338341ddaceeae4d6203f43fdf63a80d685ba5a8
[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 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <sys/mman.h>
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
32 #include <sys/stat.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <urcu/futex.h>
36 #include <urcu/uatomic.h>
37 #include <unistd.h>
38 #include <fcntl.h>
39 #include <config.h>
40
41 #include <lttng/lttng.h>
42 #include <common/common.h>
43 #include <common/compat/poll.h>
44 #include <common/compat/socket.h>
45 #include <common/defaults.h>
46 #include <common/futex.h>
47 #include <common/sessiond-comm/sessiond-comm.h>
48 #include <common/sessiond-comm/inet.h>
49 #include <common/hashtable/hashtable.h>
50 #include <common/sessiond-comm/relayd.h>
51 #include <common/uri.h>
52
53 #include "lttng-relayd.h"
54
55 /* command line options */
56 static int opt_daemon;
57 static char *opt_output_path;
58 static struct lttng_uri *control_uri = NULL;
59 static struct lttng_uri *data_uri = NULL;
60
61 const char *progname;
62 static int is_root; /* Set to 1 if the daemon is running as root */
63
64 /*
65 * Quit pipe for all threads. This permits a single cancellation point
66 * for all threads when receiving an event on the pipe.
67 */
68 static int thread_quit_pipe[2] = { -1, -1 };
69
70 /*
71 * This pipe is used to inform the worker thread that a command is queued and
72 * ready to be processed.
73 */
74 static int relay_cmd_pipe[2] = { -1, -1 };
75
76 static int dispatch_thread_exit;
77
78 static pthread_t listener_thread;
79 static pthread_t dispatcher_thread;
80 static pthread_t worker_thread;
81
82 static uint64_t last_relay_stream_id = 0;
83 static uint64_t last_relay_session_id = 0;
84
85 /*
86 * Relay command queue.
87 *
88 * The relay_thread_listener and relay_thread_dispatcher communicate with this
89 * queue.
90 */
91 static struct relay_cmd_queue relay_cmd_queue;
92
93 /* buffer allocated at startup, used to store the trace data */
94 static char *data_buffer = NULL;
95 static unsigned int data_buffer_size = 0;
96
97 /*
98 * usage function on stderr
99 */
100 static
101 void usage(void)
102 {
103 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
104 fprintf(stderr, " -h, --help Display this usage.\n");
105 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
106 fprintf(stderr, " -C, --control-port Control port listening (URI)\n");
107 fprintf(stderr, " -D, --data-port Data port listening (URI)\n");
108 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
109 }
110
111 static
112 int parse_args(int argc, char **argv)
113 {
114 int c;
115 int ret = 0;
116 char *default_address;
117
118 static struct option long_options[] = {
119 { "control-port", 1, 0, 'C' },
120 { "data-port", 1, 0, 'D' },
121 { "daemonize", 0, 0, 'd' },
122 { "output", 1, 0, 'o' },
123 { "verbose", 0, 0, 'v' },
124 { NULL, 0, 0, 0 }
125 };
126
127 while (1) {
128 int option_index = 0;
129 c = getopt_long(argc, argv, "dhv" "C:D:o:",
130 long_options, &option_index);
131 if (c == -1) {
132 break;
133 }
134
135 switch (c) {
136 case 0:
137 fprintf(stderr, "option %s", long_options[option_index].name);
138 if (optarg) {
139 fprintf(stderr, " with arg %s\n", optarg);
140 }
141 break;
142 case 'C':
143 ret = uri_parse(optarg, &control_uri);
144 if (ret < 0) {
145 ERR("Invalid control URI specified");
146 goto exit;
147 }
148 if (control_uri->port == 0) {
149 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
150 }
151 break;
152 case 'D':
153 ret = uri_parse(optarg, &data_uri);
154 if (ret < 0) {
155 ERR("Invalid data URI specified");
156 goto exit;
157 }
158 if (data_uri->port == 0) {
159 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
160 }
161 break;
162 case 'd':
163 opt_daemon = 1;
164 break;
165 case 'h':
166 usage();
167 exit(EXIT_FAILURE);
168 case 'o':
169 ret = asprintf(&opt_output_path, "%s", optarg);
170 if (ret < 0) {
171 PERROR("asprintf opt_output_path");
172 goto exit;
173 }
174 break;
175 case 'v':
176 /* Verbose level can increase using multiple -v */
177 lttng_opt_verbose += 1;
178 break;
179 default:
180 /* Unknown option or other error.
181 * Error is printed by getopt, just return */
182 ret = -1;
183 goto exit;
184 }
185 }
186
187 /* assign default values */
188 if (control_uri == NULL) {
189 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
190 DEFAULT_NETWORK_CONTROL_PORT);
191 if (ret < 0) {
192 PERROR("asprintf default data address");
193 goto exit;
194 }
195
196 ret = uri_parse(default_address, &control_uri);
197 free(default_address);
198 if (ret < 0) {
199 ERR("Invalid control URI specified");
200 goto exit;
201 }
202 }
203 if (data_uri == NULL) {
204 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
205 DEFAULT_NETWORK_DATA_PORT);
206 if (ret < 0) {
207 PERROR("asprintf default data address");
208 goto exit;
209 }
210
211 ret = uri_parse(default_address, &data_uri);
212 free(default_address);
213 if (ret < 0) {
214 ERR("Invalid data URI specified");
215 goto exit;
216 }
217 }
218
219 exit:
220 return ret;
221 }
222
223 /*
224 * Cleanup the daemon
225 */
226 static
227 void cleanup(void)
228 {
229 int i, ret;
230
231 DBG("Cleaning up");
232
233 for (i = 0; i < 2; i++) {
234 if (thread_quit_pipe[i] >= 0) {
235 ret = close(thread_quit_pipe[i]);
236 if (ret) {
237 PERROR("close");
238 }
239 }
240 }
241 }
242
243 /*
244 * Write to writable pipe used to notify a thread.
