TRANSPORT_SERVICE_VC,
1137 };
1138 /* *INDENT-ON* */
1139 
1140 static void
1142 {
1143  u32 thread_index = vlib_get_thread_index ();
1144  tcp_connection_t *tc;
1145 
1146  tc = tcp_connection_get (conn_index, thread_index);
1147  tc->timers[TCP_TIMER_KEEP] = TCP_TIMER_HANDLE_INVALID;
1148 
1149  tcp_connection_close (tc);
1150 }
1151 
1152 static void
1154 {
1155  tcp_connection_t *tc;
1156 
1157  tc = tcp_half_open_connection_get (conn_index);
1158  if (tc)
1159  {
1160  ASSERT (tc->state == TCP_STATE_SYN_SENT);
1161  session_stream_connect_notify (&tc->connection, 1 /* fail */ );
1162  TCP_DBG ("establish pop: %U", format_tcp_connection, tc, 2);
1163  }
1164  else
1165  {
1166  tc = tcp_connection_get (conn_index, vlib_get_thread_index ());
1167  /* note: the connection may have already disappeared */
1168  if (PREDICT_FALSE (tc == 0))
1169  return;
1170  TCP_DBG ("establish pop: %U", format_tcp_connection, tc, 2);
1171  ASSERT (tc->state == TCP_STATE_SYN_RCVD);
1172  /* Start cleanup. App wasn't notified yet so use delete notify as
1173  * opposed to delete to cleanup session layer state. */
1174  stream_session_delete_notify (&tc->connection);
1175  }
1176  tc->timers[TCP_TIMER_ESTABLISH] = TCP_TIMER_HANDLE_INVALID;
1178 }
1179 
1180 static void
1182 {
1183  u32 thread_index = vlib_get_thread_index ();
1184  tcp_connection_t *tc;
1185 
1186  tc = tcp_connection_get (conn_index, thread_index);
1187  if (!tc)
1188  return;
1189  tc->timers[TCP_TIMER_WAITCLOSE] = TCP_TIMER_HANDLE_INVALID;
1190 
1191  /* Session didn't come back with a close(). Send FIN either way
1192  * and switch to LAST_ACK. */
1193  if (tc->state == TCP_STATE_CLOSE_WAIT)
1194  {
1195  if (tc->flags & TCP_CONN_FINSNT)
1196  {
1197  clib_warning ("FIN was sent and still in CLOSE WAIT. Weird!");
1198  }
1199 
1200  /* Make sure we don't try to send unsent data */
1202  tcp_cong_recovery_off (tc);
1203  tc->snd_una_max = tc->snd_nxt = tc->snd_una;
1204  tcp_send_fin (tc);
1205  tc->state = TCP_STATE_LAST_ACK;
1206 
1207  /* Make sure we don't wait in LAST ACK forever */
1208  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
1209 
1210  /* Don't delete the connection yet */
1211  return;
1212  }
1213 
1214  tcp_connection_del (tc);
1215 }
1216 
1217 /* *INDENT-OFF* */
1219 {
1227 };
1228 /* *INDENT-ON* */
1229 
1230 static void
1232 {
1233  int i;
1234  u32 connection_index, timer_id;
1235 
1236  for (i = 0; i < vec_len (expired_timers); i++)
1237  {
1238  /* Get session index and timer id */
1239  connection_index = expired_timers[i] & 0x0FFFFFFF;
1240  timer_id = expired_timers[i] >> 28;
1241 
1242  TCP_EVT_DBG (TCP_EVT_TIMER_POP, connection_index, timer_id);
1243 
1244  /* Handle expiration */
1245  (*timer_expiration_handlers[timer_id]) (connection_index);
1246  }
1247 }
1248 
1249 static void
1251 {
1252  tw_timer_wheel_16t_2w_512sl_t *tw;
1253  /* *INDENT-OFF* */
1254  foreach_vlib_main (({
1255  tw = &tm->wrk_ctx[ii].timer_wheel;
1256  tw_timer_wheel_init_16t_2w_512sl (tw, tcp_expired_timers_dispatch,
1257  100e-3 /* timer period 100ms */ , ~0);
1258  tw->last_run_time = vlib_time_now (this_vlib_main);
1259  }));
1260  /* *INDENT-ON* */
1261 }
1262 
1263 static clib_error_t *
1265 {
1266  tcp_main_t *tm = vnet_get_tcp_main ();
1268  clib_error_t *error = 0;
1269  u32 num_threads;
1270  int thread;
1271  tcp_connection_t *tc __attribute__ ((unused));
1272  u32 preallocated_connections_per_thread;
1273 
1274  if ((error = vlib_call_init_function (vm, ip_main_init)))
1275  return error;
1276  if ((error = vlib_call_init_function (vm, ip4_lookup_init)))
1277  return error;
1278  if ((error = vlib_call_init_function (vm, ip6_lookup_init)))
1279  return error;
1280 
1281  /*
1282  * Registrations
1283  */
1284 
1285  ip4_register_protocol (IP_PROTOCOL_TCP, tcp4_input_node.index);
1286  ip6_register_protocol (IP_PROTOCOL_TCP, tcp6_input_node.index);
1287 
1288  /*
1289  * Initialize data structures
1290  */
1291 
1292  num_threads = 1 /* main thread */ + vtm->n_threads;
1293  vec_validate (tm->connections, num_threads - 1);
1294 
1295  /*
1296  * Preallocate connections. Assume that thread 0 won't
1297  * use preallocated threads when running multi-core
1298  */
1299  if (num_threads == 1)
1300  {
1301  thread = 0;
1302  preallocated_connections_per_thread = tm->preallocated_connections;
1303  }
1304  else
1305  {
1306  thread = 1;
1307  preallocated_connections_per_thread =
1308  tm->preallocated_connections / (num_threads - 1);
1309  }
1310  for (; thread < num_threads; thread++)
1311  {
1312  if (preallocated_connections_per_thread)
1313  pool_init_fixed (tm->connections[thread],
1314  preallocated_connections_per_thread);
1315  }
1316 
1317  /*
1318  * Use a preallocated half-open connection pool?
