Line data Source code
1 : /*
2 : * Copyright (c) 2019 Cisco and/or its affiliates.
3 : * Licensed under the Apache License, Version 2.0 (the "License");
4 : * you may not use this file except in compliance with the License.
5 : * You may obtain a copy of the License at:
6 : *
7 : * http://www.apache.org/licenses/LICENSE-2.0
8 : *
9 : * Unless required by applicable law or agreed to in writing, software
10 : * distributed under the License is distributed on an "AS IS" BASIS,
11 : * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 : * See the License for the specific language governing permissions and
13 : * limitations under the License.
14 : */
15 :
16 : #include <vnet/session/application_local.h>
17 : #include <vnet/session/session.h>
18 :
19 : typedef enum ct_segment_flags_
20 : {
21 : CT_SEGMENT_F_CLIENT_DETACHED = 1 << 0,
22 : CT_SEGMENT_F_SERVER_DETACHED = 1 << 1,
23 : } ct_segment_flags_t;
24 :
25 : typedef struct ct_segment_
26 : {
27 : u32 client_n_sessions;
28 : u32 server_n_sessions;
29 : u32 seg_ctx_index;
30 : u32 ct_seg_index;
31 : u32 segment_index;
32 : ct_segment_flags_t flags;
33 : } ct_segment_t;
34 :
35 : typedef struct ct_segments_
36 : {
37 : u32 sm_index;
38 : u32 server_wrk;
39 : u32 client_wrk;
40 : u32 fifo_pair_bytes;
41 : ct_segment_t *segments;
42 : } ct_segments_ctx_t;
43 :
44 : typedef struct ct_cleanup_req_
45 : {
46 : u32 ct_index;
47 : } ct_cleanup_req_t;
48 :
49 : typedef struct ct_worker_
50 : {
51 : ct_connection_t *connections; /**< Per-worker connection pools */
52 : u32 *pending_connects; /**< Fifo of pending ho indices */
53 : ct_cleanup_req_t *pending_cleanups; /**< Fifo of pending indices */
54 : u8 have_connects; /**< Set if connect rpc pending */
55 : u8 have_cleanups; /**< Set if cleanup rpc pending */
56 : clib_spinlock_t pending_connects_lock; /**< Lock for pending connects */
57 : u32 *new_connects; /**< Burst of connects to be done */
58 : } ct_worker_t;
59 :
60 : typedef struct ct_main_
61 : {
62 : ct_worker_t *wrk; /**< Per-worker state */
63 : u32 n_workers; /**< Number of vpp workers */
64 : u32 n_sessions; /**< Cumulative sessions counter */
65 : u32 *ho_reusable; /**< Vector of reusable ho indices */
66 : clib_spinlock_t ho_reuseable_lock; /**< Lock for reusable ho indices */
67 : clib_rwlock_t app_segs_lock; /**< RW lock for seg contexts */
68 : uword *app_segs_ctxs_table; /**< App handle to segment pool map */
69 : ct_segments_ctx_t *app_seg_ctxs; /**< Pool of ct segment contexts */
70 : u32 **fwrk_pending_connects; /**< First wrk pending half-opens */
71 : u32 fwrk_thread; /**< First worker thread */
72 : u8 fwrk_have_flush; /**< Flag for connect flush rpc */
73 : } ct_main_t;
74 :
75 : static ct_main_t ct_main;
76 :
77 : static inline ct_worker_t *
78 717065 : ct_worker_get (u32 thread_index)
79 : {
80 717065 : return &ct_main.wrk[thread_index];
81 : }
82 :
83 : static ct_connection_t *
84 57 : ct_connection_alloc (u32 thread_index)
85 : {
86 57 : ct_worker_t *wrk = ct_worker_get (thread_index);
87 : ct_connection_t *ct;
88 :
89 57 : pool_get_aligned_safe (wrk->connections, ct, CLIB_CACHE_LINE_BYTES);
90 57 : clib_memset (ct, 0, sizeof (*ct));
91 57 : ct->c_c_index = ct - wrk->connections;
92 57 : ct->c_thread_index = thread_index;
93 57 : ct->client_wrk = ~0;
94 57 : ct->server_wrk = ~0;
95 57 : ct->seg_ctx_index = ~0;
96 57 : ct->ct_seg_index = ~0;
97 57 : return ct;
98 : }
99 :
100 : static ct_connection_t *
101 716893 : ct_connection_get (u32 ct_index, u32 thread_index)
102 : {
103 716893 : ct_worker_t *wrk = ct_worker_get (thread_index);
104 :
105 716893 : if (pool_is_free_index (wrk->connections, ct_index))
106 30 : return 0;
107 716863 : return pool_elt_at_index (wrk->connections, ct_index);
108 : }
109 :
110 : static void
111 44 : ct_connection_free (ct_connection_t * ct)
112 : {
113 44 : ct_worker_t *wrk = ct_worker_get (ct->c_thread_index);
114 :
115 : if (CLIB_DEBUG)
116 : {
117 44 : clib_memset (ct, 0xfc, sizeof (*ct));
118 44 : pool_put (wrk->connections, ct);
119 44 : return;
120 : }
121 : pool_put (wrk->connections, ct);
122 : }
123 :
124 : static ct_connection_t *
125 13 : ct_half_open_alloc (void)
126 : {
127 13 : ct_main_t *cm = &ct_main;
128 : u32 *hip;
129 :
130 13 : clib_spinlock_lock (&cm->ho_reuseable_lock);
131 21 : vec_foreach (hip, cm->ho_reusable)
132 8 : pool_put_index (cm->wrk[cm->fwrk_thread].connections, *hip);
133 13 : vec_reset_length (cm->ho_reusable);
134 13 : clib_spinlock_unlock (&cm->ho_reuseable_lock);
135 :
136 13 : return ct_connection_alloc (cm->fwrk_thread);
137 : }
138 :
139 : static ct_connection_t *
140 26 : ct_half_open_get (u32 ho_index)
141 : {
142 26 : ct_main_t *cm = &ct_main;
143 26 : return ct_connection_get (ho_index, cm->fwrk_thread);
144 : }
145 :
146 : void
147 13 : ct_half_open_add_reusable (u32 ho_index)
148 : {
149 13 : ct_main_t *cm = &ct_main;
150 :
151 13 : clib_spinlock_lock (&cm->ho_reuseable_lock);
152 13 : vec_add1 (cm->ho_reusable, ho_index);
153 13 : clib_spinlock_unlock (&cm->ho_reuseable_lock);
154 13 : }
155 :
156 : session_t *
157 12 : ct_session_get_peer (session_t * s)
158 : {
159 : ct_connection_t *ct, *peer_ct;
160 12 : ct = ct_connection_get (s->connection_index, s->thread_index);
161 12 : peer_ct = ct_connection_get (ct->peer_index, s->thread_index);
162 12 : return session_get (peer_ct->c_s_index, s->thread_index);
163 : }
164 :
165 : void
166 18 : ct_session_endpoint (session_t * ll, session_endpoint_t * sep)
167 : {
168 : ct_connection_t *ct;
169 18 : ct = (ct_connection_t *) session_get_transport (ll);
170 18 : sep->transport_proto = ct->actual_tp;
171 18 : sep->port = ct->c_lcl_port;
172 18 : sep->is_ip4 = ct->c_is_ip4;
173 18 : ip_copy (&sep->ip, &ct->c_lcl_ip, ct->c_is_ip4);
174 18 : }
175 :
176 : static void
177 4 : ct_set_invalid_app_wrk (ct_connection_t *ct, u8 is_client)
178 : {
