a73b4852d574be48f25f2184e2ae793ac46e85d8
[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 uri_free(control_uri);
243 uri_free(data_uri);
244 }
245
246 /*
247 * Write to writable pipe used to notify a thread.
248 */
249 static
250 int notify_thread_pipe(int wpipe)
251 {
252 int ret;
253
254 do {
255 ret = write(wpipe, "!", 1);
256 } while (ret < 0 && errno == EINTR);
257 if (ret < 0 || ret != 1) {
258 PERROR("write poll pipe");
259 }
260
261 return ret;
262 }
263
264 /*
265 * Stop all threads by closing the thread quit pipe.
266 */
267 static
268 void stop_threads(void)
269 {
270 int ret;
271
272 /* Stopping all threads */
273 DBG("Terminating all threads");
274 ret = notify_thread_pipe(thread_quit_pipe[1]);
275 if (ret < 0) {
276 ERR("write error on thread quit pipe");
277 }
278
279 /* Dispatch thread */
280 CMM_STORE_SHARED(dispatch_thread_exit, 1);
281 futex_nto1_wake(&relay_cmd_queue.futex);
282 }
283
284 /*
285 * Signal handler for the daemon
286 *
287 * Simply stop all worker threads, leaving main() return gracefully after
288 * joining all threads and calling cleanup().
289 */
290 static
291 void sighandler(int sig)
292 {
293 switch (sig) {
294 case SIGPIPE:
295 DBG("SIGPIPE caught");
296 return;
297 case SIGINT:
298 DBG("SIGINT caught");
299 stop_threads();
300 break;
301 case SIGTERM:
302 DBG("SIGTERM caught");
303 stop_threads();
304 break;
305 default:
306 break;
307 }
308 }
309
310 /*
311 * Setup signal handler for :
312 * SIGINT, SIGTERM, SIGPIPE
313 */
314 static
315 int set_signal_handler(void)
316 {
317 int ret = 0;
318 struct sigaction sa;
319 sigset_t sigset;
320
321 if ((ret = sigemptyset(&sigset)) < 0) {
322 PERROR("sigemptyset");
323 return ret;
324 }
325
326 sa.sa_handler = sighandler;
327 sa.sa_mask = sigset;
328 sa.sa_flags = 0;
329 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
330 PERROR("sigaction");
331 return ret;
332 }
333
334 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
335 PERROR("sigaction");
336 return ret;
337 }
338
339 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
340 PERROR("sigaction");
341 return ret;
342 }
343
344 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
345
346 return ret;
347 }
348
349 /*
350 * Init thread quit pipe.
351 *
352 * Return -1 on error or 0 if all pipes are created.
353 */
354 static
355 int init_thread_quit_pipe(void)
356 {
357 int ret;
358
359 ret = utils_create_pipe_cloexec(thread_quit_pipe);
360
361 return ret;
362 }
363
364 /*
365 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
366 */
367 static
368 int create_thread_poll_set(struct lttng_poll_event *events, int size)
369 {
370 int ret;
371
372 if (events == NULL || size == 0) {
373 ret = -1;
374 goto error;
375 }
376
377 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
378 if (ret < 0) {
379 goto error;
380 }
381
382 /* Add quit pipe */
383 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
384 if (ret < 0) {
385 goto error;
386 }
387
388 return 0;
389
390 error:
391 return ret;
392 }
393
394 /*
395 * Check if the thread quit pipe was triggered.
396 *
397 * Return 1 if it was triggered else 0;
398 */
399 static
400 int check_thread_quit_pipe(int fd, uint32_t events)
401 {
402 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
403 return 1;
404 }
405
406 return 0;
407 }
408
409 /*
410 * Create and init socket from uri.
411 */
412 static
413 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
414 {
415 int ret;
416 struct lttcomm_sock *sock = NULL;
417
418 sock = lttcomm_alloc_sock_from_uri(uri);
419 if (sock == NULL) {
420 ERR("Allocating socket");
421 goto error;
422 }
423
424 ret = lttcomm_create_sock(sock);
425 if (ret < 0) {
426 goto error;
427 }
428 DBG("Listening on sock %d", sock->fd);
429
430 ret = sock->ops->bind(sock);
431 if (ret < 0) {
432 goto error;
433 }
434
435 ret = sock->ops->listen(sock, -1);
436 if (ret < 0) {
437 goto error;
438
439 }
440
441 return sock;
442
443 error:
444 if (sock) {
445 lttcomm_destroy_sock(sock);
446 }
447 return NULL;
448 }
449
450 /*
451 * Return nonzero if stream needs to be closed.
452 */
453 static
454 int close_stream_check(struct relay_stream *stream)
455 {
456
457 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
458 /*
459 * We are about to close the stream so set the data pending flag to 1
460 * which will make the end data pending command skip the stream which
461 * is now closed and ready. Note that after proceeding to a file close,
462 * the written file is ready for reading.
463 */
464 stream->data_pending_check_done = 1;
465 return 1;
466 }
467 return 0;
468 }
469
470 /*
471 * This thread manages the listening for new connections on the network
472 */
473 static
474 void *relay_thread_listener(void *data)
475 {
476 int i, ret, pollfd, err = -1;
477 int val = 1;
478 uint32_t revents, nb_fd;
479 struct lttng_poll_event events;
480 struct lttcomm_sock *control_sock, *data_sock;
481
482 DBG("[thread] Relay listener started");
483
484 control_sock = relay_init_sock(control_uri);
485 if (!control_sock) {
486 goto error_sock_control;
487 }
488
489 data_sock = relay_init_sock(data_uri);
490 if (!data_sock) {
491 goto error_sock_relay;
492 }
493
494 /*
495 * Pass 3 as size here for the thread quit pipe, control and data socket.
496 */
497 ret = create_thread_poll_set(&events, 3);
498 if (ret < 0) {
499 goto error_create_poll;
500 }
501
502 /* Add the control socket */
503 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
504 if (ret < 0) {
505 goto error_poll_add;
506 }
507
508 /* Add the data socket */
509 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
510 if (ret < 0) {
511 goto error_poll_add;
512 }
513
514 while (1) {
515 DBG("Listener accepting connections");
516
517 restart:
518 ret = lttng_poll_wait(&events, -1);
519 if (ret < 0) {
520 /*
521 * Restart interrupted system call.
522 */
523 if (errno == EINTR) {
524 goto restart;
525 }
526 goto error;
527 }
528
529 nb_fd = ret;
530
531 DBG("Relay new connection received");
532 for (i = 0; i < nb_fd; i++) {
533 /* Fetch once the poll data */
534 revents = LTTNG_POLL_GETEV(&events, i);
535 pollfd = LTTNG_POLL_GETFD(&events, i);
536
537 /* Thread quit pipe has been closed. Killing thread. */
538 ret = check_thread_quit_pipe(pollfd, revents);
539 if (ret) {
540 err = 0;
541 goto exit;
542 }
543
544 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
545 ERR("socket poll error");
546 goto error;
547 } else if (revents & LPOLLIN) {
548 /*
549 * Get allocated in this thread,
550 * enqueued to a global queue, dequeued
551 * and freed in the worker thread.
