Commit | Line | Data |
---|---|---|
d3e2ba59 JD |
1 | /* |
2 | * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com> | |
3 | * David Goulet <dgoulet@efficios.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License, version 2 only, | |
7 | * as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License along | |
15 | * with this program; if not, write to the Free Software Foundation, Inc., | |
16 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
17 | */ | |
18 | ||
19 | #define _GNU_SOURCE | |
20 | #include <getopt.h> | |
21 | #include <grp.h> | |
22 | #include <limits.h> | |
23 | #include <pthread.h> | |
24 | #include <signal.h> | |
25 | #include <stdio.h> | |
26 | #include <stdlib.h> | |
27 | #include <string.h> | |
28 | #include <sys/mman.h> | |
29 | #include <sys/mount.h> | |
30 | #include <sys/resource.h> | |
31 | #include <sys/socket.h> | |
32 | #include <sys/stat.h> | |
33 | #include <sys/types.h> | |
34 | #include <sys/wait.h> | |
35 | #include <inttypes.h> | |
36 | #include <urcu/futex.h> | |
37 | #include <urcu/uatomic.h> | |
38 | #include <unistd.h> | |
39 | #include <fcntl.h> | |
40 | #include <config.h> | |
41 | ||
42 | #include <lttng/lttng.h> | |
43 | #include <common/common.h> | |
44 | #include <common/compat/poll.h> | |
45 | #include <common/compat/socket.h> | |
46 | #include <common/defaults.h> | |
47 | #include <common/futex.h> | |
48 | #include <common/sessiond-comm/sessiond-comm.h> | |
49 | #include <common/sessiond-comm/inet.h> | |
50 | #include <common/sessiond-comm/relayd.h> | |
51 | #include <common/uri.h> | |
52 | #include <common/utils.h> | |
53 | ||
54 | #include "cmd.h" | |
55 | #include "live.h" | |
56 | #include "lttng-relayd.h" | |
57 | #include "lttng-viewer.h" | |
58 | #include "utils.h" | |
eea7556c | 59 | #include "health-relayd.h" |
d3e2ba59 JD |
60 | |
61 | static struct lttng_uri *live_uri; | |
62 | ||
63 | /* | |
64 | * Quit pipe for all threads. This permits a single cancellation point | |
65 | * for all threads when receiving an event on the pipe. | |
66 | */ | |
67 | static int live_thread_quit_pipe[2] = { -1, -1 }; | |
68 | ||
69 | /* | |
70 | * This pipe is used to inform the worker thread that a command is queued and | |
71 | * ready to be processed. | |
72 | */ | |
73 | static int live_relay_cmd_pipe[2] = { -1, -1 }; | |
74 | ||
75 | /* Shared between threads */ | |
76 | static int live_dispatch_thread_exit; | |
77 | ||
78 | static pthread_t live_listener_thread; | |
79 | static pthread_t live_dispatcher_thread; | |
80 | static pthread_t live_worker_thread; | |
81 | ||
82 | /* | |
83 | * Relay command queue. | |
84 | * | |
85 | * The live_thread_listener and live_thread_dispatcher communicate with this | |
86 | * queue. | |
87 | */ | |
88 | static struct relay_cmd_queue viewer_cmd_queue; | |
89 | ||
90 | static uint64_t last_relay_viewer_session_id; | |
91 | ||
92 | /* | |
93 | * Cleanup the daemon | |
94 | */ | |
95 | static | |
96 | void cleanup(void) | |
97 | { | |
98 | DBG("Cleaning up"); | |
99 | ||
d3e2ba59 JD |
100 | free(live_uri); |
101 | } | |
102 | ||
103 | /* | |
104 | * Write to writable pipe used to notify a thread. | |
105 | */ | |
106 | static | |
107 | int notify_thread_pipe(int wpipe) | |
108 | { | |
6cd525e8 | 109 | ssize_t ret; |
d3e2ba59 | 110 | |
6cd525e8 MD |
111 | ret = lttng_write(wpipe, "!", 1); |
112 | if (ret < 1) { | |
d3e2ba59 JD |
113 | PERROR("write poll pipe"); |
114 | } | |
115 | ||
6cd525e8 | 116 | return (int) ret; |
d3e2ba59 JD |
117 | } |
118 | ||
119 | /* | |
120 | * Stop all threads by closing the thread quit pipe. | |
121 | */ | |
122 | static | |
123 | void stop_threads(void) | |
124 | { | |
125 | int ret; | |
126 | ||
127 | /* Stopping all threads */ | |
128 | DBG("Terminating all live threads"); | |
129 | ret = notify_thread_pipe(live_thread_quit_pipe[1]); | |
130 | if (ret < 0) { | |
131 | ERR("write error on thread quit pipe"); | |
132 | } | |
133 | ||
134 | /* Dispatch thread */ | |
135 | CMM_STORE_SHARED(live_dispatch_thread_exit, 1); | |
136 | futex_nto1_wake(&viewer_cmd_queue.futex); | |
137 | } | |
138 | ||
d3e2ba59 JD |
139 | /* |
140 | * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. | |
141 | */ | |
142 | static | |
143 | int create_thread_poll_set(struct lttng_poll_event *events, int size) | |
144 | { | |
145 | int ret; | |
146 | ||
147 | if (events == NULL || size == 0) { | |
148 | ret = -1; | |
149 | goto error; | |
150 | } | |
151 | ||
152 | ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); | |
153 | if (ret < 0) { | |
154 | goto error; | |
155 | } | |
156 | ||
157 | /* Add quit pipe */ | |
158 | ret = lttng_poll_add(events, live_thread_quit_pipe[0], LPOLLIN); | |
159 | if (ret < 0) { | |
160 | goto error; | |
161 | } | |
162 | ||
163 | return 0; | |
164 | ||
165 | error: | |
166 | return ret; | |
167 | } | |
168 | ||
169 | /* | |
170 | * Check if the thread quit pipe was triggered. | |
171 | * | |
172 | * Return 1 if it was triggered else 0; | |
173 | */ | |
174 | static | |
175 | int check_thread_quit_pipe(int fd, uint32_t events) | |
176 | { | |
177 | if (fd == live_thread_quit_pipe[0] && (events & LPOLLIN)) { | |
178 | return 1; | |
179 | } | |
180 | ||
181 | return 0; | |
182 | } | |
183 | ||
184 | /* | |
185 | * Create and init socket from uri. | |
186 | */ | |
187 | static | |
188 | struct lttcomm_sock *init_socket(struct lttng_uri *uri) | |
189 | { | |
190 | int ret; | |
191 | struct lttcomm_sock *sock = NULL; | |
192 | ||
193 | sock = lttcomm_alloc_sock_from_uri(uri); | |
194 | if (sock == NULL) { | |
195 | ERR("Allocating socket"); | |
196 | goto error; | |
197 | } | |
198 | ||
199 | ret = lttcomm_create_sock(sock); | |
200 | if (ret < 0) { | |
201 | goto error; | |
202 | } | |
203 | DBG("Listening on sock %d for live", sock->fd); | |
204 | ||
205 | ret = sock->ops->bind(sock); | |
206 | if (ret < 0) { | |
207 | goto error; | |
208 | } | |
209 | ||
210 | ret = sock->ops->listen(sock, -1); | |
211 | if (ret < 0) { | |
212 | goto error; | |
213 | ||
214 | } | |
215 | ||
216 | return sock; | |
217 | ||
218 | error: | |
219 | if (sock) { | |
220 | lttcomm_destroy_sock(sock); | |
221 | } | |
222 | return NULL; | |
223 | } | |
224 | ||
225 | /* | |
226 | * This thread manages the listening for new connections on the network | |
227 | */ | |
228 | static | |
229 | void *thread_listener(void *data) | |
230 | { | |
231 | int i, ret, pollfd, err = -1; | |
232 | int val = 1; | |
233 | uint32_t revents, nb_fd; | |
234 | struct lttng_poll_event events; | |
235 | struct lttcomm_sock *live_control_sock; | |
236 | ||
237 | DBG("[thread] Relay live listener started"); | |
238 | ||
eea7556c MD |
239 | health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER); |
240 | ||
241 | health_code_update(); | |
242 | ||
d3e2ba59 JD |
243 | live_control_sock = init_socket(live_uri); |
244 | if (!live_control_sock) { | |
245 | goto error_sock_control; | |
246 | } | |
247 | ||
248 | /* | |
249 | * Pass 3 as size here for the thread quit pipe, control and data socket. | |
250 | */ | |
251 | ret = create_thread_poll_set(&events, 2); | |
252 | if (ret < 0) { | |
253 | goto error_create_poll; | |
254 | } | |
255 | ||
256 | /* Add the control socket */ | |
257 | ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP); | |
258 | if (ret < 0) { | |
259 | goto error_poll_add; | |
260 | } | |
261 | ||
262 | while (1) { | |
eea7556c MD |
263 | health_code_update(); |
264 | ||
d3e2ba59 JD |
265 | DBG("Listener accepting live viewers connections"); |
266 | ||
267 | restart: | |
eea7556c | 268 | health_poll_entry(); |
d3e2ba59 | 269 | ret = lttng_poll_wait(&events, -1); |
eea7556c | 270 | health_poll_exit(); |
d3e2ba59 JD |
271 | if (ret < 0) { |
272 | /* | |
273 | * Restart interrupted system call. | |
274 | */ | |
275 | if (errno == EINTR) { | |
276 | goto restart; | |
277 | } | |
278 | goto error; | |
279 | } | |
280 | nb_fd = ret; | |
281 | ||
282 | DBG("Relay new viewer connection received"); | |
283 | for (i = 0; i < nb_fd; i++) { | |
eea7556c MD |
284 | health_code_update(); |
285 | ||
d3e2ba59 JD |
286 | /* Fetch once the poll data */ |
287 | revents = LTTNG_POLL_GETEV(&events, i); | |
288 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
289 | ||
290 | /* Thread quit pipe has been closed. Killing thread. */ | |
291 | ret = check_thread_quit_pipe(pollfd, revents); | |
292 | if (ret) { | |
293 | err = 0; | |
294 | goto exit; | |
295 | } | |
296 | ||
297 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
298 | ERR("socket poll error"); | |
299 | goto error; | |
300 | } else if (revents & LPOLLIN) { | |
301 | /* | |
302 | * Get allocated in this thread, enqueued to a global queue, | |
303 | * dequeued and freed in the worker thread. | |
304 | */ | |