179 : ct_connection_t *peer_ct;
180 :
181 4 : peer_ct = ct_connection_get (ct->peer_index, ct->c_thread_index);
182 :
183 4 : if (is_client)
184 : {
185 4 : ct->client_wrk = APP_INVALID_INDEX;
186 4 : if (peer_ct)
187 0 : ct->client_wrk = APP_INVALID_INDEX;
188 : }
189 : else
190 : {
191 0 : ct->server_wrk = APP_INVALID_INDEX;
192 0 : if (peer_ct)
193 0 : ct->server_wrk = APP_INVALID_INDEX;
194 : }
195 4 : }
196 :
197 : static void
198 24 : ct_session_dealloc_fifos (ct_connection_t *ct, svm_fifo_t *rx_fifo,
199 : svm_fifo_t *tx_fifo)
200 : {
201 : ct_segments_ctx_t *seg_ctx;
202 24 : ct_main_t *cm = &ct_main;
203 : segment_manager_t *sm;
204 : app_worker_t *app_wrk;
205 : ct_segment_t *ct_seg;
206 : fifo_segment_t *fs;
207 : u32 seg_index;
208 : session_t *s;
209 : int cnt;
210 :
211 : /*
212 : * Cleanup fifos
213 : */
214 :
215 24 : sm = segment_manager_get (rx_fifo->segment_manager);
216 24 : seg_index = rx_fifo->segment_index;
217 :
218 24 : fs = segment_manager_get_segment_w_lock (sm, seg_index);
219 24 : fifo_segment_free_fifo (fs, rx_fifo);
220 24 : fifo_segment_free_fifo (fs, tx_fifo);
221 24 : segment_manager_segment_reader_unlock (sm);
222 :
223 : /*
224 : * Atomically update segment context with readers lock
225 : */
226 :
227 24 : clib_rwlock_reader_lock (&cm->app_segs_lock);
228 :
229 24 : seg_ctx = pool_elt_at_index (cm->app_seg_ctxs, ct->seg_ctx_index);
230 24 : ct_seg = pool_elt_at_index (seg_ctx->segments, ct->ct_seg_index);
231 :
232 24 : if (ct->flags & CT_CONN_F_CLIENT)
233 : {
234 12 : cnt =
235 12 : __atomic_sub_fetch (&ct_seg->client_n_sessions, 1, __ATOMIC_RELAXED);
236 : }
237 : else
238 : {
239 12 : cnt =
240 12 : __atomic_sub_fetch (&ct_seg->server_n_sessions, 1, __ATOMIC_RELAXED);
241 : }
242 :
243 24 : clib_rwlock_reader_unlock (&cm->app_segs_lock);
244 :
245 : /*
246 : * No need to do any app updates, return
247 : */
248 24 : ASSERT (cnt >= 0);
249 24 : if (cnt)
250 8 : return;
251 :
252 : /*
253 : * Grab exclusive lock and update flags unless some other thread
254 : * added more sessions
255 : */
256 16 : clib_rwlock_writer_lock (&cm->app_segs_lock);
257 :
258 16 : seg_ctx = pool_elt_at_index (cm->app_seg_ctxs, ct->seg_ctx_index);
259 16 : ct_seg = pool_elt_at_index (seg_ctx->segments, ct->ct_seg_index);
260 16 : if (ct->flags & CT_CONN_F_CLIENT)
261 : {
262 8 : cnt = ct_seg->client_n_sessions;
263 8 : if (cnt)
264 0 : goto done;
265 8 : ct_seg->flags |= CT_SEGMENT_F_CLIENT_DETACHED;
266 8 : s = session_get (ct->c_s_index, ct->c_thread_index);
267 8 : if (s->app_wrk_index == APP_INVALID_INDEX)
268 4 : ct_set_invalid_app_wrk (ct, 1 /* is_client */);
269 : }
270 : else
271 : {
272 8 : cnt = ct_seg->server_n_sessions;
273 8 : if (cnt)
274 0 : goto done;
275 8 : ct_seg->flags |= CT_SEGMENT_F_SERVER_DETACHED;
276 8 : s = session_get (ct->c_s_index, ct->c_thread_index);
277 8 : if (s->app_wrk_index == APP_INVALID_INDEX)
278 0 : ct_set_invalid_app_wrk (ct, 0 /* is_client */);
279 : }
280 :
281 16 : if (!(ct_seg->flags & CT_SEGMENT_F_CLIENT_DETACHED) ||
282 12 : !(ct_seg->flags & CT_SEGMENT_F_SERVER_DETACHED))
283 8 : goto done;
284 :
285 : /*
286 : * Remove segment context because both client and server detached
287 : */
288 :
289 8 : pool_put_index (seg_ctx->segments, ct->ct_seg_index);
290 :
291 : /*
292 : * No more segment indices left, remove the segments context
293 : */
294 8 : if (!pool_elts (seg_ctx->segments))
295 : {
296 8 : u64 table_handle = seg_ctx->client_wrk << 16 | seg_ctx->server_wrk;
297 8 : table_handle = (u64) seg_ctx->sm_index << 32 | table_handle;
298 8 : hash_unset (cm->app_segs_ctxs_table, table_handle);
299 8 : pool_free (seg_ctx->segments);
300 8 : pool_put_index (cm->app_seg_ctxs, ct->seg_ctx_index);
301 : }
302 :
303 : /*
304 : * Segment to be removed so notify both apps
305 : */
306 :
307 8 : app_wrk = app_worker_get_if_valid (ct->client_wrk);
308 : /* Determine if client app still needs notification, i.e., if it is
309 : * still attached. If client detached and this is the last ct session
310 : * on this segment, then its connects segment manager should also be
311 : * detached, so do not send notification */
312 8 : if (app_wrk)
313 : {
314 : segment_manager_t *csm;
315 4 : csm = app_worker_get_connect_segment_manager (app_wrk);
316 4 : if (!segment_manager_app_detached (csm))
317 4 : app_worker_del_segment_notify (app_wrk, ct->segment_handle);
318 : }
319 :
320 : /* Notify server app and free segment */
321 8 : segment_manager_lock_and_del_segment (sm, seg_index);
322 :
323 : /* Cleanup segment manager if needed. If server detaches there's a chance
324 : * the client's sessions will hold up segment removal */
325 8 : if (segment_manager_app_detached (sm) && !segment_manager_has_fifos (sm))
326 6 : segment_manager_free_safe (sm);
327 :
328 2 : done:
329 :
330 16 : clib_rwlock_writer_unlock (&cm->app_segs_lock);
331 : }
332 :
333 : static void
334 0 : ct_session_force_disconnect_server (ct_connection_t *sct)
335 : {
336 0 : sct->peer_index = ~0;
337 0 : session_transport_closing_notify (&sct->connection);
338 0 : }
339 :
340 : int
341 12 : ct_session_connect_notify (session_t *ss, session_error_t err)
342 : {
343 : u32 ss_index, opaque, thread_index;
344 : ct_connection_t *sct, *cct;
345 : app_worker_t *client_wrk;
346 : session_t *cs;
347 :
348 12 : ss_index = ss->session_index;
349 12 : thread_index = ss->thread_index;
350 12 : sct = (ct_connection_t *) session_get_transport (ss);
351 12 : client_wrk = app_worker_get (sct->client_wrk);
352 12 : opaque = sct->client_opaque;
353 :
354 12 : cct = ct_connection_get (sct->peer_index, thread_index);
355 :
356 : /* Client closed while waiting for reply from server */
357 12 : if (PREDICT_FALSE (!cct))
358 : {
359 0 : ct_session_force_disconnect_server (sct);
360 0 : return 0;
361 : }
362 :
363 12 : session_half_open_delete_notify (&cct->connection);
