Adjust the relayd protocol on version check
[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 * config_get_default_path
690 *
691 * Returns the HOME directory path. Caller MUST NOT free(3) the return pointer.
692 */
693 static
694 char *config_get_default_path(void)
695 {
696 return getenv("HOME");
697 }
698
699 static
700 char *create_output_path_auto(char *path_name)
701 {
702 int ret;
703 char *traces_path = NULL;
704 char *alloc_path = NULL;
705 char *default_path;
706
707 default_path = config_get_default_path();
708 if (default_path == NULL) {
709 ERR("Home path not found.\n \
710 Please specify an output path using -o, --output PATH");
711 goto exit;
712 }
713 alloc_path = strdup(default_path);
714 if (alloc_path == NULL) {
715 PERROR("Path allocation");
716 goto exit;
717 }
718 ret = asprintf(&traces_path, "%s/" DEFAULT_TRACE_DIR_NAME
719 "/%s", alloc_path, path_name);
720 if (ret < 0) {
721 PERROR("asprintf trace dir name");
722 goto exit;
723 }
724 exit:
725 free(alloc_path);
726 return traces_path;
727 }
728
729 static
730 char *create_output_path_noauto(char *path_name)
731 {
732 int ret;
733 char *traces_path = NULL;
734 char *full_path;
735
736 full_path = utils_expand_path(opt_output_path);
737 if (!full_path) {
738 goto exit;
739 }
740
741 ret = asprintf(&traces_path, "%s/%s", full_path, path_name);
742 if (ret < 0) {
743 PERROR("asprintf trace dir name");
744 goto exit;
745 }
746 exit:
747 free(full_path);
748 return traces_path;
749 }
750
751 /*
752 * create_output_path: create the output trace directory
753 */
754 static
755 char *create_output_path(char *path_name)
756 {
757 if (opt_output_path == NULL) {
758 return create_output_path_auto(path_name);
759 } else {
760 return create_output_path_noauto(path_name);
761 }
762 }
763
764 /*
765 * Get stream from stream id.
766 * Need to be called with RCU read-side lock held.
767 */
768 static
769 struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id,
770 struct lttng_ht *streams_ht)
771 {
772 struct lttng_ht_node_ulong *node;
773 struct lttng_ht_iter iter;
774 struct relay_stream *ret;
775
776 lttng_ht_lookup(streams_ht,
777 (void *)((unsigned long) stream_id),
778 &iter);
779 node = lttng_ht_iter_get_node_ulong(&iter);
780 if (node == NULL) {
781 DBG("Relay stream %" PRIu64 " not found", stream_id);
782 ret = NULL;
783 goto end;
784 }
785
786 ret = caa_container_of(node, struct relay_stream, stream_n);
787
788 end:
789 return ret;
790 }
791
792 static
793 void deferred_free_stream(struct rcu_head *head)
794 {
795 struct relay_stream *stream =
796 caa_container_of(head, struct relay_stream, rcu_node);
797 free(stream);
798 }
799
800 /*
801 * relay_delete_session: Free all memory associated with a session and
802 * close all the FDs
803 */
804 static
805 void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht)
806 {
807 struct lttng_ht_iter iter;
808 struct lttng_ht_node_ulong *node;
809 struct relay_stream *stream;
810 int ret;
811
812 if (!cmd->session) {
813 return;
814 }
815
816 DBG("Relay deleting session %" PRIu64, cmd->session->id);
817
818 rcu_read_lock();
819 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, node, node) {
820 node = lttng_ht_iter_get_node_ulong(&iter);
821 if (node) {
822 stream = caa_container_of(node,
823 struct relay_stream, stream_n);
824 if (stream->session == cmd->session) {
825 ret = close(stream->fd);
826 if (ret < 0) {
827 PERROR("close stream fd on delete session");
828 }
829 ret = lttng_ht_del(streams_ht, &iter);
830 assert(!ret);
831 call_rcu(&stream->rcu_node,
832 deferred_free_stream);
833 }
834 }
835 }
836 rcu_read_unlock();
837
838 free(cmd->session);
839 }
840
841 /*
842 * Handle the RELAYD_CREATE_SESSION command.
843 *
844 * On success, send back the session id or else return a negative value.