552 */
553 struct relay_command *relay_cmd;
554 struct lttcomm_sock *newsock;
555
556 relay_cmd = zmalloc(sizeof(struct relay_command));
557 if (relay_cmd == NULL) {
558 PERROR("relay command zmalloc");
559 goto error;
560 }
561
562 if (pollfd == data_sock->fd) {
563 newsock = data_sock->ops->accept(data_sock);
564 if (!newsock) {
565 PERROR("accepting data sock");
566 free(relay_cmd);
567 goto error;
568 }
569 relay_cmd->type = RELAY_DATA;
570 DBG("Relay data connection accepted, socket %d", newsock->fd);
571 } else {
572 assert(pollfd == control_sock->fd);
573 newsock = control_sock->ops->accept(control_sock);
574 if (!newsock) {
575 PERROR("accepting control sock");
576 free(relay_cmd);
577 goto error;
578 }
579 relay_cmd->type = RELAY_CONTROL;
580 DBG("Relay control connection accepted, socket %d", newsock->fd);
581 }
582 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
583 &val, sizeof(int));
584 if (ret < 0) {
585 PERROR("setsockopt inet");
586 lttcomm_destroy_sock(newsock);
587 free(relay_cmd);
588 goto error;
589 }
590 relay_cmd->sock = newsock;
591 /*
592 * Lock free enqueue the request.
593 */
594 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
595
596 /*
597 * Wake the dispatch queue futex. Implicit memory
598 * barrier with the exchange in cds_wfq_enqueue.
599 */
600 futex_nto1_wake(&relay_cmd_queue.futex);
601 }
602 }
603 }
604
605 exit:
606 error:
607 error_poll_add:
608 lttng_poll_clean(&events);
609 error_create_poll:
610 if (data_sock->fd >= 0) {
611 ret = data_sock->ops->close(data_sock);
612 if (ret) {
613 PERROR("close");
614 }
615 }
616 lttcomm_destroy_sock(data_sock);
617 error_sock_relay:
618 if (control_sock->fd >= 0) {
619 ret = control_sock->ops->close(control_sock);
620 if (ret) {
621 PERROR("close");
622 }
623 }
624 lttcomm_destroy_sock(control_sock);
625 error_sock_control:
626 if (err) {
627 DBG("Thread exited with error");
628 }
629 DBG("Relay listener thread cleanup complete");
630 stop_threads();
631 return NULL;
632 }
633
634 /*
635 * This thread manages the dispatching of the requests to worker threads
636 */
637 static
638 void *relay_thread_dispatcher(void *data)
639 {
640 int ret;
641 struct cds_wfq_node *node;
642 struct relay_command *relay_cmd = NULL;
643
644 DBG("[thread] Relay dispatcher started");
645
646 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
647 /* Atomically prepare the queue futex */
648 futex_nto1_prepare(&relay_cmd_queue.futex);
649
650 do {
651 /* Dequeue commands */
652 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
653 if (node == NULL) {
654 DBG("Woken up but nothing in the relay command queue");
655 /* Continue thread execution */
656 break;
657 }
658
659 relay_cmd = caa_container_of(node, struct relay_command, node);
660 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
661
662 /*
663 * Inform worker thread of the new request. This
664 * call is blocking so we can be assured that the data will be read
665 * at some point in time or wait to the end of the world :)
666 */
667 do {
668 ret = write(relay_cmd_pipe[1], relay_cmd,
669 sizeof(struct relay_command));
670 } while (ret < 0 && errno == EINTR);
671 free(relay_cmd);
672 if (ret < 0 || ret != sizeof(struct relay_command)) {
673 PERROR("write cmd pipe");
674 goto error;
675 }
676 } while (node != NULL);
677
678 /* Futex wait on queue. Blocking call on futex() */
679 futex_nto1_wait(&relay_cmd_queue.futex);
680 }
681
682 error:
683 DBG("Dispatch thread dying");
684 stop_threads();
685 return NULL;
686 }
687
688 /*
689 * Return the realpath(3) of the path even if the last directory token does not
690 * exist. For example, with /tmp/test1/test2, if test2/ does not exist but the
691 * /tmp/test1 does, the real path is returned. In normal time, realpath(3)
692 * fails if the end point directory does not exist.
693 */
694 static
695 char *expand_full_path(const char *path)
696 {
697 const char *end_path = path;
698 char *next, *cut_path, *expanded_path, *respath;
699
700 /* Find last token delimited by '/' */
701 while ((next = strpbrk(end_path + 1, "/"))) {
702 end_path = next;
703 }
704
705 /* Cut last token from original path */
706 cut_path = strndup(path, end_path - path);
707
708 expanded_path = malloc(PATH_MAX);
709 if (expanded_path == NULL) {
710 respath = NULL;
711 goto end;
712 }
713
714 respath = realpath(cut_path, expanded_path);
715 if (respath == NULL) {
716 switch (errno) {
717 case ENOENT:
718 ERR("%s: No such file or directory", cut_path);
719 break;
720 default:
721 PERROR("realpath");
722 break;
723 }
724 free(expanded_path);
725 } else {
726 /* Add end part to expanded path */
727 strcat(respath, end_path);
728 }
729 end:
730 free(cut_path);
731 return respath;
732 }
733
734
735 /*
736 * config_get_default_path
737 *
738 * Returns the HOME directory path. Caller MUST NOT free(3) the return pointer.
739 */
740 static
741 char *config_get_default_path(void)
742 {
743 return getenv("HOME");
744 }
745
746 /*
747 * Create recursively directory using the FULL path.
748 */
749 static
750 int mkdir_recursive(char *path, mode_t mode)
751 {
752 char *p, tmp[PATH_MAX];
753 struct stat statbuf;
754 size_t len;
755 int ret;
756
757 ret = snprintf(tmp, sizeof(tmp), "%s", path);
758 if (ret < 0) {
759 PERROR("snprintf mkdir");
760 goto error;
761 }
762
763 len = ret;
764 if (tmp[len - 1] == '/') {
765 tmp[len - 1] = 0;
766 }
767
768 for (p = tmp + 1; *p; p++) {
769 if (*p == '/') {
770 *p = 0;
771 if (tmp[strlen(tmp) - 1] == '.' &&
772 tmp[strlen(tmp) - 2] == '.' &&
773 tmp[strlen(tmp) - 3] == '/') {
774 ERR("Using '/../' is not permitted in the trace path (%s)",
775 tmp);
776 ret = -1;
777 goto error;
778 }
779 ret = stat(tmp, &statbuf);
780 if (ret < 0) {
781 ret = mkdir(tmp, mode);
782 if (ret < 0) {
783 if (errno != EEXIST) {
784 PERROR("mkdir recursive");
785 ret = -errno;
786 goto error;
787 }
788 }
789 }
790 *p = '/';
791 }
792 }
793
794 ret = mkdir(tmp, mode);
795 if (ret < 0) {
796 if (errno != EEXIST) {
797 PERROR("mkdir recursive last piece");
798 ret = -errno;
799 } else {
800 ret = 0;
801 }
802 }
803
804 error:
805 return ret;
806 }
807
808 static
809 char *create_output_path_auto(char *path_name)
810 {
811 int ret;
812 char *traces_path = NULL;
813 char *alloc_path = NULL;
814 char *default_path;
815
816 default_path = config_get_default_path();
817 if (default_path == NULL) {
818 ERR("Home path not found.\n \
819 Please specify an output path using -o, --output PATH");
820 goto exit;
821 }
822 alloc_path = strdup(default_path);
823 if (alloc_path == NULL) {
824 PERROR("Path allocation");
825 goto exit;
826 }
827 ret = asprintf(&traces_path, "%s/" DEFAULT_TRACE_DIR_NAME
828 "/%s", alloc_path, path_name);
829 if (ret < 0) {
830 PERROR("asprintf trace dir name");
831 goto exit;
832 }
833 exit:
834 free(alloc_path);
835 return traces_path;
836 }
837
838 static
839 char *create_output_path_noauto(char *path_name)
840 {
841 int ret;
842 char *traces_path = NULL;
843 char *full_path;
844
845 full_path = expand_full_path(opt_output_path);
846 ret = asprintf(&traces_path, "%s/%s", full_path, path_name);
847 if (ret < 0) {
848 PERROR("asprintf trace dir name");
849 goto exit;
850 }
851 exit:
852 free(full_path);
853 return traces_path;
854 }
855
856 /*
857 * create_output_path: create the output trace directory
858 */
859 static
860 char *create_output_path(char *path_name)
861 {
862 if (opt_output_path == NULL) {
863 return create_output_path_auto(path_name);
864 } else {
865 return create_output_path_noauto(path_name);
866 }
867 }
868
869 static
870 void deferred_free_stream(struct rcu_head *head)
871 {
872 struct relay_stream *stream =
873 caa_container_of(head, struct relay_stream, rcu_node);
874 free(stream);
875 }
876
877 /*
878 * relay_delete_session: Free all memory associated with a session and
879 * close all the FDs
880 */
881 static
882 void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht)
883 {
884 struct lttng_ht_iter iter;
885 struct lttng_ht_node_ulong *node;
886 struct relay_stream *stream;
887 int ret;
888
889 if (!cmd->session) {
890 return;
891 }
892
893 DBG("Relay deleting session %" PRIu64, cmd->session->id);
894
895 rcu_read_lock();
896 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, node, node) {
897 node = lttng_ht_iter_get_node_ulong(&iter);
898 if (node) {
899 stream = caa_container_of(node,
900 struct relay_stream, stream_n);
901 if (stream->session == cmd->session) {
902 ret = close(stream->fd);
903 if (ret < 0) {
904 PERROR("close stream fd on delete session");
905 }
906 ret = lttng_ht_del(streams_ht, &iter);
907 assert(!ret);
908 call_rcu(&stream->rcu_node,
909 deferred_free_stream);
910 }
911 }
912 }
913 rcu_read_unlock();
914
915 free(cmd->session);
916 }
917
918 /*
919 * Handle the RELAYD_CREATE_SESSION command.
