Rename data_available to data_pending
[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 <inttypes.h>
36 #include <urcu/futex.h>
37 #include <urcu/uatomic.h>
38 #include <unistd.h>
39 #include <fcntl.h>
40 #include <config.h>
41
42 #include <lttng/lttng.h>
43 #include <common/common.h>
44 #include <common/compat/poll.h>
45 #include <common/compat/socket.h>
46 #include <common/defaults.h>
47 #include <common/futex.h>
48 #include <common/sessiond-comm/sessiond-comm.h>
49 #include <common/sessiond-comm/inet.h>
50 #include <common/hashtable/hashtable.h>
51 #include <common/sessiond-comm/relayd.h>
52 #include <common/uri.h>
53 #include <common/utils.h>
54
55 #include "lttng-relayd.h"
56
57 /* command line options */
58 static int opt_daemon;
59 static char *opt_output_path;
60 static struct lttng_uri *control_uri;
61 static struct lttng_uri *data_uri;
62
63 const char *progname;
64 static int is_root; /* Set to 1 if the daemon is running as root */
65
66 /*
67 * Quit pipe for all threads. This permits a single cancellation point
68 * for all threads when receiving an event on the pipe.
69 */
70 static int thread_quit_pipe[2] = { -1, -1 };
71
72 /*
73 * This pipe is used to inform the worker thread that a command is queued and
74 * ready to be processed.
75 */
76 static int relay_cmd_pipe[2] = { -1, -1 };
77
78 /* Shared between threads */
79 static int dispatch_thread_exit;
80
81 static pthread_t listener_thread;
82 static pthread_t dispatcher_thread;
83 static pthread_t worker_thread;
84
85 static uint64_t last_relay_stream_id;
86 static uint64_t last_relay_session_id;
87
88 /*
89 * Relay command queue.
90 *
91 * The relay_thread_listener and relay_thread_dispatcher communicate with this
92 * queue.
93 */
94 static struct relay_cmd_queue relay_cmd_queue;
95
96 /* buffer allocated at startup, used to store the trace data */
97 static char *data_buffer;
98 static unsigned int data_buffer_size;
99
100 /*
101 * usage function on stderr
102 */
103 static
104 void usage(void)
105 {
106 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
107 fprintf(stderr, " -h, --help Display this usage.\n");
108 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
109 fprintf(stderr, " -C, --control-port Control port listening (URI)\n");
110 fprintf(stderr, " -D, --data-port Data port listening (URI)\n");
111 fprintf(stderr, " -o, --output Output path for traces (PATH)\n");
112 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
113 }
114
115 static
116 int parse_args(int argc, char **argv)
117 {
118 int c;
119 int ret = 0;
120 char *default_address;
121
122 static struct option long_options[] = {
123 { "control-port", 1, 0, 'C', },
124 { "data-port", 1, 0, 'D', },
125 { "daemonize", 0, 0, 'd', },
126 { "help", 0, 0, 'h', },
127 { "output", 1, 0, 'o', },
128 { "verbose", 0, 0, 'v', },
129 { NULL, 0, 0, 0, },
130 };
131
132 while (1) {
133 int option_index = 0;
134 c = getopt_long(argc, argv, "dhv" "C:D:o:",
135 long_options, &option_index);
136 if (c == -1) {
137 break;
138 }
139
140 switch (c) {
141 case 0:
142 fprintf(stderr, "option %s", long_options[option_index].name);
143 if (optarg) {
144 fprintf(stderr, " with arg %s\n", optarg);
145 }
146 break;
147 case 'C':
148 ret = uri_parse(optarg, &control_uri);
149 if (ret < 0) {
150 ERR("Invalid control URI specified");
151 goto exit;
152 }
153 if (control_uri->port == 0) {
154 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
155 }
156 break;
157 case 'D':
158 ret = uri_parse(optarg, &data_uri);
159 if (ret < 0) {
160 ERR("Invalid data URI specified");
161 goto exit;
162 }
163 if (data_uri->port == 0) {
164 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
165 }
166 break;
167 case 'd':
168 opt_daemon = 1;
169 break;
170 case 'h':
171 usage();
172 exit(EXIT_FAILURE);
173 case 'o':
174 ret = asprintf(&opt_output_path, "%s", optarg);
175 if (ret < 0) {
176 PERROR("asprintf opt_output_path");
177 goto exit;
178 }
179 break;
180 case 'v':
181 /* Verbose level can increase using multiple -v */
182 lttng_opt_verbose += 1;
183 break;
184 default:
185 /* Unknown option or other error.
186 * Error is printed by getopt, just return */
187 ret = -1;
188 goto exit;
189 }
190 }
191
192 /* assign default values */
193 if (control_uri == NULL) {
194 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
195 DEFAULT_NETWORK_CONTROL_PORT);
196 if (ret < 0) {
197 PERROR("asprintf default data address");
198 goto exit;
199 }
200
201 ret = uri_parse(default_address, &control_uri);
202 free(default_address);
203 if (ret < 0) {
204 ERR("Invalid control URI specified");
205 goto exit;
206 }
207 }
208 if (data_uri == NULL) {
209 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
210 DEFAULT_NETWORK_DATA_PORT);
211 if (ret < 0) {
212 PERROR("asprintf default data address");
213 goto exit;
214 }
215
216 ret = uri_parse(default_address, &data_uri);
217 free(default_address);
218 if (ret < 0) {
219 ERR("Invalid data URI specified");
220 goto exit;
221 }
222 }
223
224 exit:
225 return ret;
226 }
227
228 /*
229 * Cleanup the daemon
230 */
231 static
232 void cleanup(void)
233 {
234 DBG("Cleaning up");
235
236 /* free the dynamically allocated opt_output_path */
237 free(opt_output_path);
238
239 /* Close thread quit pipes */
240 utils_close_pipe(thread_quit_pipe);
241
242 /* Close relay cmd pipes */
243 utils_close_pipe(relay_cmd_pipe);
244
245 uri_free(control_uri);
246 uri_free(data_uri);
247 }
248
249 /*
250 * Write to writable pipe used to notify a thread.
251 */
252 static
253 int notify_thread_pipe(int wpipe)
254 {
255 int ret;
256
257 do {
258 ret = write(wpipe, "!", 1);
259 } while (ret < 0 && errno == EINTR);
260 if (ret < 0) {
261 PERROR("write poll pipe");
262 }
263
264 return ret;
265 }
266
267 /*
268 * Stop all threads by closing the thread quit pipe.
269 */
270 static
271 void stop_threads(void)
272 {
273 int ret;
274
275 /* Stopping all threads */
276 DBG("Terminating all threads");
277 ret = notify_thread_pipe(thread_quit_pipe[1]);
278 if (ret < 0) {
279 ERR("write error on thread quit pipe");
280 }
281
282 /* Dispatch thread */
283 CMM_STORE_SHARED(dispatch_thread_exit, 1);
284 futex_nto1_wake(&relay_cmd_queue.futex);
285 }
286
287 /*
288 * Signal handler for the daemon
289 *
290 * Simply stop all worker threads, leaving main() return gracefully after
291 * joining all threads and calling cleanup().
