Fix: call lttng_poll_set_max_size before start threads
[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/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/config.h>
58
59 #include "cmd.h"
60 #include "ctf-trace.h"
61 #include "index.h"
62 #include "utils.h"
63 #include "lttng-relayd.h"
64 #include "live.h"
65 #include "health-relayd.h"
66 #include "testpoint.h"
67 #include "viewer-stream.h"
68 #include "session.h"
69 #include "stream.h"
70 #include "connection.h"
71
72 /* command line options */
73 char *opt_output_path;
74 static int opt_daemon, opt_background;
75
76 /*
77 * We need to wait for listener and live listener threads, as well as
78 * health check thread, before being ready to signal readiness.
79 */
80 #define NR_LTTNG_RELAY_READY 3
81 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
82 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
83 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
84
85 static struct lttng_uri *control_uri;
86 static struct lttng_uri *data_uri;
87 static struct lttng_uri *live_uri;
88
89 const char *progname;
90
91 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
92 static int tracing_group_name_override;
93
94 const char * const config_section_name = "relayd";
95
96 /*
97 * Quit pipe for all threads. This permits a single cancellation point
98 * for all threads when receiving an event on the pipe.
99 */
100 int thread_quit_pipe[2] = { -1, -1 };
101
102 /*
103 * This pipe is used to inform the worker thread that a command is queued and
104 * ready to be processed.
105 */
106 static int relay_conn_pipe[2] = { -1, -1 };
107
108 /* Shared between threads */
109 static int dispatch_thread_exit;
110
111 static pthread_t listener_thread;
112 static pthread_t dispatcher_thread;
113 static pthread_t worker_thread;
114 static pthread_t health_thread;
115
116 static uint64_t last_relay_stream_id;
117
118 /*
119 * Relay command queue.
120 *
121 * The relay_thread_listener and relay_thread_dispatcher communicate with this
122 * queue.
123 */
124 static struct relay_conn_queue relay_conn_queue;
125
126 /* buffer allocated at startup, used to store the trace data */
127 static char *data_buffer;
128 static unsigned int data_buffer_size;
129
130 /* We need those values for the file/dir creation. */
131 static uid_t relayd_uid;
132 static gid_t relayd_gid;
133
134 /* Global relay stream hash table. */
135 struct lttng_ht *relay_streams_ht;
136
137 /* Global relay viewer stream hash table. */
138 struct lttng_ht *viewer_streams_ht;
139
140 /* Global hash table that stores relay index object. */
141 struct lttng_ht *indexes_ht;
142
143 /* Relayd health monitoring */
144 struct health_app *health_relayd;
145
146 static struct option long_options[] = {
147 { "control-port", 1, 0, 'C', },
148 { "data-port", 1, 0, 'D', },
149 { "live-port", 1, 0, 'L', },
150 { "daemonize", 0, 0, 'd', },
151 { "background", 0, 0, 'b', },
152 { "group", 1, 0, 'g', },
153 { "help", 0, 0, 'h', },
154 { "output", 1, 0, 'o', },
155 { "verbose", 0, 0, 'v', },
156 { "config", 1, 0, 'f' },
157 { NULL, 0, 0, 0, },
158 };
159
160 static const char *config_ignore_options[] = { "help", "config" };
161
162 /*
163 * usage function on stderr
164 */
165 static
166 void usage(void)
167 {
168 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
169 fprintf(stderr, " -h, --help Display this usage.\n");
170 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
171 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
172 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
173 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
174 fprintf(stderr, " -L, --live-port URL Live view port listening.\n");
175 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
176 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
177 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
178 fprintf(stderr, " -f --config Load daemon configuration file\n");
179 }
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
188 int set_option(int opt, const char *arg, const char *optname)
189 {
190 int ret;
191
192 switch (opt) {
193 case 0:
194 fprintf(stderr, "option %s", optname);
195 if (arg) {
196 fprintf(stderr, " with arg %s\n", arg);
197 }
198 break;
199 case 'C':
200 ret = uri_parse(arg, &control_uri);
201 if (ret < 0) {
202 ERR("Invalid control URI specified");
203 goto end;
204 }
205 if (control_uri->port == 0) {
206 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
207 }
208 break;
209 case 'D':
210 ret = uri_parse(arg, &data_uri);
211 if (ret < 0) {
212 ERR("Invalid data URI specified");
213 goto end;
214 }
215 if (data_uri->port == 0) {
216 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
217 }
218 break;
219 case 'L':
220 ret = uri_parse(arg, &live_uri);
221 if (ret < 0) {
222 ERR("Invalid live URI specified");
223 goto end;
224 }
225 if (live_uri->port == 0) {
226 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
227 }
228 break;
229 case 'd':
230 opt_daemon = 1;
231 break;
232 case 'b':
233 opt_background = 1;
234 break;
235 case 'g':
236 tracing_group_name = strdup(arg);
237 if (tracing_group_name == NULL) {
238 ret = -errno;
239 PERROR("strdup");
240 goto end;
241 }
242 tracing_group_name_override = 1;
243 break;
244 case 'h':
245 usage();
246 exit(EXIT_FAILURE);
247 case 'o':
248 ret = asprintf(&opt_output_path, "%s", arg);
249 if (ret < 0) {
250 ret = -errno;
251 PERROR("asprintf opt_output_path");
252 goto end;
253 }
254 break;
255 case 'v':
256 /* Verbose level can increase using multiple -v */
257 if (arg) {
258 lttng_opt_verbose = config_parse_value(arg);
259 } else {
260 /* Only 3 level of verbosity (-vvv). */
261 if (lttng_opt_verbose < 3) {
262 lttng_opt_verbose += 1;
263 }
264 }
265 break;
266 default:
267 /* Unknown option or other error.
268 * Error is printed by getopt, just return */
269 ret = -1;
270 goto end;
271 }
272
273 /* All good. */
274 ret = 0;
275
276 end:
277 return ret;
278 }
279
280 /*
281 * config_entry_handler_cb used to handle options read from a config file.
282 * See config_entry_handler_cb comment in common/config/config.h for the
283 * return value conventions.
284 */
285 static
286 int config_entry_handler(const struct config_entry *entry, void *unused)
287 {
288 int ret = 0, i;
289
290 if (!entry || !entry->name || !entry->value) {
291 ret = -EINVAL;
292 goto end;
293 }
294
295 /* Check if the option is to be ignored */
296 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
297 if (!strcmp(entry->name, config_ignore_options[i])) {
298 goto end;
299 }
300 }
301
302 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
303 /* Ignore if entry name is not fully matched. */
304 if (strcmp(entry->name, long_options[i].name)) {
305 continue;
306 }
307
308 /*
309 * If the option takes no argument on the command line, we have to
310 * check if the value is "true". We support non-zero numeric values,
311 * true, on and yes.
312 */
313 if (!long_options[i].has_arg) {
314 ret = config_parse_value(entry->value);
315 if (ret <= 0) {
316 if (ret) {
317 WARN("Invalid configuration value \"%s\" for option %s",
318 entry->value, entry->name);
319 }
320 /* False, skip boolean config option. */
321 goto end;
322 }
323 }
324
325 ret = set_option(long_options[i].val, entry->value, entry->name);
326 goto end;
327 }
328
329 WARN("Unrecognized option \"%s\" in daemon configuration file.",
330 entry->name);
331
332 end:
333 return ret;
334 }
335
336 static
337 int set_options(int argc, char **argv)
338 {
339 int c, ret = 0, option_index = 0, retval = 0;
340 int orig_optopt = optopt, orig_optind = optind;
341 char *default_address, *optstring;
342 const char *config_path = NULL;
343
344 optstring = utils_generate_optstring(long_options,
345 sizeof(long_options) / sizeof(struct option));
346 if (!optstring) {
347 retval = -ENOMEM;
348 goto exit;
349 }
350
351 /* Check for the --config option */
352
353 while ((c = getopt_long(argc, argv, optstring, long_options,
354 &option_index)) != -1) {
355 if (c == '?') {
356 retval = -EINVAL;
357 goto exit;
358 } else if (c != 'f') {
359 continue;
360 }
361
362 config_path = utils_expand_path(optarg);
363 if (!config_path) {
364 ERR("Failed to resolve path: %s", optarg);
365 }
366 }
367
368 ret = config_get_section_entries(config_path, config_section_name,
369 config_entry_handler, NULL);
370 if (ret) {
371 if (ret > 0) {
372 ERR("Invalid configuration option at line %i", ret);
373 }
374 retval = -1;
375 goto exit;
376 }
377
378 /* Reset getopt's global state */
379 optopt = orig_optopt;
380 optind = orig_optind;
381 while (1) {
382 c = getopt_long(argc, argv, optstring, long_options, &option_index);
383 if (c == -1) {
384 break;
385 }
386
387 ret = set_option(c, optarg, long_options[option_index].name);
388 if (ret < 0) {
389 retval = -1;
390 goto exit;
391 }
392 }
393
394 /* assign default values */
395 if (control_uri == NULL) {
396 ret = asprintf(&default_address,
397 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
398 DEFAULT_NETWORK_CONTROL_PORT);
399 if (ret < 0) {
400 PERROR("asprintf default data address");
401 retval = -1;
402 goto exit;
403 }
404
405 ret = uri_parse(default_address, &control_uri);
406 free(default_address);
407 if (ret < 0) {
408 ERR("Invalid control URI specified");
409 retval = -1;
410 goto exit;
411 }
412 }
413 if (data_uri == NULL) {
414 ret = asprintf(&default_address,
415 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
416 DEFAULT_NETWORK_DATA_PORT);
417 if (ret < 0) {
418 PERROR("asprintf default data address");
419 retval = -1;
420 goto exit;
421 }
422
423 ret = uri_parse(default_address, &data_uri);
424 free(default_address);
425 if (ret < 0) {
426 ERR("Invalid data URI specified");
427 retval = -1;
428 goto exit;
429 }
430 }
431 if (live_uri == NULL) {
432 ret = asprintf(&default_address,
433 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
434 DEFAULT_NETWORK_VIEWER_PORT);
435 if (ret < 0) {
436 PERROR("asprintf default viewer control address");
437 retval = -1;
438 goto exit;
439 }
440
441 ret = uri_parse(default_address, &live_uri);
442 free(default_address);
443 if (ret < 0) {
444 ERR("Invalid viewer control URI specified");
445 retval = -1;
446 goto exit;
447 }
448 }
449
450 exit:
451 free(optstring);
452 return retval;
453 }
454
455 /*
456 * Cleanup the daemon
457 */
458 static
459 void relayd_cleanup(struct relay_local_data *relay_ctx)
460 {
461 DBG("Cleaning up");
462
463 if (viewer_streams_ht)
464 lttng_ht_destroy(viewer_streams_ht);
465 if (relay_streams_ht)
466 lttng_ht_destroy(relay_streams_ht);
467 if (relay_ctx && relay_ctx->sessions_ht)
468 lttng_ht_destroy(relay_ctx->sessions_ht);
469 free(relay_ctx);
470
471 /* free the dynamically allocated opt_output_path */
472 free(opt_output_path);
473
474 /* Close thread quit pipes */
475 utils_close_pipe(thread_quit_pipe);
476
477 uri_free(control_uri);
478 uri_free(data_uri);
479 /* Live URI is freed in the live thread. */
480
481 if (tracing_group_name_override) {
482 free((void *) tracing_group_name);
483 }
484 }
485
486 /*
487 * Write to writable pipe used to notify a thread.
