Fix: lttng-relayd allow binding of privileged ports for non-root users
[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 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2 only,
9 * as published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * more details.
15 *
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19 */
20
21 #define _GNU_SOURCE
22 #define _LGPL_SOURCE
23 #include <getopt.h>
24 #include <grp.h>
25 #include <limits.h>
26 #include <pthread.h>
27 #include <signal.h>
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <string.h>
31 #include <sys/mman.h>
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37 #include <sys/wait.h>
38 #include <inttypes.h>
39 #include <urcu/futex.h>
40 #include <urcu/uatomic.h>
41 #include <unistd.h>
42 #include <fcntl.h>
43 #include <config.h>
44
45 #include <lttng/lttng.h>
46 #include <common/common.h>
47 #include <common/compat/poll.h>
48 #include <common/compat/socket.h>
49 #include <common/compat/endian.h>
50 #include <common/compat/getenv.h>
51 #include <common/defaults.h>
52 #include <common/daemonize.h>
53 #include <common/futex.h>
54 #include <common/sessiond-comm/sessiond-comm.h>
55 #include <common/sessiond-comm/inet.h>
56 #include <common/sessiond-comm/relayd.h>
57 #include <common/uri.h>
58 #include <common/utils.h>
59 #include <common/config/config.h>
60 #include <urcu/rculist.h>
61
62 #include "cmd.h"
63 #include "ctf-trace.h"
64 #include "index.h"
65 #include "utils.h"
66 #include "lttng-relayd.h"
67 #include "live.h"
68 #include "health-relayd.h"
69 #include "testpoint.h"
70 #include "viewer-stream.h"
71 #include "session.h"
72 #include "stream.h"
73 #include "connection.h"
74 #include "tracefile-array.h"
75
76 /* command line options */
77 char *opt_output_path;
78 static int opt_daemon, opt_background;
79
80 /*
81 * We need to wait for listener and live listener threads, as well as
82 * health check thread, before being ready to signal readiness.
83 */
84 #define NR_LTTNG_RELAY_READY 3
85 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
86
87 /* Size of receive buffer. */
88 #define RECV_DATA_BUFFER_SIZE 65536
89
90 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
91 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
92
93 static struct lttng_uri *control_uri;
94 static struct lttng_uri *data_uri;
95 static struct lttng_uri *live_uri;
96
97 const char *progname;
98
99 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
100 static int tracing_group_name_override;
101
102 const char * const config_section_name = "relayd";
103
104 /*
105 * Quit pipe for all threads. This permits a single cancellation point
106 * for all threads when receiving an event on the pipe.
107 */
108 int thread_quit_pipe[2] = { -1, -1 };
109
110 /*
111 * This pipe is used to inform the worker thread that a command is queued and
112 * ready to be processed.
113 */
114 static int relay_conn_pipe[2] = { -1, -1 };
115
116 /* Shared between threads */
117 static int dispatch_thread_exit;
118
119 static pthread_t listener_thread;
120 static pthread_t dispatcher_thread;
121 static pthread_t worker_thread;
122 static pthread_t health_thread;
123
124 /*
125 * last_relay_stream_id_lock protects last_relay_stream_id increment
126 * atomicity on 32-bit architectures.
127 */
128 static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
129 static uint64_t last_relay_stream_id;
130
131 /*
132 * Relay command queue.
133 *
134 * The relay_thread_listener and relay_thread_dispatcher communicate with this
135 * queue.
136 */
137 static struct relay_conn_queue relay_conn_queue;
138
139 /* buffer allocated at startup, used to store the trace data */
140 static char *data_buffer;
141 static unsigned int data_buffer_size;
142
143 /* Global relay stream hash table. */
144 struct lttng_ht *relay_streams_ht;
145
146 /* Global relay viewer stream hash table. */
147 struct lttng_ht *viewer_streams_ht;
148
149 /* Global relay sessions hash table. */
150 struct lttng_ht *sessions_ht;
151
152 /* Relayd health monitoring */
153 struct health_app *health_relayd;
154
155 static struct option long_options[] = {
156 { "control-port", 1, 0, 'C', },
157 { "data-port", 1, 0, 'D', },
158 { "live-port", 1, 0, 'L', },
159 { "daemonize", 0, 0, 'd', },
160 { "background", 0, 0, 'b', },
161 { "group", 1, 0, 'g', },
162 { "help", 0, 0, 'h', },
163 { "output", 1, 0, 'o', },
164 { "verbose", 0, 0, 'v', },
165 { "config", 1, 0, 'f' },
166 { NULL, 0, 0, 0, },
167 };
168
169 static const char *config_ignore_options[] = { "help", "config" };
170
171 /*
172 * usage function on stderr
173 */
174 static void usage(void)
175 {
176 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
177 fprintf(stderr, " -h, --help Display this usage.\n");
178 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
179 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
180 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
181 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
182 fprintf(stderr, " -L, --live-port URL Live view port listening.\n");
183 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
184 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
185 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
186 fprintf(stderr, " -f --config Load daemon configuration file\n");
187 }
188
189 /*
190 * Take an option from the getopt output and set it in the right variable to be
191 * used later.
192 *
193 * Return 0 on success else a negative value.
194 */
195 static int set_option(int opt, const char *arg, const char *optname)
196 {
197 int ret;
198
199 switch (opt) {
200 case 0:
201 fprintf(stderr, "option %s", optname);
202 if (arg) {
203 fprintf(stderr, " with arg %s\n", arg);
204 }
205 break;
206 case 'C':
207 if (lttng_is_setuid_setgid()) {
208 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
209 "-C, --control-port");
210 } else {
211 ret = uri_parse(arg, &control_uri);
212 if (ret < 0) {
213 ERR("Invalid control URI specified");
214 goto end;
215 }
216 if (control_uri->port == 0) {
217 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
218 }
219 }
220 break;
221 case 'D':
222 if (lttng_is_setuid_setgid()) {
223 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
224 "-D, -data-port");
225 } else {
226 ret = uri_parse(arg, &data_uri);
227 if (ret < 0) {
228 ERR("Invalid data URI specified");
229 goto end;
230 }
231 if (data_uri->port == 0) {
232 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
233 }
234 }
235 break;
236 case 'L':
237 if (lttng_is_setuid_setgid()) {
238 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
239 "-L, -live-port");
240 } else {
241 ret = uri_parse(arg, &live_uri);
242 if (ret < 0) {
243 ERR("Invalid live URI specified");
244 goto end;
245 }
246 if (live_uri->port == 0) {
247 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
248 }
249 }
250 break;
251 case 'd':
252 opt_daemon = 1;
253 break;
254 case 'b':
255 opt_background = 1;
256 break;
257 case 'g':
258 if (lttng_is_setuid_setgid()) {
259 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
260 "-g, --group");
261 } else {
262 tracing_group_name = strdup(arg);
263 if (tracing_group_name == NULL) {
264 ret = -errno;
265 PERROR("strdup");
266 goto end;
267 }
268 tracing_group_name_override = 1;
269 }
270 break;
271 case 'h':
272 usage();
273 exit(EXIT_FAILURE);
274 case 'o':
275 if (lttng_is_setuid_setgid()) {
276 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
277 "-o, --output");
278 } else {
279 ret = asprintf(&opt_output_path, "%s", arg);
280 if (ret < 0) {
281 ret = -errno;
282 PERROR("asprintf opt_output_path");
283 goto end;
284 }
285 }
286 break;
287 case 'v':
288 /* Verbose level can increase using multiple -v */
289 if (arg) {
290 lttng_opt_verbose = config_parse_value(arg);
291 } else {
292 /* Only 3 level of verbosity (-vvv). */
293 if (lttng_opt_verbose < 3) {
294 lttng_opt_verbose += 1;
295 }
296 }
297 break;
298 default:
299 /* Unknown option or other error.
300 * Error is printed by getopt, just return */
301 ret = -1;
302 goto end;
303 }
304
305 /* All good. */
306 ret = 0;
307
308 end:
309 return ret;
310 }
311
312 /*
313 * config_entry_handler_cb used to handle options read from a config file.
314 * See config_entry_handler_cb comment in common/config/config.h for the
315 * return value conventions.
316 */
317 static int config_entry_handler(const struct config_entry *entry, void *unused)
318 {
319 int ret = 0, i;
320
321 if (!entry || !entry->name || !entry->value) {
322 ret = -EINVAL;
323 goto end;
324 }
325
326 /* Check if the option is to be ignored */
327 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
328 if (!strcmp(entry->name, config_ignore_options[i])) {
329 goto end;
330 }
331 }
332
333 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
334 /* Ignore if entry name is not fully matched. */
335 if (strcmp(entry->name, long_options[i].name)) {
336 continue;
337 }
338
339 /*
340 * If the option takes no argument on the command line,
341 * we have to check if the value is "true". We support
342 * non-zero numeric values, true, on and yes.
