FD.io VPP  v19.04-6-g6f05f72
Vector Packet Processing
tcp.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016-2019 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14  */
15 
16 /**
17  * @file
18  * @brief TCP host stack utilities
19  */
20 
21 #include <vnet/tcp/tcp.h>
22 #include <vnet/session/session.h>
23 #include <vnet/fib/fib.h>
24 #include <vnet/dpo/load_balance.h>
25 #include <vnet/dpo/receive_dpo.h>
26 #include <vnet/ip/ip6_neighbor.h>
27 #include <math.h>
28 
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;
77  void *iface_ip;
78 
79  pool_get (tm->listener_pool, listener);
80  clib_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  clib_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  clib_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_AO);
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  clib_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  TCP_EVT_DBG (TCP_EVT_DELETE, tc);
204 
205  /* Cleanup local endpoint if this was an active connect */
207  tc->c_lcl_port);
208 
209  /* Check if connection is not yet fully established */
210  if (tc->state == TCP_STATE_SYN_SENT)
211  {
212  /* Try to remove the half-open connection. If this is not the owning
213  * thread, tc won't be removed. Retransmit or establish timers will
214  * eventually expire and call again cleanup on the right thread. */
216  tc->flags |= TCP_CONN_HALF_OPEN_DONE;
217  }
218  else
219  {
220  int thread_index = tc->c_thread_index;
221 
222  /* Make sure all timers are cleared */
224 
225  if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6))
226  tcp_add_del_adjacency (tc, 0);
227 
228  vec_free (tc->snd_sacks);
229  vec_free (tc->snd_sacks_fl);
230 
231  /* Poison the entry */
232  if (CLIB_DEBUG > 0)
233  clib_memset (tc, 0xFA, sizeof (*tc));
234  pool_put (tm->connections[thread_index], tc);
235  }
236 }
237 
238 /**
239  * Connection removal.
240  *
241  * This should be called only once connection enters CLOSED state. Note
242  * that it notifies the session of the removal event, so if the goal is to
243  * just remove the connection, call tcp_connection_cleanup instead.
244  */
245 void
247 {
248  session_transport_delete_notify (&tc->connection);
250 }
251 
253 tcp_connection_alloc (u8 thread_index)
254 {
255  tcp_main_t *tm = vnet_get_tcp_main ();
256  tcp_connection_t *tc;
257 
258  pool_get (tm->connections[thread_index], tc);
259  clib_memset (tc, 0, sizeof (*tc));
260  tc->c_c_index = tc - tm->connections[thread_index];
261  tc->c_thread_index = thread_index;
262  return tc;
263 }
264 
265 void
267 {
268  tcp_main_t *tm = &tcp_main;
269  if (CLIB_DEBUG)
270  {
271  u8 thread_index = tc->c_thread_index;
272  clib_memset (tc, 0xFA, sizeof (*tc));
273  pool_put (tm->connections[thread_index], tc);
274  return;
275  }
276  pool_put (tm->connections[tc->c_thread_index], tc);
277 }
278 
279 /** Notify session that connection has been reset.
280  *
281  * Switch state to closed and wait for session to call cleanup.
282  */
283 void
285 {
286  TCP_EVT_DBG (TCP_EVT_RST_RCVD, tc);
287  switch (tc->state)
288  {
289  case TCP_STATE_SYN_RCVD:
290  /* Cleanup everything. App wasn't notified yet */
291  session_transport_delete_notify (&tc->connection);
293  break;
294  case TCP_STATE_SYN_SENT:
295  session_stream_connect_notify (&tc->connection, 1 /* fail */ );
297  break;
298  case TCP_STATE_ESTABLISHED:
300  /* Set the cleanup timer, in case the session layer/app don't
301  * cleanly close the connection */
302  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME);
303  session_transport_reset_notify (&tc->connection);
304  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
305  break;
306  case TCP_STATE_CLOSE_WAIT:
307  case TCP_STATE_FIN_WAIT_1:
308  case TCP_STATE_FIN_WAIT_2:
309  case TCP_STATE_CLOSING:
310  case TCP_STATE_LAST_ACK:
312  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME);
313  /* Make sure we mark the session as closed. In some states we may
314  * be still trying to send data */
315  session_transport_closed_notify (&tc->connection);
316  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
317  break;
318  case TCP_STATE_CLOSED:
319  case TCP_STATE_TIME_WAIT:
320  break;
321  default:
322  TCP_DBG ("reset state: %u", tc->state);
323  }
324 }
325 
326 /**
327  * Begin connection closing procedure.
328  *
329  * If at the end the connection is not in CLOSED state, it is not removed.
330  * Instead, we rely on on TCP to advance through state machine to either
331  * 1) LAST_ACK (passive close) whereby when the last ACK is received
332  * tcp_connection_del is called. This notifies session of the delete and
333  * calls cleanup.
334  * 2) TIME_WAIT (active close) whereby after 2MSL the 2MSL timer triggers
335  * and cleanup is called.
336  *
337  * N.B. Half-close connections are not supported
338  */
339 void
341 {
342  TCP_EVT_DBG (TCP_EVT_CLOSE, tc);
343 
344  /* Send/Program FIN if needed and switch state */
345  switch (tc->state)
346  {
347  case TCP_STATE_SYN_SENT:
348  /* Try to cleanup. If not on the right thread, mark as half-open done.
349  * Connection will be cleaned up when establish timer pops */
351  break;
352  case TCP_STATE_SYN_RCVD:
354  tcp_send_fin (tc);
355  tcp_connection_set_state (tc, TCP_STATE_FIN_WAIT_1);
356  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_FINWAIT1_TIME);
357  break;
358  case TCP_STATE_ESTABLISHED:
359  /* If closing with unread data, reset the connection */
360  if (transport_max_rx_dequeue (&tc->connection))
361  {
362  tcp_send_reset (tc);
364  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
365  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME);
366  break;
367  }
368  if (!transport_max_tx_dequeue (&tc->connection))
369  tcp_send_fin (tc);
370  else
371  tc->flags |= TCP_CONN_FINPNDG;
372  tcp_connection_set_state (tc, TCP_STATE_FIN_WAIT_1);
373  /* Set a timer in case the peer stops responding. Otherwise the
374  * connection will be stuck here forever. */
375  ASSERT (tc->timers[TCP_TIMER_WAITCLOSE] == TCP_TIMER_HANDLE_INVALID);
376  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_FINWAIT1_TIME);
377  break;
378  case TCP_STATE_CLOSE_WAIT:
379  if (!transport_max_tx_dequeue (&tc->connection))
380  {
381  tcp_send_fin (tc);
383  tcp_connection_set_state (tc, TCP_STATE_LAST_ACK);
384  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
385  }
386  else
387  tc->flags |= TCP_CONN_FINPNDG;
388  break;
389  case TCP_STATE_FIN_WAIT_1:
390  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
391  break;
392  case TCP_STATE_CLOSED:
394  /* Delete connection but instead of doing it now wait until next
395  * dispatch cycle to give the session layer a chance to clear
396  * unhandled events */
397  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
398  break;
399  default:
400  TCP_DBG ("state: %u", tc->state);
401  }
402 }
403 
404 static void
405 tcp_session_close (u32 conn_index, u32 thread_index)
406 {
407  tcp_connection_t *tc;
408  tc = tcp_connection_get (conn_index, thread_index);
410 }
411 
412 static void
413 tcp_session_cleanup (u32 conn_index, u32 thread_index)
414 {
415  tcp_connection_t *tc;
416  tc = tcp_connection_get (conn_index, thread_index);
417  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
419 }
420 
421 /**
422  * Initialize all connection timers as invalid
423  */
424 void
426 {
427  int i;
428 
429  /* Set all to invalid */
430  for (i = 0; i < TCP_N_TIMERS; i++)
431  {
432  tc->timers[i] = TCP_TIMER_HANDLE_INVALID;
433  }
434 
435  tc->rto = TCP_RTO_INIT;
436 }
437 
438 /**
439  * Stop all connection timers
440  */
441 void
443 {
444  int i;
445  for (i = 0; i < TCP_N_TIMERS; i++)
446  {
447  tcp_timer_reset (tc, i);
448  }
449 }
450 
451 #if 0
452 typedef struct ip4_tcp_hdr
453 {
454  ip4_header_t ip;
455  tcp_header_t tcp;
456 } ip4_tcp_hdr_t;
457 
458 typedef struct ip6_tcp_hdr
459 {
460  ip6_header_t ip;
461  tcp_header_t tcp;
462 } ip6_tcp_hdr_t;
463 
464 static void
465 tcp_connection_select_lb_bucket (tcp_connection_t * tc, const dpo_id_t * dpo,
466  dpo_id_t * result)
467 {
468  const dpo_id_t *choice;
469  load_balance_t *lb;
470  int hash;
471 
472  lb = load_balance_get (dpo->dpoi_index);
473  if (tc->c_is_ip4)
474  {
475  ip4_tcp_hdr_t hdr;
476  clib_memset (&hdr, 0, sizeof (hdr));
477  hdr.ip.protocol = IP_PROTOCOL_TCP;
478  hdr.ip.address_pair.src.as_u32 = tc->c_lcl_ip.ip4.as_u32;
479  hdr.ip.address_pair.dst.as_u32 = tc->c_rmt_ip.ip4.as_u32;
480  hdr.tcp.src_port = tc->c_lcl_port;
481  hdr.tcp.dst_port = tc->c_rmt_port;
482  hash = ip4_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
483  }
484  else
485  {
486  ip6_tcp_hdr_t hdr;
487  clib_memset (&hdr, 0, sizeof (hdr));
488  hdr.ip.protocol = IP_PROTOCOL_TCP;
489  clib_memcpy_fast (&hdr.ip.src_address, &tc->c_lcl_ip.ip6,
490  sizeof (ip6_address_t));
491  clib_memcpy_fast (&hdr.ip.dst_address, &tc->c_rmt_ip.ip6,
492  sizeof (ip6_address_t));
493  hdr.tcp.src_port = tc->c_lcl_port;
494  hdr.tcp.dst_port = tc->c_rmt_port;
495  hash = ip6_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
496  }
497  choice = load_balance_get_bucket_i (lb, hash & lb->lb_n_buckets_minus_1);
498  dpo_copy (result, choice);
499 }
500 
503 {
505  u32 fib_index;
506 
507  clib_memcpy_fast (&prefix.fp_addr, &tc->c_rmt_ip, sizeof (prefix.fp_addr));
508  prefix.fp_proto = tc->c_is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6;
509  prefix.fp_len = tc->c_is_ip4 ? 32 : 128;
510  fib_index = fib_table_find (prefix.fp_proto, tc->c_fib_index);
511  return fib_table_lookup (fib_index, &prefix);
512 }
513 
514 static int
515 tcp_connection_stack_on_fib_entry (tcp_connection_t * tc)
516 {
517  dpo_id_t choice = DPO_INVALID;
518  u32 output_node_index;
519  fib_entry_t *fe;
520 
521  fe = fib_entry_get (tc->c_rmt_fei);
522  if (fe->fe_lb.dpoi_type != DPO_LOAD_BALANCE)
523  return -1;
524 
525  tcp_connection_select_lb_bucket (tc, &fe->fe_lb, &choice);
526 
527  output_node_index =
528  tc->c_is_ip4 ? tcp4_output_node.index : tcp6_output_node.index;
529  dpo_stack_from_node (output_node_index, &tc->c_rmt_dpo, &choice);
530  return 0;
531 }
532 
533 /** Stack tcp connection on peer's fib entry.
