2fcc60af00d9d3b29b9809865ed7747ab4dbde45
[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 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2 only,
9 * as published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * more details.
15 *
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19 */
20
21 #define _LGPL_SOURCE
22 #include <getopt.h>
23 #include <grp.h>
24 #include <limits.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <sys/mman.h>
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
34 #include <sys/stat.h>
35 #include <sys/types.h>
36 #include <sys/wait.h>
37 #include <inttypes.h>
38 #include <urcu/futex.h>
39 #include <urcu/uatomic.h>
40 #include <unistd.h>
41 #include <fcntl.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/compat/endian.h>
48 #include <common/compat/getenv.h>
49 #include <common/defaults.h>
50 #include <common/daemonize.h>
51 #include <common/futex.h>
52 #include <common/sessiond-comm/sessiond-comm.h>
53 #include <common/sessiond-comm/inet.h>
54 #include <common/sessiond-comm/relayd.h>
55 #include <common/uri.h>
56 #include <common/utils.h>
57 #include <common/config/session-config.h>
58 #include <urcu/rculist.h>
59
60 #include "cmd.h"
61 #include "ctf-trace.h"
62 #include "index.h"
63 #include "utils.h"
64 #include "lttng-relayd.h"
65 #include "live.h"
66 #include "health-relayd.h"
67 #include "testpoint.h"
68 #include "viewer-stream.h"
69 #include "session.h"
70 #include "stream.h"
71 #include "connection.h"
72 #include "tracefile-array.h"
73
74 static const char *help_msg =
75 #ifdef LTTNG_EMBED_HELP
76 #include <lttng-relayd.8.h>
77 #else
78 NULL
79 #endif
80 ;
81
82 /* command line options */
83 char *opt_output_path;
84 static int opt_daemon, opt_background;
85
86 /*
87 * We need to wait for listener and live listener threads, as well as
88 * health check thread, before being ready to signal readiness.
89 */
90 #define NR_LTTNG_RELAY_READY 3
91 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
92
93 /* Size of receive buffer. */
94 #define RECV_DATA_BUFFER_SIZE 65536
95
96 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
97 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
98
99 static struct lttng_uri *control_uri;
100 static struct lttng_uri *data_uri;
101 static struct lttng_uri *live_uri;
102
103 const char *progname;
104
105 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
106 static int tracing_group_name_override;
107
108 const char * const config_section_name = "relayd";
109
110 /*
111 * Quit pipe for all threads. This permits a single cancellation point
112 * for all threads when receiving an event on the pipe.
113 */
114 int thread_quit_pipe[2] = { -1, -1 };
115
116 /*
117 * This pipe is used to inform the worker thread that a command is queued and
118 * ready to be processed.
119 */
120 static int relay_conn_pipe[2] = { -1, -1 };
121
122 /* Shared between threads */
123 static int dispatch_thread_exit;
124
125 static pthread_t listener_thread;
126 static pthread_t dispatcher_thread;
127 static pthread_t worker_thread;
128 static pthread_t health_thread;
129
130 /*
131 * last_relay_stream_id_lock protects last_relay_stream_id increment
132 * atomicity on 32-bit architectures.
133 */
134 static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
135 static uint64_t last_relay_stream_id;
136
137 /*
138 * Relay command queue.
139 *
140 * The relay_thread_listener and relay_thread_dispatcher communicate with this
141 * queue.
142 */
143 static struct relay_conn_queue relay_conn_queue;
144
145 /* buffer allocated at startup, used to store the trace data */
146 static char *data_buffer;
147 static unsigned int data_buffer_size;
148
149 /* Global relay stream hash table. */
150 struct lttng_ht *relay_streams_ht;
151
152 /* Global relay viewer stream hash table. */
153 struct lttng_ht *viewer_streams_ht;
154
155 /* Global relay sessions hash table. */
156 struct lttng_ht *sessions_ht;
157
158 /* Relayd health monitoring */
159 struct health_app *health_relayd;
160
161 static struct option long_options[] = {
162 { "control-port", 1, 0, 'C', },
163 { "data-port", 1, 0, 'D', },
164 { "live-port", 1, 0, 'L', },
165 { "daemonize", 0, 0, 'd', },
166 { "background", 0, 0, 'b', },
167 { "group", 1, 0, 'g', },
168 { "help", 0, 0, 'h', },
169 { "output", 1, 0, 'o', },
170 { "verbose", 0, 0, 'v', },
171 { "config", 1, 0, 'f' },
172 { "version", 0, 0, 'V' },
173 { NULL, 0, 0, 0, },
174 };
175
176 static const char *config_ignore_options[] = { "help", "config", "version" };
177
178 /*
179 * Take an option from the getopt output and set it in the right variable to be
180 * used later.
181 *
182 * Return 0 on success else a negative value.
183 */
184 static int set_option(int opt, const char *arg, const char *optname)
185 {
186 int ret;
187
188 switch (opt) {
189 case 0:
190 fprintf(stderr, "option %s", optname);
191 if (arg) {
192 fprintf(stderr, " with arg %s\n", arg);
193 }
194 break;
195 case 'C':
196 if (lttng_is_setuid_setgid()) {
197 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
198 "-C, --control-port");
199 } else {
200 ret = uri_parse(arg, &control_uri);
201 if (ret < 0) {
202 ERR("Invalid control URI specified");
203 goto end;
204 }
205 if (control_uri->port == 0) {
206 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
207 }
208 }
209 break;
210 case 'D':
211 if (lttng_is_setuid_setgid()) {
212 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
213 "-D, -data-port");
214 } else {
215 ret = uri_parse(arg, &data_uri);
216 if (ret < 0) {
217 ERR("Invalid data URI specified");
218 goto end;
219 }
220 if (data_uri->port == 0) {
221 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
222 }
223 }
224 break;
225 case 'L':
226 if (lttng_is_setuid_setgid()) {
227 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
228 "-L, -live-port");
229 } else {
230 ret = uri_parse(arg, &live_uri);
231 if (ret < 0) {
232 ERR("Invalid live URI specified");
233 goto end;
234 }
235 if (live_uri->port == 0) {
236 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
237 }
238 }
239 break;
240 case 'd':
241 opt_daemon = 1;
242 break;
243 case 'b':
244 opt_background = 1;
245 break;
246 case 'g':
247 if (lttng_is_setuid_setgid()) {
248 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
249 "-g, --group");
250 } else {
251 tracing_group_name = strdup(arg);
252 if (tracing_group_name == NULL) {
253 ret = -errno;
254 PERROR("strdup");
255 goto end;
256 }
257 tracing_group_name_override = 1;
258 }
259 break;
260 case 'h':
261 ret = utils_show_help(8, "lttng-relayd", help_msg);
262 if (ret) {
263 ERR("Cannot show --help for `lttng-relayd`");
264 perror("exec");
265 }
266 exit(EXIT_FAILURE);
267 case 'V':
268 fprintf(stdout, "%s\n", VERSION);
269 exit(EXIT_SUCCESS);
270 case 'o':
271 if (lttng_is_setuid_setgid()) {
272 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
273 "-o, --output");
274 } else {
275 ret = asprintf(&opt_output_path, "%s", arg);
276 if (ret < 0) {
277 ret = -errno;
278 PERROR("asprintf opt_output_path");
279 goto end;
280 }
281 }
282 break;
283 case 'v':
284 /* Verbose level can increase using multiple -v */
285 if (arg) {
286 lttng_opt_verbose = config_parse_value(arg);
287 } else {
288 /* Only 3 level of verbosity (-vvv). */
289 if (lttng_opt_verbose < 3) {
290 lttng_opt_verbose += 1;
291 }
292 }
293 break;
294 default:
295 /* Unknown option or other error.
296 * Error is printed by getopt, just return */
297 ret = -1;
298 goto end;
299 }
300
301 /* All good. */
302 ret = 0;
303
304 end:
305 return ret;
306 }
307
308 /*
309 * config_entry_handler_cb used to handle options read from a config file.
310 * See config_entry_handler_cb comment in common/config/session-config.h for the
311 * return value conventions.
312 */
313 static int config_entry_handler(const struct config_entry *entry, void *unused)
314 {
315 int ret = 0, i;
316
317 if (!entry || !entry->name || !entry->value) {
318 ret = -EINVAL;
319 goto end;
320 }
321
322 /* Check if the option is to be ignored */
323 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
324 if (!strcmp(entry->name, config_ignore_options[i])) {
325 goto end;
326 }
327 }
328
329 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
330 /* Ignore if entry name is not fully matched. */
331 if (strcmp(entry->name, long_options[i].name)) {
332 continue;
333 }
334
335 /*
336 * If the option takes no argument on the command line,
337 * we have to check if the value is "true". We support
338 * non-zero numeric values, true, on and yes.
