Cleanup: Reduce scope of connections in main relayd thread
[lttng-tools.git] / src / bin / lttng-relayd / main.c
1 /*
2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _GNU_SOURCE
21 #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 #include <config.h>
43
44 #include <lttng/lttng.h>
45 #include <common/common.h>
46 #include <common/compat/poll.h>
47 #include <common/compat/socket.h>
48 #include <common/compat/endian.h>
49 #include <common/compat/getenv.h>
50 #include <common/defaults.h>
51 #include <common/daemonize.h>
52 #include <common/futex.h>
53 #include <common/sessiond-comm/sessiond-comm.h>
54 #include <common/sessiond-comm/inet.h>
55 #include <common/sessiond-comm/relayd.h>
56 #include <common/uri.h>
57 #include <common/utils.h>
58 #include <common/config/config.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
73 /* command line options */
74 char *opt_output_path;
75 static int opt_daemon, opt_background;
76
77 /*
78 * We need to wait for listener and live listener threads, as well as
79 * health check thread, before being ready to signal readiness.
80 */
81 #define NR_LTTNG_RELAY_READY 3
82 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
83 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
84 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
85
86 static struct lttng_uri *control_uri;
87 static struct lttng_uri *data_uri;
88 static struct lttng_uri *live_uri;
89
90 const char *progname;
91
92 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
93 static int tracing_group_name_override;
94
95 const char * const config_section_name = "relayd";
96
97 /*
98 * Quit pipe for all threads. This permits a single cancellation point
99 * for all threads when receiving an event on the pipe.
100 */
101 int thread_quit_pipe[2] = { -1, -1 };
102
103 /*
104 * This pipe is used to inform the worker thread that a command is queued and
105 * ready to be processed.
106 */
107 static int relay_conn_pipe[2] = { -1, -1 };
108
109 /* Shared between threads */
110 static int dispatch_thread_exit;
111
112 static pthread_t listener_thread;
113 static pthread_t dispatcher_thread;
114 static pthread_t worker_thread;
115 static pthread_t health_thread;
116
117 static uint64_t last_relay_stream_id;
118
119 /*
120 * Relay command queue.
121 *
122 * The relay_thread_listener and relay_thread_dispatcher communicate with this
123 * queue.
124 */
125 static struct relay_conn_queue relay_conn_queue;
126
127 /* buffer allocated at startup, used to store the trace data */
128 static char *data_buffer;
129 static unsigned int data_buffer_size;
130
131 /* We need those values for the file/dir creation. */
132 static uid_t relayd_uid;
133 static gid_t relayd_gid;
134
135 /* Global relay stream hash table. */
136 struct lttng_ht *relay_streams_ht;
137
138 /* Global relay viewer stream hash table. */
139 struct lttng_ht *viewer_streams_ht;
140
141 /* Global hash table that stores relay index object. */
142 struct lttng_ht *indexes_ht;
143
144 /* Relayd health monitoring */
145 struct health_app *health_relayd;
146
147 static struct option long_options[] = {
148 { "control-port", 1, 0, 'C', },
149 { "data-port", 1, 0, 'D', },
150 { "live-port", 1, 0, 'L', },
151 { "daemonize", 0, 0, 'd', },
152 { "background", 0, 0, 'b', },
153 { "group", 1, 0, 'g', },
154 { "help", 0, 0, 'h', },
155 { "output", 1, 0, 'o', },
156 { "verbose", 0, 0, 'v', },
157 { "config", 1, 0, 'f' },
158 { NULL, 0, 0, 0, },
159 };
160
161 static const char *config_ignore_options[] = { "help", "config" };
162
163 /*
164 * usage function on stderr
165 */
166 static
167 void usage(void)
168 {
169 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
170 fprintf(stderr, " -h, --help Display this usage.\n");
171 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
172 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
173 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
174 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
175 fprintf(stderr, " -L, --live-port URL Live view port listening.\n");
176 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
177 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
178 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
179 fprintf(stderr, " -f --config Load daemon configuration file\n");
180 }
181
182 /*
183 * Take an option from the getopt output and set it in the right variable to be
184 * used later.
185 *
186 * Return 0 on success else a negative value.
187 */
188 static
189 int set_option(int opt, const char *arg, const char *optname)
190 {
191 int ret;
192
193 switch (opt) {
194 case 0:
195 fprintf(stderr, "option %s", optname);
196 if (arg) {
197 fprintf(stderr, " with arg %s\n", arg);
198 }
199 break;
200 case 'C':
201 if (lttng_is_setuid_setgid()) {
202 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
203 "-C, --control-port");
204 } else {
205 ret = uri_parse(arg, &control_uri);
206 if (ret < 0) {
207 ERR("Invalid control URI specified");
208 goto end;
209 }
210 if (control_uri->port == 0) {
211 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
212 }
213 }
214 break;
215 case 'D':
216 if (lttng_is_setuid_setgid()) {
217 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
218 "-D, -data-port");
219 } else {
220 ret = uri_parse(arg, &data_uri);
221 if (ret < 0) {
222 ERR("Invalid data URI specified");
223 goto end;
224 }
225 if (data_uri->port == 0) {
226 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
227 }
228 }
229 break;
230 case 'L':
231 if (lttng_is_setuid_setgid()) {
232 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
233 "-L, -live-port");
234 } else {
235 ret = uri_parse(arg, &live_uri);
236 if (ret < 0) {
237 ERR("Invalid live URI specified");
238 goto end;
239 }
240 if (live_uri->port == 0) {
241 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
242 }
243 }
244 break;
245 case 'd':
246 opt_daemon = 1;
247 break;
248 case 'b':
249 opt_background = 1;
250 break;
251 case 'g':
252 if (lttng_is_setuid_setgid()) {
253 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
254 "-g, --group");
255 } else {
256 tracing_group_name = strdup(arg);
257 if (tracing_group_name == NULL) {
258 ret = -errno;
259 PERROR("strdup");
260 goto end;
261 }
262 tracing_group_name_override = 1;
263 }
264 break;
265 case 'h':
266 usage();
267 exit(EXIT_FAILURE);
268 case 'o':
269 if (lttng_is_setuid_setgid()) {
270 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
271 "-o, --output");
272 } else {
273 ret = asprintf(&opt_output_path, "%s", arg);
274 if (ret < 0) {
275 ret = -errno;
276 PERROR("asprintf opt_output_path");
277 goto end;
278 }
279 }
280 break;
281 case 'v':
282 /* Verbose level can increase using multiple -v */
283 if (arg) {
284 lttng_opt_verbose = config_parse_value(arg);
285 } else {
286 /* Only 3 level of verbosity (-vvv). */
287 if (lttng_opt_verbose < 3) {
288 lttng_opt_verbose += 1;
289 }
290 }
291 break;
292 default:
293 /* Unknown option or other error.
294 * Error is printed by getopt, just return */
295 ret = -1;
296 goto end;
297 }
298
299 /* All good. */
300 ret = 0;
301
302 end:
303 return ret;
304 }
305
306 /*
307 * config_entry_handler_cb used to handle options read from a config file.
308 * See config_entry_handler_cb comment in common/config/config.h for the
309 * return value conventions.
310 */
311 static
312 int config_entry_handler(const struct config_entry *entry, void *unused)
313 {
314 int ret = 0, i;
315
316 if (!entry || !entry->name || !entry->value) {
317 ret = -EINVAL;
318 goto end;
319 }
320
321 /* Check if the option is to be ignored */
322 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
323 if (!strcmp(entry->name, config_ignore_options[i])) {
324 goto end;
325 }
326 }
327
328 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
329 /* Ignore if entry name is not fully matched. */
330 if (strcmp(entry->name, long_options[i].name)) {
331 continue;
332 }
333
334 /*
335 * If the option takes no argument on the command line, we have to
336 * check if the value is "true". We support non-zero numeric values,
337 * true, on and yes.
338 */
339 if (!long_options[i].has_arg) {
340 ret = config_parse_value(entry->value);
341 if (ret <= 0) {
342 if (ret) {
343 WARN("Invalid configuration value \"%s\" for option %s",
344 entry->value, entry->name);
345 }
346 /* False, skip boolean config option. */
347 goto end;
348 }
349 }
350
351 ret = set_option(long_options[i].val, entry->value, entry->name);
352 goto end;
353 }
354
355 WARN("Unrecognized option \"%s\" in daemon configuration file.",
356 entry->name);
357
358 end:
359 return ret;
360 }
361
362 static
363 int set_options(int argc, char **argv)
364 {
365 int c, ret = 0, option_index = 0, retval = 0;
366 int orig_optopt = optopt, orig_optind = optind;
367 char *default_address, *optstring;
368 const char *config_path = NULL;
369
370 optstring = utils_generate_optstring(long_options,
371 sizeof(long_options) / sizeof(struct option));
372 if (!optstring) {
373 retval = -ENOMEM;
374 goto exit;
375 }
376
377 /* Check for the --config option */
378
379 while ((c = getopt_long(argc, argv, optstring, long_options,
380 &option_index)) != -1) {
381 if (c == '?') {
382 retval = -EINVAL;
383 goto exit;
384 } else if (c != 'f') {
385 continue;
386 }
387
388 if (lttng_is_setuid_setgid()) {
389 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
390 "-f, --config");
391 } else {
392 config_path = utils_expand_path(optarg);
393 if (!config_path) {
394 ERR("Failed to resolve path: %s", optarg);
395 }
396 }
397 }
398
399 ret = config_get_section_entries(config_path, config_section_name,
400 config_entry_handler, NULL);
401 if (ret) {
402 if (ret > 0) {
403 ERR("Invalid configuration option at line %i", ret);
404 }
405 retval = -1;
406 goto exit;
407 }
408
409 /* Reset getopt's global state */
410 optopt = orig_optopt;
411 optind = orig_optind;
412 while (1) {
413 c = getopt_long(argc, argv, optstring, long_options, &option_index);
414 if (c == -1) {
415 break;
416 }
417
418 ret = set_option(c, optarg, long_options[option_index].name);
419 if (ret < 0) {
420 retval = -1;
421 goto exit;
422 }
423 }
424
425 /* assign default values */
426 if (control_uri == NULL) {
427 ret = asprintf(&default_address,
428 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
429 DEFAULT_NETWORK_CONTROL_PORT);
430 if (ret < 0) {
431 PERROR("asprintf default data address");
432 retval = -1;
433 goto exit;
434 }
435
436 ret = uri_parse(default_address, &control_uri);
437 free(default_address);
438 if (ret < 0) {
439 ERR("Invalid control URI specified");
440 retval = -1;
441 goto exit;
442 }
443 }
444 if (data_uri == NULL) {
445 ret = asprintf(&default_address,
446 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
447 DEFAULT_NETWORK_DATA_PORT);
448 if (ret < 0) {
449 PERROR("asprintf default data address");
450 retval = -1;
451 goto exit;
452 }
453
454 ret = uri_parse(default_address, &data_uri);
455 free(default_address);
456 if (ret < 0) {
457 ERR("Invalid data URI specified");
458 retval = -1;
459 goto exit;
460 }
461 }
462 if (live_uri == NULL) {
463 ret = asprintf(&default_address,
464 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
465 DEFAULT_NETWORK_VIEWER_PORT);
466 if (ret < 0) {
467 PERROR("asprintf default viewer control address");
468 retval = -1;
469 goto exit;
470 }
471
472 ret = uri_parse(default_address, &live_uri);
473 free(default_address);
474 if (ret < 0) {
475 ERR("Invalid viewer control URI specified");
476 retval = -1;
477 goto exit;
478 }
479 }
480
481 exit:
482 free(optstring);
483 return retval;
484 }
485
486 /*
487 * Cleanup the daemon
488 */
489 static
490 void relayd_cleanup(struct relay_local_data *relay_ctx)
491 {
492 DBG("Cleaning up");
493
494 if (viewer_streams_ht)
495 lttng_ht_destroy(viewer_streams_ht);
496 if (relay_streams_ht)
497 lttng_ht_destroy(relay_streams_ht);
498 if (relay_ctx && relay_ctx->sessions_ht)
499 lttng_ht_destroy(relay_ctx->sessions_ht);
500 free(relay_ctx);
501
502 /* free the dynamically allocated opt_output_path */
503 free(opt_output_path);
504
505 /* Close thread quit pipes */
506 utils_close_pipe(thread_quit_pipe);
507
508 uri_free(control_uri);
509 uri_free(data_uri);
510 /* Live URI is freed in the live thread. */
511
512 if (tracing_group_name_override) {
513 free((void *) tracing_group_name);
514 }
515 }
516
517 /*
518 * Write to writable pipe used to notify a thread.
