d8105e4058128bdc9a902bb8051fc991971f2b90
[lttng-tools.git] / src / bin / lttng-relayd / main.c
1 /*
2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _GNU_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <pthread.h>
25 #include <signal.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <inttypes.h>
37 #include <urcu/futex.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40 #include <fcntl.h>
41 #include <config.h>
42
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/defaults.h>
48 #include <common/daemonize.h>
49 #include <common/futex.h>
50 #include <common/sessiond-comm/sessiond-comm.h>
51 #include <common/sessiond-comm/inet.h>
52 #include <common/sessiond-comm/relayd.h>
53 #include <common/uri.h>
54 #include <common/utils.h>
55 #include <common/config/config.h>
56
57 #include "cmd.h"
58 #include "ctf-trace.h"
59 #include "index.h"
60 #include "utils.h"
61 #include "lttng-relayd.h"
62 #include "live.h"
63 #include "health-relayd.h"
64 #include "testpoint.h"
65
66 /* command line options */
67 char *opt_output_path;
68 static int opt_daemon, opt_background;
69
70 /*
71 * We need to wait for listener and live listener threads, as well as
72 * health check thread, before being ready to signal readiness.
73 */
74 #define NR_LTTNG_RELAY_READY 3
75 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
76 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
77 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
78
79 static struct lttng_uri *control_uri;
80 static struct lttng_uri *data_uri;
81 static struct lttng_uri *live_uri;
82
83 const char *progname;
84
85 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
86 static int tracing_group_name_override;
87
88 const char * const config_section_name = "relayd";
89
90 /*
91 * Quit pipe for all threads. This permits a single cancellation point
92 * for all threads when receiving an event on the pipe.
93 */
94 int thread_quit_pipe[2] = { -1, -1 };
95
96 /*
97 * This pipe is used to inform the worker thread that a command is queued and
98 * ready to be processed.
99 */
100 static int relay_cmd_pipe[2] = { -1, -1 };
101
102 /* Shared between threads */
103 static int dispatch_thread_exit;
104
105 static pthread_t listener_thread;
106 static pthread_t dispatcher_thread;
107 static pthread_t worker_thread;
108 static pthread_t health_thread;
109
110 static uint64_t last_relay_stream_id;
111 static uint64_t last_relay_session_id;
112
113 /*
114 * Relay command queue.
115 *
116 * The relay_thread_listener and relay_thread_dispatcher communicate with this
117 * queue.
118 */
119 static struct relay_cmd_queue relay_cmd_queue;
120
121 /* buffer allocated at startup, used to store the trace data */
122 static char *data_buffer;
123 static unsigned int data_buffer_size;
124
125 /* We need those values for the file/dir creation. */
126 static uid_t relayd_uid;
127 static gid_t relayd_gid;
128
129 /* Global relay stream hash table. */
130 struct lttng_ht *relay_streams_ht;
131
132 /* Global relay viewer stream hash table. */
133 struct lttng_ht *viewer_streams_ht;
134
135 /* Global hash table that stores relay index object. */
136 struct lttng_ht *indexes_ht;
137
138 /* Relayd health monitoring */
139 struct health_app *health_relayd;
140
141 static struct option long_options[] = {
142 { "control-port", 1, 0, 'C', },
143 { "data-port", 1, 0, 'D', },
144 { "live-port", 1, 0, 'L', },
145 { "daemonize", 0, 0, 'd', },
146 { "background", 0, 0, 'b', },
147 { "group", 1, 0, 'g', },
148 { "help", 0, 0, 'h', },
149 { "output", 1, 0, 'o', },
150 { "verbose", 0, 0, 'v', },
151 { "config", 1, 0, 'f' },
152 { NULL, 0, 0, 0, },
153 };
154
155 static const char *config_ignore_options[] = { "help", "config" };
156
157 /*
158 * usage function on stderr
159 */
160 static
161 void usage(void)
162 {
163 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
164 fprintf(stderr, " -h, --help Display this usage.\n");
165 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
166 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
167 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
168 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
169 fprintf(stderr, " -L, --live-port URL Live view port listening.\n");
170 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
171 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
172 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
173 fprintf(stderr, " -f --config Load daemon configuration file\n");
174 }
175
176 /*
177 * Take an option from the getopt output and set it in the right variable to be
178 * used later.
179 *
180 * Return 0 on success else a negative value.
181 */
182 static
183 int set_option(int opt, const char *arg, const char *optname)
184 {
185 int ret;
186
187 switch (opt) {
188 case 0:
189 fprintf(stderr, "option %s", optname);
190 if (arg) {
191 fprintf(stderr, " with arg %s\n", arg);
192 }
193 break;
194 case 'C':
195 ret = uri_parse(arg, &control_uri);
196 if (ret < 0) {
197 ERR("Invalid control URI specified");
198 goto end;
199 }
200 if (control_uri->port == 0) {
201 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
202 }
203 break;
204 case 'D':
205 ret = uri_parse(arg, &data_uri);
206 if (ret < 0) {
207 ERR("Invalid data URI specified");
208 goto end;
209 }
210 if (data_uri->port == 0) {
211 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
212 }
213 break;
214 case 'L':
215 ret = uri_parse(arg, &live_uri);
216 if (ret < 0) {
217 ERR("Invalid live URI specified");
218 goto end;
219 }
220 if (live_uri->port == 0) {
221 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
222 }
223 break;
224 case 'd':
225 opt_daemon = 1;
226 break;
227 case 'b':
228 opt_background = 1;
229 break;
230 case 'g':
231 tracing_group_name = strdup(arg);
232 tracing_group_name_override = 1;
233 break;
234 case 'h':
235 usage();
236 exit(EXIT_FAILURE);
237 case 'o':
238 ret = asprintf(&opt_output_path, "%s", arg);
239 if (ret < 0) {
240 ret = -errno;
241 PERROR("asprintf opt_output_path");
242 goto end;
243 }
244 break;
245 case 'v':
246 /* Verbose level can increase using multiple -v */
247 if (arg) {
248 lttng_opt_verbose = config_parse_value(arg);
249 } else {
250 lttng_opt_verbose += 1;
251 }
252 break;
253 default:
254 /* Unknown option or other error.
255 * Error is printed by getopt, just return */
256 ret = -1;
257 goto end;
258 }
259
260 /* All good. */
261 ret = 0;
262
263 end:
264 return ret;
265 }
266
267 /*
268 * config_entry_handler_cb used to handle options read from a config file.
269 * See config_entry_handler_cb comment in common/config/config.h for the
270 * return value conventions.
271 */
272 static
273 int config_entry_handler(const struct config_entry *entry, void *unused)
274 {
275 int ret = 0, i;
276
277 if (!entry || !entry->name || !entry->value) {
278 ret = -EINVAL;
279 goto end;
280 }
281
282 /* Check if the option is to be ignored */
283 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
284 if (!strcmp(entry->name, config_ignore_options[i])) {
285 goto end;
286 }
287 }
288
289 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
290 /* Ignore if entry name is not fully matched. */
291 if (strcmp(entry->name, long_options[i].name)) {
292 continue;
293 }
294
295 /*
296 * If the option takes no argument on the command line, we have to
297 * check if the value is "true". We support non-zero numeric values,
298 * true, on and yes.
299 */
300 if (!long_options[i].has_arg) {
301 ret = config_parse_value(entry->value);
302 if (ret <= 0) {
303 if (ret) {
304 WARN("Invalid configuration value \"%s\" for option %s",
305 entry->value, entry->name);
306 }
307 /* False, skip boolean config option. */
308 goto end;
309 }
310 }
311
312 ret = set_option(long_options[i].val, entry->value, entry->name);
313 goto end;
314 }
315
316 WARN("Unrecognized option \"%s\" in daemon configuration file.",
317 entry->name);
318
319 end:
320 return ret;
321 }
322
323 static
324 int set_options(int argc, char **argv)
325 {
326 int c, ret = 0, option_index = 0;
327 int orig_optopt = optopt, orig_optind = optind;
328 char *default_address, *optstring;
329 const char *config_path = NULL;
330
331 optstring = utils_generate_optstring(long_options,
332 sizeof(long_options) / sizeof(struct option));
333 if (!optstring) {
334 ret = -ENOMEM;
335 goto exit;
336 }
337
338 /* Check for the --config option */
339
340 while ((c = getopt_long(argc, argv, optstring, long_options,
341 &option_index)) != -1) {
342 if (c == '?') {
343 ret = -EINVAL;
344 goto exit;
345 } else if (c != 'f') {
346 continue;
347 }
348
349 config_path = utils_expand_path(optarg);
350 if (!config_path) {
351 ERR("Failed to resolve path: %s", optarg);
352 }
353 }
354
355 ret = config_get_section_entries(config_path, config_section_name,
356 config_entry_handler, NULL);
357 if (ret) {
358 if (ret > 0) {
359 ERR("Invalid configuration option at line %i", ret);
360 ret = -1;
361 }
362 goto exit;
363 }
364
365 /* Reset getopt's global state */
366 optopt = orig_optopt;
367 optind = orig_optind;
368 while (1) {
369 c = getopt_long(argc, argv, optstring, long_options, &option_index);
370 if (c == -1) {
371 break;
372 }
373
374 ret = set_option(c, optarg, long_options[option_index].name);
375 if (ret < 0) {
376 goto exit;
377 }
378 }
379
380 /* assign default values */
381 if (control_uri == NULL) {
382 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
383 DEFAULT_NETWORK_CONTROL_PORT);
384 if (ret < 0) {
385 PERROR("asprintf default data address");
386 goto exit;
387 }
388
389 ret = uri_parse(default_address, &control_uri);
390 free(default_address);
391 if (ret < 0) {
392 ERR("Invalid control URI specified");
393 goto exit;
394 }
395 }
396 if (data_uri == NULL) {
397 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
398 DEFAULT_NETWORK_DATA_PORT);
399 if (ret < 0) {
400 PERROR("asprintf default data address");
401 goto exit;
402 }
403
404 ret = uri_parse(default_address, &data_uri);
405 free(default_address);
406 if (ret < 0) {
407 ERR("Invalid data URI specified");
408 goto exit;
409 }
410 }
411 if (live_uri == NULL) {
412 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
413 DEFAULT_NETWORK_VIEWER_PORT);
414 if (ret < 0) {
415 PERROR("asprintf default viewer control address");
416 goto exit;
417 }
418
419 ret = uri_parse(default_address, &live_uri);
420 free(default_address);
421 if (ret < 0) {
422 ERR("Invalid viewer control URI specified");
423 goto exit;
424 }
425 }
426
427 exit:
428 free(optstring);
429 return ret;
430 }
431
432 /*
433 * Cleanup the daemon
434 */
435 static
436 void cleanup(void)
437 {
438 DBG("Cleaning up");
439
440 /* free the dynamically allocated opt_output_path */
441 free(opt_output_path);
442
443 /* Close thread quit pipes */
444 utils_close_pipe(thread_quit_pipe);
445
446 uri_free(control_uri);
447 uri_free(data_uri);
448 /* Live URI is freed in the live thread. */
449
450 if (tracing_group_name_override) {
451 free((void *) tracing_group_name);
452 }
453 }
454
455 /*
456 * Write to writable pipe used to notify a thread.
