Relayd: disallow-clear option parsing
[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 static int opt_allow_clear = 1;
87
88 /*
89 * We need to wait for listener and live listener threads, as well as
90 * health check thread, before being ready to signal readiness.
91 */
92 #define NR_LTTNG_RELAY_READY 3
93 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
94
95 /* Size of receive buffer. */
96 #define RECV_DATA_BUFFER_SIZE 65536
97
98 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
99 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
100
101 static struct lttng_uri *control_uri;
102 static struct lttng_uri *data_uri;
103 static struct lttng_uri *live_uri;
104
105 const char *progname;
106
107 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
108 static int tracing_group_name_override;
109
110 const char * const config_section_name = "relayd";
111
112 /*
113 * Quit pipe for all threads. This permits a single cancellation point
114 * for all threads when receiving an event on the pipe.
115 */
116 int thread_quit_pipe[2] = { -1, -1 };
117
118 /*
119 * This pipe is used to inform the worker thread that a command is queued and
120 * ready to be processed.
121 */
122 static int relay_conn_pipe[2] = { -1, -1 };
123
124 /* Shared between threads */
125 static int dispatch_thread_exit;
126
127 static pthread_t listener_thread;
128 static pthread_t dispatcher_thread;
129 static pthread_t worker_thread;
130 static pthread_t health_thread;
131
132 /*
133 * last_relay_stream_id_lock protects last_relay_stream_id increment
134 * atomicity on 32-bit architectures.
135 */
136 static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
137 static uint64_t last_relay_stream_id;
138
139 /*
140 * Relay command queue.
141 *
142 * The relay_thread_listener and relay_thread_dispatcher communicate with this
143 * queue.
144 */
145 static struct relay_conn_queue relay_conn_queue;
146
147 /* buffer allocated at startup, used to store the trace data */
148 static char *data_buffer;
149 static unsigned int data_buffer_size;
150
151 /* Global relay stream hash table. */
152 struct lttng_ht *relay_streams_ht;
153
154 /* Global relay viewer stream hash table. */
155 struct lttng_ht *viewer_streams_ht;
156
157 /* Global relay sessions hash table. */
158 struct lttng_ht *sessions_ht;
159
160 /* Relayd health monitoring */
161 struct health_app *health_relayd;
162
163 static struct option long_options[] = {
164 { "control-port", 1, 0, 'C', },
165 { "data-port", 1, 0, 'D', },
166 { "live-port", 1, 0, 'L', },
167 { "daemonize", 0, 0, 'd', },
168 { "background", 0, 0, 'b', },
169 { "group", 1, 0, 'g', },
170 { "help", 0, 0, 'h', },
171 { "output", 1, 0, 'o', },
172 { "verbose", 0, 0, 'v', },
173 { "config", 1, 0, 'f' },
174 { "version", 0, 0, 'V' },
175 { "disallow-clear", 0, 0, 'x' },
176 { NULL, 0, 0, 0, },
177 };
178
179 static const char *config_ignore_options[] = { "help", "config", "version" };
180
181 /*
182 * Take an option from the getopt output and set it in the right variable to be
183 * used later.
184 *
185 * Return 0 on success else a negative value.
186 */
187 static int set_option(int opt, const char *arg, const char *optname)
188 {
189 int ret;
190
191 switch (opt) {
192 case 0:
193 fprintf(stderr, "option %s", optname);
194 if (arg) {
195 fprintf(stderr, " with arg %s\n", arg);
196 }
197 break;
198 case 'C':
199 if (lttng_is_setuid_setgid()) {
200 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
201 "-C, --control-port");
202 } else {
203 ret = uri_parse(arg, &control_uri);
204 if (ret < 0) {
205 ERR("Invalid control URI specified");
206 goto end;
207 }
208 if (control_uri->port == 0) {
209 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
210 }
211 }
212 break;
213 case 'D':
214 if (lttng_is_setuid_setgid()) {
215 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
216 "-D, -data-port");
217 } else {
218 ret = uri_parse(arg, &data_uri);
219 if (ret < 0) {
220 ERR("Invalid data URI specified");
221 goto end;
222 }
223 if (data_uri->port == 0) {
224 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
225 }
226 }
227 break;
228 case 'L':
229 if (lttng_is_setuid_setgid()) {
230 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
231 "-L, -live-port");
232 } else {
233 ret = uri_parse(arg, &live_uri);
234 if (ret < 0) {
235 ERR("Invalid live URI specified");
236 goto end;
237 }
238 if (live_uri->port == 0) {
239 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
240 }
241 }
242 break;
243 case 'd':
244 opt_daemon = 1;
245 break;
246 case 'b':
247 opt_background = 1;
248 break;
249 case 'g':
250 if (lttng_is_setuid_setgid()) {
251 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
252 "-g, --group");
253 } else {
254 tracing_group_name = strdup(arg);
255 if (tracing_group_name == NULL) {
256 ret = -errno;
257 PERROR("strdup");
258 goto end;
259 }
260 tracing_group_name_override = 1;
261 }
262 break;
263 case 'h':
264 ret = utils_show_help(8, "lttng-relayd", help_msg);
265 if (ret) {
266 ERR("Cannot show --help for `lttng-relayd`");
267 perror("exec");
268 }
269 exit(EXIT_FAILURE);
270 case 'V':
271 fprintf(stdout, "%s\n", VERSION);
272 exit(EXIT_SUCCESS);
273 case 'o':
274 if (lttng_is_setuid_setgid()) {
275 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
276 "-o, --output");
277 } else {
278 ret = asprintf(&opt_output_path, "%s", arg);
279 if (ret < 0) {
280 ret = -errno;
281 PERROR("asprintf opt_output_path");
282 goto end;
283 }
284 }
285 break;
286 case 'v':
287 /* Verbose level can increase using multiple -v */
288 if (arg) {
289 lttng_opt_verbose = config_parse_value(arg);
290 } else {
291 /* Only 3 level of verbosity (-vvv). */
292 if (lttng_opt_verbose < 3) {
293 lttng_opt_verbose += 1;
294 }
295 }
296 break;
297 case 'x':
298 /* Disallow clear */
299 opt_allow_clear = 0;
300 break;
301 default:
302 /* Unknown option or other error.
303 * Error is printed by getopt, just return */
304 ret = -1;
305 goto end;
306 }
307
308 /* All good. */
309 ret = 0;
310
311 end:
312 return ret;
313 }
314
315 /*
316 * config_entry_handler_cb used to handle options read from a config file.
317 * See config_entry_handler_cb comment in common/config/session-config.h for the
318 * return value conventions.
319 */
320 static int config_entry_handler(const struct config_entry *entry, void *unused)
321 {
322 int ret = 0, i;
323
324 if (!entry || !entry->name || !entry->value) {
325 ret = -EINVAL;
326 goto end;
327 }
328
329 /* Check if the option is to be ignored */
330 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
331 if (!strcmp(entry->name, config_ignore_options[i])) {
332 goto end;
333 }
334 }
335
336 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
337 /* Ignore if entry name is not fully matched. */
338 if (strcmp(entry->name, long_options[i].name)) {
339 continue;
340 }
341
342 /*
343 * If the option takes no argument on the command line,
344 * we have to check if the value is "true". We support
345 * non-zero numeric values, true, on and yes.
346 */
347 if (!long_options[i].has_arg) {
348 ret = config_parse_value(entry->value);
349 if (ret <= 0) {
350 if (ret) {
351 WARN("Invalid configuration value \"%s\" for option %s",
352 entry->value, entry->name);
353 }
354 /* False, skip boolean config option. */
355 goto end;
356 }
357 }
358
359 ret = set_option(long_options[i].val, entry->value, entry->name);
360 goto end;
361 }
362
363 WARN("Unrecognized option \"%s\" in daemon configuration file.",
364 entry->name);
365
366 end:
367 return ret;
368 }
369
370 static int set_options(int argc, char **argv)
371 {
372 int c, ret = 0, option_index = 0, retval = 0;
373 int orig_optopt = optopt, orig_optind = optind;
374 char *default_address, *optstring;
375 const char *config_path = NULL;
376 const char *value = NULL;
377
378 optstring = utils_generate_optstring(long_options,
379 sizeof(long_options) / sizeof(struct option));
380 if (!optstring) {
381 retval = -ENOMEM;
382 goto exit;
383 }
384
385 /* Check for the --config option */
386
387 while ((c = getopt_long(argc, argv, optstring, long_options,
388 &option_index)) != -1) {
389 if (c == '?') {
390 retval = -EINVAL;
391 goto exit;
392 } else if (c != 'f') {
393 continue;
394 }
395
396 if (lttng_is_setuid_setgid()) {
397 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
398 "-f, --config");
399 } else {
400 config_path = utils_expand_path(optarg);
401 if (!config_path) {
402 ERR("Failed to resolve path: %s", optarg);
403 }
404 }
405 }
406
407 ret = config_get_section_entries(config_path, config_section_name,
408 config_entry_handler, NULL);
409 if (ret) {
410 if (ret > 0) {
411 ERR("Invalid configuration option at line %i", ret);
412 }
413 retval = -1;
414 goto exit;
415 }
416
417 /* Reset getopt's global state */
418 optopt = orig_optopt;
419 optind = orig_optind;
420 while (1) {
421 c = getopt_long(argc, argv, optstring, long_options, &option_index);
422 if (c == -1) {
423 break;
424 }
425
426 ret = set_option(c, optarg, long_options[option_index].name);
427 if (ret < 0) {
428 retval = -1;
429 goto exit;
430 }
431 }
432
433 /* assign default values */
434 if (control_uri == NULL) {
435 ret = asprintf(&default_address,
436 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
437 DEFAULT_NETWORK_CONTROL_PORT);
438 if (ret < 0) {
439 PERROR("asprintf default data address");
440 retval = -1;
441 goto exit;
442 }
443
444 ret = uri_parse(default_address, &control_uri);
445 free(default_address);
446 if (ret < 0) {
447 ERR("Invalid control URI specified");
448 retval = -1;
449 goto exit;
450 }
451 }
452 if (data_uri == NULL) {
453 ret = asprintf(&default_address,
454 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
455 DEFAULT_NETWORK_DATA_PORT);
456 if (ret < 0) {
457 PERROR("asprintf default data address");
458 retval = -1;
459 goto exit;
460 }
461
462 ret = uri_parse(default_address, &data_uri);
463 free(default_address);
464 if (ret < 0) {
465 ERR("Invalid data URI specified");
466 retval = -1;
467 goto exit;
468 }
469 }
470 if (live_uri == NULL) {
471 ret = asprintf(&default_address,
472 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
473 DEFAULT_NETWORK_VIEWER_PORT);
474 if (ret < 0) {
475 PERROR("asprintf default viewer control address");
476 retval = -1;
477 goto exit;
478 }
479
480 ret = uri_parse(default_address, &live_uri);
481 free(default_address);
482 if (ret < 0) {
483 ERR("Invalid viewer control URI specified");
484 retval = -1;
485 goto exit;
486 }
487 }
488
489 if (opt_allow_clear) {
490 /* Check if env variable exists. */
491 value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
492 if (value) {
493 ret = config_parse_value(value);
494 if (ret < 0) {
495 ERR("Invalid value for %s specified", DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
496 retval = -1;
497 goto exit;
498 }
499 opt_allow_clear = !ret;
500 }
501 }
502
503 exit:
504 free(optstring);
505 return retval;
506 }
507
508 static void print_global_objects(void)
509 {
510 rcu_register_thread();
511
512 print_viewer_streams();
513 print_relay_streams();
514 print_sessions();
515
516 rcu_unregister_thread();
517 }
518
519 /*
520 * Cleanup the daemon
521 */
522 static void relayd_cleanup(void)
523 {
524 print_global_objects();
525
526 DBG("Cleaning up");
527
528 if (viewer_streams_ht)
529 lttng_ht_destroy(viewer_streams_ht);
530 if (relay_streams_ht)
531 lttng_ht_destroy(relay_streams_ht);
532 if (sessions_ht)
533 lttng_ht_destroy(sessions_ht);
534
535 /* free the dynamically allocated opt_output_path */
536 free(opt_output_path);
537
538 /* Close thread quit pipes */
539 utils_close_pipe(thread_quit_pipe);
540
541 uri_free(control_uri);
542 uri_free(data_uri);
543 /* Live URI is freed in the live thread. */
544
545 if (tracing_group_name_override) {
546 free((void *) tracing_group_name);
547 }
548 }
549
550 /*
551 * Write to writable pipe used to notify a thread.