305 | struct relay_command *relay_cmd; | |
306 | struct lttcomm_sock *newsock; | |
307 | ||
308 | relay_cmd = zmalloc(sizeof(*relay_cmd)); | |
309 | if (!relay_cmd) { | |
310 | PERROR("relay command zmalloc"); | |
311 | goto error; | |
312 | } | |
313 | ||
314 | assert(pollfd == live_control_sock->fd); | |
315 | newsock = live_control_sock->ops->accept(live_control_sock); | |
316 | if (!newsock) { | |
317 | PERROR("accepting control sock"); | |
318 | free(relay_cmd); | |
319 | goto error; | |
320 | } | |
321 | DBG("Relay viewer connection accepted socket %d", newsock->fd); | |
322 | ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val, | |
323 | sizeof(int)); | |
324 | if (ret < 0) { | |
325 | PERROR("setsockopt inet"); | |
326 | lttcomm_destroy_sock(newsock); | |
327 | free(relay_cmd); | |
328 | goto error; | |
329 | } | |
330 | relay_cmd->sock = newsock; | |
331 | ||
332 | /* | |
333 | * Lock free enqueue the request. | |
334 | */ | |
335 | cds_wfq_enqueue(&viewer_cmd_queue.queue, &relay_cmd->node); | |
336 | ||
337 | /* | |
338 | * Wake the dispatch queue futex. Implicit memory | |
339 | * barrier with the exchange in cds_wfq_enqueue. | |
340 | */ | |
341 | futex_nto1_wake(&viewer_cmd_queue.futex); | |
342 | } | |
343 | } | |
344 | } | |
345 | ||
346 | exit: | |
347 | error: | |
348 | error_poll_add: | |
349 | lttng_poll_clean(&events); | |
350 | error_create_poll: | |
351 | if (live_control_sock->fd >= 0) { | |
352 | ret = live_control_sock->ops->close(live_control_sock); | |
353 | if (ret) { | |
354 | PERROR("close"); | |
355 | } | |
356 | } | |
357 | lttcomm_destroy_sock(live_control_sock); | |
358 | error_sock_control: | |
359 | if (err) { | |
eea7556c | 360 | health_error(); |
d3e2ba59 JD |
361 | DBG("Live viewer listener thread exited with error"); |
362 | } | |
eea7556c | 363 | health_unregister(health_relayd); |
d3e2ba59 JD |
364 | DBG("Live viewer listener thread cleanup complete"); |
365 | stop_threads(); | |
366 | return NULL; | |
367 | } | |
368 | ||
369 | /* | |
370 | * This thread manages the dispatching of the requests to worker threads | |
371 | */ | |
372 | static | |
373 | void *thread_dispatcher(void *data) | |
374 | { | |
6cd525e8 MD |
375 | int err = -1; |
376 | ssize_t ret; | |
d3e2ba59 JD |
377 | struct cds_wfq_node *node; |
378 | struct relay_command *relay_cmd = NULL; | |
379 | ||
380 | DBG("[thread] Live viewer relay dispatcher started"); | |
381 | ||
eea7556c MD |
382 | health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER); |
383 | ||
384 | health_code_update(); | |
385 | ||
d3e2ba59 | 386 | while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) { |
eea7556c MD |
387 | health_code_update(); |
388 | ||
d3e2ba59 JD |
389 | /* Atomically prepare the queue futex */ |
390 | futex_nto1_prepare(&viewer_cmd_queue.futex); | |
391 | ||
392 | do { | |
eea7556c MD |
393 | health_code_update(); |
394 | ||
d3e2ba59 JD |
395 | /* Dequeue commands */ |
396 | node = cds_wfq_dequeue_blocking(&viewer_cmd_queue.queue); | |
397 | if (node == NULL) { | |
398 | DBG("Woken up but nothing in the live-viewer " | |
399 | "relay command queue"); | |
400 | /* Continue thread execution */ | |
401 | break; | |
402 | } | |
403 | ||
404 | relay_cmd = caa_container_of(node, struct relay_command, node); | |
405 | DBG("Dispatching viewer request waiting on sock %d", | |
406 | relay_cmd->sock->fd); | |
407 | ||
408 | /* | |
409 | * Inform worker thread of the new request. This call is blocking | |
410 | * so we can be assured that the data will be read at some point in | |
411 | * time or wait to the end of the world :) | |
412 | */ | |
6cd525e8 MD |
413 | ret = lttng_write(live_relay_cmd_pipe[1], relay_cmd, |
414 | sizeof(*relay_cmd)); | |
d3e2ba59 | 415 | free(relay_cmd); |
6cd525e8 | 416 | if (ret < sizeof(struct relay_command)) { |
d3e2ba59 JD |
417 | PERROR("write cmd pipe"); |
418 | goto error; | |
419 | } | |
420 | } while (node != NULL); | |
421 | ||
422 | /* Futex wait on queue. Blocking call on futex() */ | |
eea7556c | 423 | health_poll_entry(); |
d3e2ba59 | 424 | futex_nto1_wait(&viewer_cmd_queue.futex); |
eea7556c | 425 | health_poll_exit(); |
d3e2ba59 JD |
426 | } |
427 | ||
eea7556c MD |
428 | /* Normal exit, no error */ |
429 | err = 0; | |
430 | ||
d3e2ba59 | 431 | error: |
eea7556c MD |
432 | if (err) { |
433 | health_error(); | |
434 | ERR("Health error occurred in %s", __func__); | |
435 | } | |
436 | health_unregister(health_relayd); | |
d3e2ba59 JD |
437 | DBG("Live viewer dispatch thread dying"); |
438 | stop_threads(); | |
439 | return NULL; | |
440 | } | |
441 | ||
442 | /* | |
443 | * Establish connection with the viewer and check the versions. | |
444 | * | |
445 | * Return 0 on success or else negative value. | |
446 | */ | |
447 | static | |
448 | int viewer_connect(struct relay_command *cmd) | |
449 | { | |
450 | int ret; | |
451 | struct lttng_viewer_connect reply, msg; | |
452 | ||
453 | assert(cmd); | |
454 | ||
455 | cmd->version_check_done = 1; | |
456 | ||
eea7556c MD |
457 | health_code_update(); |
458 | ||
d3e2ba59 JD |
459 | /* Get version from the other side. */ |
460 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
461 | if (ret < 0 || ret != sizeof(msg)) { | |
462 | if (ret == 0) { | |
463 | /* Orderly shutdown. Not necessary to print an error. */ | |
464 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
465 | } else { | |
466 | ERR("Relay failed to receive the version values."); | |
467 | } | |
468 | ret = -1; | |
469 | goto end; | |
470 | } | |
471 | ||
eea7556c MD |
472 | health_code_update(); |
473 | ||
d3e2ba59 JD |
474 | reply.major = RELAYD_VERSION_COMM_MAJOR; |
475 | reply.minor = RELAYD_VERSION_COMM_MINOR; | |
476 | ||
477 | /* Major versions must be the same */ | |
478 | if (reply.major != be32toh(msg.major)) { | |
479 | DBG("Incompatible major versions (%u vs %u)", reply.major, | |
480 | be32toh(msg.major)); | |
481 | ret = 0; | |
482 | goto end; | |
483 | } | |
484 | ||
485 | cmd->major = reply.major; | |
486 | /* We adapt to the lowest compatible version */ | |
487 | if (reply.minor <= be32toh(msg.minor)) { | |
488 | cmd->minor = reply.minor; | |
489 | } else { | |
490 | cmd->minor = be32toh(msg.minor); | |
491 | } | |
492 | ||
493 | if (be32toh(msg.type) == VIEWER_CLIENT_COMMAND) { | |
494 | cmd->type = RELAY_VIEWER_COMMAND; | |
495 | } else if (be32toh(msg.type) == VIEWER_CLIENT_NOTIFICATION) { | |
496 | cmd->type = RELAY_VIEWER_NOTIFICATION; | |
497 | } else { | |
498 | ERR("Unknown connection type : %u", be32toh(msg.type)); | |
499 | ret = -1; | |
500 | goto end; | |
501 | } | |
502 | ||
503 | reply.major = htobe32(reply.major); | |
504 | reply.minor = htobe32(reply.minor); | |
505 | if (cmd->type == RELAY_VIEWER_COMMAND) { | |
506 | reply.viewer_session_id = htobe64(++last_relay_viewer_session_id); | |
507 | } | |
eea7556c MD |
508 | |
509 | health_code_update(); | |
510 | ||
d3e2ba59 JD |
511 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
512 | sizeof(struct lttng_viewer_connect), 0); | |
513 | if (ret < 0) { | |
514 | ERR("Relay sending version"); | |
515 | } | |
516 | ||
eea7556c MD |
517 | health_code_update(); |
518 | ||
d3e2ba59 JD |
519 | DBG("Version check done using protocol %u.%u", cmd->major, cmd->minor); |
520 | ret = 0; | |
521 | ||
522 | end: | |
523 | return ret; | |
524 | } | |
525 | ||
526 | /* | |
527 | * Send the viewer the list of current sessions. | |
528 | * | |
529 | * Return 0 on success or else a negative value. | |
530 | */ | |
531 | static | |
532 | int viewer_list_sessions(struct relay_command *cmd, | |
533 | struct lttng_ht *sessions_ht) | |
534 | { | |
535 | int ret; | |
536 | struct lttng_viewer_list_sessions session_list; | |
537 | unsigned long count; | |
538 | long approx_before, approx_after; | |
539 | struct lttng_ht_node_ulong *node; | |
540 | struct lttng_ht_iter iter; | |
541 | struct lttng_viewer_session send_session; | |
542 | struct relay_session *session; | |
543 | ||
544 | DBG("List sessions received"); | |
545 | ||
546 | if (cmd->version_check_done == 0) { | |
547 | ERR("Trying to list sessions before version check"); | |
548 | ret = -1; | |
549 | goto end_no_session; | |
550 | } | |
551 | ||
552 | rcu_read_lock(); | |
553 | cds_lfht_count_nodes(sessions_ht->ht, &approx_before, &count, &approx_after); | |
554 | session_list.sessions_count = htobe32(count); | |
555 | ||
eea7556c MD |
556 | health_code_update(); |
557 | ||
d3e2ba59 JD |
558 | ret = cmd->sock->ops->sendmsg(cmd->sock, &session_list, |
559 | sizeof(session_list), 0); | |
560 | if (ret < 0) { | |
561 | ERR("Relay sending sessions list"); | |
562 | goto end_unlock; | |
563 | } | |
564 | ||
eea7556c MD |
565 | health_code_update(); |
566 | ||
d3e2ba59 | 567 | cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, node, node) { |