245 */
246 static
247 int notify_thread_pipe(int wpipe)
248 {
249 int ret;
250
251 ret = write(wpipe, "!", 1);
252 if (ret < 0) {
253 PERROR("write poll pipe");
254 }
255
256 return ret;
257 }
258
259 /*
260 * Stop all threads by closing the thread quit pipe.
261 */
262 static
263 void stop_threads(void)
264 {
265 int ret;
266
267 /* Stopping all threads */
268 DBG("Terminating all threads");
269 ret = notify_thread_pipe(thread_quit_pipe[1]);
270 if (ret < 0) {
271 ERR("write error on thread quit pipe");
272 }
273
274 /* Dispatch thread */
275 dispatch_thread_exit = 1;
276 futex_nto1_wake(&relay_cmd_queue.futex);
277 }
278
279 /*
280 * Signal handler for the daemon
281 *
282 * Simply stop all worker threads, leaving main() return gracefully after
283 * joining all threads and calling cleanup().
284 */
285 static
286 void sighandler(int sig)
287 {
288 switch (sig) {
289 case SIGPIPE:
290 DBG("SIGPIPE caught");
291 return;
292 case SIGINT:
293 DBG("SIGINT caught");
294 stop_threads();
295 break;
296 case SIGTERM:
297 DBG("SIGTERM caught");
298 stop_threads();
299 break;
300 default:
301 break;
302 }
303 }
304
305 /*
306 * Setup signal handler for :
307 * SIGINT, SIGTERM, SIGPIPE
308 */
309 static
310 int set_signal_handler(void)
311 {
312 int ret = 0;
313 struct sigaction sa;
314 sigset_t sigset;
315
316 if ((ret = sigemptyset(&sigset)) < 0) {
317 PERROR("sigemptyset");
318 return ret;
319 }
320
321 sa.sa_handler = sighandler;
322 sa.sa_mask = sigset;
323 sa.sa_flags = 0;
324 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
325 PERROR("sigaction");
326 return ret;
327 }
328
329 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
330 PERROR("sigaction");
331 return ret;
332 }
333
334 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
335 PERROR("sigaction");
336 return ret;
337 }
338
339 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
340
341 return ret;
342 }
343
344 /*
345 * Init thread quit pipe.
346 *
347 * Return -1 on error or 0 if all pipes are created.
348 */
349 static
350 int init_thread_quit_pipe(void)
351 {
352 int ret, i;
353
354 ret = pipe(thread_quit_pipe);
355 if (ret < 0) {
356 PERROR("thread quit pipe");
357 goto error;
358 }
359
360 for (i = 0; i < 2; i++) {
361 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
362 if (ret < 0) {
363 PERROR("fcntl");
364 goto error;
365 }
366 }
367
368 error:
369 return ret;
370 }
371
372 /*
373 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
374 */
375 static
376 int create_thread_poll_set(struct lttng_poll_event *events, int size)
377 {
378 int ret;
379
380 if (events == NULL || size == 0) {
381 ret = -1;
382 goto error;
383 }
384
385 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
386 if (ret < 0) {
387 goto error;
388 }
389
390 /* Add quit pipe */
391 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
392 if (ret < 0) {
393 goto error;
394 }
395
396 return 0;
397
398 error:
399 return ret;
400 }
401
402 /*
403 * Check if the thread quit pipe was triggered.
404 *
405 * Return 1 if it was triggered else 0;
406 */
407 static
408 int check_thread_quit_pipe(int fd, uint32_t events)
409 {
410 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
411 return 1;
412 }
413
414 return 0;
415 }
416
417 /*
418 * Create and init socket from uri.
419 */
420 static
421 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
422 {
423 int ret;
424 struct lttcomm_sock *sock = NULL;
425
426 sock = lttcomm_alloc_sock_from_uri(uri);
427 if (sock == NULL) {
428 ERR("Allocating socket");
429 goto error;
430 }
431
432 ret = lttcomm_create_sock(sock);
433 if (ret < 0) {
434 goto error;
435 }
436 DBG("Listening on sock %d", sock->fd);
437
438 ret = sock->ops->bind(sock);
439 if (ret < 0) {
440 goto error;
441 }
442
443 ret = sock->ops->listen(sock, -1);
444 if (ret < 0) {
445 goto error;
446
447 }
448
449 return sock;
450
451 error:
452 if (sock) {
453 lttcomm_destroy_sock(sock);
454 }
455 return NULL;
456 }
457
458 /*
459 * This thread manages the listening for new connections on the network
460 */
461 static
462 void *relay_thread_listener(void *data)
463 {
464 int i, ret, pollfd;
465 int val = 1;
466 uint32_t revents, nb_fd;
467 struct lttng_poll_event events;
468 struct lttcomm_sock *control_sock, *data_sock;
469
470 /*
471 * Get allocated in this thread, enqueued to a global queue, dequeued and
472 * freed in the worker thread.
473 */
474 struct relay_command *relay_cmd = NULL;
475
476 DBG("[thread] Relay listener started");
477
478 control_sock = relay_init_sock(control_uri);
479 if (!control_sock) {
480 goto error_sock;
481 }
482
483 data_sock = relay_init_sock(data_uri);
484 if (!data_sock) {
485 goto error_sock;
486 }
487
488 /*
489 * Pass 3 as size here for the thread quit pipe, control and data socket.
