relay: use urcu_ref_get_unless_zero
[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 for (;;) {
981 health_code_update();
982
983 /* Atomically prepare the queue futex */
984 futex_nto1_prepare(&relay_conn_queue.futex);
985
986 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
987 break;
988 }
989
990 do {
991 health_code_update();
992
993 /* Dequeue commands */
994 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
995 &relay_conn_queue.tail);
996 if (node == NULL) {
997 DBG("Woken up but nothing in the relay command queue");
998 /* Continue thread execution */
999 break;
1000 }
1001 new_conn = caa_container_of(node, struct relay_connection, qnode);
1002
1003 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1004
1005 /*
1006 * Inform worker thread of the new request. This
1007 * call is blocking so we can be assured that
1008 * the data will be read at some point in time
1009 * or wait to the end of the world :)
1010 */
1011 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1012 if (ret < 0) {
1013 PERROR("write connection pipe");
1014 connection_put(new_conn);
1015 goto error;
1016 }
1017 } while (node != NULL);
1018
1019 /* Futex wait on queue. Blocking call on futex() */
1020 health_poll_entry();
1021 futex_nto1_wait(&relay_conn_queue.futex);
1022 health_poll_exit();
1023 }
1024
1025 /* Normal exit, no error */
1026 err = 0;
1027
1028 error:
1029 error_testpoint:
1030 if (err) {
1031 health_error();
1032 ERR("Health error occurred in %s", __func__);
1033 }
1034 health_unregister(health_relayd);
1035 DBG("Dispatch thread dying");
1036 lttng_relay_stop_threads();
1037 return NULL;
1038 }
1039
1040 /*
1041 * Set index data from the control port to a given index object.
1042 */
1043 static int set_index_control_data(struct relay_index *index,
1044 struct lttcomm_relayd_index *data,
1045 struct relay_connection *conn)
1046 {
1047 struct ctf_packet_index index_data;
1048
1049 /*
1050 * The index on disk is encoded in big endian, so we don't need
1051 * to convert the data received on the network. The data_offset
1052 * value is NEVER modified here and is updated by the data
1053 * thread.
1054 */
1055 index_data.packet_size = data->packet_size;
1056 index_data.content_size = data->content_size;
1057 index_data.timestamp_begin = data->timestamp_begin;
1058 index_data.timestamp_end = data->timestamp_end;
1059 index_data.events_discarded = data->events_discarded;
1060 index_data.stream_id = data->stream_id;
1061
1062 if (conn->minor >= 8) {
1063 index->index_data.stream_instance_id = data->stream_instance_id;
1064 index->index_data.packet_seq_num = data->packet_seq_num;
1065 }
1066
1067 return relay_index_set_data(index, &index_data);
1068 }
1069
1070 /*
1071 * Handle the RELAYD_CREATE_SESSION command.
1072 *
1073 * On success, send back the session id or else return a negative value.
1074 */
1075 static int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
1076 struct relay_connection *conn)
1077 {
1078 int ret = 0, send_ret;
1079 struct relay_session *session;
1080 struct lttcomm_relayd_status_session reply;
1081 char session_name[LTTNG_NAME_MAX];
1082 char hostname[LTTNG_HOST_NAME_MAX];
1083 uint32_t live_timer = 0;
1084 bool snapshot = false;
1085
1086 memset(session_name, 0, LTTNG_NAME_MAX);
1087 memset(hostname, 0, LTTNG_HOST_NAME_MAX);
1088
1089 memset(&reply, 0, sizeof(reply));
1090
1091 switch (conn->minor) {
1092 case 1:
1093 case 2:
1094 case 3:
1095 break;
1096 case 4: /* LTTng sessiond 2.4 */
1097 default:
1098 ret = cmd_create_session_2_4(conn, session_name,
1099 hostname, &live_timer, &snapshot);
1100 }
1101 if (ret < 0) {
1102 goto send_reply;
1103 }
1104
1105 session = session_create(session_name, hostname, live_timer,
1106 snapshot, conn->major, conn->minor);
1107 if (!session) {
1108 ret = -1;
1109 goto send_reply;
1110 }
1111 assert(!conn->session);
1112 conn->session = session;
1113 DBG("Created session %" PRIu64, session->id);
1114
1115 reply.session_id = htobe64(session->id);
1116
1117 send_reply:
1118 if (ret < 0) {
1119 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1120 } else {
1121 reply.ret_code = htobe32(LTTNG_OK);
1122 }
1123
1124 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1125 if (send_ret < 0) {
1126 ERR("Relayd sending session id");
1127 ret = send_ret;
1128 }
1129
1130 return ret;
1131 }
1132
1133 /*
1134 * When we have received all the streams and the metadata for a channel,
1135 * we make them visible to the viewer threads.
1136 */
1137 static void publish_connection_local_streams(struct relay_connection *conn)
1138 {
1139 struct relay_stream *stream;
1140 struct relay_session *session = conn->session;
1141
1142 /*
1143 * We publish all streams belonging to a session atomically wrt
1144 * session lock.
1145 */
1146 pthread_mutex_lock(&session->lock);
1147 rcu_read_lock();
1148 cds_list_for_each_entry_rcu(stream, &session->recv_list,
1149 recv_node) {
1150 stream_publish(stream);
1151 }
1152 rcu_read_unlock();
1153
1154 /*
1155 * Inform the viewer that there are new streams in the session.
1156 */
1157 if (session->viewer_attached) {
1158 uatomic_set(&session->new_streams, 1);
1159 }
1160 pthread_mutex_unlock(&session->lock);
1161 }
1162
1163 /*
1164 * relay_add_stream: allocate a new stream for a session
1165 */
1166 static int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1167 struct relay_connection *conn)
1168 {
1169 int ret;
1170 ssize_t send_ret;
1171 struct relay_session *session = conn->session;
1172 struct relay_stream *stream = NULL;
1173 struct lttcomm_relayd_status_stream reply;
1174 struct ctf_trace *trace = NULL;
1175 uint64_t stream_handle = -1ULL;
1176 char *path_name = NULL, *channel_name = NULL;
1177 uint64_t tracefile_size = 0, tracefile_count = 0;
1178
1179 if (!session || conn->version_check_done == 0) {
1180 ERR("Trying to add a stream before version check");
1181 ret = -1;
1182 goto end_no_session;
1183 }
1184
1185 switch (session->minor) {
1186 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
1187 ret = cmd_recv_stream_2_1(conn, &path_name,
1188 &channel_name);
1189 break;
1190 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
1191 default:
1192 ret = cmd_recv_stream_2_2(conn, &path_name,
1193 &channel_name, &tracefile_size, &tracefile_count);
1194 break;
1195 }
1196 if (ret < 0) {
1197 goto send_reply;
1198 }
1199
1200 trace = ctf_trace_get_by_path_or_create(session, path_name);
1201 if (!trace) {
1202 goto send_reply;
1203 }
1204 /* This stream here has one reference on the trace. */
1205
1206 pthread_mutex_lock(&last_relay_stream_id_lock);
1207 stream_handle = ++last_relay_stream_id;
1208 pthread_mutex_unlock(&last_relay_stream_id_lock);
1209
1210 /* We pass ownership of path_name and channel_name. */
1211 stream = stream_create(trace, stream_handle, path_name,
1212 channel_name, tracefile_size, tracefile_count);
1213 path_name = NULL;
1214 channel_name = NULL;
1215
1216 /*
1217 * Streams are the owners of their trace. Reference to trace is
1218 * kept within stream_create().
