blob: b3564004dd497c1df60aa1bf923360ceac76f1bf [file] [log] [blame]
Dave Barach68b0fb02017-02-28 15:15:56 -05001/*
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
Dave Barach3bbcfab2017-08-15 19:03:44 -040016/**
17 * @file
18 * @brief TCP host stack utilities
19 */
20
Dave Barach68b0fb02017-02-28 15:15:56 -050021#include <vnet/tcp/tcp.h>
22#include <vnet/session/session.h>
23#include <vnet/fib/fib.h>
Florin Corasf6359c82017-06-19 12:26:09 -040024#include <vnet/dpo/load_balance.h>
Dave Barach3bbcfab2017-08-15 19:03:44 -040025#include <vnet/dpo/receive_dpo.h>
26#include <vnet/ip/ip6_neighbor.h>
Dave Barach68b0fb02017-02-28 15:15:56 -050027#include <math.h>
28
29tcp_main_t tcp_main;
30
Florin Coras1c8ff632018-05-17 13:28:34 -070031typedef struct
32{
33 fib_protocol_t nh_proto;
34 vnet_link_t link_type;
35 ip46_address_t ip;
36 u32 sw_if_index;
37 u8 is_add;
38} tcp_add_del_adj_args_t;
39
40static void
41tcp_add_del_adj_cb (tcp_add_del_adj_args_t * args)
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 {
51 ai = adj_nbr_find (FIB_PROTOCOL_IP6, VNET_LINK_IP6, &args->ip,
52 args->sw_if_index);
53 if (ai != ADJ_INDEX_INVALID)
54 adj_unlock (ai);
55 }
56}
57
58static void
59tcp_add_del_adjacency (tcp_connection_t * tc, u8 is_add)
60{
61 tcp_add_del_adj_args_t args = {
62 .nh_proto = FIB_PROTOCOL_IP6,
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 };
68 vlib_rpc_call_main_thread (tcp_add_del_adj_cb, (u8 *) & args,
69 sizeof (args));
70}
71
Dave Barach68b0fb02017-02-28 15:15:56 -050072static u32
Florin Coras04e53442017-07-16 17:12:15 -070073tcp_connection_bind (u32 session_index, transport_endpoint_t * lcl)
Dave Barach68b0fb02017-02-28 15:15:56 -050074{
75 tcp_main_t *tm = &tcp_main;
76 tcp_connection_t *listener;
Florin Corascea194d2017-10-02 00:18:51 -070077 void *iface_ip;
Dave Barach68b0fb02017-02-28 15:15:56 -050078
79 pool_get (tm->listener_pool, listener);
80 memset (listener, 0, sizeof (*listener));
81
82 listener->c_c_index = listener - tm->listener_pool;
Florin Coras0e495682017-09-19 22:27:18 -070083 listener->c_lcl_port = lcl->port;
Dave Barach68b0fb02017-02-28 15:15:56 -050084
Florin Corascea194d2017-10-02 00:18:51 -070085 /* 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)
Florin Coras6cf30ad2017-04-04 23:08:23 -070087 {
Florin Corascea194d2017-10-02 00:18:51 -070088 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);
Florin Coras6cf30ad2017-04-04 23:08:23 -070091 }
Florin Corascea194d2017-10-02 00:18:51 -070092 ip_copy (&listener->c_lcl_ip, &lcl->ip, lcl->is_ip4);
93 listener->c_is_ip4 = lcl->is_ip4;
Florin Coras3cbc04b2017-10-02 00:18:51 -070094 listener->c_proto = TRANSPORT_PROTO_TCP;
Dave Barach68b0fb02017-02-28 15:15:56 -050095 listener->c_s_index = session_index;
Florin Corascea194d2017-10-02 00:18:51 -070096 listener->c_fib_index = lcl->fib_index;
Dave Barach68b0fb02017-02-28 15:15:56 -050097 listener->state = TCP_STATE_LISTEN;
Dave Barach68b0fb02017-02-28 15:15:56 -050098
Florin Corase69f4952017-03-07 10:06:24 -080099 tcp_connection_timers_init (listener);
100
101 TCP_EVT_DBG (TCP_EVT_BIND, listener);
102
Dave Barach68b0fb02017-02-28 15:15:56 -0500103 return listener->c_c_index;
104}
105
106u32
Florin Coras04e53442017-07-16 17:12:15 -0700107tcp_session_bind (u32 session_index, transport_endpoint_t * tep)
Dave Barach68b0fb02017-02-28 15:15:56 -0500108{
Florin Coras04e53442017-07-16 17:12:15 -0700109 return tcp_connection_bind (session_index, tep);
Dave Barach68b0fb02017-02-28 15:15:56 -0500110}
111
112static void
Florin Corase69f4952017-03-07 10:06:24 -0800113tcp_connection_unbind (u32 listener_index)
Dave Barach68b0fb02017-02-28 15:15:56 -0500114{
115 tcp_main_t *tm = vnet_get_tcp_main ();
Dave Barach2c25a622017-06-26 11:35:07 -0400116 tcp_connection_t *tc;
117
118 tc = pool_elt_at_index (tm->listener_pool, listener_index);
119
120 TCP_EVT_DBG (TCP_EVT_UNBIND, tc);
121
122 /* Poison the entry */
123 if (CLIB_DEBUG > 0)
124 memset (tc, 0xFA, sizeof (*tc));
125
Dave Barach68b0fb02017-02-28 15:15:56 -0500126 pool_put_index (tm->listener_pool, listener_index);
127}
128
129u32
Florin Corase69f4952017-03-07 10:06:24 -0800130tcp_session_unbind (u32 listener_index)
Dave Barach68b0fb02017-02-28 15:15:56 -0500131{
Florin Corase69f4952017-03-07 10:06:24 -0800132 tcp_connection_unbind (listener_index);
Dave Barach68b0fb02017-02-28 15:15:56 -0500133 return 0;
134}
135
136transport_connection_t *
137tcp_session_get_listener (u32 listener_index)
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
Florin Coras68810622017-07-24 17:40:28 -0700145/**
146 * Cleanup half-open connection
147 *
148 */
149void
150tcp_half_open_connection_del (tcp_connection_t * tc)
151{
152 tcp_main_t *tm = vnet_get_tcp_main ();
153 clib_spinlock_lock_if_init (&tm->half_open_lock);
154 pool_put_index (tm->half_open_connections, tc->c_c_index);
155 if (CLIB_DEBUG)
156 memset (tc, 0xFA, sizeof (*tc));
157 clib_spinlock_unlock_if_init (&tm->half_open_lock);
158}
159
160/**
161 * Try to cleanup half-open connection
162 *
163 * If called from a thread that doesn't own tc, the call won't have any
164 * effect.
165 *
166 * @param tc - connection to be cleaned up
167 * @return non-zero if cleanup failed.
168 */
169int
170tcp_half_open_connection_cleanup (tcp_connection_t * tc)
171{
172 /* Make sure this is the owning thread */
173 if (tc->c_thread_index != vlib_get_thread_index ())
174 return 1;
175 tcp_timer_reset (tc, TCP_TIMER_ESTABLISH);
176 tcp_timer_reset (tc, TCP_TIMER_RETRANSMIT_SYN);
177 tcp_half_open_connection_del (tc);
178 return 0;
179}
180
181tcp_connection_t *
182tcp_half_open_connection_new (void)
183{
184 tcp_main_t *tm = vnet_get_tcp_main ();
185 tcp_connection_t *tc = 0;
Florin Coras4eeeaaf2017-09-05 14:03:37 -0400186 ASSERT (vlib_get_thread_index () == 0);
Florin Coras68810622017-07-24 17:40:28 -0700187 pool_get (tm->half_open_connections, tc);
188 memset (tc, 0, sizeof (*tc));
189 tc->c_c_index = tc - tm->half_open_connections;
190 return tc;
191}
192
Dave Barach68b0fb02017-02-28 15:15:56 -0500193/**
194 * Cleans up connection state.
195 *
196 * No notifications.
197 */
198void
199tcp_connection_cleanup (tcp_connection_t * tc)
200{
201 tcp_main_t *tm = &tcp_main;
Dave Barach68b0fb02017-02-28 15:15:56 -0500202
203 /* Cleanup local endpoint if this was an active connect */
Florin Coras3cbc04b2017-10-02 00:18:51 -0700204 transport_endpoint_cleanup (TRANSPORT_PROTO_TCP, &tc->c_lcl_ip,
205 tc->c_lcl_port);
Dave Barach68b0fb02017-02-28 15:15:56 -0500206
Florin Coras68810622017-07-24 17:40:28 -0700207 /* Check if connection is not yet fully established */
Dave Barach68b0fb02017-02-28 15:15:56 -0500208 if (tc->state == TCP_STATE_SYN_SENT)
Dave Barach2c25a622017-06-26 11:35:07 -0400209 {
Florin Coras68810622017-07-24 17:40:28 -0700210 /* 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. */
213 tcp_half_open_connection_cleanup (tc);
Dave Barach2c25a622017-06-26 11:35:07 -0400214 }
Dave Barach68b0fb02017-02-28 15:15:56 -0500215 else
Dave Barach2c25a622017-06-26 11:35:07 -0400216 {
217 int thread_index = tc->c_thread_index;
Florin Coras68810622017-07-24 17:40:28 -0700218
219 /* Make sure all timers are cleared */
220 tcp_connection_timers_reset (tc);
221
Florin Coras1c8ff632018-05-17 13:28:34 -0700222 if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6))
223 tcp_add_del_adjacency (tc, 0);
224
Dave Barach2c25a622017-06-26 11:35:07 -0400225 /* Poison the entry */
226 if (CLIB_DEBUG > 0)
227 memset (tc, 0xFA, sizeof (*tc));
228 pool_put (tm->connections[thread_index], tc);
229 }
Dave Barach68b0fb02017-02-28 15:15:56 -0500230}
231
232/**
233 * Connection removal.
234 *
235 * This should be called only once connection enters CLOSED state. Note
236 * that it notifies the session of the removal event, so if the goal is to
237 * just remove the connection, call tcp_connection_cleanup instead.
238 */
239void
240tcp_connection_del (tcp_connection_t * tc)
241{
Florin Corase69f4952017-03-07 10:06:24 -0800242 TCP_EVT_DBG (TCP_EVT_DELETE, tc);
Dave Barach68b0fb02017-02-28 15:15:56 -0500243 stream_session_delete_notify (&tc->connection);
244 tcp_connection_cleanup (tc);
245}
246
Florin Coras6534b7a2017-07-18 05:38:03 -0400247tcp_connection_t *
248tcp_connection_new (u8 thread_index)
249{
250 tcp_main_t *tm = vnet_get_tcp_main ();
251 tcp_connection_t *tc;
252
253 pool_get (tm->connections[thread_index], tc);
254 memset (tc, 0, sizeof (*tc));
255 tc->c_c_index = tc - tm->connections[thread_index];
256 tc->c_thread_index = thread_index;
257 return tc;
258}
259
Florin Corasd79b41e2017-03-04 05:37:52 -0800260/** Notify session that connection has been reset.
261 *
262 * Switch state to closed and wait for session to call cleanup.