519 */
520 static
521 int notify_thread_pipe(int wpipe)
522 {
523 ssize_t ret;
524
525 ret = lttng_write(wpipe, "!", 1);
526 if (ret < 1) {
527 PERROR("write poll pipe");
528 goto end;
529 }
530 ret = 0;
531 end:
532 return ret;
533 }
534
535 static
536 int notify_health_quit_pipe(int *pipe)
537 {
538 ssize_t ret;
539
540 ret = lttng_write(pipe[1], "4", 1);
541 if (ret < 1) {
542 PERROR("write relay health quit");
543 goto end;
544 }
545 ret = 0;
546 end:
547 return ret;
548 }
549
550 /*
551 * Stop all relayd and relayd-live threads.
552 */
553 int lttng_relay_stop_threads(void)
554 {
555 int retval = 0;
556
557 /* Stopping all threads */
558 DBG("Terminating all threads");
559 if (notify_thread_pipe(thread_quit_pipe[1])) {
560 ERR("write error on thread quit pipe");
561 retval = -1;
562 }
563
564 if (notify_health_quit_pipe(health_quit_pipe)) {
565 ERR("write error on health quit pipe");
566 }
567
568 /* Dispatch thread */
569 CMM_STORE_SHARED(dispatch_thread_exit, 1);
570 futex_nto1_wake(&relay_conn_queue.futex);
571
572 if (relayd_live_stop()) {
573 ERR("Error stopping live threads");
574 retval = -1;
575 }
576 return retval;
577 }
578
579 /*
580 * Signal handler for the daemon
581 *
582 * Simply stop all worker threads, leaving main() return gracefully after
583 * joining all threads and calling cleanup().
584 */
585 static
586 void sighandler(int sig)
587 {
588 switch (sig) {
589 case SIGPIPE:
590 DBG("SIGPIPE caught");
591 return;
592 case SIGINT:
593 DBG("SIGINT caught");
594 if (lttng_relay_stop_threads()) {
595 ERR("Error stopping threads");
596 }
597 break;
598 case SIGTERM:
599 DBG("SIGTERM caught");
600 if (lttng_relay_stop_threads()) {
601 ERR("Error stopping threads");
602 }
603 break;
604 case SIGUSR1:
605 CMM_STORE_SHARED(recv_child_signal, 1);
606 break;
607 default:
608 break;
609 }
610 }
611
612 /*
613 * Setup signal handler for :
614 * SIGINT, SIGTERM, SIGPIPE
615 */
616 static
617 int set_signal_handler(void)
618 {
619 int ret = 0;
620 struct sigaction sa;
621 sigset_t sigset;
622
623 if ((ret = sigemptyset(&sigset)) < 0) {
624 PERROR("sigemptyset");
625 return ret;
626 }
627
628 sa.sa_handler = sighandler;
629 sa.sa_mask = sigset;
630 sa.sa_flags = 0;
631 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
632 PERROR("sigaction");
633 return ret;
634 }
635
636 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
637 PERROR("sigaction");
638 return ret;
639 }
640
641 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
642 PERROR("sigaction");
643 return ret;
644 }
645
646 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
647 PERROR("sigaction");
648 return ret;
649 }
650
651 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
652
653 return ret;
654 }
655
656 void lttng_relay_notify_ready(void)
657 {
658 /* Notify the parent of the fork() process that we are ready. */
659 if (opt_daemon || opt_background) {
660 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
661 kill(child_ppid, SIGUSR1);
662 }
663 }
664 }
665
666 /*
667 * Init thread quit pipe.
668 *
669 * Return -1 on error or 0 if all pipes are created.
670 */
671 static
672 int init_thread_quit_pipe(void)
673 {
674 int ret;
675
676 ret = utils_create_pipe_cloexec(thread_quit_pipe);
677
678 return ret;
679 }
680
681 /*
682 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
683 */
684 static
685 int create_thread_poll_set(struct lttng_poll_event *events, int size)
686 {
687 int ret;
688
689 if (events == NULL || size == 0) {
690 ret = -1;
691 goto error;
692 }
693
694 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
695 if (ret < 0) {
696 goto error;
697 }
698
699 /* Add quit pipe */
700 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
701 if (ret < 0) {
702 goto error;
703 }
704
705 return 0;
706
707 error:
708 return ret;
709 }
710
711 /*
712 * Check if the thread quit pipe was triggered.
713 *
714 * Return 1 if it was triggered else 0;
715 */
716 static
717 int check_thread_quit_pipe(int fd, uint32_t events)
718 {
719 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
720 return 1;
721 }
722
723 return 0;
724 }
725
726 /*
727 * Create and init socket from uri.
728 */
729 static
730 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
731 {
732 int ret;
733 struct lttcomm_sock *sock = NULL;
734
735 sock = lttcomm_alloc_sock_from_uri(uri);
736 if (sock == NULL) {
737 ERR("Allocating socket");
738 goto error;
739 }
740
741 ret = lttcomm_create_sock(sock);
742 if (ret < 0) {
743 goto error;
744 }
745 DBG("Listening on sock %d", sock->fd);
746
747 ret = sock->ops->bind(sock);
748 if (ret < 0) {
749 goto error;
750 }
751
752 ret = sock->ops->listen(sock, -1);
753 if (ret < 0) {
754 goto error;
755
756 }
757
758 return sock;
759
760 error:
761 if (sock) {
762 lttcomm_destroy_sock(sock);
763 }
764 return NULL;
765 }
766
767 /*
768 * Return nonzero if stream needs to be closed.
769 */
770 static
771 int close_stream_check(struct relay_stream *stream)
772 {
773 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
774 /*
775 * We are about to close the stream so set the data pending flag to 1
776 * which will make the end data pending command skip the stream which
777 * is now closed and ready. Note that after proceeding to a file close,
778 * the written file is ready for reading.
779 */
780 stream->data_pending_check_done = 1;
781 return 1;
782 }
783 return 0;
784 }
785
786 static void try_close_stream(struct relay_session *session,
787 struct relay_stream *stream)
788 {
789 int ret;
790 struct ctf_trace *ctf_trace;
791
792 assert(session);
793 assert(stream);
794
795 if (!close_stream_check(stream)) {
796 /* Can't close it, not ready for that. */
797 goto end;
798 }
799
800 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
801 stream->path_name);
802 assert(ctf_trace);
803
804 pthread_mutex_lock(&session->viewer_ready_lock);
805 ctf_trace->invalid_flag = 1;
806 pthread_mutex_unlock(&session->viewer_ready_lock);
807
808 ret = stream_close(session, stream);
809 if (ret || session->snapshot) {
810 /* Already close thus the ctf trace is being or has been destroyed. */
811 goto end;
812 }
813
814 ctf_trace_try_destroy(session, ctf_trace);
815
816 end:
817 return;
818 }
819
820 /*
821 * This thread manages the listening for new connections on the network
822 */
823 static
824 void *relay_thread_listener(void *data)
825 {
826 int i, ret, pollfd, err = -1;
827 uint32_t revents, nb_fd;
828 struct lttng_poll_event events;
829 struct lttcomm_sock *control_sock, *data_sock;
830
831 DBG("[thread] Relay listener started");
832
833 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
834
835 health_code_update();
836
837 control_sock = relay_init_sock(control_uri);
838 if (!control_sock) {
839 goto error_sock_control;
840 }
841
842 data_sock = relay_init_sock(data_uri);
843 if (!data_sock) {
844 goto error_sock_relay;
845 }
846
847 /*
848 * Pass 3 as size here for the thread quit pipe, control and data socket.
849 */
850 ret = create_thread_poll_set(&events, 3);
851 if (ret < 0) {
852 goto error_create_poll;
853 }
854
855 /* Add the control socket */
856 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
857 if (ret < 0) {
858 goto error_poll_add;
859 }
860
861 /* Add the data socket */
862 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
863 if (ret < 0) {
864 goto error_poll_add;
865 }
866
867 lttng_relay_notify_ready();
868
869 if (testpoint(relayd_thread_listener)) {
870 goto error_testpoint;
871 }
872
873 while (1) {
874 health_code_update();
875
876 DBG("Listener accepting connections");
877
878 restart:
879 health_poll_entry();
880 ret = lttng_poll_wait(&events, -1);
881 health_poll_exit();
882 if (ret < 0) {
883 /*
884 * Restart interrupted system call.