339 */
340 if (!long_options[i].has_arg) {
341 ret = config_parse_value(entry->value);
342 if (ret <= 0) {
343 if (ret) {
344 WARN("Invalid configuration value \"%s\" for option %s",
345 entry->value, entry->name);
346 }
347 /* False, skip boolean config option. */
348 goto end;
349 }
350 }
351
352 ret = set_option(long_options[i].val, entry->value, entry->name);
353 goto end;
354 }
355
356 WARN("Unrecognized option \"%s\" in daemon configuration file.",
357 entry->name);
358
359 end:
360 return ret;
361 }
362
363 static int set_options(int argc, char **argv)
364 {
365 int c, ret = 0, option_index = 0, retval = 0;
366 int orig_optopt = optopt, orig_optind = optind;
367 char *default_address, *optstring;
368 const char *config_path = NULL;
369
370 optstring = utils_generate_optstring(long_options,
371 sizeof(long_options) / sizeof(struct option));
372 if (!optstring) {
373 retval = -ENOMEM;
374 goto exit;
375 }
376
377 /* Check for the --config option */
378
379 while ((c = getopt_long(argc, argv, optstring, long_options,
380 &option_index)) != -1) {
381 if (c == '?') {
382 retval = -EINVAL;
383 goto exit;
384 } else if (c != 'f') {
385 continue;
386 }
387
388 if (lttng_is_setuid_setgid()) {
389 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
390 "-f, --config");
391 } else {
392 config_path = utils_expand_path(optarg);
393 if (!config_path) {
394 ERR("Failed to resolve path: %s", optarg);
395 }
396 }
397 }
398
399 ret = config_get_section_entries(config_path, config_section_name,
400 config_entry_handler, NULL);
401 if (ret) {
402 if (ret > 0) {
403 ERR("Invalid configuration option at line %i", ret);
404 }
405 retval = -1;
406 goto exit;
407 }
408
409 /* Reset getopt's global state */
410 optopt = orig_optopt;
411 optind = orig_optind;
412 while (1) {
413 c = getopt_long(argc, argv, optstring, long_options, &option_index);
414 if (c == -1) {
415 break;
416 }
417
418 ret = set_option(c, optarg, long_options[option_index].name);
419 if (ret < 0) {
420 retval = -1;
421 goto exit;
422 }
423 }
424
425 /* assign default values */
426 if (control_uri == NULL) {
427 ret = asprintf(&default_address,
428 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
429 DEFAULT_NETWORK_CONTROL_PORT);
430 if (ret < 0) {
431 PERROR("asprintf default data address");
432 retval = -1;
433 goto exit;
434 }
435
436 ret = uri_parse(default_address, &control_uri);
437 free(default_address);
438 if (ret < 0) {
439 ERR("Invalid control URI specified");
440 retval = -1;
441 goto exit;
442 }
443 }
444 if (data_uri == NULL) {
445 ret = asprintf(&default_address,
446 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
447 DEFAULT_NETWORK_DATA_PORT);
448 if (ret < 0) {
449 PERROR("asprintf default data address");
450 retval = -1;
451 goto exit;
452 }
453
454 ret = uri_parse(default_address, &data_uri);
455 free(default_address);
456 if (ret < 0) {
457 ERR("Invalid data URI specified");
458 retval = -1;
459 goto exit;
460 }
461 }
462 if (live_uri == NULL) {
463 ret = asprintf(&default_address,
464 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
465 DEFAULT_NETWORK_VIEWER_PORT);
466 if (ret < 0) {
467 PERROR("asprintf default viewer control address");
468 retval = -1;
469 goto exit;
470 }
471
472 ret = uri_parse(default_address, &live_uri);
473 free(default_address);
474 if (ret < 0) {
475 ERR("Invalid viewer control URI specified");
476 retval = -1;
477 goto exit;
478 }
479 }
480
481 exit:
482 free(optstring);
483 return retval;
484 }
485
486 static void print_global_objects(void)
487 {
488 rcu_register_thread();
489
490 print_viewer_streams();
491 print_relay_streams();
492 print_sessions();
493
494 rcu_unregister_thread();
495 }
496
497 /*
498 * Cleanup the daemon
499 */
500 static void relayd_cleanup(void)
501 {
502 print_global_objects();
503
504 DBG("Cleaning up");
505
506 if (viewer_streams_ht)
507 lttng_ht_destroy(viewer_streams_ht);
508 if (relay_streams_ht)
509 lttng_ht_destroy(relay_streams_ht);
510 if (sessions_ht)
511 lttng_ht_destroy(sessions_ht);
512
513 /* free the dynamically allocated opt_output_path */
514 free(opt_output_path);
515
516 /* Close thread quit pipes */
517 utils_close_pipe(thread_quit_pipe);
518
519 uri_free(control_uri);
520 uri_free(data_uri);
521 /* Live URI is freed in the live thread. */
522
523 if (tracing_group_name_override) {
524 free((void *) tracing_group_name);
525 }
526 }
527
528 /*
529 * Write to writable pipe used to notify a thread.
530 */
531 static int notify_thread_pipe(int wpipe)
532 {
533 ssize_t ret;
534
535 ret = lttng_write(wpipe, "!", 1);
536 if (ret < 1) {
537 PERROR("write poll pipe");
538 goto end;
539 }
540 ret = 0;
541 end:
542 return ret;
543 }
544
545 static int notify_health_quit_pipe(int *pipe)
546 {
547 ssize_t ret;
548
549 ret = lttng_write(pipe[1], "4", 1);
550 if (ret < 1) {
551 PERROR("write relay health quit");
552 goto end;
553 }
554 ret = 0;
555 end:
556 return ret;
557 }
558
559 /*
560 * Stop all relayd and relayd-live threads.
561 */
562 int lttng_relay_stop_threads(void)
563 {
564 int retval = 0;
565
566 /* Stopping all threads */
567 DBG("Terminating all threads");
568 if (notify_thread_pipe(thread_quit_pipe[1])) {
569 ERR("write error on thread quit pipe");
570 retval = -1;
571 }
572
573 if (notify_health_quit_pipe(health_quit_pipe)) {
574 ERR("write error on health quit pipe");
575 }
576
577 /* Dispatch thread */
578 CMM_STORE_SHARED(dispatch_thread_exit, 1);
579 futex_nto1_wake(&relay_conn_queue.futex);
580
581 if (relayd_live_stop()) {
582 ERR("Error stopping live threads");
583 retval = -1;
584 }
585 return retval;
586 }
587
588 /*
589 * Signal handler for the daemon
590 *
591 * Simply stop all worker threads, leaving main() return gracefully after
592 * joining all threads and calling cleanup().
593 */
594 static void sighandler(int sig)
595 {
596 switch (sig) {
597 case SIGINT:
598 DBG("SIGINT caught");
599 if (lttng_relay_stop_threads()) {
600 ERR("Error stopping threads");
601 }
602 break;
603 case SIGTERM:
604 DBG("SIGTERM caught");
605 if (lttng_relay_stop_threads()) {
606 ERR("Error stopping threads");
607 }
608 break;
609 case SIGUSR1:
610 CMM_STORE_SHARED(recv_child_signal, 1);
611 break;
612 default:
613 break;
614 }
615 }
616
617 /*
618 * Setup signal handler for :
619 * SIGINT, SIGTERM, SIGPIPE
620 */
621 static int set_signal_handler(void)
622 {
623 int ret = 0;
624 struct sigaction sa;
625 sigset_t sigset;
626
627 if ((ret = sigemptyset(&sigset)) < 0) {
628 PERROR("sigemptyset");
629 return ret;
630 }
631
632 sa.sa_mask = sigset;
633 sa.sa_flags = 0;
634
635 sa.sa_handler = sighandler;
636 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
637 PERROR("sigaction");
638 return ret;
639 }
640
641 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
642 PERROR("sigaction");
643 return ret;
644 }
645
646 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
647 PERROR("sigaction");
648 return ret;
649 }
650
651 sa.sa_handler = SIG_IGN;
652 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
653 PERROR("sigaction");
654 return ret;
655 }
656
657 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
658
659 return ret;
660 }
661
662 void lttng_relay_notify_ready(void)
663 {
664 /* Notify the parent of the fork() process that we are ready. */
665 if (opt_daemon || opt_background) {
666 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
667 kill(child_ppid, SIGUSR1);
668 }
669 }
670 }
671
672 /*
673 * Init thread quit pipe.
674 *
675 * Return -1 on error or 0 if all pipes are created.
676 */
677 static int init_thread_quit_pipe(void)
678 {
679 int ret;
680
681 ret = utils_create_pipe_cloexec(thread_quit_pipe);
682
683 return ret;
684 }
685
686 /*
687 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
688 */
689 static int create_thread_poll_set(struct lttng_poll_event *events, int size)
690 {
691 int ret;
692
693 if (events == NULL || size == 0) {
694 ret = -1;
695 goto error;
696 }
697
698 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
699 if (ret < 0) {
700 goto error;
701 }
702
703 /* Add quit pipe */
704 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
705 if (ret < 0) {
706 goto error;
707 }
708
709 return 0;
710
711 error:
712 return ret;
713 }
714
715 /*
716 * Check if the thread quit pipe was triggered.
717 *
718 * Return 1 if it was triggered else 0;
719 */
720 static int check_thread_quit_pipe(int fd, uint32_t events)
721 {
722 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
723 return 1;
724 }
725
726 return 0;
727 }
728
729 /*
730 * Create and init socket from uri.
731 */
732 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri)
733 {
734 int ret;
735 struct lttcomm_sock *sock = NULL;
736
737 sock = lttcomm_alloc_sock_from_uri(uri);
738 if (sock == NULL) {
739 ERR("Allocating socket");
740 goto error;
741 }
742
743 ret = lttcomm_create_sock(sock);
744 if (ret < 0) {
745 goto error;
746 }
747 DBG("Listening on sock %d", sock->fd);
748
749 ret = sock->ops->bind(sock);
750 if (ret < 0) {
751 goto error;
752 }
753
754 ret = sock->ops->listen(sock, -1);
755 if (ret < 0) {
756 goto error;
757
758 }
759
760 return sock;
761
762 error:
763 if (sock) {
764 lttcomm_destroy_sock(sock);
765 }
766 return NULL;
767 }
768
769 /*
770 * This thread manages the listening for new connections on the network
771 */
772 static void *relay_thread_listener(void *data)
773 {
774 int i, ret, pollfd, err = -1;
775 uint32_t revents, nb_fd;
776 struct lttng_poll_event events;
777 struct lttcomm_sock *control_sock, *data_sock;
778
779 DBG("[thread] Relay listener started");
780
781 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
782
783 health_code_update();
784
785 control_sock = relay_socket_create(control_uri);
786 if (!control_sock) {
787 goto error_sock_control;
788 }
789
790 data_sock = relay_socket_create(data_uri);
791 if (!data_sock) {
792 goto error_sock_relay;
793 }
794
795 /*
796 * Pass 3 as size here for the thread quit pipe, control and
797 * data socket.
798 */
799 ret = create_thread_poll_set(&events, 3);
800 if (ret < 0) {
801 goto error_create_poll;
802 }
803
804 /* Add the control socket */
805 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
806 if (ret < 0) {
807 goto error_poll_add;
808 }
809
810 /* Add the data socket */
811 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
812 if (ret < 0) {
813 goto error_poll_add;
814 }
815
816 lttng_relay_notify_ready();
817
818 if (testpoint(relayd_thread_listener)) {
819 goto error_testpoint;
820 }
821
822 while (1) {
823 health_code_update();
824
825 DBG("Listener accepting connections");
826
827 restart:
828 health_poll_entry();
829 ret = lttng_poll_wait(&events, -1);
830 health_poll_exit();
831 if (ret < 0) {
832 /*
833 * Restart interrupted system call.
834 */
835 if (errno == EINTR) {
836 goto restart;
837 }
838 goto error;
839 }
840
841 nb_fd = ret;
842
843 DBG("Relay new connection received");
844 for (i = 0; i < nb_fd; i++) {
845 health_code_update();
846
847 /* Fetch once the poll data */
848 revents = LTTNG_POLL_GETEV(&events, i);
849 pollfd = LTTNG_POLL_GETFD(&events, i);
850
851 if (!revents) {
852 /*
853 * No activity for this FD (poll
854 * implementation).