eea7556c MD |
568 | health_code_update(); |
569 | ||
d3e2ba59 JD |
570 | node = lttng_ht_iter_get_node_ulong(&iter); |
571 | if (!node) { | |
572 | goto end_unlock; | |
573 | } | |
574 | session = caa_container_of(node, struct relay_session, session_n); | |
575 | ||
576 | strncpy(send_session.session_name, session->session_name, | |
577 | sizeof(send_session.session_name)); | |
578 | strncpy(send_session.hostname, session->hostname, | |
579 | sizeof(send_session.hostname)); | |
580 | send_session.id = htobe64(session->id); | |
581 | send_session.live_timer = htobe32(session->live_timer); | |
582 | send_session.clients = htobe32(session->viewer_attached); | |
87b576ec | 583 | send_session.streams = htobe32(session->stream_count); |
d3e2ba59 | 584 | |
eea7556c MD |
585 | health_code_update(); |
586 | ||
d3e2ba59 JD |
587 | ret = cmd->sock->ops->sendmsg(cmd->sock, &send_session, |
588 | sizeof(send_session), 0); | |
589 | if (ret < 0) { | |
590 | ERR("Relay sending session info"); | |
591 | goto end_unlock; | |
592 | } | |
593 | } | |
eea7556c MD |
594 | health_code_update(); |
595 | ||
d3e2ba59 JD |
596 | rcu_read_unlock(); |
597 | ret = 0; | |
598 | goto end; | |
599 | ||
600 | end_unlock: | |
601 | rcu_read_unlock(); | |
602 | ||
603 | end: | |
604 | end_no_session: | |
605 | return ret; | |
606 | } | |
607 | ||
0e6830aa JD |
608 | /* |
609 | * Open index file using a given viewer stream. | |
610 | * | |
611 | * Return 0 on success or else a negative value. | |
612 | */ | |
613 | static int open_index(struct relay_viewer_stream *stream) | |
614 | { | |
615 | int ret; | |
616 | char fullpath[PATH_MAX]; | |
617 | struct lttng_packet_index_file_hdr hdr; | |
618 | ||
6b6b9a5a JD |
619 | if (stream->tracefile_count > 0) { |
620 | ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR "/%s_%" | |
621 | PRIu64 DEFAULT_INDEX_FILE_SUFFIX, stream->path_name, | |
622 | stream->channel_name, stream->tracefile_count_current); | |
623 | } else { | |
624 | ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR "/%s" | |
625 | DEFAULT_INDEX_FILE_SUFFIX, stream->path_name, | |
626 | stream->channel_name); | |
0e6830aa | 627 | } |
0e6830aa JD |
628 | if (ret < 0) { |
629 | PERROR("snprintf index path"); | |
630 | goto error; | |
631 | } | |
632 | ||
633 | DBG("Opening index file %s in read only", fullpath); | |
634 | ret = open(fullpath, O_RDONLY); | |
635 | if (ret < 0) { | |
636 | if (errno == ENOENT) { | |
637 | ret = -ENOENT; | |
638 | goto error; | |
639 | } else { | |
640 | PERROR("opening index in read-only"); | |
641 | } | |
642 | goto error; | |
643 | } | |
644 | stream->index_read_fd = ret; | |
645 | DBG("Opening index file %s in read only, (fd: %d)", fullpath, ret); | |
646 | ||
6cd525e8 MD |
647 | ret = lttng_read(stream->index_read_fd, &hdr, sizeof(hdr)); |
648 | if (ret < sizeof(hdr)) { | |
0e6830aa JD |
649 | PERROR("Reading index header"); |
650 | goto error; | |
651 | } | |
652 | if (strncmp(hdr.magic, INDEX_MAGIC, sizeof(hdr.magic)) != 0) { | |
653 | ERR("Invalid header magic"); | |
654 | ret = -1; | |
655 | goto error; | |
656 | } | |
657 | if (be32toh(hdr.index_major) != INDEX_MAJOR || | |
658 | be32toh(hdr.index_minor) != INDEX_MINOR) { | |
659 | ERR("Invalid header version"); | |
660 | ret = -1; | |
661 | goto error; | |
662 | } | |
663 | ret = 0; | |
664 | ||
665 | error: | |
0e6830aa JD |
666 | return ret; |
667 | } | |
668 | ||
d3e2ba59 JD |
669 | /* |
670 | * Allocate and init a new viewer_stream. | |
671 | * | |
672 | * Copies the values from the stream passed in parameter and insert the new | |
673 | * stream in the viewer_streams_ht. | |
674 | * | |
675 | * MUST be called with rcu_read_lock held. | |
676 | * | |
677 | * Returns 0 on success or a negative value on error. | |
678 | */ | |
679 | static | |
0e6830aa | 680 | int init_viewer_stream(struct relay_stream *stream, int seek_last) |
d3e2ba59 JD |
681 | { |
682 | int ret; | |
683 | struct relay_viewer_stream *viewer_stream; | |
684 | ||
685 | assert(stream); | |
d3e2ba59 JD |
686 | |
687 | viewer_stream = zmalloc(sizeof(*viewer_stream)); | |
688 | if (!viewer_stream) { | |
689 | PERROR("relay viewer stream zmalloc"); | |
690 | ret = -1; | |
691 | goto error; | |
692 | } | |
d3e2ba59 JD |
693 | viewer_stream->session_id = stream->session->id; |
694 | viewer_stream->stream_handle = stream->stream_handle; | |
695 | viewer_stream->path_name = strndup(stream->path_name, | |
696 | LTTNG_VIEWER_PATH_MAX); | |
697 | viewer_stream->channel_name = strndup(stream->channel_name, | |
698 | LTTNG_VIEWER_NAME_MAX); | |
d3e2ba59 JD |
699 | viewer_stream->tracefile_count = stream->tracefile_count; |
700 | viewer_stream->metadata_flag = stream->metadata_flag; | |
6b6b9a5a JD |
701 | if (seek_last) { |
702 | viewer_stream->tracefile_count_current = | |
703 | stream->tracefile_count_current; | |
704 | } else { | |
705 | viewer_stream->tracefile_count_current = | |
706 | stream->oldest_tracefile_id; | |
707 | } | |
708 | ||
709 | /* | |
710 | * The deletion of this ctf_trace object is only done in a call RCU of the | |
711 | * relay stream making it valid as long as we have the read side lock. | |
712 | */ | |
713 | viewer_stream->ctf_trace = stream->ctf_trace; | |
714 | uatomic_inc(&viewer_stream->ctf_trace->refcount); | |
d3e2ba59 | 715 | |
6b6b9a5a JD |
716 | lttng_ht_node_init_u64(&viewer_stream->stream_n, stream->stream_handle); |
717 | lttng_ht_add_unique_u64(viewer_streams_ht, &viewer_stream->stream_n); | |
718 | ||
719 | viewer_stream->index_read_fd = -1; | |
720 | viewer_stream->read_fd = -1; | |
721 | ||
722 | /* | |
723 | * This is to avoid a race between the initialization of this object and | |
724 | * the close of the given stream. If the stream is unable to find this | |
725 | * viewer stream when closing, this copy will at least take the latest | |
726 | * value. | |
727 | * We also need that for the seek_last. | |
728 | */ | |
729 | viewer_stream->total_index_received = stream->total_index_received; | |
730 | ||
731 | /* | |
732 | * If we never received an index for the current stream, delay | |
733 | * the opening of the index, otherwise open it right now. | |
734 | */ | |
735 | if (viewer_stream->tracefile_count_current == | |
736 | stream->tracefile_count_current && | |
737 | viewer_stream->total_index_received == 0) { | |
738 | viewer_stream->index_read_fd = -1; | |
739 | } else { | |
0e6830aa JD |
740 | ret = open_index(viewer_stream); |
741 | if (ret < 0) { | |
742 | goto error; | |
743 | } | |
6b6b9a5a JD |
744 | } |
745 | ||
746 | if (seek_last && viewer_stream->index_read_fd > 0) { | |
0e6830aa JD |
747 | ret = lseek(viewer_stream->index_read_fd, |
748 | viewer_stream->total_index_received * | |
749 | sizeof(struct lttng_packet_index), | |
750 | SEEK_CUR); | |
751 | if (ret < 0) { | |
752 | goto error; | |
753 | } | |
754 | viewer_stream->last_sent_index = | |
755 | viewer_stream->total_index_received; | |
756 | } | |
757 | ||
6b6b9a5a JD |
758 | ret = 0; |
759 | ||
760 | error: | |
761 | return ret; | |
762 | } | |
763 | ||
764 | /* | |
765 | * Rotate a stream to the next tracefile. | |
766 | * | |
767 | * Returns 0 on success, a negative value on error. | |
768 | */ | |
769 | static | |
770 | int rotate_viewer_stream(struct relay_viewer_stream *viewer_stream, | |
771 | struct relay_stream *stream) | |
772 | { | |
773 | int ret; | |
774 | uint64_t tracefile_id; | |
775 | ||
776 | assert(viewer_stream); | |
777 | ||
778 | tracefile_id = (viewer_stream->tracefile_count_current + 1) % | |
779 | viewer_stream->tracefile_count; | |
d3e2ba59 | 780 | |
6b6b9a5a JD |
781 | if (stream) { |
782 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); | |
783 | } | |
d3e2ba59 | 784 | /* |
6b6b9a5a JD |
785 | * The writer and the reader are not working in the same |
786 | * tracefile, we can read up to EOF, we don't care about the | |
787 | * total_index_received. | |
d3e2ba59 | 788 | */ |
6b6b9a5a JD |
789 | if (!stream || (stream->tracefile_count_current != tracefile_id)) { |
790 | viewer_stream->close_write_flag = 1; | |
791 | } else { | |
792 | /* | |
793 | * We are opening a file that is still open in write, make | |
794 | * sure we limit our reading to the number of indexes | |
795 | * received. | |
796 | */ | |
797 | viewer_stream->close_write_flag = 0; | |
798 | if (stream) { | |
799 | viewer_stream->total_index_received = | |
800 | stream->total_index_received; | |
801 | } | |
802 | } | |
803 | viewer_stream->tracefile_count_current = tracefile_id; | |
d3e2ba59 | 804 | |
6b6b9a5a | 805 | if (viewer_stream->abort_flag == 0) { |
cef0f7d5 JD |
806 | if (viewer_stream->index_read_fd > 0) { |
807 | ret = close(viewer_stream->index_read_fd); | |
808 | if (ret < 0) { | |
809 | PERROR("close index file %d", | |