457 */
458 static
459 int notify_thread_pipe(int wpipe)
460 {
461 ssize_t ret;
462
463 ret = lttng_write(wpipe, "!", 1);
464 if (ret < 1) {
465 PERROR("write poll pipe");
466 }
467
468 return ret;
469 }
470
471 static void notify_health_quit_pipe(int *pipe)
472 {
473 ssize_t ret;
474
475 ret = lttng_write(pipe[1], "4", 1);
476 if (ret < 1) {
477 PERROR("write relay health quit");
478 }
479 }
480
481 /*
482 * Stop all threads by closing the thread quit pipe.
483 */
484 static
485 void stop_threads(void)
486 {
487 int ret;
488
489 /* Stopping all threads */
490 DBG("Terminating all threads");
491 ret = notify_thread_pipe(thread_quit_pipe[1]);
492 if (ret < 0) {
493 ERR("write error on thread quit pipe");
494 }
495
496 notify_health_quit_pipe(health_quit_pipe);
497
498 /* Dispatch thread */
499 CMM_STORE_SHARED(dispatch_thread_exit, 1);
500 futex_nto1_wake(&relay_cmd_queue.futex);
501 }
502
503 /*
504 * Signal handler for the daemon
505 *
506 * Simply stop all worker threads, leaving main() return gracefully after
507 * joining all threads and calling cleanup().
508 */
509 static
510 void sighandler(int sig)
511 {
512 switch (sig) {
513 case SIGPIPE:
514 DBG("SIGPIPE caught");
515 return;
516 case SIGINT:
517 DBG("SIGINT caught");
518 stop_threads();
519 break;
520 case SIGTERM:
521 DBG("SIGTERM caught");
522 stop_threads();
523 break;
524 case SIGUSR1:
525 CMM_STORE_SHARED(recv_child_signal, 1);
526 break;
527 default:
528 break;
529 }
530 }
531
532 /*
533 * Setup signal handler for :
534 * SIGINT, SIGTERM, SIGPIPE
535 */
536 static
537 int set_signal_handler(void)
538 {
539 int ret = 0;
540 struct sigaction sa;
541 sigset_t sigset;
542
543 if ((ret = sigemptyset(&sigset)) < 0) {
544 PERROR("sigemptyset");
545 return ret;
546 }
547
548 sa.sa_handler = sighandler;
549 sa.sa_mask = sigset;
550 sa.sa_flags = 0;
551 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
552 PERROR("sigaction");
553 return ret;
554 }
555
556 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
557 PERROR("sigaction");
558 return ret;
559 }
560
561 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
562 PERROR("sigaction");
563 return ret;
564 }
565
566 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
567 PERROR("sigaction");
568 return ret;
569 }
570
571 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
572
573 return ret;
574 }
575
576 void lttng_relay_notify_ready(void)
577 {
578 /* Notify the parent of the fork() process that we are ready. */
579 if (opt_daemon || opt_background) {
580 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
581 kill(child_ppid, SIGUSR1);
582 }
583 }
584 }
585
586 /*
587 * Init thread quit pipe.
588 *
589 * Return -1 on error or 0 if all pipes are created.
590 */
591 static
592 int init_thread_quit_pipe(void)
593 {
594 int ret;
595
596 ret = utils_create_pipe_cloexec(thread_quit_pipe);
597
598 return ret;
599 }
600
601 /*
602 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
603 */
604 static
605 int create_thread_poll_set(struct lttng_poll_event *events, int size)
606 {
607 int ret;
608
609 if (events == NULL || size == 0) {
610 ret = -1;
611 goto error;
612 }
613
614 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
615 if (ret < 0) {
616 goto error;
617 }
618
619 /* Add quit pipe */
620 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
621 if (ret < 0) {
622 goto error;
623 }
624
625 return 0;
626
627 error:
628 return ret;
629 }
630
631 /*
632 * Check if the thread quit pipe was triggered.
633 *
634 * Return 1 if it was triggered else 0;
635 */
636 static
637 int check_thread_quit_pipe(int fd, uint32_t events)
638 {
639 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
640 return 1;
641 }
642
643 return 0;
644 }
645
646 /*
647 * Create and init socket from uri.
648 */
649 static
650 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
651 {
652 int ret;
653 struct lttcomm_sock *sock = NULL;
654
655 sock = lttcomm_alloc_sock_from_uri(uri);
656 if (sock == NULL) {
657 ERR("Allocating socket");
658 goto error;
659 }
660
661 ret = lttcomm_create_sock(sock);
662 if (ret < 0) {
663 goto error;
664 }
665 DBG("Listening on sock %d", sock->fd);
666
667 ret = sock->ops->bind(sock);
668 if (ret < 0) {
669 goto error;
670 }
671
672 ret = sock->ops->listen(sock, -1);
673 if (ret < 0) {
674 goto error;
675
676 }
677
678 return sock;
679
680 error:
681 if (sock) {
682 lttcomm_destroy_sock(sock);
683 }
684 return NULL;
685 }
686
687 /*
688 * Return nonzero if stream needs to be closed.
689 */
690 static
691 int close_stream_check(struct relay_stream *stream)
692 {
693
694 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
695 /*
696 * We are about to close the stream so set the data pending flag to 1
697 * which will make the end data pending command skip the stream which
698 * is now closed and ready. Note that after proceeding to a file close,
699 * the written file is ready for reading.
700 */
701 stream->data_pending_check_done = 1;
702 return 1;
703 }
704 return 0;
705 }
706
707 /*
708 * This thread manages the listening for new connections on the network
709 */
710 static
711 void *relay_thread_listener(void *data)
712 {
713 int i, ret, pollfd, err = -1;
714 int val = 1;
715 uint32_t revents, nb_fd;
716 struct lttng_poll_event events;
717 struct lttcomm_sock *control_sock, *data_sock;
718
719 DBG("[thread] Relay listener started");
720
721 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
722
723 health_code_update();
724
725 control_sock = relay_init_sock(control_uri);
726 if (!control_sock) {
727 goto error_sock_control;
728 }
729
730 data_sock = relay_init_sock(data_uri);
731 if (!data_sock) {
732 goto error_sock_relay;
733 }
734
735 /*
736 * Pass 3 as size here for the thread quit pipe, control and data socket.
737 */
738 ret = create_thread_poll_set(&events, 3);
739 if (ret < 0) {
740 goto error_create_poll;
741 }
742
743 /* Add the control socket */
744 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
745 if (ret < 0) {
746 goto error_poll_add;
747 }
748
749 /* Add the data socket */
750 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
751 if (ret < 0) {
752 goto error_poll_add;
753 }
754
755 lttng_relay_notify_ready();
756
757 if (testpoint(relayd_thread_listener)) {
758 goto error_testpoint;
759 }
760
761 while (1) {
762 health_code_update();
763
764 DBG("Listener accepting connections");
765
766 restart:
767 health_poll_entry();
768 ret = lttng_poll_wait(&events, -1);
769 health_poll_exit();
770 if (ret < 0) {
771 /*
772 * Restart interrupted system call.
773 */
774 if (errno == EINTR) {
775 goto restart;
776 }
777 goto error;
778 }
779
780 nb_fd = ret;
781
782 DBG("Relay new connection received");
783 for (i = 0; i < nb_fd; i++) {
784 health_code_update();
785
786 /* Fetch once the poll data */
787 revents = LTTNG_POLL_GETEV(&events, i);
788 pollfd = LTTNG_POLL_GETFD(&events, i);
789
790 /* Thread quit pipe has been closed. Killing thread. */
791 ret = check_thread_quit_pipe(pollfd, revents);
792 if (ret) {
793 err = 0;
794 goto exit;
795 }
796
797 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
798 ERR("socket poll error");
799 goto error;
800 } else if (revents & LPOLLIN) {
801 /*
802 * Get allocated in this thread,
803 * enqueued to a global queue, dequeued
804 * and freed in the worker thread.
805 */
806 struct relay_command *relay_cmd;
807 struct lttcomm_sock *newsock;
808
809 relay_cmd = zmalloc(sizeof(struct relay_command));
810 if (relay_cmd == NULL) {
811 PERROR("relay command zmalloc");
812 goto error;
813 }
814
815 if (pollfd == data_sock->fd) {
816 newsock = data_sock->ops->accept(data_sock);
817 if (!newsock) {
818 PERROR("accepting data sock");
819 free(relay_cmd);
820 goto error;
821 }
822 relay_cmd->type = RELAY_DATA;
823 DBG("Relay data connection accepted, socket %d", newsock->fd);
824 } else {
825 assert(pollfd == control_sock->fd);
826 newsock = control_sock->ops->accept(control_sock);
827 if (!newsock) {
828 PERROR("accepting control sock");
829 free(relay_cmd);
830 goto error;
831 }
832 relay_cmd->type = RELAY_CONTROL;
833 DBG("Relay control connection accepted, socket %d", newsock->fd);
834 }
835 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
836 &val, sizeof(int));
837 if (ret < 0) {
838 PERROR("setsockopt inet");
839 lttcomm_destroy_sock(newsock);
840 free(relay_cmd);
841 goto error;
842 }
843 relay_cmd->sock = newsock;
844 /*
845 * Lock free enqueue the request.
846 */
847 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
848
849 /*
850 * Wake the dispatch queue futex. Implicit memory
851 * barrier with the exchange in cds_wfq_enqueue.
