Commit | Line | Data |
---|---|---|
b8aa1682 JD |
1 | /* |
2 | * Copyright (C) 2012 - 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> | |
173af62f | 35 | #include <inttypes.h> |
b8aa1682 JD |
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> | |
b8aa1682 JD |
50 | #include <common/sessiond-comm/relayd.h> |
51 | #include <common/uri.h> | |
a02de639 | 52 | #include <common/utils.h> |
b8aa1682 | 53 | |
0f907de1 JD |
54 | #include "cmd.h" |
55 | #include "utils.h" | |
b8aa1682 JD |
56 | #include "lttng-relayd.h" |
57 | ||
58 | /* command line options */ | |
0f907de1 | 59 | char *opt_output_path; |
b8aa1682 | 60 | static int opt_daemon; |
095a4ae5 MD |
61 | static struct lttng_uri *control_uri; |
62 | static struct lttng_uri *data_uri; | |
b8aa1682 JD |
63 | |
64 | const char *progname; | |
65 | static int is_root; /* Set to 1 if the daemon is running as root */ | |
66 | ||
67 | /* | |
68 | * Quit pipe for all threads. This permits a single cancellation point | |
69 | * for all threads when receiving an event on the pipe. | |
70 | */ | |
71 | static int thread_quit_pipe[2] = { -1, -1 }; | |
72 | ||
73 | /* | |
74 | * This pipe is used to inform the worker thread that a command is queued and | |
75 | * ready to be processed. | |
76 | */ | |
77 | static int relay_cmd_pipe[2] = { -1, -1 }; | |
78 | ||
26c9d55e | 79 | /* Shared between threads */ |
b8aa1682 JD |
80 | static int dispatch_thread_exit; |
81 | ||
82 | static pthread_t listener_thread; | |
83 | static pthread_t dispatcher_thread; | |
84 | static pthread_t worker_thread; | |
85 | ||
095a4ae5 MD |
86 | static uint64_t last_relay_stream_id; |
87 | static uint64_t last_relay_session_id; | |
b8aa1682 JD |
88 | |
89 | /* | |
90 | * Relay command queue. | |
91 | * | |
92 | * The relay_thread_listener and relay_thread_dispatcher communicate with this | |
93 | * queue. | |
94 | */ | |
95 | static struct relay_cmd_queue relay_cmd_queue; | |
96 | ||
97 | /* buffer allocated at startup, used to store the trace data */ | |
095a4ae5 MD |
98 | static char *data_buffer; |
99 | static unsigned int data_buffer_size; | |
b8aa1682 JD |
100 | |
101 | /* | |
102 | * usage function on stderr | |
103 | */ | |
104 | static | |
105 | void usage(void) | |
106 | { | |
107 | fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); | |
108 | fprintf(stderr, " -h, --help Display this usage.\n"); | |
109 | fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); | |
110 | fprintf(stderr, " -C, --control-port Control port listening (URI)\n"); | |
111 | fprintf(stderr, " -D, --data-port Data port listening (URI)\n"); | |
25672d02 | 112 | fprintf(stderr, " -o, --output Output path for traces (PATH)\n"); |
b8aa1682 JD |
113 | fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); |
114 | } | |
115 | ||
116 | static | |
117 | int parse_args(int argc, char **argv) | |
118 | { | |
119 | int c; | |
120 | int ret = 0; | |
121 | char *default_address; | |
122 | ||
123 | static struct option long_options[] = { | |
e3678fd8 MD |
124 | { "control-port", 1, 0, 'C', }, |
125 | { "data-port", 1, 0, 'D', }, | |
126 | { "daemonize", 0, 0, 'd', }, | |
127 | { "help", 0, 0, 'h', }, | |
128 | { "output", 1, 0, 'o', }, | |
129 | { "verbose", 0, 0, 'v', }, | |
095a4ae5 | 130 | { NULL, 0, 0, 0, }, |
b8aa1682 JD |
131 | }; |
132 | ||
133 | while (1) { | |
134 | int option_index = 0; | |
135 | c = getopt_long(argc, argv, "dhv" "C:D:o:", | |
136 | long_options, &option_index); | |
137 | if (c == -1) { | |
138 | break; | |
139 | } | |
140 | ||
141 | switch (c) { | |
142 | case 0: | |
143 | fprintf(stderr, "option %s", long_options[option_index].name); | |
144 | if (optarg) { | |
145 | fprintf(stderr, " with arg %s\n", optarg); | |
146 | } | |
147 | break; | |
148 | case 'C': | |
149 | ret = uri_parse(optarg, &control_uri); | |
150 | if (ret < 0) { | |
151 | ERR("Invalid control URI specified"); | |
152 | goto exit; | |
153 | } | |
154 | if (control_uri->port == 0) { | |
155 | control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; | |
156 | } | |
157 | break; | |
158 | case 'D': | |
159 | ret = uri_parse(optarg, &data_uri); | |
160 | if (ret < 0) { | |
161 | ERR("Invalid data URI specified"); | |
162 | goto exit; | |
163 | } | |
164 | if (data_uri->port == 0) { | |
165 | data_uri->port = DEFAULT_NETWORK_DATA_PORT; | |
166 | } | |
167 | break; | |
168 | case 'd': | |
169 | opt_daemon = 1; | |
170 | break; | |
171 | case 'h': | |
172 | usage(); | |
173 | exit(EXIT_FAILURE); | |
174 | case 'o': | |
175 | ret = asprintf(&opt_output_path, "%s", optarg); | |
176 | if (ret < 0) { | |
177 | PERROR("asprintf opt_output_path"); | |
178 | goto exit; | |
179 | } | |
180 | break; | |
181 | case 'v': | |
182 | /* Verbose level can increase using multiple -v */ | |
183 | lttng_opt_verbose += 1; | |
184 | break; | |
185 | default: | |
186 | /* Unknown option or other error. | |
187 | * Error is printed by getopt, just return */ | |
188 | ret = -1; | |
189 | goto exit; | |
190 | } | |
191 | } | |
192 | ||
193 | /* assign default values */ | |
194 | if (control_uri == NULL) { | |
195 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
196 | DEFAULT_NETWORK_CONTROL_PORT); | |
197 | if (ret < 0) { | |
198 | PERROR("asprintf default data address"); | |
199 | goto exit; | |
200 | } | |
201 | ||
202 | ret = uri_parse(default_address, &control_uri); | |
203 | free(default_address); | |
204 | if (ret < 0) { | |
205 | ERR("Invalid control URI specified"); | |
206 | goto exit; | |
207 | } | |
208 | } | |
209 | if (data_uri == NULL) { | |
210 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
211 | DEFAULT_NETWORK_DATA_PORT); | |
212 | if (ret < 0) { | |
213 | PERROR("asprintf default data address"); | |
214 | goto exit; | |
215 | } | |
216 | ||
217 | ret = uri_parse(default_address, &data_uri); | |
218 | free(default_address); | |
219 | if (ret < 0) { | |
220 | ERR("Invalid data URI specified"); | |
221 | goto exit; | |
222 | } | |
223 | } | |
224 | ||
225 | exit: | |
226 | return ret; | |
227 | } | |
228 | ||
229 | /* | |
230 | * Cleanup the daemon | |
231 | */ | |
232 | static | |
233 | void cleanup(void) | |
234 | { | |
b8aa1682 JD |
235 | DBG("Cleaning up"); |
236 | ||
095a4ae5 MD |
237 | /* free the dynamically allocated opt_output_path */ |
238 | free(opt_output_path); | |
239 | ||
a02de639 CB |
240 | /* Close thread quit pipes */ |
241 | utils_close_pipe(thread_quit_pipe); | |
242 | ||
710c1f73 DG |
243 | uri_free(control_uri); |
244 | uri_free(data_uri); | |
b8aa1682 JD |
245 | } |
246 | ||
247 | /* | |
248 | * Write to writable pipe used to notify a thread. | |
249 | */ | |
250 | static | |
251 | int notify_thread_pipe(int wpipe) | |
252 | { | |
253 | int ret; | |
254 | ||
6f94560a MD |
255 | do { |
256 | ret = write(wpipe, "!", 1); | |
257 | } while (ret < 0 && errno == EINTR); | |
4cec016f | 258 | if (ret < 0 || ret != 1) { |
b8aa1682 JD |
259 | PERROR("write poll pipe"); |
260 | } | |
261 | ||
262 | return ret; | |
263 | } | |
264 | ||
265 | /* | |
266 | * Stop all threads by closing the thread quit pipe. | |
267 | */ | |
268 | static | |
269 | void stop_threads(void) | |
270 | { | |
271 | int ret; | |
272 | ||
273 | /* Stopping all threads */ | |
274 | DBG("Terminating all threads"); | |
275 | ret = notify_thread_pipe(thread_quit_pipe[1]); | |
276 | if (ret < 0) { | |
277 | ERR("write error on thread quit pipe"); | |
278 | } | |
279 | ||
280 | /* Dispatch thread */ | |
26c9d55e | 281 | CMM_STORE_SHARED(dispatch_thread_exit, 1); |
b8aa1682 JD |
282 | futex_nto1_wake(&relay_cmd_queue.futex); |
283 | } | |
284 | ||
285 | /* | |
286 | * Signal handler for the daemon | |
287 | * | |
288 | * Simply stop all worker threads, leaving main() return gracefully after | |
289 | * joining all threads and calling cleanup(). | |
290 | */ | |
291 | static | |
292 | void sighandler(int sig) | |
293 | { | |
294 | switch (sig) { | |
295 | case SIGPIPE: | |
296 | DBG("SIGPIPE caught"); | |
297 | return; | |
298 | case SIGINT: | |
299 | DBG("SIGINT caught"); | |
300 | stop_threads(); | |
301 | break; | |
302 | case SIGTERM: | |
303 | DBG("SIGTERM caught"); | |
304 | stop_threads(); | |
305 | break; | |
306 | default: | |
307 | break; | |
308 | } | |
309 | } | |
310 | ||
311 | /* | |
312 | * Setup signal handler for : | |
313 | * SIGINT, SIGTERM, SIGPIPE | |
314 | */ | |
315 | static | |
316 | int set_signal_handler(void) | |
317 | { | |
318 | int ret = 0; | |
319 | struct sigaction sa; | |
320 | sigset_t sigset; | |
321 | ||
322 | if ((ret = sigemptyset(&sigset)) < 0) { | |
323 | PERROR("sigemptyset"); | |
324 | return ret; | |
325 | } | |
326 | ||
327 | sa.sa_handler = sighandler; | |
328 | sa.sa_mask = sigset; | |
329 | sa.sa_flags = 0; | |
330 | if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) { | |
331 | PERROR("sigaction"); | |
332 | return ret; | |
333 | } | |
334 | ||
335 | if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) { | |
336 | PERROR("sigaction"); | |
337 | return ret; | |
338 | } | |
339 | ||
340 | if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { | |
341 | PERROR("sigaction"); | |
342 | return ret; | |
343 | } | |
344 | ||
345 | DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT"); | |
346 | ||
347 | return ret; | |
348 | } | |
349 | ||
350 | /* | |
351 | * Init thread quit pipe. | |
352 | * | |
353 | * Return -1 on error or 0 if all pipes are created. | |
354 | */ | |
355 | static | |
356 | int init_thread_quit_pipe(void) | |
357 | { | |
a02de639 | 358 | int ret; |
b8aa1682 | 359 | |
a02de639 | 360 | ret = utils_create_pipe_cloexec(thread_quit_pipe); |
b8aa1682 | 361 | |
b8aa1682 JD |
362 | return ret; |
363 | } | |
364 | ||
365 | /* | |
366 | * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. | |
367 | */ | |
368 | static | |
369 | int create_thread_poll_set(struct lttng_poll_event *events, int size) | |
370 | { | |
371 | int ret; | |
372 | ||
373 | if (events == NULL || size == 0) { | |
374 | ret = -1; | |
375 | goto error; | |
376 | } | |
377 | ||
378 | ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); | |
379 | if (ret < 0) { | |
380 | goto error; | |
381 | } | |
382 | ||
383 | /* Add quit pipe */ | |
384 | ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN); | |
385 | if (ret < 0) { | |
386 | goto error; | |
387 | } | |
388 | ||
389 | return 0; | |
390 | ||
391 | error: | |
392 | return ret; | |
393 | } | |
394 | ||
395 | /* | |
396 | * Check if the thread quit pipe was triggered. | |
397 | * | |
398 | * Return 1 if it was triggered else 0; | |
399 | */ | |
400 | static | |
401 | int check_thread_quit_pipe(int fd, uint32_t events) | |
402 | { | |
403 | if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { | |
404 | return 1; | |
405 | } | |
406 | ||
407 | return 0; | |
408 | } | |
409 | ||
410 | /* | |
411 | * Create and init socket from uri. | |
412 | */ | |
413 | static | |
414 | struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri) | |
415 | { | |
416 | int ret; | |
417 | struct lttcomm_sock *sock = NULL; | |
418 | ||
419 | sock = lttcomm_alloc_sock_from_uri(uri); | |
420 | if (sock == NULL) { | |
421 | ERR("Allocating socket"); | |
422 | goto error; | |
423 | } | |
424 | ||
425 | ret = lttcomm_create_sock(sock); | |
426 | if (ret < 0) { | |
427 | goto error; | |
428 | } | |
429 | DBG("Listening on sock %d", sock->fd); | |
430 | ||
431 | ret = sock->ops->bind(sock); | |
432 | if (ret < 0) { | |
433 | goto error; | |
434 | } | |
435 | ||
436 | ret = sock->ops->listen(sock, -1); | |
437 | if (ret < 0) { | |
438 | goto error; | |
439 | ||
440 | } | |
441 | ||
442 | return sock; | |
443 | ||
444 | error: | |
445 | if (sock) { | |
446 | lttcomm_destroy_sock(sock); | |
447 | } | |
448 | return NULL; | |
449 | } | |
450 | ||
173af62f DG |
451 | /* |
452 | * Return nonzero if stream needs to be closed. | |
453 | */ | |
454 | static | |
455 | int close_stream_check(struct relay_stream *stream) | |
456 | { | |
457 | ||
458 | if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) { | |
f7079f67 DG |
459 | /* |
460 | * We are about to close the stream so set the data pending flag to 1 | |
461 | * which will make the end data pending command skip the stream which | |
462 | * is now closed and ready. Note that after proceeding to a file close, | |
463 | * the written file is ready for reading. | |
464 | */ | |
465 | stream->data_pending_check_done = 1; | |
173af62f DG |
466 | return 1; |
467 | } | |
468 | return 0; | |
469 | } | |
470 | ||
b8aa1682 JD |
471 | /* |
472 | * This thread manages the listening for new connections on the network | |
473 | */ | |
474 | static | |
475 | void *relay_thread_listener(void *data) | |
476 | { | |
095a4ae5 | 477 | int i, ret, pollfd, err = -1; |
b8aa1682 JD |
478 | int val = 1; |
479 | uint32_t revents, nb_fd; | |
480 | struct lttng_poll_event events; | |
481 | struct lttcomm_sock *control_sock, *data_sock; | |
482 | ||
b8aa1682 JD |
483 | DBG("[thread] Relay listener started"); |
484 | ||
485 | control_sock = relay_init_sock(control_uri); | |
486 | if (!control_sock) { | |
095a4ae5 | 487 | goto error_sock_control; |
b8aa1682 JD |
488 | } |
489 | ||
490 | data_sock = relay_init_sock(data_uri); | |
491 | if (!data_sock) { | |
095a4ae5 | 492 | goto error_sock_relay; |
b8aa1682 JD |
493 | } |
494 | ||
495 | /* | |
496 | * Pass 3 as size here for the thread quit pipe, control and data socket. | |
497 | */ | |
498 | ret = create_thread_poll_set(&events, 3); | |
499 | if (ret < 0) { | |
500 | goto error_create_poll; | |
501 | } | |
502 | ||
503 | /* Add the control socket */ | |
504 | ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP); | |
505 | if (ret < 0) { | |
506 | goto error_poll_add; | |
507 | } | |
508 | ||
509 | /* Add the data socket */ | |
510 | ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP); | |
511 | if (ret < 0) { | |
512 | goto error_poll_add; | |
513 | } | |
514 | ||
515 | while (1) { | |
516 | DBG("Listener accepting connections"); | |
517 | ||
b8aa1682 JD |
518 | restart: |
519 | ret = lttng_poll_wait(&events, -1); | |
520 | if (ret < 0) { | |
521 | /* | |
522 | * Restart interrupted system call. | |
523 | */ | |
524 | if (errno == EINTR) { | |
525 | goto restart; | |
526 | } | |
527 | goto error; | |
528 | } | |
529 | ||
0d9c5d77 DG |
530 | nb_fd = ret; |
531 | ||
b8aa1682 JD |
532 | DBG("Relay new connection received"); |
533 | for (i = 0; i < nb_fd; i++) { | |
534 | /* Fetch once the poll data */ | |
535 | revents = LTTNG_POLL_GETEV(&events, i); | |
536 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
537 | ||
538 | /* Thread quit pipe has been closed. Killing thread. */ | |
539 | ret = check_thread_quit_pipe(pollfd, revents); | |
540 | if (ret) { | |
095a4ae5 MD |
541 | err = 0; |
542 | goto exit; | |
b8aa1682 JD |
543 | } |
544 | ||
545 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
546 | ERR("socket poll error"); | |
547 | goto error; | |
548 | } else if (revents & LPOLLIN) { | |
4b7f17b2 MD |
549 | /* |
550 | * Get allocated in this thread, | |
551 | * enqueued to a global queue, dequeued | |
552 | * and freed in the worker thread. | |
553 | */ | |
554 | struct relay_command *relay_cmd; | |
555 | struct lttcomm_sock *newsock; | |
b8aa1682 JD |
556 | |
557 | relay_cmd = zmalloc(sizeof(struct relay_command)); | |
558 | if (relay_cmd == NULL) { | |
559 | PERROR("relay command zmalloc"); | |
560 | goto error; | |
561 | } | |
562 | ||
563 | if (pollfd == data_sock->fd) { | |
564 | newsock = data_sock->ops->accept(data_sock); | |
4b7f17b2 | 565 | if (!newsock) { |
b8aa1682 | 566 | PERROR("accepting data sock"); |
4b7f17b2 | 567 | free(relay_cmd); |
b8aa1682 JD |
568 | goto error; |
569 | } | |
570 | relay_cmd->type = RELAY_DATA; | |
571 | DBG("Relay data connection accepted, socket %d", newsock->fd); | |
4b7f17b2 MD |
572 | } else { |
573 | assert(pollfd == control_sock->fd); | |
b8aa1682 | 574 | newsock = control_sock->ops->accept(control_sock); |
4b7f17b2 | 575 | if (!newsock) { |
b8aa1682 | 576 | PERROR("accepting control sock"); |
4b7f17b2 | 577 | free(relay_cmd); |
b8aa1682 JD |
578 | goto error; |
579 | } | |
580 | relay_cmd->type = RELAY_CONTROL; | |
581 | DBG("Relay control connection accepted, socket %d", newsock->fd); | |
582 | } | |
583 | ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, | |
584 | &val, sizeof(int)); | |
585 | if (ret < 0) { | |
586 | PERROR("setsockopt inet"); | |
4b7f17b2 MD |
587 | lttcomm_destroy_sock(newsock); |
588 | free(relay_cmd); | |
b8aa1682 JD |
589 | goto error; |
590 | } | |
591 | relay_cmd->sock = newsock; | |
592 | /* | |
593 | * Lock free enqueue the request. | |
594 | */ | |
595 | cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node); | |
596 | ||
597 | /* | |
598 | * Wake the dispatch queue futex. Implicit memory | |
599 | * barrier with the exchange in cds_wfq_enqueue. | |
600 | */ | |
601 | futex_nto1_wake(&relay_cmd_queue.futex); | |
602 | } | |
603 | } | |
604 | } | |
605 | ||
095a4ae5 | 606 | exit: |
b8aa1682 JD |
607 | error: |
608 | error_poll_add: | |
609 | lttng_poll_clean(&events); | |
610 | error_create_poll: | |
095a4ae5 MD |
611 | if (data_sock->fd >= 0) { |
612 | ret = data_sock->ops->close(data_sock); | |
b8aa1682 JD |
613 | if (ret) { |
614 | PERROR("close"); | |
615 | } | |
b8aa1682 | 616 | } |
095a4ae5 MD |
617 | lttcomm_destroy_sock(data_sock); |
618 | error_sock_relay: | |
619 | if (control_sock->fd >= 0) { | |
620 | ret = control_sock->ops->close(control_sock); | |
b8aa1682 JD |
621 | if (ret) { |
622 | PERROR("close"); | |
623 | } | |
b8aa1682 | 624 | } |
095a4ae5 MD |
625 | lttcomm_destroy_sock(control_sock); |
626 | error_sock_control: | |
627 | if (err) { | |
628 | DBG("Thread exited with error"); | |
629 | } | |
b8aa1682 JD |
630 | DBG("Relay listener thread cleanup complete"); |
631 | stop_threads(); | |
b8aa1682 JD |
632 | return NULL; |
633 | } | |
634 | ||
635 | /* | |
636 | * This thread manages the dispatching of the requests to worker threads | |
637 | */ | |
638 | static | |
639 | void *relay_thread_dispatcher(void *data) | |
640 | { | |
641 | int ret; | |
642 | struct cds_wfq_node *node; | |
643 | struct relay_command *relay_cmd = NULL; | |