845 */
846 static
847 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
848 struct relay_command *cmd)
849 {
850 int ret = 0, send_ret;
851 struct relay_session *session;
852 struct lttcomm_relayd_status_session reply;
853
854 assert(recv_hdr);
855 assert(cmd);
856
857 memset(&reply, 0, sizeof(reply));
858
859 session = zmalloc(sizeof(struct relay_session));
860 if (session == NULL) {
861 PERROR("relay session zmalloc");
862 ret = -1;
863 goto error;
864 }
865
866 session->id = ++last_relay_session_id;
867 session->sock = cmd->sock;
868 cmd->session = session;
869
870 reply.session_id = htobe64(session->id);
871
872 DBG("Created session %" PRIu64, session->id);
873
874 error:
875 if (ret < 0) {
876 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
877 } else {
878 reply.ret_code = htobe32(LTTNG_OK);
879 }
880
881 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
882 if (send_ret < 0) {
883 ERR("Relayd sending session id");
884 ret = send_ret;
885 }
886
887 return ret;
888 }
889
890 /*
891 * relay_add_stream: allocate a new stream for a session
892 */
893 static
894 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
895 struct relay_command *cmd, struct lttng_ht *streams_ht)
896 {
897 struct relay_session *session = cmd->session;
898 struct lttcomm_relayd_add_stream stream_info;
899 struct relay_stream *stream = NULL;
900 struct lttcomm_relayd_status_stream reply;
901 char *path = NULL, *root_path = NULL;
902 int ret, send_ret;
903
904 if (!session || cmd->version_check_done == 0) {
905 ERR("Trying to add a stream before version check");
906 ret = -1;
907 goto end_no_session;
908 }
909
910 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
911 sizeof(struct lttcomm_relayd_add_stream), 0);
912 if (ret < sizeof(struct lttcomm_relayd_add_stream)) {
913 if (ret == 0) {
914 /* Orderly shutdown. Not necessary to print an error. */
915 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
916 } else {
917 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
918 }
919 ret = -1;
920 goto end_no_session;
921 }
922 stream = zmalloc(sizeof(struct relay_stream));
923 if (stream == NULL) {
924 PERROR("relay stream zmalloc");
925 ret = -1;
926 goto end_no_session;
927 }
928
929 rcu_read_lock();
930 stream->stream_handle = ++last_relay_stream_id;
931 stream->prev_seq = -1ULL;
932 stream->session = session;
933
934 root_path = create_output_path(stream_info.pathname);
935 if (!root_path) {
936 ret = -1;
937 goto end;
938 }
939 ret = utils_mkdir_recursive(root_path, S_IRWXU | S_IRWXG);
940 if (ret < 0) {
941 ERR("relay creating output directory");
942 goto end;
943 }
944
945 ret = asprintf(&path, "%s/%s", root_path, stream_info.channel_name);
946 if (ret < 0) {
947 PERROR("asprintf stream path");
948 path = NULL;
949 goto end;
950 }
951
952 ret = open(path, O_WRONLY|O_CREAT|O_TRUNC, S_IRWXU|S_IRWXG|S_IRWXO);
953 if (ret < 0) {
954 PERROR("Relay creating trace file");
955 goto end;
956 }
957
958 stream->fd = ret;
959 DBG("Tracefile %s created", path);
960
961 lttng_ht_node_init_ulong(&stream->stream_n,
962 (unsigned long) stream->stream_handle);
963 lttng_ht_add_unique_ulong(streams_ht,
964 &stream->stream_n);
965
966 DBG("Relay new stream added %s", stream_info.channel_name);
967
968 end:
969 free(path);
970 free(root_path);
971
972 reply.handle = htobe64(stream->stream_handle);
973 /* send the session id to the client or a negative return code on error */
974 if (ret < 0) {
975 reply.ret_code = htobe32(LTTNG_ERR_UNK);
976 /* stream was not properly added to the ht, so free it */
977 free(stream);
978 } else {
979 reply.ret_code = htobe32(LTTNG_OK);
980 }
981
982 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
983 sizeof(struct lttcomm_relayd_status_stream), 0);
984 if (send_ret < 0) {
985 ERR("Relay sending stream id");
986 ret = send_ret;
987 }
988 rcu_read_unlock();
989
990 end_no_session:
991 return ret;
992 }
993
994 /*
995 * relay_close_stream: close a specific stream
996 */
997 static
998 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