855 */
856 continue;
857 }
858
859 /* Thread quit pipe has been closed. Killing thread. */
860 ret = check_thread_quit_pipe(pollfd, revents);
861 if (ret) {
862 err = 0;
863 goto exit;
864 }
865
866 if (revents & LPOLLIN) {
867 /*
868 * A new connection is requested, therefore a
869 * sessiond/consumerd connection is allocated in
870 * this thread, enqueued to a global queue and
871 * dequeued (and freed) in the worker thread.
872 */
873 int val = 1;
874 struct relay_connection *new_conn;
875 struct lttcomm_sock *newsock;
876 enum connection_type type;
877
878 if (pollfd == data_sock->fd) {
879 type = RELAY_DATA;
880 newsock = data_sock->ops->accept(data_sock);
881 DBG("Relay data connection accepted, socket %d",
882 newsock->fd);
883 } else {
884 assert(pollfd == control_sock->fd);
885 type = RELAY_CONTROL;
886 newsock = control_sock->ops->accept(control_sock);
887 DBG("Relay control connection accepted, socket %d",
888 newsock->fd);
889 }
890 if (!newsock) {
891 PERROR("accepting sock");
892 goto error;
893 }
894
895 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
896 sizeof(val));
897 if (ret < 0) {
898 PERROR("setsockopt inet");
899 lttcomm_destroy_sock(newsock);
900 goto error;
901 }
902 new_conn = connection_create(newsock, type);
903 if (!new_conn) {
904 lttcomm_destroy_sock(newsock);
905 goto error;
906 }
907
908 /* Enqueue request for the dispatcher thread. */
909 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
910 &new_conn->qnode);
911
912 /*
913 * Wake the dispatch queue futex.
914 * Implicit memory barrier with the
915 * exchange in cds_wfcq_enqueue.
916 */
917 futex_nto1_wake(&relay_conn_queue.futex);
918 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
919 ERR("socket poll error");
920 goto error;
921 } else {
922 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
923 goto error;
924 }
925 }
926 }
927
928 exit:
929 error:
930 error_poll_add:
931 error_testpoint:
932 lttng_poll_clean(&events);
933 error_create_poll:
934 if (data_sock->fd >= 0) {
935 ret = data_sock->ops->close(data_sock);
936 if (ret) {
937 PERROR("close");
938 }
939 }
940 lttcomm_destroy_sock(data_sock);
941 error_sock_relay:
942 if (control_sock->fd >= 0) {
943 ret = control_sock->ops->close(control_sock);
944 if (ret) {
945 PERROR("close");
946 }
947 }
948 lttcomm_destroy_sock(control_sock);
949 error_sock_control:
950 if (err) {
951 health_error();
952 ERR("Health error occurred in %s", __func__);
953 }
954 health_unregister(health_relayd);
955 DBG("Relay listener thread cleanup complete");
956 lttng_relay_stop_threads();
957 return NULL;
958 }
959
960 /*
961 * This thread manages the dispatching of the requests to worker threads
962 */
963 static void *relay_thread_dispatcher(void *data)
964 {
965 int err = -1;
966 ssize_t ret;
967 struct cds_wfcq_node *node;
968 struct relay_connection *new_conn = NULL;
969
970 DBG("[thread] Relay dispatcher started");
971
972 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
973
974 if (testpoint(relayd_thread_dispatcher)) {
975 goto error_testpoint;
976 }
977
978 health_code_update();
979
980 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
981 health_code_update();
982
983 /* Atomically prepare the queue futex */
984 futex_nto1_prepare(&relay_conn_queue.futex);
985
986 do {
987 health_code_update();
988
989 /* Dequeue commands */
990 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
991 &relay_conn_queue.tail);
992 if (node == NULL) {
993 DBG("Woken up but nothing in the relay command queue");
994 /* Continue thread execution */
995 break;
996 }
997 new_conn = caa_container_of(node, struct relay_connection, qnode);
998
999 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1000
1001 /*
1002 * Inform worker thread of the new request. This
1003 * call is blocking so we can be assured that
1004 * the data will be read at some point in time
1005 * or wait to the end of the world :)
1006 */
1007 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1008 if (ret < 0) {
1009 PERROR("write connection pipe");
1010 connection_put(new_conn);
1011 goto error;
1012 }
1013 } while (node != NULL);
1014
1015 /* Futex wait on queue. Blocking call on futex() */
1016 health_poll_entry();
1017 futex_nto1_wait(&relay_conn_queue.futex);
1018 health_poll_exit();
1019 }
1020
1021 /* Normal exit, no error */
1022 err = 0;
1023
1024 error:
1025 error_testpoint:
1026 if (err) {
1027 health_error();
1028 ERR("Health error occurred in %s", __func__);
1029 }
1030 health_unregister(health_relayd);
1031 DBG("Dispatch thread dying");
1032 lttng_relay_stop_threads();
1033 return NULL;
1034 }
1035
1036 /*
1037 * Set index data from the control port to a given index object.
1038 */
1039 static int set_index_control_data(struct relay_index *index,
1040 struct lttcomm_relayd_index *data,
1041 struct relay_connection *conn)
1042 {
1043 struct ctf_packet_index index_data;
1044
1045 /*
1046 * The index on disk is encoded in big endian, so we don't need
1047 * to convert the data received on the network. The data_offset
1048 * value is NEVER modified here and is updated by the data
1049 * thread.
1050 */
1051 index_data.packet_size = data->packet_size;
1052 index_data.content_size = data->content_size;
1053 index_data.timestamp_begin = data->timestamp_begin;
1054 index_data.timestamp_end = data->timestamp_end;
1055 index_data.events_discarded = data->events_discarded;
1056 index_data.stream_id = data->stream_id;
1057
1058 if (conn->minor >= 8) {
1059 index->index_data.stream_instance_id = data->stream_instance_id;
1060 index->index_data.packet_seq_num = data->packet_seq_num;
1061 }
1062
1063 return relay_index_set_data(index, &index_data);
1064 }
1065
1066 /*
1067 * Handle the RELAYD_CREATE_SESSION command.
1068 *
1069 * On success, send back the session id or else return a negative value.
1070 */
1071 static int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
1072 struct relay_connection *conn)
1073 {
1074 int ret = 0, send_ret;
1075 struct relay_session *session;
1076 struct lttcomm_relayd_status_session reply;
1077 char session_name[LTTNG_NAME_MAX];
1078 char hostname[LTTNG_HOST_NAME_MAX];
1079 uint32_t live_timer = 0;
1080 bool snapshot = false;
1081
1082 memset(session_name, 0, LTTNG_NAME_MAX);
1083 memset(hostname, 0, LTTNG_HOST_NAME_MAX);
1084
1085 memset(&reply, 0, sizeof(reply));
1086
1087 switch (conn->minor) {
1088 case 1:
1089 case 2:
1090 case 3:
1091 break;
1092 case 4: /* LTTng sessiond 2.4 */
1093 default:
1094 ret = cmd_create_session_2_4(conn, session_name,
1095 hostname, &live_timer, &snapshot);
1096 }
1097 if (ret < 0) {
1098 goto send_reply;
1099 }
1100
1101 session = session_create(session_name, hostname, live_timer,
1102 snapshot, conn->major, conn->minor);
1103 if (!session) {
1104 ret = -1;
1105 goto send_reply;
1106 }
1107 assert(!conn->session);
1108 conn->session = session;
1109 DBG("Created session %" PRIu64, session->id);
1110
1111 reply.session_id = htobe64(session->id);
1112
1113 send_reply:
1114 if (ret < 0) {
1115 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1116 } else {
1117 reply.ret_code = htobe32(LTTNG_OK);
1118 }
1119
1120 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1121 if (send_ret < 0) {
1122 ERR("Relayd sending session id");
1123 ret = send_ret;
1124 }
1125
1126 return ret;
1127 }
1128
1129 /*
1130 * When we have received all the streams and the metadata for a channel,
1131 * we make them visible to the viewer threads.
1132 */
1133 static void publish_connection_local_streams(struct relay_connection *conn)
1134 {
1135 struct relay_stream *stream;
1136 struct relay_session *session = conn->session;
1137
1138 /*
1139 * We publish all streams belonging to a session atomically wrt
1140 * session lock.
1141 */
1142 pthread_mutex_lock(&session->lock);
1143 rcu_read_lock();
1144 cds_list_for_each_entry_rcu(stream, &session->recv_list,
1145 recv_node) {
1146 stream_publish(stream);
1147 }
1148 rcu_read_unlock();
1149
1150 /*
1151 * Inform the viewer that there are new streams in the session.
1152 */
1153 if (session->viewer_attached) {
1154 uatomic_set(&session->new_streams, 1);
1155 }
1156 pthread_mutex_unlock(&session->lock);
1157 }
1158
1159 /*
1160 * relay_add_stream: allocate a new stream for a session
1161 */
1162 static int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1163 struct relay_connection *conn)
1164 {
1165 int ret;
1166 ssize_t send_ret;
1167 struct relay_session *session = conn->session;
1168 struct relay_stream *stream = NULL;
1169 struct lttcomm_relayd_status_stream reply;
1170 struct ctf_trace *trace = NULL;
1171 uint64_t stream_handle = -1ULL;
1172 char *path_name = NULL, *channel_name = NULL;
1173 uint64_t tracefile_size = 0, tracefile_count = 0;
1174
1175 if (!session || conn->version_check_done == 0) {
1176 ERR("Trying to add a stream before version check");
1177 ret = -1;
1178 goto end_no_session;
1179 }
1180
1181 switch (session->minor) {
1182 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
1183 ret = cmd_recv_stream_2_1(conn, &path_name,
1184 &channel_name);
1185 break;
1186 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
1187 default:
1188 ret = cmd_recv_stream_2_2(conn, &path_name,
1189 &channel_name, &tracefile_size, &tracefile_count);
1190 break;
1191 }
1192 if (ret < 0) {
1193 goto send_reply;
1194 }
1195
1196 trace = ctf_trace_get_by_path_or_create(session, path_name);
1197 if (!trace) {
1198 goto send_reply;
1199 }
1200 /* This stream here has one reference on the trace. */
1201
1202 pthread_mutex_lock(&last_relay_stream_id_lock);
1203 stream_handle = ++last_relay_stream_id;
1204 pthread_mutex_unlock(&last_relay_stream_id_lock);
1205
1206 /* We pass ownership of path_name and channel_name. */
1207 stream = stream_create(trace, stream_handle, path_name,
1208 channel_name, tracefile_size, tracefile_count);
1209 path_name = NULL;
1210 channel_name = NULL;
1211
1212 /*
1213 * Streams are the owners of their trace. Reference to trace is
1214 * kept within stream_create().
