Fix: Live tracing does not honor live timer after first tracefile with tracefile...
[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/compat/endian.h>
48 #include <common/defaults.h>
49 #include <common/daemonize.h>
50 #include <common/futex.h>
51 #include <common/sessiond-comm/sessiond-comm.h>
52 #include <common/sessiond-comm/inet.h>
53 #include <common/sessiond-comm/relayd.h>
54 #include <common/uri.h>
55 #include <common/utils.h>
56 #include <common/config/config.h>
57
58 #include "cmd.h"
59 #include "ctf-trace.h"
60 #include "index.h"
61 #include "utils.h"
62 #include "lttng-relayd.h"
63 #include "live.h"
64 #include "health-relayd.h"
65 #include "testpoint.h"
66 #include "viewer-stream.h"
67 #include "session.h"
68 #include "stream.h"
69 #include "connection.h"
70
71 /* command line options */
72 char *opt_output_path;
73 static int opt_daemon, opt_background;
74
75 /*
76 * We need to wait for listener and live listener threads, as well as
77 * health check thread, before being ready to signal readiness.
78 */
79 #define NR_LTTNG_RELAY_READY 3
80 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
81 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
82 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
83
84 static struct lttng_uri *control_uri;
85 static struct lttng_uri *data_uri;
86 static struct lttng_uri *live_uri;
87
88 const char *progname;
89
90 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
91 static int tracing_group_name_override;
92
93 const char * const config_section_name = "relayd";
94
95 /*
96 * Quit pipe for all threads. This permits a single cancellation point
97 * for all threads when receiving an event on the pipe.
98 */
99 int thread_quit_pipe[2] = { -1, -1 };
100
101 /*
102 * This pipe is used to inform the worker thread that a command is queued and
103 * ready to be processed.
104 */
105 static int relay_conn_pipe[2] = { -1, -1 };
106
107 /* Shared between threads */
108 static int dispatch_thread_exit;
109
110 static pthread_t listener_thread;
111 static pthread_t dispatcher_thread;
112 static pthread_t worker_thread;
113 static pthread_t health_thread;
114
115 static uint64_t last_relay_stream_id;
116
117 /*
118 * Relay command queue.
119 *
120 * The relay_thread_listener and relay_thread_dispatcher communicate with this
121 * queue.
122 */
123 static struct relay_conn_queue relay_conn_queue;
124
125 /* buffer allocated at startup, used to store the trace data */
126 static char *data_buffer;
127 static unsigned int data_buffer_size;
128
129 /* We need those values for the file/dir creation. */
130 static uid_t relayd_uid;
131 static gid_t relayd_gid;
132
133 /* Global relay stream hash table. */
134 struct lttng_ht *relay_streams_ht;
135
136 /* Global relay viewer stream hash table. */
137 struct lttng_ht *viewer_streams_ht;
138
139 /* Global hash table that stores relay index object. */
140 struct lttng_ht *indexes_ht;
141
142 /* Relayd health monitoring */
143 struct health_app *health_relayd;
144
145 static struct option long_options[] = {
146 { "control-port", 1, 0, 'C', },
147 { "data-port", 1, 0, 'D', },
148 { "live-port", 1, 0, 'L', },
149 { "daemonize", 0, 0, 'd', },
150 { "background", 0, 0, 'b', },
151 { "group", 1, 0, 'g', },
152 { "help", 0, 0, 'h', },
153 { "output", 1, 0, 'o', },
154 { "verbose", 0, 0, 'v', },
155 { "config", 1, 0, 'f' },
156 { NULL, 0, 0, 0, },
157 };
158
159 static const char *config_ignore_options[] = { "help", "config" };
160
161 /*
162 * usage function on stderr
163 */
164 static
165 void usage(void)
166 {
167 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
168 fprintf(stderr, " -h, --help Display this usage.\n");
169 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
170 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
171 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
172 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
173 fprintf(stderr, " -L, --live-port URL Live view port listening.\n");
174 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
175 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
176 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
177 fprintf(stderr, " -f --config Load daemon configuration file\n");
178 }
179
180 /*
181 * Take an option from the getopt output and set it in the right variable to be
182 * used later.
183 *
184 * Return 0 on success else a negative value.
185 */
186 static
187 int set_option(int opt, const char *arg, const char *optname)
188 {
189 int ret;
190
191 switch (opt) {
192 case 0:
193 fprintf(stderr, "option %s", optname);
194 if (arg) {
195 fprintf(stderr, " with arg %s\n", arg);
196 }
197 break;
198 case 'C':
199 ret = uri_parse(arg, &control_uri);
200 if (ret < 0) {
201 ERR("Invalid control URI specified");
202 goto end;
203 }
204 if (control_uri->port == 0) {
205 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
206 }
207 break;
208 case 'D':
209 ret = uri_parse(arg, &data_uri);
210 if (ret < 0) {
211 ERR("Invalid data URI specified");
212 goto end;
213 }
214 if (data_uri->port == 0) {
215 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
216 }
217 break;
218 case 'L':
219 ret = uri_parse(arg, &live_uri);
220 if (ret < 0) {
221 ERR("Invalid live URI specified");
222 goto end;
223 }
224 if (live_uri->port == 0) {
225 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
226 }
227 break;
228 case 'd':
229 opt_daemon = 1;
230 break;
231 case 'b':
232 opt_background = 1;
233 break;
234 case 'g':
235 tracing_group_name = strdup(arg);
236 if (tracing_group_name == NULL) {
237 ret = -errno;
238 PERROR("strdup");
239 goto end;
240 }
241 tracing_group_name_override = 1;
242 break;
243 case 'h':
244 usage();
245 exit(EXIT_FAILURE);
246 case 'o':
247 ret = asprintf(&opt_output_path, "%s", arg);
248 if (ret < 0) {
249 ret = -errno;
250 PERROR("asprintf opt_output_path");
251 goto end;
252 }
253 break;
254 case 'v':
255 /* Verbose level can increase using multiple -v */
256 if (arg) {
257 lttng_opt_verbose = config_parse_value(arg);
258 } else {
259 /* Only 3 level of verbosity (-vvv). */
260 if (lttng_opt_verbose < 3) {
261 lttng_opt_verbose += 1;
262 }
263 }
264 break;
265 default:
266 /* Unknown option or other error.
267 * Error is printed by getopt, just return */
268 ret = -1;
269 goto end;
270 }
271
272 /* All good. */
273 ret = 0;
274
275 end:
276 return ret;
277 }
278
279 /*
280 * config_entry_handler_cb used to handle options read from a config file.
281 * See config_entry_handler_cb comment in common/config/config.h for the
282 * return value conventions.
283 */
284 static
285 int config_entry_handler(const struct config_entry *entry, void *unused)
286 {
287 int ret = 0, i;
288
289 if (!entry || !entry->name || !entry->value) {
290 ret = -EINVAL;
291 goto end;
292 }
293
294 /* Check if the option is to be ignored */
295 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
296 if (!strcmp(entry->name, config_ignore_options[i])) {
297 goto end;
298 }
299 }
300
301 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
302 /* Ignore if entry name is not fully matched. */
303 if (strcmp(entry->name, long_options[i].name)) {
304 continue;
305 }
306
307 /*
308 * If the option takes no argument on the command line, we have to
309 * check if the value is "true". We support non-zero numeric values,
310 * true, on and yes.
311 */
312 if (!long_options[i].has_arg) {
313 ret = config_parse_value(entry->value);
314 if (ret <= 0) {
315 if (ret) {
316 WARN("Invalid configuration value \"%s\" for option %s",
317 entry->value, entry->name);
318 }
319 /* False, skip boolean config option. */
320 goto end;
321 }
322 }
323
324 ret = set_option(long_options[i].val, entry->value, entry->name);
325 goto end;
326 }
327
328 WARN("Unrecognized option \"%s\" in daemon configuration file.",
329 entry->name);
330
331 end:
332 return ret;
333 }
334
335 static
336 int set_options(int argc, char **argv)
337 {
338 int c, ret = 0, option_index = 0;
339 int orig_optopt = optopt, orig_optind = optind;
340 char *default_address, *optstring;
341 const char *config_path = NULL;
342
343 optstring = utils_generate_optstring(long_options,
344 sizeof(long_options) / sizeof(struct option));
345 if (!optstring) {
346 ret = -ENOMEM;
347 goto exit;
348 }
349
350 /* Check for the --config option */
351
352 while ((c = getopt_long(argc, argv, optstring, long_options,
353 &option_index)) != -1) {
354 if (c == '?') {
355 ret = -EINVAL;
356 goto exit;
357 } else if (c != 'f') {
358 continue;
359 }
360
361 config_path = utils_expand_path(optarg);
362 if (!config_path) {
363 ERR("Failed to resolve path: %s", optarg);
364 }
365 }
366
367 ret = config_get_section_entries(config_path, config_section_name,
368 config_entry_handler, NULL);
369 if (ret) {
370 if (ret > 0) {
371 ERR("Invalid configuration option at line %i", ret);
372 ret = -1;
373 }
374 goto exit;
375 }
376
377 /* Reset getopt's global state */
378 optopt = orig_optopt;
379 optind = orig_optind;
380 while (1) {
381 c = getopt_long(argc, argv, optstring, long_options, &option_index);
382 if (c == -1) {
383 break;
384 }
385
386 ret = set_option(c, optarg, long_options[option_index].name);
387 if (ret < 0) {
388 goto exit;
389 }
390 }
391
392 /* assign default values */
393 if (control_uri == NULL) {
394 ret = asprintf(&default_address,
395 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
396 DEFAULT_NETWORK_CONTROL_PORT);
397 if (ret < 0) {
398 PERROR("asprintf default data address");
399 goto exit;
400 }
401
402 ret = uri_parse(default_address, &control_uri);
403 free(default_address);
404 if (ret < 0) {
405 ERR("Invalid control URI specified");
406 goto exit;
407 }
408 }
409 if (data_uri == NULL) {
410 ret = asprintf(&default_address,
411 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
412 DEFAULT_NETWORK_DATA_PORT);
413 if (ret < 0) {
414 PERROR("asprintf default data address");
415 goto exit;
416 }
417
418 ret = uri_parse(default_address, &data_uri);
419 free(default_address);
420 if (ret < 0) {
421 ERR("Invalid data URI specified");
422 goto exit;
423 }
424 }
425 if (live_uri == NULL) {
426 ret = asprintf(&default_address,
427 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
428 DEFAULT_NETWORK_VIEWER_PORT);
429 if (ret < 0) {
430 PERROR("asprintf default viewer control address");
431 goto exit;
432 }
433
434 ret = uri_parse(default_address, &live_uri);
435 free(default_address);
436 if (ret < 0) {
437 ERR("Invalid viewer control URI specified");
438 goto exit;
439 }
440 }
441
442 exit:
443 free(optstring);
444 return ret;
445 }
446
447 /*
448 * Cleanup the daemon
449 */
450 static
451 void cleanup(void)
452 {
453 DBG("Cleaning up");
454
455 /* free the dynamically allocated opt_output_path */
456 free(opt_output_path);
457
458 /* Close thread quit pipes */
459 utils_close_pipe(thread_quit_pipe);
460
461 uri_free(control_uri);
462 uri_free(data_uri);
463 /* Live URI is freed in the live thread. */
464
465 if (tracing_group_name_override) {
466 free((void *) tracing_group_name);
467 }
468 }
469
470 /*
471 * Write to writable pipe used to notify a thread.
