blob: 0473f198e159edc6263fb7561183ecb1e0148874 [file] [log] [blame]
Florin Coras3cbc04b2017-10-02 00:18:51 -07001/*
Florin Coras288eaab2019-02-03 15:26:14 -08002 * Copyright (c) 2017-2019 Cisco and/or its affiliates.
Florin Coras3cbc04b2017-10-02 00:18:51 -07003 * 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
Florin Coras1ee78302019-02-05 15:51:15 -080016#include <vnet/session/transport.h>
Florin Coras3cbc04b2017-10-02 00:18:51 -070017#include <vnet/session/session.h>
18#include <vnet/fib/fib.h>
19
Florin Coras57660d92020-04-04 22:45:34 +000020typedef struct local_endpoint_
21{
22 transport_endpoint_t ep;
23 int refcnt;
24} local_endpoint_t;
25
Florin Coras3cbc04b2017-10-02 00:18:51 -070026/**
27 * Per-type vector of transport protocol virtual function tables
28 */
29transport_proto_vft_t *tp_vfts;
30
31/*
32 * Port allocator seed
33 */
34static u32 port_allocator_seed;
35
36/*
37 * Local endpoints table
38 */
39static transport_endpoint_table_t local_endpoints_table;
40
41/*
42 * Pool of local endpoints
43 */
Florin Coras57660d92020-04-04 22:45:34 +000044static local_endpoint_t *local_endpoints;
Florin Coras3cbc04b2017-10-02 00:18:51 -070045
46/*
47 * Local endpoints pool lock
48 */
49static clib_spinlock_t local_endpoints_lock;
50
Florin Coras1c710452017-10-17 00:03:13 -070051u8 *
52format_transport_proto (u8 * s, va_list * args)
53{
54 u32 transport_proto = va_arg (*args, u32);
Florin Coras07063b82020-03-13 04:44:51 +000055
56 if (tp_vfts[transport_proto].transport_options.name)
57 s = format (s, "%s", tp_vfts[transport_proto].transport_options.name);
58 else
59 s = format (s, "n/a");
60
Florin Coras1c710452017-10-17 00:03:13 -070061 return s;
62}
63
Florin Coras561af9b2017-12-09 10:19:43 -080064u8 *
65format_transport_proto_short (u8 * s, va_list * args)
66{
67 u32 transport_proto = va_arg (*args, u32);
Florin Coras07063b82020-03-13 04:44:51 +000068 char *short_name;
69
70 short_name = tp_vfts[transport_proto].transport_options.short_name;
71 if (short_name)
72 s = format (s, "%s", short_name);
73 else
74 s = format (s, "NA");
75
Florin Coras561af9b2017-12-09 10:19:43 -080076 return s;
77}
78
Florin Corasde9a8492018-10-24 22:18:58 -070079u8 *
80format_transport_connection (u8 * s, va_list * args)
81{
82 u32 transport_proto = va_arg (*args, u32);
83 u32 conn_index = va_arg (*args, u32);
84 u32 thread_index = va_arg (*args, u32);
85 u32 verbose = va_arg (*args, u32);
86 transport_proto_vft_t *tp_vft;
87 transport_connection_t *tc;
88 u32 indent;
89
90 tp_vft = transport_protocol_get_vft (transport_proto);
91 if (!tp_vft)
92 return s;
93
94 s = format (s, "%U", tp_vft->format_connection, conn_index, thread_index,
95 verbose);
96 tc = tp_vft->get_connection (conn_index, thread_index);
Florin Coras36d49392020-04-24 23:00:11 +000097 if (tc && verbose > 1)
Florin Corasde9a8492018-10-24 22:18:58 -070098 {
99 indent = format_get_indent (s) + 1;
Florin Coras36d49392020-04-24 23:00:11 +0000100 if (transport_connection_is_tx_paced (tc))
101 s = format (s, "%Upacer: %U\n", format_white_space, indent,
102 format_transport_pacer, &tc->pacer, tc->thread_index);
Florin Coras70f879d2020-03-13 17:54:42 +0000103 s = format (s, "%Utransport: flags 0x%x\n", format_white_space, indent,
104 tc->flags);
Florin Corasde9a8492018-10-24 22:18:58 -0700105 }
106 return s;
107}
108
109u8 *
110format_transport_listen_connection (u8 * s, va_list * args)
111{
112 u32 transport_proto = va_arg (*args, u32);
Florin Corasde9a8492018-10-24 22:18:58 -0700113 transport_proto_vft_t *tp_vft;
114
115 tp_vft = transport_protocol_get_vft (transport_proto);
116 if (!tp_vft)
117 return s;
118
Florin Coras3389dd22019-02-01 18:00:05 -0800119 s = (tp_vft->format_listener) (s, args);
Florin Corasde9a8492018-10-24 22:18:58 -0700120 return s;
121}
122
123u8 *
124format_transport_half_open_connection (u8 * s, va_list * args)
125{
126 u32 transport_proto = va_arg (*args, u32);
Florin Corasea727642021-05-07 19:39:43 -0700127 u32 ho_index = va_arg (*args, u32);
Florin Corasde9a8492018-10-24 22:18:58 -0700128 transport_proto_vft_t *tp_vft;
129
130 tp_vft = transport_protocol_get_vft (transport_proto);
131 if (!tp_vft)
132 return s;
133
Florin Corasea727642021-05-07 19:39:43 -0700134 s = format (s, "%U", tp_vft->format_half_open, ho_index);
Florin Corasde9a8492018-10-24 22:18:58 -0700135 return s;
136}
137
Florin Coras3bbbf0d2019-11-19 17:23:22 -0800138static u8
139unformat_transport_str_match (unformat_input_t * input, const char *str)
140{
141 int i;
142
143 if (strlen (str) > vec_len (input->buffer) - input->index)
144 return 0;
145
146 for (i = 0; i < strlen (str); i++)
147 {
148 if (input->buffer[i + input->index] != str[i])
149 return 0;
150 }
151 return 1;
152}
153
Florin Coras1c710452017-10-17 00:03:13 -0700154uword