263 */
264void
265tcp_connection_reset (tcp_connection_t * tc)
266{
Florin Coras6534b7a2017-07-18 05:38:03 -0400267 TCP_EVT_DBG (TCP_EVT_RST_RCVD, tc);
Florin Coras11c05492017-05-10 12:29:14 -0700268 switch (tc->state)
269 {
270 case TCP_STATE_SYN_RCVD:
271 /* Cleanup everything. App wasn't notified yet */
272 stream_session_delete_notify (&tc->connection);
273 tcp_connection_cleanup (tc);
274 break;
275 case TCP_STATE_SYN_SENT:
Florin Coras3cbc04b2017-10-02 00:18:51 -0700276 session_stream_connect_notify (&tc->connection, 1 /* fail */ );
Florin Coras6534b7a2017-07-18 05:38:03 -0400277 tcp_connection_cleanup (tc);
278 break;
Florin Coras11c05492017-05-10 12:29:14 -0700279 case TCP_STATE_ESTABLISHED:
Florin Corasf988e692017-11-27 04:34:14 -0500280 stream_session_reset_notify (&tc->connection);
281 /* fall through */
Florin Coras11c05492017-05-10 12:29:14 -0700282 case TCP_STATE_CLOSE_WAIT:
283 case TCP_STATE_FIN_WAIT_1:
284 case TCP_STATE_FIN_WAIT_2:
285 case TCP_STATE_CLOSING:
286 tc->state = TCP_STATE_CLOSED;
Florin Coras6534b7a2017-07-18 05:38:03 -0400287 TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc);
Florin Corasd79b41e2017-03-04 05:37:52 -0800288
Florin Coras11c05492017-05-10 12:29:14 -0700289 /* Make sure all timers are cleared */
290 tcp_connection_timers_reset (tc);
Dave Barach2c25a622017-06-26 11:35:07 -0400291
292 /* Wait for cleanup from session layer but not forever */
Florin Coras68810622017-07-24 17:40:28 -0700293 tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
Florin Coras11c05492017-05-10 12:29:14 -0700294 break;
295 case TCP_STATE_CLOSED:
296 return;
297 }
Florin Corasd79b41e2017-03-04 05:37:52 -0800298}
299
Dave Barach68b0fb02017-02-28 15:15:56 -0500300/**
301 * Begin connection closing procedure.
302 *
303 * If at the end the connection is not in CLOSED state, it is not removed.
304 * Instead, we rely on on TCP to advance through state machine to either
305 * 1) LAST_ACK (passive close) whereby when the last ACK is received
306 * tcp_connection_del is called. This notifies session of the delete and
307 * calls cleanup.
308 * 2) TIME_WAIT (active close) whereby after 2MSL the 2MSL timer triggers
309 * and cleanup is called.
Florin Corasd79b41e2017-03-04 05:37:52 -0800310 *
311 * N.B. Half-close connections are not supported
Dave Barach68b0fb02017-02-28 15:15:56 -0500312 */
313void
314tcp_connection_close (tcp_connection_t * tc)
315{
Florin Corase69f4952017-03-07 10:06:24 -0800316 TCP_EVT_DBG (TCP_EVT_CLOSE, tc);
317
Florin Corasb2215d62017-08-01 16:56:58 -0700318 /* Send/Program FIN if needed and switch state */
319 switch (tc->state)
320 {
321 case TCP_STATE_SYN_SENT:
322 tc->state = TCP_STATE_CLOSED;
323 break;
324 case TCP_STATE_SYN_RCVD:
325 tcp_send_fin (tc);
326 tc->state = TCP_STATE_FIN_WAIT_1;
327 break;
328 case TCP_STATE_ESTABLISHED:
329 if (!stream_session_tx_fifo_max_dequeue (&tc->connection))
330 tcp_send_fin (tc);
331 else
332 tc->flags |= TCP_CONN_FINPNDG;
333 tc->state = TCP_STATE_FIN_WAIT_1;
334 break;
335 case TCP_STATE_CLOSE_WAIT:
Florin Corasa096f2d2017-09-28 23:49:42 -0400336 tcp_connection_timers_reset (tc);
Florin Corasb2215d62017-08-01 16:56:58 -0700337 tcp_send_fin (tc);
338 tc->state = TCP_STATE_LAST_ACK;
Florin Corasa096f2d2017-09-28 23:49:42 -0400339 tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
Florin Corasb2215d62017-08-01 16:56:58 -0700340 break;
Florin Corasc87c91d2017-08-16 19:55:49 -0700341 case TCP_STATE_FIN_WAIT_1:
Florin Corasa096f2d2017-09-28 23:49:42 -0400342 tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
Florin Corasc87c91d2017-08-16 19:55:49 -0700343 break;
Florin Corasb2215d62017-08-01 16:56:58 -0700344 default:
Florin Corasa096f2d2017-09-28 23:49:42 -0400345 TCP_DBG ("state: %u", tc->state);
Florin Corasb2215d62017-08-01 16:56:58 -0700346 }
Dave Barach68b0fb02017-02-28 15:15:56 -0500347
Florin Coras6534b7a2017-07-18 05:38:03 -0400348 TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc);
Dave Barach68b0fb02017-02-28 15:15:56 -0500349
Florin Corasd79b41e2017-03-04 05:37:52 -0800350 /* If in CLOSED and WAITCLOSE timer is not set, delete connection now */
Florin Corasa096f2d2017-09-28 23:49:42 -0400351 if (!tcp_timer_is_active (tc, TCP_TIMER_WAITCLOSE)
Florin Corasd79b41e2017-03-04 05:37:52 -0800352 && tc->state == TCP_STATE_CLOSED)
Dave Barach68b0fb02017-02-28 15:15:56 -0500353 tcp_connection_del (tc);
354}
355
356void
357tcp_session_close (u32 conn_index, u32 thread_index)
358{
359 tcp_connection_t *tc;
360 tc = tcp_connection_get (conn_index, thread_index);
361 tcp_connection_close (tc);
362}
363
364void
365tcp_session_cleanup (u32 conn_index, u32 thread_index)
366{
367 tcp_connection_t *tc;
368 tc = tcp_connection_get (conn_index, thread_index);
Florin Corasa096f2d2017-09-28 23:49:42 -0400369 tcp_connection_timers_reset (tc);
Florin Corasd79b41e2017-03-04 05:37:52 -0800370
371 /* Wait for the session tx events to clear */
372 tc->state = TCP_STATE_CLOSED;
Florin Coras6534b7a2017-07-18 05:38:03 -0400373 TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc);
Florin Corasd79b41e2017-03-04 05:37:52 -0800374 tcp_timer_update (tc, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME);
Dave Barach68b0fb02017-02-28 15:15:56 -0500375}
376
Dave Barach68b0fb02017-02-28 15:15:56 -0500377/**
378 * Initialize all connection timers as invalid
379 */
380void
381tcp_connection_timers_init (tcp_connection_t * tc)
382{
383 int i;
384
385 /* Set all to invalid */
386 for (i = 0; i < TCP_N_TIMERS; i++)
387 {
388 tc->timers[i] = TCP_TIMER_HANDLE_INVALID;
389 }
390
391 tc->rto = TCP_RTO_INIT;
392}
393
394/**
395 * Stop all connection timers
396 */
397void
398tcp_connection_timers_reset (tcp_connection_t * tc)
399{
400 int i;
401 for (i = 0; i < TCP_N_TIMERS; i++)
402 {
403 tcp_timer_reset (tc, i);
404 }
405}
406
Dave Barach2c25a622017-06-26 11:35:07 -0400407#if 0
Florin Corasf6359c82017-06-19 12:26:09 -0400408typedef struct ip4_tcp_hdr
409{
410 ip4_header_t ip;
411 tcp_header_t tcp;
412} ip4_tcp_hdr_t;
413
414typedef struct ip6_tcp_hdr
415{
416 ip6_header_t ip;
417 tcp_header_t tcp;
418} ip6_tcp_hdr_t;
419
420static void
421tcp_connection_select_lb_bucket (tcp_connection_t * tc, const dpo_id_t * dpo,
422 dpo_id_t * result)
423{
424 const dpo_id_t *choice;
425 load_balance_t *lb;
426 int hash;
427
428 lb = load_balance_get (dpo->dpoi_index);
429 if (tc->c_is_ip4)
430 {
431 ip4_tcp_hdr_t hdr;
432 memset (&hdr, 0, sizeof (hdr));
433 hdr.ip.protocol = IP_PROTOCOL_TCP;
434 hdr.ip.address_pair.src.as_u32 = tc->c_lcl_ip.ip4.as_u32;
435 hdr.ip.address_pair.dst.as_u32 = tc->c_rmt_ip.ip4.as_u32;
436 hdr.tcp.src_port = tc->c_lcl_port;
437 hdr.tcp.dst_port = tc->c_rmt_port;
438 hash = ip4_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
439 }
440 else
441 {
442 ip6_tcp_hdr_t hdr;
443 memset (&hdr, 0, sizeof (hdr));
444 hdr.ip.protocol = IP_PROTOCOL_TCP;
445 clib_memcpy (&hdr.ip.src_address, &tc->c_lcl_ip.ip6,
446 sizeof (ip6_address_t));
447 clib_memcpy (&hdr.ip.dst_address, &tc->c_rmt_ip.ip6,
448 sizeof (ip6_address_t));
449 hdr.tcp.src_port = tc->c_lcl_port;
450 hdr.tcp.dst_port = tc->c_rmt_port;
451 hash = ip6_compute_flow_hash (&hdr.ip, lb->lb_hash_config);
452 }
453 choice = load_balance_get_bucket_i (lb, hash & lb->lb_n_buckets_minus_1);
454 dpo_copy (result, choice);
455}
456
457fib_node_index_t
458tcp_lookup_rmt_in_fib (tcp_connection_t * tc)
459{
460 fib_prefix_t prefix;
Florin Coras04e53442017-07-16 17:12:15 -0700461 u32 fib_index;
Florin Corasf6359c82017-06-19 12:26:09 -0400462
463 clib_memcpy (&prefix.fp_addr, &tc->c_rmt_ip, sizeof (prefix.fp_addr));
464 prefix.fp_proto = tc->c_is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6;
465 prefix.fp_len = tc->c_is_ip4 ? 32 : 128;
Florin Corascea194d2017-10-02 00:18:51 -0700466 fib_index = fib_table_find (prefix.fp_proto, tc->c_fib_index);
Florin Coras04e53442017-07-16 17:12:15 -0700467 return fib_table_lookup (fib_index, &prefix);
Florin Corasf6359c82017-06-19 12:26:09 -0400468}
469
470static int
471tcp_connection_stack_on_fib_entry (tcp_connection_t * tc)
472{
473 dpo_id_t choice = DPO_INVALID;
474 u32 output_node_index;
475 fib_entry_t *fe;
476
477 fe = fib_entry_get (tc->c_rmt_fei);
478 if (fe->fe_lb.dpoi_type != DPO_LOAD_BALANCE)
479 return -1;
480
481 tcp_connection_select_lb_bucket (tc, &fe->fe_lb, &choice);
482
483 output_node_index =
484 tc->c_is_ip4 ? tcp4_output_node.index : tcp6_output_node.index;
485 dpo_stack_from_node (output_node_index, &tc->c_rmt_dpo, &choice);
486 return 0;
487}
488
489/** Stack tcp connection on peer's fib entry.
490 *
491 * This ultimately populates the dpo the connection will use to send packets.
492 */
493static void
494tcp_connection_fib_attach (tcp_connection_t * tc)
495{
496 tc->c_rmt_fei = tcp_lookup_rmt_in_fib (tc);
497
498 ASSERT (tc->c_rmt_fei != FIB_NODE_INDEX_INVALID);
499
500 tcp_connection_stack_on_fib_entry (tc);
501}
Dave Barach2c25a622017-06-26 11:35:07 -0400502#endif /* 0 */
Florin Corasf6359c82017-06-19 12:26:09 -0400503
Florin Coras4eeeaaf2017-09-05 14:03:37 -0400504/**
505 * Initialize connection send variables.