472 */
473 static
474 int notify_thread_pipe(int wpipe)
475 {
476 ssize_t ret;
477
478 ret = lttng_write(wpipe, "!", 1);
479 if (ret < 1) {
480 PERROR("write poll pipe");
481 }
482
483 return ret;
484 }
485
486 static void notify_health_quit_pipe(int *pipe)
487 {
488 ssize_t ret;
489
490 ret = lttng_write(pipe[1], "4", 1);
491 if (ret < 1) {
492 PERROR("write relay health quit");
493 }
494 }
495
496 /*
497 * Stop all threads by closing the thread quit pipe.
498 */
499 static
500 void stop_threads(void)
501 {
502 int ret;
503
504 /* Stopping all threads */
505 DBG("Terminating all threads");
506 ret = notify_thread_pipe(thread_quit_pipe[1]);
507 if (ret < 0) {
508 ERR("write error on thread quit pipe");
509 }
510
511 notify_health_quit_pipe(health_quit_pipe);
512
513 /* Dispatch thread */
514 CMM_STORE_SHARED(dispatch_thread_exit, 1);
515 futex_nto1_wake(&relay_conn_queue.futex);
516 }
517
518 /*
519 * Signal handler for the daemon
520 *
521 * Simply stop all worker threads, leaving main() return gracefully after
522 * joining all threads and calling cleanup().
523 */
524 static
525 void sighandler(int sig)
526 {
527 switch (sig) {
528 case SIGPIPE:
529 DBG("SIGPIPE caught");
530 return;
531 case SIGINT:
532 DBG("SIGINT caught");
533 stop_threads();
534 break;
535 case SIGTERM:
536 DBG("SIGTERM caught");
537 stop_threads();
538 break;
539 case SIGUSR1:
540 CMM_STORE_SHARED(recv_child_signal, 1);
541 break;
542 default:
543 break;
544 }
545 }
546
547 /*
548 * Setup signal handler for :
549 * SIGINT, SIGTERM, SIGPIPE
550 */
551 static
552 int set_signal_handler(void)
553 {
554 int ret = 0;
555 struct sigaction sa;
556 sigset_t sigset;
557
558 if ((ret = sigemptyset(&sigset)) < 0) {
559 PERROR("sigemptyset");
560 return ret;
561 }
562
563 sa.sa_handler = sighandler;
564 sa.sa_mask = sigset;
565 sa.sa_flags = 0;
566 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
567 PERROR("sigaction");
568 return ret;
569 }
570
571 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
572 PERROR("sigaction");
573 return ret;
574 }
575
576 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
577 PERROR("sigaction");
578 return ret;
579 }
580
581 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
582 PERROR("sigaction");
583 return ret;
584 }
585
586 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
587
588 return ret;
589 }
590
591 void lttng_relay_notify_ready(void)
592 {
593 /* Notify the parent of the fork() process that we are ready. */
594 if (opt_daemon || opt_background) {
595 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
596 kill(child_ppid, SIGUSR1);
597 }
598 }
599 }
600
601 /*
602 * Init thread quit pipe.
603 *
604 * Return -1 on error or 0 if all pipes are created.
605 */
606 static
607 int init_thread_quit_pipe(void)
608 {
609 int ret;
610
611 ret = utils_create_pipe_cloexec(thread_quit_pipe);
612
613 return ret;
614 }
615
616 /*
617 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
618 */
619 static
620 int create_thread_poll_set(struct lttng_poll_event *events, int size)
621 {
622 int ret;
623
624 if (events == NULL || size == 0) {
625 ret = -1;
626 goto error;
627 }
628
629 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
630 if (ret < 0) {
631 goto error;
632 }
633
634 /* Add quit pipe */
635 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
636 if (ret < 0) {
637 goto error;
638 }
639
640 return 0;
641
642 error:
643 return ret;
644 }
645
646 /*
647 * Check if the thread quit pipe was triggered.
648 *
649 * Return 1 if it was triggered else 0;
650 */
651 static
652 int check_thread_quit_pipe(int fd, uint32_t events)
653 {
654 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
655 return 1;
656 }
657
658 return 0;
659 }
660
661 /*
662 * Create and init socket from uri.
663 */
664 static
665 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
666 {
667 int ret;
668 struct lttcomm_sock *sock = NULL;
669
670 sock = lttcomm_alloc_sock_from_uri(uri);
671 if (sock == NULL) {
672 ERR("Allocating socket");
673 goto error;
674 }
675
676 ret = lttcomm_create_sock(sock);
677 if (ret < 0) {
678 goto error;
679 }
680 DBG("Listening on sock %d", sock->fd);
681
682 ret = sock->ops->bind(sock);
683 if (ret < 0) {
684 goto error;
685 }
686
687 ret = sock->ops->listen(sock, -1);
688 if (ret < 0) {
689 goto error;
690
691 }
692
693 return sock;
694
695 error:
696 if (sock) {
697 lttcomm_destroy_sock(sock);
698 }
699 return NULL;
700 }
701
702 /*
703 * Return nonzero if stream needs to be closed.
704 */
705 static
706 int close_stream_check(struct relay_stream *stream)
707 {
708 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
709 /*
710 * We are about to close the stream so set the data pending flag to 1
711 * which will make the end data pending command skip the stream which
712 * is now closed and ready. Note that after proceeding to a file close,
713 * the written file is ready for reading.
714 */
715 stream->data_pending_check_done = 1;
716 return 1;
717 }
718 return 0;
719 }
720
721 static void try_close_stream(struct relay_session *session,
722 struct relay_stream *stream)
723 {
724 int ret;
725 struct ctf_trace *ctf_trace;
726
727 assert(session);
728 assert(stream);
729
730 if (!close_stream_check(stream)) {
731 /* Can't close it, not ready for that. */
732 goto end;
733 }
734
735 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
736 stream->path_name);
737 assert(ctf_trace);
738
739 pthread_mutex_lock(&session->viewer_ready_lock);
740 ctf_trace->invalid_flag = 1;
741 pthread_mutex_unlock(&session->viewer_ready_lock);
742
743 ret = stream_close(session, stream);
744 if (ret || session->snapshot) {
745 /* Already close thus the ctf trace is being or has been destroyed. */
746 goto end;
747 }
748
749 ctf_trace_try_destroy(session, ctf_trace);
750
751 end:
752 return;
753 }
754
755 /*
756 * This thread manages the listening for new connections on the network
757 */
758 static
759 void *relay_thread_listener(void *data)
760 {
761 int i, ret, pollfd, err = -1;
762 uint32_t revents, nb_fd;
763 struct lttng_poll_event events;
764 struct lttcomm_sock *control_sock, *data_sock;
765
766 DBG("[thread] Relay listener started");
767
768 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
769
770 health_code_update();
771
772 control_sock = relay_init_sock(control_uri);
773 if (!control_sock) {
774 goto error_sock_control;
775 }
776
777 data_sock = relay_init_sock(data_uri);
778 if (!data_sock) {
779 goto error_sock_relay;
780 }
781
782 /*
783 * Pass 3 as size here for the thread quit pipe, control and data socket.
784 */
785 ret = create_thread_poll_set(&events, 3);
786 if (ret < 0) {
787 goto error_create_poll;
788 }
789
790 /* Add the control socket */
791 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
792 if (ret < 0) {
793 goto error_poll_add;
794 }
795
796 /* Add the data socket */
797 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
798 if (ret < 0) {
799 goto error_poll_add;
800 }
801
802 lttng_relay_notify_ready();
803
804 if (testpoint(relayd_thread_listener)) {
805 goto error_testpoint;
806 }
807
808 while (1) {
809 health_code_update();
810
811 DBG("Listener accepting connections");
812
813 restart:
814 health_poll_entry();
815 ret = lttng_poll_wait(&events, -1);
816 health_poll_exit();
817 if (ret < 0) {
818 /*
819 * Restart interrupted system call.
820 */
821 if (errno == EINTR) {
822 goto restart;
823 }
824 goto error;
825 }
826
827 nb_fd = ret;
828
829 DBG("Relay new connection received");
830 for (i = 0; i < nb_fd; i++) {
831 health_code_update();
832
833 /* Fetch once the poll data */
834 revents = LTTNG_POLL_GETEV(&events, i);
835 pollfd = LTTNG_POLL_GETFD(&events, i);
836
837 /* Thread quit pipe has been closed. Killing thread. */
838 ret = check_thread_quit_pipe(pollfd, revents);
839 if (ret) {
840 err = 0;
841 goto exit;
842 }
843
844 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
845 ERR("socket poll error");
846 goto error;
847 } else if (revents & LPOLLIN) {
848 /*
849 * Get allocated in this thread, enqueued to a global queue,
850 * dequeued and freed in the worker thread.