488 */
489 static
490 int notify_thread_pipe(int wpipe)
491 {
492 ssize_t ret;
493
494 ret = lttng_write(wpipe, "!", 1);
495 if (ret < 1) {
496 PERROR("write poll pipe");
497 goto end;
498 }
499 ret = 0;
500 end:
501 return ret;
502 }
503
504 static
505 int notify_health_quit_pipe(int *pipe)
506 {
507 ssize_t ret;
508
509 ret = lttng_write(pipe[1], "4", 1);
510 if (ret < 1) {
511 PERROR("write relay health quit");
512 goto end;
513 }
514 ret = 0;
515 end:
516 return ret;
517 }
518
519 /*
520 * Stop all relayd and relayd-live threads.
521 */
522 int lttng_relay_stop_threads(void)
523 {
524 int retval = 0;
525
526 /* Stopping all threads */
527 DBG("Terminating all threads");
528 if (notify_thread_pipe(thread_quit_pipe[1])) {
529 ERR("write error on thread quit pipe");
530 retval = -1;
531 }
532
533 if (notify_health_quit_pipe(health_quit_pipe)) {
534 ERR("write error on health quit pipe");
535 }
536
537 /* Dispatch thread */
538 CMM_STORE_SHARED(dispatch_thread_exit, 1);
539 futex_nto1_wake(&relay_conn_queue.futex);
540
541 if (relayd_live_stop()) {
542 ERR("Error stopping live threads");
543 retval = -1;
544 }
545 return retval;
546 }
547
548 /*
549 * Signal handler for the daemon
550 *
551 * Simply stop all worker threads, leaving main() return gracefully after
552 * joining all threads and calling cleanup().
553 */
554 static
555 void sighandler(int sig)
556 {
557 switch (sig) {
558 case SIGPIPE:
559 DBG("SIGPIPE caught");
560 return;
561 case SIGINT:
562 DBG("SIGINT caught");
563 if (lttng_relay_stop_threads()) {
564 ERR("Error stopping threads");
565 }
566 break;
567 case SIGTERM:
568 DBG("SIGTERM caught");
569 if (lttng_relay_stop_threads()) {
570 ERR("Error stopping threads");
571 }
572 break;
573 case SIGUSR1:
574 CMM_STORE_SHARED(recv_child_signal, 1);
575 break;
576 default:
577 break;
578 }
579 }
580
581 /*
582 * Setup signal handler for :
583 * SIGINT, SIGTERM, SIGPIPE
584 */
585 static
586 int set_signal_handler(void)
587 {
588 int ret = 0;
589 struct sigaction sa;
590 sigset_t sigset;
591
592 if ((ret = sigemptyset(&sigset)) < 0) {
593 PERROR("sigemptyset");
594 return ret;
595 }
596
597 sa.sa_handler = sighandler;
598 sa.sa_mask = sigset;
599 sa.sa_flags = 0;
600 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
601 PERROR("sigaction");
602 return ret;
603 }
604
605 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
606 PERROR("sigaction");
607 return ret;
608 }
609
610 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
611 PERROR("sigaction");
612 return ret;
613 }
614
615 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
616 PERROR("sigaction");
617 return ret;
618 }
619
620 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
621
622 return ret;
623 }
624
625 void lttng_relay_notify_ready(void)
626 {
627 /* Notify the parent of the fork() process that we are ready. */
628 if (opt_daemon || opt_background) {
629 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
630 kill(child_ppid, SIGUSR1);
631 }
632 }
633 }
634
635 /*
636 * Init thread quit pipe.
637 *
638 * Return -1 on error or 0 if all pipes are created.
639 */
640 static
641 int init_thread_quit_pipe(void)
642 {
643 int ret;
644
645 ret = utils_create_pipe_cloexec(thread_quit_pipe);
646
647 return ret;
648 }
649
650 /*
651 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
652 */
653 static
654 int create_thread_poll_set(struct lttng_poll_event *events, int size)
655 {
656 int ret;
657
658 if (events == NULL || size == 0) {
659 ret = -1;
660 goto error;
661 }
662
663 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
664 if (ret < 0) {
665 goto error;
666 }
667
668 /* Add quit pipe */
669 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
670 if (ret < 0) {
671 goto error;
672 }
673
674 return 0;
675
676 error:
677 return ret;
678 }
679
680 /*
681 * Check if the thread quit pipe was triggered.
682 *
683 * Return 1 if it was triggered else 0;
684 */
685 static
686 int check_thread_quit_pipe(int fd, uint32_t events)
687 {
688 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
689 return 1;
690 }
691
692 return 0;
693 }
694
695 /*
696 * Create and init socket from uri.
697 */
698 static
699 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
700 {
701 int ret;
702 struct lttcomm_sock *sock = NULL;
703
704 sock = lttcomm_alloc_sock_from_uri(uri);
705 if (sock == NULL) {
706 ERR("Allocating socket");
707 goto error;
708 }
709
710 ret = lttcomm_create_sock(sock);
711 if (ret < 0) {
712 goto error;
713 }
714 DBG("Listening on sock %d", sock->fd);
715
716 ret = sock->ops->bind(sock);
717 if (ret < 0) {
718 goto error;
719 }
720
721 ret = sock->ops->listen(sock, -1);
722 if (ret < 0) {
723 goto error;
724
725 }
726
727 return sock;
728
729 error:
730 if (sock) {
731 lttcomm_destroy_sock(sock);
732 }
733 return NULL;
734 }
735
736 /*
737 * Return nonzero if stream needs to be closed.
738 */
739 static
740 int close_stream_check(struct relay_stream *stream)
741 {
742 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
743 /*
744 * We are about to close the stream so set the data pending flag to 1
745 * which will make the end data pending command skip the stream which
746 * is now closed and ready. Note that after proceeding to a file close,
747 * the written file is ready for reading.
748 */
749 stream->data_pending_check_done = 1;
750 return 1;
751 }
752 return 0;
753 }
754
755 static void try_close_stream(struct relay_session *session,
756 struct relay_stream *stream)
757 {
758 int ret;
759 struct ctf_trace *ctf_trace;
760
761 assert(session);
762 assert(stream);
763
764 if (!close_stream_check(stream)) {
765 /* Can't close it, not ready for that. */
766 goto end;
767 }
768
769 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
770 stream->path_name);
771 assert(ctf_trace);
772
773 pthread_mutex_lock(&session->viewer_ready_lock);
774 ctf_trace->invalid_flag = 1;
775 pthread_mutex_unlock(&session->viewer_ready_lock);
776
777 ret = stream_close(session, stream);
778 if (ret || session->snapshot) {
779 /* Already close thus the ctf trace is being or has been destroyed. */
780 goto end;
781 }
782
783 ctf_trace_try_destroy(session, ctf_trace);
784
785 end:
786 return;
787 }
788
789 /*
790 * This thread manages the listening for new connections on the network
791 */
792 static
793 void *relay_thread_listener(void *data)
794 {
795 int i, ret, pollfd, err = -1;
796 uint32_t revents, nb_fd;
797 struct lttng_poll_event events;
798 struct lttcomm_sock *control_sock, *data_sock;
799
800 DBG("[thread] Relay listener started");
801
802 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
803
804 health_code_update();
805
806 control_sock = relay_init_sock(control_uri);
807 if (!control_sock) {
808 goto error_sock_control;
809 }
810
811 data_sock = relay_init_sock(data_uri);
812 if (!data_sock) {
813 goto error_sock_relay;
814 }
815
816 /*
817 * Pass 3 as size here for the thread quit pipe, control and data socket.
818 */
819 ret = create_thread_poll_set(&events, 3);
820 if (ret < 0) {
821 goto error_create_poll;
822 }
823
824 /* Add the control socket */
825 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
826 if (ret < 0) {
827 goto error_poll_add;
828 }
829
830 /* Add the data socket */
831 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
832 if (ret < 0) {
833 goto error_poll_add;
834 }
835
836 lttng_relay_notify_ready();
837
838 if (testpoint(relayd_thread_listener)) {
839 goto error_testpoint;
840 }
841
842 while (1) {
843 health_code_update();
844
845 DBG("Listener accepting connections");
846
847 restart:
848 health_poll_entry();
849 ret = lttng_poll_wait(&events, -1);
850 health_poll_exit();
851 if (ret < 0) {
852 /*
853 * Restart interrupted system call.
854 */
855 if (errno == EINTR) {
856 goto restart;
857 }
858 goto error;
859 }
860
861 nb_fd = ret;
862
863 DBG("Relay new connection received");
864 for (i = 0; i < nb_fd; i++) {
865 health_code_update();
866
867 /* Fetch once the poll data */
868 revents = LTTNG_POLL_GETEV(&events, i);
869 pollfd = LTTNG_POLL_GETFD(&events, i);
870
871 if (!revents) {
872 /* No activity for this FD (poll implementation). */
873 continue;
874 }
875
876 /* Thread quit pipe has been closed. Killing thread. */
877 ret = check_thread_quit_pipe(pollfd, revents);
878 if (ret) {
879 err = 0;
880 goto exit;
881 }
882
883 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
884 ERR("socket poll error");
885 goto error;
886 } else if (revents & LPOLLIN) {
887 /*
888 * Get allocated in this thread, enqueued to a global queue,
889 * dequeued and freed in the worker thread.