534  *
535  * This ultimately populates the dpo the connection will use to send packets.
536  */
537 static void
538 tcp_connection_fib_attach (tcp_connection_t * tc)
539 {
540  tc->c_rmt_fei = tcp_lookup_rmt_in_fib (tc);
541 
542  ASSERT (tc->c_rmt_fei != FIB_NODE_INDEX_INVALID);
543 
544  tcp_connection_stack_on_fib_entry (tc);
545 }
546 #endif /* 0 */
547 
548 static void
550 {
551  tc->cc_algo = tcp_cc_algo_get (tcp_main.cc_algo);
552  tc->cc_algo->init (tc);
553 }
554 
555 void
557  const tcp_cc_algorithm_t * vft)
558 {
559  tcp_main_t *tm = vnet_get_tcp_main ();
560  vec_validate (tm->cc_algos, type);
561 
562  tm->cc_algos[type] = *vft;
563  hash_set_mem (tm->cc_algo_by_name, vft->name, type);
564 }
565 
568 {
569  tcp_main_t *tm = vnet_get_tcp_main ();
570  return &tm->cc_algos[type];
571 }
572 
573 /**
574  * Generate random iss as per rfc6528
575  */
576 static u32
578 {
579  tcp_main_t *tm = &tcp_main;
580  u64 tmp;
581 
582  if (tc->c_is_ip4)
583  tmp = (u64) tc->c_lcl_ip.ip4.as_u32 << 32 | (u64) tc->c_rmt_ip.ip4.as_u32;
584  else
585  tmp = tc->c_lcl_ip.ip6.as_u64[0] ^ tc->c_lcl_ip.ip6.as_u64[1]
586  ^ tc->c_rmt_ip.ip6.as_u64[0] ^ tc->c_rmt_ip.ip6.as_u64[1];
587 
588  tmp ^= tm->iss_seed.first | ((u64) tc->c_lcl_port << 16 | tc->c_rmt_port);
589  tmp ^= tm->iss_seed.second;
590  tmp = clib_xxhash (tmp) + clib_cpu_time_now ();
591  return ((tmp >> 32) ^ (tmp & 0xffffffff));
592 }
593 
594 /**
595  * Initialize connection send variables.
596  */
597 void
599 {
600  /*
601  * We use the time to randomize iss and for setting up the initial
602  * timestamp. Make sure it's updated otherwise syn and ack in the
603  * handshake may make it look as if time has flown in the opposite
604  * direction for us.
605  */
607 
608  tc->iss = tcp_generate_random_iss (tc);
609  tc->snd_una = tc->iss;
610  tc->snd_nxt = tc->iss + 1;
611  tc->snd_una_max = tc->snd_nxt;
612  tc->srtt = 0;
613 }
614 
615 void
617 {
618  u32 initial_bucket, byte_rate;
619  initial_bucket = 16 * tc->snd_mss;
620  byte_rate = 2 << 16;
621  transport_connection_tx_pacer_init (&tc->connection, byte_rate,
622  initial_bucket);
623  tc->mrtt_us = (u32) ~ 0;
624 }
625 
626 /** Initialize tcp connection variables
627  *
628  * Should be called after having received a msg from the peer, i.e., a SYN or
629  * a SYNACK, such that connection options have already been exchanged. */
630 void
632 {
634  tcp_init_mss (tc);
635  scoreboard_init (&tc->sack_sb);
636  tcp_cc_init (tc);
637  if (tc->state == TCP_STATE_SYN_RCVD)
638  tcp_init_snd_vars (tc);
639 
640  if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6))
641  tcp_add_del_adjacency (tc, 1);
642 
643  /* tcp_connection_fib_attach (tc); */
644 
645  if (transport_connection_is_tx_paced (&tc->connection)
646  || tcp_main.tx_pacing)
647  tcp_enable_pacing (tc);
648 }
649 
650 static int
651 tcp_alloc_custom_local_endpoint (tcp_main_t * tm, ip46_address_t * lcl_addr,
652  u16 * lcl_port, u8 is_ip4)
653 {
654  int index, port;
655  if (is_ip4)
656  {
657  index = tm->last_v4_address_rotor++;
658  if (tm->last_v4_address_rotor >= vec_len (tm->ip4_src_addresses))
659  tm->last_v4_address_rotor = 0;
660  lcl_addr->ip4.as_u32 = tm->ip4_src_addresses[index].as_u32;
661  }
662  else
663  {
664  index = tm->last_v6_address_rotor++;
665  if (tm->last_v6_address_rotor >= vec_len (tm->ip6_src_addresses))
666  tm->last_v6_address_rotor = 0;
667  clib_memcpy_fast (&lcl_addr->ip6, &tm->ip6_src_addresses[index],
668  sizeof (ip6_address_t));
669  }
671  if (port < 1)
672  {
673  clib_warning ("Failed to allocate src port");
674  return -1;
675  }
676  *lcl_port = port;
677  return 0;
678 }
679 
680 static int
682 {
683  tcp_main_t *tm = vnet_get_tcp_main ();
684  tcp_connection_t *tc;
685  ip46_address_t lcl_addr;
686  u16 lcl_port;
687  int rv;
688 
689  /*
690  * Allocate local endpoint
691  */
692  if ((rmt->is_ip4 && vec_len (tm->ip4_src_addresses))
693  || (!rmt->is_ip4 && vec_len (tm->ip6_src_addresses)))
694  rv = tcp_alloc_custom_local_endpoint (tm, &lcl_addr, &lcl_port,
695  rmt->is_ip4);
696  else
698  rmt, &lcl_addr, &lcl_port);
699 
700  if (rv)
701  return -1;
702 
703  /*
704  * Create connection and send SYN
705  */
706  clib_spinlock_lock_if_init (&tm->half_open_lock);
708  ip_copy (&tc->c_rmt_ip, &rmt->ip, rmt->is_ip4);
709  ip_copy (&tc->c_lcl_ip, &lcl_addr, rmt->is_ip4);
710  tc->c_rmt_port = rmt->port;
711  tc->c_lcl_port = clib_host_to_net_u16 (lcl_port);
712  tc->c_is_ip4 = rmt->is_ip4;
713  tc->c_proto = TRANSPORT_PROTO_TCP;
714  tc->c_fib_index = rmt->fib_index;
715  /* The other connection vars will be initialized after SYN ACK */
717 
718  TCP_EVT_DBG (TCP_EVT_OPEN, tc);
719  tc->state = TCP_STATE_SYN_SENT;
720  tcp_init_snd_vars (tc);
721  tcp_send_syn (tc);
722  clib_spinlock_unlock_if_init (&tm->half_open_lock);
723 
724  return tc->c_c_index;
725 }
726 
727 const char *tcp_dbg_evt_str[] = {
728 #define _(sym, str) str,
730 #undef _
731 };
732 
733 const char *tcp_fsm_states[] = {
734 #define _(sym, str) str,
736 #undef _
737 };
738 
739 u8 *
740 format_tcp_state (u8 * s, va_list * args)
741 {
742  u32 state = va_arg (*args, u32);
743 
744  if (state < TCP_N_STATES)
745  s = format (s, "%s", tcp_fsm_states[state]);
746  else
747  s = format (s, "UNKNOWN (%d (0x%x))", state, state);
748  return s;
749 }
750 
751 const char *tcp_connection_flags_str[] = {
752 #define _(sym, str) str,
754 #undef _
755 };
756 
757 static u8 *
758 format_tcp_connection_flags (u8 * s, va_list * args)
759 {
760  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
761  int i, last = -1;
762 
763  for (i = 0; i < TCP_CONN_N_FLAG_BITS; i++)
764  if (tc->flags & (1 << i))
765  last = i;
766  for (i = 0; i < last; i++)
767  {
768  if (tc->flags & (1 << i))
769  s = format (s, "%s, ", tcp_connection_flags_str[i]);
770  }
771  if (last >= 0)
772  s = format (s, "%s", tcp_connection_flags_str[last]);
773  return s;
774 }
775 
776 const char *tcp_conn_timers[] = {
777 #define _(sym, str) str,
779 #undef _
780 };
781 
782 static u8 *
783 format_tcp_timers (u8 * s, va_list * args)
784 {
785  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
786  int i, last = -1;
787 
788  for (i = 0; i < TCP_N_TIMERS; i++)
789  if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
790  last = i;
791 
792  for (i = 0; i < last; i++)
793  {
794  if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
795  s = format (s, "%s,", tcp_conn_timers[i]);
796  }
797 
798  if (last >= 0)
799  s = format (s, "%s", tcp_conn_timers[i]);
800 
801  return s;
802 }
803 
804 static u8 *
805 format_tcp_congestion_status (u8 * s, va_list * args)
806 {
807  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
808  if (tcp_in_recovery (tc))
809  s = format (s, "recovery");
810  else if (tcp_in_fastrecovery (tc))
811  s = format (s, "fastrecovery");
812  else
813  s = format (s, "none");
814  return s;
815 }
816 
817 static i32
819 {
820  return (i32) tc->rcv_wnd - (tc->rcv_nxt - tc->rcv_las);
821 }
822 
823 static u8 *
824 format_tcp_congestion (u8 * s, va_list * args)
825 {
826  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
827  u32 indent = format_get_indent (s);
828 
829  s = format (s, "%U ", format_tcp_congestion_status, tc);
830  s = format (s, "algo %s cwnd %u ssthresh %u bytes_acked %u\n",
831  tc->cc_algo->name, tc->cwnd, tc->ssthresh, tc->bytes_acked);
832  s = format (s, "%Ucc space %u prev_cwnd %u prev_ssthresh %u rtx_bytes %u\n",
834  tc->prev_cwnd, tc->prev_ssthresh, tc->snd_rxt_bytes);
835  s = format (s, "%Usnd_congestion %u dupack %u limited_transmit %u\n",
836  format_white_space, indent, tc->snd_congestion - tc->iss,
837  tc->rcv_dupacks, tc->limited_transmit - tc->iss);
838  return s;
839 }
840 
841 static u8 *
842 format_tcp_vars (u8 * s, va_list * args)
843 {
844  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