644 | ||
645 | DBG("[thread] Relay dispatcher started"); | |
646 | ||
26c9d55e | 647 | while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { |
b8aa1682 JD |
648 | /* Atomically prepare the queue futex */ |
649 | futex_nto1_prepare(&relay_cmd_queue.futex); | |
650 | ||
651 | do { | |
652 | /* Dequeue commands */ | |
653 | node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue); | |
654 | if (node == NULL) { | |
655 | DBG("Woken up but nothing in the relay command queue"); | |
656 | /* Continue thread execution */ | |
657 | break; | |
658 | } | |
659 | ||
660 | relay_cmd = caa_container_of(node, struct relay_command, node); | |
661 | DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd); | |
662 | ||
663 | /* | |
664 | * Inform worker thread of the new request. This | |
665 | * call is blocking so we can be assured that the data will be read | |
666 | * at some point in time or wait to the end of the world :) | |
667 | */ | |
6f94560a MD |
668 | do { |
669 | ret = write(relay_cmd_pipe[1], relay_cmd, | |
670 | sizeof(struct relay_command)); | |
671 | } while (ret < 0 && errno == EINTR); | |
b8aa1682 | 672 | free(relay_cmd); |
4cec016f | 673 | if (ret < 0 || ret != sizeof(struct relay_command)) { |
b8aa1682 JD |
674 | PERROR("write cmd pipe"); |
675 | goto error; | |
676 | } | |
677 | } while (node != NULL); | |
678 | ||
679 | /* Futex wait on queue. Blocking call on futex() */ | |
680 | futex_nto1_wait(&relay_cmd_queue.futex); | |
681 | } | |
682 | ||
683 | error: | |
684 | DBG("Dispatch thread dying"); | |
685 | stop_threads(); | |
686 | return NULL; | |
687 | } | |
688 | ||
de91f48a DG |
689 | /* |
690 | * Get stream from stream id. | |
691 | * Need to be called with RCU read-side lock held. | |
692 | */ | |
693 | static | |
694 | struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id, | |
695 | struct lttng_ht *streams_ht) | |
696 | { | |
697 | struct lttng_ht_node_ulong *node; | |
698 | struct lttng_ht_iter iter; | |
699 | struct relay_stream *ret; | |
700 | ||
701 | lttng_ht_lookup(streams_ht, | |
702 | (void *)((unsigned long) stream_id), | |
703 | &iter); | |
704 | node = lttng_ht_iter_get_node_ulong(&iter); | |
705 | if (node == NULL) { | |
706 | DBG("Relay stream %" PRIu64 " not found", stream_id); | |
707 | ret = NULL; | |
708 | goto end; | |
709 | } | |
710 | ||
711 | ret = caa_container_of(node, struct relay_stream, stream_n); | |
712 | ||
713 | end: | |
714 | return ret; | |
715 | } | |
716 | ||
9d1bbf21 MD |
717 | static |
718 | void deferred_free_stream(struct rcu_head *head) | |
719 | { | |
720 | struct relay_stream *stream = | |
721 | caa_container_of(head, struct relay_stream, rcu_node); | |
0f907de1 JD |
722 | free(stream->path_name); |
723 | free(stream->channel_name); | |
9d1bbf21 MD |
724 | free(stream); |
725 | } | |
726 | ||
b8aa1682 JD |
727 | /* |
728 | * relay_delete_session: Free all memory associated with a session and | |
729 | * close all the FDs | |
730 | */ | |
731 | static | |
732 | void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht) | |
733 | { | |
734 | struct lttng_ht_iter iter; | |
735 | struct lttng_ht_node_ulong *node; | |
736 | struct relay_stream *stream; | |
737 | int ret; | |
738 | ||
095a4ae5 | 739 | if (!cmd->session) { |
b8aa1682 | 740 | return; |
095a4ae5 | 741 | } |
b8aa1682 | 742 | |
77c7c900 | 743 | DBG("Relay deleting session %" PRIu64, cmd->session->id); |
5b6d8097 | 744 | |
9d1bbf21 | 745 | rcu_read_lock(); |
b8aa1682 JD |
746 | cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, node, node) { |
747 | node = lttng_ht_iter_get_node_ulong(&iter); | |
748 | if (node) { | |
749 | stream = caa_container_of(node, | |
750 | struct relay_stream, stream_n); | |
751 | if (stream->session == cmd->session) { | |
f66c074c DG |
752 | ret = close(stream->fd); |
753 | if (ret < 0) { | |
754 | PERROR("close stream fd on delete session"); | |
755 | } | |
b8aa1682 JD |
756 | ret = lttng_ht_del(streams_ht, &iter); |
757 | assert(!ret); | |
9d1bbf21 MD |
758 | call_rcu(&stream->rcu_node, |
759 | deferred_free_stream); | |
b8aa1682 JD |
760 | } |
761 | } | |
762 | } | |
9d1bbf21 | 763 | rcu_read_unlock(); |
5b6d8097 DG |
764 | |
765 | free(cmd->session); | |
b8aa1682 JD |
766 | } |
767 | ||
c5b6f4f0 DG |
768 | /* |
769 | * Handle the RELAYD_CREATE_SESSION command. | |
770 | * | |
771 | * On success, send back the session id or else return a negative value. | |
772 | */ | |
773 | static | |
774 | int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, | |
775 | struct relay_command *cmd) | |
776 | { | |
777 | int ret = 0, send_ret; | |
778 | struct relay_session *session; | |
779 | struct lttcomm_relayd_status_session reply; | |
780 | ||
781 | assert(recv_hdr); | |
782 | assert(cmd); | |
783 | ||
784 | memset(&reply, 0, sizeof(reply)); | |
785 | ||
786 | session = zmalloc(sizeof(struct relay_session)); | |
787 | if (session == NULL) { | |
788 | PERROR("relay session zmalloc"); | |
789 | ret = -1; | |
790 | goto error; | |
791 | } | |
792 | ||
793 | session->id = ++last_relay_session_id; | |
794 | session->sock = cmd->sock; | |
795 | cmd->session = session; | |
796 | ||
797 | reply.session_id = htobe64(session->id); | |
798 | ||
799 | DBG("Created session %" PRIu64, session->id); | |
800 | ||
801 | error: | |
802 | if (ret < 0) { | |
803 | reply.ret_code = htobe32(LTTNG_ERR_FATAL); | |
804 | } else { | |
805 | reply.ret_code = htobe32(LTTNG_OK); | |
806 | } | |
807 | ||
808 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
809 | if (send_ret < 0) { | |
810 | ERR("Relayd sending session id"); | |
4169f5ad | 811 | ret = send_ret; |
c5b6f4f0 DG |
812 | } |
813 | ||
814 | return ret; | |
815 | } | |
816 | ||
b8aa1682 JD |
817 | /* |
818 | * relay_add_stream: allocate a new stream for a session | |
819 | */ | |
820 | static | |
821 | int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
822 | struct relay_command *cmd, struct lttng_ht *streams_ht) | |
823 | { | |
824 | struct relay_session *session = cmd->session; | |
b8aa1682 JD |
825 | struct relay_stream *stream = NULL; |
826 | struct lttcomm_relayd_status_stream reply; | |
b8aa1682 JD |
827 | int ret, send_ret; |
828 | ||
c5b6f4f0 | 829 | if (!session || cmd->version_check_done == 0) { |
b8aa1682 JD |
830 | ERR("Trying to add a stream before version check"); |
831 | ret = -1; | |
832 | goto end_no_session; | |
833 | } | |
834 | ||
b8aa1682 JD |
835 | stream = zmalloc(sizeof(struct relay_stream)); |
836 | if (stream == NULL) { | |
837 | PERROR("relay stream zmalloc"); | |
838 | ret = -1; | |
839 | goto end_no_session; | |
840 | } | |
841 | ||
0f907de1 JD |
842 | switch (cmd->minor) { |
843 | case 1: /* LTTng sessiond 2.1 */ | |
844 | ret = cmd_recv_stream_2_1(cmd, stream); | |
845 | break; | |
846 | case 2: /* LTTng sessiond 2.2 */ | |
847 | default: | |
848 | ret = cmd_recv_stream_2_2(cmd, stream); | |
849 | break; | |
850 | } | |
851 | if (ret < 0) { | |
852 | goto err_free_stream; | |
853 | } | |
854 | ||
9d1bbf21 | 855 | rcu_read_lock(); |
b8aa1682 | 856 | stream->stream_handle = ++last_relay_stream_id; |
173af62f | 857 | stream->prev_seq = -1ULL; |
b8aa1682 JD |
858 | stream->session = session; |
859 | ||
0f907de1 | 860 | ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG); |
b8aa1682 | 861 | if (ret < 0) { |
b8aa1682 JD |
862 | ERR("relay creating output directory"); |
863 | goto end; | |
864 | } | |
865 | ||
be96a7d1 DG |
866 | /* |
867 | * No need to use run_as API here because whatever we receives, the relayd | |
868 | * uses its own credentials for the stream files. | |
869 | */ | |
0f907de1 | 870 | ret = utils_create_stream_file(stream->path_name, stream->channel_name, |
be96a7d1 | 871 | stream->tracefile_size, 0, -1, -1); |
b8aa1682 | 872 | if (ret < 0) { |
0f907de1 | 873 | ERR("Create output file"); |
b8aa1682 JD |
874 | goto end; |
875 | } | |
b8aa1682 | 876 | stream->fd = ret; |
0f907de1 JD |
877 | if (stream->tracefile_size) { |
878 | DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name); | |
879 | } else { | |
880 | DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name); | |
881 | } | |
b8aa1682 JD |
882 | |
883 | lttng_ht_node_init_ulong(&stream->stream_n, | |
884 | (unsigned long) stream->stream_handle); | |
885 | lttng_ht_add_unique_ulong(streams_ht, | |
886 | &stream->stream_n); | |
887 | ||
0f907de1 | 888 | DBG("Relay new stream added %s", stream->channel_name); |
b8aa1682 JD |
889 | |
890 | end: | |
5af40280 | 891 | reply.handle = htobe64(stream->stream_handle); |
b8aa1682 JD |
892 | /* send the session id to the client or a negative return code on error */ |
893 | if (ret < 0) { | |
f73fabfd | 894 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
5af40280 CB |
895 | /* stream was not properly added to the ht, so free it */ |
896 | free(stream); | |
b8aa1682 | 897 | } else { |
f73fabfd | 898 | reply.ret_code = htobe32(LTTNG_OK); |
b8aa1682 | 899 | } |
5af40280 | 900 | |
b8aa1682 JD |
901 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
902 | sizeof(struct lttcomm_relayd_status_stream), 0); | |
903 | if (send_ret < 0) { | |
