FD.io VPP  v19.01.3-6-g70449b9b9
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;
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  /* 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  vec_free (tc->snd_sacks);
227 
228  /* Poison the entry */
229  if (CLIB_DEBUG > 0)
230  clib_memset (tc, 0xFA, sizeof (*tc));
231  pool_put (tm->connections[thread_index], tc);
232  }
233 }
234 
235 /**
236  * Connection removal.
237  *
238  * This should be called only once connection enters CLOSED state. Note
239  * that it notifies the session of the removal event, so if the goal is to
240  * just remove the connection, call tcp_connection_cleanup instead.
241  */
242 void
244 {
245  TCP_EVT_DBG (TCP_EVT_DELETE, tc);
246  session_transport_delete_notify (&tc->connection);
248 }
249 
251 tcp_connection_alloc (u8 thread_index)
252 {
253  tcp_main_t *tm = vnet_get_tcp_main ();
254  tcp_connection_t *tc;
255 
256  pool_get (tm->connections[thread_index], tc);
257  clib_memset (tc, 0, sizeof (*tc));
258  tc->c_c_index = tc - tm->connections[thread_index];
259  tc->c_thread_index = thread_index;
260  return tc;
261 }
262 
263 void
265 {
266  tcp_main_t *tm = &tcp_main;
267  pool_put (tm->connections[tc->c_thread_index], tc);
268  if (CLIB_DEBUG > 0)
269  clib_memset (tc, 0xFA, sizeof (*tc));
270 }
271 
272 /** Notify session that connection has been reset.
273  *
274  * Switch state to closed and wait for session to call cleanup.
275  */
276 void
278 {
279  TCP_EVT_DBG (TCP_EVT_RST_RCVD, tc);
280  switch (tc->state)
281  {
282  case TCP_STATE_SYN_RCVD:
283  /* Cleanup everything. App wasn't notified yet */
284  session_transport_delete_notify (&tc->connection);
286  break;
287  case TCP_STATE_SYN_SENT:
288  session_stream_connect_notify (&tc->connection, 1 /* fail */ );
290  break;
291  case TCP_STATE_ESTABLISHED:
293  /* Set the cleanup timer, in case the session layer/app don't
294  * cleanly close the connection */
295  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME);
296  session_transport_reset_notify (&tc->connection);
297  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
298  break;
299  case TCP_STATE_CLOSE_WAIT:
300  case TCP_STATE_FIN_WAIT_1:
301  case TCP_STATE_FIN_WAIT_2:
302  case TCP_STATE_CLOSING:
303  case TCP_STATE_LAST_ACK:
305  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME);
306  /* Make sure we mark the session as closed. In some states we may
307  * be still trying to send data */
308  session_transport_closed_notify (&tc->connection);
309  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
310  break;
311  case TCP_STATE_CLOSED:
312  case TCP_STATE_TIME_WAIT:
313  break;
314  default:
315  TCP_DBG ("reset state: %u", tc->state);
316  }
317 }
318 
319 /**
320  * Begin connection closing procedure.
321  *
322  * If at the end the connection is not in CLOSED state, it is not removed.
323  * Instead, we rely on on TCP to advance through state machine to either
324  * 1) LAST_ACK (passive close) whereby when the last ACK is received
325  * tcp_connection_del is called. This notifies session of the delete and
326  * calls cleanup.
327  * 2) TIME_WAIT (active close) whereby after 2MSL the 2MSL timer triggers
328  * and cleanup is called.
329  *
330  * N.B. Half-close connections are not supported
331  */
332 void
334 {
335  TCP_EVT_DBG (TCP_EVT_CLOSE, tc);
336 
337  /* Send/Program FIN if needed and switch state */
338  switch (tc->state)
339  {
340  case TCP_STATE_SYN_SENT:
341  /* Try to cleanup. If not on the right thread, mark as half-open done.
342  * Connection will be cleaned up when establish timer pops */
344  break;
345  case TCP_STATE_SYN_RCVD:
347  tcp_send_fin (tc);
348  tcp_connection_set_state (tc, TCP_STATE_FIN_WAIT_1);
349  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_FINWAIT1_TIME);
350  break;
351  case TCP_STATE_ESTABLISHED:
352  if (!session_tx_fifo_max_dequeue (&tc->connection))
353  tcp_send_fin (tc);
354  else
355  tc->flags |= TCP_CONN_FINPNDG;
356  tcp_connection_set_state (tc, TCP_STATE_FIN_WAIT_1);
357  /* Set a timer in case the peer stops responding. Otherwise the
358  * connection will be stuck here forever. */
359  ASSERT (tc->timers[TCP_TIMER_WAITCLOSE] == TCP_TIMER_HANDLE_INVALID);
360  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_FINWAIT1_TIME);
361  break;
362  case TCP_STATE_CLOSE_WAIT:
363  if (!session_tx_fifo_max_dequeue (&tc->connection))
364  {
365  tcp_send_fin (tc);
367  tcp_connection_set_state (tc, TCP_STATE_LAST_ACK);
368  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
369  }
370  else
371  tc->flags |= TCP_CONN_FINPNDG;
372  break;
373  case TCP_STATE_FIN_WAIT_1:
374  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
375  break;
376  case TCP_STATE_CLOSED:
378  /* Delete connection but instead of doing it now wait until next
379  * dispatch cycle to give the session layer a chance to clear
380  * unhandled events */
381  tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
382  break;
383  default:
384  TCP_DBG ("state: %u", tc->state);
385  }
386 }
387 
388 static void
389 tcp_session_close (u32 conn_index, u32 thread_index)
390 {
391  tcp_connection_t *tc;
392  tc = tcp_connection_get (conn_index, thread_index);
394 }
395 
396 static void
397 tcp_session_cleanup (u32 conn_index, u32 thread_index)
398 {
399  tcp_connection_t *tc;
400  tc = tcp_connection_get (conn_index, thread_index);
401  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
403 }
404 
405 /**
406  * Initialize all connection timers as invalid
407  */
408 void
410 {
411  int i;
412 
413  /* Set all to invalid */
414  for (i = 0; i < TCP_N_TIMERS; i++)
415  {
416  tc->timers[i] = TCP_TIMER_HANDLE_INVALID;
417  }
418 
419  tc->rto = TCP_RTO_INIT;
420 }
421 
422 /**
423  * Stop all connection timers
424  */
425 void
427 {
428  int i;
429  for (i = 0; i < TCP_N_TIMERS; i++)
430  {
431  tcp_timer_reset (tc, i);
432  }
433 }
434 
435 #if 0
436 typedef struct ip4_tcp_hdr
437 {
438  ip4_header_t ip;
439  tcp_header_t tcp;
440 } ip4_tcp_hdr_t;
441 
442 typedef struct ip6_tcp_hdr
443 {
444  ip6_header_t ip;
445  tcp_header_t tcp;
446 } ip6_tcp_hdr_t;
447 
448 static void
449 tcp_connection_select_lb_bucket (tcp_connection_t * tc, const dpo_id_t * dpo,
450  dpo_id_t * result)
451 {
452  const dpo_id_t *choice;
453  load_balance_t *lb;
454  int hash;
455 
456  lb = load_balance_get (dpo->dpoi_index);
457  if (tc->c_is_ip4)
458  {
459  ip4_tcp_hdr_t hdr;
460  clib_memset (&hdr, 0, sizeof (hdr));
461  hdr.ip.protocol = IP_PROTOCOL_TCP;
462  hdr.ip.address_pair.src.as_u32 = tc->c_lcl_ip.ip4.as_u32;
463  hdr.ip.address_pair.dst.as_u32 = tc->c_rmt_ip.ip4.as_u32;
464  hdr.tcp.src_port = tc->c_lcl_port;
465  hdr.tcp.dst_port = tc->c_rmt_port;
466  hash = ip4_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
467  }
468  else
469  {
470  ip6_tcp_hdr_t hdr;
471  clib_memset (&hdr, 0, sizeof (hdr));
472  hdr.ip.protocol = IP_PROTOCOL_TCP;
473  clib_memcpy_fast (&hdr.ip.src_address, &tc->c_lcl_ip.ip6,
474  sizeof (ip6_address_t));
475  clib_memcpy_fast (&hdr.ip.dst_address, &tc->c_rmt_ip.ip6,
476  sizeof (ip6_address_t));
477  hdr.tcp.src_port = tc->c_lcl_port;
478  hdr.tcp.dst_port = tc->c_rmt_port;
479  hash = ip6_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
480  }
481  choice = load_balance_get_bucket_i (lb, hash & lb->lb_n_buckets_minus_1);
482  dpo_copy (result, choice);
483 }
484 
487 {
489  u32 fib_index;
490 
491  clib_memcpy_fast (&prefix.fp_addr, &tc->c_rmt_ip, sizeof (prefix.fp_addr));
492  prefix.fp_proto = tc->c_is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6;
493  prefix.fp_len = tc->c_is_ip4 ? 32 : 128;
494  fib_index = fib_table_find (prefix.fp_proto, tc->c_fib_index);
495  return fib_table_lookup (fib_index, &prefix);
496 }
497 
498 static int
499 tcp_connection_stack_on_fib_entry (tcp_connection_t * tc)
500 {
501  dpo_id_t choice = DPO_INVALID;
502  u32 output_node_index;
503  fib_entry_t *fe;
504 
505  fe = fib_entry_get (tc->c_rmt_fei);
506  if (fe->fe_lb.dpoi_type != DPO_LOAD_BALANCE)
507  return -1;
508 
509  tcp_connection_select_lb_bucket (tc, &fe->fe_lb, &choice);
510 
511  output_node_index =
512  tc->c_is_ip4 ? tcp4_output_node.index : tcp6_output_node.index;
513  dpo_stack_from_node (output_node_index, &tc->c_rmt_dpo, &choice);
514  return 0;
515 }
516 
517 /** Stack tcp connection on peer's fib entry.