364 12 : cct->flags &= ~CT_CONN_F_HALF_OPEN;
365 :
366 12 : if (PREDICT_FALSE (err))
367 0 : goto connect_error;
368 :
369 : /*
370 : * Alloc client session, server session assumed to be established
371 : */
372 :
373 12 : ASSERT (ss->session_state >= SESSION_STATE_READY);
374 :
375 12 : cs = session_alloc (thread_index);
376 12 : ss = session_get (ss_index, thread_index);
377 12 : cs->session_type = ss->session_type;
378 12 : cs->listener_handle = SESSION_INVALID_HANDLE;
379 12 : session_set_state (cs, SESSION_STATE_CONNECTING);
380 12 : cs->app_wrk_index = client_wrk->wrk_index;
381 12 : cs->connection_index = cct->c_c_index;
382 12 : cs->opaque = opaque;
383 12 : cct->c_s_index = cs->session_index;
384 :
385 : /* This will allocate fifos for the session. They won't be used for
386 : * exchanging data but they will be used to close the connection if
387 : * the segment manager/worker is freed */
388 12 : if ((err = app_worker_init_connected (client_wrk, cs)))
389 : {
390 0 : session_free (cs);
391 0 : ct_session_force_disconnect_server (sct);
392 0 : err = SESSION_E_ALLOC;
393 0 : goto connect_error;
394 : }
395 :
396 12 : session_set_state (cs, SESSION_STATE_CONNECTING);
397 :
398 12 : if (app_worker_connect_notify (client_wrk, cs, 0, opaque))
399 : {
400 0 : segment_manager_dealloc_fifos (cs->rx_fifo, cs->tx_fifo);
401 0 : session_free (cs);
402 0 : ct_session_force_disconnect_server (sct);
403 0 : goto cleanup_client;
404 : }
405 :
406 12 : cs = session_get (cct->c_s_index, cct->c_thread_index);
407 12 : session_set_state (cs, SESSION_STATE_READY);
408 :
409 12 : return 0;
410 :
411 0 : connect_error:
412 :
413 0 : app_worker_connect_notify (client_wrk, 0, err, cct->client_opaque);
414 :
415 0 : cleanup_client:
416 :
417 0 : if (cct->client_rx_fifo)
418 0 : ct_session_dealloc_fifos (cct, cct->client_rx_fifo, cct->client_tx_fifo);
419 0 : ct_connection_free (cct);
420 0 : return -1;
421 : }
422 :
423 : static inline ct_segment_t *
424 4 : ct_lookup_free_segment (ct_main_t *cm, segment_manager_t *sm,
425 : u32 seg_ctx_index)
426 : {
427 : uword free_bytes, max_free_bytes;
428 4 : ct_segment_t *ct_seg, *res = 0;
429 : ct_segments_ctx_t *seg_ctx;
430 : fifo_segment_t *fs;
431 : u32 max_fifos;
432 :
433 4 : seg_ctx = pool_elt_at_index (cm->app_seg_ctxs, seg_ctx_index);
434 4 : max_free_bytes = seg_ctx->fifo_pair_bytes;
435 :
436 8 : pool_foreach (ct_seg, seg_ctx->segments)
437 : {
438 : /* Client or server has detached so segment cannot be used */
439 4 : fs = segment_manager_get_segment (sm, ct_seg->segment_index);
440 4 : free_bytes = fifo_segment_available_bytes (fs);
441 4 : max_fifos = fifo_segment_size (fs) / seg_ctx->fifo_pair_bytes;
442 4 : if (free_bytes > max_free_bytes &&
443 4 : fifo_segment_num_fifos (fs) / 2 < max_fifos)
444 : {
445 4 : max_free_bytes = free_bytes;
446 4 : res = ct_seg;
447 : }
448 : }
449 :
450 4 : return res;
451 : }
452 :
453 : static ct_segment_t *
454 9 : ct_alloc_segment (ct_main_t *cm, app_worker_t *server_wrk, u64 table_handle,
455 : segment_manager_t *sm, u32 client_wrk_index)
456 : {
457 9 : u32 seg_ctx_index = ~0, sm_index, pair_bytes;
458 : segment_manager_props_t *props;
459 9 : const u32 margin = 16 << 10;
460 : ct_segments_ctx_t *seg_ctx;
461 : app_worker_t *client_wrk;
462 : u64 seg_size, seg_handle;
463 : application_t *server;
464 : ct_segment_t *ct_seg;
465 : uword *spp;
466 : int fs_index;
467 :
468 9 : server = application_get (server_wrk->app_index);
469 9 : props = application_segment_manager_properties (server);
470 9 : sm_index = segment_manager_index (sm);
471 9 : pair_bytes = props->rx_fifo_size + props->tx_fifo_size + margin;
472 :
473 : /*
474 : * Make sure another thread did not alloc a segment while acquiring the lock
475 : */
476 :
477 9 : spp = hash_get (cm->app_segs_ctxs_table, table_handle);
478 9 : if (spp)
479 : {
480 0 : seg_ctx_index = *spp;
481 0 : ct_seg = ct_lookup_free_segment (cm, sm, seg_ctx_index);
482 0 : if (ct_seg)
483 0 : return ct_seg;
484 : }
485 :
486 : /*
487 : * No segment, try to alloc one and notify the server and the client.
488 : * Make sure the segment is not used for other fifos
489 : */
490 9 : seg_size = clib_max (props->segment_size, 128 << 20);
491 : fs_index =
492 9 : segment_manager_add_segment2 (sm, seg_size, FIFO_SEGMENT_F_CUSTOM_USE);
493 9 : if (fs_index < 0)
494 0 : return 0;
495 :
496 9 : if (seg_ctx_index == ~0)
497 : {
498 9 : pool_get_zero (cm->app_seg_ctxs, seg_ctx);
499 9 : seg_ctx_index = seg_ctx - cm->app_seg_ctxs;
500 9 : hash_set (cm->app_segs_ctxs_table, table_handle, seg_ctx_index);
501 9 : seg_ctx->server_wrk = server_wrk->wrk_index;
502 9 : seg_ctx->client_wrk = client_wrk_index;
503 9 : seg_ctx->sm_index = sm_index;
504 9 : seg_ctx->fifo_pair_bytes = pair_bytes;
505 : }
506 : else
507 : {
508 0 : seg_ctx = pool_elt_at_index (cm->app_seg_ctxs, seg_ctx_index);
509 : }
510 :
511 9 : pool_get_zero (seg_ctx->segments, ct_seg);
512 9 : ct_seg->segment_index = fs_index;
513 9 : ct_seg->server_n_sessions = 0;
514 9 : ct_seg->client_n_sessions = 0;
515 9 : ct_seg->ct_seg_index = ct_seg - seg_ctx->segments;
516 9 : ct_seg->seg_ctx_index = seg_ctx_index;
517 :
518 : /* New segment, notify the server and client */
519 9 : seg_handle = segment_manager_make_segment_handle (sm_index, fs_index);
520 9 : if (app_worker_add_segment_notify (server_wrk, seg_handle))
521 0 : goto error;
522 :
523 9 : client_wrk = app_worker_get (client_wrk_index);
524 9 : if (app_worker_add_segment_notify (client_wrk, seg_handle))
525 : {
526 0 : app_worker_del_segment_notify (server_wrk, seg_handle);
527 0 : goto error;
528 : }
529 :
530 9 : return ct_seg;
531 :
532 0 : error:
533 :
534 0 : segment_manager_lock_and_del_segment (sm, fs_index);
535 0 : pool_put_index (seg_ctx->segments, ct_seg->seg_ctx_index);
536 0 : return 0;