810 | viewer_stream->index_read_fd); | |
811 | } | |
812 | viewer_stream->index_read_fd = -1; | |
6b6b9a5a | 813 | } |
cef0f7d5 JD |
814 | if (viewer_stream->read_fd > 0) { |
815 | ret = close(viewer_stream->read_fd); | |
816 | if (ret < 0) { | |
817 | PERROR("close tracefile %d", | |
818 | viewer_stream->read_fd); | |
819 | } | |
820 | viewer_stream->read_fd = -1; | |
6b6b9a5a JD |
821 | } |
822 | } else { | |
823 | viewer_stream->abort_flag = 0; | |
824 | } | |
825 | ||
cef0f7d5 | 826 | viewer_stream->index_read_fd = -1; |
6b6b9a5a JD |
827 | viewer_stream->read_fd = -1; |
828 | ||
cef0f7d5 JD |
829 | if (stream) { |
830 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); | |
831 | } | |
6b6b9a5a JD |
832 | ret = open_index(viewer_stream); |
833 | if (ret < 0) { | |
834 | goto error; | |
835 | } | |
d3e2ba59 JD |
836 | |
837 | ret = 0; | |
838 | ||
839 | error: | |
840 | return ret; | |
841 | } | |
842 | ||
843 | /* | |
844 | * Send the viewer the list of current sessions. | |
845 | */ | |
846 | static | |
847 | int viewer_attach_session(struct relay_command *cmd, | |
92c6ca54 | 848 | struct lttng_ht *sessions_ht) |
d3e2ba59 JD |
849 | { |
850 | int ret, send_streams = 0, nb_streams = 0; | |
851 | struct lttng_viewer_attach_session_request request; | |
852 | struct lttng_viewer_attach_session_response response; | |
853 | struct lttng_viewer_stream send_stream; | |
854 | struct relay_stream *stream; | |
855 | struct relay_viewer_stream *viewer_stream; | |
856 | struct lttng_ht_node_ulong *node; | |
857 | struct lttng_ht_node_u64 *node64; | |
858 | struct lttng_ht_iter iter; | |
859 | struct relay_session *session; | |
0e6830aa | 860 | int seek_last = 0; |
d3e2ba59 JD |
861 | |
862 | assert(cmd); | |
863 | assert(sessions_ht); | |
d3e2ba59 JD |
864 | |
865 | DBG("Attach session received"); | |
866 | ||
867 | if (cmd->version_check_done == 0) { | |
868 | ERR("Trying to attach session before version check"); | |
869 | ret = -1; | |
870 | goto end_no_session; | |
871 | } | |
872 | ||
eea7556c MD |
873 | health_code_update(); |
874 | ||
d3e2ba59 JD |
875 | ret = cmd->sock->ops->recvmsg(cmd->sock, &request, sizeof(request), 0); |
876 | if (ret < 0 || ret != sizeof(request)) { | |
877 | if (ret == 0) { | |
878 | /* Orderly shutdown. Not necessary to print an error. */ | |
879 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
880 | } else { | |
881 | ERR("Relay failed to receive the attach parameters."); | |
882 | } | |
883 | ret = -1; | |
884 | goto error; | |
885 | } | |
886 | ||
eea7556c MD |
887 | health_code_update(); |
888 | ||
d3e2ba59 JD |
889 | rcu_read_lock(); |
890 | lttng_ht_lookup(sessions_ht, | |
891 | (void *)((unsigned long) be64toh(request.session_id)), &iter); | |
892 | node = lttng_ht_iter_get_node_ulong(&iter); | |
893 | if (node == NULL) { | |
894 | DBG("Relay session %" PRIu64 " not found", | |
895 | be64toh(request.session_id)); | |
896 | response.status = htobe32(VIEWER_ATTACH_UNK); | |
897 | goto send_reply; | |
898 | } | |
899 | ||
900 | session = caa_container_of(node, struct relay_session, session_n); | |
b92fdc2b | 901 | if (cmd->session_id == session->id) { |
d3e2ba59 JD |
902 | /* Same viewer already attached, just send the stream list. */ |
903 | send_streams = 1; | |
904 | response.status = htobe32(VIEWER_ATTACH_OK); | |
905 | } else if (session->viewer_attached != 0) { | |
906 | DBG("Already a viewer attached"); | |
907 | response.status = htobe32(VIEWER_ATTACH_ALREADY); | |
908 | goto send_reply; | |
909 | } else if (session->live_timer == 0) { | |
910 | DBG("Not live session"); | |
911 | response.status = htobe32(VIEWER_ATTACH_NOT_LIVE); | |
912 | goto send_reply; | |
913 | } else { | |
914 | session->viewer_attached++; | |
915 | send_streams = 1; | |
916 | response.status = htobe32(VIEWER_ATTACH_OK); | |
b92fdc2b | 917 | cmd->session_id = session->id; |
d3e2ba59 JD |
918 | cmd->session = session; |
919 | } | |
920 | ||
921 | switch (be32toh(request.seek)) { | |
922 | case VIEWER_SEEK_BEGINNING: | |
923 | /* Default behaviour. */ | |
924 | break; | |
925 | case VIEWER_SEEK_LAST: | |
0e6830aa | 926 | seek_last = 1; |
d3e2ba59 JD |
927 | break; |
928 | default: | |
929 | ERR("Wrong seek parameter"); | |
930 | response.status = htobe32(VIEWER_ATTACH_SEEK_ERR); | |
931 | send_streams = 0; | |
932 | goto send_reply; | |
933 | } | |
934 | ||
935 | if (send_streams) { | |
936 | /* We should only be there if we have a session to attach to. */ | |
937 | assert(session); | |
938 | ||
939 | /* | |
940 | * Fill the viewer_streams_ht to count the number of streams | |
941 | * ready to be sent and avoid concurrency issues on the | |
942 | * relay_streams_ht and don't rely on a total session stream count. | |
943 | */ | |
944 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) { | |
945 | struct relay_viewer_stream *vstream; | |
946 | ||
eea7556c MD |
947 | health_code_update(); |
948 | ||
d3e2ba59 JD |
949 | node = lttng_ht_iter_get_node_ulong(&iter); |
950 | if (!node) { | |
951 | continue; | |
952 | } | |
953 | stream = caa_container_of(node, struct relay_stream, stream_n); | |
954 | if (stream->session != cmd->session) { | |
955 | continue; | |
956 | } | |
957 | ||
958 | /* | |
959 | * Don't send streams with no ctf_trace, they are not ready to be | |
960 | * read. | |
961 | */ | |
962 | if (!stream->ctf_trace) { | |
963 | continue; | |
964 | } | |
965 | ||
92c6ca54 | 966 | vstream = live_find_viewer_stream_by_id(stream->stream_handle); |
d3e2ba59 | 967 | if (!vstream) { |
0e6830aa | 968 | ret = init_viewer_stream(stream, seek_last); |
d3e2ba59 JD |
969 | if (ret < 0) { |
970 | goto end_unlock; | |
971 | } | |
972 | } | |
973 | nb_streams++; | |
974 | } | |
975 | response.streams_count = htobe32(nb_streams); | |
976 | } | |
977 | ||
978 | send_reply: | |
eea7556c | 979 | health_code_update(); |
d3e2ba59 JD |
980 | ret = cmd->sock->ops->sendmsg(cmd->sock, &response, sizeof(response), 0); |
981 | if (ret < 0) { | |
982 | ERR("Relay sending viewer attach response"); | |
983 | goto end_unlock; | |
984 | } | |
eea7556c | 985 | health_code_update(); |
d3e2ba59 JD |
986 | |
987 | /* | |
988 | * Unknown or busy session, just return gracefully, the viewer knows what | |
989 | * is happening. | |
990 | */ | |
991 | if (!send_streams) { | |
992 | ret = 0; | |
993 | goto end_unlock; | |
994 | } | |
995 | ||
996 | /* We should only be there if we have a session to attach to. */ | |
997 | assert(session); | |
998 | cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, node, node) { | |
eea7556c MD |
999 | health_code_update(); |
1000 | ||
d3e2ba59 JD |
1001 | node64 = lttng_ht_iter_get_node_u64(&iter); |
1002 | if (!node64) { | |
1003 | continue; | |
1004 | } | |
1005 | viewer_stream = caa_container_of(node64, struct relay_viewer_stream, | |
1006 | stream_n); | |
1007 | if (viewer_stream->session_id != cmd->session->id) { | |
1008 | continue; | |
1009 | } | |
1010 | ||
1011 | send_stream.id = htobe64(viewer_stream->stream_handle); | |
1012 | send_stream.ctf_trace_id = htobe64(viewer_stream->ctf_trace->id); | |
1013 | send_stream.metadata_flag = htobe32(viewer_stream->metadata_flag); | |
1014 | strncpy(send_stream.path_name, viewer_stream->path_name, | |
1015 | sizeof(send_stream.path_name)); | |
1016 | strncpy(send_stream.channel_name, viewer_stream->channel_name, | |
1017 | sizeof(send_stream.channel_name)); | |
1018 | ||
1019 | ret = cmd->sock->ops->sendmsg(cmd->sock, &send_stream, | |
1020 | sizeof(send_stream), 0); | |
1021 | if (ret < 0) { | |
1022 | ERR("Relay sending stream %" PRIu64, viewer_stream->stream_handle); | |
1023 | goto end_unlock; | |
1024 | } | |
1025 | DBG("Sent stream %" PRIu64 " to viewer", viewer_stream->stream_handle); | |
1026 | } | |
1027 | ret = 0; | |
1028 | ||
1029 | end_unlock: | |
1030 | rcu_read_unlock(); | |
1031 | end_no_session: | |
1032 | error: | |
1033 | return ret; | |
1034 | } | |
1035 | ||
d3e2ba59 JD |
1036 | /* |
1037 | * Get viewer stream from stream id. | |
1038 | * | |
1039 | * RCU read side lock MUST be acquired. | |
1040 | */ | |
92c6ca54 | 1041 | struct relay_viewer_stream *live_find_viewer_stream_by_id(uint64_t stream_id) |
d3e2ba59 JD |
1042 | { |
1043 | struct lttng_ht_node_u64 *node; | |
1044 | struct lttng_ht_iter iter; | |
1045 | struct relay_viewer_stream *stream = NULL; | |
1046 | ||
d3e2ba59 JD |
1047 | lttng_ht_lookup(viewer_streams_ht, &stream_id, &iter); |
1048 | node = lttng_ht_iter_get_node_u64(&iter); | |
1049 | if (node == NULL) { | |
1050 | DBG("Relay viewer stream %" PRIu64 " not found", stream_id); | |
1051 | goto end; | |
1052 | } | |
1053 | stream = caa_container_of(node, struct relay_viewer_stream, stream_n); | |
1054 | ||
1055 | end: | |