552 */
553 static int notify_thread_pipe(int wpipe)
554 {
555 ssize_t ret;
556
557 ret = lttng_write(wpipe, "!", 1);
558 if (ret < 1) {
559 PERROR("write poll pipe");
560 goto end;
561 }
562 ret = 0;
563 end:
564 return ret;
565 }
566
567 static int notify_health_quit_pipe(int *pipe)
568 {
569 ssize_t ret;
570
571 ret = lttng_write(pipe[1], "4", 1);
572 if (ret < 1) {
573 PERROR("write relay health quit");
574 goto end;
575 }
576 ret = 0;
577 end:
578 return ret;
579 }
580
581 /*
582 * Stop all relayd and relayd-live threads.
583 */
584 int lttng_relay_stop_threads(void)
585 {
586 int retval = 0;
587
588 /* Stopping all threads */
589 DBG("Terminating all threads");
590 if (notify_thread_pipe(thread_quit_pipe[1])) {
591 ERR("write error on thread quit pipe");
592 retval = -1;
593 }
594
595 if (notify_health_quit_pipe(health_quit_pipe)) {
596 ERR("write error on health quit pipe");
597 }
598
599 /* Dispatch thread */
600 CMM_STORE_SHARED(dispatch_thread_exit, 1);
601 futex_nto1_wake(&relay_conn_queue.futex);
602
603 if (relayd_live_stop()) {
604 ERR("Error stopping live threads");
605 retval = -1;
606 }
607 return retval;
608 }
609
610 /*
611 * Signal handler for the daemon
612 *
613 * Simply stop all worker threads, leaving main() return gracefully after
614 * joining all threads and calling cleanup().
615 */
616 static void sighandler(int sig)
617 {
618 switch (sig) {
619 case SIGINT:
620 DBG("SIGINT caught");
621 if (lttng_relay_stop_threads()) {
622 ERR("Error stopping threads");
623 }
624 break;
625 case SIGTERM:
626 DBG("SIGTERM caught");
627 if (lttng_relay_stop_threads()) {
628 ERR("Error stopping threads");
629 }
630 break;
631 case SIGUSR1:
632 CMM_STORE_SHARED(recv_child_signal, 1);
633 break;
634 default:
635 break;
636 }
637 }
638
639 /*
640 * Setup signal handler for :
641 * SIGINT, SIGTERM, SIGPIPE
642 */
643 static int set_signal_handler(void)
644 {
645 int ret = 0;
646 struct sigaction sa;
647 sigset_t sigset;
648
649 if ((ret = sigemptyset(&sigset)) < 0) {
650 PERROR("sigemptyset");
651 return ret;
652 }
653
654 sa.sa_mask = sigset;
655 sa.sa_flags = 0;
656
657 sa.sa_handler = sighandler;
658 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
659 PERROR("sigaction");
660 return ret;
661 }
662
663 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
664 PERROR("sigaction");
665 return ret;
666 }
667
668 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
669 PERROR("sigaction");
670 return ret;
671 }
672
673 sa.sa_handler = SIG_IGN;
674 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
675 PERROR("sigaction");
676 return ret;
677 }
678
679 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
680
681 return ret;
682 }
683
684 void lttng_relay_notify_ready(void)
685 {
686 /* Notify the parent of the fork() process that we are ready. */
687 if (opt_daemon || opt_background) {
688 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
689 kill(child_ppid, SIGUSR1);
690 }
691 }
692 }
693
694 /*
695 * Init thread quit pipe.
696 *
697 * Return -1 on error or 0 if all pipes are created.
698 */
699 static int init_thread_quit_pipe(void)
700 {
701 int ret;
702
703 ret = utils_create_pipe_cloexec(thread_quit_pipe);
704
705 return ret;
706 }
707
708 /*
709 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
710 */
711 static int create_thread_poll_set(struct lttng_poll_event *events, int size)
712 {
713 int ret;
714
715 if (events == NULL || size == 0) {
716 ret = -1;
717 goto error;
718 }
719
720 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
721 if (ret < 0) {
722 goto error;
723 }
724
725 /* Add quit pipe */
726 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
727 if (ret < 0) {
728 goto error;
729 }
730
731 return 0;
732
733 error:
734 return ret;
735 }
736
737 /*
738 * Check if the thread quit pipe was triggered.
739 *
740 * Return 1 if it was triggered else 0;
741 */
742 static int check_thread_quit_pipe(int fd, uint32_t events)
743 {
744 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
745 return 1;
746 }
747
748 return 0;
749 }
750
751 /*
752 * Create and init socket from uri.
753 */
754 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri)
755 {
756 int ret;
757 struct lttcomm_sock *sock = NULL;
758
759 sock = lttcomm_alloc_sock_from_uri(uri);
760 if (sock == NULL) {
761 ERR("Allocating socket");
762 goto error;
763 }
764
765 ret = lttcomm_create_sock(sock);
766 if (ret < 0) {
767 goto error;
768 }
769 DBG("Listening on sock %d", sock->fd);
770
771 ret = sock->ops->bind(sock);
772 if (ret < 0) {
773 PERROR("Failed to bind socket");
774 goto error;
775 }
776
777 ret = sock->ops->listen(sock, -1);
778 if (ret < 0) {
779 goto error;
780
781 }
782
783 return sock;
784
785 error:
786 if (sock) {
787 lttcomm_destroy_sock(sock);
788 }
789 return NULL;
790 }
791
792 /*
793 * This thread manages the listening for new connections on the network
794 */
795 static void *relay_thread_listener(void *data)
796 {
797 int i, ret, pollfd, err = -1;
798 uint32_t revents, nb_fd;
799 struct lttng_poll_event events;
800 struct lttcomm_sock *control_sock, *data_sock;
801
802 DBG("[thread] Relay listener started");
803
804 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
805
806 health_code_update();
807
808 control_sock = relay_socket_create(control_uri);
809 if (!control_sock) {
810 goto error_sock_control;
811 }
812
813 data_sock = relay_socket_create(data_uri);
814 if (!data_sock) {
815 goto error_sock_relay;
816 }
817
818 /*
819 * Pass 3 as size here for the thread quit pipe, control and
820 * data socket.
821 */
822 ret = create_thread_poll_set(&events, 3);
823 if (ret < 0) {
824 goto error_create_poll;
825 }
826
827 /* Add the control socket */
828 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
829 if (ret < 0) {
830 goto error_poll_add;
831 }
832
833 /* Add the data socket */
834 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
835 if (ret < 0) {
836 goto error_poll_add;
837 }
838
839 lttng_relay_notify_ready();
840
841 if (testpoint(relayd_thread_listener)) {
842 goto error_testpoint;
843 }
844
845 while (1) {
846 health_code_update();
847
848 DBG("Listener accepting connections");
849
850 restart:
851 health_poll_entry();
852 ret = lttng_poll_wait(&events, -1);
853 health_poll_exit();
854 if (ret < 0) {
855 /*
856 * Restart interrupted system call.
857 */
858 if (errno == EINTR) {
859 goto restart;
860 }
861 goto error;
862 }
863
864 nb_fd = ret;
865
866 DBG("Relay new connection received");
867 for (i = 0; i < nb_fd; i++) {
868 health_code_update();
869
870 /* Fetch once the poll data */
871 revents = LTTNG_POLL_GETEV(&events, i);
872 pollfd = LTTNG_POLL_GETFD(&events, i);
873
874 if (!revents) {
875 /*
876 * No activity for this FD (poll
877 * implementation).
878 */
879 continue;
880 }
881
882 /* Thread quit pipe has been closed. Killing thread. */
883 ret = check_thread_quit_pipe(pollfd, revents);
884 if (ret) {
885 err = 0;
886 goto exit;
887 }
888
889 if (revents & LPOLLIN) {
890 /*
891 * A new connection is requested, therefore a
892 * sessiond/consumerd connection is allocated in
893 * this thread, enqueued to a global queue and
894 * dequeued (and freed) in the worker thread.