885 */
886 if (errno == EINTR) {
887 goto restart;
888 }
889 goto error;
890 }
891
892 nb_fd = ret;
893
894 DBG("Relay new connection received");
895 for (i = 0; i < nb_fd; i++) {
896 health_code_update();
897
898 /* Fetch once the poll data */
899 revents = LTTNG_POLL_GETEV(&events, i);
900 pollfd = LTTNG_POLL_GETFD(&events, i);
901
902 if (!revents) {
903 /* No activity for this FD (poll implementation). */
904 continue;
905 }
906
907 /* Thread quit pipe has been closed. Killing thread. */
908 ret = check_thread_quit_pipe(pollfd, revents);
909 if (ret) {
910 err = 0;
911 goto exit;
912 }
913
914 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
915 ERR("socket poll error");
916 goto error;
917 } else if (revents & LPOLLIN) {
918 /*
919 * Get allocated in this thread, enqueued to a global queue,
920 * dequeued and freed in the worker thread.
921 */
922 int val = 1;
923 struct relay_connection *new_conn;
924 struct lttcomm_sock *newsock;
925
926 new_conn = connection_create();
927 if (!new_conn) {
928 goto error;
929 }
930
931 if (pollfd == data_sock->fd) {
932 new_conn->type = RELAY_DATA;
933 newsock = data_sock->ops->accept(data_sock);
934 DBG("Relay data connection accepted, socket %d",
935 newsock->fd);
936 } else {
937 assert(pollfd == control_sock->fd);
938 new_conn->type = RELAY_CONTROL;
939 newsock = control_sock->ops->accept(control_sock);
940 DBG("Relay control connection accepted, socket %d",
941 newsock->fd);
942 }
943 if (!newsock) {
944 PERROR("accepting sock");
945 connection_free(new_conn);
946 goto error;
947 }
948
949 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
950 sizeof(val));
951 if (ret < 0) {
952 PERROR("setsockopt inet");
953 lttcomm_destroy_sock(newsock);
954 connection_free(new_conn);
955 goto error;
956 }
957 new_conn->sock = newsock;
958
959 /* Enqueue request for the dispatcher thread. */
960 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
961 &new_conn->qnode);
962
963 /*
964 * Wake the dispatch queue futex. Implicit memory barrier with
965 * the exchange in cds_wfcq_enqueue.
966 */
967 futex_nto1_wake(&relay_conn_queue.futex);
968 }
969 }
970 }
971
972 exit:
973 error:
974 error_poll_add:
975 error_testpoint:
976 lttng_poll_clean(&events);
977 error_create_poll:
978 if (data_sock->fd >= 0) {
979 ret = data_sock->ops->close(data_sock);
980 if (ret) {
981 PERROR("close");
982 }
983 }
984 lttcomm_destroy_sock(data_sock);
985 error_sock_relay:
986 if (control_sock->fd >= 0) {
987 ret = control_sock->ops->close(control_sock);
988 if (ret) {
989 PERROR("close");
990 }
991 }
992 lttcomm_destroy_sock(control_sock);
993 error_sock_control:
994 if (err) {
995 health_error();
996 ERR("Health error occurred in %s", __func__);
997 }
998 health_unregister(health_relayd);
999 DBG("Relay listener thread cleanup complete");
1000 lttng_relay_stop_threads();
1001 return NULL;
1002 }
1003
1004 /*
1005 * This thread manages the dispatching of the requests to worker threads
1006 */
1007 static
1008 void *relay_thread_dispatcher(void *data)
1009 {
1010 int err = -1;
1011 ssize_t ret;
1012 struct cds_wfcq_node *node;
1013 struct relay_connection *new_conn = NULL;
1014
1015 DBG("[thread] Relay dispatcher started");
1016
1017 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
1018
1019 if (testpoint(relayd_thread_dispatcher)) {
1020 goto error_testpoint;
1021 }
1022
1023 health_code_update();
1024
1025 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1026 health_code_update();
1027
1028 /* Atomically prepare the queue futex */
1029 futex_nto1_prepare(&relay_conn_queue.futex);
1030
1031 do {
1032 health_code_update();
1033
1034 /* Dequeue commands */
1035 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1036 &relay_conn_queue.tail);
1037 if (node == NULL) {
1038 DBG("Woken up but nothing in the relay command queue");
1039 /* Continue thread execution */
1040 break;
1041 }
1042 new_conn = caa_container_of(node, struct relay_connection, qnode);
1043
1044 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1045
1046 /*
1047 * Inform worker thread of the new request. This call is blocking
1048 * so we can be assured that the data will be read at some point in
1049 * time or wait to the end of the world :)
1050 */
1051 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1052 if (ret < 0) {
1053 PERROR("write connection pipe");
1054 connection_destroy(new_conn);
1055 goto error;
1056 }
1057 } while (node != NULL);
1058
1059 /* Futex wait on queue. Blocking call on futex() */
1060 health_poll_entry();
1061 futex_nto1_wait(&relay_conn_queue.futex);
1062 health_poll_exit();
1063 }
1064
1065 /* Normal exit, no error */
1066 err = 0;
1067
1068 error:
1069 error_testpoint:
1070 if (err) {
1071 health_error();
1072 ERR("Health error occurred in %s", __func__);
1073 }
1074 health_unregister(health_relayd);
1075 DBG("Dispatch thread dying");
1076 lttng_relay_stop_threads();
1077 return NULL;
1078 }
1079
1080 static void try_close_streams(struct relay_session *session)
1081 {
1082 struct ctf_trace *ctf_trace;
1083 struct lttng_ht_iter iter;
1084
1085 assert(session);
1086
1087 pthread_mutex_lock(&session->viewer_ready_lock);
1088 rcu_read_lock();
1089 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
1090 node.node) {
1091 struct relay_stream *stream;
1092
1093 /* Close streams. */
1094 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
1095 stream_close(session, stream);
1096 }
1097
1098 ctf_trace->invalid_flag = 1;
1099 ctf_trace_try_destroy(session, ctf_trace);
1100 }
1101 rcu_read_unlock();
1102 pthread_mutex_unlock(&session->viewer_ready_lock);
1103 }
1104
1105 /*
1106 * Try to destroy a session within a connection.
1107 */
1108 static void destroy_session(struct relay_session *session,
1109 struct lttng_ht *sessions_ht)
1110 {
1111 assert(session);
1112 assert(sessions_ht);
1113
1114 /* Indicate that this session can be destroyed from now on. */
1115 session->close_flag = 1;
1116
1117 try_close_streams(session);
1118
1119 /*
1120 * This will try to delete and destroy the session if no viewer is attached
1121 * to it meaning the refcount is down to zero.
1122 */
1123 session_try_destroy(sessions_ht, session);
1124 }
1125
1126 /*
1127 * Copy index data from the control port to a given index object.
1128 */
1129 static void copy_index_control_data(struct relay_index *index,
1130 struct lttcomm_relayd_index *data)
1131 {
1132 assert(index);
1133 assert(data);
1134
1135 /*
1136 * The index on disk is encoded in big endian, so we don't need to convert
1137 * the data received on the network. The data_offset value is NEVER
1138 * modified here and is updated by the data thread.
1139 */
1140 index->index_data.packet_size = data->packet_size;
1141 index->index_data.content_size = data->content_size;
1142 index->index_data.timestamp_begin = data->timestamp_begin;
1143 index->index_data.timestamp_end = data->timestamp_end;
1144 index->index_data.events_discarded = data->events_discarded;
1145 index->index_data.stream_id = data->stream_id;
1146 }
1147
1148 /*
1149 * Handle the RELAYD_CREATE_SESSION command.
1150 *
1151 * On success, send back the session id or else return a negative value.
1152 */
1153 static
1154 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
1155 struct relay_connection *conn)
1156 {
1157 int ret = 0, send_ret;
1158 struct relay_session *session;
1159 struct lttcomm_relayd_status_session reply;
1160
1161 assert(recv_hdr);
1162 assert(conn);
1163
1164 memset(&reply, 0, sizeof(reply));
1165
1166 session = session_create();
1167 if (!session) {
1168 ret = -1;
1169 goto error;
1170 }
1171 session->minor = conn->minor;
1172 session->major = conn->major;
1173 conn->session_id = session->id;
1174 conn->session = session;
1175
1176 reply.session_id = htobe64(session->id);
1177
1178 switch (conn->minor) {
1179 case 1:
1180 case 2:
1181 case 3:
1182 break;
1183 case 4: /* LTTng sessiond 2.4 */
1184 default:
1185 ret = cmd_create_session_2_4(conn, session);
1186 }
1187
1188 lttng_ht_add_unique_u64(conn->sessions_ht, &session->session_n);
1189 DBG("Created session %" PRIu64, session->id);
1190
1191 error:
1192 if (ret < 0) {
1193 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1194 } else {
1195 reply.ret_code = htobe32(LTTNG_OK);
1196 }
1197
1198 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1199 if (send_ret < 0) {
1200 ERR("Relayd sending session id");
1201 ret = send_ret;
1202 }
1203
1204 return ret;
1205 }
1206
1207 /*
1208 * When we have received all the streams and the metadata for a channel,
1209 * we make them visible to the viewer threads.
1210 */
1211 static
1212 void set_viewer_ready_flag(struct relay_connection *conn)
1213 {
1214 struct relay_stream *stream, *tmp_stream;
1215
1216 pthread_mutex_lock(&conn->session->viewer_ready_lock);
1217 cds_list_for_each_entry_safe(stream, tmp_stream, &conn->recv_head,
1218 recv_list) {
1219 stream->viewer_ready = 1;
1220 cds_list_del(&stream->recv_list);
1221 }
1222 pthread_mutex_unlock(&conn->session->viewer_ready_lock);
1223 return;
1224 }
1225
1226 /*
1227 * Add a recv handle node to the connection recv list with the given stream
1228 * handle. A new node is allocated thus must be freed when the node is deleted
1229 * from the list.