490 */
491 ret = create_thread_poll_set(&events, 3);
492 if (ret < 0) {
493 goto error_create_poll;
494 }
495
496 /* Add the control socket */
497 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
498 if (ret < 0) {
499 goto error_poll_add;
500 }
501
502 /* Add the data socket */
503 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
504 if (ret < 0) {
505 goto error_poll_add;
506 }
507
508 while (1) {
509 DBG("Listener accepting connections");
510
511 nb_fd = LTTNG_POLL_GETNB(&events);
512
513 restart:
514 ret = lttng_poll_wait(&events, -1);
515 if (ret < 0) {
516 /*
517 * Restart interrupted system call.
518 */
519 if (errno == EINTR) {
520 goto restart;
521 }
522 goto error;
523 }
524
525 DBG("Relay new connection received");
526 for (i = 0; i < nb_fd; i++) {
527 /* Fetch once the poll data */
528 revents = LTTNG_POLL_GETEV(&events, i);
529 pollfd = LTTNG_POLL_GETFD(&events, i);
530
531 /* Thread quit pipe has been closed. Killing thread. */
532 ret = check_thread_quit_pipe(pollfd, revents);
533 if (ret) {
534 goto error;
535 }
536
537 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
538 ERR("socket poll error");
539 goto error;
540 } else if (revents & LPOLLIN) {
541 struct lttcomm_sock *newsock = NULL;
542
543 relay_cmd = zmalloc(sizeof(struct relay_command));
544 if (relay_cmd == NULL) {
545 PERROR("relay command zmalloc");
546 goto error;
547 }
548
549 if (pollfd == data_sock->fd) {
550 newsock = data_sock->ops->accept(data_sock);
551 if (newsock < 0) {
552 PERROR("accepting data sock");
553 goto error;
554 }
555 relay_cmd->type = RELAY_DATA;
556 DBG("Relay data connection accepted, socket %d", newsock->fd);
557 } else if (pollfd == control_sock->fd) {
558 newsock = control_sock->ops->accept(control_sock);
559 if (newsock < 0) {
560 PERROR("accepting control sock");
561 goto error;
562 }
563 relay_cmd->type = RELAY_CONTROL;
564 DBG("Relay control connection accepted, socket %d", newsock->fd);
565 }
566 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
567 &val, sizeof(int));
568 if (ret < 0) {
569 PERROR("setsockopt inet");
570 goto error;
571 }
572 relay_cmd->sock = newsock;
573 /*
574 * Lock free enqueue the request.
575 */
576 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
577
578 /*
579 * Wake the dispatch queue futex. Implicit memory
580 * barrier with the exchange in cds_wfq_enqueue.
581 */
582 futex_nto1_wake(&relay_cmd_queue.futex);
583 }
584 }
585 }
586
587 error:
588 error_poll_add:
589 lttng_poll_clean(&events);
590 error_create_poll:
591 if (control_sock->fd >= 0) {
592 ret = control_sock->ops->close(control_sock);
593 if (ret) {
594 PERROR("close");
595 }
596 lttcomm_destroy_sock(control_sock);
597 }
598 if (data_sock->fd >= 0) {
599 ret = data_sock->ops->close(data_sock);
600 if (ret) {
601 PERROR("close");
602 }
603 lttcomm_destroy_sock(data_sock);
604 }
605
606 DBG("Relay listener thread cleanup complete");
607 stop_threads();
608 error_sock:
609 return NULL;
610 }
611
612 /*
613 * This thread manages the dispatching of the requests to worker threads
614 */
615 static
616 void *relay_thread_dispatcher(void *data)
617 {
618 int ret;
619 struct cds_wfq_node *node;
620 struct relay_command *relay_cmd = NULL;
621
622 DBG("[thread] Relay dispatcher started");
623
624 while (!dispatch_thread_exit) {
625 /* Atomically prepare the queue futex */
626 futex_nto1_prepare(&relay_cmd_queue.futex);
627
628 do {
629 /* Dequeue commands */
630 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
631 if (node == NULL) {
632 DBG("Woken up but nothing in the relay command queue");
633 /* Continue thread execution */
634 break;
635 }
636
637 relay_cmd = caa_container_of(node, struct relay_command, node);
638 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
639
640 /*
641 * Inform worker thread of the new request. This
642 * call is blocking so we can be assured that the data will be read
643 * at some point in time or wait to the end of the world :)
644 */
645 ret = write(relay_cmd_pipe[1], relay_cmd,
646 sizeof(struct relay_command));
647 free(relay_cmd);
648 if (ret < 0) {
649 PERROR("write cmd pipe");
650 goto error;
651 }
652 } while (node != NULL);
653
654 /* Futex wait on queue. Blocking call on futex() */
655 futex_nto1_wait(&relay_cmd_queue.futex);
656 }
657
658 error:
659 DBG("Dispatch thread dying");
660 stop_threads();
661 return NULL;
662 }
663
664 /*
665 * Return the realpath(3) of the path even if the last directory token does not
666 * exist. For example, with /tmp/test1/test2, if test2/ does not exist but the
667 * /tmp/test1 does, the real path is returned. In normal time, realpath(3)
668 * fails if the end point directory does not exist.