920 *
921 * On success, send back the session id or else return a negative value.
922 */
923 static
924 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
925 struct relay_command *cmd)
926 {
927 int ret = 0, send_ret;
928 struct relay_session *session;
929 struct lttcomm_relayd_status_session reply;
930
931 assert(recv_hdr);
932 assert(cmd);
933
934 memset(&reply, 0, sizeof(reply));
935
936 session = zmalloc(sizeof(struct relay_session));
937 if (session == NULL) {
938 PERROR("relay session zmalloc");
939 ret = -1;
940 goto error;
941 }
942
943 session->id = ++last_relay_session_id;
944 session->sock = cmd->sock;
945 cmd->session = session;
946
947 reply.session_id = htobe64(session->id);
948
949 DBG("Created session %" PRIu64, session->id);
950
951 error:
952 if (ret < 0) {
953 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
954 } else {
955 reply.ret_code = htobe32(LTTNG_OK);
956 }
957
958 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
959 if (send_ret < 0) {
960 ERR("Relayd sending session id");
961 }
962
963 return ret;
964 }
965
966 /*
967 * relay_add_stream: allocate a new stream for a session
968 */
969 static
970 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
971 struct relay_command *cmd, struct lttng_ht *streams_ht)
972 {
973 struct relay_session *session = cmd->session;
974 struct lttcomm_relayd_add_stream stream_info;
975 struct relay_stream *stream = NULL;
976 struct lttcomm_relayd_status_stream reply;
977 char *path = NULL, *root_path = NULL;
978 int ret, send_ret;
979
980 if (!session || cmd->version_check_done == 0) {
981 ERR("Trying to add a stream before version check");
982 ret = -1;
983 goto end_no_session;
984 }
985
986 /* FIXME : use data_size for something ? */
987 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
988 sizeof(struct lttcomm_relayd_add_stream), 0);
989 if (ret < sizeof(struct lttcomm_relayd_add_stream)) {
990 if (ret == 0) {
991 /* Orderly shutdown. Not necessary to print an error. */
992 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
993 } else {
994 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
995 }
996 ret = -1;
997 goto end_no_session;
998 }
999 stream = zmalloc(sizeof(struct relay_stream));
1000 if (stream == NULL) {
1001 PERROR("relay stream zmalloc");
1002 ret = -1;
1003 goto end_no_session;
1004 }
1005
1006 rcu_read_lock();
1007 stream->stream_handle = ++last_relay_stream_id;
1008 stream->prev_seq = -1ULL;
1009 stream->session = session;
1010
1011 root_path = create_output_path(stream_info.pathname);
1012 if (!root_path) {
1013 ret = -1;
1014 goto end;
1015 }
1016 ret = mkdir_recursive(root_path, S_IRWXU | S_IRWXG);
1017 if (ret < 0) {
1018 ERR("relay creating output directory");
1019 goto end;
1020 }
1021
1022 ret = asprintf(&path, "%s/%s", root_path, stream_info.channel_name);
1023 if (ret < 0) {
1024 PERROR("asprintf stream path");
1025 goto end;
1026 }
1027
1028 ret = open(path, O_WRONLY|O_CREAT|O_TRUNC, S_IRWXU|S_IRWXG|S_IRWXO);
1029 if (ret < 0) {
1030 PERROR("Relay creating trace file");
1031 goto end;
1032 }
1033
1034 stream->fd = ret;
1035 DBG("Tracefile %s created", path);
1036
1037 lttng_ht_node_init_ulong(&stream->stream_n,
1038 (unsigned long) stream->stream_handle);
1039 lttng_ht_add_unique_ulong(streams_ht,
1040 &stream->stream_n);
1041
1042 DBG("Relay new stream added %s", stream_info.channel_name);
1043
1044 end:
1045 free(path);
1046 free(root_path);
1047 /* send the session id to the client or a negative return code on error */
1048 if (ret < 0) {
1049 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1050 } else {
1051 reply.ret_code = htobe32(LTTNG_OK);
1052 }
1053 reply.handle = htobe64(stream->stream_handle);
1054 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1055 sizeof(struct lttcomm_relayd_status_stream), 0);
1056 if (send_ret < 0) {
1057 ERR("Relay sending stream id");
1058 }
1059 rcu_read_unlock();
1060
1061 end_no_session:
1062 return ret;
1063 }
1064
1065 /*
1066 * relay_close_stream: close a specific stream
1067 */
1068 static
1069 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1070 struct relay_command *cmd, struct lttng_ht *streams_ht)
1071 {
1072 struct relay_session *session = cmd->session;
1073 struct lttcomm_relayd_close_stream stream_info;
1074 struct lttcomm_relayd_generic_reply reply;
1075 struct relay_stream *stream;
1076 int ret, send_ret;
1077 struct lttng_ht_node_ulong *node;
1078 struct lttng_ht_iter iter;
1079
1080 DBG("Close stream received");
1081
1082 if (!session || cmd->version_check_done == 0) {
1083 ERR("Trying to close a stream before version check");
1084 ret = -1;
1085 goto end_no_session;
1086 }
1087
1088 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1089 sizeof(struct lttcomm_relayd_close_stream), 0);
1090 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1091 if (ret == 0) {
1092 /* Orderly shutdown. Not necessary to print an error. */
1093 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1094 } else {
1095 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1096 }
1097 ret = -1;
1098 goto end_no_session;
1099 }
1100
1101 rcu_read_lock();
1102 lttng_ht_lookup(streams_ht,
1103 (void *)((unsigned long) be64toh(stream_info.stream_id)),
1104 &iter);
1105 node = lttng_ht_iter_get_node_ulong(&iter);
1106 if (node == NULL) {
1107 DBG("Relay stream %" PRIu64 " not found", be64toh(stream_info.stream_id));
1108 ret = -1;
1109 goto end_unlock;
1110 }
1111
1112 stream = caa_container_of(node, struct relay_stream, stream_n);
1113 if (!stream) {
1114 ret = -1;
1115 goto end_unlock;
1116 }
1117
1118 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1119 stream->close_flag = 1;
1120
1121 if (close_stream_check(stream)) {
1122 int delret;
1123
1124 delret = close(stream->fd);
1125 if (delret < 0) {
1126 PERROR("close stream");
1127 }
1128 delret = lttng_ht_del(streams_ht, &iter);
1129 assert(!delret);
1130 call_rcu(&stream->rcu_node,
1131 deferred_free_stream);
1132 DBG("Closed tracefile %d from close stream", stream->fd);
1133 }
1134
1135 end_unlock:
1136 rcu_read_unlock();
1137
1138 if (ret < 0) {
1139 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1140 } else {
1141 reply.ret_code = htobe32(LTTNG_OK);
1142 }
1143 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1144 sizeof(struct lttcomm_relayd_generic_reply), 0);
1145 if (send_ret < 0) {
1146 ERR("Relay sending stream id");
1147 }
1148
1149 end_no_session:
1150 return ret;
1151 }
1152
1153 /*
1154 * relay_unknown_command: send -1 if received unknown command
1155 */
1156 static
1157 void relay_unknown_command(struct relay_command *cmd)
1158 {
1159 struct lttcomm_relayd_generic_reply reply;
1160 int ret;
1161
1162 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1163 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1164 sizeof(struct lttcomm_relayd_generic_reply), 0);
1165 if (ret < 0) {
1166 ERR("Relay sending unknown command");
1167 }
1168 }
1169
1170 /*
1171 * relay_start: send an acknowledgment to the client to tell if we are
1172 * ready to receive data. We are ready if a session is established.