518  *
519  * This ultimately populates the dpo the connection will use to send packets.
520  */
521 static void
522 tcp_connection_fib_attach (tcp_connection_t * tc)
523 {
524  tc->c_rmt_fei = tcp_lookup_rmt_in_fib (tc);
525 
526  ASSERT (tc->c_rmt_fei != FIB_NODE_INDEX_INVALID);
527 
528  tcp_connection_stack_on_fib_entry (tc);
529 }
530 #endif /* 0 */
531 
532 static void
534 {
535  tc->cc_algo = tcp_cc_algo_get (tcp_main.cc_algo);
536  tc->cc_algo->init (tc);
537 }
538 
539 void
541  const tcp_cc_algorithm_t * vft)
542 {
543  tcp_main_t *tm = vnet_get_tcp_main ();
544  vec_validate (tm->cc_algos, type);
545 
546  tm->cc_algos[type] = *vft;
547 }
548 
551 {
552  tcp_main_t *tm = vnet_get_tcp_main ();
553  return &tm->cc_algos[type];
554 }
555 
556 /**
557  * Generate random iss as per rfc6528
558  */
559 static u32
561 {
562  tcp_main_t *tm = &tcp_main;
563  u64 tmp;
564 
565  if (tc->c_is_ip4)
566  tmp = (u64) tc->c_lcl_ip.ip4.as_u32 << 32 | (u64) tc->c_rmt_ip.ip4.as_u32;
567  else
568  tmp = tc->c_lcl_ip.ip6.as_u64[0] ^ tc->c_lcl_ip.ip6.as_u64[1]
569  ^ tc->c_rmt_ip.ip6.as_u64[0] ^ tc->c_rmt_ip.ip6.as_u64[1];
570 
571  tmp ^= tm->iss_seed.first | ((u64) tc->c_lcl_port << 16 | tc->c_rmt_port);
572  tmp ^= tm->iss_seed.second;
573  tmp = clib_xxhash (tmp) + clib_cpu_time_now ();
574  return ((tmp >> 32) ^ (tmp & 0xffffffff));
575 }
576 
577 /**
578  * Initialize connection send variables.
579  */
580 void
582 {
583  /*
584  * We use the time to randomize iss and for setting up the initial
585  * timestamp. Make sure it's updated otherwise syn and ack in the
586  * handshake may make it look as if time has flown in the opposite
587  * direction for us.
588  */
590 
591  tc->iss = tcp_generate_random_iss (tc);
592  tc->snd_una = tc->iss;
593  tc->snd_nxt = tc->iss + 1;
594  tc->snd_una_max = tc->snd_nxt;
595  tc->srtt = 0;
596 }
597 
598 void
600 {
601  u32 initial_bucket, byte_rate;
602  initial_bucket = 16 * tc->snd_mss;
603  byte_rate = 2 << 16;
604  transport_connection_tx_pacer_init (&tc->connection, byte_rate,
605  initial_bucket);
606  tc->mrtt_us = (u32) ~ 0;
607 }
608 
609 /** Initialize tcp connection variables
610  *
611  * Should be called after having received a msg from the peer, i.e., a SYN or
612  * a SYNACK, such that connection options have already been exchanged. */
613 void
615 {
617  tcp_init_mss (tc);
618  scoreboard_init (&tc->sack_sb);
619  tcp_cc_init (tc);
620  if (tc->state == TCP_STATE_SYN_RCVD)
621  tcp_init_snd_vars (tc);
622 
623  if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6))
624  tcp_add_del_adjacency (tc, 1);
625 
626  /* tcp_connection_fib_attach (tc); */
627 
628  if (transport_connection_is_tx_paced (&tc->connection)
629  || tcp_main.tx_pacing)
630  tcp_enable_pacing (tc);
631 }
632 
633 static int
634 tcp_alloc_custom_local_endpoint (tcp_main_t * tm, ip46_address_t * lcl_addr,
635  u16 * lcl_port, u8 is_ip4)
636 {
637  int index, port;
638  if (is_ip4)
639  {
640  index = tm->last_v4_address_rotor++;
641  if (tm->last_v4_address_rotor >= vec_len (tm->ip4_src_addresses))
642  tm->last_v4_address_rotor = 0;
643  lcl_addr->ip4.as_u32 = tm->ip4_src_addresses[index].as_u32;
644  }
645  else
646  {
647  index = tm->last_v6_address_rotor++;
648  if (tm->last_v6_address_rotor >= vec_len (tm->ip6_src_addresses))
649  tm->last_v6_address_rotor = 0;
650  clib_memcpy_fast (&lcl_addr->ip6, &tm->ip6_src_addresses[index],
651  sizeof (ip6_address_t));
652  }
654  if (port < 1)
655  {
656  clib_warning ("Failed to allocate src port");
657  return -1;
658  }
659  *lcl_port = port;
660  return 0;
661 }
662 
663 static int
665 {
666  tcp_main_t *tm = vnet_get_tcp_main ();
667  tcp_connection_t *tc;
668  ip46_address_t lcl_addr;
669  u16 lcl_port;
670  int rv;
671 
672  /*
673  * Allocate local endpoint
674  */
675  if ((rmt->is_ip4 && vec_len (tm->ip4_src_addresses))
676  || (!rmt->is_ip4 && vec_len (tm->ip6_src_addresses)))
677  rv = tcp_alloc_custom_local_endpoint (tm, &lcl_addr, &lcl_port,
678  rmt->is_ip4);
679  else
681  rmt, &lcl_addr, &lcl_port);
682 
683  if (rv)
684  return -1;
685 
686  /*
687  * Create connection and send SYN
688  */
689  clib_spinlock_lock_if_init (&tm->half_open_lock);
691  ip_copy (&tc->c_rmt_ip, &rmt->ip, rmt->is_ip4);
692  ip_copy (&tc->c_lcl_ip, &lcl_addr, rmt->is_ip4);
693  tc->c_rmt_port = rmt->port;
694  tc->c_lcl_port = clib_host_to_net_u16 (lcl_port);
695  tc->c_is_ip4 = rmt->is_ip4;
696  tc->c_proto = TRANSPORT_PROTO_TCP;
697  tc->c_fib_index = rmt->fib_index;
698  /* The other connection vars will be initialized after SYN ACK */
700 
701  TCP_EVT_DBG (TCP_EVT_OPEN, tc);
702  tc->state = TCP_STATE_SYN_SENT;
703  tcp_init_snd_vars (tc);
704  tcp_send_syn (tc);
705  clib_spinlock_unlock_if_init (&tm->half_open_lock);
706 
707  return tc->c_c_index;
708 }
709 
710 const char *tcp_dbg_evt_str[] = {
711 #define _(sym, str) str,
713 #undef _
714 };
715 
716 const char *tcp_fsm_states[] = {
717 #define _(sym, str) str,
719 #undef _
720 };
721 
722 u8 *
723 format_tcp_state (u8 * s, va_list * args)
724 {
725  u32 state = va_arg (*args, u32);
726 
727  if (state < TCP_N_STATES)
728  s = format (s, "%s", tcp_fsm_states[state]);
729  else
730  s = format (s, "UNKNOWN (%d (0x%x))", state, state);
731  return s;
732 }
733 
734 const char *tcp_connection_flags_str[] = {
735 #define _(sym, str) str,
737 #undef _
738 };
739 
740 static u8 *
741 format_tcp_connection_flags (u8 * s, va_list * args)
742 {
743  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
744  int i, last = -1;
745 
746  for (i = 0; i < TCP_CONN_N_FLAG_BITS; i++)
747  if (tc->flags & (1 << i))
748  last = i;
749  for (i = 0; i < last; i++)
750  {
751  if (tc->flags & (1 << i))
752  s = format (s, "%s, ", tcp_connection_flags_str[i]);
753  }
754  if (last >= 0)
755  s = format (s, "%s", tcp_connection_flags_str[last]);
756  return s;
757 }
758 
759 const char *tcp_conn_timers[] = {
760 #define _(sym, str) str,
762 #undef _
763 };
764 
765 static u8 *
766 format_tcp_timers (u8 * s, va_list * args)
767 {
768  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
769  int i, last = -1;
770 
771  for (i = 0; i < TCP_N_TIMERS; i++)
772  if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
773  last = i;
774 
775  for (i = 0; i < last; i++)
776  {
777  if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
778  s = format (s, "%s,", tcp_conn_timers[i]);
779  }
780 
781  if (last >= 0)
782  s = format (s, "%s", tcp_conn_timers[i]);
783 
784  return s;
785 }
786 
787 static u8 *
788 format_tcp_congestion_status (u8 * s, va_list * args)
789 {
790  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
791  if (tcp_in_recovery (tc))
792  s = format (s, "recovery");
793  else if (tcp_in_fastrecovery (tc))
794  s = format (s, "fastrecovery");
795  else
796  s = format (s, "none");
797  return s;
798 }
799 
800 static i32
802 {
803  return (i32) tc->rcv_wnd - (tc->rcv_nxt - tc->rcv_las);
804 }
805 
806 static u8 *
807 format_tcp_congestion (u8 * s, va_list * args)
808 {
809  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
810  u32 indent = format_get_indent (s);
811 
812  s = format (s, "%U ", format_tcp_congestion_status, tc);
813  s = format (s, "cwnd %u ssthresh %u rtx_bytes %u bytes_acked %u\n",
814  tc->cwnd, tc->ssthresh, tc->snd_rxt_bytes, tc->bytes_acked);
815  s = format (s, "%Ucc space %u prev_ssthresh %u snd_congestion %u"
816  " dupack %u\n", format_white_space, indent,
817  tcp_available_cc_snd_space (tc), tc->prev_ssthresh,
818  tc->snd_congestion - tc->iss, tc->rcv_dupacks);
819  s = format (s, "%Utsecr %u tsecr_last_ack %u limited_transmit %u\n",
820  format_white_space, indent, tc->rcv_opts.tsecr,
821  tc->tsecr_last_ack, tc->limited_transmit - tc->iss);