506 */
507void
508tcp_init_snd_vars (tcp_connection_t * tc)
509{
510 u32 time_now;
511
Florin Coras3cbc04b2017-10-02 00:18:51 -0700512 /*
513 * We use the time to randomize iss and for setting up the initial
514 * timestamp. Make sure it's updated otherwise syn and ack in the
515 * handshake may make it look as if time has flown in the opposite
516 * direction for us.
517 */
518 tcp_set_time_now (vlib_get_thread_index ());
Florin Coras4eeeaaf2017-09-05 14:03:37 -0400519 time_now = tcp_time_now ();
Florin Coras3cbc04b2017-10-02 00:18:51 -0700520
Florin Coras4eeeaaf2017-09-05 14:03:37 -0400521 tc->iss = random_u32 (&time_now);
522 tc->snd_una = tc->iss;
523 tc->snd_nxt = tc->iss + 1;
524 tc->snd_una_max = tc->snd_nxt;
525}
526
Dave Barach68b0fb02017-02-28 15:15:56 -0500527/** Initialize tcp connection variables
528 *
529 * Should be called after having received a msg from the peer, i.e., a SYN or
530 * a SYNACK, such that connection options have already been exchanged. */
531void
532tcp_connection_init_vars (tcp_connection_t * tc)
533{
534 tcp_connection_timers_init (tc);
Florin Corasc8343412017-05-04 14:25:50 -0700535 tcp_init_mss (tc);
Florin Coras6792ec02017-03-13 03:49:51 -0700536 scoreboard_init (&tc->sack_sb);
Dave Barach68b0fb02017-02-28 15:15:56 -0500537 tcp_cc_init (tc);
Florin Coras4eeeaaf2017-09-05 14:03:37 -0400538 if (tc->state == TCP_STATE_SYN_RCVD)
539 tcp_init_snd_vars (tc);
540
Florin Coras1c8ff632018-05-17 13:28:34 -0700541 if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6))
542 tcp_add_del_adjacency (tc, 1);
543
Dave Barach2c25a622017-06-26 11:35:07 -0400544 // tcp_connection_fib_attach (tc);
Dave Barach68b0fb02017-02-28 15:15:56 -0500545}
546
Florin Coras3cbc04b2017-10-02 00:18:51 -0700547static int
548tcp_alloc_custom_local_endpoint (tcp_main_t * tm, ip46_address_t * lcl_addr,
549 u16 * lcl_port, u8 is_ip4)
550{
551 int index, port;
552 if (is_ip4)
553 {
554 index = tm->last_v4_address_rotor++;
555 if (tm->last_v4_address_rotor >= vec_len (tm->ip4_src_addresses))
556 tm->last_v4_address_rotor = 0;
557 lcl_addr->ip4.as_u32 = tm->ip4_src_addresses[index].as_u32;
558 }
559 else
560 {
561 index = tm->last_v6_address_rotor++;
562 if (tm->last_v6_address_rotor >= vec_len (tm->ip6_src_addresses))
563 tm->last_v6_address_rotor = 0;
564 clib_memcpy (&lcl_addr->ip6, &tm->ip6_src_addresses[index],
565 sizeof (ip6_address_t));
566 }
567 port = transport_alloc_local_port (TRANSPORT_PROTO_TCP, lcl_addr);
568 if (port < 1)
569 {
570 clib_warning ("Failed to allocate src port");
571 return -1;
572 }
573 *lcl_port = port;
574 return 0;
575}
576
Dave Barach68b0fb02017-02-28 15:15:56 -0500577int
Florin Coras04e53442017-07-16 17:12:15 -0700578tcp_connection_open (transport_endpoint_t * rmt)
Dave Barach68b0fb02017-02-28 15:15:56 -0500579{
580 tcp_main_t *tm = vnet_get_tcp_main ();
581 tcp_connection_t *tc;
Dave Barach68b0fb02017-02-28 15:15:56 -0500582 ip46_address_t lcl_addr;
Florin Coras3cbc04b2017-10-02 00:18:51 -0700583 u16 lcl_port;
584 int rv;
Dave Barach68b0fb02017-02-28 15:15:56 -0500585
586 /*
Florin Coras3cbc04b2017-10-02 00:18:51 -0700587 * Allocate local endpoint
Dave Barach68b0fb02017-02-28 15:15:56 -0500588 */
Florin Coras3cbc04b2017-10-02 00:18:51 -0700589 if ((rmt->is_ip4 && vec_len (tm->ip4_src_addresses))
590 || (!rmt->is_ip4 && vec_len (tm->ip6_src_addresses)))
591 rv = tcp_alloc_custom_local_endpoint (tm, &lcl_addr, &lcl_port,
592 rmt->is_ip4);
Dave Barach68b0fb02017-02-28 15:15:56 -0500593 else
Florin Coras3cbc04b2017-10-02 00:18:51 -0700594 rv = transport_alloc_local_endpoint (TRANSPORT_PROTO_TCP,
595 rmt, &lcl_addr, &lcl_port);
Dave Barach2c25a622017-06-26 11:35:07 -0400596
Florin Coras3cbc04b2017-10-02 00:18:51 -0700597 if (rv)
598 return -1;
Dave Barach68b0fb02017-02-28 15:15:56 -0500599
600 /*
601 * Create connection and send SYN
602 */
Florin Coras68810622017-07-24 17:40:28 -0700603 clib_spinlock_lock_if_init (&tm->half_open_lock);
Florin Coras04e53442017-07-16 17:12:15 -0700604 tc = tcp_half_open_connection_new ();
Florin Coras3cbc04b2017-10-02 00:18:51 -0700605 ip_copy (&tc->c_rmt_ip, &rmt->ip, rmt->is_ip4);
606 ip_copy (&tc->c_lcl_ip, &lcl_addr, rmt->is_ip4);
Florin Coras0e495682017-09-19 22:27:18 -0700607 tc->c_rmt_port = rmt->port;
Dave Barach68b0fb02017-02-28 15:15:56 -0500608 tc->c_lcl_port = clib_host_to_net_u16 (lcl_port);
Florin Coras04e53442017-07-16 17:12:15 -0700609 tc->c_is_ip4 = rmt->is_ip4;
Florin Coras3cbc04b2017-10-02 00:18:51 -0700610 tc->c_proto = TRANSPORT_PROTO_TCP;
Florin Corascea194d2017-10-02 00:18:51 -0700611 tc->c_fib_index = rmt->fib_index;
Dave Barach68b0fb02017-02-28 15:15:56 -0500612 /* The other connection vars will be initialized after SYN ACK */
613 tcp_connection_timers_init (tc);
614
Florin Corase69f4952017-03-07 10:06:24 -0800615 TCP_EVT_DBG (TCP_EVT_OPEN, tc);
Florin Coras6534b7a2017-07-18 05:38:03 -0400616 tc->state = TCP_STATE_SYN_SENT;
Florin Coras4eeeaaf2017-09-05 14:03:37 -0400617 tcp_init_snd_vars (tc);
Florin Coras6534b7a2017-07-18 05:38:03 -0400618 tcp_send_syn (tc);
Florin Coras68810622017-07-24 17:40:28 -0700619 clib_spinlock_unlock_if_init (&tm->half_open_lock);
Florin Corase69f4952017-03-07 10:06:24 -0800620
Dave Barach68b0fb02017-02-28 15:15:56 -0500621 return tc->c_c_index;
622}
623
624int
Florin Coras04e53442017-07-16 17:12:15 -0700625tcp_session_open (transport_endpoint_t * tep)
Dave Barach68b0fb02017-02-28 15:15:56 -0500626{
Florin Coras04e53442017-07-16 17:12:15 -0700627 return tcp_connection_open (tep);
Dave Barach68b0fb02017-02-28 15:15:56 -0500628}
629
Florin Corase69f4952017-03-07 10:06:24 -0800630const char *tcp_dbg_evt_str[] = {
631#define _(sym, str) str,
632 foreach_tcp_dbg_evt
633#undef _
634};
635
636const char *tcp_fsm_states[] = {
637#define _(sym, str) str,
638 foreach_tcp_fsm_state
639#undef _
640};
641
Dave Barach68b0fb02017-02-28 15:15:56 -0500642u8 *
Florin Corase69f4952017-03-07 10:06:24 -0800643format_tcp_state (u8 * s, va_list * args)
644{
Florin Corasbb292f42017-05-19 09:49:19 -0700645 u32 state = va_arg (*args, u32);
Florin Corase69f4952017-03-07 10:06:24 -0800646
Florin Corasbb292f42017-05-19 09:49:19 -0700647 if (state < TCP_N_STATES)
648 s = format (s, "%s", tcp_fsm_states[state]);
Florin Corase69f4952017-03-07 10:06:24 -0800649 else
Florin Corasbb292f42017-05-19 09:49:19 -0700650 s = format (s, "UNKNOWN (%d (0x%x))", state, state);
Florin Corase69f4952017-03-07 10:06:24 -0800651 return s;
652}
653
Florin Corasa096f2d2017-09-28 23:49:42 -0400654const char *tcp_connection_flags_str[] = {
655#define _(sym, str) str,
656 foreach_tcp_connection_flag
657#undef _
658};
659
660u8 *
661format_tcp_connection_flags (u8 * s, va_list * args)
662{
663 tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
664 int i, last = -1;
665
666 for (i = 0; i < TCP_CONN_N_FLAG_BITS; i++)
667 if (tc->flags & (1 << i))
668 last = i;
669 for (i = 0; i < last; i++)
670 {
671 if (tc->flags & (1 << i))
672 s = format (s, "%s, ", tcp_connection_flags_str[i]);
673 }
674 if (last >= 0)
675 s = format (s, "%s", tcp_connection_flags_str[last]);
676 return s;
677}
678
Florin Corase69f4952017-03-07 10:06:24 -0800679const char *tcp_conn_timers[] = {
680#define _(sym, str) str,
681 foreach_tcp_timer
682#undef _
683};
684
685u8 *
686format_tcp_timers (u8 * s, va_list * args)
687{
688 tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
Florin Coras93992a92017-05-24 18:03:56 -0700689 int i, last = -1;
Florin Corase69f4952017-03-07 10:06:24 -0800690
691 for (i = 0; i < TCP_N_TIMERS; i++)
692 if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
693 last = i;
694
695 s = format (s, "[");
696 for (i = 0; i < last; i++)
697 {
698 if (tc->timers[i] != TCP_TIMER_HANDLE_INVALID)
699 s = format (s, "%s,", tcp_conn_timers[i]);
700 }
701
Florin Coras93992a92017-05-24 18:03:56 -0700702 if (last >= 0)
Florin Corase69f4952017-03-07 10:06:24 -0800703 s = format (s, "%s]", tcp_conn_timers[i]);
704 else
705 s = format (s, "]");
706
707 return s;
708}
709
710u8 *
Florin Corasbb292f42017-05-19 09:49:19 -0700711format_tcp_congestion_status (u8 * s, va_list * args)
712{
713 tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
714 if (tcp_in_recovery (tc))
715 s = format (s, "recovery");
716 else if (tcp_in_fastrecovery (tc))
717 s = format (s, "fastrecovery");
718 else
719 s = format (s, "none");
720 return s;
721}
722
723u8 *
724format_tcp_vars (u8 * s, va_list * args)
725{
726 tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
Florin Corasa096f2d2017-09-28 23:49:42 -0400727 s = format (s, " flags: %U timers: %U\n", format_tcp_connection_flags, tc,
728 format_tcp_timers, tc);
Dave Barach2c25a622017-06-26 11:35:07 -0400729 s = format (s, " snd_una %u snd_nxt %u snd_una_max %u",
Florin Corasbb292f42017-05-19 09:49:19 -0700730 tc->snd_una - tc->iss, tc->snd_nxt - tc->iss,
731 tc->snd_una_max - tc->iss);
732 s = format (s, " rcv_nxt %u rcv_las %u\n",
733 tc->rcv_nxt - tc->irs, tc->rcv_las - tc->irs);