845  s = format (s, " index: %u flags: %U timers: %U\n", tc->c_c_index,
847  s = format (s, " snd_una %u snd_nxt %u snd_una_max %u",
848  tc->snd_una - tc->iss, tc->snd_nxt - tc->iss,
849  tc->snd_una_max - tc->iss);
850  s = format (s, " rcv_nxt %u rcv_las %u\n",
851  tc->rcv_nxt - tc->irs, tc->rcv_las - tc->irs);
852  s = format (s, " snd_wnd %u rcv_wnd %u rcv_wscale %u ",
853  tc->snd_wnd, tc->rcv_wnd, tc->rcv_wscale);
854  s = format (s, "snd_wl1 %u snd_wl2 %u\n", tc->snd_wl1 - tc->irs,
855  tc->snd_wl2 - tc->iss);
856  s = format (s, " flight size %u out space %u rcv_wnd_av %u",
858  tcp_rcv_wnd_available (tc));
859  s = format (s, " tsval_recent %u\n", tc->tsval_recent);
860  s = format (s, " tsecr %u tsecr_last_ack %u tsval_recent_age %u",
861  tc->rcv_opts.tsecr, tc->tsecr_last_ack,
862  tcp_time_now () - tc->tsval_recent_age);
863  s = format (s, " snd_mss %u\n", tc->snd_mss);
864  s = format (s, " rto %u rto_boff %u srtt %u us %.3f rttvar %u rtt_ts %.4f",
865  tc->rto, tc->rto_boff, tc->srtt, tc->mrtt_us * 1000, tc->rttvar,
866  tc->rtt_ts);
867  s = format (s, " rtt_seq %u\n", tc->rtt_seq - tc->iss);
868  s = format (s, " cong: %U", format_tcp_congestion, tc);
869 
870  if (tc->state >= TCP_STATE_ESTABLISHED)
871  {
872  s = format (s, " sboard: %U\n", format_tcp_scoreboard, &tc->sack_sb,
873  tc);
874  }
875  if (vec_len (tc->snd_sacks))
876  s = format (s, " sacks tx: %U\n", format_tcp_sacks, tc);
877 
878  return s;
879 }
880 
881 static u8 *
882 format_tcp_connection_id (u8 * s, va_list * args)
883 {
884  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
885  if (!tc)
886  return s;
887  if (tc->c_is_ip4)
888  {
889  s = format (s, "[%d:%d][%s] %U:%d->%U:%d", tc->c_thread_index,
890  tc->c_s_index, "T", format_ip4_address, &tc->c_lcl_ip4,
891  clib_net_to_host_u16 (tc->c_lcl_port), format_ip4_address,
892  &tc->c_rmt_ip4, clib_net_to_host_u16 (tc->c_rmt_port));
893  }
894  else
895  {
896  s = format (s, "[%d:%d][%s] %U:%d->%U:%d", tc->c_thread_index,
897  tc->c_s_index, "T", format_ip6_address, &tc->c_lcl_ip6,
898  clib_net_to_host_u16 (tc->c_lcl_port), format_ip6_address,
899  &tc->c_rmt_ip6, clib_net_to_host_u16 (tc->c_rmt_port));
900  }
901 
902  return s;
903 }
904 
905 u8 *
906 format_tcp_connection (u8 * s, va_list * args)
907 {
908  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
909  u32 verbose = va_arg (*args, u32);
910 
911  if (!tc)
912  return s;
913  s = format (s, "%-50U", format_tcp_connection_id, tc);
914  if (verbose)
915  {
916  s = format (s, "%-15U", format_tcp_state, tc->state);
917  if (verbose > 1)
918  s = format (s, "\n%U", format_tcp_vars, tc);
919  }
920 
921  return s;
922 }
923 
924 static u8 *
925 format_tcp_session (u8 * s, va_list * args)
926 {
927  u32 tci = va_arg (*args, u32);
928  u32 thread_index = va_arg (*args, u32);
929  u32 verbose = va_arg (*args, u32);
930  tcp_connection_t *tc;
931 
932  tc = tcp_connection_get (tci, thread_index);
933  if (tc)
934  s = format (s, "%U", format_tcp_connection, tc, verbose);
935  else
936  s = format (s, "empty\n");
937  return s;
938 }
939 
940 static u8 *
941 format_tcp_listener_session (u8 * s, va_list * args)
942 {
943  u32 tci = va_arg (*args, u32);
944  u32 verbose = va_arg (*args, u32);
946  s = format (s, "%-50U", format_tcp_connection_id, tc);
947  if (verbose)
948  s = format (s, "%-15U", format_tcp_state, tc->state);
949  return s;
950 }
951 
952 static u8 *
953 format_tcp_half_open_session (u8 * s, va_list * args)
954 {
955  u32 tci = va_arg (*args, u32);
957  return format (s, "%U", format_tcp_connection_id, tc);
958 }
959 
960 u8 *
961 format_tcp_sacks (u8 * s, va_list * args)
962 {
963  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
964  sack_block_t *sacks = tc->snd_sacks;
965  sack_block_t *block;
966  int i, len = 0;
967 
968  len = vec_len (sacks);
969  for (i = 0; i < len - 1; i++)
970  {
971  block = &sacks[i];
972  s = format (s, " start %u end %u\n", block->start - tc->irs,
973  block->end - tc->irs);
974  }
975  if (len)
976  {
977  block = &sacks[len - 1];
978  s = format (s, " start %u end %u", block->start - tc->irs,
979  block->end - tc->irs);
980  }
981  return s;
982 }
983 
984 u8 *
985 format_tcp_rcv_sacks (u8 * s, va_list * args)
986 {
987  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
988  sack_block_t *sacks = tc->rcv_opts.sacks;
989  sack_block_t *block;
990  int i, len = 0;
991 
992  len = vec_len (sacks);
993  for (i = 0; i < len - 1; i++)
994  {
995  block = &sacks[i];
996  s = format (s, " start %u end %u\n", block->start - tc->iss,
997  block->end - tc->iss);
998  }
999  if (len)
1000  {
1001  block = &sacks[len - 1];
1002  s = format (s, " start %u end %u", block->start - tc->iss,
1003  block->end - tc->iss);
1004  }
1005  return s;
1006 }
1007 
1008 static u8 *
1009 format_tcp_sack_hole (u8 * s, va_list * args)
1010 {
1011  sack_scoreboard_hole_t *hole = va_arg (*args, sack_scoreboard_hole_t *);
1012  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1013  if (tc)
1014  s = format (s, " [%u, %u]", hole->start - tc->iss, hole->end - tc->iss);
1015  else
1016  s = format (s, " [%u, %u]", hole->start, hole->end);
1017  return s;
1018 }
1019 
1020 u8 *
1021 format_tcp_scoreboard (u8 * s, va_list * args)
1022 {
1023  sack_scoreboard_t *sb = va_arg (*args, sack_scoreboard_t *);
1024  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1025  sack_scoreboard_hole_t *hole;
1026  u32 indent = format_get_indent (s);
1027 
1028  s = format (s, "sacked_bytes %u last_sacked_bytes %u lost_bytes %u\n",
1029  sb->sacked_bytes, sb->last_sacked_bytes, sb->lost_bytes);
1030  s = format (s, "%Ulast_bytes_delivered %u high_sacked %u snd_una_adv %u\n",
1031  format_white_space, indent, sb->last_bytes_delivered,
1032  sb->high_sacked - tc->iss, sb->snd_una_adv);
1033  s = format (s, "%Ucur_rxt_hole %u high_rxt %u rescue_rxt %u",
1034  format_white_space, indent, sb->cur_rxt_hole,
1035  sb->high_rxt - tc->iss, sb->rescue_rxt - tc->iss);
1036 
1037  hole = scoreboard_first_hole (sb);
1038  if (hole)
1039  s = format (s, "\n%Uhead %u tail %u %u holes:\n%U", format_white_space,
1040  indent, sb->head, sb->tail, pool_elts (sb->holes),
1041  format_white_space, indent);
1042 
1043  while (hole)
1044  {
1045  s = format (s, "%U", format_tcp_sack_hole, hole, tc);
1046  hole = scoreboard_next_hole (sb, hole);
1047  }
1048 
1049  return s;
1050 }
1051 
1052 static transport_connection_t *
1053 tcp_session_get_transport (u32 conn_index, u32 thread_index)
1054 {
1055  tcp_connection_t *tc = tcp_connection_get (conn_index, thread_index);
1056  return &tc->connection;
1057 }
1058 
1059 static transport_connection_t *
1061 {
1063  return &tc->connection;
1064 }
1065 
1066 /**
1067  * Compute maximum segment size for session layer.
1068  *
1069  * Since the result needs to be the actual data length, it first computes
1070  * the tcp options to be used in the next burst and subtracts their
1071  * length from the connection's snd_mss.
1072  */
1073 static u16
1075 {
1076  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1077 
1078  /* Ensure snd_mss does accurately reflect the amount of data we can push
1079  * in a segment. This also makes sure that options are updated according to
1080  * the current state of the connection. */
1082 
1083  return tc->snd_mss;
1084 }
1085 
1088 {
1089  if (PREDICT_FALSE (tc->snd_wnd < tc->snd_mss))
1090  {
1091  return tc->snd_wnd <= snd_space ? tc->snd_wnd : 0;
1092  }
1093 
1094  /* If not snd_wnd constrained and we can't write at least a segment,
1095  * don't try at all */
1096  if (PREDICT_FALSE (snd_space < tc->snd_mss))
1097  return snd_space < tc->cwnd ? 0 : snd_space;
1098 
1099  /* round down to mss multiple */
1100  return snd_space - (snd_space % tc->snd_mss);
1101 }
1102 
1103 /**
1104  * Compute tx window session is allowed to fill.
