0253997f7bb9d849d968811962601d51ad8cb43e
[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 #include "tcp_keep_alive.h"
74
75 static const char *help_msg =
76 #ifdef LTTNG_EMBED_HELP
77 #include <lttng-relayd.8.h>
78 #else
79 NULL
80 #endif
81 ;
82
83 /* command line options */
84 char *opt_output_path;
85 static int opt_daemon, opt_background;
86
87 /*
88 * We need to wait for listener and live listener threads, as well as
89 * health check thread, before being ready to signal readiness.
90 */
91 #define NR_LTTNG_RELAY_READY 3
92 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
93
94 /* Size of receive buffer. */
95 #define RECV_DATA_BUFFER_SIZE 65536
96
97 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
98 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
99
100 static struct lttng_uri *control_uri;
101 static struct lttng_uri *data_uri;
102 static struct lttng_uri *live_uri;
103
104 const char *progname;
105
106 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
107 static int tracing_group_name_override;
108
109 const char * const config_section_name = "relayd";
110
111 /*
112 * Quit pipe for all threads. This permits a single cancellation point
113 * for all threads when receiving an event on the pipe.
114 */
115 int thread_quit_pipe[2] = { -1, -1 };
116
117 /*
118 * This pipe is used to inform the worker thread that a command is queued and
119 * ready to be processed.
120 */
121 static int relay_conn_pipe[2] = { -1, -1 };
122
123 /* Shared between threads */
124 static int dispatch_thread_exit;
125
126 static pthread_t listener_thread;
127 static pthread_t dispatcher_thread;
128 static pthread_t worker_thread;
129 static pthread_t health_thread;
130
131 /*
132 * last_relay_stream_id_lock protects last_relay_stream_id increment
133 * atomicity on 32-bit architectures.
134 */
135 static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
136 static uint64_t last_relay_stream_id;
137
138 /*
139 * Relay command queue.
140 *
141 * The relay_thread_listener and relay_thread_dispatcher communicate with this
142 * queue.
143 */
144 static struct relay_conn_queue relay_conn_queue;
145
146 /* buffer allocated at startup, used to store the trace data */
147 static char *data_buffer;
148 static unsigned int data_buffer_size;
149
150 /* Global relay stream hash table. */
151 struct lttng_ht *relay_streams_ht;
152
153 /* Global relay viewer stream hash table. */
154 struct lttng_ht *viewer_streams_ht;
155
156 /* Global relay sessions hash table. */
157 struct lttng_ht *sessions_ht;
158
159 /* Relayd health monitoring */
160 struct health_app *health_relayd;
161
162 static struct option long_options[] = {
163 { "control-port", 1, 0, 'C', },
164 { "data-port", 1, 0, 'D', },
165 { "live-port", 1, 0, 'L', },
166 { "daemonize", 0, 0, 'd', },
167 { "background", 0, 0, 'b', },
168 { "group", 1, 0, 'g', },
169 { "help", 0, 0, 'h', },
170 { "output", 1, 0, 'o', },
171 { "verbose", 0, 0, 'v', },
172 { "config", 1, 0, 'f' },
173 { "version", 0, 0, 'V' },
174 { NULL, 0, 0, 0, },
175 };
176
177 static const char *config_ignore_options[] = { "help", "config", "version" };
178
179 /*
180 * Take an option from the getopt output and set it in the right variable to be
181 * used later.
182 *
183 * Return 0 on success else a negative value.
184 */
185 static int set_option(int opt, const char *arg, const char *optname)
186 {
187 int ret;
188
189 switch (opt) {
190 case 0:
191 fprintf(stderr, "option %s", optname);
192 if (arg) {
193 fprintf(stderr, " with arg %s\n", arg);
194 }
195 break;
196 case 'C':
197 if (lttng_is_setuid_setgid()) {
198 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
199 "-C, --control-port");
200 } else {
201 ret = uri_parse(arg, &control_uri);
202 if (ret < 0) {
203 ERR("Invalid control URI specified");
204 goto end;
205 }
206 if (control_uri->port == 0) {
207 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
208 }
209 }
210 break;
211 case 'D':
212 if (lttng_is_setuid_setgid()) {
213 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
214 "-D, -data-port");
215 } else {
216 ret = uri_parse(arg, &data_uri);
217 if (ret < 0) {
218 ERR("Invalid data URI specified");
219 goto end;
220 }
221 if (data_uri->port == 0) {
222 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
223 }
224 }
225 break;
226 case 'L':
227 if (lttng_is_setuid_setgid()) {
228 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
229 "-L, -live-port");
230 } else {
231 ret = uri_parse(arg, &live_uri);
232 if (ret < 0) {
233 ERR("Invalid live URI specified");
234 goto end;
235 }
236 if (live_uri->port == 0) {
237 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
238 }
239 }
240 break;
241 case 'd':
242 opt_daemon = 1;
243 break;
244 case 'b':
245 opt_background = 1;
246 break;
247 case 'g':
248 if (lttng_is_setuid_setgid()) {
249 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
250 "-g, --group");
251 } else {
252 tracing_group_name = strdup(arg);
253 if (tracing_group_name == NULL) {
254 ret = -errno;
255 PERROR("strdup");
256 goto end;
257 }
258 tracing_group_name_override = 1;
259 }
260 break;
261 case 'h':
262 ret = utils_show_help(8, "lttng-relayd", help_msg);
263 if (ret) {
264 ERR("Cannot show --help for `lttng-relayd`");
265 perror("exec");
266 }
267 exit(EXIT_FAILURE);
268 case 'V':
269 fprintf(stdout, "%s\n", VERSION);
270 exit(EXIT_SUCCESS);
271 case 'o':
272 if (lttng_is_setuid_setgid()) {
273 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
274 "-o, --output");
275 } else {
276 ret = asprintf(&opt_output_path, "%s", arg);
277 if (ret < 0) {
278 ret = -errno;
279 PERROR("asprintf opt_output_path");
280 goto end;
281 }
282 }
283 break;
284 case 'v':
285 /* Verbose level can increase using multiple -v */
286 if (arg) {
287 lttng_opt_verbose = config_parse_value(arg);
288 } else {
289 /* Only 3 level of verbosity (-vvv). */
290 if (lttng_opt_verbose < 3) {
291 lttng_opt_verbose += 1;
292 }
293 }
294 break;
295 default:
296 /* Unknown option or other error.
297 * Error is printed by getopt, just return */
298 ret = -1;
299 goto end;
300 }
301
302 /* All good. */
303 ret = 0;
304
305 end:
306 return ret;
307 }
308
309 /*
310 * config_entry_handler_cb used to handle options read from a config file.
311 * See config_entry_handler_cb comment in common/config/session-config.h for the
312 * return value conventions.
313 */
314 static int config_entry_handler(const struct config_entry *entry, void *unused)
315 {
316 int ret = 0, i;
317
318 if (!entry || !entry->name || !entry->value) {
319 ret = -EINVAL;
320 goto end;
321 }
322
323 /* Check if the option is to be ignored */
324 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
325 if (!strcmp(entry->name, config_ignore_options[i])) {
326 goto end;
327 }
328 }
329
330 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
331 /* Ignore if entry name is not fully matched. */
332 if (strcmp(entry->name, long_options[i].name)) {
333 continue;
334 }
335
336 /*
337 * If the option takes no argument on the command line,
338 * we have to check if the value is "true". We support
339 * non-zero numeric values, true, on and yes.
340 */
341 if (!long_options[i].has_arg) {
342 ret = config_parse_value(entry->value);
343 if (ret <= 0) {
344 if (ret) {
345 WARN("Invalid configuration value \"%s\" for option %s",
346 entry->value, entry->name);
347 }
348 /* False, skip boolean config option. */
349 goto end;
350 }
351 }
352
353 ret = set_option(long_options[i].val, entry->value, entry->name);
354 goto end;
355 }
356
357 WARN("Unrecognized option \"%s\" in daemon configuration file.",
358 entry->name);
359
360 end:
361 return ret;
362 }
363
364 static int set_options(int argc, char **argv)
365 {
366 int c, ret = 0, option_index = 0, retval = 0;
367 int orig_optopt = optopt, orig_optind = optind;
368 char *default_address, *optstring;
369 const char *config_path = NULL;
370
371 optstring = utils_generate_optstring(long_options,
372 sizeof(long_options) / sizeof(struct option));
373 if (!optstring) {
374 retval = -ENOMEM;
375 goto exit;
376 }
377
378 /* Check for the --config option */
379
380 while ((c = getopt_long(argc, argv, optstring, long_options,
381 &option_index)) != -1) {
382 if (c == '?') {
383 retval = -EINVAL;
384 goto exit;
385 } else if (c != 'f') {
386 continue;
387 }
388
389 if (lttng_is_setuid_setgid()) {
390 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
391 "-f, --config");
392 } else {
393 config_path = utils_expand_path(optarg);
394 if (!config_path) {
395 ERR("Failed to resolve path: %s", optarg);
396 }
397 }
398 }
399
400 ret = config_get_section_entries(config_path, config_section_name,
401 config_entry_handler, NULL);
402 if (ret) {
403 if (ret > 0) {
404 ERR("Invalid configuration option at line %i", ret);
405 }
406 retval = -1;
407 goto exit;
408 }
409
410 /* Reset getopt's global state */
411 optopt = orig_optopt;
412 optind = orig_optind;
413 while (1) {
414 c = getopt_long(argc, argv, optstring, long_options, &option_index);
415 if (c == -1) {
416 break;
417 }
418
419 ret = set_option(c, optarg, long_options[option_index].name);
420 if (ret < 0) {
421 retval = -1;
422 goto exit;
423 }
424 }
425
426 /* assign default values */
427 if (control_uri == NULL) {
428 ret = asprintf(&default_address,
429 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
430 DEFAULT_NETWORK_CONTROL_PORT);
431 if (ret < 0) {
432 PERROR("asprintf default data address");
433 retval = -1;
434 goto exit;
435 }
436
437 ret = uri_parse(default_address, &control_uri);
438 free(default_address);
439 if (ret < 0) {
440 ERR("Invalid control URI specified");
441 retval = -1;
442 goto exit;
443 }
444 }
445 if (data_uri == NULL) {
446 ret = asprintf(&default_address,
447 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
448 DEFAULT_NETWORK_DATA_PORT);
449 if (ret < 0) {
450 PERROR("asprintf default data address");
451 retval = -1;
452 goto exit;
453 }
454
455 ret = uri_parse(default_address, &data_uri);
456 free(default_address);
457 if (ret < 0) {
458 ERR("Invalid data URI specified");
459 retval = -1;
460 goto exit;
461 }
462 }
463 if (live_uri == NULL) {
464 ret = asprintf(&default_address,
465 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
466 DEFAULT_NETWORK_VIEWER_PORT);
467 if (ret < 0) {
468 PERROR("asprintf default viewer control address");
469 retval = -1;
470 goto exit;
471 }
472
473 ret = uri_parse(default_address, &live_uri);
474 free(default_address);
475 if (ret < 0) {
476 ERR("Invalid viewer control URI specified");
477 retval = -1;
478 goto exit;
479 }
480 }
481
482 exit:
483 free(optstring);
484 return retval;
485 }
486
487 static void print_global_objects(void)
488 {
489 rcu_register_thread();
490
491 print_viewer_streams();
492 print_relay_streams();
493 print_sessions();
494
495 rcu_unregister_thread();
496 }
497
498 /*
499 * Cleanup the daemon
500 */
501 static void relayd_cleanup(void)
502 {
503 print_global_objects();
504
505 DBG("Cleaning up");
506
507 if (viewer_streams_ht)
508 lttng_ht_destroy(viewer_streams_ht);
509 if (relay_streams_ht)
510 lttng_ht_destroy(relay_streams_ht);
511 if (sessions_ht)
512 lttng_ht_destroy(sessions_ht);
513
514 /* free the dynamically allocated opt_output_path */
515 free(opt_output_path);
516
517 /* Close thread quit pipes */
518 utils_close_pipe(thread_quit_pipe);
519
520 uri_free(control_uri);
521 uri_free(data_uri);
522 /* Live URI is freed in the live thread. */
523
524 if (tracing_group_name_override) {
525 free((void *) tracing_group_name);
526 }
527 }
528
529 /*
530 * Write to writable pipe used to notify a thread.