1219 */
1220 ctf_trace_put(trace);
1221
1222 send_reply:
1223 memset(&reply, 0, sizeof(reply));
1224 reply.handle = htobe64(stream_handle);
1225 if (!stream) {
1226 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1227 } else {
1228 reply.ret_code = htobe32(LTTNG_OK);
1229 }
1230
1231 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1232 sizeof(struct lttcomm_relayd_status_stream), 0);
1233 if (send_ret < 0) {
1234 ERR("Relay sending stream id");
1235 ret = (int) send_ret;
1236 }
1237
1238 end_no_session:
1239 free(path_name);
1240 free(channel_name);
1241 return ret;
1242 }
1243
1244 /*
1245 * relay_close_stream: close a specific stream
1246 */
1247 static int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1248 struct relay_connection *conn)
1249 {
1250 int ret, send_ret;
1251 struct relay_session *session = conn->session;
1252 struct lttcomm_relayd_close_stream stream_info;
1253 struct lttcomm_relayd_generic_reply reply;
1254 struct relay_stream *stream;
1255
1256 DBG("Close stream received");
1257
1258 if (!session || conn->version_check_done == 0) {
1259 ERR("Trying to close a stream before version check");
1260 ret = -1;
1261 goto end_no_session;
1262 }
1263
1264 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1265 sizeof(struct lttcomm_relayd_close_stream), 0);
1266 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1267 if (ret == 0) {
1268 /* Orderly shutdown. Not necessary to print an error. */
1269 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1270 } else {
1271 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1272 }
1273 ret = -1;
1274 goto end_no_session;
1275 }
1276
1277 stream = stream_get_by_id(be64toh(stream_info.stream_id));
1278 if (!stream) {
1279 ret = -1;
1280 goto end;
1281 }
1282
1283 /*
1284 * Set last_net_seq_num before the close flag. Required by data
1285 * pending check.
1286 */
1287 pthread_mutex_lock(&stream->lock);
1288 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1289 pthread_mutex_unlock(&stream->lock);
1290
1291 /*
1292 * This is one of the conditions which may trigger a stream close
1293 * with the others being:
1294 * 1) A close command is received for a stream
1295 * 2) The control connection owning the stream is closed
1296 * 3) We have received all of the stream's data _after_ a close
1297 * request.
1298 */
1299 try_stream_close(stream);
1300 if (stream->is_metadata) {
1301 struct relay_viewer_stream *vstream;
1302
1303 vstream = viewer_stream_get_by_id(stream->stream_handle);
1304 if (vstream) {
1305 if (vstream->metadata_sent == stream->metadata_received) {
1306 /*
1307 * Since all the metadata has been sent to the
1308 * viewer and that we have a request to close
1309 * its stream, we can safely teardown the
1310 * corresponding metadata viewer stream.
1311 */
1312 viewer_stream_put(vstream);
1313 }
1314 /* Put local reference. */
1315 viewer_stream_put(vstream);
1316 }
1317 }
1318 stream_put(stream);
1319
1320 end:
1321 memset(&reply, 0, sizeof(reply));
1322 if (ret < 0) {
1323 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1324 } else {
1325 reply.ret_code = htobe32(LTTNG_OK);
1326 }
1327 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1328 sizeof(struct lttcomm_relayd_generic_reply), 0);
1329 if (send_ret < 0) {
1330 ERR("Relay sending stream id");
1331 ret = send_ret;
1332 }
1333
1334 end_no_session:
1335 return ret;
1336 }
1337
1338 /*
1339 * relay_reset_metadata: reset a metadata stream
1340 */
1341 static
1342 int relay_reset_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1343 struct relay_connection *conn)
1344 {
1345 int ret, send_ret;
1346 struct relay_session *session = conn->session;
1347 struct lttcomm_relayd_reset_metadata stream_info;
1348 struct lttcomm_relayd_generic_reply reply;
1349 struct relay_stream *stream;
1350
1351 DBG("Reset metadata received");
1352
1353 if (!session || conn->version_check_done == 0) {
1354 ERR("Trying to reset a metadata stream before version check");
1355 ret = -1;
1356 goto end_no_session;
1357 }
1358
1359 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1360 sizeof(struct lttcomm_relayd_reset_metadata), 0);
1361 if (ret < sizeof(struct lttcomm_relayd_reset_metadata)) {
1362 if (ret == 0) {
1363 /* Orderly shutdown. Not necessary to print an error. */
1364 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1365 } else {
1366 ERR("Relay didn't receive valid reset_metadata struct "
1367 "size : %d", ret);
1368 }
1369 ret = -1;
1370 goto end_no_session;
1371 }
1372 DBG("Update metadata to version %" PRIu64, be64toh(stream_info.version));
1373
1374 /* Unsupported for live sessions for now. */
1375 if (session->live_timer != 0) {
1376 ret = -1;
1377 goto end;
1378 }
1379
1380 stream = stream_get_by_id(be64toh(stream_info.stream_id));
1381 if (!stream) {
1382 ret = -1;
1383 goto end;
1384 }
1385 pthread_mutex_lock(&stream->lock);
1386 if (!stream->is_metadata) {
1387 ret = -1;
1388 goto end_unlock;
1389 }
1390
1391 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
1392 0, 0, -1, -1, stream->stream_fd->fd, NULL,
1393 &stream->stream_fd->fd);
1394 if (ret < 0) {
1395 ERR("Failed to rotate metadata file %s of channel %s",
1396 stream->path_name, stream->channel_name);
1397 goto end_unlock;
1398 }
1399
1400 end_unlock:
1401 pthread_mutex_unlock(&stream->lock);
1402 stream_put(stream);
1403
1404 end:
1405 memset(&reply, 0, sizeof(reply));
1406 if (ret < 0) {
1407 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1408 } else {
1409 reply.ret_code = htobe32(LTTNG_OK);
1410 }
1411 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1412 sizeof(struct lttcomm_relayd_generic_reply), 0);
1413 if (send_ret < 0) {
1414 ERR("Relay sending reset metadata reply");
1415 ret = send_ret;
1416 }
1417
1418 end_no_session:
1419 return ret;
1420 }
1421
1422 /*
1423 * relay_unknown_command: send -1 if received unknown command
1424 */
1425 static void relay_unknown_command(struct relay_connection *conn)
1426 {
1427 struct lttcomm_relayd_generic_reply reply;
1428 int ret;
1429
1430 memset(&reply, 0, sizeof(reply));
1431 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1432 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1433 sizeof(struct lttcomm_relayd_generic_reply), 0);
1434 if (ret < 0) {
1435 ERR("Relay sending unknown command");
1436 }
1437 }
1438
1439 /*
1440 * relay_start: send an acknowledgment to the client to tell if we are
1441 * ready to receive data. We are ready if a session is established.
1442 */
1443 static int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1444 struct relay_connection *conn)
1445 {
1446 int ret = htobe32(LTTNG_OK);
1447 struct lttcomm_relayd_generic_reply reply;
1448 struct relay_session *session = conn->session;
1449
1450 if (!session) {
1451 DBG("Trying to start the streaming without a session established");
1452 ret = htobe32(LTTNG_ERR_UNK);
1453 }
1454
1455 memset(&reply, 0, sizeof(reply));
1456 reply.ret_code = ret;
1457 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1458 sizeof(struct lttcomm_relayd_generic_reply), 0);
1459 if (ret < 0) {
1460 ERR("Relay sending start ack");
1461 }
1462
1463 return ret;
1464 }
1465
1466 /*
1467 * Append padding to the file pointed by the file descriptor fd.