155unformat_transport_proto (unformat_input_t * input, va_list * args)
156{
157 u32 *proto = va_arg (*args, u32 *);
Florin Coras07063b82020-03-13 04:44:51 +0000158 transport_proto_vft_t *tp_vft;
Florin Coras3bbbf0d2019-11-19 17:23:22 -0800159 u8 longest_match = 0, match;
Florin Coras07063b82020-03-13 04:44:51 +0000160 char *str, *str_match = 0;
161 transport_proto_t tp;
Florin Coras5bb23ec2019-08-31 09:45:13 -0700162
Florin Coras07063b82020-03-13 04:44:51 +0000163 for (tp = 0; tp < vec_len (tp_vfts); tp++)
164 {
165 tp_vft = &tp_vfts[tp];
166 str = tp_vft->transport_options.name;
167 if (!str)
168 continue;
169 if (unformat_transport_str_match (input, str))
170 {
171 match = strlen (str);
172 if (match > longest_match)
173 {
174 *proto = tp;
175 longest_match = match;
176 str_match = str;
177 }
178 }
Florin Coras5bb23ec2019-08-31 09:45:13 -0700179 }
Florin Coras07063b82020-03-13 04:44:51 +0000180 if (longest_match)
Florin Coras3bbbf0d2019-11-19 17:23:22 -0800181 {
Dave Barach4897d772020-03-26 10:56:13 -0400182 (void) unformat (input, str_match);
Florin Coras3bbbf0d2019-11-19 17:23:22 -0800183 return 1;
184 }
185
186 return 0;
Florin Coras1c710452017-10-17 00:03:13 -0700187}
Florin Coras3cbc04b2017-10-02 00:18:51 -0700188
Florin Coras07063b82020-03-13 04:44:51 +0000189u8 *
190format_transport_protos (u8 * s, va_list * args)
191{
192 transport_proto_vft_t *tp_vft;
193
194 vec_foreach (tp_vft, tp_vfts)
195 s = format (s, "%s\n", tp_vft->transport_options.name);
196
197 return s;
198}
199
Florin Coras3cbc04b2017-10-02 00:18:51 -0700200u32
201transport_endpoint_lookup (transport_endpoint_table_t * ht, u8 proto,
202 ip46_address_t * ip, u16 port)
203{
204 clib_bihash_kv_24_8_t kv;
205 int rv;
206
207 kv.key[0] = ip->as_u64[0];
208 kv.key[1] = ip->as_u64[1];
209 kv.key[2] = (u64) port << 8 | (u64) proto;
210
211 rv = clib_bihash_search_inline_24_8 (ht, &kv);
212 if (rv == 0)
213 return kv.value;
214
215 return ENDPOINT_INVALID_INDEX;
216}
217
218void
219transport_endpoint_table_add (transport_endpoint_table_t * ht, u8 proto,
220 transport_endpoint_t * te, u32 value)
221{
222 clib_bihash_kv_24_8_t kv;
223
224 kv.key[0] = te->ip.as_u64[0];
225 kv.key[1] = te->ip.as_u64[1];
226 kv.key[2] = (u64) te->port << 8 | (u64) proto;
227 kv.value = value;
228
229 clib_bihash_add_del_24_8 (ht, &kv, 1);
230}
231
232void
233transport_endpoint_table_del (transport_endpoint_table_t * ht, u8 proto,
234 transport_endpoint_t * te)
235{
236 clib_bihash_kv_24_8_t kv;
237
238 kv.key[0] = te->ip.as_u64[0];
239 kv.key[1] = te->ip.as_u64[1];
240 kv.key[2] = (u64) te->port << 8 | (u64) proto;
241
242 clib_bihash_add_del_24_8 (ht, &kv, 0);
243}
244
Florin Coras3cbc04b2017-10-02 00:18:51 -0700245void
Florin Coras561af9b2017-12-09 10:19:43 -0800246transport_register_protocol (transport_proto_t transport_proto,
247 const transport_proto_vft_t * vft,
248 fib_protocol_t fib_proto, u32 output_node)
Florin Coras3cbc04b2017-10-02 00:18:51 -0700249{
Florin Coras561af9b2017-12-09 10:19:43 -0800250 u8 is_ip4 = fib_proto == FIB_PROTOCOL_IP4;
Florin Coras3cbc04b2017-10-02 00:18:51 -0700251
Florin Coras561af9b2017-12-09 10:19:43 -0800252 vec_validate (tp_vfts, transport_proto);
253 tp_vfts[transport_proto] = *vft;
Florin Coras3cbc04b2017-10-02 00:18:51 -0700254
Florin Coras561af9b2017-12-09 10:19:43 -0800255 session_register_transport (transport_proto, vft, is_ip4, output_node);
Florin Coras3cbc04b2017-10-02 00:18:51 -0700256}
257
Florin Coras07063b82020-03-13 04:44:51 +0000258transport_proto_t
259transport_register_new_protocol (const transport_proto_vft_t * vft,
260 fib_protocol_t fib_proto, u32 output_node)
261{
262 transport_proto_t transport_proto;
263 u8 is_ip4;
264
265 transport_proto = session_add_transport_proto ();
266 is_ip4 = fib_proto == FIB_PROTOCOL_IP4;
267
268 vec_validate (tp_vfts, transport_proto);
269 tp_vfts[transport_proto] = *vft;
270
271 session_register_transport (transport_proto, vft, is_ip4, output_node);
272
273 return transport_proto;
274}
275
Florin Coras3cbc04b2017-10-02 00:18:51 -0700276/**
277 * Get transport virtual function table
278 *
279 * @param type - session type (not protocol type)
280 */
281transport_proto_vft_t *
Florin Coras561af9b2017-12-09 10:19:43 -0800282transport_protocol_get_vft (transport_proto_t transport_proto)
Florin Coras3cbc04b2017-10-02 00:18:51 -0700283{
Florin Coras561af9b2017-12-09 10:19:43 -0800284 if (transport_proto >= vec_len (tp_vfts))
Florin Coras3cbc04b2017-10-02 00:18:51 -0700285 return 0;
Florin Coras561af9b2017-12-09 10:19:43 -0800286 return &tp_vfts[transport_proto];
Florin Coras3cbc04b2017-10-02 00:18:51 -0700287}
288
Florin Coras7fb0fe12018-04-09 09:24:52 -0700289transport_service_type_t