822  return s;
823 }
824 
825 static u8 *
826 format_tcp_vars (u8 * s, va_list * args)
827 {
828  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
829  s = format (s, " index: %u flags: %U timers: %U\n", tc->c_c_index,
831  s = format (s, " snd_una %u snd_nxt %u snd_una_max %u",
832  tc->snd_una - tc->iss, tc->snd_nxt - tc->iss,
833  tc->snd_una_max - tc->iss);
834  s = format (s, " rcv_nxt %u rcv_las %u\n",
835  tc->rcv_nxt - tc->irs, tc->rcv_las - tc->irs);
836  s = format (s, " snd_wnd %u rcv_wnd %u rcv_wscale %u ",
837  tc->snd_wnd, tc->rcv_wnd, tc->rcv_wscale);
838  s = format (s, "snd_wl1 %u snd_wl2 %u\n", tc->snd_wl1 - tc->irs,
839  tc->snd_wl2 - tc->iss);
840  s = format (s, " flight size %u out space %u rcv_wnd_av %u\n",
842  tcp_rcv_wnd_available (tc));
843  s = format (s, " tsval_recent %u tsval_recent_age %u\n", tc->tsval_recent,
844  tcp_time_now () - tc->tsval_recent_age);
845  s = format (s, " rto %u rto_boff %u srtt %u us %.3f rttvar %u rtt_ts %x",
846  tc->rto, tc->rto_boff, tc->srtt, tc->mrtt_us * 1000, tc->rttvar,
847  tc->rtt_ts);
848  s = format (s, " rtt_seq %u\n", tc->rtt_seq - tc->iss);
849  s = format (s, " cong: %U", format_tcp_congestion, tc);
850 
851  if (tc->state >= TCP_STATE_ESTABLISHED)
852  {
853  s = format (s, " sboard: %U\n", format_tcp_scoreboard, &tc->sack_sb,
854  tc);
855  }
856  if (vec_len (tc->snd_sacks))
857  s = format (s, " sacks tx: %U\n", format_tcp_sacks, tc);
858 
859  return s;
860 }
861 
862 static u8 *
863 format_tcp_connection_id (u8 * s, va_list * args)
864 {
865  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
866  if (!tc)
867  return s;
868  if (tc->c_is_ip4)
869  {
870  s = format (s, "[%d:%d][%s] %U:%d->%U:%d", tc->c_thread_index,
871  tc->c_s_index, "T", format_ip4_address, &tc->c_lcl_ip4,
872  clib_net_to_host_u16 (tc->c_lcl_port), format_ip4_address,
873  &tc->c_rmt_ip4, clib_net_to_host_u16 (tc->c_rmt_port));
874  }
875  else
876  {
877  s = format (s, "[%d:%d][%s] %U:%d->%U:%d", tc->c_thread_index,
878  tc->c_s_index, "T", format_ip6_address, &tc->c_lcl_ip6,
879  clib_net_to_host_u16 (tc->c_lcl_port), format_ip6_address,
880  &tc->c_rmt_ip6, clib_net_to_host_u16 (tc->c_rmt_port));
881  }
882 
883  return s;
884 }
885 
886 u8 *
887 format_tcp_connection (u8 * s, va_list * args)
888 {
889  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
890  u32 verbose = va_arg (*args, u32);
891 
892  if (!tc)
893  return s;
894  s = format (s, "%-50U", format_tcp_connection_id, tc);
895  if (verbose)
896  {
897  s = format (s, "%-15U", format_tcp_state, tc->state);
898  if (verbose > 1)
899  s = format (s, "\n%U", format_tcp_vars, tc);
900  }
901 
902  return s;
903 }
904 
905 static u8 *
906 format_tcp_session (u8 * s, va_list * args)
907 {
908  u32 tci = va_arg (*args, u32);
909  u32 thread_index = va_arg (*args, u32);
910  u32 verbose = va_arg (*args, u32);
911  tcp_connection_t *tc;
912 
913  tc = tcp_connection_get (tci, thread_index);
914  if (tc)
915  s = format (s, "%U", format_tcp_connection, tc, verbose);
916  else
917  s = format (s, "empty\n");
918  return s;
919 }
920 
921 static u8 *
922 format_tcp_listener_session (u8 * s, va_list * args)
923 {
924  u32 tci = va_arg (*args, u32);
926  return format (s, "%U", format_tcp_connection_id, tc);
927 }
928 
929 static u8 *
930 format_tcp_half_open_session (u8 * s, va_list * args)
931 {
932  u32 tci = va_arg (*args, u32);
934  return format (s, "%U", format_tcp_connection_id, tc);
935 }
936 
937 u8 *
938 format_tcp_sacks (u8 * s, va_list * args)
939 {
940  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
941  sack_block_t *sacks = tc->snd_sacks;
942  sack_block_t *block;
943  int i, len = 0;
944 
945  len = vec_len (sacks);
946  for (i = 0; i < len - 1; i++)
947  {
948  block = &sacks[i];
949  s = format (s, " start %u end %u\n", block->start - tc->irs,
950  block->end - tc->irs);
951  }
952  if (len)
953  {
954  block = &sacks[len - 1];
955  s = format (s, " start %u end %u", block->start - tc->irs,
956  block->end - tc->irs);
957  }
958  return s;
959 }
960 
961 u8 *
962 format_tcp_rcv_sacks (u8 * s, va_list * args)
963 {
964  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
965  sack_block_t *sacks = tc->rcv_opts.sacks;
966  sack_block_t *block;
967  int i, len = 0;
968 
969  len = vec_len (sacks);
970  for (i = 0; i < len - 1; i++)
971  {
972  block = &sacks[i];
973  s = format (s, " start %u end %u\n", block->start - tc->iss,
974  block->end - tc->iss);
975  }
976  if (len)
977  {
978  block = &sacks[len - 1];
979  s = format (s, " start %u end %u", block->start - tc->iss,
980  block->end - tc->iss);
981  }
982  return s;
983 }
984 
985 static u8 *
986 format_tcp_sack_hole (u8 * s, va_list * args)
987 {
988  sack_scoreboard_hole_t *hole = va_arg (*args, sack_scoreboard_hole_t *);
989  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
990  if (tc)
991  s = format (s, " [%u, %u]", hole->start - tc->iss, hole->end - tc->iss);
992  else
993  s = format (s, " [%u, %u]", hole->start, hole->end);
994  return s;
995 }
996 
997 u8 *
998 format_tcp_scoreboard (u8 * s, va_list * args)
999 {
1000  sack_scoreboard_t *sb = va_arg (*args, sack_scoreboard_t *);
1001  tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
1002  sack_scoreboard_hole_t *hole;
1003  u32 indent = format_get_indent (s);
1004 
1005  s = format (s, "sacked_bytes %u last_sacked_bytes %u lost_bytes %u\n",
1006  sb->sacked_bytes, sb->last_sacked_bytes, sb->lost_bytes);
1007  s = format (s, "%Ulast_bytes_delivered %u high_sacked %u snd_una_adv %u\n",
1008  format_white_space, indent, sb->last_bytes_delivered,
1009  sb->high_sacked - tc->iss, sb->snd_una_adv);
1010  s = format (s, "%Ucur_rxt_hole %u high_rxt %u rescue_rxt %u",
1011  format_white_space, indent, sb->cur_rxt_hole,
1012  sb->high_rxt - tc->iss, sb->rescue_rxt - tc->iss);
1013 
1014  hole = scoreboard_first_hole (sb);
1015  if (hole)
1016  s = format (s, "\n%Uhead %u tail %u %u holes:\n%U", format_white_space,
1017  indent, sb->head, sb->tail, pool_elts (sb->holes),
1018  format_white_space, indent);
1019 
1020  while (hole)
1021  {
1022  s = format (s, "%U", format_tcp_sack_hole, hole, tc);
1023  hole = scoreboard_next_hole (sb, hole);
1024  }
1025 
1026  return s;
1027 }
1028 
1029 static transport_connection_t *
1030 tcp_session_get_transport (u32 conn_index, u32 thread_index)
1031 {
1032  tcp_connection_t *tc = tcp_connection_get (conn_index, thread_index);
1033  return &tc->connection;
1034 }
1035 
1036 static transport_connection_t *
1038 {
1040  return &tc->connection;
1041 }
1042 
1043 /**
1044  * Compute maximum segment size for session layer.
1045  *
1046  * Since the result needs to be the actual data length, it first computes
1047  * the tcp options to be used in the next burst and subtracts their
1048  * length from the connection's snd_mss.
1049  */
1050 static u16
1052 {
1053  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1054 
1055  /* Ensure snd_mss does accurately reflect the amount of data we can push
1056  * in a segment. This also makes sure that options are updated according to
1057  * the current state of the connection. */
1059 
1060  return tc->snd_mss;
1061 }
1062 
1065 {
1066  if (PREDICT_FALSE (tc->snd_wnd < tc->snd_mss))
1067  {
1068  return tc->snd_wnd <= snd_space ? tc->snd_wnd : 0;
1069  }
1070 
1071  /* If not snd_wnd constrained and we can't write at least a segment,
1072  * don't try at all */
1073  if (PREDICT_FALSE (snd_space < tc->snd_mss))
1074  return snd_space < tc->cwnd ? 0 : snd_space;
1075 
1076  /* round down to mss multiple */
1077  return snd_space - (snd_space % tc->snd_mss);
1078 }
1079 
1080 /**
1081  * Compute tx window session is allowed to fill.