895 */
896 int val = 1;
897 struct relay_connection *new_conn;
898 struct lttcomm_sock *newsock;
899 enum connection_type type;
900
901 if (pollfd == data_sock->fd) {
902 type = RELAY_DATA;
903 newsock = data_sock->ops->accept(data_sock);
904 DBG("Relay data connection accepted, socket %d",
905 newsock->fd);
906 } else {
907 assert(pollfd == control_sock->fd);
908 type = RELAY_CONTROL;
909 newsock = control_sock->ops->accept(control_sock);
910 DBG("Relay control connection accepted, socket %d",
911 newsock->fd);
912 }
913 if (!newsock) {
914 PERROR("accepting sock");
915 goto error;
916 }
917
918 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
919 sizeof(val));
920 if (ret < 0) {
921 PERROR("setsockopt inet");
922 lttcomm_destroy_sock(newsock);
923 goto error;
924 }
925
926 ret = socket_apply_keep_alive_config(newsock->fd);
927 if (ret < 0) {
928 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
929 newsock->fd);
930 lttcomm_destroy_sock(newsock);
931 goto error;
932 }
933
934 new_conn = connection_create(newsock, type);
935 if (!new_conn) {
936 lttcomm_destroy_sock(newsock);
937 goto error;
938 }
939
940 /* Enqueue request for the dispatcher thread. */
941 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
942 &new_conn->qnode);
943
944 /*
945 * Wake the dispatch queue futex.
946 * Implicit memory barrier with the
947 * exchange in cds_wfcq_enqueue.
948 */
949 futex_nto1_wake(&relay_conn_queue.futex);
950 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
951 ERR("socket poll error");
952 goto error;
953 } else {
954 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
955 goto error;
956 }
957 }
958 }
959
960 exit:
961 error:
962 error_poll_add:
963 error_testpoint:
964 lttng_poll_clean(&events);
965 error_create_poll:
966 if (data_sock->fd >= 0) {
967 ret = data_sock->ops->close(data_sock);
968 if (ret) {
969 PERROR("close");
970 }
971 }
972 lttcomm_destroy_sock(data_sock);
973 error_sock_relay:
974 if (control_sock->fd >= 0) {
975 ret = control_sock->ops->close(control_sock);
976 if (ret) {
977 PERROR("close");
978 }
979 }
980 lttcomm_destroy_sock(control_sock);
981 error_sock_control:
982 if (err) {
983 health_error();
984 ERR("Health error occurred in %s", __func__);
985 }
986 health_unregister(health_relayd);
987 DBG("Relay listener thread cleanup complete");
988 lttng_relay_stop_threads();
989 return NULL;
990 }
991
992 /*
993 * This thread manages the dispatching of the requests to worker threads
994 */
995 static void *relay_thread_dispatcher(void *data)
996 {
997 int err = -1;
998 ssize_t ret;
999 struct cds_wfcq_node *node;
1000 struct relay_connection *new_conn = NULL;
1001
1002 DBG("[thread] Relay dispatcher started");
1003
1004 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
1005
1006 if (testpoint(relayd_thread_dispatcher)) {
1007 goto error_testpoint;
1008 }
1009
1010 health_code_update();
1011
1012 for (;;) {
1013 health_code_update();
1014
1015 /* Atomically prepare the queue futex */
1016 futex_nto1_prepare(&relay_conn_queue.futex);
1017
1018 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1019 break;
1020 }
1021
1022 do {
1023 health_code_update();
1024
1025 /* Dequeue commands */
1026 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1027 &relay_conn_queue.tail);
1028 if (node == NULL) {
1029 DBG("Woken up but nothing in the relay command queue");
1030 /* Continue thread execution */
1031 break;
1032 }
1033 new_conn = caa_container_of(node, struct relay_connection, qnode);
1034
1035 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1036
1037 /*
1038 * Inform worker thread of the new request. This
1039 * call is blocking so we can be assured that
1040 * the data will be read at some point in time
1041 * or wait to the end of the world :)
1042 */
1043 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1044 if (ret < 0) {
1045 PERROR("write connection pipe");
1046 connection_put(new_conn);
1047 goto error;
1048 }
1049 } while (node != NULL);
1050
1051 /* Futex wait on queue. Blocking call on futex() */
1052 health_poll_entry();
1053 futex_nto1_wait(&relay_conn_queue.futex);
1054 health_poll_exit();
1055 }
1056
1057 /* Normal exit, no error */
1058 err = 0;
1059
1060 error:
1061 error_testpoint:
1062 if (err) {
1063 health_error();
1064 ERR("Health error occurred in %s", __func__);
1065 }
1066 health_unregister(health_relayd);
1067 DBG("Dispatch thread dying");
1068 lttng_relay_stop_threads();
1069 return NULL;
1070 }
1071
1072 /*
1073 * Set index data from the control port to a given index object.
1074 */
1075 static int set_index_control_data(struct relay_index *index,
1076 struct lttcomm_relayd_index *data,
1077 struct relay_connection *conn)
1078 {
1079 struct ctf_packet_index index_data;
1080
1081 /*
1082 * The index on disk is encoded in big endian, so we don't need
1083 * to convert the data received on the network. The data_offset
1084 * value is NEVER modified here and is updated by the data
1085 * thread.
1086 */
1087 index_data.packet_size = data->packet_size;
1088 index_data.content_size = data->content_size;
1089 index_data.timestamp_begin = data->timestamp_begin;
1090 index_data.timestamp_end = data->timestamp_end;
1091 index_data.events_discarded = data->events_discarded;
1092 index_data.stream_id = data->stream_id;
1093
1094 if (conn->minor >= 8) {
1095 index->index_data.stream_instance_id = data->stream_instance_id;
1096 index->index_data.packet_seq_num = data->packet_seq_num;
1097 }
1098
1099 return relay_index_set_data(index, &index_data);
1100 }
1101
1102 /*
1103 * Handle the RELAYD_CREATE_SESSION command.
1104 *
1105 * On success, send back the session id or else return a negative value.
1106 */
1107 static int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
1108 struct relay_connection *conn)
1109 {
1110 int ret = 0, send_ret;
1111 struct relay_session *session;
1112 struct lttcomm_relayd_status_session reply;
1113 char session_name[LTTNG_NAME_MAX];
1114 char hostname[LTTNG_HOST_NAME_MAX];
1115 uint32_t live_timer = 0;
1116 bool snapshot = false;
1117
1118 memset(session_name, 0, LTTNG_NAME_MAX);
1119 memset(hostname, 0, LTTNG_HOST_NAME_MAX);
1120
1121 memset(&reply, 0, sizeof(reply));
1122
1123 switch (conn->minor) {
1124 case 1:
1125 case 2:
1126 case 3:
1127 break;
1128 case 4: /* LTTng sessiond 2.4 */
1129 default:
1130 ret = cmd_create_session_2_4(conn, session_name,
1131 hostname, &live_timer, &snapshot);
1132 }
1133 if (ret < 0) {
1134 goto send_reply;
1135 }
1136
1137 session = session_create(session_name, hostname, live_timer,
1138 snapshot, conn->major, conn->minor);
1139 if (!session) {
1140 ret = -1;
1141 goto send_reply;
1142 }
1143 assert(!conn->session);
1144 conn->session = session;
1145 DBG("Created session %" PRIu64, session->id);
1146
1147 reply.session_id = htobe64(session->id);
1148
1149 send_reply:
1150 if (ret < 0) {
1151 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1152 } else {
1153 reply.ret_code = htobe32(LTTNG_OK);
1154 }
1155
1156 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1157 if (send_ret < 0) {
1158 ERR("Relayd sending session id");
1159 ret = send_ret;
1160 }
1161
1162 return ret;
1163 }
1164
1165 /*
1166 * When we have received all the streams and the metadata for a channel,
1167 * we make them visible to the viewer threads.
1168 */
1169 static void publish_connection_local_streams(struct relay_connection *conn)
1170 {
1171 struct relay_stream *stream;
1172 struct relay_session *session = conn->session;
1173
1174 /*
1175 * We publish all streams belonging to a session atomically wrt
1176 * session lock.
1177 */
1178 pthread_mutex_lock(&session->lock);
1179 rcu_read_lock();
1180 cds_list_for_each_entry_rcu(stream, &session->recv_list,
1181 recv_node) {
1182 stream_publish(stream);
1183 }
1184 rcu_read_unlock();
1185
1186 /*
1187 * Inform the viewer that there are new streams in the session.
1188 */
1189 if (session->viewer_attached) {
1190 uatomic_set(&session->new_streams, 1);
1191 }
1192 pthread_mutex_unlock(&session->lock);
1193 }
1194
1195 /*
1196 * relay_add_stream: allocate a new stream for a session
1197 */
1198 static int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1199 struct relay_connection *conn)
1200 {
1201 int ret;
1202 ssize_t send_ret;
1203 struct relay_session *session = conn->session;
1204 struct relay_stream *stream = NULL;
1205 struct lttcomm_relayd_status_stream reply;
1206 struct ctf_trace *trace = NULL;
1207 uint64_t stream_handle = -1ULL;
1208 char *path_name = NULL, *channel_name = NULL;
1209 uint64_t tracefile_size = 0, tracefile_count = 0;
1210
1211 if (!session || conn->version_check_done == 0) {
1212 ERR("Trying to add a stream before version check");
1213 ret = -1;
1214 goto end_no_session;
1215 }
1216
1217 switch (session->minor) {
1218 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
1219 ret = cmd_recv_stream_2_1(conn, &path_name,
1220 &channel_name);
1221 break;
1222 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
1223 default:
1224 ret = cmd_recv_stream_2_2(conn, &path_name,
1225 &channel_name, &tracefile_size, &tracefile_count);
1226 break;
1227 }
1228 if (ret < 0) {
1229 goto send_reply;
1230 }
1231
1232 trace = ctf_trace_get_by_path_or_create(session, path_name);
1233 if (!trace) {
1234 goto send_reply;
1235 }
1236 /* This stream here has one reference on the trace. */
1237
1238 pthread_mutex_lock(&last_relay_stream_id_lock);
1239 stream_handle = ++last_relay_stream_id;
1240 pthread_mutex_unlock(&last_relay_stream_id_lock);
1241
1242 /* We pass ownership of path_name and channel_name. */
1243 stream = stream_create(trace, stream_handle, path_name,
1244 channel_name, tracefile_size, tracefile_count);
1245 path_name = NULL;
1246 channel_name = NULL;
1247
1248 /*
1249 * Streams are the owners of their trace. Reference to trace is
1250 * kept within stream_create().