1056 | return stream; | |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * Send the next index for a stream. | |
1061 | * | |
1062 | * Return 0 on success or else a negative value. | |
1063 | */ | |
1064 | static | |
1065 | int viewer_get_next_index(struct relay_command *cmd, | |
92c6ca54 | 1066 | struct lttng_ht *sessions_ht) |
d3e2ba59 JD |
1067 | { |
1068 | int ret; | |
1069 | struct lttng_viewer_get_next_index request_index; | |
1070 | struct lttng_viewer_index viewer_index; | |
1071 | struct lttng_packet_index packet_index; | |
1072 | struct relay_viewer_stream *vstream; | |
1073 | struct relay_stream *rstream; | |
1074 | ||
1075 | assert(cmd); | |
d3e2ba59 JD |
1076 | assert(sessions_ht); |
1077 | ||
1078 | DBG("Viewer get next index"); | |
1079 | ||
1080 | if (cmd->version_check_done == 0) { | |
1081 | ERR("Trying to request index before version check"); | |
1082 | ret = -1; | |
1083 | goto end_no_session; | |
1084 | } | |
1085 | ||
eea7556c | 1086 | health_code_update(); |
d3e2ba59 JD |
1087 | ret = cmd->sock->ops->recvmsg(cmd->sock, &request_index, |
1088 | sizeof(request_index), 0); | |
1089 | if (ret < 0 || ret != sizeof(request_index)) { | |
1090 | ret = -1; | |
1091 | ERR("Relay didn't receive the whole packet"); | |
1092 | goto end; | |
1093 | } | |
eea7556c | 1094 | health_code_update(); |
d3e2ba59 JD |
1095 | |
1096 | rcu_read_lock(); | |
92c6ca54 | 1097 | vstream = live_find_viewer_stream_by_id(be64toh(request_index.stream_id)); |
d3e2ba59 JD |
1098 | if (!vstream) { |
1099 | ret = -1; | |
1100 | goto end_unlock; | |
1101 | } | |
1102 | ||
1103 | memset(&viewer_index, 0, sizeof(viewer_index)); | |
1104 | ||
1105 | /* | |
1106 | * The viewer should not ask for index on metadata stream. | |
1107 | */ | |
1108 | if (vstream->metadata_flag) { | |
1109 | viewer_index.status = htobe32(VIEWER_INDEX_HUP); | |
1110 | goto send_reply; | |
1111 | } | |
1112 | ||
1113 | /* First time, we open the index file */ | |
1114 | if (vstream->index_read_fd < 0) { | |
1115 | ret = open_index(vstream); | |
0e6830aa | 1116 | if (ret == -ENOENT) { |
d3e2ba59 JD |
1117 | /* |
1118 | * The index is created only when the first data packet arrives, it | |
1119 | * might not be ready at the beginning of the session | |
1120 | */ | |
1121 | viewer_index.status = htobe32(VIEWER_INDEX_RETRY); | |
1122 | goto send_reply; | |
1123 | } else if (ret < 0) { | |
1124 | viewer_index.status = htobe32(VIEWER_INDEX_ERR); | |
1125 | goto send_reply; | |
1126 | } | |
1127 | } | |
1128 | ||
1129 | rstream = relay_stream_find_by_id(vstream->stream_handle); | |
1130 | if (rstream) { | |
6b6b9a5a JD |
1131 | if (vstream->abort_flag) { |
1132 | /* Rotate on abort (overwrite). */ | |
1133 | DBG("Viewer rotate because of overwrite"); | |
1134 | ret = rotate_viewer_stream(vstream, rstream); | |
1135 | if (ret < 0) { | |
1136 | goto end_unlock; | |
1137 | } | |
1138 | } | |
6b6b9a5a | 1139 | pthread_mutex_lock(&rstream->viewer_stream_rotation_lock); |
cef0f7d5 JD |
1140 | if (rstream->tracefile_count_current == vstream->tracefile_count_current) { |
1141 | if (rstream->beacon_ts_end != -1ULL && | |
1142 | vstream->last_sent_index == rstream->total_index_received) { | |
1143 | viewer_index.status = htobe32(VIEWER_INDEX_INACTIVE); | |
1144 | viewer_index.timestamp_end = htobe64(rstream->beacon_ts_end); | |
1145 | pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock); | |
1146 | goto send_reply; | |
1147 | /* | |
1148 | * Reader and writer are working in the same tracefile, so we care | |
1149 | * about the number of index received and sent. Otherwise, we read | |
1150 | * up to EOF. | |
1151 | */ | |
1152 | } else if (rstream->total_index_received <= vstream->last_sent_index | |
1153 | && !vstream->close_write_flag) { | |
1154 | pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock); | |
1155 | /* No new index to send, retry later. */ | |
1156 | viewer_index.status = htobe32(VIEWER_INDEX_RETRY); | |
1157 | goto send_reply; | |
1158 | } | |
d3e2ba59 | 1159 | } |
6b6b9a5a JD |
1160 | pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock); |
1161 | } else if (!rstream && vstream->close_write_flag && | |
d3e2ba59 | 1162 | vstream->total_index_received == vstream->last_sent_index) { |
6b6b9a5a | 1163 | /* Last index sent and current tracefile closed in write */ |
d3e2ba59 JD |
1164 | viewer_index.status = htobe32(VIEWER_INDEX_HUP); |
1165 | goto send_reply; | |
6b6b9a5a JD |
1166 | } else { |
1167 | vstream->close_write_flag = 1; | |
d3e2ba59 JD |
1168 | } |
1169 | ||
1170 | if (!vstream->ctf_trace->metadata_received || | |
1171 | vstream->ctf_trace->metadata_received > | |
1172 | vstream->ctf_trace->metadata_sent) { | |
1173 | viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA; | |
1174 | } | |
1175 | ||
cef0f7d5 JD |
1176 | pthread_mutex_lock(&vstream->overwrite_lock); |
1177 | if (vstream->abort_flag) { | |
1178 | /* | |
1179 | * The file is being overwritten by the writer, we cannot | |
1180 | * use it. | |
1181 | */ | |
1182 | viewer_index.status = htobe32(VIEWER_INDEX_RETRY); | |
1183 | pthread_mutex_unlock(&vstream->overwrite_lock); | |
1184 | ret = rotate_viewer_stream(vstream, rstream); | |
1185 | if (ret < 0) { | |
1186 | goto end_unlock; | |
1187 | } | |
1188 | goto send_reply; | |
1189 | } | |
6cd525e8 MD |
1190 | ret = lttng_read(vstream->index_read_fd, &packet_index, |
1191 | sizeof(packet_index)); | |
cef0f7d5 | 1192 | pthread_mutex_unlock(&vstream->overwrite_lock); |
d3e2ba59 | 1193 | if (ret < sizeof(packet_index)) { |
6b6b9a5a JD |
1194 | /* |
1195 | * The tracefile is closed in write, so we read up to EOF. | |
1196 | */ | |
1197 | if (vstream->close_write_flag == 1) { | |
1198 | viewer_index.status = htobe32(VIEWER_INDEX_RETRY); | |
1199 | /* Rotate on normal EOF */ | |
1200 | ret = rotate_viewer_stream(vstream, rstream); | |
1201 | if (ret < 0) { | |
1202 | goto end_unlock; | |
1203 | } | |
1204 | } else { | |
cef0f7d5 JD |
1205 | PERROR("Relay reading index file %d", |
1206 | vstream->index_read_fd); | |
1207 | viewer_index.status = htobe32(VIEWER_INDEX_ERR); | |
6b6b9a5a JD |
1208 | } |
1209 | goto send_reply; | |
d3e2ba59 JD |
1210 | } else { |
1211 | viewer_index.status = htobe32(VIEWER_INDEX_OK); | |
1212 | vstream->last_sent_index++; | |
1213 | } | |
1214 | ||
1215 | /* | |
1216 | * Indexes are stored in big endian, no need to switch before sending. | |
1217 | */ | |
1218 | viewer_index.offset = packet_index.offset; | |
1219 | viewer_index.packet_size = packet_index.packet_size; | |
1220 | viewer_index.content_size = packet_index.content_size; | |
1221 | viewer_index.timestamp_begin = packet_index.timestamp_begin; | |
1222 | viewer_index.timestamp_end = packet_index.timestamp_end; | |
1223 | viewer_index.events_discarded = packet_index.events_discarded; | |
1224 | viewer_index.stream_id = packet_index.stream_id; | |
1225 | ||
1226 | send_reply: | |
1227 | viewer_index.flags = htobe32(viewer_index.flags); | |
eea7556c | 1228 | health_code_update(); |
d3e2ba59 JD |
1229 | ret = cmd->sock->ops->sendmsg(cmd->sock, &viewer_index, |
1230 | sizeof(viewer_index), 0); | |
1231 | if (ret < 0) { | |
1232 | ERR("Relay index to viewer"); | |
1233 | goto end_unlock; | |
1234 | } | |
eea7556c | 1235 | health_code_update(); |
d3e2ba59 JD |
1236 | |
1237 | DBG("Index %" PRIu64 "for stream %" PRIu64 "sent", | |
1238 | vstream->last_sent_index, vstream->stream_handle); | |
1239 | ||
1240 | end_unlock: | |
1241 | rcu_read_unlock(); | |
1242 | ||
1243 | end_no_session: | |
1244 | end: | |
1245 | return ret; | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * Send the next index for a stream | |
1250 | * | |
1251 | * Return 0 on success or else a negative value. | |
1252 | */ | |
1253 | static | |
92c6ca54 | 1254 | int viewer_get_packet(struct relay_command *cmd) |
d3e2ba59 JD |
1255 | { |
1256 | int ret, send_data = 0; | |
1257 | char *data = NULL; | |
1258 | uint32_t len = 0; | |
1259 | ssize_t read_len; | |
1260 | struct lttng_viewer_get_packet get_packet_info; | |
1261 | struct lttng_viewer_trace_packet reply; | |
1262 | struct relay_viewer_stream *stream; | |
1263 | ||
1264 | assert(cmd); | |
d3e2ba59 JD |
1265 | |
1266 | DBG2("Relay get data packet"); | |
1267 | ||
1268 | if (cmd->version_check_done == 0) { | |
1269 | ERR("Trying to get packet before version check"); | |
1270 | ret = -1; | |
1271 | goto end; | |
1272 | } | |
1273 | ||
eea7556c | 1274 | health_code_update(); |
d3e2ba59 JD |
1275 | ret = cmd->sock->ops->recvmsg(cmd->sock, &get_packet_info, |
1276 | sizeof(get_packet_info), 0); | |
1277 | if (ret < 0 || ret != sizeof(get_packet_info)) { | |
1278 | ret = -1; | |
1279 | ERR("Relay didn't receive the whole packet"); | |
1280 | goto end; | |
1281 | } | |
eea7556c | 1282 | health_code_update(); |
d3e2ba59 | 1283 | |