343 */
344 if (!long_options[i].has_arg) {
345 ret = config_parse_value(entry->value);
346 if (ret <= 0) {
347 if (ret) {
348 WARN("Invalid configuration value \"%s\" for option %s",
349 entry->value, entry->name);
350 }
351 /* False, skip boolean config option. */
352 goto end;
353 }
354 }
355
356 ret = set_option(long_options[i].val, entry->value, entry->name);
357 goto end;
358 }
359
360 WARN("Unrecognized option \"%s\" in daemon configuration file.",
361 entry->name);
362
363 end:
364 return ret;
365 }
366
367 static int set_options(int argc, char **argv)
368 {
369 int c, ret = 0, option_index = 0, retval = 0;
370 int orig_optopt = optopt, orig_optind = optind;
371 char *default_address, *optstring;
372 const char *config_path = NULL;
373
374 optstring = utils_generate_optstring(long_options,
375 sizeof(long_options) / sizeof(struct option));
376 if (!optstring) {
377 retval = -ENOMEM;
378 goto exit;
379 }
380
381 /* Check for the --config option */
382
383 while ((c = getopt_long(argc, argv, optstring, long_options,
384 &option_index)) != -1) {
385 if (c == '?') {
386 retval = -EINVAL;
387 goto exit;
388 } else if (c != 'f') {
389 continue;
390 }
391
392 if (lttng_is_setuid_setgid()) {
393 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
394 "-f, --config");
395 } else {
396 config_path = utils_expand_path(optarg);
397 if (!config_path) {
398 ERR("Failed to resolve path: %s", optarg);
399 }
400 }
401 }
402
403 ret = config_get_section_entries(config_path, config_section_name,
404 config_entry_handler, NULL);
405 if (ret) {
406 if (ret > 0) {
407 ERR("Invalid configuration option at line %i", ret);
408 }
409 retval = -1;
410 goto exit;
411 }
412
413 /* Reset getopt's global state */
414 optopt = orig_optopt;
415 optind = orig_optind;
416 while (1) {
417 c = getopt_long(argc, argv, optstring, long_options, &option_index);
418 if (c == -1) {
419 break;
420 }
421
422 ret = set_option(c, optarg, long_options[option_index].name);
423 if (ret < 0) {
424 retval = -1;
425 goto exit;
426 }
427 }
428
429 /* assign default values */
430 if (control_uri == NULL) {
431 ret = asprintf(&default_address,
432 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
433 DEFAULT_NETWORK_CONTROL_PORT);
434 if (ret < 0) {
435 PERROR("asprintf default data address");
436 retval = -1;
437 goto exit;
438 }
439
440 ret = uri_parse(default_address, &control_uri);
441 free(default_address);
442 if (ret < 0) {
443 ERR("Invalid control URI specified");
444 retval = -1;
445 goto exit;
446 }
447 }
448 if (data_uri == NULL) {
449 ret = asprintf(&default_address,
450 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
451 DEFAULT_NETWORK_DATA_PORT);
452 if (ret < 0) {
453 PERROR("asprintf default data address");
454 retval = -1;
455 goto exit;
456 }
457
458 ret = uri_parse(default_address, &data_uri);
459 free(default_address);
460 if (ret < 0) {
461 ERR("Invalid data URI specified");
462 retval = -1;
463 goto exit;
464 }
465 }
466 if (live_uri == NULL) {
467 ret = asprintf(&default_address,
468 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
469 DEFAULT_NETWORK_VIEWER_PORT);
470 if (ret < 0) {
471 PERROR("asprintf default viewer control address");
472 retval = -1;
473 goto exit;
474 }
475
476 ret = uri_parse(default_address, &live_uri);
477 free(default_address);
478 if (ret < 0) {
479 ERR("Invalid viewer control URI specified");
480 retval = -1;
481 goto exit;
482 }
483 }
484
485 exit:
486 free(optstring);
487 return retval;
488 }
489
490 static void print_global_objects(void)
491 {
492 rcu_register_thread();
493
494 print_viewer_streams();
495 print_relay_streams();
496 print_sessions();
497
498 rcu_unregister_thread();
499 }
500
501 /*
502 * Cleanup the daemon
503 */
504 static void relayd_cleanup(void)
505 {
506 print_global_objects();
507
508 DBG("Cleaning up");
509
510 if (viewer_streams_ht)
511 lttng_ht_destroy(viewer_streams_ht);
512 if (relay_streams_ht)
513 lttng_ht_destroy(relay_streams_ht);
514 if (sessions_ht)
515 lttng_ht_destroy(sessions_ht);
516
517 /* free the dynamically allocated opt_output_path */
518 free(opt_output_path);
519
520 /* Close thread quit pipes */
521 utils_close_pipe(thread_quit_pipe);
522
523 uri_free(control_uri);
524 uri_free(data_uri);
525 /* Live URI is freed in the live thread. */
526
527 if (tracing_group_name_override) {
528 free((void *) tracing_group_name);
529 }
530 }
531
532 /*
533 * Write to writable pipe used to notify a thread.
534 */
535 static int notify_thread_pipe(int wpipe)
536 {
537 ssize_t ret;
538
539 ret = lttng_write(wpipe, "!", 1);
540 if (ret < 1) {
541 PERROR("write poll pipe");
542 goto end;
543 }
544 ret = 0;
545 end:
546 return ret;
547 }
548
549 static int notify_health_quit_pipe(int *pipe)
550 {
551 ssize_t ret;
552
553 ret = lttng_write(pipe[1], "4", 1);
554 if (ret < 1) {
555 PERROR("write relay health quit");
556 goto end;
557 }
558 ret = 0;
559 end:
560 return ret;
561 }
562
563 /*
564 * Stop all relayd and relayd-live threads.
565 */
566 int lttng_relay_stop_threads(void)
567 {
568 int retval = 0;
569
570 /* Stopping all threads */
571 DBG("Terminating all threads");
572 if (notify_thread_pipe(thread_quit_pipe[1])) {
573 ERR("write error on thread quit pipe");
574 retval = -1;
575 }
576
577 if (notify_health_quit_pipe(health_quit_pipe)) {
578 ERR("write error on health quit pipe");
579 }
580
581 /* Dispatch thread */
582 CMM_STORE_SHARED(dispatch_thread_exit, 1);
583 futex_nto1_wake(&relay_conn_queue.futex);
584
585 if (relayd_live_stop()) {
586 ERR("Error stopping live threads");
587 retval = -1;
588 }
589 return retval;
590 }
591
592 /*
593 * Signal handler for the daemon
594 *
595 * Simply stop all worker threads, leaving main() return gracefully after
596 * joining all threads and calling cleanup().
597 */
598 static void sighandler(int sig)
599 {
600 switch (sig) {
601 case SIGPIPE:
602 DBG("SIGPIPE caught");
603 return;
604 case SIGINT:
605 DBG("SIGINT caught");
606 if (lttng_relay_stop_threads()) {
607 ERR("Error stopping threads");
608 }
609 break;
610 case SIGTERM:
611 DBG("SIGTERM caught");
612 if (lttng_relay_stop_threads()) {
613 ERR("Error stopping threads");
614 }
615 break;
616 case SIGUSR1:
617 CMM_STORE_SHARED(recv_child_signal, 1);
618 break;
619 default:
620 break;
621 }
622 }
623
624 /*
625 * Setup signal handler for :
626 * SIGINT, SIGTERM, SIGPIPE
627 */
628 static int set_signal_handler(void)
629 {
630 int ret = 0;
631 struct sigaction sa;
632 sigset_t sigset;
633
634 if ((ret = sigemptyset(&sigset)) < 0) {
635 PERROR("sigemptyset");
636 return ret;
637 }
638
639 sa.sa_handler = sighandler;
640 sa.sa_mask = sigset;
641 sa.sa_flags = 0;
642 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
643 PERROR("sigaction");
644 return ret;
645 }
646
647 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
648 PERROR("sigaction");
649 return ret;
650 }
651
652 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
653 PERROR("sigaction");
654 return ret;
655 }
656
657 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
658 PERROR("sigaction");
659 return ret;
660 }
661
662 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
663
664 return ret;
665 }
666
667 void lttng_relay_notify_ready(void)
668 {
669 /* Notify the parent of the fork() process that we are ready. */
670 if (opt_daemon || opt_background) {
671 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
672 kill(child_ppid, SIGUSR1);
673 }
674 }
675 }
676
677 /*
678 * Init thread quit pipe.
679 *
680 * Return -1 on error or 0 if all pipes are created.
681 */
682 static int init_thread_quit_pipe(void)
683 {
684 int ret;
685
686 ret = utils_create_pipe_cloexec(thread_quit_pipe);
687
688 return ret;
689 }
690
691 /*
692 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
693 */
694 static int create_thread_poll_set(struct lttng_poll_event *events, int size)
695 {
696 int ret;
697
698 if (events == NULL || size == 0) {
699 ret = -1;
700 goto error;
701 }
702
703 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
704 if (ret < 0) {
705 goto error;
706 }
707
708 /* Add quit pipe */
709 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
710 if (ret < 0) {
711 goto error;
712 }
713
714 return 0;
715
716 error:
717 return ret;
718 }
719
720 /*
721 * Check if the thread quit pipe was triggered.
722 *
723 * Return 1 if it was triggered else 0;
724 */
725 static int check_thread_quit_pipe(int fd, uint32_t events)
726 {
727 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
728 return 1;
729 }
730
731 return 0;
732 }
733
734 /*
735 * Create and init socket from uri.
736 */
737 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri)
738 {
739 int ret;
740 struct lttcomm_sock *sock = NULL;
741
742 sock = lttcomm_alloc_sock_from_uri(uri);
743 if (sock == NULL) {
744 ERR("Allocating socket");
745 goto error;
746 }
747
748 ret = lttcomm_create_sock(sock);
749 if (ret < 0) {
750 goto error;
751 }
752 DBG("Listening on sock %d", sock->fd);
753
754 ret = sock->ops->bind(sock);
755 if (ret < 0) {
756 goto error;
757 }
758
759 ret = sock->ops->listen(sock, -1);
760 if (ret < 0) {
761 goto error;
762
763 }
764
765 return sock;
766
767 error:
768 if (sock) {
769 lttcomm_destroy_sock(sock);
770 }
771 return NULL;
772 }
773
774 /*
775 * This thread manages the listening for new connections on the network
776 */
777 static void *relay_thread_listener(void *data)
778 {
779 int i, ret, pollfd, err = -1;
780 uint32_t revents, nb_fd;
781 struct lttng_poll_event events;
782 struct lttcomm_sock *control_sock, *data_sock;
783
784 DBG("[thread] Relay listener started");
785
786 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
787
788 health_code_update();
789
790 control_sock = relay_socket_create(control_uri);
791 if (!control_sock) {
792 goto error_sock_control;
793 }
794
795 data_sock = relay_socket_create(data_uri);
796 if (!data_sock) {
797 goto error_sock_relay;
798 }
799
800 /*
801 * Pass 3 as size here for the thread quit pipe, control and
802 * data socket.
803 */
804 ret = create_thread_poll_set(&events, 3);
805 if (ret < 0) {
806 goto error_create_poll;
807 }
808
809 /* Add the control socket */
810 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
811 if (ret < 0) {
812 goto error_poll_add;
813 }
814
815 /* Add the data socket */
816 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
817 if (ret < 0) {
818 goto error_poll_add;
819 }
820
821 lttng_relay_notify_ready();
822
823 if (testpoint(relayd_thread_listener)) {
824 goto error_testpoint;
825 }
826
827 while (1) {
828 health_code_update();
829
830 DBG("Listener accepting connections");
831
832 restart:
833 health_poll_entry();
834 ret = lttng_poll_wait(&events, -1);
835 health_poll_exit();
836 if (ret < 0) {
837 /*
838 * Restart interrupted system call.