851 */
852 int val = 1;
853 struct relay_connection *new_conn;
854 struct lttcomm_sock *newsock;
855
856 new_conn = connection_create();
857 if (!new_conn) {
858 goto error;
859 }
860
861 if (pollfd == data_sock->fd) {
862 new_conn->type = RELAY_DATA;
863 newsock = data_sock->ops->accept(data_sock);
864 DBG("Relay data connection accepted, socket %d",
865 newsock->fd);
866 } else {
867 assert(pollfd == control_sock->fd);
868 new_conn->type = RELAY_CONTROL;
869 newsock = control_sock->ops->accept(control_sock);
870 DBG("Relay control connection accepted, socket %d",
871 newsock->fd);
872 }
873 if (!newsock) {
874 PERROR("accepting sock");
875 connection_free(new_conn);
876 goto error;
877 }
878
879 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
880 sizeof(val));
881 if (ret < 0) {
882 PERROR("setsockopt inet");
883 lttcomm_destroy_sock(newsock);
884 connection_free(new_conn);
885 goto error;
886 }
887 new_conn->sock = newsock;
888
889 /* Enqueue request for the dispatcher thread. */
890 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
891 &new_conn->qnode);
892
893 /*
894 * Wake the dispatch queue futex. Implicit memory barrier with
895 * the exchange in cds_wfcq_enqueue.
896 */
897 futex_nto1_wake(&relay_conn_queue.futex);
898 }
899 }
900 }
901
902 exit:
903 error:
904 error_poll_add:
905 error_testpoint:
906 lttng_poll_clean(&events);
907 error_create_poll:
908 if (data_sock->fd >= 0) {
909 ret = data_sock->ops->close(data_sock);
910 if (ret) {
911 PERROR("close");
912 }
913 }
914 lttcomm_destroy_sock(data_sock);
915 error_sock_relay:
916 if (control_sock->fd >= 0) {
917 ret = control_sock->ops->close(control_sock);
918 if (ret) {
919 PERROR("close");
920 }
921 }
922 lttcomm_destroy_sock(control_sock);
923 error_sock_control:
924 if (err) {
925 health_error();
926 ERR("Health error occurred in %s", __func__);
927 }
928 health_unregister(health_relayd);
929 DBG("Relay listener thread cleanup complete");
930 stop_threads();
931 return NULL;
932 }
933
934 /*
935 * This thread manages the dispatching of the requests to worker threads
936 */
937 static
938 void *relay_thread_dispatcher(void *data)
939 {
940 int err = -1;
941 ssize_t ret;
942 struct cds_wfcq_node *node;
943 struct relay_connection *new_conn = NULL;
944
945 DBG("[thread] Relay dispatcher started");
946
947 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
948
949 if (testpoint(relayd_thread_dispatcher)) {
950 goto error_testpoint;
951 }
952
953 health_code_update();
954
955 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
956 health_code_update();
957
958 /* Atomically prepare the queue futex */
959 futex_nto1_prepare(&relay_conn_queue.futex);
960
961 do {
962 health_code_update();
963
964 /* Dequeue commands */
965 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
966 &relay_conn_queue.tail);
967 if (node == NULL) {
968 DBG("Woken up but nothing in the relay command queue");
969 /* Continue thread execution */
970 break;
971 }
972 new_conn = caa_container_of(node, struct relay_connection, qnode);
973
974 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
975
976 /*
977 * Inform worker thread of the new request. This call is blocking
978 * so we can be assured that the data will be read at some point in
979 * time or wait to the end of the world :)
980 */
981 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
982 if (ret < 0) {
983 PERROR("write connection pipe");
984 connection_destroy(new_conn);
985 goto error;
986 }
987 } while (node != NULL);
988
989 /* Futex wait on queue. Blocking call on futex() */
990 health_poll_entry();
991 futex_nto1_wait(&relay_conn_queue.futex);
992 health_poll_exit();
993 }
994
995 /* Normal exit, no error */
996 err = 0;
997
998 error:
999 error_testpoint:
1000 if (err) {
1001 health_error();
1002 ERR("Health error occurred in %s", __func__);
1003 }
1004 health_unregister(health_relayd);
1005 DBG("Dispatch thread dying");
1006 stop_threads();
1007 return NULL;
1008 }
1009
1010 static void try_close_streams(struct relay_session *session)
1011 {
1012 struct ctf_trace *ctf_trace;
1013 struct lttng_ht_iter iter;
1014
1015 assert(session);
1016
1017 pthread_mutex_lock(&session->viewer_ready_lock);
1018 rcu_read_lock();
1019 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
1020 node.node) {
1021 struct relay_stream *stream;
1022
1023 /* Close streams. */
1024 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
1025 stream_close(session, stream);
1026 }
1027
1028 ctf_trace->invalid_flag = 1;
1029 ctf_trace_try_destroy(session, ctf_trace);
1030 }
1031 rcu_read_unlock();
1032 pthread_mutex_unlock(&session->viewer_ready_lock);
1033 }
1034
1035 /*
1036 * Try to destroy a session within a connection.
1037 */
1038 static void destroy_session(struct relay_session *session,
1039 struct lttng_ht *sessions_ht)
1040 {
1041 assert(session);
1042 assert(sessions_ht);
1043
1044 /* Indicate that this session can be destroyed from now on. */
1045 session->close_flag = 1;
1046
1047 try_close_streams(session);
1048
1049 /*
1050 * This will try to delete and destroy the session if no viewer is attached
1051 * to it meaning the refcount is down to zero.
1052 */
1053 session_try_destroy(sessions_ht, session);
1054 }
1055
1056 /*
1057 * Copy index data from the control port to a given index object.
1058 */
1059 static void copy_index_control_data(struct relay_index *index,
1060 struct lttcomm_relayd_index *data)
1061 {
1062 assert(index);
1063 assert(data);
1064
1065 /*
1066 * The index on disk is encoded in big endian, so we don't need to convert
1067 * the data received on the network. The data_offset value is NEVER
1068 * modified here and is updated by the data thread.
1069 */
1070 index->index_data.packet_size = data->packet_size;
1071 index->index_data.content_size = data->content_size;
1072 index->index_data.timestamp_begin = data->timestamp_begin;
1073 index->index_data.timestamp_end = data->timestamp_end;
1074 index->index_data.events_discarded = data->events_discarded;
1075 index->index_data.stream_id = data->stream_id;
1076 }
1077
1078 /*
1079 * Handle the RELAYD_CREATE_SESSION command.
1080 *
1081 * On success, send back the session id or else return a negative value.
1082 */
1083 static
1084 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
1085 struct relay_connection *conn)
1086 {
1087 int ret = 0, send_ret;
1088 struct relay_session *session;
1089 struct lttcomm_relayd_status_session reply;
1090
1091 assert(recv_hdr);
1092 assert(conn);
1093
1094 memset(&reply, 0, sizeof(reply));
1095
1096 session = session_create();
1097 if (!session) {
1098 ret = -1;
1099 goto error;
1100 }
1101 session->minor = conn->minor;
1102 session->major = conn->major;
1103 conn->session_id = session->id;
1104 conn->session = session;
1105
1106 reply.session_id = htobe64(session->id);
1107
1108 switch (conn->minor) {
1109 case 1:
1110 case 2:
1111 case 3:
1112 break;
1113 case 4: /* LTTng sessiond 2.4 */
1114 default:
1115 ret = cmd_create_session_2_4(conn, session);
1116 }
1117
1118 lttng_ht_add_unique_u64(conn->sessions_ht, &session->session_n);
1119 DBG("Created session %" PRIu64, session->id);
1120
1121 error:
1122 if (ret < 0) {
1123 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1124 } else {
1125 reply.ret_code = htobe32(LTTNG_OK);
1126 }
1127
1128 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1129 if (send_ret < 0) {
1130 ERR("Relayd sending session id");
1131 ret = send_ret;
1132 }
1133
1134 return ret;
1135 }
1136
1137 /*
1138 * When we have received all the streams and the metadata for a channel,
1139 * we make them visible to the viewer threads.
1140 */
1141 static
1142 void set_viewer_ready_flag(struct relay_connection *conn)
1143 {
1144 struct relay_stream *stream, *tmp_stream;
1145
1146 pthread_mutex_lock(&conn->session->viewer_ready_lock);
1147 cds_list_for_each_entry_safe(stream, tmp_stream, &conn->recv_head,
1148 recv_list) {
1149 stream->viewer_ready = 1;
1150 cds_list_del(&stream->recv_list);
1151 }
1152 pthread_mutex_unlock(&conn->session->viewer_ready_lock);
1153 return;
1154 }
1155
1156 /*
1157 * Add a recv handle node to the connection recv list with the given stream
1158 * handle. A new node is allocated thus must be freed when the node is deleted
1159 * from the list.
1160 */
1161 static void queue_stream(struct relay_stream *stream,
1162 struct relay_connection *conn)
1163 {
1164 assert(conn);
1165 assert(stream);
1166
1167 cds_list_add(&stream->recv_list, &conn->recv_head);
1168 }
1169
1170 /*
1171 * relay_add_stream: allocate a new stream for a session
1172 */
1173 static
1174 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1175 struct relay_connection *conn)
1176 {
1177 int ret, send_ret;
1178 struct relay_session *session = conn->session;
1179 struct relay_stream *stream = NULL;
1180 struct lttcomm_relayd_status_stream reply;
1181 struct ctf_trace *trace;
1182
1183 if (!session || conn->version_check_done == 0) {
1184 ERR("Trying to add a stream before version check");
1185 ret = -1;
1186 goto end_no_session;
1187 }
1188
1189 stream = zmalloc(sizeof(struct relay_stream));
1190 if (stream == NULL) {
1191 PERROR("relay stream zmalloc");
1192 ret = -1;
1193 goto end_no_session;
1194 }
1195
1196 switch (conn->minor) {
1197 case 1: /* LTTng sessiond 2.1 */
1198 ret = cmd_recv_stream_2_1(conn, stream);
1199 break;
1200 case 2: /* LTTng sessiond 2.2 */
1201 default:
1202 ret = cmd_recv_stream_2_2(conn, stream);
1203 break;
1204 }
1205 if (ret < 0) {
1206 goto err_free_stream;
1207 }
1208
1209 rcu_read_lock();
1210 stream->stream_handle = ++last_relay_stream_id;
1211 stream->prev_seq = -1ULL;
1212 stream->session_id = session->id;
1213 stream->index_fd = -1;
1214 stream->read_index_fd = -1;
1215 stream->ctf_stream_id = -1ULL;
1216 lttng_ht_node_init_u64(&stream->node, stream->stream_handle);
1217 pthread_mutex_init(&stream->lock, NULL);
1218
1219 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
1220 if (ret < 0) {
1221 ERR("relay creating output directory");
1222 goto end;
1223 }
1224
1225 /*
1226 * No need to use run_as API here because whatever we receives, the relayd
1227 * uses its own credentials for the stream files.
