Fix: race with the viewer and readiness of streams
[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 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <pthread.h>
25 #include <signal.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <inttypes.h>
37 #include <urcu/futex.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40 #include <fcntl.h>
41 #include <config.h>
42
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/defaults.h>
48 #include <common/futex.h>
49 #include <common/sessiond-comm/sessiond-comm.h>
50 #include <common/sessiond-comm/inet.h>
51 #include <common/sessiond-comm/relayd.h>
52 #include <common/uri.h>
53 #include <common/utils.h>
54 #include <common/config/config.h>
55
56 #include "cmd.h"
57 #include "ctf-trace.h"
58 #include "index.h"
59 #include "utils.h"
60 #include "lttng-relayd.h"
61 #include "live.h"
62 #include "health-relayd.h"
63
64 /* command line options */
65 char *opt_output_path;
66 static int opt_daemon;
67 static struct lttng_uri *control_uri;
68 static struct lttng_uri *data_uri;
69 static struct lttng_uri *live_uri;
70
71 const char *progname;
72
73 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
74 static int tracing_group_name_override;
75
76 const char * const config_section_name = "relayd";
77
78 /*
79 * Quit pipe for all threads. This permits a single cancellation point
80 * for all threads when receiving an event on the pipe.
81 */
82 static int thread_quit_pipe[2] = { -1, -1 };
83
84 /*
85 * This pipe is used to inform the worker thread that a command is queued and
86 * ready to be processed.
87 */
88 static int relay_cmd_pipe[2] = { -1, -1 };
89
90 /* Shared between threads */
91 static int dispatch_thread_exit;
92
93 static pthread_t listener_thread;
94 static pthread_t dispatcher_thread;
95 static pthread_t worker_thread;
96 static pthread_t health_thread;
97
98 static uint64_t last_relay_stream_id;
99 static uint64_t last_relay_session_id;
100
101 /*
102 * Relay command queue.
103 *
104 * The relay_thread_listener and relay_thread_dispatcher communicate with this
105 * queue.
106 */
107 static struct relay_cmd_queue relay_cmd_queue;
108
109 /* buffer allocated at startup, used to store the trace data */
110 static char *data_buffer;
111 static unsigned int data_buffer_size;
112
113 /* We need those values for the file/dir creation. */
114 static uid_t relayd_uid;
115 static gid_t relayd_gid;
116
117 /* Global relay stream hash table. */
118 struct lttng_ht *relay_streams_ht;
119
120 /* Global relay viewer stream hash table. */
121 struct lttng_ht *viewer_streams_ht;
122
123 /* Global hash table that stores relay index object. */
124 struct lttng_ht *indexes_ht;
125
126 /* Relayd health monitoring */
127 struct health_app *health_relayd;
128
129 static struct option long_options[] = {
130 { "control-port", 1, 0, 'C', },
131 { "data-port", 1, 0, 'D', },
132 { "live-port", 1, 0, 'L', },
133 { "daemonize", 0, 0, 'd', },
134 { "group", 1, 0, 'g', },
135 { "help", 0, 0, 'h', },
136 { "output", 1, 0, 'o', },
137 { "verbose", 0, 0, 'v', },
138 { "config", 1, 0, 'f' },
139 { NULL, 0, 0, 0, },
140 };
141
142 static const char *config_ignore_options[] = { "help", "config" };
143
144 /*
145 * usage function on stderr
146 */
147 static
148 void usage(void)
149 {
150 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
151 fprintf(stderr, " -h, --help Display this usage.\n");
152 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
153 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
154 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
155 fprintf(stderr, " -L, --live-port URL Live view port listening.\n");
156 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
157 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
158 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
159 fprintf(stderr, " -f --config Load daemon configuration file\n");
160 }
161
162 /*
163 * Take an option from the getopt output and set it in the right variable to be
164 * used later.
165 *
166 * Return 0 on success else a negative value.
167 */
168 static
169 int set_option(int opt, const char *arg, const char *optname)
170 {
171 int ret;
172
173 switch (opt) {
174 case 0:
175 fprintf(stderr, "option %s", optname);
176 if (arg) {
177 fprintf(stderr, " with arg %s\n", arg);
178 }
179 break;
180 case 'C':
181 ret = uri_parse(arg, &control_uri);
182 if (ret < 0) {
183 ERR("Invalid control URI specified");
184 goto end;
185 }
186 if (control_uri->port == 0) {
187 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
188 }
189 break;
190 case 'D':
191 ret = uri_parse(arg, &data_uri);
192 if (ret < 0) {
193 ERR("Invalid data URI specified");
194 goto end;
195 }
196 if (data_uri->port == 0) {
197 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
198 }
199 break;
200 case 'L':
201 ret = uri_parse(arg, &live_uri);
202 if (ret < 0) {
203 ERR("Invalid live URI specified");
204 goto end;
205 }
206 if (live_uri->port == 0) {
207 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
208 }
209 break;
210 case 'd':
211 opt_daemon = 1;
212 break;
213 case 'g':
214 tracing_group_name = strdup(arg);
215 tracing_group_name_override = 1;
216 break;
217 case 'h':
218 usage();
219 exit(EXIT_FAILURE);
220 case 'o':
221 ret = asprintf(&opt_output_path, "%s", arg);
222 if (ret < 0) {
223 ret = -errno;
224 PERROR("asprintf opt_output_path");
225 goto end;
226 }
227 break;
228 case 'v':
229 /* Verbose level can increase using multiple -v */
230 if (arg) {
231 lttng_opt_verbose = config_parse_value(arg);
232 } else {
233 lttng_opt_verbose += 1;
234 }
235 break;
236 default:
237 /* Unknown option or other error.
238 * Error is printed by getopt, just return */
239 ret = -1;
240 goto end;
241 }
242
243 /* All good. */
244 ret = 0;
245
246 end:
247 return ret;
248 }
249
250 /*
251 * config_entry_handler_cb used to handle options read from a config file.
252 * See config_entry_handler_cb comment in common/config/config.h for the
253 * return value conventions.
254 */
255 static
256 int config_entry_handler(const struct config_entry *entry, void *unused)
257 {
258 int ret = 0, i;
259
260 if (!entry || !entry->name || !entry->value) {
261 ret = -EINVAL;
262 goto end;
263 }
264
265 /* Check if the option is to be ignored */
266 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
267 if (!strcmp(entry->name, config_ignore_options[i])) {
268 goto end;
269 }
270 }
271
272 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
273 /* Ignore if entry name is not fully matched. */
274 if (strcmp(entry->name, long_options[i].name)) {
275 continue;
276 }
277
278 /*
279 * If the option takes no argument on the command line, we have to
280 * check if the value is "true". We support non-zero numeric values,
281 * true, on and yes.
282 */
283 if (!long_options[i].has_arg) {
284 ret = config_parse_value(entry->value);
285 if (ret <= 0) {
286 if (ret) {
287 WARN("Invalid configuration value \"%s\" for option %s",
288 entry->value, entry->name);
289 }
290 /* False, skip boolean config option. */
291 goto end;
292 }
293 }
294
295 ret = set_option(long_options[i].val, entry->value, entry->name);
296 goto end;
297 }
298
299 WARN("Unrecognized option \"%s\" in daemon configuration file.",
300 entry->name);
301
302 end:
303 return ret;
304 }
305
306 static
307 int set_options(int argc, char **argv)
308 {
309 int c, ret = 0, option_index = 0;
310 int orig_optopt = optopt, orig_optind = optind;
311 char *default_address, *optstring;
312 const char *config_path = NULL;
313
314 optstring = utils_generate_optstring(long_options,
315 sizeof(long_options) / sizeof(struct option));
316 if (!optstring) {
317 ret = -ENOMEM;
318 goto exit;
319 }
320
321 /* Check for the --config option */
322
323 while ((c = getopt_long(argc, argv, optstring, long_options,
324 &option_index)) != -1) {
325 if (c == '?') {
326 ret = -EINVAL;
327 goto exit;
328 } else if (c != 'f') {
329 continue;
330 }
331
332 config_path = utils_expand_path(optarg);
333 if (!config_path) {
334 ERR("Failed to resolve path: %s", optarg);
335 }
336 }
337
338 ret = config_get_section_entries(config_path, config_section_name,
339 config_entry_handler, NULL);
340 if (ret) {
341 if (ret > 0) {
342 ERR("Invalid configuration option at line %i", ret);
343 ret = -1;
344 }
345 goto exit;
346 }
347
348 /* Reset getopt's global state */
349 optopt = orig_optopt;
350 optind = orig_optind;
351 while (1) {
352 c = getopt_long(argc, argv, optstring, long_options, &option_index);
353 if (c == -1) {
354 break;
355 }
356
357 ret = set_option(c, optarg, long_options[option_index].name);
358 if (ret < 0) {
359 goto exit;
360 }
361 }
362
363 /* assign default values */
364 if (control_uri == NULL) {
365 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
366 DEFAULT_NETWORK_CONTROL_PORT);
367 if (ret < 0) {
368 PERROR("asprintf default data address");
369 goto exit;
370 }
371
372 ret = uri_parse(default_address, &control_uri);
373 free(default_address);
374 if (ret < 0) {
375 ERR("Invalid control URI specified");
376 goto exit;
377 }
378 }
379 if (data_uri == NULL) {
380 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
381 DEFAULT_NETWORK_DATA_PORT);
382 if (ret < 0) {
383 PERROR("asprintf default data address");
384 goto exit;
385 }
386
387 ret = uri_parse(default_address, &data_uri);
388 free(default_address);
389 if (ret < 0) {
390 ERR("Invalid data URI specified");
391 goto exit;
392 }
393 }
394 if (live_uri == NULL) {
395 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
396 DEFAULT_NETWORK_VIEWER_PORT);
397 if (ret < 0) {
398 PERROR("asprintf default viewer control address");
399 goto exit;
400 }
401
402 ret = uri_parse(default_address, &live_uri);
403 free(default_address);
404 if (ret < 0) {
405 ERR("Invalid viewer control URI specified");
406 goto exit;
407 }
408 }
409
410 exit:
411 free(optstring);
412 return ret;
413 }
414
415 /*
416 * Cleanup the daemon
417 */
418 static
419 void cleanup(void)
420 {
421 DBG("Cleaning up");
422
423 /* free the dynamically allocated opt_output_path */
424 free(opt_output_path);
425
426 /* Close thread quit pipes */
427 utils_close_pipe(thread_quit_pipe);
428
429 uri_free(control_uri);
430 uri_free(data_uri);
431 /* Live URI is freed in the live thread. */
432
433 if (tracing_group_name_override) {
434 free((void *) tracing_group_name);
435 }
436 }
437
438 /*
439 * Write to writable pipe used to notify a thread.
