50dff9e5ba371abecb9c56269674073d5cd36f40
[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, retval = 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 retval = -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 retval = -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 }
373 retval = -1;
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 retval = -1;
389 goto exit;
390 }
391 }
392
393 /* assign default values */
394 if (control_uri == NULL) {
395 ret = asprintf(&default_address,
396 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
397 DEFAULT_NETWORK_CONTROL_PORT);
398 if (ret < 0) {
399 PERROR("asprintf default data address");
400 retval = -1;
401 goto exit;
402 }
403
404 ret = uri_parse(default_address, &control_uri);
405 free(default_address);
406 if (ret < 0) {
407 ERR("Invalid control URI specified");
408 retval = -1;
409 goto exit;
410 }
411 }
412 if (data_uri == NULL) {
413 ret = asprintf(&default_address,
414 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
415 DEFAULT_NETWORK_DATA_PORT);
416 if (ret < 0) {
417 PERROR("asprintf default data address");
418 retval = -1;
419 goto exit;
420 }
421
422 ret = uri_parse(default_address, &data_uri);
423 free(default_address);
424 if (ret < 0) {
425 ERR("Invalid data URI specified");
426 retval = -1;
427 goto exit;
428 }
429 }
430 if (live_uri == NULL) {
431 ret = asprintf(&default_address,
432 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
433 DEFAULT_NETWORK_VIEWER_PORT);
434 if (ret < 0) {
435 PERROR("asprintf default viewer control address");
436 retval = -1;
437 goto exit;
438 }
439
440 ret = uri_parse(default_address, &live_uri);
441 free(default_address);
442 if (ret < 0) {
443 ERR("Invalid viewer control URI specified");
444 retval = -1;
445 goto exit;
446 }
447 }
448
449 exit:
450 free(optstring);
451 return retval;
452 }
453
454 /*
455 * Cleanup the daemon
456 */
457 static
458 void relayd_cleanup(struct relay_local_data *relay_ctx)
459 {
460 DBG("Cleaning up");
461
462 if (viewer_streams_ht)
463 lttng_ht_destroy(viewer_streams_ht);
464 if (relay_streams_ht)
465 lttng_ht_destroy(relay_streams_ht);
466 if (relay_ctx && relay_ctx->sessions_ht)
467 lttng_ht_destroy(relay_ctx->sessions_ht);
468 free(relay_ctx);
469
470 /* free the dynamically allocated opt_output_path */
471 free(opt_output_path);
472
473 /* Close thread quit pipes */
474 utils_close_pipe(thread_quit_pipe);
475
476 uri_free(control_uri);
477 uri_free(data_uri);
478 /* Live URI is freed in the live thread. */
479
480 if (tracing_group_name_override) {
481 free((void *) tracing_group_name);
482 }
483 }
484
485 /*
486 * Write to writable pipe used to notify a thread.
487 */
488 static
489 int notify_thread_pipe(int wpipe)
490 {
491 ssize_t ret;
492
493 ret = lttng_write(wpipe, "!", 1);
494 if (ret < 1) {
495 PERROR("write poll pipe");
496 goto end;
497 }
498 ret = 0;
499 end:
500 return ret;
501 }
502
503 static
504 int notify_health_quit_pipe(int *pipe)
505 {
506 ssize_t ret;
507
508 ret = lttng_write(pipe[1], "4", 1);
509 if (ret < 1) {
510 PERROR("write relay health quit");
511 goto end;
512 }
513 ret = 0;
514 end:
515 return ret;
516 }
517
518 /*
519 * Stop all relayd and relayd-live threads.
520 */
521 int lttng_relay_stop_threads(void)
522 {
523 int retval = 0;
524
525 /* Stopping all threads */
526 DBG("Terminating all threads");
527 if (notify_thread_pipe(thread_quit_pipe[1])) {
528 ERR("write error on thread quit pipe");
529 retval = -1;
530 }
531
532 if (notify_health_quit_pipe(health_quit_pipe)) {
533 ERR("write error on health quit pipe");
534 }
535
536 /* Dispatch thread */
537 CMM_STORE_SHARED(dispatch_thread_exit, 1);
538 futex_nto1_wake(&relay_conn_queue.futex);
539
540 if (relayd_live_stop()) {
541 ERR("Error stopping live threads");
542 retval = -1;
543 }
544 return retval;
545 }
546
547 /*
548 * Signal handler for the daemon
549 *
550 * Simply stop all worker threads, leaving main() return gracefully after
551 * joining all threads and calling cleanup().
552 */
553 static
554 void sighandler(int sig)
555 {
556 switch (sig) {
557 case SIGPIPE:
558 DBG("SIGPIPE caught");
559 return;
560 case SIGINT:
561 DBG("SIGINT caught");
562 if (lttng_relay_stop_threads()) {
563 ERR("Error stopping threads");
564 }
565 break;
566 case SIGTERM:
567 DBG("SIGTERM caught");
568 if (lttng_relay_stop_threads()) {
569 ERR("Error stopping threads");
570 }
571 break;
572 case SIGUSR1:
573 CMM_STORE_SHARED(recv_child_signal, 1);
574 break;
575 default:
576 break;
577 }
578 }
579
580 /*
581 * Setup signal handler for :
582 * SIGINT, SIGTERM, SIGPIPE
583 */
584 static
585 int set_signal_handler(void)
586 {
587 int ret = 0;
588 struct sigaction sa;
589 sigset_t sigset;
590
591 if ((ret = sigemptyset(&sigset)) < 0) {
592 PERROR("sigemptyset");
593 return ret;
594 }
595
596 sa.sa_handler = sighandler;
597 sa.sa_mask = sigset;
598 sa.sa_flags = 0;
599 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
600 PERROR("sigaction");
601 return ret;
602 }
603
604 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
605 PERROR("sigaction");
606 return ret;
607 }
608
609 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
610 PERROR("sigaction");
611 return ret;
612 }
613
614 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
615 PERROR("sigaction");
616 return ret;
617 }
618
619 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
620
621 return ret;
622 }
623
624 void lttng_relay_notify_ready(void)
625 {
626 /* Notify the parent of the fork() process that we are ready. */
627 if (opt_daemon || opt_background) {
628 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
629 kill(child_ppid, SIGUSR1);
630 }
631 }
632 }
633
634 /*
635 * Init thread quit pipe.
636 *
637 * Return -1 on error or 0 if all pipes are created.
638 */
639 static
640 int init_thread_quit_pipe(void)
641 {
642 int ret;
643
644 ret = utils_create_pipe_cloexec(thread_quit_pipe);
645
646 return ret;
647 }
648
649 /*
650 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
651 */
652 static
653 int create_thread_poll_set(struct lttng_poll_event *events, int size)
654 {
655 int ret;
656
657 if (events == NULL || size == 0) {
658 ret = -1;
659 goto error;
660 }
661
662 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
663 if (ret < 0) {
664 goto error;
665 }
666
667 /* Add quit pipe */
668 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
669 if (ret < 0) {
670 goto error;
671 }
672
673 return 0;
674
675 error:
676 return ret;
677 }
678
679 /*
680 * Check if the thread quit pipe was triggered.
681 *
682 * Return 1 if it was triggered else 0;
683 */
684 static
685 int check_thread_quit_pipe(int fd, uint32_t events)
686 {
687 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
688 return 1;
689 }
690
691 return 0;
692 }
693
694 /*
695 * Create and init socket from uri.
696 */
697 static
698 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
699 {
700 int ret;
701 struct lttcomm_sock *sock = NULL;
702
703 sock = lttcomm_alloc_sock_from_uri(uri);
704 if (sock == NULL) {
705 ERR("Allocating socket");
706 goto error;
707 }
708
709 ret = lttcomm_create_sock(sock);
710 if (ret < 0) {
711 goto error;
712 }
713 DBG("Listening on sock %d", sock->fd);
714
715 ret = sock->ops->bind(sock);
716 if (ret < 0) {
717 goto error;
718 }
719
720 ret = sock->ops->listen(sock, -1);
721 if (ret < 0) {
722 goto error;
723
724 }
725
726 return sock;
727
728 error:
729 if (sock) {
730 lttcomm_destroy_sock(sock);
731 }
732 return NULL;
733 }
734
735 /*
736 * Return nonzero if stream needs to be closed.
737 */
738 static
739 int close_stream_check(struct relay_stream *stream)
740 {
741 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
742 /*
743 * We are about to close the stream so set the data pending flag to 1
744 * which will make the end data pending command skip the stream which
745 * is now closed and ready. Note that after proceeding to a file close,
746 * the written file is ready for reading.