1228 */
1229 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1230 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1231 if (ret < 0) {
1232 ERR("Create output file");
1233 goto end;
1234 }
1235 stream->fd = ret;
1236 if (stream->tracefile_size) {
1237 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1238 } else {
1239 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1240 }
1241
1242 trace = ctf_trace_find_by_path(session->ctf_traces_ht, stream->path_name);
1243 if (!trace) {
1244 trace = ctf_trace_create(stream->path_name);
1245 if (!trace) {
1246 ret = -1;
1247 goto end;
1248 }
1249 ctf_trace_add(session->ctf_traces_ht, trace);
1250 }
1251 ctf_trace_get_ref(trace);
1252
1253 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1254 stream->metadata_flag = 1;
1255 /* Assign quick reference to the metadata stream in the trace. */
1256 trace->metadata_stream = stream;
1257 }
1258
1259 /*
1260 * Add the stream in the recv list of the connection. Once the end stream
1261 * message is received, this list is emptied and streams are set with the
1262 * viewer ready flag.
1263 */
1264 queue_stream(stream, conn);
1265
1266 /*
1267 * Both in the ctf_trace object and the global stream ht since the data
1268 * side of the relayd does not have the concept of session.
1269 */
1270 lttng_ht_add_unique_u64(relay_streams_ht, &stream->node);
1271 cds_list_add_tail(&stream->trace_list, &trace->stream_list);
1272
1273 session->stream_count++;
1274
1275 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1276 stream->stream_handle);
1277
1278 end:
1279 memset(&reply, 0, sizeof(reply));
1280 reply.handle = htobe64(stream->stream_handle);
1281 /* send the session id to the client or a negative return code on error */
1282 if (ret < 0) {
1283 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1284 /* stream was not properly added to the ht, so free it */
1285 free(stream);
1286 } else {
1287 reply.ret_code = htobe32(LTTNG_OK);
1288 }
1289
1290 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1291 sizeof(struct lttcomm_relayd_status_stream), 0);
1292 if (send_ret < 0) {
1293 ERR("Relay sending stream id");
1294 ret = send_ret;
1295 }
1296 rcu_read_unlock();
1297
1298 end_no_session:
1299 return ret;
1300
1301 err_free_stream:
1302 free(stream->path_name);
1303 free(stream->channel_name);
1304 free(stream);
1305 return ret;
1306 }
1307
1308 /*
1309 * relay_close_stream: close a specific stream
1310 */
1311 static
1312 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1313 struct relay_connection *conn)
1314 {
1315 int ret, send_ret;
1316 struct relay_session *session = conn->session;
1317 struct lttcomm_relayd_close_stream stream_info;
1318 struct lttcomm_relayd_generic_reply reply;
1319 struct relay_stream *stream;
1320
1321 DBG("Close stream received");
1322
1323 if (!session || conn->version_check_done == 0) {
1324 ERR("Trying to close a stream before version check");
1325 ret = -1;
1326 goto end_no_session;
1327 }
1328
1329 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1330 sizeof(struct lttcomm_relayd_close_stream), 0);
1331 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1332 if (ret == 0) {
1333 /* Orderly shutdown. Not necessary to print an error. */
1334 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1335 } else {
1336 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1337 }
1338 ret = -1;
1339 goto end_no_session;
1340 }
1341
1342 rcu_read_lock();
1343 stream = stream_find_by_id(relay_streams_ht,
1344 be64toh(stream_info.stream_id));
1345 if (!stream) {
1346 ret = -1;
1347 goto end_unlock;
1348 }
1349
1350 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1351 stream->close_flag = 1;
1352 session->stream_count--;
1353
1354 /* Check if we can close it or else the data will do it. */
1355 try_close_stream(session, stream);
1356
1357 end_unlock:
1358 rcu_read_unlock();
1359
1360 memset(&reply, 0, sizeof(reply));
1361 if (ret < 0) {
1362 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1363 } else {
1364 reply.ret_code = htobe32(LTTNG_OK);
1365 }
1366 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1367 sizeof(struct lttcomm_relayd_generic_reply), 0);
1368 if (send_ret < 0) {
1369 ERR("Relay sending stream id");
1370 ret = send_ret;
1371 }
1372
1373 end_no_session:
1374 return ret;
1375 }
1376
1377 /*
1378 * relay_unknown_command: send -1 if received unknown command
1379 */
1380 static
1381 void relay_unknown_command(struct relay_connection *conn)
1382 {
1383 struct lttcomm_relayd_generic_reply reply;
1384 int ret;
1385
1386 memset(&reply, 0, sizeof(reply));
1387 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1388 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1389 sizeof(struct lttcomm_relayd_generic_reply), 0);
1390 if (ret < 0) {
1391 ERR("Relay sending unknown command");
1392 }
1393 }
1394
1395 /*
1396 * relay_start: send an acknowledgment to the client to tell if we are
1397 * ready to receive data. We are ready if a session is established.
1398 */
1399 static
1400 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1401 struct relay_connection *conn)
1402 {
1403 int ret = htobe32(LTTNG_OK);
1404 struct lttcomm_relayd_generic_reply reply;
1405 struct relay_session *session = conn->session;
1406
1407 if (!session) {
1408 DBG("Trying to start the streaming without a session established");
1409 ret = htobe32(LTTNG_ERR_UNK);
1410 }
1411
1412 memset(&reply, 0, sizeof(reply));
1413 reply.ret_code = ret;
1414 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1415 sizeof(struct lttcomm_relayd_generic_reply), 0);
1416 if (ret < 0) {
1417 ERR("Relay sending start ack");
1418 }
1419
1420 return ret;
1421 }
1422
1423 /*
1424 * Append padding to the file pointed by the file descriptor fd.
1425 */
1426 static int write_padding_to_file(int fd, uint32_t size)
1427 {
1428 ssize_t ret = 0;
1429 char *zeros;
1430
1431 if (size == 0) {
1432 goto end;
1433 }
1434
1435 zeros = zmalloc(size);
1436 if (zeros == NULL) {
1437 PERROR("zmalloc zeros for padding");
1438 ret = -1;
1439 goto end;
1440 }
1441
1442 ret = lttng_write(fd, zeros, size);
1443 if (ret < size) {
1444 PERROR("write padding to file");
1445 }
1446
1447 free(zeros);
1448
1449 end:
1450 return ret;
1451 }
1452
1453 /*
1454 * relay_recv_metadata: receive the metada for the session.
1455 */
1456 static
1457 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1458 struct relay_connection *conn)
1459 {
1460 int ret = htobe32(LTTNG_OK);
1461 ssize_t size_ret;
1462 struct relay_session *session = conn->session;
1463 struct lttcomm_relayd_metadata_payload *metadata_struct;
1464 struct relay_stream *metadata_stream;
1465 uint64_t data_size, payload_size;
1466 struct ctf_trace *ctf_trace;
1467
1468 if (!session) {
1469 ERR("Metadata sent before version check");
1470 ret = -1;
1471 goto end;
1472 }
1473
1474 data_size = payload_size = be64toh(recv_hdr->data_size);
1475 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1476 ERR("Incorrect data size");
1477 ret = -1;
1478 goto end;
1479 }
1480 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1481
1482 if (data_buffer_size < data_size) {
1483 /* In case the realloc fails, we can free the memory */
1484 char *tmp_data_ptr;
1485
1486 tmp_data_ptr = realloc(data_buffer, data_size);
1487 if (!tmp_data_ptr) {
1488 ERR("Allocating data buffer");
1489 free(data_buffer);
1490 ret = -1;
1491 goto end;
1492 }
1493 data_buffer = tmp_data_ptr;
1494 data_buffer_size = data_size;
1495 }
1496 memset(data_buffer, 0, data_size);
1497 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1498 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
1499 if (ret < 0 || ret != data_size) {
1500 if (ret == 0) {
1501 /* Orderly shutdown. Not necessary to print an error. */
1502 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1503 } else {
1504 ERR("Relay didn't receive the whole metadata");
1505 }
1506 ret = -1;
1507 goto end;
1508 }
1509 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1510
1511 rcu_read_lock();
1512 metadata_stream = stream_find_by_id(relay_streams_ht,
1513 be64toh(metadata_struct->stream_id));
1514 if (!metadata_stream) {
1515 ret = -1;
1516 goto end_unlock;
1517 }
1518
1519 size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
1520 payload_size);
1521 if (size_ret < payload_size) {
1522 ERR("Relay error writing metadata on file");
1523 ret = -1;
1524 goto end_unlock;
1525 }
1526
1527 ret = write_padding_to_file(metadata_stream->fd,
1528 be32toh(metadata_struct->padding_size));
1529 if (ret < 0) {
1530 goto end_unlock;
1531 }
1532
1533 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1534 metadata_stream->path_name);
1535 assert(ctf_trace);
1536 ctf_trace->metadata_received +=
1537 payload_size + be32toh(metadata_struct->padding_size);
1538
1539 DBG2("Relay metadata written");
1540
1541 end_unlock:
1542 rcu_read_unlock();
1543 end:
1544 return ret;
1545 }
1546
1547 /*
1548 * relay_send_version: send relayd version number
1549 */
1550 static
1551 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1552 struct relay_connection *conn)
1553 {
1554 int ret;
1555 struct lttcomm_relayd_version reply, msg;
1556
1557 assert(conn);
1558
1559 conn->version_check_done = 1;
1560
1561 /* Get version from the other side. */
1562 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1563 if (ret < 0 || ret != sizeof(msg)) {
1564 if (ret == 0) {
1565 /* Orderly shutdown. Not necessary to print an error. */
1566 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1567 } else {
1568 ERR("Relay failed to receive the version values.");
1569 }
1570 ret = -1;
1571 goto end;
1572 }
1573
1574 memset(&reply, 0, sizeof(reply));
1575 reply.major = RELAYD_VERSION_COMM_MAJOR;
1576 reply.minor = RELAYD_VERSION_COMM_MINOR;
1577
1578 /* Major versions must be the same */
1579 if (reply.major != be32toh(msg.major)) {
1580 DBG("Incompatible major versions (%u vs %u), deleting session",
1581 reply.major, be32toh(msg.major));
1582 destroy_session(conn->session, conn->sessions_ht);
1583 ret = 0;
1584 goto end;
1585 }
1586
1587 conn->major = reply.major;
1588 /* We adapt to the lowest compatible version */
1589 if (reply.minor <= be32toh(msg.minor)) {
1590 conn->minor = reply.minor;
1591 } else {
1592 conn->minor = be32toh(msg.minor);
1593 }
1594
1595 reply.major = htobe32(reply.major);
1596 reply.minor = htobe32(reply.minor);
1597 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1598 sizeof(struct lttcomm_relayd_version), 0);
1599 if (ret < 0) {
1600 ERR("Relay sending version");
1601 }
1602
1603 DBG("Version check done using protocol %u.%u", conn->major,
1604 conn->minor);
1605
1606 end:
1607 return ret;
1608 }
1609
1610 /*
1611 * Check for data pending for a given stream id from the session daemon.