1105  *
1106  * Takes into account available send space, snd_mss and the congestion
1107  * state of the connection. If possible, the value returned is a multiple
1108  * of snd_mss.
1109  *
1110  * @param tc tcp connection
1111  * @return number of bytes session is allowed to write
1112  */
1113 static inline u32
1115 {
1116  int snd_space, snt_limited;
1117 
1119  || tc->state == TCP_STATE_CLOSED))
1120  return 0;
1121 
1122  snd_space = tcp_available_output_snd_space (tc);
1123 
1124  /* If we haven't gotten dupacks or if we did and have gotten sacked
1125  * bytes then we can still send as per Limited Transmit (RFC3042) */
1126  if (PREDICT_FALSE (tc->rcv_dupacks != 0
1127  && (tcp_opts_sack_permitted (tc)
1128  && tc->sack_sb.last_sacked_bytes == 0)))
1129  {
1130  if (tc->rcv_dupacks == 1 && tc->limited_transmit != tc->snd_nxt)
1131  tc->limited_transmit = tc->snd_nxt;
1132  ASSERT (seq_leq (tc->limited_transmit, tc->snd_nxt));
1133 
1134  snt_limited = tc->snd_nxt - tc->limited_transmit;
1135  snd_space = clib_max (2 * tc->snd_mss - snt_limited, 0);
1136  }
1137  return tcp_round_snd_space (tc, snd_space);
1138 }
1139 
1140 u32
1142 {
1143  return tcp_snd_space_inline (tc);
1144 }
1145 
1146 static u32
1148 {
1149  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1150  return clib_min (tcp_snd_space_inline (tc),
1151  tc->snd_wnd - (tc->snd_nxt - tc->snd_una));
1152 }
1153 
1154 static u32
1156 {
1157  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1158 
1159  ASSERT (seq_geq (tc->snd_nxt, tc->snd_una));
1160 
1161  /* This still works if fast retransmit is on */
1162  return (tc->snd_nxt - tc->snd_una);
1163 }
1164 
1165 static void
1166 tcp_update_time (f64 now, u8 thread_index)
1167 {
1168  tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index);
1169 
1170  tcp_set_time_now (wrk);
1171  tw_timer_expire_timers_16t_2w_512sl (&wrk->timer_wheel, now);
1172  tcp_do_fastretransmits (wrk);
1173  tcp_send_acks (wrk);
1175 }
1176 
1177 static void
1179 {
1180  tcp_connection_t *tc = (tcp_connection_t *) tconn;
1181  if (tc->flags & TCP_CONN_PSH_PENDING)
1182  return;
1183  tc->flags |= TCP_CONN_PSH_PENDING;
1184  tc->psh_seq = tc->snd_una + transport_max_tx_dequeue (tconn) - 1;
1185 }
1186 
1187 /* *INDENT-OFF* */
1189  .enable = vnet_tcp_enable_disable,
1190  .start_listen = tcp_session_bind,
1191  .stop_listen = tcp_session_unbind,
1192  .push_header = tcp_session_push_header,
1193  .get_connection = tcp_session_get_transport,
1194  .get_listener = tcp_session_get_listener,
1195  .get_half_open = tcp_half_open_session_get_transport,
1196  .connect = tcp_session_open,
1197  .close = tcp_session_close,
1198  .cleanup = tcp_session_cleanup,
1199  .send_mss = tcp_session_send_mss,
1200  .send_space = tcp_session_send_space,
1201  .update_time = tcp_update_time,
1202  .tx_fifo_offset = tcp_session_tx_fifo_offset,
1203  .flush_data = tcp_session_flush_data,
1204  .format_connection = format_tcp_session,
1205  .format_listener = format_tcp_listener_session,
1206  .format_half_open = format_tcp_half_open_session,
1207  .tx_type = TRANSPORT_TX_PEEK,
1208  .service_type = TRANSPORT_SERVICE_VC,
1209 };
1210 /* *INDENT-ON* */
1211 
1212 void
1214 {
1215  f64 srtt;
1216  u64 rate;
1217 
1218  if (!transport_connection_is_tx_paced (&tc->connection))
1219  return;
1220 
1221  srtt = clib_min ((f64) tc->srtt * TCP_TICK, tc->mrtt_us);
1222  /* TODO should constrain to interface's max throughput but
1223  * we don't have link speeds for sw ifs ..*/
1224  rate = tc->cwnd / srtt;
1225  transport_connection_tx_pacer_update (&tc->connection, rate);
1226 }
1227 
1228 void
1230  u32 start_bucket)
1231 {
1232  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1233  u32 byte_rate = window / ((f64) TCP_TICK * tc->srtt);
1234  u64 last_time = wrk->vm->clib_time.last_cpu_time;
1235  transport_connection_tx_pacer_reset (&tc->connection, byte_rate,
1236  start_bucket, last_time);
1237 }
1238 
1239 static void
1241 {
1242  u32 thread_index = vlib_get_thread_index ();
1243  tcp_connection_t *tc;
1244 
1245  tc = tcp_connection_get (conn_index, thread_index);
1246  tc->timers[TCP_TIMER_KEEP] = TCP_TIMER_HANDLE_INVALID;
1247 
1248  tcp_connection_close (tc);
1249 }
1250 
1251 static void
1253 {
1254  tcp_connection_t *tc;
1255 
1256  tc = tcp_connection_get (conn_index, vlib_get_thread_index ());
1257  /* note: the connection may have already disappeared */
1258  if (PREDICT_FALSE (tc == 0))
1259  return;
1260  ASSERT (tc->state == TCP_STATE_SYN_RCVD);
1261  tc->timers[TCP_TIMER_ESTABLISH] = TCP_TIMER_HANDLE_INVALID;
1262  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1264  /* Start cleanup. Do NOT delete the session until we do the connection
1265  * cleanup. Otherwise, we end up with a dangling session index in the
1266  * tcp connection. */
1267  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
1268 }
1269 
1270 static void
1272 {
1273  tcp_connection_t *tc;
1274 
1275  tc = tcp_half_open_connection_get (conn_index);
1276  if (!tc)
1277  return;
1278 
1279  ASSERT (tc->state == TCP_STATE_SYN_SENT);
1280  /* Notify app if we haven't tried to clean this up already */
1281  if (!(tc->flags & TCP_CONN_HALF_OPEN_DONE))
1282  session_stream_connect_notify (&tc->connection, 1 /* fail */ );
1283 
1284  tc->timers[TCP_TIMER_ESTABLISH_AO] = TCP_TIMER_HANDLE_INVALID;
1286 }
1287 
1288 static void
1290 {
1291  u32 thread_index = vlib_get_thread_index ();
1292  tcp_connection_t *tc;
1293 
1294  tc = tcp_connection_get (conn_index, thread_index);
1295  if (!tc)
1296  return;
1297  tc->timers[TCP_TIMER_WAITCLOSE] = TCP_TIMER_HANDLE_INVALID;
1298 
1299  switch (tc->state)
1300  {
1301  case TCP_STATE_CLOSE_WAIT:
1303  session_transport_closed_notify (&tc->connection);
1304 
1305  if (!(tc->flags & TCP_CONN_FINPNDG))
1306  {
1307  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1308  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
1309  break;
1310  }
1311 
1312  /* Session didn't come back with a close. Send FIN either way
1313  * and switch to LAST_ACK. */
1314  tcp_cong_recovery_off (tc);
1315  /* Make sure we don't try to send unsent data */
1316  tc->snd_nxt = tc->snd_una;
1317  tcp_send_fin (tc);
1318  tcp_connection_set_state (tc, TCP_STATE_LAST_ACK);
1319 
1320  /* Make sure we don't wait in LAST ACK forever */
1321  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
1322 
1323  /* Don't delete the connection yet */
1324  break;
1325  case TCP_STATE_FIN_WAIT_1:
1327  if (tc->flags & TCP_CONN_FINPNDG)
1328  {
1329  /* If FIN pending, we haven't sent everything, but we did try.