747 */
748 stream->data_pending_check_done = 1;
749 return 1;
750 }
751 return 0;
752 }
753
754 static void try_close_stream(struct relay_session *session,
755 struct relay_stream *stream)
756 {
757 int ret;
758 struct ctf_trace *ctf_trace;
759
760 assert(session);
761 assert(stream);
762
763 if (!close_stream_check(stream)) {
764 /* Can't close it, not ready for that. */
765 goto end;
766 }
767
768 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
769 stream->path_name);
770 assert(ctf_trace);
771
772 pthread_mutex_lock(&session->viewer_ready_lock);
773 ctf_trace->invalid_flag = 1;
774 pthread_mutex_unlock(&session->viewer_ready_lock);
775
776 ret = stream_close(session, stream);
777 if (ret || session->snapshot) {
778 /* Already close thus the ctf trace is being or has been destroyed. */
779 goto end;
780 }
781
782 ctf_trace_try_destroy(session, ctf_trace);
783
784 end:
785 return;
786 }
787
788 /*
789 * This thread manages the listening for new connections on the network
790 */
791 static
792 void *relay_thread_listener(void *data)
793 {
794 int i, ret, pollfd, err = -1;
795 uint32_t revents, nb_fd;
796 struct lttng_poll_event events;
797 struct lttcomm_sock *control_sock, *data_sock;
798
799 DBG("[thread] Relay listener started");
800
801 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
802
803 health_code_update();
804
805 control_sock = relay_init_sock(control_uri);
806 if (!control_sock) {
807 goto error_sock_control;
808 }
809
810 data_sock = relay_init_sock(data_uri);
811 if (!data_sock) {
812 goto error_sock_relay;
813 }
814
815 /*
816 * Pass 3 as size here for the thread quit pipe, control and data socket.
817 */
818 ret = create_thread_poll_set(&events, 3);
819 if (ret < 0) {
820 goto error_create_poll;
821 }
822
823 /* Add the control socket */
824 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
825 if (ret < 0) {
826 goto error_poll_add;
827 }
828
829 /* Add the data socket */
830 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
831 if (ret < 0) {
832 goto error_poll_add;
833 }
834
835 lttng_relay_notify_ready();
836
837 if (testpoint(relayd_thread_listener)) {
838 goto error_testpoint;
839 }
840
841 while (1) {
842 health_code_update();
843
844 DBG("Listener accepting connections");
845
846 restart:
847 health_poll_entry();
848 ret = lttng_poll_wait(&events, -1);
849 health_poll_exit();
850 if (ret < 0) {
851 /*
852 * Restart interrupted system call.
853 */
854 if (errno == EINTR) {
855 goto restart;
856 }
857 goto error;
858 }
859
860 nb_fd = ret;
861
862 DBG("Relay new connection received");
863 for (i = 0; i < nb_fd; i++) {
864 health_code_update();
865
866 /* Fetch once the poll data */
867 revents = LTTNG_POLL_GETEV(&events, i);
868 pollfd = LTTNG_POLL_GETFD(&events, i);
869
870 if (!revents) {
871 /* No activity for this FD (poll implementation). */
872 continue;
873 }
874
875 /* Thread quit pipe has been closed. Killing thread. */
876 ret = check_thread_quit_pipe(pollfd, revents);
877 if (ret) {
878 err = 0;
879 goto exit;
880 }
881
882 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
883 ERR("socket poll error");
884 goto error;
885 } else if (revents & LPOLLIN) {
886 /*
887 * Get allocated in this thread, enqueued to a global queue,
888 * dequeued and freed in the worker thread.
889 */
890 int val = 1;
891 struct relay_connection *new_conn;
892 struct lttcomm_sock *newsock;
893
894 new_conn = connection_create();
895 if (!new_conn) {
896 goto error;
897 }
898
899 if (pollfd == data_sock->fd) {
900 new_conn->type = RELAY_DATA;
901 newsock = data_sock->ops->accept(data_sock);
902 DBG("Relay data connection accepted, socket %d",
903 newsock->fd);
904 } else {
905 assert(pollfd == control_sock->fd);
906 new_conn->type = RELAY_CONTROL;
907 newsock = control_sock->ops->accept(control_sock);
908 DBG("Relay control connection accepted, socket %d",
909 newsock->fd);
910 }
911 if (!newsock) {
912 PERROR("accepting sock");
913 connection_free(new_conn);
914 goto error;
915 }
916
917 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
918 sizeof(val));
919 if (ret < 0) {
920 PERROR("setsockopt inet");
921 lttcomm_destroy_sock(newsock);
922 connection_free(new_conn);
923 goto error;
924 }
925 new_conn->sock = newsock;
926
927 /* Enqueue request for the dispatcher thread. */
928 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
929 &new_conn->qnode);
930
931 /*
932 * Wake the dispatch queue futex. Implicit memory barrier with
933 * the exchange in cds_wfcq_enqueue.
934 */
935 futex_nto1_wake(&relay_conn_queue.futex);
936 }
937 }
938 }
939
940 exit:
941 error:
942 error_poll_add:
943 error_testpoint:
944 lttng_poll_clean(&events);
945 error_create_poll:
946 if (data_sock->fd >= 0) {
947 ret = data_sock->ops->close(data_sock);
948 if (ret) {
949 PERROR("close");
950 }
951 }
952 lttcomm_destroy_sock(data_sock);
953 error_sock_relay:
954 if (control_sock->fd >= 0) {
955 ret = control_sock->ops->close(control_sock);
956 if (ret) {
957 PERROR("close");
958 }
959 }
960 lttcomm_destroy_sock(control_sock);
961 error_sock_control:
962 if (err) {
963 health_error();
964 ERR("Health error occurred in %s", __func__);
965 }
966 health_unregister(health_relayd);
967 DBG("Relay listener thread cleanup complete");
968 lttng_relay_stop_threads();
969 return NULL;
970 }
971
972 /*
973 * This thread manages the dispatching of the requests to worker threads
974 */
975 static
976 void *relay_thread_dispatcher(void *data)
977 {
978 int err = -1;
979 ssize_t ret;
980 struct cds_wfcq_node *node;
981 struct relay_connection *new_conn = NULL;
982
983 DBG("[thread] Relay dispatcher started");
984
985 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
986
987 if (testpoint(relayd_thread_dispatcher)) {
988 goto error_testpoint;
989 }
990
991 health_code_update();
992
993 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
994 health_code_update();
995
996 /* Atomically prepare the queue futex */
997 futex_nto1_prepare(&relay_conn_queue.futex);
998
999 do {
1000 health_code_update();
1001
1002 /* Dequeue commands */
1003 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1004 &relay_conn_queue.tail);
1005 if (node == NULL) {
1006 DBG("Woken up but nothing in the relay command queue");
1007 /* Continue thread execution */
1008 break;
1009 }
1010 new_conn = caa_container_of(node, struct relay_connection, qnode);
1011
1012 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1013
1014 /*
1015 * Inform worker thread of the new request. This call is blocking
1016 * so we can be assured that the data will be read at some point in
1017 * time or wait to the end of the world :)
1018 */
1019 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1020 if (ret < 0) {
1021 PERROR("write connection pipe");
1022 connection_destroy(new_conn);
1023 goto error;
1024 }
1025 } while (node != NULL);
1026
1027 /* Futex wait on queue. Blocking call on futex() */
1028 health_poll_entry();
1029 futex_nto1_wait(&relay_conn_queue.futex);
1030 health_poll_exit();
1031 }
1032
1033 /* Normal exit, no error */
1034 err = 0;
1035
1036 error:
1037 error_testpoint:
1038 if (err) {
1039 health_error();
1040 ERR("Health error occurred in %s", __func__);
1041 }
1042 health_unregister(health_relayd);
1043 DBG("Dispatch thread dying");
1044 lttng_relay_stop_threads();
1045 return NULL;
1046 }
1047
1048 static void try_close_streams(struct relay_session *session)
1049 {
1050 struct ctf_trace *ctf_trace;
1051 struct lttng_ht_iter iter;
1052
1053 assert(session);
1054
1055 pthread_mutex_lock(&session->viewer_ready_lock);
1056 rcu_read_lock();
1057 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
1058 node.node) {
1059 struct relay_stream *stream;
1060
1061 /* Close streams. */
1062 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
1063 stream_close(session, stream);
1064 }
1065
1066 ctf_trace->invalid_flag = 1;
1067 ctf_trace_try_destroy(session, ctf_trace);
1068 }
1069 rcu_read_unlock();
1070 pthread_mutex_unlock(&session->viewer_ready_lock);
1071 }
1072
1073 /*
1074 * Try to destroy a session within a connection.
1075 */
1076 static void destroy_session(struct relay_session *session,
1077 struct lttng_ht *sessions_ht)
1078 {
1079 assert(session);
1080 assert(sessions_ht);
1081
1082 /* Indicate that this session can be destroyed from now on. */
1083 session->close_flag = 1;
1084
1085 try_close_streams(session);
1086
1087 /*
1088 * This will try to delete and destroy the session if no viewer is attached
1089 * to it meaning the refcount is down to zero.
1090 */
1091 session_try_destroy(sessions_ht, session);
1092 }
1093
1094 /*
1095 * Copy index data from the control port to a given index object.
1096 */
1097 static void copy_index_control_data(struct relay_index *index,
1098 struct lttcomm_relayd_index *data)
1099 {
1100 assert(index);
1101 assert(data);
1102
1103 /*
1104 * The index on disk is encoded in big endian, so we don't need to convert
1105 * the data received on the network. The data_offset value is NEVER
1106 * modified here and is updated by the data thread.