290transport_protocol_service_type (transport_proto_t tp)
291{
Nathan Skrzypczake971bc92019-06-19 13:42:37 +0200292 return tp_vfts[tp].transport_options.service_type;
Florin Coras7fb0fe12018-04-09 09:24:52 -0700293}
294
Florin Corasf08f26d2018-05-10 13:20:47 -0700295transport_tx_fn_type_t
296transport_protocol_tx_fn_type (transport_proto_t tp)
297{
Nathan Skrzypczake971bc92019-06-19 13:42:37 +0200298 return tp_vfts[tp].transport_options.tx_type;
Florin Corasf08f26d2018-05-10 13:20:47 -0700299}
300
Florin Coras1ee78302019-02-05 15:51:15 -0800301void
302transport_cleanup (transport_proto_t tp, u32 conn_index, u8 thread_index)
Florin Coras4edc37e2019-02-04 23:01:34 -0800303{
Florin Coras1ee78302019-02-05 15:51:15 -0800304 tp_vfts[tp].cleanup (conn_index, thread_index);
Florin Coras4edc37e2019-02-04 23:01:34 -0800305}
306
Florin Corasd50ff7f2020-04-16 04:30:22 +0000307void
308transport_cleanup_half_open (transport_proto_t tp, u32 conn_index)
309{
Florin Coras05bc33c2021-05-19 19:33:19 -0700310 if (tp_vfts[tp].cleanup_ho)
Florin Corasd50ff7f2020-04-16 04:30:22 +0000311 tp_vfts[tp].cleanup_ho (conn_index);
312}
313
Florin Coras1ee78302019-02-05 15:51:15 -0800314int
315transport_connect (transport_proto_t tp, transport_endpoint_cfg_t * tep)
Florin Coras4edc37e2019-02-04 23:01:34 -0800316{
Florin Coras1ee78302019-02-05 15:51:15 -0800317 return tp_vfts[tp].connect (tep);
318}
319
320void
liuyacan534468e2021-05-09 03:50:40 +0000321transport_half_close (transport_proto_t tp, u32 conn_index, u8 thread_index)
322{
323 if (tp_vfts[tp].half_close)
324 tp_vfts[tp].half_close (conn_index, thread_index);
325}
326
327void
Florin Coras1ee78302019-02-05 15:51:15 -0800328transport_close (transport_proto_t tp, u32 conn_index, u8 thread_index)
329{
330 tp_vfts[tp].close (conn_index, thread_index);
331}
332
Florin Corasdfb3b872019-08-16 17:48:44 -0700333void
334transport_reset (transport_proto_t tp, u32 conn_index, u8 thread_index)
335{
336 if (tp_vfts[tp].reset)
337 tp_vfts[tp].reset (conn_index, thread_index);
338 else
339 tp_vfts[tp].close (conn_index, thread_index);
340}
341
Florin Coras1ee78302019-02-05 15:51:15 -0800342u32
343transport_start_listen (transport_proto_t tp, u32 session_index,
Florin Coras0bce71e2022-02-10 11:57:06 -0800344 transport_endpoint_cfg_t *tep)
Florin Coras1ee78302019-02-05 15:51:15 -0800345{
346 return tp_vfts[tp].start_listen (session_index, tep);
347}
348
349u32
350transport_stop_listen (transport_proto_t tp, u32 conn_index)
351{
352 return tp_vfts[tp].stop_listen (conn_index);
Florin Coras4edc37e2019-02-04 23:01:34 -0800353}
354
Florin Coras40903ac2018-06-10 14:41:23 -0700355u8
356transport_protocol_is_cl (transport_proto_t tp)
357{
Nathan Skrzypczake971bc92019-06-19 13:42:37 +0200358 return (tp_vfts[tp].transport_options.service_type == TRANSPORT_SERVICE_CL);
Florin Coras40903ac2018-06-10 14:41:23 -0700359}
360
Aloys Augustincdb71702019-04-08 17:54:39 +0200361always_inline void
362default_get_transport_endpoint (transport_connection_t * tc,
Florin Coras09d18c22019-04-24 11:10:02 -0700363 transport_endpoint_t * tep, u8 is_lcl)
Aloys Augustincdb71702019-04-08 17:54:39 +0200364{
365 if (is_lcl)
366 {
Florin Coras09d18c22019-04-24 11:10:02 -0700367 tep->port = tc->lcl_port;
368 tep->is_ip4 = tc->is_ip4;
369 clib_memcpy_fast (&tep->ip, &tc->lcl_ip, sizeof (tc->lcl_ip));
Aloys Augustincdb71702019-04-08 17:54:39 +0200370 }
371 else
372 {
Florin Coras09d18c22019-04-24 11:10:02 -0700373 tep->port = tc->rmt_port;
374 tep->is_ip4 = tc->is_ip4;
375 clib_memcpy_fast (&tep->ip, &tc->rmt_ip, sizeof (tc->rmt_ip));
Aloys Augustincdb71702019-04-08 17:54:39 +0200376 }
377}
378
379void
380transport_get_endpoint (transport_proto_t tp, u32 conn_index,
Florin Coras09d18c22019-04-24 11:10:02 -0700381 u32 thread_index, transport_endpoint_t * tep,
382 u8 is_lcl)
Aloys Augustincdb71702019-04-08 17:54:39 +0200383{
384 if (tp_vfts[tp].get_transport_endpoint)
Florin Coras09d18c22019-04-24 11:10:02 -0700385 tp_vfts[tp].get_transport_endpoint (conn_index, thread_index, tep,
386 is_lcl);
Aloys Augustincdb71702019-04-08 17:54:39 +0200387 else
388 {
389 transport_connection_t *tc;
390 tc = transport_get_connection (tp, conn_index, thread_index);
Florin Coras09d18c22019-04-24 11:10:02 -0700391 default_get_transport_endpoint (tc, tep, is_lcl);
Aloys Augustincdb71702019-04-08 17:54:39 +0200392 }
393}
394
395void
396transport_get_listener_endpoint (transport_proto_t tp, u32 conn_index,
Florin Coras09d18c22019-04-24 11:10:02 -0700397 transport_endpoint_t * tep, u8 is_lcl)
Aloys Augustincdb71702019-04-08 17:54:39 +0200398{
399 if (tp_vfts[tp].get_transport_listener_endpoint)
Florin Coras09d18c22019-04-24 11:10:02 -0700400 tp_vfts[tp].get_transport_listener_endpoint (conn_index, tep, is_lcl);