1082  *
1083  * Takes into account available send space, snd_mss and the congestion
1084  * state of the connection. If possible, the value returned is a multiple
1085  * of snd_mss.
1086  *
1087  * @param tc tcp connection
1088  * @return number of bytes session is allowed to write
1089  */
1090 static inline u32
1092 {
1093  int snd_space, snt_limited;
1094 
1096  || tc->state == TCP_STATE_CLOSED))
1097  return 0;
1098 
1099  snd_space = tcp_available_output_snd_space (tc);
1100 
1101  /* If we haven't gotten dupacks or if we did and have gotten sacked
1102  * bytes then we can still send as per Limited Transmit (RFC3042) */
1103  if (PREDICT_FALSE (tc->rcv_dupacks != 0
1104  && (tcp_opts_sack_permitted (tc)
1105  && tc->sack_sb.last_sacked_bytes == 0)))
1106  {
1107  if (tc->rcv_dupacks == 1 && tc->limited_transmit != tc->snd_nxt)
1108  tc->limited_transmit = tc->snd_nxt;
1109  ASSERT (seq_leq (tc->limited_transmit, tc->snd_nxt));
1110 
1111  snt_limited = tc->snd_nxt - tc->limited_transmit;
1112  snd_space = clib_max (2 * tc->snd_mss - snt_limited, 0);
1113  }
1114  return tcp_round_snd_space (tc, snd_space);
1115 }
1116 
1117 u32
1119 {
1120  return tcp_snd_space_inline (tc);
1121 }
1122 
1123 static u32
1125 {
1126  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1127  return clib_min (tcp_snd_space_inline (tc),
1128  tc->snd_wnd - (tc->snd_nxt - tc->snd_una));
1129 }
1130 
1131 static u32
1133 {
1134  tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
1135 
1136  ASSERT (seq_geq (tc->snd_nxt, tc->snd_una));
1137 
1138  /* This still works if fast retransmit is on */
1139  return (tc->snd_nxt - tc->snd_una);
1140 }
1141 
1142 static void
1143 tcp_update_time (f64 now, u8 thread_index)
1144 {
1145  tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index);
1146 
1147  tcp_set_time_now (wrk);
1148  tw_timer_expire_timers_16t_2w_512sl (&wrk->timer_wheel, now);
1149  tcp_do_fastretransmits (wrk);
1150  tcp_send_acks (wrk);
1152 }
1153 
1154 static u32
1156 {
1157  tcp_connection_t *tc = (tcp_connection_t *) tconn;
1158  return tcp_push_header (tc, b);
1159 }
1160 
1161 static void
1163 {
1164  tcp_connection_t *tc = (tcp_connection_t *) tconn;
1165  if (tc->flags & TCP_CONN_PSH_PENDING)
1166  return;
1167  tc->flags |= TCP_CONN_PSH_PENDING;
1168  tc->psh_seq = tc->snd_una_max + transport_max_tx_dequeue (tconn) - 1;
1169 }
1170 
1171 /* *INDENT-OFF* */
1173  .enable = vnet_tcp_enable_disable,
1174  .bind = tcp_session_bind,
1175  .unbind = tcp_session_unbind,
1176  .push_header = tcp_session_push_header,
1177  .get_connection = tcp_session_get_transport,
1178  .get_listener = tcp_session_get_listener,
1179  .get_half_open = tcp_half_open_session_get_transport,
1180  .open = tcp_session_open,
1181  .close = tcp_session_close,
1182  .cleanup = tcp_session_cleanup,
1183  .send_mss = tcp_session_send_mss,
1184  .send_space = tcp_session_send_space,
1185  .update_time = tcp_update_time,
1186  .tx_fifo_offset = tcp_session_tx_fifo_offset,
1187  .flush_data = tcp_session_flush_data,
1188  .format_connection = format_tcp_session,
1189  .format_listener = format_tcp_listener_session,
1190  .format_half_open = format_tcp_half_open_session,
1191  .tx_type = TRANSPORT_TX_PEEK,
1192  .service_type = TRANSPORT_SERVICE_VC,
1193 };
1194 /* *INDENT-ON* */
1195 
1196 void
1198 {
1199  f64 srtt;
1200  u64 rate;
1201 
1202  if (!transport_connection_is_tx_paced (&tc->connection))
1203  return;
1204 
1205  srtt = clib_min ((f64) tc->srtt * TCP_TICK, tc->mrtt_us);
1206  /* TODO should constrain to interface's max throughput but
1207  * we don't have link speeds for sw ifs ..*/
1208  rate = tc->cwnd / srtt;
1209  transport_connection_tx_pacer_update (&tc->connection, rate);
1210 }
1211 
1212 void
1214  u32 start_bucket)
1215 {
1216  tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1217  u32 byte_rate = window / ((f64) TCP_TICK * tc->srtt);
1218  u64 last_time = wrk->vm->clib_time.last_cpu_time;
1219  transport_connection_tx_pacer_reset (&tc->connection, byte_rate,
1220  start_bucket, last_time);
1221 }
1222 
1223 static void
1225 {
1226  u32 thread_index = vlib_get_thread_index ();
1227  tcp_connection_t *tc;
1228 
1229  tc = tcp_connection_get (conn_index, thread_index);
1230  tc->timers[TCP_TIMER_KEEP] = TCP_TIMER_HANDLE_INVALID;
1231 
1232  tcp_connection_close (tc);
1233 }
1234 
1235 static void
1237 {
1238  tcp_connection_t *tc;
1239 
1240  tc = tcp_connection_get (conn_index, vlib_get_thread_index ());
1241  /* note: the connection may have already disappeared */
1242  if (PREDICT_FALSE (tc == 0))
1243  return;
1244  ASSERT (tc->state == TCP_STATE_SYN_RCVD);
1245  /* Start cleanup. App wasn't notified yet so use delete notify as
1246  * opposed to delete to cleanup session layer state. */
1247  session_transport_delete_notify (&tc->connection);
1248  tc->timers[TCP_TIMER_ESTABLISH] = TCP_TIMER_HANDLE_INVALID;
1250 }
1251 
1252 static void
1254 {
1255  tcp_connection_t *tc;
1256 
1257  tc = tcp_half_open_connection_get (conn_index);
1258  if (!tc)
1259  return;
1260 
1261  ASSERT (tc->state == TCP_STATE_SYN_SENT);
1262  /* Notify app if we haven't tried to clean this up already */
1263  if (!(tc->flags & TCP_CONN_HALF_OPEN_DONE))
1264  session_stream_connect_notify (&tc->connection, 1 /* fail */ );
1265 
1266  tc->timers[TCP_TIMER_ESTABLISH_AO] = TCP_TIMER_HANDLE_INVALID;
1268 }
1269 
1270 static void
1272 {
1273  u32 thread_index = vlib_get_thread_index (), rto;
1274  tcp_connection_t *tc;
1275 
1276  tc = tcp_connection_get (conn_index, thread_index);
1277  if (!tc)
1278  return;
1279  tc->timers[TCP_TIMER_WAITCLOSE] = TCP_TIMER_HANDLE_INVALID;
1280 
1281  switch (tc->state)
1282  {
1283  case TCP_STATE_CLOSE_WAIT:
1285  session_transport_closed_notify (&tc->connection);
1286 
1287  if (!(tc->flags & TCP_CONN_FINPNDG))
1288  {
1289  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1290  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
1291  break;
1292  }
1293 
1294  /* Session didn't come back with a close. Send FIN either way
1295  * and switch to LAST_ACK. */
1296  tcp_cong_recovery_off (tc);
1297  /* Make sure we don't try to send unsent data */
1298  tc->snd_una_max = tc->snd_nxt = tc->snd_una;
1299  tcp_send_fin (tc);
1300  tcp_connection_set_state (tc, TCP_STATE_LAST_ACK);
1301 
1302  /* Make sure we don't wait in LAST ACK forever */
1303  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
1304 
1305  /* Don't delete the connection yet */
1306  break;
1307  case TCP_STATE_FIN_WAIT_1:
1309  if (tc->flags & TCP_CONN_FINPNDG)
1310  {
1311  /* If FIN pending send it before closing and wait as long as
1312  * the rto timeout would wait. Notify session layer that transport
1313  * is closed. We haven't sent everything but we did try. */
1314  tcp_cong_recovery_off (tc);
1315  tcp_send_fin (tc);
1316  rto = clib_max ((tc->rto >> tc->rto_boff) * TCP_TO_TIMER_TICK, 1);
1317  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE,
1318  clib_min (rto, TCP_2MSL_TIME));
1319  session_transport_closed_notify (&tc->connection);
1320  }
1321  else
1322  {
1323  /* We've sent the fin but no progress. Close the connection and
1324  * to make sure everything is flushed, setup a cleanup timer */
1325  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1326  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
1327  }
1328  break;
1329  case TCP_STATE_LAST_ACK:
1330  case TCP_STATE_CLOSING:
1332  tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1333  tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
1334  session_transport_closed_notify (&tc->connection);
1335  break;
1336  default:
1337  tcp_connection_del (tc);
1338  break;
1339  }
1340 }
1341 
1342 /* *INDENT-OFF* */
1344 {
1353 };
1354 /* *INDENT-ON* */
1355 