531 */
532 static int notify_thread_pipe(int wpipe)
533 {
534 ssize_t ret;
535
536 ret = lttng_write(wpipe, "!", 1);
537 if (ret < 1) {
538 PERROR("write poll pipe");
539 goto end;
540 }
541 ret = 0;
542 end:
543 return ret;
544 }
545
546 static int notify_health_quit_pipe(int *pipe)
547 {
548 ssize_t ret;
549
550 ret = lttng_write(pipe[1], "4", 1);
551 if (ret < 1) {
552 PERROR("write relay health quit");
553 goto end;
554 }
555 ret = 0;
556 end:
557 return ret;
558 }
559
560 /*
561 * Stop all relayd and relayd-live threads.
562 */
563 int lttng_relay_stop_threads(void)
564 {
565 int retval = 0;
566
567 /* Stopping all threads */
568 DBG("Terminating all threads");
569 if (notify_thread_pipe(thread_quit_pipe[1])) {
570 ERR("write error on thread quit pipe");
571 retval = -1;
572 }
573
574 if (notify_health_quit_pipe(health_quit_pipe)) {
575 ERR("write error on health quit pipe");
576 }
577
578 /* Dispatch thread */
579 CMM_STORE_SHARED(dispatch_thread_exit, 1);
580 futex_nto1_wake(&relay_conn_queue.futex);
581
582 if (relayd_live_stop()) {
583 ERR("Error stopping live threads");
584 retval = -1;
585 }
586 return retval;
587 }
588
589 /*
590 * Signal handler for the daemon
591 *
592 * Simply stop all worker threads, leaving main() return gracefully after
593 * joining all threads and calling cleanup().
594 */
595 static void sighandler(int sig)
596 {
597 switch (sig) {
598 case SIGINT:
599 DBG("SIGINT caught");
600 if (lttng_relay_stop_threads()) {
601 ERR("Error stopping threads");
602 }
603 break;
604 case SIGTERM:
605 DBG("SIGTERM caught");
606 if (lttng_relay_stop_threads()) {
607 ERR("Error stopping threads");
608 }
609 break;
610 case SIGUSR1:
611 CMM_STORE_SHARED(recv_child_signal, 1);
612 break;
613 default:
614 break;
615 }
616 }
617
618 /*
619 * Setup signal handler for :
620 * SIGINT, SIGTERM, SIGPIPE
621 */
622 static int set_signal_handler(void)
623 {
624 int ret = 0;
625 struct sigaction sa;
626 sigset_t sigset;
627
628 if ((ret = sigemptyset(&sigset)) < 0) {
629 PERROR("sigemptyset");
630 return ret;
631 }
632
633 sa.sa_mask = sigset;
634 sa.sa_flags = 0;
635
636 sa.sa_handler = sighandler;
637 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
638 PERROR("sigaction");
639 return ret;
640 }
641
642 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
643 PERROR("sigaction");
644 return ret;
645 }
646
647 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
648 PERROR("sigaction");
649 return ret;
650 }
651
652 sa.sa_handler = SIG_IGN;
653 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
654 PERROR("sigaction");
655 return ret;
656 }
657
658 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
659
660 return ret;
661 }
662
663 void lttng_relay_notify_ready(void)
664 {
665 /* Notify the parent of the fork() process that we are ready. */
666 if (opt_daemon || opt_background) {
667 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
668 kill(child_ppid, SIGUSR1);
669 }
670 }
671 }
672
673 /*
674 * Init thread quit pipe.
675 *
676 * Return -1 on error or 0 if all pipes are created.
677 */
678 static int init_thread_quit_pipe(void)
679 {
680 int ret;
681
682 ret = utils_create_pipe_cloexec(thread_quit_pipe);
683
684 return ret;
685 }
686
687 /*
688 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
689 */
690 static int create_thread_poll_set(struct lttng_poll_event *events, int size)
691 {
692 int ret;
693
694 if (events == NULL || size == 0) {
695 ret = -1;
696 goto error;
697 }
698
699 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
700 if (ret < 0) {
701 goto error;
702 }
703
704 /* Add quit pipe */
705 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
706 if (ret < 0) {
707 goto error;
708 }
709
710 return 0;
711
712 error:
713 return ret;
714 }
715
716 /*
717 * Check if the thread quit pipe was triggered.
718 *
719 * Return 1 if it was triggered else 0;
720 */
721 static int check_thread_quit_pipe(int fd, uint32_t events)
722 {
723 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
724 return 1;
725 }
726
727 return 0;
728 }
729
730 /*
731 * Create and init socket from uri.
732 */
733 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri)
734 {
735 int ret;
736 struct lttcomm_sock *sock = NULL;
737
738 sock = lttcomm_alloc_sock_from_uri(uri);
739 if (sock == NULL) {
740 ERR("Allocating socket");
741 goto error;
742 }
743
744 ret = lttcomm_create_sock(sock);
745 if (ret < 0) {
746 goto error;
747 }
748 DBG("Listening on sock %d", sock->fd);
749
750 ret = sock->ops->bind(sock);
751 if (ret < 0) {
752 PERROR("Failed to bind socket");
753 goto error;
754 }
755
756 ret = sock->ops->listen(sock, -1);
757 if (ret < 0) {
758 goto error;
759
760 }
761
762 return sock;
763
764 error:
765 if (sock) {
766 lttcomm_destroy_sock(sock);
767 }
768 return NULL;
769 }
770
771 /*
772 * This thread manages the listening for new connections on the network
773 */
774 static void *relay_thread_listener(void *data)
775 {
776 int i, ret, pollfd, err = -1;
777 uint32_t revents, nb_fd;
778 struct lttng_poll_event events;
779 struct lttcomm_sock *control_sock, *data_sock;
780
781 DBG("[thread] Relay listener started");
782
783 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
784
785 health_code_update();
786
787 control_sock = relay_socket_create(control_uri);
788 if (!control_sock) {
789 goto error_sock_control;
790 }
791
792 data_sock = relay_socket_create(data_uri);
793 if (!data_sock) {
794 goto error_sock_relay;
795 }
796
797 /*
798 * Pass 3 as size here for the thread quit pipe, control and
799 * data socket.
800 */
801 ret = create_thread_poll_set(&events, 3);
802 if (ret < 0) {
803 goto error_create_poll;
804 }
805
806 /* Add the control socket */
807 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
808 if (ret < 0) {
809 goto error_poll_add;
810 }
811
812 /* Add the data socket */
813 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
814 if (ret < 0) {
815 goto error_poll_add;
816 }
817
818 lttng_relay_notify_ready();
819
820 if (testpoint(relayd_thread_listener)) {
821 goto error_testpoint;
822 }
823
824 while (1) {
825 health_code_update();
826
827 DBG("Listener accepting connections");
828
829 restart:
830 health_poll_entry();
831 ret = lttng_poll_wait(&events, -1);
832 health_poll_exit();
833 if (ret < 0) {
834 /*
835 * Restart interrupted system call.
836 */
837 if (errno == EINTR) {
838 goto restart;
839 }
840 goto error;
841 }
842
843 nb_fd = ret;
844
845 DBG("Relay new connection received");
846 for (i = 0; i < nb_fd; i++) {
847 health_code_update();
848
849 /* Fetch once the poll data */
850 revents = LTTNG_POLL_GETEV(&events, i);
851 pollfd = LTTNG_POLL_GETFD(&events, i);
852
853 if (!revents) {
854 /*
855 * No activity for this FD (poll
856 * implementation).
857 */
858 continue;
859 }
860
861 /* Thread quit pipe has been closed. Killing thread. */
862 ret = check_thread_quit_pipe(pollfd, revents);
863 if (ret) {
864 err = 0;
865 goto exit;
866 }
867
868 if (revents & LPOLLIN) {
869 /*
870 * A new connection is requested, therefore a
871 * sessiond/consumerd connection is allocated in
872 * this thread, enqueued to a global queue and
873 * dequeued (and freed) in the worker thread.
874 */
875 int val = 1;
876 struct relay_connection *new_conn;
877 struct lttcomm_sock *newsock;
878 enum connection_type type;
879
880 if (pollfd == data_sock->fd) {
881 type = RELAY_DATA;
882 newsock = data_sock->ops->accept(data_sock);
883 DBG("Relay data connection accepted, socket %d",
884 newsock->fd);
885 } else {
886 assert(pollfd == control_sock->fd);
887 type = RELAY_CONTROL;
888 newsock = control_sock->ops->accept(control_sock);
889 DBG("Relay control connection accepted, socket %d",
890 newsock->fd);
891 }
892 if (!newsock) {
893 PERROR("accepting sock");
894 goto error;
895 }
896
897 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
898 sizeof(val));
899 if (ret < 0) {
900 PERROR("setsockopt inet");
901 lttcomm_destroy_sock(newsock);
902 goto error;
903 }
904
905 ret = socket_apply_keep_alive_config(newsock->fd);
906 if (ret < 0) {
907 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
908 newsock->fd);
909 lttcomm_destroy_sock(newsock);
910 goto error;
911 }
912
913 new_conn = connection_create(newsock, type);
914 if (!new_conn) {
915 lttcomm_destroy_sock(newsock);
916 goto error;
917 }
918
919 /* Enqueue request for the dispatcher thread. */
920 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
921 &new_conn->qnode);
922
923 /*
924 * Wake the dispatch queue futex.