0233a6a5 DG |
1284 | /* From this point on, the error label can be reached. */ |
1285 | memset(&reply, 0, sizeof(reply)); | |
1286 | ||
d3e2ba59 | 1287 | rcu_read_lock(); |
92c6ca54 | 1288 | stream = live_find_viewer_stream_by_id(be64toh(get_packet_info.stream_id)); |
d3e2ba59 JD |
1289 | if (!stream) { |
1290 | goto error; | |
1291 | } | |
1292 | assert(stream->ctf_trace); | |
1293 | ||
1294 | /* | |
1295 | * First time we read this stream, we need open the tracefile, we should | |
1296 | * only arrive here if an index has already been sent to the viewer, so the | |
1297 | * tracefile must exist, if it does not it is a fatal error. | |
1298 | */ | |
1299 | if (stream->read_fd < 0) { | |
1300 | char fullpath[PATH_MAX]; | |
1301 | ||
6b6b9a5a JD |
1302 | if (stream->tracefile_count > 0) { |
1303 | ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name, | |
1304 | stream->channel_name, | |
1305 | stream->tracefile_count_current); | |
1306 | } else { | |
1307 | ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name, | |
1308 | stream->channel_name); | |
1309 | } | |
d3e2ba59 JD |
1310 | if (ret < 0) { |
1311 | goto error; | |
1312 | } | |
1313 | ret = open(fullpath, O_RDONLY); | |
1314 | if (ret < 0) { | |
1315 | PERROR("Relay opening trace file"); | |
1316 | goto error; | |
1317 | } | |
1318 | stream->read_fd = ret; | |
1319 | } | |
1320 | ||
d3e2ba59 JD |
1321 | if (!stream->ctf_trace->metadata_received || |
1322 | stream->ctf_trace->metadata_received > | |
1323 | stream->ctf_trace->metadata_sent) { | |
1324 | reply.status = htobe32(VIEWER_GET_PACKET_ERR); | |
1325 | reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA; | |
d3e2ba59 JD |
1326 | goto send_reply; |
1327 | } | |
1328 | ||
1329 | len = be32toh(get_packet_info.len); | |
1330 | data = zmalloc(len); | |
1331 | if (!data) { | |
1332 | PERROR("relay data zmalloc"); | |
1333 | goto error; | |
1334 | } | |
1335 | ||
1336 | ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET); | |
1337 | if (ret < 0) { | |
6b6b9a5a JD |
1338 | /* |
1339 | * If the read fd was closed by the streaming side, the | |
1340 | * abort_flag will be set to 1, otherwise it is an error. | |
1341 | */ | |
1342 | if (stream->abort_flag == 0) { | |
1343 | PERROR("lseek"); | |
1344 | goto error; | |
1345 | } | |
1346 | reply.status = htobe32(VIEWER_GET_PACKET_EOF); | |
1347 | goto send_reply; | |
d3e2ba59 | 1348 | } |
6cd525e8 MD |
1349 | read_len = lttng_read(stream->read_fd, data, len); |
1350 | if (read_len < len) { | |
6b6b9a5a JD |
1351 | /* |
1352 | * If the read fd was closed by the streaming side, the | |
1353 | * abort_flag will be set to 1, otherwise it is an error. | |
1354 | */ | |
1355 | if (stream->abort_flag == 0) { | |
1356 | PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64, | |
1357 | stream->read_fd, | |
1358 | be64toh(get_packet_info.offset)); | |
1359 | goto error; | |
1360 | } else { | |
1361 | reply.status = htobe32(VIEWER_GET_PACKET_EOF); | |
1362 | goto send_reply; | |
1363 | } | |
d3e2ba59 JD |
1364 | } |
1365 | reply.status = htobe32(VIEWER_GET_PACKET_OK); | |
1366 | reply.len = htobe32(len); | |
1367 | send_data = 1; | |
1368 | goto send_reply; | |
1369 | ||
1370 | error: | |
1371 | reply.status = htobe32(VIEWER_GET_PACKET_ERR); | |
1372 | ||
1373 | send_reply: | |
1374 | reply.flags = htobe32(reply.flags); | |
eea7556c MD |
1375 | |
1376 | health_code_update(); | |
d3e2ba59 JD |
1377 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); |
1378 | if (ret < 0) { | |
1379 | ERR("Relay data header to viewer"); | |
1380 | goto end_unlock; | |
1381 | } | |
eea7556c | 1382 | health_code_update(); |
d3e2ba59 JD |
1383 | |
1384 | if (send_data) { | |
eea7556c | 1385 | health_code_update(); |
d3e2ba59 JD |
1386 | ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0); |
1387 | if (ret < 0) { | |
1388 | ERR("Relay send data to viewer"); | |
1389 | goto end_unlock; | |
1390 | } | |
eea7556c | 1391 | health_code_update(); |
d3e2ba59 JD |
1392 | } |
1393 | ||
1394 | DBG("Sent %u bytes for stream %" PRIu64, len, | |
1395 | be64toh(get_packet_info.stream_id)); | |
1396 | ||
1397 | end_unlock: | |
1398 | free(data); | |
1399 | rcu_read_unlock(); | |
1400 | ||
1401 | end: | |
1402 | return ret; | |
1403 | } | |
1404 | ||
1405 | /* | |
1406 | * Send the session's metadata | |
1407 | * | |
1408 | * Return 0 on success else a negative value. | |
1409 | */ | |
1410 | static | |
92c6ca54 | 1411 | int viewer_get_metadata(struct relay_command *cmd) |
d3e2ba59 JD |
1412 | { |
1413 | int ret = 0; | |
1414 | ssize_t read_len; | |
1415 | uint64_t len = 0; | |
1416 | char *data = NULL; | |
1417 | struct lttng_viewer_get_metadata request; | |
1418 | struct lttng_viewer_metadata_packet reply; | |
1419 | struct relay_viewer_stream *stream; | |
1420 | ||
1421 | assert(cmd); | |
d3e2ba59 JD |
1422 | |
1423 | DBG("Relay get metadata"); | |
1424 | ||
1425 | if (cmd->version_check_done == 0) { | |
1426 | ERR("Trying to get metadata before version check"); | |
1427 | ret = -1; | |
1428 | goto end; | |
1429 | } | |
1430 | ||
eea7556c | 1431 | health_code_update(); |
d3e2ba59 JD |
1432 | ret = cmd->sock->ops->recvmsg(cmd->sock, &request, |
1433 | sizeof(request), 0); | |
1434 | if (ret < 0 || ret != sizeof(request)) { | |
1435 | ret = -1; | |
1436 | ERR("Relay didn't receive the whole packet"); | |
1437 | goto end; | |
1438 | } | |
eea7556c | 1439 | health_code_update(); |
d3e2ba59 JD |
1440 | |
1441 | rcu_read_lock(); | |
92c6ca54 | 1442 | stream = live_find_viewer_stream_by_id(be64toh(request.stream_id)); |
d3e2ba59 JD |
1443 | if (!stream || !stream->metadata_flag) { |
1444 | ERR("Invalid metadata stream"); | |
1445 | goto error; | |
1446 | } | |
1447 | assert(stream->ctf_trace); | |
1448 | assert(stream->ctf_trace->metadata_sent <= | |
1449 | stream->ctf_trace->metadata_received); | |
1450 | ||
1451 | len = stream->ctf_trace->metadata_received - | |
1452 | stream->ctf_trace->metadata_sent; | |
1453 | if (len == 0) { | |
1454 | reply.status = htobe32(VIEWER_NO_NEW_METADATA); | |
1455 | goto send_reply; | |
1456 | } | |
1457 | ||
1458 | /* first time, we open the metadata file */ | |
1459 | if (stream->read_fd < 0) { | |
1460 | char fullpath[PATH_MAX]; | |
1461 | ||
1462 | ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name, | |
1463 | stream->channel_name); | |
1464 | if (ret < 0) { | |
1465 | goto error; | |
1466 | } | |
1467 | ret = open(fullpath, O_RDONLY); | |
1468 | if (ret < 0) { | |
1469 | PERROR("Relay opening metadata file"); | |
1470 | goto error; | |
1471 | } | |
1472 | stream->read_fd = ret; | |
1473 | } | |
1474 | ||
1475 | reply.len = htobe64(len); | |
1476 | data = zmalloc(len); | |
1477 | if (!data) { | |
1478 | PERROR("viewer metadata zmalloc"); | |
1479 | goto error; | |
1480 | } | |
1481 | ||
6cd525e8 MD |
1482 | read_len = lttng_read(stream->read_fd, data, len); |
1483 | if (read_len < len) { | |
d3e2ba59 JD |
1484 | PERROR("Relay reading metadata file"); |
1485 | goto error; | |
1486 | } | |
1487 | stream->ctf_trace->metadata_sent += read_len; | |
1488 | reply.status = htobe32(VIEWER_METADATA_OK); | |
1489 | goto send_reply; | |
1490 | ||
1491 | error: | |
1492 | reply.status = htobe32(VIEWER_METADATA_ERR); | |
1493 | ||
1494 | send_reply: | |
eea7556c | 1495 | health_code_update(); |
d3e2ba59 JD |
1496 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); |
1497 | if (ret < 0) { | |
1498 | ERR("Relay data header to viewer"); | |
1499 | goto end_unlock; | |
1500 | } | |
eea7556c | 1501 | health_code_update(); |
d3e2ba59 JD |
1502 | |
1503 | if (len > 0) { | |
1504 | ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0); | |
1505 | if (ret < 0) { | |
1506 | ERR("Relay send data to viewer"); | |
1507 | goto end_unlock; | |
1508 | } | |
1509 | } | |
1510 | ||
1511 | DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len, | |
1512 | be64toh(request.stream_id)); | |
1513 | ||
1514 | DBG("Metadata sent"); | |
1515 | ||
1516 | end_unlock: | |
1517 | free(data); | |
1518 | rcu_read_unlock(); | |
1519 | end: | |
1520 | return ret; | |
1521 | } | |
1522 | ||
1523 | /* | |
1524 | * live_relay_unknown_command: send -1 if received unknown command | |
1525 | */ | |
1526 | static | |
1527 | void live_relay_unknown_command(struct relay_command *cmd) | |
1528 | { | |
1529 | struct lttcomm_relayd_generic_reply reply; | |
1530 | int ret; | |
1531 | ||
1532 | reply.ret_code = htobe32(LTTNG_ERR_UNK); | |
1533 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1534 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1535 | if (ret < 0) { | |
1536 | ERR("Relay sending unknown command"); | |
1537 | } | |
1538 | } | |
1539 | ||
1540 | /* | |
1541 | * Process the commands received on the control socket | |
1542 | */ | |
1543 | static | |