734 s = format (s, " snd_wnd %u rcv_wnd %u snd_wl1 %u snd_wl2 %u\n",
735 tc->snd_wnd, tc->rcv_wnd, tc->snd_wl1 - tc->irs,
736 tc->snd_wl2 - tc->iss);
Florin Coras93992a92017-05-24 18:03:56 -0700737 s = format (s, " flight size %u send space %u rcv_wnd_av %d\n",
Florin Coras4eeeaaf2017-09-05 14:03:37 -0400738 tcp_flight_size (tc), tcp_available_output_snd_space (tc),
Florin Coras93992a92017-05-24 18:03:56 -0700739 tcp_rcv_wnd_available (tc));
Florin Corasbb292f42017-05-19 09:49:19 -0700740 s = format (s, " cong %U ", format_tcp_congestion_status, tc);
741 s = format (s, "cwnd %u ssthresh %u rtx_bytes %u bytes_acked %u\n",
Florin Coras93992a92017-05-24 18:03:56 -0700742 tc->cwnd, tc->ssthresh, tc->snd_rxt_bytes, tc->bytes_acked);
Dave Barach2c25a622017-06-26 11:35:07 -0400743 s = format (s, " prev_ssthresh %u snd_congestion %u dupack %u",
Florin Coras93992a92017-05-24 18:03:56 -0700744 tc->prev_ssthresh, tc->snd_congestion - tc->iss,
745 tc->rcv_dupacks);
Dave Barach2c25a622017-06-26 11:35:07 -0400746 s = format (s, " limited_transmit %u\n", tc->limited_transmit - tc->iss);
747 s = format (s, " tsecr %u tsecr_last_ack %u\n", tc->rcv_opts.tsecr,
748 tc->tsecr_last_ack);
Florin Corasbb292f42017-05-19 09:49:19 -0700749 s = format (s, " rto %u rto_boff %u srtt %u rttvar %u rtt_ts %u ", tc->rto,
750 tc->rto_boff, tc->srtt, tc->rttvar, tc->rtt_ts);
751 s = format (s, "rtt_seq %u\n", tc->rtt_seq);
Dave Barach2c25a622017-06-26 11:35:07 -0400752 s = format (s, " tsval_recent %u tsval_recent_age %u\n", tc->tsval_recent,
753 tcp_time_now () - tc->tsval_recent_age);
Florin Corasde706082017-10-11 01:43:15 -0700754 if (tc->state >= TCP_STATE_ESTABLISHED)
755 s = format (s, " scoreboard: %U\n", format_tcp_scoreboard, &tc->sack_sb,
756 tc);
Florin Corasbb292f42017-05-19 09:49:19 -0700757 if (vec_len (tc->snd_sacks))
758 s = format (s, " sacks tx: %U\n", format_tcp_sacks, tc);
759
760 return s;
761}
762
763u8 *
764format_tcp_connection_id (u8 * s, va_list * args)
Florin Corase69f4952017-03-07 10:06:24 -0800765{
766 tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
Florin Corasa5464812017-04-19 13:00:05 -0700767 if (!tc)
768 return s;
Florin Corase69f4952017-03-07 10:06:24 -0800769 if (tc->c_is_ip4)
770 {
771 s = format (s, "[#%d][%s] %U:%d->%U:%d", tc->c_thread_index, "T",
772 format_ip4_address, &tc->c_lcl_ip4,
773 clib_net_to_host_u16 (tc->c_lcl_port), format_ip4_address,
774 &tc->c_rmt_ip4, clib_net_to_host_u16 (tc->c_rmt_port));
775 }
776 else
777 {
778 s = format (s, "[#%d][%s] %U:%d->%U:%d", tc->c_thread_index, "T",
779 format_ip6_address, &tc->c_lcl_ip6,
780 clib_net_to_host_u16 (tc->c_lcl_port), format_ip6_address,
781 &tc->c_rmt_ip6, clib_net_to_host_u16 (tc->c_rmt_port));
782 }
783
784 return s;
785}
786
787u8 *
Florin Corasbb292f42017-05-19 09:49:19 -0700788format_tcp_connection (u8 * s, va_list * args)
Florin Corase69f4952017-03-07 10:06:24 -0800789{
790 tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
Florin Corasbb292f42017-05-19 09:49:19 -0700791 u32 verbose = va_arg (*args, u32);
792
Florin Corasc87c91d2017-08-16 19:55:49 -0700793 if (!tc)
794 return s;
Florin Corasbb292f42017-05-19 09:49:19 -0700795 s = format (s, "%-50U", format_tcp_connection_id, tc);
796 if (verbose)
797 {
798 s = format (s, "%-15U", format_tcp_state, tc->state);
799 if (verbose > 1)
Florin Corasa096f2d2017-09-28 23:49:42 -0400800 s = format (s, "\n%U", format_tcp_vars, tc);
Florin Corasbb292f42017-05-19 09:49:19 -0700801 }
Florin Coras3eb50622017-07-13 01:24:57 -0400802
Florin Corase69f4952017-03-07 10:06:24 -0800803 return s;
804}
805
806u8 *
807format_tcp_session (u8 * s, va_list * args)
Dave Barach68b0fb02017-02-28 15:15:56 -0500808{
809 u32 tci = va_arg (*args, u32);
810 u32 thread_index = va_arg (*args, u32);
Florin Corasbb292f42017-05-19 09:49:19 -0700811 u32 verbose = va_arg (*args, u32);
Dave Barach68b0fb02017-02-28 15:15:56 -0500812 tcp_connection_t *tc;
813
814 tc = tcp_connection_get (tci, thread_index);
Florin Coras6cf30ad2017-04-04 23:08:23 -0700815 if (tc)
Florin Coras93992a92017-05-24 18:03:56 -0700816 s = format (s, "%U", format_tcp_connection, tc, verbose);
Florin Coras6cf30ad2017-04-04 23:08:23 -0700817 else
Florin Coras1f152cd2017-08-18 19:28:03 -0700818 s = format (s, "empty\n");
Florin Coras93992a92017-05-24 18:03:56 -0700819 return s;
Dave Barach68b0fb02017-02-28 15:15:56 -0500820}
821
822u8 *
Florin Corase69f4952017-03-07 10:06:24 -0800823format_tcp_listener_session (u8 * s, va_list * args)
Dave Barach68b0fb02017-02-28 15:15:56 -0500824{
825 u32 tci = va_arg (*args, u32);
826 tcp_connection_t *tc = tcp_listener_get (tci);
Florin Corasbb292f42017-05-19 09:49:19 -0700827 return format (s, "%U", format_tcp_connection_id, tc);
Dave Barach68b0fb02017-02-28 15:15:56 -0500828}
829
830u8 *
Florin Corase69f4952017-03-07 10:06:24 -0800831format_tcp_half_open_session (u8 * s, va_list * args)
Dave Barach68b0fb02017-02-28 15:15:56 -0500832{
833 u32 tci = va_arg (*args, u32);
834 tcp_connection_t *tc = tcp_half_open_connection_get (tci);
Florin Corasbb292f42017-05-19 09:49:19 -0700835 return format (s, "%U", format_tcp_connection_id, tc);
Dave Barach68b0fb02017-02-28 15:15:56 -0500836}
837
Florin Coras06d11012017-05-17 14:21:51 -0700838u8 *
839format_tcp_sacks (u8 * s, va_list * args)
840{
841 tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
842 sack_block_t *sacks = tc->snd_sacks;
843 sack_block_t *block;
Dave Barach2c25a622017-06-26 11:35:07 -0400844 int i, len = 0;
845
846 len = vec_len (sacks);
847 for (i = 0; i < len - 1; i++)
848 {
849 block = &sacks[i];
850 s = format (s, " start %u end %u\n", block->start - tc->irs,
851 block->end - tc->irs);
852 }
853 if (len)
854 {
855 block = &sacks[len - 1];
856 s = format (s, " start %u end %u", block->start - tc->irs,
857 block->end - tc->irs);
858 }
Florin Coras06d11012017-05-17 14:21:51 -0700859 return s;
860}
861
862u8 *
Florin Coras3eb50622017-07-13 01:24:57 -0400863format_tcp_rcv_sacks (u8 * s, va_list * args)
864{
865 tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
866 sack_block_t *sacks = tc->rcv_opts.sacks;
867 sack_block_t *block;
868 int i, len = 0;
869
870 len = vec_len (sacks);
871 for (i = 0; i < len - 1; i++)
872 {
873 block = &sacks[i];
874 s = format (s, " start %u end %u\n", block->start - tc->iss,
875 block->end - tc->iss);
876 }
877 if (len)
878 {
879 block = &sacks[len - 1];
880 s = format (s, " start %u end %u", block->start - tc->iss,
881 block->end - tc->iss);
882 }
883 return s;
884}
885
886u8 *
Florin Coras06d11012017-05-17 14:21:51 -0700887format_tcp_sack_hole (u8 * s, va_list * args)
888{
889 sack_scoreboard_hole_t *hole = va_arg (*args, sack_scoreboard_hole_t *);
Florin Coras1f152cd2017-08-18 19:28:03 -0700890 tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
891 if (tc)
892 s = format (s, " [%u, %u]", hole->start - tc->iss, hole->end - tc->iss);
893 else
894 s = format (s, " [%u, %u]", hole->start, hole->end);
Florin Coras06d11012017-05-17 14:21:51 -0700895 return s;
896}
897
898u8 *
899format_tcp_scoreboard (u8 * s, va_list * args)
900{
901 sack_scoreboard_t *sb = va_arg (*args, sack_scoreboard_t *);
Florin Coras1f152cd2017-08-18 19:28:03 -0700902 tcp_connection_t *tc = va_arg (*args, tcp_connection_t *);
Florin Coras06d11012017-05-17 14:21:51 -0700903 sack_scoreboard_hole_t *hole;
Florin Coras93992a92017-05-24 18:03:56 -0700904 s = format (s, "sacked_bytes %u last_sacked_bytes %u lost_bytes %u\n",
905 sb->sacked_bytes, sb->last_sacked_bytes, sb->lost_bytes);
906 s = format (s, " last_bytes_delivered %u high_sacked %u snd_una_adv %u\n",
907 sb->last_bytes_delivered, sb->high_sacked, sb->snd_una_adv);
908 s = format (s, " cur_rxt_hole %u high_rxt %u rescue_rxt %u",
909 sb->cur_rxt_hole, sb->high_rxt, sb->rescue_rxt);
910
Florin Coras06d11012017-05-17 14:21:51 -0700911 hole = scoreboard_first_hole (sb);
Florin Coras93992a92017-05-24 18:03:56 -0700912 if (hole)
913 s = format (s, "\n head %u tail %u holes:\n", sb->head, sb->tail);
914
Florin Coras06d11012017-05-17 14:21:51 -0700915 while (hole)
916 {
Florin Coras1f152cd2017-08-18 19:28:03 -0700917 s = format (s, "%U", format_tcp_sack_hole, hole, tc);
Florin Coras06d11012017-05-17 14:21:51 -0700918 hole = scoreboard_next_hole (sb, hole);
919 }
Florin Coras3eb50622017-07-13 01:24:57 -0400920
Florin Coras06d11012017-05-17 14:21:51 -0700921 return s;
922}
923
Dave Barach68b0fb02017-02-28 15:15:56 -0500924transport_connection_t *
925tcp_session_get_transport (u32 conn_index, u32 thread_index)
926{
927 tcp_connection_t *tc = tcp_connection_get (conn_index, thread_index);
928 return &tc->connection;
929}
930
931transport_connection_t *
932tcp_half_open_session_get_transport (u32 conn_index)
933{
934 tcp_connection_t *tc = tcp_half_open_connection_get (conn_index);
935 return &tc->connection;
936}
937
Florin Corasc8343412017-05-04 14:25:50 -0700938/**
939 * Compute maximum segment size for session layer.