1173 */
1174 static
1175 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1176 struct relay_command *cmd)
1177 {
1178 int ret = htobe32(LTTNG_OK);
1179 struct lttcomm_relayd_generic_reply reply;
1180 struct relay_session *session = cmd->session;
1181
1182 if (!session) {
1183 DBG("Trying to start the streaming without a session established");
1184 ret = htobe32(LTTNG_ERR_UNK);
1185 }
1186
1187 reply.ret_code = ret;
1188 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1189 sizeof(struct lttcomm_relayd_generic_reply), 0);
1190 if (ret < 0) {
1191 ERR("Relay sending start ack");
1192 }
1193
1194 return ret;
1195 }
1196
1197 /*
1198 * Get stream from stream id.
1199 * Need to be called with RCU read-side lock held.
1200 */
1201 static
1202 struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id,
1203 struct lttng_ht *streams_ht)
1204 {
1205 struct lttng_ht_node_ulong *node;
1206 struct lttng_ht_iter iter;
1207 struct relay_stream *ret;
1208
1209 lttng_ht_lookup(streams_ht,
1210 (void *)((unsigned long) stream_id),
1211 &iter);
1212 node = lttng_ht_iter_get_node_ulong(&iter);
1213 if (node == NULL) {
1214 DBG("Relay stream %" PRIu64 " not found", stream_id);
1215 ret = NULL;
1216 goto end;
1217 }
1218
1219 ret = caa_container_of(node, struct relay_stream, stream_n);
1220
1221 end:
1222 return ret;
1223 }
1224
1225 /*
1226 * Append padding to the file pointed by the file descriptor fd.
1227 */
1228 static int write_padding_to_file(int fd, uint32_t size)
1229 {
1230 int ret = 0;
1231 char *zeros;
1232
1233 if (size == 0) {
1234 goto end;
1235 }
1236
1237 zeros = zmalloc(size);
1238 if (zeros == NULL) {
1239 PERROR("zmalloc zeros for padding");
1240 ret = -1;
1241 goto end;
1242 }
1243
1244 do {
1245 ret = write(fd, zeros, size);
1246 } while (ret < 0 && errno == EINTR);
1247 if (ret < 0 || ret != size) {
1248 PERROR("write padding to file");
1249 }
1250
1251 free(zeros);
1252
1253 end:
1254 return ret;
1255 }
1256
1257 /*
1258 * relay_recv_metadata: receive the metada for the session.
1259 */
1260 static
1261 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1262 struct relay_command *cmd, struct lttng_ht *streams_ht)
1263 {
1264 int ret = htobe32(LTTNG_OK);
1265 struct relay_session *session = cmd->session;
1266 struct lttcomm_relayd_metadata_payload *metadata_struct;
1267 struct relay_stream *metadata_stream;
1268 uint64_t data_size, payload_size;
1269
1270 if (!session) {
1271 ERR("Metadata sent before version check");
1272 ret = -1;
1273 goto end;
1274 }
1275
1276 data_size = payload_size = be64toh(recv_hdr->data_size);
1277 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1278 ERR("Incorrect data size");
1279 ret = -1;
1280 goto end;
1281 }
1282 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1283
1284 if (data_buffer_size < data_size) {
1285 /* In case the realloc fails, we can free the memory */
1286 char *tmp_data_ptr;
1287
1288 tmp_data_ptr = realloc(data_buffer, data_size);
1289 if (!tmp_data_ptr) {
1290 ERR("Allocating data buffer");
1291 free(data_buffer);
1292 ret = -1;
1293 goto end;
1294 }
1295 data_buffer = tmp_data_ptr;
1296 data_buffer_size = data_size;
1297 }
1298 memset(data_buffer, 0, data_size);
1299 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1300 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1301 if (ret < 0 || ret != data_size) {
1302 if (ret == 0) {
1303 /* Orderly shutdown. Not necessary to print an error. */
1304 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1305 } else {
1306 ERR("Relay didn't receive the whole metadata");
1307 }
1308 ret = -1;
1309 goto end;
1310 }
1311 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1312
1313 rcu_read_lock();
1314 metadata_stream = relay_stream_from_stream_id(
1315 be64toh(metadata_struct->stream_id), streams_ht);
1316 if (!metadata_stream) {
1317 ret = -1;
1318 goto end_unlock;
1319 }
1320
1321 do {
1322 ret = write(metadata_stream->fd, metadata_struct->payload,
1323 payload_size);
1324 } while (ret < 0 && errno == EINTR);
1325 if (ret < 0 || ret != payload_size) {
1326 ERR("Relay error writing metadata on file");
1327 ret = -1;
1328 goto end_unlock;
1329 }
1330
1331 ret = write_padding_to_file(metadata_stream->fd,
1332 be32toh(metadata_struct->padding_size));
1333 if (ret < 0) {
1334 goto end_unlock;
1335 }
1336
1337 DBG2("Relay metadata written");
1338
1339 end_unlock:
1340 rcu_read_unlock();
1341 end:
1342 return ret;
1343 }
1344
1345 /*
1346 * relay_send_version: send relayd version number
1347 */
1348 static
1349 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1350 struct relay_command *cmd)
1351 {
1352 int ret;
1353 struct lttcomm_relayd_version reply, msg;
1354
1355 assert(cmd);
1356
1357 cmd->version_check_done = 1;
1358
1359 /* Get version from the other side. */
1360 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1361 if (ret < 0 || ret != sizeof(msg)) {
1362 if (ret == 0) {
1363 /* Orderly shutdown. Not necessary to print an error. */
1364 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1365 } else {
1366 ERR("Relay failed to receive the version values.");
1367 }
1368 ret = -1;
1369 goto end;
1370 }
1371
1372 /*
1373 * For now, we just ignore the received version but after 2.1 stable
1374 * release, a check must be done to see if we either adapt to the other
1375 * side version (which MUST be lower than us) or keep the latest data
1376 * structure considering that the other side will adapt.