1251 */
1252 ctf_trace_put(trace);
1253
1254 send_reply:
1255 memset(&reply, 0, sizeof(reply));
1256 reply.handle = htobe64(stream_handle);
1257 if (!stream) {
1258 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1259 } else {
1260 reply.ret_code = htobe32(LTTNG_OK);
1261 }
1262
1263 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1264 sizeof(struct lttcomm_relayd_status_stream), 0);
1265 if (send_ret < 0) {
1266 ERR("Relay sending stream id");
1267 ret = (int) send_ret;
1268 }
1269
1270 end_no_session:
1271 free(path_name);
1272 free(channel_name);
1273 return ret;
1274 }
1275
1276 /*
1277 * relay_close_stream: close a specific stream
1278 */
1279 static int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1280 struct relay_connection *conn)
1281 {
1282 int ret, send_ret;
1283 struct relay_session *session = conn->session;
1284 struct lttcomm_relayd_close_stream stream_info;
1285 struct lttcomm_relayd_generic_reply reply;
1286 struct relay_stream *stream;
1287
1288 DBG("Close stream received");
1289
1290 if (!session || conn->version_check_done == 0) {
1291 ERR("Trying to close a stream before version check");
1292 ret = -1;
1293 goto end_no_session;
1294 }
1295
1296 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1297 sizeof(struct lttcomm_relayd_close_stream), 0);
1298 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1299 if (ret == 0) {
1300 /* Orderly shutdown. Not necessary to print an error. */
1301 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1302 } else {
1303 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1304 }
1305 ret = -1;
1306 goto end_no_session;
1307 }
1308
1309 stream = stream_get_by_id(be64toh(stream_info.stream_id));
1310 if (!stream) {
1311 ret = -1;
1312 goto end;
1313 }
1314
1315 /*
1316 * Set last_net_seq_num before the close flag. Required by data
1317 * pending check.
1318 */
1319 pthread_mutex_lock(&stream->lock);
1320 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1321 pthread_mutex_unlock(&stream->lock);
1322
1323 /*
1324 * This is one of the conditions which may trigger a stream close
1325 * with the others being:
1326 * 1) A close command is received for a stream
1327 * 2) The control connection owning the stream is closed
1328 * 3) We have received all of the stream's data _after_ a close
1329 * request.
1330 */
1331 try_stream_close(stream);
1332 if (stream->is_metadata) {
1333 struct relay_viewer_stream *vstream;
1334
1335 vstream = viewer_stream_get_by_id(stream->stream_handle);
1336 if (vstream) {
1337 if (vstream->metadata_sent == stream->metadata_received) {
1338 /*
1339 * Since all the metadata has been sent to the
1340 * viewer and that we have a request to close
1341 * its stream, we can safely teardown the
1342 * corresponding metadata viewer stream.
1343 */
1344 viewer_stream_put(vstream);
1345 }
1346 /* Put local reference. */
1347 viewer_stream_put(vstream);
1348 }
1349 }
1350 stream_put(stream);
1351
1352 end:
1353 memset(&reply, 0, sizeof(reply));
1354 if (ret < 0) {
1355 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1356 } else {
1357 reply.ret_code = htobe32(LTTNG_OK);
1358 }
1359 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1360 sizeof(struct lttcomm_relayd_generic_reply), 0);
1361 if (send_ret < 0) {
1362 ERR("Relay sending stream id");
1363 ret = send_ret;
1364 }
1365
1366 end_no_session:
1367 return ret;
1368 }
1369
1370 /*
1371 * relay_reset_metadata: reset a metadata stream
1372 */
1373 static
1374 int relay_reset_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1375 struct relay_connection *conn)
1376 {
1377 int ret, send_ret;
1378 struct relay_session *session = conn->session;
1379 struct lttcomm_relayd_reset_metadata stream_info;
1380 struct lttcomm_relayd_generic_reply reply;
1381 struct relay_stream *stream;
1382
1383 DBG("Reset metadata received");
1384
1385 if (!session || conn->version_check_done == 0) {
1386 ERR("Trying to reset a metadata stream before version check");
1387 ret = -1;
1388 goto end_no_session;
1389 }
1390
1391 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1392 sizeof(struct lttcomm_relayd_reset_metadata), 0);
1393 if (ret < sizeof(struct lttcomm_relayd_reset_metadata)) {
1394 if (ret == 0) {
1395 /* Orderly shutdown. Not necessary to print an error. */
1396 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1397 } else {
1398 ERR("Relay didn't receive valid reset_metadata struct "
1399 "size : %d", ret);
1400 }
1401 ret = -1;
1402 goto end_no_session;
1403 }
1404 DBG("Update metadata to version %" PRIu64, be64toh(stream_info.version));
1405
1406 /* Unsupported for live sessions for now. */
1407 if (session->live_timer != 0) {
1408 ret = -1;
1409 goto end;
1410 }
1411
1412 stream = stream_get_by_id(be64toh(stream_info.stream_id));
1413 if (!stream) {
1414 ret = -1;
1415 goto end;
1416 }
1417 pthread_mutex_lock(&stream->lock);
1418 if (!stream->is_metadata) {
1419 ret = -1;
1420 goto end_unlock;
1421 }
1422
1423 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
1424 0, 0, -1, -1, stream->stream_fd->fd, NULL,
1425 &stream->stream_fd->fd);
1426 if (ret < 0) {
1427 ERR("Failed to rotate metadata file %s of channel %s",
1428 stream->path_name, stream->channel_name);
1429 goto end_unlock;
1430 }
1431
1432 end_unlock:
1433 pthread_mutex_unlock(&stream->lock);
1434 stream_put(stream);
1435
1436 end:
1437 memset(&reply, 0, sizeof(reply));
1438 if (ret < 0) {
1439 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1440 } else {
1441 reply.ret_code = htobe32(LTTNG_OK);
1442 }
1443 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1444 sizeof(struct lttcomm_relayd_generic_reply), 0);
1445 if (send_ret < 0) {
1446 ERR("Relay sending reset metadata reply");
1447 ret = send_ret;
1448 }
1449
1450 end_no_session:
1451 return ret;
1452 }
1453
1454 /*
1455 * relay_unknown_command: send -1 if received unknown command
1456 */
1457 static void relay_unknown_command(struct relay_connection *conn)
1458 {
1459 struct lttcomm_relayd_generic_reply reply;
1460 int ret;
1461
1462 memset(&reply, 0, sizeof(reply));
1463 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1464 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1465 sizeof(struct lttcomm_relayd_generic_reply), 0);
1466 if (ret < 0) {
1467 ERR("Relay sending unknown command");
1468 }
1469 }
1470
1471 /*
1472 * relay_start: send an acknowledgment to the client to tell if we are
1473 * ready to receive data. We are ready if a session is established.
1474 */
1475 static int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1476 struct relay_connection *conn)
1477 {
1478 int ret = htobe32(LTTNG_OK);
1479 struct lttcomm_relayd_generic_reply reply;
1480 struct relay_session *session = conn->session;
1481
1482 if (!session) {
1483 DBG("Trying to start the streaming without a session established");
1484 ret = htobe32(LTTNG_ERR_UNK);
1485 }
1486
1487 memset(&reply, 0, sizeof(reply));
1488 reply.ret_code = ret;
1489 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1490 sizeof(struct lttcomm_relayd_generic_reply), 0);
1491 if (ret < 0) {
1492 ERR("Relay sending start ack");
1493 }
1494
1495 return ret;
1496 }
1497
1498 /*
1499 * Append padding to the file pointed by the file descriptor fd.
1500 */
1501 static int write_padding_to_file(int fd, uint32_t size)
1502 {
1503 ssize_t ret = 0;
1504 char *zeros;
1505
1506 if (size == 0) {
1507 goto end;
1508 }
1509
1510 zeros = zmalloc(size);
1511 if (zeros == NULL) {
1512 PERROR("zmalloc zeros for padding");
1513 ret = -1;
1514 goto end;
1515 }
1516
1517 ret = lttng_write(fd, zeros, size);
1518 if (ret < size) {
1519 PERROR("write padding to file");
1520 }
1521
1522 free(zeros);
1523
1524 end:
1525 return ret;
1526 }
1527
1528 /*
1529 * relay_recv_metadata: receive the metadata for the session.
1530 */
1531 static int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1532 struct relay_connection *conn)
1533 {
1534 int ret = 0;
1535 ssize_t size_ret;
1536 struct relay_session *session = conn->session;
1537 struct lttcomm_relayd_metadata_payload *metadata_struct;
1538 struct relay_stream *metadata_stream;
1539 uint64_t data_size, payload_size;
1540
1541 if (!session) {
1542 ERR("Metadata sent before version check");
1543 ret = -1;
1544 goto end;
1545 }
1546
1547 data_size = payload_size = be64toh(recv_hdr->data_size);
1548 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1549 ERR("Incorrect data size");
1550 ret = -1;
1551 goto end;
1552 }
1553 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1554
1555 if (data_buffer_size < data_size) {
1556 /* In case the realloc fails, we can free the memory */
1557 char *tmp_data_ptr;
1558
1559 tmp_data_ptr = realloc(data_buffer, data_size);
1560 if (!tmp_data_ptr) {
1561 ERR("Allocating data buffer");
1562 free(data_buffer);
1563 ret = -1;
1564 goto end;
1565 }
1566 data_buffer = tmp_data_ptr;
1567 data_buffer_size = data_size;
1568 }
1569 memset(data_buffer, 0, data_size);
1570 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1571 size_ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
1572 if (size_ret < 0 || size_ret != data_size) {
1573 if (size_ret == 0) {
1574 /* Orderly shutdown. Not necessary to print an error. */
1575 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1576 } else {
1577 ERR("Relay didn't receive the whole metadata");
1578 }
1579 ret = -1;
1580 goto end;
1581 }
1582 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1583
1584 metadata_stream = stream_get_by_id(be64toh(metadata_struct->stream_id));
1585 if (!metadata_stream) {
1586 ret = -1;
1587 goto end;
1588 }
1589
1590 pthread_mutex_lock(&metadata_stream->lock);
1591
1592 size_ret = lttng_write(metadata_stream->stream_fd->fd, metadata_struct->payload,
1593 payload_size);
1594 if (size_ret < payload_size) {
1595 ERR("Relay error writing metadata on file");
1596 ret = -1;
1597 goto end_put;
1598 }
1599
1600 size_ret = write_padding_to_file(metadata_stream->stream_fd->fd,
1601 be32toh(metadata_struct->padding_size));
1602 if (size_ret < 0) {
1603 goto end_put;
1604 }
1605
1606 metadata_stream->metadata_received +=
1607 payload_size + be32toh(metadata_struct->padding_size);
1608 DBG2("Relay metadata written. Updated metadata_received %" PRIu64,
1609 metadata_stream->metadata_received);
1610
1611 end_put:
1612 pthread_mutex_unlock(&metadata_stream->lock);
1613 stream_put(metadata_stream);
1614 end:
1615 return ret;
1616 }
1617
1618 /*
1619 * relay_send_version: send relayd version number
1620 */
1621 static int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1622 struct relay_connection *conn)
1623 {
1624 int ret;
1625 struct lttcomm_relayd_version reply, msg;
1626 bool compatible = true;
1627
1628 conn->version_check_done = 1;
1629
1630 /* Get version from the other side. */
1631 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1632 if (ret < 0 || ret != sizeof(msg)) {
1633 if (ret == 0) {
1634 /* Orderly shutdown. Not necessary to print an error. */
1635 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1636 } else {
1637 ERR("Relay failed to receive the version values.");
1638 }
1639 ret = -1;
1640 goto end;
1641 }
1642
1643 memset(&reply, 0, sizeof(reply));
1644 reply.major = RELAYD_VERSION_COMM_MAJOR;
1645 reply.minor = RELAYD_VERSION_COMM_MINOR;
1646
1647 /* Major versions must be the same */
1648 if (reply.major != be32toh(msg.major)) {
1649 DBG("Incompatible major versions (%u vs %u), deleting session",
1650 reply.major, be32toh(msg.major));
1651 compatible = false;
1652 }
1653
1654 conn->major = reply.major;
1655 /* We adapt to the lowest compatible version */
1656 if (reply.minor <= be32toh(msg.minor)) {
1657 conn->minor = reply.minor;
1658 } else {
1659 conn->minor = be32toh(msg.minor);
1660 }
1661
1662 reply.major = htobe32(reply.major);
1663 reply.minor = htobe32(reply.minor);
1664 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1665 sizeof(struct lttcomm_relayd_version), 0);
1666 if (ret < 0) {
1667 ERR("Relay sending version");
1668 }
1669
1670 if (!compatible) {
1671 ret = -1;
1672 goto end;
1673 }
1674
1675 DBG("Version check done using protocol %u.%u", conn->major,
1676 conn->minor);
1677
1678 end:
1679 return ret;
1680 }
1681
1682 /*
1683 * Check for data pending for a given stream id from the session daemon.