999 struct relay_command *cmd, struct lttng_ht *streams_ht)
1000 {
1001 struct relay_session *session = cmd->session;
1002 struct lttcomm_relayd_close_stream stream_info;
1003 struct lttcomm_relayd_generic_reply reply;
1004 struct relay_stream *stream;
1005 int ret, send_ret;
1006 struct lttng_ht_iter iter;
1007
1008 DBG("Close stream received");
1009
1010 if (!session || cmd->version_check_done == 0) {
1011 ERR("Trying to close a stream before version check");
1012 ret = -1;
1013 goto end_no_session;
1014 }
1015
1016 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1017 sizeof(struct lttcomm_relayd_close_stream), 0);
1018 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1019 if (ret == 0) {
1020 /* Orderly shutdown. Not necessary to print an error. */
1021 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1022 } else {
1023 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1024 }
1025 ret = -1;
1026 goto end_no_session;
1027 }
1028
1029 rcu_read_lock();
1030 stream = relay_stream_from_stream_id(be64toh(stream_info.stream_id),
1031 streams_ht);
1032 if (!stream) {
1033 ret = -1;
1034 goto end_unlock;
1035 }
1036
1037 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1038 stream->close_flag = 1;
1039
1040 if (close_stream_check(stream)) {
1041 int delret;
1042
1043 delret = close(stream->fd);
1044 if (delret < 0) {
1045 PERROR("close stream");
1046 }
1047 iter.iter.node = &stream->stream_n.node;
1048 delret = lttng_ht_del(streams_ht, &iter);
1049 assert(!delret);
1050 call_rcu(&stream->rcu_node,
1051 deferred_free_stream);
1052 DBG("Closed tracefile %d from close stream", stream->fd);
1053 }
1054
1055 end_unlock:
1056 rcu_read_unlock();
1057
1058 if (ret < 0) {
1059 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1060 } else {
1061 reply.ret_code = htobe32(LTTNG_OK);
1062 }
1063 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1064 sizeof(struct lttcomm_relayd_generic_reply), 0);
1065 if (send_ret < 0) {
1066 ERR("Relay sending stream id");
1067 ret = send_ret;
1068 }
1069
1070 end_no_session:
1071 return ret;
1072 }
1073
1074 /*
1075 * relay_unknown_command: send -1 if received unknown command
1076 */
1077 static
1078 void relay_unknown_command(struct relay_command *cmd)
1079 {
1080 struct lttcomm_relayd_generic_reply reply;
1081 int ret;
1082
1083 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1084 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1085 sizeof(struct lttcomm_relayd_generic_reply), 0);
1086 if (ret < 0) {
1087 ERR("Relay sending unknown command");
1088 }
1089 }
1090
1091 /*
1092 * relay_start: send an acknowledgment to the client to tell if we are
1093 * ready to receive data. We are ready if a session is established.
1094 */
1095 static
1096 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1097 struct relay_command *cmd)
1098 {
1099 int ret = htobe32(LTTNG_OK);
1100 struct lttcomm_relayd_generic_reply reply;
1101 struct relay_session *session = cmd->session;
1102
1103 if (!session) {
1104 DBG("Trying to start the streaming without a session established");
1105 ret = htobe32(LTTNG_ERR_UNK);
1106 }
1107
1108 reply.ret_code = ret;
1109 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1110 sizeof(struct lttcomm_relayd_generic_reply), 0);
1111 if (ret < 0) {
1112 ERR("Relay sending start ack");
1113 }
1114
1115 return ret;
1116 }
1117
1118 /*
1119 * Append padding to the file pointed by the file descriptor fd.
1120 */
1121 static int write_padding_to_file(int fd, uint32_t size)
1122 {
1123 int ret = 0;
1124 char *zeros;
1125
1126 if (size == 0) {
1127 goto end;
1128 }
1129
1130 zeros = zmalloc(size);
1131 if (zeros == NULL) {
1132 PERROR("zmalloc zeros for padding");
1133 ret = -1;
1134 goto end;
1135 }
1136
1137 do {
1138 ret = write(fd, zeros, size);
1139 } while (ret < 0 && errno == EINTR);
1140 if (ret < 0 || ret != size) {
1141 PERROR("write padding to file");
1142 }
1143
1144 free(zeros);
1145
1146 end:
1147 return ret;
1148 }
1149
1150 /*
1151 * relay_recv_metadata: receive the metada for the session.