Aloys Augustincdb71702019-04-08 17:54:39 +0200401 else
402 {
403 transport_connection_t *tc;
404 tc = transport_get_listener (tp, conn_index);
Florin Coras09d18c22019-04-24 11:10:02 -0700405 default_get_transport_endpoint (tc, tep, is_lcl);
Aloys Augustincdb71702019-04-08 17:54:39 +0200406 }
407}
408
Florin Coras04ae8272021-04-12 19:55:37 -0700409int
410transport_connection_attribute (transport_proto_t tp, u32 conn_index,
411 u8 thread_index, u8 is_get,
412 transport_endpt_attr_t *attr)
413{
414 if (!tp_vfts[tp].attribute)
415 return -1;
416
417 return tp_vfts[tp].attribute (conn_index, thread_index, is_get, attr);
418}
419
Florin Coras3cbc04b2017-10-02 00:18:51 -0700420#define PORT_MASK ((1 << 16)- 1)
421
422void
Florin Coras05ead782022-03-23 16:35:05 -0700423transport_endpoint_free (u32 tepi)
Florin Coras3cbc04b2017-10-02 00:18:51 -0700424{
Florin Coras3cbc04b2017-10-02 00:18:51 -0700425 pool_put_index (local_endpoints, tepi);
Florin Coras05ead782022-03-23 16:35:05 -0700426}
427
428static void
429transport_endpoint_pool_realloc_rpc (void *rpc_args)
430{
431 pool_realloc_safe_aligned (local_endpoints, 0);
Florin Coras3cbc04b2017-10-02 00:18:51 -0700432}
433
Florin Coras57660d92020-04-04 22:45:34 +0000434always_inline local_endpoint_t *
Florin Coras05ead782022-03-23 16:35:05 -0700435transport_endpoint_alloc (void)
Florin Coras3cbc04b2017-10-02 00:18:51 -0700436{
Florin Coras57660d92020-04-04 22:45:34 +0000437 local_endpoint_t *lep;
Florin Coras05ead782022-03-23 16:35:05 -0700438
439 ASSERT (vlib_get_thread_index () <= transport_cl_thread ());
440 pool_get_aligned_safe (local_endpoints, lep, transport_cl_thread (),
441 transport_endpoint_pool_realloc_rpc, 0);
Florin Coras57660d92020-04-04 22:45:34 +0000442 return lep;
Florin Coras3cbc04b2017-10-02 00:18:51 -0700443}
444
445void
446transport_endpoint_cleanup (u8 proto, ip46_address_t * lcl_ip, u16 port)
447{
Florin Coras57660d92020-04-04 22:45:34 +0000448 local_endpoint_t *lep;
449 u32 lepi;
Florin Coras3cbc04b2017-10-02 00:18:51 -0700450
451 /* Cleanup local endpoint if this was an active connect */
Florin Coras57660d92020-04-04 22:45:34 +0000452 lepi = transport_endpoint_lookup (&local_endpoints_table, proto, lcl_ip,
Florin Coras3cbc04b2017-10-02 00:18:51 -0700453 clib_net_to_host_u16 (port));
Florin Corasf55183a2022-03-24 17:02:08 -0700454 if (lepi == ENDPOINT_INVALID_INDEX)
455 return;
456
457 lep = pool_elt_at_index (local_endpoints, lepi);
458 if (!clib_atomic_sub_fetch (&lep->refcnt, 1))
Florin Coras3cbc04b2017-10-02 00:18:51 -0700459 {
Florin Corasf55183a2022-03-24 17:02:08 -0700460 transport_endpoint_table_del (&local_endpoints_table, proto, &lep->ep);
461
462 /* All workers can free connections. Synchronize access to pool */
463 clib_spinlock_lock (&local_endpoints_lock);
464 transport_endpoint_free (lepi);
465 clib_spinlock_unlock (&local_endpoints_lock);
Florin Coras3cbc04b2017-10-02 00:18:51 -0700466 }
467}
468
Florin Corasf55183a2022-03-24 17:02:08 -0700469static int
470transport_endpoint_mark_used (u8 proto, ip46_address_t *ip, u16 port)
Florin Coras5665ced2018-10-25 18:03:45 -0700471{
Florin Coras57660d92020-04-04 22:45:34 +0000472 local_endpoint_t *lep;
Florin Corasf55183a2022-03-24 17:02:08 -0700473 u32 tei;
Florin Coras05ead782022-03-23 16:35:05 -0700474
475 ASSERT (vlib_get_thread_index () <= transport_cl_thread ());
476
Florin Corasf55183a2022-03-24 17:02:08 -0700477 tei = transport_endpoint_lookup (&local_endpoints_table, proto, ip, port);
478 if (tei != ENDPOINT_INVALID_INDEX)
479 return SESSION_E_PORTINUSE;
480
Florin Coras05ead782022-03-23 16:35:05 -0700481 /* Pool reallocs with worker barrier */
482 lep = transport_endpoint_alloc ();
Florin Coras57660d92020-04-04 22:45:34 +0000483 clib_memcpy_fast (&lep->ep.ip, ip, sizeof (*ip));
484 lep->ep.port = port;
485 lep->refcnt = 1;
Florin Corasf55183a2022-03-24 17:02:08 -0700486
Florin Coras57660d92020-04-04 22:45:34 +0000487 transport_endpoint_table_add (&local_endpoints_table, proto, &lep->ep,
488 lep - local_endpoints);
Florin Corasf55183a2022-03-24 17:02:08 -0700489
490 return 0;
Florin Coras5665ced2018-10-25 18:03:45 -0700491}
492
Florin Coras57660d92020-04-04 22:45:34 +0000493void
494transport_share_local_endpoint (u8 proto, ip46_address_t * lcl_ip, u16 port)
495{
496 local_endpoint_t *lep;
497 u32 lepi;
498
499 lepi = transport_endpoint_lookup (&local_endpoints_table, proto, lcl_ip,
500 clib_net_to_host_u16 (port));
501 if (lepi != ENDPOINT_INVALID_INDEX)
502 {
503 lep = pool_elt_at_index (local_endpoints, lepi);
504 clib_atomic_add_fetch (&lep->refcnt, 1);
505 }
506}
507
Florin Coras3cbc04b2017-10-02 00:18:51 -0700508/**
509 * Allocate local port and add if successful add entry to local endpoint
510 * table to mark the pair as used.