292 */
293 static
294 void sighandler(int sig)
295 {
296 switch (sig) {
297 case SIGPIPE:
298 DBG("SIGPIPE caught");
299 return;
300 case SIGINT:
301 DBG("SIGINT caught");
302 stop_threads();
303 break;
304 case SIGTERM:
305 DBG("SIGTERM caught");
306 stop_threads();
307 break;
308 default:
309 break;
310 }
311 }
312
313 /*
314 * Setup signal handler for :
315 * SIGINT, SIGTERM, SIGPIPE
316 */
317 static
318 int set_signal_handler(void)
319 {
320 int ret = 0;
321 struct sigaction sa;
322 sigset_t sigset;
323
324 if ((ret = sigemptyset(&sigset)) < 0) {
325 PERROR("sigemptyset");
326 return ret;
327 }
328
329 sa.sa_handler = sighandler;
330 sa.sa_mask = sigset;
331 sa.sa_flags = 0;
332 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
333 PERROR("sigaction");
334 return ret;
335 }
336
337 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
338 PERROR("sigaction");
339 return ret;
340 }
341
342 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
343 PERROR("sigaction");
344 return ret;
345 }
346
347 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
348
349 return ret;
350 }
351
352 /*
353 * Init thread quit pipe.
354 *
355 * Return -1 on error or 0 if all pipes are created.
356 */
357 static
358 int init_thread_quit_pipe(void)
359 {
360 int ret;
361
362 ret = utils_create_pipe_cloexec(thread_quit_pipe);
363
364 return ret;
365 }
366
367 /*
368 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
369 */
370 static
371 int create_thread_poll_set(struct lttng_poll_event *events, int size)
372 {
373 int ret;
374
375 if (events == NULL || size == 0) {
376 ret = -1;
377 goto error;
378 }
379
380 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
381 if (ret < 0) {
382 goto error;
383 }
384
385 /* Add quit pipe */
386 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
387 if (ret < 0) {
388 goto error;
389 }
390
391 return 0;
392
393 error:
394 return ret;
395 }
396
397 /*
398 * Check if the thread quit pipe was triggered.
399 *
400 * Return 1 if it was triggered else 0;
401 */
402 static
403 int check_thread_quit_pipe(int fd, uint32_t events)
404 {
405 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
406 return 1;
407 }
408
409 return 0;
410 }
411
412 /*
413 * Create and init socket from uri.
414 */
415 static
416 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
417 {
418 int ret;
419 struct lttcomm_sock *sock = NULL;
420
421 sock = lttcomm_alloc_sock_from_uri(uri);
422 if (sock == NULL) {
423 ERR("Allocating socket");
424 goto error;
425 }
426
427 ret = lttcomm_create_sock(sock);
428 if (ret < 0) {
429 goto error;
430 }
431 DBG("Listening on sock %d", sock->fd);
432
433 ret = sock->ops->bind(sock);
434 if (ret < 0) {
435 goto error;
436 }
437
438 ret = sock->ops->listen(sock, -1);
439 if (ret < 0) {
440 goto error;
441
442 }
443
444 return sock;
445
446 error:
447 if (sock) {
448 lttcomm_destroy_sock(sock);
449 }
450 return NULL;
451 }
452
453 /*
454 * Return nonzero if stream needs to be closed.
455 */
456 static
457 int close_stream_check(struct relay_stream *stream)
458 {
459
460 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
461 return 1;
462 }
463 return 0;
464 }
465
466 /*
467 * This thread manages the listening for new connections on the network
468 */
469 static
470 void *relay_thread_listener(void *data)
471 {
472 int i, ret, pollfd, err = -1;
473 int val = 1;
474 uint32_t revents, nb_fd;
475 struct lttng_poll_event events;
476 struct lttcomm_sock *control_sock, *data_sock;
477
478 DBG("[thread] Relay listener started");
479
480 control_sock = relay_init_sock(control_uri);
481 if (!control_sock) {
482 goto error_sock_control;
483 }
484
485 data_sock = relay_init_sock(data_uri);
486 if (!data_sock) {
487 goto error_sock_relay;
488 }
489
490 /*
491 * Pass 3 as size here for the thread quit pipe, control and data socket.
492 */
493 ret = create_thread_poll_set(&events, 3);
494 if (ret < 0) {
495 goto error_create_poll;
496 }
497
498 /* Add the control socket */
499 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
500 if (ret < 0) {
501 goto error_poll_add;
502 }
503
504 /* Add the data socket */
505 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
506 if (ret < 0) {
507 goto error_poll_add;
508 }
509
510 while (1) {
511 DBG("Listener accepting connections");
512
513 nb_fd = LTTNG_POLL_GETNB(&events);
514
515 restart:
516 ret = lttng_poll_wait(&events, -1);
517 if (ret < 0) {
518 /*
519 * Restart interrupted system call.
520 */
521 if (errno == EINTR) {
522 goto restart;
523 }
524 goto error;
525 }
526
527 DBG("Relay new connection received");
528 for (i = 0; i < nb_fd; i++) {
529 /* Fetch once the poll data */
530 revents = LTTNG_POLL_GETEV(&events, i);
531 pollfd = LTTNG_POLL_GETFD(&events, i);
532
533 /* Thread quit pipe has been closed. Killing thread. */
534 ret = check_thread_quit_pipe(pollfd, revents);
535 if (ret) {
536 err = 0;
537 goto exit;
538 }
539
540 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
541 ERR("socket poll error");
542 goto error;
543 } else if (revents & LPOLLIN) {
544 /*
545 * Get allocated in this thread,
546 * enqueued to a global queue, dequeued
547 * and freed in the worker thread.
548 */
549 struct relay_command *relay_cmd;
550 struct lttcomm_sock *newsock;
551
552 relay_cmd = zmalloc(sizeof(struct relay_command));
553 if (relay_cmd == NULL) {
554 PERROR("relay command zmalloc");
555 goto error;
556 }
557
558 if (pollfd == data_sock->fd) {
559 newsock = data_sock->ops->accept(data_sock);
560 if (!newsock) {
561 PERROR("accepting data sock");
562 free(relay_cmd);
563 goto error;
564 }
565 relay_cmd->type = RELAY_DATA;
566 DBG("Relay data connection accepted, socket %d", newsock->fd);
567 } else {
568 assert(pollfd == control_sock->fd);
569 newsock = control_sock->ops->accept(control_sock);
570 if (!newsock) {
571 PERROR("accepting control sock");
572 free(relay_cmd);
573 goto error;
574 }
575 relay_cmd->type = RELAY_CONTROL;
576 DBG("Relay control connection accepted, socket %d", newsock->fd);
577 }
578 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
579 &val, sizeof(int));
580 if (ret < 0) {
581 PERROR("setsockopt inet");
582 lttcomm_destroy_sock(newsock);
583 free(relay_cmd);
584 goto error;
585 }
586 relay_cmd->sock = newsock;
587 /*
588 * Lock free enqueue the request.