925 * Implicit memory barrier with the
926 * exchange in cds_wfcq_enqueue.
927 */
928 futex_nto1_wake(&relay_conn_queue.futex);
929 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
930 ERR("socket poll error");
931 goto error;
932 } else {
933 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
934 goto error;
935 }
936 }
937 }
938
939 exit:
940 error:
941 error_poll_add:
942 error_testpoint:
943 lttng_poll_clean(&events);
944 error_create_poll:
945 if (data_sock->fd >= 0) {
946 ret = data_sock->ops->close(data_sock);
947 if (ret) {
948 PERROR("close");
949 }
950 }
951 lttcomm_destroy_sock(data_sock);
952 error_sock_relay:
953 if (control_sock->fd >= 0) {
954 ret = control_sock->ops->close(control_sock);
955 if (ret) {
956 PERROR("close");
957 }
958 }
959 lttcomm_destroy_sock(control_sock);
960 error_sock_control:
961 if (err) {
962 health_error();
963 ERR("Health error occurred in %s", __func__);
964 }
965 health_unregister(health_relayd);
966 DBG("Relay listener thread cleanup complete");
967 lttng_relay_stop_threads();
968 return NULL;
969 }
970
971 /*
972 * This thread manages the dispatching of the requests to worker threads
973 */
974 static void *relay_thread_dispatcher(void *data)
975 {
976 int err = -1;
977 ssize_t ret;
978 struct cds_wfcq_node *node;
979 struct relay_connection *new_conn = NULL;
980
981 DBG("[thread] Relay dispatcher started");
982
983 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
984
985 if (testpoint(relayd_thread_dispatcher)) {
986 goto error_testpoint;
987 }
988
989 health_code_update();
990
991 for (;;) {
992 health_code_update();
993
994 /* Atomically prepare the queue futex */
995 futex_nto1_prepare(&relay_conn_queue.futex);
996
997 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
998 break;
999 }
1000
1001 do {
1002 health_code_update();
1003
1004 /* Dequeue commands */
1005 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1006 &relay_conn_queue.tail);
1007 if (node == NULL) {
1008 DBG("Woken up but nothing in the relay command queue");
1009 /* Continue thread execution */
1010 break;
1011 }
1012 new_conn = caa_container_of(node, struct relay_connection, qnode);
1013
1014 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1015
1016 /*
1017 * Inform worker thread of the new request. This
1018 * call is blocking so we can be assured that
1019 * the data will be read at some point in time
1020 * or wait to the end of the world :)
1021 */
1022 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1023 if (ret < 0) {
1024 PERROR("write connection pipe");
1025 connection_put(new_conn);
1026 goto error;
1027 }
1028 } while (node != NULL);
1029
1030 /* Futex wait on queue. Blocking call on futex() */
1031 health_poll_entry();
1032 futex_nto1_wait(&relay_conn_queue.futex);
1033 health_poll_exit();
1034 }
1035
1036 /* Normal exit, no error */
1037 err = 0;
1038
1039 error:
1040 error_testpoint:
1041 if (err) {
1042 health_error();
1043 ERR("Health error occurred in %s", __func__);
1044 }
1045 health_unregister(health_relayd);
1046 DBG("Dispatch thread dying");
1047 lttng_relay_stop_threads();
1048 return NULL;
1049 }
1050
1051 /*
1052 * Set index data from the control port to a given index object.
1053 */
1054 static int set_index_control_data(struct relay_index *index,
1055 struct lttcomm_relayd_index *data,
1056 struct relay_connection *conn)
1057 {
1058 struct ctf_packet_index index_data;
1059
1060 /*
1061 * The index on disk is encoded in big endian, so we don't need
1062 * to convert the data received on the network. The data_offset
1063 * value is NEVER modified here and is updated by the data
1064 * thread.
1065 */
1066 index_data.packet_size = data->packet_size;
1067 index_data.content_size = data->content_size;
1068 index_data.timestamp_begin = data->timestamp_begin;
1069 index_data.timestamp_end = data->timestamp_end;
1070 index_data.events_discarded = data->events_discarded;
1071 index_data.stream_id = data->stream_id;
1072
1073 if (conn->minor >= 8) {
1074 index->index_data.stream_instance_id = data->stream_instance_id;
1075 index->index_data.packet_seq_num = data->packet_seq_num;
1076 }
1077
1078 return relay_index_set_data(index, &index_data);
1079 }
1080
1081 /*
1082 * Handle the RELAYD_CREATE_SESSION command.
1083 *
1084 * On success, send back the session id or else return a negative value.
1085 */
1086 static int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
1087 struct relay_connection *conn)
1088 {
1089 int ret = 0, send_ret;
1090 struct relay_session *session;
1091 struct lttcomm_relayd_status_session reply;
1092 char session_name[LTTNG_NAME_MAX];
1093 char hostname[LTTNG_HOST_NAME_MAX];
1094 uint32_t live_timer = 0;
1095 bool snapshot = false;
1096
1097 memset(session_name, 0, LTTNG_NAME_MAX);
1098 memset(hostname, 0, LTTNG_HOST_NAME_MAX);
1099
1100 memset(&reply, 0, sizeof(reply));
1101
1102 switch (conn->minor) {
1103 case 1:
1104 case 2:
1105 case 3:
1106 break;
1107 case 4: /* LTTng sessiond 2.4 */
1108 default:
1109 ret = cmd_create_session_2_4(conn, session_name,
1110 hostname, &live_timer, &snapshot);
1111 }
1112 if (ret < 0) {
1113 goto send_reply;
1114 }
1115
1116 session = session_create(session_name, hostname, live_timer,
1117 snapshot, conn->major, conn->minor);
1118 if (!session) {
1119 ret = -1;
1120 goto send_reply;
1121 }
1122 assert(!conn->session);
1123 conn->session = session;
1124 DBG("Created session %" PRIu64, session->id);
1125
1126 reply.session_id = htobe64(session->id);
1127
1128 send_reply:
1129 if (ret < 0) {
1130 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1131 } else {
1132 reply.ret_code = htobe32(LTTNG_OK);
1133 }
1134
1135 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1136 if (send_ret < 0) {
1137 ERR("Relayd sending session id");
1138 ret = send_ret;
1139 }
1140
1141 return ret;
1142 }
1143
1144 /*
1145 * When we have received all the streams and the metadata for a channel,
1146 * we make them visible to the viewer threads.
1147 */
1148 static void publish_connection_local_streams(struct relay_connection *conn)
1149 {
1150 struct relay_stream *stream;
1151 struct relay_session *session = conn->session;
1152
1153 /*
1154 * We publish all streams belonging to a session atomically wrt
1155 * session lock.
1156 */
1157 pthread_mutex_lock(&session->lock);
1158 rcu_read_lock();
1159 cds_list_for_each_entry_rcu(stream, &session->recv_list,
1160 recv_node) {
1161 stream_publish(stream);
1162 }
1163 rcu_read_unlock();
1164
1165 /*
1166 * Inform the viewer that there are new streams in the session.
1167 */
1168 if (session->viewer_attached) {
1169 uatomic_set(&session->new_streams, 1);
1170 }
1171 pthread_mutex_unlock(&session->lock);
1172 }
1173
1174 /*
1175 * relay_add_stream: allocate a new stream for a session
1176 */
1177 static int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1178 struct relay_connection *conn)
1179 {
1180 int ret;
1181 ssize_t send_ret;
1182 struct relay_session *session = conn->session;
1183 struct relay_stream *stream = NULL;
1184 struct lttcomm_relayd_status_stream reply;
1185 struct ctf_trace *trace = NULL;
1186 uint64_t stream_handle = -1ULL;
1187 char *path_name = NULL, *channel_name = NULL;
1188 uint64_t tracefile_size = 0, tracefile_count = 0;
1189
1190 if (!session || conn->version_check_done == 0) {
1191 ERR("Trying to add a stream before version check");
1192 ret = -1;
1193 goto end_no_session;
1194 }
1195
1196 switch (session->minor) {
1197 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
1198 ret = cmd_recv_stream_2_1(conn, &path_name,
1199 &channel_name);
1200 break;
1201 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
1202 default:
1203 ret = cmd_recv_stream_2_2(conn, &path_name,
1204 &channel_name, &tracefile_size, &tracefile_count);
1205 break;
1206 }
1207 if (ret < 0) {
1208 goto send_reply;
1209 }
1210
1211 trace = ctf_trace_get_by_path_or_create(session, path_name);
1212 if (!trace) {
1213 goto send_reply;
1214 }
1215 /* This stream here has one reference on the trace. */
1216
1217 pthread_mutex_lock(&last_relay_stream_id_lock);
1218 stream_handle = ++last_relay_stream_id;
1219 pthread_mutex_unlock(&last_relay_stream_id_lock);
1220
1221 /* We pass ownership of path_name and channel_name. */
1222 stream = stream_create(trace, stream_handle, path_name,
1223 channel_name, tracefile_size, tracefile_count);
1224 path_name = NULL;
1225 channel_name = NULL;
1226
1227 /*
1228 * Streams are the owners of their trace. Reference to trace is
1229 * kept within stream_create().