1684 */
1685 static int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1686 struct relay_connection *conn)
1687 {
1688 struct relay_session *session = conn->session;
1689 struct lttcomm_relayd_data_pending msg;
1690 struct lttcomm_relayd_generic_reply reply;
1691 struct relay_stream *stream;
1692 int ret;
1693 uint64_t last_net_seq_num, stream_id;
1694
1695 DBG("Data pending command received");
1696
1697 if (!session || conn->version_check_done == 0) {
1698 ERR("Trying to check for data before version check");
1699 ret = -1;
1700 goto end_no_session;
1701 }
1702
1703 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1704 if (ret < sizeof(msg)) {
1705 if (ret == 0) {
1706 /* Orderly shutdown. Not necessary to print an error. */
1707 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1708 } else {
1709 ERR("Relay didn't receive valid data_pending struct size : %d",
1710 ret);
1711 }
1712 ret = -1;
1713 goto end_no_session;
1714 }
1715
1716 stream_id = be64toh(msg.stream_id);
1717 last_net_seq_num = be64toh(msg.last_net_seq_num);
1718
1719 stream = stream_get_by_id(stream_id);
1720 if (stream == NULL) {
1721 ret = -1;
1722 goto end;
1723 }
1724
1725 pthread_mutex_lock(&stream->lock);
1726
1727 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1728 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1729 last_net_seq_num);
1730
1731 /* Avoid wrapping issue */
1732 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1733 /* Data has in fact been written and is NOT pending */
1734 ret = 0;
1735 } else {
1736 /* Data still being streamed thus pending */
1737 ret = 1;
1738 }
1739
1740 stream->data_pending_check_done = true;
1741 pthread_mutex_unlock(&stream->lock);
1742
1743 stream_put(stream);
1744 end:
1745
1746 memset(&reply, 0, sizeof(reply));
1747 reply.ret_code = htobe32(ret);
1748 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1749 if (ret < 0) {
1750 ERR("Relay data pending ret code failed");
1751 }
1752
1753 end_no_session:
1754 return ret;
1755 }
1756
1757 /*
1758 * Wait for the control socket to reach a quiescent state.
1759 *
1760 * Note that for now, when receiving this command from the session
1761 * daemon, this means that every subsequent commands or data received on
1762 * the control socket has been handled. So, this is why we simply return
1763 * OK here.
1764 */
1765 static int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1766 struct relay_connection *conn)
1767 {
1768 int ret;
1769 uint64_t stream_id;
1770 struct relay_stream *stream;
1771 struct lttcomm_relayd_quiescent_control msg;
1772 struct lttcomm_relayd_generic_reply reply;
1773
1774 DBG("Checking quiescent state on control socket");
1775
1776 if (!conn->session || conn->version_check_done == 0) {
1777 ERR("Trying to check for data before version check");
1778 ret = -1;
1779 goto end_no_session;
1780 }
1781
1782 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1783 if (ret < sizeof(msg)) {
1784 if (ret == 0) {
1785 /* Orderly shutdown. Not necessary to print an error. */
1786 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1787 } else {
1788 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1789 ret);
1790 }
1791 ret = -1;
1792 goto end_no_session;
1793 }
1794
1795 stream_id = be64toh(msg.stream_id);
1796 stream = stream_get_by_id(stream_id);
1797 if (!stream) {
1798 goto reply;
1799 }
1800 pthread_mutex_lock(&stream->lock);
1801 stream->data_pending_check_done = true;
1802 pthread_mutex_unlock(&stream->lock);
1803 DBG("Relay quiescent control pending flag set to %" PRIu64, stream_id);
1804 stream_put(stream);
1805 reply:
1806 memset(&reply, 0, sizeof(reply));
1807 reply.ret_code = htobe32(LTTNG_OK);
1808 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1809 if (ret < 0) {
1810 ERR("Relay data quiescent control ret code failed");
1811 }
1812
1813 end_no_session:
1814 return ret;
1815 }
1816
1817 /*
1818 * Initialize a data pending command. This means that a consumer is about
1819 * to ask for data pending for each stream it holds. Simply iterate over
1820 * all streams of a session and set the data_pending_check_done flag.
1821 *
1822 * This command returns to the client a LTTNG_OK code.
1823 */
1824 static int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1825 struct relay_connection *conn)
1826 {
1827 int ret;
1828 struct lttng_ht_iter iter;
1829 struct lttcomm_relayd_begin_data_pending msg;
1830 struct lttcomm_relayd_generic_reply reply;
1831 struct relay_stream *stream;
1832 uint64_t session_id;
1833
1834 assert(recv_hdr);
1835 assert(conn);
1836
1837 DBG("Init streams for data pending");
1838
1839 if (!conn->session || conn->version_check_done == 0) {
1840 ERR("Trying to check for data before version check");
1841 ret = -1;
1842 goto end_no_session;
1843 }
1844
1845 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1846 if (ret < sizeof(msg)) {
1847 if (ret == 0) {
1848 /* Orderly shutdown. Not necessary to print an error. */
1849 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1850 } else {
1851 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1852 ret);
1853 }
1854 ret = -1;
1855 goto end_no_session;
1856 }
1857
1858 session_id = be64toh(msg.session_id);
1859
1860 /*
1861 * Iterate over all streams to set the begin data pending flag.
1862 * For now, the streams are indexed by stream handle so we have
1863 * to iterate over all streams to find the one associated with
1864 * the right session_id.
1865 */
1866 rcu_read_lock();
1867 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1868 node.node) {
1869 if (!stream_get(stream)) {
1870 continue;
1871 }
1872 if (stream->trace->session->id == session_id) {
1873 pthread_mutex_lock(&stream->lock);
1874 stream->data_pending_check_done = false;
1875 pthread_mutex_unlock(&stream->lock);
1876 DBG("Set begin data pending flag to stream %" PRIu64,
1877 stream->stream_handle);
1878 }
1879 stream_put(stream);
1880 }
1881 rcu_read_unlock();
1882
1883 memset(&reply, 0, sizeof(reply));
1884 /* All good, send back reply. */
1885 reply.ret_code = htobe32(LTTNG_OK);
1886
1887 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1888 if (ret < 0) {
1889 ERR("Relay begin data pending send reply failed");
1890 }
1891
1892 end_no_session:
1893 return ret;
1894 }
1895
1896 /*
1897 * End data pending command. This will check, for a given session id, if
1898 * each stream associated with it has its data_pending_check_done flag
1899 * set. If not, this means that the client lost track of the stream but
1900 * the data is still being streamed on our side. In this case, we inform
1901 * the client that data is in flight.
1902 *
1903 * Return to the client if there is data in flight or not with a ret_code.
1904 */
1905 static int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1906 struct relay_connection *conn)
1907 {
1908 int ret;
1909 struct lttng_ht_iter iter;
1910 struct lttcomm_relayd_end_data_pending msg;
1911 struct lttcomm_relayd_generic_reply reply;
1912 struct relay_stream *stream;
1913 uint64_t session_id;
1914 uint32_t is_data_inflight = 0;
1915
1916 DBG("End data pending command");
1917
1918 if (!conn->session || conn->version_check_done == 0) {
1919 ERR("Trying to check for data before version check");
1920 ret = -1;
1921 goto end_no_session;
1922 }
1923
1924 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1925 if (ret < sizeof(msg)) {
1926 if (ret == 0) {
1927 /* Orderly shutdown. Not necessary to print an error. */
1928 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1929 } else {
1930 ERR("Relay didn't receive valid end data_pending struct size: %d",
1931 ret);
1932 }
1933 ret = -1;
1934 goto end_no_session;
1935 }
1936
1937 session_id = be64toh(msg.session_id);
1938
1939 /*
1940 * Iterate over all streams to see if the begin data pending
1941 * flag is set.
1942 */
1943 rcu_read_lock();
1944 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1945 node.node) {
1946 if (!stream_get(stream)) {
1947 continue;
1948 }
1949 if (stream->trace->session->id != session_id) {
1950 stream_put(stream);
1951 continue;
1952 }
1953 pthread_mutex_lock(&stream->lock);
1954 if (!stream->data_pending_check_done) {
1955 if (!stream->closed || !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
1956 is_data_inflight = 1;
1957 DBG("Data is still in flight for stream %" PRIu64,
1958 stream->stream_handle);
1959 pthread_mutex_unlock(&stream->lock);
1960 stream_put(stream);
1961 break;
1962 }
1963 }
1964 pthread_mutex_unlock(&stream->lock);
1965 stream_put(stream);
1966 }
1967 rcu_read_unlock();
1968
1969 memset(&reply, 0, sizeof(reply));
1970 /* All good, send back reply. */
1971 reply.ret_code = htobe32(is_data_inflight);
1972
1973 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1974 if (ret < 0) {
1975 ERR("Relay end data pending send reply failed");
1976 }
1977
1978 end_no_session:
1979 return ret;
1980 }
1981
1982 /*
1983 * Receive an index for a specific stream.