589 */
590 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
591
592 /*
593 * Wake the dispatch queue futex. Implicit memory
594 * barrier with the exchange in cds_wfq_enqueue.
595 */
596 futex_nto1_wake(&relay_cmd_queue.futex);
597 }
598 }
599 }
600
601 exit:
602 error:
603 error_poll_add:
604 lttng_poll_clean(&events);
605 error_create_poll:
606 if (data_sock->fd >= 0) {
607 ret = data_sock->ops->close(data_sock);
608 if (ret) {
609 PERROR("close");
610 }
611 }
612 lttcomm_destroy_sock(data_sock);
613 error_sock_relay:
614 if (control_sock->fd >= 0) {
615 ret = control_sock->ops->close(control_sock);
616 if (ret) {
617 PERROR("close");
618 }
619 }
620 lttcomm_destroy_sock(control_sock);
621 error_sock_control:
622 if (err) {
623 DBG("Thread exited with error");
624 }
625 DBG("Relay listener thread cleanup complete");
626 stop_threads();
627 return NULL;
628 }
629
630 /*
631 * This thread manages the dispatching of the requests to worker threads
632 */
633 static
634 void *relay_thread_dispatcher(void *data)
635 {
636 int ret;
637 struct cds_wfq_node *node;
638 struct relay_command *relay_cmd = NULL;
639
640 DBG("[thread] Relay dispatcher started");
641
642 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
643 /* Atomically prepare the queue futex */
644 futex_nto1_prepare(&relay_cmd_queue.futex);
645
646 do {
647 /* Dequeue commands */
648 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
649 if (node == NULL) {
650 DBG("Woken up but nothing in the relay command queue");
651 /* Continue thread execution */
652 break;
653 }
654
655 relay_cmd = caa_container_of(node, struct relay_command, node);
656 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
657
658 /*
659 * Inform worker thread of the new request. This
660 * call is blocking so we can be assured that the data will be read
661 * at some point in time or wait to the end of the world :)
662 */
663 do {
664 ret = write(relay_cmd_pipe[1], relay_cmd,
665 sizeof(struct relay_command));
666 } while (ret < 0 && errno == EINTR);
667 free(relay_cmd);
668 if (ret < 0) {
669 PERROR("write cmd pipe");
670 goto error;
671 }
672 } while (node != NULL);
673
674 /* Futex wait on queue. Blocking call on futex() */
675 futex_nto1_wait(&relay_cmd_queue.futex);
676 }
677
678 error:
679 DBG("Dispatch thread dying");
680 stop_threads();
681 return NULL;
682 }
683
684 /*
685 * Return the realpath(3) of the path even if the last directory token does not
686 * exist. For example, with /tmp/test1/test2, if test2/ does not exist but the
687 * /tmp/test1 does, the real path is returned. In normal time, realpath(3)
688 * fails if the end point directory does not exist.
689 */
690 static
691 char *expand_full_path(const char *path)
692 {
693 const char *end_path = path;
694 char *next, *cut_path, *expanded_path, *respath;
695
696 /* Find last token delimited by '/' */
697 while ((next = strpbrk(end_path + 1, "/"))) {
698 end_path = next;
699 }
700
701 /* Cut last token from original path */
702 cut_path = strndup(path, end_path - path);
703
704 expanded_path = malloc(PATH_MAX);
705 if (expanded_path == NULL) {
706 respath = NULL;
707 goto end;
708 }
709
710 respath = realpath(cut_path, expanded_path);
711 if (respath == NULL) {
712 switch (errno) {
713 case ENOENT:
714 ERR("%s: No such file or directory", cut_path);
715 break;
716 default:
717 PERROR("realpath");
718 break;
719 }
720 free(expanded_path);
721 } else {
722 /* Add end part to expanded path */
723 strcat(respath, end_path);
724 }
725 end:
726 free(cut_path);
727 return respath;
728 }
729
730
731 /*
732 * config_get_default_path
733 *
734 * Returns the HOME directory path. Caller MUST NOT free(3) the return pointer.
735 */
736 static
737 char *config_get_default_path(void)
738 {
739 return getenv("HOME");
740 }
741
742 /*
743 * Create recursively directory using the FULL path.
744 */
745 static
746 int mkdir_recursive(char *path, mode_t mode)
747 {
748 char *p, tmp[PATH_MAX];
749 struct stat statbuf;
750 size_t len;
751 int ret;
752
753 ret = snprintf(tmp, sizeof(tmp), "%s", path);
754 if (ret < 0) {
755 PERROR("snprintf mkdir");
756 goto error;
757 }
758
759 len = ret;
760 if (tmp[len - 1] == '/') {
761 tmp[len - 1] = 0;
762 }
763
764 for (p = tmp + 1; *p; p++) {
765 if (*p == '/') {
766 *p = 0;
767 if (tmp[strlen(tmp) - 1] == '.' &&
768 tmp[strlen(tmp) - 2] == '.' &&
769 tmp[strlen(tmp) - 3] == '/') {
770 ERR("Using '/../' is not permitted in the trace path (%s)",
771 tmp);
772 ret = -1;
773 goto error;
774 }
775 ret = stat(tmp, &statbuf);
776 if (ret < 0) {
777 ret = mkdir(tmp, mode);
778 if (ret < 0) {
779 if (errno != EEXIST) {
780 PERROR("mkdir recursive");
781 ret = -errno;
782 goto error;
783 }
784 }
785 }
786 *p = '/';
787 }
788 }
789
790 ret = mkdir(tmp, mode);
791 if (ret < 0) {
792 if (errno != EEXIST) {
793 PERROR("mkdir recursive last piece");
794 ret = -errno;
795 } else {
796 ret = 0;
797 }
798 }
799
800 error:
801 return ret;
802 }
803
804 static
805 char *create_output_path_auto(char *path_name)
806 {
807 int ret;
808 char *traces_path = NULL;
809 char *alloc_path = NULL;
810 char *default_path;
811
812 default_path = config_get_default_path();
813 if (default_path == NULL) {
814 ERR("Home path not found.\n \
815 Please specify an output path using -o, --output PATH");
816 goto exit;
817 }
818 alloc_path = strdup(default_path);
819 if (alloc_path == NULL) {
820 PERROR("Path allocation");
821 goto exit;
822 }
823 ret = asprintf(&traces_path, "%s/" DEFAULT_TRACE_DIR_NAME
824 "/%s", alloc_path, path_name);
825 if (ret < 0) {
826 PERROR("asprintf trace dir name");
827 goto exit;
828 }
829 exit:
830 free(alloc_path);
831 return traces_path;
832 }
833
834 static
835 char *create_output_path_noauto(char *path_name)
836 {
837 int ret;
838 char *traces_path = NULL;
839 char *full_path;
840
841 full_path = expand_full_path(opt_output_path);
842 ret = asprintf(&traces_path, "%s/%s", full_path, path_name);
843 if (ret < 0) {
844 PERROR("asprintf trace dir name");
845 goto exit;
846 }
847 exit:
848 free(full_path);
849 return traces_path;
850 }
851
852 /*
853 * create_output_path: create the output trace directory
854 */
855 static
856 char *create_output_path(char *path_name)
857 {
858 if (opt_output_path == NULL) {
859 return create_output_path_auto(path_name);
860 } else {
861 return create_output_path_noauto(path_name);
862 }
863 }
864
865 static
866 void deferred_free_stream(struct rcu_head *head)
867 {
868 struct relay_stream *stream =
869 caa_container_of(head, struct relay_stream, rcu_node);
870 free(stream);