1107 */
1108 index->index_data.packet_size = data->packet_size;
1109 index->index_data.content_size = data->content_size;
1110 index->index_data.timestamp_begin = data->timestamp_begin;
1111 index->index_data.timestamp_end = data->timestamp_end;
1112 index->index_data.events_discarded = data->events_discarded;
1113 index->index_data.stream_id = data->stream_id;
1114 }
1115
1116 /*
1117 * Handle the RELAYD_CREATE_SESSION command.
1118 *
1119 * On success, send back the session id or else return a negative value.
1120 */
1121 static
1122 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
1123 struct relay_connection *conn)
1124 {
1125 int ret = 0, send_ret;
1126 struct relay_session *session;
1127 struct lttcomm_relayd_status_session reply;
1128
1129 assert(recv_hdr);
1130 assert(conn);
1131
1132 memset(&reply, 0, sizeof(reply));
1133
1134 session = session_create();
1135 if (!session) {
1136 ret = -1;
1137 goto error;
1138 }
1139 session->minor = conn->minor;
1140 session->major = conn->major;
1141 conn->session_id = session->id;
1142 conn->session = session;
1143
1144 reply.session_id = htobe64(session->id);
1145
1146 switch (conn->minor) {
1147 case 1:
1148 case 2:
1149 case 3:
1150 break;
1151 case 4: /* LTTng sessiond 2.4 */
1152 default:
1153 ret = cmd_create_session_2_4(conn, session);
1154 }
1155
1156 lttng_ht_add_unique_u64(conn->sessions_ht, &session->session_n);
1157 DBG("Created session %" PRIu64, session->id);
1158
1159 error:
1160 if (ret < 0) {
1161 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1162 } else {
1163 reply.ret_code = htobe32(LTTNG_OK);
1164 }
1165
1166 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1167 if (send_ret < 0) {
1168 ERR("Relayd sending session id");
1169 ret = send_ret;
1170 }
1171
1172 return ret;
1173 }
1174
1175 /*
1176 * When we have received all the streams and the metadata for a channel,
1177 * we make them visible to the viewer threads.
1178 */
1179 static
1180 void set_viewer_ready_flag(struct relay_connection *conn)
1181 {
1182 struct relay_stream *stream, *tmp_stream;
1183
1184 pthread_mutex_lock(&conn->session->viewer_ready_lock);
1185 cds_list_for_each_entry_safe(stream, tmp_stream, &conn->recv_head,
1186 recv_list) {
1187 stream->viewer_ready = 1;
1188 cds_list_del(&stream->recv_list);
1189 }
1190 pthread_mutex_unlock(&conn->session->viewer_ready_lock);
1191 return;
1192 }
1193
1194 /*
1195 * Add a recv handle node to the connection recv list with the given stream
1196 * handle. A new node is allocated thus must be freed when the node is deleted
1197 * from the list.
1198 */
1199 static void queue_stream(struct relay_stream *stream,
1200 struct relay_connection *conn)
1201 {
1202 assert(conn);
1203 assert(stream);
1204
1205 cds_list_add(&stream->recv_list, &conn->recv_head);
1206 }
1207
1208 /*
1209 * relay_add_stream: allocate a new stream for a session
1210 */
1211 static
1212 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1213 struct relay_connection *conn)
1214 {
1215 int ret, send_ret;
1216 struct relay_session *session = conn->session;
1217 struct relay_stream *stream = NULL;
1218 struct lttcomm_relayd_status_stream reply;
1219 struct ctf_trace *trace;
1220
1221 if (!session || conn->version_check_done == 0) {
1222 ERR("Trying to add a stream before version check");
1223 ret = -1;
1224 goto end_no_session;
1225 }
1226
1227 stream = zmalloc(sizeof(struct relay_stream));
1228 if (stream == NULL) {
1229 PERROR("relay stream zmalloc");
1230 ret = -1;
1231 goto end_no_session;
1232 }
1233
1234 switch (conn->minor) {
1235 case 1: /* LTTng sessiond 2.1 */
1236 ret = cmd_recv_stream_2_1(conn, stream);
1237 break;
1238 case 2: /* LTTng sessiond 2.2 */
1239 default:
1240 ret = cmd_recv_stream_2_2(conn, stream);
1241 break;
1242 }
1243 if (ret < 0) {
1244 goto err_free_stream;
1245 }
1246
1247 rcu_read_lock();
1248 stream->stream_handle = ++last_relay_stream_id;
1249 stream->prev_seq = -1ULL;
1250 stream->session_id = session->id;
1251 stream->index_fd = -1;
1252 stream->read_index_fd = -1;
1253 stream->ctf_stream_id = -1ULL;
1254 lttng_ht_node_init_u64(&stream->node, stream->stream_handle);
1255 pthread_mutex_init(&stream->lock, NULL);
1256
1257 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG,
1258 -1, -1);
1259 if (ret < 0) {
1260 ERR("relay creating output directory");
1261 goto err_free_stream;
1262 }
1263
1264 /*
1265 * No need to use run_as API here because whatever we receives, the relayd
1266 * uses its own credentials for the stream files.
1267 */
1268 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1269 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1270 if (ret < 0) {
1271 ERR("Create output file");
1272 goto err_free_stream;
1273 }
1274 stream->fd = ret;
1275 if (stream->tracefile_size) {
1276 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1277 } else {
1278 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1279 }
1280
1281 trace = ctf_trace_find_by_path(session->ctf_traces_ht, stream->path_name);
1282 if (!trace) {
1283 trace = ctf_trace_create(stream->path_name);
1284 if (!trace) {
1285 ret = -1;
1286 goto end;
1287 }
1288 ctf_trace_add(session->ctf_traces_ht, trace);
1289 }
1290 ctf_trace_get_ref(trace);
1291
1292 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1293 stream->metadata_flag = 1;
1294 /* Assign quick reference to the metadata stream in the trace. */
1295 trace->metadata_stream = stream;
1296 }
1297
1298 /*
1299 * Add the stream in the recv list of the connection. Once the end stream
1300 * message is received, this list is emptied and streams are set with the
1301 * viewer ready flag.
1302 */
1303 queue_stream(stream, conn);
1304
1305 /*
1306 * Both in the ctf_trace object and the global stream ht since the data
1307 * side of the relayd does not have the concept of session.
1308 */
1309 lttng_ht_add_unique_u64(relay_streams_ht, &stream->node);
1310 cds_list_add_tail(&stream->trace_list, &trace->stream_list);
1311
1312 session->stream_count++;
1313
1314 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1315 stream->stream_handle);
1316
1317 end:
1318 memset(&reply, 0, sizeof(reply));
1319 reply.handle = htobe64(stream->stream_handle);
1320 /* send the session id to the client or a negative return code on error */
1321 if (ret < 0) {
1322 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1323 /* stream was not properly added to the ht, so free it */
1324 stream_destroy(stream);
1325 } else {
1326 reply.ret_code = htobe32(LTTNG_OK);
1327 }
1328
1329 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1330 sizeof(struct lttcomm_relayd_status_stream), 0);
1331 if (send_ret < 0) {
1332 ERR("Relay sending stream id");
1333 ret = send_ret;
1334 }
1335 rcu_read_unlock();
1336
1337 end_no_session:
1338 return ret;
1339
1340 err_free_stream:
1341 stream_destroy(stream);
1342 return ret;
1343 }
1344
1345 /*
1346 * relay_close_stream: close a specific stream
1347 */
1348 static
1349 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1350 struct relay_connection *conn)
1351 {
1352 int ret, send_ret;
1353 struct relay_session *session = conn->session;
1354 struct lttcomm_relayd_close_stream stream_info;
1355 struct lttcomm_relayd_generic_reply reply;
1356 struct relay_stream *stream;
1357
1358 DBG("Close stream received");
1359
1360 if (!session || conn->version_check_done == 0) {
1361 ERR("Trying to close a stream before version check");
1362 ret = -1;
1363 goto end_no_session;
1364 }
1365
1366 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1367 sizeof(struct lttcomm_relayd_close_stream), 0);
1368 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1369 if (ret == 0) {
1370 /* Orderly shutdown. Not necessary to print an error. */
1371 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1372 } else {
1373 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1374 }
1375 ret = -1;
1376 goto end_no_session;
1377 }
1378
1379 rcu_read_lock();
1380 stream = stream_find_by_id(relay_streams_ht,
1381 be64toh(stream_info.stream_id));
1382 if (!stream) {
1383 ret = -1;
1384 goto end_unlock;
1385 }
1386
1387 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1388 stream->close_flag = 1;
1389 session->stream_count--;
1390
1391 /* Check if we can close it or else the data will do it. */
1392 try_close_stream(session, stream);
1393
1394 end_unlock:
1395 rcu_read_unlock();
1396
1397 memset(&reply, 0, sizeof(reply));
1398 if (ret < 0) {
1399 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1400 } else {
1401 reply.ret_code = htobe32(LTTNG_OK);
1402 }
1403 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1404 sizeof(struct lttcomm_relayd_generic_reply), 0);
1405 if (send_ret < 0) {
1406 ERR("Relay sending stream id");
1407 ret = send_ret;
1408 }
1409
1410 end_no_session:
1411 return ret;
1412 }
1413
1414 /*
1415 * relay_unknown_command: send -1 if received unknown command
1416 */
1417 static
1418 void relay_unknown_command(struct relay_connection *conn)
1419 {
1420 struct lttcomm_relayd_generic_reply reply;
1421 int ret;
1422
1423 memset(&reply, 0, sizeof(reply));
1424 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1425 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1426 sizeof(struct lttcomm_relayd_generic_reply), 0);
1427 if (ret < 0) {
1428 ERR("Relay sending unknown command");
1429 }
1430 }
1431
1432 /*
1433 * relay_start: send an acknowledgment to the client to tell if we are
1434 * ready to receive data. We are ready if a session is established.