852 */
853 futex_nto1_wake(&relay_cmd_queue.futex);
854 }
855 }
856 }
857
858 exit:
859 error:
860 error_poll_add:
861 error_testpoint:
862 lttng_poll_clean(&events);
863 error_create_poll:
864 if (data_sock->fd >= 0) {
865 ret = data_sock->ops->close(data_sock);
866 if (ret) {
867 PERROR("close");
868 }
869 }
870 lttcomm_destroy_sock(data_sock);
871 error_sock_relay:
872 if (control_sock->fd >= 0) {
873 ret = control_sock->ops->close(control_sock);
874 if (ret) {
875 PERROR("close");
876 }
877 }
878 lttcomm_destroy_sock(control_sock);
879 error_sock_control:
880 if (err) {
881 health_error();
882 ERR("Health error occurred in %s", __func__);
883 }
884 health_unregister(health_relayd);
885 DBG("Relay listener thread cleanup complete");
886 stop_threads();
887 return NULL;
888 }
889
890 /*
891 * This thread manages the dispatching of the requests to worker threads
892 */
893 static
894 void *relay_thread_dispatcher(void *data)
895 {
896 int err = -1;
897 ssize_t ret;
898 struct cds_wfq_node *node;
899 struct relay_command *relay_cmd = NULL;
900
901 DBG("[thread] Relay dispatcher started");
902
903 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
904
905 if (testpoint(relayd_thread_dispatcher)) {
906 goto error_testpoint;
907 }
908
909 health_code_update();
910
911 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
912 health_code_update();
913
914 /* Atomically prepare the queue futex */
915 futex_nto1_prepare(&relay_cmd_queue.futex);
916
917 do {
918 health_code_update();
919
920 /* Dequeue commands */
921 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
922 if (node == NULL) {
923 DBG("Woken up but nothing in the relay command queue");
924 /* Continue thread execution */
925 break;
926 }
927
928 relay_cmd = caa_container_of(node, struct relay_command, node);
929 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
930
931 /*
932 * Inform worker thread of the new request. This
933 * call is blocking so we can be assured that the data will be read
934 * at some point in time or wait to the end of the world :)
935 */
936 ret = lttng_write(relay_cmd_pipe[1], relay_cmd,
937 sizeof(struct relay_command));
938 free(relay_cmd);
939 if (ret < sizeof(struct relay_command)) {
940 PERROR("write cmd pipe");
941 goto error;
942 }
943 } while (node != NULL);
944
945 /* Futex wait on queue. Blocking call on futex() */
946 health_poll_entry();
947 futex_nto1_wait(&relay_cmd_queue.futex);
948 health_poll_exit();
949 }
950
951 /* Normal exit, no error */
952 err = 0;
953
954 error:
955 error_testpoint:
956 if (err) {
957 health_error();
958 ERR("Health error occurred in %s", __func__);
959 }
960 health_unregister(health_relayd);
961 DBG("Dispatch thread dying");
962 stop_threads();
963 return NULL;
964 }
965
966 /*
967 * Get stream from stream id.
968 * Need to be called with RCU read-side lock held.
969 */
970 struct relay_stream *relay_stream_find_by_id(uint64_t stream_id)
971 {
972 struct lttng_ht_node_ulong *node;
973 struct lttng_ht_iter iter;
974 struct relay_stream *ret;
975
976 lttng_ht_lookup(relay_streams_ht,
977 (void *)((unsigned long) stream_id),
978 &iter);
979 node = lttng_ht_iter_get_node_ulong(&iter);
980 if (node == NULL) {
981 DBG("Relay stream %" PRIu64 " not found", stream_id);
982 ret = NULL;
983 goto end;
984 }
985
986 ret = caa_container_of(node, struct relay_stream, stream_n);
987
988 end:
989 return ret;
990 }
991
992 static
993 void deferred_free_stream(struct rcu_head *head)
994 {
995 struct relay_stream *stream =
996 caa_container_of(head, struct relay_stream, rcu_node);
997
998 free(stream->path_name);
999 free(stream->channel_name);
1000 free(stream);
1001 }
1002
1003 static
1004 void deferred_free_session(struct rcu_head *head)
1005 {
1006 struct relay_session *session =
1007 caa_container_of(head, struct relay_session, rcu_node);
1008 free(session);
1009 }
1010
1011 /*
1012 * Close a given stream. The stream is freed using a call RCU.
1013 *
1014 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
1015 * BEFORE the stream lock MUST be acquired.
1016 */
1017 static void destroy_stream(struct relay_stream *stream)
1018 {
1019 int delret;
1020 struct relay_viewer_stream *vstream;
1021 struct lttng_ht_iter iter;
1022
1023 assert(stream);
1024
1025 delret = close(stream->fd);
1026 if (delret < 0) {
1027 PERROR("close stream");
1028 }
1029
1030 if (stream->index_fd >= 0) {
1031 delret = close(stream->index_fd);
1032 if (delret < 0) {
1033 PERROR("close stream index_fd");
1034 }
1035 }
1036
1037 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
1038 if (vstream) {
1039 /*
1040 * Set the last good value into the viewer stream. This is done
1041 * right before the stream gets deleted from the hash table. The
1042 * lookup failure on the live thread side of a stream indicates
1043 * that the viewer stream index received value should be used.
1044 */
1045 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
1046 vstream->total_index_received = stream->total_index_received;
1047 vstream->tracefile_count_last = stream->tracefile_count_current;
1048 vstream->close_write_flag = 1;
1049 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
1050 }
1051
1052 /* Cleanup index of that stream. */
1053 relay_index_destroy_by_stream_id(stream->stream_handle);
1054
1055 iter.iter.node = &stream->stream_n.node;
1056 delret = lttng_ht_del(relay_streams_ht, &iter);
1057 assert(!delret);
1058 iter.iter.node = &stream->ctf_trace_node.node;
1059 delret = lttng_ht_del(stream->ctf_traces_ht, &iter);
1060 assert(!delret);
1061
1062 if (stream->ctf_trace) {
1063 ctf_trace_try_destroy(stream->ctf_trace);
1064 }
1065
1066 call_rcu(&stream->rcu_node, deferred_free_stream);
1067 DBG("Closed tracefile %d from close stream", stream->fd);
1068 }
1069
1070 /*
1071 * relay_delete_session: Free all memory associated with a session and
1072 * close all the FDs
1073 */
1074 static
1075 void relay_delete_session(struct relay_command *cmd,
1076 struct lttng_ht *sessions_ht)
1077 {
1078 struct lttng_ht_iter iter;
1079 struct lttng_ht_node_ulong *node;
1080 struct relay_stream *stream;
1081 int ret;
1082
1083 if (!cmd->session) {
1084 return;
1085 }
1086
1087 DBG("Relay deleting session %" PRIu64, cmd->session->id);
1088
1089 rcu_read_lock();
1090 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
1091 node = lttng_ht_iter_get_node_ulong(&iter);
1092 if (!node) {
1093 continue;
1094 }
1095 stream = caa_container_of(node, struct relay_stream, stream_n);
1096 if (stream->session == cmd->session) {
1097 destroy_stream(stream);
1098 cmd->session->stream_count--;
1099 assert(cmd->session->stream_count >= 0);
1100 }
1101 }
1102
1103 /* Make this session not visible anymore. */
1104 iter.iter.node = &cmd->session->session_n.node;
1105 ret = lttng_ht_del(sessions_ht, &iter);
1106 assert(!ret);
1107 call_rcu(&cmd->session->rcu_node, deferred_free_session);
1108 rcu_read_unlock();
1109 }
1110
1111 /*
1112 * Copy index data from the control port to a given index object.
1113 */
1114 static void copy_index_control_data(struct relay_index *index,
1115 struct lttcomm_relayd_index *data)
1116 {
1117 assert(index);
1118 assert(data);
1119
1120 /*
1121 * The index on disk is encoded in big endian, so we don't need to convert
1122 * the data received on the network. The data_offset value is NEVER
1123 * modified here and is updated by the data thread.
1124 */
1125 index->index_data.packet_size = data->packet_size;
1126 index->index_data.content_size = data->content_size;
1127 index->index_data.timestamp_begin = data->timestamp_begin;
1128 index->index_data.timestamp_end = data->timestamp_end;
1129 index->index_data.events_discarded = data->events_discarded;
1130 index->index_data.stream_id = data->stream_id;
1131 }
1132
1133 /*
1134 * Handle the RELAYD_CREATE_SESSION command.
1135 *
1136 * On success, send back the session id or else return a negative value.
1137 */
1138 static
1139 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
1140 struct relay_command *cmd,
1141 struct lttng_ht *sessions_ht)
1142 {
1143 int ret = 0, send_ret;
1144 struct relay_session *session;
1145 struct lttcomm_relayd_status_session reply;
1146
1147 assert(recv_hdr);
1148 assert(cmd);
1149
1150 memset(&reply, 0, sizeof(reply));
1151
1152 session = zmalloc(sizeof(struct relay_session));
1153 if (session == NULL) {
1154 PERROR("relay session zmalloc");
1155 ret = -1;
1156 goto error;
1157 }
1158
1159 session->id = ++last_relay_session_id;
1160 session->sock = cmd->sock;
1161 session->minor = cmd->minor;
1162 session->major = cmd->major;
1163 cmd->session = session;
1164
1165 reply.session_id = htobe64(session->id);
1166
1167 switch (cmd->minor) {
1168 case 1:
1169 case 2:
1170 case 3:
1171 break;
1172 case 4: /* LTTng sessiond 2.4 */
1173 default:
1174 ret = cmd_create_session_2_4(cmd, session);
1175 break;
1176 }
1177
1178 lttng_ht_node_init_ulong(&session->session_n,
1179 (unsigned long) session->id);
1180 lttng_ht_add_unique_ulong(sessions_ht,
1181 &session->session_n);
1182
1183 DBG("Created session %" PRIu64, session->id);
1184
1185 error:
1186 if (ret < 0) {
1187 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1188 } else {
1189 reply.ret_code = htobe32(LTTNG_OK);
1190 }
1191
1192 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1193 if (send_ret < 0) {
1194 ERR("Relayd sending session id");
1195 ret = send_ret;
1196 }
1197
1198 return ret;
1199 }
1200
1201 /*
1202 * When we have received all the streams and the metadata for a channel,
1203 * we make them visible to the viewer threads.
1204 */
1205 static
1206 void set_viewer_ready_flag(struct relay_command *cmd)
1207 {
1208 struct relay_stream_recv_handle *node, *tmp_node;
1209
1210 cds_list_for_each_entry_safe(node, tmp_node, &cmd->recv_head, node) {
1211 struct relay_stream *stream;
1212
1213 rcu_read_lock();
1214 stream = relay_stream_find_by_id(node->id);
1215 if (!stream) {
1216 /*
1217 * Stream is most probably being cleaned up by the data thread thus
1218 * simply continue to the next one.
1219 */
1220 rcu_read_unlock();
1221 continue;
1222 }
1223
1224 stream->viewer_ready = 1;
1225 rcu_read_unlock();
1226
1227 /* Clean stream handle node. */
1228 cds_list_del(&node->node);
1229 free(node);
1230 }
1231
1232 return;
1233 }
1234
1235 /*
1236 * Add a recv handle node to the connection recv list with the given stream
1237 * handle. A new node is allocated thus must be freed when the node is deleted
1238 * from the list.