1984 *
1985 * Return 0 on success else a negative value.
1986 */
1987 static int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1988 struct relay_connection *conn)
1989 {
1990 int ret, send_ret;
1991 struct relay_session *session = conn->session;
1992 struct lttcomm_relayd_index index_info;
1993 struct relay_index *index;
1994 struct lttcomm_relayd_generic_reply reply;
1995 struct relay_stream *stream;
1996 uint64_t net_seq_num;
1997 size_t msg_len;
1998
1999 assert(conn);
2000
2001 DBG("Relay receiving index");
2002
2003 if (!session || conn->version_check_done == 0) {
2004 ERR("Trying to close a stream before version check");
2005 ret = -1;
2006 goto end_no_session;
2007 }
2008
2009 msg_len = lttcomm_relayd_index_len(
2010 lttng_to_index_major(conn->major, conn->minor),
2011 lttng_to_index_minor(conn->major, conn->minor));
2012 ret = conn->sock->ops->recvmsg(conn->sock, &index_info,
2013 msg_len, 0);
2014 if (ret < msg_len) {
2015 if (ret == 0) {
2016 /* Orderly shutdown. Not necessary to print an error. */
2017 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2018 } else {
2019 ERR("Relay didn't receive valid index struct size : %d", ret);
2020 }
2021 ret = -1;
2022 goto end_no_session;
2023 }
2024
2025 net_seq_num = be64toh(index_info.net_seq_num);
2026
2027 stream = stream_get_by_id(be64toh(index_info.relay_stream_id));
2028 if (!stream) {
2029 ERR("stream_get_by_id not found");
2030 ret = -1;
2031 goto end;
2032 }
2033 pthread_mutex_lock(&stream->lock);
2034
2035 /* Live beacon handling */
2036 if (index_info.packet_size == 0) {
2037 DBG("Received live beacon for stream %" PRIu64,
2038 stream->stream_handle);
2039
2040 /*
2041 * Only flag a stream inactive when it has already
2042 * received data and no indexes are in flight.
2043 */
2044 if (stream->index_received_seqcount > 0
2045 && stream->indexes_in_flight == 0) {
2046 stream->beacon_ts_end =
2047 be64toh(index_info.timestamp_end);
2048 }
2049 ret = 0;
2050 goto end_stream_put;
2051 } else {
2052 stream->beacon_ts_end = -1ULL;
2053 }
2054
2055 if (stream->ctf_stream_id == -1ULL) {
2056 stream->ctf_stream_id = be64toh(index_info.stream_id);
2057 }
2058 index = relay_index_get_by_id_or_create(stream, net_seq_num);
2059 if (!index) {
2060 ret = -1;
2061 ERR("relay_index_get_by_id_or_create index NULL");
2062 goto end_stream_put;
2063 }
2064 if (set_index_control_data(index, &index_info, conn)) {
2065 ERR("set_index_control_data error");
2066 relay_index_put(index);
2067 ret = -1;
2068 goto end_stream_put;
2069 }
2070 ret = relay_index_try_flush(index);
2071 if (ret == 0) {
2072 tracefile_array_commit_seq(stream->tfa);
2073 stream->index_received_seqcount++;
2074 } else if (ret > 0) {
2075 /* no flush. */
2076 ret = 0;
2077 } else {
2078 /*
2079 * ret < 0
2080 *
2081 * relay_index_try_flush is responsible for the self-reference
2082 * put of the index object on error.
2083 */
2084 ERR("relay_index_try_flush error %d", ret);
2085 ret = -1;
2086 }
2087
2088 end_stream_put:
2089 pthread_mutex_unlock(&stream->lock);
2090 stream_put(stream);
2091
2092 end:
2093
2094 memset(&reply, 0, sizeof(reply));
2095 if (ret < 0) {
2096 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2097 } else {
2098 reply.ret_code = htobe32(LTTNG_OK);
2099 }
2100 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2101 if (send_ret < 0) {
2102 ERR("Relay sending close index id reply");
2103 ret = send_ret;
2104 }
2105
2106 end_no_session:
2107 return ret;
2108 }
2109
2110 /*
2111 * Receive the streams_sent message.
2112 *
2113 * Return 0 on success else a negative value.
2114 */
2115 static int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
2116 struct relay_connection *conn)
2117 {
2118 int ret, send_ret;
2119 struct lttcomm_relayd_generic_reply reply;
2120
2121 assert(conn);
2122
2123 DBG("Relay receiving streams_sent");
2124
2125 if (!conn->session || conn->version_check_done == 0) {
2126 ERR("Trying to close a stream before version check");
2127 ret = -1;
2128 goto end_no_session;
2129 }
2130
2131 /*
2132 * Publish every pending stream in the connection recv list which are
2133 * now ready to be used by the viewer.
2134 */
2135 publish_connection_local_streams(conn);
2136
2137 memset(&reply, 0, sizeof(reply));
2138 reply.ret_code = htobe32(LTTNG_OK);
2139 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2140 if (send_ret < 0) {
2141 ERR("Relay sending sent_stream reply");
2142 ret = send_ret;
2143 } else {
2144 /* Success. */
2145 ret = 0;
2146 }
2147
2148 end_no_session:
2149 return ret;
2150 }
2151
2152 /*
2153 * Process the commands received on the control socket
2154 */
2155 static int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
2156 struct relay_connection *conn)
2157 {
2158 int ret = 0;
2159
2160 switch (be32toh(recv_hdr->cmd)) {
2161 case RELAYD_CREATE_SESSION:
2162 ret = relay_create_session(recv_hdr, conn);
2163 break;
2164 case RELAYD_ADD_STREAM:
2165 ret = relay_add_stream(recv_hdr, conn);
2166 break;
2167 case RELAYD_START_DATA:
2168 ret = relay_start(recv_hdr, conn);
2169 break;
2170 case RELAYD_SEND_METADATA:
2171 ret = relay_recv_metadata(recv_hdr, conn);
2172 break;
2173 case RELAYD_VERSION:
2174 ret = relay_send_version(recv_hdr, conn);
2175 break;
2176 case RELAYD_CLOSE_STREAM:
2177 ret = relay_close_stream(recv_hdr, conn);
2178 break;
2179 case RELAYD_DATA_PENDING:
2180 ret = relay_data_pending(recv_hdr, conn);
2181 break;
2182 case RELAYD_QUIESCENT_CONTROL:
2183 ret = relay_quiescent_control(recv_hdr, conn);
2184 break;
2185 case RELAYD_BEGIN_DATA_PENDING:
2186 ret = relay_begin_data_pending(recv_hdr, conn);
2187 break;
2188 case RELAYD_END_DATA_PENDING:
2189 ret = relay_end_data_pending(recv_hdr, conn);
2190 break;
2191 case RELAYD_SEND_INDEX:
2192 ret = relay_recv_index(recv_hdr, conn);
2193 break;
2194 case RELAYD_STREAMS_SENT:
2195 ret = relay_streams_sent(recv_hdr, conn);
2196 break;
2197 case RELAYD_RESET_METADATA:
2198 ret = relay_reset_metadata(recv_hdr, conn);
2199 break;
2200 case RELAYD_UPDATE_SYNC_INFO:
2201 default:
2202 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2203 relay_unknown_command(conn);
2204 ret = -1;
2205 goto end;
2206 }
2207
2208 end:
2209 return ret;
2210 }
2211
2212 /*
2213 * Handle index for a data stream.
2214 *
2215 * Called with the stream lock held.
2216 *
2217 * Return 0 on success else a negative value.
2218 */
2219 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2220 int rotate_index)
2221 {
2222 int ret = 0;
2223 uint64_t data_offset;
2224 struct relay_index *index;
2225
2226 /* Get data offset because we are about to update the index. */
2227 data_offset = htobe64(stream->tracefile_size_current);
2228
2229 DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64,
2230 stream->stream_handle, net_seq_num, stream->tracefile_size_current);
2231
2232 /*
2233 * Lookup for an existing index for that stream id/sequence
2234 * number. If it exists, the control thread has already received the
2235 * data for it, thus we need to write it to disk.
2236 */
2237 index = relay_index_get_by_id_or_create(stream, net_seq_num);
2238 if (!index) {
2239 ret = -1;
2240 goto end;
2241 }
2242
2243 if (rotate_index || !stream->index_file) {
2244 uint32_t major, minor;
2245
2246 /* Put ref on previous index_file. */
2247 if (stream->index_file) {
2248 lttng_index_file_put(stream->index_file);
2249 stream->index_file = NULL;
2250 }
2251 major = stream->trace->session->major;
2252 minor = stream->trace->session->minor;
2253 stream->index_file = lttng_index_file_create(stream->path_name,
2254 stream->channel_name,
2255 -1, -1, stream->tracefile_size,
2256 tracefile_array_get_file_index_head(stream->tfa),
2257 lttng_to_index_major(major, minor),
2258 lttng_to_index_minor(major, minor));
2259 if (!stream->index_file) {
2260 ret = -1;
2261 /* Put self-ref for this index due to error. */
2262 relay_index_put(index);
2263 index = NULL;
2264 goto end;
2265 }
2266 }
2267
2268 if (relay_index_set_file(index, stream->index_file, data_offset)) {
2269 ret = -1;
2270 /* Put self-ref for this index due to error. */
2271 relay_index_put(index);
2272 index = NULL;
2273 goto end;
2274 }
2275
2276 ret = relay_index_try_flush(index);
2277 if (ret == 0) {
2278 tracefile_array_commit_seq(stream->tfa);
2279 stream->index_received_seqcount++;
2280 } else if (ret > 0) {
2281 /* No flush. */
2282 ret = 0;
2283 } else {
2284 /*
2285 * ret < 0
2286 *
2287 * relay_index_try_flush is responsible for the self-reference
2288 * put of the index object on error.