537 : }
538 :
539 : static int
540 13 : ct_init_accepted_session (app_worker_t *server_wrk, ct_connection_t *ct,
541 : session_t *ls, session_t *ll)
542 : {
543 : segment_manager_props_t *props;
544 : u64 seg_handle, table_handle;
545 13 : u32 sm_index, fs_index = ~0;
546 : ct_segments_ctx_t *seg_ctx;
547 13 : ct_main_t *cm = &ct_main;
548 : application_t *server;
549 : segment_manager_t *sm;
550 : ct_segment_t *ct_seg;
551 : fifo_segment_t *fs;
552 : uword *spp;
553 : int rv;
554 :
555 13 : sm = app_worker_get_listen_segment_manager (server_wrk, ll);
556 13 : sm_index = segment_manager_index (sm);
557 13 : server = application_get (server_wrk->app_index);
558 13 : props = application_segment_manager_properties (server);
559 :
560 13 : table_handle = ct->client_wrk << 16 | server_wrk->wrk_index;
561 13 : table_handle = (u64) sm_index << 32 | table_handle;
562 :
563 : /*
564 : * Check if we already have a segment that can hold the fifos
565 : */
566 :
567 13 : clib_rwlock_reader_lock (&cm->app_segs_lock);
568 :
569 13 : spp = hash_get (cm->app_segs_ctxs_table, table_handle);
570 13 : if (spp)
571 : {
572 4 : ct_seg = ct_lookup_free_segment (cm, sm, *spp);
573 4 : if (ct_seg)
574 : {
575 4 : ct->seg_ctx_index = ct_seg->seg_ctx_index;
576 4 : ct->ct_seg_index = ct_seg->ct_seg_index;
577 4 : fs_index = ct_seg->segment_index;
578 4 : ct_seg->flags &=
579 : ~(CT_SEGMENT_F_SERVER_DETACHED | CT_SEGMENT_F_CLIENT_DETACHED);
580 4 : __atomic_add_fetch (&ct_seg->server_n_sessions, 1, __ATOMIC_RELAXED);
581 4 : __atomic_add_fetch (&ct_seg->client_n_sessions, 1, __ATOMIC_RELAXED);
582 : }
583 : }
584 :
585 13 : clib_rwlock_reader_unlock (&cm->app_segs_lock);
586 :
587 : /*
588 : * If not, grab exclusive lock and allocate segment
589 : */
590 13 : if (fs_index == ~0)
591 : {
592 9 : clib_rwlock_writer_lock (&cm->app_segs_lock);
593 :
594 : ct_seg =
595 9 : ct_alloc_segment (cm, server_wrk, table_handle, sm, ct->client_wrk);
596 9 : if (!ct_seg)
597 : {
598 0 : clib_rwlock_writer_unlock (&cm->app_segs_lock);
599 0 : return -1;
600 : }
601 :
602 9 : ct->seg_ctx_index = ct_seg->seg_ctx_index;
603 9 : ct->ct_seg_index = ct_seg->ct_seg_index;
604 9 : ct_seg->server_n_sessions += 1;
605 9 : ct_seg->client_n_sessions += 1;
606 9 : fs_index = ct_seg->segment_index;
607 :
608 9 : clib_rwlock_writer_unlock (&cm->app_segs_lock);
609 : }
610 :
611 : /*
612 : * Allocate and initialize the fifos
613 : */
614 13 : fs = segment_manager_get_segment_w_lock (sm, fs_index);
615 13 : rv = segment_manager_try_alloc_fifos (
616 13 : fs, ls->thread_index, props->rx_fifo_size, props->tx_fifo_size,
617 : &ls->rx_fifo, &ls->tx_fifo);
618 13 : if (rv)
619 : {
620 0 : segment_manager_segment_reader_unlock (sm);
621 :
622 0 : clib_rwlock_reader_lock (&cm->app_segs_lock);
623 :
624 0 : seg_ctx = pool_elt_at_index (cm->app_seg_ctxs, ct->seg_ctx_index);
625 0 : ct_seg = pool_elt_at_index (seg_ctx->segments, ct->ct_seg_index);
626 0 : __atomic_sub_fetch (&ct_seg->server_n_sessions, 1, __ATOMIC_RELAXED);
627 0 : __atomic_sub_fetch (&ct_seg->client_n_sessions, 1, __ATOMIC_RELAXED);
628 :
629 0 : clib_rwlock_reader_unlock (&cm->app_segs_lock);
630 :
631 0 : return rv;
632 : }
633 :
634 13 : ls->rx_fifo->shr->master_session_index = ls->session_index;
635 13 : ls->tx_fifo->shr->master_session_index = ls->session_index;
636 13 : ls->rx_fifo->master_thread_index = ls->thread_index;
637 13 : ls->tx_fifo->master_thread_index = ls->thread_index;
638 :
639 13 : seg_handle = segment_manager_segment_handle (sm, fs);
640 13 : segment_manager_segment_reader_unlock (sm);
641 :
642 13 : ct->segment_handle = seg_handle;
643 :
644 13 : return 0;
645 : }
646 :
647 : static void
648 13 : ct_accept_one (u32 thread_index, u32 ho_index)
649 : {
650 : ct_connection_t *sct, *cct, *ho;
651 : transport_connection_t *ll_ct;
652 : app_worker_t *server_wrk;
653 : u32 cct_index, ll_index;
654 : session_t *ss, *ll;
655 :
656 : /*
657 : * Alloc client ct and initialize from ho
658 : */
659 13 : cct = ct_connection_alloc (thread_index);
660 13 : cct_index = cct->c_c_index;
661 :
662 13 : ho = ct_half_open_get (ho_index);
663 :
664 : /* Unlikely but half-open session and transport could have been freed */
665 13 : if (PREDICT_FALSE (!ho))
666 : {
667 0 : ct_connection_free (cct);
668 0 : return;
669 : }
670 :
671 13 : clib_memcpy (cct, ho, sizeof (*ho));
672 13 : cct->c_c_index = cct_index;
673 13 : cct->c_thread_index = thread_index;
674 13 : cct->flags |= CT_CONN_F_HALF_OPEN;
675 :
676 : /* Notify session layer that half-open is on a different thread
677 : * and mark ho connection index reusable. Avoids another rpc
678 : */
679 13 : session_half_open_migrate_notify (&cct->connection);
680 13 : session_half_open_migrated_notify (&cct->connection);
681 13 : ct_half_open_add_reusable (ho_index);
682 :
683 : /*
684 : * Alloc and init server transport
685 : */
686 :
687 13 : ll_index = cct->peer_index;
688 13 : ll = listen_session_get (ll_index);
689 13 : sct = ct_connection_alloc (thread_index);
690 : /* Transport not necessarily ct but it might, so grab after sct alloc */
691 13 : ll_ct = listen_session_get_transport (ll);
692 :
693 : /* Make sure cct is valid after sct alloc */
694 13 : cct = ct_connection_get (cct_index, thread_index);
695 :
696 13 : sct->c_rmt_port = 0;
697 13 : sct->c_lcl_port = ll_ct->lcl_port;
698 13 : sct->c_is_ip4 = cct->c_is_ip4;
699 13 : clib_memcpy (&sct->c_lcl_ip, &cct->c_rmt_ip, sizeof (cct->c_rmt_ip));
700 13 : sct->client_wrk = cct->client_wrk;
701 13 : sct->c_proto = TRANSPORT_PROTO_NONE;
702 13 : sct->client_opaque = cct->client_opaque;
703 13 : sct->actual_tp = cct->actual_tp;
704 :
705 13 : sct->peer_index = cct->c_c_index;
706 13 : cct->peer_index = sct->c_c_index;
707 :
708 : /*
709 : * Accept server session. Client session is created only after
710 : * server confirms accept.