1230 */
1231 static void queue_stream(struct relay_stream *stream,
1232 struct relay_connection *conn)
1233 {
1234 assert(conn);
1235 assert(stream);
1236
1237 cds_list_add(&stream->recv_list, &conn->recv_head);
1238 }
1239
1240 /*
1241 * relay_add_stream: allocate a new stream for a session
1242 */
1243 static
1244 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1245 struct relay_connection *conn)
1246 {
1247 int ret, send_ret;
1248 struct relay_session *session = conn->session;
1249 struct relay_stream *stream = NULL;
1250 struct lttcomm_relayd_status_stream reply;
1251 struct ctf_trace *trace;
1252
1253 if (!session || conn->version_check_done == 0) {
1254 ERR("Trying to add a stream before version check");
1255 ret = -1;
1256 goto end_no_session;
1257 }
1258
1259 stream = zmalloc(sizeof(struct relay_stream));
1260 if (stream == NULL) {
1261 PERROR("relay stream zmalloc");
1262 ret = -1;
1263 goto end_no_session;
1264 }
1265
1266 switch (conn->minor) {
1267 case 1: /* LTTng sessiond 2.1 */
1268 ret = cmd_recv_stream_2_1(conn, stream);
1269 break;
1270 case 2: /* LTTng sessiond 2.2 */
1271 default:
1272 ret = cmd_recv_stream_2_2(conn, stream);
1273 break;
1274 }
1275 if (ret < 0) {
1276 goto err_free_stream;
1277 }
1278
1279 rcu_read_lock();
1280 stream->stream_handle = ++last_relay_stream_id;
1281 stream->prev_seq = -1ULL;
1282 stream->session_id = session->id;
1283 stream->index_fd = -1;
1284 stream->read_index_fd = -1;
1285 stream->ctf_stream_id = -1ULL;
1286 lttng_ht_node_init_u64(&stream->node, stream->stream_handle);
1287 pthread_mutex_init(&stream->lock, NULL);
1288
1289 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
1290 if (ret < 0) {
1291 ERR("relay creating output directory");
1292 goto end;
1293 }
1294
1295 /*
1296 * No need to use run_as API here because whatever we receives, the relayd
1297 * uses its own credentials for the stream files.
1298 */
1299 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1300 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1301 if (ret < 0) {
1302 ERR("Create output file");
1303 goto end;
1304 }
1305 stream->fd = ret;
1306 if (stream->tracefile_size) {
1307 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1308 } else {
1309 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1310 }
1311
1312 trace = ctf_trace_find_by_path(session->ctf_traces_ht, stream->path_name);
1313 if (!trace) {
1314 trace = ctf_trace_create(stream->path_name);
1315 if (!trace) {
1316 ret = -1;
1317 goto end;
1318 }
1319 ctf_trace_add(session->ctf_traces_ht, trace);
1320 }
1321 ctf_trace_get_ref(trace);
1322
1323 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1324 stream->metadata_flag = 1;
1325 /* Assign quick reference to the metadata stream in the trace. */
1326 trace->metadata_stream = stream;
1327 }
1328
1329 /*
1330 * Add the stream in the recv list of the connection. Once the end stream
1331 * message is received, this list is emptied and streams are set with the
1332 * viewer ready flag.
1333 */
1334 queue_stream(stream, conn);
1335
1336 /*
1337 * Both in the ctf_trace object and the global stream ht since the data
1338 * side of the relayd does not have the concept of session.
1339 */
1340 lttng_ht_add_unique_u64(relay_streams_ht, &stream->node);
1341 cds_list_add_tail(&stream->trace_list, &trace->stream_list);
1342
1343 session->stream_count++;
1344
1345 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1346 stream->stream_handle);
1347
1348 end:
1349 memset(&reply, 0, sizeof(reply));
1350 reply.handle = htobe64(stream->stream_handle);
1351 /* send the session id to the client or a negative return code on error */
1352 if (ret < 0) {
1353 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1354 /* stream was not properly added to the ht, so free it */
1355 free(stream);
1356 } else {
1357 reply.ret_code = htobe32(LTTNG_OK);
1358 }
1359
1360 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1361 sizeof(struct lttcomm_relayd_status_stream), 0);
1362 if (send_ret < 0) {
1363 ERR("Relay sending stream id");
1364 ret = send_ret;
1365 }
1366 rcu_read_unlock();
1367
1368 end_no_session:
1369 return ret;
1370
1371 err_free_stream:
1372 free(stream->path_name);
1373 free(stream->channel_name);
1374 free(stream);
1375 return ret;
1376 }
1377
1378 /*
1379 * relay_close_stream: close a specific stream
1380 */
1381 static
1382 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1383 struct relay_connection *conn)
1384 {
1385 int ret, send_ret;
1386 struct relay_session *session = conn->session;
1387 struct lttcomm_relayd_close_stream stream_info;
1388 struct lttcomm_relayd_generic_reply reply;
1389 struct relay_stream *stream;
1390
1391 DBG("Close stream received");
1392
1393 if (!session || conn->version_check_done == 0) {
1394 ERR("Trying to close a stream before version check");
1395 ret = -1;
1396 goto end_no_session;
1397 }
1398
1399 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1400 sizeof(struct lttcomm_relayd_close_stream), 0);
1401 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1402 if (ret == 0) {
1403 /* Orderly shutdown. Not necessary to print an error. */
1404 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1405 } else {
1406 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1407 }
1408 ret = -1;
1409 goto end_no_session;
1410 }
1411
1412 rcu_read_lock();
1413 stream = stream_find_by_id(relay_streams_ht,
1414 be64toh(stream_info.stream_id));
1415 if (!stream) {
1416 ret = -1;
1417 goto end_unlock;
1418 }
1419
1420 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1421 stream->close_flag = 1;
1422 session->stream_count--;
1423
1424 /* Check if we can close it or else the data will do it. */
1425 try_close_stream(session, stream);
1426
1427 end_unlock:
1428 rcu_read_unlock();
1429
1430 memset(&reply, 0, sizeof(reply));
1431 if (ret < 0) {
1432 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1433 } else {
1434 reply.ret_code = htobe32(LTTNG_OK);
1435 }
1436 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1437 sizeof(struct lttcomm_relayd_generic_reply), 0);
1438 if (send_ret < 0) {
1439 ERR("Relay sending stream id");
1440 ret = send_ret;
1441 }
1442
1443 end_no_session:
1444 return ret;
1445 }
1446
1447 /*
1448 * relay_unknown_command: send -1 if received unknown command
1449 */
1450 static
1451 void relay_unknown_command(struct relay_connection *conn)
1452 {
1453 struct lttcomm_relayd_generic_reply reply;
1454 int ret;
1455
1456 memset(&reply, 0, sizeof(reply));
1457 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1458 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1459 sizeof(struct lttcomm_relayd_generic_reply), 0);
1460 if (ret < 0) {
1461 ERR("Relay sending unknown command");
1462 }
1463 }
1464
1465 /*
1466 * relay_start: send an acknowledgment to the client to tell if we are
1467 * ready to receive data. We are ready if a session is established.
1468 */
1469 static
1470 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1471 struct relay_connection *conn)
1472 {
1473 int ret = htobe32(LTTNG_OK);
1474 struct lttcomm_relayd_generic_reply reply;
1475 struct relay_session *session = conn->session;
1476
1477 if (!session) {
1478 DBG("Trying to start the streaming without a session established");
1479 ret = htobe32(LTTNG_ERR_UNK);
1480 }
1481
1482 memset(&reply, 0, sizeof(reply));
1483 reply.ret_code = ret;
1484 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1485 sizeof(struct lttcomm_relayd_generic_reply), 0);
1486 if (ret < 0) {
1487 ERR("Relay sending start ack");
1488 }
1489
1490 return ret;
1491 }
1492
1493 /*
1494 * Append padding to the file pointed by the file descriptor fd.
1495 */
1496 static int write_padding_to_file(int fd, uint32_t size)
1497 {
1498 ssize_t ret = 0;
1499 char *zeros;
1500
1501 if (size == 0) {
1502 goto end;
1503 }
1504
1505 zeros = zmalloc(size);
1506 if (zeros == NULL) {
1507 PERROR("zmalloc zeros for padding");
1508 ret = -1;
1509 goto end;
1510 }
1511
1512 ret = lttng_write(fd, zeros, size);
1513 if (ret < size) {
1514 PERROR("write padding to file");
1515 }
1516
1517 free(zeros);
1518
1519 end:
1520 return ret;
1521 }
1522
1523 /*
1524 * relay_recv_metadata: receive the metada for the session.
1525 */
1526 static
1527 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1528 struct relay_connection *conn)
1529 {
1530 int ret = htobe32(LTTNG_OK);
1531 ssize_t size_ret;
1532 struct relay_session *session = conn->session;
1533 struct lttcomm_relayd_metadata_payload *metadata_struct;
1534 struct relay_stream *metadata_stream;
1535 uint64_t data_size, payload_size;
1536 struct ctf_trace *ctf_trace;
1537
1538 if (!session) {
1539 ERR("Metadata sent before version check");
1540 ret = -1;
1541 goto end;
1542 }
1543
1544 data_size = payload_size = be64toh(recv_hdr->data_size);
1545 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1546 ERR("Incorrect data size");
1547 ret = -1;
1548 goto end;
1549 }
1550 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1551
1552 if (data_buffer_size < data_size) {
1553 /* In case the realloc fails, we can free the memory */
1554 char *tmp_data_ptr;
1555
1556 tmp_data_ptr = realloc(data_buffer, data_size);
1557 if (!tmp_data_ptr) {
1558 ERR("Allocating data buffer");
1559 free(data_buffer);
1560 ret = -1;
1561 goto end;
1562 }
1563 data_buffer = tmp_data_ptr;
1564 data_buffer_size = data_size;
1565 }
1566 memset(data_buffer, 0, data_size);
1567 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1568 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
1569 if (ret < 0 || ret != data_size) {
1570 if (ret == 0) {
1571 /* Orderly shutdown. Not necessary to print an error. */
1572 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1573 } else {
1574 ERR("Relay didn't receive the whole metadata");
1575 }
1576 ret = -1;
1577 goto end;
1578 }
1579 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1580
1581 rcu_read_lock();
1582 metadata_stream = stream_find_by_id(relay_streams_ht,
1583 be64toh(metadata_struct->stream_id));
1584 if (!metadata_stream) {
1585 ret = -1;
1586 goto end_unlock;
1587 }
1588
1589 size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
1590 payload_size);
1591 if (size_ret < payload_size) {
1592 ERR("Relay error writing metadata on file");
1593 ret = -1;
1594 goto end_unlock;
1595 }
1596
1597 ret = write_padding_to_file(metadata_stream->fd,
1598 be32toh(metadata_struct->padding_size));
1599 if (ret < 0) {
1600 goto end_unlock;
1601 }
1602
1603 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1604 metadata_stream->path_name);
1605 assert(ctf_trace);
1606 ctf_trace->metadata_received +=
1607 payload_size + be32toh(metadata_struct->padding_size);
1608
1609 DBG2("Relay metadata written");
1610
1611 end_unlock:
1612 rcu_read_unlock();
1613 end:
1614 return ret;
1615 }
1616
1617 /*
1618 * relay_send_version: send relayd version number
1619 */
1620 static
1621 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
1627 assert(conn);
1628
1629 conn->version_check_done = 1;
1630
1631 /* Get version from the other side. */
1632 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1633 if (ret < 0 || ret != sizeof(msg)) {
1634 if (ret == 0) {
1635 /* Orderly shutdown. Not necessary to print an error. */
1636 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1637 } else {
1638 ERR("Relay failed to receive the version values.");
1639 }
1640 ret = -1;
1641 goto end;
1642 }
1643
1644 memset(&reply, 0, sizeof(reply));
1645 reply.major = RELAYD_VERSION_COMM_MAJOR;
1646 reply.minor = RELAYD_VERSION_COMM_MINOR;
1647
1648 /* Major versions must be the same */
1649 if (reply.major != be32toh(msg.major)) {
1650 DBG("Incompatible major versions (%u vs %u), deleting session",
1651 reply.major, be32toh(msg.major));
1652 destroy_session(conn->session, conn->sessions_ht);
1653 ret = 0;
1654 goto end;
1655 }
1656
1657 conn->major = reply.major;
1658 /* We adapt to the lowest compatible version */
1659 if (reply.minor <= be32toh(msg.minor)) {
1660 conn->minor = reply.minor;
1661 } else {
1662 conn->minor = be32toh(msg.minor);
1663 }
1664
1665 reply.major = htobe32(reply.major);
1666 reply.minor = htobe32(reply.minor);
1667 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1668 sizeof(struct lttcomm_relayd_version), 0);
1669 if (ret < 0) {
1670 ERR("Relay sending version");
1671 }
1672
1673 DBG("Version check done using protocol %u.%u", conn->major,
1674 conn->minor);
1675
1676 end:
1677 return ret;
1678 }
1679
1680 /*
1681 * Check for data pending for a given stream id from the session daemon.