1319  */
1320  if (tm->preallocated_half_open_connections)
1321  pool_init_fixed (tm->half_open_connections,
1322  tm->preallocated_half_open_connections);
1323 
1324  /* Initialize clocks per tick for TCP timestamp. Used to compute
1325  * monotonically increasing timestamps. */
1326  tm->tstamp_ticks_per_clock = vm->clib_time.seconds_per_clock
1328 
1329  if (num_threads > 1)
1330  {
1331  clib_spinlock_init (&tm->half_open_lock);
1332  }
1333 
1334  vec_validate (tm->wrk_ctx, num_threads - 1);
1336 
1337  tm->bytes_per_buffer = vlib_buffer_free_list_buffer_size
1339 
1340  return error;
1341 }
1342 
1343 clib_error_t *
1345 {
1346  if (is_en)
1347  {
1348  if (tcp_main.is_enabled)
1349  return 0;
1350 
1351  return tcp_main_enable (vm);
1352  }
1353  else
1354  {
1355  tcp_main.is_enabled = 0;
1356  }
1357 
1358  return 0;
1359 }
1360 
1361 void
1362 tcp_punt_unknown (vlib_main_t * vm, u8 is_ip4, u8 is_add)
1363 {
1364  tcp_main_t *tm = &tcp_main;
1365  if (is_ip4)
1366  tm->punt_unknown4 = is_add;
1367  else
1368  tm->punt_unknown6 = is_add;
1369 }
1370 
1371 static clib_error_t *
1373 {
1374  tcp_main_t *tm = vnet_get_tcp_main ();
1375  ip_main_t *im = &ip_main;
1376  ip_protocol_info_t *pi;
1377 
1378  /* Session layer, and by implication tcp, are disabled by default */
1379  tm->is_enabled = 0;
1380 
1381  /* Register with IP for header parsing */
1382  pi = ip_get_protocol_info (im, IP_PROTOCOL_TCP);
1383  if (pi == 0)
1384  return clib_error_return (0, "TCP protocol info AWOL");
1387 
1388  /* Register as transport with session layer */
1393 
1394  tcp_api_reference ();
1395  return 0;
1396 }
1397 
1399 
1400 static clib_error_t *
1402 {
1403  tcp_main_t *tm = vnet_get_tcp_main ();
1404 
1405  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1406  {
1407  if (unformat (input, "preallocated-connections %d",
1408  &tm->preallocated_connections))
1409  ;
1410  else if (unformat (input, "preallocated-half-open-connections %d",
1411  &tm->preallocated_half_open_connections))
1412  ;
1413  else if (unformat (input, "buffer-fail-fraction %f",
1414  &tm->buffer_fail_fraction))
1415  ;
1416  else if (unformat (input, "max-rx-fifo %U", unformat_memory_size,
1417  &tm->max_rx_fifo))
1418  ;
1419  else
1420  return clib_error_return (0, "unknown input `%U'",
1421  format_unformat_error, input);
1422  }
1423  return 0;
1424 }
1425 
1427 
1428 
1429 /**
1430  * \brief Configure an ipv4 source address range
1431  * @param vm vlib_main_t pointer
1432  * @param start first ipv4 address in the source address range
1433  * @param end last ipv4 address in the source address range
1434  * @param table_id VRF / table ID, 0 for the default FIB
1435  * @return 0 if all OK, else an error indication from api_errno.h
1436  */
1437 
1438 int
1440  ip4_address_t * start,
1441  ip4_address_t * end, u32 table_id)
1442 {
1443  tcp_main_t *tm = vnet_get_tcp_main ();
1444  vnet_main_t *vnm = vnet_get_main ();
1445  u32 start_host_byte_order, end_host_byte_order;
1447  vnet_sw_interface_t *si;
1448  fib_node_index_t fei;
1449  u32 fib_index = 0;
1450  u32 sw_if_index;
1451  int rv;
1452  int vnet_proxy_arp_add_del (ip4_address_t * lo_addr,
1453  ip4_address_t * hi_addr, u32 fib_index,
1454  int is_del);
1455 
1456  memset (&prefix, 0, sizeof (prefix));
1457 
1458  fib_index = fib_table_find (FIB_PROTOCOL_IP4, table_id);
1459 
1460  if (fib_index == ~0)
1461  return VNET_API_ERROR_NO_SUCH_FIB;
1462 
1463  start_host_byte_order = clib_net_to_host_u32 (start->as_u32);
1464  end_host_byte_order = clib_net_to_host_u32 (end->as_u32);
1465 
1466  /* sanity check for reversed args or some such */
1467  if ((end_host_byte_order - start_host_byte_order) > (10 << 10))
1468  return VNET_API_ERROR_INVALID_ARGUMENT;
1469 
1470  /* Lookup the last address, to identify the interface involved */
1471  prefix.fp_len = 32;
1472  prefix.fp_proto = FIB_PROTOCOL_IP4;
1473  memcpy (&prefix.fp_addr.ip4, end, sizeof (ip4_address_t));
1474 
1475  fei = fib_table_lookup (fib_index, &prefix);
1476 
1477  /* Couldn't find route to destination. Bail out. */
1478  if (fei == FIB_NODE_INDEX_INVALID)
1479  return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1480 
1481  sw_if_index = fib_entry_get_resolving_interface (fei);
1482 
1483  /* Enable proxy arp on the interface */
1484  si = vnet_get_sw_interface (vnm, sw_if_index);
1486 
1487  /* Configure proxy arp across the range */
1488  rv = vnet_proxy_arp_add_del (start, end, fib_index, 0 /* is_del */ );
1489 
1490  if (rv)
1491  return rv;
1492 
1493  do
1494  {
1495  dpo_id_t dpo = DPO_INVALID;
1496 
1497  vec_add1 (tm->ip4_src_addresses, start[0]);
1498 
1499  /* Add local adjacencies for the range */
1500 