1377 */
1378
1379 ret = sscanf(VERSION, "%10u.%10u", &reply.major, &reply.minor);
1380 if (ret < 2) {
1381 ERR("Error in scanning version");
1382 ret = -1;
1383 goto end;
1384 }
1385 reply.major = htobe32(reply.major);
1386 reply.minor = htobe32(reply.minor);
1387 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1388 sizeof(struct lttcomm_relayd_version), 0);
1389 if (ret < 0) {
1390 ERR("Relay sending version");
1391 }
1392 DBG("Version check done (%u.%u)", be32toh(reply.major),
1393 be32toh(reply.minor));
1394
1395 end:
1396 return ret;
1397 }
1398
1399 /*
1400 * Check for data pending for a given stream id from the session daemon.
1401 */
1402 static
1403 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1404 struct relay_command *cmd, struct lttng_ht *streams_ht)
1405 {
1406 struct relay_session *session = cmd->session;
1407 struct lttcomm_relayd_data_pending msg;
1408 struct lttcomm_relayd_generic_reply reply;
1409 struct relay_stream *stream;
1410 int ret;
1411 struct lttng_ht_node_ulong *node;
1412 struct lttng_ht_iter iter;
1413 uint64_t last_net_seq_num, stream_id;
1414
1415 DBG("Data pending command received");
1416
1417 if (!session || cmd->version_check_done == 0) {
1418 ERR("Trying to check for data before version check");
1419 ret = -1;
1420 goto end_no_session;
1421 }
1422
1423 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1424 if (ret < sizeof(msg)) {
1425 if (ret == 0) {
1426 /* Orderly shutdown. Not necessary to print an error. */
1427 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1428 } else {
1429 ERR("Relay didn't receive valid data_pending struct size : %d",
1430 ret);
1431 }
1432 ret = -1;
1433 goto end_no_session;
1434 }
1435
1436 stream_id = be64toh(msg.stream_id);
1437 last_net_seq_num = be64toh(msg.last_net_seq_num);
1438
1439 rcu_read_lock();
1440 lttng_ht_lookup(streams_ht, (void *)((unsigned long) stream_id), &iter);
1441 node = lttng_ht_iter_get_node_ulong(&iter);
1442 if (node == NULL) {
1443 DBG("Relay stream %" PRIu64 " not found", stream_id);
1444 ret = -1;
1445 goto end_unlock;
1446 }
1447
1448 stream = caa_container_of(node, struct relay_stream, stream_n);
1449 assert(stream);
1450
1451 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1452 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1453 last_net_seq_num);
1454
1455 /* Avoid wrapping issue */
1456 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1457 /* Data has in fact been written and is NOT pending */
1458 ret = 0;
1459 } else {
1460 /* Data still being streamed thus pending */
1461 ret = 1;
1462 }
1463
1464 /* Pending check is now done. */
1465 stream->data_pending_check_done = 1;
1466
1467 end_unlock:
1468 rcu_read_unlock();
1469
1470 reply.ret_code = htobe32(ret);
1471 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1472 if (ret < 0) {
1473 ERR("Relay data pending ret code failed");
1474 }
1475
1476 end_no_session:
1477 return ret;
1478 }
1479
1480 /*
1481 * Wait for the control socket to reach a quiescent state.
1482 *
1483 * Note that for now, when receiving this command from the session daemon, this
1484 * means that every subsequent commands or data received on the control socket
1485 * has been handled. So, this is why we simply return OK here.
1486 */
1487 static
1488 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1489 struct relay_command *cmd, struct lttng_ht *streams_ht)
1490 {
1491 int ret;
1492 uint64_t stream_id;
1493 struct relay_stream *stream;
1494 struct lttng_ht_iter iter;
1495 struct lttcomm_relayd_quiescent_control msg;
1496 struct lttcomm_relayd_generic_reply reply;
1497
1498 DBG("Checking quiescent state on control socket");
1499
1500 if (!cmd->session || cmd->version_check_done == 0) {
1501 ERR("Trying to check for data before version check");
1502 ret = -1;
1503 goto end_no_session;
1504 }
1505
1506 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1507 if (ret < sizeof(msg)) {
1508 if (ret == 0) {
1509 /* Orderly shutdown. Not necessary to print an error. */
1510 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1511 } else {
1512 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1513 ret);
1514 }
1515 ret = -1;
1516 goto end_no_session;
1517 }
1518
1519 stream_id = be64toh(msg.stream_id);
1520
1521 rcu_read_lock();
1522 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) {
1523 if (stream->stream_handle == stream_id) {
1524 stream->data_pending_check_done = 1;
1525 DBG("Relay quiescent control pending flag set to %" PRIu64,
1526 stream_id);
1527 break;
1528 }
1529 }
1530 rcu_read_unlock();
1531
1532 reply.ret_code = htobe32(LTTNG_OK);
1533 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1534 if (ret < 0) {
1535 ERR("Relay data quiescent control ret code failed");
1536 }
1537
1538 end_no_session:
1539 return ret;
1540 }
1541
1542 /*
1543 * Initialize a data pending command. This means that a client is about to ask
1544 * for data pending for each stream he/she holds. Simply iterate over all
1545 * streams of a session and set the data_pending_check_done flag.
1546 *
1547 * This command returns to the client a LTTNG_OK code.
1548 */
1549 static
1550 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1551 struct relay_command *cmd, struct lttng_ht *streams_ht)
1552 {
1553 int ret;
1554 struct lttng_ht_iter iter;
1555 struct lttcomm_relayd_begin_data_pending msg;
1556 struct lttcomm_relayd_generic_reply reply;
1557 struct relay_stream *stream;
1558 uint64_t session_id;
1559
1560 assert(recv_hdr);
1561 assert(cmd);
1562 assert(streams_ht);
1563
1564 DBG("Init streams for data pending");
1565
1566 if (!cmd->session || cmd->version_check_done == 0) {
1567 ERR("Trying to check for data before version check");
1568 ret = -1;
1569 goto end_no_session;
1570 }
1571
1572 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1573 if (ret < sizeof(msg)) {
1574 if (ret == 0) {
1575 /* Orderly shutdown. Not necessary to print an error. */
1576 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1577 } else {
1578 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1579 ret);
1580 }
1581 ret = -1;
1582 goto end_no_session;
1583 }
1584
1585 session_id = be64toh(msg.session_id);
1586
1587 /*
1588 * Iterate over all streams to set the begin data pending flag. For now, the
1589 * streams are indexed by stream handle so we have to iterate over all
1590 * streams to find the one associated with the right session_id.
1591 */
1592 rcu_read_lock();
1593 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) {
1594 if (stream->session->id == session_id) {
1595 stream->data_pending_check_done = 0;
1596 DBG("Set begin data pending flag to stream %" PRIu64,
1597 stream->stream_handle);
1598 }
1599 }
1600 rcu_read_unlock();
1601
1602 /* All good, send back reply. */
1603 reply.ret_code = htobe32(LTTNG_OK);
1604
1605 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1606 if (ret < 0) {
1607 ERR("Relay begin data pending send reply failed");
1608 }
1609
1610 end_no_session:
1611 return ret;
1612 }
1613
1614 /*
1615 * End data pending command. This will check, for a given session id, if each
1616 * stream associated with it has its data_pending_check_done flag set. If not,
1617 * this means that the client lost track of the stream but the data is still
1618 * being streamed on our side. In this case, we inform the client that data is
1619 * inflight.