1682 */
1683 static
1684 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1685 struct relay_connection *conn)
1686 {
1687 struct relay_session *session = conn->session;
1688 struct lttcomm_relayd_data_pending msg;
1689 struct lttcomm_relayd_generic_reply reply;
1690 struct relay_stream *stream;
1691 int ret;
1692 uint64_t last_net_seq_num, stream_id;
1693
1694 DBG("Data pending command received");
1695
1696 if (!session || conn->version_check_done == 0) {
1697 ERR("Trying to check for data before version check");
1698 ret = -1;
1699 goto end_no_session;
1700 }
1701
1702 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1703 if (ret < sizeof(msg)) {
1704 if (ret == 0) {
1705 /* Orderly shutdown. Not necessary to print an error. */
1706 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1707 } else {
1708 ERR("Relay didn't receive valid data_pending struct size : %d",
1709 ret);
1710 }
1711 ret = -1;
1712 goto end_no_session;
1713 }
1714
1715 stream_id = be64toh(msg.stream_id);
1716 last_net_seq_num = be64toh(msg.last_net_seq_num);
1717
1718 rcu_read_lock();
1719 stream = stream_find_by_id(relay_streams_ht, stream_id);
1720 if (stream == NULL) {
1721 ret = -1;
1722 goto end_unlock;
1723 }
1724
1725 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1726 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1727 last_net_seq_num);
1728
1729 /* Avoid wrapping issue */
1730 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1731 /* Data has in fact been written and is NOT pending */
1732 ret = 0;
1733 } else {
1734 /* Data still being streamed thus pending */
1735 ret = 1;
1736 }
1737
1738 /* Pending check is now done. */
1739 stream->data_pending_check_done = 1;
1740
1741 end_unlock:
1742 rcu_read_unlock();
1743
1744 memset(&reply, 0, sizeof(reply));
1745 reply.ret_code = htobe32(ret);
1746 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1747 if (ret < 0) {
1748 ERR("Relay data pending ret code failed");
1749 }
1750
1751 end_no_session:
1752 return ret;
1753 }
1754
1755 /*
1756 * Wait for the control socket to reach a quiescent state.
1757 *
1758 * Note that for now, when receiving this command from the session daemon, this
1759 * means that every subsequent commands or data received on the control socket
1760 * has been handled. So, this is why we simply return OK here.
1761 */
1762 static
1763 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1764 struct relay_connection *conn)
1765 {
1766 int ret;
1767 uint64_t stream_id;
1768 struct relay_stream *stream;
1769 struct lttng_ht_iter iter;
1770 struct lttcomm_relayd_quiescent_control msg;
1771 struct lttcomm_relayd_generic_reply reply;
1772
1773 DBG("Checking quiescent state on control socket");
1774
1775 if (!conn->session || conn->version_check_done == 0) {
1776 ERR("Trying to check for data before version check");
1777 ret = -1;
1778 goto end_no_session;
1779 }
1780
1781 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1782 if (ret < sizeof(msg)) {
1783 if (ret == 0) {
1784 /* Orderly shutdown. Not necessary to print an error. */
1785 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1786 } else {
1787 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1788 ret);
1789 }
1790 ret = -1;
1791 goto end_no_session;
1792 }
1793
1794 stream_id = be64toh(msg.stream_id);
1795
1796 rcu_read_lock();
1797 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1798 node.node) {
1799 if (stream->stream_handle == stream_id) {
1800 stream->data_pending_check_done = 1;
1801 DBG("Relay quiescent control pending flag set to %" PRIu64,
1802 stream_id);
1803 break;
1804 }
1805 }
1806 rcu_read_unlock();
1807
1808 memset(&reply, 0, sizeof(reply));
1809 reply.ret_code = htobe32(LTTNG_OK);
1810 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1811 if (ret < 0) {
1812 ERR("Relay data quiescent control ret code failed");
1813 }
1814
1815 end_no_session:
1816 return ret;
1817 }
1818
1819 /*
1820 * Initialize a data pending command. This means that a client is about to ask
1821 * for data pending for each stream he/she holds. Simply iterate over all
1822 * streams of a session and set the data_pending_check_done flag.
1823 *
1824 * This command returns to the client a LTTNG_OK code.
1825 */
1826 static
1827 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1828 struct relay_connection *conn)
1829 {
1830 int ret;
1831 struct lttng_ht_iter iter;
1832 struct lttcomm_relayd_begin_data_pending msg;
1833 struct lttcomm_relayd_generic_reply reply;
1834 struct relay_stream *stream;
1835 uint64_t session_id;
1836
1837 assert(recv_hdr);
1838 assert(conn);
1839
1840 DBG("Init streams for data pending");
1841
1842 if (!conn->session || conn->version_check_done == 0) {
1843 ERR("Trying to check for data before version check");
1844 ret = -1;
1845 goto end_no_session;
1846 }
1847
1848 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1849 if (ret < sizeof(msg)) {
1850 if (ret == 0) {
1851 /* Orderly shutdown. Not necessary to print an error. */
1852 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1853 } else {
1854 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1855 ret);
1856 }
1857 ret = -1;
1858 goto end_no_session;
1859 }
1860
1861 session_id = be64toh(msg.session_id);
1862
1863 /*
1864 * Iterate over all streams to set the begin data pending flag. For now, the
1865 * streams are indexed by stream handle so we have to iterate over all
1866 * streams to find the one associated with the right session_id.
1867 */
1868 rcu_read_lock();
1869 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1870 node.node) {
1871 if (stream->session_id == session_id) {
1872 stream->data_pending_check_done = 0;
1873 DBG("Set begin data pending flag to stream %" PRIu64,
1874 stream->stream_handle);
1875 }
1876 }
1877 rcu_read_unlock();
1878
1879 memset(&reply, 0, sizeof(reply));
1880 /* All good, send back reply. */
1881 reply.ret_code = htobe32(LTTNG_OK);
1882
1883 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1884 if (ret < 0) {
1885 ERR("Relay begin data pending send reply failed");
1886 }
1887
1888 end_no_session:
1889 return ret;
1890 }
1891
1892 /*
1893 * End data pending command. This will check, for a given session id, if each
1894 * stream associated with it has its data_pending_check_done flag set. If not,
1895 * this means that the client lost track of the stream but the data is still
1896 * being streamed on our side. In this case, we inform the client that data is
1897 * inflight.
1898 *
1899 * Return to the client if there is data in flight or not with a ret_code.
1900 */
1901 static
1902 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1903 struct relay_connection *conn)
1904 {
1905 int ret;
1906 struct lttng_ht_iter iter;
1907 struct lttcomm_relayd_end_data_pending msg;
1908 struct lttcomm_relayd_generic_reply reply;
1909 struct relay_stream *stream;
1910 uint64_t session_id;
1911 uint32_t is_data_inflight = 0;
1912
1913 assert(recv_hdr);
1914 assert(conn);
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 /* Iterate over all streams to see if the begin data pending flag is set. */
1940 rcu_read_lock();
1941 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1942 node.node) {
1943 if (stream->session_id == session_id &&
1944 !stream->data_pending_check_done && !stream->terminated_flag) {
1945 is_data_inflight = 1;
1946 DBG("Data is still in flight for stream %" PRIu64,
1947 stream->stream_handle);
1948 break;
1949 }
1950 }
1951 rcu_read_unlock();
1952
1953 memset(&reply, 0, sizeof(reply));
1954 /* All good, send back reply. */
1955 reply.ret_code = htobe32(is_data_inflight);
1956
1957 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1958 if (ret < 0) {
1959 ERR("Relay end data pending send reply failed");
1960 }
1961
1962 end_no_session:
1963 return ret;
1964 }
1965
1966 /*
1967 * Receive an index for a specific stream.
1968 *
1969 * Return 0 on success else a negative value.
1970 */
1971 static
1972 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1973 struct relay_connection *conn)
1974 {
1975 int ret, send_ret, index_created = 0;
1976 struct relay_session *session = conn->session;
1977 struct lttcomm_relayd_index index_info;
1978 struct relay_index *index, *wr_index = NULL;
1979 struct lttcomm_relayd_generic_reply reply;
1980 struct relay_stream *stream;
1981 uint64_t net_seq_num;
1982
1983 assert(conn);
1984
1985 DBG("Relay receiving index");
1986
1987 if (!session || conn->version_check_done == 0) {
1988 ERR("Trying to close a stream before version check");
1989 ret = -1;
1990 goto end_no_session;
1991 }
1992
1993 ret = conn->sock->ops->recvmsg(conn->sock, &index_info,
1994 sizeof(index_info), 0);
1995 if (ret < sizeof(index_info)) {
1996 if (ret == 0) {
1997 /* Orderly shutdown. Not necessary to print an error. */
1998 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1999 } else {
2000 ERR("Relay didn't receive valid index struct size : %d", ret);
2001 }
2002 ret = -1;
2003 goto end_no_session;
2004 }
2005
2006 net_seq_num = be64toh(index_info.net_seq_num);
2007
2008 rcu_read_lock();
2009 stream = stream_find_by_id(relay_streams_ht,
2010 be64toh(index_info.relay_stream_id));
2011 if (!stream) {
2012 ret = -1;
2013 goto end_rcu_unlock;
2014 }
2015
2016 /* Live beacon handling */
2017 if (index_info.packet_size == 0) {
2018 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
2019
2020 /*
2021 * Only flag a stream inactive when it has already received data
2022 * and no indexes are in flight.