1612 */
1613 static
1614 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1615 struct relay_connection *conn)
1616 {
1617 struct relay_session *session = conn->session;
1618 struct lttcomm_relayd_data_pending msg;
1619 struct lttcomm_relayd_generic_reply reply;
1620 struct relay_stream *stream;
1621 int ret;
1622 uint64_t last_net_seq_num, stream_id;
1623
1624 DBG("Data pending command received");
1625
1626 if (!session || conn->version_check_done == 0) {
1627 ERR("Trying to check for data before version check");
1628 ret = -1;
1629 goto end_no_session;
1630 }
1631
1632 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1633 if (ret < sizeof(msg)) {
1634 if (ret == 0) {
1635 /* Orderly shutdown. Not necessary to print an error. */
1636 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1637 } else {
1638 ERR("Relay didn't receive valid data_pending struct size : %d",
1639 ret);
1640 }
1641 ret = -1;
1642 goto end_no_session;
1643 }
1644
1645 stream_id = be64toh(msg.stream_id);
1646 last_net_seq_num = be64toh(msg.last_net_seq_num);
1647
1648 rcu_read_lock();
1649 stream = stream_find_by_id(relay_streams_ht, stream_id);
1650 if (stream == NULL) {
1651 ret = -1;
1652 goto end_unlock;
1653 }
1654
1655 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1656 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1657 last_net_seq_num);
1658
1659 /* Avoid wrapping issue */
1660 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1661 /* Data has in fact been written and is NOT pending */
1662 ret = 0;
1663 } else {
1664 /* Data still being streamed thus pending */
1665 ret = 1;
1666 }
1667
1668 /* Pending check is now done. */
1669 stream->data_pending_check_done = 1;
1670
1671 end_unlock:
1672 rcu_read_unlock();
1673
1674 memset(&reply, 0, sizeof(reply));
1675 reply.ret_code = htobe32(ret);
1676 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1677 if (ret < 0) {
1678 ERR("Relay data pending ret code failed");
1679 }
1680
1681 end_no_session:
1682 return ret;
1683 }
1684
1685 /*
1686 * Wait for the control socket to reach a quiescent state.
1687 *
1688 * Note that for now, when receiving this command from the session daemon, this
1689 * means that every subsequent commands or data received on the control socket
1690 * has been handled. So, this is why we simply return OK here.
1691 */
1692 static
1693 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1694 struct relay_connection *conn)
1695 {
1696 int ret;
1697 uint64_t stream_id;
1698 struct relay_stream *stream;
1699 struct lttng_ht_iter iter;
1700 struct lttcomm_relayd_quiescent_control msg;
1701 struct lttcomm_relayd_generic_reply reply;
1702
1703 DBG("Checking quiescent state on control socket");
1704
1705 if (!conn->session || conn->version_check_done == 0) {
1706 ERR("Trying to check for data before version check");
1707 ret = -1;
1708 goto end_no_session;
1709 }
1710
1711 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1712 if (ret < sizeof(msg)) {
1713 if (ret == 0) {
1714 /* Orderly shutdown. Not necessary to print an error. */
1715 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1716 } else {
1717 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1718 ret);
1719 }
1720 ret = -1;
1721 goto end_no_session;
1722 }
1723
1724 stream_id = be64toh(msg.stream_id);
1725
1726 rcu_read_lock();
1727 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1728 node.node) {
1729 if (stream->stream_handle == stream_id) {
1730 stream->data_pending_check_done = 1;
1731 DBG("Relay quiescent control pending flag set to %" PRIu64,
1732 stream_id);
1733 break;
1734 }
1735 }
1736 rcu_read_unlock();
1737
1738 memset(&reply, 0, sizeof(reply));
1739 reply.ret_code = htobe32(LTTNG_OK);
1740 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1741 if (ret < 0) {
1742 ERR("Relay data quiescent control ret code failed");
1743 }
1744
1745 end_no_session:
1746 return ret;
1747 }
1748
1749 /*
1750 * Initialize a data pending command. This means that a client is about to ask
1751 * for data pending for each stream he/she holds. Simply iterate over all
1752 * streams of a session and set the data_pending_check_done flag.
1753 *
1754 * This command returns to the client a LTTNG_OK code.
1755 */
1756 static
1757 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1758 struct relay_connection *conn)
1759 {
1760 int ret;
1761 struct lttng_ht_iter iter;
1762 struct lttcomm_relayd_begin_data_pending msg;
1763 struct lttcomm_relayd_generic_reply reply;
1764 struct relay_stream *stream;
1765 uint64_t session_id;
1766
1767 assert(recv_hdr);
1768 assert(conn);
1769
1770 DBG("Init streams for data pending");
1771
1772 if (!conn->session || conn->version_check_done == 0) {
1773 ERR("Trying to check for data before version check");
1774 ret = -1;
1775 goto end_no_session;
1776 }
1777
1778 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1779 if (ret < sizeof(msg)) {
1780 if (ret == 0) {
1781 /* Orderly shutdown. Not necessary to print an error. */
1782 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1783 } else {
1784 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1785 ret);
1786 }
1787 ret = -1;
1788 goto end_no_session;
1789 }
1790
1791 session_id = be64toh(msg.session_id);
1792
1793 /*
1794 * Iterate over all streams to set the begin data pending flag. For now, the
1795 * streams are indexed by stream handle so we have to iterate over all
1796 * streams to find the one associated with the right session_id.
1797 */
1798 rcu_read_lock();
1799 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1800 node.node) {
1801 if (stream->session_id == session_id) {
1802 stream->data_pending_check_done = 0;
1803 DBG("Set begin data pending flag to stream %" PRIu64,
1804 stream->stream_handle);
1805 }
1806 }
1807 rcu_read_unlock();
1808
1809 memset(&reply, 0, sizeof(reply));
1810 /* All good, send back reply. */
1811 reply.ret_code = htobe32(LTTNG_OK);
1812
1813 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1814 if (ret < 0) {
1815 ERR("Relay begin data pending send reply failed");
1816 }
1817
1818 end_no_session:
1819 return ret;
1820 }
1821
1822 /*
1823 * End data pending command. This will check, for a given session id, if each
1824 * stream associated with it has its data_pending_check_done flag set. If not,
1825 * this means that the client lost track of the stream but the data is still
1826 * being streamed on our side. In this case, we inform the client that data is
1827 * inflight.
1828 *
1829 * Return to the client if there is data in flight or not with a ret_code.
1830 */
1831 static
1832 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1833 struct relay_connection *conn)
1834 {
1835 int ret;
1836 struct lttng_ht_iter iter;
1837 struct lttcomm_relayd_end_data_pending msg;
1838 struct lttcomm_relayd_generic_reply reply;
1839 struct relay_stream *stream;
1840 uint64_t session_id;
1841 uint32_t is_data_inflight = 0;
1842
1843 assert(recv_hdr);
1844 assert(conn);
1845
1846 DBG("End data pending command");
1847
1848 if (!conn->session || conn->version_check_done == 0) {
1849 ERR("Trying to check for data before version check");
1850 ret = -1;
1851 goto end_no_session;
1852 }
1853
1854 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1855 if (ret < sizeof(msg)) {
1856 if (ret == 0) {
1857 /* Orderly shutdown. Not necessary to print an error. */
1858 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1859 } else {
1860 ERR("Relay didn't receive valid end data_pending struct size: %d",
1861 ret);
1862 }
1863 ret = -1;
1864 goto end_no_session;
1865 }
1866
1867 session_id = be64toh(msg.session_id);
1868
1869 /* Iterate over all streams to see if the begin data pending flag is set. */
1870 rcu_read_lock();
1871 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1872 node.node) {
1873 if (stream->session_id == session_id &&
1874 !stream->data_pending_check_done && !stream->terminated_flag) {
1875 is_data_inflight = 1;
1876 DBG("Data is still in flight for stream %" PRIu64,
1877 stream->stream_handle);
1878 break;
1879 }
1880 }
1881 rcu_read_unlock();
1882
1883 memset(&reply, 0, sizeof(reply));
1884 /* All good, send back reply. */
1885 reply.ret_code = htobe32(is_data_inflight);
1886
1887 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1888 if (ret < 0) {
1889 ERR("Relay end data pending send reply failed");
1890 }
1891
1892 end_no_session:
1893 return ret;
1894 }
1895
1896 /*
1897 * Receive an index for a specific stream.
1898 *
1899 * Return 0 on success else a negative value.