1356 static void
1358 {
1359  int i;
1360  u32 connection_index, timer_id;
1361 
1362  for (i = 0; i < vec_len (expired_timers); i++)
1363  {
1364  /* Get session index and timer id */
1365  connection_index = expired_timers[i] & 0x0FFFFFFF;
1366  timer_id = expired_timers[i] >> 28;
1367 
1368  TCP_EVT_DBG (TCP_EVT_TIMER_POP, connection_index, timer_id);
1369 
1370  /* Handle expiration */
1371  (*timer_expiration_handlers[timer_id]) (connection_index);
1372  }
1373 }
1374 
1375 static void
1377 {
1378  tw_timer_wheel_16t_2w_512sl_t *tw;
1379  /* *INDENT-OFF* */
1380  foreach_vlib_main (({
1381  tw = &tm->wrk_ctx[ii].timer_wheel;
1382  tw_timer_wheel_init_16t_2w_512sl (tw, tcp_expired_timers_dispatch,
1383  100e-3 /* timer period 100ms */ , ~0);
1384  tw->last_run_time = vlib_time_now (this_vlib_main);
1385  }));
1386  /* *INDENT-ON* */
1387 }
1388 
1389 static void
1391 {
1392  u32 default_seed = random_default_seed ();
1393  u64 time_now = clib_cpu_time_now ();
1394 
1395  tm->iss_seed.first = (u64) random_u32 (&default_seed) << 32;
1396  tm->iss_seed.second = random_u64 (&time_now);
1397 }
1398 
1399 static clib_error_t *
1401 {
1403  u32 num_threads, n_workers, prealloc_conn_per_wrk;
1404  tcp_connection_t *tc __attribute__ ((unused));
1405  tcp_main_t *tm = vnet_get_tcp_main ();
1406  clib_error_t *error = 0;
1407  int thread;
1408 
1409  if ((error = vlib_call_init_function (vm, ip_main_init)))
1410  return error;
1411  if ((error = vlib_call_init_function (vm, ip4_lookup_init)))
1412  return error;
1413  if ((error = vlib_call_init_function (vm, ip6_lookup_init)))
1414  return error;
1415 
1416  /*
1417  * Registrations
1418  */
1419 
1420  ip4_register_protocol (IP_PROTOCOL_TCP, tcp4_input_node.index);
1421  ip6_register_protocol (IP_PROTOCOL_TCP, tcp6_input_node.index);
1422 
1423  /*
1424  * Initialize data structures
1425  */
1426 
1427  num_threads = 1 /* main thread */ + vtm->n_threads;
1428  vec_validate (tm->connections, num_threads - 1);
1429  vec_validate (tm->wrk_ctx, num_threads - 1);
1430  n_workers = num_threads == 1 ? 1 : vtm->n_threads;
1431  prealloc_conn_per_wrk = tm->preallocated_connections / n_workers;
1432 
1433  for (thread = 0; thread < num_threads; thread++)
1434  {
1435  vec_validate (tm->wrk_ctx[thread].pending_fast_rxt, 255);
1436  vec_validate (tm->wrk_ctx[thread].ongoing_fast_rxt, 255);
1437  vec_validate (tm->wrk_ctx[thread].postponed_fast_rxt, 255);
1438  vec_validate (tm->wrk_ctx[thread].pending_deq_acked, 255);
1439  vec_validate (tm->wrk_ctx[thread].pending_acks, 255);
1440  vec_validate (tm->wrk_ctx[thread].pending_disconnects, 255);
1441  vec_reset_length (tm->wrk_ctx[thread].pending_fast_rxt);
1442  vec_reset_length (tm->wrk_ctx[thread].ongoing_fast_rxt);
1443  vec_reset_length (tm->wrk_ctx[thread].postponed_fast_rxt);
1444  vec_reset_length (tm->wrk_ctx[thread].pending_deq_acked);
1445  vec_reset_length (tm->wrk_ctx[thread].pending_acks);
1446  vec_reset_length (tm->wrk_ctx[thread].pending_disconnects);
1447  tm->wrk_ctx[thread].vm = vlib_mains[thread];
1448 
1449  /*
1450  * Preallocate connections. Assume that thread 0 won't
1451  * use preallocated threads when running multi-core
1452  */
1453  if ((thread > 0 || num_threads == 1) && prealloc_conn_per_wrk)
1454  pool_init_fixed (tm->connections[thread], prealloc_conn_per_wrk);
1455  }
1456 
1457  /*
1458  * Use a preallocated half-open connection pool?
1459  */
1460  if (tm->preallocated_half_open_connections)
1461  pool_init_fixed (tm->half_open_connections,
1462  tm->preallocated_half_open_connections);
1463 
1464  /* Initialize clocks per tick for TCP timestamp. Used to compute
1465  * monotonically increasing timestamps. */
1466  tm->tstamp_ticks_per_clock = vm->clib_time.seconds_per_clock
1468 
1469  if (num_threads > 1)
1470  {
1471  clib_spinlock_init (&tm->half_open_lock);
1472  }
1473 
1476 
1477  tm->bytes_per_buffer = VLIB_BUFFER_DATA_SIZE;
1478 
1479  return error;
1480 }
1481 
1482 clib_error_t *
1484 {
1485  if (is_en)
1486  {
1487  if (tcp_main.is_enabled)
1488  return 0;
1489 
1490  return tcp_main_enable (vm);
1491  }
1492  else
1493  {
1494  tcp_main.is_enabled = 0;
1495  }
1496 
1497  return 0;
1498 }
1499 
1500 void
1501 tcp_punt_unknown (vlib_main_t * vm, u8 is_ip4, u8 is_add)
1502 {
1503  tcp_main_t *tm = &tcp_main;
1504  if (is_ip4)
1505  tm->punt_unknown4 = is_add;
1506  else
1507  tm->punt_unknown6 = is_add;
1508 }
1509 
1510 static clib_error_t *
1512 {
1513  tcp_main_t *tm = vnet_get_tcp_main ();
1514  ip_main_t *im = &ip_main;
1515  ip_protocol_info_t *pi;
1516 
1517  /* Session layer, and by implication tcp, are disabled by default */
1518  tm->is_enabled = 0;
1519 
1520  /* Register with IP for header parsing */
1521  pi = ip_get_protocol_info (im, IP_PROTOCOL_TCP);
1522  if (pi == 0)
1523  return clib_error_return (0, "TCP protocol info AWOL");
1526 
1527  /* Register as transport with session layer */
1532 
1533  tcp_api_reference ();
1534  tm->tx_pacing = 1;
1535  tm->cc_algo = TCP_CC_NEWRENO;
1536  return 0;
1537 }
1538 
1540 
1541 uword
1543 {
1544  uword *result = va_arg (*va, uword *);
1545 
1546  if (unformat (input, "newreno"))
1547  *result = TCP_CC_NEWRENO;
1548  else if (unformat (input, "cubic"))
1549  *result = TCP_CC_CUBIC;
1550  else
1551  return 0;
1552 
1553  return 1;
1554 }
1555 
1556 uword
1558 {
1559  tcp_main_t *tm = vnet_get_tcp_main ();
1560  tcp_cc_algorithm_t *cc_alg;
1561  unformat_input_t sub_input;
1562  int found = 0;
1563 
1564  vec_foreach (cc_alg, tm->cc_algos)
1565  {
1566  if (!unformat (input, cc_alg->name))
1567  continue;
1568 
1569  if (cc_alg->unformat_cfg
1570  && unformat (input, "%U", unformat_vlib_cli_sub_input, &sub_input))
1571  {
1572  if (cc_alg->unformat_cfg (&sub_input))
1573  found = 1;
1574  }
1575  }
1576  return found;
1577 }
1578 
1579 static clib_error_t *
1581 {
1582  tcp_main_t *tm = vnet_get_tcp_main ();
1583 
1584  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1585  {
1586  if (unformat (input, "preallocated-connections %d",
1587  &tm->preallocated_connections))
1588  ;
1589  else if (unformat (input, "preallocated-half-open-connections %d",
1590  &tm->preallocated_half_open_connections))
1591  ;
1592  else if (unformat (input, "buffer-fail-fraction %f",
1593  &tm->buffer_fail_fraction))
1594  ;
1595  else if (unformat (input, "max-rx-fifo %U", unformat_memory_size,
1596  &tm->max_rx_fifo))
1597  ;
1598  else if (unformat (input, "no-tx-pacing"))
1599  tm->tx_pacing = 0;
1600  else if (unformat (input, "cc-algo %U", unformat_tcp_cc_algo,
1601  &tm->cc_algo))
1602  ;
1603  else if (unformat (input, "%U", unformat_tcp_cc_algo_cfg))
1604  ;
1605  else
1606  return clib_error_return (0, "unknown input `%U'",
1607  format_unformat_error, input);
1608  }
1609  return 0;
1610 }
1611 
1613 
1614 
1615 /**
1616  * \brief Configure an ipv4 source address range
1617  * @param vm vlib_main_t pointer
1618  * @param start first ipv4 address in the source address range
1619  * @param end last ipv4 address in the source address range
1620  * @param table_id VRF / table ID, 0 for the default FIB
1621  * @return 0 if all OK, else an error indication from api_errno.h
1622  */
1623 
1624 int
1626  ip4_address_t * start,
1627  ip4_address_t * end, u32 table_id)
1628 {
1629  tcp_main_t *tm = vnet_get_tcp_main ();
1630  vnet_main_t *vnm = vnet_get_main ();
1631  u32 start_host_byte_order, end_host_byte_order;
1633  vnet_sw_interface_t *si;
1634  fib_node_index_t fei;
1635  u32 fib_index = 0;
1636  u32 sw_if_index;
1637  int rv;
1638  int vnet_proxy_arp_add_del (ip4_address_t * lo_addr,
1639  ip4_address_t * hi_addr, u32 fib_index,
1640  int is_del);
1641 
1642  clib_memset (&prefix, 0, sizeof (prefix));
1643 
1644  fib_index = fib_table_find (FIB_PROTOCOL_IP4, table_id);