2023 */
2024 if (stream->total_index_received > 0 && stream->indexes_in_flight == 0) {
2025 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
2026 }
2027 ret = 0;
2028 goto end_rcu_unlock;
2029 } else {
2030 stream->beacon_ts_end = -1ULL;
2031 }
2032
2033 index = relay_index_find(stream->stream_handle, net_seq_num);
2034 if (!index) {
2035 /* A successful creation will add the object to the HT. */
2036 index = relay_index_create(stream->stream_handle, net_seq_num);
2037 if (!index) {
2038 goto end_rcu_unlock;
2039 }
2040 index_created = 1;
2041 stream->indexes_in_flight++;
2042 }
2043
2044 copy_index_control_data(index, &index_info);
2045 if (stream->ctf_stream_id == -1ULL) {
2046 stream->ctf_stream_id = be64toh(index_info.stream_id);
2047 }
2048
2049 if (index_created) {
2050 /*
2051 * Try to add the relay index object to the hash table. If an object
2052 * already exist, destroy back the index created, set the data in this
2053 * object and write it on disk.
2054 */
2055 relay_index_add(index, &wr_index);
2056 if (wr_index) {
2057 copy_index_control_data(wr_index, &index_info);
2058 free(index);
2059 }
2060 } else {
2061 /* The index already exists so write it on disk. */
2062 wr_index = index;
2063 }
2064
2065 /* Do we have a writable ready index to write on disk. */
2066 if (wr_index) {
2067 ret = relay_index_write(wr_index->fd, wr_index);
2068 if (ret < 0) {
2069 goto end_rcu_unlock;
2070 }
2071 stream->total_index_received++;
2072 stream->indexes_in_flight--;
2073 assert(stream->indexes_in_flight >= 0);
2074 }
2075
2076 end_rcu_unlock:
2077 rcu_read_unlock();
2078
2079 memset(&reply, 0, sizeof(reply));
2080 if (ret < 0) {
2081 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2082 } else {
2083 reply.ret_code = htobe32(LTTNG_OK);
2084 }
2085 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2086 if (send_ret < 0) {
2087 ERR("Relay sending close index id reply");
2088 ret = send_ret;
2089 }
2090
2091 end_no_session:
2092 return ret;
2093 }
2094
2095 /*
2096 * Receive the streams_sent message.
2097 *
2098 * Return 0 on success else a negative value.
2099 */
2100 static
2101 int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
2102 struct relay_connection *conn)
2103 {
2104 int ret, send_ret;
2105 struct lttcomm_relayd_generic_reply reply;
2106
2107 assert(conn);
2108
2109 DBG("Relay receiving streams_sent");
2110
2111 if (!conn->session || conn->version_check_done == 0) {
2112 ERR("Trying to close a stream before version check");
2113 ret = -1;
2114 goto end_no_session;
2115 }
2116
2117 /*
2118 * Flag every pending stream in the connection recv list that they are
2119 * ready to be used by the viewer.
2120 */
2121 set_viewer_ready_flag(conn);
2122
2123 /*
2124 * Inform the viewer that there are new streams in the session.
2125 */
2126 if (conn->session->viewer_refcount) {
2127 uatomic_set(&conn->session->new_streams, 1);
2128 }
2129
2130 memset(&reply, 0, sizeof(reply));
2131 reply.ret_code = htobe32(LTTNG_OK);
2132 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2133 if (send_ret < 0) {
2134 ERR("Relay sending sent_stream reply");
2135 ret = send_ret;
2136 } else {
2137 /* Success. */
2138 ret = 0;
2139 }
2140
2141 end_no_session:
2142 return ret;
2143 }
2144
2145 /*
2146 * Process the commands received on the control socket
2147 */
2148 static
2149 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
2150 struct relay_connection *conn)
2151 {
2152 int ret = 0;
2153
2154 switch (be32toh(recv_hdr->cmd)) {
2155 case RELAYD_CREATE_SESSION:
2156 ret = relay_create_session(recv_hdr, conn);
2157 break;
2158 case RELAYD_ADD_STREAM:
2159 ret = relay_add_stream(recv_hdr, conn);
2160 break;
2161 case RELAYD_START_DATA:
2162 ret = relay_start(recv_hdr, conn);
2163 break;
2164 case RELAYD_SEND_METADATA:
2165 ret = relay_recv_metadata(recv_hdr, conn);
2166 break;
2167 case RELAYD_VERSION:
2168 ret = relay_send_version(recv_hdr, conn);
2169 break;
2170 case RELAYD_CLOSE_STREAM:
2171 ret = relay_close_stream(recv_hdr, conn);
2172 break;
2173 case RELAYD_DATA_PENDING:
2174 ret = relay_data_pending(recv_hdr, conn);
2175 break;
2176 case RELAYD_QUIESCENT_CONTROL:
2177 ret = relay_quiescent_control(recv_hdr, conn);
2178 break;
2179 case RELAYD_BEGIN_DATA_PENDING:
2180 ret = relay_begin_data_pending(recv_hdr, conn);
2181 break;
2182 case RELAYD_END_DATA_PENDING:
2183 ret = relay_end_data_pending(recv_hdr, conn);
2184 break;
2185 case RELAYD_SEND_INDEX:
2186 ret = relay_recv_index(recv_hdr, conn);
2187 break;
2188 case RELAYD_STREAMS_SENT:
2189 ret = relay_streams_sent(recv_hdr, conn);
2190 break;
2191 case RELAYD_UPDATE_SYNC_INFO:
2192 default:
2193 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2194 relay_unknown_command(conn);
2195 ret = -1;
2196 goto end;
2197 }
2198
2199 end:
2200 return ret;
2201 }
2202
2203 /*
2204 * Handle index for a data stream.
2205 *
2206 * RCU read side lock MUST be acquired.
2207 *
2208 * Return 0 on success else a negative value.
2209 */
2210 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2211 int rotate_index)
2212 {
2213 int ret = 0, index_created = 0;
2214 uint64_t stream_id, data_offset;
2215 struct relay_index *index, *wr_index = NULL;
2216
2217 assert(stream);
2218
2219 stream_id = stream->stream_handle;
2220 /* Get data offset because we are about to update the index. */
2221 data_offset = htobe64(stream->tracefile_size_current);
2222
2223 /*
2224 * Lookup for an existing index for that stream id/sequence number. If on
2225 * exists, the control thread already received the data for it thus we need
2226 * to write it on disk.
2227 */
2228 index = relay_index_find(stream_id, net_seq_num);
2229 if (!index) {
2230 /* A successful creation will add the object to the HT. */
2231 index = relay_index_create(stream_id, net_seq_num);
2232 if (!index) {
2233 ret = -1;
2234 goto error;
2235 }
2236 index_created = 1;
2237 stream->indexes_in_flight++;
2238 }
2239
2240 if (rotate_index || stream->index_fd < 0) {
2241 index->to_close_fd = stream->index_fd;
2242 ret = index_create_file(stream->path_name, stream->channel_name,
2243 relayd_uid, relayd_gid, stream->tracefile_size,
2244 stream->tracefile_count_current);
2245 if (ret < 0) {
2246 /* This will close the stream's index fd if one. */
2247 relay_index_free_safe(index);
2248 goto error;
2249 }
2250 stream->index_fd = ret;
2251 }
2252 index->fd = stream->index_fd;
2253 index->index_data.offset = data_offset;
2254
2255 if (index_created) {
2256 /*
2257 * Try to add the relay index object to the hash table. If an object
2258 * already exist, destroy back the index created and set the data.
2259 */
2260 relay_index_add(index, &wr_index);
2261 if (wr_index) {
2262 /* Copy back data from the created index. */
2263 wr_index->fd = index->fd;
2264 wr_index->to_close_fd = index->to_close_fd;
2265 wr_index->index_data.offset = data_offset;
2266 free(index);
2267 }
2268 } else {
2269 /* The index already exists so write it on disk. */
2270 wr_index = index;
2271 }
2272
2273 /* Do we have a writable ready index to write on disk. */
2274 if (wr_index) {
2275 ret = relay_index_write(wr_index->fd, wr_index);
2276 if (ret < 0) {
2277 goto error;
2278 }
2279 stream->total_index_received++;
2280 stream->indexes_in_flight--;
2281 assert(stream->indexes_in_flight >= 0);
2282 }
2283
2284 error:
2285 return ret;
2286 }
2287
2288 /*
2289 * relay_process_data: Process the data received on the data socket
2290 */
2291 static
2292 int relay_process_data(struct relay_connection *conn)
2293 {
2294 int ret = 0, rotate_index = 0;
2295 ssize_t size_ret;
2296 struct relay_stream *stream;
2297 struct lttcomm_relayd_data_hdr data_hdr;
2298 uint64_t stream_id;
2299 uint64_t net_seq_num;
2300 uint32_t data_size;
2301 struct relay_session *session;
2302
2303 assert(conn);
2304
2305 ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr,
2306 sizeof(struct lttcomm_relayd_data_hdr), 0);
2307 if (ret <= 0) {
2308 if (ret == 0) {
2309 /* Orderly shutdown. Not necessary to print an error. */
2310 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2311 } else {
2312 ERR("Unable to receive data header on sock %d", conn->sock->fd);
2313 }
2314 ret = -1;
2315 goto end;
2316 }
2317
2318 stream_id = be64toh(data_hdr.stream_id);
2319
2320 rcu_read_lock();
2321 stream = stream_find_by_id(relay_streams_ht, stream_id);
2322 if (!stream) {
2323 ret = -1;
2324 goto end_rcu_unlock;
2325 }
2326
2327 session = session_find_by_id(conn->sessions_ht, stream->session_id);
2328 assert(session);
2329
2330 data_size = be32toh(data_hdr.data_size);
2331 if (data_buffer_size < data_size) {
2332 char *tmp_data_ptr;
2333
2334 tmp_data_ptr = realloc(data_buffer, data_size);
2335 if (!tmp_data_ptr) {
2336 ERR("Allocating data buffer");
2337 free(data_buffer);
2338 ret = -1;
2339 goto end_rcu_unlock;
2340 }
2341 data_buffer = tmp_data_ptr;
2342 data_buffer_size = data_size;
2343 }
2344 memset(data_buffer, 0, data_size);
2345
2346 net_seq_num = be64toh(data_hdr.net_seq_num);
2347
2348 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2349 data_size, stream_id, net_seq_num);
2350 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
2351 if (ret <= 0) {
2352 if (ret == 0) {
2353 /* Orderly shutdown. Not necessary to print an error. */
2354 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2355 }
2356 ret = -1;
2357 goto end_rcu_unlock;
2358 }
2359
2360 /* Check if a rotation is needed. */
2361 if (stream->tracefile_size > 0 &&
2362 (stream->tracefile_size_current + data_size) >
2363 stream->tracefile_size) {
2364 struct relay_viewer_stream *vstream;
2365 uint64_t new_id;
2366
2367 new_id = (stream->tracefile_count_current + 1) %
2368 stream->tracefile_count;
2369 /*
2370 * When we wrap-around back to 0, we start overwriting old
2371 * trace data.