904 | ERR("Relay sending stream id"); | |
4169f5ad | 905 | ret = send_ret; |
b8aa1682 | 906 | } |
9d1bbf21 | 907 | rcu_read_unlock(); |
b8aa1682 JD |
908 | |
909 | end_no_session: | |
910 | return ret; | |
0f907de1 JD |
911 | |
912 | err_free_stream: | |
913 | free(stream->path_name); | |
914 | free(stream->channel_name); | |
915 | free(stream); | |
916 | return ret; | |
b8aa1682 JD |
917 | } |
918 | ||
173af62f DG |
919 | /* |
920 | * relay_close_stream: close a specific stream | |
921 | */ | |
922 | static | |
923 | int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
924 | struct relay_command *cmd, struct lttng_ht *streams_ht) | |
925 | { | |
926 | struct relay_session *session = cmd->session; | |
927 | struct lttcomm_relayd_close_stream stream_info; | |
928 | struct lttcomm_relayd_generic_reply reply; | |
929 | struct relay_stream *stream; | |
930 | int ret, send_ret; | |
173af62f DG |
931 | struct lttng_ht_iter iter; |
932 | ||
933 | DBG("Close stream received"); | |
934 | ||
c5b6f4f0 | 935 | if (!session || cmd->version_check_done == 0) { |
173af62f DG |
936 | ERR("Trying to close a stream before version check"); |
937 | ret = -1; | |
938 | goto end_no_session; | |
939 | } | |
940 | ||
941 | ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info, | |
7c5aef62 | 942 | sizeof(struct lttcomm_relayd_close_stream), 0); |
173af62f | 943 | if (ret < sizeof(struct lttcomm_relayd_close_stream)) { |
a6cd2b97 DG |
944 | if (ret == 0) { |
945 | /* Orderly shutdown. Not necessary to print an error. */ | |
946 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
947 | } else { | |
948 | ERR("Relay didn't receive valid add_stream struct size : %d", ret); | |
949 | } | |
173af62f DG |
950 | ret = -1; |
951 | goto end_no_session; | |
952 | } | |
953 | ||
954 | rcu_read_lock(); | |
de91f48a DG |
955 | stream = relay_stream_from_stream_id(be64toh(stream_info.stream_id), |
956 | streams_ht); | |
173af62f DG |
957 | if (!stream) { |
958 | ret = -1; | |
959 | goto end_unlock; | |
960 | } | |
961 | ||
8e2583a4 | 962 | stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); |
173af62f DG |
963 | stream->close_flag = 1; |
964 | ||
965 | if (close_stream_check(stream)) { | |
966 | int delret; | |
967 | ||
f66c074c DG |
968 | delret = close(stream->fd); |
969 | if (delret < 0) { | |
970 | PERROR("close stream"); | |
971 | } | |
de91f48a | 972 | iter.iter.node = &stream->stream_n.node; |
173af62f DG |
973 | delret = lttng_ht_del(streams_ht, &iter); |
974 | assert(!delret); | |
975 | call_rcu(&stream->rcu_node, | |
976 | deferred_free_stream); | |
977 | DBG("Closed tracefile %d from close stream", stream->fd); | |
978 | } | |
979 | ||
980 | end_unlock: | |
981 | rcu_read_unlock(); | |
982 | ||
983 | if (ret < 0) { | |
f73fabfd | 984 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
173af62f | 985 | } else { |
f73fabfd | 986 | reply.ret_code = htobe32(LTTNG_OK); |
173af62f DG |
987 | } |
988 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
989 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
990 | if (send_ret < 0) { | |
991 | ERR("Relay sending stream id"); | |
4169f5ad | 992 | ret = send_ret; |
173af62f DG |
993 | } |
994 | ||
995 | end_no_session: | |
996 | return ret; | |
997 | } | |
998 | ||
b8aa1682 JD |
999 | /* |
1000 | * relay_unknown_command: send -1 if received unknown command | |
1001 | */ | |
1002 | static | |
1003 | void relay_unknown_command(struct relay_command *cmd) | |
1004 | { | |
1005 | struct lttcomm_relayd_generic_reply reply; | |
1006 | int ret; | |
1007 | ||
f73fabfd | 1008 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1009 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1010 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1011 | if (ret < 0) { | |
1012 | ERR("Relay sending unknown command"); | |
1013 | } | |
1014 | } | |
1015 | ||
1016 | /* | |
1017 | * relay_start: send an acknowledgment to the client to tell if we are | |
1018 | * ready to receive data. We are ready if a session is established. | |
1019 | */ | |
1020 | static | |
1021 | int relay_start(struct lttcomm_relayd_hdr *recv_hdr, | |
1022 | struct relay_command *cmd) | |
1023 | { | |
f73fabfd | 1024 | int ret = htobe32(LTTNG_OK); |
b8aa1682 JD |
1025 | struct lttcomm_relayd_generic_reply reply; |
1026 | struct relay_session *session = cmd->session; | |
1027 | ||
1028 | if (!session) { | |
1029 | DBG("Trying to start the streaming without a session established"); | |
f73fabfd | 1030 | ret = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1031 | } |
1032 | ||
1033 | reply.ret_code = ret; | |
1034 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1035 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1036 | if (ret < 0) { | |
1037 | ERR("Relay sending start ack"); | |
1038 | } | |
1039 | ||
1040 | return ret; | |
1041 | } | |
1042 | ||
1d4dfdef DG |
1043 | /* |
1044 | * Append padding to the file pointed by the file descriptor fd. | |
1045 | */ | |
1046 | static int write_padding_to_file(int fd, uint32_t size) | |
1047 | { | |
1048 | int ret = 0; | |
1049 | char *zeros; | |
1050 | ||
1051 | if (size == 0) { | |
1052 | goto end; | |
1053 | } | |
1054 | ||
1055 | zeros = zmalloc(size); | |
1056 | if (zeros == NULL) { | |
1057 | PERROR("zmalloc zeros for padding"); | |
1058 | ret = -1; | |
1059 | goto end; | |
1060 | } | |
1061 | ||
1062 | do { | |
1063 | ret = write(fd, zeros, size); | |
1064 | } while (ret < 0 && errno == EINTR); | |
4cec016f | 1065 | if (ret < 0 || ret != size) { |
1d4dfdef DG |
1066 | PERROR("write padding to file"); |
1067 | } | |
1068 | ||
e986c7a1 DG |
1069 | free(zeros); |
1070 | ||
1d4dfdef DG |
1071 | end: |
1072 | return ret; | |
1073 | } | |
1074 | ||
b8aa1682 JD |
1075 | /* |
1076 | * relay_recv_metadata: receive the metada for the session. | |
1077 | */ | |
1078 | static | |
1079 | int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, | |
1080 | struct relay_command *cmd, struct lttng_ht *streams_ht) | |
1081 | { | |
f73fabfd | 1082 | int ret = htobe32(LTTNG_OK); |
b8aa1682 JD |
1083 | struct relay_session *session = cmd->session; |
1084 | struct lttcomm_relayd_metadata_payload *metadata_struct; | |
1085 | struct relay_stream *metadata_stream; | |
1086 | uint64_t data_size, payload_size; | |
1087 | ||
1088 | if (!session) { | |
1089 | ERR("Metadata sent before version check"); | |
1090 | ret = -1; | |
1091 | goto end; | |
1092 | } | |
1093 | ||
f6416125 MD |
1094 | data_size = payload_size = be64toh(recv_hdr->data_size); |
1095 | if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { | |
1096 | ERR("Incorrect data size"); | |
1097 | ret = -1; | |
1098 | goto end; | |
1099 | } | |
1100 | payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); | |
1101 | ||
b8aa1682 | 1102 | if (data_buffer_size < data_size) { |
d7b3776f | 1103 | /* In case the realloc fails, we can free the memory */ |
c617c0c6 MD |
1104 | char *tmp_data_ptr; |
1105 | ||
1106 | tmp_data_ptr = realloc(data_buffer, data_size); | |
1107 | if (!tmp_data_ptr) { | |
b8aa1682 | 1108 | ERR("Allocating data buffer"); |
c617c0c6 | 1109 | free(data_buffer); |
b8aa1682 JD |
1110 | ret = -1; |
1111 | goto end; | |
1112 | } | |
c617c0c6 | 1113 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
1114 | data_buffer_size = data_size; |
1115 | } | |
1116 | memset(data_buffer, 0, data_size); | |
77c7c900 | 1117 | DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size); |
7c5aef62 | 1118 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 1119 | if (ret < 0 || ret != data_size) { |
a6cd2b97 DG |
1120 | if (ret == 0) { |
1121 | /* Orderly shutdown. Not necessary to print an error. */ | |
1122 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1123 | } else { | |
1124 | ERR("Relay didn't receive the whole metadata"); | |
1125 | } | |
b8aa1682 | 1126 | ret = -1; |
b8aa1682 JD |
1127 | goto end; |
1128 | } | |
1129 | metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; | |
9d1bbf21 MD |
1130 | |
1131 | rcu_read_lock(); | |
b8aa1682 JD |
1132 | metadata_stream = relay_stream_from_stream_id( |
1133 | be64toh(metadata_struct->stream_id), streams_ht); | |
1134 | if (!metadata_stream) { | |
1135 | ret = -1; | |
9d1bbf21 | 1136 | goto end_unlock; |
b8aa1682 JD |
1137 | } |
1138 | ||
6f94560a MD |
1139 | do { |
1140 | ret = write(metadata_stream->fd, metadata_struct->payload, | |
1141 | payload_size); | |
1142 | } while (ret < 0 && errno == EINTR); | |
4cec016f | 1143 | if (ret < 0 || ret != payload_size) { |
b8aa1682 JD |
1144 | ERR("Relay error writing metadata on file"); |
1145 | ret = -1; | |
9d1bbf21 | 1146 | goto end_unlock; |
b8aa1682 | 1147 | } |
1d4dfdef DG |
1148 | |
1149 | ret = write_padding_to_file(metadata_stream->fd, | |
1150 | be32toh(metadata_struct->padding_size)); | |
1151 | if (ret < 0) { | |
1152 | goto end_unlock; | |
1153 | } | |
1154 | ||
b8aa1682 JD |
1155 | DBG2("Relay metadata written"); |
1156 | ||
9d1bbf21 | 1157 | end_unlock: |
6e3c5836 | 1158 | rcu_read_unlock(); |
b8aa1682 JD |
1159 | end: |
1160 | return ret; | |
1161 | } | |
1162 | ||
1163 | /* | |