1544 | int process_control(struct lttng_viewer_cmd *recv_hdr, | |
92c6ca54 | 1545 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
d3e2ba59 JD |
1546 | { |
1547 | int ret = 0; | |
1548 | ||
1549 | switch (be32toh(recv_hdr->cmd)) { | |
1550 | case VIEWER_CONNECT: | |
1551 | ret = viewer_connect(cmd); | |
1552 | break; | |
1553 | case VIEWER_LIST_SESSIONS: | |
1554 | ret = viewer_list_sessions(cmd, sessions_ht); | |
1555 | break; | |
1556 | case VIEWER_ATTACH_SESSION: | |
92c6ca54 | 1557 | ret = viewer_attach_session(cmd, sessions_ht); |
d3e2ba59 JD |
1558 | break; |
1559 | case VIEWER_GET_NEXT_INDEX: | |
92c6ca54 | 1560 | ret = viewer_get_next_index(cmd, sessions_ht); |
d3e2ba59 JD |
1561 | break; |
1562 | case VIEWER_GET_PACKET: | |
92c6ca54 | 1563 | ret = viewer_get_packet(cmd); |
d3e2ba59 JD |
1564 | break; |
1565 | case VIEWER_GET_METADATA: | |
92c6ca54 | 1566 | ret = viewer_get_metadata(cmd); |
d3e2ba59 JD |
1567 | break; |
1568 | default: | |
1569 | ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd)); | |
1570 | live_relay_unknown_command(cmd); | |
1571 | ret = -1; | |
1572 | goto end; | |
1573 | } | |
1574 | ||
1575 | end: | |
1576 | return ret; | |
1577 | } | |
1578 | ||
1579 | static | |
1580 | void cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) | |
1581 | { | |
1582 | int ret; | |
1583 | ||
1584 | assert(events); | |
1585 | ||
1586 | lttng_poll_del(events, pollfd); | |
1587 | ||
1588 | ret = close(pollfd); | |
1589 | if (ret < 0) { | |
1590 | ERR("Closing pollfd %d", pollfd); | |
1591 | } | |
1592 | } | |
1593 | ||
1594 | /* | |
1595 | * Create and add connection to the given hash table. | |
1596 | * | |
1597 | * Return poll add value or else -1 on error. | |
1598 | */ | |
1599 | static | |
1600 | int add_connection(int fd, struct lttng_poll_event *events, | |
1601 | struct lttng_ht *relay_connections_ht) | |
1602 | { | |
1603 | int ret; | |
1604 | struct relay_command *relay_connection; | |
1605 | ||
1606 | assert(events); | |
1607 | assert(relay_connections_ht); | |
1608 | ||
1609 | relay_connection = zmalloc(sizeof(struct relay_command)); | |
1610 | if (relay_connection == NULL) { | |
1611 | PERROR("Relay command zmalloc"); | |
1612 | goto error; | |
1613 | } | |
1614 | ||
6cd525e8 MD |
1615 | ret = lttng_read(fd, relay_connection, sizeof(*relay_connection)); |
1616 | if (ret < sizeof(*relay_connection)) { | |
d3e2ba59 JD |
1617 | PERROR("read relay cmd pipe"); |
1618 | goto error_read; | |
1619 | } | |
1620 | ||
1621 | lttng_ht_node_init_ulong(&relay_connection->sock_n, | |
1622 | (unsigned long) relay_connection->sock->fd); | |
1623 | rcu_read_lock(); | |
1624 | lttng_ht_add_unique_ulong(relay_connections_ht, | |
1625 | &relay_connection->sock_n); | |
1626 | rcu_read_unlock(); | |
1627 | ||
1628 | return lttng_poll_add(events, relay_connection->sock->fd, | |
1629 | LPOLLIN | LPOLLRDHUP); | |
1630 | ||
1631 | error_read: | |
1632 | free(relay_connection); | |
1633 | error: | |
1634 | return -1; | |
1635 | } | |
1636 | ||
1637 | static | |
1638 | void deferred_free_connection(struct rcu_head *head) | |
1639 | { | |
1640 | struct relay_command *relay_connection = | |
1641 | caa_container_of(head, struct relay_command, rcu_node); | |
1642 | ||
1643 | if (relay_connection->session && | |
1644 | relay_connection->session->viewer_attached > 0) { | |
1645 | relay_connection->session->viewer_attached--; | |
1646 | } | |
1647 | lttcomm_destroy_sock(relay_connection->sock); | |
1648 | free(relay_connection); | |
1649 | } | |
1650 | ||
1651 | static | |
1652 | void deferred_free_viewer_stream(struct rcu_head *head) | |
1653 | { | |
1654 | struct relay_viewer_stream *stream = | |
1655 | caa_container_of(head, struct relay_viewer_stream, rcu_node); | |
1656 | ||
1657 | if (stream->ctf_trace) { | |
1658 | uatomic_dec(&stream->ctf_trace->refcount); | |
1659 | assert(uatomic_read(&stream->ctf_trace->refcount) >= 0); | |
1660 | if (uatomic_read(&stream->ctf_trace->refcount) == 0) { | |
1661 | DBG("Freeing ctf_trace %" PRIu64, stream->ctf_trace->id); | |
1662 | free(stream->ctf_trace); | |
1663 | } | |
1664 | } | |
1665 | ||
1666 | free(stream->path_name); | |
1667 | free(stream->channel_name); | |
1668 | free(stream); | |
1669 | } | |
1670 | ||
1671 | static | |
b92fdc2b | 1672 | void viewer_del_streams(uint64_t session_id) |
d3e2ba59 JD |
1673 | { |
1674 | int ret; | |
1675 | struct relay_viewer_stream *stream; | |
1676 | struct lttng_ht_node_u64 *node; | |
1677 | struct lttng_ht_iter iter; | |
1678 | ||
d3e2ba59 JD |
1679 | rcu_read_lock(); |
1680 | cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, node, node) { | |
eea7556c MD |
1681 | health_code_update(); |
1682 | ||
d3e2ba59 JD |
1683 | node = lttng_ht_iter_get_node_u64(&iter); |
1684 | if (!node) { | |
1685 | continue; | |
1686 | } | |
1687 | ||
1688 | stream = caa_container_of(node, struct relay_viewer_stream, stream_n); | |
b92fdc2b | 1689 | if (stream->session_id != session_id) { |
d3e2ba59 JD |
1690 | continue; |
1691 | } | |
1692 | ||
6b6b9a5a | 1693 | if (stream->read_fd >= 0) { |
d3e2ba59 JD |
1694 | ret = close(stream->read_fd); |
1695 | if (ret < 0) { | |
1696 | PERROR("close read_fd"); | |
1697 | } | |
1698 | } | |
6b6b9a5a | 1699 | if (stream->index_read_fd >= 0) { |
d3e2ba59 JD |
1700 | ret = close(stream->index_read_fd); |
1701 | if (ret < 0) { | |
1702 | PERROR("close index_read_fd"); | |
1703 | } | |
1704 | } | |
1705 | if (stream->metadata_flag && stream->ctf_trace) { | |
1706 | stream->ctf_trace->metadata_sent = 0; | |
1707 | } | |
1708 | ret = lttng_ht_del(viewer_streams_ht, &iter); | |
1709 | assert(!ret); | |
1710 | call_rcu(&stream->rcu_node, deferred_free_viewer_stream); | |
1711 | } | |
1712 | rcu_read_unlock(); | |
1713 | } | |
1714 | ||
1715 | /* | |
1716 | * Delete and free a connection. | |
1717 | * | |
1718 | * RCU read side lock MUST be acquired. | |
1719 | */ | |
1720 | static | |
1721 | void del_connection(struct lttng_ht *relay_connections_ht, | |
92c6ca54 | 1722 | struct lttng_ht_iter *iter, struct relay_command *relay_connection) |
d3e2ba59 JD |
1723 | { |
1724 | int ret; | |
1725 | ||
1726 | assert(relay_connections_ht); | |
1727 | assert(iter); | |
1728 | assert(relay_connection); | |
d3e2ba59 JD |
1729 | |
1730 | ret = lttng_ht_del(relay_connections_ht, iter); | |
1731 | assert(!ret); | |
1732 | ||
b92fdc2b | 1733 | viewer_del_streams(relay_connection->session_id); |
d3e2ba59 JD |
1734 | |
1735 | call_rcu(&relay_connection->rcu_node, deferred_free_connection); | |
1736 | } | |
1737 | ||
1738 | /* | |
1739 | * This thread does the actual work | |
1740 | */ | |
1741 | static | |
1742 | void *thread_worker(void *data) | |
1743 | { | |
1744 | int ret, err = -1; | |
1745 | uint32_t nb_fd; | |
1746 | struct relay_command *relay_connection; | |
1747 | struct lttng_poll_event events; | |
1748 | struct lttng_ht *relay_connections_ht; | |
1749 | struct lttng_ht_node_ulong *node; | |
1750 | struct lttng_ht_iter iter; | |
1751 | struct lttng_viewer_cmd recv_hdr; | |
1752 | struct relay_local_data *relay_ctx = (struct relay_local_data *) data; | |
1753 | struct lttng_ht *sessions_ht = relay_ctx->sessions_ht; | |
d3e2ba59 JD |
1754 | |
1755 | DBG("[thread] Live viewer relay worker started"); | |
1756 | ||
1757 | rcu_register_thread(); | |
1758 | ||
eea7556c MD |
1759 | health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER); |
1760 | ||
d3e2ba59 JD |
1761 | /* table of connections indexed on socket */ |
1762 | relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
1763 | if (!relay_connections_ht) { | |
1764 | goto relay_connections_ht_error; | |
1765 | } | |
1766 | ||
1767 | ret = create_thread_poll_set(&events, 2); | |
1768 | if (ret < 0) { | |
1769 | goto error_poll_create; | |
1770 | } | |
1771 | ||
1772 | ret = lttng_poll_add(&events, live_relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); | |
1773 | if (ret < 0) { | |
1774 | goto error; | |
1775 | } | |
1776 | ||
1777 | restart: | |
1778 | while (1) { | |
1779 | int i; | |
1780 | ||
eea7556c MD |
1781 | health_code_update(); |
1782 | ||
d3e2ba59 JD |
1783 | /* Infinite blocking call, waiting for transmission */ |
1784 | DBG3("Relayd live viewer worker thread polling..."); | |
eea7556c | 1785 | health_poll_entry(); |
d3e2ba59 | 1786 | ret = lttng_poll_wait(&events, -1); |
eea7556c | 1787 | health_poll_exit(); |
d3e2ba59 JD |
1788 | if (ret < 0) { |
1789 | /* | |
1790 | * Restart interrupted system call. | |
1791 | */ | |
1792 | if (errno == EINTR) { | |
1793 | goto restart; | |
1794 | } | |
1795 | goto error; | |
1796 | } | |
1797 | ||
1798 | nb_fd = ret; | |
1799 | ||
1800 | /* | |
1801 | * Process control. The control connection is prioritised so we don't | |
1802 | * starve it with high throughput tracing data on the data | |