871 }
872
873 /*
874 * relay_delete_session: Free all memory associated with a session and
875 * close all the FDs
876 */
877 static
878 void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht)
879 {
880 struct lttng_ht_iter iter;
881 struct lttng_ht_node_ulong *node;
882 struct relay_stream *stream;
883 int ret;
884
885 if (!cmd->session) {
886 return;
887 }
888
889 DBG("Relay deleting session %" PRIu64, cmd->session->id);
890
891 lttcomm_destroy_sock(cmd->session->sock);
892
893 rcu_read_lock();
894 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, node, node) {
895 node = lttng_ht_iter_get_node_ulong(&iter);
896 if (node) {
897 stream = caa_container_of(node,
898 struct relay_stream, stream_n);
899 if (stream->session == cmd->session) {
900 close(stream->fd);
901 ret = lttng_ht_del(streams_ht, &iter);
902 assert(!ret);
903 call_rcu(&stream->rcu_node,
904 deferred_free_stream);
905 }
906 }
907 }
908 rcu_read_unlock();
909
910 free(cmd->session);
911 }
912
913 /*
914 * relay_add_stream: allocate a new stream for a session
915 */
916 static
917 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
918 struct relay_command *cmd, struct lttng_ht *streams_ht)
919 {
920 struct relay_session *session = cmd->session;
921 struct lttcomm_relayd_add_stream stream_info;
922 struct relay_stream *stream = NULL;
923 struct lttcomm_relayd_status_stream reply;
924 char *path = NULL, *root_path = NULL;
925 int ret, send_ret;
926
927 if (!session || session->version_check_done == 0) {
928 ERR("Trying to add a stream before version check");
929 ret = -1;
930 goto end_no_session;
931 }
932
933 /* FIXME : use data_size for something ? */
934 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
935 sizeof(struct lttcomm_relayd_add_stream), MSG_WAITALL);
936 if (ret < sizeof(struct lttcomm_relayd_add_stream)) {
937 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
938 ret = -1;
939 goto end_no_session;
940 }
941 stream = zmalloc(sizeof(struct relay_stream));
942 if (stream == NULL) {
943 PERROR("relay stream zmalloc");
944 ret = -1;
945 goto end_no_session;
946 }
947
948 rcu_read_lock();
949 stream->stream_handle = ++last_relay_stream_id;
950 stream->prev_seq = -1ULL;
951 stream->session = session;
952
953 root_path = create_output_path(stream_info.pathname);
954 if (!root_path) {
955 ret = -1;
956 goto end;
957 }
958 ret = mkdir_recursive(root_path, S_IRWXU | S_IRWXG);
959 if (ret < 0) {
960 ERR("relay creating output directory");
961 goto end;
962 }
963
964 ret = asprintf(&path, "%s/%s", root_path, stream_info.channel_name);
965 if (ret < 0) {
966 PERROR("asprintf stream path");
967 goto end;
968 }
969
970 ret = open(path, O_WRONLY|O_CREAT|O_TRUNC, S_IRWXU|S_IRWXG|S_IRWXO);
971 if (ret < 0) {
972 PERROR("Relay creating trace file");
973 goto end;
974 }
975
976 stream->fd = ret;
977 DBG("Tracefile %s created", path);
978
979 lttng_ht_node_init_ulong(&stream->stream_n,
980 (unsigned long) stream->stream_handle);
981 lttng_ht_add_unique_ulong(streams_ht,
982 &stream->stream_n);
983
984 DBG("Relay new stream added %s", stream_info.channel_name);
985
986 end:
987 free(path);
988 free(root_path);
989 /* send the session id to the client or a negative return code on error */
990 if (ret < 0) {
991 reply.ret_code = htobe32(LTTNG_ERR_UNK);
992 } else {
993 reply.ret_code = htobe32(LTTNG_OK);
994 }
995 reply.handle = htobe64(stream->stream_handle);
996 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
997 sizeof(struct lttcomm_relayd_status_stream), 0);
998 if (send_ret < 0) {
999 ERR("Relay sending stream id");
1000 }
1001 rcu_read_unlock();
1002
1003 end_no_session:
1004 return ret;
1005 }
1006
1007 /*
1008 * relay_close_stream: close a specific stream
1009 */
1010 static
1011 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1012 struct relay_command *cmd, struct lttng_ht *streams_ht)
1013 {
1014 struct relay_session *session = cmd->session;
1015 struct lttcomm_relayd_close_stream stream_info;
1016 struct lttcomm_relayd_generic_reply reply;
1017 struct relay_stream *stream;
1018 int ret, send_ret;
1019 struct lttng_ht_node_ulong *node;
1020 struct lttng_ht_iter iter;
1021
1022 DBG("Close stream received");
1023
1024 if (!session || session->version_check_done == 0) {
1025 ERR("Trying to close a stream before version check");
1026 ret = -1;
1027 goto end_no_session;
1028 }
1029
1030 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1031 sizeof(struct lttcomm_relayd_close_stream), MSG_WAITALL);
1032 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1033 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1034 ret = -1;
1035 goto end_no_session;
1036 }
1037
1038 rcu_read_lock();
1039 lttng_ht_lookup(streams_ht,
1040 (void *)((unsigned long) be64toh(stream_info.stream_id)),
1041 &iter);
1042 node = lttng_ht_iter_get_node_ulong(&iter);
1043 if (node == NULL) {
1044 DBG("Relay stream %" PRIu64 " not found", be64toh(stream_info.stream_id));
1045 ret = -1;
1046 goto end_unlock;
1047 }
1048
1049 stream = caa_container_of(node, struct relay_stream, stream_n);
1050 if (!stream) {
1051 ret = -1;
1052 goto end_unlock;
1053 }
1054
1055 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1056 stream->close_flag = 1;
1057
1058 if (close_stream_check(stream)) {
1059 int delret;
1060
1061 close(stream->fd);
1062 delret = lttng_ht_del(streams_ht, &iter);
1063 assert(!delret);
1064 call_rcu(&stream->rcu_node,
1065 deferred_free_stream);
1066 DBG("Closed tracefile %d from close stream", stream->fd);
1067 }
1068
1069 end_unlock:
1070 rcu_read_unlock();
1071
1072 if (ret < 0) {
1073 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1074 } else {
1075 reply.ret_code = htobe32(LTTNG_OK);
1076 }
1077 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1078 sizeof(struct lttcomm_relayd_generic_reply), 0);
1079 if (send_ret < 0) {
1080 ERR("Relay sending stream id");
1081 }
1082
1083 end_no_session:
1084 return ret;
1085 }
1086
1087 /*
1088 * relay_unknown_command: send -1 if received unknown command
1089 */
1090 static
1091 void relay_unknown_command(struct relay_command *cmd)
1092 {
1093 struct lttcomm_relayd_generic_reply reply;
1094 int ret;
1095
1096 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1097 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1098 sizeof(struct lttcomm_relayd_generic_reply), 0);
1099 if (ret < 0) {
1100 ERR("Relay sending unknown command");
1101 }
1102 }
1103
1104 /*
1105 * relay_start: send an acknowledgment to the client to tell if we are
1106 * ready to receive data. We are ready if a session is established.