1230 */
1231 ctf_trace_put(trace);
1232
1233 send_reply:
1234 memset(&reply, 0, sizeof(reply));
1235 reply.handle = htobe64(stream_handle);
1236 if (!stream) {
1237 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1238 } else {
1239 reply.ret_code = htobe32(LTTNG_OK);
1240 }
1241
1242 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1243 sizeof(struct lttcomm_relayd_status_stream), 0);
1244 if (send_ret < 0) {
1245 ERR("Relay sending stream id");
1246 ret = (int) send_ret;
1247 }
1248
1249 end_no_session:
1250 free(path_name);
1251 free(channel_name);
1252 return ret;
1253 }
1254
1255 /*
1256 * relay_close_stream: close a specific stream
1257 */
1258 static int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1259 struct relay_connection *conn)
1260 {
1261 int ret, send_ret;
1262 struct relay_session *session = conn->session;
1263 struct lttcomm_relayd_close_stream stream_info;
1264 struct lttcomm_relayd_generic_reply reply;
1265 struct relay_stream *stream;
1266
1267 DBG("Close stream received");
1268
1269 if (!session || conn->version_check_done == 0) {
1270 ERR("Trying to close a stream before version check");
1271 ret = -1;
1272 goto end_no_session;
1273 }
1274
1275 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1276 sizeof(struct lttcomm_relayd_close_stream), 0);
1277 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1278 if (ret == 0) {
1279 /* Orderly shutdown. Not necessary to print an error. */
1280 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1281 } else {
1282 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1283 }
1284 ret = -1;
1285 goto end_no_session;
1286 }
1287
1288 stream = stream_get_by_id(be64toh(stream_info.stream_id));
1289 if (!stream) {
1290 ret = -1;
1291 goto end;
1292 }
1293
1294 /*
1295 * Set last_net_seq_num before the close flag. Required by data
1296 * pending check.
1297 */
1298 pthread_mutex_lock(&stream->lock);
1299 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1300 pthread_mutex_unlock(&stream->lock);
1301
1302 /*
1303 * This is one of the conditions which may trigger a stream close
1304 * with the others being:
1305 * 1) A close command is received for a stream
1306 * 2) The control connection owning the stream is closed
1307 * 3) We have received all of the stream's data _after_ a close
1308 * request.
1309 */
1310 try_stream_close(stream);
1311 if (stream->is_metadata) {
1312 struct relay_viewer_stream *vstream;
1313
1314 vstream = viewer_stream_get_by_id(stream->stream_handle);
1315 if (vstream) {
1316 if (vstream->metadata_sent == stream->metadata_received) {
1317 /*
1318 * Since all the metadata has been sent to the
1319 * viewer and that we have a request to close
1320 * its stream, we can safely teardown the
1321 * corresponding metadata viewer stream.
1322 */
1323 viewer_stream_put(vstream);
1324 }
1325 /* Put local reference. */
1326 viewer_stream_put(vstream);
1327 }
1328 }
1329 stream_put(stream);
1330
1331 end:
1332 memset(&reply, 0, sizeof(reply));
1333 if (ret < 0) {
1334 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1335 } else {
1336 reply.ret_code = htobe32(LTTNG_OK);
1337 }
1338 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1339 sizeof(struct lttcomm_relayd_generic_reply), 0);
1340 if (send_ret < 0) {
1341 ERR("Relay sending stream id");
1342 ret = send_ret;
1343 }
1344
1345 end_no_session:
1346 return ret;
1347 }
1348
1349 /*
1350 * relay_reset_metadata: reset a metadata stream
1351 */
1352 static
1353 int relay_reset_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1354 struct relay_connection *conn)
1355 {
1356 int ret, send_ret;
1357 struct relay_session *session = conn->session;
1358 struct lttcomm_relayd_reset_metadata stream_info;
1359 struct lttcomm_relayd_generic_reply reply;
1360 struct relay_stream *stream;
1361
1362 DBG("Reset metadata received");
1363
1364 if (!session || conn->version_check_done == 0) {
1365 ERR("Trying to reset a metadata stream before version check");
1366 ret = -1;
1367 goto end_no_session;
1368 }
1369
1370 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1371 sizeof(struct lttcomm_relayd_reset_metadata), 0);
1372 if (ret < sizeof(struct lttcomm_relayd_reset_metadata)) {
1373 if (ret == 0) {
1374 /* Orderly shutdown. Not necessary to print an error. */
1375 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1376 } else {
1377 ERR("Relay didn't receive valid reset_metadata struct "
1378 "size : %d", ret);
1379 }
1380 ret = -1;
1381 goto end_no_session;
1382 }
1383 DBG("Update metadata to version %" PRIu64, be64toh(stream_info.version));
1384
1385 /* Unsupported for live sessions for now. */
1386 if (session->live_timer != 0) {
1387 ret = -1;
1388 goto end;
1389 }
1390
1391 stream = stream_get_by_id(be64toh(stream_info.stream_id));
1392 if (!stream) {
1393 ret = -1;
1394 goto end;
1395 }
1396 pthread_mutex_lock(&stream->lock);
1397 if (!stream->is_metadata) {
1398 ret = -1;
1399 goto end_unlock;
1400 }
1401
1402 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
1403 0, 0, -1, -1, stream->stream_fd->fd, NULL,
1404 &stream->stream_fd->fd);
1405 if (ret < 0) {
1406 ERR("Failed to rotate metadata file %s of channel %s",
1407 stream->path_name, stream->channel_name);
1408 goto end_unlock;
1409 }
1410
1411 end_unlock:
1412 pthread_mutex_unlock(&stream->lock);
1413 stream_put(stream);
1414
1415 end:
1416 memset(&reply, 0, sizeof(reply));
1417 if (ret < 0) {
1418 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1419 } else {
1420 reply.ret_code = htobe32(LTTNG_OK);
1421 }
1422 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1423 sizeof(struct lttcomm_relayd_generic_reply), 0);
1424 if (send_ret < 0) {
1425 ERR("Relay sending reset metadata reply");
1426 ret = send_ret;
1427 }
1428
1429 end_no_session:
1430 return ret;
1431 }
1432
1433 /*
1434 * relay_unknown_command: send -1 if received unknown command
1435 */
1436 static void relay_unknown_command(struct relay_connection *conn)
1437 {
1438 struct lttcomm_relayd_generic_reply reply;
1439 int ret;
1440
1441 memset(&reply, 0, sizeof(reply));
1442 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1443 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1444 sizeof(struct lttcomm_relayd_generic_reply), 0);
1445 if (ret < 0) {
1446 ERR("Relay sending unknown command");
1447 }
1448 }
1449
1450 /*
1451 * relay_start: send an acknowledgment to the client to tell if we are
1452 * ready to receive data. We are ready if a session is established.
1453 */
1454 static int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1455 struct relay_connection *conn)
1456 {
1457 int ret = htobe32(LTTNG_OK);
1458 struct lttcomm_relayd_generic_reply reply;
1459 struct relay_session *session = conn->session;
1460
1461 if (!session) {
1462 DBG("Trying to start the streaming without a session established");
1463 ret = htobe32(LTTNG_ERR_UNK);
1464 }
1465
1466 memset(&reply, 0, sizeof(reply));
1467 reply.ret_code = ret;
1468 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1469 sizeof(struct lttcomm_relayd_generic_reply), 0);
1470 if (ret < 0) {
1471 ERR("Relay sending start ack");
1472 }
1473
1474 return ret;
1475 }
1476
1477 /*
1478 * Append padding to the file pointed by the file descriptor fd.
1479 */
1480 static int write_padding_to_file(int fd, uint32_t size)
1481 {
1482 ssize_t ret = 0;
1483 char *zeros;
1484
1485 if (size == 0) {
1486 goto end;
1487 }
1488
1489 zeros = zmalloc(size);
1490 if (zeros == NULL) {
1491 PERROR("zmalloc zeros for padding");
1492 ret = -1;
1493 goto end;
1494 }
1495
1496 ret = lttng_write(fd, zeros, size);
1497 if (ret < size) {
1498 PERROR("write padding to file");
1499 }
1500
1501 free(zeros);
1502
1503 end:
1504 return ret;
1505 }
1506
1507 /*
1508 * relay_recv_metadata: receive the metadata for the session.
1509 */
1510 static int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1511 struct relay_connection *conn)
1512 {
1513 int ret = 0;
1514 ssize_t size_ret;
1515 struct relay_session *session = conn->session;
1516 struct lttcomm_relayd_metadata_payload *metadata_struct;
1517 struct relay_stream *metadata_stream;
1518 uint64_t data_size, payload_size;
1519
1520 if (!session) {
1521 ERR("Metadata sent before version check");
1522 ret = -1;
1523 goto end;
1524 }
1525
1526 data_size = payload_size = be64toh(recv_hdr->data_size);
1527 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1528 ERR("Incorrect data size");
1529 ret = -1;
1530 goto end;
1531 }
1532 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1533
1534 if (data_buffer_size < data_size) {
1535 /* In case the realloc fails, we can free the memory */
1536 char *tmp_data_ptr;
1537
1538 tmp_data_ptr = realloc(data_buffer, data_size);
1539 if (!tmp_data_ptr) {
1540 ERR("Allocating data buffer");
1541 free(data_buffer);
1542 ret = -1;
1543 goto end;
1544 }
1545 data_buffer = tmp_data_ptr;
1546 data_buffer_size = data_size;
1547 }
1548 memset(data_buffer, 0, data_size);
1549 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1550 size_ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
1551 if (size_ret < 0 || size_ret != data_size) {
1552 if (size_ret == 0) {
1553 /* Orderly shutdown. Not necessary to print an error. */
1554 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1555 } else {
1556 ERR("Relay didn't receive the whole metadata");
1557 }
1558 ret = -1;
1559 goto end;
1560 }
1561 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1562
1563 metadata_stream = stream_get_by_id(be64toh(metadata_struct->stream_id));
1564 if (!metadata_stream) {
1565 ret = -1;
1566 goto end;
1567 }
1568
1569 pthread_mutex_lock(&metadata_stream->lock);
1570
1571 size_ret = lttng_write(metadata_stream->stream_fd->fd, metadata_struct->payload,
1572 payload_size);
1573 if (size_ret < payload_size) {
1574 ERR("Relay error writing metadata on file");
1575 ret = -1;
1576 goto end_put;
1577 }
1578
1579 size_ret = write_padding_to_file(metadata_stream->stream_fd->fd,
1580 be32toh(metadata_struct->padding_size));
1581 if (size_ret < 0) {
1582 goto end_put;
1583 }
1584
1585 metadata_stream->metadata_received +=
1586 payload_size + be32toh(metadata_struct->padding_size);
1587 DBG2("Relay metadata written. Updated metadata_received %" PRIu64,
1588 metadata_stream->metadata_received);
1589
1590 end_put:
1591 pthread_mutex_unlock(&metadata_stream->lock);
1592 stream_put(metadata_stream);
1593 end:
1594 return ret;
1595 }
1596
1597 /*
1598 * relay_send_version: send relayd version number
1599 */
1600 static int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1601 struct relay_connection *conn)
1602 {
1603 int ret;
1604 struct lttcomm_relayd_version reply, msg;
1605 bool compatible = true;
1606
1607 conn->version_check_done = 1;
1608
1609 /* Get version from the other side. */
1610 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1611 if (ret < 0 || ret != sizeof(msg)) {
1612 if (ret == 0) {
1613 /* Orderly shutdown. Not necessary to print an error. */
1614 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1615 } else {
1616 ERR("Relay failed to receive the version values.");
1617 }
1618 ret = -1;
1619 goto end;
1620 }
1621
1622 memset(&reply, 0, sizeof(reply));
1623 reply.major = RELAYD_VERSION_COMM_MAJOR;
1624 reply.minor = RELAYD_VERSION_COMM_MINOR;
1625
1626 /* Major versions must be the same */
1627 if (reply.major != be32toh(msg.major)) {
1628 DBG("Incompatible major versions (%u vs %u), deleting session",
1629 reply.major, be32toh(msg.major));
1630 compatible = false;
1631 }
1632
1633 conn->major = reply.major;
1634 /* We adapt to the lowest compatible version */
1635 if (reply.minor <= be32toh(msg.minor)) {
1636 conn->minor = reply.minor;
1637 } else {
1638 conn->minor = be32toh(msg.minor);
1639 }
1640
1641 reply.major = htobe32(reply.major);
1642 reply.minor = htobe32(reply.minor);
1643 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1644 sizeof(struct lttcomm_relayd_version), 0);
1645 if (ret < 0) {
1646 ERR("Relay sending version");
1647 }
1648
1649 if (!compatible) {
1650 ret = -1;
1651 goto end;
1652 }
1653
1654 DBG("Version check done using protocol %u.%u", conn->major,
1655 conn->minor);
1656
1657 end:
1658 return ret;
1659 }
1660
1661 /*
1662 * Check for data pending for a given stream id from the session daemon.