1435 */
1436 static
1437 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1438 struct relay_connection *conn)
1439 {
1440 int ret = htobe32(LTTNG_OK);
1441 struct lttcomm_relayd_generic_reply reply;
1442 struct relay_session *session = conn->session;
1443
1444 if (!session) {
1445 DBG("Trying to start the streaming without a session established");
1446 ret = htobe32(LTTNG_ERR_UNK);
1447 }
1448
1449 memset(&reply, 0, sizeof(reply));
1450 reply.ret_code = ret;
1451 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1452 sizeof(struct lttcomm_relayd_generic_reply), 0);
1453 if (ret < 0) {
1454 ERR("Relay sending start ack");
1455 }
1456
1457 return ret;
1458 }
1459
1460 /*
1461 * Append padding to the file pointed by the file descriptor fd.
1462 */
1463 static int write_padding_to_file(int fd, uint32_t size)
1464 {
1465 ssize_t ret = 0;
1466 char *zeros;
1467
1468 if (size == 0) {
1469 goto end;
1470 }
1471
1472 zeros = zmalloc(size);
1473 if (zeros == NULL) {
1474 PERROR("zmalloc zeros for padding");
1475 ret = -1;
1476 goto end;
1477 }
1478
1479 ret = lttng_write(fd, zeros, size);
1480 if (ret < size) {
1481 PERROR("write padding to file");
1482 }
1483
1484 free(zeros);
1485
1486 end:
1487 return ret;
1488 }
1489
1490 /*
1491 * relay_recv_metadata: receive the metada for the session.
1492 */
1493 static
1494 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1495 struct relay_connection *conn)
1496 {
1497 int ret = htobe32(LTTNG_OK);
1498 ssize_t size_ret;
1499 struct relay_session *session = conn->session;
1500 struct lttcomm_relayd_metadata_payload *metadata_struct;
1501 struct relay_stream *metadata_stream;
1502 uint64_t data_size, payload_size;
1503 struct ctf_trace *ctf_trace;
1504
1505 if (!session) {
1506 ERR("Metadata sent before version check");
1507 ret = -1;
1508 goto end;
1509 }
1510
1511 data_size = payload_size = be64toh(recv_hdr->data_size);
1512 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1513 ERR("Incorrect data size");
1514 ret = -1;
1515 goto end;
1516 }
1517 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1518
1519 if (data_buffer_size < data_size) {
1520 /* In case the realloc fails, we can free the memory */
1521 char *tmp_data_ptr;
1522
1523 tmp_data_ptr = realloc(data_buffer, data_size);
1524 if (!tmp_data_ptr) {
1525 ERR("Allocating data buffer");
1526 free(data_buffer);
1527 ret = -1;
1528 goto end;
1529 }
1530 data_buffer = tmp_data_ptr;
1531 data_buffer_size = data_size;
1532 }
1533 memset(data_buffer, 0, data_size);
1534 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1535 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
1536 if (ret < 0 || ret != data_size) {
1537 if (ret == 0) {
1538 /* Orderly shutdown. Not necessary to print an error. */
1539 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1540 } else {
1541 ERR("Relay didn't receive the whole metadata");
1542 }
1543 ret = -1;
1544 goto end;
1545 }
1546 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1547
1548 rcu_read_lock();
1549 metadata_stream = stream_find_by_id(relay_streams_ht,
1550 be64toh(metadata_struct->stream_id));
1551 if (!metadata_stream) {
1552 ret = -1;
1553 goto end_unlock;
1554 }
1555
1556 size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
1557 payload_size);
1558 if (size_ret < payload_size) {
1559 ERR("Relay error writing metadata on file");
1560 ret = -1;
1561 goto end_unlock;
1562 }
1563
1564 ret = write_padding_to_file(metadata_stream->fd,
1565 be32toh(metadata_struct->padding_size));
1566 if (ret < 0) {
1567 goto end_unlock;
1568 }
1569
1570 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1571 metadata_stream->path_name);
1572 assert(ctf_trace);
1573 ctf_trace->metadata_received +=
1574 payload_size + be32toh(metadata_struct->padding_size);
1575
1576 DBG2("Relay metadata written");
1577
1578 end_unlock:
1579 rcu_read_unlock();
1580 end:
1581 return ret;
1582 }
1583
1584 /*
1585 * relay_send_version: send relayd version number
1586 */
1587 static
1588 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1589 struct relay_connection *conn)
1590 {
1591 int ret;
1592 struct lttcomm_relayd_version reply, msg;
1593
1594 assert(conn);
1595
1596 conn->version_check_done = 1;
1597
1598 /* Get version from the other side. */
1599 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1600 if (ret < 0 || ret != sizeof(msg)) {
1601 if (ret == 0) {
1602 /* Orderly shutdown. Not necessary to print an error. */
1603 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1604 } else {
1605 ERR("Relay failed to receive the version values.");
1606 }
1607 ret = -1;
1608 goto end;
1609 }
1610
1611 memset(&reply, 0, sizeof(reply));
1612 reply.major = RELAYD_VERSION_COMM_MAJOR;
1613 reply.minor = RELAYD_VERSION_COMM_MINOR;
1614
1615 /* Major versions must be the same */
1616 if (reply.major != be32toh(msg.major)) {
1617 DBG("Incompatible major versions (%u vs %u), deleting session",
1618 reply.major, be32toh(msg.major));
1619 destroy_session(conn->session, conn->sessions_ht);
1620 ret = 0;
1621 goto end;
1622 }
1623
1624 conn->major = reply.major;
1625 /* We adapt to the lowest compatible version */
1626 if (reply.minor <= be32toh(msg.minor)) {
1627 conn->minor = reply.minor;
1628 } else {
1629 conn->minor = be32toh(msg.minor);
1630 }
1631
1632 reply.major = htobe32(reply.major);
1633 reply.minor = htobe32(reply.minor);
1634 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1635 sizeof(struct lttcomm_relayd_version), 0);
1636 if (ret < 0) {
1637 ERR("Relay sending version");
1638 }
1639
1640 DBG("Version check done using protocol %u.%u", conn->major,
1641 conn->minor);
1642
1643 end:
1644 return ret;
1645 }
1646
1647 /*
1648 * Check for data pending for a given stream id from the session daemon.
1649 */
1650 static
1651 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1652 struct relay_connection *conn)
1653 {
1654 struct relay_session *session = conn->session;
1655 struct lttcomm_relayd_data_pending msg;
1656 struct lttcomm_relayd_generic_reply reply;
1657 struct relay_stream *stream;
1658 int ret;
1659 uint64_t last_net_seq_num, stream_id;
1660
1661 DBG("Data pending command received");
1662
1663 if (!session || conn->version_check_done == 0) {
1664 ERR("Trying to check for data before version check");
1665 ret = -1;
1666 goto end_no_session;
1667 }
1668
1669 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1670 if (ret < sizeof(msg)) {
1671 if (ret == 0) {
1672 /* Orderly shutdown. Not necessary to print an error. */
1673 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1674 } else {
1675 ERR("Relay didn't receive valid data_pending struct size : %d",
1676 ret);
1677 }
1678 ret = -1;
1679 goto end_no_session;
1680 }
1681
1682 stream_id = be64toh(msg.stream_id);
1683 last_net_seq_num = be64toh(msg.last_net_seq_num);
1684
1685 rcu_read_lock();
1686 stream = stream_find_by_id(relay_streams_ht, stream_id);
1687 if (stream == NULL) {
1688 ret = -1;
1689 goto end_unlock;
1690 }
1691
1692 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1693 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1694 last_net_seq_num);
1695
1696 /* Avoid wrapping issue */
1697 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1698 /* Data has in fact been written and is NOT pending */
1699 ret = 0;
1700 } else {
1701 /* Data still being streamed thus pending */
1702 ret = 1;
1703 }
1704
1705 /* Pending check is now done. */
1706 stream->data_pending_check_done = 1;
1707
1708 end_unlock:
1709 rcu_read_unlock();
1710
1711 memset(&reply, 0, sizeof(reply));
1712 reply.ret_code = htobe32(ret);
1713 ret = conn->sock->ops->sendmsg(conn->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_connection *conn)
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 (!conn->session || conn->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 = conn->sock->ops->recvmsg(conn->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", conn->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 node.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 memset(&reply, 0, sizeof(reply));
1776 reply.ret_code = htobe32(LTTNG_OK);
1777 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1778 if (ret < 0) {
1779 ERR("Relay data quiescent control ret code failed");
1780 }
1781
1782 end_no_session:
1783 return ret;
1784 }
1785
1786 /*
1787 * Initialize a data pending command. This means that a client is about to ask
1788 * for data pending for each stream he/she holds. Simply iterate over all
1789 * streams of a session and set the data_pending_check_done flag.