1107 */
1108 static
1109 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1110 struct relay_command *cmd)
1111 {
1112 int ret = htobe32(LTTNG_OK);
1113 struct lttcomm_relayd_generic_reply reply;
1114 struct relay_session *session = cmd->session;
1115
1116 if (!session) {
1117 DBG("Trying to start the streaming without a session established");
1118 ret = htobe32(LTTNG_ERR_UNK);
1119 }
1120
1121 reply.ret_code = ret;
1122 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1123 sizeof(struct lttcomm_relayd_generic_reply), 0);
1124 if (ret < 0) {
1125 ERR("Relay sending start ack");
1126 }
1127
1128 return ret;
1129 }
1130
1131 /*
1132 * Get stream from stream id.
1133 * Need to be called with RCU read-side lock held.
1134 */
1135 static
1136 struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id,
1137 struct lttng_ht *streams_ht)
1138 {
1139 struct lttng_ht_node_ulong *node;
1140 struct lttng_ht_iter iter;
1141 struct relay_stream *ret;
1142
1143 lttng_ht_lookup(streams_ht,
1144 (void *)((unsigned long) stream_id),
1145 &iter);
1146 node = lttng_ht_iter_get_node_ulong(&iter);
1147 if (node == NULL) {
1148 DBG("Relay stream %" PRIu64 " not found", stream_id);
1149 ret = NULL;
1150 goto end;
1151 }
1152
1153 ret = caa_container_of(node, struct relay_stream, stream_n);
1154
1155 end:
1156 return ret;
1157 }
1158
1159 /*
1160 * Append padding to the file pointed by the file descriptor fd.
1161 */
1162 static int write_padding_to_file(int fd, uint32_t size)
1163 {
1164 int ret = 0;
1165 char *zeros;
1166
1167 if (size == 0) {
1168 goto end;
1169 }
1170
1171 zeros = zmalloc(size);
1172 if (zeros == NULL) {
1173 PERROR("zmalloc zeros for padding");
1174 ret = -1;
1175 goto end;
1176 }
1177
1178 do {
1179 ret = write(fd, zeros, size);
1180 } while (ret < 0 && errno == EINTR);
1181 if (ret < 0) {
1182 PERROR("write padding to file");
1183 }
1184
1185 free(zeros);
1186
1187 end:
1188 return ret;
1189 }
1190
1191 /*
1192 * relay_recv_metadata: receive the metada for the session.
1193 */
1194 static
1195 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1196 struct relay_command *cmd, struct lttng_ht *streams_ht)
1197 {
1198 int ret = htobe32(LTTNG_OK);
1199 struct relay_session *session = cmd->session;
1200 struct lttcomm_relayd_metadata_payload *metadata_struct;
1201 struct relay_stream *metadata_stream;
1202 uint64_t data_size, payload_size;
1203
1204 if (!session) {
1205 ERR("Metadata sent before version check");
1206 ret = -1;
1207 goto end;
1208 }
1209
1210 data_size = payload_size = be64toh(recv_hdr->data_size);
1211 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1212 ERR("Incorrect data size");
1213 ret = -1;
1214 goto end;
1215 }
1216 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1217
1218 if (data_buffer_size < data_size) {
1219 /* In case the realloc fails, we can free the memory */
1220 char *tmp_data_ptr = data_buffer;
1221 data_buffer = realloc(data_buffer, data_size);
1222 if (!data_buffer) {
1223 ERR("Allocating data buffer");
1224 free(tmp_data_ptr);
1225 ret = -1;
1226 goto end;
1227 }
1228 data_buffer_size = data_size;
1229 }
1230 memset(data_buffer, 0, data_size);
1231 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1232 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size,
1233 MSG_WAITALL);
1234 if (ret < 0 || ret != data_size) {
1235 ret = -1;
1236 ERR("Relay didn't receive the whole metadata");
1237 goto end;
1238 }
1239 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1240
1241 rcu_read_lock();
1242 metadata_stream = relay_stream_from_stream_id(
1243 be64toh(metadata_struct->stream_id), streams_ht);
1244 if (!metadata_stream) {
1245 ret = -1;
1246 goto end_unlock;
1247 }
1248
1249 do {
1250 ret = write(metadata_stream->fd, metadata_struct->payload,
1251 payload_size);
1252 } while (ret < 0 && errno == EINTR);
1253 if (ret < payload_size) {
1254 ERR("Relay error writing metadata on file");
1255 ret = -1;
1256 goto end_unlock;
1257 }
1258
1259 ret = write_padding_to_file(metadata_stream->fd,
1260 be32toh(metadata_struct->padding_size));
1261 if (ret < 0) {
1262 goto end_unlock;
1263 }
1264
1265 DBG2("Relay metadata written");
1266
1267 end_unlock:
1268 rcu_read_unlock();
1269 end:
1270 return ret;
1271 }
1272
1273 /*
1274 * relay_send_version: send relayd version number
1275 */
1276 static
1277 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1278 struct relay_command *cmd)
1279 {
1280 int ret;
1281 struct lttcomm_relayd_version reply;
1282 struct relay_session *session;
1283
1284 if (cmd->session == NULL) {
1285 session = zmalloc(sizeof(struct relay_session));
1286 if (session == NULL) {
1287 PERROR("relay session zmalloc");
1288 ret = -1;
1289 goto end;
1290 }
1291 session->id = ++last_relay_session_id;
1292 DBG("Created session %" PRIu64, session->id);
1293 cmd->session = session;
1294 } else {
1295 session = cmd->session;
1296 }
1297 session->version_check_done = 1;
1298
1299 ret = sscanf(VERSION, "%u.%u", &reply.major, &reply.minor);
1300 if (ret < 2) {
1301 ERR("Error in scanning version");
1302 ret = -1;
1303 goto end;
1304 }
1305 reply.major = htobe32(reply.major);
1306 reply.minor = htobe32(reply.minor);
1307 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1308 sizeof(struct lttcomm_relayd_version), 0);
1309 if (ret < 0) {
1310 ERR("Relay sending version");
1311 }
1312 DBG("Version check done (%u.%u)", be32toh(reply.major),
1313 be32toh(reply.minor));
1314
1315 end:
1316 return ret;
1317 }
1318
1319 /*
1320 * Check for data pending for a given stream id from the session daemon.