940 *
941 * Since the result needs to be the actual data length, it first computes
942 * the tcp options to be used in the next burst and subtracts their
943 * length from the connection's snd_mss.
944 */
Dave Barach68b0fb02017-02-28 15:15:56 -0500945u16
946tcp_session_send_mss (transport_connection_t * trans_conn)
947{
948 tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
Florin Corasc8343412017-05-04 14:25:50 -0700949
950 /* Ensure snd_mss does accurately reflect the amount of data we can push
951 * in a segment. This also makes sure that options are updated according to
952 * the current state of the connection. */
953 tcp_update_snd_mss (tc);
954
Dave Barach68b0fb02017-02-28 15:15:56 -0500955 return tc->snd_mss;
956}
957
Florin Coras3af90fc2017-05-03 21:09:42 -0700958always_inline u32
959tcp_round_snd_space (tcp_connection_t * tc, u32 snd_space)
960{
Dave Barach2c25a622017-06-26 11:35:07 -0400961 if (PREDICT_FALSE (tc->snd_wnd < tc->snd_mss))
Florin Coras3af90fc2017-05-03 21:09:42 -0700962 {
Florin Corasdb84e572017-05-09 18:54:52 -0700963 return tc->snd_wnd <= snd_space ? tc->snd_wnd : 0;
Florin Coras3af90fc2017-05-03 21:09:42 -0700964 }
965
Florin Coras1f152cd2017-08-18 19:28:03 -0700966 /* If not snd_wnd constrained and we can't write at least a segment,
967 * don't try at all */
Dave Barach2c25a622017-06-26 11:35:07 -0400968 if (PREDICT_FALSE (snd_space < tc->snd_mss))
Florin Coras9d063042017-09-14 03:08:00 -0400969 return snd_space < tc->cwnd ? 0 : snd_space;
Florin Coras3af90fc2017-05-03 21:09:42 -0700970
971 /* round down to mss multiple */
972 return snd_space - (snd_space % tc->snd_mss);
973}
974
Florin Coras6792ec02017-03-13 03:49:51 -0700975/**
976 * Compute tx window session is allowed to fill.
Florin Corasbb292f42017-05-19 09:49:19 -0700977 *
978 * Takes into account available send space, snd_mss and the congestion
979 * state of the connection. If possible, the value returned is a multiple
980 * of snd_mss.
981 *
982 * @param tc tcp connection
983 * @return number of bytes session is allowed to write
Florin Coras6792ec02017-03-13 03:49:51 -0700984 */
Dave Barach68b0fb02017-02-28 15:15:56 -0500985u32
Florin Corasbb292f42017-05-19 09:49:19 -0700986tcp_snd_space (tcp_connection_t * tc)
Dave Barach68b0fb02017-02-28 15:15:56 -0500987{
Florin Corasf03a59a2017-06-09 21:07:32 -0700988 int snd_space, snt_limited;
Florin Coras6792ec02017-03-13 03:49:51 -0700989
Florin Corasf03a59a2017-06-09 21:07:32 -0700990 if (PREDICT_TRUE (tcp_in_cong_recovery (tc) == 0))
Florin Coras6792ec02017-03-13 03:49:51 -0700991 {
Florin Coras1f152cd2017-08-18 19:28:03 -0700992 snd_space = tcp_available_output_snd_space (tc);
Florin Corasf03a59a2017-06-09 21:07:32 -0700993
994 /* If we haven't gotten dupacks or if we did and have gotten sacked
995 * bytes then we can still send as per Limited Transmit (RFC3042) */
996 if (PREDICT_FALSE (tc->rcv_dupacks != 0
997 && (tcp_opts_sack_permitted (tc)
998 && tc->sack_sb.last_sacked_bytes == 0)))
999 {
1000 if (tc->rcv_dupacks == 1 && tc->limited_transmit != tc->snd_nxt)
1001 tc->limited_transmit = tc->snd_nxt;
1002 ASSERT (seq_leq (tc->limited_transmit, tc->snd_nxt));
1003
1004 snt_limited = tc->snd_nxt - tc->limited_transmit;
1005 snd_space = clib_max (2 * tc->snd_mss - snt_limited, 0);
1006 }
Florin Coras3af90fc2017-05-03 21:09:42 -07001007 return tcp_round_snd_space (tc, snd_space);
1008 }
Florin Coras6792ec02017-03-13 03:49:51 -07001009
Florin Coras3af90fc2017-05-03 21:09:42 -07001010 if (tcp_in_recovery (tc))
1011 {
1012 tc->snd_nxt = tc->snd_una_max;
Florin Coras1f152cd2017-08-18 19:28:03 -07001013 snd_space = tcp_available_snd_wnd (tc) - tc->snd_rxt_bytes
Florin Coras3af90fc2017-05-03 21:09:42 -07001014 - (tc->snd_una_max - tc->snd_congestion);
Florin Corasc8343412017-05-04 14:25:50 -07001015 if (snd_space <= 0 || (tc->snd_una_max - tc->snd_una) >= tc->snd_wnd)
Florin Coras6792ec02017-03-13 03:49:51 -07001016 return 0;
Florin Coras3af90fc2017-05-03 21:09:42 -07001017 return tcp_round_snd_space (tc, snd_space);
Florin Coras6792ec02017-03-13 03:49:51 -07001018 }
1019
Florin Coras1f152cd2017-08-18 19:28:03 -07001020 /* RFC 5681: When previously unsent data is available and the new value of
1021 * cwnd and the receiver's advertised window allow, a TCP SHOULD send 1*SMSS
1022 * bytes of previously unsent data. */
1023 if (tcp_in_fastrecovery (tc) && !tcp_fastrecovery_sent_1_smss (tc))
Florin Coras6792ec02017-03-13 03:49:51 -07001024 {
Florin Coras1f152cd2017-08-18 19:28:03 -07001025 if (tcp_available_output_snd_space (tc) < tc->snd_mss)
1026 return 0;
Florin Coras6792ec02017-03-13 03:49:51 -07001027 tcp_fastrecovery_1_smss_on (tc);
1028 return tc->snd_mss;
1029 }
1030
1031 return 0;
Dave Barach68b0fb02017-02-28 15:15:56 -05001032}
1033
1034u32
Florin Corasbb292f42017-05-19 09:49:19 -07001035tcp_session_send_space (transport_connection_t * trans_conn)
1036{
1037 tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
Florin Coras1f152cd2017-08-18 19:28:03 -07001038 return clib_min (tcp_snd_space (tc),
1039 tc->snd_wnd - (tc->snd_nxt - tc->snd_una));
Florin Corasbb292f42017-05-19 09:49:19 -07001040}
1041
Florin Coras93992a92017-05-24 18:03:56 -07001042i32
1043tcp_rcv_wnd_available (tcp_connection_t * tc)
1044{
1045 return (i32) tc->rcv_wnd - (tc->rcv_nxt - tc->rcv_las);
1046}
1047
Florin Corasbb292f42017-05-19 09:49:19 -07001048u32
Florin Corasd79b41e2017-03-04 05:37:52 -08001049tcp_session_tx_fifo_offset (transport_connection_t * trans_conn)
Dave Barach68b0fb02017-02-28 15:15:56 -05001050{
1051 tcp_connection_t *tc = (tcp_connection_t *) trans_conn;
Florin Coras6792ec02017-03-13 03:49:51 -07001052
1053 ASSERT (seq_geq (tc->snd_nxt, tc->snd_una));
1054
1055 /* This still works if fast retransmit is on */
Florin Corasd79b41e2017-03-04 05:37:52 -08001056 return (tc->snd_nxt - tc->snd_una);
Dave Barach68b0fb02017-02-28 15:15:56 -05001057}
1058
Florin Coras561af9b2017-12-09 10:19:43 -08001059void
1060tcp_update_time (f64 now, u8 thread_index)
1061{
1062 tcp_set_time_now (thread_index);
1063 tw_timer_expire_timers_16t_2w_512sl (&tcp_main.timer_wheels[thread_index],
1064 now);
1065 tcp_flush_frames_to_output (thread_index);
1066}
1067
Dave Barach68b0fb02017-02-28 15:15:56 -05001068/* *INDENT-OFF* */
Florin Coras04e53442017-07-16 17:12:15 -07001069const static transport_proto_vft_t tcp_proto = {
Florin Coras561af9b2017-12-09 10:19:43 -08001070 .enable = vnet_tcp_enable_disable,
Florin Coras04e53442017-07-16 17:12:15 -07001071 .bind = tcp_session_bind,
Florin Corase69f4952017-03-07 10:06:24 -08001072 .unbind = tcp_session_unbind,
Dave Barach68b0fb02017-02-28 15:15:56 -05001073 .push_header = tcp_push_header,
1074 .get_connection = tcp_session_get_transport,
1075 .get_listener = tcp_session_get_listener,
1076 .get_half_open = tcp_half_open_session_get_transport,
Florin Coras04e53442017-07-16 17:12:15 -07001077 .open = tcp_session_open,
Dave Barach68b0fb02017-02-28 15:15:56 -05001078 .close = tcp_session_close,
1079 .cleanup = tcp_session_cleanup,
1080 .send_mss = tcp_session_send_mss,
1081 .send_space = tcp_session_send_space,
Florin Coras561af9b2017-12-09 10:19:43 -08001082 .update_time = tcp_update_time,
Florin Corasd79b41e2017-03-04 05:37:52 -08001083 .tx_fifo_offset = tcp_session_tx_fifo_offset,
Florin Corase69f4952017-03-07 10:06:24 -08001084 .format_connection = format_tcp_session,
1085 .format_listener = format_tcp_listener_session,
1086 .format_half_open = format_tcp_half_open_session,
Florin Coras371ca502018-02-21 12:07:41 -08001087 .tx_type = TRANSPORT_TX_PEEK,