669 */
670 static
671 char *expand_full_path(const char *path)
672 {
673 const char *end_path = path;
674 char *next, *cut_path, *expanded_path;
675
676 /* Find last token delimited by '/' */
677 while ((next = strpbrk(end_path + 1, "/"))) {
678 end_path = next;
679 }
680
681 /* Cut last token from original path */
682 cut_path = strndup(path, end_path - path);
683
684 expanded_path = malloc(PATH_MAX);
685 if (expanded_path == NULL) {
686 goto error;
687 }
688
689 expanded_path = realpath((char *)cut_path, expanded_path);
690 if (expanded_path == NULL) {
691 switch (errno) {
692 case ENOENT:
693 ERR("%s: No such file or directory", cut_path);
694 break;
695 default:
696 PERROR("realpath");
697 break;
698 }
699 goto error;
700 }
701
702 /* Add end part to expanded path */
703 strcat(expanded_path, end_path);
704
705 free(cut_path);
706 return expanded_path;
707
708 error:
709 free(cut_path);
710 return NULL;
711 }
712
713
714 /*
715 * config_get_default_path
716 *
717 * Returns the HOME directory path. Caller MUST NOT free(3) the return pointer.
718 */
719 static
720 char *config_get_default_path(void)
721 {
722 return getenv("HOME");
723 }
724
725 /*
726 * Create recursively directory using the FULL path.
727 */
728 static
729 int mkdir_recursive(char *path, mode_t mode)
730 {
731 char *p, tmp[PATH_MAX];
732 struct stat statbuf;
733 size_t len;
734 int ret;
735
736 ret = snprintf(tmp, sizeof(tmp), "%s", path);
737 if (ret < 0) {
738 PERROR("snprintf mkdir");
739 goto error;
740 }
741
742 len = ret;
743 if (tmp[len - 1] == '/') {
744 tmp[len - 1] = 0;
745 }
746
747 for (p = tmp + 1; *p; p++) {
748 if (*p == '/') {
749 *p = 0;
750 if (tmp[strlen(tmp) - 1] == '.' &&
751 tmp[strlen(tmp) - 2] == '.' &&
752 tmp[strlen(tmp) - 3] == '/') {
753 ERR("Using '/../' is not permitted in the trace path (%s)",
754 tmp);
755 ret = -1;
756 goto error;
757 }
758 ret = stat(tmp, &statbuf);
759 if (ret < 0) {
760 ret = mkdir(tmp, mode);
761 if (ret < 0) {
762 if (!(errno == EEXIST)) {
763 PERROR("mkdir recursive");
764 ret = -errno;
765 goto error;
766 }
767 }
768 }
769 *p = '/';
770 }
771 }
772
773 ret = mkdir(tmp, mode);
774 if (ret < 0) {
775 if (!(errno == EEXIST)) {
776 PERROR("mkdir recursive last piece");
777 ret = -errno;
778 } else {
779 ret = 0;
780 }
781 }
782
783 error:
784 return ret;
785 }
786
787 /*
788 * create_output_path: create the output trace directory
789 */
790 static
791 char *create_output_path(char *path_name)
792 {
793 int ret = 0;
794 char *alloc_path = NULL;
795 char *traces_path = NULL;
796 char *full_path = NULL;
797
798 /* Auto output path */
799 if (opt_output_path == NULL) {
800 alloc_path = strdup(config_get_default_path());
801 if (alloc_path == NULL) {
802 ERR("Home path not found.\n \
803 Please specify an output path using -o, --output PATH");
804 ret = -1;
805 goto exit;
806 }
807
808 ret = asprintf(&traces_path, "%s/" DEFAULT_TRACE_DIR_NAME
809 "/%s", alloc_path, path_name);
810 if (ret < 0) {
811 PERROR("asprintf trace dir name");
812 goto exit;
813 }
814 } else {
815 full_path = expand_full_path(opt_output_path);
816 ret = asprintf(&traces_path, "%s/%s", full_path, path_name);
817 if (ret < 0) {
818 PERROR("asprintf trace dir name");
819 goto exit;
820 }
821 }
822 free(alloc_path);
823 free(full_path);
824
825 exit:
826 return traces_path;
827 }
828
829 /*
830 * relay_delete_session: Free all memory associated with a session and
831 * close all the FDs
832 */
833 static
834 void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht)
835 {
836 struct lttng_ht_iter iter;
837 struct lttng_ht_node_ulong *node;
838 struct relay_stream *stream;
839 int ret;
840
841 if (!cmd->session)
842 return;
843
844 DBG("Relay deleting session %lu", cmd->session->id);
845 free(cmd->session->sock);
846
847 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, node, node) {
848 node = lttng_ht_iter_get_node_ulong(&iter);
849 if (node) {
850 stream = caa_container_of(node,
851 struct relay_stream, stream_n);
852 if (stream->session == cmd->session) {
853 close(stream->fd);
854 ret = lttng_ht_del(streams_ht, &iter);
855 assert(!ret);
856 free(stream);
857 }
858 }
859 }
860 }
861
862 /*
863 * relay_add_stream: allocate a new stream for a session
864 */
865 static
866 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
867 struct relay_command *cmd, struct lttng_ht *streams_ht)