1239 */
1240 static void queue_stream_handle(uint64_t handle, struct relay_command *cmd)
1241 {
1242 struct relay_stream_recv_handle *node;
1243
1244 assert(cmd);
1245
1246 node = zmalloc(sizeof(*node));
1247 if (!node) {
1248 PERROR("zmalloc queue stream handle");
1249 return;
1250 }
1251
1252 node->id = handle;
1253 cds_list_add(&node->node, &cmd->recv_head);
1254 }
1255
1256 /*
1257 * relay_add_stream: allocate a new stream for a session
1258 */
1259 static
1260 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1261 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1262 {
1263 struct relay_session *session = cmd->session;
1264 struct relay_stream *stream = NULL;
1265 struct lttcomm_relayd_status_stream reply;
1266 int ret, send_ret;
1267
1268 if (!session || cmd->version_check_done == 0) {
1269 ERR("Trying to add a stream before version check");
1270 ret = -1;
1271 goto end_no_session;
1272 }
1273
1274 stream = zmalloc(sizeof(struct relay_stream));
1275 if (stream == NULL) {
1276 PERROR("relay stream zmalloc");
1277 ret = -1;
1278 goto end_no_session;
1279 }
1280
1281 switch (cmd->minor) {
1282 case 1: /* LTTng sessiond 2.1 */
1283 ret = cmd_recv_stream_2_1(cmd, stream);
1284 break;
1285 case 2: /* LTTng sessiond 2.2 */
1286 default:
1287 ret = cmd_recv_stream_2_2(cmd, stream);
1288 break;
1289 }
1290 if (ret < 0) {
1291 goto err_free_stream;
1292 }
1293
1294 rcu_read_lock();
1295 stream->stream_handle = ++last_relay_stream_id;
1296 stream->prev_seq = -1ULL;
1297 stream->session = session;
1298 stream->index_fd = -1;
1299 stream->read_index_fd = -1;
1300 stream->ctf_trace = NULL;
1301 pthread_mutex_init(&stream->lock, NULL);
1302
1303 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
1304 if (ret < 0) {
1305 ERR("relay creating output directory");
1306 goto end;
1307 }
1308
1309 /*
1310 * No need to use run_as API here because whatever we receives, the relayd
1311 * uses its own credentials for the stream files.
1312 */
1313 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1314 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1315 if (ret < 0) {
1316 ERR("Create output file");
1317 goto end;
1318 }
1319 stream->fd = ret;
1320 if (stream->tracefile_size) {
1321 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1322 } else {
1323 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1324 }
1325
1326 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1327 stream->metadata_flag = 1;
1328 /*
1329 * When we receive a new metadata stream, we create a new
1330 * ctf_trace and we assign this ctf_trace to all streams with
1331 * the same path.
1332 *
1333 * If later on we receive a new stream for the same ctf_trace,
1334 * we copy the information from the first hit in the HT to the
1335 * new stream.
1336 */
1337 stream->ctf_trace = ctf_trace_create();
1338 if (!stream->ctf_trace) {
1339 ret = -1;
1340 goto end;
1341 }
1342 stream->ctf_trace->refcount++;
1343 stream->ctf_trace->metadata_stream = stream;
1344 }
1345 ctf_trace_assign(cmd->ctf_traces_ht, stream);
1346 stream->ctf_traces_ht = cmd->ctf_traces_ht;
1347
1348 /*
1349 * Add the stream handle in the recv list of the connection. Once the end
1350 * stream message is received, this list is emptied and streams are set
1351 * with the viewer ready flag.
1352 */
1353 queue_stream_handle(stream->stream_handle, cmd);
1354
1355 lttng_ht_node_init_ulong(&stream->stream_n,
1356 (unsigned long) stream->stream_handle);
1357 lttng_ht_add_unique_ulong(relay_streams_ht,
1358 &stream->stream_n);
1359
1360 lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name);
1361 lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node);
1362 session->stream_count++;
1363
1364 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1365 stream->stream_handle);
1366
1367 end:
1368 reply.handle = htobe64(stream->stream_handle);
1369 /* send the session id to the client or a negative return code on error */
1370 if (ret < 0) {
1371 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1372 /* stream was not properly added to the ht, so free it */
1373 free(stream);
1374 } else {
1375 reply.ret_code = htobe32(LTTNG_OK);
1376 }
1377
1378 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1379 sizeof(struct lttcomm_relayd_status_stream), 0);
1380 if (send_ret < 0) {
1381 ERR("Relay sending stream id");
1382 ret = send_ret;
1383 }
1384 rcu_read_unlock();
1385
1386 end_no_session:
1387 return ret;
1388
1389 err_free_stream:
1390 free(stream->path_name);
1391 free(stream->channel_name);
1392 free(stream);
1393 return ret;
1394 }
1395
1396 /*
1397 * relay_close_stream: close a specific stream
1398 */
1399 static
1400 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1401 struct relay_command *cmd)
1402 {
1403 int ret, send_ret;
1404 struct relay_session *session = cmd->session;
1405 struct lttcomm_relayd_close_stream stream_info;
1406 struct lttcomm_relayd_generic_reply reply;
1407 struct relay_stream *stream;
1408
1409 DBG("Close stream received");
1410
1411 if (!session || cmd->version_check_done == 0) {
1412 ERR("Trying to close a stream before version check");
1413 ret = -1;
1414 goto end_no_session;
1415 }
1416
1417 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1418 sizeof(struct lttcomm_relayd_close_stream), 0);
1419 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1420 if (ret == 0) {
1421 /* Orderly shutdown. Not necessary to print an error. */
1422 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1423 } else {
1424 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1425 }
1426 ret = -1;
1427 goto end_no_session;
1428 }
1429
1430 rcu_read_lock();
1431 stream = relay_stream_find_by_id(be64toh(stream_info.stream_id));
1432 if (!stream) {
1433 ret = -1;
1434 goto end_unlock;
1435 }
1436
1437 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1438 stream->close_flag = 1;
1439 session->stream_count--;
1440 assert(session->stream_count >= 0);
1441
1442 if (close_stream_check(stream)) {
1443 destroy_stream(stream);
1444 }
1445
1446 end_unlock:
1447 rcu_read_unlock();
1448
1449 if (ret < 0) {
1450 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1451 } else {
1452 reply.ret_code = htobe32(LTTNG_OK);
1453 }
1454 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1455 sizeof(struct lttcomm_relayd_generic_reply), 0);
1456 if (send_ret < 0) {
1457 ERR("Relay sending stream id");
1458 ret = send_ret;
1459 }
1460
1461 end_no_session:
1462 return ret;
1463 }
1464
1465 /*
1466 * relay_unknown_command: send -1 if received unknown command
1467 */
1468 static
1469 void relay_unknown_command(struct relay_command *cmd)
1470 {
1471 struct lttcomm_relayd_generic_reply reply;
1472 int ret;
1473
1474 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1475 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1476 sizeof(struct lttcomm_relayd_generic_reply), 0);
1477 if (ret < 0) {
1478 ERR("Relay sending unknown command");
1479 }
1480 }
1481
1482 /*
1483 * relay_start: send an acknowledgment to the client to tell if we are
1484 * ready to receive data. We are ready if a session is established.
1485 */
1486 static
1487 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1488 struct relay_command *cmd)
1489 {
1490 int ret = htobe32(LTTNG_OK);
1491 struct lttcomm_relayd_generic_reply reply;
1492 struct relay_session *session = cmd->session;
1493
1494 if (!session) {
1495 DBG("Trying to start the streaming without a session established");
1496 ret = htobe32(LTTNG_ERR_UNK);
1497 }
1498
1499 reply.ret_code = ret;
1500 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1501 sizeof(struct lttcomm_relayd_generic_reply), 0);
1502 if (ret < 0) {
1503 ERR("Relay sending start ack");
1504 }
1505
1506 return ret;
1507 }
1508
1509 /*
1510 * Append padding to the file pointed by the file descriptor fd.
1511 */
1512 static int write_padding_to_file(int fd, uint32_t size)
1513 {
1514 ssize_t ret = 0;
1515 char *zeros;
1516
1517 if (size == 0) {
1518 goto end;
1519 }
1520
1521 zeros = zmalloc(size);
1522 if (zeros == NULL) {
1523 PERROR("zmalloc zeros for padding");
1524 ret = -1;
1525 goto end;
1526 }
1527
1528 ret = lttng_write(fd, zeros, size);
1529 if (ret < size) {
1530 PERROR("write padding to file");
1531 }
1532
1533 free(zeros);
1534
1535 end:
1536 return ret;
1537 }
1538
1539 /*
1540 * relay_recv_metadata: receive the metada for the session.
1541 */
1542 static
1543 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1544 struct relay_command *cmd)
1545 {
1546 int ret = htobe32(LTTNG_OK);
1547 ssize_t size_ret;
1548 struct relay_session *session = cmd->session;
1549 struct lttcomm_relayd_metadata_payload *metadata_struct;
1550 struct relay_stream *metadata_stream;
1551 uint64_t data_size, payload_size;
1552
1553 if (!session) {
1554 ERR("Metadata sent before version check");
1555 ret = -1;
1556 goto end;
1557 }
1558
1559 data_size = payload_size = be64toh(recv_hdr->data_size);
1560 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1561 ERR("Incorrect data size");
1562 ret = -1;
1563 goto end;
1564 }
1565 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1566
1567 if (data_buffer_size < data_size) {
1568 /* In case the realloc fails, we can free the memory */
1569 char *tmp_data_ptr;
1570
1571 tmp_data_ptr = realloc(data_buffer, data_size);
1572 if (!tmp_data_ptr) {
1573 ERR("Allocating data buffer");
1574 free(data_buffer);
1575 ret = -1;
1576 goto end;
1577 }
1578 data_buffer = tmp_data_ptr;
1579 data_buffer_size = data_size;
1580 }
1581 memset(data_buffer, 0, data_size);
1582 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1583 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1584 if (ret < 0 || ret != data_size) {
1585 if (ret == 0) {
1586 /* Orderly shutdown. Not necessary to print an error. */
1587 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1588 } else {
1589 ERR("Relay didn't receive the whole metadata");
1590 }
1591 ret = -1;
1592 goto end;
1593 }
1594 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1595
1596 rcu_read_lock();
1597 metadata_stream = relay_stream_find_by_id(
1598 be64toh(metadata_struct->stream_id));
1599 if (!metadata_stream) {
1600 ret = -1;
1601 goto end_unlock;
1602 }
1603
1604 size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
1605 payload_size);
1606 if (size_ret < payload_size) {
1607 ERR("Relay error writing metadata on file");
1608 ret = -1;
1609 goto end_unlock;
1610 }
1611
1612 ret = write_padding_to_file(metadata_stream->fd,
1613 be32toh(metadata_struct->padding_size));
1614 if (ret < 0) {
1615 goto end_unlock;
1616 }
1617 metadata_stream->ctf_trace->metadata_received +=
1618 payload_size + be32toh(metadata_struct->padding_size);
1619
1620 DBG2("Relay metadata written");
1621
1622 end_unlock:
1623 rcu_read_unlock();
1624 end:
1625 return ret;
1626 }
1627
1628 /*
1629 * relay_send_version: send relayd version number
1630 */
1631 static
1632 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1633 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1634 {
1635 int ret;
1636 struct lttcomm_relayd_version reply, msg;
1637
1638 assert(cmd);
1639
1640 cmd->version_check_done = 1;
1641
1642 /* Get version from the other side. */
1643 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1644 if (ret < 0 || ret != sizeof(msg)) {
1645 if (ret == 0) {
1646 /* Orderly shutdown. Not necessary to print an error. */
1647 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1648 } else {
1649 ERR("Relay failed to receive the version values.");
1650 }
1651 ret = -1;
1652 goto end;
1653 }
1654
1655 reply.major = RELAYD_VERSION_COMM_MAJOR;
1656 reply.minor = RELAYD_VERSION_COMM_MINOR;
1657
1658 /* Major versions must be the same */
1659 if (reply.major != be32toh(msg.major)) {
1660 DBG("Incompatible major versions (%u vs %u), deleting session",
1661 reply.major, be32toh(msg.major));
1662 relay_delete_session(cmd, sessions_ht);
1663 ret = 0;
1664 goto end;
1665 }
1666
1667 cmd->major = reply.major;
1668 /* We adapt to the lowest compatible version */
1669 if (reply.minor <= be32toh(msg.minor)) {
1670 cmd->minor = reply.minor;
1671 } else {
1672 cmd->minor = be32toh(msg.minor);
1673 }
1674
1675 reply.major = htobe32(reply.major);
1676 reply.minor = htobe32(reply.minor);
1677 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1678 sizeof(struct lttcomm_relayd_version), 0);
1679 if (ret < 0) {
1680 ERR("Relay sending version");
1681 }
1682
1683 DBG("Version check done using protocol %u.%u", cmd->major,
1684 cmd->minor);
1685
1686 end:
1687 return ret;
1688 }
1689
1690 /*
1691 * Check for data pending for a given stream id from the session daemon.