1900 */
1901 static
1902 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1903 struct relay_connection *conn)
1904 {
1905 int ret, send_ret, index_created = 0;
1906 struct relay_session *session = conn->session;
1907 struct lttcomm_relayd_index index_info;
1908 struct relay_index *index, *wr_index = NULL;
1909 struct lttcomm_relayd_generic_reply reply;
1910 struct relay_stream *stream;
1911 uint64_t net_seq_num;
1912
1913 assert(conn);
1914
1915 DBG("Relay receiving index");
1916
1917 if (!session || conn->version_check_done == 0) {
1918 ERR("Trying to close a stream before version check");
1919 ret = -1;
1920 goto end_no_session;
1921 }
1922
1923 ret = conn->sock->ops->recvmsg(conn->sock, &index_info,
1924 sizeof(index_info), 0);
1925 if (ret < sizeof(index_info)) {
1926 if (ret == 0) {
1927 /* Orderly shutdown. Not necessary to print an error. */
1928 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1929 } else {
1930 ERR("Relay didn't receive valid index struct size : %d", ret);
1931 }
1932 ret = -1;
1933 goto end_no_session;
1934 }
1935
1936 net_seq_num = be64toh(index_info.net_seq_num);
1937
1938 rcu_read_lock();
1939 stream = stream_find_by_id(relay_streams_ht,
1940 be64toh(index_info.relay_stream_id));
1941 if (!stream) {
1942 ret = -1;
1943 goto end_rcu_unlock;
1944 }
1945
1946 /* Live beacon handling */
1947 if (index_info.packet_size == 0) {
1948 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1949
1950 /*
1951 * Only flag a stream inactive when it has already received data
1952 * and no indexes are in flight.
1953 */
1954 if (stream->total_index_received > 0 && stream->indexes_in_flight == 0) {
1955 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1956 }
1957 ret = 0;
1958 goto end_rcu_unlock;
1959 } else {
1960 stream->beacon_ts_end = -1ULL;
1961 }
1962
1963 index = relay_index_find(stream->stream_handle, net_seq_num);
1964 if (!index) {
1965 /* A successful creation will add the object to the HT. */
1966 index = relay_index_create(stream->stream_handle, net_seq_num);
1967 if (!index) {
1968 goto end_rcu_unlock;
1969 }
1970 index_created = 1;
1971 stream->indexes_in_flight++;
1972 }
1973
1974 copy_index_control_data(index, &index_info);
1975 if (stream->ctf_stream_id == -1ULL) {
1976 stream->ctf_stream_id = be64toh(index_info.stream_id);
1977 }
1978
1979 if (index_created) {
1980 /*
1981 * Try to add the relay index object to the hash table. If an object
1982 * already exist, destroy back the index created, set the data in this
1983 * object and write it on disk.
1984 */
1985 relay_index_add(index, &wr_index);
1986 if (wr_index) {
1987 copy_index_control_data(wr_index, &index_info);
1988 free(index);
1989 }
1990 } else {
1991 /* The index already exists so write it on disk. */
1992 wr_index = index;
1993 }
1994
1995 /* Do we have a writable ready index to write on disk. */
1996 if (wr_index) {
1997 ret = relay_index_write(wr_index->fd, wr_index);
1998 if (ret < 0) {
1999 goto end_rcu_unlock;
2000 }
2001 stream->total_index_received++;
2002 stream->indexes_in_flight--;
2003 assert(stream->indexes_in_flight >= 0);
2004 }
2005
2006 end_rcu_unlock:
2007 rcu_read_unlock();
2008
2009 memset(&reply, 0, sizeof(reply));
2010 if (ret < 0) {
2011 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2012 } else {
2013 reply.ret_code = htobe32(LTTNG_OK);
2014 }
2015 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2016 if (send_ret < 0) {
2017 ERR("Relay sending close index id reply");
2018 ret = send_ret;
2019 }
2020
2021 end_no_session:
2022 return ret;
2023 }
2024
2025 /*
2026 * Receive the streams_sent message.
2027 *
2028 * Return 0 on success else a negative value.
2029 */
2030 static
2031 int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
2032 struct relay_connection *conn)
2033 {
2034 int ret, send_ret;
2035 struct lttcomm_relayd_generic_reply reply;
2036
2037 assert(conn);
2038
2039 DBG("Relay receiving streams_sent");
2040
2041 if (!conn->session || conn->version_check_done == 0) {
2042 ERR("Trying to close a stream before version check");
2043 ret = -1;
2044 goto end_no_session;
2045 }
2046
2047 /*
2048 * Flag every pending stream in the connection recv list that they are
2049 * ready to be used by the viewer.
2050 */
2051 set_viewer_ready_flag(conn);
2052
2053 /*
2054 * Inform the viewer that there are new streams in the session.
2055 */
2056 if (conn->session->viewer_refcount) {
2057 uatomic_set(&conn->session->new_streams, 1);
2058 }
2059
2060 memset(&reply, 0, sizeof(reply));
2061 reply.ret_code = htobe32(LTTNG_OK);
2062 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2063 if (send_ret < 0) {
2064 ERR("Relay sending sent_stream reply");
2065 ret = send_ret;
2066 } else {
2067 /* Success. */
2068 ret = 0;
2069 }
2070
2071 end_no_session:
2072 return ret;
2073 }
2074
2075 /*
2076 * Process the commands received on the control socket
2077 */
2078 static
2079 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
2080 struct relay_connection *conn)
2081 {
2082 int ret = 0;
2083
2084 switch (be32toh(recv_hdr->cmd)) {
2085 case RELAYD_CREATE_SESSION:
2086 ret = relay_create_session(recv_hdr, conn);
2087 break;
2088 case RELAYD_ADD_STREAM:
2089 ret = relay_add_stream(recv_hdr, conn);
2090 break;
2091 case RELAYD_START_DATA:
2092 ret = relay_start(recv_hdr, conn);
2093 break;
2094 case RELAYD_SEND_METADATA:
2095 ret = relay_recv_metadata(recv_hdr, conn);
2096 break;
2097 case RELAYD_VERSION:
2098 ret = relay_send_version(recv_hdr, conn);
2099 break;
2100 case RELAYD_CLOSE_STREAM:
2101 ret = relay_close_stream(recv_hdr, conn);
2102 break;
2103 case RELAYD_DATA_PENDING:
2104 ret = relay_data_pending(recv_hdr, conn);
2105 break;
2106 case RELAYD_QUIESCENT_CONTROL:
2107 ret = relay_quiescent_control(recv_hdr, conn);
2108 break;
2109 case RELAYD_BEGIN_DATA_PENDING:
2110 ret = relay_begin_data_pending(recv_hdr, conn);
2111 break;
2112 case RELAYD_END_DATA_PENDING:
2113 ret = relay_end_data_pending(recv_hdr, conn);
2114 break;
2115 case RELAYD_SEND_INDEX:
2116 ret = relay_recv_index(recv_hdr, conn);
2117 break;
2118 case RELAYD_STREAMS_SENT:
2119 ret = relay_streams_sent(recv_hdr, conn);
2120 break;
2121 case RELAYD_UPDATE_SYNC_INFO:
2122 default:
2123 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2124 relay_unknown_command(conn);
2125 ret = -1;
2126 goto end;
2127 }
2128
2129 end:
2130 return ret;
2131 }
2132
2133 /*
2134 * Handle index for a data stream.
2135 *
2136 * RCU read side lock MUST be acquired.
2137 *
2138 * Return 0 on success else a negative value.
2139 */
2140 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2141 int rotate_index)
2142 {
2143 int ret = 0, index_created = 0;
2144 uint64_t stream_id, data_offset;
2145 struct relay_index *index, *wr_index = NULL;
2146
2147 assert(stream);
2148
2149 stream_id = stream->stream_handle;
2150 /* Get data offset because we are about to update the index. */
2151 data_offset = htobe64(stream->tracefile_size_current);
2152
2153 /*
2154 * Lookup for an existing index for that stream id/sequence number. If on
2155 * exists, the control thread already received the data for it thus we need
2156 * to write it on disk.
2157 */
2158 index = relay_index_find(stream_id, net_seq_num);
2159 if (!index) {
2160 /* A successful creation will add the object to the HT. */
2161 index = relay_index_create(stream_id, net_seq_num);
2162 if (!index) {
2163 ret = -1;
2164 goto error;
2165 }
2166 index_created = 1;
2167 stream->indexes_in_flight++;
2168 }
2169
2170 if (rotate_index || stream->index_fd < 0) {
2171 index->to_close_fd = stream->index_fd;
2172 ret = index_create_file(stream->path_name, stream->channel_name,
2173 relayd_uid, relayd_gid, stream->tracefile_size,
2174 stream->tracefile_count_current);
2175 if (ret < 0) {
2176 /* This will close the stream's index fd if one. */
2177 relay_index_free_safe(index);
2178 goto error;
2179 }
2180 stream->index_fd = ret;
2181 }
2182 index->fd = stream->index_fd;
2183 index->index_data.offset = data_offset;
2184
2185 if (index_created) {
2186 /*
2187 * Try to add the relay index object to the hash table. If an object
2188 * already exist, destroy back the index created and set the data.
2189 */
2190 relay_index_add(index, &wr_index);
2191 if (wr_index) {
2192 /* Copy back data from the created index. */
2193 wr_index->fd = index->fd;
2194 wr_index->to_close_fd = index->to_close_fd;
2195 wr_index->index_data.offset = data_offset;
2196 free(index);
2197 }
2198 } else {
2199 /* The index already exists so write it on disk. */
2200 wr_index = index;
2201 }
2202
2203 /* Do we have a writable ready index to write on disk. */
2204 if (wr_index) {
2205 ret = relay_index_write(wr_index->fd, wr_index);
2206 if (ret < 0) {
2207 goto error;
2208 }
2209 stream->total_index_received++;
2210 stream->indexes_in_flight--;
2211 assert(stream->indexes_in_flight >= 0);
2212 }
2213
2214 error:
2215 return ret;
2216 }
2217
2218 /*
2219 * relay_process_data: Process the data received on the data socket
2220 */
2221 static
2222 int relay_process_data(struct relay_connection *conn)
2223 {
2224 int ret = 0, rotate_index = 0;
2225 ssize_t size_ret;
2226 struct relay_stream *stream;
2227 struct lttcomm_relayd_data_hdr data_hdr;
2228 uint64_t stream_id;
2229 uint64_t net_seq_num;
2230 uint32_t data_size;
2231 struct relay_session *session;
2232
2233 assert(conn);
2234
2235 ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr,
2236 sizeof(struct lttcomm_relayd_data_hdr), 0);
2237 if (ret <= 0) {
2238 if (ret == 0) {
2239 /* Orderly shutdown. Not necessary to print an error. */
2240 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2241 } else {
2242 ERR("Unable to receive data header on sock %d", conn->sock->fd);
2243 }
2244 ret = -1;
2245 goto end;
2246 }
2247
2248 stream_id = be64toh(data_hdr.stream_id);
2249
2250 rcu_read_lock();
2251 stream = stream_find_by_id(relay_streams_ht, stream_id);
2252 if (!stream) {
2253 ret = -1;
2254 goto end_rcu_unlock;
2255 }
2256
2257 session = session_find_by_id(conn->sessions_ht, stream->session_id);
2258 assert(session);
2259
2260 data_size = be32toh(data_hdr.data_size);
2261 if (data_buffer_size < data_size) {
2262 char *tmp_data_ptr;
2263
2264 tmp_data_ptr = realloc(data_buffer, data_size);
2265 if (!tmp_data_ptr) {
2266 ERR("Allocating data buffer");
2267 free(data_buffer);
2268 ret = -1;
2269 goto end_rcu_unlock;
2270 }
2271 data_buffer = tmp_data_ptr;
2272 data_buffer_size = data_size;
2273 }
2274 memset(data_buffer, 0, data_size);
2275
2276 net_seq_num = be64toh(data_hdr.net_seq_num);
2277
2278 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2279 data_size, stream_id, net_seq_num);
2280 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
2281 if (ret <= 0) {
2282 if (ret == 0) {
2283 /* Orderly shutdown. Not necessary to print an error. */
2284 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2285 }
2286 ret = -1;
2287 goto end_rcu_unlock;
2288 }
2289
2290 /* Check if a rotation is needed. */
2291 if (stream->tracefile_size > 0 &&
2292 (stream->tracefile_size_current + data_size) >
2293 stream->tracefile_size) {
2294 struct relay_viewer_stream *vstream;
2295 uint64_t new_id;
2296
2297 new_id = (stream->tracefile_count_current + 1) %
2298 stream->tracefile_count;
2299 /*
2300 * When we wrap-around back to 0, we start overwriting old
2301 * trace data.