440 */
441 static
442 int notify_thread_pipe(int wpipe)
443 {
444 ssize_t ret;
445
446 ret = lttng_write(wpipe, "!", 1);
447 if (ret < 1) {
448 PERROR("write poll pipe");
449 }
450
451 return ret;
452 }
453
454 static void notify_health_quit_pipe(int *pipe)
455 {
456 ssize_t ret;
457
458 ret = lttng_write(pipe[1], "4", 1);
459 if (ret < 1) {
460 PERROR("write relay health quit");
461 }
462 }
463
464 /*
465 * Stop all threads by closing the thread quit pipe.
466 */
467 static
468 void stop_threads(void)
469 {
470 int ret;
471
472 /* Stopping all threads */
473 DBG("Terminating all threads");
474 ret = notify_thread_pipe(thread_quit_pipe[1]);
475 if (ret < 0) {
476 ERR("write error on thread quit pipe");
477 }
478
479 notify_health_quit_pipe(health_quit_pipe);
480
481 /* Dispatch thread */
482 CMM_STORE_SHARED(dispatch_thread_exit, 1);
483 futex_nto1_wake(&relay_cmd_queue.futex);
484 }
485
486 /*
487 * Signal handler for the daemon
488 *
489 * Simply stop all worker threads, leaving main() return gracefully after
490 * joining all threads and calling cleanup().
491 */
492 static
493 void sighandler(int sig)
494 {
495 switch (sig) {
496 case SIGPIPE:
497 DBG("SIGPIPE caught");
498 return;
499 case SIGINT:
500 DBG("SIGINT caught");
501 stop_threads();
502 break;
503 case SIGTERM:
504 DBG("SIGTERM caught");
505 stop_threads();
506 break;
507 default:
508 break;
509 }
510 }
511
512 /*
513 * Setup signal handler for :
514 * SIGINT, SIGTERM, SIGPIPE
515 */
516 static
517 int set_signal_handler(void)
518 {
519 int ret = 0;
520 struct sigaction sa;
521 sigset_t sigset;
522
523 if ((ret = sigemptyset(&sigset)) < 0) {
524 PERROR("sigemptyset");
525 return ret;
526 }
527
528 sa.sa_handler = sighandler;
529 sa.sa_mask = sigset;
530 sa.sa_flags = 0;
531 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
532 PERROR("sigaction");
533 return ret;
534 }
535
536 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
537 PERROR("sigaction");
538 return ret;
539 }
540
541 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
542 PERROR("sigaction");
543 return ret;
544 }
545
546 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
547
548 return ret;
549 }
550
551 /*
552 * Init thread quit pipe.
553 *
554 * Return -1 on error or 0 if all pipes are created.
555 */
556 static
557 int init_thread_quit_pipe(void)
558 {
559 int ret;
560
561 ret = utils_create_pipe_cloexec(thread_quit_pipe);
562
563 return ret;
564 }
565
566 /*
567 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
568 */
569 static
570 int create_thread_poll_set(struct lttng_poll_event *events, int size)
571 {
572 int ret;
573
574 if (events == NULL || size == 0) {
575 ret = -1;
576 goto error;
577 }
578
579 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
580 if (ret < 0) {
581 goto error;
582 }
583
584 /* Add quit pipe */
585 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
586 if (ret < 0) {
587 goto error;
588 }
589
590 return 0;
591
592 error:
593 return ret;
594 }
595
596 /*
597 * Check if the thread quit pipe was triggered.
598 *
599 * Return 1 if it was triggered else 0;
600 */
601 static
602 int check_thread_quit_pipe(int fd, uint32_t events)
603 {
604 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
605 return 1;
606 }
607
608 return 0;
609 }
610
611 /*
612 * Create and init socket from uri.
613 */
614 static
615 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
616 {
617 int ret;
618 struct lttcomm_sock *sock = NULL;
619
620 sock = lttcomm_alloc_sock_from_uri(uri);
621 if (sock == NULL) {
622 ERR("Allocating socket");
623 goto error;
624 }
625
626 ret = lttcomm_create_sock(sock);
627 if (ret < 0) {
628 goto error;
629 }
630 DBG("Listening on sock %d", sock->fd);
631
632 ret = sock->ops->bind(sock);
633 if (ret < 0) {
634 goto error;
635 }
636
637 ret = sock->ops->listen(sock, -1);
638 if (ret < 0) {
639 goto error;
640
641 }
642
643 return sock;
644
645 error:
646 if (sock) {
647 lttcomm_destroy_sock(sock);
648 }
649 return NULL;
650 }
651
652 /*
653 * Return nonzero if stream needs to be closed.
654 */
655 static
656 int close_stream_check(struct relay_stream *stream)
657 {
658
659 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
660 /*
661 * We are about to close the stream so set the data pending flag to 1
662 * which will make the end data pending command skip the stream which
663 * is now closed and ready. Note that after proceeding to a file close,
664 * the written file is ready for reading.
665 */
666 stream->data_pending_check_done = 1;
667 return 1;
668 }
669 return 0;
670 }
671
672 /*
673 * This thread manages the listening for new connections on the network
674 */
675 static
676 void *relay_thread_listener(void *data)
677 {
678 int i, ret, pollfd, err = -1;
679 int val = 1;
680 uint32_t revents, nb_fd;
681 struct lttng_poll_event events;
682 struct lttcomm_sock *control_sock, *data_sock;
683
684 DBG("[thread] Relay listener started");
685
686 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
687
688 health_code_update();
689
690 control_sock = relay_init_sock(control_uri);
691 if (!control_sock) {
692 goto error_sock_control;
693 }
694
695 data_sock = relay_init_sock(data_uri);
696 if (!data_sock) {
697 goto error_sock_relay;
698 }
699
700 /*
701 * Pass 3 as size here for the thread quit pipe, control and data socket.
702 */
703 ret = create_thread_poll_set(&events, 3);
704 if (ret < 0) {
705 goto error_create_poll;
706 }
707
708 /* Add the control socket */
709 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
710 if (ret < 0) {
711 goto error_poll_add;
712 }
713
714 /* Add the data socket */
715 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
716 if (ret < 0) {
717 goto error_poll_add;
718 }
719
720 while (1) {
721 health_code_update();
722
723 DBG("Listener accepting connections");
724
725 restart:
726 health_poll_entry();
727 ret = lttng_poll_wait(&events, -1);
728 health_poll_exit();
729 if (ret < 0) {
730 /*
731 * Restart interrupted system call.
732 */
733 if (errno == EINTR) {
734 goto restart;
735 }
736 goto error;
737 }
738
739 nb_fd = ret;
740
741 DBG("Relay new connection received");
742 for (i = 0; i < nb_fd; i++) {
743 health_code_update();
744
745 /* Fetch once the poll data */
746 revents = LTTNG_POLL_GETEV(&events, i);
747 pollfd = LTTNG_POLL_GETFD(&events, i);
748
749 /* Thread quit pipe has been closed. Killing thread. */
750 ret = check_thread_quit_pipe(pollfd, revents);
751 if (ret) {
752 err = 0;
753 goto exit;
754 }
755
756 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
757 ERR("socket poll error");
758 goto error;
759 } else if (revents & LPOLLIN) {
760 /*
761 * Get allocated in this thread,
762 * enqueued to a global queue, dequeued
763 * and freed in the worker thread.
764 */
765 struct relay_command *relay_cmd;
766 struct lttcomm_sock *newsock;
767
768 relay_cmd = zmalloc(sizeof(struct relay_command));
769 if (relay_cmd == NULL) {
770 PERROR("relay command zmalloc");
771 goto error;
772 }
773
774 if (pollfd == data_sock->fd) {
775 newsock = data_sock->ops->accept(data_sock);
776 if (!newsock) {
777 PERROR("accepting data sock");
778 free(relay_cmd);
779 goto error;
780 }
781 relay_cmd->type = RELAY_DATA;
782 DBG("Relay data connection accepted, socket %d", newsock->fd);
783 } else {
784 assert(pollfd == control_sock->fd);
785 newsock = control_sock->ops->accept(control_sock);
786 if (!newsock) {
787 PERROR("accepting control sock");
788 free(relay_cmd);
789 goto error;
790 }
791 relay_cmd->type = RELAY_CONTROL;
792 DBG("Relay control connection accepted, socket %d", newsock->fd);
793 }
794 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
795 &val, sizeof(int));
796 if (ret < 0) {
797 PERROR("setsockopt inet");
798 lttcomm_destroy_sock(newsock);
799 free(relay_cmd);
800 goto error;
801 }
802 relay_cmd->sock = newsock;
803 /*
804 * Lock free enqueue the request.
805 */
806 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
807
808 /*
809 * Wake the dispatch queue futex. Implicit memory
810 * barrier with the exchange in cds_wfq_enqueue.
811 */
812 futex_nto1_wake(&relay_cmd_queue.futex);
813 }
814 }
815 }
816
817 exit:
818 error:
819 error_poll_add:
820 lttng_poll_clean(&events);
821 error_create_poll:
822 if (data_sock->fd >= 0) {
823 ret = data_sock->ops->close(data_sock);
824 if (ret) {
825 PERROR("close");
826 }
827 }
828 lttcomm_destroy_sock(data_sock);
829 error_sock_relay:
830 if (control_sock->fd >= 0) {
831 ret = control_sock->ops->close(control_sock);
832 if (ret) {
833 PERROR("close");
834 }
835 }
836 lttcomm_destroy_sock(control_sock);
837 error_sock_control:
838 if (err) {
839 health_error();
840 ERR("Health error occurred in %s", __func__);
841 }
842 health_unregister(health_relayd);
843 DBG("Relay listener thread cleanup complete");
844 stop_threads();
845 return NULL;
846 }
847
848 /*
849 * This thread manages the dispatching of the requests to worker threads
850 */
851 static
852 void *relay_thread_dispatcher(void *data)
853 {
854 int err = -1;
855 ssize_t ret;
856 struct cds_wfq_node *node;
857 struct relay_command *relay_cmd = NULL;
858
859 DBG("[thread] Relay dispatcher started");
860
861 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
862
863 health_code_update();
864
865 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
866 health_code_update();
867
868 /* Atomically prepare the queue futex */
869 futex_nto1_prepare(&relay_cmd_queue.futex);
870
871 do {
872 health_code_update();
873
874 /* Dequeue commands */
875 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
876 if (node == NULL) {
877 DBG("Woken up but nothing in the relay command queue");
878 /* Continue thread execution */
879 break;
880 }
881
882 relay_cmd = caa_container_of(node, struct relay_command, node);
883 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
884
885 /*
886 * Inform worker thread of the new request. This
887 * call is blocking so we can be assured that the data will be read
888 * at some point in time or wait to the end of the world :)
889 */
890 ret = lttng_write(relay_cmd_pipe[1], relay_cmd,
891 sizeof(struct relay_command));
892 free(relay_cmd);
893 if (ret < sizeof(struct relay_command)) {
894 PERROR("write cmd pipe");
895 goto error;
896 }
897 } while (node != NULL);
898
899 /* Futex wait on queue. Blocking call on futex() */
900 health_poll_entry();
901 futex_nto1_wait(&relay_cmd_queue.futex);
902 health_poll_exit();
903 }
904
905 /* Normal exit, no error */
906 err = 0;
907
908 error:
909 if (err) {
910 health_error();
911 ERR("Health error occurred in %s", __func__);
912 }
913 health_unregister(health_relayd);
914 DBG("Dispatch thread dying");
915 stop_threads();
916 return NULL;
917 }
918
919 /*
920 * Get stream from stream id.