1468 */
1469 static int write_padding_to_file(int fd, uint32_t size)
1470 {
1471 ssize_t ret = 0;
1472 char *zeros;
1473
1474 if (size == 0) {
1475 goto end;
1476 }
1477
1478 zeros = zmalloc(size);
1479 if (zeros == NULL) {
1480 PERROR("zmalloc zeros for padding");
1481 ret = -1;
1482 goto end;
1483 }
1484
1485 ret = lttng_write(fd, zeros, size);
1486 if (ret < size) {
1487 PERROR("write padding to file");
1488 }
1489
1490 free(zeros);
1491
1492 end:
1493 return ret;
1494 }
1495
1496 /*
1497 * relay_recv_metadata: receive the metadata for the session.
1498 */
1499 static int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1500 struct relay_connection *conn)
1501 {
1502 int ret = 0;
1503 ssize_t size_ret;
1504 struct relay_session *session = conn->session;
1505 struct lttcomm_relayd_metadata_payload *metadata_struct;
1506 struct relay_stream *metadata_stream;
1507 uint64_t data_size, payload_size;
1508
1509 if (!session) {
1510 ERR("Metadata sent before version check");
1511 ret = -1;
1512 goto end;
1513 }
1514
1515 data_size = payload_size = be64toh(recv_hdr->data_size);
1516 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1517 ERR("Incorrect data size");
1518 ret = -1;
1519 goto end;
1520 }
1521 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1522
1523 if (data_buffer_size < data_size) {
1524 /* In case the realloc fails, we can free the memory */
1525 char *tmp_data_ptr;
1526
1527 tmp_data_ptr = realloc(data_buffer, data_size);
1528 if (!tmp_data_ptr) {
1529 ERR("Allocating data buffer");
1530 free(data_buffer);
1531 ret = -1;
1532 goto end;
1533 }
1534 data_buffer = tmp_data_ptr;
1535 data_buffer_size = data_size;
1536 }
1537 memset(data_buffer, 0, data_size);
1538 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1539 size_ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
1540 if (size_ret < 0 || size_ret != data_size) {
1541 if (size_ret == 0) {
1542 /* Orderly shutdown. Not necessary to print an error. */
1543 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1544 } else {
1545 ERR("Relay didn't receive the whole metadata");
1546 }
1547 ret = -1;
1548 goto end;
1549 }
1550 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1551
1552 metadata_stream = stream_get_by_id(be64toh(metadata_struct->stream_id));
1553 if (!metadata_stream) {
1554 ret = -1;
1555 goto end;
1556 }
1557
1558 pthread_mutex_lock(&metadata_stream->lock);
1559
1560 size_ret = lttng_write(metadata_stream->stream_fd->fd, metadata_struct->payload,
1561 payload_size);
1562 if (size_ret < payload_size) {
1563 ERR("Relay error writing metadata on file");
1564 ret = -1;
1565 goto end_put;
1566 }
1567
1568 size_ret = write_padding_to_file(metadata_stream->stream_fd->fd,
1569 be32toh(metadata_struct->padding_size));
1570 if (size_ret < 0) {
1571 goto end_put;
1572 }
1573
1574 metadata_stream->metadata_received +=
1575 payload_size + be32toh(metadata_struct->padding_size);
1576 DBG2("Relay metadata written. Updated metadata_received %" PRIu64,
1577 metadata_stream->metadata_received);
1578
1579 end_put:
1580 pthread_mutex_unlock(&metadata_stream->lock);
1581 stream_put(metadata_stream);
1582 end:
1583 return ret;
1584 }
1585
1586 /*
1587 * relay_send_version: send relayd version number
1588 */
1589 static int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1590 struct relay_connection *conn)
1591 {
1592 int ret;
1593 struct lttcomm_relayd_version reply, msg;
1594
1595 conn->version_check_done = 1;
1596
1597 /* Get version from the other side. */
1598 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1599 if (ret < 0 || ret != sizeof(msg)) {
1600 if (ret == 0) {
1601 /* Orderly shutdown. Not necessary to print an error. */
1602 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1603 } else {
1604 ERR("Relay failed to receive the version values.");
1605 }
1606 ret = -1;
1607 goto end;
1608 }
1609
1610 memset(&reply, 0, sizeof(reply));
1611 reply.major = RELAYD_VERSION_COMM_MAJOR;
1612 reply.minor = RELAYD_VERSION_COMM_MINOR;
1613
1614 /* Major versions must be the same */
1615 if (reply.major != be32toh(msg.major)) {
1616 DBG("Incompatible major versions (%u vs %u), deleting session",
1617 reply.major, be32toh(msg.major));
1618 connection_put(conn);
1619 ret = 0;
1620 goto end;
1621 }
1622
1623 conn->major = reply.major;
1624 /* We adapt to the lowest compatible version */
1625 if (reply.minor <= be32toh(msg.minor)) {
1626 conn->minor = reply.minor;
1627 } else {
1628 conn->minor = be32toh(msg.minor);
1629 }
1630
1631 reply.major = htobe32(reply.major);
1632 reply.minor = htobe32(reply.minor);
1633 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1634 sizeof(struct lttcomm_relayd_version), 0);
1635 if (ret < 0) {
1636 ERR("Relay sending version");
1637 }
1638
1639 DBG("Version check done using protocol %u.%u", conn->major,
1640 conn->minor);
1641
1642 end:
1643 return ret;
1644 }
1645
1646 /*
1647 * Check for data pending for a given stream id from the session daemon.
1648 */
1649 static int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1650 struct relay_connection *conn)
1651 {
1652 struct relay_session *session = conn->session;
1653 struct lttcomm_relayd_data_pending msg;
1654 struct lttcomm_relayd_generic_reply reply;
1655 struct relay_stream *stream;
1656 int ret;
1657 uint64_t last_net_seq_num, stream_id;
1658
1659 DBG("Data pending command received");
1660
1661 if (!session || conn->version_check_done == 0) {
1662 ERR("Trying to check for data before version check");
1663 ret = -1;
1664 goto end_no_session;
1665 }
1666
1667 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1668 if (ret < sizeof(msg)) {
1669 if (ret == 0) {
1670 /* Orderly shutdown. Not necessary to print an error. */
1671 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1672 } else {
1673 ERR("Relay didn't receive valid data_pending struct size : %d",
1674 ret);
1675 }
1676 ret = -1;
1677 goto end_no_session;
1678 }
1679
1680 stream_id = be64toh(msg.stream_id);
1681 last_net_seq_num = be64toh(msg.last_net_seq_num);
1682
1683 stream = stream_get_by_id(stream_id);
1684 if (stream == NULL) {
1685 ret = -1;
1686 goto end;
1687 }
1688
1689 pthread_mutex_lock(&stream->lock);
1690
1691 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1692 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1693 last_net_seq_num);
1694
1695 /* Avoid wrapping issue */
1696 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1697 /* Data has in fact been written and is NOT pending */
1698 ret = 0;
1699 } else {
1700 /* Data still being streamed thus pending */
1701 ret = 1;
1702 }
1703
1704 stream->data_pending_check_done = true;
1705 pthread_mutex_unlock(&stream->lock);
1706
1707 stream_put(stream);
1708 end:
1709
1710 memset(&reply, 0, sizeof(reply));
1711 reply.ret_code = htobe32(ret);
1712 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1713 if (ret < 0) {
1714 ERR("Relay data pending ret code failed");
1715 }
1716
1717 end_no_session:
1718 return ret;
1719 }
1720
1721 /*
1722 * Wait for the control socket to reach a quiescent state.
1723 *
1724 * Note that for now, when receiving this command from the session
1725 * daemon, this means that every subsequent commands or data received on
1726 * the control socket has been handled. So, this is why we simply return
1727 * OK here.