1152 */
1153 static
1154 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1155 struct relay_command *cmd, struct lttng_ht *streams_ht)
1156 {
1157 int ret = htobe32(LTTNG_OK);
1158 struct relay_session *session = cmd->session;
1159 struct lttcomm_relayd_metadata_payload *metadata_struct;
1160 struct relay_stream *metadata_stream;
1161 uint64_t data_size, payload_size;
1162
1163 if (!session) {
1164 ERR("Metadata sent before version check");
1165 ret = -1;
1166 goto end;
1167 }
1168
1169 data_size = payload_size = be64toh(recv_hdr->data_size);
1170 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1171 ERR("Incorrect data size");
1172 ret = -1;
1173 goto end;
1174 }
1175 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1176
1177 if (data_buffer_size < data_size) {
1178 /* In case the realloc fails, we can free the memory */
1179 char *tmp_data_ptr;
1180
1181 tmp_data_ptr = realloc(data_buffer, data_size);
1182 if (!tmp_data_ptr) {
1183 ERR("Allocating data buffer");
1184 free(data_buffer);
1185 ret = -1;
1186 goto end;
1187 }
1188 data_buffer = tmp_data_ptr;
1189 data_buffer_size = data_size;
1190 }
1191 memset(data_buffer, 0, data_size);
1192 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1193 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1194 if (ret < 0 || ret != data_size) {
1195 if (ret == 0) {
1196 /* Orderly shutdown. Not necessary to print an error. */
1197 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1198 } else {
1199 ERR("Relay didn't receive the whole metadata");
1200 }
1201 ret = -1;
1202 goto end;
1203 }
1204 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1205
1206 rcu_read_lock();
1207 metadata_stream = relay_stream_from_stream_id(
1208 be64toh(metadata_struct->stream_id), streams_ht);
1209 if (!metadata_stream) {
1210 ret = -1;
1211 goto end_unlock;
1212 }
1213
1214 do {
1215 ret = write(metadata_stream->fd, metadata_struct->payload,
1216 payload_size);
1217 } while (ret < 0 && errno == EINTR);
1218 if (ret < 0 || ret != payload_size) {
1219 ERR("Relay error writing metadata on file");
1220 ret = -1;
1221 goto end_unlock;
1222 }
1223
1224 ret = write_padding_to_file(metadata_stream->fd,
1225 be32toh(metadata_struct->padding_size));
1226 if (ret < 0) {
1227 goto end_unlock;
1228 }
1229
1230 DBG2("Relay metadata written");
1231
1232 end_unlock:
1233 rcu_read_unlock();
1234 end:
1235 return ret;
1236 }
1237
1238 /*
1239 * relay_send_version: send relayd version number
1240 */
1241 static
1242 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1243 struct relay_command *cmd, struct lttng_ht *streams_ht)
1244 {
1245 int ret;
1246 struct lttcomm_relayd_version reply, msg;
1247
1248 assert(cmd);
1249
1250 cmd->version_check_done = 1;
1251
1252 /* Get version from the other side. */
1253 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1254 if (ret < 0 || ret != sizeof(msg)) {
1255 if (ret == 0) {
1256 /* Orderly shutdown. Not necessary to print an error. */
1257 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1258 } else {
1259 ERR("Relay failed to receive the version values.");
1260 }
1261 ret = -1;
1262 goto end;
1263 }
1264
1265 ret = sscanf(VERSION, "%10u.%10u", &reply.major, &reply.minor);
1266 if (ret < 2) {
1267 ERR("Error in scanning version");
1268 ret = -1;
1269 goto end;
1270 }
1271 reply.major = htobe32(reply.major);
1272 reply.minor = htobe32(reply.minor);
1273 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1274 sizeof(struct lttcomm_relayd_version), 0);
1275 if (ret < 0) {
1276 ERR("Relay sending version");
1277 }
1278
1279 /* Major versions must be the same */
1280 if (reply.major != be32toh(msg.major)) {
1281 DBG("Incompatible major versions, deleting session");
1282 relay_delete_session(cmd, streams_ht);
1283 ret = 0;
1284 goto end;
1285 }
1286
1287 cmd->session->major = reply.major;
1288 /* We adapt to the lowest compatible version */
1289 if (reply.minor <= be32toh(msg.minor)) {
1290 cmd->session->minor = reply.minor;
1291 } else {
1292 cmd->session->minor = be32toh(msg.minor);
1293 }
1294
1295 DBG("Version check done using protocol %u.%u", cmd->session->major,
1296 cmd->session->minor);
1297
1298 end:
1299 return ret;
1300 }
1301
1302 /*
1303 * Check for data pending for a given stream id from the session daemon.