868 {
869 struct relay_session *session = cmd->session;
870 struct lttcomm_relayd_add_stream stream_info;
871 struct relay_stream *stream = NULL;
872 struct lttcomm_relayd_status_stream reply;
873 char *path = NULL, *root_path = NULL;
874 int ret, send_ret;
875
876 if (!session || session->version_check_done == 0) {
877 ERR("Trying to add a stream before version check");
878 ret = -1;
879 goto end_no_session;
880 }
881
882 /* FIXME : use data_size for something ? */
883 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
884 sizeof(struct lttcomm_relayd_add_stream), MSG_WAITALL);
885 if (ret < sizeof(struct lttcomm_relayd_add_stream)) {
886 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
887 ret = -1;
888 goto end_no_session;
889 }
890 stream = zmalloc(sizeof(struct relay_stream));
891 if (stream == NULL) {
892 PERROR("relay stream zmalloc");
893 ret = -1;
894 goto end_no_session;
895 }
896
897 stream->stream_handle = ++last_relay_stream_id;
898 stream->seq = 0;
899 stream->session = session;
900
901 root_path = create_output_path(stream_info.pathname);
902 if (!root_path) {
903 ret = -1;
904 goto end;
905 }
906 ret = mkdir_recursive(root_path, S_IRWXU | S_IRWXG);
907 if (ret < 0) {
908 free(root_path);
909 ERR("relay creating output directory");
910 goto end;
911 }
912
913 ret = asprintf(&path, "%s/%s", root_path, stream_info.channel_name);
914 if (ret < 0) {
915 PERROR("asprintf stream path");
916 goto end;
917 }
918
919 ret = open(path, O_WRONLY|O_CREAT|O_TRUNC, S_IRWXU|S_IRWXG|S_IRWXO);
920 if (ret < 0) {
921 PERROR("Relay creating trace file");
922 goto end;
923 }
924
925 stream->fd = ret;
926 DBG("Tracefile %s created", path);
927
928 lttng_ht_node_init_ulong(&stream->stream_n,
929 (unsigned long) stream->stream_handle);
930 lttng_ht_add_unique_ulong(streams_ht,
931 &stream->stream_n);
932
933 DBG("Relay new stream added %s", stream_info.channel_name);
934
935 end:
936 free(path);
937 free(root_path);
938 /* send the session id to the client or a negative return code on error */
939 if (ret < 0) {
940 reply.ret_code = htobe32(LTTCOMM_ERR);
941 } else {
942 reply.ret_code = htobe32(LTTCOMM_OK);
943 }
944 reply.handle = htobe64(stream->stream_handle);
945 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
946 sizeof(struct lttcomm_relayd_status_stream), 0);
947 if (send_ret < 0) {
948 ERR("Relay sending stream id");
949 }
950
951 end_no_session:
952 return ret;
953 }
954
955 /*
956 * relay_unknown_command: send -1 if received unknown command
957 */
958 static
959 void relay_unknown_command(struct relay_command *cmd)
960 {
961 struct lttcomm_relayd_generic_reply reply;
962 int ret;
963
964 reply.ret_code = htobe32(LTTCOMM_ERR);
965 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
966 sizeof(struct lttcomm_relayd_generic_reply), 0);
967 if (ret < 0) {
968 ERR("Relay sending unknown command");
969 }
970 }
971
972 /*
973 * relay_start: send an acknowledgment to the client to tell if we are
974 * ready to receive data. We are ready if a session is established.
975 */
976 static
977 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
978 struct relay_command *cmd)
979 {
980 int ret = htobe32(LTTCOMM_OK);
981 struct lttcomm_relayd_generic_reply reply;
982 struct relay_session *session = cmd->session;
983
984 if (!session) {
985 DBG("Trying to start the streaming without a session established");
986 ret = htobe32(LTTCOMM_ERR);
987 }
988
989 reply.ret_code = ret;
990 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
991 sizeof(struct lttcomm_relayd_generic_reply), 0);
992 if (ret < 0) {
993 ERR("Relay sending start ack");
994 }
995
996 return ret;
997 }
998
999 /*
1000 * Get stream from stream id.
1001 */
1002 static
1003 struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id,
1004 struct lttng_ht *streams_ht)
1005 {
1006 struct lttng_ht_node_ulong *node;
1007 struct lttng_ht_iter iter;
1008 struct relay_stream *ret;
1009
1010 lttng_ht_lookup(streams_ht,
1011 (void *)((unsigned long) stream_id),
1012 &iter);
1013 node = lttng_ht_iter_get_node_ulong(&iter);
1014 if (node == NULL) {
1015 DBG("Relay stream %lu not found", stream_id);
1016 ret = NULL;
1017 goto end;
1018 }
1019
1020 ret = caa_container_of(node, struct relay_stream, stream_n);
1021
1022 end:
1023 return ret;
1024 }
1025
1026 /*
1027 * relay_recv_metadata: receive the metada for the session.