1620 *
1621 * Return to the client if there is data in flight or not with a ret_code.
1622 */
1623 static
1624 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1625 struct relay_command *cmd, struct lttng_ht *streams_ht)
1626 {
1627 int ret;
1628 struct lttng_ht_iter iter;
1629 struct lttcomm_relayd_end_data_pending msg;
1630 struct lttcomm_relayd_generic_reply reply;
1631 struct relay_stream *stream;
1632 uint64_t session_id;
1633 uint32_t is_data_inflight = 0;
1634
1635 assert(recv_hdr);
1636 assert(cmd);
1637 assert(streams_ht);
1638
1639 DBG("End data pending command");
1640
1641 if (!cmd->session || cmd->version_check_done == 0) {
1642 ERR("Trying to check for data before version check");
1643 ret = -1;
1644 goto end_no_session;
1645 }
1646
1647 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1648 if (ret < sizeof(msg)) {
1649 if (ret == 0) {
1650 /* Orderly shutdown. Not necessary to print an error. */
1651 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1652 } else {
1653 ERR("Relay didn't receive valid end data_pending struct size: %d",
1654 ret);
1655 }
1656 ret = -1;
1657 goto end_no_session;
1658 }
1659
1660 session_id = be64toh(msg.session_id);
1661
1662 /* Iterate over all streams to see if the begin data pending flag is set. */
1663 rcu_read_lock();
1664 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) {
1665 if (stream->session->id == session_id &&
1666 !stream->data_pending_check_done) {
1667 is_data_inflight = 1;
1668 DBG("Data is still in flight for stream %" PRIu64,
1669 stream->stream_handle);
1670 break;
1671 }
1672 }
1673 rcu_read_unlock();
1674
1675 /* All good, send back reply. */
1676 reply.ret_code = htobe32(is_data_inflight);
1677
1678 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1679 if (ret < 0) {
1680 ERR("Relay end data pending send reply failed");
1681 }
1682
1683 end_no_session:
1684 return ret;
1685 }
1686
1687 /*
1688 * relay_process_control: Process the commands received on the control socket
1689 */
1690 static
1691 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1692 struct relay_command *cmd, struct lttng_ht *streams_ht)
1693 {
1694 int ret = 0;
1695
1696 switch (be32toh(recv_hdr->cmd)) {
1697 case RELAYD_CREATE_SESSION:
1698 ret = relay_create_session(recv_hdr, cmd);
1699 break;
1700 case RELAYD_ADD_STREAM:
1701 ret = relay_add_stream(recv_hdr, cmd, streams_ht);
1702 break;
1703 case RELAYD_START_DATA:
1704 ret = relay_start(recv_hdr, cmd);
1705 break;
1706 case RELAYD_SEND_METADATA:
1707 ret = relay_recv_metadata(recv_hdr, cmd, streams_ht);
1708 break;
1709 case RELAYD_VERSION:
1710 ret = relay_send_version(recv_hdr, cmd);
1711 break;
1712 case RELAYD_CLOSE_STREAM:
1713 ret = relay_close_stream(recv_hdr, cmd, streams_ht);
1714 break;
1715 case RELAYD_DATA_PENDING:
1716 ret = relay_data_pending(recv_hdr, cmd, streams_ht);
1717 break;
1718 case RELAYD_QUIESCENT_CONTROL:
1719 ret = relay_quiescent_control(recv_hdr, cmd, streams_ht);
1720 break;
1721 case RELAYD_BEGIN_DATA_PENDING:
1722 ret = relay_begin_data_pending(recv_hdr, cmd, streams_ht);
1723 break;
1724 case RELAYD_END_DATA_PENDING:
1725 ret = relay_end_data_pending(recv_hdr, cmd, streams_ht);
1726 break;
1727 case RELAYD_UPDATE_SYNC_INFO:
1728 default:
1729 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1730 relay_unknown_command(cmd);
1731 ret = -1;
1732 goto end;
1733 }
1734
1735 end:
1736 return ret;
1737 }
1738
1739 /*
1740 * relay_process_data: Process the data received on the data socket
1741 */
1742 static
1743 int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht)
1744 {
1745 int ret = 0;
1746 struct relay_stream *stream;
1747 struct lttcomm_relayd_data_hdr data_hdr;
1748 uint64_t stream_id;
1749 uint64_t net_seq_num;
1750 uint32_t data_size;
1751
1752 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1753 sizeof(struct lttcomm_relayd_data_hdr), 0);
1754 if (ret <= 0) {
1755 if (ret == 0) {
1756 /* Orderly shutdown. Not necessary to print an error. */
1757 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1758 } else {
1759 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
1760 }
1761 ret = -1;
1762 goto end;
1763 }
1764
1765 stream_id = be64toh(data_hdr.stream_id);
1766
1767 rcu_read_lock();
1768 stream = relay_stream_from_stream_id(stream_id, streams_ht);
1769 if (!stream) {
1770 ret = -1;
1771 goto end_unlock;
1772 }
1773
1774 data_size = be32toh(data_hdr.data_size);
1775 if (data_buffer_size < data_size) {
1776 char *tmp_data_ptr;
1777
1778 tmp_data_ptr = realloc(data_buffer, data_size);
1779 if (!tmp_data_ptr) {
1780 ERR("Allocating data buffer");
1781 free(data_buffer);
1782 ret = -1;
1783 goto end_unlock;
1784 }
1785 data_buffer = tmp_data_ptr;
1786 data_buffer_size = data_size;
1787 }
1788 memset(data_buffer, 0, data_size);
1789
1790 net_seq_num = be64toh(data_hdr.net_seq_num);
1791
1792 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
1793 data_size, stream_id, net_seq_num);
1794 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1795 if (ret <= 0) {
1796 if (ret == 0) {
1797 /* Orderly shutdown. Not necessary to print an error. */
1798 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1799 }
1800 ret = -1;
1801 goto end_unlock;
1802 }
1803
1804 do {
1805 ret = write(stream->fd, data_buffer, data_size);
1806 } while (ret < 0 && errno == EINTR);
1807 if (ret < 0 || ret != data_size) {
1808 ERR("Relay error writing data to file");
1809 ret = -1;
1810 goto end_unlock;
1811 }
1812
1813 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
1814 ret, stream->stream_handle);
1815
1816 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
1817 if (ret < 0) {
1818 goto end_unlock;
1819 }
1820
1821 stream->prev_seq = net_seq_num;
1822
1823 /* Check if we need to close the FD */
1824 if (close_stream_check(stream)) {
1825 int cret;
1826 struct lttng_ht_iter iter;
1827
1828 cret = close(stream->fd);
1829 if (cret < 0) {
1830 PERROR("close stream process data");
1831 }
1832 iter.iter.node = &stream->stream_n.node;
1833 ret = lttng_ht_del(streams_ht, &iter);
1834 assert(!ret);
1835 call_rcu(&stream->rcu_node,
1836 deferred_free_stream);
1837 DBG("Closed tracefile %d after recv data", stream->fd);
1838 }
1839
1840 end_unlock:
1841 rcu_read_unlock();
1842 end:
1843 return ret;
1844 }
1845
1846 static
1847 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
1848 {
1849 int ret;
1850
1851 lttng_poll_del(events, pollfd);
1852
1853 ret = close(pollfd);
1854 if (ret < 0) {
1855 ERR("Closing pollfd %d", pollfd);
1856 }
1857 }
1858
1859 static