1692 */
1693 static
1694 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1695 struct relay_command *cmd)
1696 {
1697 struct relay_session *session = cmd->session;
1698 struct lttcomm_relayd_data_pending msg;
1699 struct lttcomm_relayd_generic_reply reply;
1700 struct relay_stream *stream;
1701 int ret;
1702 uint64_t last_net_seq_num, stream_id;
1703
1704 DBG("Data pending command received");
1705
1706 if (!session || cmd->version_check_done == 0) {
1707 ERR("Trying to check for data before version check");
1708 ret = -1;
1709 goto end_no_session;
1710 }
1711
1712 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1713 if (ret < sizeof(msg)) {
1714 if (ret == 0) {
1715 /* Orderly shutdown. Not necessary to print an error. */
1716 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1717 } else {
1718 ERR("Relay didn't receive valid data_pending struct size : %d",
1719 ret);
1720 }
1721 ret = -1;
1722 goto end_no_session;
1723 }
1724
1725 stream_id = be64toh(msg.stream_id);
1726 last_net_seq_num = be64toh(msg.last_net_seq_num);
1727
1728 rcu_read_lock();
1729 stream = relay_stream_find_by_id(stream_id);
1730 if (stream == NULL) {
1731 ret = -1;
1732 goto end_unlock;
1733 }
1734
1735 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1736 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1737 last_net_seq_num);
1738
1739 /* Avoid wrapping issue */
1740 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1741 /* Data has in fact been written and is NOT pending */
1742 ret = 0;
1743 } else {
1744 /* Data still being streamed thus pending */
1745 ret = 1;
1746 }
1747
1748 /* Pending check is now done. */
1749 stream->data_pending_check_done = 1;
1750
1751 end_unlock:
1752 rcu_read_unlock();
1753
1754 reply.ret_code = htobe32(ret);
1755 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1756 if (ret < 0) {
1757 ERR("Relay data pending ret code failed");
1758 }
1759
1760 end_no_session:
1761 return ret;
1762 }
1763
1764 /*
1765 * Wait for the control socket to reach a quiescent state.
1766 *
1767 * Note that for now, when receiving this command from the session daemon, this
1768 * means that every subsequent commands or data received on the control socket
1769 * has been handled. So, this is why we simply return OK here.
1770 */
1771 static
1772 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1773 struct relay_command *cmd)
1774 {
1775 int ret;
1776 uint64_t stream_id;
1777 struct relay_stream *stream;
1778 struct lttng_ht_iter iter;
1779 struct lttcomm_relayd_quiescent_control msg;
1780 struct lttcomm_relayd_generic_reply reply;
1781
1782 DBG("Checking quiescent state on control socket");
1783
1784 if (!cmd->session || cmd->version_check_done == 0) {
1785 ERR("Trying to check for data before version check");
1786 ret = -1;
1787 goto end_no_session;
1788 }
1789
1790 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1791 if (ret < sizeof(msg)) {
1792 if (ret == 0) {
1793 /* Orderly shutdown. Not necessary to print an error. */
1794 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1795 } else {
1796 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1797 ret);
1798 }
1799 ret = -1;
1800 goto end_no_session;
1801 }
1802
1803 stream_id = be64toh(msg.stream_id);
1804
1805 rcu_read_lock();
1806 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1807 stream_n.node) {
1808 if (stream->stream_handle == stream_id) {
1809 stream->data_pending_check_done = 1;
1810 DBG("Relay quiescent control pending flag set to %" PRIu64,
1811 stream_id);
1812 break;
1813 }
1814 }
1815 rcu_read_unlock();
1816
1817 reply.ret_code = htobe32(LTTNG_OK);
1818 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1819 if (ret < 0) {
1820 ERR("Relay data quiescent control ret code failed");
1821 }
1822
1823 end_no_session:
1824 return ret;
1825 }
1826
1827 /*
1828 * Initialize a data pending command. This means that a client is about to ask
1829 * for data pending for each stream he/she holds. Simply iterate over all
1830 * streams of a session and set the data_pending_check_done flag.
1831 *
1832 * This command returns to the client a LTTNG_OK code.
1833 */
1834 static
1835 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1836 struct relay_command *cmd)
1837 {
1838 int ret;
1839 struct lttng_ht_iter iter;
1840 struct lttcomm_relayd_begin_data_pending msg;
1841 struct lttcomm_relayd_generic_reply reply;
1842 struct relay_stream *stream;
1843 uint64_t session_id;
1844
1845 assert(recv_hdr);
1846 assert(cmd);
1847
1848 DBG("Init streams for data pending");
1849
1850 if (!cmd->session || cmd->version_check_done == 0) {
1851 ERR("Trying to check for data before version check");
1852 ret = -1;
1853 goto end_no_session;
1854 }
1855
1856 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1857 if (ret < sizeof(msg)) {
1858 if (ret == 0) {
1859 /* Orderly shutdown. Not necessary to print an error. */
1860 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1861 } else {
1862 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1863 ret);
1864 }
1865 ret = -1;
1866 goto end_no_session;
1867 }
1868
1869 session_id = be64toh(msg.session_id);
1870
1871 /*
1872 * Iterate over all streams to set the begin data pending flag. For now, the
1873 * streams are indexed by stream handle so we have to iterate over all
1874 * streams to find the one associated with the right session_id.
1875 */
1876 rcu_read_lock();
1877 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1878 stream_n.node) {
1879 if (stream->session->id == session_id) {
1880 stream->data_pending_check_done = 0;
1881 DBG("Set begin data pending flag to stream %" PRIu64,
1882 stream->stream_handle);
1883 }
1884 }
1885 rcu_read_unlock();
1886
1887 /* All good, send back reply. */
1888 reply.ret_code = htobe32(LTTNG_OK);
1889
1890 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1891 if (ret < 0) {
1892 ERR("Relay begin data pending send reply failed");
1893 }
1894
1895 end_no_session:
1896 return ret;
1897 }
1898
1899 /*
1900 * End data pending command. This will check, for a given session id, if each
1901 * stream associated with it has its data_pending_check_done flag set. If not,
1902 * this means that the client lost track of the stream but the data is still
1903 * being streamed on our side. In this case, we inform the client that data is
1904 * inflight.
1905 *
1906 * Return to the client if there is data in flight or not with a ret_code.
1907 */
1908 static
1909 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1910 struct relay_command *cmd)
1911 {
1912 int ret;
1913 struct lttng_ht_iter iter;
1914 struct lttcomm_relayd_end_data_pending msg;
1915 struct lttcomm_relayd_generic_reply reply;
1916 struct relay_stream *stream;
1917 uint64_t session_id;
1918 uint32_t is_data_inflight = 0;
1919
1920 assert(recv_hdr);
1921 assert(cmd);
1922
1923 DBG("End data pending command");
1924
1925 if (!cmd->session || cmd->version_check_done == 0) {
1926 ERR("Trying to check for data before version check");
1927 ret = -1;
1928 goto end_no_session;
1929 }
1930
1931 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1932 if (ret < sizeof(msg)) {
1933 if (ret == 0) {
1934 /* Orderly shutdown. Not necessary to print an error. */
1935 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1936 } else {
1937 ERR("Relay didn't receive valid end data_pending struct size: %d",
1938 ret);
1939 }
1940 ret = -1;
1941 goto end_no_session;
1942 }
1943
1944 session_id = be64toh(msg.session_id);
1945
1946 /* Iterate over all streams to see if the begin data pending flag is set. */
1947 rcu_read_lock();
1948 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1949 stream_n.node) {
1950 if (stream->session->id == session_id &&
1951 !stream->data_pending_check_done) {
1952 is_data_inflight = 1;
1953 DBG("Data is still in flight for stream %" PRIu64,
1954 stream->stream_handle);
1955 break;
1956 }
1957 }
1958 rcu_read_unlock();
1959
1960 /* All good, send back reply. */
1961 reply.ret_code = htobe32(is_data_inflight);
1962
1963 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1964 if (ret < 0) {
1965 ERR("Relay end data pending send reply failed");
1966 }
1967
1968 end_no_session:
1969 return ret;
1970 }
1971
1972 /*
1973 * Receive an index for a specific stream.
1974 *
1975 * Return 0 on success else a negative value.
1976 */
1977 static
1978 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1979 struct relay_command *cmd)
1980 {
1981 int ret, send_ret, index_created = 0;
1982 struct relay_session *session = cmd->session;
1983 struct lttcomm_relayd_index index_info;
1984 struct relay_index *index, *wr_index = NULL;
1985 struct lttcomm_relayd_generic_reply reply;
1986 struct relay_stream *stream;
1987 uint64_t net_seq_num;
1988
1989 assert(cmd);
1990
1991 DBG("Relay receiving index");
1992
1993 if (!session || cmd->version_check_done == 0) {
1994 ERR("Trying to close a stream before version check");
1995 ret = -1;
1996 goto end_no_session;
1997 }
1998
1999 ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info,
2000 sizeof(index_info), 0);
2001 if (ret < sizeof(index_info)) {
2002 if (ret == 0) {
2003 /* Orderly shutdown. Not necessary to print an error. */
2004 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2005 } else {
2006 ERR("Relay didn't receive valid index struct size : %d", ret);
2007 }
2008 ret = -1;
2009 goto end_no_session;
2010 }
2011
2012 net_seq_num = be64toh(index_info.net_seq_num);
2013
2014 rcu_read_lock();
2015 stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id));
2016 if (!stream) {
2017 ret = -1;
2018 goto end_rcu_unlock;
2019 }
2020
2021 /* Live beacon handling */
2022 if (index_info.packet_size == 0) {
2023 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
2024
2025 /*
2026 * Only flag a stream inactive when it has already received data.