921 * Need to be called with RCU read-side lock held.
922 */
923 struct relay_stream *relay_stream_find_by_id(uint64_t stream_id)
924 {
925 struct lttng_ht_node_ulong *node;
926 struct lttng_ht_iter iter;
927 struct relay_stream *ret;
928
929 lttng_ht_lookup(relay_streams_ht,
930 (void *)((unsigned long) stream_id),
931 &iter);
932 node = lttng_ht_iter_get_node_ulong(&iter);
933 if (node == NULL) {
934 DBG("Relay stream %" PRIu64 " not found", stream_id);
935 ret = NULL;
936 goto end;
937 }
938
939 ret = caa_container_of(node, struct relay_stream, stream_n);
940
941 end:
942 return ret;
943 }
944
945 static
946 void deferred_free_stream(struct rcu_head *head)
947 {
948 struct relay_stream *stream =
949 caa_container_of(head, struct relay_stream, rcu_node);
950
951 free(stream->path_name);
952 free(stream->channel_name);
953 free(stream);
954 }
955
956 static
957 void deferred_free_session(struct rcu_head *head)
958 {
959 struct relay_session *session =
960 caa_container_of(head, struct relay_session, rcu_node);
961 free(session);
962 }
963
964 /*
965 * Close a given stream. The stream is freed using a call RCU.
966 *
967 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
968 * BEFORE the stream lock MUST be acquired.
969 */
970 static void destroy_stream(struct relay_stream *stream)
971 {
972 int delret;
973 struct relay_viewer_stream *vstream;
974 struct lttng_ht_iter iter;
975
976 assert(stream);
977
978 delret = close(stream->fd);
979 if (delret < 0) {
980 PERROR("close stream");
981 }
982
983 if (stream->index_fd >= 0) {
984 delret = close(stream->index_fd);
985 if (delret < 0) {
986 PERROR("close stream index_fd");
987 }
988 }
989
990 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
991 if (vstream) {
992 /*
993 * Set the last good value into the viewer stream. This is done
994 * right before the stream gets deleted from the hash table. The
995 * lookup failure on the live thread side of a stream indicates
996 * that the viewer stream index received value should be used.
997 */
998 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
999 vstream->total_index_received = stream->total_index_received;
1000 vstream->tracefile_count_last = stream->tracefile_count_current;
1001 vstream->close_write_flag = 1;
1002 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
1003 }
1004
1005 /* Cleanup index of that stream. */
1006 relay_index_destroy_by_stream_id(stream->stream_handle);
1007
1008 iter.iter.node = &stream->stream_n.node;
1009 delret = lttng_ht_del(relay_streams_ht, &iter);
1010 assert(!delret);
1011 iter.iter.node = &stream->ctf_trace_node.node;
1012 delret = lttng_ht_del(stream->ctf_traces_ht, &iter);
1013 assert(!delret);
1014
1015 if (stream->ctf_trace) {
1016 ctf_trace_try_destroy(stream->ctf_trace);
1017 }
1018
1019 call_rcu(&stream->rcu_node, deferred_free_stream);
1020 DBG("Closed tracefile %d from close stream", stream->fd);
1021 }
1022
1023 /*
1024 * relay_delete_session: Free all memory associated with a session and
1025 * close all the FDs
1026 */
1027 static
1028 void relay_delete_session(struct relay_command *cmd,
1029 struct lttng_ht *sessions_ht)
1030 {
1031 struct lttng_ht_iter iter;
1032 struct lttng_ht_node_ulong *node;
1033 struct relay_stream *stream;
1034 int ret;
1035
1036 if (!cmd->session) {
1037 return;
1038 }
1039
1040 DBG("Relay deleting session %" PRIu64, cmd->session->id);
1041
1042 rcu_read_lock();
1043 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
1044 node = lttng_ht_iter_get_node_ulong(&iter);
1045 if (!node) {
1046 continue;
1047 }
1048 stream = caa_container_of(node, struct relay_stream, stream_n);
1049 if (stream->session == cmd->session) {
1050 destroy_stream(stream);
1051 cmd->session->stream_count--;
1052 assert(cmd->session->stream_count >= 0);
1053 }
1054 }
1055
1056 /* Make this session not visible anymore. */
1057 iter.iter.node = &cmd->session->session_n.node;
1058 ret = lttng_ht_del(sessions_ht, &iter);
1059 assert(!ret);
1060 call_rcu(&cmd->session->rcu_node, deferred_free_session);
1061 rcu_read_unlock();
1062 }
1063
1064 /*
1065 * Copy index data from the control port to a given index object.
1066 */
1067 static void copy_index_control_data(struct relay_index *index,
1068 struct lttcomm_relayd_index *data)
1069 {
1070 assert(index);
1071 assert(data);
1072
1073 /*
1074 * The index on disk is encoded in big endian, so we don't need to convert
1075 * the data received on the network. The data_offset value is NEVER
1076 * modified here and is updated by the data thread.
1077 */
1078 index->index_data.packet_size = data->packet_size;
1079 index->index_data.content_size = data->content_size;
1080 index->index_data.timestamp_begin = data->timestamp_begin;
1081 index->index_data.timestamp_end = data->timestamp_end;
1082 index->index_data.events_discarded = data->events_discarded;
1083 index->index_data.stream_id = data->stream_id;
1084 }
1085
1086 /*
1087 * Handle the RELAYD_CREATE_SESSION command.
1088 *
1089 * On success, send back the session id or else return a negative value.
1090 */
1091 static
1092 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
1093 struct relay_command *cmd,
1094 struct lttng_ht *sessions_ht)
1095 {
1096 int ret = 0, send_ret;
1097 struct relay_session *session;
1098 struct lttcomm_relayd_status_session reply;
1099
1100 assert(recv_hdr);
1101 assert(cmd);
1102
1103 memset(&reply, 0, sizeof(reply));
1104
1105 session = zmalloc(sizeof(struct relay_session));
1106 if (session == NULL) {
1107 PERROR("relay session zmalloc");
1108 ret = -1;
1109 goto error;
1110 }
1111
1112 session->id = ++last_relay_session_id;
1113 session->sock = cmd->sock;
1114 session->minor = cmd->minor;
1115 session->major = cmd->major;
1116 cmd->session = session;
1117
1118 reply.session_id = htobe64(session->id);
1119
1120 switch (cmd->minor) {
1121 case 4: /* LTTng sessiond 2.4 */
1122 default:
1123 ret = cmd_create_session_2_4(cmd, session);
1124 break;
1125 }
1126
1127 lttng_ht_node_init_ulong(&session->session_n,
1128 (unsigned long) session->id);
1129 lttng_ht_add_unique_ulong(sessions_ht,
1130 &session->session_n);
1131
1132 DBG("Created session %" PRIu64, session->id);
1133
1134 error:
1135 if (ret < 0) {
1136 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1137 } else {
1138 reply.ret_code = htobe32(LTTNG_OK);
1139 }
1140
1141 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1142 if (send_ret < 0) {
1143 ERR("Relayd sending session id");
1144 ret = send_ret;
1145 }
1146
1147 return ret;
1148 }
1149
1150 /*
1151 * When we have received all the streams and the metadata for a channel,
1152 * we make them visible to the viewer threads.
1153 */
1154 static
1155 void set_viewer_ready_flag(struct relay_command *cmd)
1156 {
1157 struct relay_stream_recv_handle *node, *tmp_node;
1158
1159 cds_list_for_each_entry_safe(node, tmp_node, &cmd->recv_head, node) {
1160 struct relay_stream *stream;
1161
1162 rcu_read_lock();
1163 stream = relay_stream_find_by_id(node->id);
1164 if (!stream) {
1165 /*
1166 * Stream is most probably being cleaned up by the data thread thus
1167 * simply continue to the next one.
1168 */
1169 continue;
1170 }
1171
1172 /*
1173 * If any of the streams in the list doesn't have a ctf_trace assigned,
1174 * it means that we never received the metadata stream, so we have to
1175 * wait until it arrives to make the streams available to the viewer.
1176 */
1177 if (!stream->ctf_trace) {
1178 goto end;
1179 }
1180
1181 stream->viewer_ready = 1;
1182 rcu_read_unlock();
1183
1184 /* Clean stream handle node. */
1185 cds_list_del(&node->node);
1186 free(node);
1187 }
1188
1189 end:
1190 return;
1191 }
1192
1193 /*
1194 * Add a recv handle node to the connection recv list with the given stream
1195 * handle. A new node is allocated thus must be freed when the node is deleted
1196 * from the list.
1197 */
1198 static void queue_stream_handle(uint64_t handle, struct relay_command *cmd)
1199 {
1200 struct relay_stream_recv_handle *node;
1201
1202 assert(cmd);
1203
1204 node = zmalloc(sizeof(*node));
1205 if (!node) {
1206 PERROR("zmalloc queue stream handle");
1207 return;
1208 }
1209
1210 node->id = handle;
1211 cds_list_add(&node->node, &cmd->recv_head);
1212 }
1213
1214 /*
1215 * relay_add_stream: allocate a new stream for a session
1216 */
1217 static
1218 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1219 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1220 {
1221 struct relay_session *session = cmd->session;
1222 struct relay_stream *stream = NULL;
1223 struct lttcomm_relayd_status_stream reply;
1224 int ret, send_ret;
1225
1226 if (!session || cmd->version_check_done == 0) {
1227 ERR("Trying to add a stream before version check");
1228 ret = -1;
1229 goto end_no_session;
1230 }
1231
1232 stream = zmalloc(sizeof(struct relay_stream));
1233 if (stream == NULL) {
1234 PERROR("relay stream zmalloc");
1235 ret = -1;
1236 goto end_no_session;
1237 }
1238
1239 switch (cmd->minor) {
1240 case 1: /* LTTng sessiond 2.1 */
1241 ret = cmd_recv_stream_2_1(cmd, stream);
1242 break;
1243 case 2: /* LTTng sessiond 2.2 */
1244 default:
1245 ret = cmd_recv_stream_2_2(cmd, stream);
1246 break;
1247 }
1248 if (ret < 0) {
1249 goto err_free_stream;
1250 }
1251
1252 rcu_read_lock();
1253 stream->stream_handle = ++last_relay_stream_id;
1254 stream->prev_seq = -1ULL;
1255 stream->session = session;
1256 stream->index_fd = -1;
1257 stream->read_index_fd = -1;
1258 stream->ctf_trace = NULL;
1259 pthread_mutex_init(&stream->lock, NULL);
1260
1261 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
1262 if (ret < 0) {
1263 ERR("relay creating output directory");
1264 goto end;
1265 }
1266
1267 /*
1268 * No need to use run_as API here because whatever we receives, the relayd
1269 * uses its own credentials for the stream files.