1803 | * connection. | |
1804 | */ | |
1805 | for (i = 0; i < nb_fd; i++) { | |
1806 | /* Fetch once the poll data */ | |
1807 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); | |
1808 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
1809 | ||
eea7556c MD |
1810 | health_code_update(); |
1811 | ||
d3e2ba59 JD |
1812 | /* Thread quit pipe has been closed. Killing thread. */ |
1813 | ret = check_thread_quit_pipe(pollfd, revents); | |
1814 | if (ret) { | |
1815 | err = 0; | |
1816 | goto exit; | |
1817 | } | |
1818 | ||
1819 | /* Inspect the relay cmd pipe for new connection */ | |
1820 | if (pollfd == live_relay_cmd_pipe[0]) { | |
1821 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
1822 | ERR("Relay live pipe error"); | |
1823 | goto error; | |
1824 | } else if (revents & LPOLLIN) { | |
1825 | DBG("Relay live viewer command received"); | |
1826 | ret = add_connection(live_relay_cmd_pipe[0], | |
1827 | &events, relay_connections_ht); | |
1828 | if (ret < 0) { | |
1829 | goto error; | |
1830 | } | |
1831 | } | |
1832 | } else if (revents) { | |
1833 | rcu_read_lock(); | |
1834 | lttng_ht_lookup(relay_connections_ht, | |
1835 | (void *)((unsigned long) pollfd), &iter); | |
1836 | node = lttng_ht_iter_get_node_ulong(&iter); | |
1837 | if (node == NULL) { | |
1838 | DBG2("Relay viewer sock %d not found", pollfd); | |
1839 | rcu_read_unlock(); | |
1840 | goto error; | |
1841 | } | |
1842 | relay_connection = caa_container_of(node, struct relay_command, | |
1843 | sock_n); | |
1844 | ||
1845 | if (revents & (LPOLLERR)) { | |
d3e2ba59 JD |
1846 | cleanup_poll_connection(&events, pollfd); |
1847 | del_connection(relay_connections_ht, &iter, | |
92c6ca54 | 1848 | relay_connection); |
d3e2ba59 JD |
1849 | } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { |
1850 | DBG("Viewer socket %d hung up", pollfd); | |
1851 | cleanup_poll_connection(&events, pollfd); | |
1852 | del_connection(relay_connections_ht, &iter, | |
92c6ca54 | 1853 | relay_connection); |
d3e2ba59 JD |
1854 | } else if (revents & LPOLLIN) { |
1855 | ret = relay_connection->sock->ops->recvmsg( | |
1856 | relay_connection->sock, &recv_hdr, | |
1857 | sizeof(struct lttng_viewer_cmd), | |
1858 | 0); | |
1859 | /* connection closed */ | |
1860 | if (ret <= 0) { | |
1861 | cleanup_poll_connection(&events, pollfd); | |
aaec7998 | 1862 | del_connection(relay_connections_ht, &iter, |
92c6ca54 | 1863 | relay_connection); |
d3e2ba59 JD |
1864 | DBG("Viewer control connection closed with %d", |
1865 | pollfd); | |
1866 | } else { | |
1867 | if (relay_connection->session) { | |
1868 | DBG2("Relay viewer worker receiving data for " | |
1869 | "session: %" PRIu64, | |
1870 | relay_connection->session->id); | |
1871 | } | |
1872 | ret = process_control(&recv_hdr, relay_connection, | |
92c6ca54 | 1873 | sessions_ht); |
d3e2ba59 JD |
1874 | if (ret < 0) { |
1875 | /* Clear the session on error. */ | |
1876 | cleanup_poll_connection(&events, pollfd); | |
1877 | del_connection(relay_connections_ht, &iter, | |
92c6ca54 | 1878 | relay_connection); |
d3e2ba59 JD |
1879 | DBG("Viewer connection closed with %d", pollfd); |
1880 | } | |
1881 | } | |
1882 | } | |
1883 | rcu_read_unlock(); | |
1884 | } | |
1885 | } | |
1886 | } | |
1887 | ||
1888 | exit: | |
1889 | error: | |
1890 | lttng_poll_clean(&events); | |
1891 | ||
1892 | /* empty the hash table and free the memory */ | |
1893 | rcu_read_lock(); | |
1894 | cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { | |
eea7556c MD |
1895 | health_code_update(); |
1896 | ||
d3e2ba59 JD |
1897 | node = lttng_ht_iter_get_node_ulong(&iter); |
1898 | if (!node) { | |
1899 | continue; | |
1900 | } | |
1901 | ||
1902 | relay_connection = caa_container_of(node, struct relay_command, | |
1903 | sock_n); | |
92c6ca54 | 1904 | del_connection(relay_connections_ht, &iter, relay_connection); |
d3e2ba59 JD |
1905 | } |
1906 | rcu_read_unlock(); | |
1907 | error_poll_create: | |
1908 | lttng_ht_destroy(relay_connections_ht); | |
1909 | relay_connections_ht_error: | |
1910 | /* Close relay cmd pipes */ | |
1911 | utils_close_pipe(live_relay_cmd_pipe); | |
1912 | if (err) { | |
1913 | DBG("Viewer worker thread exited with error"); | |
1914 | } | |
1915 | DBG("Viewer worker thread cleanup complete"); | |
eea7556c MD |
1916 | if (err) { |
1917 | health_error(); | |
1918 | ERR("Health error occurred in %s", __func__); | |
1919 | } | |
1920 | health_unregister(health_relayd); | |
d3e2ba59 JD |
1921 | stop_threads(); |
1922 | rcu_unregister_thread(); | |
1923 | return NULL; | |
1924 | } | |
1925 | ||
1926 | /* | |
1927 | * Create the relay command pipe to wake thread_manage_apps. | |
1928 | * Closed in cleanup(). | |
1929 | */ | |
1930 | static int create_relay_cmd_pipe(void) | |
1931 | { | |
1932 | int ret; | |
1933 | ||
1934 | ret = utils_create_pipe_cloexec(live_relay_cmd_pipe); | |
1935 | ||
1936 | return ret; | |
1937 | } | |
1938 | ||
aaec7998 | 1939 | void live_stop_threads(void) |
d3e2ba59 JD |
1940 | { |
1941 | int ret; | |
1942 | void *status; | |
1943 | ||
1944 | stop_threads(); | |
1945 | ||
1946 | ret = pthread_join(live_listener_thread, &status); | |
1947 | if (ret != 0) { | |
1948 | PERROR("pthread_join live listener"); | |
1949 | goto error; /* join error, exit without cleanup */ | |
1950 | } | |
1951 | ||
1952 | ret = pthread_join(live_worker_thread, &status); | |
1953 | if (ret != 0) { | |
1954 | PERROR("pthread_join live worker"); | |
1955 | goto error; /* join error, exit without cleanup */ | |
1956 | } | |
1957 | ||
1958 | ret = pthread_join(live_dispatcher_thread, &status); | |
1959 | if (ret != 0) { | |
1960 | PERROR("pthread_join live dispatcher"); | |
1961 | goto error; /* join error, exit without cleanup */ | |
1962 | } | |
1963 | ||
1964 | cleanup(); | |
1965 | ||
1966 | error: | |
1967 | return; | |
1968 | } | |
1969 | ||
1970 | /* | |
1971 | * main | |
1972 | */ | |
1973 | int live_start_threads(struct lttng_uri *uri, | |
42415026 | 1974 | struct relay_local_data *relay_ctx, int quit_pipe[2]) |
d3e2ba59 JD |
1975 | { |
1976 | int ret = 0; | |
1977 | void *status; | |
1978 | int is_root; | |
1979 | ||
1980 | assert(uri); | |
1981 | live_uri = uri; | |
1982 | ||
42415026 DG |
1983 | live_thread_quit_pipe[0] = quit_pipe[0]; |
1984 | live_thread_quit_pipe[1] = quit_pipe[1]; | |
d3e2ba59 JD |
1985 | |
1986 | /* Check if daemon is UID = 0 */ | |
1987 | is_root = !getuid(); | |
1988 | ||
1989 | if (!is_root) { | |
1990 | if (live_uri->port < 1024) { | |
1991 | ERR("Need to be root to use ports < 1024"); | |
1992 | ret = -1; | |
1993 | goto exit; | |
1994 | } | |
1995 | } | |
1996 | ||
1997 | /* Setup the thread apps communication pipe. */ | |
1998 | if ((ret = create_relay_cmd_pipe()) < 0) { | |
1999 | goto exit; | |
2000 | } | |
2001 | ||
2002 | /* Init relay command queue. */ | |
2003 | cds_wfq_init(&viewer_cmd_queue.queue); | |
2004 | ||
2005 | /* Set up max poll set size */ | |
2006 | lttng_poll_set_max_size(); | |
2007 | ||
2008 | /* Setup the dispatcher thread */ | |
2009 | ret = pthread_create(&live_dispatcher_thread, NULL, | |
2010 | thread_dispatcher, (void *) NULL); | |
2011 | if (ret != 0) { | |
2012 | PERROR("pthread_create viewer dispatcher"); | |
2013 | goto exit_dispatcher; | |
2014 | } | |
2015 | ||
2016 | /* Setup the worker thread */ | |
2017 | ret = pthread_create(&live_worker_thread, NULL, | |
2018 | thread_worker, relay_ctx); | |
2019 | if (ret != 0) { | |
2020 | PERROR("pthread_create viewer worker"); | |
2021 | goto exit_worker; | |
2022 | } | |
2023 | ||
2024 | /* Setup the listener thread */ | |
2025 | ret = pthread_create(&live_listener_thread, NULL, | |
2026 | thread_listener, (void *) NULL); | |
2027 | if (ret != 0) { | |
2028 | PERROR("pthread_create viewer listener"); | |
2029 | goto exit_listener; | |
2030 | } | |
2031 | ||
2032 | ret = 0; | |
2033 | goto end; | |
2034 | ||
2035 | exit_listener: | |
2036 | ret = pthread_join(live_listener_thread, &status); | |
2037 | if (ret != 0) { | |
2038 | PERROR("pthread_join live listener"); | |
2039 | goto error; /* join error, exit without cleanup */ | |
2040 | } | |
2041 | ||
2042 | exit_worker: | |
2043 | ret = pthread_join(live_worker_thread, &status); | |
2044 | if (ret != 0) { | |
2045 | PERROR("pthread_join live worker"); | |
2046 | goto error; /* join error, exit without cleanup */ | |
2047 | } | |
2048 | ||
2049 | exit_dispatcher: | |
2050 | ret = pthread_join(live_dispatcher_thread, &status); | |
2051 | if (ret != 0) { | |
2052 | PERROR("pthread_join live dispatcher"); | |
2053 | goto error; /* join error, exit without cleanup */ | |
2054 | } | |
2055 | ||
2056 | exit: | |
2057 | cleanup(); | |
2058 | ||
2059 | end: | |
2060 | error: | |
2061 | return ret; | |
2062 | } |