1164 | * relay_send_version: send relayd version number | |
1165 | */ | |
1166 | static | |
1167 | int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, | |
d4519fa3 | 1168 | struct relay_command *cmd, struct lttng_ht *streams_ht) |
b8aa1682 | 1169 | { |
7f51dcba | 1170 | int ret; |
092b6259 | 1171 | struct lttcomm_relayd_version reply, msg; |
b8aa1682 | 1172 | |
c5b6f4f0 DG |
1173 | assert(cmd); |
1174 | ||
1175 | cmd->version_check_done = 1; | |
b8aa1682 | 1176 | |
092b6259 | 1177 | /* Get version from the other side. */ |
7c5aef62 | 1178 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
092b6259 | 1179 | if (ret < 0 || ret != sizeof(msg)) { |
a6cd2b97 DG |
1180 | if (ret == 0) { |
1181 | /* Orderly shutdown. Not necessary to print an error. */ | |
1182 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1183 | } else { | |
1184 | ERR("Relay failed to receive the version values."); | |
1185 | } | |
092b6259 | 1186 | ret = -1; |
092b6259 DG |
1187 | goto end; |
1188 | } | |
1189 | ||
c617c0c6 | 1190 | ret = sscanf(VERSION, "%10u.%10u", &reply.major, &reply.minor); |
7f51dcba MD |
1191 | if (ret < 2) { |
1192 | ERR("Error in scanning version"); | |
1193 | ret = -1; | |
1194 | goto end; | |
1195 | } | |
d4519fa3 JD |
1196 | |
1197 | /* Major versions must be the same */ | |
1198 | if (reply.major != be32toh(msg.major)) { | |
6151a90f JD |
1199 | DBG("Incompatible major versions (%u vs %u), deleting session", |
1200 | reply.major, be32toh(msg.major)); | |
d4519fa3 JD |
1201 | relay_delete_session(cmd, streams_ht); |
1202 | ret = 0; | |
1203 | goto end; | |
1204 | } | |
1205 | ||
0f907de1 JD |
1206 | cmd->major = reply.major; |
1207 | /* We adapt to the lowest compatible version */ | |
1208 | if (reply.minor <= be32toh(msg.minor)) { | |
1209 | cmd->minor = reply.minor; | |
1210 | } else { | |
1211 | cmd->minor = be32toh(msg.minor); | |
1212 | } | |
1213 | ||
6151a90f JD |
1214 | reply.major = htobe32(reply.major); |
1215 | reply.minor = htobe32(reply.minor); | |
1216 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1217 | sizeof(struct lttcomm_relayd_version), 0); | |
1218 | if (ret < 0) { | |
1219 | ERR("Relay sending version"); | |
1220 | } | |
1221 | ||
0f907de1 JD |
1222 | DBG("Version check done using protocol %u.%u", cmd->major, |
1223 | cmd->minor); | |
b8aa1682 JD |
1224 | |
1225 | end: | |
1226 | return ret; | |
1227 | } | |
1228 | ||
c8f59ee5 | 1229 | /* |
6d805429 | 1230 | * Check for data pending for a given stream id from the session daemon. |
c8f59ee5 DG |
1231 | */ |
1232 | static | |
6d805429 | 1233 | int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, |
c8f59ee5 DG |
1234 | struct relay_command *cmd, struct lttng_ht *streams_ht) |
1235 | { | |
1236 | struct relay_session *session = cmd->session; | |
6d805429 | 1237 | struct lttcomm_relayd_data_pending msg; |
c8f59ee5 DG |
1238 | struct lttcomm_relayd_generic_reply reply; |
1239 | struct relay_stream *stream; | |
1240 | int ret; | |
c8f59ee5 DG |
1241 | uint64_t last_net_seq_num, stream_id; |
1242 | ||
6d805429 | 1243 | DBG("Data pending command received"); |
c8f59ee5 | 1244 | |
c5b6f4f0 | 1245 | if (!session || cmd->version_check_done == 0) { |
c8f59ee5 DG |
1246 | ERR("Trying to check for data before version check"); |
1247 | ret = -1; | |
1248 | goto end_no_session; | |
1249 | } | |
1250 | ||
7c5aef62 | 1251 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
c8f59ee5 | 1252 | if (ret < sizeof(msg)) { |
a6cd2b97 DG |
1253 | if (ret == 0) { |
1254 | /* Orderly shutdown. Not necessary to print an error. */ | |
1255 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1256 | } else { | |
1257 | ERR("Relay didn't receive valid data_pending struct size : %d", | |
1258 | ret); | |
1259 | } | |
c8f59ee5 DG |
1260 | ret = -1; |
1261 | goto end_no_session; | |
1262 | } | |
1263 | ||
1264 | stream_id = be64toh(msg.stream_id); | |
1265 | last_net_seq_num = be64toh(msg.last_net_seq_num); | |
1266 | ||
1267 | rcu_read_lock(); | |
de91f48a DG |
1268 | stream = relay_stream_from_stream_id(stream_id, streams_ht); |
1269 | if (stream == NULL) { | |
c8f59ee5 DG |
1270 | ret = -1; |
1271 | goto end_unlock; | |
1272 | } | |
1273 | ||
6d805429 | 1274 | DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 |
c8f59ee5 DG |
1275 | " and last_seq %" PRIu64, stream_id, stream->prev_seq, |
1276 | last_net_seq_num); | |
1277 | ||
33832e64 | 1278 | /* Avoid wrapping issue */ |
39df6d9f | 1279 | if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { |
6d805429 | 1280 | /* Data has in fact been written and is NOT pending */ |
c8f59ee5 | 1281 | ret = 0; |
6d805429 DG |
1282 | } else { |
1283 | /* Data still being streamed thus pending */ | |
1284 | ret = 1; | |
c8f59ee5 DG |
1285 | } |
1286 | ||
f7079f67 DG |
1287 | /* Pending check is now done. */ |
1288 | stream->data_pending_check_done = 1; | |
1289 | ||
c8f59ee5 DG |
1290 | end_unlock: |
1291 | rcu_read_unlock(); | |
1292 | ||
1293 | reply.ret_code = htobe32(ret); | |
1294 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1295 | if (ret < 0) { | |
6d805429 | 1296 | ERR("Relay data pending ret code failed"); |
c8f59ee5 DG |
1297 | } |
1298 | ||
1299 | end_no_session: | |
1300 | return ret; | |
1301 | } | |
1302 | ||
1303 | /* | |
1304 | * Wait for the control socket to reach a quiescent state. | |
1305 | * | |
1306 | * Note that for now, when receiving this command from the session daemon, this | |
1307 | * means that every subsequent commands or data received on the control socket | |
1308 | * has been handled. So, this is why we simply return OK here. | |
1309 | */ | |
1310 | static | |
1311 | int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, | |
ad7051c0 | 1312 | struct relay_command *cmd, struct lttng_ht *streams_ht) |
c8f59ee5 DG |
1313 | { |
1314 | int ret; | |
ad7051c0 DG |
1315 | uint64_t stream_id; |
1316 | struct relay_stream *stream; | |
1317 | struct lttng_ht_iter iter; | |
1318 | struct lttcomm_relayd_quiescent_control msg; | |
c8f59ee5 DG |
1319 | struct lttcomm_relayd_generic_reply reply; |
1320 | ||
1321 | DBG("Checking quiescent state on control socket"); | |
1322 | ||
ad7051c0 DG |
1323 | if (!cmd->session || cmd->version_check_done == 0) { |
1324 | ERR("Trying to check for data before version check"); | |
1325 | ret = -1; | |
1326 | goto end_no_session; | |
1327 | } | |
1328 | ||
1329 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1330 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1331 | if (ret == 0) { |
1332 | /* Orderly shutdown. Not necessary to print an error. */ | |
1333 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1334 | } else { | |
1335 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1336 | ret); | |
1337 | } | |
ad7051c0 DG |
1338 | ret = -1; |
1339 | goto end_no_session; | |
1340 | } | |
1341 | ||
1342 | stream_id = be64toh(msg.stream_id); | |
1343 | ||
1344 | rcu_read_lock(); | |
1345 | cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) { | |
1346 | if (stream->stream_handle == stream_id) { | |
1347 | stream->data_pending_check_done = 1; | |
1348 | DBG("Relay quiescent control pending flag set to %" PRIu64, | |
1349 | stream_id); | |
1350 | break; | |
1351 | } | |
1352 | } | |
1353 | rcu_read_unlock(); | |
1354 | ||
c8f59ee5 DG |
1355 | reply.ret_code = htobe32(LTTNG_OK); |
1356 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1357 | if (ret < 0) { | |
6d805429 | 1358 | ERR("Relay data quiescent control ret code failed"); |
c8f59ee5 DG |
1359 | } |
1360 | ||
ad7051c0 | 1361 | end_no_session: |
c8f59ee5 DG |
1362 | return ret; |
1363 | } | |
1364 | ||
f7079f67 DG |
1365 | /* |
1366 | * Initialize a data pending command. This means that a client is about to ask | |
1367 | * for data pending for each stream he/she holds. Simply iterate over all | |
1368 | * streams of a session and set the data_pending_check_done flag. | |
1369 | * | |
1370 | * This command returns to the client a LTTNG_OK code. | |
1371 | */ | |
1372 | static | |
1373 | int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
1374 | struct relay_command *cmd, struct lttng_ht *streams_ht) | |
1375 | { | |
1376 | int ret; | |
1377 | struct lttng_ht_iter iter; | |
1378 | struct lttcomm_relayd_begin_data_pending msg; | |
1379 | struct lttcomm_relayd_generic_reply reply; | |
1380 | struct relay_stream *stream; | |
1381 | uint64_t session_id; | |
1382 | ||
1383 | assert(recv_hdr); | |
1384 | assert(cmd); | |
1385 | assert(streams_ht); | |
1386 | ||
1387 | DBG("Init streams for data pending"); | |
1388 | ||
1389 | if (!cmd->session || cmd->version_check_done == 0) { | |
1390 | ERR("Trying to check for data before version check"); | |
1391 | ret = -1; | |
1392 | goto end_no_session; | |
1393 | } | |
1394 | ||
1395 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1396 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1397 | if (ret == 0) { |
1398 | /* Orderly shutdown. Not necessary to print an error. */ | |
1399 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1400 | } else { | |
1401 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1402 | ret); | |