1790 *
1791 * This command returns to the client a LTTNG_OK code.
1792 */
1793 static
1794 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1795 struct relay_connection *conn)
1796 {
1797 int ret;
1798 struct lttng_ht_iter iter;
1799 struct lttcomm_relayd_begin_data_pending msg;
1800 struct lttcomm_relayd_generic_reply reply;
1801 struct relay_stream *stream;
1802 uint64_t session_id;
1803
1804 assert(recv_hdr);
1805 assert(conn);
1806
1807 DBG("Init streams for data pending");
1808
1809 if (!conn->session || conn->version_check_done == 0) {
1810 ERR("Trying to check for data before version check");
1811 ret = -1;
1812 goto end_no_session;
1813 }
1814
1815 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1816 if (ret < sizeof(msg)) {
1817 if (ret == 0) {
1818 /* Orderly shutdown. Not necessary to print an error. */
1819 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1820 } else {
1821 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1822 ret);
1823 }
1824 ret = -1;
1825 goto end_no_session;
1826 }
1827
1828 session_id = be64toh(msg.session_id);
1829
1830 /*
1831 * Iterate over all streams to set the begin data pending flag. For now, the
1832 * streams are indexed by stream handle so we have to iterate over all
1833 * streams to find the one associated with the right session_id.
1834 */
1835 rcu_read_lock();
1836 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1837 node.node) {
1838 if (stream->session_id == session_id) {
1839 stream->data_pending_check_done = 0;
1840 DBG("Set begin data pending flag to stream %" PRIu64,
1841 stream->stream_handle);
1842 }
1843 }
1844 rcu_read_unlock();
1845
1846 memset(&reply, 0, sizeof(reply));
1847 /* All good, send back reply. */
1848 reply.ret_code = htobe32(LTTNG_OK);
1849
1850 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1851 if (ret < 0) {
1852 ERR("Relay begin data pending send reply failed");
1853 }
1854
1855 end_no_session:
1856 return ret;
1857 }
1858
1859 /*
1860 * End data pending command. This will check, for a given session id, if each
1861 * stream associated with it has its data_pending_check_done flag set. If not,
1862 * this means that the client lost track of the stream but the data is still
1863 * being streamed on our side. In this case, we inform the client that data is
1864 * inflight.
1865 *
1866 * Return to the client if there is data in flight or not with a ret_code.
1867 */
1868 static
1869 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1870 struct relay_connection *conn)
1871 {
1872 int ret;
1873 struct lttng_ht_iter iter;
1874 struct lttcomm_relayd_end_data_pending msg;
1875 struct lttcomm_relayd_generic_reply reply;
1876 struct relay_stream *stream;
1877 uint64_t session_id;
1878 uint32_t is_data_inflight = 0;
1879
1880 assert(recv_hdr);
1881 assert(conn);
1882
1883 DBG("End data pending command");
1884
1885 if (!conn->session || conn->version_check_done == 0) {
1886 ERR("Trying to check for data before version check");
1887 ret = -1;
1888 goto end_no_session;
1889 }
1890
1891 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1892 if (ret < sizeof(msg)) {
1893 if (ret == 0) {
1894 /* Orderly shutdown. Not necessary to print an error. */
1895 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1896 } else {
1897 ERR("Relay didn't receive valid end data_pending struct size: %d",
1898 ret);
1899 }
1900 ret = -1;
1901 goto end_no_session;
1902 }
1903
1904 session_id = be64toh(msg.session_id);
1905
1906 /* Iterate over all streams to see if the begin data pending flag is set. */
1907 rcu_read_lock();
1908 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1909 node.node) {
1910 if (stream->session_id == session_id &&
1911 !stream->data_pending_check_done && !stream->terminated_flag) {
1912 is_data_inflight = 1;
1913 DBG("Data is still in flight for stream %" PRIu64,
1914 stream->stream_handle);
1915 break;
1916 }
1917 }
1918 rcu_read_unlock();
1919
1920 memset(&reply, 0, sizeof(reply));
1921 /* All good, send back reply. */
1922 reply.ret_code = htobe32(is_data_inflight);
1923
1924 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1925 if (ret < 0) {
1926 ERR("Relay end data pending send reply failed");
1927 }
1928
1929 end_no_session:
1930 return ret;
1931 }
1932
1933 /*
1934 * Receive an index for a specific stream.
1935 *
1936 * Return 0 on success else a negative value.
1937 */
1938 static
1939 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1940 struct relay_connection *conn)
1941 {
1942 int ret, send_ret, index_created = 0;
1943 struct relay_session *session = conn->session;
1944 struct lttcomm_relayd_index index_info;
1945 struct relay_index *index, *wr_index = NULL;
1946 struct lttcomm_relayd_generic_reply reply;
1947 struct relay_stream *stream;
1948 uint64_t net_seq_num;
1949
1950 assert(conn);
1951
1952 DBG("Relay receiving index");
1953
1954 if (!session || conn->version_check_done == 0) {
1955 ERR("Trying to close a stream before version check");
1956 ret = -1;
1957 goto end_no_session;
1958 }
1959
1960 ret = conn->sock->ops->recvmsg(conn->sock, &index_info,
1961 sizeof(index_info), 0);
1962 if (ret < sizeof(index_info)) {
1963 if (ret == 0) {
1964 /* Orderly shutdown. Not necessary to print an error. */
1965 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1966 } else {
1967 ERR("Relay didn't receive valid index struct size : %d", ret);
1968 }
1969 ret = -1;
1970 goto end_no_session;
1971 }
1972
1973 net_seq_num = be64toh(index_info.net_seq_num);
1974
1975 rcu_read_lock();
1976 stream = stream_find_by_id(relay_streams_ht,
1977 be64toh(index_info.relay_stream_id));
1978 if (!stream) {
1979 ret = -1;
1980 goto end_rcu_unlock;
1981 }
1982
1983 /* Live beacon handling */
1984 if (index_info.packet_size == 0) {
1985 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1986
1987 /*
1988 * Only flag a stream inactive when it has already received data
1989 * and no indexes are in flight.
1990 */
1991 if (stream->total_index_received > 0 && stream->indexes_in_flight == 0) {
1992 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1993 }
1994 ret = 0;
1995 goto end_rcu_unlock;
1996 } else {
1997 stream->beacon_ts_end = -1ULL;
1998 }
1999
2000 index = relay_index_find(stream->stream_handle, net_seq_num);
2001 if (!index) {
2002 /* A successful creation will add the object to the HT. */
2003 index = relay_index_create(stream->stream_handle, net_seq_num);
2004 if (!index) {
2005 goto end_rcu_unlock;
2006 }
2007 index_created = 1;
2008 stream->indexes_in_flight++;
2009 }
2010
2011 copy_index_control_data(index, &index_info);
2012 if (stream->ctf_stream_id == -1ULL) {
2013 stream->ctf_stream_id = be64toh(index_info.stream_id);
2014 }
2015
2016 if (index_created) {
2017 /*
2018 * Try to add the relay index object to the hash table. If an object
2019 * already exist, destroy back the index created, set the data in this
2020 * object and write it on disk.
2021 */
2022 relay_index_add(index, &wr_index);
2023 if (wr_index) {
2024 copy_index_control_data(wr_index, &index_info);
2025 free(index);
2026 }
2027 } else {
2028 /* The index already exists so write it on disk. */
2029 wr_index = index;
2030 }
2031
2032 /* Do we have a writable ready index to write on disk. */
2033 if (wr_index) {
2034 ret = relay_index_write(wr_index->fd, wr_index);
2035 if (ret < 0) {
2036 goto end_rcu_unlock;
2037 }
2038 stream->total_index_received++;
2039 stream->indexes_in_flight--;
2040 assert(stream->indexes_in_flight >= 0);
2041 }
2042
2043 end_rcu_unlock:
2044 rcu_read_unlock();
2045
2046 memset(&reply, 0, sizeof(reply));
2047 if (ret < 0) {
2048 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2049 } else {
2050 reply.ret_code = htobe32(LTTNG_OK);
2051 }
2052 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2053 if (send_ret < 0) {
2054 ERR("Relay sending close index id reply");
2055 ret = send_ret;
2056 }
2057
2058 end_no_session:
2059 return ret;
2060 }
2061
2062 /*
2063 * Receive the streams_sent message.
2064 *
2065 * Return 0 on success else a negative value.
2066 */
2067 static
2068 int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
2069 struct relay_connection *conn)
2070 {
2071 int ret, send_ret;
2072 struct lttcomm_relayd_generic_reply reply;
2073
2074 assert(conn);
2075
2076 DBG("Relay receiving streams_sent");
2077
2078 if (!conn->session || conn->version_check_done == 0) {
2079 ERR("Trying to close a stream before version check");
2080 ret = -1;
2081 goto end_no_session;
2082 }
2083
2084 /*
2085 * Flag every pending stream in the connection recv list that they are
2086 * ready to be used by the viewer.
2087 */
2088 set_viewer_ready_flag(conn);
2089
2090 /*
2091 * Inform the viewer that there are new streams in the session.