839 */
840 if (errno == EINTR) {
841 goto restart;
842 }
843 goto error;
844 }
845
846 nb_fd = ret;
847
848 DBG("Relay new connection received");
849 for (i = 0; i < nb_fd; i++) {
850 health_code_update();
851
852 /* Fetch once the poll data */
853 revents = LTTNG_POLL_GETEV(&events, i);
854 pollfd = LTTNG_POLL_GETFD(&events, i);
855
856 if (!revents) {
857 /*
858 * No activity for this FD (poll
859 * implementation).
860 */
861 continue;
862 }
863
864 /* Thread quit pipe has been closed. Killing thread. */
865 ret = check_thread_quit_pipe(pollfd, revents);
866 if (ret) {
867 err = 0;
868 goto exit;
869 }
870
871 if (revents & LPOLLIN) {
872 /*
873 * A new connection is requested, therefore a
874 * sessiond/consumerd connection is allocated in
875 * this thread, enqueued to a global queue and
876 * dequeued (and freed) in the worker thread.
877 */
878 int val = 1;
879 struct relay_connection *new_conn;
880 struct lttcomm_sock *newsock;
881 enum connection_type type;
882
883 if (pollfd == data_sock->fd) {
884 type = RELAY_DATA;
885 newsock = data_sock->ops->accept(data_sock);
886 DBG("Relay data connection accepted, socket %d",
887 newsock->fd);
888 } else {
889 assert(pollfd == control_sock->fd);
890 type = RELAY_CONTROL;
891 newsock = control_sock->ops->accept(control_sock);
892 DBG("Relay control connection accepted, socket %d",
893 newsock->fd);
894 }
895 if (!newsock) {
896 PERROR("accepting sock");
897 goto error;
898 }
899
900 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
901 sizeof(val));
902 if (ret < 0) {
903 PERROR("setsockopt inet");
904 lttcomm_destroy_sock(newsock);
905 goto error;
906 }
907 new_conn = connection_create(newsock, type);
908 if (!new_conn) {
909 lttcomm_destroy_sock(newsock);
910 goto error;
911 }
912
913 /* Enqueue request for the dispatcher thread. */
914 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
915 &new_conn->qnode);
916
917 /*
918 * Wake the dispatch queue futex.
919 * Implicit memory barrier with the
920 * exchange in cds_wfcq_enqueue.
921 */
922 futex_nto1_wake(&relay_conn_queue.futex);
923 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
924 ERR("socket poll error");
925 goto error;
926 } else {
927 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
928 goto error;
929 }
930 }
931 }
932
933 exit:
934 error:
935 error_poll_add:
936 error_testpoint:
937 lttng_poll_clean(&events);
938 error_create_poll:
939 if (data_sock->fd >= 0) {
940 ret = data_sock->ops->close(data_sock);
941 if (ret) {
942 PERROR("close");
943 }
944 }
945 lttcomm_destroy_sock(data_sock);
946 error_sock_relay:
947 if (control_sock->fd >= 0) {
948 ret = control_sock->ops->close(control_sock);
949 if (ret) {
950 PERROR("close");
951 }
952 }
953 lttcomm_destroy_sock(control_sock);
954 error_sock_control:
955 if (err) {
956 health_error();
957 ERR("Health error occurred in %s", __func__);
958 }
959 health_unregister(health_relayd);
960 DBG("Relay listener thread cleanup complete");
961 lttng_relay_stop_threads();
962 return NULL;
963 }
964
965 /*
966 * This thread manages the dispatching of the requests to worker threads
967 */
968 static void *relay_thread_dispatcher(void *data)
969 {
970 int err = -1;
971 ssize_t ret;
972 struct cds_wfcq_node *node;
973 struct relay_connection *new_conn = NULL;
974
975 DBG("[thread] Relay dispatcher started");
976
977 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
978
979 if (testpoint(relayd_thread_dispatcher)) {
980 goto error_testpoint;
981 }
982
983 health_code_update();
984
985 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
986 health_code_update();
987
988 /* Atomically prepare the queue futex */
989 futex_nto1_prepare(&relay_conn_queue.futex);
990
991 do {
992 health_code_update();
993
994 /* Dequeue commands */
995 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
996 &relay_conn_queue.tail);
997 if (node == NULL) {
998 DBG("Woken up but nothing in the relay command queue");
999 /* Continue thread execution */
1000 break;
1001 }
1002 new_conn = caa_container_of(node, struct relay_connection, qnode);
1003
1004 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1005
1006 /*
1007 * Inform worker thread of the new request. This
1008 * call is blocking so we can be assured that
1009 * the data will be read at some point in time
1010 * or wait to the end of the world :)
1011 */
1012 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1013 if (ret < 0) {
1014 PERROR("write connection pipe");
1015 connection_put(new_conn);
1016 goto error;
1017 }
1018 } while (node != NULL);
1019
1020 /* Futex wait on queue. Blocking call on futex() */
1021 health_poll_entry();
1022 futex_nto1_wait(&relay_conn_queue.futex);
1023 health_poll_exit();
1024 }
1025
1026 /* Normal exit, no error */
1027 err = 0;
1028
1029 error:
1030 error_testpoint:
1031 if (err) {
1032 health_error();
1033 ERR("Health error occurred in %s", __func__);
1034 }
1035 health_unregister(health_relayd);
1036 DBG("Dispatch thread dying");
1037 lttng_relay_stop_threads();
1038 return NULL;
1039 }
1040
1041 /*
1042 * Set index data from the control port to a given index object.
1043 */
1044 static int set_index_control_data(struct relay_index *index,
1045 struct lttcomm_relayd_index *data)
1046 {
1047 struct ctf_packet_index index_data;
1048
1049 /*
1050 * The index on disk is encoded in big endian, so we don't need
1051 * to convert the data received on the network. The data_offset
1052 * value is NEVER modified here and is updated by the data
1053 * thread.
1054 */
1055 index_data.packet_size = data->packet_size;
1056 index_data.content_size = data->content_size;
1057 index_data.timestamp_begin = data->timestamp_begin;
1058 index_data.timestamp_end = data->timestamp_end;
1059 index_data.events_discarded = data->events_discarded;
1060 index_data.stream_id = data->stream_id;
1061 return relay_index_set_data(index, &index_data);
1062 }
1063
1064 /*
1065 * Handle the RELAYD_CREATE_SESSION command.
1066 *
1067 * On success, send back the session id or else return a negative value.
1068 */
1069 static int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
1070 struct relay_connection *conn)
1071 {
1072 int ret = 0, send_ret;
1073 struct relay_session *session;
1074 struct lttcomm_relayd_status_session reply;
1075 char session_name[NAME_MAX];
1076 char hostname[HOST_NAME_MAX];
1077 uint32_t live_timer = 0;
1078 bool snapshot = false;
1079
1080 memset(session_name, 0, NAME_MAX);
1081 memset(hostname, 0, HOST_NAME_MAX);
1082
1083 memset(&reply, 0, sizeof(reply));
1084
1085 switch (conn->minor) {
1086 case 1:
1087 case 2:
1088 case 3:
1089 break;
1090 case 4: /* LTTng sessiond 2.4 */
1091 default:
1092 ret = cmd_create_session_2_4(conn, session_name,
1093 hostname, &live_timer, &snapshot);
1094 }
1095 if (ret < 0) {
1096 goto send_reply;
1097 }
1098
1099 session = session_create(session_name, hostname, live_timer,
1100 snapshot, conn->major, conn->minor);
1101 if (!session) {
1102 ret = -1;
1103 goto send_reply;
1104 }
1105 assert(!conn->session);
1106 conn->session = session;
1107 DBG("Created session %" PRIu64, session->id);
1108
1109 reply.session_id = htobe64(session->id);
1110
1111 send_reply:
1112 if (ret < 0) {
1113 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1114 } else {
1115 reply.ret_code = htobe32(LTTNG_OK);
1116 }
1117
1118 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1119 if (send_ret < 0) {
1120 ERR("Relayd sending session id");
1121 ret = send_ret;
1122 }
1123
1124 return ret;
1125 }
1126
1127 /*
1128 * When we have received all the streams and the metadata for a channel,
1129 * we make them visible to the viewer threads.
1130 */
1131 static void publish_connection_local_streams(struct relay_connection *conn)
1132 {
1133 struct relay_stream *stream;
1134 struct relay_session *session = conn->session;
1135
1136 /*
1137 * We publish all streams belonging to a session atomically wrt
1138 * session lock.
1139 */
1140 pthread_mutex_lock(&session->lock);
1141 rcu_read_lock();
1142 cds_list_for_each_entry_rcu(stream, &session->recv_list,
1143 recv_node) {
1144 stream_publish(stream);
1145 }
1146 rcu_read_unlock();
1147
1148 /*
1149 * Inform the viewer that there are new streams in the session.
1150 */
1151 if (session->viewer_attached) {
1152 uatomic_set(&session->new_streams, 1);
1153 }
1154 pthread_mutex_unlock(&session->lock);
1155 }
1156
1157 /*
1158 * relay_add_stream: allocate a new stream for a session
1159 */
1160 static int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1161 struct relay_connection *conn)
1162 {
1163 int ret;
1164 ssize_t send_ret;
1165 struct relay_session *session = conn->session;
1166 struct relay_stream *stream = NULL;
1167 struct lttcomm_relayd_status_stream reply;
1168 struct ctf_trace *trace = NULL;
1169 uint64_t stream_handle = -1ULL;
1170 char *path_name = NULL, *channel_name = NULL;
1171 uint64_t tracefile_size = 0, tracefile_count = 0;
1172
1173 if (!session || conn->version_check_done == 0) {
1174 ERR("Trying to add a stream before version check");
1175 ret = -1;
1176 goto end_no_session;
1177 }
1178
1179 switch (session->minor) {
1180 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
1181 ret = cmd_recv_stream_2_1(conn, &path_name,
1182 &channel_name);
1183 break;
1184 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
1185 default:
1186 ret = cmd_recv_stream_2_2(conn, &path_name,
1187 &channel_name, &tracefile_size, &tracefile_count);
1188 break;
1189 }
1190 if (ret < 0) {
1191 goto send_reply;
1192 }
1193
1194 trace = ctf_trace_get_by_path_or_create(session, path_name);
1195 if (!trace) {
1196 goto send_reply;
1197 }
1198 /* This stream here has one reference on the trace. */
1199
1200 pthread_mutex_lock(&last_relay_stream_id_lock);
1201 stream_handle = ++last_relay_stream_id;
1202 pthread_mutex_unlock(&last_relay_stream_id_lock);
1203
1204 /* We pass ownership of path_name and channel_name. */
1205 stream = stream_create(trace, stream_handle, path_name,
1206 channel_name, tracefile_size, tracefile_count);
1207 path_name = NULL;
1208 channel_name = NULL;
1209
1210 /*
1211 * Streams are the owners of their trace. Reference to trace is
1212 * kept within stream_create().