1403 | } | |
f7079f67 DG |
1404 | ret = -1; |
1405 | goto end_no_session; | |
1406 | } | |
1407 | ||
1408 | session_id = be64toh(msg.session_id); | |
1409 | ||
1410 | /* | |
1411 | * Iterate over all streams to set the begin data pending flag. For now, the | |
1412 | * streams are indexed by stream handle so we have to iterate over all | |
1413 | * streams to find the one associated with the right session_id. | |
1414 | */ | |
1415 | rcu_read_lock(); | |
1416 | cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) { | |
1417 | if (stream->session->id == session_id) { | |
1418 | stream->data_pending_check_done = 0; | |
1419 | DBG("Set begin data pending flag to stream %" PRIu64, | |
1420 | stream->stream_handle); | |
1421 | } | |
1422 | } | |
1423 | rcu_read_unlock(); | |
1424 | ||
1425 | /* All good, send back reply. */ | |
1426 | reply.ret_code = htobe32(LTTNG_OK); | |
1427 | ||
1428 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1429 | if (ret < 0) { | |
1430 | ERR("Relay begin data pending send reply failed"); | |
1431 | } | |
1432 | ||
1433 | end_no_session: | |
1434 | return ret; | |
1435 | } | |
1436 | ||
1437 | /* | |
1438 | * End data pending command. This will check, for a given session id, if each | |
1439 | * stream associated with it has its data_pending_check_done flag set. If not, | |
1440 | * this means that the client lost track of the stream but the data is still | |
1441 | * being streamed on our side. In this case, we inform the client that data is | |
1442 | * inflight. | |
1443 | * | |
1444 | * Return to the client if there is data in flight or not with a ret_code. | |
1445 | */ | |
1446 | static | |
1447 | int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
1448 | struct relay_command *cmd, struct lttng_ht *streams_ht) | |
1449 | { | |
1450 | int ret; | |
1451 | struct lttng_ht_iter iter; | |
1452 | struct lttcomm_relayd_end_data_pending msg; | |
1453 | struct lttcomm_relayd_generic_reply reply; | |
1454 | struct relay_stream *stream; | |
1455 | uint64_t session_id; | |
1456 | uint32_t is_data_inflight = 0; | |
1457 | ||
1458 | assert(recv_hdr); | |
1459 | assert(cmd); | |
1460 | assert(streams_ht); | |
1461 | ||
1462 | DBG("End data pending command"); | |
1463 | ||
1464 | if (!cmd->session || cmd->version_check_done == 0) { | |
1465 | ERR("Trying to check for data before version check"); | |
1466 | ret = -1; | |
1467 | goto end_no_session; | |
1468 | } | |
1469 | ||
1470 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1471 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1472 | if (ret == 0) { |
1473 | /* Orderly shutdown. Not necessary to print an error. */ | |
1474 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1475 | } else { | |
1476 | ERR("Relay didn't receive valid end data_pending struct size: %d", | |
1477 | ret); | |
1478 | } | |
f7079f67 DG |
1479 | ret = -1; |
1480 | goto end_no_session; | |
1481 | } | |
1482 | ||
1483 | session_id = be64toh(msg.session_id); | |
1484 | ||
1485 | /* Iterate over all streams to see if the begin data pending flag is set. */ | |
1486 | rcu_read_lock(); | |
1487 | cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) { | |
1488 | if (stream->session->id == session_id && | |
1489 | !stream->data_pending_check_done) { | |
1490 | is_data_inflight = 1; | |
1491 | DBG("Data is still in flight for stream %" PRIu64, | |
1492 | stream->stream_handle); | |
1493 | break; | |
1494 | } | |
1495 | } | |
1496 | rcu_read_unlock(); | |
1497 | ||
1498 | /* All good, send back reply. */ | |
1499 | reply.ret_code = htobe32(is_data_inflight); | |
1500 | ||
1501 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1502 | if (ret < 0) { | |
1503 | ERR("Relay end data pending send reply failed"); | |
1504 | } | |
1505 | ||
1506 | end_no_session: | |
1507 | return ret; | |
1508 | } | |
1509 | ||
b8aa1682 JD |
1510 | /* |
1511 | * relay_process_control: Process the commands received on the control socket | |
1512 | */ | |
1513 | static | |
1514 | int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, | |
1515 | struct relay_command *cmd, struct lttng_ht *streams_ht) | |
1516 | { | |
1517 | int ret = 0; | |
1518 | ||
1519 | switch (be32toh(recv_hdr->cmd)) { | |
b8aa1682 JD |
1520 | case RELAYD_CREATE_SESSION: |
1521 | ret = relay_create_session(recv_hdr, cmd); | |
1522 | break; | |
b8aa1682 JD |
1523 | case RELAYD_ADD_STREAM: |
1524 | ret = relay_add_stream(recv_hdr, cmd, streams_ht); | |
1525 | break; | |
1526 | case RELAYD_START_DATA: | |
1527 | ret = relay_start(recv_hdr, cmd); | |
1528 | break; | |
1529 | case RELAYD_SEND_METADATA: | |
1530 | ret = relay_recv_metadata(recv_hdr, cmd, streams_ht); | |
1531 | break; | |
1532 | case RELAYD_VERSION: | |
d4519fa3 | 1533 | ret = relay_send_version(recv_hdr, cmd, streams_ht); |
b8aa1682 | 1534 | break; |
173af62f DG |
1535 | case RELAYD_CLOSE_STREAM: |
1536 | ret = relay_close_stream(recv_hdr, cmd, streams_ht); | |
1537 | break; | |
6d805429 DG |
1538 | case RELAYD_DATA_PENDING: |
1539 | ret = relay_data_pending(recv_hdr, cmd, streams_ht); | |
c8f59ee5 DG |
1540 | break; |
1541 | case RELAYD_QUIESCENT_CONTROL: | |
ad7051c0 | 1542 | ret = relay_quiescent_control(recv_hdr, cmd, streams_ht); |
c8f59ee5 | 1543 | break; |
f7079f67 DG |
1544 | case RELAYD_BEGIN_DATA_PENDING: |
1545 | ret = relay_begin_data_pending(recv_hdr, cmd, streams_ht); | |
1546 | break; | |
1547 | case RELAYD_END_DATA_PENDING: | |
1548 | ret = relay_end_data_pending(recv_hdr, cmd, streams_ht); | |
1549 | break; | |
b8aa1682 JD |
1550 | case RELAYD_UPDATE_SYNC_INFO: |
1551 | default: | |
1552 | ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); | |
1553 | relay_unknown_command(cmd); | |
1554 | ret = -1; | |
1555 | goto end; | |
1556 | } | |
1557 | ||
1558 | end: | |
1559 | return ret; | |
1560 | } | |
1561 | ||
1562 | /* | |
1563 | * relay_process_data: Process the data received on the data socket | |
1564 | */ | |
1565 | static | |
1566 | int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht) | |
1567 | { | |
1568 | int ret = 0; | |
1569 | struct relay_stream *stream; | |
1570 | struct lttcomm_relayd_data_hdr data_hdr; | |
1571 | uint64_t stream_id; | |
173af62f | 1572 | uint64_t net_seq_num; |
b8aa1682 JD |
1573 | uint32_t data_size; |
1574 | ||
1575 | ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr, | |
7c5aef62 | 1576 | sizeof(struct lttcomm_relayd_data_hdr), 0); |
b8aa1682 | 1577 | if (ret <= 0) { |
a6cd2b97 DG |
1578 | if (ret == 0) { |
1579 | /* Orderly shutdown. Not necessary to print an error. */ | |
1580 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1581 | } else { | |
1582 | ERR("Unable to receive data header on sock %d", cmd->sock->fd); | |
1583 | } | |
b8aa1682 JD |
1584 | ret = -1; |
1585 | goto end; | |
1586 | } | |
1587 | ||
1588 | stream_id = be64toh(data_hdr.stream_id); | |
9d1bbf21 MD |
1589 | |
1590 | rcu_read_lock(); | |
b8aa1682 JD |
1591 | stream = relay_stream_from_stream_id(stream_id, streams_ht); |
1592 | if (!stream) { | |
1593 | ret = -1; | |
9d1bbf21 | 1594 | goto end_unlock; |
b8aa1682 JD |
1595 | } |
1596 | ||
1597 | data_size = be32toh(data_hdr.data_size); | |
1598 | if (data_buffer_size < data_size) { | |
c617c0c6 MD |
1599 | char *tmp_data_ptr; |
1600 | ||
1601 | tmp_data_ptr = realloc(data_buffer, data_size); | |
1602 | if (!tmp_data_ptr) { | |
b8aa1682 | 1603 | ERR("Allocating data buffer"); |
c617c0c6 | 1604 | free(data_buffer); |
b8aa1682 | 1605 | ret = -1; |
9d1bbf21 | 1606 | goto end_unlock; |
b8aa1682 | 1607 | } |
c617c0c6 | 1608 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
1609 | data_buffer_size = data_size; |
1610 | } | |
1611 | memset(data_buffer, 0, data_size); | |
1612 | ||
173af62f DG |
1613 | net_seq_num = be64toh(data_hdr.net_seq_num); |
1614 | ||
77c7c900 | 1615 | DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, |
173af62f | 1616 | data_size, stream_id, net_seq_num); |
7c5aef62 | 1617 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 1618 | if (ret <= 0) { |
a6cd2b97 DG |
1619 | if (ret == 0) { |
1620 | /* Orderly shutdown. Not necessary to print an error. */ | |
1621 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1622 | } | |
b8aa1682 | 1623 | ret = -1; |
9d1bbf21 | 1624 | goto end_unlock; |
b8aa1682 JD |
1625 | } |
1626 | ||
0f907de1 JD |
1627 | if (stream->tracefile_size > 0 && |
1628 | (stream->tracefile_size_current + data_size) > | |
1629 | stream->tracefile_size) { | |
1630 | ret = utils_rotate_stream_file(stream->path_name, | |
1631 | stream->channel_name, stream->tracefile_size, | |
be96a7d1 | 1632 | stream->tracefile_count, -1, -1, |
0f907de1 JD |
1633 | stream->fd, &(stream->tracefile_count_current)); |
1634 | if (ret < 0) { | |
1635 | ERR("Rotating output file"); | |
1636 | goto end; | |
1637 | } | |
1638 | stream->fd = ret; | |
1639 | } | |
1640 | stream->tracefile_size_current += data_size; | |
6f94560a MD |
1641 | do { |
1642 | ret = write(stream->fd, data_buffer, data_size); | |
1643 | } while (ret < 0 && errno == EINTR); | |
4cec016f | 1644 | if (ret < 0 || ret != data_size) { |