1304 */
1305 static
1306 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1307 struct relay_command *cmd, struct lttng_ht *streams_ht)
1308 {
1309 struct relay_session *session = cmd->session;
1310 struct lttcomm_relayd_data_pending msg;
1311 struct lttcomm_relayd_generic_reply reply;
1312 struct relay_stream *stream;
1313 int ret;
1314 uint64_t last_net_seq_num, stream_id;
1315
1316 DBG("Data pending command received");
1317
1318 if (!session || cmd->version_check_done == 0) {
1319 ERR("Trying to check for data before version check");
1320 ret = -1;
1321 goto end_no_session;
1322 }
1323
1324 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1325 if (ret < sizeof(msg)) {
1326 if (ret == 0) {
1327 /* Orderly shutdown. Not necessary to print an error. */
1328 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1329 } else {
1330 ERR("Relay didn't receive valid data_pending struct size : %d",
1331 ret);
1332 }
1333 ret = -1;
1334 goto end_no_session;
1335 }
1336
1337 stream_id = be64toh(msg.stream_id);
1338 last_net_seq_num = be64toh(msg.last_net_seq_num);
1339
1340 rcu_read_lock();
1341 stream = relay_stream_from_stream_id(stream_id, streams_ht);
1342 if (stream == NULL) {
1343 ret = -1;
1344 goto end_unlock;
1345 }
1346
1347 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1348 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1349 last_net_seq_num);
1350
1351 /* Avoid wrapping issue */
1352 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1353 /* Data has in fact been written and is NOT pending */
1354 ret = 0;
1355 } else {
1356 /* Data still being streamed thus pending */
1357 ret = 1;
1358 }
1359
1360 /* Pending check is now done. */
1361 stream->data_pending_check_done = 1;
1362
1363 end_unlock:
1364 rcu_read_unlock();
1365
1366 reply.ret_code = htobe32(ret);
1367 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1368 if (ret < 0) {
1369 ERR("Relay data pending ret code failed");
1370 }
1371
1372 end_no_session:
1373 return ret;
1374 }
1375
1376 /*
1377 * Wait for the control socket to reach a quiescent state.
1378 *
1379 * Note that for now, when receiving this command from the session daemon, this
1380 * means that every subsequent commands or data received on the control socket
1381 * has been handled. So, this is why we simply return OK here.
1382 */
1383 static
1384 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1385 struct relay_command *cmd, struct lttng_ht *streams_ht)
1386 {
1387 int ret;
1388 uint64_t stream_id;
1389 struct relay_stream *stream;
1390 struct lttng_ht_iter iter;
1391 struct lttcomm_relayd_quiescent_control msg;
1392 struct lttcomm_relayd_generic_reply reply;
1393
1394 DBG("Checking quiescent state on control socket");
1395
1396 if (!cmd->session || cmd->version_check_done == 0) {
1397 ERR("Trying to check for data before version check");
1398 ret = -1;
1399 goto end_no_session;
1400 }
1401
1402 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1403 if (ret < sizeof(msg)) {
1404 if (ret == 0) {
1405 /* Orderly shutdown. Not necessary to print an error. */
1406 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1407 } else {
1408 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1409 ret);
1410 }
1411 ret = -1;
1412 goto end_no_session;
1413 }
1414
1415 stream_id = be64toh(msg.stream_id);
1416
1417 rcu_read_lock();
1418 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) {
1419 if (stream->stream_handle == stream_id) {
1420 stream->data_pending_check_done = 1;
1421 DBG("Relay quiescent control pending flag set to %" PRIu64,
1422 stream_id);
1423 break;
1424 }
1425 }
1426 rcu_read_unlock();
1427
1428 reply.ret_code = htobe32(LTTNG_OK);
1429 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1430 if (ret < 0) {
1431 ERR("Relay data quiescent control ret code failed");
1432 }
1433
1434 end_no_session:
1435 return ret;
1436 }
1437
1438 /*
1439 * Initialize a data pending command. This means that a client is about to ask
1440 * for data pending for each stream he/she holds. Simply iterate over all
1441 * streams of a session and set the data_pending_check_done flag.
1442 *
1443 * This command returns to the client a LTTNG_OK code.
1444 */
1445 static
1446 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1447 struct relay_command *cmd, struct lttng_ht *streams_ht)
1448 {
1449 int ret;
1450 struct lttng_ht_iter iter;
1451 struct lttcomm_relayd_begin_data_pending msg;
1452 struct lttcomm_relayd_generic_reply reply;
1453 struct relay_stream *stream;
1454 uint64_t session_id;
1455
1456 assert(recv_hdr);
1457 assert(cmd);
1458 assert(streams_ht);
1459
1460 DBG("Init streams for data pending");
1461
1462 if (!cmd->session || cmd->version_check_done == 0) {
1463 ERR("Trying to check for data before version check");
1464 ret = -1;
1465 goto end_no_session;
1466 }
1467
1468 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1469 if (ret < sizeof(msg)) {
1470 if (ret == 0) {
1471 /* Orderly shutdown. Not necessary to print an error. */
1472 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1473 } else {
1474 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1475 ret);
1476 }
1477 ret = -1;
1478 goto end_no_session;
1479 }
1480
1481 session_id = be64toh(msg.session_id);
1482
1483 /*
1484 * Iterate over all streams to set the begin data pending flag. For now, the
1485 * streams are indexed by stream handle so we have to iterate over all
1486 * streams to find the one associated with the right session_id.