890 */
891 int val = 1;
892 struct relay_connection *new_conn;
893 struct lttcomm_sock *newsock;
894
895 new_conn = connection_create();
896 if (!new_conn) {
897 goto error;
898 }
899
900 if (pollfd == data_sock->fd) {
901 new_conn->type = RELAY_DATA;
902 newsock = data_sock->ops->accept(data_sock);
903 DBG("Relay data connection accepted, socket %d",
904 newsock->fd);
905 } else {
906 assert(pollfd == control_sock->fd);
907 new_conn->type = RELAY_CONTROL;
908 newsock = control_sock->ops->accept(control_sock);
909 DBG("Relay control connection accepted, socket %d",
910 newsock->fd);
911 }
912 if (!newsock) {
913 PERROR("accepting sock");
914 connection_free(new_conn);
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 connection_free(new_conn);
924 goto error;
925 }
926 new_conn->sock = newsock;
927
928 /* Enqueue request for the dispatcher thread. */
929 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
930 &new_conn->qnode);
931
932 /*
933 * Wake the dispatch queue futex. Implicit memory barrier with
934 * the exchange in cds_wfcq_enqueue.
935 */
936 futex_nto1_wake(&relay_conn_queue.futex);
937 }
938 }
939 }
940
941 exit:
942 error:
943 error_poll_add:
944 error_testpoint:
945 lttng_poll_clean(&events);
946 error_create_poll:
947 if (data_sock->fd >= 0) {
948 ret = data_sock->ops->close(data_sock);
949 if (ret) {
950 PERROR("close");
951 }
952 }
953 lttcomm_destroy_sock(data_sock);
954 error_sock_relay:
955 if (control_sock->fd >= 0) {
956 ret = control_sock->ops->close(control_sock);
957 if (ret) {
958 PERROR("close");
959 }
960 }
961 lttcomm_destroy_sock(control_sock);
962 error_sock_control:
963 if (err) {
964 health_error();
965 ERR("Health error occurred in %s", __func__);
966 }
967 health_unregister(health_relayd);
968 DBG("Relay listener thread cleanup complete");
969 lttng_relay_stop_threads();
970 return NULL;
971 }
972
973 /*
974 * This thread manages the dispatching of the requests to worker threads
975 */
976 static
977 void *relay_thread_dispatcher(void *data)
978 {
979 int err = -1;
980 ssize_t ret;
981 struct cds_wfcq_node *node;
982 struct relay_connection *new_conn = NULL;
983
984 DBG("[thread] Relay dispatcher started");
985
986 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
987
988 if (testpoint(relayd_thread_dispatcher)) {
989 goto error_testpoint;
990 }
991
992 health_code_update();
993
994 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
995 health_code_update();
996
997 /* Atomically prepare the queue futex */
998 futex_nto1_prepare(&relay_conn_queue.futex);
999
1000 do {
1001 health_code_update();
1002
1003 /* Dequeue commands */
1004 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1005 &relay_conn_queue.tail);
1006 if (node == NULL) {
1007 DBG("Woken up but nothing in the relay command queue");
1008 /* Continue thread execution */
1009 break;
1010 }
1011 new_conn = caa_container_of(node, struct relay_connection, qnode);
1012
1013 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1014
1015 /*
1016 * Inform worker thread of the new request. This call is blocking
1017 * so we can be assured that the data will be read at some point in
1018 * time or wait to the end of the world :)
1019 */
1020 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1021 if (ret < 0) {
1022 PERROR("write connection pipe");
1023 connection_destroy(new_conn);
1024 goto error;
1025 }
1026 } while (node != NULL);
1027
1028 /* Futex wait on queue. Blocking call on futex() */
1029 health_poll_entry();
1030 futex_nto1_wait(&relay_conn_queue.futex);
1031 health_poll_exit();
1032 }
1033
1034 /* Normal exit, no error */
1035 err = 0;
1036
1037 error:
1038 error_testpoint:
1039 if (err) {
1040 health_error();
1041 ERR("Health error occurred in %s", __func__);
1042 }
1043 health_unregister(health_relayd);
1044 DBG("Dispatch thread dying");
1045 lttng_relay_stop_threads();
1046 return NULL;
1047 }
1048
1049 static void try_close_streams(struct relay_session *session)
1050 {
1051 struct ctf_trace *ctf_trace;
1052 struct lttng_ht_iter iter;
1053
1054 assert(session);
1055
1056 pthread_mutex_lock(&session->viewer_ready_lock);
1057 rcu_read_lock();
1058 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
1059 node.node) {
1060 struct relay_stream *stream;
1061
1062 /* Close streams. */
1063 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
1064 stream_close(session, stream);
1065 }
1066
1067 ctf_trace->invalid_flag = 1;
1068 ctf_trace_try_destroy(session, ctf_trace);
1069 }
1070 rcu_read_unlock();
1071 pthread_mutex_unlock(&session->viewer_ready_lock);
1072 }
1073
1074 /*
1075 * Try to destroy a session within a connection.
1076 */
1077 static void destroy_session(struct relay_session *session,
1078 struct lttng_ht *sessions_ht)
1079 {
1080 assert(session);
1081 assert(sessions_ht);
1082
1083 /* Indicate that this session can be destroyed from now on. */
1084 session->close_flag = 1;
1085
1086 try_close_streams(session);
1087
1088 /*
1089 * This will try to delete and destroy the session if no viewer is attached
1090 * to it meaning the refcount is down to zero.
1091 */
1092 session_try_destroy(sessions_ht, session);
1093 }
1094
1095 /*
1096 * Copy index data from the control port to a given index object.
1097 */
1098 static void copy_index_control_data(struct relay_index *index,
1099 struct lttcomm_relayd_index *data)
1100 {
1101 assert(index);
1102 assert(data);
1103
1104 /*
1105 * The index on disk is encoded in big endian, so we don't need to convert
1106 * the data received on the network. The data_offset value is NEVER
1107 * modified here and is updated by the data thread.
1108 */
1109 index->index_data.packet_size = data->packet_size;
1110 index->index_data.content_size = data->content_size;
1111 index->index_data.timestamp_begin = data->timestamp_begin;
1112 index->index_data.timestamp_end = data->timestamp_end;
1113 index->index_data.events_discarded = data->events_discarded;
1114 index->index_data.stream_id = data->stream_id;
1115 }
1116
1117 /*
1118 * Handle the RELAYD_CREATE_SESSION command.
1119 *
1120 * On success, send back the session id or else return a negative value.
1121 */
1122 static
1123 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
1124 struct relay_connection *conn)
1125 {
1126 int ret = 0, send_ret;
1127 struct relay_session *session;
1128 struct lttcomm_relayd_status_session reply;
1129
1130 assert(recv_hdr);
1131 assert(conn);
1132
1133 memset(&reply, 0, sizeof(reply));
1134
1135 session = session_create();
1136 if (!session) {
1137 ret = -1;
1138 goto error;
1139 }
1140 session->minor = conn->minor;
1141 session->major = conn->major;
1142 conn->session_id = session->id;
1143 conn->session = session;
1144
1145 reply.session_id = htobe64(session->id);
1146
1147 switch (conn->minor) {
1148 case 1:
1149 case 2:
1150 case 3:
1151 break;
1152 case 4: /* LTTng sessiond 2.4 */
1153 default:
1154 ret = cmd_create_session_2_4(conn, session);
1155 }
1156
1157 lttng_ht_add_unique_u64(conn->sessions_ht, &session->session_n);
1158 DBG("Created session %" PRIu64, session->id);
1159
1160 error:
1161 if (ret < 0) {
1162 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1163 } else {
1164 reply.ret_code = htobe32(LTTNG_OK);
1165 }
1166
1167 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1168 if (send_ret < 0) {
1169 ERR("Relayd sending session id");
1170 ret = send_ret;
1171 }
1172
1173 return ret;
1174 }
1175
1176 /*
1177 * When we have received all the streams and the metadata for a channel,
1178 * we make them visible to the viewer threads.
1179 */
1180 static
1181 void set_viewer_ready_flag(struct relay_connection *conn)
1182 {
1183 struct relay_stream *stream, *tmp_stream;
1184
1185 pthread_mutex_lock(&conn->session->viewer_ready_lock);
1186 cds_list_for_each_entry_safe(stream, tmp_stream, &conn->recv_head,
1187 recv_list) {
1188 stream->viewer_ready = 1;
1189 cds_list_del(&stream->recv_list);
1190 }
1191 pthread_mutex_unlock(&conn->session->viewer_ready_lock);
1192 return;
1193 }
1194
1195 /*
1196 * Add a recv handle node to the connection recv list with the given stream
1197 * handle. A new node is allocated thus must be freed when the node is deleted
1198 * from the list.
1199 */
1200 static void queue_stream(struct relay_stream *stream,
1201 struct relay_connection *conn)
1202 {
1203 assert(conn);
1204 assert(stream);
1205
1206 cds_list_add(&stream->recv_list, &conn->recv_head);
1207 }
1208
1209 /*
1210 * relay_add_stream: allocate a new stream for a session
1211 */
1212 static
1213 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1214 struct relay_connection *conn)
1215 {
1216 int ret, send_ret;
1217 struct relay_session *session = conn->session;
1218 struct relay_stream *stream = NULL;
1219 struct lttcomm_relayd_status_stream reply;
1220 struct ctf_trace *trace;
1221
1222 if (!session || conn->version_check_done == 0) {
1223 ERR("Trying to add a stream before version check");
1224 ret = -1;
1225 goto end_no_session;
1226 }
1227
1228 stream = zmalloc(sizeof(struct relay_stream));
1229 if (stream == NULL) {
1230 PERROR("relay stream zmalloc");
1231 ret = -1;
1232 goto end_no_session;
1233 }
1234
1235 switch (conn->minor) {
1236 case 1: /* LTTng sessiond 2.1 */
1237 ret = cmd_recv_stream_2_1(conn, stream);
1238 break;
1239 case 2: /* LTTng sessiond 2.2 */
1240 default:
1241 ret = cmd_recv_stream_2_2(conn, stream);
1242 break;
1243 }
1244 if (ret < 0) {
1245 goto err_free_stream;
1246 }
1247
1248 rcu_read_lock();
1249 stream->stream_handle = ++last_relay_stream_id;
1250 stream->prev_seq = -1ULL;
1251 stream->session_id = session->id;
1252 stream->index_fd = -1;
1253 stream->read_index_fd = -1;
1254 stream->ctf_stream_id = -1ULL;
1255 lttng_ht_node_init_u64(&stream->node, stream->stream_handle);
1256 pthread_mutex_init(&stream->lock, NULL);
1257
1258 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
1259 if (ret < 0) {
1260 ERR("relay creating output directory");
1261 goto end;
1262 }
1263
1264 /*
1265 * No need to use run_as API here because whatever we receives, the relayd
1266 * uses its own credentials for the stream files.