1215 */
1216 ctf_trace_put(trace);
1217
1218 send_reply:
1219 memset(&reply, 0, sizeof(reply));
1220 reply.handle = htobe64(stream_handle);
1221 if (!stream) {
1222 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1223 } else {
1224 reply.ret_code = htobe32(LTTNG_OK);
1225 }
1226
1227 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1228 sizeof(struct lttcomm_relayd_status_stream), 0);
1229 if (send_ret < 0) {
1230 ERR("Relay sending stream id");
1231 ret = (int) send_ret;
1232 }
1233
1234 end_no_session:
1235 free(path_name);
1236 free(channel_name);
1237 return ret;
1238 }
1239
1240 /*
1241 * relay_close_stream: close a specific stream
1242 */
1243 static int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1244 struct relay_connection *conn)
1245 {
1246 int ret, send_ret;
1247 struct relay_session *session = conn->session;
1248 struct lttcomm_relayd_close_stream stream_info;
1249 struct lttcomm_relayd_generic_reply reply;
1250 struct relay_stream *stream;
1251
1252 DBG("Close stream received");
1253
1254 if (!session || conn->version_check_done == 0) {
1255 ERR("Trying to close a stream before version check");
1256 ret = -1;
1257 goto end_no_session;
1258 }
1259
1260 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1261 sizeof(struct lttcomm_relayd_close_stream), 0);
1262 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1263 if (ret == 0) {
1264 /* Orderly shutdown. Not necessary to print an error. */
1265 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1266 } else {
1267 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1268 }
1269 ret = -1;
1270 goto end_no_session;
1271 }
1272
1273 stream = stream_get_by_id(be64toh(stream_info.stream_id));
1274 if (!stream) {
1275 ret = -1;
1276 goto end;
1277 }
1278
1279 /*
1280 * Set last_net_seq_num before the close flag. Required by data
1281 * pending check.
1282 */
1283 pthread_mutex_lock(&stream->lock);
1284 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1285 pthread_mutex_unlock(&stream->lock);
1286
1287 /*
1288 * This is one of the conditions which may trigger a stream close
1289 * with the others being:
1290 * 1) A close command is received for a stream
1291 * 2) The control connection owning the stream is closed
1292 * 3) We have received all of the stream's data _after_ a close
1293 * request.
1294 */
1295 try_stream_close(stream);
1296 if (stream->is_metadata) {
1297 struct relay_viewer_stream *vstream;
1298
1299 vstream = viewer_stream_get_by_id(stream->stream_handle);
1300 if (vstream) {
1301 if (vstream->metadata_sent == stream->metadata_received) {
1302 /*
1303 * Since all the metadata has been sent to the
1304 * viewer and that we have a request to close
1305 * its stream, we can safely teardown the
1306 * corresponding metadata viewer stream.
1307 */
1308 viewer_stream_put(vstream);
1309 }
1310 /* Put local reference. */
1311 viewer_stream_put(vstream);
1312 }
1313 }
1314 stream_put(stream);
1315
1316 end:
1317 memset(&reply, 0, sizeof(reply));
1318 if (ret < 0) {
1319 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1320 } else {
1321 reply.ret_code = htobe32(LTTNG_OK);
1322 }
1323 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1324 sizeof(struct lttcomm_relayd_generic_reply), 0);
1325 if (send_ret < 0) {
1326 ERR("Relay sending stream id");
1327 ret = send_ret;
1328 }
1329
1330 end_no_session:
1331 return ret;
1332 }
1333
1334 /*
1335 * relay_reset_metadata: reset a metadata stream
1336 */
1337 static
1338 int relay_reset_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1339 struct relay_connection *conn)
1340 {
1341 int ret, send_ret;
1342 struct relay_session *session = conn->session;
1343 struct lttcomm_relayd_reset_metadata stream_info;
1344 struct lttcomm_relayd_generic_reply reply;
1345 struct relay_stream *stream;
1346
1347 DBG("Reset metadata received");
1348
1349 if (!session || conn->version_check_done == 0) {
1350 ERR("Trying to reset a metadata stream before version check");
1351 ret = -1;
1352 goto end_no_session;
1353 }
1354
1355 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1356 sizeof(struct lttcomm_relayd_reset_metadata), 0);
1357 if (ret < sizeof(struct lttcomm_relayd_reset_metadata)) {
1358 if (ret == 0) {
1359 /* Orderly shutdown. Not necessary to print an error. */
1360 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1361 } else {
1362 ERR("Relay didn't receive valid reset_metadata struct "
1363 "size : %d", ret);
1364 }
1365 ret = -1;
1366 goto end_no_session;
1367 }
1368 DBG("Update metadata to version %" PRIu64, be64toh(stream_info.version));
1369
1370 /* Unsupported for live sessions for now. */
1371 if (session->live_timer != 0) {
1372 ret = -1;
1373 goto end;
1374 }
1375
1376 stream = stream_get_by_id(be64toh(stream_info.stream_id));
1377 if (!stream) {
1378 ret = -1;
1379 goto end;
1380 }
1381 pthread_mutex_lock(&stream->lock);
1382 if (!stream->is_metadata) {
1383 ret = -1;
1384 goto end_unlock;
1385 }
1386
1387 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
1388 0, 0, -1, -1, stream->stream_fd->fd, NULL,
1389 &stream->stream_fd->fd);
1390 if (ret < 0) {
1391 ERR("Failed to rotate metadata file %s of channel %s",
1392 stream->path_name, stream->channel_name);
1393 goto end_unlock;
1394 }
1395
1396 end_unlock:
1397 pthread_mutex_unlock(&stream->lock);
1398 stream_put(stream);
1399
1400 end:
1401 memset(&reply, 0, sizeof(reply));
1402 if (ret < 0) {
1403 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1404 } else {
1405 reply.ret_code = htobe32(LTTNG_OK);
1406 }
1407 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1408 sizeof(struct lttcomm_relayd_generic_reply), 0);
1409 if (send_ret < 0) {
1410 ERR("Relay sending reset metadata reply");
1411 ret = send_ret;
1412 }
1413
1414 end_no_session:
1415 return ret;
1416 }
1417
1418 /*
1419 * relay_unknown_command: send -1 if received unknown command
1420 */
1421 static void relay_unknown_command(struct relay_connection *conn)
1422 {
1423 struct lttcomm_relayd_generic_reply reply;
1424 int ret;
1425
1426 memset(&reply, 0, sizeof(reply));
1427 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1428 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1429 sizeof(struct lttcomm_relayd_generic_reply), 0);
1430 if (ret < 0) {
1431 ERR("Relay sending unknown command");
1432 }
1433 }
1434
1435 /*
1436 * relay_start: send an acknowledgment to the client to tell if we are
1437 * ready to receive data. We are ready if a session is established.
1438 */
1439 static int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1440 struct relay_connection *conn)
1441 {
1442 int ret = htobe32(LTTNG_OK);
1443 struct lttcomm_relayd_generic_reply reply;
1444 struct relay_session *session = conn->session;
1445
1446 if (!session) {
1447 DBG("Trying to start the streaming without a session established");
1448 ret = htobe32(LTTNG_ERR_UNK);
1449 }
1450
1451 memset(&reply, 0, sizeof(reply));
1452 reply.ret_code = ret;
1453 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1454 sizeof(struct lttcomm_relayd_generic_reply), 0);
1455 if (ret < 0) {
1456 ERR("Relay sending start ack");
1457 }
1458
1459 return ret;
1460 }
1461
1462 /*
1463 * Append padding to the file pointed by the file descriptor fd.
1464 */
1465 static int write_padding_to_file(int fd, uint32_t size)
1466 {
1467 ssize_t ret = 0;
1468 char *zeros;
1469
1470 if (size == 0) {
1471 goto end;
1472 }
1473
1474 zeros = zmalloc(size);
1475 if (zeros == NULL) {
1476 PERROR("zmalloc zeros for padding");
1477 ret = -1;
1478 goto end;
1479 }
1480
1481 ret = lttng_write(fd, zeros, size);
1482 if (ret < size) {
1483 PERROR("write padding to file");
1484 }
1485
1486 free(zeros);
1487
1488 end:
1489 return ret;
1490 }
1491
1492 /*
1493 * relay_recv_metadata: receive the metadata for the session.
1494 */
1495 static int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1496 struct relay_connection *conn)
1497 {
1498 int ret = 0;
1499 ssize_t size_ret;
1500 struct relay_session *session = conn->session;
1501 struct lttcomm_relayd_metadata_payload *metadata_struct;
1502 struct relay_stream *metadata_stream;
1503 uint64_t data_size, payload_size;
1504
1505 if (!session) {
1506 ERR("Metadata sent before version check");
1507 ret = -1;
1508 goto end;
1509 }
1510
1511 data_size = payload_size = be64toh(recv_hdr->data_size);
1512 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1513 ERR("Incorrect data size");
1514 ret = -1;
1515 goto end;
1516 }
1517 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1518
1519 if (data_buffer_size < data_size) {
1520 /* In case the realloc fails, we can free the memory */
1521 char *tmp_data_ptr;
1522
1523 tmp_data_ptr = realloc(data_buffer, data_size);
1524 if (!tmp_data_ptr) {
1525 ERR("Allocating data buffer");
1526 free(data_buffer);
1527 ret = -1;
1528 goto end;
1529 }
1530 data_buffer = tmp_data_ptr;
1531 data_buffer_size = data_size;
1532 }
1533 memset(data_buffer, 0, data_size);
1534 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1535 size_ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
1536 if (size_ret < 0 || size_ret != data_size) {
1537 if (size_ret == 0) {
1538 /* Orderly shutdown. Not necessary to print an error. */
1539 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1540 } else {
1541 ERR("Relay didn't receive the whole metadata");
1542 }
1543 ret = -1;
1544 goto end;
1545 }
1546 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1547
1548 metadata_stream = stream_get_by_id(be64toh(metadata_struct->stream_id));
1549 if (!metadata_stream) {
1550 ret = -1;
1551 goto end;
1552 }
1553
1554 pthread_mutex_lock(&metadata_stream->lock);
1555
1556 size_ret = lttng_write(metadata_stream->stream_fd->fd, metadata_struct->payload,
1557 payload_size);
1558 if (size_ret < payload_size) {
1559 ERR("Relay error writing metadata on file");
1560 ret = -1;
1561 goto end_put;
1562 }
1563
1564 size_ret = write_padding_to_file(metadata_stream->stream_fd->fd,
1565 be32toh(metadata_struct->padding_size));
1566 if (size_ret < 0) {
1567 goto end_put;
1568 }
1569
1570 metadata_stream->metadata_received +=
1571 payload_size + be32toh(metadata_struct->padding_size);
1572 DBG2("Relay metadata written. Updated metadata_received %" PRIu64,
1573 metadata_stream->metadata_received);
1574
1575 end_put:
1576 pthread_mutex_unlock(&metadata_stream->lock);
1577 stream_put(metadata_stream);
1578 end:
1579 return ret;
1580 }
1581
1582 /*
1583 * relay_send_version: send relayd version number
1584 */
1585 static int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1586 struct relay_connection *conn)
1587 {
1588 int ret;
1589 struct lttcomm_relayd_version reply, msg;
1590
1591 conn->version_check_done = 1;
1592
1593 /* Get version from the other side. */
1594 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1595 if (ret < 0 || ret != sizeof(msg)) {
1596 if (ret == 0) {
1597 /* Orderly shutdown. Not necessary to print an error. */
1598 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1599 } else {
1600 ERR("Relay failed to receive the version values.");
1601 }
1602 ret = -1;
1603 goto end;
1604 }
1605
1606 memset(&reply, 0, sizeof(reply));
1607 reply.major = RELAYD_VERSION_COMM_MAJOR;
1608 reply.minor = RELAYD_VERSION_COMM_MINOR;
1609
1610 /* Major versions must be the same */
1611 if (reply.major != be32toh(msg.major)) {
1612 DBG("Incompatible major versions (%u vs %u), deleting session",
1613 reply.major, be32toh(msg.major));
1614 connection_put(conn);
1615 ret = 0;
1616 goto end;
1617 }
1618
1619 conn->major = reply.major;
1620 /* We adapt to the lowest compatible version */
1621 if (reply.minor <= be32toh(msg.minor)) {
1622 conn->minor = reply.minor;
1623 } else {
1624 conn->minor = be32toh(msg.minor);
1625 }
1626
1627 reply.major = htobe32(reply.major);
1628 reply.minor = htobe32(reply.minor);
1629 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1630 sizeof(struct lttcomm_relayd_version), 0);
1631 if (ret < 0) {
1632 ERR("Relay sending version");
1633 }
1634
1635 DBG("Version check done using protocol %u.%u", conn->major,
1636 conn->minor);
1637
1638 end:
1639 return ret;
1640 }
1641
1642 /*
1643 * Check for data pending for a given stream id from the session daemon.