1487 */
1488 rcu_read_lock();
1489 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) {
1490 if (stream->session->id == session_id) {
1491 stream->data_pending_check_done = 0;
1492 DBG("Set begin data pending flag to stream %" PRIu64,
1493 stream->stream_handle);
1494 }
1495 }
1496 rcu_read_unlock();
1497
1498 /* All good, send back reply. */
1499 reply.ret_code = htobe32(LTTNG_OK);
1500
1501 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1502 if (ret < 0) {
1503 ERR("Relay begin data pending send reply failed");
1504 }
1505
1506 end_no_session:
1507 return ret;
1508 }
1509
1510 /*
1511 * End data pending command. This will check, for a given session id, if each
1512 * stream associated with it has its data_pending_check_done flag set. If not,
1513 * this means that the client lost track of the stream but the data is still
1514 * being streamed on our side. In this case, we inform the client that data is
1515 * inflight.
1516 *
1517 * Return to the client if there is data in flight or not with a ret_code.
1518 */
1519 static
1520 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1521 struct relay_command *cmd, struct lttng_ht *streams_ht)
1522 {
1523 int ret;
1524 struct lttng_ht_iter iter;
1525 struct lttcomm_relayd_end_data_pending msg;
1526 struct lttcomm_relayd_generic_reply reply;
1527 struct relay_stream *stream;
1528 uint64_t session_id;
1529 uint32_t is_data_inflight = 0;
1530
1531 assert(recv_hdr);
1532 assert(cmd);
1533 assert(streams_ht);
1534
1535 DBG("End data pending command");
1536
1537 if (!cmd->session || cmd->version_check_done == 0) {
1538 ERR("Trying to check for data before version check");
1539 ret = -1;
1540 goto end_no_session;
1541 }
1542
1543 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1544 if (ret < sizeof(msg)) {
1545 if (ret == 0) {
1546 /* Orderly shutdown. Not necessary to print an error. */
1547 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1548 } else {
1549 ERR("Relay didn't receive valid end data_pending struct size: %d",
1550 ret);
1551 }
1552 ret = -1;
1553 goto end_no_session;
1554 }
1555
1556 session_id = be64toh(msg.session_id);
1557
1558 /* Iterate over all streams to see if the begin data pending flag is set. */
1559 rcu_read_lock();
1560 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) {
1561 if (stream->session->id == session_id &&
1562 !stream->data_pending_check_done) {
1563 is_data_inflight = 1;
1564 DBG("Data is still in flight for stream %" PRIu64,
1565 stream->stream_handle);
1566 break;
1567 }
1568 }
1569 rcu_read_unlock();
1570
1571 /* All good, send back reply. */
1572 reply.ret_code = htobe32(is_data_inflight);
1573
1574 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1575 if (ret < 0) {
1576 ERR("Relay end data pending send reply failed");
1577 }
1578
1579 end_no_session:
1580 return ret;
1581 }
1582
1583 /*
1584 * relay_process_control: Process the commands received on the control socket
1585 */
1586 static
1587 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1588 struct relay_command *cmd, struct lttng_ht *streams_ht)
1589 {
1590 int ret = 0;
1591
1592 switch (be32toh(recv_hdr->cmd)) {
1593 case RELAYD_CREATE_SESSION:
1594 ret = relay_create_session(recv_hdr, cmd);
1595 break;
1596 case RELAYD_ADD_STREAM:
1597 ret = relay_add_stream(recv_hdr, cmd, streams_ht);
1598 break;
1599 case RELAYD_START_DATA:
1600 ret = relay_start(recv_hdr, cmd);
1601 break;
1602 case RELAYD_SEND_METADATA:
1603 ret = relay_recv_metadata(recv_hdr, cmd, streams_ht);
1604 break;
1605 case RELAYD_VERSION:
1606 ret = relay_send_version(recv_hdr, cmd, streams_ht);
1607 break;
1608 case RELAYD_CLOSE_STREAM:
1609 ret = relay_close_stream(recv_hdr, cmd, streams_ht);
1610 break;
1611 case RELAYD_DATA_PENDING:
1612 ret = relay_data_pending(recv_hdr, cmd, streams_ht);