1267 */
1268 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1269 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1270 if (ret < 0) {
1271 ERR("Create output file");
1272 goto end;
1273 }
1274 stream->fd = ret;
1275 if (stream->tracefile_size) {
1276 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1277 } else {
1278 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1279 }
1280
1281 trace = ctf_trace_find_by_path(session->ctf_traces_ht, stream->path_name);
1282 if (!trace) {
1283 trace = ctf_trace_create(stream->path_name);
1284 if (!trace) {
1285 ret = -1;
1286 goto end;
1287 }
1288 ctf_trace_add(session->ctf_traces_ht, trace);
1289 }
1290 ctf_trace_get_ref(trace);
1291
1292 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1293 stream->metadata_flag = 1;
1294 /* Assign quick reference to the metadata stream in the trace. */
1295 trace->metadata_stream = stream;
1296 }
1297
1298 /*
1299 * Add the stream in the recv list of the connection. Once the end stream
1300 * message is received, this list is emptied and streams are set with the
1301 * viewer ready flag.
1302 */
1303 queue_stream(stream, conn);
1304
1305 /*
1306 * Both in the ctf_trace object and the global stream ht since the data
1307 * side of the relayd does not have the concept of session.
1308 */
1309 lttng_ht_add_unique_u64(relay_streams_ht, &stream->node);
1310 cds_list_add_tail(&stream->trace_list, &trace->stream_list);
1311
1312 session->stream_count++;
1313
1314 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1315 stream->stream_handle);
1316
1317 end:
1318 memset(&reply, 0, sizeof(reply));
1319 reply.handle = htobe64(stream->stream_handle);
1320 /* send the session id to the client or a negative return code on error */
1321 if (ret < 0) {
1322 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1323 /* stream was not properly added to the ht, so free it */
1324 free(stream);
1325 } else {
1326 reply.ret_code = htobe32(LTTNG_OK);
1327 }
1328
1329 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1330 sizeof(struct lttcomm_relayd_status_stream), 0);
1331 if (send_ret < 0) {
1332 ERR("Relay sending stream id");
1333 ret = send_ret;
1334 }
1335 rcu_read_unlock();
1336
1337 end_no_session:
1338 return ret;
1339
1340 err_free_stream:
1341 free(stream->path_name);
1342 free(stream->channel_name);
1343 free(stream);
1344 return ret;
1345 }
1346
1347 /*
1348 * relay_close_stream: close a specific stream
1349 */
1350 static
1351 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1352 struct relay_connection *conn)
1353 {
1354 int ret, send_ret;
1355 struct relay_session *session = conn->session;
1356 struct lttcomm_relayd_close_stream stream_info;
1357 struct lttcomm_relayd_generic_reply reply;
1358 struct relay_stream *stream;
1359
1360 DBG("Close stream received");
1361
1362 if (!session || conn->version_check_done == 0) {
1363 ERR("Trying to close a stream before version check");
1364 ret = -1;
1365 goto end_no_session;
1366 }
1367
1368 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1369 sizeof(struct lttcomm_relayd_close_stream), 0);
1370 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1371 if (ret == 0) {
1372 /* Orderly shutdown. Not necessary to print an error. */
1373 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1374 } else {
1375 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1376 }
1377 ret = -1;
1378 goto end_no_session;
1379 }
1380
1381 rcu_read_lock();
1382 stream = stream_find_by_id(relay_streams_ht,
1383 be64toh(stream_info.stream_id));
1384 if (!stream) {
1385 ret = -1;
1386 goto end_unlock;
1387 }
1388
1389 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1390 stream->close_flag = 1;
1391 session->stream_count--;
1392
1393 /* Check if we can close it or else the data will do it. */
1394 try_close_stream(session, stream);
1395
1396 end_unlock:
1397 rcu_read_unlock();
1398
1399 memset(&reply, 0, sizeof(reply));
1400 if (ret < 0) {
1401 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1402 } else {
1403 reply.ret_code = htobe32(LTTNG_OK);
1404 }
1405 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1406 sizeof(struct lttcomm_relayd_generic_reply), 0);
1407 if (send_ret < 0) {
1408 ERR("Relay sending stream id");
1409 ret = send_ret;
1410 }
1411
1412 end_no_session:
1413 return ret;
1414 }
1415
1416 /*
1417 * relay_unknown_command: send -1 if received unknown command
1418 */
1419 static
1420 void relay_unknown_command(struct relay_connection *conn)
1421 {
1422 struct lttcomm_relayd_generic_reply reply;
1423 int ret;
1424
1425 memset(&reply, 0, sizeof(reply));
1426 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1427 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1428 sizeof(struct lttcomm_relayd_generic_reply), 0);
1429 if (ret < 0) {
1430 ERR("Relay sending unknown command");
1431 }
1432 }
1433
1434 /*
1435 * relay_start: send an acknowledgment to the client to tell if we are
1436 * ready to receive data. We are ready if a session is established.
1437 */
1438 static
1439 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1440 struct relay_connection *conn)
1441 {
1442 int ret = htobe32(LTTNG_OK);
1443 struct lttcomm_relayd_generic_reply reply;
1444 struct relay_session *session = conn->session;
1445
1446 if (!session) {
1447 DBG("Trying to start the streaming without a session established");
1448 ret = htobe32(LTTNG_ERR_UNK);
1449 }
1450
1451 memset(&reply, 0, sizeof(reply));
1452 reply.ret_code = ret;
1453 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1454 sizeof(struct lttcomm_relayd_generic_reply), 0);
1455 if (ret < 0) {
1456 ERR("Relay sending start ack");
1457 }
1458
1459 return ret;
1460 }
1461
1462 /*
1463 * Append padding to the file pointed by the file descriptor fd.
1464 */
1465 static int write_padding_to_file(int fd, uint32_t size)
1466 {
1467 ssize_t ret = 0;
1468 char *zeros;
1469
1470 if (size == 0) {
1471 goto end;
1472 }
1473
1474 zeros = zmalloc(size);
1475 if (zeros == NULL) {
1476 PERROR("zmalloc zeros for padding");
1477 ret = -1;
1478 goto end;
1479 }
1480
1481 ret = lttng_write(fd, zeros, size);
1482 if (ret < size) {
1483 PERROR("write padding to file");
1484 }
1485
1486 free(zeros);
1487
1488 end:
1489 return ret;
1490 }
1491
1492 /*
1493 * relay_recv_metadata: receive the metada for the session.
1494 */
1495 static
1496 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1497 struct relay_connection *conn)
1498 {
1499 int ret = htobe32(LTTNG_OK);
1500 ssize_t size_ret;
1501 struct relay_session *session = conn->session;
1502 struct lttcomm_relayd_metadata_payload *metadata_struct;
1503 struct relay_stream *metadata_stream;
1504 uint64_t data_size, payload_size;
1505 struct ctf_trace *ctf_trace;
1506
1507 if (!session) {
1508 ERR("Metadata sent before version check");
1509 ret = -1;
1510 goto end;
1511 }
1512
1513 data_size = payload_size = be64toh(recv_hdr->data_size);
1514 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1515 ERR("Incorrect data size");
1516 ret = -1;
1517 goto end;
1518 }
1519 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1520
1521 if (data_buffer_size < data_size) {
1522 /* In case the realloc fails, we can free the memory */
1523 char *tmp_data_ptr;
1524
1525 tmp_data_ptr = realloc(data_buffer, data_size);
1526 if (!tmp_data_ptr) {
1527 ERR("Allocating data buffer");
1528 free(data_buffer);
1529 ret = -1;
1530 goto end;
1531 }
1532 data_buffer = tmp_data_ptr;
1533 data_buffer_size = data_size;
1534 }
1535 memset(data_buffer, 0, data_size);
1536 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1537 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
1538 if (ret < 0 || ret != data_size) {
1539 if (ret == 0) {
1540 /* Orderly shutdown. Not necessary to print an error. */
1541 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1542 } else {
1543 ERR("Relay didn't receive the whole metadata");
1544 }
1545 ret = -1;
1546 goto end;
1547 }
1548 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1549
1550 rcu_read_lock();
1551 metadata_stream = stream_find_by_id(relay_streams_ht,
1552 be64toh(metadata_struct->stream_id));
1553 if (!metadata_stream) {
1554 ret = -1;
1555 goto end_unlock;
1556 }
1557
1558 size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
1559 payload_size);
1560 if (size_ret < payload_size) {
1561 ERR("Relay error writing metadata on file");
1562 ret = -1;
1563 goto end_unlock;
1564 }
1565
1566 ret = write_padding_to_file(metadata_stream->fd,
1567 be32toh(metadata_struct->padding_size));
1568 if (ret < 0) {
1569 goto end_unlock;
1570 }
1571
1572 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1573 metadata_stream->path_name);
1574 assert(ctf_trace);
1575 ctf_trace->metadata_received +=
1576 payload_size + be32toh(metadata_struct->padding_size);
1577
1578 DBG2("Relay metadata written");
1579
1580 end_unlock:
1581 rcu_read_unlock();
1582 end:
1583 return ret;
1584 }
1585
1586 /*
1587 * relay_send_version: send relayd version number
1588 */
1589 static
1590 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1591 struct relay_connection *conn)
1592 {
1593 int ret;
1594 struct lttcomm_relayd_version reply, msg;
1595
1596 assert(conn);
1597
1598 conn->version_check_done = 1;
1599
1600 /* Get version from the other side. */
1601 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1602 if (ret < 0 || ret != sizeof(msg)) {
1603 if (ret == 0) {
1604 /* Orderly shutdown. Not necessary to print an error. */
1605 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1606 } else {
1607 ERR("Relay failed to receive the version values.");
1608 }
1609 ret = -1;
1610 goto end;
1611 }
1612
1613 memset(&reply, 0, sizeof(reply));
1614 reply.major = RELAYD_VERSION_COMM_MAJOR;
1615 reply.minor = RELAYD_VERSION_COMM_MINOR;
1616
1617 /* Major versions must be the same */
1618 if (reply.major != be32toh(msg.major)) {
1619 DBG("Incompatible major versions (%u vs %u), deleting session",
1620 reply.major, be32toh(msg.major));
1621 destroy_session(conn->session, conn->sessions_ht);
1622 ret = 0;
1623 goto end;
1624 }
1625
1626 conn->major = reply.major;
1627 /* We adapt to the lowest compatible version */
1628 if (reply.minor <= be32toh(msg.minor)) {
1629 conn->minor = reply.minor;
1630 } else {
1631 conn->minor = be32toh(msg.minor);
1632 }
1633
1634 reply.major = htobe32(reply.major);
1635 reply.minor = htobe32(reply.minor);
1636 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1637 sizeof(struct lttcomm_relayd_version), 0);
1638 if (ret < 0) {
1639 ERR("Relay sending version");
1640 }
1641
1642 DBG("Version check done using protocol %u.%u", conn->major,
1643 conn->minor);
1644
1645 end:
1646 return ret;
1647 }
1648
1649 /*
1650 * Check for data pending for a given stream id from the session daemon.