1644 */
1645 static int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1646 struct relay_connection *conn)
1647 {
1648 struct relay_session *session = conn->session;
1649 struct lttcomm_relayd_data_pending msg;
1650 struct lttcomm_relayd_generic_reply reply;
1651 struct relay_stream *stream;
1652 int ret;
1653 uint64_t last_net_seq_num, stream_id;
1654
1655 DBG("Data pending command received");
1656
1657 if (!session || conn->version_check_done == 0) {
1658 ERR("Trying to check for data before version check");
1659 ret = -1;
1660 goto end_no_session;
1661 }
1662
1663 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1664 if (ret < sizeof(msg)) {
1665 if (ret == 0) {
1666 /* Orderly shutdown. Not necessary to print an error. */
1667 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1668 } else {
1669 ERR("Relay didn't receive valid data_pending struct size : %d",
1670 ret);
1671 }
1672 ret = -1;
1673 goto end_no_session;
1674 }
1675
1676 stream_id = be64toh(msg.stream_id);
1677 last_net_seq_num = be64toh(msg.last_net_seq_num);
1678
1679 stream = stream_get_by_id(stream_id);
1680 if (stream == NULL) {
1681 ret = -1;
1682 goto end;
1683 }
1684
1685 pthread_mutex_lock(&stream->lock);
1686
1687 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1688 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1689 last_net_seq_num);
1690
1691 /* Avoid wrapping issue */
1692 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1693 /* Data has in fact been written and is NOT pending */
1694 ret = 0;
1695 } else {
1696 /* Data still being streamed thus pending */
1697 ret = 1;
1698 }
1699
1700 stream->data_pending_check_done = true;
1701 pthread_mutex_unlock(&stream->lock);
1702
1703 stream_put(stream);
1704 end:
1705
1706 memset(&reply, 0, sizeof(reply));
1707 reply.ret_code = htobe32(ret);
1708 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1709 if (ret < 0) {
1710 ERR("Relay data pending ret code failed");
1711 }
1712
1713 end_no_session:
1714 return ret;
1715 }
1716
1717 /*
1718 * Wait for the control socket to reach a quiescent state.
1719 *
1720 * Note that for now, when receiving this command from the session
1721 * daemon, this means that every subsequent commands or data received on
1722 * the control socket has been handled. So, this is why we simply return
1723 * OK here.
1724 */
1725 static int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1726 struct relay_connection *conn)
1727 {
1728 int ret;
1729 uint64_t stream_id;
1730 struct relay_stream *stream;
1731 struct lttcomm_relayd_quiescent_control msg;
1732 struct lttcomm_relayd_generic_reply reply;
1733
1734 DBG("Checking quiescent state on control socket");
1735
1736 if (!conn->session || conn->version_check_done == 0) {
1737 ERR("Trying to check for data before version check");
1738 ret = -1;
1739 goto end_no_session;
1740 }
1741
1742 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1743 if (ret < sizeof(msg)) {
1744 if (ret == 0) {
1745 /* Orderly shutdown. Not necessary to print an error. */
1746 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1747 } else {
1748 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1749 ret);
1750 }
1751 ret = -1;
1752 goto end_no_session;
1753 }
1754
1755 stream_id = be64toh(msg.stream_id);
1756 stream = stream_get_by_id(stream_id);
1757 if (!stream) {
1758 goto reply;
1759 }
1760 pthread_mutex_lock(&stream->lock);
1761 stream->data_pending_check_done = true;
1762 pthread_mutex_unlock(&stream->lock);
1763 DBG("Relay quiescent control pending flag set to %" PRIu64, stream_id);
1764 stream_put(stream);
1765 reply:
1766 memset(&reply, 0, sizeof(reply));
1767 reply.ret_code = htobe32(LTTNG_OK);
1768 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1769 if (ret < 0) {
1770 ERR("Relay data quiescent control ret code failed");
1771 }
1772
1773 end_no_session:
1774 return ret;
1775 }
1776
1777 /*
1778 * Initialize a data pending command. This means that a consumer is about
1779 * to ask for data pending for each stream it holds. Simply iterate over
1780 * all streams of a session and set the data_pending_check_done flag.
1781 *
1782 * This command returns to the client a LTTNG_OK code.
1783 */
1784 static int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1785 struct relay_connection *conn)
1786 {
1787 int ret;
1788 struct lttng_ht_iter iter;
1789 struct lttcomm_relayd_begin_data_pending msg;
1790 struct lttcomm_relayd_generic_reply reply;
1791 struct relay_stream *stream;
1792 uint64_t session_id;
1793
1794 assert(recv_hdr);
1795 assert(conn);
1796
1797 DBG("Init streams for data pending");
1798
1799 if (!conn->session || conn->version_check_done == 0) {
1800 ERR("Trying to check for data before version check");
1801 ret = -1;
1802 goto end_no_session;
1803 }
1804
1805 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1806 if (ret < sizeof(msg)) {
1807 if (ret == 0) {
1808 /* Orderly shutdown. Not necessary to print an error. */
1809 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1810 } else {
1811 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1812 ret);
1813 }
1814 ret = -1;
1815 goto end_no_session;
1816 }
1817
1818 session_id = be64toh(msg.session_id);
1819
1820 /*
1821 * Iterate over all streams to set the begin data pending flag.
1822 * For now, the streams are indexed by stream handle so we have
1823 * to iterate over all streams to find the one associated with
1824 * the right session_id.
1825 */
1826 rcu_read_lock();
1827 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1828 node.node) {
1829 if (!stream_get(stream)) {
1830 continue;
1831 }
1832 if (stream->trace->session->id == session_id) {
1833 pthread_mutex_lock(&stream->lock);
1834 stream->data_pending_check_done = false;
1835 pthread_mutex_unlock(&stream->lock);
1836 DBG("Set begin data pending flag to stream %" PRIu64,
1837 stream->stream_handle);
1838 }
1839 stream_put(stream);
1840 }
1841 rcu_read_unlock();
1842
1843 memset(&reply, 0, sizeof(reply));
1844 /* All good, send back reply. */
1845 reply.ret_code = htobe32(LTTNG_OK);
1846
1847 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1848 if (ret < 0) {
1849 ERR("Relay begin data pending send reply failed");
1850 }
1851
1852 end_no_session:
1853 return ret;
1854 }
1855
1856 /*
1857 * End data pending command. This will check, for a given session id, if
1858 * each stream associated with it has its data_pending_check_done flag
1859 * set. If not, this means that the client lost track of the stream but
1860 * the data is still being streamed on our side. In this case, we inform
1861 * the client that data is in flight.
1862 *
1863 * Return to the client if there is data in flight or not with a ret_code.
1864 */
1865 static int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1866 struct relay_connection *conn)
1867 {
1868 int ret;
1869 struct lttng_ht_iter iter;
1870 struct lttcomm_relayd_end_data_pending msg;
1871 struct lttcomm_relayd_generic_reply reply;
1872 struct relay_stream *stream;
1873 uint64_t session_id;
1874 uint32_t is_data_inflight = 0;
1875
1876 DBG("End data pending command");
1877
1878 if (!conn->session || conn->version_check_done == 0) {
1879 ERR("Trying to check for data before version check");
1880 ret = -1;
1881 goto end_no_session;
1882 }
1883
1884 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1885 if (ret < sizeof(msg)) {
1886 if (ret == 0) {
1887 /* Orderly shutdown. Not necessary to print an error. */
1888 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1889 } else {
1890 ERR("Relay didn't receive valid end data_pending struct size: %d",
1891 ret);
1892 }
1893 ret = -1;
1894 goto end_no_session;
1895 }
1896
1897 session_id = be64toh(msg.session_id);
1898
1899 /*
1900 * Iterate over all streams to see if the begin data pending
1901 * flag is set.