1613 break;
1614 case RELAYD_QUIESCENT_CONTROL:
1615 ret = relay_quiescent_control(recv_hdr, cmd, streams_ht);
1616 break;
1617 case RELAYD_BEGIN_DATA_PENDING:
1618 ret = relay_begin_data_pending(recv_hdr, cmd, streams_ht);
1619 break;
1620 case RELAYD_END_DATA_PENDING:
1621 ret = relay_end_data_pending(recv_hdr, cmd, streams_ht);
1622 break;
1623 case RELAYD_UPDATE_SYNC_INFO:
1624 default:
1625 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1626 relay_unknown_command(cmd);
1627 ret = -1;
1628 goto end;
1629 }
1630
1631 end:
1632 return ret;
1633 }
1634
1635 /*
1636 * relay_process_data: Process the data received on the data socket
1637 */
1638 static
1639 int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht)
1640 {
1641 int ret = 0;
1642 struct relay_stream *stream;
1643 struct lttcomm_relayd_data_hdr data_hdr;
1644 uint64_t stream_id;
1645 uint64_t net_seq_num;
1646 uint32_t data_size;
1647
1648 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1649 sizeof(struct lttcomm_relayd_data_hdr), 0);
1650 if (ret <= 0) {
1651 if (ret == 0) {
1652 /* Orderly shutdown. Not necessary to print an error. */
1653 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1654 } else {
1655 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
1656 }
1657 ret = -1;
1658 goto end;
1659 }
1660
1661 stream_id = be64toh(data_hdr.stream_id);
1662
1663 rcu_read_lock();
1664 stream = relay_stream_from_stream_id(stream_id, streams_ht);
1665 if (!stream) {
1666 ret = -1;
1667 goto end_unlock;
1668 }
1669
1670 data_size = be32toh(data_hdr.data_size);
1671 if (data_buffer_size < data_size) {
1672 char *tmp_data_ptr;
1673
1674 tmp_data_ptr = realloc(data_buffer, data_size);
1675 if (!tmp_data_ptr) {
1676 ERR("Allocating data buffer");
1677 free(data_buffer);
1678 ret = -1;
1679 goto end_unlock;
1680 }
1681 data_buffer = tmp_data_ptr;
1682 data_buffer_size = data_size;
1683 }
1684 memset(data_buffer, 0, data_size);
1685
1686 net_seq_num = be64toh(data_hdr.net_seq_num);
1687
1688 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
1689 data_size, stream_id, net_seq_num);
1690 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1691 if (ret <= 0) {
1692 if (ret == 0) {
1693 /* Orderly shutdown. Not necessary to print an error. */
1694 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1695 }
1696 ret = -1;
1697 goto end_unlock;
1698 }
1699
1700 do {
1701 ret = write(stream->fd, data_buffer, data_size);
1702 } while (ret < 0 && errno == EINTR);
1703 if (ret < 0 || ret != data_size) {
1704 ERR("Relay error writing data to file");
1705 ret = -1;
1706 goto end_unlock;
1707 }
1708
1709 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
1710 ret, stream->stream_handle);
1711
1712 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
1713 if (ret < 0) {
1714 goto end_unlock;
1715 }
1716
1717 stream->prev_seq = net_seq_num;
1718
1719 /* Check if we need to close the FD */
1720 if (close_stream_check(stream)) {
1721 int cret;
1722 struct lttng_ht_iter iter;
1723
1724 cret = close(stream->fd);
1725 if (cret < 0) {
1726 PERROR("close stream process data");
1727 }
1728 iter.iter.node = &stream->stream_n.node;
1729 ret = lttng_ht_del(streams_ht, &iter);
1730 assert(!ret);
1731 call_rcu(&stream->rcu_node,
1732 deferred_free_stream);
1733 DBG("Closed tracefile %d after recv data", stream->fd);
1734 }
1735
1736 end_unlock:
1737 rcu_read_unlock();
1738 end:
1739 return ret;
1740 }
1741
1742 static
1743 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
1744 {
1745 int ret;
1746
1747 lttng_poll_del(events, pollfd);
1748
1749 ret = close(pollfd);
1750 if (ret < 0) {
1751 ERR("Closing pollfd %d", pollfd);
1752 }
1753 }
1754
1755 static
1756 int relay_add_connection(int fd, struct lttng_poll_event *events,
1757 struct lttng_ht *relay_connections_ht)
1758 {