1501  receive_dpo_add_or_lock (DPO_PROTO_IP4, ~0 /* sw_if_index */ ,
1502  NULL, &dpo);
1503  prefix.fp_len = 32;
1504  prefix.fp_proto = FIB_PROTOCOL_IP4;
1505  prefix.fp_addr.ip4.as_u32 = start->as_u32;
1506 
1508  &prefix,
1510  FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
1511  dpo_reset (&dpo);
1512 
1513  start_host_byte_order++;
1514  start->as_u32 = clib_host_to_net_u32 (start_host_byte_order);
1515  }
1516  while (start_host_byte_order <= end_host_byte_order);
1517 
1518  return 0;
1519 }
1520 
1521 /**
1522  * \brief Configure an ipv6 source address range
1523  * @param vm vlib_main_t pointer
1524  * @param start first ipv6 address in the source address range
1525  * @param end last ipv6 address in the source address range
1526  * @param table_id VRF / table ID, 0 for the default FIB
1527  * @return 0 if all OK, else an error indication from api_errno.h
1528  */
1529 
1530 int
1532  ip6_address_t * start,
1533  ip6_address_t * end, u32 table_id)
1534 {
1535  tcp_main_t *tm = vnet_get_tcp_main ();
1537  u32 fib_index = 0;
1538  fib_node_index_t fei;
1539  u32 sw_if_index;
1540 
1541  memset (&prefix, 0, sizeof (prefix));
1542 
1543  fib_index = fib_table_find (FIB_PROTOCOL_IP6, table_id);
1544 
1545  if (fib_index == ~0)
1546  return VNET_API_ERROR_NO_SUCH_FIB;
1547 
1548  while (1)
1549  {
1550  int i;
1551  ip6_address_t tmp;
1552  dpo_id_t dpo = DPO_INVALID;
1553 
1554  /* Remember this address */
1555  vec_add1 (tm->ip6_src_addresses, start[0]);
1556 
1557  /* Lookup the prefix, to identify the interface involved */
1558  prefix.fp_len = 128;
1559  prefix.fp_proto = FIB_PROTOCOL_IP6;
1560  memcpy (&prefix.fp_addr.ip6, start, sizeof (ip6_address_t));
1561 
1562  fei = fib_table_lookup (fib_index, &prefix);
1563 
1564  /* Couldn't find route to destination. Bail out. */
1565  if (fei == FIB_NODE_INDEX_INVALID)
1566  return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1567 
1568  sw_if_index = fib_entry_get_resolving_interface (fei);
1569 
1570  if (sw_if_index == (u32) ~ 0)
1571  return VNET_API_ERROR_NO_MATCHING_INTERFACE;
1572 
1573  /* Add a proxy neighbor discovery entry for this address */
1574  ip6_neighbor_proxy_add_del (sw_if_index, start, 0 /* is_del */ );
1575 
1576  /* Add a receive adjacency for this address */
1577  receive_dpo_add_or_lock (DPO_PROTO_IP6, ~0 /* sw_if_index */ ,
1578  NULL, &dpo);
1579 
1581  &prefix,
1583  FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
1584  dpo_reset (&dpo);
1585 
1586  /* Done with the entire range? */
1587  if (!memcmp (start, end, sizeof (start[0])))
1588  break;
1589 
1590  /* Increment the address. DGMS. */
1591  tmp = start[0];
1592  for (i = 15; i >= 0; i--)
1593  {
1594  tmp.as_u8[i] += 1;
1595  if (tmp.as_u8[i] != 0)
1596  break;
1597  }
1598  start[0] = tmp;
1599  }
1600  return 0;
1601 }
1602 
1603 static clib_error_t *
1605  unformat_input_t * input, vlib_cli_command_t * cmd_arg)
1606 {
1607  ip4_address_t v4start, v4end;
1608  ip6_address_t v6start, v6end;
1609  u32 table_id = 0;
1610  int v4set = 0;
1611  int v6set = 0;
1612  int rv;
1613 
1614  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1615  {
1616  if (unformat (input, "%U - %U", unformat_ip4_address, &v4start,
1617  unformat_ip4_address, &v4end))
1618  v4set = 1;
1619  else if (unformat (input, "%U", unformat_ip4_address, &v4start))
1620  {
1621  memcpy (&v4end, &v4start, sizeof (v4start));
1622  v4set = 1;
1623  }
1624  else if (unformat (input, "%U - %U", unformat_ip6_address, &v6start,
1625  unformat_ip6_address, &v6end))
1626  v6set = 1;
1627  else if (unformat (input, "%U", unformat_ip6_address, &v6start))
1628  {
1629  memcpy (&v6end, &v6start, sizeof (v6start));
1630  v6set = 1;
1631  }
1632  else if (unformat (input, "fib-table %d", &table_id))
1633  ;
1634  else
1635  break;
1636  }
1637 
1638  if (!v4set && !v6set)
1639  return clib_error_return (0, "at least one v4 or v6 address required");
1640 
1641  if (v4set)
1642  {
1643  rv = tcp_configure_v4_source_address_range (vm, &v4start, &v4end,
1644  table_id);
1645  switch (rv)
1646  {
1647  case 0:
1648  break;
1649 
1650  case VNET_API_ERROR_NO_SUCH_FIB:
1651  return clib_error_return (0, "Invalid table-id %d", table_id);
1652 
1653  case VNET_API_ERROR_INVALID_ARGUMENT:
1654  return clib_error_return (0, "Invalid address range %U - %U",
1655  format_ip4_address, &v4start,
1656  format_ip4_address, &v4end);
1657  default:
1658  return clib_error_return (0, "error %d", rv);
1659  break;
1660  }
1661  }
1662  if (v6set)
1663  {
1664  rv = tcp_configure_v6_source_address_range (vm, &v6start, &v6end,
1665  table_id);
1666  switch (rv)
1667  {
1668  case 0:
1669  break;
1670 
1671  case VNET_API_ERROR_NO_SUCH_FIB:
1672  return clib_error_return (0, "Invalid table-id %d", table_id);
1673 
1674  default:
1675  return clib_error_return (0, "error %d", rv);
1676  break;
1677  }
1678  }
1679  return 0;
1680 }
1681 
1682 /* *INDENT-OFF* */
1683 VLIB_CLI_COMMAND (tcp_src_address_command, static) =
1684 {
1685  .path = "tcp src-address",
1686  .short_help = "tcp src-address <ip-addr> [- <ip-addr>] add src address range",
1687  .function = tcp_src_address,
1688 };
1689 /* *INDENT-ON* */
1690 
1691 static u8 *
1693 {
1694 #if TCP_SCOREBOARD_TRACE
1695 
1696  scoreboard_trace_elt_t *block;
1697  int i = 0;
1698 
1699  if (!sb->trace)
1700  return s;
1701 
1702  s = format (s, "scoreboard trace:");
1703  vec_foreach (block, sb->trace)
1704  {
1705  s = format (s, "{%u, %u, %u, %u, %u}, ", block->start, block->end,
1706  block->ack, block->snd_una_max, block->group);
1707  if ((++i % 3) == 0)
1708  s = format (s, "\n");
1709  }
1710  return s;
1711 #else
1712  return 0;
1713 #endif
1714 }
1715 
1716 static clib_error_t *
1718  vlib_cli_command_t * cmd_arg)
1719 {
1720  transport_connection_t *tconn = 0;
1721  tcp_connection_t *tc;
1722  u8 *s = 0;
1723  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1724  {
1725  if (unformat (input, "%U", unformat_transport_connection, &tconn,
1727  ;
1728  else
1729  return clib_error_return (0, "unknown input `%U'",
1730  format_unformat_error, input);
1731  }
1732 
1733  if (!TCP_SCOREBOARD_TRACE)
1734  {
1735  vlib_cli_output (vm, "scoreboard tracing not enabled");
1736  return 0;
1737  }
1738 
1739  tc = tcp_get_connection_from_transport (tconn);
1740  s = tcp_scoreboard_dump_trace (s, &tc->sack_sb);
1741  vlib_cli_output (vm, "%v", s);
1742  return 0;
1743 }
1744 
1745 /* *INDENT-OFF* */
1746 VLIB_CLI_COMMAND (tcp_show_scoreboard_trace_command, static) =
1747 {
1748  .path = "show tcp scoreboard trace",
1749  .short_help = "show tcp scoreboard trace <connection>",
1750  .function = tcp_show_scoreboard_trace_fn,
1751 };
1752 /* *INDENT-ON* */
1753 
1754 u8 *
1756 {
1757  int i, trace_len;
1759  u32 next_ack, left, group, has_new_ack = 0;
1760  tcp_connection_t _dummy_tc, *dummy_tc = &_dummy_tc;
1761  sack_block_t *block;
1762 
1763  if (!tc)
1764  return s;
1765 
1766  memset (dummy_tc, 0, sizeof (*dummy_tc));
1767  tcp_connection_timers_init (dummy_tc);
1768  scoreboard_init (&dummy_tc->sack_sb);
1769  dummy_tc->rcv_opts.flags |= TCP_OPTS_FLAG_SACK;
1770 
1771 #if TCP_SCOREBOARD_TRACE
1772  trace = tc->sack_sb.trace;
1773  trace_len = vec_len (tc->sack_sb.trace);
1774 #else
1775  trace = 0;
1776  trace_len = 0;
1777 #endif
1778 
1779  for (i = 0; i < trace_len; i++)
1780  {
1781  if (trace[i].ack != 0)
1782  {
1783  dummy_tc->snd_una = trace[i].ack - 1448;
1784  dummy_tc->snd_una_max = trace[i].ack;
1785  }
1786  }
1787 
1788  left = 0;
1789  while (left < trace_len)
1790  {
1791  group = trace[left].group;
1792  vec_reset_length (dummy_tc->rcv_opts.sacks);
1793  has_new_ack = 0;
1794  while (trace[left].group == group)
1795  {
1796  if (trace[left].ack != 0)
1797  {
1798  if (verbose)
1799  s = format (s, "Adding ack %u, snd_una_max %u, segs: ",
1800  trace[left].ack, trace[left].snd_una_max);
1801  dummy_tc->snd_una_max = trace[left].snd_una_max;
1802  next_ack = trace[left].ack;
1803  has_new_ack = 1;
1804  }
1805  else
1806  {
1807  if (verbose)
1808  s = format (s, "[%u, %u], ", trace[left].start,
1809  trace[left].end);
1810  vec_add2 (dummy_tc->rcv_opts.sacks, block, 1);
1811  block->start = trace[left].start;
1812  block->end = trace[left].end;
1813  }
1814  left++;
1815  }
1816 
1817  /* Push segments */
1818  tcp_rcv_sacks (dummy_tc, next_ack);
1819  if (has_new_ack)
1820  dummy_tc->snd_una = next_ack + dummy_tc->sack_sb.snd_una_adv;
1821 
1822  if (verbose)
1823  s = format (s, "result: %U", format_tcp_scoreboard,
1824  &dummy_tc->sack_sb);
1825 
1826  }
1827  s = format (s, "result: %U", format_tcp_scoreboard, &dummy_tc->sack_sb);
1828 
1829  return s;
1830 }
1831 
1832 static clib_error_t *
1834  vlib_cli_command_t * cmd_arg)
1835 {
1836  transport_connection_t *tconn = 0;
1837  tcp_connection_t *tc = 0;
1838  u8 *str = 0;
1839  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1840  {
1841  if (unformat (input, "%U", unformat_transport_connection, &tconn,
1843  ;
1844  else
1845  return clib_error_return (0, "unknown input `%U'",
1846  format_unformat_error, input);
1847  }
1848 
1849  if (!TCP_SCOREBOARD_TRACE)
1850  {
1851  vlib_cli_output (vm, "scoreboard tracing not enabled");
1852  return 0;
1853  }
1854 
1855  tc = tcp_get_connection_from_transport (tconn);
1856  if (!tc)
1857  {
1858  vlib_cli_output (vm, "connection not found");
1859  return 0;
1860  }
1861  str = tcp_scoreboard_replay (str, tc, 1);