711 : */
712 13 : ss = session_alloc (thread_index);
713 13 : ll = listen_session_get (ll_index);
714 26 : ss->session_type = session_type_from_proto_and_ip (TRANSPORT_PROTO_NONE,
715 13 : sct->c_is_ip4);
716 13 : ss->connection_index = sct->c_c_index;
717 13 : ss->listener_handle = listen_session_get_handle (ll);
718 13 : session_set_state (ss, SESSION_STATE_CREATED);
719 :
720 13 : server_wrk = application_listener_select_worker (ll);
721 13 : ss->app_wrk_index = server_wrk->wrk_index;
722 :
723 13 : sct->c_s_index = ss->session_index;
724 13 : sct->server_wrk = ss->app_wrk_index;
725 :
726 13 : if (ct_init_accepted_session (server_wrk, sct, ss, ll))
727 : {
728 0 : ct_session_connect_notify (ss, SESSION_E_ALLOC);
729 0 : ct_connection_free (sct);
730 0 : session_free (ss);
731 0 : return;
732 : }
733 :
734 13 : cct->server_wrk = sct->server_wrk;
735 13 : cct->seg_ctx_index = sct->seg_ctx_index;
736 13 : cct->ct_seg_index = sct->ct_seg_index;
737 13 : cct->client_rx_fifo = ss->tx_fifo;
738 13 : cct->client_tx_fifo = ss->rx_fifo;
739 13 : cct->client_rx_fifo->refcnt++;
740 13 : cct->client_tx_fifo->refcnt++;
741 13 : cct->segment_handle = sct->segment_handle;
742 :
743 13 : session_set_state (ss, SESSION_STATE_ACCEPTING);
744 13 : if (app_worker_accept_notify (server_wrk, ss))
745 : {
746 0 : ct_session_connect_notify (ss, SESSION_E_REFUSED);
747 0 : ct_session_dealloc_fifos (sct, ss->rx_fifo, ss->tx_fifo);
748 0 : ct_connection_free (sct);
749 0 : session_free (ss);
750 : }
751 : }
752 :
753 : static void
754 13 : ct_accept_rpc_wrk_handler (void *rpc_args)
755 : {
756 : u32 thread_index, n_connects, i, n_pending;
757 13 : const u32 max_connects = 32;
758 : ct_worker_t *wrk;
759 13 : u8 need_rpc = 0;
760 :
761 13 : thread_index = pointer_to_uword (rpc_args);
762 13 : wrk = ct_worker_get (thread_index);
763 :
764 : /* Connects could be handled without worker barrier so grab lock */
765 13 : clib_spinlock_lock (&wrk->pending_connects_lock);
766 :
767 13 : n_pending = clib_fifo_elts (wrk->pending_connects);
768 13 : n_connects = clib_min (n_pending, max_connects);
769 13 : vec_validate (wrk->new_connects, n_connects);
770 :
771 26 : for (i = 0; i < n_connects; i++)
772 13 : clib_fifo_sub1 (wrk->pending_connects, wrk->new_connects[i]);
773 :
774 13 : if (n_pending == n_connects)
775 13 : wrk->have_connects = 0;
776 : else
777 0 : need_rpc = 1;
778 :
779 13 : clib_spinlock_unlock (&wrk->pending_connects_lock);
780 :
781 26 : for (i = 0; i < n_connects; i++)
782 13 : ct_accept_one (thread_index, wrk->new_connects[i]);
783 :
784 13 : if (need_rpc)
785 0 : session_send_rpc_evt_to_thread_force (
786 : thread_index, ct_accept_rpc_wrk_handler,
787 0 : uword_to_pointer (thread_index, void *));
788 13 : }
789 :
790 : static void
791 13 : ct_fwrk_flush_connects (void *rpc_args)
792 : {
793 : u32 thread_index, fwrk_index, n_workers;
794 13 : ct_main_t *cm = &ct_main;
795 : ct_worker_t *wrk;
796 : u8 need_rpc;
797 :
798 13 : fwrk_index = cm->fwrk_thread;
799 13 : n_workers = vec_len (cm->fwrk_pending_connects);
800 :
801 26 : for (thread_index = fwrk_index; thread_index < n_workers; thread_index++)
802 : {
803 13 : if (!vec_len (cm->fwrk_pending_connects[thread_index]))
804 0 : continue;
805 :
806 13 : wrk = ct_worker_get (thread_index);
807 :
808 : /* Connects can be done without worker barrier, grab dst worker lock */
809 13 : if (thread_index != fwrk_index)
810 0 : clib_spinlock_lock (&wrk->pending_connects_lock);
811 :
812 13 : clib_fifo_add (wrk->pending_connects,
813 : cm->fwrk_pending_connects[thread_index],
814 : vec_len (cm->fwrk_pending_connects[thread_index]));
815 13 : if (!wrk->have_connects)
816 : {
817 13 : wrk->have_connects = 1;
818 13 : need_rpc = 1;
819 : }
820 :
821 13 : if (thread_index != fwrk_index)
822 0 : clib_spinlock_unlock (&wrk->pending_connects_lock);
823 :
824 13 : vec_reset_length (cm->fwrk_pending_connects[thread_index]);
825 :
826 13 : if (need_rpc)
827 13 : session_send_rpc_evt_to_thread_force (
828 : thread_index, ct_accept_rpc_wrk_handler,
829 13 : uword_to_pointer (thread_index, void *));
830 : }
831 :
832 13 : cm->fwrk_have_flush = 0;
833 13 : }
834 :
835 : static void
836 13 : ct_program_connect_to_wrk (u32 ho_index)
837 : {
838 13 : ct_main_t *cm = &ct_main;
839 : u32 thread_index;
840 :
841 : /* Simple round-robin policy for spreading sessions over workers. We skip
842 : * thread index 0, i.e., offset the index by 1, when we have workers as it
843 : * is the one dedicated to main thread. Note that n_workers does not include
844 : * main thread */
845 13 : cm->n_sessions += 1;
846 13 : thread_index = cm->n_workers ? (cm->n_sessions % cm->n_workers) + 1 : 0;
847 :
848 : /* Pospone flushing of connect request to dst worker until after session
849 : * layer fully initializes the half-open session. */
850 13 : vec_add1 (cm->fwrk_pending_connects[thread_index], ho_index);
851 13 : if (!cm->fwrk_have_flush)
852 : {
853 13 : session_send_rpc_evt_to_thread_force (
854 : cm->fwrk_thread, ct_fwrk_flush_connects,
855 13 : uword_to_pointer (thread_index, void *));
856 13 : cm->fwrk_have_flush = 1;
857 : }
858 13 : }
859 :
860 : static int
861 13 : ct_connect (app_worker_t *client_wrk, session_t *ll,
862 : session_endpoint_cfg_t *sep)
863 : {
864 : ct_connection_t *ho;
865 : u32 ho_index;
866 :
867 : /*
868 : * Alloc and init client half-open transport
869 : */
870 :
871 13 : ho = ct_half_open_alloc ();
872 13 : ho_index = ho->c_c_index;
873 13 : ho->c_rmt_port = sep->port;
874 13 : ho->c_lcl_port = 0;
875 13 : ho->c_is_ip4 = sep->is_ip4;
876 13 : ho->client_opaque = sep->opaque;
877 13 : ho->client_wrk = client_wrk->wrk_index;
878 13 : ho->peer_index = ll->session_index;
879 13 : ho->c_proto = TRANSPORT_PROTO_NONE;
880 13 : ho->c_flags |= TRANSPORT_CONNECTION_F_NO_LOOKUP;
881 13 : clib_memcpy (&ho->c_rmt_ip, &sep->ip, sizeof (sep->ip));
882 13 : ho->flags |= CT_CONN_F_CLIENT;
883 13 : ho->c_s_index = ~0;
884 13 : ho->actual_tp = sep->original_tp;
885 :
886 : /*
887 : * Program connect on a worker, connected reply comes
888 : * after server accepts the connection.