1270 */
1271 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1272 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1273 if (ret < 0) {
1274 ERR("Create output file");
1275 goto end;
1276 }
1277 stream->fd = ret;
1278 if (stream->tracefile_size) {
1279 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1280 } else {
1281 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1282 }
1283
1284 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1285 stream->metadata_flag = 1;
1286 /*
1287 * When we receive a new metadata stream, we create a new
1288 * ctf_trace and we assign this ctf_trace to all streams with
1289 * the same path.
1290 *
1291 * If later on we receive a new stream for the same ctf_trace,
1292 * we copy the information from the first hit in the HT to the
1293 * new stream.
1294 */
1295 stream->ctf_trace = ctf_trace_create();
1296 if (!stream->ctf_trace) {
1297 ret = -1;
1298 goto end;
1299 }
1300 stream->ctf_trace->refcount++;
1301 stream->ctf_trace->metadata_stream = stream;
1302 }
1303 ctf_trace_assign(cmd->ctf_traces_ht, stream);
1304 stream->ctf_traces_ht = cmd->ctf_traces_ht;
1305
1306 /*
1307 * Add the stream handle in the recv list of the connection. Once the end
1308 * stream message is received, this list is emptied and streams are set
1309 * with the viewer ready flag.
1310 */
1311 queue_stream_handle(stream->stream_handle, cmd);
1312
1313 lttng_ht_node_init_ulong(&stream->stream_n,
1314 (unsigned long) stream->stream_handle);
1315 lttng_ht_add_unique_ulong(relay_streams_ht,
1316 &stream->stream_n);
1317
1318 lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name);
1319 lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node);
1320 session->stream_count++;
1321
1322 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1323 stream->stream_handle);
1324
1325 end:
1326 reply.handle = htobe64(stream->stream_handle);
1327 /* send the session id to the client or a negative return code on error */
1328 if (ret < 0) {
1329 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1330 /* stream was not properly added to the ht, so free it */
1331 free(stream);
1332 } else {
1333 reply.ret_code = htobe32(LTTNG_OK);
1334 }
1335
1336 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1337 sizeof(struct lttcomm_relayd_status_stream), 0);
1338 if (send_ret < 0) {
1339 ERR("Relay sending stream id");
1340 ret = send_ret;
1341 }
1342 rcu_read_unlock();
1343
1344 end_no_session:
1345 return ret;
1346
1347 err_free_stream:
1348 free(stream->path_name);
1349 free(stream->channel_name);
1350 free(stream);
1351 return ret;
1352 }
1353
1354 /*
1355 * relay_close_stream: close a specific stream
1356 */
1357 static
1358 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1359 struct relay_command *cmd)
1360 {
1361 int ret, send_ret;
1362 struct relay_session *session = cmd->session;
1363 struct lttcomm_relayd_close_stream stream_info;
1364 struct lttcomm_relayd_generic_reply reply;
1365 struct relay_stream *stream;
1366
1367 DBG("Close stream received");
1368
1369 if (!session || cmd->version_check_done == 0) {
1370 ERR("Trying to close a stream before version check");
1371 ret = -1;
1372 goto end_no_session;
1373 }
1374
1375 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1376 sizeof(struct lttcomm_relayd_close_stream), 0);
1377 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1378 if (ret == 0) {
1379 /* Orderly shutdown. Not necessary to print an error. */
1380 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1381 } else {
1382 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1383 }
1384 ret = -1;
1385 goto end_no_session;
1386 }
1387
1388 rcu_read_lock();
1389 stream = relay_stream_find_by_id(be64toh(stream_info.stream_id));
1390 if (!stream) {
1391 ret = -1;
1392 goto end_unlock;
1393 }
1394
1395 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1396 stream->close_flag = 1;
1397 session->stream_count--;
1398 assert(session->stream_count >= 0);
1399
1400 if (close_stream_check(stream)) {
1401 destroy_stream(stream);
1402 }
1403
1404 end_unlock:
1405 rcu_read_unlock();
1406
1407 if (ret < 0) {
1408 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1409 } else {
1410 reply.ret_code = htobe32(LTTNG_OK);
1411 }
1412 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1413 sizeof(struct lttcomm_relayd_generic_reply), 0);
1414 if (send_ret < 0) {
1415 ERR("Relay sending stream id");
1416 ret = send_ret;
1417 }
1418
1419 end_no_session:
1420 return ret;
1421 }
1422
1423 /*
1424 * relay_unknown_command: send -1 if received unknown command
1425 */
1426 static
1427 void relay_unknown_command(struct relay_command *cmd)
1428 {
1429 struct lttcomm_relayd_generic_reply reply;
1430 int ret;
1431
1432 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1433 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1434 sizeof(struct lttcomm_relayd_generic_reply), 0);
1435 if (ret < 0) {
1436 ERR("Relay sending unknown command");
1437 }
1438 }
1439
1440 /*
1441 * relay_start: send an acknowledgment to the client to tell if we are
1442 * ready to receive data. We are ready if a session is established.
1443 */
1444 static
1445 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1446 struct relay_command *cmd)
1447 {
1448 int ret = htobe32(LTTNG_OK);
1449 struct lttcomm_relayd_generic_reply reply;
1450 struct relay_session *session = cmd->session;
1451
1452 if (!session) {
1453 DBG("Trying to start the streaming without a session established");
1454 ret = htobe32(LTTNG_ERR_UNK);
1455 }
1456
1457 reply.ret_code = ret;
1458 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1459 sizeof(struct lttcomm_relayd_generic_reply), 0);
1460 if (ret < 0) {
1461 ERR("Relay sending start ack");
1462 }
1463
1464 return ret;
1465 }
1466
1467 /*
1468 * Append padding to the file pointed by the file descriptor fd.
1469 */
1470 static int write_padding_to_file(int fd, uint32_t size)
1471 {
1472 ssize_t ret = 0;
1473 char *zeros;
1474
1475 if (size == 0) {
1476 goto end;
1477 }
1478
1479 zeros = zmalloc(size);
1480 if (zeros == NULL) {
1481 PERROR("zmalloc zeros for padding");
1482 ret = -1;
1483 goto end;
1484 }
1485
1486 ret = lttng_write(fd, zeros, size);
1487 if (ret < size) {
1488 PERROR("write padding to file");
1489 }
1490
1491 free(zeros);
1492
1493 end:
1494 return ret;
1495 }
1496
1497 /*
1498 * relay_recv_metadata: receive the metada for the session.
1499 */
1500 static
1501 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1502 struct relay_command *cmd)
1503 {
1504 int ret = htobe32(LTTNG_OK);
1505 ssize_t size_ret;
1506 struct relay_session *session = cmd->session;
1507 struct lttcomm_relayd_metadata_payload *metadata_struct;
1508 struct relay_stream *metadata_stream;
1509 uint64_t data_size, payload_size;
1510
1511 if (!session) {
1512 ERR("Metadata sent before version check");
1513 ret = -1;
1514 goto end;
1515 }
1516
1517 data_size = payload_size = be64toh(recv_hdr->data_size);
1518 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1519 ERR("Incorrect data size");
1520 ret = -1;
1521 goto end;
1522 }
1523 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1524
1525 if (data_buffer_size < data_size) {
1526 /* In case the realloc fails, we can free the memory */
1527 char *tmp_data_ptr;
1528
1529 tmp_data_ptr = realloc(data_buffer, data_size);
1530 if (!tmp_data_ptr) {
1531 ERR("Allocating data buffer");
1532 free(data_buffer);
1533 ret = -1;
1534 goto end;
1535 }
1536 data_buffer = tmp_data_ptr;
1537 data_buffer_size = data_size;
1538 }
1539 memset(data_buffer, 0, data_size);
1540 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1541 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1542 if (ret < 0 || ret != data_size) {
1543 if (ret == 0) {
1544 /* Orderly shutdown. Not necessary to print an error. */
1545 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1546 } else {
1547 ERR("Relay didn't receive the whole metadata");
1548 }
1549 ret = -1;
1550 goto end;
1551 }
1552 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1553
1554 rcu_read_lock();
1555 metadata_stream = relay_stream_find_by_id(
1556 be64toh(metadata_struct->stream_id));
1557 if (!metadata_stream) {
1558 ret = -1;
1559 goto end_unlock;
1560 }
1561
1562 size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
1563 payload_size);
1564 if (size_ret < payload_size) {
1565 ERR("Relay error writing metadata on file");
1566 ret = -1;
1567 goto end_unlock;
1568 }
1569
1570 ret = write_padding_to_file(metadata_stream->fd,
1571 be32toh(metadata_struct->padding_size));
1572 if (ret < 0) {
1573 goto end_unlock;
1574 }
1575 metadata_stream->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_command *cmd, struct lttng_ht *sessions_ht)
1592 {
1593 int ret;
1594 struct lttcomm_relayd_version reply, msg;
1595
1596 assert(cmd);
1597
1598 cmd->version_check_done = 1;
1599
1600 /* Get version from the other side. */
1601 ret = cmd->sock->ops->recvmsg(cmd->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", cmd->sock->fd);
1606 } else {
1607 ERR("Relay failed to receive the version values.");
1608 }
1609 ret = -1;
1610 goto end;
1611 }
1612
1613 reply.major = RELAYD_VERSION_COMM_MAJOR;
1614 reply.minor = RELAYD_VERSION_COMM_MINOR;
1615
1616 /* Major versions must be the same */
1617 if (reply.major != be32toh(msg.major)) {
1618 DBG("Incompatible major versions (%u vs %u), deleting session",
1619 reply.major, be32toh(msg.major));
1620 relay_delete_session(cmd, sessions_ht);
1621 ret = 0;
1622 goto end;
1623 }
1624
1625 cmd->major = reply.major;
1626 /* We adapt to the lowest compatible version */
1627 if (reply.minor <= be32toh(msg.minor)) {
1628 cmd->minor = reply.minor;
1629 } else {
1630 cmd->minor = be32toh(msg.minor);
1631 }
1632
1633 reply.major = htobe32(reply.major);
1634 reply.minor = htobe32(reply.minor);
1635 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1636 sizeof(struct lttcomm_relayd_version), 0);
1637 if (ret < 0) {
1638 ERR("Relay sending version");
1639 }
1640
1641 DBG("Version check done using protocol %u.%u", cmd->major,
1642 cmd->minor);
1643
1644 end:
1645 return ret;
1646 }
1647
1648 /*
1649 * Check for data pending for a given stream id from the session daemon.