1862  vlib_cli_output (vm, "%v", str);
1863  return 0;
1864 }
1865 
1866 /* *INDENT-OFF* */
1867 VLIB_CLI_COMMAND (tcp_replay_scoreboard_command, static) =
1868 {
1869  .path = "tcp replay scoreboard",
1870  .short_help = "tcp replay scoreboard <connection>",
1871  .function = tcp_scoreboard_trace_fn,
1872 };
1873 /* *INDENT-ON* */
1874 
1875 static clib_error_t *
1877  vlib_cli_command_t * cmd_arg)
1878 {
1879  tcp_main_t *tm = vnet_get_tcp_main ();
1881  return clib_error_return (0, "unknown input `%U'", format_unformat_error,
1882  input);
1883  vlib_cli_output (vm, "IPv4 TCP punt: %s",
1884  tm->punt_unknown4 ? "enabled" : "disabled");
1885  vlib_cli_output (vm, "IPv6 TCP punt: %s",
1886  tm->punt_unknown6 ? "enabled" : "disabled");
1887  return 0;
1888 }
1889 /* *INDENT-OFF* */
1890 VLIB_CLI_COMMAND (show_tcp_punt_command, static) =
1891 {
1892  .path = "show tcp punt",
1893  .short_help = "show tcp punt",
1894  .function = show_tcp_punt_fn,
1895 };
1896 /* *INDENT-ON* */
1897 
1898 /*
1899  * fd.io coding-style-patch-verification: ON
1900  *
1901  * Local Variables:
1902  * eval: (c-set-style "gnu")
1903  * End:
1904  */
static void tcp_session_close(u32 conn_index, u32 thread_index)
Definition: tcp.c:365
#define vec_validate(V, I)
Make sure vector is long enough for given index (no header, unspecified alignment) ...
Definition: vec.h:437
#define tcp_in_cong_recovery(tc)
Definition: tcp.h:350
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:530
static int tcp_session_open(transport_endpoint_t *tep)
Definition: tcp.c:656
fib_protocol_t fp_proto
protocol type
Definition: fib_types.h:212
#define TCP_2MSL_TIME
Definition: tcp.h:101
static vlib_cli_command_t trace
(constructor) VLIB_CLI_COMMAND (trace)
Definition: vlib_api_cli.c:862
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:56
An entry in a FIB table.
Definition: fib_entry.h:458
#define clib_min(x, y)
Definition: clib.h:291
static u8 * format_tcp_listener_session(u8 *s, va_list *args)
Definition: tcp.c:861
virtual circuit service
const char * tcp_fsm_states[]
Definition: tcp.c:667
static clib_error_t * tcp_config_fn(vlib_main_t *vm, unformat_input_t *input)
Definition: tcp.c:1401
#define tcp_in_recovery(tc)
Definition: tcp.h:344
void ip_copy(ip46_address_t *dst, ip46_address_t *src, u8 is_ip4)
Definition: ip.c:81
#define seq_leq(_s1, _s2)
Definition: tcp.h:555
static u32 tcp_session_push_header(transport_connection_t *tconn, vlib_buffer_t *b)
Definition: tcp.c:1110
struct _sack_block sack_block_t
u8 * format_tcp_rcv_sacks(u8 *s, va_list *args)
Definition: tcp.c:901
void ip6_register_protocol(u32 protocol, u32 node_index)
Definition: ip6_forward.c:1427
struct _scoreboard_trace_elt scoreboard_trace_elt_t
u8 * format_tcp_sacks(u8 *s, va_list *args)
Definition: tcp.c:877
vnet_main_t * vnet_get_main(void)
Definition: misc.c:47
struct _transport_connection transport_connection_t
static clib_error_t * ip4_lookup_init(vlib_main_t *vm)
Definition: ip4_forward.c:903
const char * tcp_connection_flags_str[]
Definition: tcp.c:685
vlib_node_registration_t tcp4_output_node
(constructor) VLIB_REGISTER_NODE (tcp4_output_node)
Definition: tcp_output.c:2060
void scoreboard_init(sack_scoreboard_t *sb)
Definition: tcp_input.c:786
uword unformat_transport_connection(unformat_input_t *input, va_list *args)
Definition: session_cli.c:171
static u32 tcp_session_send_space(transport_connection_t *trans_conn)
Definition: tcp.c:1081
#define PREDICT_TRUE(x)
Definition: clib.h:108
static transport_connection_t * tcp_half_open_session_get_transport(u32 conn_index)
Definition: tcp.c:972
tcp_connection_t * tcp_connection_new(u8 thread_index)
Definition: tcp.c:249
u8 as_u8[16]
Definition: ip6_packet.h:48
static void tcp_timer_establish_handler(u32 conn_index)
Definition: tcp.c:1153
void ip_set(ip46_address_t *dst, void *src, u8 is_ip4)
Definition: ip.c:90
static u32 tcp_session_tx_fifo_offset(transport_connection_t *trans_conn)
Definition: tcp.c:1089
#define NULL
Definition: clib.h:57
static u32 ip4_compute_flow_hash(const ip4_header_t *ip, flow_hash_config_t flow_hash_config)
Definition: ip4.h:296
struct _sack_scoreboard sack_scoreboard_t
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:579
static f64 vlib_time_now(vlib_main_t *vm)
Definition: main.h:227
static tcp_connection_t * tcp_half_open_connection_new(void)
Definition: tcp.c:182
static u8 * format_tcp_timers(u8 *s, va_list *args)
Definition: tcp.c:717
static tcp_connection_t * tcp_half_open_connection_get(u32 conn_index)
Definition: tcp.h:523
sack_scoreboard_hole_t * scoreboard_next_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:564
flow_hash_config_t lb_hash_config
the hash config to use when selecting a bucket.