1330  * Notify session layer that transport is closed. */
1331  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1332  session_transport_closed_notify (&tc->connection);
1333  tcp_send_reset (tc);
1334  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
1335  }
1336  else
1337  {
1338  /* We've sent the fin but no progress. Close the connection and
1339  * to make sure everything is flushed, setup a cleanup timer */
1340  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1341  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
1342  }
1343  break;
1344  case TCP_STATE_LAST_ACK:
1345  case TCP_STATE_CLOSING:
1347  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1348  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
1349  session_transport_closed_notify (&tc->connection);
1350  break;
1351  default:
1352  tcp_connection_del (tc);
1353  break;
1354  }
1355 }
1356 
1357 /* *INDENT-OFF* */
1359 {
1368 };
1369 /* *INDENT-ON* */
1370 
1371 static void
1373 {
1374  int i;
1375  u32 connection_index, timer_id;
1376 
1377  for (i = 0; i < vec_len (expired_timers); i++)
1378  {
1379  /* Get session index and timer id */
1380  connection_index = expired_timers[i] & 0x0FFFFFFF;
1381  timer_id = expired_timers[i] >> 28;
1382 
1383  TCP_EVT_DBG (TCP_EVT_TIMER_POP, connection_index, timer_id);
1384 
1385  /* Handle expiration */
1386  (*timer_expiration_handlers[timer_id]) (connection_index);
1387  }
1388 }
1389 
1390 static void
1392 {
1393  tw_timer_wheel_16t_2w_512sl_t *tw;
1394  /* *INDENT-OFF* */
1395  foreach_vlib_main (({
1396  tw = &tm->wrk_ctx[ii].timer_wheel;
1397  tw_timer_wheel_init_16t_2w_512sl (tw, tcp_expired_timers_dispatch,
1398  100e-3 /* timer period 100ms */ , ~0);
1399  tw->last_run_time = vlib_time_now (this_vlib_main);
1400  }));
1401  /* *INDENT-ON* */
1402 }
1403 
1404 static void
1406 {
1407  u32 default_seed = random_default_seed ();
1408  u64 time_now = clib_cpu_time_now ();
1409 
1410  tm->iss_seed.first = (u64) random_u32 (&default_seed) << 32;
1411  tm->iss_seed.second = random_u64 (&time_now);
1412 }
1413 
1414 static clib_error_t *
1416 {
1418  u32 num_threads, n_workers, prealloc_conn_per_wrk;
1419  tcp_connection_t *tc __attribute__ ((unused));
1420  tcp_main_t *tm = vnet_get_tcp_main ();
1421  clib_error_t *error = 0;
1422  int thread;
1423 
1424  if ((error = vlib_call_init_function (vm, ip_main_init)))
1425  return error;
1426  if ((error = vlib_call_init_function (vm, ip4_lookup_init)))
1427  return error;
1428  if ((error = vlib_call_init_function (vm, ip6_lookup_init)))
1429  return error;
1430 
1431  /*
1432  * Registrations
1433  */
1434 
1435  ip4_register_protocol (IP_PROTOCOL_TCP, tcp4_input_node.index);
1436  ip6_register_protocol (IP_PROTOCOL_TCP, tcp6_input_node.index);
1437 
1438  /*
1439  * Initialize data structures
1440  */
1441 
1442  num_threads = 1 /* main thread */ + vtm->n_threads;
1443  vec_validate (tm->connections, num_threads - 1);
1444  vec_validate (tm->wrk_ctx, num_threads - 1);
1445  n_workers = num_threads == 1 ? 1 : vtm->n_threads;
1446  prealloc_conn_per_wrk = tm->preallocated_connections / n_workers;
1447 
1448  for (thread = 0; thread < num_threads; thread++)
1449  {
1450  vec_validate (tm->wrk_ctx[thread].pending_fast_rxt, 255);
1451  vec_validate (tm->wrk_ctx[thread].ongoing_fast_rxt, 255);
1452  vec_validate (tm->wrk_ctx[thread].postponed_fast_rxt, 255);
1453  vec_validate (tm->wrk_ctx[thread].pending_deq_acked, 255);
1454  vec_validate (tm->wrk_ctx[thread].pending_acks, 255);
1455  vec_validate (tm->wrk_ctx[thread].pending_disconnects, 255);
1456  vec_reset_length (tm->wrk_ctx[thread].pending_fast_rxt);
1457  vec_reset_length (tm->wrk_ctx[thread].ongoing_fast_rxt);
1458  vec_reset_length (tm->wrk_ctx[thread].postponed_fast_rxt);
1459  vec_reset_length (tm->wrk_ctx[thread].pending_deq_acked);
1460  vec_reset_length (tm->wrk_ctx[thread].pending_acks);
1461  vec_reset_length (tm->wrk_ctx[thread].pending_disconnects);
1462  tm->wrk_ctx[thread].vm = vlib_mains[thread];
1463 
1464  /*
1465  * Preallocate connections. Assume that thread 0 won't
1466  * use preallocated threads when running multi-core
1467  */
1468  if ((thread > 0 || num_threads == 1) && prealloc_conn_per_wrk)
1469  pool_init_fixed (tm->connections[thread], prealloc_conn_per_wrk);
1470  }
1471 
1472  /*
1473  * Use a preallocated half-open connection pool?
1474  */
1475  if (tm->preallocated_half_open_connections)
1476  pool_init_fixed (tm->half_open_connections,
1477  tm->preallocated_half_open_connections);
1478 
1479  /* Initialize clocks per tick for TCP timestamp. Used to compute
1480  * monotonically increasing timestamps. */
1481  tm->tstamp_ticks_per_clock = vm->clib_time.seconds_per_clock
1483 
1484  if (num_threads > 1)
1485  {
1486  clib_spinlock_init (&tm->half_open_lock);
1487  }
1488 
1491 
1492  tm->bytes_per_buffer = vlib_buffer_get_default_data_size (vm);
1493 
1494  return error;
1495 }
1496 
1497 clib_error_t *
1499 {
1500  if (is_en)
1501  {
1502  if (tcp_main.is_enabled)
1503  return 0;
1504 
1505  return tcp_main_enable (vm);
1506  }
1507  else
1508  {
1509  tcp_main.is_enabled = 0;
1510  }
1511 
1512  return 0;
1513 }
1514 
1515 void
1517 {
1518  tcp_main_t *tm = &tcp_main;
1519  if (is_ip4)
1520  tm->punt_unknown4 = is_add;
1521  else
1522  tm->punt_unknown6 = is_add;
1523 }
1524 
1525 static clib_error_t *
1527 {
1528  tcp_main_t *tm = vnet_get_tcp_main ();
1529  ip_main_t *im = &ip_main;
1530  ip_protocol_info_t *pi;
1531 
1532  /* Session layer, and by implication tcp, are disabled by default */
1533  tm->is_enabled = 0;
1534 
1535  /* Register with IP for header parsing */
1536  pi = ip_get_protocol_info (im, IP_PROTOCOL_TCP);
1537  if (pi == 0)
1538  return clib_error_return (0, "TCP protocol info AWOL");
1541 
1542  /* Register as transport with session layer */
1547 
1548  tcp_api_reference ();
1549  tm->cc_algo_by_name = hash_create_string (0, sizeof (uword));
1550  tm->tx_pacing = 1;
1551  tm->cc_algo = TCP_CC_NEWRENO;
1552  tm->default_mtu = 1460;
1553  return 0;
1554 }
1555 
1557 
1558 uword
1560 {
1561  uword *result = va_arg (*va, uword *);
1562  tcp_main_t *tm = &tcp_main;
1563  char *cc_algo_name;
1564  u8 found = 0;
1565  uword *p;
1566 
1567  if (unformat (input, "%s", &cc_algo_name)
1568  && ((p = hash_get_mem (tm->cc_algo_by_name, cc_algo_name))))
1569  {
1570  *result = *p;
1571  found = 1;
1572  }
1573 
1574  vec_free (cc_algo_name);
1575  return found;
1576 }
1577 
1578 uword
1580 {
1581  tcp_main_t *tm = vnet_get_tcp_main ();
1582  tcp_cc_algorithm_t *cc_alg;
1583  unformat_input_t sub_input;
1584  int found = 0;
1585 
1586  vec_foreach (cc_alg, tm->cc_algos)
1587  {
1588  if (!unformat (input, cc_alg->name))
1589  continue;
1590 
1591  if (cc_alg->unformat_cfg
1592  && unformat (input, "%U", unformat_vlib_cli_sub_input, &sub_input))
1593  {
1594  if (cc_alg->unformat_cfg (&sub_input))
1595  found = 1;
1596  }
1597  }
1598  return found;
1599 }
1600 
1601 static clib_error_t *
1603 {
1604  tcp_main_t *tm = vnet_get_tcp_main ();
1605 
1606  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1607  {
1608  if (unformat (input, "preallocated-connections %d",
1609  &tm->preallocated_connections))
1610  ;
1611  else if (unformat (input, "preallocated-half-open-connections %d",
1612  &tm->preallocated_half_open_connections))
1613  ;
1614  else if (unformat (input, "buffer-fail-fraction %f",
1615  &tm->buffer_fail_fraction))
1616  ;
1617  else if (unformat (input, "max-rx-fifo %U", unformat_memory_size,
1618  &tm->max_rx_fifo))
1619  ;
1620  else if (unformat (input, "mtu %d", &tm->default_mtu))
1621  ;
1622  else if (unformat (input, "no-tx-pacing"))
1623  tm->tx_pacing = 0;
1624  else if (unformat (input, "cc-algo %U", unformat_tcp_cc_algo,
1625  &tm->cc_algo))
1626  ;
1627  else if (unformat (input, "%U", unformat_tcp_cc_algo_cfg))
1628  ;
1629  else
1630  return clib_error_return (0, "unknown input `%U'",
1631  format_unformat_error, input);
1632  }
1633  return 0;
1634 }
1635 
1637 
1638 
1639 /**
1640  * \brief Configure an ipv4 source address range
1641  * @param vm vlib_main_t pointer
1642  * @param start first ipv4 address in the source address range
1643  * @param end last ipv4 address in the source address range
1644  * @param table_id VRF / table ID, 0 for the default FIB
1645  * @return 0 if all OK, else an error indication from api_errno.h
1646  */
1647 
1648 int
1650  ip4_address_t * start,
1651  ip4_address_t * end, u32 table_id)
1652 {
1653  tcp_main_t *tm = vnet_get_tcp_main ();
1654  vnet_main_t *vnm = vnet_get_main ();
1655  u32 start_host_byte_order, end_host_byte_order;
1657  vnet_sw_interface_t *si;
1658  fib_node_index_t fei;
1659  u32 fib_index = 0;
1660  u32 sw_if_index;
1661  int rv;
1662  int vnet_proxy_arp_add_del (ip4_address_t * lo_addr,
1663  ip4_address_t * hi_addr, u32 fib_index,
1664  int is_del);
1665 
1666  clib_memset (&prefix, 0, sizeof (prefix));
1667 
1668  fib_index = fib_table_find (FIB_PROTOCOL_IP4, table_id);
1669 
1670  if (fib_index == ~0)
1671  return VNET_API_ERROR_NO_SUCH_FIB;
1672 
1673  start_host_byte_order = clib_net_to_host_u32 (start->as_u32);
1674  end_host_byte_order = clib_net_to_host_u32 (end->as_u32);
1675 
1676  /* sanity check for reversed args or some such */
1677  if ((end_host_byte_order - start_host_byte_order) > (10 << 10))