511 */
512int
513transport_alloc_local_port (u8 proto, ip46_address_t * ip)
514{
Florin Coras3cbc04b2017-10-02 00:18:51 -0700515 u16 min = 1024, max = 65535; /* XXX configurable ? */
516 int tries, limit;
517
518 limit = max - min;
519
Florin Coras05ead782022-03-23 16:35:05 -0700520 /* Only support active opens from one of ctrl threads */
521 ASSERT (vlib_get_thread_index () <= transport_cl_thread ());
Florin Coras3cbc04b2017-10-02 00:18:51 -0700522
523 /* Search for first free slot */
524 for (tries = 0; tries < limit; tries++)
525 {
526 u16 port = 0;
527
528 /* Find a port in the specified range */
529 while (1)
530 {
531 port = random_u32 (&port_allocator_seed) & PORT_MASK;
532 if (PREDICT_TRUE (port >= min && port < max))
533 break;
534 }
535
Florin Corasf55183a2022-03-24 17:02:08 -0700536 if (!transport_endpoint_mark_used (proto, ip, port))
537 return port;
Florin Coras3cbc04b2017-10-02 00:18:51 -0700538 }
539 return -1;
540}
541
Florin Coras00e01d32019-10-21 16:07:46 -0700542static session_error_t
Florin Coras5665ced2018-10-25 18:03:45 -0700543transport_get_interface_ip (u32 sw_if_index, u8 is_ip4, ip46_address_t * addr)
Florin Coras3cbc04b2017-10-02 00:18:51 -0700544{
Florin Coras5665ced2018-10-25 18:03:45 -0700545 if (is_ip4)
Florin Coras3cbc04b2017-10-02 00:18:51 -0700546 {
547 ip4_address_t *ip4;
548 ip4 = ip_interface_get_first_ip (sw_if_index, 1);
Florin Corasfc804d92018-01-26 01:27:01 -0800549 if (!ip4)
Florin Coras00e01d32019-10-21 16:07:46 -0700550 return SESSION_E_NOIP;
Florin Coras5665ced2018-10-25 18:03:45 -0700551 addr->ip4.as_u32 = ip4->as_u32;
Florin Coras3cbc04b2017-10-02 00:18:51 -0700552 }
553 else
554 {
555 ip6_address_t *ip6;
556 ip6 = ip_interface_get_first_ip (sw_if_index, 0);
557 if (ip6 == 0)
Florin Coras00e01d32019-10-21 16:07:46 -0700558 return SESSION_E_NOIP;
Dave Barach178cf492018-11-13 16:34:13 -0500559 clib_memcpy_fast (&addr->ip6, ip6, sizeof (*ip6));
Florin Coras5665ced2018-10-25 18:03:45 -0700560 }
561 return 0;
562}
563
Florin Coras00e01d32019-10-21 16:07:46 -0700564static session_error_t
Florin Coras596c45b2021-09-09 12:04:17 -0700565transport_find_local_ip_for_remote (u32 *sw_if_index,
566 transport_endpoint_t *rmt,
567 ip46_address_t *lcl_addr)
Florin Coras5665ced2018-10-25 18:03:45 -0700568{
569 fib_node_index_t fei;
570 fib_prefix_t prefix;
571
Florin Coras596c45b2021-09-09 12:04:17 -0700572 if (*sw_if_index == ENDPOINT_INVALID_INDEX)
Florin Coras5665ced2018-10-25 18:03:45 -0700573 {
574 /* Find a FIB path to the destination */
Dave Barach178cf492018-11-13 16:34:13 -0500575 clib_memcpy_fast (&prefix.fp_addr, &rmt->ip, sizeof (rmt->ip));
Florin Coras5665ced2018-10-25 18:03:45 -0700576 prefix.fp_proto = rmt->is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6;
577 prefix.fp_len = rmt->is_ip4 ? 32 : 128;
578
579 ASSERT (rmt->fib_index != ENDPOINT_INVALID_INDEX);
580 fei = fib_table_lookup (rmt->fib_index, &prefix);
581
582 /* Couldn't find route to destination. Bail out. */
583 if (fei == FIB_NODE_INDEX_INVALID)
Florin Coras00e01d32019-10-21 16:07:46 -0700584 return SESSION_E_NOROUTE;
Florin Coras5665ced2018-10-25 18:03:45 -0700585
Florin Coras596c45b2021-09-09 12:04:17 -0700586 *sw_if_index = fib_entry_get_resolving_interface (fei);
587 if (*sw_if_index == ENDPOINT_INVALID_INDEX)
Florin Coras00e01d32019-10-21 16:07:46 -0700588 return SESSION_E_NOINTF;
Florin Coras3cbc04b2017-10-02 00:18:51 -0700589 }
590
Florin Coras5665ced2018-10-25 18:03:45 -0700591 clib_memset (lcl_addr, 0, sizeof (*lcl_addr));
Florin Coras596c45b2021-09-09 12:04:17 -0700592 return transport_get_interface_ip (*sw_if_index, rmt->is_ip4, lcl_addr);
Florin Coras5665ced2018-10-25 18:03:45 -0700593}
594
595int
596transport_alloc_local_endpoint (u8 proto, transport_endpoint_cfg_t * rmt_cfg,
597 ip46_address_t * lcl_addr, u16 * lcl_port)
598{
599 transport_endpoint_t *rmt = (transport_endpoint_t *) rmt_cfg;
Florin Coras00e01d32019-10-21 16:07:46 -0700600 session_error_t error;
Florin Coras5665ced2018-10-25 18:03:45 -0700601 int port;
Florin Coras5665ced2018-10-25 18:03:45 -0700602
603 /*
604 * Find the local address
605 */
606 if (ip_is_zero (&rmt_cfg->peer.ip, rmt_cfg->peer.is_ip4))
Florin Coras3cbc04b2017-10-02 00:18:51 -0700607 {
Florin Coras596c45b2021-09-09 12:04:17 -0700608 error = transport_find_local_ip_for_remote (&rmt_cfg->peer.sw_if_index,
Florin Coras5665ced2018-10-25 18:03:45 -0700609 rmt, lcl_addr);
610 if (error)
Florin Coras00e01d32019-10-21 16:07:46 -0700611 return error;