1860 int relay_add_connection(int fd, struct lttng_poll_event *events,
1861 struct lttng_ht *relay_connections_ht)
1862 {
1863 struct relay_command *relay_connection;
1864 int ret;
1865
1866 relay_connection = zmalloc(sizeof(struct relay_command));
1867 if (relay_connection == NULL) {
1868 PERROR("Relay command zmalloc");
1869 goto error;
1870 }
1871 do {
1872 ret = read(fd, relay_connection, sizeof(struct relay_command));
1873 } while (ret < 0 && errno == EINTR);
1874 if (ret < 0 || ret < sizeof(struct relay_command)) {
1875 PERROR("read relay cmd pipe");
1876 goto error_read;
1877 }
1878
1879 lttng_ht_node_init_ulong(&relay_connection->sock_n,
1880 (unsigned long) relay_connection->sock->fd);
1881 rcu_read_lock();
1882 lttng_ht_add_unique_ulong(relay_connections_ht,
1883 &relay_connection->sock_n);
1884 rcu_read_unlock();
1885 return lttng_poll_add(events,
1886 relay_connection->sock->fd,
1887 LPOLLIN | LPOLLRDHUP);
1888
1889 error_read:
1890 free(relay_connection);
1891 error:
1892 return -1;
1893 }
1894
1895 static
1896 void deferred_free_connection(struct rcu_head *head)
1897 {
1898 struct relay_command *relay_connection =
1899 caa_container_of(head, struct relay_command, rcu_node);
1900
1901 lttcomm_destroy_sock(relay_connection->sock);
1902 free(relay_connection);
1903 }
1904
1905 static
1906 void relay_del_connection(struct lttng_ht *relay_connections_ht,
1907 struct lttng_ht *streams_ht, struct lttng_ht_iter *iter,
1908 struct relay_command *relay_connection)
1909 {
1910 int ret;
1911
1912 ret = lttng_ht_del(relay_connections_ht, iter);
1913 assert(!ret);
1914 if (relay_connection->type == RELAY_CONTROL) {
1915 relay_delete_session(relay_connection, streams_ht);
1916 }
1917
1918 call_rcu(&relay_connection->rcu_node,
1919 deferred_free_connection);
1920 }
1921
1922 /*
1923 * This thread does the actual work
1924 */
1925 static
1926 void *relay_thread_worker(void *data)
1927 {
1928 int ret, err = -1, last_seen_data_fd = -1;
1929 uint32_t nb_fd;
1930 struct relay_command *relay_connection;
1931 struct lttng_poll_event events;
1932 struct lttng_ht *relay_connections_ht;
1933 struct lttng_ht_node_ulong *node;
1934 struct lttng_ht_iter iter;
1935 struct lttng_ht *streams_ht;
1936 struct lttcomm_relayd_hdr recv_hdr;
1937
1938 DBG("[thread] Relay worker started");
1939
1940 rcu_register_thread();
1941
1942 /* table of connections indexed on socket */
1943 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1944 if (!relay_connections_ht) {
1945 goto relay_connections_ht_error;
1946 }
1947
1948 /* tables of streams indexed by stream ID */
1949 streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1950 if (!streams_ht) {
1951 goto streams_ht_error;
1952 }
1953
1954 ret = create_thread_poll_set(&events, 2);
1955 if (ret < 0) {
1956 goto error_poll_create;
1957 }
1958
1959 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1960 if (ret < 0) {
1961 goto error;
1962 }
1963
1964 restart:
1965 while (1) {
1966 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
1967
1968 /* Infinite blocking call, waiting for transmission */
1969 DBG3("Relayd worker thread polling...");
1970 ret = lttng_poll_wait(&events, -1);
1971 if (ret < 0) {
1972 /*
1973 * Restart interrupted system call.
1974 */
1975 if (errno == EINTR) {
1976 goto restart;
1977 }
1978 goto error;
1979 }
1980
1981 nb_fd = ret;
1982
1983 /*
1984 * Process control. The control connection is prioritised so we don't
1985 * starve it with high throughout put tracing data on the data
1986 * connection.
1987 */
1988 for (i = 0; i < nb_fd; i++) {
1989 /* Fetch once the poll data */
1990 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1991 int pollfd = LTTNG_POLL_GETFD(&events, i);
1992
1993 /* Thread quit pipe has been closed. Killing thread. */
1994 ret = check_thread_quit_pipe(pollfd, revents);
1995 if (ret) {
1996 err = 0;
1997 goto exit;
1998 }
1999
2000 /* Inspect the relay cmd pipe for new connection */
2001 if (pollfd == relay_cmd_pipe[0]) {
2002 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2003 ERR("Relay pipe error");
2004 goto error;
2005 } else if (revents & LPOLLIN) {
2006 DBG("Relay command received");
2007 ret = relay_add_connection(relay_cmd_pipe[0],
2008 &events, relay_connections_ht);
2009 if (ret < 0) {
2010 goto error;
2011 }
2012 }
2013 } else if (revents) {
2014 rcu_read_lock();
2015 lttng_ht_lookup(relay_connections_ht,
2016 (void *)((unsigned long) pollfd),
2017 &iter);
2018 node = lttng_ht_iter_get_node_ulong(&iter);
2019 if (node == NULL) {
2020 DBG2("Relay sock %d not found", pollfd);
2021 rcu_read_unlock();
2022 goto error;
2023 }
2024 relay_connection = caa_container_of(node,
2025 struct relay_command, sock_n);
2026
2027 if (revents & (LPOLLERR)) {
2028 ERR("POLL ERROR");
2029 relay_cleanup_poll_connection(&events, pollfd);
2030 relay_del_connection(relay_connections_ht,
2031 streams_ht, &iter,
2032 relay_connection);
2033 if (last_seen_data_fd == pollfd) {
2034 last_seen_data_fd = last_notdel_data_fd;
2035 }
2036 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2037 DBG("Socket %d hung up", pollfd);
2038 relay_cleanup_poll_connection(&events, pollfd);
2039 relay_del_connection(relay_connections_ht,
2040 streams_ht, &iter,
2041 relay_connection);
2042 if (last_seen_data_fd == pollfd) {
2043 last_seen_data_fd = last_notdel_data_fd;
2044 }
2045 } else if (revents & LPOLLIN) {
2046 /* control socket */
2047 if (relay_connection->type == RELAY_CONTROL) {
2048 ret = relay_connection->sock->ops->recvmsg(
2049 relay_connection->sock, &recv_hdr,
2050 sizeof(struct lttcomm_relayd_hdr), 0);
2051 /* connection closed */
2052 if (ret <= 0) {
2053 relay_cleanup_poll_connection(&events, pollfd);
2054 relay_del_connection(relay_connections_ht,
2055 streams_ht, &iter,
2056 relay_connection);
2057 DBG("Control connection closed with %d", pollfd);
2058 } else {
2059 if (relay_connection->session) {
2060 DBG2("Relay worker receiving data for session : %" PRIu64,
2061 relay_connection->session->id);
2062 }
2063 ret = relay_process_control(&recv_hdr,
2064 relay_connection,
2065 streams_ht);
2066 if (ret < 0) {
2067 /* Clear the session on error. */
2068 relay_cleanup_poll_connection(&events, pollfd);
2069 relay_del_connection(relay_connections_ht,
2070 streams_ht, &iter,
2071 relay_connection);
2072 DBG("Connection closed with %d", pollfd);
2073 }
2074 seen_control = 1;
2075 }
2076 } else {
2077 /*
2078 * Flag the last seen data fd not deleted. It will be
2079 * used as the last seen fd if any fd gets deleted in
2080 * this first loop.