1663 */
1664 static int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1665 struct relay_connection *conn)
1666 {
1667 struct relay_session *session = conn->session;
1668 struct lttcomm_relayd_data_pending msg;
1669 struct lttcomm_relayd_generic_reply reply;
1670 struct relay_stream *stream;
1671 int ret;
1672 uint64_t last_net_seq_num, stream_id;
1673
1674 DBG("Data pending command received");
1675
1676 if (!session || conn->version_check_done == 0) {
1677 ERR("Trying to check for data before version check");
1678 ret = -1;
1679 goto end_no_session;
1680 }
1681
1682 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1683 if (ret < sizeof(msg)) {
1684 if (ret == 0) {
1685 /* Orderly shutdown. Not necessary to print an error. */
1686 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1687 } else {
1688 ERR("Relay didn't receive valid data_pending struct size : %d",
1689 ret);
1690 }
1691 ret = -1;
1692 goto end_no_session;
1693 }
1694
1695 stream_id = be64toh(msg.stream_id);
1696 last_net_seq_num = be64toh(msg.last_net_seq_num);
1697
1698 stream = stream_get_by_id(stream_id);
1699 if (stream == NULL) {
1700 ret = -1;
1701 goto end;
1702 }
1703
1704 pthread_mutex_lock(&stream->lock);
1705
1706 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1707 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1708 last_net_seq_num);
1709
1710 /* Avoid wrapping issue */
1711 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1712 /* Data has in fact been written and is NOT pending */
1713 ret = 0;
1714 } else {
1715 /* Data still being streamed thus pending */
1716 ret = 1;
1717 }
1718
1719 stream->data_pending_check_done = true;
1720 pthread_mutex_unlock(&stream->lock);
1721
1722 stream_put(stream);
1723 end:
1724
1725 memset(&reply, 0, sizeof(reply));
1726 reply.ret_code = htobe32(ret);
1727 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1728 if (ret < 0) {
1729 ERR("Relay data pending ret code failed");
1730 }
1731
1732 end_no_session:
1733 return ret;
1734 }
1735
1736 /*
1737 * Wait for the control socket to reach a quiescent state.
1738 *
1739 * Note that for now, when receiving this command from the session
1740 * daemon, this means that every subsequent commands or data received on
1741 * the control socket has been handled. So, this is why we simply return
1742 * OK here.
1743 */
1744 static int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1745 struct relay_connection *conn)
1746 {
1747 int ret;
1748 uint64_t stream_id;
1749 struct relay_stream *stream;
1750 struct lttcomm_relayd_quiescent_control msg;
1751 struct lttcomm_relayd_generic_reply reply;
1752
1753 DBG("Checking quiescent state on control socket");
1754
1755 if (!conn->session || conn->version_check_done == 0) {
1756 ERR("Trying to check for data before version check");
1757 ret = -1;
1758 goto end_no_session;
1759 }
1760
1761 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1762 if (ret < sizeof(msg)) {
1763 if (ret == 0) {
1764 /* Orderly shutdown. Not necessary to print an error. */
1765 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1766 } else {
1767 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1768 ret);
1769 }
1770 ret = -1;
1771 goto end_no_session;
1772 }
1773
1774 stream_id = be64toh(msg.stream_id);
1775 stream = stream_get_by_id(stream_id);
1776 if (!stream) {
1777 goto reply;
1778 }
1779 pthread_mutex_lock(&stream->lock);
1780 stream->data_pending_check_done = true;
1781 pthread_mutex_unlock(&stream->lock);
1782 DBG("Relay quiescent control pending flag set to %" PRIu64, stream_id);
1783 stream_put(stream);
1784 reply:
1785 memset(&reply, 0, sizeof(reply));
1786 reply.ret_code = htobe32(LTTNG_OK);
1787 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1788 if (ret < 0) {
1789 ERR("Relay data quiescent control ret code failed");
1790 }
1791
1792 end_no_session:
1793 return ret;
1794 }
1795
1796 /*
1797 * Initialize a data pending command. This means that a consumer is about
1798 * to ask for data pending for each stream it holds. Simply iterate over
1799 * all streams of a session and set the data_pending_check_done flag.
1800 *
1801 * This command returns to the client a LTTNG_OK code.
1802 */
1803 static int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1804 struct relay_connection *conn)
1805 {
1806 int ret;
1807 struct lttng_ht_iter iter;
1808 struct lttcomm_relayd_begin_data_pending msg;
1809 struct lttcomm_relayd_generic_reply reply;
1810 struct relay_stream *stream;
1811 uint64_t session_id;
1812
1813 assert(recv_hdr);
1814 assert(conn);
1815
1816 DBG("Init streams for data pending");
1817
1818 if (!conn->session || conn->version_check_done == 0) {
1819 ERR("Trying to check for data before version check");
1820 ret = -1;
1821 goto end_no_session;
1822 }
1823
1824 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1825 if (ret < sizeof(msg)) {
1826 if (ret == 0) {
1827 /* Orderly shutdown. Not necessary to print an error. */
1828 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1829 } else {
1830 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1831 ret);
1832 }
1833 ret = -1;
1834 goto end_no_session;
1835 }
1836
1837 session_id = be64toh(msg.session_id);
1838
1839 /*
1840 * Iterate over all streams to set the begin data pending flag.
1841 * For now, the streams are indexed by stream handle so we have
1842 * to iterate over all streams to find the one associated with
1843 * the right session_id.
1844 */
1845 rcu_read_lock();
1846 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1847 node.node) {
1848 if (!stream_get(stream)) {
1849 continue;
1850 }
1851 if (stream->trace->session->id == session_id) {
1852 pthread_mutex_lock(&stream->lock);
1853 stream->data_pending_check_done = false;
1854 pthread_mutex_unlock(&stream->lock);
1855 DBG("Set begin data pending flag to stream %" PRIu64,
1856 stream->stream_handle);
1857 }
1858 stream_put(stream);
1859 }
1860 rcu_read_unlock();
1861
1862 memset(&reply, 0, sizeof(reply));
1863 /* All good, send back reply. */
1864 reply.ret_code = htobe32(LTTNG_OK);
1865
1866 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1867 if (ret < 0) {
1868 ERR("Relay begin data pending send reply failed");
1869 }
1870
1871 end_no_session:
1872 return ret;
1873 }
1874
1875 /*
1876 * End data pending command. This will check, for a given session id, if
1877 * each stream associated with it has its data_pending_check_done flag
1878 * set. If not, this means that the client lost track of the stream but
1879 * the data is still being streamed on our side. In this case, we inform
1880 * the client that data is in flight.
1881 *
1882 * Return to the client if there is data in flight or not with a ret_code.
1883 */
1884 static int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1885 struct relay_connection *conn)
1886 {
1887 int ret;
1888 struct lttng_ht_iter iter;
1889 struct lttcomm_relayd_end_data_pending msg;
1890 struct lttcomm_relayd_generic_reply reply;
1891 struct relay_stream *stream;
1892 uint64_t session_id;
1893 uint32_t is_data_inflight = 0;
1894
1895 DBG("End data pending command");
1896
1897 if (!conn->session || conn->version_check_done == 0) {
1898 ERR("Trying to check for data before version check");
1899 ret = -1;
1900 goto end_no_session;
1901 }
1902
1903 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1904 if (ret < sizeof(msg)) {
1905 if (ret == 0) {
1906 /* Orderly shutdown. Not necessary to print an error. */
1907 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1908 } else {
1909 ERR("Relay didn't receive valid end data_pending struct size: %d",
1910 ret);
1911 }
1912 ret = -1;
1913 goto end_no_session;
1914 }
1915
1916 session_id = be64toh(msg.session_id);
1917
1918 /*
1919 * Iterate over all streams to see if the begin data pending
1920 * flag is set.
1921 */
1922 rcu_read_lock();
1923 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1924 node.node) {
1925 if (!stream_get(stream)) {
1926 continue;
1927 }
1928 if (stream->trace->session->id != session_id) {
1929 stream_put(stream);
1930 continue;
1931 }
1932 pthread_mutex_lock(&stream->lock);
1933 if (!stream->data_pending_check_done) {
1934 if (!stream->closed || !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
1935 is_data_inflight = 1;
1936 DBG("Data is still in flight for stream %" PRIu64,
1937 stream->stream_handle);
1938 pthread_mutex_unlock(&stream->lock);
1939 stream_put(stream);
1940 break;
1941 }
1942 }
1943 pthread_mutex_unlock(&stream->lock);
1944 stream_put(stream);
1945 }
1946 rcu_read_unlock();
1947
1948 memset(&reply, 0, sizeof(reply));
1949 /* All good, send back reply. */
1950 reply.ret_code = htobe32(is_data_inflight);
1951
1952 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1953 if (ret < 0) {
1954 ERR("Relay end data pending send reply failed");
1955 }
1956
1957 end_no_session:
1958 return ret;
1959 }
1960
1961 /*
1962 * Receive an index for a specific stream.