1678  return VNET_API_ERROR_INVALID_ARGUMENT;
1679 
1680  /* Lookup the last address, to identify the interface involved */
1681  prefix.fp_len = 32;
1682  prefix.fp_proto = FIB_PROTOCOL_IP4;
1683  memcpy (&prefix.fp_addr.ip4, end, sizeof (ip4_address_t));
1684 
1685  fei = fib_table_lookup (fib_index, &prefix);
1686 
1687  /* Couldn't find route to destination. Bail out. */
1688  if (fei == FIB_NODE_INDEX_INVALID)
1689  return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1690 
1691  sw_if_index = fib_entry_get_resolving_interface (fei);
1692 
1693  /* Enable proxy arp on the interface */
1694  si = vnet_get_sw_interface (vnm, sw_if_index);
1696 
1697  /* Configure proxy arp across the range */
1698  rv = vnet_proxy_arp_add_del (start, end, fib_index, 0 /* is_del */ );
1699 
1700  if (rv)
1701  return rv;
1702 
1703  do
1704  {
1705  dpo_id_t dpo = DPO_INVALID;
1706 
1707  vec_add1 (tm->ip4_src_addresses, start[0]);
1708 
1709  /* Add local adjacencies for the range */
1710 
1711  receive_dpo_add_or_lock (DPO_PROTO_IP4, ~0 /* sw_if_index */ ,
1712  NULL, &dpo);
1713  prefix.fp_len = 32;
1714  prefix.fp_proto = FIB_PROTOCOL_IP4;
1715  prefix.fp_addr.ip4.as_u32 = start->as_u32;
1716 
1718  &prefix,
1720  FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
1721  dpo_reset (&dpo);
1722 
1723  start_host_byte_order++;
1724  start->as_u32 = clib_host_to_net_u32 (start_host_byte_order);
1725  }
1726  while (start_host_byte_order <= end_host_byte_order);
1727 
1728  return 0;
1729 }
1730 
1731 /**
1732  * \brief Configure an ipv6 source address range
1733  * @param vm vlib_main_t pointer
1734  * @param start first ipv6 address in the source address range
1735  * @param end last ipv6 address in the source address range
1736  * @param table_id VRF / table ID, 0 for the default FIB
1737  * @return 0 if all OK, else an error indication from api_errno.h
1738  */
1739 
1740 int
1742  ip6_address_t * start,
1743  ip6_address_t * end, u32 table_id)
1744 {
1745  tcp_main_t *tm = vnet_get_tcp_main ();
1747  u32 fib_index = 0;
1748  fib_node_index_t fei;
1749  u32 sw_if_index;
1750 
1751  clib_memset (&prefix, 0, sizeof (prefix));
1752 
1753  fib_index = fib_table_find (FIB_PROTOCOL_IP6, table_id);
1754 
1755  if (fib_index == ~0)
1756  return VNET_API_ERROR_NO_SUCH_FIB;
1757 
1758  while (1)
1759  {
1760  int i;
1761  ip6_address_t tmp;
1762  dpo_id_t dpo = DPO_INVALID;
1763 
1764  /* Remember this address */
1765  vec_add1 (tm->ip6_src_addresses, start[0]);
1766 
1767  /* Lookup the prefix, to identify the interface involved */
1768  prefix.fp_len = 128;
1769  prefix.fp_proto = FIB_PROTOCOL_IP6;
1770  memcpy (&prefix.fp_addr.ip6, start, sizeof (ip6_address_t));
1771 
1772  fei = fib_table_lookup (fib_index, &prefix);
1773 
1774  /* Couldn't find route to destination. Bail out. */
1775  if (fei == FIB_NODE_INDEX_INVALID)
1776  return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1777 
1778  sw_if_index = fib_entry_get_resolving_interface (fei);
1779 
1780  if (sw_if_index == (u32) ~ 0)
1781  return VNET_API_ERROR_NO_MATCHING_INTERFACE;
1782 
1783  /* Add a proxy neighbor discovery entry for this address */
1784  ip6_neighbor_proxy_add_del (sw_if_index, start, 0 /* is_del */ );
1785 
1786  /* Add a receive adjacency for this address */
1787  receive_dpo_add_or_lock (DPO_PROTO_IP6, ~0 /* sw_if_index */ ,
1788  NULL, &dpo);
1789 
1791  &prefix,
1793  FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
1794  dpo_reset (&dpo);
1795 
1796  /* Done with the entire range? */
1797  if (!memcmp (start, end, sizeof (start[0])))
1798  break;
1799 
1800  /* Increment the address. DGMS. */
1801  tmp = start[0];
1802  for (i = 15; i >= 0; i--)
1803  {
1804  tmp.as_u8[i] += 1;
1805  if (tmp.as_u8[i] != 0)
1806  break;
1807  }
1808  start[0] = tmp;
1809  }
1810  return 0;
1811 }
1812 
1813 static clib_error_t *
1815  unformat_input_t * input, vlib_cli_command_t * cmd_arg)
1816 {
1817  ip4_address_t v4start, v4end;
1818  ip6_address_t v6start, v6end;
1819  u32 table_id = 0;
1820  int v4set = 0;
1821  int v6set = 0;
1822  int rv;
1823 
1824  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1825  {
1826  if (unformat (input, "%U - %U", unformat_ip4_address, &v4start,
1827  unformat_ip4_address, &v4end))
1828  v4set = 1;
1829  else if (unformat (input, "%U", unformat_ip4_address, &v4start))
1830  {
1831  memcpy (&v4end, &v4start, sizeof (v4start));
1832  v4set = 1;
1833  }
1834  else if (unformat (input, "%U - %U", unformat_ip6_address, &v6start,
1835  unformat_ip6_address, &v6end))
1836  v6set = 1;
1837  else if (unformat (input, "%U", unformat_ip6_address, &v6start))
1838  {
1839  memcpy (&v6end, &v6start, sizeof (v6start));
1840  v6set = 1;
1841  }
1842  else if (unformat (input, "fib-table %d", &table_id))
1843  ;
1844  else
1845  break;
1846  }
1847 
1848  if (!v4set && !v6set)
1849  return clib_error_return (0, "at least one v4 or v6 address required");
1850 
1851  if (v4set)
1852  {
1853  rv = tcp_configure_v4_source_address_range (vm, &v4start, &v4end,
1854  table_id);
1855  switch (rv)
1856  {
1857  case 0:
1858  break;
1859 
1860  case VNET_API_ERROR_NO_SUCH_FIB:
1861  return clib_error_return (0, "Invalid table-id %d", table_id);
1862 
1863  case VNET_API_ERROR_INVALID_ARGUMENT:
1864  return clib_error_return (0, "Invalid address range %U - %U",
1865  format_ip4_address, &v4start,
1866  format_ip4_address, &v4end);
1867  default:
1868  return clib_error_return (0, "error %d", rv);
1869  break;
1870  }
1871  }
1872  if (v6set)
1873  {
1874  rv = tcp_configure_v6_source_address_range (vm, &v6start, &v6end,
1875  table_id);
1876  switch (rv)
1877  {
1878  case 0:
1879  break;
1880 
1881  case VNET_API_ERROR_NO_SUCH_FIB:
1882  return clib_error_return (0, "Invalid table-id %d", table_id);
1883 
1884  default:
1885  return clib_error_return (0, "error %d", rv);
1886  break;
1887  }
1888  }
1889  return 0;
1890 }
1891 
1892 /* *INDENT-OFF* */
1893 VLIB_CLI_COMMAND (tcp_src_address_command, static) =
1894 {
1895  .path = "tcp src-address",
1896  .short_help = "tcp src-address <ip-addr> [- <ip-addr>] add src address range",
1897  .function = tcp_src_address,
1898 };
1899 /* *INDENT-ON* */
1900 
1901 static u8 *
1903 {
1904 #if TCP_SCOREBOARD_TRACE
1905 
1906  scoreboard_trace_elt_t *block;
1907  int i = 0;
1908 
1909  if (!sb->trace)
1910  return s;
1911 
1912  s = format (s, "scoreboard trace:");
1913  vec_foreach (block, sb->trace)
1914  {
1915  s = format (s, "{%u, %u, %u, %u, %u}, ", block->start, block->end,
1916  block->ack, block->snd_una_max, block->group);
1917  if ((++i % 3) == 0)
1918  s = format (s, "\n");
1919  }
1920  return s;
1921 #else
1922  return 0;
1923 #endif
1924 }
1925 
1926 static clib_error_t *
1928  vlib_cli_command_t * cmd_arg)
1929 {
1930  transport_connection_t *tconn = 0;
1931  tcp_connection_t *tc;
1932  u8 *s = 0;
1933  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1934  {
1935  if (unformat (input, "%U", unformat_transport_connection, &tconn,
1937  ;
1938  else
1939  return clib_error_return (0, "unknown input `%U'",
1940  format_unformat_error, input);
1941  }
1942 
1943  if (!TCP_SCOREBOARD_TRACE)
1944  {
1945  vlib_cli_output (vm, "scoreboard tracing not enabled");
1946  return 0;
1947  }
1948 
1949  tc = tcp_get_connection_from_transport (tconn);
1950  s = tcp_scoreboard_dump_trace (s, &tc->sack_sb);
1951  vlib_cli_output (vm, "%v", s);
1952  return 0;
1953 }
1954 
1955 /* *INDENT-OFF* */
1956 VLIB_CLI_COMMAND (tcp_show_scoreboard_trace_command, static) =
1957 {
1958  .path = "show tcp scoreboard trace",
1959  .short_help = "show tcp scoreboard trace <connection>",
1960  .function = tcp_show_scoreboard_trace_fn,
1961 };
1962 /* *INDENT-ON* */
1963 
1964 u8 *
1966 {
1967  int i, trace_len;
1969  u32 next_ack, left, group, has_new_ack = 0;
1970  tcp_connection_t _dummy_tc, *dummy_tc = &_dummy_tc;
1971  sack_block_t *block;
1972 
1973  if (!TCP_SCOREBOARD_TRACE)
1974  {
1975  s = format (s, "scoreboard tracing not enabled");
1976  return s;
1977  }
1978 
1979  if (!tc)
1980  return s;
1981 
1982  clib_memset (dummy_tc, 0, sizeof (*dummy_tc));
1983  tcp_connection_timers_init (dummy_tc);
1984  scoreboard_init (&dummy_tc->sack_sb);
1985  dummy_tc->rcv_opts.flags |= TCP_OPTS_FLAG_SACK;
1986 
1987 #if TCP_SCOREBOARD_TRACE
1988  trace = tc->sack_sb.trace;
1989  trace_len = vec_len (tc->sack_sb.trace);
1990 #endif
1991 
1992  for (i = 0; i < trace_len; i++)
1993  {
1994  if (trace[i].ack != 0)
1995  {
1996  dummy_tc->snd_una = trace[i].ack - 1448;
1997  dummy_tc->snd_una_max = trace[i].ack;
1998  }
1999  }
2000 
2001  left = 0;
2002  while (left < trace_len)
2003  {
2004  group = trace[left].group;
2005  vec_reset_length (dummy_tc->rcv_opts.sacks);
2006  has_new_ack = 0;
2007  while (trace[left].group == group)
2008  {
2009  if (trace[left].ack != 0)
2010  {
2011  if (verbose)
2012  s = format (s, "Adding ack %u, snd_una_max %u, segs: ",
2013  trace[left].ack, trace[left].snd_una_max);
2014  dummy_tc->snd_una_max = trace[left].snd_una_max;
2015  next_ack = trace[left].ack;
2016  has_new_ack = 1;
2017  }
2018  else
2019  {
2020  if (verbose)
2021  s = format (s, "[%u, %u], ", trace[left].start,
2022  trace[left].end);
2023  vec_add2 (dummy_tc->rcv_opts.sacks, block, 1);
2024  block->start = trace[left].start;