1645 
1646  if (fib_index == ~0)
1647  return VNET_API_ERROR_NO_SUCH_FIB;
1648 
1649  start_host_byte_order = clib_net_to_host_u32 (start->as_u32);
1650  end_host_byte_order = clib_net_to_host_u32 (end->as_u32);
1651 
1652  /* sanity check for reversed args or some such */
1653  if ((end_host_byte_order - start_host_byte_order) > (10 << 10))
1654  return VNET_API_ERROR_INVALID_ARGUMENT;
1655 
1656  /* Lookup the last address, to identify the interface involved */
1657  prefix.fp_len = 32;
1658  prefix.fp_proto = FIB_PROTOCOL_IP4;
1659  memcpy (&prefix.fp_addr.ip4, end, sizeof (ip4_address_t));
1660 
1661  fei = fib_table_lookup (fib_index, &prefix);
1662 
1663  /* Couldn't find route to destination. Bail out. */
1664  if (fei == FIB_NODE_INDEX_INVALID)
1665  return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1666 
1667  sw_if_index = fib_entry_get_resolving_interface (fei);
1668 
1669  /* Enable proxy arp on the interface */
1670  si = vnet_get_sw_interface (vnm, sw_if_index);
1672 
1673  /* Configure proxy arp across the range */
1674  rv = vnet_proxy_arp_add_del (start, end, fib_index, 0 /* is_del */ );
1675 
1676  if (rv)
1677  return rv;
1678 
1679  do
1680  {
1681  dpo_id_t dpo = DPO_INVALID;
1682 
1683  vec_add1 (tm->ip4_src_addresses, start[0]);
1684 
1685  /* Add local adjacencies for the range */
1686 
1687  receive_dpo_add_or_lock (DPO_PROTO_IP4, ~0 /* sw_if_index */ ,
1688  NULL, &dpo);
1689  prefix.fp_len = 32;
1690  prefix.fp_proto = FIB_PROTOCOL_IP4;
1691  prefix.fp_addr.ip4.as_u32 = start->as_u32;
1692 
1694  &prefix,
1696  FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
1697  dpo_reset (&dpo);
1698 
1699  start_host_byte_order++;
1700  start->as_u32 = clib_host_to_net_u32 (start_host_byte_order);
1701  }
1702  while (start_host_byte_order <= end_host_byte_order);
1703 
1704  return 0;
1705 }
1706 
1707 /**
1708  * \brief Configure an ipv6 source address range
1709  * @param vm vlib_main_t pointer
1710  * @param start first ipv6 address in the source address range
1711  * @param end last ipv6 address in the source address range
1712  * @param table_id VRF / table ID, 0 for the default FIB
1713  * @return 0 if all OK, else an error indication from api_errno.h
1714  */
1715 
1716 int
1718  ip6_address_t * start,
1719  ip6_address_t * end, u32 table_id)
1720 {
1721  tcp_main_t *tm = vnet_get_tcp_main ();
1723  u32 fib_index = 0;
1724  fib_node_index_t fei;
1725  u32 sw_if_index;
1726 
1727  clib_memset (&prefix, 0, sizeof (prefix));
1728 
1729  fib_index = fib_table_find (FIB_PROTOCOL_IP6, table_id);
1730 
1731  if (fib_index == ~0)
1732  return VNET_API_ERROR_NO_SUCH_FIB;
1733 
1734  while (1)
1735  {
1736  int i;
1737  ip6_address_t tmp;
1738  dpo_id_t dpo = DPO_INVALID;
1739 
1740  /* Remember this address */
1741  vec_add1 (tm->ip6_src_addresses, start[0]);
1742 
1743  /* Lookup the prefix, to identify the interface involved */
1744  prefix.fp_len = 128;
1745  prefix.fp_proto = FIB_PROTOCOL_IP6;
1746  memcpy (&prefix.fp_addr.ip6, start, sizeof (ip6_address_t));
1747 
1748  fei = fib_table_lookup (fib_index, &prefix);
1749 
1750  /* Couldn't find route to destination. Bail out. */
1751  if (fei == FIB_NODE_INDEX_INVALID)
1752  return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1753 
1754  sw_if_index = fib_entry_get_resolving_interface (fei);
1755 
1756  if (sw_if_index == (u32) ~ 0)
1757  return VNET_API_ERROR_NO_MATCHING_INTERFACE;
1758 
1759  /* Add a proxy neighbor discovery entry for this address */
1760  ip6_neighbor_proxy_add_del (sw_if_index, start, 0 /* is_del */ );
1761 
1762  /* Add a receive adjacency for this address */
1763  receive_dpo_add_or_lock (DPO_PROTO_IP6, ~0 /* sw_if_index */ ,
1764  NULL, &dpo);
1765 
1767  &prefix,
1769  FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
1770  dpo_reset (&dpo);
1771 
1772  /* Done with the entire range? */
1773  if (!memcmp (start, end, sizeof (start[0])))
1774  break;
1775 
1776  /* Increment the address. DGMS. */
1777  tmp = start[0];
1778  for (i = 15; i >= 0; i--)
1779  {
1780  tmp.as_u8[i] += 1;
1781  if (tmp.as_u8[i] != 0)
1782  break;
1783  }
1784  start[0] = tmp;
1785  }
1786  return 0;
1787 }
1788 
1789 static clib_error_t *
1791  unformat_input_t * input, vlib_cli_command_t * cmd_arg)
1792 {
1793  ip4_address_t v4start, v4end;
1794  ip6_address_t v6start, v6end;
1795  u32 table_id = 0;
1796  int v4set = 0;
1797  int v6set = 0;
1798  int rv;
1799 
1800  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1801  {
1802  if (unformat (input, "%U - %U", unformat_ip4_address, &v4start,
1803  unformat_ip4_address, &v4end))
1804  v4set = 1;
1805  else if (unformat (input, "%U", unformat_ip4_address, &v4start))
1806  {
1807  memcpy (&v4end, &v4start, sizeof (v4start));
1808  v4set = 1;
1809  }
1810  else if (unformat (input, "%U - %U", unformat_ip6_address, &v6start,
1811  unformat_ip6_address, &v6end))
1812  v6set = 1;
1813  else if (unformat (input, "%U", unformat_ip6_address, &v6start))
1814  {
1815  memcpy (&v6end, &v6start, sizeof (v6start));
1816  v6set = 1;
1817  }
1818  else if (unformat (input, "fib-table %d", &table_id))
1819  ;
1820  else
1821  break;
1822  }
1823 
1824  if (!v4set && !v6set)
1825  return clib_error_return (0, "at least one v4 or v6 address required");
1826 
1827  if (v4set)
1828  {
1829  rv = tcp_configure_v4_source_address_range (vm, &v4start, &v4end,
1830  table_id);
1831  switch (rv)
1832  {
1833  case 0:
1834  break;
1835 
1836  case VNET_API_ERROR_NO_SUCH_FIB:
1837  return clib_error_return (0, "Invalid table-id %d", table_id);
1838 
1839  case VNET_API_ERROR_INVALID_ARGUMENT:
1840  return clib_error_return (0, "Invalid address range %U - %U",
1841  format_ip4_address, &v4start,
1842  format_ip4_address, &v4end);
1843  default:
1844  return clib_error_return (0, "error %d", rv);
1845  break;
1846  }
1847  }
1848  if (v6set)
1849  {
1850  rv = tcp_configure_v6_source_address_range (vm, &v6start, &v6end,
1851  table_id);
1852  switch (rv)
1853  {
1854  case 0:
1855  break;
1856 
1857  case VNET_API_ERROR_NO_SUCH_FIB:
1858  return clib_error_return (0, "Invalid table-id %d", table_id);
1859 
1860  default:
1861  return clib_error_return (0, "error %d", rv);
1862  break;
1863  }
1864  }
1865  return 0;
1866 }
1867 
1868 /* *INDENT-OFF* */
1869 VLIB_CLI_COMMAND (tcp_src_address_command, static) =
1870 {
1871  .path = "tcp src-address",
1872  .short_help = "tcp src-address <ip-addr> [- <ip-addr>] add src address range",
1873  .function = tcp_src_address,
1874 };
1875 /* *INDENT-ON* */
1876 
1877 static u8 *
1879 {
1880 #if TCP_SCOREBOARD_TRACE
1881 
1882  scoreboard_trace_elt_t *block;
1883  int i = 0;
1884 
1885  if (!sb->trace)
1886  return s;
1887 
1888  s = format (s, "scoreboard trace:");
1889  vec_foreach (block, sb->trace)
1890  {
1891  s = format (s, "{%u, %u, %u, %u, %u}, ", block->start, block->end,
1892  block->ack, block->snd_una_max, block->group);
1893  if ((++i % 3) == 0)
1894  s = format (s, "\n");
1895  }
1896  return s;
1897 #else
1898  return 0;
1899 #endif
1900 }
1901 
1902 static clib_error_t *
1904  vlib_cli_command_t * cmd_arg)
1905 {
1906  transport_connection_t *tconn = 0;
1907  tcp_connection_t *tc;
1908  u8 *s = 0;
1909  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1910  {
1911  if (unformat (input, "%U", unformat_transport_connection, &tconn,
1913  ;
1914  else
1915  return clib_error_return (0, "unknown input `%U'",
1916  format_unformat_error, input);
1917  }
1918 
1919  if (!TCP_SCOREBOARD_TRACE)
1920  {
1921  vlib_cli_output (vm, "scoreboard tracing not enabled");
1922  return 0;
1923  }
1924 
1925  tc = tcp_get_connection_from_transport (tconn);
1926  s = tcp_scoreboard_dump_trace (s, &tc->sack_sb);
1927  vlib_cli_output (vm, "%v", s);
1928  return 0;
1929 }