1028 */
1029 static
1030 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1031 struct relay_command *cmd, struct lttng_ht *streams_ht)
1032 {
1033 int ret = htobe32(LTTCOMM_OK);
1034 struct relay_session *session = cmd->session;
1035 struct lttcomm_relayd_metadata_payload *metadata_struct;
1036 struct relay_stream *metadata_stream;
1037 uint64_t data_size, payload_size;
1038
1039 if (!session) {
1040 ERR("Metadata sent before version check");
1041 ret = -1;
1042 goto end;
1043 }
1044
1045 data_size = be64toh(recv_hdr->data_size);
1046 payload_size = data_size - sizeof(uint64_t);
1047 if (data_buffer_size < data_size) {
1048 data_buffer = realloc(data_buffer, data_size);
1049 if (!data_buffer) {
1050 ERR("Allocating data buffer");
1051 ret = -1;
1052 goto end;
1053 }
1054 data_buffer_size = data_size;
1055 }
1056 memset(data_buffer, 0, data_size);
1057 DBG2("Relay receiving metadata, waiting for %lu bytes", data_size);
1058 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, MSG_WAITALL);
1059 if (ret < 0 || ret != data_size) {
1060 ret = -1;
1061 ERR("Relay didn't receive the whole metadata");
1062 goto end;
1063 }
1064 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1065 metadata_stream = relay_stream_from_stream_id(
1066 be64toh(metadata_struct->stream_id), streams_ht);
1067 if (!metadata_stream) {
1068 ret = -1;
1069 goto end;
1070 }
1071
1072 ret = write(metadata_stream->fd, metadata_struct->payload,
1073 payload_size);
1074 if (ret < (payload_size)) {
1075 ERR("Relay error writing metadata on file");
1076 ret = -1;
1077 goto end;
1078 }
1079 DBG2("Relay metadata written");
1080
1081 end:
1082 return ret;
1083 }
1084
1085 /*
1086 * relay_send_version: send relayd version number
1087 */
1088 static
1089 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1090 struct relay_command *cmd)
1091 {
1092 int ret = htobe32(LTTCOMM_OK);
1093 struct lttcomm_relayd_version reply;
1094 struct relay_session *session = NULL;
1095
1096 if (cmd->session == NULL) {
1097 session = zmalloc(sizeof(struct relay_session));
1098 if (session == NULL) {
1099 PERROR("relay session zmalloc");
1100 ret = -1;
1101 goto end;
1102 }
1103 session->id = ++last_relay_session_id;
1104 DBG("Created session %lu", session->id);
1105 cmd->session = session;
1106 }
1107 session->version_check_done = 1;
1108
1109 sscanf(VERSION, "%u.%u", &reply.major, &reply.minor);
1110 reply.major = htobe32(reply.major);
1111 reply.minor = htobe32(reply.minor);
1112 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1113 sizeof(struct lttcomm_relayd_version), 0);
1114 if (ret < 0) {
1115 ERR("Relay sending version");
1116 }
1117 DBG("Version check done");
1118
1119 end:
1120 return ret;
1121 }
1122
1123 /*
1124 * relay_process_control: Process the commands received on the control socket
1125 */
1126 static
1127 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1128 struct relay_command *cmd, struct lttng_ht *streams_ht)
1129 {
1130 int ret = 0;
1131
1132 switch (be32toh(recv_hdr->cmd)) {
1133 /*
1134 case RELAYD_CREATE_SESSION:
1135 ret = relay_create_session(recv_hdr, cmd);
1136 break;
1137 */
1138 case RELAYD_ADD_STREAM:
1139 ret = relay_add_stream(recv_hdr, cmd, streams_ht);
1140 break;
1141 case RELAYD_START_DATA:
1142 ret = relay_start(recv_hdr, cmd);
1143 break;
1144 case RELAYD_SEND_METADATA:
1145 ret = relay_recv_metadata(recv_hdr, cmd, streams_ht);
1146 break;
1147 case RELAYD_VERSION:
1148 ret = relay_send_version(recv_hdr, cmd);
1149 break;
1150 case RELAYD_UPDATE_SYNC_INFO:
1151 default:
1152 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1153 relay_unknown_command(cmd);
1154 ret = -1;
1155 goto end;
1156 }
1157
1158 end:
1159 return ret;
1160 }
1161
1162 /*
1163 * relay_process_data: Process the data received on the data socket
1164 */
1165 static
1166 int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht)
1167 {
1168 int ret = 0;
1169 struct relay_stream *stream;
1170 struct lttcomm_relayd_data_hdr data_hdr;
1171 uint64_t stream_id;
1172 uint32_t data_size;
1173
1174 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1175 sizeof(struct lttcomm_relayd_data_hdr), MSG_WAITALL);
1176 if (ret <= 0) {
1177 ERR("Connections seems to be closed");
1178 ret = -1;
1179 goto end;
1180 }
1181
1182 stream_id = be64toh(data_hdr.stream_id);
1183 stream = relay_stream_from_stream_id(stream_id, streams_ht);
1184 if (!stream) {
1185 ret = -1;
1186 goto end;
1187 }
1188
1189 data_size = be32toh(data_hdr.data_size);
1190 if (data_buffer_size < data_size) {
1191 data_buffer = realloc(data_buffer, data_size);
1192 if (!data_buffer) {
1193 ERR("Allocating data buffer");
1194 ret = -1;
1195 goto end;
1196 }
1197 data_buffer_size = data_size;
1198 }
1199 memset(data_buffer, 0, data_size);
1200
1201 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, MSG_WAITALL);
1202 if (ret <= 0) {
1203 ret = -1;
1204 goto end;
1205 }
1206
1207 ret = write(stream->fd, data_buffer, data_size);
1208 if (ret < data_size) {
1209 ERR("Relay error writing data to file");
1210 ret = -1;
1211 goto end;
1212 }
1213 DBG2("Relay wrote %d bytes to tracefile for stream id %lu", ret, stream->stream_handle);
1214
1215 end:
1216 return ret;
1217 }
1218
1219 static
1220 void relay_cleanup_connection(struct lttng_ht *relay_connections_ht, struct lttng_poll_event *events,
1221 struct lttng_ht *streams_ht, int pollfd, struct lttng_ht_iter *iter)
1222 {
1223 int ret;
1224
1225 ret = lttng_ht_del(relay_connections_ht, iter);
1226 assert(!ret);
1227 lttng_poll_del(events, pollfd);
1228
1229 ret = close(pollfd);
1230 if (ret < 0) {
1231 ERR("Closing pollfd %d", pollfd);
1232 }
1233 }
1234
1235 static
1236 int relay_add_connection(int fd, struct lttng_poll_event *events,
1237 struct lttng_ht *relay_connections_ht)
1238 {
1239 int ret;
1240 struct relay_command *relay_connection;
1241
1242 relay_connection = zmalloc(sizeof(struct relay_command));