1651 */
1652 static
1653 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1654 struct relay_connection *conn)
1655 {
1656 struct relay_session *session = conn->session;
1657 struct lttcomm_relayd_data_pending msg;
1658 struct lttcomm_relayd_generic_reply reply;
1659 struct relay_stream *stream;
1660 int ret;
1661 uint64_t last_net_seq_num, stream_id;
1662
1663 DBG("Data pending command received");
1664
1665 if (!session || conn->version_check_done == 0) {
1666 ERR("Trying to check for data before version check");
1667 ret = -1;
1668 goto end_no_session;
1669 }
1670
1671 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1672 if (ret < sizeof(msg)) {
1673 if (ret == 0) {
1674 /* Orderly shutdown. Not necessary to print an error. */
1675 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1676 } else {
1677 ERR("Relay didn't receive valid data_pending struct size : %d",
1678 ret);
1679 }
1680 ret = -1;
1681 goto end_no_session;
1682 }
1683
1684 stream_id = be64toh(msg.stream_id);
1685 last_net_seq_num = be64toh(msg.last_net_seq_num);
1686
1687 rcu_read_lock();
1688 stream = stream_find_by_id(relay_streams_ht, stream_id);
1689 if (stream == NULL) {
1690 ret = -1;
1691 goto end_unlock;
1692 }
1693
1694 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1695 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1696 last_net_seq_num);
1697
1698 /* Avoid wrapping issue */
1699 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1700 /* Data has in fact been written and is NOT pending */
1701 ret = 0;
1702 } else {
1703 /* Data still being streamed thus pending */
1704 ret = 1;
1705 }
1706
1707 /* Pending check is now done. */
1708 stream->data_pending_check_done = 1;
1709
1710 end_unlock:
1711 rcu_read_unlock();
1712
1713 memset(&reply, 0, sizeof(reply));
1714 reply.ret_code = htobe32(ret);
1715 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1716 if (ret < 0) {
1717 ERR("Relay data pending ret code failed");
1718 }
1719
1720 end_no_session:
1721 return ret;
1722 }
1723
1724 /*
1725 * Wait for the control socket to reach a quiescent state.
1726 *
1727 * Note that for now, when receiving this command from the session daemon, this
1728 * means that every subsequent commands or data received on the control socket
1729 * has been handled. So, this is why we simply return OK here.
1730 */
1731 static
1732 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1733 struct relay_connection *conn)
1734 {
1735 int ret;
1736 uint64_t stream_id;
1737 struct relay_stream *stream;
1738 struct lttng_ht_iter iter;
1739 struct lttcomm_relayd_quiescent_control msg;
1740 struct lttcomm_relayd_generic_reply reply;
1741
1742 DBG("Checking quiescent state on control socket");
1743
1744 if (!conn->session || conn->version_check_done == 0) {
1745 ERR("Trying to check for data before version check");
1746 ret = -1;
1747 goto end_no_session;
1748 }
1749
1750 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1751 if (ret < sizeof(msg)) {
1752 if (ret == 0) {
1753 /* Orderly shutdown. Not necessary to print an error. */
1754 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1755 } else {
1756 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1757 ret);
1758 }
1759 ret = -1;
1760 goto end_no_session;
1761 }
1762
1763 stream_id = be64toh(msg.stream_id);
1764
1765 rcu_read_lock();
1766 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1767 node.node) {
1768 if (stream->stream_handle == stream_id) {
1769 stream->data_pending_check_done = 1;
1770 DBG("Relay quiescent control pending flag set to %" PRIu64,
1771 stream_id);
1772 break;
1773 }
1774 }
1775 rcu_read_unlock();
1776
1777 memset(&reply, 0, sizeof(reply));
1778 reply.ret_code = htobe32(LTTNG_OK);
1779 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1780 if (ret < 0) {
1781 ERR("Relay data quiescent control ret code failed");
1782 }
1783
1784 end_no_session:
1785 return ret;
1786 }
1787
1788 /*
1789 * Initialize a data pending command. This means that a client is about to ask
1790 * for data pending for each stream he/she holds. Simply iterate over all
1791 * streams of a session and set the data_pending_check_done flag.
1792 *
1793 * This command returns to the client a LTTNG_OK code.
1794 */
1795 static
1796 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1797 struct relay_connection *conn)
1798 {
1799 int ret;
1800 struct lttng_ht_iter iter;
1801 struct lttcomm_relayd_begin_data_pending msg;
1802 struct lttcomm_relayd_generic_reply reply;
1803 struct relay_stream *stream;
1804 uint64_t session_id;
1805
1806 assert(recv_hdr);
1807 assert(conn);
1808
1809 DBG("Init streams for data pending");
1810
1811 if (!conn->session || conn->version_check_done == 0) {
1812 ERR("Trying to check for data before version check");
1813 ret = -1;
1814 goto end_no_session;
1815 }
1816
1817 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1818 if (ret < sizeof(msg)) {
1819 if (ret == 0) {
1820 /* Orderly shutdown. Not necessary to print an error. */
1821 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1822 } else {
1823 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1824 ret);
1825 }
1826 ret = -1;
1827 goto end_no_session;
1828 }
1829
1830 session_id = be64toh(msg.session_id);
1831
1832 /*
1833 * Iterate over all streams to set the begin data pending flag. For now, the
1834 * streams are indexed by stream handle so we have to iterate over all
1835 * streams to find the one associated with the right session_id.
1836 */
1837 rcu_read_lock();
1838 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1839 node.node) {
1840 if (stream->session_id == session_id) {
1841 stream->data_pending_check_done = 0;
1842 DBG("Set begin data pending flag to stream %" PRIu64,
1843 stream->stream_handle);
1844 }
1845 }
1846 rcu_read_unlock();
1847
1848 memset(&reply, 0, sizeof(reply));
1849 /* All good, send back reply. */
1850 reply.ret_code = htobe32(LTTNG_OK);
1851
1852 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1853 if (ret < 0) {
1854 ERR("Relay begin data pending send reply failed");
1855 }
1856
1857 end_no_session:
1858 return ret;
1859 }
1860
1861 /*
1862 * End data pending command. This will check, for a given session id, if each
1863 * stream associated with it has its data_pending_check_done flag set. If not,
1864 * this means that the client lost track of the stream but the data is still
1865 * being streamed on our side. In this case, we inform the client that data is
1866 * inflight.
1867 *
1868 * Return to the client if there is data in flight or not with a ret_code.
1869 */
1870 static
1871 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1872 struct relay_connection *conn)
1873 {
1874 int ret;
1875 struct lttng_ht_iter iter;
1876 struct lttcomm_relayd_end_data_pending msg;
1877 struct lttcomm_relayd_generic_reply reply;
1878 struct relay_stream *stream;
1879 uint64_t session_id;
1880 uint32_t is_data_inflight = 0;
1881
1882 assert(recv_hdr);
1883 assert(conn);
1884
1885 DBG("End data pending command");
1886
1887 if (!conn->session || conn->version_check_done == 0) {
1888 ERR("Trying to check for data before version check");
1889 ret = -1;
1890 goto end_no_session;
1891 }
1892
1893 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1894 if (ret < sizeof(msg)) {
1895 if (ret == 0) {
1896 /* Orderly shutdown. Not necessary to print an error. */
1897 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1898 } else {
1899 ERR("Relay didn't receive valid end data_pending struct size: %d",
1900 ret);
1901 }
1902 ret = -1;
1903 goto end_no_session;
1904 }
1905
1906 session_id = be64toh(msg.session_id);
1907
1908 /* Iterate over all streams to see if the begin data pending flag is set. */
1909 rcu_read_lock();
1910 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1911 node.node) {
1912 if (stream->session_id == session_id &&
1913 !stream->data_pending_check_done && !stream->terminated_flag) {
1914 is_data_inflight = 1;
1915 DBG("Data is still in flight for stream %" PRIu64,
1916 stream->stream_handle);
1917 break;
1918 }
1919 }
1920 rcu_read_unlock();
1921
1922 memset(&reply, 0, sizeof(reply));
1923 /* All good, send back reply. */
1924 reply.ret_code = htobe32(is_data_inflight);
1925
1926 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1927 if (ret < 0) {
1928 ERR("Relay end data pending send reply failed");
1929 }
1930
1931 end_no_session:
1932 return ret;
1933 }
1934
1935 /*
1936 * Receive an index for a specific stream.
1937 *
1938 * Return 0 on success else a negative value.
1939 */
1940 static
1941 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1942 struct relay_connection *conn)
1943 {
1944 int ret, send_ret, index_created = 0;
1945 struct relay_session *session = conn->session;
1946 struct lttcomm_relayd_index index_info;
1947 struct relay_index *index, *wr_index = NULL;
1948 struct lttcomm_relayd_generic_reply reply;
1949 struct relay_stream *stream;
1950 uint64_t net_seq_num;
1951
1952 assert(conn);
1953
1954 DBG("Relay receiving index");
1955
1956 if (!session || conn->version_check_done == 0) {
1957 ERR("Trying to close a stream before version check");
1958 ret = -1;
1959 goto end_no_session;
1960 }
1961
1962 ret = conn->sock->ops->recvmsg(conn->sock, &index_info,
1963 sizeof(index_info), 0);
1964 if (ret < sizeof(index_info)) {
1965 if (ret == 0) {
1966 /* Orderly shutdown. Not necessary to print an error. */
1967 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1968 } else {
1969 ERR("Relay didn't receive valid index struct size : %d", ret);
1970 }
1971 ret = -1;
1972 goto end_no_session;
1973 }
1974
1975 net_seq_num = be64toh(index_info.net_seq_num);
1976
1977 rcu_read_lock();
1978 stream = stream_find_by_id(relay_streams_ht,
1979 be64toh(index_info.relay_stream_id));
1980 if (!stream) {
1981 ret = -1;
1982 goto end_rcu_unlock;
1983 }
1984
1985 /* Live beacon handling */
1986 if (index_info.packet_size == 0) {
1987 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1988
1989 /*
1990 * Only flag a stream inactive when it has already received data
1991 * and no indexes are in flight.