1963 *
1964 * Return 0 on success else a negative value.
1965 */
1966 static int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1967 struct relay_connection *conn)
1968 {
1969 int ret, send_ret;
1970 struct relay_session *session = conn->session;
1971 struct lttcomm_relayd_index index_info;
1972 struct relay_index *index;
1973 struct lttcomm_relayd_generic_reply reply;
1974 struct relay_stream *stream;
1975 uint64_t net_seq_num;
1976 size_t msg_len;
1977
1978 assert(conn);
1979
1980 DBG("Relay receiving index");
1981
1982 if (!session || conn->version_check_done == 0) {
1983 ERR("Trying to close a stream before version check");
1984 ret = -1;
1985 goto end_no_session;
1986 }
1987
1988 msg_len = lttcomm_relayd_index_len(
1989 lttng_to_index_major(conn->major, conn->minor),
1990 lttng_to_index_minor(conn->major, conn->minor));
1991 ret = conn->sock->ops->recvmsg(conn->sock, &index_info,
1992 msg_len, 0);
1993 if (ret < msg_len) {
1994 if (ret == 0) {
1995 /* Orderly shutdown. Not necessary to print an error. */
1996 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1997 } else {
1998 ERR("Relay didn't receive valid index struct size : %d", ret);
1999 }
2000 ret = -1;
2001 goto end_no_session;
2002 }
2003
2004 net_seq_num = be64toh(index_info.net_seq_num);
2005
2006 stream = stream_get_by_id(be64toh(index_info.relay_stream_id));
2007 if (!stream) {
2008 ERR("stream_get_by_id not found");
2009 ret = -1;
2010 goto end;
2011 }
2012 pthread_mutex_lock(&stream->lock);
2013
2014 /* Live beacon handling */
2015 if (index_info.packet_size == 0) {
2016 DBG("Received live beacon for stream %" PRIu64,
2017 stream->stream_handle);
2018
2019 /*
2020 * Only flag a stream inactive when it has already
2021 * received data and no indexes are in flight.
2022 */
2023 if (stream->index_received_seqcount > 0
2024 && stream->indexes_in_flight == 0) {
2025 stream->beacon_ts_end =
2026 be64toh(index_info.timestamp_end);
2027 }
2028 ret = 0;
2029 goto end_stream_put;
2030 } else {
2031 stream->beacon_ts_end = -1ULL;
2032 }
2033
2034 if (stream->ctf_stream_id == -1ULL) {
2035 stream->ctf_stream_id = be64toh(index_info.stream_id);
2036 }
2037 index = relay_index_get_by_id_or_create(stream, net_seq_num);
2038 if (!index) {
2039 ret = -1;
2040 ERR("relay_index_get_by_id_or_create index NULL");
2041 goto end_stream_put;
2042 }
2043 if (set_index_control_data(index, &index_info, conn)) {
2044 ERR("set_index_control_data error");
2045 relay_index_put(index);
2046 ret = -1;
2047 goto end_stream_put;
2048 }
2049 ret = relay_index_try_flush(index);
2050 if (ret == 0) {
2051 tracefile_array_commit_seq(stream->tfa);
2052 stream->index_received_seqcount++;
2053 } else if (ret > 0) {
2054 /* no flush. */
2055 ret = 0;
2056 } else {
2057 ERR("relay_index_try_flush error %d", ret);
2058 relay_index_put(index);
2059 ret = -1;
2060 }
2061
2062 end_stream_put:
2063 pthread_mutex_unlock(&stream->lock);
2064 stream_put(stream);
2065
2066 end:
2067
2068 memset(&reply, 0, sizeof(reply));
2069 if (ret < 0) {
2070 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2071 } else {
2072 reply.ret_code = htobe32(LTTNG_OK);
2073 }
2074 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2075 if (send_ret < 0) {
2076 ERR("Relay sending close index id reply");
2077 ret = send_ret;
2078 }
2079
2080 end_no_session:
2081 return ret;
2082 }
2083
2084 /*
2085 * Receive the streams_sent message.
2086 *
2087 * Return 0 on success else a negative value.
2088 */
2089 static int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
2090 struct relay_connection *conn)
2091 {
2092 int ret, send_ret;
2093 struct lttcomm_relayd_generic_reply reply;
2094
2095 assert(conn);
2096
2097 DBG("Relay receiving streams_sent");
2098
2099 if (!conn->session || conn->version_check_done == 0) {
2100 ERR("Trying to close a stream before version check");
2101 ret = -1;
2102 goto end_no_session;
2103 }
2104
2105 /*
2106 * Publish every pending stream in the connection recv list which are
2107 * now ready to be used by the viewer.
2108 */
2109 publish_connection_local_streams(conn);
2110
2111 memset(&reply, 0, sizeof(reply));
2112 reply.ret_code = htobe32(LTTNG_OK);
2113 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2114 if (send_ret < 0) {
2115 ERR("Relay sending sent_stream reply");
2116 ret = send_ret;
2117 } else {
2118 /* Success. */
2119 ret = 0;
2120 }
2121
2122 end_no_session:
2123 return ret;
2124 }
2125
2126 /*
2127 * Process the commands received on the control socket
2128 */
2129 static int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
2130 struct relay_connection *conn)
2131 {
2132 int ret = 0;
2133
2134 switch (be32toh(recv_hdr->cmd)) {
2135 case RELAYD_CREATE_SESSION:
2136 ret = relay_create_session(recv_hdr, conn);
2137 break;
2138 case RELAYD_ADD_STREAM:
2139 ret = relay_add_stream(recv_hdr, conn);
2140 break;
2141 case RELAYD_START_DATA:
2142 ret = relay_start(recv_hdr, conn);
2143 break;
2144 case RELAYD_SEND_METADATA:
2145 ret = relay_recv_metadata(recv_hdr, conn);
2146 break;
2147 case RELAYD_VERSION:
2148 ret = relay_send_version(recv_hdr, conn);
2149 break;
2150 case RELAYD_CLOSE_STREAM:
2151 ret = relay_close_stream(recv_hdr, conn);
2152 break;
2153 case RELAYD_DATA_PENDING:
2154 ret = relay_data_pending(recv_hdr, conn);
2155 break;
2156 case RELAYD_QUIESCENT_CONTROL:
2157 ret = relay_quiescent_control(recv_hdr, conn);
2158 break;
2159 case RELAYD_BEGIN_DATA_PENDING:
2160 ret = relay_begin_data_pending(recv_hdr, conn);
2161 break;
2162 case RELAYD_END_DATA_PENDING:
2163 ret = relay_end_data_pending(recv_hdr, conn);
2164 break;
2165 case RELAYD_SEND_INDEX:
2166 ret = relay_recv_index(recv_hdr, conn);
2167 break;
2168 case RELAYD_STREAMS_SENT:
2169 ret = relay_streams_sent(recv_hdr, conn);
2170 break;
2171 case RELAYD_RESET_METADATA:
2172 ret = relay_reset_metadata(recv_hdr, conn);
2173 break;
2174 case RELAYD_UPDATE_SYNC_INFO:
2175 default:
2176 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2177 relay_unknown_command(conn);
2178 ret = -1;
2179 goto end;
2180 }
2181
2182 end:
2183 return ret;
2184 }
2185
2186 /*
2187 * Handle index for a data stream.
2188 *
2189 * Called with the stream lock held.
2190 *
2191 * Return 0 on success else a negative value.
2192 */
2193 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2194 int rotate_index)
2195 {
2196 int ret = 0;
2197 uint64_t data_offset;
2198 struct relay_index *index;
2199
2200 /* Get data offset because we are about to update the index. */
2201 data_offset = htobe64(stream->tracefile_size_current);
2202
2203 DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64,
2204 stream->stream_handle, net_seq_num, stream->tracefile_size_current);
2205
2206 /*
2207 * Lookup for an existing index for that stream id/sequence
2208 * number. If it exists, the control thread has already received the
2209 * data for it, thus we need to write it to disk.
2210 */
2211 index = relay_index_get_by_id_or_create(stream, net_seq_num);
2212 if (!index) {
2213 ret = -1;
2214 goto end;
2215 }
2216
2217 if (rotate_index || !stream->index_file) {
2218 uint32_t major, minor;
2219
2220 /* Put ref on previous index_file. */
2221 if (stream->index_file) {
2222 lttng_index_file_put(stream->index_file);
2223 stream->index_file = NULL;
2224 }
2225 major = stream->trace->session->major;
2226 minor = stream->trace->session->minor;
2227 stream->index_file = lttng_index_file_create(stream->path_name,
2228 stream->channel_name,
2229 -1, -1, stream->tracefile_size,
2230 tracefile_array_get_file_index_head(stream->tfa),
2231 lttng_to_index_major(major, minor),
2232 lttng_to_index_minor(major, minor));
2233 if (!stream->index_file) {
2234 ret = -1;
2235 /* Put self-ref for this index due to error. */
2236 relay_index_put(index);
2237 index = NULL;
2238 goto end;
2239 }
2240 }
2241
2242 if (relay_index_set_file(index, stream->index_file, data_offset)) {
2243 ret = -1;
2244 /* Put self-ref for this index due to error. */
2245 relay_index_put(index);
2246 index = NULL;
2247 goto end;
2248 }
2249
2250 ret = relay_index_try_flush(index);
2251 if (ret == 0) {
2252 tracefile_array_commit_seq(stream->tfa);
2253 stream->index_received_seqcount++;
2254 } else if (ret > 0) {
2255 /* No flush. */
2256 ret = 0;
2257 } else {
2258 /* Put self-ref for this index due to error. */
2259 relay_index_put(index);
2260 index = NULL;
2261 ret = -1;
2262 }
2263 end:
2264 return ret;
2265 }
2266
2267 /*
2268 * relay_process_data: Process the data received on the data socket
2269 */
2270 static int relay_process_data(struct relay_connection *conn)
2271 {
2272 int ret = 0, rotate_index = 0;
2273 ssize_t size_ret;
2274 struct relay_stream *stream;
2275 struct lttcomm_relayd_data_hdr data_hdr;
2276 uint64_t stream_id;
2277 uint64_t net_seq_num;
2278 uint32_t data_size;
2279 struct relay_session *session;
2280 bool new_stream = false, close_requested = false;
2281 size_t chunk_size = RECV_DATA_BUFFER_SIZE;
2282 size_t recv_off = 0;
2283 char data_buffer[chunk_size];
2284
2285 ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr,
2286 sizeof(struct lttcomm_relayd_data_hdr), 0);
2287 if (ret <= 0) {
2288 if (ret == 0) {
2289 /* Orderly shutdown. Not necessary to print an error. */
2290 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2291 } else {
2292 ERR("Unable to receive data header on sock %d", conn->sock->fd);
2293 }
2294 ret = -1;
2295 goto end;
2296 }
2297
2298 stream_id = be64toh(data_hdr.stream_id);
2299 stream = stream_get_by_id(stream_id);
2300 if (!stream) {
2301 ERR("relay_process_data: Cannot find stream %" PRIu64, stream_id);
2302 ret = -1;
2303 goto end;
2304 }
2305 session = stream->trace->session;
2306 data_size = be32toh(data_hdr.data_size);
2307
2308 net_seq_num = be64toh(data_hdr.net_seq_num);
2309
2310 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2311 data_size, stream_id, net_seq_num);
2312
2313 pthread_mutex_lock(&stream->lock);
2314
2315 /* Check if a rotation is needed. */
2316 if (stream->tracefile_size > 0 &&
2317 (stream->tracefile_size_current + data_size) >
2318 stream->tracefile_size) {
2319 uint64_t old_id, new_id;
2320
2321 old_id = tracefile_array_get_file_index_head(stream->tfa);
2322 tracefile_array_file_rotate(stream->tfa);
2323
2324 /* new_id is updated by utils_rotate_stream_file. */
2325 new_id = old_id;
2326
2327 ret = utils_rotate_stream_file(stream->path_name,
2328 stream->channel_name, stream->tracefile_size,
2329 stream->tracefile_count, -1,
2330 -1, stream->stream_fd->fd,
2331 &new_id, &stream->stream_fd->fd);
2332 if (ret < 0) {
2333 ERR("Rotating stream output file");
2334 goto end_stream_unlock;
2335 }
2336 /*
2337 * Reset current size because we just performed a stream
2338 * rotation.