2092 */
2093 if (conn->session->viewer_refcount) {
2094 uatomic_set(&conn->session->new_streams, 1);
2095 }
2096
2097 memset(&reply, 0, sizeof(reply));
2098 reply.ret_code = htobe32(LTTNG_OK);
2099 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2100 if (send_ret < 0) {
2101 ERR("Relay sending sent_stream reply");
2102 ret = send_ret;
2103 } else {
2104 /* Success. */
2105 ret = 0;
2106 }
2107
2108 end_no_session:
2109 return ret;
2110 }
2111
2112 /*
2113 * Process the commands received on the control socket
2114 */
2115 static
2116 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
2117 struct relay_connection *conn)
2118 {
2119 int ret = 0;
2120
2121 switch (be32toh(recv_hdr->cmd)) {
2122 case RELAYD_CREATE_SESSION:
2123 ret = relay_create_session(recv_hdr, conn);
2124 break;
2125 case RELAYD_ADD_STREAM:
2126 ret = relay_add_stream(recv_hdr, conn);
2127 break;
2128 case RELAYD_START_DATA:
2129 ret = relay_start(recv_hdr, conn);
2130 break;
2131 case RELAYD_SEND_METADATA:
2132 ret = relay_recv_metadata(recv_hdr, conn);
2133 break;
2134 case RELAYD_VERSION:
2135 ret = relay_send_version(recv_hdr, conn);
2136 break;
2137 case RELAYD_CLOSE_STREAM:
2138 ret = relay_close_stream(recv_hdr, conn);
2139 break;
2140 case RELAYD_DATA_PENDING:
2141 ret = relay_data_pending(recv_hdr, conn);
2142 break;
2143 case RELAYD_QUIESCENT_CONTROL:
2144 ret = relay_quiescent_control(recv_hdr, conn);
2145 break;
2146 case RELAYD_BEGIN_DATA_PENDING:
2147 ret = relay_begin_data_pending(recv_hdr, conn);
2148 break;
2149 case RELAYD_END_DATA_PENDING:
2150 ret = relay_end_data_pending(recv_hdr, conn);
2151 break;
2152 case RELAYD_SEND_INDEX:
2153 ret = relay_recv_index(recv_hdr, conn);
2154 break;
2155 case RELAYD_STREAMS_SENT:
2156 ret = relay_streams_sent(recv_hdr, conn);
2157 break;
2158 case RELAYD_UPDATE_SYNC_INFO:
2159 default:
2160 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2161 relay_unknown_command(conn);
2162 ret = -1;
2163 goto end;
2164 }
2165
2166 end:
2167 return ret;
2168 }
2169
2170 /*
2171 * Handle index for a data stream.
2172 *
2173 * RCU read side lock MUST be acquired.
2174 *
2175 * Return 0 on success else a negative value.
2176 */
2177 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2178 int rotate_index)
2179 {
2180 int ret = 0, index_created = 0;
2181 uint64_t stream_id, data_offset;
2182 struct relay_index *index, *wr_index = NULL;
2183
2184 assert(stream);
2185
2186 stream_id = stream->stream_handle;
2187 /* Get data offset because we are about to update the index. */
2188 data_offset = htobe64(stream->tracefile_size_current);
2189
2190 /*
2191 * Lookup for an existing index for that stream id/sequence number. If on
2192 * exists, the control thread already received the data for it thus we need
2193 * to write it on disk.
2194 */
2195 index = relay_index_find(stream_id, net_seq_num);
2196 if (!index) {
2197 /* A successful creation will add the object to the HT. */
2198 index = relay_index_create(stream_id, net_seq_num);
2199 if (!index) {
2200 ret = -1;
2201 goto error;
2202 }
2203 index_created = 1;
2204 stream->indexes_in_flight++;
2205 }
2206
2207 if (rotate_index || stream->index_fd < 0) {
2208 index->to_close_fd = stream->index_fd;
2209 ret = index_create_file(stream->path_name, stream->channel_name,
2210 relayd_uid, relayd_gid, stream->tracefile_size,
2211 stream->tracefile_count_current);
2212 if (ret < 0) {
2213 /* This will close the stream's index fd if one. */
2214 relay_index_free_safe(index);
2215 goto error;
2216 }
2217 stream->index_fd = ret;
2218 }
2219 index->fd = stream->index_fd;
2220 index->index_data.offset = data_offset;
2221
2222 if (index_created) {
2223 /*
2224 * Try to add the relay index object to the hash table. If an object
2225 * already exist, destroy back the index created and set the data.
2226 */
2227 relay_index_add(index, &wr_index);
2228 if (wr_index) {
2229 /* Copy back data from the created index. */
2230 wr_index->fd = index->fd;
2231 wr_index->to_close_fd = index->to_close_fd;
2232 wr_index->index_data.offset = data_offset;
2233 free(index);
2234 }
2235 } else {
2236 /* The index already exists so write it on disk. */
2237 wr_index = index;
2238 }
2239
2240 /* Do we have a writable ready index to write on disk. */
2241 if (wr_index) {
2242 ret = relay_index_write(wr_index->fd, wr_index);
2243 if (ret < 0) {
2244 goto error;
2245 }
2246 stream->total_index_received++;
2247 stream->indexes_in_flight--;
2248 assert(stream->indexes_in_flight >= 0);
2249 }
2250
2251 error:
2252 return ret;
2253 }
2254
2255 /*
2256 * relay_process_data: Process the data received on the data socket
2257 */
2258 static
2259 int relay_process_data(struct relay_connection *conn)
2260 {
2261 int ret = 0, rotate_index = 0;
2262 ssize_t size_ret;
2263 struct relay_stream *stream;
2264 struct lttcomm_relayd_data_hdr data_hdr;
2265 uint64_t stream_id;
2266 uint64_t net_seq_num;
2267 uint32_t data_size;
2268 struct relay_session *session;
2269
2270 assert(conn);
2271
2272 ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr,
2273 sizeof(struct lttcomm_relayd_data_hdr), 0);
2274 if (ret <= 0) {
2275 if (ret == 0) {
2276 /* Orderly shutdown. Not necessary to print an error. */
2277 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2278 } else {
2279 ERR("Unable to receive data header on sock %d", conn->sock->fd);
2280 }
2281 ret = -1;
2282 goto end;
2283 }
2284
2285 stream_id = be64toh(data_hdr.stream_id);
2286
2287 rcu_read_lock();
2288 stream = stream_find_by_id(relay_streams_ht, stream_id);
2289 if (!stream) {
2290 ret = -1;
2291 goto end_rcu_unlock;
2292 }
2293
2294 session = session_find_by_id(conn->sessions_ht, stream->session_id);
2295 assert(session);
2296
2297 data_size = be32toh(data_hdr.data_size);
2298 if (data_buffer_size < data_size) {
2299 char *tmp_data_ptr;
2300
2301 tmp_data_ptr = realloc(data_buffer, data_size);
2302 if (!tmp_data_ptr) {
2303 ERR("Allocating data buffer");
2304 free(data_buffer);
2305 ret = -1;
2306 goto end_rcu_unlock;
2307 }
2308 data_buffer = tmp_data_ptr;
2309 data_buffer_size = data_size;
2310 }
2311 memset(data_buffer, 0, data_size);
2312
2313 net_seq_num = be64toh(data_hdr.net_seq_num);
2314
2315 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2316 data_size, stream_id, net_seq_num);
2317 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
2318 if (ret <= 0) {
2319 if (ret == 0) {
2320 /* Orderly shutdown. Not necessary to print an error. */
2321 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2322 }
2323 ret = -1;
2324 goto end_rcu_unlock;
2325 }
2326
2327 /* Check if a rotation is needed. */
2328 if (stream->tracefile_size > 0 &&
2329 (stream->tracefile_size_current + data_size) >
2330 stream->tracefile_size) {
2331 struct relay_viewer_stream *vstream;
2332 uint64_t new_id;
2333
2334 new_id = (stream->tracefile_count_current + 1) %
2335 stream->tracefile_count;
2336 /*
2337 * When we wrap-around back to 0, we start overwriting old
2338 * trace data.
2339 */
2340 if (!stream->tracefile_overwrite && new_id == 0) {
2341 stream->tracefile_overwrite = 1;
2342 }
2343 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
2344 if (stream->tracefile_overwrite) {
2345 stream->oldest_tracefile_id =
2346 (stream->oldest_tracefile_id + 1) %
2347 stream->tracefile_count;
2348 }
2349 vstream = viewer_stream_find_by_id(stream->stream_handle);
2350 if (vstream) {
2351 /*
2352 * The viewer is reading a file about to be
2353 * overwritten. Close the FDs it is
2354 * currently using and let it handle the fault.
2355 */
2356 if (vstream->tracefile_count_current == new_id) {
2357 pthread_mutex_lock(&vstream->overwrite_lock);
2358 vstream->abort_flag = 1;
2359 pthread_mutex_unlock(&vstream->overwrite_lock);
2360 DBG("Streaming side setting abort_flag on stream %s_%" PRIu64 "\n",
2361 stream->channel_name, new_id);
2362 } else if (vstream->tracefile_count_current ==
2363 stream->tracefile_count_current) {
2364 /*
2365 * The reader and writer were in the
2366 * same trace file, inform the viewer
2367 * that no new index will ever be added
2368 * to this file.