1902 */
1903 rcu_read_lock();
1904 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1905 node.node) {
1906 if (!stream_get(stream)) {
1907 continue;
1908 }
1909 if (stream->trace->session->id != session_id) {
1910 stream_put(stream);
1911 continue;
1912 }
1913 pthread_mutex_lock(&stream->lock);
1914 if (!stream->data_pending_check_done) {
1915 if (!stream->closed || !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
1916 is_data_inflight = 1;
1917 DBG("Data is still in flight for stream %" PRIu64,
1918 stream->stream_handle);
1919 pthread_mutex_unlock(&stream->lock);
1920 stream_put(stream);
1921 break;
1922 }
1923 }
1924 pthread_mutex_unlock(&stream->lock);
1925 stream_put(stream);
1926 }
1927 rcu_read_unlock();
1928
1929 memset(&reply, 0, sizeof(reply));
1930 /* All good, send back reply. */
1931 reply.ret_code = htobe32(is_data_inflight);
1932
1933 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1934 if (ret < 0) {
1935 ERR("Relay end data pending send reply failed");
1936 }
1937
1938 end_no_session:
1939 return ret;
1940 }
1941
1942 /*
1943 * Receive an index for a specific stream.
1944 *
1945 * Return 0 on success else a negative value.
1946 */
1947 static int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1948 struct relay_connection *conn)
1949 {
1950 int ret, send_ret;
1951 struct relay_session *session = conn->session;
1952 struct lttcomm_relayd_index index_info;
1953 struct relay_index *index;
1954 struct lttcomm_relayd_generic_reply reply;
1955 struct relay_stream *stream;
1956 uint64_t net_seq_num;
1957 size_t msg_len;
1958
1959 assert(conn);
1960
1961 DBG("Relay receiving index");
1962
1963 if (!session || conn->version_check_done == 0) {
1964 ERR("Trying to close a stream before version check");
1965 ret = -1;
1966 goto end_no_session;
1967 }
1968
1969 msg_len = lttcomm_relayd_index_len(
1970 lttng_to_index_major(conn->major, conn->minor),
1971 lttng_to_index_minor(conn->major, conn->minor));
1972 ret = conn->sock->ops->recvmsg(conn->sock, &index_info,
1973 msg_len, 0);
1974 if (ret < msg_len) {
1975 if (ret == 0) {
1976 /* Orderly shutdown. Not necessary to print an error. */
1977 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1978 } else {
1979 ERR("Relay didn't receive valid index struct size : %d", ret);
1980 }
1981 ret = -1;
1982 goto end_no_session;
1983 }
1984
1985 net_seq_num = be64toh(index_info.net_seq_num);
1986
1987 stream = stream_get_by_id(be64toh(index_info.relay_stream_id));
1988 if (!stream) {
1989 ERR("stream_get_by_id not found");
1990 ret = -1;
1991 goto end;
1992 }
1993 pthread_mutex_lock(&stream->lock);
1994
1995 /* Live beacon handling */
1996 if (index_info.packet_size == 0) {
1997 DBG("Received live beacon for stream %" PRIu64,
1998 stream->stream_handle);
1999
2000 /*
2001 * Only flag a stream inactive when it has already
2002 * received data and no indexes are in flight.
2003 */
2004 if (stream->index_received_seqcount > 0
2005 && stream->indexes_in_flight == 0) {
2006 stream->beacon_ts_end =
2007 be64toh(index_info.timestamp_end);
2008 }
2009 ret = 0;
2010 goto end_stream_put;
2011 } else {
2012 stream->beacon_ts_end = -1ULL;
2013 }
2014
2015 if (stream->ctf_stream_id == -1ULL) {
2016 stream->ctf_stream_id = be64toh(index_info.stream_id);
2017 }
2018 index = relay_index_get_by_id_or_create(stream, net_seq_num);
2019 if (!index) {
2020 ret = -1;
2021 ERR("relay_index_get_by_id_or_create index NULL");
2022 goto end_stream_put;
2023 }
2024 if (set_index_control_data(index, &index_info, conn)) {
2025 ERR("set_index_control_data error");
2026 relay_index_put(index);
2027 ret = -1;
2028 goto end_stream_put;
2029 }
2030 ret = relay_index_try_flush(index);
2031 if (ret == 0) {
2032 tracefile_array_commit_seq(stream->tfa);
2033 stream->index_received_seqcount++;
2034 } else if (ret > 0) {
2035 /* no flush. */
2036 ret = 0;
2037 } else {
2038 ERR("relay_index_try_flush error %d", ret);
2039 relay_index_put(index);
2040 ret = -1;
2041 }
2042
2043 end_stream_put:
2044 pthread_mutex_unlock(&stream->lock);
2045 stream_put(stream);
2046
2047 end:
2048
2049 memset(&reply, 0, sizeof(reply));
2050 if (ret < 0) {
2051 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2052 } else {
2053 reply.ret_code = htobe32(LTTNG_OK);
2054 }
2055 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2056 if (send_ret < 0) {
2057 ERR("Relay sending close index id reply");
2058 ret = send_ret;
2059 }
2060
2061 end_no_session:
2062 return ret;
2063 }
2064
2065 /*
2066 * Receive the streams_sent message.
2067 *
2068 * Return 0 on success else a negative value.
2069 */
2070 static int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
2071 struct relay_connection *conn)
2072 {
2073 int ret, send_ret;
2074 struct lttcomm_relayd_generic_reply reply;
2075
2076 assert(conn);
2077
2078 DBG("Relay receiving streams_sent");
2079
2080 if (!conn->session || conn->version_check_done == 0) {
2081 ERR("Trying to close a stream before version check");
2082 ret = -1;
2083 goto end_no_session;
2084 }
2085
2086 /*
2087 * Publish every pending stream in the connection recv list which are
2088 * now ready to be used by the viewer.
2089 */
2090 publish_connection_local_streams(conn);
2091
2092 memset(&reply, 0, sizeof(reply));
2093 reply.ret_code = htobe32(LTTNG_OK);
2094 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2095 if (send_ret < 0) {
2096 ERR("Relay sending sent_stream reply");
2097 ret = send_ret;
2098 } else {
2099 /* Success. */
2100 ret = 0;
2101 }
2102
2103 end_no_session:
2104 return ret;
2105 }
2106
2107 /*
2108 * Process the commands received on the control socket
2109 */
2110 static int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
2111 struct relay_connection *conn)
2112 {
2113 int ret = 0;
2114
2115 switch (be32toh(recv_hdr->cmd)) {
2116 case RELAYD_CREATE_SESSION:
2117 ret = relay_create_session(recv_hdr, conn);
2118 break;
2119 case RELAYD_ADD_STREAM:
2120 ret = relay_add_stream(recv_hdr, conn);
2121 break;
2122 case RELAYD_START_DATA:
2123 ret = relay_start(recv_hdr, conn);
2124 break;
2125 case RELAYD_SEND_METADATA:
2126 ret = relay_recv_metadata(recv_hdr, conn);
2127 break;
2128 case RELAYD_VERSION:
2129 ret = relay_send_version(recv_hdr, conn);
2130 break;
2131 case RELAYD_CLOSE_STREAM:
2132 ret = relay_close_stream(recv_hdr, conn);
2133 break;
2134 case RELAYD_DATA_PENDING:
2135 ret = relay_data_pending(recv_hdr, conn);
2136 break;
2137 case RELAYD_QUIESCENT_CONTROL:
2138 ret = relay_quiescent_control(recv_hdr, conn);
2139 break;
2140 case RELAYD_BEGIN_DATA_PENDING:
2141 ret = relay_begin_data_pending(recv_hdr, conn);
2142 break;
2143 case RELAYD_END_DATA_PENDING:
2144 ret = relay_end_data_pending(recv_hdr, conn);
2145 break;
2146 case RELAYD_SEND_INDEX:
2147 ret = relay_recv_index(recv_hdr, conn);
2148 break;
2149 case RELAYD_STREAMS_SENT:
2150 ret = relay_streams_sent(recv_hdr, conn);
2151 break;
2152 case RELAYD_RESET_METADATA:
2153 ret = relay_reset_metadata(recv_hdr, conn);
2154 break;
2155 case RELAYD_UPDATE_SYNC_INFO:
2156 default:
2157 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2158 relay_unknown_command(conn);
2159 ret = -1;
2160 goto end;
2161 }
2162
2163 end:
2164 return ret;
2165 }
2166
2167 /*
2168 * Handle index for a data stream.
2169 *
2170 * Called with the stream lock held.
2171 *
2172 * Return 0 on success else a negative value.
2173 */
2174 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2175 int rotate_index)
2176 {
2177 int ret = 0;
2178 uint64_t data_offset;
2179 struct relay_index *index;
2180
2181 /* Get data offset because we are about to update the index. */
2182 data_offset = htobe64(stream->tracefile_size_current);
2183
2184 DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64,
2185 stream->stream_handle, net_seq_num, stream->tracefile_size_current);
2186
2187 /*
2188 * Lookup for an existing index for that stream id/sequence
2189 * number. If it exists, the control thread has already received the
2190 * data for it, thus we need to write it to disk.