1213 */
1214 ctf_trace_put(trace);
1215
1216 send_reply:
1217 memset(&reply, 0, sizeof(reply));
1218 reply.handle = htobe64(stream_handle);
1219 if (!stream) {
1220 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1221 } else {
1222 reply.ret_code = htobe32(LTTNG_OK);
1223 }
1224
1225 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1226 sizeof(struct lttcomm_relayd_status_stream), 0);
1227 if (send_ret < 0) {
1228 ERR("Relay sending stream id");
1229 ret = (int) send_ret;
1230 }
1231
1232 end_no_session:
1233 free(path_name);
1234 free(channel_name);
1235 return ret;
1236 }
1237
1238 /*
1239 * relay_close_stream: close a specific stream
1240 */
1241 static int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1242 struct relay_connection *conn)
1243 {
1244 int ret, send_ret;
1245 struct relay_session *session = conn->session;
1246 struct lttcomm_relayd_close_stream stream_info;
1247 struct lttcomm_relayd_generic_reply reply;
1248 struct relay_stream *stream;
1249
1250 DBG("Close stream received");
1251
1252 if (!session || conn->version_check_done == 0) {
1253 ERR("Trying to close a stream before version check");
1254 ret = -1;
1255 goto end_no_session;
1256 }
1257
1258 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
1259 sizeof(struct lttcomm_relayd_close_stream), 0);
1260 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1261 if (ret == 0) {
1262 /* Orderly shutdown. Not necessary to print an error. */
1263 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1264 } else {
1265 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1266 }
1267 ret = -1;
1268 goto end_no_session;
1269 }
1270
1271 stream = stream_get_by_id(be64toh(stream_info.stream_id));
1272 if (!stream) {
1273 ret = -1;
1274 goto end;
1275 }
1276
1277 /*
1278 * Set last_net_seq_num before the close flag. Required by data
1279 * pending check.
1280 */
1281 pthread_mutex_lock(&stream->lock);
1282 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1283 pthread_mutex_unlock(&stream->lock);
1284
1285 /*
1286 * This is one of the conditions which may trigger a stream close
1287 * with the others being:
1288 * 1) A close command is received for a stream
1289 * 2) The control connection owning the stream is closed
1290 * 3) We have received all of the stream's data _after_ a close
1291 * request.
1292 */
1293 try_stream_close(stream);
1294 if (stream->is_metadata) {
1295 struct relay_viewer_stream *vstream;
1296
1297 vstream = viewer_stream_get_by_id(stream->stream_handle);
1298 if (vstream) {
1299 if (vstream->metadata_sent == stream->metadata_received) {
1300 /*
1301 * Since all the metadata has been sent to the
1302 * viewer and that we have a request to close
1303 * its stream, we can safely teardown the
1304 * corresponding metadata viewer stream.
1305 */
1306 viewer_stream_put(vstream);
1307 }
1308 /* Put local reference. */
1309 viewer_stream_put(vstream);
1310 }
1311 }
1312 stream_put(stream);
1313
1314 end:
1315 memset(&reply, 0, sizeof(reply));
1316 if (ret < 0) {
1317 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1318 } else {
1319 reply.ret_code = htobe32(LTTNG_OK);
1320 }
1321 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1322 sizeof(struct lttcomm_relayd_generic_reply), 0);
1323 if (send_ret < 0) {
1324 ERR("Relay sending stream id");
1325 ret = send_ret;
1326 }
1327
1328 end_no_session:
1329 return ret;
1330 }
1331
1332 /*
1333 * relay_unknown_command: send -1 if received unknown command
1334 */
1335 static void relay_unknown_command(struct relay_connection *conn)
1336 {
1337 struct lttcomm_relayd_generic_reply reply;
1338 int ret;
1339
1340 memset(&reply, 0, sizeof(reply));
1341 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1342 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1343 sizeof(struct lttcomm_relayd_generic_reply), 0);
1344 if (ret < 0) {
1345 ERR("Relay sending unknown command");
1346 }
1347 }
1348
1349 /*
1350 * relay_start: send an acknowledgment to the client to tell if we are
1351 * ready to receive data. We are ready if a session is established.
1352 */
1353 static int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1354 struct relay_connection *conn)
1355 {
1356 int ret = htobe32(LTTNG_OK);
1357 struct lttcomm_relayd_generic_reply reply;
1358 struct relay_session *session = conn->session;
1359
1360 if (!session) {
1361 DBG("Trying to start the streaming without a session established");
1362 ret = htobe32(LTTNG_ERR_UNK);
1363 }
1364
1365 memset(&reply, 0, sizeof(reply));
1366 reply.ret_code = ret;
1367 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1368 sizeof(struct lttcomm_relayd_generic_reply), 0);
1369 if (ret < 0) {
1370 ERR("Relay sending start ack");
1371 }
1372
1373 return ret;
1374 }
1375
1376 /*
1377 * Append padding to the file pointed by the file descriptor fd.
1378 */
1379 static int write_padding_to_file(int fd, uint32_t size)
1380 {
1381 ssize_t ret = 0;
1382 char *zeros;
1383
1384 if (size == 0) {
1385 goto end;
1386 }
1387
1388 zeros = zmalloc(size);
1389 if (zeros == NULL) {
1390 PERROR("zmalloc zeros for padding");
1391 ret = -1;
1392 goto end;
1393 }
1394
1395 ret = lttng_write(fd, zeros, size);
1396 if (ret < size) {
1397 PERROR("write padding to file");
1398 }
1399
1400 free(zeros);
1401
1402 end:
1403 return ret;
1404 }
1405
1406 /*
1407 * relay_recv_metadata: receive the metadata for the session.
1408 */
1409 static int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1410 struct relay_connection *conn)
1411 {
1412 int ret = 0;
1413 ssize_t size_ret;
1414 struct relay_session *session = conn->session;
1415 struct lttcomm_relayd_metadata_payload *metadata_struct;
1416 struct relay_stream *metadata_stream;
1417 uint64_t data_size, payload_size;
1418
1419 if (!session) {
1420 ERR("Metadata sent before version check");
1421 ret = -1;
1422 goto end;
1423 }
1424
1425 data_size = payload_size = be64toh(recv_hdr->data_size);
1426 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1427 ERR("Incorrect data size");
1428 ret = -1;
1429 goto end;
1430 }
1431 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1432
1433 if (data_buffer_size < data_size) {
1434 /* In case the realloc fails, we can free the memory */
1435 char *tmp_data_ptr;
1436
1437 tmp_data_ptr = realloc(data_buffer, data_size);
1438 if (!tmp_data_ptr) {
1439 ERR("Allocating data buffer");
1440 free(data_buffer);
1441 ret = -1;
1442 goto end;
1443 }
1444 data_buffer = tmp_data_ptr;
1445 data_buffer_size = data_size;
1446 }
1447 memset(data_buffer, 0, data_size);
1448 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1449 size_ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
1450 if (size_ret < 0 || size_ret != data_size) {
1451 if (size_ret == 0) {
1452 /* Orderly shutdown. Not necessary to print an error. */
1453 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1454 } else {
1455 ERR("Relay didn't receive the whole metadata");
1456 }
1457 ret = -1;
1458 goto end;
1459 }
1460 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1461
1462 metadata_stream = stream_get_by_id(be64toh(metadata_struct->stream_id));
1463 if (!metadata_stream) {
1464 ret = -1;
1465 goto end;
1466 }
1467
1468 pthread_mutex_lock(&metadata_stream->lock);
1469
1470 size_ret = lttng_write(metadata_stream->stream_fd->fd, metadata_struct->payload,
1471 payload_size);
1472 if (size_ret < payload_size) {
1473 ERR("Relay error writing metadata on file");
1474 ret = -1;
1475 goto end_put;
1476 }
1477
1478 size_ret = write_padding_to_file(metadata_stream->stream_fd->fd,
1479 be32toh(metadata_struct->padding_size));
1480 if (size_ret < 0) {
1481 goto end_put;
1482 }
1483
1484 metadata_stream->metadata_received +=
1485 payload_size + be32toh(metadata_struct->padding_size);
1486 DBG2("Relay metadata written. Updated metadata_received %" PRIu64,
1487 metadata_stream->metadata_received);
1488
1489 end_put:
1490 pthread_mutex_unlock(&metadata_stream->lock);
1491 stream_put(metadata_stream);
1492 end:
1493 return ret;
1494 }
1495
1496 /*
1497 * relay_send_version: send relayd version number
1498 */
1499 static int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1500 struct relay_connection *conn)
1501 {
1502 int ret;
1503 struct lttcomm_relayd_version reply, msg;
1504
1505 conn->version_check_done = 1;
1506
1507 /* Get version from the other side. */
1508 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1509 if (ret < 0 || ret != sizeof(msg)) {
1510 if (ret == 0) {
1511 /* Orderly shutdown. Not necessary to print an error. */
1512 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1513 } else {
1514 ERR("Relay failed to receive the version values.");
1515 }
1516 ret = -1;
1517 goto end;
1518 }
1519
1520 memset(&reply, 0, sizeof(reply));
1521 reply.major = RELAYD_VERSION_COMM_MAJOR;
1522 reply.minor = RELAYD_VERSION_COMM_MINOR;
1523
1524 /* Major versions must be the same */
1525 if (reply.major != be32toh(msg.major)) {
1526 DBG("Incompatible major versions (%u vs %u), deleting session",
1527 reply.major, be32toh(msg.major));
1528 connection_put(conn);
1529 ret = 0;
1530 goto end;
1531 }
1532
1533 conn->major = reply.major;
1534 /* We adapt to the lowest compatible version */
1535 if (reply.minor <= be32toh(msg.minor)) {
1536 conn->minor = reply.minor;
1537 } else {
1538 conn->minor = be32toh(msg.minor);
1539 }
1540
1541 reply.major = htobe32(reply.major);
1542 reply.minor = htobe32(reply.minor);
1543 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1544 sizeof(struct lttcomm_relayd_version), 0);
1545 if (ret < 0) {
1546 ERR("Relay sending version");
1547 }
1548
1549 DBG("Version check done using protocol %u.%u", conn->major,
1550 conn->minor);
1551
1552 end:
1553 return ret;
1554 }
1555
1556 /*
1557 * Check for data pending for a given stream id from the session daemon.