2025  block->end = trace[left].end;
2026  }
2027  left++;
2028  }
2029 
2030  /* Push segments */
2031  tcp_rcv_sacks (dummy_tc, next_ack);
2032  if (has_new_ack)
2033  dummy_tc->snd_una = next_ack + dummy_tc->sack_sb.snd_una_adv;
2034 
2035  if (verbose)
2036  s = format (s, "result: %U", format_tcp_scoreboard,
2037  &dummy_tc->sack_sb);
2038 
2039  }
2040  s = format (s, "result: %U", format_tcp_scoreboard, &dummy_tc->sack_sb);
2041 
2042  return s;
2043 }
2044 
2045 static clib_error_t *
2047  vlib_cli_command_t * cmd_arg)
2048 {
2049  transport_connection_t *tconn = 0;
2050  tcp_connection_t *tc = 0;
2051  u8 *str = 0;
2052  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
2053  {
2054  if (unformat (input, "%U", unformat_transport_connection, &tconn,
2056  ;
2057  else
2058  return clib_error_return (0, "unknown input `%U'",
2059  format_unformat_error, input);
2060  }
2061 
2062  if (!TCP_SCOREBOARD_TRACE)
2063  {
2064  vlib_cli_output (vm, "scoreboard tracing not enabled");
2065  return 0;
2066  }
2067 
2068  tc = tcp_get_connection_from_transport (tconn);
2069  if (!tc)
2070  {
2071  vlib_cli_output (vm, "connection not found");
2072  return 0;
2073  }
2074  str = tcp_scoreboard_replay (str, tc, 1);
2075  vlib_cli_output (vm, "%v", str);
2076  return 0;
2077 }
2078 
2079 /* *INDENT-OFF* */
2080 VLIB_CLI_COMMAND (tcp_replay_scoreboard_command, static) =
2081 {
2082  .path = "tcp replay scoreboard",
2083  .short_help = "tcp replay scoreboard <connection>",
2084  .function = tcp_scoreboard_trace_fn,
2085 };
2086 /* *INDENT-ON* */
2087 
2088 static clib_error_t *
2090  vlib_cli_command_t * cmd_arg)
2091 {
2092  tcp_main_t *tm = vnet_get_tcp_main ();
2094  return clib_error_return (0, "unknown input `%U'", format_unformat_error,
2095  input);
2096  vlib_cli_output (vm, "IPv4 TCP punt: %s",
2097  tm->punt_unknown4 ? "enabled" : "disabled");
2098  vlib_cli_output (vm, "IPv6 TCP punt: %s",
2099  tm->punt_unknown6 ? "enabled" : "disabled");
2100  return 0;
2101 }
2102 /* *INDENT-OFF* */
2103 VLIB_CLI_COMMAND (show_tcp_punt_command, static) =
2104 {
2105  .path = "show tcp punt",
2106  .short_help = "show tcp punt",
2107  .function = show_tcp_punt_fn,
2108 };
2109 /* *INDENT-ON* */
2110 
2111 /*
2112  * fd.io coding-style-patch-verification: ON
2113  *
2114  * Local Variables:
2115  * eval: (c-set-style "gnu")
2116  * End:
2117  */
static void tcp_session_close(u32 conn_index, u32 thread_index)
Definition: tcp.c:405
#define vec_validate(V, I)
Make sure vector is long enough for given index (no header, unspecified alignment) ...
Definition: vec.h:439
u32 sw_if_index
Definition: ipsec_gre.api:37
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
fib_protocol_t fp_proto
protocol type
Definition: fib_types.h:212
#define ENDPOINT_INVALID_INDEX
#define TCP_2MSL_TIME
Definition: tcp.h:103
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:62
An entry in a FIB table.
Definition: fib_entry.h:462
#define clib_min(x, y)
Definition: clib.h:295
static u8 * format_tcp_listener_session(u8 *s, va_list *args)
Definition: tcp.c:941
const char * tcp_fsm_states[]
Definition: tcp.c:733
static clib_error_t * tcp_config_fn(vlib_main_t *vm, unformat_input_t *input)
Definition: tcp.c:1602
#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
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:640
struct _sack_block sack_block_t
u8 * format_tcp_rcv_sacks(u8 *s, va_list *args)
Definition: tcp.c:985
void ip6_register_protocol(u32 protocol, u32 node_index)
Definition: ip6_forward.c:1466
struct _scoreboard_trace_elt scoreboard_trace_elt_t
u8 * format_tcp_sacks(u8 *s, va_list *args)
Definition: tcp.c:961
vnet_main_t * vnet_get_main(void)
Definition: misc.c:47
static clib_error_t * ip4_lookup_init(vlib_main_t *vm)
Definition: ip4_forward.c:909
const char * tcp_connection_flags_str[]
Definition: tcp.c:751
vlib_node_registration_t tcp4_output_node
(constructor) VLIB_REGISTER_NODE (tcp4_output_node)
Definition: tcp_output.c:2154
void scoreboard_init(sack_scoreboard_t *sb)
Definition: tcp_input.c:892
uword unformat_transport_connection(unformat_input_t *input, va_list *args)
Definition: session_cli.c:195
static u32 tcp_session_send_space(transport_connection_t *trans_conn)
Definition: tcp.c:1147
static transport_connection_t * tcp_half_open_session_get_transport(u32 conn_index)
Definition: tcp.c:1060
u8 as_u8[16]
Definition: ip6_packet.h:48
unsigned long u64
Definition: types.h:89
static void tcp_timer_establish_handler(u32 conn_index)
Definition: tcp.c:1252
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:1155
#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:296
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:746
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:651
static f64 vlib_time_now(vlib_main_t *vm)
Definition: main.h:255
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:783
static tcp_connection_t * tcp_half_open_connection_get(u32 conn_index)
Definition: tcp.h:601
sack_scoreboard_hole_t * scoreboard_next_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:666
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:98
struct _tcp_main tcp_main_t
void tcp_connection_timers_reset(tcp_connection_t *tc)
Stop all connection timers.
Definition: tcp.c:442
vlib_node_registration_t tcp6_output_node
(constructor) VLIB_REGISTER_NODE (tcp6_output_node)
Definition: tcp_output.c:2174
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:522
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:829
#define vec_add2(V, P, N)
Add N elements to end of vector V, return pointer to new elements in P.
Definition: vec.h:560
int i
int ip6_neighbor_proxy_add_del(u32 sw_if_index, ip6_address_t *addr, u8 is_del)
void tcp_update_burst_snd_vars(tcp_connection_t *tc)
Update burst send vars.
Definition: tcp_output.c:408
static u32 format_get_indent(u8 *s)
Definition: format.h:72
#define hash_set_mem(h, key, value)
Definition: hash.h:275
clib_memset(h->entries, 0, sizeof(h->entries[0])*entries)
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: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:747
clib_time_t clib_time
Definition: main.h:72
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:1074
Definition: ip.h:109
static tcp_connection_t * tcp_get_connection_from_transport(transport_connection_t *tconn)
Definition: tcp.h:566
static void tcp_initialize_iss_seed(tcp_main_t *tm)
Definition: tcp.c:1405
void ip4_register_protocol(u32 protocol, u32 node_index)
Definition: ip4_forward.c:1699
static void tcp_initialize_timer_wheels(tcp_main_t *tm)
Definition: tcp.c:1391
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:236
struct _tcp_header tcp_header_t
fib_protocol_t nh_proto
Definition: tcp.c:33
int tcp_half_open_connection_cleanup(tcp_connection_t *tc)
Try to cleanup half-open connection.
Definition: tcp.c:170
vlib_main_t ** vlib_mains
Definition: buffer.c:321
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:567
#define TCP_RTO_INIT
Definition: tcp.h:114
format_function_t format_ip4_address
Definition: format.h:75
#define TCP_TICK
TCP tick period (s)
Definition: tcp.h:27
unformat_function_t * unformat_pg_edit
Definition: ip.h:90
static void tcp_connection_set_state(tcp_connection_t *tc, tcp_state_t state)
Definition: tcp.h:572
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:598
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:163
#define TCP_CLOSEWAIT_TIME
Definition: tcp.h:104
#define always_inline
Definition: clib.h:98
static u8 * format_tcp_vars(u8 *s, va_list *args)
Definition: tcp.c:842
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:425
static u32 tcp_available_output_snd_space(const tcp_connection_t *tc)
Definition: tcp.h:732
u8 * format_white_space(u8 *s, va_list *va)
Definition: std-formats.c:113
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
tw_timer_wheel_16t_2w_512sl_t timer_wheel
worker timer wheel
Definition: tcp.h:383
#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:318
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:1902
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:472
static u32 tcp_time_now(void)
Definition: tcp.h:785
u32 tcp_snd_space(tcp_connection_t *tc)
Definition: tcp.c:1141
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:241
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 i32 tcp_rcv_wnd_available(tcp_connection_t *tc)
Definition: tcp.c:818
#define hash_create_string(elts, value_bytes)
Definition: hash.h:690
void transport_connection_tx_pacer_init(transport_connection_t *tc, u32 rate_bytes_per_sec, u32 initial_bucket)
Initialize tx pacer for connection.
Definition: transport.c:583
static void tcp_cc_init(tcp_connection_t *tc)
Definition: tcp.c:549
struct _transport_proto_vft transport_proto_vft_t
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:828
void tcp_do_fastretransmits(tcp_worker_ctx_t *wrk)
Definition: tcp_input.c:1311
uword unformat_tcp_cc_algo_cfg(unformat_input_t *input, va_list *va)
Definition: tcp.c:1579
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:1358
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:1415
Definition: fib_entry.h:279
vlib_main_t * vm
convenience pointer to this thread&#39;s vlib main
Definition: tcp.h:413
#define ADJ_INDEX_INVALID
Invalid ADJ index - used when no adj is known likewise blazoned capitals INVALID speak volumes where ...