1321 */
1322 static
1323 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1324 struct relay_command *cmd, struct lttng_ht *streams_ht)
1325 {
1326 struct relay_session *session = cmd->session;
1327 struct lttcomm_relayd_data_pending msg;
1328 struct lttcomm_relayd_generic_reply reply;
1329 struct relay_stream *stream;
1330 int ret;
1331 struct lttng_ht_node_ulong *node;
1332 struct lttng_ht_iter iter;
1333 uint64_t last_net_seq_num, stream_id;
1334
1335 DBG("Data pending command received");
1336
1337 if (!session || session->version_check_done == 0) {
1338 ERR("Trying to check for data before version check");
1339 ret = -1;
1340 goto end_no_session;
1341 }
1342
1343 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), MSG_WAITALL);
1344 if (ret < sizeof(msg)) {
1345 ERR("Relay didn't receive valid data_pending struct size : %d", ret);
1346 ret = -1;
1347 goto end_no_session;
1348 }
1349
1350 stream_id = be64toh(msg.stream_id);
1351 last_net_seq_num = be64toh(msg.last_net_seq_num);
1352
1353 rcu_read_lock();
1354 lttng_ht_lookup(streams_ht, (void *)((unsigned long) stream_id), &iter);
1355 node = lttng_ht_iter_get_node_ulong(&iter);
1356 if (node == NULL) {
1357 DBG("Relay stream %" PRIu64 " not found", stream_id);
1358 ret = -1;
1359 goto end_unlock;
1360 }
1361
1362 stream = caa_container_of(node, struct relay_stream, stream_n);
1363 assert(stream);
1364
1365 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1366 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1367 last_net_seq_num);
1368
1369 /* Avoid wrapping issue */
1370 if (((int64_t) (stream->prev_seq - last_net_seq_num)) <= 0) {
1371 /* Data has in fact been written and is NOT pending */
1372 ret = 0;
1373 } else {
1374 /* Data still being streamed thus pending */
1375 ret = 1;
1376 }
1377
1378 end_unlock:
1379 rcu_read_unlock();
1380
1381 reply.ret_code = htobe32(ret);
1382 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1383 if (ret < 0) {
1384 ERR("Relay data pending ret code failed");
1385 }
1386
1387 end_no_session:
1388 return ret;
1389 }
1390
1391 /*
1392 * Wait for the control socket to reach a quiescent state.
1393 *
1394 * Note that for now, when receiving this command from the session daemon, this
1395 * means that every subsequent commands or data received on the control socket
1396 * has been handled. So, this is why we simply return OK here.
1397 */
1398 static
1399 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1400 struct relay_command *cmd)
1401 {
1402 int ret;
1403 struct lttcomm_relayd_generic_reply reply;
1404
1405 DBG("Checking quiescent state on control socket");
1406
1407 reply.ret_code = htobe32(LTTNG_OK);
1408 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1409 if (ret < 0) {
1410 ERR("Relay data quiescent control ret code failed");
1411 }
1412
1413 return ret;
1414 }
1415
1416 /*
1417 * relay_process_control: Process the commands received on the control socket
1418 */
1419 static
1420 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1421 struct relay_command *cmd, struct lttng_ht *streams_ht)
1422 {
1423 int ret = 0;
1424
1425 switch (be32toh(recv_hdr->cmd)) {
1426 /*
1427 case RELAYD_CREATE_SESSION:
1428 ret = relay_create_session(recv_hdr, cmd);
1429 break;
1430 */
1431 case RELAYD_ADD_STREAM:
1432 ret = relay_add_stream(recv_hdr, cmd, streams_ht);
1433 break;
1434 case RELAYD_START_DATA:
1435 ret = relay_start(recv_hdr, cmd);
1436 break;
1437 case RELAYD_SEND_METADATA:
1438 ret = relay_recv_metadata(recv_hdr, cmd, streams_ht);
1439 break;
1440 case RELAYD_VERSION:
1441 ret = relay_send_version(recv_hdr, cmd);
1442 break;
1443 case RELAYD_CLOSE_STREAM:
1444 ret = relay_close_stream(recv_hdr, cmd, streams_ht);
1445 break;
1446 case RELAYD_DATA_PENDING:
1447 ret = relay_data_pending(recv_hdr, cmd, streams_ht);
1448 break;
1449 case RELAYD_QUIESCENT_CONTROL:
1450 ret = relay_quiescent_control(recv_hdr, cmd);
1451 break;
1452 case RELAYD_UPDATE_SYNC_INFO:
1453 default:
1454 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1455 relay_unknown_command(cmd);
1456 ret = -1;
1457 goto end;
1458 }
1459
1460 end:
1461 return ret;
1462 }
1463
1464 /*
1465 * relay_process_data: Process the data received on the data socket
1466 */
1467 static
1468 int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht)
1469 {
1470 int ret = 0;
1471 struct relay_stream *stream;
1472 struct lttcomm_relayd_data_hdr data_hdr;
1473 uint64_t stream_id;
1474 uint64_t net_seq_num;
1475 uint32_t data_size;
1476
1477 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1478 sizeof(struct lttcomm_relayd_data_hdr), MSG_WAITALL);
1479 if (ret <= 0) {
1480 ERR("Connections seems to be closed");
1481 ret = -1;
1482 goto end;
1483 }
1484
1485 stream_id = be64toh(data_hdr.stream_id);
1486
1487 rcu_read_lock();
1488 stream = relay_stream_from_stream_id(stream_id, streams_ht);
1489 if (!stream) {
1490 ret = -1;
1491 goto end_unlock;
1492 }
1493
1494 data_size = be32toh(data_hdr.data_size);
1495 if (data_buffer_size < data_size) {
1496 char *tmp_data_ptr = data_buffer;