2027 */
2028 if (stream->total_index_received > 0) {
2029 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
2030 }
2031 ret = 0;
2032 goto end_rcu_unlock;
2033 } else {
2034 stream->beacon_ts_end = -1ULL;
2035 }
2036
2037 index = relay_index_find(stream->stream_handle, net_seq_num);
2038 if (!index) {
2039 /* A successful creation will add the object to the HT. */
2040 index = relay_index_create(stream->stream_handle, net_seq_num);
2041 if (!index) {
2042 goto end_rcu_unlock;
2043 }
2044 index_created = 1;
2045 }
2046
2047 copy_index_control_data(index, &index_info);
2048
2049 if (index_created) {
2050 /*
2051 * Try to add the relay index object to the hash table. If an object
2052 * already exist, destroy back the index created, set the data in this
2053 * object and write it on disk.
2054 */
2055 relay_index_add(index, &wr_index);
2056 if (wr_index) {
2057 copy_index_control_data(wr_index, &index_info);
2058 free(index);
2059 }
2060 } else {
2061 /* The index already exists so write it on disk. */
2062 wr_index = index;
2063 }
2064
2065 /* Do we have a writable ready index to write on disk. */
2066 if (wr_index) {
2067 /* Starting at 2.4, create the index file if none available. */
2068 if (cmd->minor >= 4 && stream->index_fd < 0) {
2069 ret = index_create_file(stream->path_name, stream->channel_name,
2070 relayd_uid, relayd_gid, stream->tracefile_size,
2071 stream->tracefile_count_current);
2072 if (ret < 0) {
2073 goto end_rcu_unlock;
2074 }
2075 stream->index_fd = ret;
2076 }
2077
2078 ret = relay_index_write(wr_index->fd, wr_index);
2079 if (ret < 0) {
2080 goto end_rcu_unlock;
2081 }
2082 stream->total_index_received++;
2083 }
2084
2085 end_rcu_unlock:
2086 rcu_read_unlock();
2087
2088 if (ret < 0) {
2089 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2090 } else {
2091 reply.ret_code = htobe32(LTTNG_OK);
2092 }
2093 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
2094 if (send_ret < 0) {
2095 ERR("Relay sending close index id reply");
2096 ret = send_ret;
2097 }
2098
2099 end_no_session:
2100 return ret;
2101 }
2102
2103 /*
2104 * Receive the streams_sent message.
2105 *
2106 * Return 0 on success else a negative value.
2107 */
2108 static
2109 int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
2110 struct relay_command *cmd)
2111 {
2112 int ret, send_ret;
2113 struct lttcomm_relayd_generic_reply reply;
2114
2115 assert(cmd);
2116
2117 DBG("Relay receiving streams_sent");
2118
2119 if (!cmd->session || cmd->version_check_done == 0) {
2120 ERR("Trying to close a stream before version check");
2121 ret = -1;
2122 goto end_no_session;
2123 }
2124
2125 /*
2126 * Flag every pending stream in the connection recv list that they are
2127 * ready to be used by the viewer.
2128 */
2129 set_viewer_ready_flag(cmd);
2130
2131 reply.ret_code = htobe32(LTTNG_OK);
2132 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
2133 if (send_ret < 0) {
2134 ERR("Relay sending sent_stream reply");
2135 ret = send_ret;
2136 } else {
2137 /* Success. */
2138 ret = 0;
2139 }
2140
2141 end_no_session:
2142 return ret;
2143 }
2144
2145 /*
2146 * Process the commands received on the control socket
2147 */
2148 static
2149 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
2150 struct relay_command *cmd, struct relay_local_data *ctx)
2151 {
2152 int ret = 0;
2153
2154 switch (be32toh(recv_hdr->cmd)) {
2155 case RELAYD_CREATE_SESSION:
2156 ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht);
2157 break;
2158 case RELAYD_ADD_STREAM:
2159 ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht);
2160 break;
2161 case RELAYD_START_DATA:
2162 ret = relay_start(recv_hdr, cmd);
2163 break;
2164 case RELAYD_SEND_METADATA:
2165 ret = relay_recv_metadata(recv_hdr, cmd);
2166 break;
2167 case RELAYD_VERSION:
2168 ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht);
2169 break;
2170 case RELAYD_CLOSE_STREAM:
2171 ret = relay_close_stream(recv_hdr, cmd);
2172 break;
2173 case RELAYD_DATA_PENDING:
2174 ret = relay_data_pending(recv_hdr, cmd);
2175 break;
2176 case RELAYD_QUIESCENT_CONTROL:
2177 ret = relay_quiescent_control(recv_hdr, cmd);
2178 break;
2179 case RELAYD_BEGIN_DATA_PENDING:
2180 ret = relay_begin_data_pending(recv_hdr, cmd);
2181 break;
2182 case RELAYD_END_DATA_PENDING:
2183 ret = relay_end_data_pending(recv_hdr, cmd);
2184 break;
2185 case RELAYD_SEND_INDEX:
2186 ret = relay_recv_index(recv_hdr, cmd);
2187 break;
2188 case RELAYD_STREAMS_SENT:
2189 ret = relay_streams_sent(recv_hdr, cmd);
2190 break;
2191 case RELAYD_UPDATE_SYNC_INFO:
2192 default:
2193 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2194 relay_unknown_command(cmd);
2195 ret = -1;
2196 goto end;
2197 }
2198
2199 end:
2200 return ret;
2201 }
2202
2203 /*
2204 * Handle index for a data stream.
2205 *
2206 * RCU read side lock MUST be acquired.
2207 *
2208 * Return 0 on success else a negative value.
2209 */
2210 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2211 int rotate_index)
2212 {
2213 int ret = 0, index_created = 0;
2214 uint64_t stream_id, data_offset;
2215 struct relay_index *index, *wr_index = NULL;
2216
2217 assert(stream);
2218
2219 stream_id = stream->stream_handle;
2220 /* Get data offset because we are about to update the index. */
2221 data_offset = htobe64(stream->tracefile_size_current);
2222
2223 /*
2224 * Lookup for an existing index for that stream id/sequence number. If on
2225 * exists, the control thread already received the data for it thus we need
2226 * to write it on disk.
2227 */
2228 index = relay_index_find(stream_id, net_seq_num);
2229 if (!index) {
2230 /* A successful creation will add the object to the HT. */
2231 index = relay_index_create(stream_id, net_seq_num);
2232 if (!index) {
2233 ret = -1;
2234 goto error;
2235 }
2236 index_created = 1;
2237 }
2238
2239 if (rotate_index || stream->index_fd < 0) {
2240 index->to_close_fd = stream->index_fd;
2241 ret = index_create_file(stream->path_name, stream->channel_name,
2242 relayd_uid, relayd_gid, stream->tracefile_size,
2243 stream->tracefile_count_current);
2244 if (ret < 0) {
2245 /* This will close the stream's index fd if one. */
2246 relay_index_free_safe(index);
2247 goto error;
2248 }
2249 stream->index_fd = ret;
2250 }
2251 index->fd = stream->index_fd;
2252 index->index_data.offset = data_offset;
2253
2254 if (index_created) {
2255 /*
2256 * Try to add the relay index object to the hash table. If an object
2257 * already exist, destroy back the index created and set the data.
2258 */
2259 relay_index_add(index, &wr_index);
2260 if (wr_index) {
2261 /* Copy back data from the created index. */
2262 wr_index->fd = index->fd;
2263 wr_index->to_close_fd = index->to_close_fd;
2264 wr_index->index_data.offset = data_offset;
2265 free(index);
2266 }
2267 } else {
2268 /* The index already exists so write it on disk. */
2269 wr_index = index;
2270 }
2271
2272 /* Do we have a writable ready index to write on disk. */
2273 if (wr_index) {
2274 ret = relay_index_write(wr_index->fd, wr_index);
2275 if (ret < 0) {
2276 goto error;
2277 }
2278 stream->total_index_received++;
2279 }
2280
2281 error:
2282 return ret;
2283 }
2284
2285 /*
2286 * relay_process_data: Process the data received on the data socket
2287 */
2288 static
2289 int relay_process_data(struct relay_command *cmd)
2290 {
2291 int ret = 0, rotate_index = 0;
2292 ssize_t size_ret;
2293 struct relay_stream *stream;
2294 struct lttcomm_relayd_data_hdr data_hdr;
2295 uint64_t stream_id;
2296 uint64_t net_seq_num;
2297 uint32_t data_size;
2298
2299 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
2300 sizeof(struct lttcomm_relayd_data_hdr), 0);
2301 if (ret <= 0) {
2302 if (ret == 0) {
2303 /* Orderly shutdown. Not necessary to print an error. */
2304 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2305 } else {
2306 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
2307 }
2308 ret = -1;
2309 goto end;
2310 }
2311
2312 stream_id = be64toh(data_hdr.stream_id);
2313
2314 rcu_read_lock();
2315 stream = relay_stream_find_by_id(stream_id);
2316 if (!stream) {
2317 ret = -1;
2318 goto end_rcu_unlock;
2319 }
2320
2321 data_size = be32toh(data_hdr.data_size);
2322 if (data_buffer_size < data_size) {
2323 char *tmp_data_ptr;
2324
2325 tmp_data_ptr = realloc(data_buffer, data_size);
2326 if (!tmp_data_ptr) {
2327 ERR("Allocating data buffer");
2328 free(data_buffer);
2329 ret = -1;
2330 goto end_rcu_unlock;
2331 }
2332 data_buffer = tmp_data_ptr;
2333 data_buffer_size = data_size;
2334 }
2335 memset(data_buffer, 0, data_size);
2336
2337 net_seq_num = be64toh(data_hdr.net_seq_num);
2338
2339 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2340 data_size, stream_id, net_seq_num);
2341 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
2342 if (ret <= 0) {
2343 if (ret == 0) {
2344 /* Orderly shutdown. Not necessary to print an error. */
2345 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2346 }
2347 ret = -1;
2348 goto end_rcu_unlock;
2349 }
2350
2351 /* Check if a rotation is needed. */
2352 if (stream->tracefile_size > 0 &&
2353 (stream->tracefile_size_current + data_size) >
2354 stream->tracefile_size) {
2355 struct relay_viewer_stream *vstream;
2356 uint64_t new_id;
2357
2358 new_id = (stream->tracefile_count_current + 1) %
2359 stream->tracefile_count;
2360 /*
2361 * When we wrap-around back to 0, we start overwriting old
2362 * trace data.
2363 */
2364 if (!stream->tracefile_overwrite && new_id == 0) {
2365 stream->tracefile_overwrite = 1;
2366 }
2367 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
2368 if (stream->tracefile_overwrite) {
2369 stream->oldest_tracefile_id =
2370 (stream->oldest_tracefile_id + 1) %
2371 stream->tracefile_count;
2372 }
2373 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
2374 if (vstream) {
2375 /*
2376 * The viewer is reading a file about to be
2377 * overwritten. Close the FDs it is
2378 * currently using and let it handle the fault.