Florin Coras3cbc04b2017-10-02 00:18:51 -0700612 }
Florin Coras5665ced2018-10-25 18:03:45 -0700613 else
614 {
615 /* Assume session layer vetted this address */
Dave Barach178cf492018-11-13 16:34:13 -0500616 clib_memcpy_fast (lcl_addr, &rmt_cfg->peer.ip,
617 sizeof (rmt_cfg->peer.ip));
Florin Coras5665ced2018-10-25 18:03:45 -0700618 }
619
620 /*
621 * Allocate source port
622 */
623 if (rmt_cfg->peer.port == 0)
624 {
625 port = transport_alloc_local_port (proto, lcl_addr);
626 if (port < 1)
Florin Coras00e01d32019-10-21 16:07:46 -0700627 return SESSION_E_NOPORT;
Florin Coras5665ced2018-10-25 18:03:45 -0700628 *lcl_port = port;
629 }
630 else
631 {
632 port = clib_net_to_host_u16 (rmt_cfg->peer.port);
Florin Coras57660d92020-04-04 22:45:34 +0000633 *lcl_port = port;
Florin Coras5665ced2018-10-25 18:03:45 -0700634
Florin Corasf55183a2022-03-24 17:02:08 -0700635 return transport_endpoint_mark_used (proto, lcl_addr, port);
Florin Coras5665ced2018-10-25 18:03:45 -0700636 }
637
Florin Coras3cbc04b2017-10-02 00:18:51 -0700638 return 0;
639}
640
Florin Corasa8e71c82019-10-22 19:01:39 -0700641u8 *
642format_clib_us_time (u8 * s, va_list * args)
643{
644 clib_us_time_t t = va_arg (*args, clib_us_time_t);
645 if (t < 1e3)
646 s = format (s, "%u us", t);
647 else
648 s = format (s, "%.3f s", (f64) t * CLIB_US_TIME_PERIOD);
649 return s;
650}
Florin Corasd67f1122018-05-21 17:47:40 -0700651
652u8 *
653format_transport_pacer (u8 * s, va_list * args)
654{
655 spacer_t *pacer = va_arg (*args, spacer_t *);
Florin Corasa8e71c82019-10-22 19:01:39 -0700656 u32 thread_index = va_arg (*args, int);
657 clib_us_time_t now, diff;
Florin Corasd67f1122018-05-21 17:47:40 -0700658
Florin Corasa8e71c82019-10-22 19:01:39 -0700659 now = transport_us_time_now (thread_index);
660 diff = now - pacer->last_update;
Florin Corasa39f4722020-11-12 10:20:05 -0800661 s = format (s, "rate %lu bucket %ld t/p %.3f last_update %U burst %u",
Florin Corasbe237bf2019-09-27 08:16:40 -0700662 pacer->bytes_per_sec, pacer->bucket, pacer->tokens_per_period,
Florin Coras7808df22020-11-02 18:00:32 -0800663 format_clib_us_time, diff, pacer->max_burst);
Florin Corasd67f1122018-05-21 17:47:40 -0700664 return s;
665}
666
667static inline u32
Florin Corasa8e71c82019-10-22 19:01:39 -0700668spacer_max_burst (spacer_t * pacer, clib_us_time_t time_now)
Florin Corasd67f1122018-05-21 17:47:40 -0700669{
Florin Corasa8e71c82019-10-22 19:01:39 -0700670 u64 n_periods = (time_now - pacer->last_update);
Florin Coras93dd58c2022-01-09 17:20:28 -0800671 i64 inc;
Florin Corasd67f1122018-05-21 17:47:40 -0700672
Florin Coras11e9e352019-11-13 19:09:47 -0800673 if ((inc = (f32) n_periods * pacer->tokens_per_period) > 10)
Florin Corase55a6d72018-10-31 23:09:22 -0700674 {
Florin Corasa8e71c82019-10-22 19:01:39 -0700675 pacer->last_update = time_now;
Florin Coras93dd58c2022-01-09 17:20:28 -0800676 pacer->bucket = clib_min (pacer->bucket + inc, (i64) pacer->max_burst);
Florin Corase55a6d72018-10-31 23:09:22 -0700677 }
678
Florin Coras2b4f74f2022-01-09 19:03:09 -0800679 return pacer->bucket >= 0 ? pacer->max_burst : 0;
Florin Corasd67f1122018-05-21 17:47:40 -0700680}
681
682static inline void
683spacer_update_bucket (spacer_t * pacer, u32 bytes)
684{
Florin Corasd67f1122018-05-21 17:47:40 -0700685 pacer->bucket -= bytes;
686}
687
688static inline void
Florin Coras11e9e352019-11-13 19:09:47 -0800689spacer_set_pace_rate (spacer_t * pacer, u64 rate_bytes_per_sec,
Florin Corasa39f4722020-11-12 10:20:05 -0800690 clib_us_time_t rtt, clib_time_type_t sec_per_loop)
Florin Corasd67f1122018-05-21 17:47:40 -0700691{
Florin Coras7808df22020-11-02 18:00:32 -0800692 clib_us_time_t max_time;
693
Florin Corasd67f1122018-05-21 17:47:40 -0700694 ASSERT (rate_bytes_per_sec != 0);
Florin Coras7c8f8282019-09-15 16:28:45 -0700695 pacer->bytes_per_sec = rate_bytes_per_sec;
Florin Corasa8e71c82019-10-22 19:01:39 -0700696 pacer->tokens_per_period = rate_bytes_per_sec * CLIB_US_TIME_PERIOD;
Florin Coras7808df22020-11-02 18:00:32 -0800697
698 /* Allow a min number of bursts per rtt, if their size is acceptable. Goal
699 * is to spread the sending of data over the rtt but to also allow for some
700 * coalescing that can potentially
701 * 1) reduce load on session layer by reducing scheduling frequency for a
702 * session and
703 * 2) optimize sending when tso if available
704 *
705 * Max "time-length" of a burst cannot be less than 1us or more than 1ms.