2289 */
2290 ERR("relay_index_try_flush error %d", ret);
2291 ret = -1;
2292 }
2293 end:
2294 return ret;
2295 }
2296
2297 /*
2298 * relay_process_data: Process the data received on the data socket
2299 */
2300 static int relay_process_data(struct relay_connection *conn)
2301 {
2302 int ret = 0, rotate_index = 0;
2303 ssize_t size_ret;
2304 struct relay_stream *stream;
2305 struct lttcomm_relayd_data_hdr data_hdr;
2306 uint64_t stream_id;
2307 uint64_t net_seq_num;
2308 uint32_t data_size;
2309 struct relay_session *session;
2310 bool new_stream = false, close_requested = false;
2311 size_t chunk_size = RECV_DATA_BUFFER_SIZE;
2312 size_t recv_off = 0;
2313 char data_buffer[chunk_size];
2314
2315 ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr,
2316 sizeof(struct lttcomm_relayd_data_hdr), 0);
2317 if (ret <= 0) {
2318 if (ret == 0) {
2319 /* Orderly shutdown. Not necessary to print an error. */
2320 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2321 } else {
2322 ERR("Unable to receive data header on sock %d", conn->sock->fd);
2323 }
2324 ret = -1;
2325 goto end;
2326 }
2327
2328 stream_id = be64toh(data_hdr.stream_id);
2329 stream = stream_get_by_id(stream_id);
2330 if (!stream) {
2331 ERR("relay_process_data: Cannot find stream %" PRIu64, stream_id);
2332 ret = -1;
2333 goto end;
2334 }
2335 session = stream->trace->session;
2336 data_size = be32toh(data_hdr.data_size);
2337
2338 net_seq_num = be64toh(data_hdr.net_seq_num);
2339
2340 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2341 data_size, stream_id, net_seq_num);
2342
2343 pthread_mutex_lock(&stream->lock);
2344
2345 /* Check if a rotation is needed. */
2346 if (stream->tracefile_size > 0 &&
2347 (stream->tracefile_size_current + data_size) >
2348 stream->tracefile_size) {
2349 uint64_t old_id, new_id;
2350
2351 old_id = tracefile_array_get_file_index_head(stream->tfa);
2352 tracefile_array_file_rotate(stream->tfa);
2353
2354 /* new_id is updated by utils_rotate_stream_file. */
2355 new_id = old_id;
2356
2357 ret = utils_rotate_stream_file(stream->path_name,
2358 stream->channel_name, stream->tracefile_size,
2359 stream->tracefile_count, -1,
2360 -1, stream->stream_fd->fd,
2361 &new_id, &stream->stream_fd->fd);
2362 if (ret < 0) {
2363 ERR("Rotating stream output file");
2364 goto end_stream_unlock;
2365 }
2366 /*
2367 * Reset current size because we just performed a stream
2368 * rotation.
2369 */
2370 stream->tracefile_size_current = 0;
2371 rotate_index = 1;
2372 }
2373
2374 /*
2375 * Index are handled in protocol version 2.4 and above. Also,
2376 * snapshot and index are NOT supported.
2377 */
2378 if (session->minor >= 4 && !session->snapshot) {
2379 ret = handle_index_data(stream, net_seq_num, rotate_index);
2380 if (ret < 0) {
2381 ERR("handle_index_data: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
2382 stream->stream_handle, net_seq_num, ret);
2383 goto end_stream_unlock;
2384 }
2385 }
2386
2387 for (recv_off = 0; recv_off < data_size; recv_off += chunk_size) {
2388 size_t recv_size = min(data_size - recv_off, chunk_size);
2389
2390 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_size, 0);
2391 if (ret <= 0) {
2392 if (ret == 0) {
2393 /* Orderly shutdown. Not necessary to print an error. */
2394 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2395 } else {
2396 ERR("Socket %d error %d", conn->sock->fd, ret);
2397 }
2398 ret = -1;
2399 goto end_stream_unlock;
2400 }
2401
2402 /* Write data to stream output fd. */
2403 size_ret = lttng_write(stream->stream_fd->fd, data_buffer,
2404 recv_size);
2405 if (size_ret < recv_size) {
2406 ERR("Relay error writing data to file");
2407 ret = -1;
2408 goto end_stream_unlock;
2409 }
2410
2411 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64,
2412 size_ret, stream->stream_handle);
2413 }
2414
2415 ret = write_padding_to_file(stream->stream_fd->fd,
2416 be32toh(data_hdr.padding_size));
2417 if (ret < 0) {
2418 ERR("write_padding_to_file: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
2419 stream->stream_handle, net_seq_num, ret);
2420 goto end_stream_unlock;
2421 }
2422 stream->tracefile_size_current +=
2423 data_size + be32toh(data_hdr.padding_size);
2424 if (stream->prev_seq == -1ULL) {
2425 new_stream = true;
2426 }
2427
2428 stream->prev_seq = net_seq_num;
2429
2430 end_stream_unlock:
2431 close_requested = stream->close_requested;
2432 pthread_mutex_unlock(&stream->lock);
2433 if (close_requested) {
2434 try_stream_close(stream);
2435 }
2436
2437 if (new_stream) {
2438 pthread_mutex_lock(&session->lock);
2439 uatomic_set(&session->new_streams, 1);
2440 pthread_mutex_unlock(&session->lock);
2441 }
2442 stream_put(stream);
2443 end:
2444 return ret;
2445 }
2446
2447 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2448 {
2449 int ret;
2450
2451 (void) lttng_poll_del(events, pollfd);
2452
2453 ret = close(pollfd);
2454 if (ret < 0) {
2455 ERR("Closing pollfd %d", pollfd);
2456 }
2457 }
2458
2459 static void relay_thread_close_connection(struct lttng_poll_event *events,
2460 int pollfd, struct relay_connection *conn)
2461 {
2462 const char *type_str;
2463
2464 switch (conn->type) {
2465 case RELAY_DATA:
2466 type_str = "Data";
2467 break;
2468 case RELAY_CONTROL:
2469 type_str = "Control";
2470 break;
2471 case RELAY_VIEWER_COMMAND:
2472 type_str = "Viewer Command";
2473 break;
2474 case RELAY_VIEWER_NOTIFICATION:
2475 type_str = "Viewer Notification";
2476 break;
2477 default:
2478 type_str = "Unknown";
2479 }
2480 cleanup_connection_pollfd(events, pollfd);
2481 connection_put(conn);
2482 DBG("%s connection closed with %d", type_str, pollfd);
2483 }
2484
2485 /*
2486 * This thread does the actual work
2487 */
2488 static void *relay_thread_worker(void *data)
2489 {
2490 int ret, err = -1, last_seen_data_fd = -1;
2491 uint32_t nb_fd;
2492 struct lttng_poll_event events;
2493 struct lttng_ht *relay_connections_ht;
2494 struct lttng_ht_iter iter;
2495 struct lttcomm_relayd_hdr recv_hdr;
2496 struct relay_connection *destroy_conn = NULL;
2497
2498 DBG("[thread] Relay worker started");
2499
2500 rcu_register_thread();
2501
2502 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2503
2504 if (testpoint(relayd_thread_worker)) {
2505 goto error_testpoint;
2506 }
2507
2508 health_code_update();
2509
2510 /* table of connections indexed on socket */
2511 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2512 if (!relay_connections_ht) {
2513 goto relay_connections_ht_error;
2514 }
2515
2516 ret = create_thread_poll_set(&events, 2);
2517 if (ret < 0) {
2518 goto error_poll_create;
2519 }
2520
2521 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2522 if (ret < 0) {
2523 goto error;
2524 }
2525
2526 restart:
2527 while (1) {
2528 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2529
2530 health_code_update();
2531
2532 /* Infinite blocking call, waiting for transmission */
2533 DBG3("Relayd worker thread polling...");
2534 health_poll_entry();
2535 ret = lttng_poll_wait(&events, -1);
2536 health_poll_exit();
2537 if (ret < 0) {
2538 /*
2539 * Restart interrupted system call.
2540 */
2541 if (errno == EINTR) {
2542 goto restart;
2543 }
2544 goto error;
2545 }
2546
2547 nb_fd = ret;
2548
2549 /*
2550 * Process control. The control connection is
2551 * prioritized so we don't starve it with high
2552 * throughput tracing data on the data connection.
2553 */
2554 for (i = 0; i < nb_fd; i++) {
2555 /* Fetch once the poll data */
2556 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2557 int pollfd = LTTNG_POLL_GETFD(&events, i);
2558
2559 health_code_update();
2560
2561 if (!revents) {
2562 /*
2563 * No activity for this FD (poll
2564 * implementation).
2565 */
2566 continue;
2567 }
2568
2569 /* Thread quit pipe has been closed. Killing thread. */
2570 ret = check_thread_quit_pipe(pollfd, revents);
2571 if (ret) {
2572 err = 0;
2573 goto exit;
2574 }
2575
2576 /* Inspect the relay conn pipe for new connection */
2577 if (pollfd == relay_conn_pipe[0]) {
2578 if (revents & LPOLLIN) {
2579 struct relay_connection *conn;
2580
2581 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
2582 if (ret < 0) {
2583 goto error;
2584 }
2585 lttng_poll_add(&events, conn->sock->fd,
2586 LPOLLIN | LPOLLRDHUP);
2587 connection_ht_add(relay_connections_ht, conn);
2588 DBG("Connection socket %d added", conn->sock->fd);
2589 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2590 ERR("Relay connection pipe error");
2591 goto error;
2592 } else {
2593 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2594 goto error;
2595 }
2596 } else {
2597 struct relay_connection *ctrl_conn;
2598
2599 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
2600 /* If not found, there is a synchronization issue. */
2601 assert(ctrl_conn);
2602
2603 if (ctrl_conn->type == RELAY_DATA) {
2604 if (revents & LPOLLIN) {
2605 /*
2606 * Flag the last seen data fd not deleted. It will be
2607 * used as the last seen fd if any fd gets deleted in
2608 * this first loop.
2609 */
2610 last_notdel_data_fd = pollfd;
2611 }
2612 goto put_ctrl_connection;
2613 }
2614 assert(ctrl_conn->type == RELAY_CONTROL);
2615
2616 if (revents & LPOLLIN) {
2617 ret = ctrl_conn->sock->ops->recvmsg(ctrl_conn->sock,
2618 &recv_hdr, sizeof(recv_hdr), 0);
2619 if (ret <= 0) {
2620 /* Connection closed */
2621 relay_thread_close_connection(&events, pollfd,
2622 ctrl_conn);
2623 } else {
2624 ret = relay_process_control(&recv_hdr, ctrl_conn);
2625 if (ret < 0) {
2626 /* Clear the session on error. */
2627 relay_thread_close_connection(&events,
2628 pollfd, ctrl_conn);
2629 }
2630 seen_control = 1;
2631 }
2632 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2633 relay_thread_close_connection(&events,
2634 pollfd, ctrl_conn);
2635 if (last_seen_data_fd == pollfd) {
2636 last_seen_data_fd = last_notdel_data_fd;
2637 }
2638 } else {
2639 ERR("Unexpected poll events %u for control sock %d",
2640 revents, pollfd);
2641 connection_put(ctrl_conn);
2642 goto error;
2643 }
2644 put_ctrl_connection:
2645 connection_put(ctrl_conn);
2646 }
2647 }
2648
2649 /*
2650 * The last loop handled a control request, go back to poll to make
2651 * sure we prioritise the control socket.