1088 .service_type = TRANSPORT_SERVICE_VC,
Dave Barach68b0fb02017-02-28 15:15:56 -05001089};
1090/* *INDENT-ON* */
1091
1092void
1093tcp_timer_keep_handler (u32 conn_index)
1094{
Damjan Marion586afd72017-04-05 19:18:20 +02001095 u32 thread_index = vlib_get_thread_index ();
Dave Barach68b0fb02017-02-28 15:15:56 -05001096 tcp_connection_t *tc;
1097
Damjan Marion586afd72017-04-05 19:18:20 +02001098 tc = tcp_connection_get (conn_index, thread_index);
Dave Barach68b0fb02017-02-28 15:15:56 -05001099 tc->timers[TCP_TIMER_KEEP] = TCP_TIMER_HANDLE_INVALID;
1100
1101 tcp_connection_close (tc);
1102}
1103
1104void
1105tcp_timer_establish_handler (u32 conn_index)
1106{
1107 tcp_connection_t *tc;
Dave Barach68b0fb02017-02-28 15:15:56 -05001108
1109 tc = tcp_half_open_connection_get (conn_index);
Florin Corasab0289a2017-08-14 11:25:25 -07001110 if (tc)
1111 {
1112 ASSERT (tc->state == TCP_STATE_SYN_SENT);
Florin Coras3cbc04b2017-10-02 00:18:51 -07001113 session_stream_connect_notify (&tc->connection, 1 /* fail */ );
Florin Coras9d063042017-09-14 03:08:00 -04001114 TCP_DBG ("establish pop: %U", format_tcp_connection, tc, 2);
Florin Corasab0289a2017-08-14 11:25:25 -07001115 }
1116 else
1117 {
1118 tc = tcp_connection_get (conn_index, vlib_get_thread_index ());
Dave Barachb7f1faa2017-08-29 11:43:37 -04001119 /* note: the connection may have already disappeared */
1120 if (PREDICT_FALSE (tc == 0))
1121 return;
Florin Coras9d063042017-09-14 03:08:00 -04001122 TCP_DBG ("establish pop: %U", format_tcp_connection, tc, 2);
Florin Corasab0289a2017-08-14 11:25:25 -07001123 ASSERT (tc->state == TCP_STATE_SYN_RCVD);
Florin Coras4eeeaaf2017-09-05 14:03:37 -04001124 /* Start cleanup. App wasn't notified yet so use delete notify as
1125 * opposed to delete to cleanup session layer state. */
1126 stream_session_delete_notify (&tc->connection);
Florin Corasab0289a2017-08-14 11:25:25 -07001127 }
rootc9d1c5b2017-08-15 12:58:31 -04001128 tc->timers[TCP_TIMER_ESTABLISH] = TCP_TIMER_HANDLE_INVALID;
Dave Barach68b0fb02017-02-28 15:15:56 -05001129 tcp_connection_cleanup (tc);
1130}
1131
1132void
Florin Corasd79b41e2017-03-04 05:37:52 -08001133tcp_timer_waitclose_handler (u32 conn_index)
Dave Barach68b0fb02017-02-28 15:15:56 -05001134{
Damjan Marion586afd72017-04-05 19:18:20 +02001135 u32 thread_index = vlib_get_thread_index ();
Dave Barach68b0fb02017-02-28 15:15:56 -05001136 tcp_connection_t *tc;
1137
Damjan Marion586afd72017-04-05 19:18:20 +02001138 tc = tcp_connection_get (conn_index, thread_index);
Florin Coras68810622017-07-24 17:40:28 -07001139 if (!tc)
1140 return;
Florin Corasd79b41e2017-03-04 05:37:52 -08001141 tc->timers[TCP_TIMER_WAITCLOSE] = TCP_TIMER_HANDLE_INVALID;
1142
1143 /* Session didn't come back with a close(). Send FIN either way
1144 * and switch to LAST_ACK. */
1145 if (tc->state == TCP_STATE_CLOSE_WAIT)
1146 {
1147 if (tc->flags & TCP_CONN_FINSNT)
1148 {
1149 clib_warning ("FIN was sent and still in CLOSE WAIT. Weird!");
1150 }
1151
1152 tcp_send_fin (tc);
1153 tc->state = TCP_STATE_LAST_ACK;
1154
1155 /* Make sure we don't wait in LAST ACK forever */
1156 tcp_timer_set (tc, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
1157
1158 /* Don't delete the connection yet */
1159 return;
1160 }
Dave Barach68b0fb02017-02-28 15:15:56 -05001161
1162 tcp_connection_del (tc);
1163}
1164
1165/* *INDENT-OFF* */
1166static timer_expiration_handler *timer_expiration_handlers[TCP_N_TIMERS] =
1167{
1168 tcp_timer_retransmit_handler,
1169 tcp_timer_delack_handler,
Florin Coras3e350af2017-03-30 02:54:28 -07001170 tcp_timer_persist_handler,
Dave Barach68b0fb02017-02-28 15:15:56 -05001171 tcp_timer_keep_handler,
Florin Corasd79b41e2017-03-04 05:37:52 -08001172 tcp_timer_waitclose_handler,
Dave Barach68b0fb02017-02-28 15:15:56 -05001173 tcp_timer_retransmit_syn_handler,
1174 tcp_timer_establish_handler
1175};
1176/* *INDENT-ON* */
1177
1178static void
1179tcp_expired_timers_dispatch (u32 * expired_timers)
1180{
1181 int i;
1182 u32 connection_index, timer_id;
1183
1184 for (i = 0; i < vec_len (expired_timers); i++)
1185 {
1186 /* Get session index and timer id */
1187 connection_index = expired_timers[i] & 0x0FFFFFFF;
1188 timer_id = expired_timers[i] >> 28;
1189
Florin Corase69f4952017-03-07 10:06:24 -08001190 TCP_EVT_DBG (TCP_EVT_TIMER_POP, connection_index, timer_id);
1191
Dave Barach68b0fb02017-02-28 15:15:56 -05001192 /* Handle expiration */
1193 (*timer_expiration_handlers[timer_id]) (connection_index);
1194 }
1195}
1196
1197void
1198tcp_initialize_timer_wheels (tcp_main_t * tm)
1199{
1200 tw_timer_wheel_16t_2w_512sl_t *tw;
Florin Corasa5464812017-04-19 13:00:05 -07001201 /* *INDENT-OFF* */
1202 foreach_vlib_main (({
1203 tw = &tm->timer_wheels[ii];
Dave Barach68b0fb02017-02-28 15:15:56 -05001204 tw_timer_wheel_init_16t_2w_512sl (tw, tcp_expired_timers_dispatch,
1205 100e-3 /* timer period 100ms */ , ~0);
Florin Corasa5464812017-04-19 13:00:05 -07001206 tw->last_run_time = vlib_time_now (this_vlib_main);
1207 }));
1208 /* *INDENT-ON* */
Dave Barach68b0fb02017-02-28 15:15:56 -05001209}
1210
1211clib_error_t *
Florin Corasa0b34a72017-03-07 01:20:52 -08001212tcp_main_enable (vlib_main_t * vm)
Dave Barach68b0fb02017-02-28 15:15:56 -05001213{
Dave Barach68b0fb02017-02-28 15:15:56 -05001214 tcp_main_t *tm = vnet_get_tcp_main ();
1215 vlib_thread_main_t *vtm = vlib_get_thread_main ();
1216 clib_error_t *error = 0;
1217 u32 num_threads;
Dave Barachb7f1faa2017-08-29 11:43:37 -04001218 int thread;
Dave Barach2c25a622017-06-26 11:35:07 -04001219 tcp_connection_t *tc __attribute__ ((unused));
Florin Coras66b11312017-07-31 17:18:03 -07001220 u32 preallocated_connections_per_thread;
Dave Barach68b0fb02017-02-28 15:15:56 -05001221
Dave Barach68b0fb02017-02-28 15:15:56 -05001222 if ((error = vlib_call_init_function (vm, ip_main_init)))
1223 return error;
1224 if ((error = vlib_call_init_function (vm, ip4_lookup_init)))
1225 return error;
1226 if ((error = vlib_call_init_function (vm, ip6_lookup_init)))
1227 return error;
1228
1229 /*
1230 * Registrations
1231 */
1232
Dave Barach68b0fb02017-02-28 15:15:56 -05001233 ip4_register_protocol (IP_PROTOCOL_TCP, tcp4_input_node.index);
rootc9d1c5b2017-08-15 12:58:31 -04001234 ip6_register_protocol (IP_PROTOCOL_TCP, tcp6_input_node.index);
Dave Barach68b0fb02017-02-28 15:15:56 -05001235
Dave Barach68b0fb02017-02-28 15:15:56 -05001236 /*
1237 * Initialize data structures
1238 */
1239
1240 num_threads = 1 /* main thread */ + vtm->n_threads;
1241 vec_validate (tm->connections, num_threads - 1);
1242
Dave Barach2c25a622017-06-26 11:35:07 -04001243 /*
Florin Coras66b11312017-07-31 17:18:03 -07001244 * Preallocate connections. Assume that thread 0 won't
1245 * use preallocated threads when running multi-core
Dave Barach2c25a622017-06-26 11:35:07 -04001246 */
Florin Coras66b11312017-07-31 17:18:03 -07001247 if (num_threads == 1)
Dave Barach2c25a622017-06-26 11:35:07 -04001248 {
Florin Coras66b11312017-07-31 17:18:03 -07001249 thread = 0;
1250 preallocated_connections_per_thread = tm->preallocated_connections;
1251 }
1252 else
1253 {
1254 thread = 1;
1255 preallocated_connections_per_thread =
1256 tm->preallocated_connections / (num_threads - 1);
1257 }
1258 for (; thread < num_threads; thread++)
1259 {
Dave Barachb7f1faa2017-08-29 11:43:37 -04001260 if (preallocated_connections_per_thread)
1261 pool_init_fixed (tm->connections[thread],
1262 preallocated_connections_per_thread);
Dave Barach2c25a622017-06-26 11:35:07 -04001263 }
1264
1265 /*
Dave Barachb7f1faa2017-08-29 11:43:37 -04001266 * Use a preallocated half-open connection pool?