Definition: load_balance.h:134
static_always_inline void clib_spinlock_unlock_if_init(clib_spinlock_t *p)
Definition: lock.h:98
struct _tcp_main tcp_main_t
struct _transport_proto_vft transport_proto_vft_t
void tcp_connection_timers_reset(tcp_connection_t *tc)
Stop all connection timers.
Definition: tcp.c:404
vlib_node_registration_t tcp6_output_node
(constructor) VLIB_REGISTER_NODE (tcp6_output_node)
Definition: tcp_output.c:2081
void dpo_copy(dpo_id_t *dst, const dpo_id_t *src)
atomic copy a data-plane object.
Definition: dpo.c:261
#define vec_add1(V, E)
Add 1 element to end of vector (unspecified alignment).
Definition: vec.h:523
timer_expiration_handler tcp_timer_retransmit_handler
static heap_elt_t * last(heap_header_t *h)
Definition: heap.c:53
#define vec_add2(V, P, N)
Add N elements to end of vector V, return pointer to new elements in P.
Definition: vec.h:562
int i
int ip6_neighbor_proxy_add_del(u32 sw_if_index, ip6_address_t *addr, u8 is_del)
void tcp_update_burst_snd_vars(tcp_connection_t *tc)
Update burst send vars.
Definition: tcp_output.c:417
static vnet_sw_interface_t * vnet_get_sw_interface(vnet_main_t *vnm, u32 sw_if_index)
void * ip_interface_get_first_ip(u32 sw_if_index, u8 is_ip4)
Definition: ip.c:133
struct _tcp_connection tcp_connection_t
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:419
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:643
clib_time_t clib_time
Definition: main.h:63
static u32 tcp_available_snd_wnd(const tcp_connection_t *tc)
Definition: tcp.h:622
static u16 tcp_session_send_mss(transport_connection_t *trans_conn)
Compute maximum segment size for session layer.
Definition: tcp.c:986
Definition: ip.h:109
static tcp_connection_t * tcp_get_connection_from_transport(transport_connection_t *tconn)
Definition: tcp.h:496
void ip4_register_protocol(u32 protocol, u32 node_index)
Definition: ip4_forward.c:1684
static void tcp_initialize_timer_wheels(tcp_main_t *tm)
Definition: tcp.c:1250
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:228
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:170
static void tcp_half_open_connection_del(tcp_connection_t *tc)
Cleanup half-open connection.
Definition: tcp.c:150
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:528
#define TCP_RTO_INIT
Definition: tcp.h:111
format_function_t format_ip4_address
Definition: format.h:75
memset(h->entries, 0, sizeof(h->entries[0])*entries)
unformat_function_t * unformat_pg_edit
Definition: ip.h:90
unformat_function_t unformat_ip4_address
Definition: format.h:70
void tcp_init_snd_vars(tcp_connection_t *tc)
Initialize connection send variables.
Definition: tcp.c:539
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:163
u32 sw_if_index
Definition: vxlan_gbp.api:39
#define always_inline
Definition: clib.h:94
static u8 * format_tcp_vars(u8 *s, va_list *args)
Definition: tcp.c:761
u16 lb_n_buckets_minus_1
number of buckets in the load-balance - 1.
Definition: load_balance.h:99
void tcp_connection_timers_init(tcp_connection_t *tc)
Initialize all connection timers as invalid.
Definition: tcp.c:387
static u32 tcp_available_output_snd_space(const tcp_connection_t *tc)
Definition: tcp.h:628
tcp_main_t tcp_main
Definition: tcp.c:29
static void tcp_connection_unbind(u32 listener_index)
Definition: tcp.c:113
Aggregrate type for a prefix.
Definition: fib_types.h:203
#define clib_error_return(e, args...)
Definition: error.h:99
void stream_session_delete_notify(transport_connection_t *tc)
Notification from transport that connection is being deleted.
Definition: session.c:799
void adj_unlock(adj_index_t adj_index)
Release a reference counting lock on the adjacency.
Definition: adj.c:252
vhost_vring_state_t state
Definition: vhost_user.h:120
static u8 * tcp_scoreboard_dump_trace(u8 *s, sack_scoreboard_t *sb)
Definition: tcp.c:1692
timer_expiration_handler tcp_timer_retransmit_syn_handler
pthread_t thread[MAX_CONNS]
Definition: main.c:142
static u32 tcp_time_now(void)
Definition: tcp.h:677
u32 tcp_snd_space(tcp_connection_t *tc)
Definition: tcp.c:1075
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:1064
u16 fp_len
The mask length.
Definition: fib_types.h:207
#define vlib_call_init_function(vm, x)
Definition: init.h:260
void tcp_api_reference(void)
Definition: tcp_api.c:109
#define TCP_EVT_DBG(_evt, _args...)
Definition: tcp_debug.h:238
fib_node_index_t fib_table_lookup(u32 fib_index, const fib_prefix_t *prefix)
Perfom a longest prefix match in the non-forwarding table.
Definition: fib_table.c:66
static u32 tcp_connection_bind(u32 session_index, transport_endpoint_t *lcl)
Definition: tcp.c:73
static int tcp_connection_open(transport_endpoint_t *rmt)
Definition: tcp.c:609
static i32 tcp_rcv_wnd_available(tcp_connection_t *tc)
Definition: tcp.c:755
static void tcp_cc_init(tcp_connection_t *tc)
Definition: tcp.c:511
static void clib_spinlock_init(clib_spinlock_t *p)
Definition: lock.h:57
static void tcp_timer_set(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:706
static u32 tcp_session_unbind(u32 listener_index)
Definition: tcp.c:130
static timer_expiration_handler * timer_expiration_handlers[TCP_N_TIMERS]
Definition: tcp.c:1218
The identity of a DPO is a combination of its type and its instance number/index of objects of that t...