1930 
1931 /* *INDENT-OFF* */
1932 VLIB_CLI_COMMAND (tcp_show_scoreboard_trace_command, static) =
1933 {
1934  .path = "show tcp scoreboard trace",
1935  .short_help = "show tcp scoreboard trace <connection>",
1936  .function = tcp_show_scoreboard_trace_fn,
1937 };
1938 /* *INDENT-ON* */
1939 
1940 u8 *
1942 {
1943  int i, trace_len;
1945  u32 next_ack, left, group, has_new_ack = 0;
1946  tcp_connection_t _dummy_tc, *dummy_tc = &_dummy_tc;
1947  sack_block_t *block;
1948 
1949  if (!TCP_SCOREBOARD_TRACE)
1950  {
1951  s = format (s, "scoreboard tracing not enabled");
1952  return s;
1953  }
1954 
1955  if (!tc)
1956  return s;
1957 
1958  clib_memset (dummy_tc, 0, sizeof (*dummy_tc));
1959  tcp_connection_timers_init (dummy_tc);
1960  scoreboard_init (&dummy_tc->sack_sb);
1961  dummy_tc->rcv_opts.flags |= TCP_OPTS_FLAG_SACK;
1962 
1963 #if TCP_SCOREBOARD_TRACE
1964  trace = tc->sack_sb.trace;
1965  trace_len = vec_len (tc->sack_sb.trace);
1966 #endif
1967 
1968  for (i = 0; i < trace_len; i++)
1969  {
1970  if (trace[i].ack != 0)
1971  {
1972  dummy_tc->snd_una = trace[i].ack - 1448;
1973  dummy_tc->snd_una_max = trace[i].ack;
1974  }
1975  }
1976 
1977  left = 0;
1978  while (left < trace_len)
1979  {
1980  group = trace[left].group;
1981  vec_reset_length (dummy_tc->rcv_opts.sacks);
1982  has_new_ack = 0;
1983  while (trace[left].group == group)
1984  {
1985  if (trace[left].ack != 0)
1986  {
1987  if (verbose)
1988  s = format (s, "Adding ack %u, snd_una_max %u, segs: ",
1989  trace[left].ack, trace[left].snd_una_max);
1990  dummy_tc->snd_una_max = trace[left].snd_una_max;
1991  next_ack = trace[left].ack;
1992  has_new_ack = 1;
1993  }
1994  else
1995  {
1996  if (verbose)
1997  s = format (s, "[%u, %u], ", trace[left].start,
1998  trace[left].end);
1999  vec_add2 (dummy_tc->rcv_opts.sacks, block, 1);
2000  block->start = trace[left].start;
2001  block->end = trace[left].end;
2002  }
2003  left++;
2004  }
2005 
2006  /* Push segments */
2007  tcp_rcv_sacks (dummy_tc, next_ack);
2008  if (has_new_ack)
2009  dummy_tc->snd_una = next_ack + dummy_tc->sack_sb.snd_una_adv;
2010 
2011  if (verbose)
2012  s = format (s, "result: %U", format_tcp_scoreboard,
2013  &dummy_tc->sack_sb);
2014 
2015  }
2016  s = format (s, "result: %U", format_tcp_scoreboard, &dummy_tc->sack_sb);
2017 
2018  return s;
2019 }
2020 
2021 static clib_error_t *
2023  vlib_cli_command_t * cmd_arg)
2024 {
2025  transport_connection_t *tconn = 0;
2026  tcp_connection_t *tc = 0;
2027  u8 *str = 0;
2028  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
2029  {
2030  if (unformat (input, "%U", unformat_transport_connection, &tconn,
2032  ;
2033  else
2034  return clib_error_return (0, "unknown input `%U'",
2035  format_unformat_error, input);
2036  }
2037 
2038  if (!TCP_SCOREBOARD_TRACE)
2039  {
2040  vlib_cli_output (vm, "scoreboard tracing not enabled");
2041  return 0;
2042  }
2043 
2044  tc = tcp_get_connection_from_transport (tconn);
2045  if (!tc)
2046  {
2047  vlib_cli_output (vm, "connection not found");
2048  return 0;
2049  }
2050  str = tcp_scoreboard_replay (str, tc, 1);
2051  vlib_cli_output (vm, "%v", str);
2052  return 0;
2053 }
2054 
2055 /* *INDENT-OFF* */
2056 VLIB_CLI_COMMAND (tcp_replay_scoreboard_command, static) =
2057 {
2058  .path = "tcp replay scoreboard",
2059  .short_help = "tcp replay scoreboard <connection>",
2060  .function = tcp_scoreboard_trace_fn,
2061 };
2062 /* *INDENT-ON* */
2063 
2064 static clib_error_t *
2066  vlib_cli_command_t * cmd_arg)
2067 {
2068  tcp_main_t *tm = vnet_get_tcp_main ();
2070  return clib_error_return (0, "unknown input `%U'", format_unformat_error,
2071  input);
2072  vlib_cli_output (vm, "IPv4 TCP punt: %s",
2073  tm->punt_unknown4 ? "enabled" : "disabled");
2074  vlib_cli_output (vm, "IPv6 TCP punt: %s",
2075  tm->punt_unknown6 ? "enabled" : "disabled");
2076  return 0;
2077 }
2078 /* *INDENT-OFF* */
2079 VLIB_CLI_COMMAND (show_tcp_punt_command, static) =
2080 {
2081  .path = "show tcp punt",
2082  .short_help = "show tcp punt",
2083  .function = show_tcp_punt_fn,
2084 };
2085 /* *INDENT-ON* */
2086 
2087 /*
2088  * fd.io coding-style-patch-verification: ON
2089  *
2090  * Local Variables:
2091  * eval: (c-set-style "gnu")
2092  * End:
2093  */
static void tcp_session_close(u32 conn_index, u32 thread_index)
Definition: tcp.c:389
#define vec_validate(V, I)
Make sure vector is long enough for given index (no header, unspecified alignment) ...
Definition: vec.h:439
void dpo_stack_from_node(u32 child_node_index, dpo_id_t *dpo, const dpo_id_t *parent)
Stack one DPO object on another, and thus establish a child parent relationship.
Definition: dpo.c:530
fib_protocol_t fp_proto
protocol type
Definition: fib_types.h:212
#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:922
virtual circuit service
const char * tcp_fsm_states[]
Definition: tcp.c:716
static clib_error_t * tcp_config_fn(vlib_main_t *vm, unformat_input_t *input)
Definition: tcp.c:1580
#define tcp_in_recovery(tc)
Definition: tcp.h:363
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:646
static u32 tcp_session_push_header(transport_connection_t *tconn, vlib_buffer_t *b)
Definition: tcp.c:1155
struct _sack_block sack_block_t
u8 * format_tcp_rcv_sacks(u8 *s, va_list *args)
Definition: tcp.c:962
void ip6_register_protocol(u32 protocol, u32 node_index)
Definition: ip6_forward.c:1461
struct _scoreboard_trace_elt scoreboard_trace_elt_t
u8 * format_tcp_sacks(u8 *s, va_list *args)
Definition: tcp.c:938
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:906
#define TCP_TO_TIMER_TICK
Definition: tcp.h:98
const char * tcp_connection_flags_str[]
Definition: tcp.c:734
vlib_node_registration_t tcp4_output_node
(constructor) VLIB_REGISTER_NODE (tcp4_output_node)
Definition: tcp_output.c:2226
void scoreboard_init(sack_scoreboard_t *sb)
Definition: tcp_input.c:845
uword unformat_transport_connection(unformat_input_t *input, va_list *args)
Definition: session_cli.c:184
static u32 tcp_session_send_space(transport_connection_t *trans_conn)
Definition: tcp.c:1124
static transport_connection_t * tcp_half_open_session_get_transport(u32 conn_index)
Definition: tcp.c:1037
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:1236
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:1132
#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
clib_memset(h->entries, 0, sizeof(h->entries[0]) *entries)
struct _sack_scoreboard sack_scoreboard_t
void session_transport_delete_notify(transport_connection_t *tc)
Notification from transport that connection is being deleted.
Definition: session.c:820
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:634
static f64 vlib_time_now(vlib_main_t *vm)
Definition: main.h:232
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:766
static tcp_connection_t * tcp_half_open_connection_get(u32 conn_index)
Definition: tcp.h:606
sack_scoreboard_hole_t * scoreboard_next_hole(sack_scoreboard_t *sb, sack_scoreboard_hole_t *hole)
Definition: tcp_input.c:623
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
struct _transport_proto_vft transport_proto_vft_t
void tcp_connection_timers_reset(tcp_connection_t *tc)
Stop all connection timers.
Definition: tcp.c:426
vlib_node_registration_t tcp6_output_node
(constructor) VLIB_REGISTER_NODE (tcp6_output_node)
Definition: tcp_output.c:2249
static u8 transport_connection_is_tx_paced(transport_connection_t *tc)
Check if transport connection is paced.
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:525
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
void session_transport_reset_notify(transport_connection_t *tc)
Notify application that connection has been reset.
Definition: session.c:904
#define vec_add2(V, P, N)
Add N elements to end of vector V, return pointer to new elements in P.