2081 */
2082 last_notdel_data_fd = pollfd;
2083 }
2084 }
2085 rcu_read_unlock();
2086 }
2087 }
2088
2089 /*
2090 * The last loop handled a control request, go back to poll to make
2091 * sure we prioritise the control socket.
2092 */
2093 if (seen_control) {
2094 continue;
2095 }
2096
2097 if (last_seen_data_fd >= 0) {
2098 for (i = 0; i < nb_fd; i++) {
2099 int pollfd = LTTNG_POLL_GETFD(&events, i);
2100 if (last_seen_data_fd == pollfd) {
2101 idx = i;
2102 break;
2103 }
2104 }
2105 }
2106
2107 /* Process data connection. */
2108 for (i = idx + 1; i < nb_fd; i++) {
2109 /* Fetch the poll data. */
2110 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2111 int pollfd = LTTNG_POLL_GETFD(&events, i);
2112
2113 /* Skip the command pipe. It's handled in the first loop. */
2114 if (pollfd == relay_cmd_pipe[0]) {
2115 continue;
2116 }
2117
2118 if (revents) {
2119 rcu_read_lock();
2120 lttng_ht_lookup(relay_connections_ht,
2121 (void *)((unsigned long) pollfd),
2122 &iter);
2123 node = lttng_ht_iter_get_node_ulong(&iter);
2124 if (node == NULL) {
2125 /* Skip it. Might be removed before. */
2126 rcu_read_unlock();
2127 continue;
2128 }
2129 relay_connection = caa_container_of(node,
2130 struct relay_command, sock_n);
2131
2132 if (revents & LPOLLIN) {
2133 if (relay_connection->type != RELAY_DATA) {
2134 continue;
2135 }
2136
2137 ret = relay_process_data(relay_connection, streams_ht);
2138 /* connection closed */
2139 if (ret < 0) {
2140 relay_cleanup_poll_connection(&events, pollfd);
2141 relay_del_connection(relay_connections_ht,
2142 streams_ht, &iter,
2143 relay_connection);
2144 DBG("Data connection closed with %d", pollfd);
2145 /*
2146 * Every goto restart call sets the last seen fd where
2147 * here we don't really care since we gracefully
2148 * continue the loop after the connection is deleted.
2149 */
2150 } else {
2151 /* Keep last seen port. */
2152 last_seen_data_fd = pollfd;
2153 rcu_read_unlock();
2154 goto restart;
2155 }
2156 }
2157 rcu_read_unlock();
2158 }
2159 }
2160 last_seen_data_fd = -1;
2161 }
2162
2163 exit:
2164 error:
2165 lttng_poll_clean(&events);
2166
2167 /* empty the hash table and free the memory */
2168 rcu_read_lock();
2169 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2170 node = lttng_ht_iter_get_node_ulong(&iter);
2171 if (node) {
2172 relay_connection = caa_container_of(node,
2173 struct relay_command, sock_n);
2174 relay_del_connection(relay_connections_ht,
2175 streams_ht, &iter,
2176 relay_connection);
2177 }
2178 }
2179 rcu_read_unlock();
2180 error_poll_create:
2181 lttng_ht_destroy(streams_ht);
2182 streams_ht_error:
2183 lttng_ht_destroy(relay_connections_ht);
2184 relay_connections_ht_error:
2185 /* Close relay cmd pipes */
2186 utils_close_pipe(relay_cmd_pipe);
2187 if (err) {
2188 DBG("Thread exited with error");
2189 }
2190 DBG("Worker thread cleanup complete");
2191 free(data_buffer);
2192 stop_threads();
2193 rcu_unregister_thread();
2194 return NULL;
2195 }
2196
2197 /*
2198 * Create the relay command pipe to wake thread_manage_apps.
2199 * Closed in cleanup().
2200 */
2201 static int create_relay_cmd_pipe(void)
2202 {
2203 int ret;
2204
2205 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2206
2207 return ret;
2208 }
2209
2210 /*
2211 * main
2212 */
2213 int main(int argc, char **argv)
2214 {
2215 int ret = 0;
2216 void *status;
2217
2218 /* Create thread quit pipe */
2219 if ((ret = init_thread_quit_pipe()) < 0) {
2220 goto error;
2221 }
2222
2223 /* Parse arguments */
2224 progname = argv[0];
2225 if ((ret = parse_args(argc, argv)) < 0) {
2226 goto exit;
2227 }
2228
2229 if ((ret = set_signal_handler()) < 0) {
2230 goto exit;
2231 }
2232
2233 /* Daemonize */
2234 if (opt_daemon) {
2235 ret = daemon(0, 0);
2236 if (ret < 0) {
2237 PERROR("daemon");
2238 goto exit;
2239 }
2240 }
2241
2242 /* Check if daemon is UID = 0 */
2243 is_root = !getuid();
2244
2245 if (!is_root) {
2246 if (control_uri->port < 1024 || data_uri->port < 1024) {
2247 ERR("Need to be root to use ports < 1024");
2248 ret = -1;
2249 goto exit;
2250 }
2251 }
2252
2253 /* Setup the thread apps communication pipe. */
2254 if ((ret = create_relay_cmd_pipe()) < 0) {
2255 goto exit;
2256 }
2257
2258 /* Init relay command queue. */
2259 cds_wfq_init(&relay_cmd_queue.queue);
2260
2261 /* Set up max poll set size */
2262 lttng_poll_set_max_size();
2263
2264 /* Setup the dispatcher thread */
2265 ret = pthread_create(&dispatcher_thread, NULL,
2266 relay_thread_dispatcher, (void *) NULL);
2267 if (ret != 0) {
2268 PERROR("pthread_create dispatcher");
2269 goto exit_dispatcher;
2270 }
2271
2272 /* Setup the worker thread */
2273 ret = pthread_create(&worker_thread, NULL,
2274 relay_thread_worker, (void *) NULL);
2275 if (ret != 0) {
2276 PERROR("pthread_create worker");
2277 goto exit_worker;
2278 }
2279
2280 /* Setup the listener thread */
2281 ret = pthread_create(&listener_thread, NULL,
2282 relay_thread_listener, (void *) NULL);
2283 if (ret != 0) {
2284 PERROR("pthread_create listener");
2285 goto exit_listener;
2286 }
2287
2288 exit_listener:
2289 ret = pthread_join(listener_thread, &status);
2290 if (ret != 0) {
2291 PERROR("pthread_join");
2292 goto error; /* join error, exit without cleanup */
2293 }
2294
2295 exit_worker:
2296 ret = pthread_join(worker_thread, &status);
2297 if (ret != 0) {
2298 PERROR("pthread_join");
2299 goto error; /* join error, exit without cleanup */
2300 }
2301
2302 exit_dispatcher:
2303 ret = pthread_join(dispatcher_thread, &status);
2304 if (ret != 0) {
2305 PERROR("pthread_join");
2306 goto error; /* join error, exit without cleanup */
2307 }
2308
2309 exit:
2310 cleanup();
2311 if (!ret) {
2312 exit(EXIT_SUCCESS);
2313 }
2314
2315 error:
2316 exit(EXIT_FAILURE);
2317 }
This page took 0.130989 seconds and 3 git commands to generate.