Definition: dpo.h:168
static clib_error_t * tcp_main_enable(vlib_main_t *vm)
Definition: tcp.c:1264
Definition: fib_entry.h:279
#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:464
timer_expiration_handler tcp_timer_persist_handler
static transport_connection_t * tcp_session_get_transport(u32 conn_index, u32 thread_index)
Definition: tcp.c:965
static ip_protocol_info_t * ip_get_protocol_info(ip_main_t *im, u32 protocol)
Definition: ip.h:136
u32 tcp_push_header(tcp_connection_t *tconn, vlib_buffer_t *b)
Definition: tcp_output.c:1194
#define tcp_in_fastrecovery(tc)
Definition: tcp.h:343
ip46_address_t fp_addr
The address type is not deriveable from the fp_addr member.
Definition: fib_types.h:226
dpo_type_t dpoi_type
the type
Definition: dpo.h:172
format_function_t * format_header
Definition: ip.h:81
static const dpo_id_t * load_balance_get_bucket_i(const load_balance_t *lb, u32 bucket)
Definition: load_balance.h:209
void tcp_rcv_sacks(tcp_connection_t *tc, u32 ack)
Definition: tcp_input.c:829
void tcp_send_syn(tcp_connection_t *tc)
Send SYN.
Definition: tcp_output.c:987
struct _unformat_input_t unformat_input_t
unsigned short u16
Definition: types.h:57
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:137
#define VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX
Definition: buffer.h:440
static u32 ip6_compute_flow_hash(const ip6_header_t *ip, flow_hash_config_t flow_hash_config)
Definition: ip6.h:435
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:274
static void tcp_expired_timers_dispatch(u32 *expired_timers)
Definition: tcp.c:1231
#define TCP_TIMER_HANDLE_INVALID
Definition: tcp.h:93
static void tcp_session_cleanup(u32 conn_index, u32 thread_index)
Definition: tcp.c:373
The FIB DPO provieds;.
Definition: load_balance.h:84
#define TCP_CLEANUP_TIME
Definition: tcp.h:104
static u32 tcp_flight_size(const tcp_connection_t *tc)
Our estimate of the number of bytes in flight (pipe size)
Definition: tcp.h:580
#define PREDICT_FALSE(x)
Definition: clib.h:107
#define VLIB_CONFIG_FUNCTION(x, n,...)
Definition: init.h:172
reliable transport protos
vnet_sw_interface_flags_t flags
Definition: interface.h:720
u8 * format_tcp_scoreboard(u8 *s, va_list *args)
Definition: tcp.c:937
clib_error_t * vnet_tcp_enable_disable(vlib_main_t *vm, u8 is_en)
Definition: tcp.c:1344
f64 seconds_per_clock
Definition: time.h:57
static void tcp_timer_waitclose_handler(u32 conn_index)
Definition: tcp.c:1181
void( timer_expiration_handler)(u32 index)
Definition: tcp.h:86
#define foreach_vlib_main(body)
Definition: threads.h:234
#define TCP_DBG(_fmt, _args...)
Definition: tcp_debug.h:89
u32 fib_entry_get_resolving_interface(fib_node_index_t entry_index)
Definition: fib_entry.c:1409
static void tcp_timer_reset(tcp_connection_t *tc, u8 timer_id)
Definition: tcp.h:717
#define tcp_fastrecovery_sent_1_smss(tc)
Definition: tcp.h:346
unformat_function_t unformat_ip6_address
Definition: format.h:91
clib_error_t * ip_main_init(vlib_main_t *vm)
Definition: ip_init.c:45
fib_node_index_t fib_table_entry_special_dpo_update(u32 fib_index, const fib_prefix_t *prefix, fib_source_t source, fib_entry_flag_t flags, const dpo_id_t *dpo)
Update a &#39;special&#39; entry to the FIB that links to the DPO passed A special entry is an entry that the...
Definition: fib_table.c:346
#define UNFORMAT_END_OF_INPUT
Definition: format.h:144
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:1439
static_always_inline uword vlib_get_thread_index(void)
Definition: threads.h:211
format_function_t format_ip6_address
Definition: format.h:93
vlib_main_t * vm
Definition: buffer.c:294
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:1604
void tcp_connection_close(tcp_connection_t *tc)
Begin connection closing procedure.
Definition: tcp.c:315
#define ENDPOINT_INVALID_INDEX
Definition: transport.h:116
#define clib_warning(format, args...)
Definition: error.h:59
#define clib_memcpy(a, b, c)
Definition: string.h:75
format_function_t format_tcp_header
Definition: format.h:101
const char * tcp_dbg_evt_str[]
Definition: tcp.c:661
static u8 * format_tcp_connection_flags(u8 *s, va_list *args)
Definition: tcp.c:692
u32 fib_node_index_t
A typedef of a node index.
Definition: fib_types.h:30
void tcp_flush_frames_to_output(u8 thread_index)
Flush v4 and v6 tcp and ip-lookup tx frames for thread index.
Definition: tcp_output.c:1058
vlib_node_registration_t tcp6_input_node
(constructor) VLIB_REGISTER_NODE (tcp6_input_node)
Definition: tcp_input.c:3417
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:45
int vnet_proxy_arp_add_del(ip4_address_t *lo_addr, ip4_address_t *hi_addr, u32 fib_index, int is_del)
Definition: arp.c:1994
#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:235
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:173
#define VLIB_CLI_COMMAND(x,...)
Definition: cli.h:155
static void tcp_timer_update(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:730
signed int i32
Definition: types.h:77
#define pool_put_index(p, i)
Free pool element with given index.
Definition: pool.h:299
#define ASSERT(truth)
u32 session_tx_fifo_max_dequeue(transport_connection_t *tc)
Definition: session.c:466
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:1833
unformat_function_t unformat_pg_tcp_header
Definition: format.h:103
enum vnet_link_t_ vnet_link_t
Link Type: A description of the protocol of packets on the link.
static load_balance_t * load_balance_get(index_t lbi)
Definition: load_balance.h:200
#define seq_geq(_s1, _s2)
Definition: tcp.h:557
static u8 * format_tcp_congestion_status(u8 *s, va_list *args)
Definition: tcp.c:742
void stream_session_reset_notify(transport_connection_t *tc)
Notify application that connection has been reset.