Dave Barach2c25a622017-06-26 11:35:07 -04001267 */
Dave Barachb7f1faa2017-08-29 11:43:37 -04001268 if (tm->preallocated_half_open_connections)
1269 pool_init_fixed (tm->half_open_connections,
1270 tm->preallocated_half_open_connections);
Dave Barach2c25a622017-06-26 11:35:07 -04001271
Dave Barach68b0fb02017-02-28 15:15:56 -05001272 /* Initialize per worker thread tx buffers (used for control messages) */
1273 vec_validate (tm->tx_buffers, num_threads - 1);
1274
1275 /* Initialize timer wheels */
1276 vec_validate (tm->timer_wheels, num_threads - 1);
1277 tcp_initialize_timer_wheels (tm);
1278
Dave Barach68b0fb02017-02-28 15:15:56 -05001279 /* Initialize clocks per tick for TCP timestamp. Used to compute
1280 * monotonically increasing timestamps. */
1281 tm->tstamp_ticks_per_clock = vm->clib_time.seconds_per_clock
1282 / TCP_TSTAMP_RESOLUTION;
1283
Florin Coras04e53442017-07-16 17:12:15 -07001284 if (num_threads > 1)
Florin Coras68810622017-07-24 17:40:28 -07001285 {
1286 clib_spinlock_init (&tm->half_open_lock);
Florin Coras68810622017-07-24 17:40:28 -07001287 }
Florin Coras66b11312017-07-31 17:18:03 -07001288
1289 vec_validate (tm->tx_frames[0], num_threads - 1);
1290 vec_validate (tm->tx_frames[1], num_threads - 1);
Florin Coras9d063042017-09-14 03:08:00 -04001291 vec_validate (tm->ip_lookup_tx_frames[0], num_threads - 1);
1292 vec_validate (tm->ip_lookup_tx_frames[1], num_threads - 1);
Florin Coras66b11312017-07-31 17:18:03 -07001293
Florin Corasb2215d62017-08-01 16:56:58 -07001294 tm->bytes_per_buffer = vlib_buffer_free_list_buffer_size
1295 (vm, VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX);
Florin Coras82d3ec82017-08-14 08:10:42 -07001296
1297 vec_validate (tm->time_now, num_threads - 1);
Dave Barach68b0fb02017-02-28 15:15:56 -05001298 return error;
1299}
1300
Florin Corasa0b34a72017-03-07 01:20:52 -08001301clib_error_t *
1302vnet_tcp_enable_disable (vlib_main_t * vm, u8 is_en)
1303{
1304 if (is_en)
1305 {
1306 if (tcp_main.is_enabled)
1307 return 0;
1308
1309 return tcp_main_enable (vm);
1310 }
1311 else
1312 {
1313 tcp_main.is_enabled = 0;
1314 }
1315
1316 return 0;
1317}
1318
Pierre Pfister7fe51f32017-09-20 08:48:36 +02001319void
1320tcp_punt_unknown (vlib_main_t * vm, u8 is_ip4, u8 is_add)
1321{
1322 tcp_main_t *tm = &tcp_main;
1323 if (is_ip4)
1324 tm->punt_unknown4 = is_add;
1325 else
1326 tm->punt_unknown6 = is_add;
1327}
1328
Florin Corasa0b34a72017-03-07 01:20:52 -08001329clib_error_t *
1330tcp_init (vlib_main_t * vm)
1331{
1332 tcp_main_t *tm = vnet_get_tcp_main ();
Florin Coras40364272017-11-16 09:57:50 -08001333 ip_main_t *im = &ip_main;
1334 ip_protocol_info_t *pi;
1335
1336 /* Session layer, and by implication tcp, are disabled by default */
Florin Corasa0b34a72017-03-07 01:20:52 -08001337 tm->is_enabled = 0;
Florin Coras40364272017-11-16 09:57:50 -08001338
1339 /* Register with IP for header parsing */
1340 pi = ip_get_protocol_info (im, IP_PROTOCOL_TCP);
1341 if (pi == 0)
1342 return clib_error_return (0, "TCP protocol info AWOL");
1343 pi->format_header = format_tcp_header;
1344 pi->unformat_pg_edit = unformat_pg_tcp_header;
1345
Florin Coras561af9b2017-12-09 10:19:43 -08001346 /* Register as transport with session layer */
1347 transport_register_protocol (TRANSPORT_PROTO_TCP, &tcp_proto,
1348 FIB_PROTOCOL_IP4, tcp4_output_node.index);
1349 transport_register_protocol (TRANSPORT_PROTO_TCP, &tcp_proto,
1350 FIB_PROTOCOL_IP6, tcp6_output_node.index);
1351
Dave Barach3bbcfab2017-08-15 19:03:44 -04001352 tcp_api_reference ();
Florin Corasa0b34a72017-03-07 01:20:52 -08001353 return 0;
1354}
1355
Dave Barach68b0fb02017-02-28 15:15:56 -05001356VLIB_INIT_FUNCTION (tcp_init);
1357
Dave Barach2c25a622017-06-26 11:35:07 -04001358static clib_error_t *
1359tcp_config_fn (vlib_main_t * vm, unformat_input_t * input)
1360{
1361 tcp_main_t *tm = vnet_get_tcp_main ();
1362
1363 while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1364 {
1365 if (unformat
1366 (input, "preallocated-connections %d",
1367 &tm->preallocated_connections))
1368 ;
1369 else if (unformat (input, "preallocated-half-open-connections %d",
1370 &tm->preallocated_half_open_connections))
1371 ;
Florin Corasf988e692017-11-27 04:34:14 -05001372 else if (unformat (input, "buffer-fail-fraction %f",
1373 &tm->buffer_fail_fraction))
1374 ;
Dave Barach2c25a622017-06-26 11:35:07 -04001375 else
1376 return clib_error_return (0, "unknown input `%U'",
1377 format_unformat_error, input);
1378 }
1379 return 0;
1380}
1381
1382VLIB_CONFIG_FUNCTION (tcp_config_fn, "tcp");
1383
Dave Barach3bbcfab2017-08-15 19:03:44 -04001384
1385/**
1386 * \brief Configure an ipv4 source address range
1387 * @param vm vlib_main_t pointer
1388 * @param start first ipv4 address in the source address range
1389 * @param end last ipv4 address in the source address range
1390 * @param table_id VRF / table ID, 0 for the default FIB
1391 * @return 0 if all OK, else an error indication from api_errno.h
1392 */
1393
1394int
1395tcp_configure_v4_source_address_range (vlib_main_t * vm,
1396 ip4_address_t * start,
1397 ip4_address_t * end, u32 table_id)
1398{
1399 tcp_main_t *tm = vnet_get_tcp_main ();
1400 vnet_main_t *vnm = vnet_get_main ();
1401 u32 start_host_byte_order, end_host_byte_order;
1402 fib_prefix_t prefix;
1403 vnet_sw_interface_t *si;
1404 fib_node_index_t fei;
1405 u32 fib_index = 0;
1406 u32 sw_if_index;
1407 int rv;
1408 int vnet_proxy_arp_add_del (ip4_address_t * lo_addr,
1409 ip4_address_t * hi_addr, u32 fib_index,
1410 int is_del);
1411
1412 memset (&prefix, 0, sizeof (prefix));
1413
1414 fib_index = fib_table_find (FIB_PROTOCOL_IP4, table_id);
1415
1416 if (fib_index == ~0)
1417 return VNET_API_ERROR_NO_SUCH_FIB;
1418
1419 start_host_byte_order = clib_net_to_host_u32 (start->as_u32);
1420 end_host_byte_order = clib_net_to_host_u32 (end->as_u32);
1421
1422 /* sanity check for reversed args or some such */
1423 if ((end_host_byte_order - start_host_byte_order) > (10 << 10))
1424 return VNET_API_ERROR_INVALID_ARGUMENT;
1425
1426 /* Lookup the last address, to identify the interface involved */
1427 prefix.fp_len = 32;
1428 prefix.fp_proto = FIB_PROTOCOL_IP4;
1429 memcpy (&prefix.fp_addr.ip4, end, sizeof (ip4_address_t));
1430
1431 fei = fib_table_lookup (fib_index, &prefix);
1432
1433 /* Couldn't find route to destination. Bail out. */
1434 if (fei == FIB_NODE_INDEX_INVALID)
1435 return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1436
1437 sw_if_index = fib_entry_get_resolving_interface (fei);
1438
1439 /* Enable proxy arp on the interface */
1440 si = vnet_get_sw_interface (vnm, sw_if_index);
1441 si->flags |= VNET_SW_INTERFACE_FLAG_PROXY_ARP;
1442
1443 /* Configure proxy arp across the range */
1444 rv = vnet_proxy_arp_add_del (start, end, fib_index, 0 /* is_del */ );
1445
1446 if (rv)
1447 return rv;
1448
1449 do
1450 {
1451 dpo_id_t dpo = DPO_INVALID;
1452
1453 vec_add1 (tm->ip4_src_addresses, start[0]);
1454
1455 /* Add local adjacencies for the range */
1456
1457 receive_dpo_add_or_lock (DPO_PROTO_IP4, ~0 /* sw_if_index */ ,
1458 NULL, &dpo);
1459 prefix.fp_len = 32;
1460 prefix.fp_proto = FIB_PROTOCOL_IP4;
1461 prefix.fp_addr.ip4.as_u32 = start->as_u32;
1462
1463 fib_table_entry_special_dpo_update (fib_index,
1464 &prefix,
1465 FIB_SOURCE_API,
1466 FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
1467 dpo_reset (&dpo);
1468
1469 start_host_byte_order++;
1470 start->as_u32 = clib_host_to_net_u32 (start_host_byte_order);
1471 }
1472 while (start_host_byte_order <= end_host_byte_order);
1473
1474 return 0;
1475}
1476
1477/**
1478 * \brief Configure an ipv6 source address range
1479 * @param vm vlib_main_t pointer
1480 * @param start first ipv6 address in the source address range
1481 * @param end last ipv6 address in the source address range
1482 * @param table_id VRF / table ID, 0 for the default FIB
1483 * @return 0 if all OK, else an error indication from api_errno.h
1484 */
1485
1486int
1487tcp_configure_v6_source_address_range (vlib_main_t * vm,
1488 ip6_address_t * start,
1489 ip6_address_t * end, u32 table_id)
1490{
1491 tcp_main_t *tm = vnet_get_tcp_main ();
1492 fib_prefix_t prefix;
1493 u32 fib_index = 0;
1494 fib_node_index_t fei;
1495 u32 sw_if_index;
1496
1497 memset (&prefix, 0, sizeof (prefix));
1498
1499 fib_index = fib_table_find (FIB_PROTOCOL_IP6, table_id);
1500
1501 if (fib_index == ~0)
1502 return VNET_API_ERROR_NO_SUCH_FIB;
1503
1504 while (1)
1505 {
1506 int i;
1507 ip6_address_t tmp;
1508 dpo_id_t dpo = DPO_INVALID;
1509
1510 /* Remember this address */
1511 vec_add1 (tm->ip6_src_addresses, start[0]);
1512
1513 /* Lookup the prefix, to identify the interface involved */
1514 prefix.fp_len = 128;
1515 prefix.fp_proto = FIB_PROTOCOL_IP6;
1516 memcpy (&prefix.fp_addr.ip6, start, sizeof (ip6_address_t));
1517
1518 fei = fib_table_lookup (fib_index, &prefix);
1519
1520 /* Couldn't find route to destination. Bail out. */
1521 if (fei == FIB_NODE_INDEX_INVALID)
1522 return VNET_API_ERROR_NEXT_HOP_NOT_IN_FIB;
1523
1524 sw_if_index = fib_entry_get_resolving_interface (fei);
1525
1526 if (sw_if_index == (u32) ~ 0)
1527 return VNET_API_ERROR_NO_MATCHING_INTERFACE;
1528
1529 /* Add a proxy neighbor discovery entry for this address */
1530 ip6_neighbor_proxy_add_del (sw_if_index, start, 0 /* is_del */ );
1531
1532 /* Add a receive adjacency for this address */
1533 receive_dpo_add_or_lock (DPO_PROTO_IP6, ~0 /* sw_if_index */ ,
1534 NULL, &dpo);
1535
1536 fib_table_entry_special_dpo_update (fib_index,
1537 &prefix,
1538 FIB_SOURCE_API,