1558 */
1559 static int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1560 struct relay_connection *conn)
1561 {
1562 struct relay_session *session = conn->session;
1563 struct lttcomm_relayd_data_pending msg;
1564 struct lttcomm_relayd_generic_reply reply;
1565 struct relay_stream *stream;
1566 int ret;
1567 uint64_t last_net_seq_num, stream_id;
1568
1569 DBG("Data pending command received");
1570
1571 if (!session || conn->version_check_done == 0) {
1572 ERR("Trying to check for data before version check");
1573 ret = -1;
1574 goto end_no_session;
1575 }
1576
1577 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1578 if (ret < sizeof(msg)) {
1579 if (ret == 0) {
1580 /* Orderly shutdown. Not necessary to print an error. */
1581 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1582 } else {
1583 ERR("Relay didn't receive valid data_pending struct size : %d",
1584 ret);
1585 }
1586 ret = -1;
1587 goto end_no_session;
1588 }
1589
1590 stream_id = be64toh(msg.stream_id);
1591 last_net_seq_num = be64toh(msg.last_net_seq_num);
1592
1593 stream = stream_get_by_id(stream_id);
1594 if (stream == NULL) {
1595 ret = -1;
1596 goto end;
1597 }
1598
1599 pthread_mutex_lock(&stream->lock);
1600
1601 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1602 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1603 last_net_seq_num);
1604
1605 /* Avoid wrapping issue */
1606 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1607 /* Data has in fact been written and is NOT pending */
1608 ret = 0;
1609 } else {
1610 /* Data still being streamed thus pending */
1611 ret = 1;
1612 }
1613
1614 stream->data_pending_check_done = true;
1615 pthread_mutex_unlock(&stream->lock);
1616
1617 stream_put(stream);
1618 end:
1619
1620 memset(&reply, 0, sizeof(reply));
1621 reply.ret_code = htobe32(ret);
1622 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1623 if (ret < 0) {
1624 ERR("Relay data pending ret code failed");
1625 }
1626
1627 end_no_session:
1628 return ret;
1629 }
1630
1631 /*
1632 * Wait for the control socket to reach a quiescent state.
1633 *
1634 * Note that for now, when receiving this command from the session
1635 * daemon, this means that every subsequent commands or data received on
1636 * the control socket has been handled. So, this is why we simply return
1637 * OK here.
1638 */
1639 static int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1640 struct relay_connection *conn)
1641 {
1642 int ret;
1643 uint64_t stream_id;
1644 struct relay_stream *stream;
1645 struct lttcomm_relayd_quiescent_control msg;
1646 struct lttcomm_relayd_generic_reply reply;
1647
1648 DBG("Checking quiescent state on control socket");
1649
1650 if (!conn->session || conn->version_check_done == 0) {
1651 ERR("Trying to check for data before version check");
1652 ret = -1;
1653 goto end_no_session;
1654 }
1655
1656 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1657 if (ret < sizeof(msg)) {
1658 if (ret == 0) {
1659 /* Orderly shutdown. Not necessary to print an error. */
1660 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1661 } else {
1662 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1663 ret);
1664 }
1665 ret = -1;
1666 goto end_no_session;
1667 }
1668
1669 stream_id = be64toh(msg.stream_id);
1670 stream = stream_get_by_id(stream_id);
1671 if (!stream) {
1672 goto reply;
1673 }
1674 pthread_mutex_lock(&stream->lock);
1675 stream->data_pending_check_done = true;
1676 pthread_mutex_unlock(&stream->lock);
1677 DBG("Relay quiescent control pending flag set to %" PRIu64, stream_id);
1678 stream_put(stream);
1679 reply:
1680 memset(&reply, 0, sizeof(reply));
1681 reply.ret_code = htobe32(LTTNG_OK);
1682 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1683 if (ret < 0) {
1684 ERR("Relay data quiescent control ret code failed");
1685 }
1686
1687 end_no_session:
1688 return ret;
1689 }
1690
1691 /*
1692 * Initialize a data pending command. This means that a consumer is about
1693 * to ask for data pending for each stream it holds. Simply iterate over
1694 * all streams of a session and set the data_pending_check_done flag.
1695 *
1696 * This command returns to the client a LTTNG_OK code.
1697 */
1698 static int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1699 struct relay_connection *conn)
1700 {
1701 int ret;
1702 struct lttng_ht_iter iter;
1703 struct lttcomm_relayd_begin_data_pending msg;
1704 struct lttcomm_relayd_generic_reply reply;
1705 struct relay_stream *stream;
1706 uint64_t session_id;
1707
1708 assert(recv_hdr);
1709 assert(conn);
1710
1711 DBG("Init streams for data pending");
1712
1713 if (!conn->session || conn->version_check_done == 0) {
1714 ERR("Trying to check for data before version check");
1715 ret = -1;
1716 goto end_no_session;
1717 }
1718
1719 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1720 if (ret < sizeof(msg)) {
1721 if (ret == 0) {
1722 /* Orderly shutdown. Not necessary to print an error. */
1723 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1724 } else {
1725 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1726 ret);
1727 }
1728 ret = -1;
1729 goto end_no_session;
1730 }
1731
1732 session_id = be64toh(msg.session_id);
1733
1734 /*
1735 * Iterate over all streams to set the begin data pending flag.
1736 * For now, the streams are indexed by stream handle so we have
1737 * to iterate over all streams to find the one associated with
1738 * the right session_id.
1739 */
1740 rcu_read_lock();
1741 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1742 node.node) {
1743 if (!stream_get(stream)) {
1744 continue;
1745 }
1746 if (stream->trace->session->id == session_id) {
1747 pthread_mutex_lock(&stream->lock);
1748 stream->data_pending_check_done = false;
1749 pthread_mutex_unlock(&stream->lock);
1750 DBG("Set begin data pending flag to stream %" PRIu64,
1751 stream->stream_handle);
1752 }
1753 stream_put(stream);
1754 }
1755 rcu_read_unlock();
1756
1757 memset(&reply, 0, sizeof(reply));
1758 /* All good, send back reply. */
1759 reply.ret_code = htobe32(LTTNG_OK);
1760
1761 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1762 if (ret < 0) {
1763 ERR("Relay begin data pending send reply failed");
1764 }
1765
1766 end_no_session:
1767 return ret;
1768 }
1769
1770 /*
1771 * End data pending command. This will check, for a given session id, if
1772 * each stream associated with it has its data_pending_check_done flag
1773 * set. If not, this means that the client lost track of the stream but
1774 * the data is still being streamed on our side. In this case, we inform
1775 * the client that data is in flight.
1776 *
1777 * Return to the client if there is data in flight or not with a ret_code.
1778 */
1779 static int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1780 struct relay_connection *conn)
1781 {
1782 int ret;
1783 struct lttng_ht_iter iter;
1784 struct lttcomm_relayd_end_data_pending msg;
1785 struct lttcomm_relayd_generic_reply reply;
1786 struct relay_stream *stream;
1787 uint64_t session_id;
1788 uint32_t is_data_inflight = 0;
1789
1790 DBG("End data pending command");
1791
1792 if (!conn->session || conn->version_check_done == 0) {
1793 ERR("Trying to check for data before version check");
1794 ret = -1;
1795 goto end_no_session;
1796 }
1797
1798 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
1799 if (ret < sizeof(msg)) {
1800 if (ret == 0) {
1801 /* Orderly shutdown. Not necessary to print an error. */
1802 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1803 } else {
1804 ERR("Relay didn't receive valid end data_pending struct size: %d",
1805 ret);
1806 }
1807 ret = -1;
1808 goto end_no_session;
1809 }
1810
1811 session_id = be64toh(msg.session_id);
1812
1813 /*
1814 * Iterate over all streams to see if the begin data pending
1815 * flag is set.
1816 */
1817 rcu_read_lock();
1818 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1819 node.node) {
1820 if (!stream_get(stream)) {
1821 continue;
1822 }
1823 if (stream->trace->session->id != session_id) {
1824 stream_put(stream);
1825 continue;
1826 }
1827 pthread_mutex_lock(&stream->lock);
1828 if (!stream->data_pending_check_done) {
1829 if (!stream->closed || !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
1830 is_data_inflight = 1;
1831 DBG("Data is still in flight for stream %" PRIu64,
1832 stream->stream_handle);
1833 pthread_mutex_unlock(&stream->lock);
1834 stream_put(stream);
1835 break;
1836 }
1837 }
1838 pthread_mutex_unlock(&stream->lock);
1839 stream_put(stream);
1840 }
1841 rcu_read_unlock();
1842
1843 memset(&reply, 0, sizeof(reply));
1844 /* All good, send back reply. */
1845 reply.ret_code = htobe32(is_data_inflight);
1846
1847 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1848 if (ret < 0) {
1849 ERR("Relay end data pending send reply failed");
1850 }
1851
1852 end_no_session:
1853 return ret;
1854 }
1855
1856 /*
1857 * Receive an index for a specific stream.
1858 *
1859 * Return 0 on success else a negative value.