1992 */
1993 if (stream->total_index_received > 0 && stream->indexes_in_flight == 0) {
1994 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1995 }
1996 ret = 0;
1997 goto end_rcu_unlock;
1998 } else {
1999 stream->beacon_ts_end = -1ULL;
2000 }
2001
2002 index = relay_index_find(stream->stream_handle, net_seq_num);
2003 if (!index) {
2004 /* A successful creation will add the object to the HT. */
2005 index = relay_index_create(stream->stream_handle, net_seq_num);
2006 if (!index) {
2007 goto end_rcu_unlock;
2008 }
2009 index_created = 1;
2010 stream->indexes_in_flight++;
2011 }
2012
2013 copy_index_control_data(index, &index_info);
2014 if (stream->ctf_stream_id == -1ULL) {
2015 stream->ctf_stream_id = be64toh(index_info.stream_id);
2016 }
2017
2018 if (index_created) {
2019 /*
2020 * Try to add the relay index object to the hash table. If an object
2021 * already exist, destroy back the index created, set the data in this
2022 * object and write it on disk.
2023 */
2024 relay_index_add(index, &wr_index);
2025 if (wr_index) {
2026 copy_index_control_data(wr_index, &index_info);
2027 free(index);
2028 }
2029 } else {
2030 /* The index already exists so write it on disk. */
2031 wr_index = index;
2032 }
2033
2034 /* Do we have a writable ready index to write on disk. */
2035 if (wr_index) {
2036 ret = relay_index_write(wr_index->fd, wr_index);
2037 if (ret < 0) {
2038 goto end_rcu_unlock;
2039 }
2040 stream->total_index_received++;
2041 stream->indexes_in_flight--;
2042 assert(stream->indexes_in_flight >= 0);
2043 }
2044
2045 end_rcu_unlock:
2046 rcu_read_unlock();
2047
2048 memset(&reply, 0, sizeof(reply));
2049 if (ret < 0) {
2050 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2051 } else {
2052 reply.ret_code = htobe32(LTTNG_OK);
2053 }
2054 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2055 if (send_ret < 0) {
2056 ERR("Relay sending close index id reply");
2057 ret = send_ret;
2058 }
2059
2060 end_no_session:
2061 return ret;
2062 }
2063
2064 /*
2065 * Receive the streams_sent message.
2066 *
2067 * Return 0 on success else a negative value.
2068 */
2069 static
2070 int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
2071 struct relay_connection *conn)
2072 {
2073 int ret, send_ret;
2074 struct lttcomm_relayd_generic_reply reply;
2075
2076 assert(conn);
2077
2078 DBG("Relay receiving streams_sent");
2079
2080 if (!conn->session || conn->version_check_done == 0) {
2081 ERR("Trying to close a stream before version check");
2082 ret = -1;
2083 goto end_no_session;
2084 }
2085
2086 /*
2087 * Flag every pending stream in the connection recv list that they are
2088 * ready to be used by the viewer.
2089 */
2090 set_viewer_ready_flag(conn);
2091
2092 /*
2093 * Inform the viewer that there are new streams in the session.
2094 */
2095 if (conn->session->viewer_refcount) {
2096 uatomic_set(&conn->session->new_streams, 1);
2097 }
2098
2099 memset(&reply, 0, sizeof(reply));
2100 reply.ret_code = htobe32(LTTNG_OK);
2101 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2102 if (send_ret < 0) {
2103 ERR("Relay sending sent_stream reply");
2104 ret = send_ret;
2105 } else {
2106 /* Success. */
2107 ret = 0;
2108 }
2109
2110 end_no_session:
2111 return ret;
2112 }
2113
2114 /*
2115 * Process the commands received on the control socket
2116 */
2117 static
2118 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
2119 struct relay_connection *conn)
2120 {
2121 int ret = 0;
2122
2123 switch (be32toh(recv_hdr->cmd)) {
2124 case RELAYD_CREATE_SESSION:
2125 ret = relay_create_session(recv_hdr, conn);
2126 break;
2127 case RELAYD_ADD_STREAM:
2128 ret = relay_add_stream(recv_hdr, conn);
2129 break;
2130 case RELAYD_START_DATA:
2131 ret = relay_start(recv_hdr, conn);
2132 break;
2133 case RELAYD_SEND_METADATA:
2134 ret = relay_recv_metadata(recv_hdr, conn);
2135 break;
2136 case RELAYD_VERSION:
2137 ret = relay_send_version(recv_hdr, conn);
2138 break;
2139 case RELAYD_CLOSE_STREAM:
2140 ret = relay_close_stream(recv_hdr, conn);
2141 break;
2142 case RELAYD_DATA_PENDING:
2143 ret = relay_data_pending(recv_hdr, conn);
2144 break;
2145 case RELAYD_QUIESCENT_CONTROL:
2146 ret = relay_quiescent_control(recv_hdr, conn);
2147 break;
2148 case RELAYD_BEGIN_DATA_PENDING:
2149 ret = relay_begin_data_pending(recv_hdr, conn);
2150 break;
2151 case RELAYD_END_DATA_PENDING:
2152 ret = relay_end_data_pending(recv_hdr, conn);
2153 break;
2154 case RELAYD_SEND_INDEX:
2155 ret = relay_recv_index(recv_hdr, conn);
2156 break;
2157 case RELAYD_STREAMS_SENT:
2158 ret = relay_streams_sent(recv_hdr, conn);
2159 break;
2160 case RELAYD_UPDATE_SYNC_INFO:
2161 default:
2162 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2163 relay_unknown_command(conn);
2164 ret = -1;
2165 goto end;
2166 }
2167
2168 end:
2169 return ret;
2170 }
2171
2172 /*
2173 * Handle index for a data stream.
2174 *
2175 * RCU read side lock MUST be acquired.
2176 *
2177 * Return 0 on success else a negative value.
2178 */
2179 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2180 int rotate_index)
2181 {
2182 int ret = 0, index_created = 0;
2183 uint64_t stream_id, data_offset;
2184 struct relay_index *index, *wr_index = NULL;
2185
2186 assert(stream);
2187
2188 stream_id = stream->stream_handle;
2189 /* Get data offset because we are about to update the index. */
2190 data_offset = htobe64(stream->tracefile_size_current);
2191
2192 /*
2193 * Lookup for an existing index for that stream id/sequence number. If on
2194 * exists, the control thread already received the data for it thus we need
2195 * to write it on disk.
2196 */
2197 index = relay_index_find(stream_id, net_seq_num);
2198 if (!index) {
2199 /* A successful creation will add the object to the HT. */
2200 index = relay_index_create(stream_id, net_seq_num);
2201 if (!index) {
2202 ret = -1;
2203 goto error;
2204 }
2205 index_created = 1;
2206 stream->indexes_in_flight++;
2207 }
2208
2209 if (rotate_index || stream->index_fd < 0) {
2210 index->to_close_fd = stream->index_fd;
2211 ret = index_create_file(stream->path_name, stream->channel_name,
2212 relayd_uid, relayd_gid, stream->tracefile_size,
2213 stream->tracefile_count_current);
2214 if (ret < 0) {
2215 /* This will close the stream's index fd if one. */
2216 relay_index_free_safe(index);
2217 goto error;
2218 }
2219 stream->index_fd = ret;
2220 }
2221 index->fd = stream->index_fd;
2222 index->index_data.offset = data_offset;
2223
2224 if (index_created) {
2225 /*
2226 * Try to add the relay index object to the hash table. If an object
2227 * already exist, destroy back the index created and set the data.
2228 */
2229 relay_index_add(index, &wr_index);
2230 if (wr_index) {
2231 /* Copy back data from the created index. */
2232 wr_index->fd = index->fd;
2233 wr_index->to_close_fd = index->to_close_fd;
2234 wr_index->index_data.offset = data_offset;
2235 free(index);
2236 }
2237 } else {
2238 /* The index already exists so write it on disk. */
2239 wr_index = index;
2240 }
2241
2242 /* Do we have a writable ready index to write on disk. */
2243 if (wr_index) {
2244 ret = relay_index_write(wr_index->fd, wr_index);
2245 if (ret < 0) {
2246 goto error;
2247 }
2248 stream->total_index_received++;
2249 stream->indexes_in_flight--;
2250 assert(stream->indexes_in_flight >= 0);
2251 }
2252
2253 error:
2254 return ret;
2255 }
2256
2257 /*
2258 * relay_process_data: Process the data received on the data socket
2259 */
2260 static
2261 int relay_process_data(struct relay_connection *conn)
2262 {
2263 int ret = 0, rotate_index = 0;
2264 ssize_t size_ret;
2265 struct relay_stream *stream;
2266 struct lttcomm_relayd_data_hdr data_hdr;
2267 uint64_t stream_id;
2268 uint64_t net_seq_num;
2269 uint32_t data_size;
2270 struct relay_session *session;
2271
2272 assert(conn);
2273
2274 ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr,
2275 sizeof(struct lttcomm_relayd_data_hdr), 0);
2276 if (ret <= 0) {
2277 if (ret == 0) {
2278 /* Orderly shutdown. Not necessary to print an error. */
2279 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2280 } else {
2281 ERR("Unable to receive data header on sock %d", conn->sock->fd);
2282 }
2283 ret = -1;
2284 goto end;
2285 }
2286
2287 stream_id = be64toh(data_hdr.stream_id);
2288
2289 rcu_read_lock();
2290 stream = stream_find_by_id(relay_streams_ht, stream_id);
2291 if (!stream) {
2292 ret = -1;
2293 goto end_rcu_unlock;
2294 }
2295
2296 session = session_find_by_id(conn->sessions_ht, stream->session_id);
2297 assert(session);
2298
2299 data_size = be32toh(data_hdr.data_size);
2300 if (data_buffer_size < data_size) {
2301 char *tmp_data_ptr;
2302
2303 tmp_data_ptr = realloc(data_buffer, data_size);
2304 if (!tmp_data_ptr) {
2305 ERR("Allocating data buffer");
2306 free(data_buffer);
2307 ret = -1;
2308 goto end_rcu_unlock;
2309 }
2310 data_buffer = tmp_data_ptr;
2311 data_buffer_size = data_size;
2312 }
2313 memset(data_buffer, 0, data_size);
2314
2315 net_seq_num = be64toh(data_hdr.net_seq_num);
2316
2317 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2318 data_size, stream_id, net_seq_num);
2319 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
2320 if (ret <= 0) {
2321 if (ret == 0) {
2322 /* Orderly shutdown. Not necessary to print an error. */
2323 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2324 }
2325 ret = -1;
2326 goto end_rcu_unlock;
2327 }
2328
2329 /* Check if a rotation is needed. */
2330 if (stream->tracefile_size > 0 &&
2331 (stream->tracefile_size_current + data_size) >
2332 stream->tracefile_size) {
2333 struct relay_viewer_stream *vstream;
2334 uint64_t new_id;
2335
2336 new_id = (stream->tracefile_count_current + 1) %
2337 stream->tracefile_count;
2338 /*
2339 * When we wrap-around back to 0, we start overwriting old
2340 * trace data.