b8aa1682 JD |
1645 | ERR("Relay error writing data to file"); |
1646 | ret = -1; | |
9d1bbf21 | 1647 | goto end_unlock; |
b8aa1682 | 1648 | } |
1d4dfdef | 1649 | |
5ab7344e JD |
1650 | DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, |
1651 | ret, stream->stream_handle); | |
1652 | ||
1d4dfdef DG |
1653 | ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); |
1654 | if (ret < 0) { | |
1655 | goto end_unlock; | |
1656 | } | |
1657 | ||
173af62f DG |
1658 | stream->prev_seq = net_seq_num; |
1659 | ||
1660 | /* Check if we need to close the FD */ | |
1661 | if (close_stream_check(stream)) { | |
f66c074c | 1662 | int cret; |
173af62f DG |
1663 | struct lttng_ht_iter iter; |
1664 | ||
f66c074c DG |
1665 | cret = close(stream->fd); |
1666 | if (cret < 0) { | |
1667 | PERROR("close stream process data"); | |
1668 | } | |
173af62f DG |
1669 | iter.iter.node = &stream->stream_n.node; |
1670 | ret = lttng_ht_del(streams_ht, &iter); | |
1671 | assert(!ret); | |
1672 | call_rcu(&stream->rcu_node, | |
1673 | deferred_free_stream); | |
1674 | DBG("Closed tracefile %d after recv data", stream->fd); | |
1675 | } | |
1676 | ||
9d1bbf21 MD |
1677 | end_unlock: |
1678 | rcu_read_unlock(); | |
b8aa1682 JD |
1679 | end: |
1680 | return ret; | |
1681 | } | |
1682 | ||
1683 | static | |
9d1bbf21 | 1684 | void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) |
b8aa1682 JD |
1685 | { |
1686 | int ret; | |
1687 | ||
b8aa1682 JD |
1688 | lttng_poll_del(events, pollfd); |
1689 | ||
1690 | ret = close(pollfd); | |
1691 | if (ret < 0) { | |
1692 | ERR("Closing pollfd %d", pollfd); | |
1693 | } | |
1694 | } | |
1695 | ||
1696 | static | |
1697 | int relay_add_connection(int fd, struct lttng_poll_event *events, | |
1698 | struct lttng_ht *relay_connections_ht) | |
1699 | { | |
b8aa1682 | 1700 | struct relay_command *relay_connection; |
9d1bbf21 | 1701 | int ret; |
b8aa1682 JD |
1702 | |
1703 | relay_connection = zmalloc(sizeof(struct relay_command)); | |
1704 | if (relay_connection == NULL) { | |
1705 | PERROR("Relay command zmalloc"); | |
9d1bbf21 | 1706 | goto error; |
b8aa1682 | 1707 | } |
f921c78f DG |
1708 | do { |
1709 | ret = read(fd, relay_connection, sizeof(struct relay_command)); | |
1710 | } while (ret < 0 && errno == EINTR); | |
cdf0f8fc | 1711 | if (ret < 0 || ret < sizeof(struct relay_command)) { |
b8aa1682 | 1712 | PERROR("read relay cmd pipe"); |
9d1bbf21 | 1713 | goto error_read; |
b8aa1682 JD |
1714 | } |
1715 | ||
1716 | lttng_ht_node_init_ulong(&relay_connection->sock_n, | |
1717 | (unsigned long) relay_connection->sock->fd); | |
9d1bbf21 | 1718 | rcu_read_lock(); |
b8aa1682 JD |
1719 | lttng_ht_add_unique_ulong(relay_connections_ht, |
1720 | &relay_connection->sock_n); | |
9d1bbf21 MD |
1721 | rcu_read_unlock(); |
1722 | return lttng_poll_add(events, | |
b8aa1682 JD |
1723 | relay_connection->sock->fd, |
1724 | LPOLLIN | LPOLLRDHUP); | |
1725 | ||
9d1bbf21 MD |
1726 | error_read: |
1727 | free(relay_connection); | |
1728 | error: | |
1729 | return -1; | |
1730 | } | |
1731 | ||
1732 | static | |
1733 | void deferred_free_connection(struct rcu_head *head) | |
1734 | { | |
1735 | struct relay_command *relay_connection = | |
1736 | caa_container_of(head, struct relay_command, rcu_node); | |
5b6d8097 DG |
1737 | |
1738 | lttcomm_destroy_sock(relay_connection->sock); | |
9d1bbf21 MD |
1739 | free(relay_connection); |
1740 | } | |
1741 | ||
1742 | static | |
1743 | void relay_del_connection(struct lttng_ht *relay_connections_ht, | |
1744 | struct lttng_ht *streams_ht, struct lttng_ht_iter *iter, | |
1745 | struct relay_command *relay_connection) | |
1746 | { | |
1747 | int ret; | |
1748 | ||
1749 | ret = lttng_ht_del(relay_connections_ht, iter); | |
1750 | assert(!ret); | |
1751 | if (relay_connection->type == RELAY_CONTROL) { | |
1752 | relay_delete_session(relay_connection, streams_ht); | |
1753 | } | |
5b6d8097 | 1754 | |
9d1bbf21 MD |
1755 | call_rcu(&relay_connection->rcu_node, |
1756 | deferred_free_connection); | |
b8aa1682 JD |
1757 | } |
1758 | ||
1759 | /* | |
1760 | * This thread does the actual work | |
1761 | */ | |
1762 | static | |
1763 | void *relay_thread_worker(void *data) | |
1764 | { | |
beaad64c DG |
1765 | int ret, err = -1, last_seen_data_fd = -1; |
1766 | uint32_t nb_fd; | |
b8aa1682 JD |
1767 | struct relay_command *relay_connection; |
1768 | struct lttng_poll_event events; | |
1769 | struct lttng_ht *relay_connections_ht; | |
1770 | struct lttng_ht_node_ulong *node; | |
1771 | struct lttng_ht_iter iter; | |
1772 | struct lttng_ht *streams_ht; | |
1773 | struct lttcomm_relayd_hdr recv_hdr; | |
1774 | ||
1775 | DBG("[thread] Relay worker started"); | |
1776 | ||
9d1bbf21 MD |
1777 | rcu_register_thread(); |
1778 | ||
b8aa1682 JD |
1779 | /* table of connections indexed on socket */ |
1780 | relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
095a4ae5 MD |
1781 | if (!relay_connections_ht) { |
1782 | goto relay_connections_ht_error; | |
1783 | } | |
b8aa1682 JD |
1784 | |
1785 | /* tables of streams indexed by stream ID */ | |
1786 | streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
095a4ae5 MD |
1787 | if (!streams_ht) { |
1788 | goto streams_ht_error; | |
1789 | } | |
b8aa1682 JD |
1790 | |
1791 | ret = create_thread_poll_set(&events, 2); | |
1792 | if (ret < 0) { | |
1793 | goto error_poll_create; | |
1794 | } | |
1795 | ||
1796 | ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); | |
1797 | if (ret < 0) { | |
1798 | goto error; | |
1799 | } | |
1800 | ||
beaad64c | 1801 | restart: |
b8aa1682 | 1802 | while (1) { |
beaad64c DG |
1803 | int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; |
1804 | ||
b8aa1682 | 1805 | /* Infinite blocking call, waiting for transmission */ |
87c1611d | 1806 | DBG3("Relayd worker thread polling..."); |
b8aa1682 JD |
1807 | ret = lttng_poll_wait(&events, -1); |
1808 | if (ret < 0) { | |
1809 | /* | |
1810 | * Restart interrupted system call. | |
1811 | */ | |
1812 | if (errno == EINTR) { | |
1813 | goto restart; | |
1814 | } | |
1815 | goto error; | |
1816 | } | |
1817 | ||
0d9c5d77 DG |
1818 | nb_fd = ret; |
1819 | ||
beaad64c DG |
1820 | /* |
1821 | * Process control. The control connection is prioritised so we don't | |
1822 | * starve it with high throughout put tracing data on the data | |
1823 | * connection. | |
1824 | */ | |
b8aa1682 JD |
1825 | for (i = 0; i < nb_fd; i++) { |
1826 | /* Fetch once the poll data */ | |
beaad64c DG |
1827 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); |
1828 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
b8aa1682 JD |
1829 | |
1830 | /* Thread quit pipe has been closed. Killing thread. */ | |
1831 | ret = check_thread_quit_pipe(pollfd, revents); | |
1832 | if (ret) { | |
095a4ae5 MD |
1833 | err = 0; |
1834 | goto exit; | |
b8aa1682 JD |
1835 | } |
1836 | ||
1837 | /* Inspect the relay cmd pipe for new connection */ | |
1838 | if (pollfd == relay_cmd_pipe[0]) { | |
1839 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
1840 | ERR("Relay pipe error"); | |
1841 | goto error; | |
1842 | } else if (revents & LPOLLIN) { | |
1843 | DBG("Relay command received"); | |
1844 | ret = relay_add_connection(relay_cmd_pipe[0], | |
1845 | &events, relay_connections_ht); | |
1846 | if (ret < 0) { | |
1847 | goto error; | |
1848 | } | |
1849 | } | |
beaad64c | 1850 | } else if (revents) { |
9d1bbf21 | 1851 | rcu_read_lock(); |
b8aa1682 JD |
1852 | lttng_ht_lookup(relay_connections_ht, |
1853 | (void *)((unsigned long) pollfd), | |
1854 | &iter); | |
1855 | node = lttng_ht_iter_get_node_ulong(&iter); | |
1856 | if (node == NULL) { | |
1857 | DBG2("Relay sock %d not found", pollfd); | |
9d1bbf21 | 1858 | rcu_read_unlock(); |
b8aa1682 JD |
1859 | goto error; |
1860 | } | |
1861 | relay_connection = caa_container_of(node, | |
1862 | struct relay_command, sock_n); | |
1863 | ||
1864 | if (revents & (LPOLLERR)) { | |
1865 | ERR("POLL ERROR"); | |
9d1bbf21 MD |
1866 | relay_cleanup_poll_connection(&events, pollfd); |
1867 | relay_del_connection(relay_connections_ht, | |
1868 | streams_ht, &iter, | |
1869 | relay_connection); | |
beaad64c DG |
1870 | if (last_seen_data_fd == pollfd) { |
1871 | last_seen_data_fd = last_notdel_data_fd; | |
1872 | } | |
b8aa1682 JD |
1873 | } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { |
1874 | DBG("Socket %d hung up", pollfd); | |
9d1bbf21 MD |
1875 | relay_cleanup_poll_connection(&events, pollfd); |
1876 | relay_del_connection(relay_connections_ht, | |
1877 | streams_ht, &iter, | |
1878 | relay_connection); | |
beaad64c DG |
1879 | if (last_seen_data_fd == pollfd) { |
1880 | last_seen_data_fd = last_notdel_data_fd; | |
1881 | } | |
b8aa1682 JD |
1882 | } else if (revents & LPOLLIN) { |
1883 | /* control socket */ | |
1884 | if (relay_connection->type == RELAY_CONTROL) { | |
1885 | ret = relay_connection->sock->ops->recvmsg( | |
1886 | relay_connection->sock, &recv_hdr, | |
7c5aef62 | 1887 | sizeof(struct lttcomm_relayd_hdr), 0); |
b8aa1682 JD |
1888 | /* connection closed */ |
1889 | if (ret <= 0) { | |