1860 */
1861 static int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1862 struct relay_connection *conn)
1863 {
1864 int ret, send_ret;
1865 struct relay_session *session = conn->session;
1866 struct lttcomm_relayd_index index_info;
1867 struct relay_index *index;
1868 struct lttcomm_relayd_generic_reply reply;
1869 struct relay_stream *stream;
1870 uint64_t net_seq_num;
1871
1872 assert(conn);
1873
1874 DBG("Relay receiving index");
1875
1876 if (!session || conn->version_check_done == 0) {
1877 ERR("Trying to close a stream before version check");
1878 ret = -1;
1879 goto end_no_session;
1880 }
1881
1882 ret = conn->sock->ops->recvmsg(conn->sock, &index_info,
1883 sizeof(index_info), 0);
1884 if (ret < sizeof(index_info)) {
1885 if (ret == 0) {
1886 /* Orderly shutdown. Not necessary to print an error. */
1887 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1888 } else {
1889 ERR("Relay didn't receive valid index struct size : %d", ret);
1890 }
1891 ret = -1;
1892 goto end_no_session;
1893 }
1894
1895 net_seq_num = be64toh(index_info.net_seq_num);
1896
1897 stream = stream_get_by_id(be64toh(index_info.relay_stream_id));
1898 if (!stream) {
1899 ERR("stream_get_by_id not found");
1900 ret = -1;
1901 goto end;
1902 }
1903 pthread_mutex_lock(&stream->lock);
1904
1905 /* Live beacon handling */
1906 if (index_info.packet_size == 0) {
1907 DBG("Received live beacon for stream %" PRIu64,
1908 stream->stream_handle);
1909
1910 /*
1911 * Only flag a stream inactive when it has already
1912 * received data and no indexes are in flight.
1913 */
1914 if (stream->index_received_seqcount > 0
1915 && stream->indexes_in_flight == 0) {
1916 stream->beacon_ts_end =
1917 be64toh(index_info.timestamp_end);
1918 }
1919 ret = 0;
1920 goto end_stream_put;
1921 } else {
1922 stream->beacon_ts_end = -1ULL;
1923 }
1924
1925 if (stream->ctf_stream_id == -1ULL) {
1926 stream->ctf_stream_id = be64toh(index_info.stream_id);
1927 }
1928 index = relay_index_get_by_id_or_create(stream, net_seq_num);
1929 if (!index) {
1930 ret = -1;
1931 ERR("relay_index_get_by_id_or_create index NULL");
1932 goto end_stream_put;
1933 }
1934 if (set_index_control_data(index, &index_info)) {
1935 ERR("set_index_control_data error");
1936 relay_index_put(index);
1937 ret = -1;
1938 goto end_stream_put;
1939 }
1940 ret = relay_index_try_flush(index);
1941 if (ret == 0) {
1942 tracefile_array_commit_seq(stream->tfa);
1943 stream->index_received_seqcount++;
1944 } else if (ret > 0) {
1945 /* no flush. */
1946 ret = 0;
1947 } else {
1948 ERR("relay_index_try_flush error %d", ret);
1949 relay_index_put(index);
1950 ret = -1;
1951 }
1952
1953 end_stream_put:
1954 pthread_mutex_unlock(&stream->lock);
1955 stream_put(stream);
1956
1957 end:
1958
1959 memset(&reply, 0, sizeof(reply));
1960 if (ret < 0) {
1961 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1962 } else {
1963 reply.ret_code = htobe32(LTTNG_OK);
1964 }
1965 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1966 if (send_ret < 0) {
1967 ERR("Relay sending close index id reply");
1968 ret = send_ret;
1969 }
1970
1971 end_no_session:
1972 return ret;
1973 }
1974
1975 /*
1976 * Receive the streams_sent message.
1977 *
1978 * Return 0 on success else a negative value.
1979 */
1980 static int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
1981 struct relay_connection *conn)
1982 {
1983 int ret, send_ret;
1984 struct lttcomm_relayd_generic_reply reply;
1985
1986 assert(conn);
1987
1988 DBG("Relay receiving streams_sent");
1989
1990 if (!conn->session || conn->version_check_done == 0) {
1991 ERR("Trying to close a stream before version check");
1992 ret = -1;
1993 goto end_no_session;
1994 }
1995
1996 /*
1997 * Publish every pending stream in the connection recv list which are
1998 * now ready to be used by the viewer.
1999 */
2000 publish_connection_local_streams(conn);
2001
2002 memset(&reply, 0, sizeof(reply));
2003 reply.ret_code = htobe32(LTTNG_OK);
2004 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2005 if (send_ret < 0) {
2006 ERR("Relay sending sent_stream reply");
2007 ret = send_ret;
2008 } else {
2009 /* Success. */
2010 ret = 0;
2011 }
2012
2013 end_no_session:
2014 return ret;
2015 }
2016
2017 /*
2018 * Process the commands received on the control socket
2019 */
2020 static int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
2021 struct relay_connection *conn)
2022 {
2023 int ret = 0;
2024
2025 switch (be32toh(recv_hdr->cmd)) {
2026 case RELAYD_CREATE_SESSION:
2027 ret = relay_create_session(recv_hdr, conn);
2028 break;
2029 case RELAYD_ADD_STREAM:
2030 ret = relay_add_stream(recv_hdr, conn);
2031 break;
2032 case RELAYD_START_DATA:
2033 ret = relay_start(recv_hdr, conn);
2034 break;
2035 case RELAYD_SEND_METADATA:
2036 ret = relay_recv_metadata(recv_hdr, conn);
2037 break;
2038 case RELAYD_VERSION:
2039 ret = relay_send_version(recv_hdr, conn);
2040 break;
2041 case RELAYD_CLOSE_STREAM:
2042 ret = relay_close_stream(recv_hdr, conn);
2043 break;
2044 case RELAYD_DATA_PENDING:
2045 ret = relay_data_pending(recv_hdr, conn);
2046 break;
2047 case RELAYD_QUIESCENT_CONTROL:
2048 ret = relay_quiescent_control(recv_hdr, conn);
2049 break;
2050 case RELAYD_BEGIN_DATA_PENDING:
2051 ret = relay_begin_data_pending(recv_hdr, conn);
2052 break;
2053 case RELAYD_END_DATA_PENDING:
2054 ret = relay_end_data_pending(recv_hdr, conn);
2055 break;
2056 case RELAYD_SEND_INDEX:
2057 ret = relay_recv_index(recv_hdr, conn);
2058 break;
2059 case RELAYD_STREAMS_SENT:
2060 ret = relay_streams_sent(recv_hdr, conn);
2061 break;
2062 case RELAYD_UPDATE_SYNC_INFO:
2063 default:
2064 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2065 relay_unknown_command(conn);
2066 ret = -1;
2067 goto end;
2068 }
2069
2070 end:
2071 return ret;
2072 }
2073
2074 /*
2075 * Handle index for a data stream.
2076 *
2077 * Called with the stream lock held.
2078 *
2079 * Return 0 on success else a negative value.
2080 */
2081 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2082 int rotate_index)
2083 {
2084 int ret = 0;
2085 uint64_t data_offset;
2086 struct relay_index *index;
2087
2088 /* Get data offset because we are about to update the index. */
2089 data_offset = htobe64(stream->tracefile_size_current);
2090
2091 DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64,
2092 stream->stream_handle, net_seq_num, stream->tracefile_size_current);
2093
2094 /*
2095 * Lookup for an existing index for that stream id/sequence
2096 * number. If it exists, the control thread has already received the
2097 * data for it, thus we need to write it to disk.
2098 */
2099 index = relay_index_get_by_id_or_create(stream, net_seq_num);
2100 if (!index) {
2101 ret = -1;
2102 goto end;
2103 }
2104
2105 if (rotate_index || !stream->index_fd) {
2106 int fd;
2107
2108 /* Put ref on previous index_fd. */
2109 if (stream->index_fd) {
2110 stream_fd_put(stream->index_fd);
2111 stream->index_fd = NULL;
2112 }
2113
2114 fd = index_create_file(stream->path_name, stream->channel_name,
2115 -1, -1, stream->tracefile_size,
2116 tracefile_array_get_file_index_head(stream->tfa));
2117 if (fd < 0) {
2118 ret = -1;
2119 /* Put self-ref for this index due to error. */
2120 relay_index_put(index);
2121 goto end;
2122 }
2123 stream->index_fd = stream_fd_create(fd);
2124 if (!stream->index_fd) {
2125 ret = -1;
2126 if (close(fd)) {
2127 PERROR("Error closing FD %d", fd);
2128 }
2129 /* Put self-ref for this index due to error. */
2130 relay_index_put(index);
2131 /* Will put the local ref. */
2132 goto end;
2133 }
2134 }
2135
2136 if (relay_index_set_fd(index, stream->index_fd, data_offset)) {
2137 ret = -1;
2138 /* Put self-ref for this index due to error. */
2139 relay_index_put(index);
2140 goto end;
2141 }
2142
2143 ret = relay_index_try_flush(index);
2144 if (ret == 0) {
2145 tracefile_array_commit_seq(stream->tfa);
2146 stream->index_received_seqcount++;
2147 } else if (ret > 0) {
2148 /* No flush. */
2149 ret = 0;
2150 } else {
2151 /* Put self-ref for this index due to error. */
2152 relay_index_put(index);
2153 ret = -1;
2154 }
2155 end:
2156 return ret;
2157 }
2158
2159 /*
2160 * relay_process_data: Process the data received on the data socket
2161 */
2162 static int relay_process_data(struct relay_connection *conn)
2163 {
2164 int ret = 0, rotate_index = 0;
2165 ssize_t size_ret;
2166 struct relay_stream *stream;
2167 struct lttcomm_relayd_data_hdr data_hdr;
2168 uint64_t stream_id;
2169 uint64_t net_seq_num;
2170 uint32_t data_size;
2171 struct relay_session *session;
2172 bool new_stream = false, close_requested = false;
2173 size_t chunk_size = RECV_DATA_BUFFER_SIZE;
2174 size_t recv_off = 0;
2175 char data_buffer[chunk_size];
2176
2177 ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr,
2178 sizeof(struct lttcomm_relayd_data_hdr), 0);
2179 if (ret <= 0) {
2180 if (ret == 0) {
2181 /* Orderly shutdown. Not necessary to print an error. */
2182 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2183 } else {
2184 ERR("Unable to receive data header on sock %d", conn->sock->fd);
2185 }
2186 ret = -1;
2187 goto end;
2188 }
2189
2190 stream_id = be64toh(data_hdr.stream_id);
2191 stream = stream_get_by_id(stream_id);
2192 if (!stream) {
2193 ERR("relay_process_data: Cannot find stream %" PRIu64, stream_id);
2194 ret = -1;
2195 goto end;
2196 }
2197 session = stream->trace->session;
2198 data_size = be32toh(data_hdr.data_size);
2199
2200 net_seq_num = be64toh(data_hdr.net_seq_num);
2201
2202 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2203 data_size, stream_id, net_seq_num);
2204
2205 pthread_mutex_lock(&stream->lock);
2206
2207 /* Check if a rotation is needed. */
2208 if (stream->tracefile_size > 0 &&
2209 (stream->tracefile_size_current + data_size) >
2210 stream->tracefile_size) {
2211 uint64_t old_id, new_id;
2212
2213 old_id = tracefile_array_get_file_index_head(stream->tfa);
2214 tracefile_array_file_rotate(stream->tfa);
2215
2216 /* new_id is updated by utils_rotate_stream_file. */
2217 new_id = old_id;
2218
2219 ret = utils_rotate_stream_file(stream->path_name,
2220 stream->channel_name, stream->tracefile_size,
2221 stream->tracefile_count, -1,
2222 -1, stream->stream_fd->fd,
2223 &new_id, &stream->stream_fd->fd);
2224 if (ret < 0) {
2225 ERR("Rotating stream output file");
2226 goto end_stream_unlock;
2227 }
2228 /*
2229 * Reset current size because we just performed a stream
2230 * rotation.