2302 */
2303 if (!stream->tracefile_overwrite && new_id == 0) {
2304 stream->tracefile_overwrite = 1;
2305 }
2306 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
2307 if (stream->tracefile_overwrite) {
2308 stream->oldest_tracefile_id =
2309 (stream->oldest_tracefile_id + 1) %
2310 stream->tracefile_count;
2311 }
2312 vstream = viewer_stream_find_by_id(stream->stream_handle);
2313 if (vstream) {
2314 /*
2315 * The viewer is reading a file about to be
2316 * overwritten. Close the FDs it is
2317 * currently using and let it handle the fault.
2318 */
2319 if (vstream->tracefile_count_current == new_id) {
2320 pthread_mutex_lock(&vstream->overwrite_lock);
2321 vstream->abort_flag = 1;
2322 pthread_mutex_unlock(&vstream->overwrite_lock);
2323 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2324 stream->channel_name, new_id);
2325 } else if (vstream->tracefile_count_current ==
2326 stream->tracefile_count_current) {
2327 /*
2328 * The reader and writer were in the
2329 * same trace file, inform the viewer
2330 * that no new index will ever be added
2331 * to this file.
2332 */
2333 vstream->close_write_flag = 1;
2334 }
2335 }
2336 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2337 stream->tracefile_size, stream->tracefile_count,
2338 relayd_uid, relayd_gid, stream->fd,
2339 &(stream->tracefile_count_current), &stream->fd);
2340 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
2341 if (ret < 0) {
2342 ERR("Rotating stream output file");
2343 goto end_rcu_unlock;
2344 }
2345 /* Reset current size because we just perform a stream rotation. */
2346 stream->tracefile_size_current = 0;
2347 rotate_index = 1;
2348 }
2349
2350 /*
2351 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2352 * index are NOT supported.
2353 */
2354 if (session->minor >= 4 && !session->snapshot) {
2355 ret = handle_index_data(stream, net_seq_num, rotate_index);
2356 if (ret < 0) {
2357 goto end_rcu_unlock;
2358 }
2359 }
2360
2361 /* Write data to stream output fd. */
2362 size_ret = lttng_write(stream->fd, data_buffer, data_size);
2363 if (size_ret < data_size) {
2364 ERR("Relay error writing data to file");
2365 ret = -1;
2366 goto end_rcu_unlock;
2367 }
2368
2369 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2370 ret, stream->stream_handle);
2371
2372 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2373 if (ret < 0) {
2374 goto end_rcu_unlock;
2375 }
2376 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2377
2378 stream->prev_seq = net_seq_num;
2379
2380 try_close_stream(session, stream);
2381
2382 end_rcu_unlock:
2383 rcu_read_unlock();
2384 end:
2385 return ret;
2386 }
2387
2388 static
2389 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2390 {
2391 int ret;
2392
2393 assert(events);
2394
2395 (void) lttng_poll_del(events, pollfd);
2396
2397 ret = close(pollfd);
2398 if (ret < 0) {
2399 ERR("Closing pollfd %d", pollfd);
2400 }
2401 }
2402
2403 static void destroy_connection(struct lttng_ht *relay_connections_ht,
2404 struct relay_connection *conn)
2405 {
2406 assert(relay_connections_ht);
2407 assert(conn);
2408
2409 connection_delete(relay_connections_ht, conn);
2410
2411 /* For the control socket, we try to destroy the session. */
2412 if (conn->type == RELAY_CONTROL && conn->session) {
2413 destroy_session(conn->session, conn->sessions_ht);
2414 }
2415
2416 connection_destroy(conn);
2417 }
2418
2419 /*
2420 * This thread does the actual work
2421 */
2422 static
2423 void *relay_thread_worker(void *data)
2424 {
2425 int ret, err = -1, last_seen_data_fd = -1;
2426 uint32_t nb_fd;
2427 struct relay_connection *conn;
2428 struct lttng_poll_event events;
2429 struct lttng_ht *relay_connections_ht;
2430 struct lttng_ht_iter iter;
2431 struct lttcomm_relayd_hdr recv_hdr;
2432 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2433 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2434 struct relay_index *index;
2435
2436 DBG("[thread] Relay worker started");
2437
2438 rcu_register_thread();
2439
2440 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2441
2442 if (testpoint(relayd_thread_worker)) {
2443 goto error_testpoint;
2444 }
2445
2446 health_code_update();
2447
2448 /* table of connections indexed on socket */
2449 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2450 if (!relay_connections_ht) {
2451 goto relay_connections_ht_error;
2452 }
2453
2454 /* Tables of received indexes indexed by index handle and net_seq_num. */
2455 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2456 if (!indexes_ht) {
2457 goto indexes_ht_error;
2458 }
2459
2460 ret = create_thread_poll_set(&events, 2);
2461 if (ret < 0) {
2462 goto error_poll_create;
2463 }
2464
2465 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2466 if (ret < 0) {
2467 goto error;
2468 }
2469
2470 restart:
2471 while (1) {
2472 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2473
2474 health_code_update();
2475
2476 /* Infinite blocking call, waiting for transmission */
2477 DBG3("Relayd worker thread polling...");
2478 health_poll_entry();
2479 ret = lttng_poll_wait(&events, -1);
2480 health_poll_exit();
2481 if (ret < 0) {
2482 /*
2483 * Restart interrupted system call.
2484 */
2485 if (errno == EINTR) {
2486 goto restart;
2487 }
2488 goto error;
2489 }
2490
2491 nb_fd = ret;
2492
2493 /*
2494 * Process control. The control connection is prioritised so we don't
2495 * starve it with high throughout put tracing data on the data
2496 * connection.
2497 */
2498 for (i = 0; i < nb_fd; i++) {
2499 /* Fetch once the poll data */
2500 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2501 int pollfd = LTTNG_POLL_GETFD(&events, i);
2502
2503 health_code_update();
2504
2505 /* Thread quit pipe has been closed. Killing thread. */
2506 ret = check_thread_quit_pipe(pollfd, revents);
2507 if (ret) {
2508 err = 0;
2509 goto exit;
2510 }
2511
2512 /* Inspect the relay conn pipe for new connection */
2513 if (pollfd == relay_conn_pipe[0]) {
2514 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2515 ERR("Relay connection pipe error");
2516 goto error;
2517 } else if (revents & LPOLLIN) {
2518 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
2519 if (ret < 0) {
2520 goto error;
2521 }
2522 conn->sessions_ht = sessions_ht;
2523 connection_init(conn);
2524 lttng_poll_add(&events, conn->sock->fd,
2525 LPOLLIN | LPOLLRDHUP);
2526 rcu_read_lock();
2527 lttng_ht_add_unique_ulong(relay_connections_ht,
2528 &conn->sock_n);
2529 rcu_read_unlock();
2530 DBG("Connection socket %d added", conn->sock->fd);
2531 }
2532 } else {
2533 rcu_read_lock();
2534 conn = connection_find_by_sock(relay_connections_ht, pollfd);
2535 /* If not found, there is a synchronization issue. */
2536 assert(conn);
2537
2538 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2539 cleanup_connection_pollfd(&events, pollfd);
2540 destroy_connection(relay_connections_ht, conn);
2541 if (last_seen_data_fd == pollfd) {
2542 last_seen_data_fd = last_notdel_data_fd;
2543 }
2544 } else if (revents & LPOLLIN) {
2545 if (conn->type == RELAY_CONTROL) {
2546 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2547 sizeof(recv_hdr), 0);
2548 if (ret <= 0) {
2549 /* Connection closed */
2550 cleanup_connection_pollfd(&events, pollfd);
2551 destroy_connection(relay_connections_ht, conn);
2552 DBG("Control connection closed with %d", pollfd);
2553 } else {
2554 ret = relay_process_control(&recv_hdr, conn);
2555 if (ret < 0) {
2556 /* Clear the session on error. */
2557 cleanup_connection_pollfd(&events, pollfd);
2558 destroy_connection(relay_connections_ht, conn);
2559 DBG("Connection closed with %d", pollfd);
2560 }
2561 seen_control = 1;
2562 }
2563 } else {
2564 /*
2565 * Flag the last seen data fd not deleted. It will be
2566 * used as the last seen fd if any fd gets deleted in
2567 * this first loop.
2568 */
2569 last_notdel_data_fd = pollfd;
2570 }
2571 } else {
2572 ERR("Unknown poll events %u for sock %d", revents, pollfd);
2573 }
2574 rcu_read_unlock();
2575 }
2576 }
2577
2578 /*
2579 * The last loop handled a control request, go back to poll to make
2580 * sure we prioritise the control socket.