1243 if (relay_connection == NULL) {
1244 PERROR("Relay command zmalloc");
1245 ret = -1;
1246 goto end;
1247 }
1248 ret = read(fd, relay_connection, sizeof(struct relay_command));
1249 if (ret < 0 || ret < sizeof(relay_connection)) {
1250 PERROR("read relay cmd pipe");
1251 ret = -1;
1252 goto end;
1253 }
1254
1255 lttng_ht_node_init_ulong(&relay_connection->sock_n,
1256 (unsigned long) relay_connection->sock->fd);
1257 lttng_ht_add_unique_ulong(relay_connections_ht,
1258 &relay_connection->sock_n);
1259 ret = lttng_poll_add(events,
1260 relay_connection->sock->fd,
1261 LPOLLIN | LPOLLRDHUP);
1262
1263 end:
1264 return ret;
1265 }
1266
1267 /*
1268 * This thread does the actual work
1269 */
1270 static
1271 void *relay_thread_worker(void *data)
1272 {
1273 int i, ret, pollfd;
1274 uint32_t revents, nb_fd;
1275 struct relay_command *relay_connection;
1276 struct lttng_poll_event events;
1277 struct lttng_ht *relay_connections_ht;
1278 struct lttng_ht_node_ulong *node;
1279 struct lttng_ht_iter iter;
1280 struct lttng_ht *streams_ht;
1281 struct lttcomm_relayd_hdr recv_hdr;
1282
1283 DBG("[thread] Relay worker started");
1284
1285 /* table of connections indexed on socket */
1286 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1287
1288 /* tables of streams indexed by stream ID */
1289 streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1290
1291 ret = create_thread_poll_set(&events, 2);
1292 if (ret < 0) {
1293 goto error_poll_create;
1294 }
1295
1296 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1297 if (ret < 0) {
1298 goto error;
1299 }
1300
1301 while (1) {
1302 /* Zeroed the events structure */
1303 lttng_poll_reset(&events);
1304
1305 nb_fd = LTTNG_POLL_GETNB(&events);
1306
1307 /* Infinite blocking call, waiting for transmission */
1308 restart:
1309 ret = lttng_poll_wait(&events, -1);
1310 if (ret < 0) {
1311 /*
1312 * Restart interrupted system call.
1313 */
1314 if (errno == EINTR) {
1315 goto restart;
1316 }
1317 goto error;
1318 }
1319
1320 for (i = 0; i < nb_fd; i++) {
1321 /* Fetch once the poll data */
1322 revents = LTTNG_POLL_GETEV(&events, i);
1323 pollfd = LTTNG_POLL_GETFD(&events, i);
1324
1325 /* Thread quit pipe has been closed. Killing thread. */
1326 ret = check_thread_quit_pipe(pollfd, revents);
1327 if (ret) {
1328 goto error;
1329 }
1330
1331 /* Inspect the relay cmd pipe for new connection */
1332 if (pollfd == relay_cmd_pipe[0]) {
1333 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1334 ERR("Relay pipe error");
1335 goto error;
1336 } else if (revents & LPOLLIN) {
1337 DBG("Relay command received");
1338 ret = relay_add_connection(relay_cmd_pipe[0],
1339 &events, relay_connections_ht);
1340 if (ret < 0) {
1341 goto error;
1342 }
1343 }
1344 } else if (revents > 0) {
1345 lttng_ht_lookup(relay_connections_ht,
1346 (void *)((unsigned long) pollfd),
1347 &iter);
1348 node = lttng_ht_iter_get_node_ulong(&iter);
1349 if (node == NULL) {
1350 DBG2("Relay sock %d not found", pollfd);
1351 goto error;
1352 }
1353 relay_connection = caa_container_of(node,
1354 struct relay_command, sock_n);
1355
1356 if (revents & (LPOLLERR)) {
1357 ERR("POLL ERROR");
1358 relay_cleanup_connection(relay_connections_ht,
1359 &events, streams_ht, pollfd, &iter);
1360 free(relay_connection);
1361 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
1362 DBG("Socket %d hung up", pollfd);
1363 relay_cleanup_connection(relay_connections_ht,
1364 &events, streams_ht, pollfd, &iter);
1365 if (relay_connection->type == RELAY_CONTROL) {
1366 relay_delete_session(relay_connection, streams_ht);
1367 }
1368 free(relay_connection);
1369 } else if (revents & LPOLLIN) {
1370 /* control socket */
1371 if (relay_connection->type == RELAY_CONTROL) {
1372 ret = relay_connection->sock->ops->recvmsg(
1373 relay_connection->sock, &recv_hdr,
1374 sizeof(struct lttcomm_relayd_hdr), MSG_WAITALL);
1375 /* connection closed */
1376 if (ret <= 0) {
1377 relay_cleanup_connection(relay_connections_ht,
1378 &events, streams_ht, pollfd, &iter);
1379 relay_delete_session(relay_connection, streams_ht);
1380 free(relay_connection);
1381 DBG("Control connection closed with %d", pollfd);
1382 } else {
1383 if (relay_connection->session) {
1384 DBG2("Relay worker receiving data for session : %lu",
1385 relay_connection->session->id);
1386 }
1387 ret = relay_process_control(&recv_hdr,
1388 relay_connection,
1389 streams_ht);
1390 /*
1391 * there was an error in processing a control
1392 * command: clear the session
1393 * */
1394 if (ret < 0) {
1395 relay_cleanup_connection(relay_connections_ht,
1396 &events, streams_ht, pollfd, &iter);
1397 free(relay_connection);
1398 DBG("Connection closed with %d", pollfd);
1399 }
1400 }
1401 /* data socket */
1402 } else if (relay_connection->type == RELAY_DATA) {
1403 ret = relay_process_data(relay_connection, streams_ht);
1404 /* connection closed */
1405 if (ret < 0) {
1406 relay_cleanup_connection(relay_connections_ht,
1407 &events, streams_ht, pollfd, &iter);
1408 relay_delete_session(relay_connection, streams_ht);
1409 DBG("Data connection closed with %d", pollfd);
1410 }
1411 }
1412 }
1413 }
1414 }
1415 }
1416
1417 error:
1418 lttng_poll_clean(&events);
1419
1420 /* empty the hash table and free the memory */
1421 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
1422 node = lttng_ht_iter_get_node_ulong(&iter);
1423 if (node) {
1424 relay_connection = caa_container_of(node,
1425 struct relay_command, sock_n);
1426 free(relay_connection);
1427 }
1428 ret = lttng_ht_del(relay_connections_ht, &iter);
1429 assert(!ret);
1430 }
1431 error_poll_create:
1432 free(data_buffer);
1433 lttng_ht_destroy(relay_connections_ht);
1434 DBG("Worker thread cleanup complete");
1435 stop_threads();
1436 return NULL;
1437 }
1438
1439 /*
1440 * Create the relay command pipe to wake thread_manage_apps.