889 : */
890 13 : ct_program_connect_to_wrk (ho_index);
891 :
892 13 : return ho_index;
893 : }
894 :
895 : static u32
896 18 : ct_start_listen (u32 app_listener_index, transport_endpoint_cfg_t *tep)
897 : {
898 : session_endpoint_cfg_t *sep;
899 : ct_connection_t *ct;
900 :
901 18 : sep = (session_endpoint_cfg_t *) tep;
902 18 : ct = ct_connection_alloc (0);
903 18 : ct->server_wrk = sep->app_wrk_index;
904 18 : ct->c_is_ip4 = sep->is_ip4;
905 18 : clib_memcpy (&ct->c_lcl_ip, &sep->ip, sizeof (sep->ip));
906 18 : ct->c_lcl_port = sep->port;
907 18 : ct->c_s_index = app_listener_index;
908 18 : ct->actual_tp = sep->transport_proto;
909 18 : return ct->c_c_index;
910 : }
911 :
912 : static u32
913 18 : ct_stop_listen (u32 ct_index)
914 : {
915 : ct_connection_t *ct;
916 18 : ct = ct_connection_get (ct_index, 0);
917 18 : ct_connection_free (ct);
918 18 : return 0;
919 : }
920 :
921 : static transport_connection_t *
922 79 : ct_listener_get (u32 ct_index)
923 : {
924 79 : return (transport_connection_t *) ct_connection_get (ct_index, 0);
925 : }
926 :
927 : static transport_connection_t *
928 13 : ct_session_half_open_get (u32 ct_index)
929 : {
930 13 : return (transport_connection_t *) ct_half_open_get (ct_index);
931 : }
932 :
933 : static void
934 1 : ct_session_cleanup (u32 conn_index, u32 thread_index)
935 : {
936 : ct_connection_t *ct, *peer_ct;
937 :
938 1 : ct = ct_connection_get (conn_index, thread_index);
939 1 : if (!ct)
940 0 : return;
941 :
942 1 : peer_ct = ct_connection_get (ct->peer_index, thread_index);
943 1 : if (peer_ct)
944 1 : peer_ct->peer_index = ~0;
945 :
946 1 : ct_connection_free (ct);
947 : }
948 :
949 : static void
950 0 : ct_cleanup_ho (u32 ho_index)
951 : {
952 : ct_connection_t *ho;
953 :
954 0 : ho = ct_half_open_get (ho_index);
955 0 : ct_connection_free (ho);
956 0 : }
957 :
958 : static int
959 17 : ct_session_connect (transport_endpoint_cfg_t * tep)
960 : {
961 : session_endpoint_cfg_t *sep_ext;
962 17 : session_endpoint_t _sep, *sep = &_sep;
963 : app_worker_t *app_wrk;
964 : session_handle_t lh;
965 : application_t *app;
966 : app_listener_t *al;
967 : u32 table_index;
968 : session_t *ll;
969 : u8 fib_proto;
970 :
971 17 : sep_ext = (session_endpoint_cfg_t *) tep;
972 17 : _sep = *(session_endpoint_t *) tep;
973 17 : app_wrk = app_worker_get (sep_ext->app_wrk_index);
974 17 : app = application_get (app_wrk->app_index);
975 :
976 17 : sep->transport_proto = sep_ext->original_tp;
977 17 : table_index = application_local_session_table (app);
978 17 : lh = session_lookup_local_endpoint (table_index, sep);
979 17 : if (lh == SESSION_DROP_HANDLE)
980 3 : return SESSION_E_FILTERED;
981 :
982 14 : if (lh == SESSION_INVALID_HANDLE)
983 1 : goto global_scope;
984 :
985 13 : ll = listen_session_get_from_handle (lh);
986 13 : al = app_listener_get_w_session (ll);
987 :
988 : /*
989 : * Break loop if rule in local table points to connecting app. This
990 : * can happen if client is a generic proxy. Route connect through
991 : * global table instead.
992 : */
993 13 : if (al->app_index == app->app_index)
994 0 : goto global_scope;
995 :
996 13 : return ct_connect (app_wrk, ll, sep_ext);
997 :
998 : /*
999 : * If nothing found, check the global scope for locally attached
1000 : * destinations. Make sure first that we're allowed to.
1001 : */
1002 :
1003 1 : global_scope:
1004 1 : if (session_endpoint_is_local (sep))
1005 1 : return SESSION_E_NOROUTE;
1006 :
1007 0 : if (!application_has_global_scope (app))
1008 0 : return SESSION_E_SCOPE;
1009 :
1010 0 : fib_proto = session_endpoint_fib_proto (sep);
1011 0 : table_index = session_lookup_get_index_for_fib (fib_proto, sep->fib_index);
1012 0 : ll = session_lookup_listener_wildcard (table_index, sep);
1013 :
1014 : /* Avoid connecting app to own listener */
1015 0 : if (ll && ll->app_index != app->app_index)
1016 0 : return ct_connect (app_wrk, ll, sep_ext);
1017 :
1018 : /* Failed to connect but no error */
1019 0 : return SESSION_E_LOCAL_CONNECT;
1020 : }
1021 :
1022 : static inline int
1023 12 : ct_close_is_reset (ct_connection_t *ct, session_t *s)
1024 : {
1025 12 : if (ct->flags & CT_CONN_F_CLIENT)
1026 6 : return (svm_fifo_max_dequeue (ct->client_rx_fifo) > 0);
1027 : else
1028 6 : return (svm_fifo_max_dequeue (s->rx_fifo) > 0);
1029 : }
1030 :
1031 : static void
1032 12 : ct_session_cleanup_server_session (session_t *s)
1033 : {
1034 : ct_connection_t *ct;
1035 :
1036 12 : ct = (ct_connection_t *) session_get_transport (s);
1037 12 : ct_session_dealloc_fifos (ct, s->rx_fifo, s->tx_fifo);
1038 12 : session_free (s);
1039 12 : ct_connection_free (ct);
1040 12 : }
1041 :
1042 : static void
1043 25 : ct_session_postponed_cleanup (ct_connection_t *ct)
1044 : {
1045 : ct_connection_t *peer_ct;
1046 : app_worker_t *app_wrk;
1047 : session_t *s;
1048 :
1049 25 : s = session_get (ct->c_s_index, ct->c_thread_index);
1050 25 : app_wrk = app_worker_get_if_valid (s->app_wrk_index);
1051 :
1052 25 : peer_ct = ct_connection_get (ct->peer_index, ct->c_thread_index);
1053 25 : if (peer_ct)
1054 : {
1055 12 : if (ct_close_is_reset (ct, s))
1056 2 : session_transport_reset_notify (&peer_ct->connection);
1057 : else
1058 10 : session_transport_closing_notify (&peer_ct->connection);
1059 : }
1060 25 : session_transport_closed_notify (&ct->connection);
1061 :
1062 : /* It would be cleaner to call session_transport_delete_notify
1063 : * but then we can't control session cleanup lower */
1064 25 : session_set_state (s, SESSION_STATE_TRANSPORT_DELETED);
1065 25 : if (app_wrk)
1066 20 : app_worker_cleanup_notify (app_wrk, s, SESSION_CLEANUP_TRANSPORT);
1067 :
1068 25 : if (ct->flags & CT_CONN_F_CLIENT)
1069 : {
1070 : /* Normal free for client session as the fifos are allocated through