2652 */
2653 if (seen_control) {
2654 continue;
2655 }
2656
2657 if (last_seen_data_fd >= 0) {
2658 for (i = 0; i < nb_fd; i++) {
2659 int pollfd = LTTNG_POLL_GETFD(&events, i);
2660
2661 health_code_update();
2662
2663 if (last_seen_data_fd == pollfd) {
2664 idx = i;
2665 break;
2666 }
2667 }
2668 }
2669
2670 /* Process data connection. */
2671 for (i = idx + 1; i < nb_fd; i++) {
2672 /* Fetch the poll data. */
2673 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2674 int pollfd = LTTNG_POLL_GETFD(&events, i);
2675 struct relay_connection *data_conn;
2676
2677 health_code_update();
2678
2679 if (!revents) {
2680 /* No activity for this FD (poll implementation). */
2681 continue;
2682 }
2683
2684 /* Skip the command pipe. It's handled in the first loop. */
2685 if (pollfd == relay_conn_pipe[0]) {
2686 continue;
2687 }
2688
2689 data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
2690 if (!data_conn) {
2691 /* Skip it. Might be removed before. */
2692 continue;
2693 }
2694 if (data_conn->type == RELAY_CONTROL) {
2695 goto put_data_connection;
2696 }
2697 assert(data_conn->type == RELAY_DATA);
2698
2699 if (revents & LPOLLIN) {
2700 ret = relay_process_data(data_conn);
2701 /* Connection closed */
2702 if (ret < 0) {
2703 relay_thread_close_connection(&events, pollfd,
2704 data_conn);
2705 /*
2706 * Every goto restart call sets the last seen fd where
2707 * here we don't really care since we gracefully
2708 * continue the loop after the connection is deleted.
2709 */
2710 } else {
2711 /* Keep last seen port. */
2712 last_seen_data_fd = pollfd;
2713 connection_put(data_conn);
2714 goto restart;
2715 }
2716 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2717 relay_thread_close_connection(&events, pollfd,
2718 data_conn);
2719 } else {
2720 ERR("Unknown poll events %u for data sock %d",
2721 revents, pollfd);
2722 }
2723 put_data_connection:
2724 connection_put(data_conn);
2725 }
2726 last_seen_data_fd = -1;
2727 }
2728
2729 /* Normal exit, no error */
2730 ret = 0;
2731
2732 exit:
2733 error:
2734 /* Cleanup reamaining connection object. */
2735 rcu_read_lock();
2736 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
2737 destroy_conn,
2738 sock_n.node) {
2739 health_code_update();
2740
2741 if (session_abort(destroy_conn->session)) {
2742 assert(0);
2743 }
2744
2745 /*
2746 * No need to grab another ref, because we own
2747 * destroy_conn.
2748 */
2749 relay_thread_close_connection(&events, destroy_conn->sock->fd,
2750 destroy_conn);
2751 }
2752 rcu_read_unlock();
2753
2754 lttng_poll_clean(&events);
2755 error_poll_create:
2756 lttng_ht_destroy(relay_connections_ht);
2757 relay_connections_ht_error:
2758 /* Close relay conn pipes */
2759 utils_close_pipe(relay_conn_pipe);
2760 if (err) {
2761 DBG("Thread exited with error");
2762 }
2763 DBG("Worker thread cleanup complete");
2764 error_testpoint:
2765 if (err) {
2766 health_error();
2767 ERR("Health error occurred in %s", __func__);
2768 }
2769 health_unregister(health_relayd);
2770 rcu_unregister_thread();
2771 lttng_relay_stop_threads();
2772 return NULL;
2773 }
2774
2775 /*
2776 * Create the relay command pipe to wake thread_manage_apps.
2777 * Closed in cleanup().
2778 */
2779 static int create_relay_conn_pipe(void)
2780 {
2781 int ret;
2782
2783 ret = utils_create_pipe_cloexec(relay_conn_pipe);
2784
2785 return ret;
2786 }
2787
2788 /*
2789 * main
2790 */
2791 int main(int argc, char **argv)
2792 {
2793 int ret = 0, retval = 0;
2794 void *status;
2795
2796 /* Parse arguments */
2797 progname = argv[0];
2798 if (set_options(argc, argv)) {
2799 retval = -1;
2800 goto exit_options;
2801 }
2802
2803 if (set_signal_handler()) {
2804 retval = -1;
2805 goto exit_options;
2806 }
2807
2808 if (!opt_allow_clear) {
2809 DBG("Clear command disallowed.");
2810 }
2811
2812 /* Try to create directory if -o, --output is specified. */
2813 if (opt_output_path) {
2814 if (*opt_output_path != '/') {
2815 ERR("Please specify an absolute path for -o, --output PATH");
2816 retval = -1;
2817 goto exit_options;
2818 }
2819
2820 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG,
2821 -1, -1);
2822 if (ret < 0) {
2823 ERR("Unable to create %s", opt_output_path);
2824 retval = -1;
2825 goto exit_options;
2826 }
2827 }
2828
2829 /* Daemonize */
2830 if (opt_daemon || opt_background) {
2831 int i;
2832
2833 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2834 !opt_background);
2835 if (ret < 0) {
2836 retval = -1;
2837 goto exit_options;
2838 }
2839
2840 /*
2841 * We are in the child. Make sure all other file
2842 * descriptors are closed, in case we are called with
2843 * more opened file descriptors than the standard ones.
2844 */
2845 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2846 (void) close(i);
2847 }
2848 }
2849
2850 /* Initialize thread health monitoring */
2851 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2852 if (!health_relayd) {
2853 PERROR("health_app_create error");
2854 retval = -1;
2855 goto exit_health_app_create;
2856 }
2857
2858 /* Create thread quit pipe */
2859 if (init_thread_quit_pipe()) {
2860 retval = -1;
2861 goto exit_init_data;
2862 }
2863
2864 /* Setup the thread apps communication pipe. */
2865 if (create_relay_conn_pipe()) {
2866 retval = -1;
2867 goto exit_init_data;
2868 }
2869
2870 /* Init relay command queue. */
2871 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
2872
2873 /* Initialize communication library */
2874 lttcomm_init();
2875 lttcomm_inet_init();
2876
2877 /* tables of sessions indexed by session ID */
2878 sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2879 if (!sessions_ht) {
2880 retval = -1;
2881 goto exit_init_data;
2882 }
2883
2884 /* tables of streams indexed by stream ID */
2885 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2886 if (!relay_streams_ht) {
2887 retval = -1;
2888 goto exit_init_data;
2889 }
2890
2891 /* tables of streams indexed by stream ID */
2892 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2893 if (!viewer_streams_ht) {
2894 retval = -1;
2895 goto exit_init_data;
2896 }
2897
2898 ret = utils_create_pipe(health_quit_pipe);
2899 if (ret) {
2900 retval = -1;
2901 goto exit_health_quit_pipe;
2902 }
2903
2904 /* Create thread to manage the client socket */
2905 ret = pthread_create(&health_thread, default_pthread_attr(),
2906 thread_manage_health, (void *) NULL);
2907 if (ret) {
2908 errno = ret;
2909 PERROR("pthread_create health");
2910 retval = -1;
2911 goto exit_health_thread;
2912 }
2913
2914 /* Setup the dispatcher thread */
2915 ret = pthread_create(&dispatcher_thread, default_pthread_attr(),
2916 relay_thread_dispatcher, (void *) NULL);
2917 if (ret) {
2918 errno = ret;
2919 PERROR("pthread_create dispatcher");
2920 retval = -1;
2921 goto exit_dispatcher_thread;
2922 }
2923
2924 /* Setup the worker thread */
2925 ret = pthread_create(&worker_thread, default_pthread_attr(),
2926 relay_thread_worker, NULL);
2927 if (ret) {
2928 errno = ret;
2929 PERROR("pthread_create worker");
2930 retval = -1;
2931 goto exit_worker_thread;
2932 }
2933
2934 /* Setup the listener thread */
2935 ret = pthread_create(&listener_thread, default_pthread_attr(),
2936 relay_thread_listener, (void *) NULL);
2937 if (ret) {
2938 errno = ret;
2939 PERROR("pthread_create listener");
2940 retval = -1;
2941 goto exit_listener_thread;
2942 }
2943
2944 ret = relayd_live_create(live_uri);
2945 if (ret) {
2946 ERR("Starting live viewer threads");
2947 retval = -1;
2948 goto exit_live;
2949 }
2950
2951 /*
2952 * This is where we start awaiting program completion (e.g. through
2953 * signal that asks threads to teardown).
2954 */
2955
2956 ret = relayd_live_join();
2957 if (ret) {
2958 retval = -1;
2959 }
2960 exit_live:
2961
2962 ret = pthread_join(listener_thread, &status);
2963 if (ret) {
2964 errno = ret;
2965 PERROR("pthread_join listener_thread");
2966 retval = -1;
2967 }
2968
2969 exit_listener_thread:
2970 ret = pthread_join(worker_thread, &status);
2971 if (ret) {
2972 errno = ret;
2973 PERROR("pthread_join worker_thread");
2974 retval = -1;
2975 }
2976
2977 exit_worker_thread:
2978 ret = pthread_join(dispatcher_thread, &status);
2979 if (ret) {
2980 errno = ret;
2981 PERROR("pthread_join dispatcher_thread");
2982 retval = -1;
2983 }
2984 exit_dispatcher_thread:
2985
2986 ret = pthread_join(health_thread, &status);
2987 if (ret) {
2988 errno = ret;
2989 PERROR("pthread_join health_thread");
2990 retval = -1;
2991 }
2992 exit_health_thread:
2993
2994 utils_close_pipe(health_quit_pipe);
2995 exit_health_quit_pipe:
2996
2997 exit_init_data:
2998 health_app_destroy(health_relayd);
2999 exit_health_app_create:
3000 exit_options:
3001 /*
3002 * Wait for all pending call_rcu work to complete before tearing
3003 * down data structures. call_rcu worker may be trying to
3004 * perform lookups in those structures.
3005 */
3006 rcu_barrier();
3007 relayd_cleanup();
3008
3009 /* Ensure all prior call_rcu are done. */
3010 rcu_barrier();
3011
3012 if (!retval) {
3013 exit(EXIT_SUCCESS);
3014 } else {
3015 exit(EXIT_FAILURE);
3016 }
3017 }
This page took 0.140604 seconds and 5 git commands to generate.