Definition: session.c:839
static uword ip6_address_is_link_local_unicast(const ip6_address_t *a)
Definition: ip6_packet.h:321
From the control plane API.
Definition: fib_entry.h:74
void tcp_cc_algo_register(tcp_cc_algorithm_type_e type, const tcp_cc_algorithm_t *vft)
Definition: tcp.c:518
u8 * tcp_scoreboard_replay(u8 *s, tcp_connection_t *tc, u8 verbose)
Definition: tcp.c:1755
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:1531
struct _tcp_cc_algorithm tcp_cc_algorithm_t
Definition: tcp.h:252
#define clib_max(x, y)
Definition: clib.h:284
static clib_error_t * tcp_init(vlib_main_t *vm)
Definition: tcp.c:1372
static void tcp_timer_keep_handler(u32 conn_index)
Definition: tcp.c:1141
dpo_id_t fe_lb
The load-balance used for forwarding.
Definition: fib_entry.h:484
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:1117
void tcp_punt_unknown(vlib_main_t *vm, u8 is_ip4, u8 is_add)
Definition: tcp.c:1362
#define tcp_fastrecovery_1_smss_on(tc)
Definition: tcp.h:347
void tcp_connection_reset(tcp_connection_t *tc)
Notify session that connection has been reset.
Definition: tcp.c:266
vlib_node_registration_t tcp4_input_node
(constructor) VLIB_REGISTER_NODE (tcp4_input_node)
Definition: tcp_input.c:3394
static u32 tcp_session_bind(u32 session_index, transport_endpoint_t *tep)
Definition: tcp.c:107
index_t dpoi_index
the index of objects of that type
Definition: dpo.h:184
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:42
void tcp_send_fin(tcp_connection_t *tc)
Send FIN.
Definition: tcp_output.c:1071
#define foreach_tcp_connection_flag
TCP connection flags.
Definition: tcp.h:114
#define foreach_tcp_timer
TCP timers.
Definition: tcp.h:69
static u8 * format_tcp_connection_id(u8 *s, va_list *args)
Definition: tcp.c:802
#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:1783
static u32 tcp_set_time_now(u32 thread_index)
Definition: tcp.h:683
sack_scoreboard_hole_t * scoreboard_first_hole(sack_scoreboard_t *sb)
Definition: tcp_input.c:580
static u32 tcp_round_snd_space(tcp_connection_t *tc, u32 snd_space)
Definition: tcp.c:999
struct _transport_endpoint transport_endpoint_t
void tcp_connection_init_vars(tcp_connection_t *tc)
Initialize tcp connection variables.
Definition: tcp.c:563
#define DPO_INVALID
An initialiser for DPOs declared on the stack.
Definition: dpo.h:195
const char * tcp_conn_timers[]
Definition: tcp.c:710
static u8 * format_tcp_session(u8 *s, va_list *args)
Definition: tcp.c:845
static u32 vlib_buffer_free_list_buffer_size(vlib_main_t *vm, vlib_buffer_free_list_index_t index)
Definition: buffer_funcs.h:689
typedef prefix
Definition: ip_types.api:40
left
unformat_function_t unformat_memory_size
Definition: format.h:295
u8 * format_tcp_connection(u8 *s, va_list *args)
Definition: tcp.c:826
static u8 * format_tcp_sack_hole(u8 *s, va_list *args)
Definition: tcp.c:925
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:477
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:161
int session_stream_connect_notify(transport_connection_t *tc, u8 is_fail)
Definition: session.c:594
static vlib_thread_main_t * vlib_get_thread_main()
Definition: global_funcs.h:32
void tcp_connection_cleanup(tcp_connection_t *tc)
Cleans up connection state.
Definition: tcp.c:199
void tcp_connection_del(tcp_connection_t *tc)
Connection removal.
Definition: tcp.c:241
void tcp_init_mss(tcp_connection_t *tc)
Definition: tcp_output.c:437
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:214
void dpo_reset(dpo_id_t *dpo)
reset a DPO ID The DPO will be unlocked.
Definition: dpo.c:231
#define vec_foreach(var, vec)
Vector iterator.
int transport_alloc_local_endpoint(u8 proto, transport_endpoint_t *rmt, ip46_address_t *lcl_addr, u16 *lcl_port)
Definition: transport.c:301
#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:869
#define tcp_opts_sack_permitted(_to)
Definition: tcp_packet.h:160
static u32 tcp_snd_space_inline(tcp_connection_t *tc)
Compute tx window session is allowed to fill.
Definition: tcp.c:1026
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:256
u8 * format_tcp_state(u8 *s, va_list *args)
Definition: tcp.c:674
static clib_error_t * ip6_lookup_init(vlib_main_t *vm)
Definition: ip6_forward.c:2593
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:1876
static void tcp_update_time(f64 now, u8 thread_index)
Definition: tcp.c:1100
static tcp_main_t * vnet_get_tcp_main()
Definition: tcp.h:450
static void tcp_cong_recovery_off(tcp_connection_t *tc)
Definition: tcp.h:354
timer_expiration_handler tcp_timer_delack_handler
void vlib_cli_output(vlib_main_t *vm, char *fmt,...)
Definition: cli.c:725
static_always_inline void clib_spinlock_lock_if_init(clib_spinlock_t *p)
Definition: lock.h:82
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:1717
uword unformat(unformat_input_t *i, const char *fmt,...)
Definition: unformat.c:972
static u8 tcp_timer_is_active(tcp_connection_t *tc, tcp_timers_e timer)
Definition: tcp.h:803
static tcp_connection_t * tcp_listener_get(u32 tli)
Definition: tcp.h:517
static uword unformat_check_input(unformat_input_t *i)
Definition: format.h:170
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:97