1497 data_buffer = realloc(data_buffer, data_size);
1498 if (!data_buffer) {
1499 ERR("Allocating data buffer");
1500 free(tmp_data_ptr);
1501 ret = -1;
1502 goto end_unlock;
1503 }
1504 data_buffer_size = data_size;
1505 }
1506 memset(data_buffer, 0, data_size);
1507
1508 net_seq_num = be64toh(data_hdr.net_seq_num);
1509
1510 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
1511 data_size, stream_id, net_seq_num);
1512 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, MSG_WAITALL);
1513 if (ret <= 0) {
1514 ret = -1;
1515 goto end_unlock;
1516 }
1517
1518 do {
1519 ret = write(stream->fd, data_buffer, data_size);
1520 } while (ret < 0 && errno == EINTR);
1521 if (ret < data_size) {
1522 ERR("Relay error writing data to file");
1523 ret = -1;
1524 goto end_unlock;
1525 }
1526
1527 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
1528 ret, stream->stream_handle);
1529
1530 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
1531 if (ret < 0) {
1532 goto end_unlock;
1533 }
1534
1535 stream->prev_seq = net_seq_num;
1536
1537 /* Check if we need to close the FD */
1538 if (close_stream_check(stream)) {
1539 struct lttng_ht_iter iter;
1540
1541 close(stream->fd);
1542 iter.iter.node = &stream->stream_n.node;
1543 ret = lttng_ht_del(streams_ht, &iter);
1544 assert(!ret);
1545 call_rcu(&stream->rcu_node,
1546 deferred_free_stream);
1547 DBG("Closed tracefile %d after recv data", stream->fd);
1548 }
1549
1550 end_unlock:
1551 rcu_read_unlock();
1552 end:
1553 return ret;
1554 }
1555
1556 static
1557 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
1558 {
1559 int ret;
1560
1561 lttng_poll_del(events, pollfd);
1562
1563 ret = close(pollfd);
1564 if (ret < 0) {
1565 ERR("Closing pollfd %d", pollfd);
1566 }
1567 }
1568
1569 static
1570 int relay_add_connection(int fd, struct lttng_poll_event *events,
1571 struct lttng_ht *relay_connections_ht)
1572 {
1573 struct relay_command *relay_connection;
1574 int ret;
1575
1576 relay_connection = zmalloc(sizeof(struct relay_command));
1577 if (relay_connection == NULL) {
1578 PERROR("Relay command zmalloc");
1579 goto error;
1580 }
1581 ret = read(fd, relay_connection, sizeof(struct relay_command));
1582 if (ret < 0 || ret < sizeof(struct relay_command)) {
1583 PERROR("read relay cmd pipe");
1584 goto error_read;
1585 }
1586
1587 lttng_ht_node_init_ulong(&relay_connection->sock_n,
1588 (unsigned long) relay_connection->sock->fd);
1589 rcu_read_lock();
1590 lttng_ht_add_unique_ulong(relay_connections_ht,
1591 &relay_connection->sock_n);
1592 rcu_read_unlock();
1593 return lttng_poll_add(events,
1594 relay_connection->sock->fd,
1595 LPOLLIN | LPOLLRDHUP);
1596
1597 error_read:
1598 free(relay_connection);
1599 error:
1600 return -1;
1601 }
1602
1603 static
1604 void deferred_free_connection(struct rcu_head *head)
1605 {
1606 struct relay_command *relay_connection =
1607 caa_container_of(head, struct relay_command, rcu_node);
1608
1609 lttcomm_destroy_sock(relay_connection->sock);
1610 free(relay_connection);
1611 }
1612
1613 static
1614 void relay_del_connection(struct lttng_ht *relay_connections_ht,
1615 struct lttng_ht *streams_ht, struct lttng_ht_iter *iter,
1616 struct relay_command *relay_connection)
1617 {
1618 int ret;
1619
1620 ret = lttng_ht_del(relay_connections_ht, iter);
1621 assert(!ret);
1622 if (relay_connection->type == RELAY_CONTROL) {
1623 relay_delete_session(relay_connection, streams_ht);
1624 }
1625
1626 call_rcu(&relay_connection->rcu_node,
1627 deferred_free_connection);
1628 }
1629
1630 /*
1631 * This thread does the actual work
1632 */
1633 static
1634 void *relay_thread_worker(void *data)
1635 {
1636 int i, ret, pollfd, err = -1;
1637 uint32_t revents, nb_fd;
1638 struct relay_command *relay_connection;
1639 struct lttng_poll_event events;
1640 struct lttng_ht *relay_connections_ht;
1641 struct lttng_ht_node_ulong *node;
1642 struct lttng_ht_iter iter;
1643 struct lttng_ht *streams_ht;
1644 struct lttcomm_relayd_hdr recv_hdr;
1645
1646 DBG("[thread] Relay worker started");
1647
1648 rcu_register_thread();
1649
1650 /* table of connections indexed on socket */
1651 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1652 if (!relay_connections_ht) {
1653 goto relay_connections_ht_error;
1654 }
1655
1656 /* tables of streams indexed by stream ID */
1657 streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1658 if (!streams_ht) {
1659 goto streams_ht_error;
1660 }
1661
1662 ret = create_thread_poll_set(&events, 2);
1663 if (ret < 0) {
1664 goto error_poll_create;
1665 }
1666
1667 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1668 if (ret < 0) {
1669 goto error;
1670 }
1671
1672 while (1) {
1673 /* Zeroed the events structure */
1674 lttng_poll_reset(&events);
1675
1676 nb_fd = LTTNG_POLL_GETNB(&events);
1677
1678 /* Infinite blocking call, waiting for transmission */
1679 restart:
1680 DBG3("Relayd worker thread polling...");
1681 ret = lttng_poll_wait(&events, -1);
1682 if (ret < 0) {
1683 /*
1684 * Restart interrupted system call.