1650 */
1651 static
1652 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1653 struct relay_command *cmd)
1654 {
1655 struct relay_session *session = cmd->session;
1656 struct lttcomm_relayd_data_pending msg;
1657 struct lttcomm_relayd_generic_reply reply;
1658 struct relay_stream *stream;
1659 int ret;
1660 uint64_t last_net_seq_num, stream_id;
1661
1662 DBG("Data pending command received");
1663
1664 if (!session || cmd->version_check_done == 0) {
1665 ERR("Trying to check for data before version check");
1666 ret = -1;
1667 goto end_no_session;
1668 }
1669
1670 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1671 if (ret < sizeof(msg)) {
1672 if (ret == 0) {
1673 /* Orderly shutdown. Not necessary to print an error. */
1674 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1675 } else {
1676 ERR("Relay didn't receive valid data_pending struct size : %d",
1677 ret);
1678 }
1679 ret = -1;
1680 goto end_no_session;
1681 }
1682
1683 stream_id = be64toh(msg.stream_id);
1684 last_net_seq_num = be64toh(msg.last_net_seq_num);
1685
1686 rcu_read_lock();
1687 stream = relay_stream_find_by_id(stream_id);
1688 if (stream == NULL) {
1689 ret = -1;
1690 goto end_unlock;
1691 }
1692
1693 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1694 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1695 last_net_seq_num);
1696
1697 /* Avoid wrapping issue */
1698 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1699 /* Data has in fact been written and is NOT pending */
1700 ret = 0;
1701 } else {
1702 /* Data still being streamed thus pending */
1703 ret = 1;
1704 }
1705
1706 /* Pending check is now done. */
1707 stream->data_pending_check_done = 1;
1708
1709 end_unlock:
1710 rcu_read_unlock();
1711
1712 reply.ret_code = htobe32(ret);
1713 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1714 if (ret < 0) {
1715 ERR("Relay data pending ret code failed");
1716 }
1717
1718 end_no_session:
1719 return ret;
1720 }
1721
1722 /*
1723 * Wait for the control socket to reach a quiescent state.
1724 *
1725 * Note that for now, when receiving this command from the session daemon, this
1726 * means that every subsequent commands or data received on the control socket
1727 * has been handled. So, this is why we simply return OK here.
1728 */
1729 static
1730 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1731 struct relay_command *cmd)
1732 {
1733 int ret;
1734 uint64_t stream_id;
1735 struct relay_stream *stream;
1736 struct lttng_ht_iter iter;
1737 struct lttcomm_relayd_quiescent_control msg;
1738 struct lttcomm_relayd_generic_reply reply;
1739
1740 DBG("Checking quiescent state on control socket");
1741
1742 if (!cmd->session || cmd->version_check_done == 0) {
1743 ERR("Trying to check for data before version check");
1744 ret = -1;
1745 goto end_no_session;
1746 }
1747
1748 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1749 if (ret < sizeof(msg)) {
1750 if (ret == 0) {
1751 /* Orderly shutdown. Not necessary to print an error. */
1752 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1753 } else {
1754 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1755 ret);
1756 }
1757 ret = -1;
1758 goto end_no_session;
1759 }
1760
1761 stream_id = be64toh(msg.stream_id);
1762
1763 rcu_read_lock();
1764 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1765 stream_n.node) {
1766 if (stream->stream_handle == stream_id) {
1767 stream->data_pending_check_done = 1;
1768 DBG("Relay quiescent control pending flag set to %" PRIu64,
1769 stream_id);
1770 break;
1771 }
1772 }
1773 rcu_read_unlock();
1774
1775 reply.ret_code = htobe32(LTTNG_OK);
1776 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1777 if (ret < 0) {
1778 ERR("Relay data quiescent control ret code failed");
1779 }
1780
1781 end_no_session:
1782 return ret;
1783 }
1784
1785 /*
1786 * Initialize a data pending command. This means that a client is about to ask
1787 * for data pending for each stream he/she holds. Simply iterate over all
1788 * streams of a session and set the data_pending_check_done flag.
1789 *
1790 * This command returns to the client a LTTNG_OK code.
1791 */
1792 static
1793 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1794 struct relay_command *cmd)
1795 {
1796 int ret;
1797 struct lttng_ht_iter iter;
1798 struct lttcomm_relayd_begin_data_pending msg;
1799 struct lttcomm_relayd_generic_reply reply;
1800 struct relay_stream *stream;
1801 uint64_t session_id;
1802
1803 assert(recv_hdr);
1804 assert(cmd);
1805
1806 DBG("Init streams for data pending");
1807
1808 if (!cmd->session || cmd->version_check_done == 0) {
1809 ERR("Trying to check for data before version check");
1810 ret = -1;
1811 goto end_no_session;
1812 }
1813
1814 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1815 if (ret < sizeof(msg)) {
1816 if (ret == 0) {
1817 /* Orderly shutdown. Not necessary to print an error. */
1818 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1819 } else {
1820 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1821 ret);
1822 }
1823 ret = -1;
1824 goto end_no_session;
1825 }
1826
1827 session_id = be64toh(msg.session_id);
1828
1829 /*
1830 * Iterate over all streams to set the begin data pending flag. For now, the
1831 * streams are indexed by stream handle so we have to iterate over all
1832 * streams to find the one associated with the right session_id.
1833 */
1834 rcu_read_lock();
1835 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1836 stream_n.node) {
1837 if (stream->session->id == session_id) {
1838 stream->data_pending_check_done = 0;
1839 DBG("Set begin data pending flag to stream %" PRIu64,
1840 stream->stream_handle);
1841 }
1842 }
1843 rcu_read_unlock();
1844
1845 /* All good, send back reply. */
1846 reply.ret_code = htobe32(LTTNG_OK);
1847
1848 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1849 if (ret < 0) {
1850 ERR("Relay begin data pending send reply failed");
1851 }
1852
1853 end_no_session:
1854 return ret;
1855 }
1856
1857 /*
1858 * End data pending command. This will check, for a given session id, if each
1859 * stream associated with it has its data_pending_check_done flag set. If not,
1860 * this means that the client lost track of the stream but the data is still
1861 * being streamed on our side. In this case, we inform the client that data is
1862 * inflight.
1863 *
1864 * Return to the client if there is data in flight or not with a ret_code.
1865 */
1866 static
1867 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1868 struct relay_command *cmd)
1869 {
1870 int ret;
1871 struct lttng_ht_iter iter;
1872 struct lttcomm_relayd_end_data_pending msg;
1873 struct lttcomm_relayd_generic_reply reply;
1874 struct relay_stream *stream;
1875 uint64_t session_id;
1876 uint32_t is_data_inflight = 0;
1877
1878 assert(recv_hdr);
1879 assert(cmd);
1880
1881 DBG("End data pending command");
1882
1883 if (!cmd->session || cmd->version_check_done == 0) {
1884 ERR("Trying to check for data before version check");
1885 ret = -1;
1886 goto end_no_session;
1887 }
1888
1889 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1890 if (ret < sizeof(msg)) {
1891 if (ret == 0) {
1892 /* Orderly shutdown. Not necessary to print an error. */
1893 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1894 } else {
1895 ERR("Relay didn't receive valid end data_pending struct size: %d",
1896 ret);
1897 }
1898 ret = -1;
1899 goto end_no_session;
1900 }
1901
1902 session_id = be64toh(msg.session_id);
1903
1904 /* Iterate over all streams to see if the begin data pending flag is set. */
1905 rcu_read_lock();
1906 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1907 stream_n.node) {
1908 if (stream->session->id == session_id &&
1909 !stream->data_pending_check_done) {
1910 is_data_inflight = 1;
1911 DBG("Data is still in flight for stream %" PRIu64,
1912 stream->stream_handle);
1913 break;
1914 }
1915 }
1916 rcu_read_unlock();
1917
1918 /* All good, send back reply. */
1919 reply.ret_code = htobe32(is_data_inflight);
1920
1921 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1922 if (ret < 0) {
1923 ERR("Relay end data pending send reply failed");
1924 }
1925
1926 end_no_session:
1927 return ret;
1928 }
1929
1930 /*
1931 * Receive an index for a specific stream.
1932 *
1933 * Return 0 on success else a negative value.
1934 */
1935 static
1936 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1937 struct relay_command *cmd)
1938 {
1939 int ret, send_ret, index_created = 0;
1940 struct relay_session *session = cmd->session;
1941 struct lttcomm_relayd_index index_info;
1942 struct relay_index *index, *wr_index = NULL;
1943 struct lttcomm_relayd_generic_reply reply;
1944 struct relay_stream *stream;
1945 uint64_t net_seq_num;
1946
1947 assert(cmd);
1948
1949 DBG("Relay receiving index");
1950
1951 if (!session || cmd->version_check_done == 0) {
1952 ERR("Trying to close a stream before version check");
1953 ret = -1;
1954 goto end_no_session;
1955 }
1956
1957 ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info,
1958 sizeof(index_info), 0);
1959 if (ret < sizeof(index_info)) {
1960 if (ret == 0) {
1961 /* Orderly shutdown. Not necessary to print an error. */
1962 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1963 } else {
1964 ERR("Relay didn't receive valid index struct size : %d", ret);
1965 }
1966 ret = -1;
1967 goto end_no_session;
1968 }
1969
1970 net_seq_num = be64toh(index_info.net_seq_num);
1971
1972 rcu_read_lock();
1973 stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id));
1974 if (!stream) {
1975 ret = -1;
1976 goto end_rcu_unlock;
1977 }
1978
1979 /* Live beacon handling */
1980 if (index_info.packet_size == 0) {
1981 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1982
1983 /*
1984 * Only flag a stream inactive when it has already received data.