1728 */
1729 static int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1730 struct relay_connection *conn)
1731 {
1732 int ret;
1733 uint64_t stream_id;
1734 struct relay_stream *stream;
1735 struct lttcomm_relayd_quiescent_control msg;
1736 struct lttcomm_relayd_generic_reply reply;
1737
1738 DBG("Checking quiescent state on control socket");
1739
1740 if (!conn->session || conn->version_check_done == 0) {
1741 ERR("Trying to check for data before version check");
1742 ret = -1;
1743 goto end_no_session;
1744 }
1745
1746 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1747 if (ret < sizeof(msg)) {
1748 if (ret == 0) {
1749 /* Orderly shutdown. Not necessary to print an error. */
1750 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1751 } else {
1752 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1753 ret);
1754 }
1755 ret = -1;
1756 goto end_no_session;
1757 }
1758
1759 stream_id = be64toh(msg.stream_id);
1760 stream = stream_get_by_id(stream_id);
1761 if (!stream) {
1762 goto reply;
1763 }
1764 pthread_mutex_lock(&stream->lock);
1765 stream->data_pending_check_done = true;
1766 pthread_mutex_unlock(&stream->lock);
1767 DBG("Relay quiescent control pending flag set to %" PRIu64, stream_id);
1768 stream_put(stream);
1769 reply:
1770 memset(&reply, 0, sizeof(reply));
1771 reply.ret_code = htobe32(LTTNG_OK);
1772 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1773 if (ret < 0) {
1774 ERR("Relay data quiescent control ret code failed");
1775 }
1776
1777 end_no_session:
1778 return ret;
1779 }
1780
1781 /*
1782 * Initialize a data pending command. This means that a consumer is about
1783 * to ask for data pending for each stream it holds. Simply iterate over
1784 * all streams of a session and set the data_pending_check_done flag.
1785 *
1786 * This command returns to the client a LTTNG_OK code.
1787 */
1788 static int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1789 struct relay_connection *conn)
1790 {
1791 int ret;
1792 struct lttng_ht_iter iter;
1793 struct lttcomm_relayd_begin_data_pending msg;
1794 struct lttcomm_relayd_generic_reply reply;
1795 struct relay_stream *stream;
1796 uint64_t session_id;
1797
1798 assert(recv_hdr);
1799 assert(conn);
1800
1801 DBG("Init streams for data pending");
1802
1803 if (!conn->session || conn->version_check_done == 0) {
1804 ERR("Trying to check for data before version check");
1805 ret = -1;
1806 goto end_no_session;
1807 }
1808
1809 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1810 if (ret < sizeof(msg)) {
1811 if (ret == 0) {
1812 /* Orderly shutdown. Not necessary to print an error. */
1813 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1814 } else {
1815 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1816 ret);
1817 }
1818 ret = -1;
1819 goto end_no_session;
1820 }
1821
1822 session_id = be64toh(msg.session_id);
1823
1824 /*
1825 * Iterate over all streams to set the begin data pending flag.
1826 * For now, the streams are indexed by stream handle so we have
1827 * to iterate over all streams to find the one associated with
1828 * the right session_id.
1829 */
1830 rcu_read_lock();
1831 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1832 node.node) {
1833 if (!stream_get(stream)) {
1834 continue;
1835 }
1836 if (stream->trace->session->id == session_id) {
1837 pthread_mutex_lock(&stream->lock);
1838 stream->data_pending_check_done = false;
1839 pthread_mutex_unlock(&stream->lock);
1840 DBG("Set begin data pending flag to stream %" PRIu64,
1841 stream->stream_handle);
1842 }
1843 stream_put(stream);
1844 }
1845 rcu_read_unlock();
1846
1847 memset(&reply, 0, sizeof(reply));
1848 /* All good, send back reply. */
1849 reply.ret_code = htobe32(LTTNG_OK);
1850
1851 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1852 if (ret < 0) {
1853 ERR("Relay begin data pending send reply failed");
1854 }
1855
1856 end_no_session:
1857 return ret;
1858 }
1859
1860 /*
1861 * End data pending command. This will check, for a given session id, if
1862 * each stream associated with it has its data_pending_check_done flag
1863 * set. If not, this means that the client lost track of the stream but
1864 * the data is still being streamed on our side. In this case, we inform
1865 * the client that data is in flight.
1866 *
1867 * Return to the client if there is data in flight or not with a ret_code.
1868 */
1869 static int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1870 struct relay_connection *conn)
1871 {
1872 int ret;
1873 struct lttng_ht_iter iter;
1874 struct lttcomm_relayd_end_data_pending msg;
1875 struct lttcomm_relayd_generic_reply reply;
1876 struct relay_stream *stream;
1877 uint64_t session_id;
1878 uint32_t is_data_inflight = 0;
1879
1880 DBG("End data pending command");
1881
1882 if (!conn->session || conn->version_check_done == 0) {
1883 ERR("Trying to check for data before version check");
1884 ret = -1;
1885 goto end_no_session;
1886 }
1887
1888 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1889 if (ret < sizeof(msg)) {
1890 if (ret == 0) {
1891 /* Orderly shutdown. Not necessary to print an error. */
1892 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1893 } else {
1894 ERR("Relay didn't receive valid end data_pending struct size: %d",
1895 ret);
1896 }
1897 ret = -1;
1898 goto end_no_session;
1899 }
1900
1901 session_id = be64toh(msg.session_id);
1902
1903 /*
1904 * Iterate over all streams to see if the begin data pending
1905 * flag is set.
1906 */
1907 rcu_read_lock();
1908 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1909 node.node) {
1910 if (!stream_get(stream)) {
1911 continue;
1912 }
1913 if (stream->trace->session->id != session_id) {
1914 stream_put(stream);
1915 continue;
1916 }
1917 pthread_mutex_lock(&stream->lock);
1918 if (!stream->data_pending_check_done) {
1919 if (!stream->closed || !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
1920 is_data_inflight = 1;
1921 DBG("Data is still in flight for stream %" PRIu64,
1922 stream->stream_handle);
1923 pthread_mutex_unlock(&stream->lock);
1924 stream_put(stream);
1925 break;
1926 }
1927 }
1928 pthread_mutex_unlock(&stream->lock);
1929 stream_put(stream);
1930 }
1931 rcu_read_unlock();
1932
1933 memset(&reply, 0, sizeof(reply));
1934 /* All good, send back reply. */
1935 reply.ret_code = htobe32(is_data_inflight);
1936
1937 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1938 if (ret < 0) {
1939 ERR("Relay end data pending send reply failed");
1940 }
1941
1942 end_no_session:
1943 return ret;
1944 }
1945
1946 /*
1947 * Receive an index for a specific stream.
1948 *
1949 * Return 0 on success else a negative value.