2339 */
2340 stream->tracefile_size_current = 0;
2341 rotate_index = 1;
2342 }
2343
2344 /*
2345 * Index are handled in protocol version 2.4 and above. Also,
2346 * snapshot and index are NOT supported.
2347 */
2348 if (session->minor >= 4 && !session->snapshot) {
2349 ret = handle_index_data(stream, net_seq_num, rotate_index);
2350 if (ret < 0) {
2351 ERR("handle_index_data: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
2352 stream->stream_handle, net_seq_num, ret);
2353 goto end_stream_unlock;
2354 }
2355 }
2356
2357 for (recv_off = 0; recv_off < data_size; recv_off += chunk_size) {
2358 size_t recv_size = min(data_size - recv_off, chunk_size);
2359
2360 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_size, 0);
2361 if (ret <= 0) {
2362 if (ret == 0) {
2363 /* Orderly shutdown. Not necessary to print an error. */
2364 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2365 } else {
2366 ERR("Socket %d error %d", conn->sock->fd, ret);
2367 }
2368 ret = -1;
2369 goto end_stream_unlock;
2370 }
2371
2372 /* Write data to stream output fd. */
2373 size_ret = lttng_write(stream->stream_fd->fd, data_buffer,
2374 recv_size);
2375 if (size_ret < recv_size) {
2376 ERR("Relay error writing data to file");
2377 ret = -1;
2378 goto end_stream_unlock;
2379 }
2380
2381 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64,
2382 size_ret, stream->stream_handle);
2383 }
2384
2385 ret = write_padding_to_file(stream->stream_fd->fd,
2386 be32toh(data_hdr.padding_size));
2387 if (ret < 0) {
2388 ERR("write_padding_to_file: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
2389 stream->stream_handle, net_seq_num, ret);
2390 goto end_stream_unlock;
2391 }
2392 stream->tracefile_size_current +=
2393 data_size + be32toh(data_hdr.padding_size);
2394 if (stream->prev_seq == -1ULL) {
2395 new_stream = true;
2396 }
2397
2398 stream->prev_seq = net_seq_num;
2399
2400 end_stream_unlock:
2401 close_requested = stream->close_requested;
2402 pthread_mutex_unlock(&stream->lock);
2403 if (close_requested) {
2404 try_stream_close(stream);
2405 }
2406
2407 if (new_stream) {
2408 pthread_mutex_lock(&session->lock);
2409 uatomic_set(&session->new_streams, 1);
2410 pthread_mutex_unlock(&session->lock);
2411 }
2412 stream_put(stream);
2413 end:
2414 return ret;
2415 }
2416
2417 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2418 {
2419 int ret;
2420
2421 (void) lttng_poll_del(events, pollfd);
2422
2423 ret = close(pollfd);
2424 if (ret < 0) {
2425 ERR("Closing pollfd %d", pollfd);
2426 }
2427 }
2428
2429 static void relay_thread_close_connection(struct lttng_poll_event *events,
2430 int pollfd, struct relay_connection *conn)
2431 {
2432 const char *type_str;
2433
2434 switch (conn->type) {
2435 case RELAY_DATA:
2436 type_str = "Data";
2437 break;
2438 case RELAY_CONTROL:
2439 type_str = "Control";
2440 break;
2441 case RELAY_VIEWER_COMMAND:
2442 type_str = "Viewer Command";
2443 break;
2444 case RELAY_VIEWER_NOTIFICATION:
2445 type_str = "Viewer Notification";
2446 break;
2447 default:
2448 type_str = "Unknown";
2449 }
2450 cleanup_connection_pollfd(events, pollfd);
2451 connection_put(conn);
2452 DBG("%s connection closed with %d", type_str, pollfd);
2453 }
2454
2455 /*
2456 * This thread does the actual work
2457 */
2458 static void *relay_thread_worker(void *data)
2459 {
2460 int ret, err = -1, last_seen_data_fd = -1;
2461 uint32_t nb_fd;
2462 struct lttng_poll_event events;
2463 struct lttng_ht *relay_connections_ht;
2464 struct lttng_ht_iter iter;
2465 struct lttcomm_relayd_hdr recv_hdr;
2466 struct relay_connection *destroy_conn = NULL;
2467
2468 DBG("[thread] Relay worker started");
2469
2470 rcu_register_thread();
2471
2472 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2473
2474 if (testpoint(relayd_thread_worker)) {
2475 goto error_testpoint;
2476 }
2477
2478 health_code_update();
2479
2480 /* table of connections indexed on socket */
2481 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2482 if (!relay_connections_ht) {
2483 goto relay_connections_ht_error;
2484 }
2485
2486 ret = create_thread_poll_set(&events, 2);
2487 if (ret < 0) {
2488 goto error_poll_create;
2489 }
2490
2491 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2492 if (ret < 0) {
2493 goto error;
2494 }
2495
2496 restart:
2497 while (1) {
2498 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2499
2500 health_code_update();
2501
2502 /* Infinite blocking call, waiting for transmission */
2503 DBG3("Relayd worker thread polling...");
2504 health_poll_entry();
2505 ret = lttng_poll_wait(&events, -1);
2506 health_poll_exit();
2507 if (ret < 0) {
2508 /*
2509 * Restart interrupted system call.
2510 */
2511 if (errno == EINTR) {
2512 goto restart;
2513 }
2514 goto error;
2515 }
2516
2517 nb_fd = ret;
2518
2519 /*
2520 * Process control. The control connection is
2521 * prioritized so we don't starve it with high
2522 * throughput tracing data on the data connection.
2523 */
2524 for (i = 0; i < nb_fd; i++) {
2525 /* Fetch once the poll data */
2526 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2527 int pollfd = LTTNG_POLL_GETFD(&events, i);
2528
2529 health_code_update();
2530
2531 if (!revents) {
2532 /*
2533 * No activity for this FD (poll
2534 * implementation).
2535 */
2536 continue;
2537 }
2538
2539 /* Thread quit pipe has been closed. Killing thread. */
2540 ret = check_thread_quit_pipe(pollfd, revents);
2541 if (ret) {
2542 err = 0;
2543 goto exit;
2544 }
2545
2546 /* Inspect the relay conn pipe for new connection */
2547 if (pollfd == relay_conn_pipe[0]) {
2548 if (revents & LPOLLIN) {
2549 struct relay_connection *conn;
2550
2551 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
2552 if (ret < 0) {
2553 goto error;
2554 }
2555 lttng_poll_add(&events, conn->sock->fd,
2556 LPOLLIN | LPOLLRDHUP);
2557 connection_ht_add(relay_connections_ht, conn);
2558 DBG("Connection socket %d added", conn->sock->fd);
2559 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2560 ERR("Relay connection pipe error");
2561 goto error;
2562 } else {
2563 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2564 goto error;
2565 }
2566 } else {
2567 struct relay_connection *ctrl_conn;
2568
2569 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
2570 /* If not found, there is a synchronization issue. */
2571 assert(ctrl_conn);
2572
2573 if (ctrl_conn->type == RELAY_DATA) {
2574 if (revents & LPOLLIN) {
2575 /*
2576 * Flag the last seen data fd not deleted. It will be
2577 * used as the last seen fd if any fd gets deleted in
2578 * this first loop.
2579 */
2580 last_notdel_data_fd = pollfd;
2581 }
2582 goto put_ctrl_connection;
2583 }
2584 assert(ctrl_conn->type == RELAY_CONTROL);
2585
2586 if (revents & LPOLLIN) {
2587 ret = ctrl_conn->sock->ops->recvmsg(ctrl_conn->sock,
2588 &recv_hdr, sizeof(recv_hdr), 0);
2589 if (ret <= 0) {
2590 /* Connection closed */
2591 relay_thread_close_connection(&events, pollfd,
2592 ctrl_conn);
2593 } else {
2594 ret = relay_process_control(&recv_hdr, ctrl_conn);
2595 if (ret < 0) {
2596 /* Clear the session on error. */
2597 relay_thread_close_connection(&events,
2598 pollfd, ctrl_conn);
2599 }
2600 seen_control = 1;
2601 }
2602 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2603 relay_thread_close_connection(&events,
2604 pollfd, ctrl_conn);
2605 if (last_seen_data_fd == pollfd) {
2606 last_seen_data_fd = last_notdel_data_fd;
2607 }
2608 } else {
2609 ERR("Unexpected poll events %u for control sock %d",
2610 revents, pollfd);
2611 connection_put(ctrl_conn);
2612 goto error;
2613 }
2614 put_ctrl_connection:
2615 connection_put(ctrl_conn);
2616 }
2617 }
2618
2619 /*
2620 * The last loop handled a control request, go back to poll to make
2621 * sure we prioritise the control socket.