Definition: adj_types.h:36
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:514
timer_expiration_handler tcp_timer_persist_handler
static transport_connection_t * tcp_session_get_transport(u32 conn_index, u32 thread_index)
Definition: tcp.c:1053
static ip_protocol_info_t * ip_get_protocol_info(ip_main_t *im, u32 protocol)
Definition: ip.h:136
#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
void tcp_connection_tx_pacer_reset(tcp_connection_t *tc, u32 window, u32 start_bucket)
Definition: tcp.c:1229
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:228
void tcp_rcv_sacks(tcp_connection_t *tc, u32 ack)
Definition: tcp_input.c:937
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
tcp_connection_t * tcp_connection_alloc(u8 thread_index)
Definition: tcp.c:253
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
static u32 transport_max_rx_dequeue(transport_connection_t *tc)
Definition: session.h:392
void tcp_connection_tx_pacer_update(tcp_connection_t *tc)
Definition: tcp.c:1213
static u32 ip6_compute_flow_hash(const ip6_header_t *ip, flow_hash_config_t flow_hash_config)
Definition: ip6.h:478
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:286
static void tcp_expired_timers_dispatch(u32 *expired_timers)
Definition: tcp.c:1372
#define TCP_TIMER_HANDLE_INVALID
Definition: tcp.h:95
static void tcp_session_cleanup(u32 conn_index, u32 thread_index)
Definition: tcp.c:413
The FIB DPO provieds;.
Definition: load_balance.h:106
#define TCP_CLEANUP_TIME
Definition: tcp.h:107
static u32 tcp_flight_size(const tcp_connection_t *tc)
Our estimate of the number of bytes in flight (pipe size)
Definition: tcp.h:665
#define PREDICT_FALSE(x)
Definition: clib.h:111
#define VLIB_CONFIG_FUNCTION(x, n,...)
Definition: init.h:172
vnet_sw_interface_flags_t flags
Definition: interface.h:684
u8 * format_tcp_scoreboard(u8 *s, va_list *args)
Definition: tcp.c:1021
clib_error_t * vnet_tcp_enable_disable(vlib_main_t *vm, u8 is_en)
Definition: tcp.c:1498
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:186
static void tcp_timer_waitclose_handler(u32 conn_index)
Definition: tcp.c:1289
#define TCP_FINWAIT1_TIME
Definition: tcp.h:106
void( timer_expiration_handler)(u32 index)
Definition: tcp.h:88
#define foreach_vlib_main(body)
Definition: threads.h:235
void transport_connection_tx_pacer_update(transport_connection_t *tc, u64 bytes_per_sec)
Update tx pacer pacing rate.
Definition: transport.c:595
#define TCP_DBG(_fmt, _args...)
Definition: tcp_debug.h:89
u8 len
Definition: ip_types.api:49
u32 fib_entry_get_resolving_interface(fib_node_index_t entry_index)
Definition: fib_entry.c:1451
static void tcp_timer_reset(tcp_connection_t *tc, u8 timer_id)
Definition: tcp.h:839
void tcp_connection_free(tcp_connection_t *tc)
Definition: tcp.c:266
unformat_function_t unformat_ip6_address
Definition: format.h:91
static_always_inline u32 vlib_buffer_get_default_data_size(vlib_main_t *vm)
Definition: buffer_funcs.h:96
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:1649
static_always_inline uword vlib_get_thread_index(void)
Definition: threads.h:212
format_function_t format_ip6_address
Definition: format.h:93
vlib_main_t * vm
Definition: buffer.c:312
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:1814
void tcp_connection_close(tcp_connection_t *tc)
Begin connection closing procedure.
Definition: tcp.c:340
void tcp_send_reset(tcp_connection_t *tc)
Build and set reset packet for connection.
Definition: tcp_output.c:841
#define vec_free(V)
Free vector&#39;s memory (no header).
Definition: vec.h:341
virtual circuit service
static void tcp_session_flush_data(transport_connection_t *tconn)
Definition: tcp.c:1178
#define clib_warning(format, args...)
Definition: error.h:59
static u8 * format_tcp_congestion(u8 *s, va_list *args)
Definition: tcp.c:824
format_function_t format_tcp_header
Definition: format.h:101
struct _transport_connection transport_connection_t
const char * tcp_dbg_evt_str[]
Definition: tcp.c:727
static u8 * format_tcp_connection_flags(u8 *s, va_list *args)
Definition: tcp.c:758
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:1559
vlib_node_registration_t tcp6_input_node
(constructor) VLIB_REGISTER_NODE (tcp6_input_node)
Definition: tcp_input.c:3592
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:50
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:1998
#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:341
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:248
#define VLIB_CLI_COMMAND(x,...)
Definition: cli.h:155
static u32 transport_max_tx_dequeue(transport_connection_t *tc)
Definition: session.h:385
static void tcp_timer_update(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:852
signed int i32
Definition: types.h:77
#define pool_put_index(p, i)
Free pool element with given index.
Definition: pool.h:311
#define ASSERT(truth)
u64 last_cpu_time
Definition: time.h:50
static int tcp_session_open(transport_endpoint_cfg_t *rmt)
Definition: tcp.c:681
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:2046
unformat_function_t unformat_pg_tcp_header
Definition: format.h:103
u8 is_add
Definition: ipsec_gre.api:36
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:1013
static load_balance_t * load_balance_get(index_t lbi)
Definition: load_balance.h:219
#define seq_geq(_s1, _s2)
Definition: tcp.h:642
static u8 * format_tcp_congestion_status(u8 *s, va_list *args)
Definition: tcp.c:805
static u32 tcp_generate_random_iss(tcp_connection_t *tc)
Generate random iss as per rfc6528.
Definition: tcp.c:577
reliable transport protos
static uword ip6_address_is_link_local_unicast(const ip6_address_t *a)
Definition: ip6_packet.h:328
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:556
u8 * tcp_scoreboard_replay(u8 *s, tcp_connection_t *tc, u8 verbose)
Definition: tcp.c:1965
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:1741
struct _tcp_cc_algorithm tcp_cc_algorithm_t
Definition: tcp.h:239
#define clib_max(x, y)
Definition: clib.h:288
static clib_error_t * tcp_init(vlib_main_t *vm)
Definition: tcp.c:1526
static void tcp_timer_keep_handler(u32 conn_index)
Definition: tcp.c:1240
dpo_id_t fe_lb
The load-balance used for forwarding.
Definition: fib_entry.h:488
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:1188
void tcp_punt_unknown(vlib_main_t *vm, u8 is_ip4, u8 is_add)
Definition: tcp.c:1516
void transport_connection_tx_pacer_reset(transport_connection_t *tc, u32 rate_bytes_per_sec, u32 start_bucket, u64 time_now)
Definition: transport.c:572
void tcp_connection_reset(tcp_connection_t *tc)
Notify session that connection has been reset.
Definition: tcp.c:284
vlib_node_registration_t tcp4_input_node
(constructor) VLIB_REGISTER_NODE (tcp4_input_node)
Definition: tcp_input.c:3572
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:43
void tcp_send_fin(tcp_connection_t *tc)
Send FIN.
Definition: tcp_output.c:1025
#define foreach_tcp_connection_flag
TCP connection flags.
Definition: tcp.h:117
#define foreach_tcp_timer
TCP timers.
Definition: tcp.h:70
static u8 * format_tcp_connection_id(u8 *s, va_list *args)
Definition: tcp.c:882
#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:1823
uword unformat_vlib_cli_sub_input(unformat_input_t *i, va_list *args)
Definition: cli.c:154
static void tcp_timer_establish_ao_handler(u32 conn_index)
Definition: tcp.c:1271
sack_scoreboard_hole_t * scoreboard_first_hole(sack_scoreboard_t *sb)
Definition: tcp_input.c:682
static u32 tcp_round_snd_space(tcp_connection_t *tc, u32 snd_space)
Definition: tcp.c:1087
static tcp_worker_ctx_t * tcp_get_worker(u32 thread_index)
Definition: tcp.h:520
void session_transport_closed_notify(transport_connection_t *tc)
Notification from transport that session can be closed.
Definition: session.c:807
u64 uword
Definition: types.h:112
void tcp_connection_init_vars(tcp_connection_t *tc)
Initialize tcp connection variables.
Definition: tcp.c:631
#define DPO_INVALID
An initialiser for DPOs declared on the stack.
Definition: dpo.h:195
const char * tcp_conn_timers[]
Definition: tcp.c:776
static u8 * format_tcp_session(u8 *s, va_list *args)
Definition: tcp.c:925
typedef prefix
Definition: ip_types.api:35
left
#define hash_get_mem(h, key)
Definition: hash.h:269
unformat_function_t unformat_memory_size
Definition: format.h:295
u8 * format_tcp_connection(u8 *s, va_list *args)
Definition: tcp.c:906
static u8 * format_tcp_sack_hole(u8 *s, va_list *args)
Definition: tcp.c:1009
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:547
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:147
int session_stream_connect_notify(transport_connection_t *tc, u8 is_fail)
Definition: session.c:598
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:246
void tcp_init_mss(tcp_connection_t *tc)
Definition: tcp_output.c:428
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:231
#define vec_foreach(var, vec)
Vector iterator.
#define TCP_TSTAMP_RESOLUTION
Time stamp resolution.
Definition: tcp.h:29
static u8 * format_tcp_half_open_session(u8 *s, va_list *args)
Definition: tcp.c:953
#define tcp_opts_sack_permitted(_to)
Definition: tcp_packet.h:160
void tcp_send_acks(tcp_worker_ctx_t *wrk)
Definition: tcp_output.c:1193
static u32 tcp_snd_space_inline(tcp_connection_t *tc)
Compute tx window session is allowed to fill.
Definition: tcp.c:1114
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:375
u8 * format_tcp_state(u8 *s, va_list *args)
Definition: tcp.c:740
static clib_error_t * ip6_lookup_init(vlib_main_t *vm)
Definition: ip6_forward.c:2684
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:2089
static void tcp_update_time(f64 now, u8 thread_index)
Definition: tcp.c:1166
void tcp_enable_pacing(tcp_connection_t *tc)
Definition: tcp.c:616
static tcp_main_t * vnet_get_tcp_main()
Definition: tcp.h:514
static void tcp_cong_recovery_off(tcp_connection_t *tc)
Definition: tcp.h:357
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 u32 tcp_set_time_now(tcp_worker_ctx_t *wrk)
Definition: tcp.h:803
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:1927
uword unformat(unformat_input_t *i, const char *fmt,...)
Definition: unformat.c:972
static tcp_connection_t * tcp_listener_get(u32 tli)
Definition: tcp.h:595
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:99
static uword pool_elts(void *v)
Number of active elements in a pool.
Definition: pool.h:128