1950 */
1951 static int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1952 struct relay_connection *conn)
1953 {
1954 int ret, send_ret;
1955 struct relay_session *session = conn->session;
1956 struct lttcomm_relayd_index index_info;
1957 struct relay_index *index;
1958 struct lttcomm_relayd_generic_reply reply;
1959 struct relay_stream *stream;
1960 uint64_t net_seq_num;
1961 size_t msg_len;
1962
1963 assert(conn);
1964
1965 DBG("Relay receiving index");
1966
1967 if (!session || conn->version_check_done == 0) {
1968 ERR("Trying to close a stream before version check");
1969 ret = -1;
1970 goto end_no_session;
1971 }
1972
1973 msg_len = lttcomm_relayd_index_len(
1974 lttng_to_index_major(conn->major, conn->minor),
1975 lttng_to_index_minor(conn->major, conn->minor));
1976 ret = conn->sock->ops->recvmsg(conn->sock, &index_info,
1977 msg_len, 0);
1978 if (ret < msg_len) {
1979 if (ret == 0) {
1980 /* Orderly shutdown. Not necessary to print an error. */
1981 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1982 } else {
1983 ERR("Relay didn't receive valid index struct size : %d", ret);
1984 }
1985 ret = -1;
1986 goto end_no_session;
1987 }
1988
1989 net_seq_num = be64toh(index_info.net_seq_num);
1990
1991 stream = stream_get_by_id(be64toh(index_info.relay_stream_id));
1992 if (!stream) {
1993 ERR("stream_get_by_id not found");
1994 ret = -1;
1995 goto end;
1996 }
1997 pthread_mutex_lock(&stream->lock);
1998
1999 /* Live beacon handling */
2000 if (index_info.packet_size == 0) {
2001 DBG("Received live beacon for stream %" PRIu64,
2002 stream->stream_handle);
2003
2004 /*
2005 * Only flag a stream inactive when it has already
2006 * received data and no indexes are in flight.
2007 */
2008 if (stream->index_received_seqcount > 0
2009 && stream->indexes_in_flight == 0) {
2010 stream->beacon_ts_end =
2011 be64toh(index_info.timestamp_end);
2012 }
2013 ret = 0;
2014 goto end_stream_put;
2015 } else {
2016 stream->beacon_ts_end = -1ULL;
2017 }
2018
2019 if (stream->ctf_stream_id == -1ULL) {
2020 stream->ctf_stream_id = be64toh(index_info.stream_id);
2021 }
2022 index = relay_index_get_by_id_or_create(stream, net_seq_num);
2023 if (!index) {
2024 ret = -1;
2025 ERR("relay_index_get_by_id_or_create index NULL");
2026 goto end_stream_put;
2027 }
2028 if (set_index_control_data(index, &index_info, conn)) {
2029 ERR("set_index_control_data error");
2030 relay_index_put(index);
2031 ret = -1;
2032 goto end_stream_put;
2033 }
2034 ret = relay_index_try_flush(index);
2035 if (ret == 0) {
2036 tracefile_array_commit_seq(stream->tfa);
2037 stream->index_received_seqcount++;
2038 } else if (ret > 0) {
2039 /* no flush. */
2040 ret = 0;
2041 } else {
2042 ERR("relay_index_try_flush error %d", ret);
2043 relay_index_put(index);
2044 ret = -1;
2045 }
2046
2047 end_stream_put:
2048 pthread_mutex_unlock(&stream->lock);
2049 stream_put(stream);
2050
2051 end:
2052
2053 memset(&reply, 0, sizeof(reply));
2054 if (ret < 0) {
2055 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2056 } else {
2057 reply.ret_code = htobe32(LTTNG_OK);
2058 }
2059 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2060 if (send_ret < 0) {
2061 ERR("Relay sending close index id reply");
2062 ret = send_ret;
2063 }
2064
2065 end_no_session:
2066 return ret;
2067 }
2068
2069 /*
2070 * Receive the streams_sent message.
2071 *
2072 * Return 0 on success else a negative value.
2073 */
2074 static int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
2075 struct relay_connection *conn)
2076 {
2077 int ret, send_ret;
2078 struct lttcomm_relayd_generic_reply reply;
2079
2080 assert(conn);
2081
2082 DBG("Relay receiving streams_sent");
2083
2084 if (!conn->session || conn->version_check_done == 0) {
2085 ERR("Trying to close a stream before version check");
2086 ret = -1;
2087 goto end_no_session;
2088 }
2089
2090 /*
2091 * Publish every pending stream in the connection recv list which are
2092 * now ready to be used by the viewer.
2093 */
2094 publish_connection_local_streams(conn);
2095
2096 memset(&reply, 0, sizeof(reply));
2097 reply.ret_code = htobe32(LTTNG_OK);
2098 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2099 if (send_ret < 0) {
2100 ERR("Relay sending sent_stream reply");
2101 ret = send_ret;
2102 } else {
2103 /* Success. */
2104 ret = 0;
2105 }
2106
2107 end_no_session:
2108 return ret;
2109 }
2110
2111 /*
2112 * Process the commands received on the control socket
2113 */
2114 static int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
2115 struct relay_connection *conn)
2116 {
2117 int ret = 0;
2118
2119 switch (be32toh(recv_hdr->cmd)) {
2120 case RELAYD_CREATE_SESSION:
2121 ret = relay_create_session(recv_hdr, conn);
2122 break;
2123 case RELAYD_ADD_STREAM:
2124 ret = relay_add_stream(recv_hdr, conn);
2125 break;
2126 case RELAYD_START_DATA:
2127 ret = relay_start(recv_hdr, conn);
2128 break;
2129 case RELAYD_SEND_METADATA:
2130 ret = relay_recv_metadata(recv_hdr, conn);
2131 break;
2132 case RELAYD_VERSION:
2133 ret = relay_send_version(recv_hdr, conn);
2134 break;
2135 case RELAYD_CLOSE_STREAM:
2136 ret = relay_close_stream(recv_hdr, conn);
2137 break;
2138 case RELAYD_DATA_PENDING:
2139 ret = relay_data_pending(recv_hdr, conn);
2140 break;
2141 case RELAYD_QUIESCENT_CONTROL:
2142 ret = relay_quiescent_control(recv_hdr, conn);
2143 break;
2144 case RELAYD_BEGIN_DATA_PENDING:
2145 ret = relay_begin_data_pending(recv_hdr, conn);
2146 break;
2147 case RELAYD_END_DATA_PENDING:
2148 ret = relay_end_data_pending(recv_hdr, conn);
2149 break;
2150 case RELAYD_SEND_INDEX:
2151 ret = relay_recv_index(recv_hdr, conn);
2152 break;
2153 case RELAYD_STREAMS_SENT:
2154 ret = relay_streams_sent(recv_hdr, conn);
2155 break;
2156 case RELAYD_RESET_METADATA:
2157 ret = relay_reset_metadata(recv_hdr, conn);
2158 break;
2159 case RELAYD_UPDATE_SYNC_INFO:
2160 default:
2161 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2162 relay_unknown_command(conn);
2163 ret = -1;
2164 goto end;
2165 }
2166
2167 end:
2168 return ret;
2169 }
2170
2171 /*
2172 * Handle index for a data stream.
2173 *
2174 * Called with the stream lock held.
2175 *
2176 * Return 0 on success else a negative value.
2177 */
2178 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2179 int rotate_index)
2180 {
2181 int ret = 0;
2182 uint64_t data_offset;
2183 struct relay_index *index;
2184
2185 /* Get data offset because we are about to update the index. */
2186 data_offset = htobe64(stream->tracefile_size_current);
2187
2188 DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64,
2189 stream->stream_handle, net_seq_num, stream->tracefile_size_current);
2190
2191 /*
2192 * Lookup for an existing index for that stream id/sequence
2193 * number. If it exists, the control thread has already received the
2194 * data for it, thus we need to write it to disk.