2372 */
2373 if (!stream->tracefile_overwrite && new_id == 0) {
2374 stream->tracefile_overwrite = 1;
2375 }
2376 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
2377 if (stream->tracefile_overwrite) {
2378 stream->oldest_tracefile_id =
2379 (stream->oldest_tracefile_id + 1) %
2380 stream->tracefile_count;
2381 }
2382 vstream = viewer_stream_find_by_id(stream->stream_handle);
2383 if (vstream) {
2384 /*
2385 * The viewer is reading a file about to be
2386 * overwritten. Close the FDs it is
2387 * currently using and let it handle the fault.
2388 */
2389 if (vstream->tracefile_count_current == new_id) {
2390 pthread_mutex_lock(&vstream->overwrite_lock);
2391 vstream->abort_flag = 1;
2392 pthread_mutex_unlock(&vstream->overwrite_lock);
2393 DBG("Streaming side setting abort_flag on stream %s_%" PRIu64 "\n",
2394 stream->channel_name, new_id);
2395 } else if (vstream->tracefile_count_current ==
2396 stream->tracefile_count_current) {
2397 /*
2398 * The reader and writer were in the
2399 * same trace file, inform the viewer
2400 * that no new index will ever be added
2401 * to this file.
2402 */
2403 vstream->close_write_flag = 1;
2404 }
2405 }
2406 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2407 stream->tracefile_size, stream->tracefile_count,
2408 relayd_uid, relayd_gid, stream->fd,
2409 &(stream->tracefile_count_current), &stream->fd);
2410 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
2411 if (ret < 0) {
2412 ERR("Rotating stream output file");
2413 goto end_rcu_unlock;
2414 }
2415 /* Reset current size because we just perform a stream rotation. */
2416 stream->tracefile_size_current = 0;
2417 rotate_index = 1;
2418 }
2419
2420 /*
2421 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2422 * index are NOT supported.
2423 */
2424 if (session->minor >= 4 && !session->snapshot) {
2425 ret = handle_index_data(stream, net_seq_num, rotate_index);
2426 if (ret < 0) {
2427 goto end_rcu_unlock;
2428 }
2429 }
2430
2431 /* Write data to stream output fd. */
2432 size_ret = lttng_write(stream->fd, data_buffer, data_size);
2433 if (size_ret < data_size) {
2434 ERR("Relay error writing data to file");
2435 ret = -1;
2436 goto end_rcu_unlock;
2437 }
2438
2439 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2440 ret, stream->stream_handle);
2441
2442 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2443 if (ret < 0) {
2444 goto end_rcu_unlock;
2445 }
2446 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2447
2448 stream->prev_seq = net_seq_num;
2449
2450 try_close_stream(session, stream);
2451
2452 end_rcu_unlock:
2453 rcu_read_unlock();
2454 end:
2455 return ret;
2456 }
2457
2458 static
2459 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2460 {
2461 int ret;
2462
2463 assert(events);
2464
2465 (void) lttng_poll_del(events, pollfd);
2466
2467 ret = close(pollfd);
2468 if (ret < 0) {
2469 ERR("Closing pollfd %d", pollfd);
2470 }
2471 }
2472
2473 static void destroy_connection(struct lttng_ht *relay_connections_ht,
2474 struct relay_connection *conn)
2475 {
2476 assert(relay_connections_ht);
2477 assert(conn);
2478
2479 connection_delete(relay_connections_ht, conn);
2480
2481 /* For the control socket, we try to destroy the session. */
2482 if (conn->type == RELAY_CONTROL && conn->session) {
2483 destroy_session(conn->session, conn->sessions_ht);
2484 }
2485
2486 connection_destroy(conn);
2487 }
2488
2489 /*
2490 * This thread does the actual work
2491 */
2492 static
2493 void *relay_thread_worker(void *data)
2494 {
2495 int ret, err = -1, last_seen_data_fd = -1;
2496 uint32_t nb_fd;
2497 struct lttng_poll_event events;
2498 struct lttng_ht *relay_connections_ht;
2499 struct lttng_ht_iter iter;
2500 struct lttcomm_relayd_hdr recv_hdr;
2501 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2502 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2503 struct relay_index *index;
2504 struct relay_connection *destroy_conn = NULL;
2505
2506 DBG("[thread] Relay worker started");
2507
2508 rcu_register_thread();
2509
2510 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2511
2512 if (testpoint(relayd_thread_worker)) {
2513 goto error_testpoint;
2514 }
2515
2516 health_code_update();
2517
2518 /* table of connections indexed on socket */
2519 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2520 if (!relay_connections_ht) {
2521 goto relay_connections_ht_error;
2522 }
2523
2524 /* Tables of received indexes indexed by index handle and net_seq_num. */
2525 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2526 if (!indexes_ht) {
2527 goto indexes_ht_error;
2528 }
2529
2530 ret = create_thread_poll_set(&events, 2);
2531 if (ret < 0) {
2532 goto error_poll_create;
2533 }
2534
2535 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2536 if (ret < 0) {
2537 goto error;
2538 }
2539
2540 restart:
2541 while (1) {
2542 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2543
2544 health_code_update();
2545
2546 /* Infinite blocking call, waiting for transmission */
2547 DBG3("Relayd worker thread polling...");
2548 health_poll_entry();
2549 ret = lttng_poll_wait(&events, -1);
2550 health_poll_exit();
2551 if (ret < 0) {
2552 /*
2553 * Restart interrupted system call.
2554 */
2555 if (errno == EINTR) {
2556 goto restart;
2557 }
2558 goto error;
2559 }
2560
2561 nb_fd = ret;
2562
2563 /*
2564 * Process control. The control connection is prioritised so we don't
2565 * starve it with high throughout put tracing data on the data
2566 * connection.
2567 */
2568 for (i = 0; i < nb_fd; i++) {
2569 /* Fetch once the poll data */
2570 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2571 int pollfd = LTTNG_POLL_GETFD(&events, i);
2572
2573 health_code_update();
2574
2575 if (!revents) {
2576 /* No activity for this FD (poll implementation). */
2577 continue;
2578 }
2579
2580 /* Thread quit pipe has been closed. Killing thread. */
2581 ret = check_thread_quit_pipe(pollfd, revents);
2582 if (ret) {
2583 err = 0;
2584 goto exit;
2585 }
2586
2587 /* Inspect the relay conn pipe for new connection */
2588 if (pollfd == relay_conn_pipe[0]) {
2589 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2590 ERR("Relay connection pipe error");
2591 goto error;
2592 } else if (revents & LPOLLIN) {
2593 struct relay_connection *conn;
2594
2595 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
2596 if (ret < 0) {
2597 goto error;
2598 }
2599 conn->sessions_ht = sessions_ht;
2600 connection_init(conn);
2601 lttng_poll_add(&events, conn->sock->fd,
2602 LPOLLIN | LPOLLRDHUP);
2603 rcu_read_lock();
2604 lttng_ht_add_unique_ulong(relay_connections_ht,
2605 &conn->sock_n);
2606 rcu_read_unlock();
2607 DBG("Connection socket %d added", conn->sock->fd);
2608 }
2609 } else {
2610 struct relay_connection *ctrl_conn;
2611
2612 rcu_read_lock();
2613 ctrl_conn = connection_find_by_sock(relay_connections_ht, pollfd);
2614 /* If not found, there is a synchronization issue. */
2615 assert(ctrl_conn);
2616
2617 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2618 cleanup_connection_pollfd(&events, pollfd);
2619 destroy_connection(relay_connections_ht, ctrl_conn);
2620 if (last_seen_data_fd == pollfd) {
2621 last_seen_data_fd = last_notdel_data_fd;
2622 }
2623 } else if (revents & LPOLLIN) {
2624 if (ctrl_conn->type == RELAY_CONTROL) {
2625 ret = ctrl_conn->sock->ops->recvmsg(ctrl_conn->sock, &recv_hdr,
2626 sizeof(recv_hdr), 0);
2627 if (ret <= 0) {
2628 /* Connection closed */
2629 cleanup_connection_pollfd(&events, pollfd);
2630 destroy_connection(relay_connections_ht, ctrl_conn);
2631 DBG("Control connection closed with %d", pollfd);
2632 } else {
2633 ret = relay_process_control(&recv_hdr, ctrl_conn);
2634 if (ret < 0) {
2635 /* Clear the session on error. */
2636 cleanup_connection_pollfd(&events, pollfd);
2637 destroy_connection(relay_connections_ht, ctrl_conn);
2638 DBG("Connection closed with %d", pollfd);
2639 }
2640 seen_control = 1;
2641 }
2642 } else {
2643 /*
2644 * Flag the last seen data fd not deleted. It will be
2645 * used as the last seen fd if any fd gets deleted in
2646 * this first loop.
2647 */
2648 last_notdel_data_fd = pollfd;
2649 }
2650 } else {
2651 ERR("Unknown poll events %u for sock %d", revents, pollfd);
2652 }
2653 rcu_read_unlock();
2654 }
2655 }
2656
2657 /*
2658 * The last loop handled a control request, go back to poll to make
2659 * sure we prioritise the control socket.