1985 */
1986 if (stream->total_index_received > 0) {
1987 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1988 }
1989 ret = 0;
1990 goto end_rcu_unlock;
1991 } else {
1992 stream->beacon_ts_end = -1ULL;
1993 }
1994
1995 index = relay_index_find(stream->stream_handle, net_seq_num);
1996 if (!index) {
1997 /* A successful creation will add the object to the HT. */
1998 index = relay_index_create(stream->stream_handle, net_seq_num);
1999 if (!index) {
2000 goto end_rcu_unlock;
2001 }
2002 index_created = 1;
2003 }
2004
2005 copy_index_control_data(index, &index_info);
2006
2007 if (index_created) {
2008 /*
2009 * Try to add the relay index object to the hash table. If an object
2010 * already exist, destroy back the index created, set the data in this
2011 * object and write it on disk.
2012 */
2013 relay_index_add(index, &wr_index);
2014 if (wr_index) {
2015 copy_index_control_data(wr_index, &index_info);
2016 free(index);
2017 }
2018 } else {
2019 /* The index already exists so write it on disk. */
2020 wr_index = index;
2021 }
2022
2023 /* Do we have a writable ready index to write on disk. */
2024 if (wr_index) {
2025 /* Starting at 2.4, create the index file if none available. */
2026 if (cmd->minor >= 4 && stream->index_fd < 0) {
2027 ret = index_create_file(stream->path_name, stream->channel_name,
2028 relayd_uid, relayd_gid, stream->tracefile_size,
2029 stream->tracefile_count_current);
2030 if (ret < 0) {
2031 goto end_rcu_unlock;
2032 }
2033 stream->index_fd = ret;
2034 }
2035
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 }
2042
2043 end_rcu_unlock:
2044 rcu_read_unlock();
2045
2046 if (ret < 0) {
2047 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2048 } else {
2049 reply.ret_code = htobe32(LTTNG_OK);
2050 }
2051 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
2052 if (send_ret < 0) {
2053 ERR("Relay sending close index id reply");
2054 ret = send_ret;
2055 }
2056
2057 end_no_session:
2058 return ret;
2059 }
2060
2061 /*
2062 * Receive the streams_sent message.
2063 *
2064 * Return 0 on success else a negative value.
2065 */
2066 static
2067 int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
2068 struct relay_command *cmd)
2069 {
2070 int ret, send_ret;
2071 struct lttcomm_relayd_generic_reply reply;
2072
2073 assert(cmd);
2074
2075 DBG("Relay receiving streams_sent");
2076
2077 if (!cmd->session || cmd->version_check_done == 0) {
2078 ERR("Trying to close a stream before version check");
2079 ret = -1;
2080 goto end_no_session;
2081 }
2082
2083 /*
2084 * Flag every pending stream in the connection recv list that they are
2085 * ready to be used by the viewer.
2086 */
2087 set_viewer_ready_flag(cmd);
2088
2089 reply.ret_code = htobe32(LTTNG_OK);
2090 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
2091 if (send_ret < 0) {
2092 ERR("Relay sending sent_stream reply");
2093 ret = send_ret;
2094 } else {
2095 /* Success. */
2096 ret = 0;
2097 }
2098
2099 end_no_session:
2100 return ret;
2101 }
2102
2103 /*
2104 * Process the commands received on the control socket
2105 */
2106 static
2107 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
2108 struct relay_command *cmd, struct relay_local_data *ctx)
2109 {
2110 int ret = 0;
2111
2112 switch (be32toh(recv_hdr->cmd)) {
2113 case RELAYD_CREATE_SESSION:
2114 ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht);
2115 break;
2116 case RELAYD_ADD_STREAM:
2117 ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht);
2118 break;
2119 case RELAYD_START_DATA:
2120 ret = relay_start(recv_hdr, cmd);
2121 break;
2122 case RELAYD_SEND_METADATA:
2123 ret = relay_recv_metadata(recv_hdr, cmd);
2124 break;
2125 case RELAYD_VERSION:
2126 ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht);
2127 break;
2128 case RELAYD_CLOSE_STREAM:
2129 ret = relay_close_stream(recv_hdr, cmd);
2130 break;
2131 case RELAYD_DATA_PENDING:
2132 ret = relay_data_pending(recv_hdr, cmd);
2133 break;
2134 case RELAYD_QUIESCENT_CONTROL:
2135 ret = relay_quiescent_control(recv_hdr, cmd);
2136 break;
2137 case RELAYD_BEGIN_DATA_PENDING:
2138 ret = relay_begin_data_pending(recv_hdr, cmd);
2139 break;
2140 case RELAYD_END_DATA_PENDING:
2141 ret = relay_end_data_pending(recv_hdr, cmd);
2142 break;
2143 case RELAYD_SEND_INDEX:
2144 ret = relay_recv_index(recv_hdr, cmd);
2145 break;
2146 case RELAYD_STREAMS_SENT:
2147 ret = relay_streams_sent(recv_hdr, cmd);
2148 break;
2149 case RELAYD_UPDATE_SYNC_INFO:
2150 default:
2151 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2152 relay_unknown_command(cmd);
2153 ret = -1;
2154 goto end;
2155 }
2156
2157 end:
2158 return ret;
2159 }
2160
2161 /*
2162 * Handle index for a data stream.
2163 *
2164 * RCU read side lock MUST be acquired.
2165 *
2166 * Return 0 on success else a negative value.
2167 */
2168 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2169 int rotate_index)
2170 {
2171 int ret = 0, index_created = 0;
2172 uint64_t stream_id, data_offset;
2173 struct relay_index *index, *wr_index = NULL;
2174
2175 assert(stream);
2176
2177 stream_id = stream->stream_handle;
2178 /* Get data offset because we are about to update the index. */
2179 data_offset = htobe64(stream->tracefile_size_current);
2180
2181 /*
2182 * Lookup for an existing index for that stream id/sequence number. If on
2183 * exists, the control thread already received the data for it thus we need
2184 * to write it on disk.
2185 */
2186 index = relay_index_find(stream_id, net_seq_num);
2187 if (!index) {
2188 /* A successful creation will add the object to the HT. */
2189 index = relay_index_create(stream_id, net_seq_num);
2190 if (!index) {
2191 ret = -1;
2192 goto error;
2193 }
2194 index_created = 1;
2195 }
2196
2197 if (rotate_index || stream->index_fd < 0) {
2198 index->to_close_fd = stream->index_fd;
2199 ret = index_create_file(stream->path_name, stream->channel_name,
2200 relayd_uid, relayd_gid, stream->tracefile_size,
2201 stream->tracefile_count_current);
2202 if (ret < 0) {
2203 /* This will close the stream's index fd if one. */
2204 relay_index_free_safe(index);
2205 goto error;
2206 }
2207 stream->index_fd = ret;
2208 }
2209 index->fd = stream->index_fd;
2210 index->index_data.offset = data_offset;
2211
2212 if (index_created) {
2213 /*
2214 * Try to add the relay index object to the hash table. If an object
2215 * already exist, destroy back the index created and set the data.
2216 */
2217 relay_index_add(index, &wr_index);
2218 if (wr_index) {
2219 /* Copy back data from the created index. */
2220 wr_index->fd = index->fd;
2221 wr_index->to_close_fd = index->to_close_fd;
2222 wr_index->index_data.offset = data_offset;
2223 free(index);
2224 }
2225 } else {
2226 /* The index already exists so write it on disk. */
2227 wr_index = index;
2228 }
2229
2230 /* Do we have a writable ready index to write on disk. */
2231 if (wr_index) {
2232 ret = relay_index_write(wr_index->fd, wr_index);
2233 if (ret < 0) {
2234 goto error;
2235 }
2236 stream->total_index_received++;
2237 }
2238
2239 error:
2240 return ret;
2241 }
2242
2243 /*
2244 * relay_process_data: Process the data received on the data socket
2245 */
2246 static
2247 int relay_process_data(struct relay_command *cmd)
2248 {
2249 int ret = 0, rotate_index = 0;
2250 ssize_t size_ret;
2251 struct relay_stream *stream;
2252 struct lttcomm_relayd_data_hdr data_hdr;
2253 uint64_t stream_id;
2254 uint64_t net_seq_num;
2255 uint32_t data_size;
2256
2257 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
2258 sizeof(struct lttcomm_relayd_data_hdr), 0);
2259 if (ret <= 0) {
2260 if (ret == 0) {
2261 /* Orderly shutdown. Not necessary to print an error. */
2262 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2263 } else {
2264 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
2265 }
2266 ret = -1;
2267 goto end;
2268 }
2269
2270 stream_id = be64toh(data_hdr.stream_id);
2271
2272 rcu_read_lock();
2273 stream = relay_stream_find_by_id(stream_id);
2274 if (!stream) {
2275 ret = -1;
2276 goto end_rcu_unlock;
2277 }
2278
2279 data_size = be32toh(data_hdr.data_size);
2280 if (data_buffer_size < data_size) {
2281 char *tmp_data_ptr;
2282
2283 tmp_data_ptr = realloc(data_buffer, data_size);
2284 if (!tmp_data_ptr) {
2285 ERR("Allocating data buffer");
2286 free(data_buffer);
2287 ret = -1;
2288 goto end_rcu_unlock;
2289 }
2290 data_buffer = tmp_data_ptr;
2291 data_buffer_size = data_size;
2292 }
2293 memset(data_buffer, 0, data_size);
2294
2295 net_seq_num = be64toh(data_hdr.net_seq_num);
2296
2297 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2298 data_size, stream_id, net_seq_num);
2299 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
2300 if (ret <= 0) {
2301 if (ret == 0) {
2302 /* Orderly shutdown. Not necessary to print an error. */
2303 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2304 }
2305 ret = -1;
2306 goto end_rcu_unlock;
2307 }
2308
2309 /* Check if a rotation is needed. */
2310 if (stream->tracefile_size > 0 &&
2311 (stream->tracefile_size_current + data_size) >
2312 stream->tracefile_size) {
2313 struct relay_viewer_stream *vstream;
2314 uint64_t new_id;
2315
2316 new_id = (stream->tracefile_count_current + 1) %
2317 stream->tracefile_count;
2318 /*
2319 * When we wrap-around back to 0, we start overwriting old
2320 * trace data.
2321 */
2322 if (!stream->tracefile_overwrite && new_id == 0) {
2323 stream->tracefile_overwrite = 1;
2324 }
2325 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
2326 if (stream->tracefile_overwrite) {
2327 stream->oldest_tracefile_id =
2328 (stream->oldest_tracefile_id + 1) %
2329 stream->tracefile_count;
2330 }
2331 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
2332 if (vstream) {
2333 /*
2334 * The viewer is reading a file about to be
2335 * overwritten. Close the FDs it is
2336 * currently using and let it handle the fault.