2369 */
2370 vstream->close_write_flag = 1;
2371 }
2372 }
2373 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2374 stream->tracefile_size, stream->tracefile_count,
2375 relayd_uid, relayd_gid, stream->fd,
2376 &(stream->tracefile_count_current), &stream->fd);
2377 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
2378 if (ret < 0) {
2379 ERR("Rotating stream output file");
2380 goto end_rcu_unlock;
2381 }
2382 /* Reset current size because we just perform a stream rotation. */
2383 stream->tracefile_size_current = 0;
2384 rotate_index = 1;
2385 }
2386
2387 /*
2388 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2389 * index are NOT supported.
2390 */
2391 if (session->minor >= 4 && !session->snapshot) {
2392 ret = handle_index_data(stream, net_seq_num, rotate_index);
2393 if (ret < 0) {
2394 goto end_rcu_unlock;
2395 }
2396 }
2397
2398 /* Write data to stream output fd. */
2399 size_ret = lttng_write(stream->fd, data_buffer, data_size);
2400 if (size_ret < data_size) {
2401 ERR("Relay error writing data to file");
2402 ret = -1;
2403 goto end_rcu_unlock;
2404 }
2405
2406 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2407 ret, stream->stream_handle);
2408
2409 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2410 if (ret < 0) {
2411 goto end_rcu_unlock;
2412 }
2413 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2414
2415 stream->prev_seq = net_seq_num;
2416
2417 try_close_stream(session, stream);
2418
2419 end_rcu_unlock:
2420 rcu_read_unlock();
2421 end:
2422 return ret;
2423 }
2424
2425 static
2426 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2427 {
2428 int ret;
2429
2430 assert(events);
2431
2432 (void) lttng_poll_del(events, pollfd);
2433
2434 ret = close(pollfd);
2435 if (ret < 0) {
2436 ERR("Closing pollfd %d", pollfd);
2437 }
2438 }
2439
2440 static void destroy_connection(struct lttng_ht *relay_connections_ht,
2441 struct relay_connection *conn)
2442 {
2443 assert(relay_connections_ht);
2444 assert(conn);
2445
2446 connection_delete(relay_connections_ht, conn);
2447
2448 /* For the control socket, we try to destroy the session. */
2449 if (conn->type == RELAY_CONTROL && conn->session) {
2450 destroy_session(conn->session, conn->sessions_ht);
2451 }
2452
2453 connection_destroy(conn);
2454 }
2455
2456 /*
2457 * This thread does the actual work
2458 */
2459 static
2460 void *relay_thread_worker(void *data)
2461 {
2462 int ret, err = -1, last_seen_data_fd = -1;
2463 uint32_t nb_fd;
2464 struct relay_connection *conn;
2465 struct lttng_poll_event events;
2466 struct lttng_ht *relay_connections_ht;
2467 struct lttng_ht_iter iter;
2468 struct lttcomm_relayd_hdr recv_hdr;
2469 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2470 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2471 struct relay_index *index;
2472
2473 DBG("[thread] Relay worker started");
2474
2475 rcu_register_thread();
2476
2477 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2478
2479 if (testpoint(relayd_thread_worker)) {
2480 goto error_testpoint;
2481 }
2482
2483 health_code_update();
2484
2485 /* table of connections indexed on socket */
2486 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2487 if (!relay_connections_ht) {
2488 goto relay_connections_ht_error;
2489 }
2490
2491 /* Tables of received indexes indexed by index handle and net_seq_num. */
2492 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2493 if (!indexes_ht) {
2494 goto indexes_ht_error;
2495 }
2496
2497 ret = create_thread_poll_set(&events, 2);
2498 if (ret < 0) {
2499 goto error_poll_create;
2500 }
2501
2502 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2503 if (ret < 0) {
2504 goto error;
2505 }
2506
2507 restart:
2508 while (1) {
2509 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2510
2511 health_code_update();
2512
2513 /* Infinite blocking call, waiting for transmission */
2514 DBG3("Relayd worker thread polling...");
2515 health_poll_entry();
2516 ret = lttng_poll_wait(&events, -1);
2517 health_poll_exit();
2518 if (ret < 0) {
2519 /*
2520 * Restart interrupted system call.
2521 */
2522 if (errno == EINTR) {
2523 goto restart;
2524 }
2525 goto error;
2526 }
2527
2528 nb_fd = ret;
2529
2530 /*
2531 * Process control. The control connection is prioritised so we don't
2532 * starve it with high throughout put tracing data on the data
2533 * connection.
2534 */
2535 for (i = 0; i < nb_fd; i++) {
2536 /* Fetch once the poll data */
2537 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2538 int pollfd = LTTNG_POLL_GETFD(&events, i);
2539
2540 health_code_update();
2541
2542 if (!revents) {
2543 /* No activity for this FD (poll implementation). */
2544 continue;
2545 }
2546
2547 /* Thread quit pipe has been closed. Killing thread. */
2548 ret = check_thread_quit_pipe(pollfd, revents);
2549 if (ret) {
2550 err = 0;
2551 goto exit;
2552 }
2553
2554 /* Inspect the relay conn pipe for new connection */
2555 if (pollfd == relay_conn_pipe[0]) {
2556 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2557 ERR("Relay connection pipe error");
2558 goto error;
2559 } else if (revents & LPOLLIN) {
2560 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
2561 if (ret < 0) {
2562 goto error;
2563 }
2564 conn->sessions_ht = sessions_ht;
2565 connection_init(conn);
2566 lttng_poll_add(&events, conn->sock->fd,
2567 LPOLLIN | LPOLLRDHUP);
2568 rcu_read_lock();
2569 lttng_ht_add_unique_ulong(relay_connections_ht,
2570 &conn->sock_n);
2571 rcu_read_unlock();
2572 DBG("Connection socket %d added", conn->sock->fd);
2573 }
2574 } else {
2575 rcu_read_lock();
2576 conn = connection_find_by_sock(relay_connections_ht, pollfd);
2577 /* If not found, there is a synchronization issue. */
2578 assert(conn);
2579
2580 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2581 cleanup_connection_pollfd(&events, pollfd);
2582 destroy_connection(relay_connections_ht, conn);
2583 if (last_seen_data_fd == pollfd) {
2584 last_seen_data_fd = last_notdel_data_fd;
2585 }
2586 } else if (revents & LPOLLIN) {
2587 if (conn->type == RELAY_CONTROL) {
2588 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2589 sizeof(recv_hdr), 0);
2590 if (ret <= 0) {
2591 /* Connection closed */
2592 cleanup_connection_pollfd(&events, pollfd);
2593 destroy_connection(relay_connections_ht, conn);
2594 DBG("Control connection closed with %d", pollfd);
2595 } else {
2596 ret = relay_process_control(&recv_hdr, conn);
2597 if (ret < 0) {
2598 /* Clear the session on error. */
2599 cleanup_connection_pollfd(&events, pollfd);
2600 destroy_connection(relay_connections_ht, conn);
2601 DBG("Connection closed with %d", pollfd);
2602 }
2603 seen_control = 1;
2604 }
2605 } else {
2606 /*
2607 * Flag the last seen data fd not deleted. It will be
2608 * used as the last seen fd if any fd gets deleted in
2609 * this first loop.
2610 */
2611 last_notdel_data_fd = pollfd;
2612 }
2613 } else {
2614 ERR("Unknown poll events %u for sock %d", revents, pollfd);
2615 }
2616 rcu_read_unlock();
2617 }
2618 }
2619
2620 /*
2621 * The last loop handled a control request, go back to poll to make
2622 * sure we prioritise the control socket.
2623 */
2624 if (seen_control) {
2625 continue;
2626 }
2627
2628 if (last_seen_data_fd >= 0) {
2629 for (i = 0; i < nb_fd; i++) {
2630 int pollfd = LTTNG_POLL_GETFD(&events, i);
2631
2632 health_code_update();
2633
2634 if (last_seen_data_fd == pollfd) {
2635 idx = i;
2636 break;
2637 }
2638 }
2639 }
2640
2641 /* Process data connection. */
2642 for (i = idx + 1; i < nb_fd; i++) {
2643 /* Fetch the poll data. */
2644 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2645 int pollfd = LTTNG_POLL_GETFD(&events, i);
2646
2647 health_code_update();
2648
2649 if (!revents) {
2650 /* No activity for this FD (poll implementation). */
2651 continue;
2652 }
2653
2654 /* Skip the command pipe. It's handled in the first loop. */
2655 if (pollfd == relay_conn_pipe[0]) {
2656 continue;
2657 }
2658
2659 rcu_read_lock();
2660 conn = connection_find_by_sock(relay_connections_ht, pollfd);
2661 if (!conn) {
2662 /* Skip it. Might be removed before. */
2663 rcu_read_unlock();
2664 continue;
2665 }
2666
2667 if (revents & LPOLLIN) {
2668 if (conn->type != RELAY_DATA) {
2669 rcu_read_unlock();
2670 continue;
2671 }
2672
2673 ret = relay_process_data(conn);
2674 /* Connection closed */
2675 if (ret < 0) {
2676 cleanup_connection_pollfd(&events, pollfd);
2677 destroy_connection(relay_connections_ht, conn);
2678 DBG("Data connection closed with %d", pollfd);
2679 /*
2680 * Every goto restart call sets the last seen fd where
2681 * here we don't really care since we gracefully
2682 * continue the loop after the connection is deleted.