706 */
Florin Corasa39f4722020-11-12 10:20:05 -0800707 max_time = clib_max (rtt / TRANSPORT_PACER_BURSTS_PER_RTT,
708 (clib_us_time_t) (sec_per_loop * CLIB_US_TIME_FREQ));
Florin Coras7808df22020-11-02 18:00:32 -0800709 max_time = clib_clamp (max_time, 1 /* 1us */ , 1000 /* 1ms */ );
710 pacer->max_burst = (rate_bytes_per_sec * max_time) * CLIB_US_TIME_PERIOD;
711 pacer->max_burst = clib_clamp (pacer->max_burst, TRANSPORT_PACER_MIN_BURST,
712 TRANSPORT_PACER_MAX_BURST);
Florin Corasd67f1122018-05-21 17:47:40 -0700713}
714
Florin Coras52814732019-06-12 15:38:19 -0700715static inline u64
716spacer_pace_rate (spacer_t * pacer)
717{
Florin Coras7c8f8282019-09-15 16:28:45 -0700718 return pacer->bytes_per_sec;
Florin Coras52814732019-06-12 15:38:19 -0700719}
720
Florin Coras36ebcff2019-09-12 18:36:44 -0700721static inline void
Florin Corasa8e71c82019-10-22 19:01:39 -0700722spacer_reset (spacer_t * pacer, clib_us_time_t time_now, u64 bucket)
Florin Coras36ebcff2019-09-12 18:36:44 -0700723{
Florin Corasa8e71c82019-10-22 19:01:39 -0700724 pacer->last_update = time_now;
725 pacer->bucket = bucket;
Florin Coras36ebcff2019-09-12 18:36:44 -0700726}
727
Florin Corasd67f1122018-05-21 17:47:40 -0700728void
Florin Corasc44a5582018-11-01 16:30:54 -0700729transport_connection_tx_pacer_reset (transport_connection_t * tc,
Florin Coras11e9e352019-11-13 19:09:47 -0800730 u64 rate_bytes_per_sec, u32 start_bucket,
731 clib_us_time_t rtt)
Florin Corasd67f1122018-05-21 17:47:40 -0700732{
Florin Corasa39f4722020-11-12 10:20:05 -0800733 spacer_set_pace_rate (&tc->pacer, rate_bytes_per_sec, rtt,
734 transport_seconds_per_loop (tc->thread_index));
Florin Corasa8e71c82019-10-22 19:01:39 -0700735 spacer_reset (&tc->pacer, transport_us_time_now (tc->thread_index),
736 start_bucket);
737}
738
739void
Florin Coras11e9e352019-11-13 19:09:47 -0800740transport_connection_tx_pacer_reset_bucket (transport_connection_t * tc,
741 u32 bucket)
Florin Corasa8e71c82019-10-22 19:01:39 -0700742{
Florin Coras11e9e352019-11-13 19:09:47 -0800743 spacer_reset (&tc->pacer, transport_us_time_now (tc->thread_index), bucket);
Florin Corasc44a5582018-11-01 16:30:54 -0700744}
745
746void
747transport_connection_tx_pacer_init (transport_connection_t * tc,
Florin Corasa8e71c82019-10-22 19:01:39 -0700748 u64 rate_bytes_per_sec,
Florin Corasc44a5582018-11-01 16:30:54 -0700749 u32 initial_bucket)
750{
Florin Corasc44a5582018-11-01 16:30:54 -0700751 tc->flags |= TRANSPORT_CONNECTION_F_IS_TX_PACED;
752 transport_connection_tx_pacer_reset (tc, rate_bytes_per_sec,
Florin Coras11e9e352019-11-13 19:09:47 -0800753 initial_bucket, 1e6);
Florin Corasd67f1122018-05-21 17:47:40 -0700754}
755
756void
757transport_connection_tx_pacer_update (transport_connection_t * tc,
Florin Coras11e9e352019-11-13 19:09:47 -0800758 u64 bytes_per_sec, clib_us_time_t rtt)
Florin Corasd67f1122018-05-21 17:47:40 -0700759{
Florin Corasa39f4722020-11-12 10:20:05 -0800760 spacer_set_pace_rate (&tc->pacer, bytes_per_sec, rtt,
761 transport_seconds_per_loop (tc->thread_index));
Florin Corasd67f1122018-05-21 17:47:40 -0700762}
763
764u32
Florin Corasa8e71c82019-10-22 19:01:39 -0700765transport_connection_tx_pacer_burst (transport_connection_t * tc)
Florin Corase55a6d72018-10-31 23:09:22 -0700766{
Florin Corasa8e71c82019-10-22 19:01:39 -0700767 return spacer_max_burst (&tc->pacer,
768 transport_us_time_now (tc->thread_index));
Florin Coras36ebcff2019-09-12 18:36:44 -0700769}
770
Florin Coras52814732019-06-12 15:38:19 -0700771u64
772transport_connection_tx_pacer_rate (transport_connection_t * tc)
773{
774 return spacer_pace_rate (&tc->pacer);
775}
776
Florin Corasd67f1122018-05-21 17:47:40 -0700777void
Florin Coras00482232019-08-02 12:52:00 -0700778transport_connection_update_tx_bytes (transport_connection_t * tc, u32 bytes)