2195 */
2196 index = relay_index_get_by_id_or_create(stream, net_seq_num);
2197 if (!index) {
2198 ret = -1;
2199 goto end;
2200 }
2201
2202 if (rotate_index || !stream->index_file) {
2203 uint32_t major, minor;
2204
2205 /* Put ref on previous index_file. */
2206 if (stream->index_file) {
2207 lttng_index_file_put(stream->index_file);
2208 stream->index_file = NULL;
2209 }
2210 major = stream->trace->session->major;
2211 minor = stream->trace->session->minor;
2212 stream->index_file = lttng_index_file_create(stream->path_name,
2213 stream->channel_name,
2214 -1, -1, stream->tracefile_size,
2215 tracefile_array_get_file_index_head(stream->tfa),
2216 lttng_to_index_major(major, minor),
2217 lttng_to_index_minor(major, minor));
2218 if (!stream->index_file) {
2219 ret = -1;
2220 /* Put self-ref for this index due to error. */
2221 relay_index_put(index);
2222 index = NULL;
2223 goto end;
2224 }
2225 }
2226
2227 if (relay_index_set_file(index, stream->index_file, data_offset)) {
2228 ret = -1;
2229 /* Put self-ref for this index due to error. */
2230 relay_index_put(index);
2231 index = NULL;
2232 goto end;
2233 }
2234
2235 ret = relay_index_try_flush(index);
2236 if (ret == 0) {
2237 tracefile_array_commit_seq(stream->tfa);
2238 stream->index_received_seqcount++;
2239 } else if (ret > 0) {
2240 /* No flush. */
2241 ret = 0;
2242 } else {
2243 /* Put self-ref for this index due to error. */
2244 relay_index_put(index);
2245 index = NULL;
2246 ret = -1;
2247 }
2248 end:
2249 return ret;
2250 }
2251
2252 /*
2253 * relay_process_data: Process the data received on the data socket
2254 */
2255 static int relay_process_data(struct relay_connection *conn)
2256 {
2257 int ret = 0, rotate_index = 0;
2258 ssize_t size_ret;
2259 struct relay_stream *stream;
2260 struct lttcomm_relayd_data_hdr data_hdr;
2261 uint64_t stream_id;
2262 uint64_t net_seq_num;
2263 uint32_t data_size;
2264 struct relay_session *session;
2265 bool new_stream = false, close_requested = false;
2266 size_t chunk_size = RECV_DATA_BUFFER_SIZE;
2267 size_t recv_off = 0;
2268 char data_buffer[chunk_size];
2269
2270 ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr,
2271 sizeof(struct lttcomm_relayd_data_hdr), 0);
2272 if (ret <= 0) {
2273 if (ret == 0) {
2274 /* Orderly shutdown. Not necessary to print an error. */
2275 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2276 } else {
2277 ERR("Unable to receive data header on sock %d", conn->sock->fd);
2278 }
2279 ret = -1;
2280 goto end;
2281 }
2282
2283 stream_id = be64toh(data_hdr.stream_id);
2284 stream = stream_get_by_id(stream_id);
2285 if (!stream) {
2286 ERR("relay_process_data: Cannot find stream %" PRIu64, stream_id);
2287 ret = -1;
2288 goto end;
2289 }
2290 session = stream->trace->session;
2291 data_size = be32toh(data_hdr.data_size);
2292
2293 net_seq_num = be64toh(data_hdr.net_seq_num);
2294
2295 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2296 data_size, stream_id, net_seq_num);
2297
2298 pthread_mutex_lock(&stream->lock);
2299
2300 /* Check if a rotation is needed. */
2301 if (stream->tracefile_size > 0 &&
2302 (stream->tracefile_size_current + data_size) >
2303 stream->tracefile_size) {
2304 uint64_t old_id, new_id;
2305
2306 old_id = tracefile_array_get_file_index_head(stream->tfa);
2307 tracefile_array_file_rotate(stream->tfa);
2308
2309 /* new_id is updated by utils_rotate_stream_file. */
2310 new_id = old_id;
2311
2312 ret = utils_rotate_stream_file(stream->path_name,
2313 stream->channel_name, stream->tracefile_size,
2314 stream->tracefile_count, -1,
2315 -1, stream->stream_fd->fd,
2316 &new_id, &stream->stream_fd->fd);
2317 if (ret < 0) {
2318 ERR("Rotating stream output file");
2319 goto end_stream_unlock;
2320 }
2321 /*
2322 * Reset current size because we just performed a stream
2323 * rotation.
2324 */
2325 stream->tracefile_size_current = 0;
2326 rotate_index = 1;
2327 }
2328
2329 /*
2330 * Index are handled in protocol version 2.4 and above. Also,
2331 * snapshot and index are NOT supported.
2332 */
2333 if (session->minor >= 4 && !session->snapshot) {
2334 ret = handle_index_data(stream, net_seq_num, rotate_index);
2335 if (ret < 0) {
2336 ERR("handle_index_data: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
2337 stream->stream_handle, net_seq_num, ret);
2338 goto end_stream_unlock;
2339 }
2340 }
2341
2342 for (recv_off = 0; recv_off < data_size; recv_off += chunk_size) {
2343 size_t recv_size = min(data_size - recv_off, chunk_size);
2344
2345 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_size, 0);
2346 if (ret <= 0) {
2347 if (ret == 0) {
2348 /* Orderly shutdown. Not necessary to print an error. */
2349 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2350 } else {
2351 ERR("Socket %d error %d", conn->sock->fd, ret);
2352 }
2353 ret = -1;
2354 goto end_stream_unlock;
2355 }
2356
2357 /* Write data to stream output fd. */
2358 size_ret = lttng_write(stream->stream_fd->fd, data_buffer,
2359 recv_size);
2360 if (size_ret < recv_size) {
2361 ERR("Relay error writing data to file");
2362 ret = -1;
2363 goto end_stream_unlock;
2364 }
2365
2366 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64,
2367 size_ret, stream->stream_handle);
2368 }
2369
2370 ret = write_padding_to_file(stream->stream_fd->fd,
2371 be32toh(data_hdr.padding_size));
2372 if (ret < 0) {
2373 ERR("write_padding_to_file: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
2374 stream->stream_handle, net_seq_num, ret);
2375 goto end_stream_unlock;
2376 }
2377 stream->tracefile_size_current +=
2378 data_size + be32toh(data_hdr.padding_size);
2379 if (stream->prev_seq == -1ULL) {
2380 new_stream = true;
2381 }
2382
2383 stream->prev_seq = net_seq_num;
2384
2385 end_stream_unlock:
2386 close_requested = stream->close_requested;
2387 pthread_mutex_unlock(&stream->lock);
2388 if (close_requested) {
2389 try_stream_close(stream);
2390 }
2391
2392 if (new_stream) {
2393 pthread_mutex_lock(&session->lock);
2394 uatomic_set(&session->new_streams, 1);
2395 pthread_mutex_unlock(&session->lock);
2396 }
2397 stream_put(stream);
2398 end:
2399 return ret;
2400 }
2401
2402 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2403 {
2404 int ret;
2405
2406 (void) lttng_poll_del(events, pollfd);
2407
2408 ret = close(pollfd);
2409 if (ret < 0) {
2410 ERR("Closing pollfd %d", pollfd);
2411 }
2412 }
2413
2414 static void relay_thread_close_connection(struct lttng_poll_event *events,
2415 int pollfd, struct relay_connection *conn)
2416 {
2417 const char *type_str;
2418
2419 switch (conn->type) {
2420 case RELAY_DATA:
2421 type_str = "Data";
2422 break;
2423 case RELAY_CONTROL:
2424 type_str = "Control";
2425 break;
2426 case RELAY_VIEWER_COMMAND:
2427 type_str = "Viewer Command";
2428 break;
2429 case RELAY_VIEWER_NOTIFICATION:
2430 type_str = "Viewer Notification";
2431 break;
2432 default:
2433 type_str = "Unknown";
2434 }
2435 cleanup_connection_pollfd(events, pollfd);
2436 connection_put(conn);
2437 DBG("%s connection closed with %d", type_str, pollfd);
2438 }
2439
2440 /*
2441 * This thread does the actual work
2442 */
2443 static void *relay_thread_worker(void *data)
2444 {
2445 int ret, err = -1, last_seen_data_fd = -1;
2446 uint32_t nb_fd;
2447 struct lttng_poll_event events;
2448 struct lttng_ht *relay_connections_ht;
2449 struct lttng_ht_iter iter;
2450 struct lttcomm_relayd_hdr recv_hdr;
2451 struct relay_connection *destroy_conn = NULL;
2452
2453 DBG("[thread] Relay worker started");
2454
2455 rcu_register_thread();
2456
2457 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2458
2459 if (testpoint(relayd_thread_worker)) {
2460 goto error_testpoint;
2461 }
2462
2463 health_code_update();
2464
2465 /* table of connections indexed on socket */
2466 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2467 if (!relay_connections_ht) {
2468 goto relay_connections_ht_error;
2469 }
2470
2471 ret = create_thread_poll_set(&events, 2);
2472 if (ret < 0) {
2473 goto error_poll_create;
2474 }
2475
2476 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2477 if (ret < 0) {
2478 goto error;
2479 }
2480
2481 restart:
2482 while (1) {
2483 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2484
2485 health_code_update();
2486
2487 /* Infinite blocking call, waiting for transmission */
2488 DBG3("Relayd worker thread polling...");
2489 health_poll_entry();
2490 ret = lttng_poll_wait(&events, -1);
2491 health_poll_exit();
2492 if (ret < 0) {
2493 /*
2494 * Restart interrupted system call.