2337 */
2338 if (vstream->tracefile_count_current == new_id) {
2339 pthread_mutex_lock(&vstream->overwrite_lock);
2340 vstream->abort_flag = 1;
2341 pthread_mutex_unlock(&vstream->overwrite_lock);
2342 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2343 stream->channel_name, new_id);
2344 } else if (vstream->tracefile_count_current ==
2345 stream->tracefile_count_current) {
2346 /*
2347 * The reader and writer were in the
2348 * same trace file, inform the viewer
2349 * that no new index will ever be added
2350 * to this file.
2351 */
2352 vstream->close_write_flag = 1;
2353 }
2354 }
2355 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2356 stream->tracefile_size, stream->tracefile_count,
2357 relayd_uid, relayd_gid, stream->fd,
2358 &(stream->tracefile_count_current), &stream->fd);
2359 stream->total_index_received = 0;
2360 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
2361 if (ret < 0) {
2362 ERR("Rotating stream output file");
2363 goto end_rcu_unlock;
2364 }
2365 /* Reset current size because we just perform a stream rotation. */
2366 stream->tracefile_size_current = 0;
2367 rotate_index = 1;
2368 }
2369
2370 /*
2371 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2372 * index are NOT supported.
2373 */
2374 if (stream->session->minor >= 4 && !stream->session->snapshot) {
2375 ret = handle_index_data(stream, net_seq_num, rotate_index);
2376 if (ret < 0) {
2377 goto end_rcu_unlock;
2378 }
2379 }
2380
2381 /* Write data to stream output fd. */
2382 size_ret = lttng_write(stream->fd, data_buffer, data_size);
2383 if (size_ret < data_size) {
2384 ERR("Relay error writing data to file");
2385 ret = -1;
2386 goto end_rcu_unlock;
2387 }
2388
2389 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2390 ret, stream->stream_handle);
2391
2392 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2393 if (ret < 0) {
2394 goto end_rcu_unlock;
2395 }
2396 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2397
2398 stream->prev_seq = net_seq_num;
2399
2400 /* Check if we need to close the FD */
2401 if (close_stream_check(stream)) {
2402 destroy_stream(stream);
2403 }
2404
2405 end_rcu_unlock:
2406 rcu_read_unlock();
2407 end:
2408 return ret;
2409 }
2410
2411 static
2412 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
2413 {
2414 int ret;
2415
2416 lttng_poll_del(events, pollfd);
2417
2418 ret = close(pollfd);
2419 if (ret < 0) {
2420 ERR("Closing pollfd %d", pollfd);
2421 }
2422 }
2423
2424 static
2425 int relay_add_connection(int fd, struct lttng_poll_event *events,
2426 struct lttng_ht *relay_connections_ht)
2427 {
2428 struct relay_command *relay_connection;
2429 ssize_t ret;
2430
2431 relay_connection = zmalloc(sizeof(struct relay_command));
2432 if (relay_connection == NULL) {
2433 PERROR("Relay command zmalloc");
2434 goto error;
2435 }
2436 ret = lttng_read(fd, relay_connection, sizeof(struct relay_command));
2437 if (ret < sizeof(struct relay_command)) {
2438 PERROR("read relay cmd pipe");
2439 goto error_read;
2440 }
2441 CDS_INIT_LIST_HEAD(&relay_connection->recv_head);
2442
2443 /*
2444 * Only used by the control side and the reference is copied inside each
2445 * stream from that connection. Thus a destroy HT must be done after every
2446 * stream has been destroyed.
2447 */
2448 if (relay_connection->type == RELAY_CONTROL) {
2449 relay_connection->ctf_traces_ht = lttng_ht_new(0,
2450 LTTNG_HT_TYPE_STRING);
2451 if (!relay_connection->ctf_traces_ht) {
2452 goto error_read;
2453 }
2454 }
2455
2456 lttng_ht_node_init_ulong(&relay_connection->sock_n,
2457 (unsigned long) relay_connection->sock->fd);
2458 rcu_read_lock();
2459 lttng_ht_add_unique_ulong(relay_connections_ht,
2460 &relay_connection->sock_n);
2461 rcu_read_unlock();
2462 return lttng_poll_add(events,
2463 relay_connection->sock->fd,
2464 LPOLLIN | LPOLLRDHUP);
2465
2466 error_read:
2467 free(relay_connection);
2468 error:
2469 return -1;
2470 }
2471
2472 static
2473 void deferred_free_connection(struct rcu_head *head)
2474 {
2475 struct relay_command *relay_connection =
2476 caa_container_of(head, struct relay_command, rcu_node);
2477
2478 lttcomm_destroy_sock(relay_connection->sock);
2479 free(relay_connection);
2480 }
2481
2482 static
2483 void relay_del_connection(struct lttng_ht *relay_connections_ht,
2484 struct lttng_ht_iter *iter, struct relay_command *relay_connection,
2485 struct lttng_ht *sessions_ht)
2486 {
2487 int ret;
2488
2489 ret = lttng_ht_del(relay_connections_ht, iter);
2490 assert(!ret);
2491
2492 if (relay_connection->type == RELAY_CONTROL) {
2493 struct relay_stream_recv_handle *node, *tmp_node;
2494
2495 relay_delete_session(relay_connection, sessions_ht);
2496 lttng_ht_destroy(relay_connection->ctf_traces_ht);
2497
2498 /* Clean up recv list. */
2499 cds_list_for_each_entry_safe(node, tmp_node,
2500 &relay_connection->recv_head, node) {
2501 cds_list_del(&node->node);
2502 free(node);
2503 }
2504 }
2505
2506 call_rcu(&relay_connection->rcu_node, deferred_free_connection);
2507 }
2508
2509 /*
2510 * This thread does the actual work
2511 */
2512 static
2513 void *relay_thread_worker(void *data)
2514 {
2515 int ret, err = -1, last_seen_data_fd = -1;
2516 uint32_t nb_fd;
2517 struct relay_command *relay_connection;
2518 struct lttng_poll_event events;
2519 struct lttng_ht *relay_connections_ht;
2520 struct lttng_ht_node_ulong *node;
2521 struct lttng_ht_iter iter;
2522 struct lttcomm_relayd_hdr recv_hdr;
2523 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2524 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2525
2526 DBG("[thread] Relay worker started");
2527
2528 rcu_register_thread();
2529
2530 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2531
2532 health_code_update();
2533
2534 /* table of connections indexed on socket */
2535 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2536 if (!relay_connections_ht) {
2537 goto relay_connections_ht_error;
2538 }
2539
2540 /* Tables of received indexes indexed by index handle and net_seq_num. */
2541 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2542 if (!indexes_ht) {
2543 goto indexes_ht_error;
2544 }
2545
2546 ret = create_thread_poll_set(&events, 2);
2547 if (ret < 0) {
2548 goto error_poll_create;
2549 }
2550
2551 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
2552 if (ret < 0) {
2553 goto error;
2554 }
2555
2556 restart:
2557 while (1) {
2558 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2559
2560 health_code_update();
2561
2562 /* Infinite blocking call, waiting for transmission */
2563 DBG3("Relayd worker thread polling...");
2564 health_poll_entry();
2565 ret = lttng_poll_wait(&events, -1);
2566 health_poll_exit();
2567 if (ret < 0) {
2568 /*
2569 * Restart interrupted system call.
2570 */
2571 if (errno == EINTR) {
2572 goto restart;
2573 }
2574 goto error;
2575 }
2576
2577 nb_fd = ret;
2578
2579 /*
2580 * Process control. The control connection is prioritised so we don't
2581 * starve it with high throughout put tracing data on the data
2582 * connection.
2583 */
2584 for (i = 0; i < nb_fd; i++) {
2585 /* Fetch once the poll data */
2586 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2587 int pollfd = LTTNG_POLL_GETFD(&events, i);
2588
2589 health_code_update();
2590
2591 /* Thread quit pipe has been closed. Killing thread. */
2592 ret = check_thread_quit_pipe(pollfd, revents);
2593 if (ret) {
2594 err = 0;
2595 goto exit;
2596 }
2597
2598 /* Inspect the relay cmd pipe for new connection */
2599 if (pollfd == relay_cmd_pipe[0]) {
2600 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2601 ERR("Relay pipe error");
2602 goto error;
2603 } else if (revents & LPOLLIN) {
2604 DBG("Relay command received");
2605 ret = relay_add_connection(relay_cmd_pipe[0],
2606 &events, relay_connections_ht);
2607 if (ret < 0) {
2608 goto error;
2609 }
2610 }
2611 } else if (revents) {
2612 rcu_read_lock();
2613 lttng_ht_lookup(relay_connections_ht,
2614 (void *)((unsigned long) pollfd),
2615 &iter);
2616 node = lttng_ht_iter_get_node_ulong(&iter);
2617 if (node == NULL) {
2618 DBG2("Relay sock %d not found", pollfd);
2619 rcu_read_unlock();
2620 goto error;
2621 }
2622 relay_connection = caa_container_of(node,
2623 struct relay_command, sock_n);
2624
2625 if (revents & (LPOLLERR)) {
2626 ERR("POLL ERROR");
2627 relay_cleanup_poll_connection(&events, pollfd);
2628 relay_del_connection(relay_connections_ht,
2629 &iter, relay_connection, sessions_ht);
2630 if (last_seen_data_fd == pollfd) {
2631 last_seen_data_fd = last_notdel_data_fd;
2632 }
2633 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2634 DBG("Socket %d hung up", pollfd);
2635 relay_cleanup_poll_connection(&events, pollfd);
2636 relay_del_connection(relay_connections_ht,
2637 &iter, relay_connection, sessions_ht);
2638 if (last_seen_data_fd == pollfd) {
2639 last_seen_data_fd = last_notdel_data_fd;
2640 }
2641 } else if (revents & LPOLLIN) {
2642 /* control socket */
2643 if (relay_connection->type == RELAY_CONTROL) {
2644 ret = relay_connection->sock->ops->recvmsg(
2645 relay_connection->sock, &recv_hdr,
2646 sizeof(struct lttcomm_relayd_hdr), 0);
2647 /* connection closed */
2648 if (ret <= 0) {
2649 relay_cleanup_poll_connection(&events, pollfd);
2650 relay_del_connection(relay_connections_ht,
2651 &iter, relay_connection, sessions_ht);
2652 DBG("Control connection closed with %d", pollfd);
2653 } else {
2654 if (relay_connection->session) {
2655 DBG2("Relay worker receiving data for session : %" PRIu64,
2656 relay_connection->session->id);
2657 }
2658 ret = relay_process_control(&recv_hdr,
2659 relay_connection, relay_ctx);
2660 if (ret < 0) {
2661 /* Clear the session on error. */
2662 relay_cleanup_poll_connection(&events, pollfd);
2663 relay_del_connection(relay_connections_ht,
2664 &iter, relay_connection, sessions_ht);
2665 DBG("Connection closed with %d", pollfd);
2666 }
2667 seen_control = 1;
2668 }
2669 } else {
2670 /*
2671 * Flag the last seen data fd not deleted. It will be
2672 * used as the last seen fd if any fd gets deleted in
2673 * this first loop.