2341 */
2342 if (!stream->tracefile_overwrite && new_id == 0) {
2343 stream->tracefile_overwrite = 1;
2344 }
2345 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
2346 if (stream->tracefile_overwrite) {
2347 stream->oldest_tracefile_id =
2348 (stream->oldest_tracefile_id + 1) %
2349 stream->tracefile_count;
2350 }
2351 vstream = viewer_stream_find_by_id(stream->stream_handle);
2352 if (vstream) {
2353 /*
2354 * The viewer is reading a file about to be
2355 * overwritten. Close the FDs it is
2356 * currently using and let it handle the fault.
2357 */
2358 if (vstream->tracefile_count_current == new_id) {
2359 pthread_mutex_lock(&vstream->overwrite_lock);
2360 vstream->abort_flag = 1;
2361 pthread_mutex_unlock(&vstream->overwrite_lock);
2362 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2363 stream->channel_name, new_id);
2364 } else if (vstream->tracefile_count_current ==
2365 stream->tracefile_count_current) {
2366 /*
2367 * The reader and writer were in the
2368 * same trace file, inform the viewer
2369 * that no new index will ever be added
2370 * to this file.
2371 */
2372 vstream->close_write_flag = 1;
2373 }
2374 }
2375 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2376 stream->tracefile_size, stream->tracefile_count,
2377 relayd_uid, relayd_gid, stream->fd,
2378 &(stream->tracefile_count_current), &stream->fd);
2379 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
2380 if (ret < 0) {
2381 ERR("Rotating stream output file");
2382 goto end_rcu_unlock;
2383 }
2384 /* Reset current size because we just perform a stream rotation. */
2385 stream->tracefile_size_current = 0;
2386 rotate_index = 1;
2387 }
2388
2389 /*
2390 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2391 * index are NOT supported.
2392 */
2393 if (session->minor >= 4 && !session->snapshot) {
2394 ret = handle_index_data(stream, net_seq_num, rotate_index);
2395 if (ret < 0) {
2396 goto end_rcu_unlock;
2397 }
2398 }
2399
2400 /* Write data to stream output fd. */
2401 size_ret = lttng_write(stream->fd, data_buffer, data_size);
2402 if (size_ret < data_size) {
2403 ERR("Relay error writing data to file");
2404 ret = -1;
2405 goto end_rcu_unlock;
2406 }
2407
2408 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2409 ret, stream->stream_handle);
2410
2411 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2412 if (ret < 0) {
2413 goto end_rcu_unlock;
2414 }
2415 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2416
2417 stream->prev_seq = net_seq_num;
2418
2419 try_close_stream(session, stream);
2420
2421 end_rcu_unlock:
2422 rcu_read_unlock();
2423 end:
2424 return ret;
2425 }
2426
2427 static
2428 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2429 {
2430 int ret;
2431
2432 assert(events);
2433
2434 (void) lttng_poll_del(events, pollfd);
2435
2436 ret = close(pollfd);
2437 if (ret < 0) {
2438 ERR("Closing pollfd %d", pollfd);
2439 }
2440 }
2441
2442 static void destroy_connection(struct lttng_ht *relay_connections_ht,
2443 struct relay_connection *conn)
2444 {
2445 assert(relay_connections_ht);
2446 assert(conn);
2447
2448 connection_delete(relay_connections_ht, conn);
2449
2450 /* For the control socket, we try to destroy the session. */
2451 if (conn->type == RELAY_CONTROL && conn->session) {
2452 destroy_session(conn->session, conn->sessions_ht);
2453 }
2454
2455 connection_destroy(conn);
2456 }
2457
2458 /*
2459 * This thread does the actual work
2460 */
2461 static
2462 void *relay_thread_worker(void *data)
2463 {
2464 int ret, err = -1, last_seen_data_fd = -1;
2465 uint32_t nb_fd;
2466 struct relay_connection *conn;
2467 struct lttng_poll_event events;
2468 struct lttng_ht *relay_connections_ht;
2469 struct lttng_ht_iter iter;
2470 struct lttcomm_relayd_hdr recv_hdr;
2471 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2472 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2473 struct relay_index *index;
2474
2475 DBG("[thread] Relay worker started");
2476
2477 rcu_register_thread();
2478
2479 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2480
2481 if (testpoint(relayd_thread_worker)) {
2482 goto error_testpoint;
2483 }
2484
2485 health_code_update();
2486
2487 /* table of connections indexed on socket */
2488 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2489 if (!relay_connections_ht) {
2490 goto relay_connections_ht_error;
2491 }
2492
2493 /* Tables of received indexes indexed by index handle and net_seq_num. */
2494 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2495 if (!indexes_ht) {
2496 goto indexes_ht_error;
2497 }
2498
2499 ret = create_thread_poll_set(&events, 2);
2500 if (ret < 0) {
2501 goto error_poll_create;
2502 }
2503
2504 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2505 if (ret < 0) {
2506 goto error;
2507 }
2508
2509 restart:
2510 while (1) {
2511 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2512
2513 health_code_update();
2514
2515 /* Infinite blocking call, waiting for transmission */
2516 DBG3("Relayd worker thread polling...");
2517 health_poll_entry();
2518 ret = lttng_poll_wait(&events, -1);
2519 health_poll_exit();
2520 if (ret < 0) {
2521 /*
2522 * Restart interrupted system call.
2523 */
2524 if (errno == EINTR) {
2525 goto restart;
2526 }
2527 goto error;
2528 }
2529
2530 nb_fd = ret;
2531
2532 /*
2533 * Process control. The control connection is prioritised so we don't
2534 * starve it with high throughout put tracing data on the data
2535 * connection.
2536 */
2537 for (i = 0; i < nb_fd; i++) {
2538 /* Fetch once the poll data */
2539 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2540 int pollfd = LTTNG_POLL_GETFD(&events, i);
2541
2542 health_code_update();
2543
2544 if (!revents) {
2545 /* No activity for this FD (poll implementation). */
2546 continue;
2547 }
2548
2549 /* Thread quit pipe has been closed. Killing thread. */
2550 ret = check_thread_quit_pipe(pollfd, revents);
2551 if (ret) {
2552 err = 0;
2553 goto exit;
2554 }
2555
2556 /* Inspect the relay conn pipe for new connection */
2557 if (pollfd == relay_conn_pipe[0]) {
2558 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2559 ERR("Relay connection pipe error");
2560 goto error;
2561 } else if (revents & LPOLLIN) {
2562 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
2563 if (ret < 0) {
2564 goto error;
2565 }
2566 conn->sessions_ht = sessions_ht;
2567 connection_init(conn);
2568 lttng_poll_add(&events, conn->sock->fd,
2569 LPOLLIN | LPOLLRDHUP);
2570 rcu_read_lock();
2571 lttng_ht_add_unique_ulong(relay_connections_ht,
2572 &conn->sock_n);
2573 rcu_read_unlock();
2574 DBG("Connection socket %d added", conn->sock->fd);
2575 }
2576 } else {
2577 rcu_read_lock();
2578 conn = connection_find_by_sock(relay_connections_ht, pollfd);
2579 /* If not found, there is a synchronization issue. */
2580 assert(conn);
2581
2582 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2583 cleanup_connection_pollfd(&events, pollfd);
2584 destroy_connection(relay_connections_ht, conn);
2585 if (last_seen_data_fd == pollfd) {
2586 last_seen_data_fd = last_notdel_data_fd;
2587 }
2588 } else if (revents & LPOLLIN) {
2589 if (conn->type == RELAY_CONTROL) {
2590 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2591 sizeof(recv_hdr), 0);
2592 if (ret <= 0) {
2593 /* Connection closed */
2594 cleanup_connection_pollfd(&events, pollfd);
2595 destroy_connection(relay_connections_ht, conn);
2596 DBG("Control connection closed with %d", pollfd);
2597 } else {
2598 ret = relay_process_control(&recv_hdr, conn);
2599 if (ret < 0) {
2600 /* Clear the session on error. */
2601 cleanup_connection_pollfd(&events, pollfd);
2602 destroy_connection(relay_connections_ht, conn);
2603 DBG("Connection closed with %d", pollfd);
2604 }
2605 seen_control = 1;
2606 }
2607 } else {
2608 /*
2609 * Flag the last seen data fd not deleted. It will be
2610 * used as the last seen fd if any fd gets deleted in
2611 * this first loop.
2612 */
2613 last_notdel_data_fd = pollfd;
2614 }
2615 } else {
2616 ERR("Unknown poll events %u for sock %d", revents, pollfd);
2617 }
2618 rcu_read_unlock();
2619 }
2620 }
2621
2622 /*
2623 * The last loop handled a control request, go back to poll to make
2624 * sure we prioritise the control socket.