1539 FIB_ENTRY_FLAG_EXCLUSIVE, &dpo);
1540 dpo_reset (&dpo);
1541
1542 /* Done with the entire range? */
1543 if (!memcmp (start, end, sizeof (start[0])))
1544 break;
1545
1546 /* Increment the address. DGMS. */
1547 tmp = start[0];
1548 for (i = 15; i >= 0; i--)
1549 {
1550 tmp.as_u8[i] += 1;
1551 if (tmp.as_u8[i] != 0)
1552 break;
1553 }
1554 start[0] = tmp;
1555 }
1556 return 0;
1557}
1558
Dave Barach2c25a622017-06-26 11:35:07 -04001559static clib_error_t *
1560tcp_src_address (vlib_main_t * vm,
1561 unformat_input_t * input, vlib_cli_command_t * cmd_arg)
1562{
Dave Barach2c25a622017-06-26 11:35:07 -04001563 ip4_address_t v4start, v4end;
1564 ip6_address_t v6start, v6end;
Dave Barach3bbcfab2017-08-15 19:03:44 -04001565 u32 table_id = 0;
Dave Barach2c25a622017-06-26 11:35:07 -04001566 int v4set = 0;
1567 int v6set = 0;
Dave Barach3bbcfab2017-08-15 19:03:44 -04001568 int rv;
Dave Barach2c25a622017-06-26 11:35:07 -04001569
1570 while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1571 {
1572 if (unformat (input, "%U - %U", unformat_ip4_address, &v4start,
1573 unformat_ip4_address, &v4end))
1574 v4set = 1;
1575 else if (unformat (input, "%U", unformat_ip4_address, &v4start))
1576 {
1577 memcpy (&v4end, &v4start, sizeof (v4start));
1578 v4set = 1;
1579 }
1580 else if (unformat (input, "%U - %U", unformat_ip6_address, &v6start,
Dave Barach3bbcfab2017-08-15 19:03:44 -04001581 unformat_ip6_address, &v6end))
Dave Barach2c25a622017-06-26 11:35:07 -04001582 v6set = 1;
1583 else if (unformat (input, "%U", unformat_ip6_address, &v6start))
1584 {
Yoann Desmouceaux6b297aa2017-09-20 10:34:22 +02001585 memcpy (&v6end, &v6start, sizeof (v6start));
Dave Barach2c25a622017-06-26 11:35:07 -04001586 v6set = 1;
1587 }
Dave Barach3bbcfab2017-08-15 19:03:44 -04001588 else if (unformat (input, "fib-table %d", &table_id))
1589 ;
Dave Barach2c25a622017-06-26 11:35:07 -04001590 else
1591 break;
1592 }
1593
1594 if (!v4set && !v6set)
1595 return clib_error_return (0, "at least one v4 or v6 address required");
1596
1597 if (v4set)
1598 {
Dave Barach3bbcfab2017-08-15 19:03:44 -04001599 rv = tcp_configure_v4_source_address_range (vm, &v4start, &v4end,
1600 table_id);
1601 switch (rv)
Dave Barach2c25a622017-06-26 11:35:07 -04001602 {
Dave Barach3bbcfab2017-08-15 19:03:44 -04001603 case 0:
1604 break;
1605
1606 case VNET_API_ERROR_NO_SUCH_FIB:
1607 return clib_error_return (0, "Invalid table-id %d", table_id);
1608
1609 case VNET_API_ERROR_INVALID_ARGUMENT:
1610 return clib_error_return (0, "Invalid address range %U - %U",
1611 format_ip4_address, &v4start,
1612 format_ip4_address, &v4end);
1613 default:
1614 return clib_error_return (0, "error %d", rv);
1615 break;
Dave Barach2c25a622017-06-26 11:35:07 -04001616 }
Dave Barach2c25a622017-06-26 11:35:07 -04001617 }
1618 if (v6set)
1619 {
Dave Barach3bbcfab2017-08-15 19:03:44 -04001620 rv = tcp_configure_v6_source_address_range (vm, &v6start, &v6end,
1621 table_id);
1622 switch (rv)
1623 {
1624 case 0:
1625 break;
1626
1627 case VNET_API_ERROR_NO_SUCH_FIB:
1628 return clib_error_return (0, "Invalid table-id %d", table_id);
1629
1630 default:
1631 return clib_error_return (0, "error %d", rv);
1632 break;
1633 }
Dave Barach2c25a622017-06-26 11:35:07 -04001634 }
1635 return 0;
1636}
1637
1638/* *INDENT-OFF* */
1639VLIB_CLI_COMMAND (tcp_src_address_command, static) =
1640{
1641 .path = "tcp src-address",
1642 .short_help = "tcp src-address <ip-addr> [- <ip-addr>] add src address range",
1643 .function = tcp_src_address,
1644};
1645/* *INDENT-ON* */
1646
Florin Coras3eb50622017-07-13 01:24:57 -04001647static u8 *
1648tcp_scoreboard_dump_trace (u8 * s, sack_scoreboard_t * sb)
1649{
1650#if TCP_SCOREBOARD_TRACE
Dave Barach2c25a622017-06-26 11:35:07 -04001651
Florin Coras3eb50622017-07-13 01:24:57 -04001652 scoreboard_trace_elt_t *block;
1653 int i = 0;
1654
1655 if (!sb->trace)
1656 return s;
1657
1658 s = format (s, "scoreboard trace:");
1659 vec_foreach (block, sb->trace)
1660 {
1661 s = format (s, "{%u, %u, %u, %u, %u}, ", block->start, block->end,
1662 block->ack, block->snd_una_max, block->group);
1663 if ((++i % 3) == 0)
1664 s = format (s, "\n");
1665 }
1666 return s;
1667#else
1668 return 0;
1669#endif
1670}
1671
1672static clib_error_t *
1673tcp_show_scoreboard_trace_fn (vlib_main_t * vm, unformat_input_t * input,
1674 vlib_cli_command_t * cmd_arg)
1675{
1676 transport_connection_t *tconn = 0;
1677 tcp_connection_t *tc;
1678 u8 *s = 0;
1679 while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1680 {
1681 if (unformat (input, "%U", unformat_transport_connection, &tconn,
1682 TRANSPORT_PROTO_TCP))
1683 ;
1684 else
1685 return clib_error_return (0, "unknown input `%U'",
1686 format_unformat_error, input);
1687 }
1688
1689 if (!TCP_SCOREBOARD_TRACE)
1690 {
1691 vlib_cli_output (vm, "scoreboard tracing not enabled");
1692 return 0;
1693 }
1694
1695 tc = tcp_get_connection_from_transport (tconn);
1696 s = tcp_scoreboard_dump_trace (s, &tc->sack_sb);
1697 vlib_cli_output (vm, "%v", s);
1698 return 0;
1699}
1700
1701/* *INDENT-OFF* */
1702VLIB_CLI_COMMAND (tcp_show_scoreboard_trace_command, static) =
1703{
1704 .path = "show tcp scoreboard trace",
1705 .short_help = "show tcp scoreboard trace <connection>",
1706 .function = tcp_show_scoreboard_trace_fn,
1707};
1708/* *INDENT-ON* */
1709
1710u8 *
1711tcp_scoreboard_replay (u8 * s, tcp_connection_t * tc, u8 verbose)
1712{
1713 int i, trace_len;
1714 scoreboard_trace_elt_t *trace;
1715 u32 next_ack, left, group, has_new_ack = 0;
1716 tcp_connection_t _dummy_tc, *dummy_tc = &_dummy_tc;
1717 sack_block_t *block;
1718
1719 if (!tc)
1720 return s;
1721
1722 memset (dummy_tc, 0, sizeof (*dummy_tc));
1723 tcp_connection_timers_init (dummy_tc);
1724 scoreboard_init (&dummy_tc->sack_sb);
1725 dummy_tc->rcv_opts.flags |= TCP_OPTS_FLAG_SACK;
1726
1727#if TCP_SCOREBOARD_TRACE
1728 trace = tc->sack_sb.trace;
1729 trace_len = vec_len (tc->sack_sb.trace);
1730#else
1731 trace = 0;
1732 trace_len = 0;
1733#endif
1734
1735 for (i = 0; i < trace_len; i++)
1736 {
1737 if (trace[i].ack != 0)
1738 {
1739 dummy_tc->snd_una = trace[i].ack - 1448;
1740 dummy_tc->snd_una_max = trace[i].ack;
1741 }
1742 }
1743
1744 left = 0;
1745 while (left < trace_len)
1746 {
1747 group = trace[left].group;
1748 vec_reset_length (dummy_tc->rcv_opts.sacks);
1749 has_new_ack = 0;
1750 while (trace[left].group == group)
1751 {
1752 if (trace[left].ack != 0)
1753 {
1754 if (verbose)
1755 s = format (s, "Adding ack %u, snd_una_max %u, segs: ",
1756 trace[left].ack, trace[left].snd_una_max);
1757 dummy_tc->snd_una_max = trace[left].snd_una_max;
1758 next_ack = trace[left].ack;
1759 has_new_ack = 1;
1760 }
1761 else
1762 {
1763 if (verbose)
1764 s = format (s, "[%u, %u], ", trace[left].start,
1765 trace[left].end);
1766 vec_add2 (dummy_tc->rcv_opts.sacks, block, 1);
1767 block->start = trace[left].start;
1768 block->end = trace[left].end;
1769 }
1770 left++;
1771 }
1772
1773 /* Push segments */
1774 tcp_rcv_sacks (dummy_tc, next_ack);
1775 if (has_new_ack)
1776 dummy_tc->snd_una = next_ack + dummy_tc->sack_sb.snd_una_adv;
1777
1778 if (verbose)
1779 s = format (s, "result: %U", format_tcp_scoreboard,
1780 &dummy_tc->sack_sb);
1781
1782 }
1783 s = format (s, "result: %U", format_tcp_scoreboard, &dummy_tc->sack_sb);
1784
1785 return s;
1786}
1787
1788static clib_error_t *
1789tcp_scoreboard_trace_fn (vlib_main_t * vm, unformat_input_t * input,
1790 vlib_cli_command_t * cmd_arg)
1791{
1792 transport_connection_t *tconn = 0;
1793 tcp_connection_t *tc = 0;
1794 u8 *str = 0;
1795 while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1796 {
1797 if (unformat (input, "%U", unformat_transport_connection, &tconn,
1798 TRANSPORT_PROTO_TCP))
1799 ;
1800 else
1801 return clib_error_return (0, "unknown input `%U'",
1802 format_unformat_error, input);
1803 }
1804
1805 if (!TCP_SCOREBOARD_TRACE)
1806 {
1807 vlib_cli_output (vm, "scoreboard tracing not enabled");
1808 return 0;
1809 }
1810
1811 tc = tcp_get_connection_from_transport (tconn);
1812 if (!tc)
1813 {
1814 vlib_cli_output (vm, "connection not found");
1815 return 0;
1816 }
1817 str = tcp_scoreboard_replay (str, tc, 1);
1818 vlib_cli_output (vm, "%v", str);
1819 return 0;
1820}
1821
1822/* *INDENT-OFF* */
1823VLIB_CLI_COMMAND (tcp_replay_scoreboard_command, static) =
1824{
1825 .path = "tcp replay scoreboard",
1826 .short_help = "tcp replay scoreboard <connection>",
1827 .function = tcp_scoreboard_trace_fn,
1828};
1829/* *INDENT-ON* */
Dave Barach2c25a622017-06-26 11:35:07 -04001830
Pierre Pfister7fe51f32017-09-20 08:48:36 +02001831static clib_error_t *
1832show_tcp_punt_fn (vlib_main_t * vm, unformat_input_t * input,
1833 vlib_cli_command_t * cmd_arg)
1834{
1835 tcp_main_t *tm = vnet_get_tcp_main ();
1836 if (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
1837 return clib_error_return (0, "unknown input `%U'", format_unformat_error,
1838 input);
1839 vlib_cli_output (vm, "IPv4 TCP punt: %s",
1840 tm->punt_unknown4 ? "enabled" : "disabled");
1841 vlib_cli_output (vm, "IPv6 TCP punt: %s",
1842 tm->punt_unknown6 ? "enabled" : "disabled");
1843 return 0;
1844}
1845/* *INDENT-OFF* */
1846VLIB_CLI_COMMAND (show_tcp_punt_command, static) =
1847{
1848 .path = "show tcp punt",
1849 .short_help = "show tcp punt",
1850 .function = show_tcp_punt_fn,
1851};
1852/* *INDENT-ON* */
1853
Dave Barach68b0fb02017-02-28 15:15:56 -05001854/*
1855 * fd.io coding-style-patch-verification: ON
1856 *
1857 * Local Variables:
1858 * eval: (c-set-style "gnu")
1859 * End:
1860 */