2622 */
2623 if (seen_control) {
2624 continue;
2625 }
2626
2627 if (last_seen_data_fd >= 0) {
2628 for (i = 0; i < nb_fd; i++) {
2629 int pollfd = LTTNG_POLL_GETFD(&events, i);
2630
2631 health_code_update();
2632
2633 if (last_seen_data_fd == pollfd) {
2634 idx = i;
2635 break;
2636 }
2637 }
2638 }
2639
2640 /* Process data connection. */
2641 for (i = idx + 1; i < nb_fd; i++) {
2642 /* Fetch the poll data. */
2643 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2644 int pollfd = LTTNG_POLL_GETFD(&events, i);
2645 struct relay_connection *data_conn;
2646
2647 health_code_update();
2648
2649 if (!revents) {
2650 /* No activity for this FD (poll implementation). */
2651 continue;
2652 }
2653
2654 /* Skip the command pipe. It's handled in the first loop. */
2655 if (pollfd == relay_conn_pipe[0]) {
2656 continue;
2657 }
2658
2659 data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
2660 if (!data_conn) {
2661 /* Skip it. Might be removed before. */
2662 continue;
2663 }
2664 if (data_conn->type == RELAY_CONTROL) {
2665 goto put_data_connection;
2666 }
2667 assert(data_conn->type == RELAY_DATA);
2668
2669 if (revents & LPOLLIN) {
2670 ret = relay_process_data(data_conn);
2671 /* Connection closed */
2672 if (ret < 0) {
2673 relay_thread_close_connection(&events, pollfd,
2674 data_conn);
2675 /*
2676 * Every goto restart call sets the last seen fd where
2677 * here we don't really care since we gracefully
2678 * continue the loop after the connection is deleted.
2679 */
2680 } else {
2681 /* Keep last seen port. */
2682 last_seen_data_fd = pollfd;
2683 connection_put(data_conn);
2684 goto restart;
2685 }
2686 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2687 relay_thread_close_connection(&events, pollfd,
2688 data_conn);
2689 } else {
2690 ERR("Unknown poll events %u for data sock %d",
2691 revents, pollfd);
2692 }
2693 put_data_connection:
2694 connection_put(data_conn);
2695 }
2696 last_seen_data_fd = -1;
2697 }
2698
2699 /* Normal exit, no error */
2700 ret = 0;
2701
2702 exit:
2703 error:
2704 /* Cleanup reamaining connection object. */
2705 rcu_read_lock();
2706 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
2707 destroy_conn,
2708 sock_n.node) {
2709 health_code_update();
2710
2711 if (session_abort(destroy_conn->session)) {
2712 assert(0);
2713 }
2714
2715 /*
2716 * No need to grab another ref, because we own
2717 * destroy_conn.
2718 */
2719 relay_thread_close_connection(&events, destroy_conn->sock->fd,
2720 destroy_conn);
2721 }
2722 rcu_read_unlock();
2723
2724 lttng_poll_clean(&events);
2725 error_poll_create:
2726 lttng_ht_destroy(relay_connections_ht);
2727 relay_connections_ht_error:
2728 /* Close relay conn pipes */
2729 utils_close_pipe(relay_conn_pipe);
2730 if (err) {
2731 DBG("Thread exited with error");
2732 }
2733 DBG("Worker thread cleanup complete");
2734 error_testpoint:
2735 if (err) {
2736 health_error();
2737 ERR("Health error occurred in %s", __func__);
2738 }
2739 health_unregister(health_relayd);
2740 rcu_unregister_thread();
2741 lttng_relay_stop_threads();
2742 return NULL;
2743 }
2744
2745 /*
2746 * Create the relay command pipe to wake thread_manage_apps.
2747 * Closed in cleanup().
2748 */
2749 static int create_relay_conn_pipe(void)
2750 {
2751 int ret;
2752
2753 ret = utils_create_pipe_cloexec(relay_conn_pipe);
2754
2755 return ret;
2756 }
2757
2758 /*
2759 * main
2760 */
2761 int main(int argc, char **argv)
2762 {
2763 int ret = 0, retval = 0;
2764 void *status;
2765
2766 /* Parse arguments */
2767 progname = argv[0];
2768 if (set_options(argc, argv)) {
2769 retval = -1;
2770 goto exit_options;
2771 }
2772
2773 if (set_signal_handler()) {
2774 retval = -1;
2775 goto exit_options;
2776 }
2777
2778 /* Try to create directory if -o, --output is specified. */
2779 if (opt_output_path) {
2780 if (*opt_output_path != '/') {
2781 ERR("Please specify an absolute path for -o, --output PATH");
2782 retval = -1;
2783 goto exit_options;
2784 }
2785
2786 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG,
2787 -1, -1);
2788 if (ret < 0) {
2789 ERR("Unable to create %s", opt_output_path);
2790 retval = -1;
2791 goto exit_options;
2792 }
2793 }
2794
2795 /* Daemonize */
2796 if (opt_daemon || opt_background) {
2797 int i;
2798
2799 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2800 !opt_background);
2801 if (ret < 0) {
2802 retval = -1;
2803 goto exit_options;
2804 }
2805
2806 /*
2807 * We are in the child. Make sure all other file
2808 * descriptors are closed, in case we are called with
2809 * more opened file descriptors than the standard ones.
2810 */
2811 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2812 (void) close(i);
2813 }
2814 }
2815
2816 /* Initialize thread health monitoring */
2817 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2818 if (!health_relayd) {
2819 PERROR("health_app_create error");
2820 retval = -1;
2821 goto exit_health_app_create;
2822 }
2823
2824 /* Create thread quit pipe */
2825 if (init_thread_quit_pipe()) {
2826 retval = -1;
2827 goto exit_init_data;
2828 }
2829
2830 /* Setup the thread apps communication pipe. */
2831 if (create_relay_conn_pipe()) {
2832 retval = -1;
2833 goto exit_init_data;
2834 }
2835
2836 /* Init relay command queue. */
2837 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
2838
2839 /* Initialize communication library */
2840 lttcomm_init();
2841 lttcomm_inet_init();
2842
2843 /* tables of sessions indexed by session ID */
2844 sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2845 if (!sessions_ht) {
2846 retval = -1;
2847 goto exit_init_data;
2848 }
2849
2850 /* tables of streams indexed by stream ID */
2851 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2852 if (!relay_streams_ht) {
2853 retval = -1;
2854 goto exit_init_data;
2855 }
2856
2857 /* tables of streams indexed by stream ID */
2858 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2859 if (!viewer_streams_ht) {
2860 retval = -1;
2861 goto exit_init_data;
2862 }
2863
2864 ret = utils_create_pipe(health_quit_pipe);
2865 if (ret) {
2866 retval = -1;
2867 goto exit_health_quit_pipe;
2868 }
2869
2870 /* Create thread to manage the client socket */
2871 ret = pthread_create(&health_thread, default_pthread_attr(),
2872 thread_manage_health, (void *) NULL);
2873 if (ret) {
2874 errno = ret;
2875 PERROR("pthread_create health");
2876 retval = -1;
2877 goto exit_health_thread;
2878 }
2879
2880 /* Setup the dispatcher thread */
2881 ret = pthread_create(&dispatcher_thread, default_pthread_attr(),
2882 relay_thread_dispatcher, (void *) NULL);
2883 if (ret) {
2884 errno = ret;
2885 PERROR("pthread_create dispatcher");
2886 retval = -1;
2887 goto exit_dispatcher_thread;
2888 }
2889
2890 /* Setup the worker thread */
2891 ret = pthread_create(&worker_thread, default_pthread_attr(),
2892 relay_thread_worker, NULL);
2893 if (ret) {
2894 errno = ret;
2895 PERROR("pthread_create worker");
2896 retval = -1;
2897 goto exit_worker_thread;
2898 }
2899
2900 /* Setup the listener thread */
2901 ret = pthread_create(&listener_thread, default_pthread_attr(),
2902 relay_thread_listener, (void *) NULL);
2903 if (ret) {
2904 errno = ret;
2905 PERROR("pthread_create listener");
2906 retval = -1;
2907 goto exit_listener_thread;
2908 }
2909
2910 ret = relayd_live_create(live_uri);
2911 if (ret) {
2912 ERR("Starting live viewer threads");
2913 retval = -1;
2914 goto exit_live;
2915 }
2916
2917 /*
2918 * This is where we start awaiting program completion (e.g. through
2919 * signal that asks threads to teardown).
2920 */
2921
2922 ret = relayd_live_join();
2923 if (ret) {
2924 retval = -1;
2925 }
2926 exit_live:
2927
2928 ret = pthread_join(listener_thread, &status);
2929 if (ret) {
2930 errno = ret;
2931 PERROR("pthread_join listener_thread");
2932 retval = -1;
2933 }
2934
2935 exit_listener_thread:
2936 ret = pthread_join(worker_thread, &status);
2937 if (ret) {
2938 errno = ret;
2939 PERROR("pthread_join worker_thread");
2940 retval = -1;
2941 }
2942
2943 exit_worker_thread:
2944 ret = pthread_join(dispatcher_thread, &status);
2945 if (ret) {
2946 errno = ret;
2947 PERROR("pthread_join dispatcher_thread");
2948 retval = -1;
2949 }
2950 exit_dispatcher_thread:
2951
2952 ret = pthread_join(health_thread, &status);
2953 if (ret) {
2954 errno = ret;
2955 PERROR("pthread_join health_thread");
2956 retval = -1;
2957 }
2958 exit_health_thread:
2959
2960 utils_close_pipe(health_quit_pipe);
2961 exit_health_quit_pipe:
2962
2963 exit_init_data:
2964 health_app_destroy(health_relayd);
2965 exit_health_app_create:
2966 exit_options:
2967 /*
2968 * Wait for all pending call_rcu work to complete before tearing
2969 * down data structures. call_rcu worker may be trying to
2970 * perform lookups in those structures.
2971 */
2972 rcu_barrier();
2973 relayd_cleanup();
2974
2975 /* Ensure all prior call_rcu are done. */
2976 rcu_barrier();
2977
2978 if (!retval) {
2979 exit(EXIT_SUCCESS);
2980 } else {
2981 exit(EXIT_FAILURE);
2982 }
2983 }
This page took 0.131712 seconds and 4 git commands to generate.