2660 */
2661 if (seen_control) {
2662 continue;
2663 }
2664
2665 if (last_seen_data_fd >= 0) {
2666 for (i = 0; i < nb_fd; i++) {
2667 int pollfd = LTTNG_POLL_GETFD(&events, i);
2668
2669 health_code_update();
2670
2671 if (last_seen_data_fd == pollfd) {
2672 idx = i;
2673 break;
2674 }
2675 }
2676 }
2677
2678 /* Process data connection. */
2679 for (i = idx + 1; i < nb_fd; i++) {
2680 /* Fetch the poll data. */
2681 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2682 int pollfd = LTTNG_POLL_GETFD(&events, i);
2683 struct relay_connection *data_conn;
2684
2685 health_code_update();
2686
2687 if (!revents) {
2688 /* No activity for this FD (poll implementation). */
2689 continue;
2690 }
2691
2692 /* Skip the command pipe. It's handled in the first loop. */
2693 if (pollfd == relay_conn_pipe[0]) {
2694 continue;
2695 }
2696
2697 rcu_read_lock();
2698 data_conn = connection_find_by_sock(relay_connections_ht, pollfd);
2699 if (!data_conn) {
2700 /* Skip it. Might be removed before. */
2701 rcu_read_unlock();
2702 continue;
2703 }
2704
2705 if (revents & LPOLLIN) {
2706 if (data_conn->type != RELAY_DATA) {
2707 rcu_read_unlock();
2708 continue;
2709 }
2710
2711 ret = relay_process_data(data_conn);
2712 /* Connection closed */
2713 if (ret < 0) {
2714 cleanup_connection_pollfd(&events, pollfd);
2715 destroy_connection(relay_connections_ht, data_conn);
2716 DBG("Data connection closed with %d", pollfd);
2717 /*
2718 * Every goto restart call sets the last seen fd where
2719 * here we don't really care since we gracefully
2720 * continue the loop after the connection is deleted.
2721 */
2722 } else {
2723 /* Keep last seen port. */
2724 last_seen_data_fd = pollfd;
2725 rcu_read_unlock();
2726 goto restart;
2727 }
2728 }
2729 rcu_read_unlock();
2730 }
2731 last_seen_data_fd = -1;
2732 }
2733
2734 /* Normal exit, no error */
2735 ret = 0;
2736
2737 exit:
2738 error:
2739 lttng_poll_clean(&events);
2740
2741 /* Cleanup reamaining connection object. */
2742 rcu_read_lock();
2743 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
2744 destroy_conn,
2745 sock_n.node) {
2746 health_code_update();
2747 destroy_connection(relay_connections_ht, destroy_conn);
2748 }
2749 rcu_read_unlock();
2750 error_poll_create:
2751 rcu_read_lock();
2752 cds_lfht_for_each_entry(indexes_ht->ht, &iter.iter, index,
2753 index_n.node) {
2754 health_code_update();
2755 relay_index_delete(index);
2756 relay_index_free_safe(index);
2757 }
2758 rcu_read_unlock();
2759 lttng_ht_destroy(indexes_ht);
2760 indexes_ht_error:
2761 lttng_ht_destroy(relay_connections_ht);
2762 relay_connections_ht_error:
2763 /* Close relay conn pipes */
2764 utils_close_pipe(relay_conn_pipe);
2765 if (err) {
2766 DBG("Thread exited with error");
2767 }
2768 DBG("Worker thread cleanup complete");
2769 free(data_buffer);
2770 error_testpoint:
2771 if (err) {
2772 health_error();
2773 ERR("Health error occurred in %s", __func__);
2774 }
2775 health_unregister(health_relayd);
2776 rcu_unregister_thread();
2777 lttng_relay_stop_threads();
2778 return NULL;
2779 }
2780
2781 /*
2782 * Create the relay command pipe to wake thread_manage_apps.
2783 * Closed in cleanup().
2784 */
2785 static int create_relay_conn_pipe(void)
2786 {
2787 int ret;
2788
2789 ret = utils_create_pipe_cloexec(relay_conn_pipe);
2790
2791 return ret;
2792 }
2793
2794 /*
2795 * main
2796 */
2797 int main(int argc, char **argv)
2798 {
2799 int ret = 0, retval = 0;
2800 void *status;
2801 struct relay_local_data *relay_ctx = NULL;
2802
2803 /* Parse arguments */
2804 progname = argv[0];
2805 if (set_options(argc, argv)) {
2806 retval = -1;
2807 goto exit_options;
2808 }
2809
2810 if (set_signal_handler()) {
2811 retval = -1;
2812 goto exit_options;
2813 }
2814
2815 /* Try to create directory if -o, --output is specified. */
2816 if (opt_output_path) {
2817 if (*opt_output_path != '/') {
2818 ERR("Please specify an absolute path for -o, --output PATH");
2819 retval = -1;
2820 goto exit_options;
2821 }
2822
2823 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2824 if (ret < 0) {
2825 ERR("Unable to create %s", opt_output_path);
2826 retval = -1;
2827 goto exit_options;
2828 }
2829 }
2830
2831 /* Daemonize */
2832 if (opt_daemon || opt_background) {
2833 int i;
2834
2835 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2836 !opt_background);
2837 if (ret < 0) {
2838 retval = -1;
2839 goto exit_options;
2840 }
2841
2842 /*
2843 * We are in the child. Make sure all other file
2844 * descriptors are closed, in case we are called with
2845 * more opened file descriptors than the standard ones.
2846 */
2847 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2848 (void) close(i);
2849 }
2850 }
2851
2852
2853 /* Initialize thread health monitoring */
2854 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2855 if (!health_relayd) {
2856 PERROR("health_app_create error");
2857 retval = -1;
2858 goto exit_health_app_create;
2859 }
2860
2861 /* Create thread quit pipe */
2862 if (init_thread_quit_pipe()) {
2863 retval = -1;
2864 goto exit_init_data;
2865 }
2866
2867 /* We need those values for the file/dir creation. */
2868 relayd_uid = getuid();
2869 relayd_gid = getgid();
2870
2871 /* Check if daemon is UID = 0 */
2872 if (relayd_uid == 0) {
2873 if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) {
2874 ERR("Need to be root to use ports < 1024");
2875 retval = -1;
2876 goto exit_init_data;
2877 }
2878 }
2879
2880 /* Setup the thread apps communication pipe. */
2881 if (create_relay_conn_pipe()) {
2882 retval = -1;
2883 goto exit_init_data;
2884 }
2885
2886 /* Init relay command queue. */
2887 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
2888
2889 /* Set up max poll set size */
2890 if (lttng_poll_set_max_size()) {
2891 retval = -1;
2892 goto exit_init_data;
2893 }
2894
2895 /* Initialize communication library */
2896 lttcomm_init();
2897 lttcomm_inet_init();
2898
2899 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2900 if (!relay_ctx) {
2901 PERROR("relay_ctx");
2902 retval = -1;
2903 goto exit_init_data;
2904 }
2905
2906 /* tables of sessions indexed by session ID */
2907 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2908 if (!relay_ctx->sessions_ht) {
2909 retval = -1;
2910 goto exit_init_data;
2911 }
2912
2913 /* tables of streams indexed by stream ID */
2914 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2915 if (!relay_streams_ht) {
2916 retval = -1;
2917 goto exit_init_data;
2918 }
2919
2920 /* tables of streams indexed by stream ID */
2921 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2922 if (!viewer_streams_ht) {
2923 retval = -1;
2924 goto exit_init_data;
2925 }
2926
2927 ret = utils_create_pipe(health_quit_pipe);
2928 if (ret) {
2929 retval = -1;
2930 goto exit_health_quit_pipe;
2931 }
2932
2933 /* Create thread to manage the client socket */
2934 ret = pthread_create(&health_thread, NULL,
2935 thread_manage_health, (void *) NULL);
2936 if (ret) {
2937 errno = ret;
2938 PERROR("pthread_create health");
2939 retval = -1;
2940 goto exit_health_thread;
2941 }
2942
2943 /* Setup the dispatcher thread */
2944 ret = pthread_create(&dispatcher_thread, NULL,
2945 relay_thread_dispatcher, (void *) NULL);
2946 if (ret) {
2947 errno = ret;
2948 PERROR("pthread_create dispatcher");
2949 retval = -1;
2950 goto exit_dispatcher_thread;
2951 }
2952
2953 /* Setup the worker thread */
2954 ret = pthread_create(&worker_thread, NULL,
2955 relay_thread_worker, (void *) relay_ctx);
2956 if (ret) {
2957 errno = ret;
2958 PERROR("pthread_create worker");
2959 retval = -1;
2960 goto exit_worker_thread;
2961 }
2962
2963 /* Setup the listener thread */
2964 ret = pthread_create(&listener_thread, NULL,
2965 relay_thread_listener, (void *) NULL);
2966 if (ret) {
2967 errno = ret;
2968 PERROR("pthread_create listener");
2969 retval = -1;
2970 goto exit_listener_thread;
2971 }
2972
2973 ret = relayd_live_create(live_uri, relay_ctx);
2974 if (ret) {
2975 ERR("Starting live viewer threads");
2976 retval = -1;
2977 goto exit_live;
2978 }
2979
2980 /*
2981 * This is where we start awaiting program completion (e.g. through
2982 * signal that asks threads to teardown).
2983 */
2984
2985 ret = relayd_live_join();
2986 if (ret) {
2987 retval = -1;
2988 }
2989 exit_live:
2990
2991 ret = pthread_join(listener_thread, &status);
2992 if (ret) {
2993 errno = ret;
2994 PERROR("pthread_join listener_thread");
2995 retval = -1;
2996 }
2997
2998 exit_listener_thread:
2999 ret = pthread_join(worker_thread, &status);
3000 if (ret) {
3001 errno = ret;
3002 PERROR("pthread_join worker_thread");
3003 retval = -1;
3004 }
3005
3006 exit_worker_thread:
3007 ret = pthread_join(dispatcher_thread, &status);
3008 if (ret) {
3009 errno = ret;
3010 PERROR("pthread_join dispatcher_thread");
3011 retval = -1;
3012 }
3013 exit_dispatcher_thread:
3014
3015 ret = pthread_join(health_thread, &status);
3016 if (ret) {
3017 errno = ret;
3018 PERROR("pthread_join health_thread");
3019 retval = -1;
3020 }
3021 exit_health_thread:
3022
3023 utils_close_pipe(health_quit_pipe);
3024 exit_health_quit_pipe:
3025
3026 exit_init_data:
3027 health_app_destroy(health_relayd);
3028 exit_health_app_create:
3029 exit_options:
3030 relayd_cleanup(relay_ctx);
3031
3032 if (!retval) {
3033 exit(EXIT_SUCCESS);
3034 } else {
3035 exit(EXIT_FAILURE);
3036 }
3037 }
This page took 0.179719 seconds and 4 git commands to generate.