9d1bbf21 MD |
1890 | relay_cleanup_poll_connection(&events, pollfd); |
1891 | relay_del_connection(relay_connections_ht, | |
1892 | streams_ht, &iter, | |
1893 | relay_connection); | |
b8aa1682 JD |
1894 | DBG("Control connection closed with %d", pollfd); |
1895 | } else { | |
1896 | if (relay_connection->session) { | |
77c7c900 | 1897 | DBG2("Relay worker receiving data for session : %" PRIu64, |
b8aa1682 JD |
1898 | relay_connection->session->id); |
1899 | } | |
1900 | ret = relay_process_control(&recv_hdr, | |
1901 | relay_connection, | |
1902 | streams_ht); | |
b8aa1682 | 1903 | if (ret < 0) { |
beaad64c | 1904 | /* Clear the session on error. */ |
9d1bbf21 MD |
1905 | relay_cleanup_poll_connection(&events, pollfd); |
1906 | relay_del_connection(relay_connections_ht, | |
1907 | streams_ht, &iter, | |
1908 | relay_connection); | |
b8aa1682 JD |
1909 | DBG("Connection closed with %d", pollfd); |
1910 | } | |
beaad64c | 1911 | seen_control = 1; |
b8aa1682 | 1912 | } |
beaad64c DG |
1913 | } else { |
1914 | /* | |
1915 | * Flag the last seen data fd not deleted. It will be | |
1916 | * used as the last seen fd if any fd gets deleted in | |
1917 | * this first loop. | |
1918 | */ | |
1919 | last_notdel_data_fd = pollfd; | |
1920 | } | |
1921 | } | |
1922 | rcu_read_unlock(); | |
1923 | } | |
1924 | } | |
1925 | ||
1926 | /* | |
1927 | * The last loop handled a control request, go back to poll to make | |
1928 | * sure we prioritise the control socket. | |
1929 | */ | |
1930 | if (seen_control) { | |
1931 | continue; | |
1932 | } | |
1933 | ||
1934 | if (last_seen_data_fd >= 0) { | |
1935 | for (i = 0; i < nb_fd; i++) { | |
1936 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
1937 | if (last_seen_data_fd == pollfd) { | |
1938 | idx = i; | |
1939 | break; | |
1940 | } | |
1941 | } | |
1942 | } | |
1943 | ||
1944 | /* Process data connection. */ | |
1945 | for (i = idx + 1; i < nb_fd; i++) { | |
1946 | /* Fetch the poll data. */ | |
1947 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); | |
1948 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
1949 | ||
1950 | /* Skip the command pipe. It's handled in the first loop. */ | |
1951 | if (pollfd == relay_cmd_pipe[0]) { | |
1952 | continue; | |
1953 | } | |
1954 | ||
1955 | if (revents) { | |
1956 | rcu_read_lock(); | |
1957 | lttng_ht_lookup(relay_connections_ht, | |
1958 | (void *)((unsigned long) pollfd), | |
1959 | &iter); | |
1960 | node = lttng_ht_iter_get_node_ulong(&iter); | |
1961 | if (node == NULL) { | |
1962 | /* Skip it. Might be removed before. */ | |
1963 | rcu_read_unlock(); | |
1964 | continue; | |
1965 | } | |
1966 | relay_connection = caa_container_of(node, | |
1967 | struct relay_command, sock_n); | |
1968 | ||
1969 | if (revents & LPOLLIN) { | |
1970 | if (relay_connection->type != RELAY_DATA) { | |
1971 | continue; | |
1972 | } | |
1973 | ||
1974 | ret = relay_process_data(relay_connection, streams_ht); | |
1975 | /* connection closed */ | |
1976 | if (ret < 0) { | |
1977 | relay_cleanup_poll_connection(&events, pollfd); | |
1978 | relay_del_connection(relay_connections_ht, | |
1979 | streams_ht, &iter, | |
1980 | relay_connection); | |
1981 | DBG("Data connection closed with %d", pollfd); | |
1982 | /* | |
1983 | * Every goto restart call sets the last seen fd where | |
1984 | * here we don't really care since we gracefully | |
1985 | * continue the loop after the connection is deleted. | |
1986 | */ | |
1987 | } else { | |
1988 | /* Keep last seen port. */ | |
1989 | last_seen_data_fd = pollfd; | |
1990 | rcu_read_unlock(); | |
1991 | goto restart; | |
b8aa1682 JD |
1992 | } |
1993 | } | |
9d1bbf21 | 1994 | rcu_read_unlock(); |
b8aa1682 JD |
1995 | } |
1996 | } | |
beaad64c | 1997 | last_seen_data_fd = -1; |
b8aa1682 JD |
1998 | } |
1999 | ||
095a4ae5 | 2000 | exit: |
b8aa1682 JD |
2001 | error: |
2002 | lttng_poll_clean(&events); | |
2003 | ||
2004 | /* empty the hash table and free the memory */ | |
9d1bbf21 | 2005 | rcu_read_lock(); |
b8aa1682 JD |
2006 | cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { |
2007 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2008 | if (node) { | |
2009 | relay_connection = caa_container_of(node, | |
2010 | struct relay_command, sock_n); | |
9d1bbf21 MD |
2011 | relay_del_connection(relay_connections_ht, |
2012 | streams_ht, &iter, | |
2013 | relay_connection); | |
b8aa1682 | 2014 | } |
b8aa1682 | 2015 | } |
9d1bbf21 | 2016 | rcu_read_unlock(); |
b8aa1682 | 2017 | error_poll_create: |
095a4ae5 MD |
2018 | lttng_ht_destroy(streams_ht); |
2019 | streams_ht_error: | |
b8aa1682 | 2020 | lttng_ht_destroy(relay_connections_ht); |
095a4ae5 | 2021 | relay_connections_ht_error: |
6620da75 DG |
2022 | /* Close relay cmd pipes */ |
2023 | utils_close_pipe(relay_cmd_pipe); | |
095a4ae5 MD |
2024 | if (err) { |
2025 | DBG("Thread exited with error"); | |
2026 | } | |
b8aa1682 | 2027 | DBG("Worker thread cleanup complete"); |
095a4ae5 | 2028 | free(data_buffer); |
b8aa1682 | 2029 | stop_threads(); |
9d1bbf21 | 2030 | rcu_unregister_thread(); |
b8aa1682 JD |
2031 | return NULL; |
2032 | } | |
2033 | ||
2034 | /* | |
2035 | * Create the relay command pipe to wake thread_manage_apps. | |
2036 | * Closed in cleanup(). | |
2037 | */ | |
2038 | static int create_relay_cmd_pipe(void) | |
2039 | { | |
a02de639 | 2040 | int ret; |
b8aa1682 | 2041 | |
a02de639 | 2042 | ret = utils_create_pipe_cloexec(relay_cmd_pipe); |
b8aa1682 | 2043 | |
b8aa1682 JD |
2044 | return ret; |
2045 | } | |
2046 | ||
2047 | /* | |
2048 | * main | |
2049 | */ | |
2050 | int main(int argc, char **argv) | |
2051 | { | |
2052 | int ret = 0; | |
2053 | void *status; | |
2054 | ||
2055 | /* Create thread quit pipe */ | |
2056 | if ((ret = init_thread_quit_pipe()) < 0) { | |
2057 | goto error; | |
2058 | } | |
2059 | ||
2060 | /* Parse arguments */ | |
2061 | progname = argv[0]; | |
c617c0c6 | 2062 | if ((ret = parse_args(argc, argv)) < 0) { |
a02de639 | 2063 | goto exit; |
b8aa1682 JD |
2064 | } |
2065 | ||
2066 | if ((ret = set_signal_handler()) < 0) { | |
2067 | goto exit; | |
2068 | } | |
2069 | ||
4d513a50 DG |
2070 | /* Try to create directory if -o, --output is specified. */ |
2071 | if (opt_output_path) { | |
2072 | ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); | |
2073 | if (ret < 0) { | |
2074 | ERR("Unable to create %s", opt_output_path); | |
2075 | goto exit; | |
2076 | } | |
2077 | } | |
2078 | ||
b8aa1682 JD |
2079 | /* Daemonize */ |
2080 | if (opt_daemon) { | |
2081 | ret = daemon(0, 0); | |
2082 | if (ret < 0) { | |
2083 | PERROR("daemon"); | |
a02de639 | 2084 | goto exit; |
b8aa1682 JD |
2085 | } |
2086 | } | |
2087 | ||
2088 | /* Check if daemon is UID = 0 */ | |
2089 | is_root = !getuid(); | |
2090 | ||
2091 | if (!is_root) { | |
2092 | if (control_uri->port < 1024 || data_uri->port < 1024) { | |
2093 | ERR("Need to be root to use ports < 1024"); | |
2094 | ret = -1; | |
a02de639 | 2095 | goto exit; |
b8aa1682 JD |
2096 | } |
2097 | } | |
2098 | ||
2099 | /* Setup the thread apps communication pipe. */ | |
2100 | if ((ret = create_relay_cmd_pipe()) < 0) { | |
2101 | goto exit; | |
2102 | } | |
2103 | ||
2104 | /* Init relay command queue. */ | |
2105 | cds_wfq_init(&relay_cmd_queue.queue); | |
2106 | ||
2107 | /* Set up max poll set size */ | |
2108 | lttng_poll_set_max_size(); | |
2109 | ||
2110 | /* Setup the dispatcher thread */ | |
2111 | ret = pthread_create(&dispatcher_thread, NULL, | |
2112 | relay_thread_dispatcher, (void *) NULL); | |
2113 | if (ret != 0) { | |
2114 | PERROR("pthread_create dispatcher"); | |
2115 | goto exit_dispatcher; | |
2116 | } | |
2117 | ||
2118 | /* Setup the worker thread */ | |
2119 | ret = pthread_create(&worker_thread, NULL, | |
2120 | relay_thread_worker, (void *) NULL); | |
2121 | if (ret != 0) { | |
2122 | PERROR("pthread_create worker"); | |
2123 | goto exit_worker; | |
2124 | } | |
2125 | ||
2126 | /* Setup the listener thread */ | |
2127 | ret = pthread_create(&listener_thread, NULL, | |
2128 | relay_thread_listener, (void *) NULL); | |
2129 | if (ret != 0) { | |
2130 | PERROR("pthread_create listener"); | |
2131 | goto exit_listener; | |
2132 | } | |
2133 | ||
2134 | exit_listener: | |
2135 | ret = pthread_join(listener_thread, &status); | |
2136 | if (ret != 0) { | |
2137 | PERROR("pthread_join"); | |
2138 | goto error; /* join error, exit without cleanup */ | |
2139 | } | |
2140 | ||
2141 | exit_worker: | |
2142 | ret = pthread_join(worker_thread, &status); | |
2143 | if (ret != 0) { | |
2144 | PERROR("pthread_join"); | |
2145 | goto error; /* join error, exit without cleanup */ | |
2146 | } | |
2147 | ||
2148 | exit_dispatcher: | |
2149 | ret = pthread_join(dispatcher_thread, &status); | |
2150 | if (ret != 0) { | |
2151 | PERROR("pthread_join"); | |
2152 | goto error; /* join error, exit without cleanup */ | |
2153 | } | |
2154 | ||
2155 | exit: | |
2156 | cleanup(); | |
2157 | if (!ret) { | |
2158 | exit(EXIT_SUCCESS); | |
2159 | } | |
a02de639 | 2160 | |
b8aa1682 JD |
2161 | error: |
2162 | exit(EXIT_FAILURE); | |
2163 | } |