2231 */
2232 stream->tracefile_size_current = 0;
2233 rotate_index = 1;
2234 }
2235
2236 /*
2237 * Index are handled in protocol version 2.4 and above. Also,
2238 * snapshot and index are NOT supported.
2239 */
2240 if (session->minor >= 4 && !session->snapshot) {
2241 ret = handle_index_data(stream, net_seq_num, rotate_index);
2242 if (ret < 0) {
2243 ERR("handle_index_data: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
2244 stream->stream_handle, net_seq_num, ret);
2245 goto end_stream_unlock;
2246 }
2247 }
2248
2249 for (recv_off = 0; recv_off < data_size; recv_off += chunk_size) {
2250 size_t recv_size = min(data_size - recv_off, chunk_size);
2251
2252 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_size, 0);
2253 if (ret <= 0) {
2254 if (ret == 0) {
2255 /* Orderly shutdown. Not necessary to print an error. */
2256 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
2257 } else {
2258 ERR("Socket %d error %d", conn->sock->fd, ret);
2259 }
2260 ret = -1;
2261 goto end_stream_unlock;
2262 }
2263
2264 /* Write data to stream output fd. */
2265 size_ret = lttng_write(stream->stream_fd->fd, data_buffer,
2266 recv_size);
2267 if (size_ret < recv_size) {
2268 ERR("Relay error writing data to file");
2269 ret = -1;
2270 goto end_stream_unlock;
2271 }
2272
2273 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64,
2274 size_ret, stream->stream_handle);
2275 }
2276
2277 ret = write_padding_to_file(stream->stream_fd->fd,
2278 be32toh(data_hdr.padding_size));
2279 if (ret < 0) {
2280 ERR("write_padding_to_file: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
2281 stream->stream_handle, net_seq_num, ret);
2282 goto end_stream_unlock;
2283 }
2284 stream->tracefile_size_current +=
2285 data_size + be32toh(data_hdr.padding_size);
2286 if (stream->prev_seq == -1ULL) {
2287 new_stream = true;
2288 }
2289
2290 stream->prev_seq = net_seq_num;
2291
2292 end_stream_unlock:
2293 close_requested = stream->close_requested;
2294 pthread_mutex_unlock(&stream->lock);
2295 if (close_requested) {
2296 try_stream_close(stream);
2297 }
2298
2299 if (new_stream) {
2300 pthread_mutex_lock(&session->lock);
2301 uatomic_set(&session->new_streams, 1);
2302 pthread_mutex_unlock(&session->lock);
2303 }
2304 stream_put(stream);
2305 end:
2306 return ret;
2307 }
2308
2309 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2310 {
2311 int ret;
2312
2313 (void) lttng_poll_del(events, pollfd);
2314
2315 ret = close(pollfd);
2316 if (ret < 0) {
2317 ERR("Closing pollfd %d", pollfd);
2318 }
2319 }
2320
2321 static void relay_thread_close_connection(struct lttng_poll_event *events,
2322 int pollfd, struct relay_connection *conn)
2323 {
2324 const char *type_str;
2325
2326 switch (conn->type) {
2327 case RELAY_DATA:
2328 type_str = "Data";
2329 break;
2330 case RELAY_CONTROL:
2331 type_str = "Control";
2332 break;
2333 case RELAY_VIEWER_COMMAND:
2334 type_str = "Viewer Command";
2335 break;
2336 case RELAY_VIEWER_NOTIFICATION:
2337 type_str = "Viewer Notification";
2338 break;
2339 default:
2340 type_str = "Unknown";
2341 }
2342 cleanup_connection_pollfd(events, pollfd);
2343 connection_put(conn);
2344 DBG("%s connection closed with %d", type_str, pollfd);
2345 }
2346
2347 /*
2348 * This thread does the actual work
2349 */
2350 static void *relay_thread_worker(void *data)
2351 {
2352 int ret, err = -1, last_seen_data_fd = -1;
2353 uint32_t nb_fd;
2354 struct lttng_poll_event events;
2355 struct lttng_ht *relay_connections_ht;
2356 struct lttng_ht_iter iter;
2357 struct lttcomm_relayd_hdr recv_hdr;
2358 struct relay_connection *destroy_conn = NULL;
2359
2360 DBG("[thread] Relay worker started");
2361
2362 rcu_register_thread();
2363
2364 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2365
2366 if (testpoint(relayd_thread_worker)) {
2367 goto error_testpoint;
2368 }
2369
2370 health_code_update();
2371
2372 /* table of connections indexed on socket */
2373 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2374 if (!relay_connections_ht) {
2375 goto relay_connections_ht_error;
2376 }
2377
2378 ret = create_thread_poll_set(&events, 2);
2379 if (ret < 0) {
2380 goto error_poll_create;
2381 }
2382
2383 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2384 if (ret < 0) {
2385 goto error;
2386 }
2387
2388 restart:
2389 while (1) {
2390 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2391
2392 health_code_update();
2393
2394 /* Infinite blocking call, waiting for transmission */
2395 DBG3("Relayd worker thread polling...");
2396 health_poll_entry();
2397 ret = lttng_poll_wait(&events, -1);
2398 health_poll_exit();
2399 if (ret < 0) {
2400 /*
2401 * Restart interrupted system call.
2402 */
2403 if (errno == EINTR) {
2404 goto restart;
2405 }
2406 goto error;
2407 }
2408
2409 nb_fd = ret;
2410
2411 /*
2412 * Process control. The control connection is
2413 * prioritized so we don't starve it with high
2414 * throughput tracing data on the data connection.
2415 */
2416 for (i = 0; i < nb_fd; i++) {
2417 /* Fetch once the poll data */
2418 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2419 int pollfd = LTTNG_POLL_GETFD(&events, i);
2420
2421 health_code_update();
2422
2423 if (!revents) {
2424 /*
2425 * No activity for this FD (poll
2426 * implementation).
2427 */
2428 continue;
2429 }
2430
2431 /* Thread quit pipe has been closed. Killing thread. */
2432 ret = check_thread_quit_pipe(pollfd, revents);
2433 if (ret) {
2434 err = 0;
2435 goto exit;
2436 }
2437
2438 /* Inspect the relay conn pipe for new connection */
2439 if (pollfd == relay_conn_pipe[0]) {
2440 if (revents & LPOLLIN) {
2441 struct relay_connection *conn;
2442
2443 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
2444 if (ret < 0) {
2445 goto error;
2446 }
2447 lttng_poll_add(&events, conn->sock->fd,
2448 LPOLLIN | LPOLLRDHUP);
2449 connection_ht_add(relay_connections_ht, conn);
2450 DBG("Connection socket %d added", conn->sock->fd);
2451 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2452 ERR("Relay connection pipe error");
2453 goto error;
2454 } else {
2455 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2456 goto error;
2457 }
2458 } else {
2459 struct relay_connection *ctrl_conn;
2460
2461 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
2462 /* If not found, there is a synchronization issue. */
2463 assert(ctrl_conn);
2464
2465 if (ctrl_conn->type == RELAY_DATA) {
2466 if (revents & LPOLLIN) {
2467 /*
2468 * Flag the last seen data fd not deleted. It will be
2469 * used as the last seen fd if any fd gets deleted in
2470 * this first loop.
2471 */
2472 last_notdel_data_fd = pollfd;
2473 }
2474 goto put_ctrl_connection;
2475 }
2476 assert(ctrl_conn->type == RELAY_CONTROL);
2477
2478 if (revents & LPOLLIN) {
2479 ret = ctrl_conn->sock->ops->recvmsg(ctrl_conn->sock,
2480 &recv_hdr, sizeof(recv_hdr), 0);
2481 if (ret <= 0) {
2482 /* Connection closed */
2483 relay_thread_close_connection(&events, pollfd,
2484 ctrl_conn);
2485 } else {
2486 ret = relay_process_control(&recv_hdr, ctrl_conn);
2487 if (ret < 0) {
2488 /* Clear the session on error. */
2489 relay_thread_close_connection(&events,
2490 pollfd, ctrl_conn);
2491 }
2492 seen_control = 1;
2493 }
2494 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2495 relay_thread_close_connection(&events,
2496 pollfd, ctrl_conn);
2497 if (last_seen_data_fd == pollfd) {
2498 last_seen_data_fd = last_notdel_data_fd;
2499 }
2500 } else {
2501 ERR("Unexpected poll events %u for control sock %d",
2502 revents, pollfd);
2503 connection_put(ctrl_conn);
2504 goto error;
2505 }
2506 put_ctrl_connection:
2507 connection_put(ctrl_conn);
2508 }
2509 }
2510
2511 /*
2512 * The last loop handled a control request, go back to poll to make
2513 * sure we prioritise the control socket.