1441 * Closed in cleanup().
1442 */
1443 static int create_relay_cmd_pipe(void)
1444 {
1445 int ret, i;
1446
1447 ret = pipe(relay_cmd_pipe);
1448 if (ret < 0) {
1449 PERROR("relay cmd pipe");
1450 goto error;
1451 }
1452
1453 for (i = 0; i < 2; i++) {
1454 ret = fcntl(relay_cmd_pipe[i], F_SETFD, FD_CLOEXEC);
1455 if (ret < 0) {
1456 PERROR("fcntl relay_cmd_pipe");
1457 goto error;
1458 }
1459 }
1460
1461 error:
1462 return ret;
1463 }
1464
1465 /*
1466 * main
1467 */
1468 int main(int argc, char **argv)
1469 {
1470 int ret = 0;
1471 void *status;
1472
1473 /* Create thread quit pipe */
1474 if ((ret = init_thread_quit_pipe()) < 0) {
1475 goto error;
1476 }
1477
1478 /* Parse arguments */
1479 progname = argv[0];
1480 if ((ret = parse_args(argc, argv) < 0)) {
1481 goto error;
1482 }
1483
1484 if ((ret = set_signal_handler()) < 0) {
1485 goto exit;
1486 }
1487
1488 /* Daemonize */
1489 if (opt_daemon) {
1490 ret = daemon(0, 0);
1491 if (ret < 0) {
1492 PERROR("daemon");
1493 goto error;
1494 }
1495 }
1496
1497 /* Check if daemon is UID = 0 */
1498 is_root = !getuid();
1499
1500 if (!is_root) {
1501 if (control_uri->port < 1024 || data_uri->port < 1024) {
1502 ERR("Need to be root to use ports < 1024");
1503 ret = -1;
1504 goto error;
1505 }
1506 }
1507
1508 /* Setup the thread apps communication pipe. */
1509 if ((ret = create_relay_cmd_pipe()) < 0) {
1510 goto exit;
1511 }
1512
1513 /* Init relay command queue. */
1514 cds_wfq_init(&relay_cmd_queue.queue);
1515
1516 /* Set up max poll set size */
1517 lttng_poll_set_max_size();
1518
1519 /* Setup the dispatcher thread */
1520 ret = pthread_create(&dispatcher_thread, NULL,
1521 relay_thread_dispatcher, (void *) NULL);
1522 if (ret != 0) {
1523 PERROR("pthread_create dispatcher");
1524 goto exit_dispatcher;
1525 }
1526
1527 /* Setup the worker thread */
1528 ret = pthread_create(&worker_thread, NULL,
1529 relay_thread_worker, (void *) NULL);
1530 if (ret != 0) {
1531 PERROR("pthread_create worker");
1532 goto exit_worker;
1533 }
1534
1535 /* Setup the listener thread */
1536 ret = pthread_create(&listener_thread, NULL,
1537 relay_thread_listener, (void *) NULL);
1538 if (ret != 0) {
1539 PERROR("pthread_create listener");
1540 goto exit_listener;
1541 }
1542
1543 exit_listener:
1544 ret = pthread_join(listener_thread, &status);
1545 if (ret != 0) {
1546 PERROR("pthread_join");
1547 goto error; /* join error, exit without cleanup */
1548 }
1549
1550 exit_worker:
1551 ret = pthread_join(worker_thread, &status);
1552 if (ret != 0) {
1553 PERROR("pthread_join");
1554 goto error; /* join error, exit without cleanup */
1555 }
1556
1557 exit_dispatcher:
1558 ret = pthread_join(dispatcher_thread, &status);
1559 if (ret != 0) {
1560 PERROR("pthread_join");
1561 goto error; /* join error, exit without cleanup */
1562 }
1563
1564 exit:
1565 cleanup();
1566 if (!ret) {
1567 exit(EXIT_SUCCESS);
1568 }
1569 error:
1570 exit(EXIT_FAILURE);
1571 }
This page took 0.126863 seconds and 3 git commands to generate.