2683 */
2684 } else {
2685 /* Keep last seen port. */
2686 last_seen_data_fd = pollfd;
2687 rcu_read_unlock();
2688 goto restart;
2689 }
2690 }
2691 rcu_read_unlock();
2692 }
2693 last_seen_data_fd = -1;
2694 }
2695
2696 /* Normal exit, no error */
2697 ret = 0;
2698
2699 exit:
2700 error:
2701 lttng_poll_clean(&events);
2702
2703 /* Cleanup reamaining connection object. */
2704 rcu_read_lock();
2705 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2706 sock_n.node) {
2707 health_code_update();
2708 destroy_connection(relay_connections_ht, conn);
2709 }
2710 rcu_read_unlock();
2711 error_poll_create:
2712 rcu_read_lock();
2713 cds_lfht_for_each_entry(indexes_ht->ht, &iter.iter, index,
2714 index_n.node) {
2715 health_code_update();
2716 relay_index_delete(index);
2717 relay_index_free_safe(index);
2718 }
2719 rcu_read_unlock();
2720 lttng_ht_destroy(indexes_ht);
2721 indexes_ht_error:
2722 lttng_ht_destroy(relay_connections_ht);
2723 relay_connections_ht_error:
2724 /* Close relay conn pipes */
2725 utils_close_pipe(relay_conn_pipe);
2726 if (err) {
2727 DBG("Thread exited with error");
2728 }
2729 DBG("Worker thread cleanup complete");
2730 free(data_buffer);
2731 error_testpoint:
2732 if (err) {
2733 health_error();
2734 ERR("Health error occurred in %s", __func__);
2735 }
2736 health_unregister(health_relayd);
2737 rcu_unregister_thread();
2738 lttng_relay_stop_threads();
2739 return NULL;
2740 }
2741
2742 /*
2743 * Create the relay command pipe to wake thread_manage_apps.
2744 * Closed in cleanup().
2745 */
2746 static int create_relay_conn_pipe(void)
2747 {
2748 int ret;
2749
2750 ret = utils_create_pipe_cloexec(relay_conn_pipe);
2751
2752 return ret;
2753 }
2754
2755 /*
2756 * main
2757 */
2758 int main(int argc, char **argv)
2759 {
2760 int ret = 0, retval = 0;
2761 void *status;
2762 struct relay_local_data *relay_ctx = NULL;
2763
2764 /* Parse arguments */
2765 progname = argv[0];
2766 if (set_options(argc, argv)) {
2767 retval = -1;
2768 goto exit_options;
2769 }
2770
2771 if (set_signal_handler()) {
2772 retval = -1;
2773 goto exit_options;
2774 }
2775
2776 /* Try to create directory if -o, --output is specified. */
2777 if (opt_output_path) {
2778 if (*opt_output_path != '/') {
2779 ERR("Please specify an absolute path for -o, --output PATH");
2780 retval = -1;
2781 goto exit_options;
2782 }
2783
2784 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG,
2785 -1, -1);
2786 if (ret < 0) {
2787 ERR("Unable to create %s", opt_output_path);
2788 retval = -1;
2789 goto exit_options;
2790 }
2791 }
2792
2793 /* Daemonize */
2794 if (opt_daemon || opt_background) {
2795 int i;
2796
2797 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2798 !opt_background);
2799 if (ret < 0) {
2800 retval = -1;
2801 goto exit_options;
2802 }
2803
2804 /*
2805 * We are in the child. Make sure all other file
2806 * descriptors are closed, in case we are called with
2807 * more opened file descriptors than the standard ones.
2808 */
2809 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2810 (void) close(i);
2811 }
2812 }
2813
2814
2815 /* Initialize thread health monitoring */
2816 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2817 if (!health_relayd) {
2818 PERROR("health_app_create error");
2819 retval = -1;
2820 goto exit_health_app_create;
2821 }
2822
2823 /* Create thread quit pipe */
2824 if (init_thread_quit_pipe()) {
2825 retval = -1;
2826 goto exit_init_data;
2827 }
2828
2829 /* We need those values for the file/dir creation. */
2830 relayd_uid = getuid();
2831 relayd_gid = getgid();
2832
2833 /* Check if daemon is UID = 0 */
2834 if (relayd_uid == 0) {
2835 if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) {
2836 ERR("Need to be root to use ports < 1024");
2837 retval = -1;
2838 goto exit_init_data;
2839 }
2840 }
2841
2842 /* Setup the thread apps communication pipe. */
2843 if (create_relay_conn_pipe()) {
2844 retval = -1;
2845 goto exit_init_data;
2846 }
2847
2848 /* Init relay command queue. */
2849 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
2850
2851 /* Set up max poll set size */
2852 lttng_poll_set_max_size();
2853
2854 /* Initialize communication library */
2855 lttcomm_init();
2856 lttcomm_inet_init();
2857
2858 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2859 if (!relay_ctx) {
2860 PERROR("relay_ctx");
2861 retval = -1;
2862 goto exit_init_data;
2863 }
2864
2865 /* tables of sessions indexed by session ID */
2866 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2867 if (!relay_ctx->sessions_ht) {
2868 retval = -1;
2869 goto exit_init_data;
2870 }
2871
2872 /* tables of streams indexed by stream ID */
2873 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2874 if (!relay_streams_ht) {
2875 retval = -1;
2876 goto exit_init_data;
2877 }
2878
2879 /* tables of streams indexed by stream ID */
2880 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2881 if (!viewer_streams_ht) {
2882 retval = -1;
2883 goto exit_init_data;
2884 }
2885
2886 ret = utils_create_pipe(health_quit_pipe);
2887 if (ret) {
2888 retval = -1;
2889 goto exit_health_quit_pipe;
2890 }
2891
2892 /* Create thread to manage the client socket */
2893 ret = pthread_create(&health_thread, NULL,
2894 thread_manage_health, (void *) NULL);
2895 if (ret) {
2896 errno = ret;
2897 PERROR("pthread_create health");
2898 retval = -1;
2899 goto exit_health_thread;
2900 }
2901
2902 /* Setup the dispatcher thread */
2903 ret = pthread_create(&dispatcher_thread, NULL,
2904 relay_thread_dispatcher, (void *) NULL);
2905 if (ret) {
2906 errno = ret;
2907 PERROR("pthread_create dispatcher");
2908 retval = -1;
2909 goto exit_dispatcher_thread;
2910 }
2911
2912 /* Setup the worker thread */
2913 ret = pthread_create(&worker_thread, NULL,
2914 relay_thread_worker, (void *) relay_ctx);
2915 if (ret) {
2916 errno = ret;
2917 PERROR("pthread_create worker");
2918 retval = -1;
2919 goto exit_worker_thread;
2920 }
2921
2922 /* Setup the listener thread */
2923 ret = pthread_create(&listener_thread, NULL,
2924 relay_thread_listener, (void *) NULL);
2925 if (ret) {
2926 errno = ret;
2927 PERROR("pthread_create listener");
2928 retval = -1;
2929 goto exit_listener_thread;
2930 }
2931
2932 ret = relayd_live_create(live_uri, relay_ctx);
2933 if (ret) {
2934 ERR("Starting live viewer threads");
2935 retval = -1;
2936 goto exit_live;
2937 }
2938
2939 /*
2940 * This is where we start awaiting program completion (e.g. through
2941 * signal that asks threads to teardown).
2942 */
2943
2944 ret = relayd_live_join();
2945 if (ret) {
2946 retval = -1;
2947 }
2948 exit_live:
2949
2950 ret = pthread_join(listener_thread, &status);
2951 if (ret) {
2952 errno = ret;
2953 PERROR("pthread_join listener_thread");
2954 retval = -1;
2955 }
2956
2957 exit_listener_thread:
2958 ret = pthread_join(worker_thread, &status);
2959 if (ret) {
2960 errno = ret;
2961 PERROR("pthread_join worker_thread");
2962 retval = -1;
2963 }
2964
2965 exit_worker_thread:
2966 ret = pthread_join(dispatcher_thread, &status);
2967 if (ret) {
2968 errno = ret;
2969 PERROR("pthread_join dispatcher_thread");
2970 retval = -1;
2971 }
2972 exit_dispatcher_thread:
2973
2974 ret = pthread_join(health_thread, &status);
2975 if (ret) {
2976 errno = ret;
2977 PERROR("pthread_join health_thread");
2978 retval = -1;
2979 }
2980 exit_health_thread:
2981
2982 utils_close_pipe(health_quit_pipe);
2983 exit_health_quit_pipe:
2984
2985 exit_init_data:
2986 health_app_destroy(health_relayd);
2987 exit_health_app_create:
2988 exit_options:
2989 relayd_cleanup(relay_ctx);
2990
2991 if (!retval) {
2992 exit(EXIT_SUCCESS);
2993 } else {
2994 exit(EXIT_FAILURE);
2995 }
2996 }
This page took 0.086655 seconds and 3 git commands to generate.