2191 */
2192 index = relay_index_get_by_id_or_create(stream, net_seq_num);
2193 if (!index) {
2194 ret = -1;
2195 goto end;
2196 }
2197
2198 if (rotate_index || !stream->index_file) {
2199 uint32_t major, minor;
2200
2201 /* Put ref on previous index_file. */
2202 if (stream->index_file) {
2203 lttng_index_file_put(stream->index_file);
2204 stream->index_file = NULL;
2205 }
2206 major = stream->trace->session->major;
2207 minor = stream->trace->session->minor;
2208 stream->index_file = lttng_index_file_create(stream->path_name,
2209 stream->channel_name,
2210 -1, -1, stream->tracefile_size,
2211 tracefile_array_get_file_index_head(stream->tfa),
2212 lttng_to_index_major(major, minor),
2213 lttng_to_index_minor(major, minor));
2214 if (!stream->index_file) {
2215 ret = -1;
2216 /* Put self-ref for this index due to error. */
2217 relay_index_put(index);
2218 index = NULL;
2219 goto end;
2220 }
2221 }
2222
2223 if (relay_index_set_file(index, stream->index_file, data_offset)) {
2224 ret = -1;
2225 /* Put self-ref for this index due to error. */
2226 relay_index_put(index);
2227 index = NULL;
2228 goto end;
2229 }
2230
2231 ret = relay_index_try_flush(index);
2232 if (ret == 0) {
2233 tracefile_array_commit_seq(stream->tfa);
2234 stream->index_received_seqcount++;
2235 } else if (ret > 0) {
2236 /* No flush. */
2237 ret = 0;
2238 } else {
2239 /* Put self-ref for this index due to error. */
2240 relay_index_put(index);
2241 index = NULL;
2242 ret = -1;
2243 }
2244 end:
2245 return ret;
2246 }
2247
2248 /*
2249 * relay_process_data: Process the data received on the data socket
2250 */
2251 static int relay_process_data(struct relay_connection *conn)
2252 {
2253 int ret = 0, rotate_index = 0;
2254 ssize_t size_ret;
2255 struct relay_stream *stream;
2256 struct lttcomm_relayd_data_hdr data_hdr;
2257 uint64_t stream_id;
2258 uint64_t net_seq_num;
2259 uint32_t data_size;
2260 struct relay_session *session;
2261 bool new_stream = false, close_requested = false;
2262 size_t chunk_size = RECV_DATA_BUFFER_SIZE;
2263 size_t recv_off = 0;
2264 char data_buffer[chunk_size];
2265
2266 ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr,
2267 sizeof(struct lttcomm_relayd_data_hdr), 0);
2268 if (ret <= 0) {
2269 if (ret == 0) {
2270 /* Orderly shutdown. Not necessary to print an error. */
2271 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2272 } else {
2273 ERR("Unable to receive data header on sock %d", conn->sock->fd);
2274 }
2275 ret = -1;
2276 goto end;
2277 }
2278
2279 stream_id = be64toh(data_hdr.stream_id);
2280 stream = stream_get_by_id(stream_id);
2281 if (!stream) {
2282 ERR("relay_process_data: Cannot find stream %" PRIu64, stream_id);
2283 ret = -1;
2284 goto end;
2285 }
2286 session = stream->trace->session;
2287 data_size = be32toh(data_hdr.data_size);
2288
2289 net_seq_num = be64toh(data_hdr.net_seq_num);
2290
2291 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2292 data_size, stream_id, net_seq_num);
2293
2294 pthread_mutex_lock(&stream->lock);
2295
2296 /* Check if a rotation is needed. */
2297 if (stream->tracefile_size > 0 &&
2298 (stream->tracefile_size_current + data_size) >
2299 stream->tracefile_size) {
2300 uint64_t old_id, new_id;
2301
2302 old_id = tracefile_array_get_file_index_head(stream->tfa);
2303 tracefile_array_file_rotate(stream->tfa);
2304
2305 /* new_id is updated by utils_rotate_stream_file. */
2306 new_id = old_id;
2307
2308 ret = utils_rotate_stream_file(stream->path_name,
2309 stream->channel_name, stream->tracefile_size,
2310 stream->tracefile_count, -1,
2311 -1, stream->stream_fd->fd,
2312 &new_id, &stream->stream_fd->fd);
2313 if (ret < 0) {
2314 ERR("Rotating stream output file");
2315 goto end_stream_unlock;
2316 }
2317 /*
2318 * Reset current size because we just performed a stream
2319 * rotation.
2320 */
2321 stream->tracefile_size_current = 0;
2322 rotate_index = 1;
2323 }
2324
2325 /*
2326 * Index are handled in protocol version 2.4 and above. Also,
2327 * snapshot and index are NOT supported.
2328 */
2329 if (session->minor >= 4 && !session->snapshot) {
2330 ret = handle_index_data(stream, net_seq_num, rotate_index);
2331 if (ret < 0) {
2332 ERR("handle_index_data: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
2333 stream->stream_handle, net_seq_num, ret);
2334 goto end_stream_unlock;
2335 }
2336 }
2337
2338 for (recv_off = 0; recv_off < data_size; recv_off += chunk_size) {
2339 size_t recv_size = min(data_size - recv_off, chunk_size);
2340
2341 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_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", conn->sock->fd);
2346 } else {
2347 ERR("Socket %d error %d", conn->sock->fd, ret);
2348 }
2349 ret = -1;
2350 goto end_stream_unlock;
2351 }
2352
2353 /* Write data to stream output fd. */
2354 size_ret = lttng_write(stream->stream_fd->fd, data_buffer,
2355 recv_size);
2356 if (size_ret < recv_size) {
2357 ERR("Relay error writing data to file");
2358 ret = -1;
2359 goto end_stream_unlock;
2360 }
2361
2362 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64,
2363 size_ret, stream->stream_handle);
2364 }
2365
2366 ret = write_padding_to_file(stream->stream_fd->fd,
2367 be32toh(data_hdr.padding_size));
2368 if (ret < 0) {
2369 ERR("write_padding_to_file: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
2370 stream->stream_handle, net_seq_num, ret);
2371 goto end_stream_unlock;
2372 }
2373 stream->tracefile_size_current +=
2374 data_size + be32toh(data_hdr.padding_size);
2375 if (stream->prev_seq == -1ULL) {
2376 new_stream = true;
2377 }
2378
2379 stream->prev_seq = net_seq_num;
2380
2381 end_stream_unlock:
2382 close_requested = stream->close_requested;
2383 pthread_mutex_unlock(&stream->lock);
2384 if (close_requested) {
2385 try_stream_close(stream);
2386 }
2387
2388 if (new_stream) {
2389 pthread_mutex_lock(&session->lock);
2390 uatomic_set(&session->new_streams, 1);
2391 pthread_mutex_unlock(&session->lock);
2392 }
2393 stream_put(stream);
2394 end:
2395 return ret;
2396 }
2397
2398 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2399 {
2400 int ret;
2401
2402 (void) lttng_poll_del(events, pollfd);
2403
2404 ret = close(pollfd);
2405 if (ret < 0) {
2406 ERR("Closing pollfd %d", pollfd);
2407 }
2408 }
2409
2410 static void relay_thread_close_connection(struct lttng_poll_event *events,
2411 int pollfd, struct relay_connection *conn)
2412 {
2413 const char *type_str;
2414
2415 switch (conn->type) {
2416 case RELAY_DATA:
2417 type_str = "Data";
2418 break;
2419 case RELAY_CONTROL:
2420 type_str = "Control";
2421 break;
2422 case RELAY_VIEWER_COMMAND:
2423 type_str = "Viewer Command";
2424 break;
2425 case RELAY_VIEWER_NOTIFICATION:
2426 type_str = "Viewer Notification";
2427 break;
2428 default:
2429 type_str = "Unknown";
2430 }
2431 cleanup_connection_pollfd(events, pollfd);
2432 connection_put(conn);
2433 DBG("%s connection closed with %d", type_str, pollfd);
2434 }
2435
2436 /*
2437 * This thread does the actual work
2438 */
2439 static void *relay_thread_worker(void *data)
2440 {
2441 int ret, err = -1, last_seen_data_fd = -1;
2442 uint32_t nb_fd;
2443 struct lttng_poll_event events;
2444 struct lttng_ht *relay_connections_ht;
2445 struct lttng_ht_iter iter;
2446 struct lttcomm_relayd_hdr recv_hdr;
2447 struct relay_connection *destroy_conn = NULL;
2448
2449 DBG("[thread] Relay worker started");
2450
2451 rcu_register_thread();
2452
2453 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2454
2455 if (testpoint(relayd_thread_worker)) {
2456 goto error_testpoint;
2457 }
2458
2459 health_code_update();
2460
2461 /* table of connections indexed on socket */
2462 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2463 if (!relay_connections_ht) {
2464 goto relay_connections_ht_error;
2465 }
2466
2467 ret = create_thread_poll_set(&events, 2);
2468 if (ret < 0) {
2469 goto error_poll_create;
2470 }
2471
2472 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2473 if (ret < 0) {
2474 goto error;
2475 }
2476
2477 restart:
2478 while (1) {
2479 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2480
2481 health_code_update();
2482
2483 /* Infinite blocking call, waiting for transmission */
2484 DBG3("Relayd worker thread polling...");
2485 health_poll_entry();
2486 ret = lttng_poll_wait(&events, -1);
2487 health_poll_exit();
2488 if (ret < 0) {
2489 /*
2490 * Restart interrupted system call.
2491 */
2492 if (errno == EINTR) {
2493 goto restart;
2494 }
2495 goto error;
2496 }
2497
2498 nb_fd = ret;
2499
2500 /*
2501 * Process control. The control connection is
2502 * prioritized so we don't starve it with high
2503 * throughput tracing data on the data connection.
2504 */
2505 for (i = 0; i < nb_fd; i++) {
2506 /* Fetch once the poll data */
2507 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2508 int pollfd = LTTNG_POLL_GETFD(&events, i);
2509
2510 health_code_update();
2511
2512 if (!revents) {
2513 /*
2514 * No activity for this FD (poll
2515 * implementation).
2516 */
2517 continue;
2518 }
2519
2520 /* Thread quit pipe has been closed. Killing thread. */
2521 ret = check_thread_quit_pipe(pollfd, revents);
2522 if (ret) {
2523 err = 0;
2524 goto exit;
2525 }
2526
2527 /* Inspect the relay conn pipe for new connection */
2528 if (pollfd == relay_conn_pipe[0]) {
2529 if (revents & LPOLLIN) {
2530 struct relay_connection *conn;
2531
2532 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
2533 if (ret < 0) {
2534 goto error;
2535 }
2536 lttng_poll_add(&events, conn->sock->fd,
2537 LPOLLIN | LPOLLRDHUP);
2538 connection_ht_add(relay_connections_ht, conn);
2539 DBG("Connection socket %d added", conn->sock->fd);
2540 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2541 ERR("Relay connection pipe error");
2542 goto error;
2543 } else {
2544 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2545 goto error;
2546 }
2547 } else {
2548 struct relay_connection *ctrl_conn;
2549
2550 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
2551 /* If not found, there is a synchronization issue. */
2552 assert(ctrl_conn);
2553
2554 if (ctrl_conn->type == RELAY_DATA) {
2555 if (revents & LPOLLIN) {
2556 /*
2557 * Flag the last seen data fd not deleted. It will be
2558 * used as the last seen fd if any fd gets deleted in
2559 * this first loop.
2560 */
2561 last_notdel_data_fd = pollfd;
2562 }
2563 goto put_ctrl_connection;
2564 }
2565 assert(ctrl_conn->type == RELAY_CONTROL);
2566
2567 if (revents & LPOLLIN) {
2568 ret = ctrl_conn->sock->ops->recvmsg(ctrl_conn->sock,
2569 &recv_hdr, sizeof(recv_hdr), 0);
2570 if (ret <= 0) {
2571 /* Connection closed */
2572 relay_thread_close_connection(&events, pollfd,
2573 ctrl_conn);
2574 } else {
2575 ret = relay_process_control(&recv_hdr, ctrl_conn);
2576 if (ret < 0) {
2577 /* Clear the session on error. */
2578 relay_thread_close_connection(&events,
2579 pollfd, ctrl_conn);
2580 }
2581 seen_control = 1;
2582 }
2583 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2584 relay_thread_close_connection(&events,
2585 pollfd, ctrl_conn);
2586 if (last_seen_data_fd == pollfd) {
2587 last_seen_data_fd = last_notdel_data_fd;
2588 }
2589 } else {
2590 ERR("Unexpected poll events %u for control sock %d",
2591 revents, pollfd);
2592 connection_put(ctrl_conn);
2593 goto error;
2594 }
2595 put_ctrl_connection:
2596 connection_put(ctrl_conn);
2597 }
2598 }
2599
2600 /*
2601 * The last loop handled a control request, go back to poll to make
2602 * sure we prioritise the control socket.