Definition: vec.h:564
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:424
static u32 format_get_indent(u8 *s)
Definition: format.h:72
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: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:753
clib_time_t clib_time
Definition: main.h:65
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:1051
Definition: ip.h:109
static tcp_connection_t * tcp_get_connection_from_transport(transport_connection_t *tconn)
Definition: tcp.h:571
static void tcp_initialize_iss_seed(tcp_main_t *tm)
Definition: tcp.c:1390
void ip4_register_protocol(u32 protocol, u32 node_index)
Definition: ip4_forward.c:1700
static void tcp_initialize_timer_wheels(tcp_main_t *tm)
Definition: tcp.c:1376
#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:310
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:550
#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:577
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:581
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:163
#define TCP_CLOSEWAIT_TIME
Definition: tcp.h:104
u32 sw_if_index
Definition: vxlan_gbp.api:37
#define always_inline
Definition: clib.h:98
static u8 * format_tcp_vars(u8 *s, va_list *args)
Definition: tcp.c:826
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:409
static u32 tcp_available_output_snd_space(const tcp_connection_t *tc)
Definition: tcp.h:738
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:402
#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:1878
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:431
static u32 tcp_time_now(void)
Definition: tcp.h:791
u32 tcp_snd_space(tcp_connection_t *tc)
Definition: tcp.c:1118
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 i32 tcp_rcv_wnd_available(tcp_connection_t *tc)
Definition: tcp.c:801
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:555
static void tcp_cc_init(tcp_connection_t *tc)
Definition: tcp.c:533
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:833
void tcp_do_fastretransmits(tcp_worker_ctx_t *wrk)
Definition: tcp_input.c:1255
uword unformat_tcp_cc_algo_cfg(unformat_input_t *input, va_list *va)
Definition: tcp.c:1557
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:1343
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:1400
Definition: fib_entry.h:279
vlib_main_t * vm
convenience pointer to this thread&#39;s vlib main
Definition: tcp.h:432
#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:511
timer_expiration_handler tcp_timer_persist_handler
static transport_connection_t * tcp_session_get_transport(u32 conn_index, u32 thread_index)
Definition: tcp.c:1030
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:1216
#define tcp_in_fastrecovery(tc)
Definition: tcp.h:362
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:1213
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:888
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
tcp_connection_t * tcp_connection_alloc(u8 thread_index)
Definition: tcp.c:251
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
void tcp_connection_tx_pacer_update(tcp_connection_t *tc)
Definition: tcp.c:1197
static u32 ip6_compute_flow_hash(const ip6_header_t *ip, flow_hash_config_t flow_hash_config)
Definition: ip6.h:452
#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:1357
#define TCP_TIMER_HANDLE_INVALID
Definition: tcp.h:95
static void tcp_session_cleanup(u32 conn_index, u32 thread_index)
Definition: tcp.c:397
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:671
#define PREDICT_FALSE(x)
Definition: clib.h:111
#define VLIB_CONFIG_FUNCTION(x, n,...)
Definition: init.h:172
reliable transport protos
vnet_sw_interface_flags_t flags
Definition: interface.h:706
u8 * format_tcp_scoreboard(u8 *s, va_list *args)
Definition: tcp.c:998
clib_error_t * vnet_tcp_enable_disable(vlib_main_t *vm, u8 is_en)
Definition: tcp.c:1483
f64 seconds_per_clock
Definition: time.h:57
static void tcp_timer_waitclose_handler(u32 conn_index)
Definition: tcp.c:1271
#define TCP_FINWAIT1_TIME
Definition: tcp.h:106
#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:567
#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:844
void tcp_connection_free(tcp_connection_t *tc)
Definition: tcp.c:264
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:1625
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:301
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:1790
void tcp_connection_close(tcp_connection_t *tc)
Begin connection closing procedure.
Definition: tcp.c:333
#define vec_free(V)
Free vector&#39;s memory (no header).
Definition: vec.h:341
#define ENDPOINT_INVALID_INDEX
Definition: transport.h:153
static void tcp_session_flush_data(transport_connection_t *tconn)
Definition: tcp.c:1162
#define clib_warning(format, args...)
Definition: error.h:59
static u8 * format_tcp_congestion(u8 *s, va_list *args)
Definition: tcp.c:807
format_function_t format_tcp_header
Definition: format.h:101
const char * tcp_dbg_evt_str[]
Definition: tcp.c:710
static u8 * format_tcp_connection_flags(u8 *s, va_list *args)
Definition: tcp.c:741
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:1542
vlib_node_registration_t tcp6_input_node
(constructor) VLIB_REGISTER_NODE (tcp6_input_node)
Definition: tcp_input.c:3569
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:1999
#define VLIB_BUFFER_DATA_SIZE
Definition: buffer.h:51
#define pool_init_fixed(pool, max_elts)
initialize a fixed-size, preallocated pool
Definition: pool.h:86
void transport_endpoint_cleanup(u8 proto, ip46_address_t *lcl_ip, u16 port)
Definition: transport.c:300
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:238
#define VLIB_CLI_COMMAND(x,...)
Definition: cli.h:155
static u32 transport_max_tx_dequeue(transport_connection_t *tc)
Definition: session.h:523
static void tcp_timer_update(tcp_connection_t *tc, u8 timer_id, u32 interval)
Definition: tcp.h:857
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:664
u32 session_tx_fifo_max_dequeue(transport_connection_t *tc)
Definition: session.c:498
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:2022
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.
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:1072
static load_balance_t * load_balance_get(index_t lbi)
Definition: load_balance.h:219
#define seq_geq(_s1, _s2)
Definition: tcp.h:648
static u8 * format_tcp_congestion_status(u8 *s, va_list *args)
Definition: tcp.c:788
static u32 tcp_generate_random_iss(tcp_connection_t *tc)
Generate random iss as per rfc6528.
Definition: tcp.c:560
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:540
u8 * tcp_scoreboard_replay(u8 *s, tcp_connection_t *tc, u8 verbose)
Definition: tcp.c:1941
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:1717
struct _tcp_cc_algorithm tcp_cc_algorithm_t
Definition: tcp.h:260
#define clib_max(x, y)
Definition: clib.h:288
static clib_error_t * tcp_init(vlib_main_t *vm)
Definition: tcp.c:1511
static void tcp_timer_keep_handler(u32 conn_index)
Definition: tcp.c:1224
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:1172
void tcp_punt_unknown(vlib_main_t *vm, u8 is_ip4, u8 is_add)
Definition: tcp.c:1501
void transport_connection_tx_pacer_reset(transport_connection_t *tc, u32 rate_bytes_per_sec, u32 start_bucket, u64 time_now)
Definition: transport.c:538
void tcp_connection_reset(tcp_connection_t *tc)
Notify session that connection has been reset.
Definition: tcp.c:277
vlib_node_registration_t tcp4_input_node
(constructor) VLIB_REGISTER_NODE (tcp4_input_node)
Definition: tcp_input.c:3546
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:1084
#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:863
#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:1806
uword unformat_vlib_cli_sub_input(unformat_input_t *i, va_list *args)
Definition: cli.c:158
void() timer_expiration_handler(u32 index)
Definition: tcp.h:88
static void tcp_timer_establish_ao_handler(u32 conn_index)
Definition: tcp.c:1253
sack_scoreboard_hole_t * scoreboard_first_hole(sack_scoreboard_t *sb)
Definition: tcp_input.c:639
static u32 tcp_round_snd_space(tcp_connection_t *tc, u32 snd_space)
Definition: tcp.c:1064
static tcp_worker_ctx_t * tcp_get_worker(u32 thread_index)
Definition: tcp.h:525
void session_transport_closed_notify(transport_connection_t *tc)
Notification from transport that session can be closed.
Definition: session.c:875
u64 uword
Definition: types.h:112
void tcp_connection_init_vars(tcp_connection_t *tc)
Initialize tcp connection variables.
Definition: tcp.c:614
#define DPO_INVALID
An initialiser for DPOs declared on the stack.
Definition: dpo.h:195
const char * tcp_conn_timers[]
Definition: tcp.c:759
static u8 * format_tcp_session(u8 *s, va_list *args)
Definition: tcp.c:906
typedef prefix
Definition: ip_types.api:35
left
unformat_function_t unformat_memory_size
Definition: format.h:295
u8 * format_tcp_connection(u8 *s, va_list *args)
Definition: tcp.c:887
static u8 * format_tcp_sack_hole(u8 *s, va_list *args)
Definition: tcp.c:986
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:552
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:168
int session_stream_connect_notify(transport_connection_t *tc, u8 is_fail)
Definition: session.c:630
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:243
void tcp_init_mss(tcp_connection_t *tc)
Definition: tcp_output.c:444
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:930
#define tcp_opts_sack_permitted(_to)
Definition: tcp_packet.h:160
void tcp_send_acks(tcp_worker_ctx_t *wrk)
Definition: tcp_output.c:1280
static u32 tcp_snd_space_inline(tcp_connection_t *tc)
Compute tx window session is allowed to fill.
Definition: tcp.c:1091
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:334
u8 * format_tcp_state(u8 *s, va_list *args)
Definition: tcp.c:723
static clib_error_t * ip6_lookup_init(vlib_main_t *vm)
Definition: ip6_forward.c:2627
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:2065
static void tcp_update_time(f64 now, u8 thread_index)
Definition: tcp.c:1143
void tcp_enable_pacing(tcp_connection_t *tc)
Definition: tcp.c:599
static tcp_main_t * vnet_get_tcp_main()
Definition: tcp.h:519
static void tcp_cong_recovery_off(tcp_connection_t *tc)
Definition: tcp.h:376
timer_expiration_handler tcp_timer_delack_handler
void vlib_cli_output(vlib_main_t *vm, char *fmt,...)
Definition: cli.c:762
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:809
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:1903
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:600
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