1685 */
1686 if (errno == EINTR) {
1687 goto restart;
1688 }
1689 goto error;
1690 }
1691
1692 for (i = 0; i < nb_fd; i++) {
1693 /* Fetch once the poll data */
1694 revents = LTTNG_POLL_GETEV(&events, i);
1695 pollfd = LTTNG_POLL_GETFD(&events, i);
1696
1697 /* Thread quit pipe has been closed. Killing thread. */
1698 ret = check_thread_quit_pipe(pollfd, revents);
1699 if (ret) {
1700 err = 0;
1701 goto exit;
1702 }
1703
1704 /* Inspect the relay cmd pipe for new connection */
1705 if (pollfd == relay_cmd_pipe[0]) {
1706 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1707 ERR("Relay pipe error");
1708 goto error;
1709 } else if (revents & LPOLLIN) {
1710 DBG("Relay command received");
1711 ret = relay_add_connection(relay_cmd_pipe[0],
1712 &events, relay_connections_ht);
1713 if (ret < 0) {
1714 goto error;
1715 }
1716 }
1717 } else if (revents > 0) {
1718 rcu_read_lock();
1719 lttng_ht_lookup(relay_connections_ht,
1720 (void *)((unsigned long) pollfd),
1721 &iter);
1722 node = lttng_ht_iter_get_node_ulong(&iter);
1723 if (node == NULL) {
1724 DBG2("Relay sock %d not found", pollfd);
1725 rcu_read_unlock();
1726 goto error;
1727 }
1728 relay_connection = caa_container_of(node,
1729 struct relay_command, sock_n);
1730
1731 if (revents & (LPOLLERR)) {
1732 ERR("POLL ERROR");
1733 relay_cleanup_poll_connection(&events, pollfd);
1734 relay_del_connection(relay_connections_ht,
1735 streams_ht, &iter,
1736 relay_connection);
1737 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
1738 DBG("Socket %d hung up", pollfd);
1739 relay_cleanup_poll_connection(&events, pollfd);
1740 relay_del_connection(relay_connections_ht,
1741 streams_ht, &iter,
1742 relay_connection);
1743 } else if (revents & LPOLLIN) {
1744 /* control socket */
1745 if (relay_connection->type == RELAY_CONTROL) {
1746 ret = relay_connection->sock->ops->recvmsg(
1747 relay_connection->sock, &recv_hdr,
1748 sizeof(struct lttcomm_relayd_hdr), MSG_WAITALL);
1749 /* connection closed */
1750 if (ret <= 0) {
1751 relay_cleanup_poll_connection(&events, pollfd);
1752 relay_del_connection(relay_connections_ht,
1753 streams_ht, &iter,
1754 relay_connection);
1755 DBG("Control connection closed with %d", pollfd);
1756 } else {
1757 if (relay_connection->session) {
1758 DBG2("Relay worker receiving data for session : %" PRIu64,
1759 relay_connection->session->id);
1760 }
1761 ret = relay_process_control(&recv_hdr,
1762 relay_connection,
1763 streams_ht);
1764 /*
1765 * there was an error in processing a control
1766 * command: clear the session
1767 * */
1768 if (ret < 0) {
1769 relay_cleanup_poll_connection(&events, pollfd);
1770 relay_del_connection(relay_connections_ht,
1771 streams_ht, &iter,
1772 relay_connection);
1773 DBG("Connection closed with %d", pollfd);
1774 }
1775 }
1776 /* data socket */
1777 } else if (relay_connection->type == RELAY_DATA) {
1778 ret = relay_process_data(relay_connection, streams_ht);
1779 /* connection closed */
1780 if (ret < 0) {
1781 relay_cleanup_poll_connection(&events, pollfd);
1782 relay_del_connection(relay_connections_ht,
1783 streams_ht, &iter,
1784 relay_connection);
1785 DBG("Data connection closed with %d", pollfd);
1786 }
1787 }
1788 }
1789 rcu_read_unlock();
1790 }
1791 }
1792 }
1793
1794 exit:
1795 error:
1796 lttng_poll_clean(&events);
1797
1798 /* empty the hash table and free the memory */
1799 rcu_read_lock();
1800 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
1801 node = lttng_ht_iter_get_node_ulong(&iter);
1802 if (node) {
1803 relay_connection = caa_container_of(node,
1804 struct relay_command, sock_n);
1805 relay_del_connection(relay_connections_ht,
1806 streams_ht, &iter,
1807 relay_connection);
1808 }
1809 }
1810 rcu_read_unlock();
1811 error_poll_create:
1812 lttng_ht_destroy(streams_ht);
1813 streams_ht_error:
1814 lttng_ht_destroy(relay_connections_ht);
1815 relay_connections_ht_error:
1816 if (err) {
1817 DBG("Thread exited with error");
1818 }
1819 DBG("Worker thread cleanup complete");
1820 free(data_buffer);
1821 stop_threads();
1822 rcu_unregister_thread();
1823 return NULL;
1824 }
1825
1826 /*
1827 * Create the relay command pipe to wake thread_manage_apps.
1828 * Closed in cleanup().
1829 */
1830 static int create_relay_cmd_pipe(void)
1831 {
1832 int ret;
1833
1834 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
1835
1836 return ret;
1837 }
1838
1839 /*
1840 * main
1841 */
1842 int main(int argc, char **argv)
1843 {
1844 int ret = 0;
1845 void *status;
1846
1847 /* Create thread quit pipe */
1848 if ((ret = init_thread_quit_pipe()) < 0) {
1849 goto error;
1850 }
1851
1852 /* Parse arguments */
1853 progname = argv[0];
1854 if ((ret = parse_args(argc, argv) < 0)) {
1855 goto exit;
1856 }
1857
1858 if ((ret = set_signal_handler()) < 0) {
1859 goto exit;
1860 }
1861
1862 /* Daemonize */
1863 if (opt_daemon) {
1864 ret = daemon(0, 0);
1865 if (ret < 0) {
1866 PERROR("daemon");
1867 goto exit;
1868 }
1869 }
1870
1871 /* Check if daemon is UID = 0 */
1872 is_root = !getuid();
1873
1874 if (!is_root) {
1875 if (control_uri->port < 1024 || data_uri->port < 1024) {
1876 ERR("Need to be root to use ports < 1024");
1877 ret = -1;
1878 goto exit;
1879 }
1880 }
1881
1882 /* Setup the thread apps communication pipe. */
1883 if ((ret = create_relay_cmd_pipe()) < 0) {
1884 goto exit;
1885 }
1886
1887 /* Init relay command queue. */
1888 cds_wfq_init(&relay_cmd_queue.queue);
1889
1890 /* Set up max poll set size */
1891 lttng_poll_set_max_size();
1892
1893 /* Setup the dispatcher thread */
1894 ret = pthread_create(&dispatcher_thread, NULL,
1895 relay_thread_dispatcher, (void *) NULL);
1896 if (ret != 0) {
1897 PERROR("pthread_create dispatcher");
1898 goto exit_dispatcher;
1899 }
1900
1901 /* Setup the worker thread */
1902 ret = pthread_create(&worker_thread, NULL,
1903 relay_thread_worker, (void *) NULL);
1904 if (ret != 0) {
1905 PERROR("pthread_create worker");
1906 goto exit_worker;
1907 }
1908
1909 /* Setup the listener thread */
1910 ret = pthread_create(&listener_thread, NULL,
1911 relay_thread_listener, (void *) NULL);
1912 if (ret != 0) {
1913 PERROR("pthread_create listener");
1914 goto exit_listener;
1915 }
1916
1917 exit_listener:
1918 ret = pthread_join(listener_thread, &status);
1919 if (ret != 0) {
1920 PERROR("pthread_join");
1921 goto error; /* join error, exit without cleanup */
1922 }
1923
1924 exit_worker:
1925 ret = pthread_join(worker_thread, &status);
1926 if (ret != 0) {
1927 PERROR("pthread_join");
1928 goto error; /* join error, exit without cleanup */
1929 }
1930
1931 exit_dispatcher:
1932 ret = pthread_join(dispatcher_thread, &status);
1933 if (ret != 0) {
1934 PERROR("pthread_join");
1935 goto error; /* join error, exit without cleanup */
1936 }
1937
1938 exit:
1939 cleanup();
1940 if (!ret) {
1941 exit(EXIT_SUCCESS);
1942 }
1943
1944 error:
1945 exit(EXIT_FAILURE);
1946 }
This page took 0.086651 seconds and 4 git commands to generate.