1071 : * the connects segment manager in a segment that's not shared with
1072 : * the server */
1073 12 : ct_session_dealloc_fifos (ct, ct->client_rx_fifo, ct->client_tx_fifo);
1074 12 : session_program_cleanup (s);
1075 12 : ct_connection_free (ct);
1076 : }
1077 : else
1078 : {
1079 : /* Manual session and fifo segment cleanup to avoid implicit
1080 : * segment manager cleanups and notifications */
1081 13 : if (app_wrk)
1082 : {
1083 : /* Remove custom cleanup notify infra when/if switching to normal
1084 : * session cleanup. Note that ct is freed in the cb function */
1085 12 : app_worker_cleanup_notify_custom (app_wrk, s,
1086 : SESSION_CLEANUP_SESSION,
1087 : ct_session_cleanup_server_session);
1088 : }
1089 : else
1090 : {
1091 1 : ct_connection_free (ct);
1092 : }
1093 : }
1094 25 : }
1095 :
1096 : static void
1097 20 : ct_handle_cleanups (void *args)
1098 : {
1099 20 : uword thread_index = pointer_to_uword (args);
1100 20 : const u32 max_cleanups = 100;
1101 : ct_cleanup_req_t *req;
1102 : ct_connection_t *ct;
1103 20 : u32 n_to_handle = 0;
1104 : ct_worker_t *wrk;
1105 : session_t *s;
1106 :
1107 20 : wrk = ct_worker_get (thread_index);
1108 20 : wrk->have_cleanups = 0;
1109 20 : n_to_handle = clib_fifo_elts (wrk->pending_cleanups);
1110 20 : n_to_handle = clib_min (n_to_handle, max_cleanups);
1111 :
1112 45 : while (n_to_handle)
1113 : {
1114 25 : clib_fifo_sub2 (wrk->pending_cleanups, req);
1115 25 : ct = ct_connection_get (req->ct_index, thread_index);
1116 25 : s = session_get (ct->c_s_index, ct->c_thread_index);
1117 25 : if (!svm_fifo_has_event (s->tx_fifo))
1118 25 : ct_session_postponed_cleanup (ct);
1119 : else
1120 0 : clib_fifo_add1 (wrk->pending_cleanups, *req);
1121 25 : n_to_handle -= 1;
1122 : }
1123 :
1124 20 : if (clib_fifo_elts (wrk->pending_cleanups))
1125 : {
1126 0 : wrk->have_cleanups = 1;
1127 0 : session_send_rpc_evt_to_thread_force (
1128 : thread_index, ct_handle_cleanups,
1129 : uword_to_pointer (thread_index, void *));
1130 : }
1131 20 : }
1132 :
1133 : static void
1134 25 : ct_program_cleanup (ct_connection_t *ct)
1135 : {
1136 : ct_cleanup_req_t *req;
1137 : uword thread_index;
1138 : ct_worker_t *wrk;
1139 :
1140 25 : thread_index = ct->c_thread_index;
1141 25 : wrk = ct_worker_get (ct->c_thread_index);
1142 :
1143 25 : clib_fifo_add2 (wrk->pending_cleanups, req);
1144 25 : req->ct_index = ct->c_c_index;
1145 :
1146 25 : if (wrk->have_cleanups)
1147 5 : return;
1148 :
1149 20 : wrk->have_cleanups = 1;
1150 20 : session_send_rpc_evt_to_thread_force (
1151 : thread_index, ct_handle_cleanups, uword_to_pointer (thread_index, void *));
1152 : }
1153 :
1154 : static void
1155 25 : ct_session_close (u32 ct_index, u32 thread_index)
1156 : {
1157 : ct_connection_t *ct, *peer_ct;
1158 : session_t *s;
1159 :
1160 25 : ct = ct_connection_get (ct_index, thread_index);
1161 25 : s = session_get (ct->c_s_index, ct->c_thread_index);
1162 25 : peer_ct = ct_connection_get (ct->peer_index, thread_index);
1163 25 : if (peer_ct)
1164 : {
1165 12 : peer_ct->peer_index = ~0;
1166 : /* Make sure session was allocated */
1167 12 : if (peer_ct->flags & CT_CONN_F_HALF_OPEN)
1168 : {
1169 0 : ct_session_connect_notify (s, SESSION_E_REFUSED);
1170 0 : ct->peer_index = ~0;
1171 : }
1172 12 : else if (peer_ct->c_s_index == ~0)
1173 : {
1174 : /* should not happen */
1175 0 : clib_warning ("ct peer without session");
1176 0 : ct_connection_free (peer_ct);
1177 : }
1178 : }
1179 :
1180 : /* Do not send closed notify to make sure pending tx events are
1181 : * still delivered and program cleanup */
1182 25 : ct_program_cleanup (ct);
1183 25 : }
1184 :
1185 : static transport_connection_t *
1186 574349 : ct_session_get (u32 ct_index, u32 thread_index)
1187 : {
1188 574349 : return (transport_connection_t *) ct_connection_get (ct_index,
1189 : thread_index);
1190 : }
1191 :
1192 : static u8 *
1193 0 : format_ct_connection_id (u8 * s, va_list * args)
1194 : {
1195 0 : ct_connection_t *ct = va_arg (*args, ct_connection_t *);
1196 0 : if (!ct)
1197 0 : return s;
1198 0 : if (ct->c_is_ip4)
1199 : {
1200 0 : s = format (s, "[%d:%d][CT:%U] %U:%d->%U:%d", ct->c_thread_index,
1201 0 : ct->c_s_index, format_transport_proto_short, ct->actual_tp,
1202 : format_ip4_address, &ct->c_lcl_ip4,
1203 0 : clib_net_to_host_u16 (ct->c_lcl_port), format_ip4_address,
1204 0 : &ct->c_rmt_ip4, clib_net_to_host_u16 (ct->c_rmt_port));
1205 : }
1206 : else
1207 : {
1208 0 : s = format (s, "[%d:%d][CT:%U] %U:%d->%U:%d", ct->c_thread_index,
1209 0 : ct->c_s_index, format_transport_proto_short, ct->actual_tp,
1210 : format_ip6_address, &ct->c_lcl_ip6,
1211 0 : clib_net_to_host_u16 (ct->c_lcl_port), format_ip6_address,
1212 0 : &ct->c_rmt_ip6, clib_net_to_host_u16 (ct->c_rmt_port));
1213 : }
1214 :
1215 0 : return s;
1216 : }
1217 :
1218 : static int
1219 137279 : ct_custom_tx (void *session, transport_send_params_t * sp)
1220 : {
1221 137279 : session_t *s = (session_t *) session;
1222 137279 : if (session_has_transport (s))
1223 0 : return 0;
1224 : /* If event enqueued towards peer, remove from scheduler and remove
1225 : * session tx flag, i.e., accept new tx events. Unset fifo flag now to
1226 : * avoid missing events if peer did not clear fifo flag yet, which is
1227 : * interpreted as successful notification and session is descheduled. */
1228 137279 : svm_fifo_unset_event (s->tx_fifo);
1229 137279 : if (!ct_session_tx (s))
1230 137279 : sp->flags = TRANSPORT_SND_F_DESCHED;
1231 :
1232 : /* The scheduler uses packet count as a means of upper bounding the amount
1233 : * of work done per dispatch. So make it look like we have sent something */
1234 137279 : return 1;
1235 : }
1236 :
1237 : static int
1238 4987 : ct_app_rx_evt (transport_connection_t * tc)