2625 */
2626 if (seen_control) {
2627 continue;
2628 }
2629
2630 if (last_seen_data_fd >= 0) {
2631 for (i = 0; i < nb_fd; i++) {
2632 int pollfd = LTTNG_POLL_GETFD(&events, i);
2633
2634 health_code_update();
2635
2636 if (last_seen_data_fd == pollfd) {
2637 idx = i;
2638 break;
2639 }
2640 }
2641 }
2642
2643 /* Process data connection. */
2644 for (i = idx + 1; i < nb_fd; i++) {
2645 /* Fetch the poll data. */
2646 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2647 int pollfd = LTTNG_POLL_GETFD(&events, i);
2648
2649 health_code_update();
2650
2651 if (!revents) {
2652 /* No activity for this FD (poll implementation). */
2653 continue;
2654 }
2655
2656 /* Skip the command pipe. It's handled in the first loop. */
2657 if (pollfd == relay_conn_pipe[0]) {
2658 continue;
2659 }
2660
2661 rcu_read_lock();
2662 conn = connection_find_by_sock(relay_connections_ht, pollfd);
2663 if (!conn) {
2664 /* Skip it. Might be removed before. */
2665 rcu_read_unlock();
2666 continue;
2667 }
2668
2669 if (revents & LPOLLIN) {
2670 if (conn->type != RELAY_DATA) {
2671 rcu_read_unlock();
2672 continue;
2673 }
2674
2675 ret = relay_process_data(conn);
2676 /* Connection closed */
2677 if (ret < 0) {
2678 cleanup_connection_pollfd(&events, pollfd);
2679 destroy_connection(relay_connections_ht, conn);
2680 DBG("Data connection closed with %d", pollfd);
2681 /*
2682 * Every goto restart call sets the last seen fd where
2683 * here we don't really care since we gracefully
2684 * continue the loop after the connection is deleted.
2685 */
2686 } else {
2687 /* Keep last seen port. */
2688 last_seen_data_fd = pollfd;
2689 rcu_read_unlock();
2690 goto restart;
2691 }
2692 }
2693 rcu_read_unlock();
2694 }
2695 last_seen_data_fd = -1;
2696 }
2697
2698 /* Normal exit, no error */
2699 ret = 0;
2700
2701 exit:
2702 error:
2703 lttng_poll_clean(&events);
2704
2705 /* Cleanup reamaining connection object. */
2706 rcu_read_lock();
2707 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2708 sock_n.node) {
2709 health_code_update();
2710 destroy_connection(relay_connections_ht, conn);
2711 }
2712 rcu_read_unlock();
2713 error_poll_create:
2714 rcu_read_lock();
2715 cds_lfht_for_each_entry(indexes_ht->ht, &iter.iter, index,
2716 index_n.node) {
2717 health_code_update();
2718 relay_index_delete(index);
2719 relay_index_free_safe(index);
2720 }
2721 rcu_read_unlock();
2722 lttng_ht_destroy(indexes_ht);
2723 indexes_ht_error:
2724 lttng_ht_destroy(relay_connections_ht);
2725 relay_connections_ht_error:
2726 /* Close relay conn pipes */
2727 utils_close_pipe(relay_conn_pipe);
2728 if (err) {
2729 DBG("Thread exited with error");
2730 }
2731 DBG("Worker thread cleanup complete");
2732 free(data_buffer);
2733 error_testpoint:
2734 if (err) {
2735 health_error();
2736 ERR("Health error occurred in %s", __func__);
2737 }
2738 health_unregister(health_relayd);
2739 rcu_unregister_thread();
2740 lttng_relay_stop_threads();
2741 return NULL;
2742 }
2743
2744 /*
2745 * Create the relay command pipe to wake thread_manage_apps.
2746 * Closed in cleanup().
2747 */
2748 static int create_relay_conn_pipe(void)
2749 {
2750 int ret;
2751
2752 ret = utils_create_pipe_cloexec(relay_conn_pipe);
2753
2754 return ret;
2755 }
2756
2757 /*
2758 * main
2759 */
2760 int main(int argc, char **argv)
2761 {
2762 int ret = 0, retval = 0;
2763 void *status;
2764 struct relay_local_data *relay_ctx = NULL;
2765
2766 /* Parse arguments */
2767 progname = argv[0];
2768 if (set_options(argc, argv)) {
2769 retval = -1;
2770 goto exit_options;
2771 }
2772
2773 if (set_signal_handler()) {
2774 retval = -1;
2775 goto exit_options;
2776 }
2777
2778 /* Try to create directory if -o, --output is specified. */
2779 if (opt_output_path) {
2780 if (*opt_output_path != '/') {
2781 ERR("Please specify an absolute path for -o, --output PATH");
2782 retval = -1;
2783 goto exit_options;
2784 }
2785
2786 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2787 if (ret < 0) {
2788 ERR("Unable to create %s", opt_output_path);
2789 retval = -1;
2790 goto exit_options;
2791 }
2792 }
2793
2794 /* Daemonize */
2795 if (opt_daemon || opt_background) {
2796 int i;
2797
2798 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2799 !opt_background);
2800 if (ret < 0) {
2801 retval = -1;
2802 goto exit_options;
2803 }
2804
2805 /*
2806 * We are in the child. Make sure all other file
2807 * descriptors are closed, in case we are called with
2808 * more opened file descriptors than the standard ones.
2809 */
2810 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2811 (void) close(i);
2812 }
2813 }
2814
2815
2816 /* Initialize thread health monitoring */
2817 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2818 if (!health_relayd) {
2819 PERROR("health_app_create error");
2820 retval = -1;
2821 goto exit_health_app_create;
2822 }
2823
2824 /* Create thread quit pipe */
2825 if (init_thread_quit_pipe()) {
2826 retval = -1;
2827 goto exit_init_data;
2828 }
2829
2830 /* We need those values for the file/dir creation. */
2831 relayd_uid = getuid();
2832 relayd_gid = getgid();
2833
2834 /* Check if daemon is UID = 0 */
2835 if (relayd_uid == 0) {
2836 if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) {
2837 ERR("Need to be root to use ports < 1024");
2838 retval = -1;
2839 goto exit_init_data;
2840 }
2841 }
2842
2843 /* Setup the thread apps communication pipe. */
2844 if (create_relay_conn_pipe()) {
2845 retval = -1;
2846 goto exit_init_data;
2847 }
2848
2849 /* Init relay command queue. */
2850 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
2851
2852 /* Set up max poll set size */
2853 if (lttng_poll_set_max_size()) {
2854 retval = -1;
2855 goto exit_init_data;
2856 }
2857
2858 /* Initialize communication library */
2859 lttcomm_init();
2860 lttcomm_inet_init();
2861
2862 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2863 if (!relay_ctx) {
2864 PERROR("relay_ctx");
2865 retval = -1;
2866 goto exit_init_data;
2867 }
2868
2869 /* tables of sessions indexed by session ID */
2870 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2871 if (!relay_ctx->sessions_ht) {
2872 retval = -1;
2873 goto exit_init_data;
2874 }
2875
2876 /* tables of streams indexed by stream ID */
2877 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2878 if (!relay_streams_ht) {
2879 retval = -1;
2880 goto exit_init_data;
2881 }
2882
2883 /* tables of streams indexed by stream ID */
2884 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2885 if (!viewer_streams_ht) {
2886 retval = -1;
2887 goto exit_init_data;
2888 }
2889
2890 ret = utils_create_pipe(health_quit_pipe);
2891 if (ret) {
2892 retval = -1;
2893 goto exit_health_quit_pipe;
2894 }
2895
2896 /* Create thread to manage the client socket */
2897 ret = pthread_create(&health_thread, NULL,
2898 thread_manage_health, (void *) NULL);
2899 if (ret) {
2900 errno = ret;
2901 PERROR("pthread_create health");
2902 retval = -1;
2903 goto exit_health_thread;
2904 }
2905
2906 /* Setup the dispatcher thread */
2907 ret = pthread_create(&dispatcher_thread, NULL,
2908 relay_thread_dispatcher, (void *) NULL);
2909 if (ret) {
2910 errno = ret;
2911 PERROR("pthread_create dispatcher");
2912 retval = -1;
2913 goto exit_dispatcher_thread;
2914 }
2915
2916 /* Setup the worker thread */
2917 ret = pthread_create(&worker_thread, NULL,
2918 relay_thread_worker, (void *) relay_ctx);
2919 if (ret) {
2920 errno = ret;
2921 PERROR("pthread_create worker");
2922 retval = -1;
2923 goto exit_worker_thread;
2924 }
2925
2926 /* Setup the listener thread */
2927 ret = pthread_create(&listener_thread, NULL,
2928 relay_thread_listener, (void *) NULL);
2929 if (ret) {
2930 errno = ret;
2931 PERROR("pthread_create listener");
2932 retval = -1;
2933 goto exit_listener_thread;
2934 }
2935
2936 ret = relayd_live_create(live_uri, relay_ctx);
2937 if (ret) {
2938 ERR("Starting live viewer threads");
2939 retval = -1;
2940 goto exit_live;
2941 }
2942
2943 /*
2944 * This is where we start awaiting program completion (e.g. through
2945 * signal that asks threads to teardown).
2946 */
2947
2948 ret = relayd_live_join();
2949 if (ret) {
2950 retval = -1;
2951 }
2952 exit_live:
2953
2954 ret = pthread_join(listener_thread, &status);
2955 if (ret) {
2956 errno = ret;
2957 PERROR("pthread_join listener_thread");
2958 retval = -1;
2959 }
2960
2961 exit_listener_thread:
2962 ret = pthread_join(worker_thread, &status);
2963 if (ret) {
2964 errno = ret;
2965 PERROR("pthread_join worker_thread");
2966 retval = -1;
2967 }
2968
2969 exit_worker_thread:
2970 ret = pthread_join(dispatcher_thread, &status);
2971 if (ret) {
2972 errno = ret;
2973 PERROR("pthread_join dispatcher_thread");
2974 retval = -1;
2975 }
2976 exit_dispatcher_thread:
2977
2978 ret = pthread_join(health_thread, &status);
2979 if (ret) {
2980 errno = ret;
2981 PERROR("pthread_join health_thread");
2982 retval = -1;
2983 }
2984 exit_health_thread:
2985
2986 utils_close_pipe(health_quit_pipe);
2987 exit_health_quit_pipe:
2988
2989 exit_init_data:
2990 health_app_destroy(health_relayd);
2991 exit_health_app_create:
2992 exit_options:
2993 relayd_cleanup(relay_ctx);
2994
2995 if (!retval) {
2996 exit(EXIT_SUCCESS);
2997 } else {
2998 exit(EXIT_FAILURE);
2999 }
3000 }
This page took 0.163152 seconds and 4 git commands to generate.