2514 */
2515 if (seen_control) {
2516 continue;
2517 }
2518
2519 if (last_seen_data_fd >= 0) {
2520 for (i = 0; i < nb_fd; i++) {
2521 int pollfd = LTTNG_POLL_GETFD(&events, i);
2522
2523 health_code_update();
2524
2525 if (last_seen_data_fd == pollfd) {
2526 idx = i;
2527 break;
2528 }
2529 }
2530 }
2531
2532 /* Process data connection. */
2533 for (i = idx + 1; i < nb_fd; i++) {
2534 /* Fetch the poll data. */
2535 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2536 int pollfd = LTTNG_POLL_GETFD(&events, i);
2537 struct relay_connection *data_conn;
2538
2539 health_code_update();
2540
2541 if (!revents) {
2542 /* No activity for this FD (poll implementation). */
2543 continue;
2544 }
2545
2546 /* Skip the command pipe. It's handled in the first loop. */
2547 if (pollfd == relay_conn_pipe[0]) {
2548 continue;
2549 }
2550
2551 data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
2552 if (!data_conn) {
2553 /* Skip it. Might be removed before. */
2554 continue;
2555 }
2556 if (data_conn->type == RELAY_CONTROL) {
2557 goto put_data_connection;
2558 }
2559 assert(data_conn->type == RELAY_DATA);
2560
2561 if (revents & LPOLLIN) {
2562 ret = relay_process_data(data_conn);
2563 /* Connection closed */
2564 if (ret < 0) {
2565 relay_thread_close_connection(&events, pollfd,
2566 data_conn);
2567 /*
2568 * Every goto restart call sets the last seen fd where
2569 * here we don't really care since we gracefully
2570 * continue the loop after the connection is deleted.
2571 */
2572 } else {
2573 /* Keep last seen port. */
2574 last_seen_data_fd = pollfd;
2575 connection_put(data_conn);
2576 goto restart;
2577 }
2578 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2579 relay_thread_close_connection(&events, pollfd,
2580 data_conn);
2581 } else {
2582 ERR("Unknown poll events %u for data sock %d",
2583 revents, pollfd);
2584 }
2585 put_data_connection:
2586 connection_put(data_conn);
2587 }
2588 last_seen_data_fd = -1;
2589 }
2590
2591 /* Normal exit, no error */
2592 ret = 0;
2593
2594 exit:
2595 error:
2596 /* Cleanup reamaining connection object. */
2597 rcu_read_lock();
2598 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
2599 destroy_conn,
2600 sock_n.node) {
2601 health_code_update();
2602 /*
2603 * No need to grab another ref, because we own
2604 * destroy_conn.
2605 */
2606 relay_thread_close_connection(&events, destroy_conn->sock->fd,
2607 destroy_conn);
2608 }
2609 rcu_read_unlock();
2610
2611 lttng_poll_clean(&events);
2612 error_poll_create:
2613 lttng_ht_destroy(relay_connections_ht);
2614 relay_connections_ht_error:
2615 /* Close relay conn pipes */
2616 utils_close_pipe(relay_conn_pipe);
2617 if (err) {
2618 DBG("Thread exited with error");
2619 }
2620 DBG("Worker thread cleanup complete");
2621 error_testpoint:
2622 if (err) {
2623 health_error();
2624 ERR("Health error occurred in %s", __func__);
2625 }
2626 health_unregister(health_relayd);
2627 rcu_unregister_thread();
2628 lttng_relay_stop_threads();
2629 return NULL;
2630 }
2631
2632 /*
2633 * Create the relay command pipe to wake thread_manage_apps.
2634 * Closed in cleanup().
2635 */
2636 static int create_relay_conn_pipe(void)
2637 {
2638 int ret;
2639
2640 ret = utils_create_pipe_cloexec(relay_conn_pipe);
2641
2642 return ret;
2643 }
2644
2645 /*
2646 * main
2647 */
2648 int main(int argc, char **argv)
2649 {
2650 int ret = 0, retval = 0;
2651 void *status;
2652
2653 /* Parse arguments */
2654 progname = argv[0];
2655 if (set_options(argc, argv)) {
2656 retval = -1;
2657 goto exit_options;
2658 }
2659
2660 if (set_signal_handler()) {
2661 retval = -1;
2662 goto exit_options;
2663 }
2664
2665 /* Try to create directory if -o, --output is specified. */
2666 if (opt_output_path) {
2667 if (*opt_output_path != '/') {
2668 ERR("Please specify an absolute path for -o, --output PATH");
2669 retval = -1;
2670 goto exit_options;
2671 }
2672
2673 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG,
2674 -1, -1);
2675 if (ret < 0) {
2676 ERR("Unable to create %s", opt_output_path);
2677 retval = -1;
2678 goto exit_options;
2679 }
2680 }
2681
2682 /* Daemonize */
2683 if (opt_daemon || opt_background) {
2684 int i;
2685
2686 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2687 !opt_background);
2688 if (ret < 0) {
2689 retval = -1;
2690 goto exit_options;
2691 }
2692
2693 /*
2694 * We are in the child. Make sure all other file
2695 * descriptors are closed, in case we are called with
2696 * more opened file descriptors than the standard ones.
2697 */
2698 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2699 (void) close(i);
2700 }
2701 }
2702
2703
2704 /* Initialize thread health monitoring */
2705 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2706 if (!health_relayd) {
2707 PERROR("health_app_create error");
2708 retval = -1;
2709 goto exit_health_app_create;
2710 }
2711
2712 /* Create thread quit pipe */
2713 if (init_thread_quit_pipe()) {
2714 retval = -1;
2715 goto exit_init_data;
2716 }
2717
2718 /* Setup the thread apps communication pipe. */
2719 if (create_relay_conn_pipe()) {
2720 retval = -1;
2721 goto exit_init_data;
2722 }
2723
2724 /* Init relay command queue. */
2725 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
2726
2727 /* Set up max poll set size */
2728 if (lttng_poll_set_max_size()) {
2729 retval = -1;
2730 goto exit_init_data;
2731 }
2732
2733 /* Initialize communication library */
2734 lttcomm_init();
2735 lttcomm_inet_init();
2736
2737 /* tables of sessions indexed by session ID */
2738 sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2739 if (!sessions_ht) {
2740 retval = -1;
2741 goto exit_init_data;
2742 }
2743
2744 /* tables of streams indexed by stream ID */
2745 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2746 if (!relay_streams_ht) {
2747 retval = -1;
2748 goto exit_init_data;
2749 }
2750
2751 /* tables of streams indexed by stream ID */
2752 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2753 if (!viewer_streams_ht) {
2754 retval = -1;
2755 goto exit_init_data;
2756 }
2757
2758 ret = utils_create_pipe(health_quit_pipe);
2759 if (ret) {
2760 retval = -1;
2761 goto exit_health_quit_pipe;
2762 }
2763
2764 /* Create thread to manage the client socket */
2765 ret = pthread_create(&health_thread, NULL,
2766 thread_manage_health, (void *) NULL);
2767 if (ret) {
2768 errno = ret;
2769 PERROR("pthread_create health");
2770 retval = -1;
2771 goto exit_health_thread;
2772 }
2773
2774 /* Setup the dispatcher thread */
2775 ret = pthread_create(&dispatcher_thread, NULL,
2776 relay_thread_dispatcher, (void *) NULL);
2777 if (ret) {
2778 errno = ret;
2779 PERROR("pthread_create dispatcher");
2780 retval = -1;
2781 goto exit_dispatcher_thread;
2782 }
2783
2784 /* Setup the worker thread */
2785 ret = pthread_create(&worker_thread, NULL,
2786 relay_thread_worker, NULL);
2787 if (ret) {
2788 errno = ret;
2789 PERROR("pthread_create worker");
2790 retval = -1;
2791 goto exit_worker_thread;
2792 }
2793
2794 /* Setup the listener thread */
2795 ret = pthread_create(&listener_thread, NULL,
2796 relay_thread_listener, (void *) NULL);
2797 if (ret) {
2798 errno = ret;
2799 PERROR("pthread_create listener");
2800 retval = -1;
2801 goto exit_listener_thread;
2802 }
2803
2804 ret = relayd_live_create(live_uri);
2805 if (ret) {
2806 ERR("Starting live viewer threads");
2807 retval = -1;
2808 goto exit_live;
2809 }
2810
2811 /*
2812 * This is where we start awaiting program completion (e.g. through
2813 * signal that asks threads to teardown).
2814 */
2815
2816 ret = relayd_live_join();
2817 if (ret) {
2818 retval = -1;
2819 }
2820 exit_live:
2821
2822 ret = pthread_join(listener_thread, &status);
2823 if (ret) {
2824 errno = ret;
2825 PERROR("pthread_join listener_thread");
2826 retval = -1;
2827 }
2828
2829 exit_listener_thread:
2830 ret = pthread_join(worker_thread, &status);
2831 if (ret) {
2832 errno = ret;
2833 PERROR("pthread_join worker_thread");
2834 retval = -1;
2835 }
2836
2837 exit_worker_thread:
2838 ret = pthread_join(dispatcher_thread, &status);
2839 if (ret) {
2840 errno = ret;
2841 PERROR("pthread_join dispatcher_thread");
2842 retval = -1;
2843 }
2844 exit_dispatcher_thread:
2845
2846 ret = pthread_join(health_thread, &status);
2847 if (ret) {
2848 errno = ret;
2849 PERROR("pthread_join health_thread");
2850 retval = -1;
2851 }
2852 exit_health_thread:
2853
2854 utils_close_pipe(health_quit_pipe);
2855 exit_health_quit_pipe:
2856
2857 exit_init_data:
2858 health_app_destroy(health_relayd);
2859 exit_health_app_create:
2860 exit_options:
2861 relayd_cleanup();
2862
2863 /* Ensure all prior call_rcu are done. */
2864 rcu_barrier();
2865
2866 if (!retval) {
2867 exit(EXIT_SUCCESS);
2868 } else {
2869 exit(EXIT_FAILURE);
2870 }
2871 }
This page took 0.125652 seconds and 4 git commands to generate.