Florin Corasd67f1122018-05-21 17:47:40 -0700779{
Florin Corasd67f1122018-05-21 17:47:40 -0700780 if (transport_connection_is_tx_paced (tc))
781 spacer_update_bucket (&tc->pacer, bytes);
782}
783
784void
Florin Corase55a6d72018-10-31 23:09:22 -0700785transport_connection_tx_pacer_update_bytes (transport_connection_t * tc,
786 u32 bytes)
787{
788 spacer_update_bucket (&tc->pacer, bytes);
789}
790
791void
Florin Coras8f10b902021-04-02 18:32:00 -0700792transport_update_pacer_time (u32 thread_index, clib_time_type_t now)
793{
794 session_wrk_update_time (session_main_get_worker (thread_index), now);
795}
796
797void
Florin Coras70f879d2020-03-13 17:54:42 +0000798transport_connection_reschedule (transport_connection_t * tc)
799{
800 tc->flags &= ~TRANSPORT_CONNECTION_F_DESCHED;
Florin Coras2b4f74f2022-01-09 19:03:09 -0800801 transport_connection_tx_pacer_reset_bucket (tc, 0 /* bucket */);
Florin Coras70f879d2020-03-13 17:54:42 +0000802 if (transport_max_tx_dequeue (tc))
803 sesssion_reschedule_tx (tc);
804 else
805 {
806 session_t *s = session_get (tc->s_index, tc->thread_index);
807 svm_fifo_unset_event (s->tx_fifo);
808 if (svm_fifo_max_dequeue_cons (s->tx_fifo))
809 if (svm_fifo_set_event (s->tx_fifo))
810 sesssion_reschedule_tx (tc);
811 }
812}
813
814void
Florin Coras8c4fa012020-11-06 16:59:08 -0800815transport_fifos_init_ooo (transport_connection_t * tc)
816{
817 session_t *s = session_get (tc->s_index, tc->thread_index);
818 svm_fifo_init_ooo_lookup (s->rx_fifo, 0 /* ooo enq */ );
819 svm_fifo_init_ooo_lookup (s->tx_fifo, 1 /* ooo deq */ );
820}
821
822void
Florin Corasa8e71c82019-10-22 19:01:39 -0700823transport_update_time (clib_time_type_t time_now, u8 thread_index)
Florin Coras561af9b2017-12-09 10:19:43 -0800824{
825 transport_proto_vft_t *vft;
826 vec_foreach (vft, tp_vfts)
827 {
828 if (vft->update_time)
829 (vft->update_time) (time_now, thread_index);
830 }
831}
832
833void
834transport_enable_disable (vlib_main_t * vm, u8 is_en)
835{
836 transport_proto_vft_t *vft;
837 vec_foreach (vft, tp_vfts)
838 {
839 if (vft->enable)
840 (vft->enable) (vm, is_en);
Florin Coras5384cca2022-01-20 18:10:26 -0800841
842 if (vft->update_time)
843 session_register_update_time_fn (vft->update_time, is_en);
Florin Coras561af9b2017-12-09 10:19:43 -0800844 }
845}
846
847void
Florin Coras3cbc04b2017-10-02 00:18:51 -0700848transport_init (void)
849{
850 vlib_thread_main_t *vtm = vlib_get_thread_main ();
Florin Coras31c99552019-03-01 13:00:58 -0800851 session_main_t *smm = vnet_get_session_main ();
Florin Coras3cbc04b2017-10-02 00:18:51 -0700852 u32 num_threads;
853
Florin Coras93e65802017-11-29 00:07:11 -0500854 if (smm->local_endpoints_table_buckets == 0)
855 smm->local_endpoints_table_buckets = 250000;
856 if (smm->local_endpoints_table_memory == 0)
857 smm->local_endpoints_table_memory = 512 << 20;
858
Florin Coras3cbc04b2017-10-02 00:18:51 -0700859 /* Initialize [port-allocator] random number seed */
860 port_allocator_seed = (u32) clib_cpu_time_now ();
861
862 clib_bihash_init_24_8 (&local_endpoints_table, "local endpoints table",
Florin Coras93e65802017-11-29 00:07:11 -0500863 smm->local_endpoints_table_buckets,
864 smm->local_endpoints_table_memory);
Florin Coras05ead782022-03-23 16:35:05 -0700865 clib_spinlock_init (&local_endpoints_lock);
866
Florin Coras3cbc04b2017-10-02 00:18:51 -0700867 num_threads = 1 /* main thread */ + vtm->n_threads;
868 if (num_threads > 1)
Florin Corasfb50bc32021-05-18 00:28:59 -0700869 {
Florin Corasfb50bc32021-05-18 00:28:59 -0700870 /* Main not polled if there are workers */
871 smm->transport_cl_thread = 1;
872 }
Florin Coras3cbc04b2017-10-02 00:18:51 -0700873}
874
875/*
876 * fd.io coding-style-patch-verification: ON
877 *
878 * Local Variables:
879 * eval: (c-set-style "gnu")
880 * End:
881 */