2495 */
2496 if (errno == EINTR) {
2497 goto restart;
2498 }
2499 goto error;
2500 }
2501
2502 nb_fd = ret;
2503
2504 /*
2505 * Process control. The control connection is
2506 * prioritized so we don't starve it with high
2507 * throughput tracing data on the data connection.
2508 */
2509 for (i = 0; i < nb_fd; i++) {
2510 /* Fetch once the poll data */
2511 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2512 int pollfd = LTTNG_POLL_GETFD(&events, i);
2513
2514 health_code_update();
2515
2516 if (!revents) {
2517 /*
2518 * No activity for this FD (poll
2519 * implementation).
2520 */
2521 continue;
2522 }
2523
2524 /* Thread quit pipe has been closed. Killing thread. */
2525 ret = check_thread_quit_pipe(pollfd, revents);
2526 if (ret) {
2527 err = 0;
2528 goto exit;
2529 }
2530
2531 /* Inspect the relay conn pipe for new connection */
2532 if (pollfd == relay_conn_pipe[0]) {
2533 if (revents & LPOLLIN) {
2534 struct relay_connection *conn;
2535
2536 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
2537 if (ret < 0) {
2538 goto error;
2539 }
2540 lttng_poll_add(&events, conn->sock->fd,
2541 LPOLLIN | LPOLLRDHUP);
2542 connection_ht_add(relay_connections_ht, conn);
2543 DBG("Connection socket %d added", conn->sock->fd);
2544 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2545 ERR("Relay connection pipe error");
2546 goto error;
2547 } else {
2548 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2549 goto error;
2550 }
2551 } else {
2552 struct relay_connection *ctrl_conn;
2553
2554 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
2555 /* If not found, there is a synchronization issue. */
2556 assert(ctrl_conn);
2557
2558 if (ctrl_conn->type == RELAY_DATA) {
2559 if (revents & LPOLLIN) {
2560 /*
2561 * Flag the last seen data fd not deleted. It will be
2562 * used as the last seen fd if any fd gets deleted in
2563 * this first loop.
2564 */
2565 last_notdel_data_fd = pollfd;
2566 }
2567 goto put_ctrl_connection;
2568 }
2569 assert(ctrl_conn->type == RELAY_CONTROL);
2570
2571 if (revents & LPOLLIN) {
2572 ret = ctrl_conn->sock->ops->recvmsg(ctrl_conn->sock,
2573 &recv_hdr, sizeof(recv_hdr), 0);
2574 if (ret <= 0) {
2575 /* Connection closed */
2576 relay_thread_close_connection(&events, pollfd,
2577 ctrl_conn);
2578 } else {
2579 ret = relay_process_control(&recv_hdr, ctrl_conn);
2580 if (ret < 0) {
2581 /* Clear the session on error. */
2582 relay_thread_close_connection(&events,
2583 pollfd, ctrl_conn);
2584 }
2585 seen_control = 1;
2586 }
2587 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2588 relay_thread_close_connection(&events,
2589 pollfd, ctrl_conn);
2590 if (last_seen_data_fd == pollfd) {
2591 last_seen_data_fd = last_notdel_data_fd;
2592 }
2593 } else {
2594 ERR("Unexpected poll events %u for control sock %d",
2595 revents, pollfd);
2596 connection_put(ctrl_conn);
2597 goto error;
2598 }
2599 put_ctrl_connection:
2600 connection_put(ctrl_conn);
2601 }
2602 }
2603
2604 /*
2605 * The last loop handled a control request, go back to poll to make
2606 * sure we prioritise the control socket.
2607 */
2608 if (seen_control) {
2609 continue;
2610 }
2611
2612 if (last_seen_data_fd >= 0) {
2613 for (i = 0; i < nb_fd; i++) {
2614 int pollfd = LTTNG_POLL_GETFD(&events, i);
2615
2616 health_code_update();
2617
2618 if (last_seen_data_fd == pollfd) {
2619 idx = i;
2620 break;
2621 }
2622 }
2623 }
2624
2625 /* Process data connection. */
2626 for (i = idx + 1; i < nb_fd; i++) {
2627 /* Fetch the poll data. */
2628 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2629 int pollfd = LTTNG_POLL_GETFD(&events, i);
2630 struct relay_connection *data_conn;
2631
2632 health_code_update();
2633
2634 if (!revents) {
2635 /* No activity for this FD (poll implementation). */
2636 continue;
2637 }
2638
2639 /* Skip the command pipe. It's handled in the first loop. */
2640 if (pollfd == relay_conn_pipe[0]) {
2641 continue;
2642 }
2643
2644 data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
2645 if (!data_conn) {
2646 /* Skip it. Might be removed before. */
2647 continue;
2648 }
2649 if (data_conn->type == RELAY_CONTROL) {
2650 goto put_data_connection;
2651 }
2652 assert(data_conn->type == RELAY_DATA);
2653
2654 if (revents & LPOLLIN) {
2655 ret = relay_process_data(data_conn);
2656 /* Connection closed */
2657 if (ret < 0) {
2658 relay_thread_close_connection(&events, pollfd,
2659 data_conn);
2660 /*
2661 * Every goto restart call sets the last seen fd where
2662 * here we don't really care since we gracefully
2663 * continue the loop after the connection is deleted.
2664 */
2665 } else {
2666 /* Keep last seen port. */
2667 last_seen_data_fd = pollfd;
2668 connection_put(data_conn);
2669 goto restart;
2670 }
2671 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2672 relay_thread_close_connection(&events, pollfd,
2673 data_conn);
2674 } else {
2675 ERR("Unknown poll events %u for data sock %d",
2676 revents, pollfd);
2677 }
2678 put_data_connection:
2679 connection_put(data_conn);
2680 }
2681 last_seen_data_fd = -1;
2682 }
2683
2684 /* Normal exit, no error */
2685 ret = 0;
2686
2687 exit:
2688 error:
2689 /* Cleanup reamaining connection object. */
2690 rcu_read_lock();
2691 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
2692 destroy_conn,
2693 sock_n.node) {
2694 health_code_update();
2695
2696 if (session_abort(destroy_conn->session)) {
2697 assert(0);
2698 }
2699
2700 /*
2701 * No need to grab another ref, because we own
2702 * destroy_conn.