2603 */
2604 if (seen_control) {
2605 continue;
2606 }
2607
2608 if (last_seen_data_fd >= 0) {
2609 for (i = 0; i < nb_fd; i++) {
2610 int pollfd = LTTNG_POLL_GETFD(&events, i);
2611
2612 health_code_update();
2613
2614 if (last_seen_data_fd == pollfd) {
2615 idx = i;
2616 break;
2617 }
2618 }
2619 }
2620
2621 /* Process data connection. */
2622 for (i = idx + 1; i < nb_fd; i++) {
2623 /* Fetch the poll data. */
2624 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2625 int pollfd = LTTNG_POLL_GETFD(&events, i);
2626 struct relay_connection *data_conn;
2627
2628 health_code_update();
2629
2630 if (!revents) {
2631 /* No activity for this FD (poll implementation). */
2632 continue;
2633 }
2634
2635 /* Skip the command pipe. It's handled in the first loop. */
2636 if (pollfd == relay_conn_pipe[0]) {
2637 continue;
2638 }
2639
2640 data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
2641 if (!data_conn) {
2642 /* Skip it. Might be removed before. */
2643 continue;
2644 }
2645 if (data_conn->type == RELAY_CONTROL) {
2646 goto put_data_connection;
2647 }
2648 assert(data_conn->type == RELAY_DATA);
2649
2650 if (revents & LPOLLIN) {
2651 ret = relay_process_data(data_conn);
2652 /* Connection closed */
2653 if (ret < 0) {
2654 relay_thread_close_connection(&events, pollfd,
2655 data_conn);
2656 /*
2657 * Every goto restart call sets the last seen fd where
2658 * here we don't really care since we gracefully
2659 * continue the loop after the connection is deleted.
2660 */
2661 } else {
2662 /* Keep last seen port. */
2663 last_seen_data_fd = pollfd;
2664 connection_put(data_conn);
2665 goto restart;
2666 }
2667 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2668 relay_thread_close_connection(&events, pollfd,
2669 data_conn);
2670 } else {
2671 ERR("Unknown poll events %u for data sock %d",
2672 revents, pollfd);
2673 }
2674 put_data_connection:
2675 connection_put(data_conn);
2676 }
2677 last_seen_data_fd = -1;
2678 }
2679
2680 /* Normal exit, no error */
2681 ret = 0;
2682
2683 exit:
2684 error:
2685 /* Cleanup reamaining connection object. */
2686 rcu_read_lock();
2687 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
2688 destroy_conn,
2689 sock_n.node) {
2690 health_code_update();
2691
2692 if (session_abort(destroy_conn->session)) {
2693 assert(0);
2694 }
2695
2696 /*
2697 * No need to grab another ref, because we own
2698 * destroy_conn.
2699 */
2700 relay_thread_close_connection(&events, destroy_conn->sock->fd,
2701 destroy_conn);
2702 }
2703 rcu_read_unlock();
2704
2705 lttng_poll_clean(&events);
2706 error_poll_create:
2707 lttng_ht_destroy(relay_connections_ht);
2708 relay_connections_ht_error:
2709 /* Close relay conn pipes */
2710 utils_close_pipe(relay_conn_pipe);
2711 if (err) {
2712 DBG("Thread exited with error");
2713 }
2714 DBG("Worker thread cleanup complete");
2715 error_testpoint:
2716 if (err) {
2717 health_error();
2718 ERR("Health error occurred in %s", __func__);
2719 }
2720 health_unregister(health_relayd);
2721 rcu_unregister_thread();
2722 lttng_relay_stop_threads();
2723 return NULL;
2724 }
2725
2726 /*
2727 * Create the relay command pipe to wake thread_manage_apps.
2728 * Closed in cleanup().
2729 */
2730 static int create_relay_conn_pipe(void)
2731 {
2732 int ret;
2733
2734 ret = utils_create_pipe_cloexec(relay_conn_pipe);
2735
2736 return ret;
2737 }
2738
2739 /*
2740 * main
2741 */
2742 int main(int argc, char **argv)
2743 {
2744 int ret = 0, retval = 0;
2745 void *status;
2746
2747 /* Parse arguments */
2748 progname = argv[0];
2749 if (set_options(argc, argv)) {
2750 retval = -1;
2751 goto exit_options;
2752 }
2753
2754 if (set_signal_handler()) {
2755 retval = -1;
2756 goto exit_options;
2757 }
2758
2759 /* Try to create directory if -o, --output is specified. */
2760 if (opt_output_path) {
2761 if (*opt_output_path != '/') {
2762 ERR("Please specify an absolute path for -o, --output PATH");
2763 retval = -1;
2764 goto exit_options;
2765 }
2766
2767 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG,
2768 -1, -1);
2769 if (ret < 0) {
2770 ERR("Unable to create %s", opt_output_path);
2771 retval = -1;
2772 goto exit_options;
2773 }
2774 }
2775
2776 /* Daemonize */
2777 if (opt_daemon || opt_background) {
2778 int i;
2779
2780 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2781 !opt_background);
2782 if (ret < 0) {
2783 retval = -1;
2784 goto exit_options;
2785 }
2786
2787 /*
2788 * We are in the child. Make sure all other file
2789 * descriptors are closed, in case we are called with
2790 * more opened file descriptors than the standard ones.
2791 */
2792 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2793 (void) close(i);
2794 }
2795 }
2796
2797 /* Initialize thread health monitoring */
2798 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2799 if (!health_relayd) {
2800 PERROR("health_app_create error");
2801 retval = -1;
2802 goto exit_health_app_create;
2803 }
2804
2805 /* Create thread quit pipe */
2806 if (init_thread_quit_pipe()) {
2807 retval = -1;
2808 goto exit_init_data;
2809 }
2810
2811 /* Setup the thread apps communication pipe. */
2812 if (create_relay_conn_pipe()) {
2813 retval = -1;
2814 goto exit_init_data;
2815 }
2816
2817 /* Init relay command queue. */
2818 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
2819
2820 /* Initialize communication library */
2821 lttcomm_init();
2822 lttcomm_inet_init();
2823
2824 /* tables of sessions indexed by session ID */
2825 sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2826 if (!sessions_ht) {
2827 retval = -1;
2828 goto exit_init_data;
2829 }
2830
2831 /* tables of streams indexed by stream ID */
2832 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2833 if (!relay_streams_ht) {
2834 retval = -1;
2835 goto exit_init_data;
2836 }
2837
2838 /* tables of streams indexed by stream ID */
2839 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2840 if (!viewer_streams_ht) {
2841 retval = -1;
2842 goto exit_init_data;
2843 }
2844
2845 ret = utils_create_pipe(health_quit_pipe);
2846 if (ret) {
2847 retval = -1;
2848 goto exit_health_quit_pipe;
2849 }
2850
2851 /* Create thread to manage the client socket */
2852 ret = pthread_create(&health_thread, default_pthread_attr(),
2853 thread_manage_health, (void *) NULL);
2854 if (ret) {
2855 errno = ret;
2856 PERROR("pthread_create health");
2857 retval = -1;
2858 goto exit_health_thread;
2859 }
2860
2861 /* Setup the dispatcher thread */
2862 ret = pthread_create(&dispatcher_thread, default_pthread_attr(),
2863 relay_thread_dispatcher, (void *) NULL);
2864 if (ret) {
2865 errno = ret;
2866 PERROR("pthread_create dispatcher");
2867 retval = -1;
2868 goto exit_dispatcher_thread;
2869 }
2870
2871 /* Setup the worker thread */
2872 ret = pthread_create(&worker_thread, default_pthread_attr(),
2873 relay_thread_worker, NULL);
2874 if (ret) {
2875 errno = ret;
2876 PERROR("pthread_create worker");
2877 retval = -1;
2878 goto exit_worker_thread;
2879 }
2880
2881 /* Setup the listener thread */
2882 ret = pthread_create(&listener_thread, default_pthread_attr(),
2883 relay_thread_listener, (void *) NULL);
2884 if (ret) {
2885 errno = ret;
2886 PERROR("pthread_create listener");
2887 retval = -1;
2888 goto exit_listener_thread;
2889 }
2890
2891 ret = relayd_live_create(live_uri);
2892 if (ret) {
2893 ERR("Starting live viewer threads");
2894 retval = -1;
2895 goto exit_live;
2896 }
2897
2898 /*
2899 * This is where we start awaiting program completion (e.g. through
2900 * signal that asks threads to teardown).
2901 */
2902
2903 ret = relayd_live_join();
2904 if (ret) {
2905 retval = -1;
2906 }
2907 exit_live:
2908
2909 ret = pthread_join(listener_thread, &status);
2910 if (ret) {
2911 errno = ret;
2912 PERROR("pthread_join listener_thread");
2913 retval = -1;
2914 }
2915
2916 exit_listener_thread:
2917 ret = pthread_join(worker_thread, &status);
2918 if (ret) {
2919 errno = ret;
2920 PERROR("pthread_join worker_thread");
2921 retval = -1;
2922 }
2923
2924 exit_worker_thread:
2925 ret = pthread_join(dispatcher_thread, &status);
2926 if (ret) {
2927 errno = ret;
2928 PERROR("pthread_join dispatcher_thread");
2929 retval = -1;
2930 }
2931 exit_dispatcher_thread:
2932
2933 ret = pthread_join(health_thread, &status);
2934 if (ret) {
2935 errno = ret;
2936 PERROR("pthread_join health_thread");
2937 retval = -1;
2938 }
2939 exit_health_thread:
2940
2941 utils_close_pipe(health_quit_pipe);
2942 exit_health_quit_pipe:
2943
2944 exit_init_data:
2945 health_app_destroy(health_relayd);
2946 exit_health_app_create:
2947 exit_options:
2948 /*
2949 * Wait for all pending call_rcu work to complete before tearing
2950 * down data structures. call_rcu worker may be trying to
2951 * perform lookups in those structures.
2952 */
2953 rcu_barrier();
2954 relayd_cleanup();
2955
2956 /* Ensure all prior call_rcu are done. */
2957 rcu_barrier();
2958
2959 if (!retval) {
2960 exit(EXIT_SUCCESS);
2961 } else {
2962 exit(EXIT_FAILURE);
2963 }
2964 }
This page took 0.163132 seconds and 3 git commands to generate.