2674 */
2675 last_notdel_data_fd = pollfd;
2676 }
2677 }
2678 rcu_read_unlock();
2679 }
2680 }
2681
2682 /*
2683 * The last loop handled a control request, go back to poll to make
2684 * sure we prioritise the control socket.
2685 */
2686 if (seen_control) {
2687 continue;
2688 }
2689
2690 if (last_seen_data_fd >= 0) {
2691 for (i = 0; i < nb_fd; i++) {
2692 int pollfd = LTTNG_POLL_GETFD(&events, i);
2693
2694 health_code_update();
2695
2696 if (last_seen_data_fd == pollfd) {
2697 idx = i;
2698 break;
2699 }
2700 }
2701 }
2702
2703 /* Process data connection. */
2704 for (i = idx + 1; i < nb_fd; i++) {
2705 /* Fetch the poll data. */
2706 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2707 int pollfd = LTTNG_POLL_GETFD(&events, i);
2708
2709 health_code_update();
2710
2711 /* Skip the command pipe. It's handled in the first loop. */
2712 if (pollfd == relay_cmd_pipe[0]) {
2713 continue;
2714 }
2715
2716 if (revents) {
2717 rcu_read_lock();
2718 lttng_ht_lookup(relay_connections_ht,
2719 (void *)((unsigned long) pollfd),
2720 &iter);
2721 node = lttng_ht_iter_get_node_ulong(&iter);
2722 if (node == NULL) {
2723 /* Skip it. Might be removed before. */
2724 rcu_read_unlock();
2725 continue;
2726 }
2727 relay_connection = caa_container_of(node,
2728 struct relay_command, sock_n);
2729
2730 if (revents & LPOLLIN) {
2731 if (relay_connection->type != RELAY_DATA) {
2732 continue;
2733 }
2734
2735 ret = relay_process_data(relay_connection);
2736 /* connection closed */
2737 if (ret < 0) {
2738 relay_cleanup_poll_connection(&events, pollfd);
2739 relay_del_connection(relay_connections_ht,
2740 &iter, relay_connection, sessions_ht);
2741 DBG("Data connection closed with %d", pollfd);
2742 /*
2743 * Every goto restart call sets the last seen fd where
2744 * here we don't really care since we gracefully
2745 * continue the loop after the connection is deleted.
2746 */
2747 } else {
2748 /* Keep last seen port. */
2749 last_seen_data_fd = pollfd;
2750 rcu_read_unlock();
2751 goto restart;
2752 }
2753 }
2754 rcu_read_unlock();
2755 }
2756 }
2757 last_seen_data_fd = -1;
2758 }
2759
2760 /* Normal exit, no error */
2761 ret = 0;
2762
2763 exit:
2764 error:
2765 lttng_poll_clean(&events);
2766
2767 /* empty the hash table and free the memory */
2768 rcu_read_lock();
2769 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2770 health_code_update();
2771
2772 node = lttng_ht_iter_get_node_ulong(&iter);
2773 if (node) {
2774 relay_connection = caa_container_of(node,
2775 struct relay_command, sock_n);
2776 relay_del_connection(relay_connections_ht,
2777 &iter, relay_connection, sessions_ht);
2778 }
2779 }
2780 rcu_read_unlock();
2781 error_poll_create:
2782 lttng_ht_destroy(indexes_ht);
2783 indexes_ht_error:
2784 lttng_ht_destroy(relay_connections_ht);
2785 relay_connections_ht_error:
2786 /* Close relay cmd pipes */
2787 utils_close_pipe(relay_cmd_pipe);
2788 if (err) {
2789 DBG("Thread exited with error");
2790 }
2791 DBG("Worker thread cleanup complete");
2792 free(data_buffer);
2793 if (err) {
2794 health_error();
2795 ERR("Health error occurred in %s", __func__);
2796 }
2797 health_unregister(health_relayd);
2798 rcu_unregister_thread();
2799 stop_threads();
2800 return NULL;
2801 }
2802
2803 /*
2804 * Create the relay command pipe to wake thread_manage_apps.
2805 * Closed in cleanup().
2806 */
2807 static int create_relay_cmd_pipe(void)
2808 {
2809 int ret;
2810
2811 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2812
2813 return ret;
2814 }
2815
2816 /*
2817 * main
2818 */
2819 int main(int argc, char **argv)
2820 {
2821 int ret = 0;
2822 void *status;
2823 struct relay_local_data *relay_ctx;
2824
2825 /* Create thread quit pipe */
2826 if ((ret = init_thread_quit_pipe()) < 0) {
2827 goto error;
2828 }
2829
2830 /* Parse arguments */
2831 progname = argv[0];
2832 if ((ret = set_options(argc, argv)) < 0) {
2833 goto exit;
2834 }
2835
2836 if ((ret = set_signal_handler()) < 0) {
2837 goto exit;
2838 }
2839
2840 /* Try to create directory if -o, --output is specified. */
2841 if (opt_output_path) {
2842 if (*opt_output_path != '/') {
2843 ERR("Please specify an absolute path for -o, --output PATH");
2844 goto exit;
2845 }
2846
2847 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2848 if (ret < 0) {
2849 ERR("Unable to create %s", opt_output_path);
2850 goto exit;
2851 }
2852 }
2853
2854 /* Daemonize */
2855 if (opt_daemon) {
2856 ret = daemon(0, 0);
2857 if (ret < 0) {
2858 PERROR("daemon");
2859 goto exit;
2860 }
2861 }
2862
2863 /* We need those values for the file/dir creation. */
2864 relayd_uid = getuid();
2865 relayd_gid = getgid();
2866
2867 /* Check if daemon is UID = 0 */
2868 if (relayd_uid == 0) {
2869 if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) {
2870 ERR("Need to be root to use ports < 1024");
2871 ret = -1;
2872 goto exit;
2873 }
2874 }
2875
2876 /* Setup the thread apps communication pipe. */
2877 if ((ret = create_relay_cmd_pipe()) < 0) {
2878 goto exit;
2879 }
2880
2881 /* Init relay command queue. */
2882 cds_wfq_init(&relay_cmd_queue.queue);
2883
2884 /* Set up max poll set size */
2885 lttng_poll_set_max_size();
2886
2887 /* Initialize communication library */
2888 lttcomm_init();
2889
2890 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2891 if (!relay_ctx) {
2892 PERROR("relay_ctx");
2893 goto exit;
2894 }
2895
2896 /* tables of sessions indexed by session ID */
2897 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2898 if (!relay_ctx->sessions_ht) {
2899 goto exit_relay_ctx_sessions;
2900 }
2901
2902 /* tables of streams indexed by stream ID */
2903 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2904 if (!relay_streams_ht) {
2905 goto exit_relay_ctx_streams;
2906 }
2907
2908 /* tables of streams indexed by stream ID */
2909 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2910 if (!viewer_streams_ht) {
2911 goto exit_relay_ctx_viewer_streams;
2912 }
2913
2914 /* Initialize thread health monitoring */
2915 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2916 if (!health_relayd) {
2917 PERROR("health_app_create error");
2918 goto exit_health_app_create;
2919 }
2920
2921 ret = utils_create_pipe(health_quit_pipe);
2922 if (ret < 0) {
2923 goto error_health_pipe;
2924 }
2925
2926 /* Create thread to manage the client socket */
2927 ret = pthread_create(&health_thread, NULL,
2928 thread_manage_health, (void *) NULL);
2929 if (ret != 0) {
2930 PERROR("pthread_create health");
2931 goto health_error;
2932 }
2933
2934 /* Setup the dispatcher thread */
2935 ret = pthread_create(&dispatcher_thread, NULL,
2936 relay_thread_dispatcher, (void *) NULL);
2937 if (ret != 0) {
2938 PERROR("pthread_create dispatcher");
2939 goto exit_dispatcher;
2940 }
2941
2942 /* Setup the worker thread */
2943 ret = pthread_create(&worker_thread, NULL,
2944 relay_thread_worker, (void *) relay_ctx);
2945 if (ret != 0) {
2946 PERROR("pthread_create worker");
2947 goto exit_worker;
2948 }
2949
2950 /* Setup the listener thread */
2951 ret = pthread_create(&listener_thread, NULL,
2952 relay_thread_listener, (void *) NULL);
2953 if (ret != 0) {
2954 PERROR("pthread_create listener");
2955 goto exit_listener;
2956 }
2957
2958 ret = live_start_threads(live_uri, relay_ctx, thread_quit_pipe);
2959 if (ret != 0) {
2960 ERR("Starting live viewer threads");
2961 goto exit_live;
2962 }
2963
2964 exit_live:
2965 ret = pthread_join(listener_thread, &status);
2966 if (ret != 0) {
2967 PERROR("pthread_join");
2968 goto error; /* join error, exit without cleanup */
2969 }
2970
2971 exit_listener:
2972 ret = pthread_join(worker_thread, &status);
2973 if (ret != 0) {
2974 PERROR("pthread_join");
2975 goto error; /* join error, exit without cleanup */
2976 }
2977
2978 exit_worker:
2979 ret = pthread_join(dispatcher_thread, &status);
2980 if (ret != 0) {
2981 PERROR("pthread_join");
2982 goto error; /* join error, exit without cleanup */
2983 }
2984
2985 exit_dispatcher:
2986 ret = pthread_join(health_thread, &status);
2987 if (ret != 0) {
2988 PERROR("pthread_join health thread");
2989 goto error; /* join error, exit without cleanup */
2990 }
2991
2992 /*
2993 * Stop live threads only after joining other threads.
2994 */
2995 live_stop_threads();
2996
2997 health_error:
2998 utils_close_pipe(health_quit_pipe);
2999
3000 error_health_pipe:
3001 health_app_destroy(health_relayd);
3002
3003 exit_health_app_create:
3004 lttng_ht_destroy(viewer_streams_ht);
3005
3006 exit_relay_ctx_viewer_streams:
3007 lttng_ht_destroy(relay_streams_ht);
3008
3009 exit_relay_ctx_streams:
3010 lttng_ht_destroy(relay_ctx->sessions_ht);
3011
3012 exit_relay_ctx_sessions:
3013 free(relay_ctx);
3014
3015 exit:
3016 cleanup();
3017 if (!ret) {
3018 exit(EXIT_SUCCESS);
3019 }
3020
3021 error:
3022 exit(EXIT_FAILURE);
3023 }
This page took 0.166871 seconds and 4 git commands to generate.