2379 */
2380 if (vstream->tracefile_count_current == new_id) {
2381 pthread_mutex_lock(&vstream->overwrite_lock);
2382 vstream->abort_flag = 1;
2383 pthread_mutex_unlock(&vstream->overwrite_lock);
2384 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2385 stream->channel_name, new_id);
2386 } else if (vstream->tracefile_count_current ==
2387 stream->tracefile_count_current) {
2388 /*
2389 * The reader and writer were in the
2390 * same trace file, inform the viewer
2391 * that no new index will ever be added
2392 * to this file.
2393 */
2394 vstream->close_write_flag = 1;
2395 }
2396 }
2397 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2398 stream->tracefile_size, stream->tracefile_count,
2399 relayd_uid, relayd_gid, stream->fd,
2400 &(stream->tracefile_count_current), &stream->fd);
2401 stream->total_index_received = 0;
2402 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
2403 if (ret < 0) {
2404 ERR("Rotating stream output file");
2405 goto end_rcu_unlock;
2406 }
2407 /* Reset current size because we just perform a stream rotation. */
2408 stream->tracefile_size_current = 0;
2409 rotate_index = 1;
2410 }
2411
2412 /*
2413 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2414 * index are NOT supported.
2415 */
2416 if (stream->session->minor >= 4 && !stream->session->snapshot) {
2417 ret = handle_index_data(stream, net_seq_num, rotate_index);
2418 if (ret < 0) {
2419 goto end_rcu_unlock;
2420 }
2421 }
2422
2423 /* Write data to stream output fd. */
2424 size_ret = lttng_write(stream->fd, data_buffer, data_size);
2425 if (size_ret < data_size) {
2426 ERR("Relay error writing data to file");
2427 ret = -1;
2428 goto end_rcu_unlock;
2429 }
2430
2431 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2432 ret, stream->stream_handle);
2433
2434 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2435 if (ret < 0) {
2436 goto end_rcu_unlock;
2437 }
2438 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2439
2440 stream->prev_seq = net_seq_num;
2441
2442 /* Check if we need to close the FD */
2443 if (close_stream_check(stream)) {
2444 destroy_stream(stream);
2445 }
2446
2447 end_rcu_unlock:
2448 rcu_read_unlock();
2449 end:
2450 return ret;
2451 }
2452
2453 static
2454 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
2455 {
2456 int ret;
2457
2458 lttng_poll_del(events, pollfd);
2459
2460 ret = close(pollfd);
2461 if (ret < 0) {
2462 ERR("Closing pollfd %d", pollfd);
2463 }
2464 }
2465
2466 static
2467 int relay_add_connection(int fd, struct lttng_poll_event *events,
2468 struct lttng_ht *relay_connections_ht)
2469 {
2470 struct relay_command *relay_connection;
2471 ssize_t ret;
2472
2473 relay_connection = zmalloc(sizeof(struct relay_command));
2474 if (relay_connection == NULL) {
2475 PERROR("Relay command zmalloc");
2476 goto error;
2477 }
2478 ret = lttng_read(fd, relay_connection, sizeof(struct relay_command));
2479 if (ret < sizeof(struct relay_command)) {
2480 PERROR("read relay cmd pipe");
2481 goto error_read;
2482 }
2483 CDS_INIT_LIST_HEAD(&relay_connection->recv_head);
2484
2485 /*
2486 * Only used by the control side and the reference is copied inside each
2487 * stream from that connection. Thus a destroy HT must be done after every
2488 * stream has been destroyed.
2489 */
2490 if (relay_connection->type == RELAY_CONTROL) {
2491 relay_connection->ctf_traces_ht = lttng_ht_new(0,
2492 LTTNG_HT_TYPE_STRING);
2493 if (!relay_connection->ctf_traces_ht) {
2494 goto error_read;
2495 }
2496 }
2497
2498 lttng_ht_node_init_ulong(&relay_connection->sock_n,
2499 (unsigned long) relay_connection->sock->fd);
2500 rcu_read_lock();
2501 lttng_ht_add_unique_ulong(relay_connections_ht,
2502 &relay_connection->sock_n);
2503 rcu_read_unlock();
2504 return lttng_poll_add(events,
2505 relay_connection->sock->fd,
2506 LPOLLIN | LPOLLRDHUP);
2507
2508 error_read:
2509 free(relay_connection);
2510 error:
2511 return -1;
2512 }
2513
2514 static
2515 void deferred_free_connection(struct rcu_head *head)
2516 {
2517 struct relay_command *relay_connection =
2518 caa_container_of(head, struct relay_command, rcu_node);
2519
2520 lttcomm_destroy_sock(relay_connection->sock);
2521 free(relay_connection);
2522 }
2523
2524 static
2525 void relay_del_connection(struct lttng_ht *relay_connections_ht,
2526 struct lttng_ht_iter *iter, struct relay_command *relay_connection,
2527 struct lttng_ht *sessions_ht)
2528 {
2529 int ret;
2530
2531 ret = lttng_ht_del(relay_connections_ht, iter);
2532 assert(!ret);
2533
2534 if (relay_connection->type == RELAY_CONTROL) {
2535 struct relay_stream_recv_handle *node, *tmp_node;
2536
2537 relay_delete_session(relay_connection, sessions_ht);
2538 lttng_ht_destroy(relay_connection->ctf_traces_ht);
2539
2540 /* Clean up recv list. */
2541 cds_list_for_each_entry_safe(node, tmp_node,
2542 &relay_connection->recv_head, node) {
2543 cds_list_del(&node->node);
2544 free(node);
2545 }
2546 }
2547
2548 call_rcu(&relay_connection->rcu_node, deferred_free_connection);
2549 }
2550
2551 /*
2552 * This thread does the actual work
2553 */
2554 static
2555 void *relay_thread_worker(void *data)
2556 {
2557 int ret, err = -1, last_seen_data_fd = -1;
2558 uint32_t nb_fd;
2559 struct relay_command *relay_connection;
2560 struct lttng_poll_event events;
2561 struct lttng_ht *relay_connections_ht;
2562 struct lttng_ht_node_ulong *node;
2563 struct lttng_ht_iter iter;
2564 struct lttcomm_relayd_hdr recv_hdr;
2565 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2566 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2567
2568 DBG("[thread] Relay worker started");
2569
2570 rcu_register_thread();
2571
2572 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2573
2574 if (testpoint(relayd_thread_worker)) {
2575 goto error_testpoint;
2576 }
2577
2578 health_code_update();
2579
2580 /* table of connections indexed on socket */
2581 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2582 if (!relay_connections_ht) {
2583 goto relay_connections_ht_error;
2584 }
2585
2586 /* Tables of received indexes indexed by index handle and net_seq_num. */
2587 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2588 if (!indexes_ht) {
2589 goto indexes_ht_error;
2590 }
2591
2592 ret = create_thread_poll_set(&events, 2);
2593 if (ret < 0) {
2594 goto error_poll_create;
2595 }
2596
2597 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
2598 if (ret < 0) {
2599 goto error;
2600 }
2601
2602 restart:
2603 while (1) {
2604 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2605
2606 health_code_update();
2607
2608 /* Infinite blocking call, waiting for transmission */
2609 DBG3("Relayd worker thread polling...");
2610 health_poll_entry();
2611 ret = lttng_poll_wait(&events, -1);
2612 health_poll_exit();
2613 if (ret < 0) {
2614 /*
2615 * Restart interrupted system call.
2616 */
2617 if (errno == EINTR) {
2618 goto restart;
2619 }
2620 goto error;
2621 }
2622
2623 nb_fd = ret;
2624
2625 /*
2626 * Process control. The control connection is prioritised so we don't
2627 * starve it with high throughout put tracing data on the data
2628 * connection.
2629 */
2630 for (i = 0; i < nb_fd; i++) {
2631 /* Fetch once the poll data */
2632 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2633 int pollfd = LTTNG_POLL_GETFD(&events, i);
2634
2635 health_code_update();
2636
2637 /* Thread quit pipe has been closed. Killing thread. */
2638 ret = check_thread_quit_pipe(pollfd, revents);
2639 if (ret) {
2640 err = 0;
2641 goto exit;
2642 }
2643
2644 /* Inspect the relay cmd pipe for new connection */
2645 if (pollfd == relay_cmd_pipe[0]) {
2646 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2647 ERR("Relay pipe error");
2648 goto error;
2649 } else if (revents & LPOLLIN) {
2650 DBG("Relay command received");
2651 ret = relay_add_connection(relay_cmd_pipe[0],
2652 &events, relay_connections_ht);
2653 if (ret < 0) {
2654 goto error;
2655 }
2656 }
2657 } else if (revents) {
2658 rcu_read_lock();
2659 lttng_ht_lookup(relay_connections_ht,
2660 (void *)((unsigned long) pollfd),
2661 &iter);
2662 node = lttng_ht_iter_get_node_ulong(&iter);
2663 if (node == NULL) {
2664 DBG2("Relay sock %d not found", pollfd);
2665 rcu_read_unlock();
2666 goto error;
2667 }
2668 relay_connection = caa_container_of(node,
2669 struct relay_command, sock_n);
2670
2671 if (revents & (LPOLLERR)) {
2672 ERR("POLL ERROR");
2673 relay_cleanup_poll_connection(&events, pollfd);
2674 relay_del_connection(relay_connections_ht,
2675 &iter, relay_connection, sessions_ht);
2676 if (last_seen_data_fd == pollfd) {
2677 last_seen_data_fd = last_notdel_data_fd;
2678 }
2679 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2680 DBG("Socket %d hung up", pollfd);
2681 relay_cleanup_poll_connection(&events, pollfd);
2682 relay_del_connection(relay_connections_ht,
2683 &iter, relay_connection, sessions_ht);
2684 if (last_seen_data_fd == pollfd) {
2685 last_seen_data_fd = last_notdel_data_fd;
2686 }
2687 } else if (revents & LPOLLIN) {
2688 /* control socket */
2689 if (relay_connection->type == RELAY_CONTROL) {
2690 ret = relay_connection->sock->ops->recvmsg(
2691 relay_connection->sock, &recv_hdr,
2692 sizeof(struct lttcomm_relayd_hdr), 0);
2693 /* connection closed */
2694 if (ret <= 0) {
2695 relay_cleanup_poll_connection(&events, pollfd);
2696 relay_del_connection(relay_connections_ht,
2697 &iter, relay_connection, sessions_ht);
2698 DBG("Control connection closed with %d", pollfd);
2699 } else {
2700 if (relay_connection->session) {
2701 DBG2("Relay worker receiving data for session : %" PRIu64,
2702 relay_connection->session->id);
2703 }
2704 ret = relay_process_control(&recv_hdr,
2705 relay_connection, relay_ctx);
2706 if (ret < 0) {
2707 /* Clear the session on error. */
2708 relay_cleanup_poll_connection(&events, pollfd);
2709 relay_del_connection(relay_connections_ht,
2710 &iter, relay_connection, sessions_ht);
2711 DBG("Connection closed with %d", pollfd);
2712 }
2713 seen_control = 1;
2714 }
2715 } else {
2716 /*
2717 * Flag the last seen data fd not deleted. It will be
2718 * used as the last seen fd if any fd gets deleted in
2719 * this first loop.