2581 */
2582 if (seen_control) {
2583 continue;
2584 }
2585
2586 if (last_seen_data_fd >= 0) {
2587 for (i = 0; i < nb_fd; i++) {
2588 int pollfd = LTTNG_POLL_GETFD(&events, i);
2589
2590 health_code_update();
2591
2592 if (last_seen_data_fd == pollfd) {
2593 idx = i;
2594 break;
2595 }
2596 }
2597 }
2598
2599 /* Process data connection. */
2600 for (i = idx + 1; i < nb_fd; i++) {
2601 /* Fetch the poll data. */
2602 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2603 int pollfd = LTTNG_POLL_GETFD(&events, i);
2604
2605 health_code_update();
2606
2607 /* Skip the command pipe. It's handled in the first loop. */
2608 if (pollfd == relay_conn_pipe[0]) {
2609 continue;
2610 }
2611
2612 if (revents) {
2613 rcu_read_lock();
2614 conn = connection_find_by_sock(relay_connections_ht, pollfd);
2615 if (!conn) {
2616 /* Skip it. Might be removed before. */
2617 rcu_read_unlock();
2618 continue;
2619 }
2620
2621 if (revents & LPOLLIN) {
2622 if (conn->type != RELAY_DATA) {
2623 rcu_read_unlock();
2624 continue;
2625 }
2626
2627 ret = relay_process_data(conn);
2628 /* Connection closed */
2629 if (ret < 0) {
2630 cleanup_connection_pollfd(&events, pollfd);
2631 destroy_connection(relay_connections_ht, conn);
2632 DBG("Data connection closed with %d", pollfd);
2633 /*
2634 * Every goto restart call sets the last seen fd where
2635 * here we don't really care since we gracefully
2636 * continue the loop after the connection is deleted.
2637 */
2638 } else {
2639 /* Keep last seen port. */
2640 last_seen_data_fd = pollfd;
2641 rcu_read_unlock();
2642 goto restart;
2643 }
2644 }
2645 rcu_read_unlock();
2646 }
2647 }
2648 last_seen_data_fd = -1;
2649 }
2650
2651 /* Normal exit, no error */
2652 ret = 0;
2653
2654 exit:
2655 error:
2656 lttng_poll_clean(&events);
2657
2658 /* Cleanup reamaining connection object. */
2659 rcu_read_lock();
2660 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2661 sock_n.node) {
2662 health_code_update();
2663 destroy_connection(relay_connections_ht, conn);
2664 }
2665 rcu_read_unlock();
2666 error_poll_create:
2667 rcu_read_lock();
2668 cds_lfht_for_each_entry(indexes_ht->ht, &iter.iter, index,
2669 index_n.node) {
2670 health_code_update();
2671 relay_index_delete(index);
2672 relay_index_free_safe(index);
2673 }
2674 rcu_read_unlock();
2675 lttng_ht_destroy(indexes_ht);
2676 indexes_ht_error:
2677 lttng_ht_destroy(relay_connections_ht);
2678 relay_connections_ht_error:
2679 /* Close relay conn pipes */
2680 utils_close_pipe(relay_conn_pipe);
2681 if (err) {
2682 DBG("Thread exited with error");
2683 }
2684 DBG("Worker thread cleanup complete");
2685 free(data_buffer);
2686 error_testpoint:
2687 if (err) {
2688 health_error();
2689 ERR("Health error occurred in %s", __func__);
2690 }
2691 health_unregister(health_relayd);
2692 rcu_unregister_thread();
2693 stop_threads();
2694 return NULL;
2695 }
2696
2697 /*
2698 * Create the relay command pipe to wake thread_manage_apps.
2699 * Closed in cleanup().
2700 */
2701 static int create_relay_conn_pipe(void)
2702 {
2703 int ret;
2704
2705 ret = utils_create_pipe_cloexec(relay_conn_pipe);
2706
2707 return ret;
2708 }
2709
2710 /*
2711 * main
2712 */
2713 int main(int argc, char **argv)
2714 {
2715 int ret = 0;
2716 void *status;
2717 struct relay_local_data *relay_ctx;
2718
2719 /* Parse arguments */
2720 progname = argv[0];
2721 if ((ret = set_options(argc, argv)) < 0) {
2722 goto exit;
2723 }
2724
2725 if ((ret = set_signal_handler()) < 0) {
2726 goto exit;
2727 }
2728
2729 /* Try to create directory if -o, --output is specified. */
2730 if (opt_output_path) {
2731 if (*opt_output_path != '/') {
2732 ERR("Please specify an absolute path for -o, --output PATH");
2733 goto exit;
2734 }
2735
2736 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2737 if (ret < 0) {
2738 ERR("Unable to create %s", opt_output_path);
2739 goto exit;
2740 }
2741 }
2742
2743 /* Daemonize */
2744 if (opt_daemon || opt_background) {
2745 int i;
2746
2747 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2748 !opt_background);
2749 if (ret < 0) {
2750 goto exit;
2751 }
2752
2753 /*
2754 * We are in the child. Make sure all other file
2755 * descriptors are closed, in case we are called with
2756 * more opened file descriptors than the standard ones.
2757 */
2758 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2759 (void) close(i);
2760 }
2761 }
2762
2763 /* Create thread quit pipe */
2764 if ((ret = init_thread_quit_pipe()) < 0) {
2765 goto error;
2766 }
2767
2768 /* We need those values for the file/dir creation. */
2769 relayd_uid = getuid();
2770 relayd_gid = getgid();
2771
2772 /* Check if daemon is UID = 0 */
2773 if (relayd_uid == 0) {
2774 if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) {
2775 ERR("Need to be root to use ports < 1024");
2776 ret = -1;
2777 goto exit;
2778 }
2779 }
2780
2781 /* Setup the thread apps communication pipe. */
2782 if ((ret = create_relay_conn_pipe()) < 0) {
2783 goto exit;
2784 }
2785
2786 /* Init relay command queue. */
2787 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
2788
2789 /* Set up max poll set size */
2790 lttng_poll_set_max_size();
2791
2792 /* Initialize communication library */
2793 lttcomm_init();
2794 lttcomm_inet_init();
2795
2796 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2797 if (!relay_ctx) {
2798 PERROR("relay_ctx");
2799 goto exit;
2800 }
2801
2802 /* tables of sessions indexed by session ID */
2803 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2804 if (!relay_ctx->sessions_ht) {
2805 goto exit_relay_ctx_sessions;
2806 }
2807
2808 /* tables of streams indexed by stream ID */
2809 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2810 if (!relay_streams_ht) {
2811 goto exit_relay_ctx_streams;
2812 }
2813
2814 /* tables of streams indexed by stream ID */
2815 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2816 if (!viewer_streams_ht) {
2817 goto exit_relay_ctx_viewer_streams;
2818 }
2819
2820 /* Initialize thread health monitoring */
2821 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2822 if (!health_relayd) {
2823 PERROR("health_app_create error");
2824 goto exit_health_app_create;
2825 }
2826
2827 ret = utils_create_pipe(health_quit_pipe);
2828 if (ret < 0) {
2829 goto error_health_pipe;
2830 }
2831
2832 /* Create thread to manage the client socket */
2833 ret = pthread_create(&health_thread, NULL,
2834 thread_manage_health, (void *) NULL);
2835 if (ret != 0) {
2836 PERROR("pthread_create health");
2837 goto health_error;
2838 }
2839
2840 /* Setup the dispatcher thread */
2841 ret = pthread_create(&dispatcher_thread, NULL,
2842 relay_thread_dispatcher, (void *) NULL);
2843 if (ret != 0) {
2844 PERROR("pthread_create dispatcher");
2845 goto exit_dispatcher;
2846 }
2847
2848 /* Setup the worker thread */
2849 ret = pthread_create(&worker_thread, NULL,
2850 relay_thread_worker, (void *) relay_ctx);
2851 if (ret != 0) {
2852 PERROR("pthread_create worker");
2853 goto exit_worker;
2854 }
2855
2856 /* Setup the listener thread */
2857 ret = pthread_create(&listener_thread, NULL,
2858 relay_thread_listener, (void *) NULL);
2859 if (ret != 0) {
2860 PERROR("pthread_create listener");
2861 goto exit_listener;
2862 }
2863
2864 ret = live_start_threads(live_uri, relay_ctx);
2865 if (ret != 0) {
2866 ERR("Starting live viewer threads");
2867 goto exit_live;
2868 }
2869
2870 exit_live:
2871 ret = pthread_join(listener_thread, &status);
2872 if (ret != 0) {
2873 PERROR("pthread_join");
2874 goto error; /* join error, exit without cleanup */
2875 }
2876
2877 exit_listener:
2878 ret = pthread_join(worker_thread, &status);
2879 if (ret != 0) {
2880 PERROR("pthread_join");
2881 goto error; /* join error, exit without cleanup */
2882 }
2883
2884 exit_worker:
2885 ret = pthread_join(dispatcher_thread, &status);
2886 if (ret != 0) {
2887 PERROR("pthread_join");
2888 goto error; /* join error, exit without cleanup */
2889 }
2890
2891 exit_dispatcher:
2892 ret = pthread_join(health_thread, &status);
2893 if (ret != 0) {
2894 PERROR("pthread_join health thread");
2895 goto error; /* join error, exit without cleanup */
2896 }
2897
2898 /*
2899 * Stop live threads only after joining other threads.
2900 */
2901 live_stop_threads();
2902
2903 health_error:
2904 utils_close_pipe(health_quit_pipe);
2905
2906 error_health_pipe:
2907 health_app_destroy(health_relayd);
2908
2909 exit_health_app_create:
2910 lttng_ht_destroy(viewer_streams_ht);
2911
2912 exit_relay_ctx_viewer_streams:
2913 lttng_ht_destroy(relay_streams_ht);
2914
2915 exit_relay_ctx_streams:
2916 lttng_ht_destroy(relay_ctx->sessions_ht);
2917
2918 exit_relay_ctx_sessions:
2919 free(relay_ctx);
2920
2921 exit:
2922 cleanup();
2923 if (!ret) {
2924 exit(EXIT_SUCCESS);
2925 }
2926
2927 error:
2928 exit(EXIT_FAILURE);
2929 }
This page took 0.098848 seconds and 4 git commands to generate.