1239 : {
1240 4987 : ct_connection_t *ct = (ct_connection_t *) tc, *peer_ct;
1241 : session_t *ps, *s;
1242 :
1243 4987 : s = session_get (ct->c_s_index, ct->c_thread_index);
1244 4987 : if (session_has_transport (s) || s->session_state < SESSION_STATE_READY)
1245 0 : return -1;
1246 4987 : peer_ct = ct_connection_get (ct->peer_index, tc->thread_index);
1247 4987 : if (!peer_ct || (peer_ct->flags & CT_CONN_F_HALF_OPEN))
1248 0 : return -1;
1249 4987 : ps = session_get (peer_ct->c_s_index, peer_ct->c_thread_index);
1250 4987 : if (ps->session_state >= SESSION_STATE_TRANSPORT_CLOSING)
1251 0 : return -1;
1252 4987 : return session_dequeue_notify (ps);
1253 : }
1254 :
1255 : static u8 *
1256 0 : format_ct_listener (u8 * s, va_list * args)
1257 : {
1258 0 : u32 tc_index = va_arg (*args, u32);
1259 0 : u32 __clib_unused thread_index = va_arg (*args, u32);
1260 0 : u32 __clib_unused verbose = va_arg (*args, u32);
1261 0 : ct_connection_t *ct = ct_connection_get (tc_index, 0);
1262 0 : s = format (s, "%-" SESSION_CLI_ID_LEN "U", format_ct_connection_id, ct);
1263 0 : if (verbose)
1264 0 : s = format (s, "%-" SESSION_CLI_STATE_LEN "s", "LISTEN");
1265 0 : return s;
1266 : }
1267 :
1268 : static u8 *
1269 0 : format_ct_half_open (u8 *s, va_list *args)
1270 : {
1271 0 : u32 ho_index = va_arg (*args, u32);
1272 0 : u32 verbose = va_arg (*args, u32);
1273 0 : ct_connection_t *ct = ct_half_open_get (ho_index);
1274 0 : s = format (s, "%-" SESSION_CLI_ID_LEN "U", format_ct_connection_id, ct);
1275 0 : if (verbose)
1276 0 : s = format (s, "%-" SESSION_CLI_STATE_LEN "s", "HALF-OPEN");
1277 0 : return s;
1278 : }
1279 :
1280 : static u8 *
1281 0 : format_ct_connection (u8 * s, va_list * args)
1282 : {
1283 0 : ct_connection_t *ct = va_arg (*args, ct_connection_t *);
1284 0 : u32 verbose = va_arg (*args, u32);
1285 :
1286 0 : if (!ct)
1287 0 : return s;
1288 0 : s = format (s, "%-" SESSION_CLI_ID_LEN "U", format_ct_connection_id, ct);
1289 0 : if (verbose)
1290 : {
1291 0 : s = format (s, "%-" SESSION_CLI_STATE_LEN "s", "ESTABLISHED");
1292 0 : if (verbose > 1)
1293 : {
1294 0 : s = format (s, "\n");
1295 : }
1296 : }
1297 0 : return s;
1298 : }
1299 :
1300 : static u8 *
1301 0 : format_ct_session (u8 * s, va_list * args)
1302 : {
1303 0 : u32 ct_index = va_arg (*args, u32);
1304 0 : u32 thread_index = va_arg (*args, u32);
1305 0 : u32 verbose = va_arg (*args, u32);
1306 : ct_connection_t *ct;
1307 :
1308 0 : ct = ct_connection_get (ct_index, thread_index);
1309 0 : if (!ct)
1310 : {
1311 0 : s = format (s, "empty\n");
1312 0 : return s;
1313 : }
1314 :
1315 0 : s = format (s, "%U", format_ct_connection, ct, verbose);
1316 0 : return s;
1317 : }
1318 :
1319 : clib_error_t *
1320 57 : ct_enable_disable (vlib_main_t * vm, u8 is_en)
1321 : {
1322 57 : vlib_thread_main_t *vtm = &vlib_thread_main;
1323 57 : ct_main_t *cm = &ct_main;
1324 : ct_worker_t *wrk;
1325 :
1326 57 : cm->n_workers = vlib_num_workers ();
1327 57 : cm->fwrk_thread = transport_cl_thread ();
1328 57 : vec_validate (cm->wrk, vtm->n_vlib_mains);
1329 192 : vec_foreach (wrk, cm->wrk)
1330 135 : clib_spinlock_init (&wrk->pending_connects_lock);
1331 57 : clib_spinlock_init (&cm->ho_reuseable_lock);
1332 57 : clib_rwlock_init (&cm->app_segs_lock);
1333 57 : vec_validate (cm->fwrk_pending_connects, cm->n_workers);
1334 57 : return 0;
1335 : }
1336 :
1337 : /* *INDENT-OFF* */
1338 : static const transport_proto_vft_t cut_thru_proto = {
1339 : .enable = ct_enable_disable,
1340 : .start_listen = ct_start_listen,
1341 : .stop_listen = ct_stop_listen,
1342 : .get_connection = ct_session_get,
1343 : .get_listener = ct_listener_get,
1344 : .get_half_open = ct_session_half_open_get,
1345 : .cleanup = ct_session_cleanup,
1346 : .cleanup_ho = ct_cleanup_ho,
1347 : .connect = ct_session_connect,
1348 : .close = ct_session_close,
1349 : .custom_tx = ct_custom_tx,
1350 : .app_rx_evt = ct_app_rx_evt,
1351 : .format_listener = format_ct_listener,
1352 : .format_half_open = format_ct_half_open,
1353 : .format_connection = format_ct_session,
1354 : .transport_options = {
1355 : .name = "ct",
1356 : .short_name = "C",
1357 : .tx_type = TRANSPORT_TX_INTERNAL,
1358 : .service_type = TRANSPORT_SERVICE_VC,
1359 : },
1360 : };
1361 : /* *INDENT-ON* */
1362 :
1363 : static inline int
1364 137279 : ct_session_can_tx (session_t *s)
1365 : {
1366 137281 : return (s->session_state == SESSION_STATE_READY ||
1367 137279 : s->session_state == SESSION_STATE_CLOSING ||
1368 0 : s->session_state == SESSION_STATE_APP_CLOSED);
1369 : }
1370 :
1371 : int
1372 137279 : ct_session_tx (session_t * s)
1373 : {
1374 : ct_connection_t *ct, *peer_ct;
1375 : session_t *peer_s;
1376 :
1377 137279 : if (!ct_session_can_tx (s))
1378 0 : return 0;
1379 137279 : ct = (ct_connection_t *) session_get_transport (s);
1380 137279 : peer_ct = ct_connection_get (ct->peer_index, ct->c_thread_index);
1381 137279 : if (!peer_ct)
1382 0 : return 0;
1383 137279 : peer_s = session_get (peer_ct->c_s_index, peer_ct->c_thread_index);
1384 137279 : if (peer_s->session_state >= SESSION_STATE_TRANSPORT_CLOSING)
1385 2 : return 0;
1386 137277 : return session_enqueue_notify (peer_s);
1387 : }
1388 :
1389 : static clib_error_t *
1390 575 : ct_transport_init (vlib_main_t * vm)
1391 : {
1392 575 : transport_register_protocol (TRANSPORT_PROTO_NONE, &cut_thru_proto,
1393 : FIB_PROTOCOL_IP4, ~0);
1394 575 : transport_register_protocol (TRANSPORT_PROTO_NONE, &cut_thru_proto,
1395 : FIB_PROTOCOL_IP6, ~0);
1396 575 : return 0;
1397 : }
1398 :
1399 78911 : VLIB_INIT_FUNCTION (ct_transport_init);
1400 :
1401 : /*
1402 : * fd.io coding-style-patch-verification: ON
1403 : *
1404 : * Local Variables:
1405 : * eval: (c-set-style "gnu")
1406 : * End:
1407 : */
|