2703 */
2704 relay_thread_close_connection(&events, destroy_conn->sock->fd,
2705 destroy_conn);
2706 }
2707 rcu_read_unlock();
2708
2709 lttng_poll_clean(&events);
2710 error_poll_create:
2711 lttng_ht_destroy(relay_connections_ht);
2712 relay_connections_ht_error:
2713 /* Close relay conn pipes */
2714 utils_close_pipe(relay_conn_pipe);
2715 if (err) {
2716 DBG("Thread exited with error");
2717 }
2718 DBG("Worker thread cleanup complete");
2719 error_testpoint:
2720 if (err) {
2721 health_error();
2722 ERR("Health error occurred in %s", __func__);
2723 }
2724 health_unregister(health_relayd);
2725 rcu_unregister_thread();
2726 lttng_relay_stop_threads();
2727 return NULL;
2728 }
2729
2730 /*
2731 * Create the relay command pipe to wake thread_manage_apps.
2732 * Closed in cleanup().
2733 */
2734 static int create_relay_conn_pipe(void)
2735 {
2736 int ret;
2737
2738 ret = utils_create_pipe_cloexec(relay_conn_pipe);
2739
2740 return ret;
2741 }
2742
2743 /*
2744 * main
2745 */
2746 int main(int argc, char **argv)
2747 {
2748 int ret = 0, retval = 0;
2749 void *status;
2750
2751 /* Parse arguments */
2752 progname = argv[0];
2753 if (set_options(argc, argv)) {
2754 retval = -1;
2755 goto exit_options;
2756 }
2757
2758 if (set_signal_handler()) {
2759 retval = -1;
2760 goto exit_options;
2761 }
2762
2763 /* Try to create directory if -o, --output is specified. */
2764 if (opt_output_path) {
2765 if (*opt_output_path != '/') {
2766 ERR("Please specify an absolute path for -o, --output PATH");
2767 retval = -1;
2768 goto exit_options;
2769 }
2770
2771 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG,
2772 -1, -1);
2773 if (ret < 0) {
2774 ERR("Unable to create %s", opt_output_path);
2775 retval = -1;
2776 goto exit_options;
2777 }
2778 }
2779
2780 /* Daemonize */
2781 if (opt_daemon || opt_background) {
2782 int i;
2783
2784 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2785 !opt_background);
2786 if (ret < 0) {
2787 retval = -1;
2788 goto exit_options;
2789 }
2790
2791 /*
2792 * We are in the child. Make sure all other file
2793 * descriptors are closed, in case we are called with
2794 * more opened file descriptors than the standard ones.
2795 */
2796 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2797 (void) close(i);
2798 }
2799 }
2800
2801 /* Initialize thread health monitoring */
2802 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2803 if (!health_relayd) {
2804 PERROR("health_app_create error");
2805 retval = -1;
2806 goto exit_health_app_create;
2807 }
2808
2809 /* Create thread quit pipe */
2810 if (init_thread_quit_pipe()) {
2811 retval = -1;
2812 goto exit_init_data;
2813 }
2814
2815 /* Setup the thread apps communication pipe. */
2816 if (create_relay_conn_pipe()) {
2817 retval = -1;
2818 goto exit_init_data;
2819 }
2820
2821 /* Init relay command queue. */
2822 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
2823
2824 /* Initialize communication library */
2825 lttcomm_init();
2826 lttcomm_inet_init();
2827
2828 /* tables of sessions indexed by session ID */
2829 sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2830 if (!sessions_ht) {
2831 retval = -1;
2832 goto exit_init_data;
2833 }
2834
2835 /* tables of streams indexed by stream ID */
2836 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2837 if (!relay_streams_ht) {
2838 retval = -1;
2839 goto exit_init_data;
2840 }
2841
2842 /* tables of streams indexed by stream ID */
2843 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2844 if (!viewer_streams_ht) {
2845 retval = -1;
2846 goto exit_init_data;
2847 }
2848
2849 ret = utils_create_pipe(health_quit_pipe);
2850 if (ret) {
2851 retval = -1;
2852 goto exit_health_quit_pipe;
2853 }
2854
2855 /* Create thread to manage the client socket */
2856 ret = pthread_create(&health_thread, default_pthread_attr(),
2857 thread_manage_health, (void *) NULL);
2858 if (ret) {
2859 errno = ret;
2860 PERROR("pthread_create health");
2861 retval = -1;
2862 goto exit_health_thread;
2863 }
2864
2865 /* Setup the dispatcher thread */
2866 ret = pthread_create(&dispatcher_thread, default_pthread_attr(),
2867 relay_thread_dispatcher, (void *) NULL);
2868 if (ret) {
2869 errno = ret;
2870 PERROR("pthread_create dispatcher");
2871 retval = -1;
2872 goto exit_dispatcher_thread;
2873 }
2874
2875 /* Setup the worker thread */
2876 ret = pthread_create(&worker_thread, default_pthread_attr(),
2877 relay_thread_worker, NULL);
2878 if (ret) {
2879 errno = ret;
2880 PERROR("pthread_create worker");
2881 retval = -1;
2882 goto exit_worker_thread;
2883 }
2884
2885 /* Setup the listener thread */
2886 ret = pthread_create(&listener_thread, default_pthread_attr(),
2887 relay_thread_listener, (void *) NULL);
2888 if (ret) {
2889 errno = ret;
2890 PERROR("pthread_create listener");
2891 retval = -1;
2892 goto exit_listener_thread;
2893 }
2894
2895 ret = relayd_live_create(live_uri);
2896 if (ret) {
2897 ERR("Starting live viewer threads");
2898 retval = -1;
2899 goto exit_live;
2900 }
2901
2902 /*
2903 * This is where we start awaiting program completion (e.g. through
2904 * signal that asks threads to teardown).
2905 */
2906
2907 ret = relayd_live_join();
2908 if (ret) {
2909 retval = -1;
2910 }
2911 exit_live:
2912
2913 ret = pthread_join(listener_thread, &status);
2914 if (ret) {
2915 errno = ret;
2916 PERROR("pthread_join listener_thread");
2917 retval = -1;
2918 }
2919
2920 exit_listener_thread:
2921 ret = pthread_join(worker_thread, &status);
2922 if (ret) {
2923 errno = ret;
2924 PERROR("pthread_join worker_thread");
2925 retval = -1;
2926 }
2927
2928 exit_worker_thread:
2929 ret = pthread_join(dispatcher_thread, &status);
2930 if (ret) {
2931 errno = ret;
2932 PERROR("pthread_join dispatcher_thread");
2933 retval = -1;
2934 }
2935 exit_dispatcher_thread:
2936
2937 ret = pthread_join(health_thread, &status);
2938 if (ret) {
2939 errno = ret;
2940 PERROR("pthread_join health_thread");
2941 retval = -1;
2942 }
2943 exit_health_thread:
2944
2945 utils_close_pipe(health_quit_pipe);
2946 exit_health_quit_pipe:
2947
2948 exit_init_data:
2949 health_app_destroy(health_relayd);
2950 exit_health_app_create:
2951 exit_options:
2952 /*
2953 * Wait for all pending call_rcu work to complete before tearing
2954 * down data structures. call_rcu worker may be trying to
2955 * perform lookups in those structures.
2956 */
2957 rcu_barrier();
2958 relayd_cleanup();
2959
2960 /* Ensure all prior call_rcu are done. */
2961 rcu_barrier();
2962
2963 if (!retval) {
2964 exit(EXIT_SUCCESS);
2965 } else {
2966 exit(EXIT_FAILURE);
2967 }
2968 }
This page took 0.09018 seconds and 4 git commands to generate.