2720 */
2721 last_notdel_data_fd = pollfd;
2722 }
2723 }
2724 rcu_read_unlock();
2725 }
2726 }
2727
2728 /*
2729 * The last loop handled a control request, go back to poll to make
2730 * sure we prioritise the control socket.
2731 */
2732 if (seen_control) {
2733 continue;
2734 }
2735
2736 if (last_seen_data_fd >= 0) {
2737 for (i = 0; i < nb_fd; i++) {
2738 int pollfd = LTTNG_POLL_GETFD(&events, i);
2739
2740 health_code_update();
2741
2742 if (last_seen_data_fd == pollfd) {
2743 idx = i;
2744 break;
2745 }
2746 }
2747 }
2748
2749 /* Process data connection. */
2750 for (i = idx + 1; i < nb_fd; i++) {
2751 /* Fetch the poll data. */
2752 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2753 int pollfd = LTTNG_POLL_GETFD(&events, i);
2754
2755 health_code_update();
2756
2757 /* Skip the command pipe. It's handled in the first loop. */
2758 if (pollfd == relay_cmd_pipe[0]) {
2759 continue;
2760 }
2761
2762 if (revents) {
2763 rcu_read_lock();
2764 lttng_ht_lookup(relay_connections_ht,
2765 (void *)((unsigned long) pollfd),
2766 &iter);
2767 node = lttng_ht_iter_get_node_ulong(&iter);
2768 if (node == NULL) {
2769 /* Skip it. Might be removed before. */
2770 rcu_read_unlock();
2771 continue;
2772 }
2773 relay_connection = caa_container_of(node,
2774 struct relay_command, sock_n);
2775
2776 if (revents & LPOLLIN) {
2777 if (relay_connection->type != RELAY_DATA) {
2778 continue;
2779 }
2780
2781 ret = relay_process_data(relay_connection);
2782 /* connection closed */
2783 if (ret < 0) {
2784 relay_cleanup_poll_connection(&events, pollfd);
2785 relay_del_connection(relay_connections_ht,
2786 &iter, relay_connection, sessions_ht);
2787 DBG("Data connection closed with %d", pollfd);
2788 /*
2789 * Every goto restart call sets the last seen fd where
2790 * here we don't really care since we gracefully
2791 * continue the loop after the connection is deleted.
2792 */
2793 } else {
2794 /* Keep last seen port. */
2795 last_seen_data_fd = pollfd;
2796 rcu_read_unlock();
2797 goto restart;
2798 }
2799 }
2800 rcu_read_unlock();
2801 }
2802 }
2803 last_seen_data_fd = -1;
2804 }
2805
2806 /* Normal exit, no error */
2807 ret = 0;
2808
2809 exit:
2810 error:
2811 lttng_poll_clean(&events);
2812
2813 /* empty the hash table and free the memory */
2814 rcu_read_lock();
2815 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2816 health_code_update();
2817
2818 node = lttng_ht_iter_get_node_ulong(&iter);
2819 if (node) {
2820 relay_connection = caa_container_of(node,
2821 struct relay_command, sock_n);
2822 relay_del_connection(relay_connections_ht,
2823 &iter, relay_connection, sessions_ht);
2824 }
2825 }
2826 rcu_read_unlock();
2827 error_poll_create:
2828 lttng_ht_destroy(indexes_ht);
2829 indexes_ht_error:
2830 lttng_ht_destroy(relay_connections_ht);
2831 relay_connections_ht_error:
2832 /* Close relay cmd pipes */
2833 utils_close_pipe(relay_cmd_pipe);
2834 if (err) {
2835 DBG("Thread exited with error");
2836 }
2837 DBG("Worker thread cleanup complete");
2838 free(data_buffer);
2839 error_testpoint:
2840 if (err) {
2841 health_error();
2842 ERR("Health error occurred in %s", __func__);
2843 }
2844 health_unregister(health_relayd);
2845 rcu_unregister_thread();
2846 stop_threads();
2847 return NULL;
2848 }
2849
2850 /*
2851 * Create the relay command pipe to wake thread_manage_apps.
2852 * Closed in cleanup().
2853 */
2854 static int create_relay_cmd_pipe(void)
2855 {
2856 int ret;
2857
2858 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2859
2860 return ret;
2861 }
2862
2863 /*
2864 * main
2865 */
2866 int main(int argc, char **argv)
2867 {
2868 int ret = 0;
2869 void *status;
2870 struct relay_local_data *relay_ctx;
2871
2872 /* Parse arguments */
2873 progname = argv[0];
2874 if ((ret = set_options(argc, argv)) < 0) {
2875 goto exit;
2876 }
2877
2878 if ((ret = set_signal_handler()) < 0) {
2879 goto exit;
2880 }
2881
2882 /* Try to create directory if -o, --output is specified. */
2883 if (opt_output_path) {
2884 if (*opt_output_path != '/') {
2885 ERR("Please specify an absolute path for -o, --output PATH");
2886 goto exit;
2887 }
2888
2889 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2890 if (ret < 0) {
2891 ERR("Unable to create %s", opt_output_path);
2892 goto exit;
2893 }
2894 }
2895
2896 /* Daemonize */
2897 if (opt_daemon || opt_background) {
2898 int i;
2899
2900 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2901 !opt_background);
2902 if (ret < 0) {
2903 goto exit;
2904 }
2905
2906 /*
2907 * We are in the child. Make sure all other file
2908 * descriptors are closed, in case we are called with
2909 * more opened file descriptors than the standard ones.
2910 */
2911 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2912 (void) close(i);
2913 }
2914 }
2915
2916 /* Create thread quit pipe */
2917 if ((ret = init_thread_quit_pipe()) < 0) {
2918 goto error;
2919 }
2920
2921 /* We need those values for the file/dir creation. */
2922 relayd_uid = getuid();
2923 relayd_gid = getgid();
2924
2925 /* Check if daemon is UID = 0 */
2926 if (relayd_uid == 0) {
2927 if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) {
2928 ERR("Need to be root to use ports < 1024");
2929 ret = -1;
2930 goto exit;
2931 }
2932 }
2933
2934 /* Setup the thread apps communication pipe. */
2935 if ((ret = create_relay_cmd_pipe()) < 0) {
2936 goto exit;
2937 }
2938
2939 /* Init relay command queue. */
2940 cds_wfq_init(&relay_cmd_queue.queue);
2941
2942 /* Set up max poll set size */
2943 lttng_poll_set_max_size();
2944
2945 /* Initialize communication library */
2946 lttcomm_init();
2947 lttcomm_inet_init();
2948
2949 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2950 if (!relay_ctx) {
2951 PERROR("relay_ctx");
2952 goto exit;
2953 }
2954
2955 /* tables of sessions indexed by session ID */
2956 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2957 if (!relay_ctx->sessions_ht) {
2958 goto exit_relay_ctx_sessions;
2959 }
2960
2961 /* tables of streams indexed by stream ID */
2962 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2963 if (!relay_streams_ht) {
2964 goto exit_relay_ctx_streams;
2965 }
2966
2967 /* tables of streams indexed by stream ID */
2968 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2969 if (!viewer_streams_ht) {
2970 goto exit_relay_ctx_viewer_streams;
2971 }
2972
2973 /* Initialize thread health monitoring */
2974 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2975 if (!health_relayd) {
2976 PERROR("health_app_create error");
2977 goto exit_health_app_create;
2978 }
2979
2980 ret = utils_create_pipe(health_quit_pipe);
2981 if (ret < 0) {
2982 goto error_health_pipe;
2983 }
2984
2985 /* Create thread to manage the client socket */
2986 ret = pthread_create(&health_thread, NULL,
2987 thread_manage_health, (void *) NULL);
2988 if (ret != 0) {
2989 PERROR("pthread_create health");
2990 goto health_error;
2991 }
2992
2993 /* Setup the dispatcher thread */
2994 ret = pthread_create(&dispatcher_thread, NULL,
2995 relay_thread_dispatcher, (void *) NULL);
2996 if (ret != 0) {
2997 PERROR("pthread_create dispatcher");
2998 goto exit_dispatcher;
2999 }
3000
3001 /* Setup the worker thread */
3002 ret = pthread_create(&worker_thread, NULL,
3003 relay_thread_worker, (void *) relay_ctx);
3004 if (ret != 0) {
3005 PERROR("pthread_create worker");
3006 goto exit_worker;
3007 }
3008
3009 /* Setup the listener thread */
3010 ret = pthread_create(&listener_thread, NULL,
3011 relay_thread_listener, (void *) NULL);
3012 if (ret != 0) {
3013 PERROR("pthread_create listener");
3014 goto exit_listener;
3015 }
3016
3017 ret = live_start_threads(live_uri, relay_ctx);
3018 if (ret != 0) {
3019 ERR("Starting live viewer threads");
3020 goto exit_live;
3021 }
3022
3023 exit_live:
3024 ret = pthread_join(listener_thread, &status);
3025 if (ret != 0) {
3026 PERROR("pthread_join");
3027 goto error; /* join error, exit without cleanup */
3028 }
3029
3030 exit_listener:
3031 ret = pthread_join(worker_thread, &status);
3032 if (ret != 0) {
3033 PERROR("pthread_join");
3034 goto error; /* join error, exit without cleanup */
3035 }
3036
3037 exit_worker:
3038 ret = pthread_join(dispatcher_thread, &status);
3039 if (ret != 0) {
3040 PERROR("pthread_join");
3041 goto error; /* join error, exit without cleanup */
3042 }
3043
3044 exit_dispatcher:
3045 ret = pthread_join(health_thread, &status);
3046 if (ret != 0) {
3047 PERROR("pthread_join health thread");
3048 goto error; /* join error, exit without cleanup */
3049 }
3050
3051 /*
3052 * Stop live threads only after joining other threads.
3053 */
3054 live_stop_threads();
3055
3056 health_error:
3057 utils_close_pipe(health_quit_pipe);
3058
3059 error_health_pipe:
3060 health_app_destroy(health_relayd);
3061
3062 exit_health_app_create:
3063 lttng_ht_destroy(viewer_streams_ht);
3064
3065 exit_relay_ctx_viewer_streams:
3066 lttng_ht_destroy(relay_streams_ht);
3067
3068 exit_relay_ctx_streams:
3069 lttng_ht_destroy(relay_ctx->sessions_ht);
3070
3071 exit_relay_ctx_sessions:
3072 free(relay_ctx);
3073
3074 exit:
3075 cleanup();
3076 if (!ret) {
3077 exit(EXIT_SUCCESS);
3078 }
3079
3080 error:
3081 exit(EXIT_FAILURE);
3082 }
This page took 0.179166 seconds and 3 git commands to generate.