1759 struct relay_command *relay_connection;
1760 int ret;
1761
1762 relay_connection = zmalloc(sizeof(struct relay_command));
1763 if (relay_connection == NULL) {
1764 PERROR("Relay command zmalloc");
1765 goto error;
1766 }
1767 do {
1768 ret = read(fd, relay_connection, sizeof(struct relay_command));
1769 } while (ret < 0 && errno == EINTR);
1770 if (ret < 0 || ret < sizeof(struct relay_command)) {
1771 PERROR("read relay cmd pipe");
1772 goto error_read;
1773 }
1774
1775 lttng_ht_node_init_ulong(&relay_connection->sock_n,
1776 (unsigned long) relay_connection->sock->fd);
1777 rcu_read_lock();
1778 lttng_ht_add_unique_ulong(relay_connections_ht,
1779 &relay_connection->sock_n);
1780 rcu_read_unlock();
1781 return lttng_poll_add(events,
1782 relay_connection->sock->fd,
1783 LPOLLIN | LPOLLRDHUP);
1784
1785 error_read:
1786 free(relay_connection);
1787 error:
1788 return -1;
1789 }
1790
1791 static
1792 void deferred_free_connection(struct rcu_head *head)
1793 {
1794 struct relay_command *relay_connection =
1795 caa_container_of(head, struct relay_command, rcu_node);
1796
1797 lttcomm_destroy_sock(relay_connection->sock);
1798 free(relay_connection);
1799 }
1800
1801 static
1802 void relay_del_connection(struct lttng_ht *relay_connections_ht,
1803 struct lttng_ht *streams_ht, struct lttng_ht_iter *iter,
1804 struct relay_command *relay_connection)
1805 {
1806 int ret;
1807
1808 ret = lttng_ht_del(relay_connections_ht, iter);
1809 assert(!ret);
1810 if (relay_connection->type == RELAY_CONTROL) {
1811 relay_delete_session(relay_connection, streams_ht);
1812 }
1813
1814 call_rcu(&relay_connection->rcu_node,
1815 deferred_free_connection);
1816 }
1817
1818 /*
1819 * This thread does the actual work
1820 */
1821 static
1822 void *relay_thread_worker(void *data)
1823 {
1824 int ret, err = -1, last_seen_data_fd = -1;
1825 uint32_t nb_fd;
1826 struct relay_command *relay_connection;
1827 struct lttng_poll_event events;
1828 struct lttng_ht *relay_connections_ht;
1829 struct lttng_ht_node_ulong *node;
1830 struct lttng_ht_iter iter;
1831 struct lttng_ht *streams_ht;
1832 struct lttcomm_relayd_hdr recv_hdr;
1833
1834 DBG("[thread] Relay worker started");
1835
1836 rcu_register_thread();
1837
1838 /* table of connections indexed on socket */
1839 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1840 if (!relay_connections_ht) {
1841 goto relay_connections_ht_error;
1842 }
1843
1844 /* tables of streams indexed by stream ID */
1845 streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1846 if (!streams_ht) {
1847 goto streams_ht_error;
1848 }
1849
1850 ret = create_thread_poll_set(&events, 2);
1851 if (ret < 0) {
1852 goto error_poll_create;
1853 }
1854
1855 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1856 if (ret < 0) {
1857 goto error;
1858 }
1859
1860 restart:
1861 while (1) {
1862 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
1863
1864 /* Infinite blocking call, waiting for transmission */
1865 DBG3("Relayd worker thread polling...");
1866 ret = lttng_poll_wait(&events, -1);
1867 if (ret < 0) {
1868 /*
1869 * Restart interrupted system call.
1870 */
1871 if (errno == EINTR) {
1872 goto restart;
1873 }
1874 goto error;
1875 }
1876
1877 nb_fd = ret;
1878
1879 /*
1880 * Process control. The control connection is prioritised so we don't
1881 * starve it with high throughout put tracing data on the data
1882 * connection.
1883 */
1884 for (i = 0; i < nb_fd; i++) {
1885 /* Fetch once the poll data */
1886 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1887 int pollfd = LTTNG_POLL_GETFD(&events, i);
1888
1889 /* Thread quit pipe has been closed. Killing thread. */
1890 ret = check_thread_quit_pipe(pollfd, revents);
1891 if (ret) {
1892 err = 0;
1893 goto exit;
1894 }
1895
1896 /* Inspect the relay cmd pipe for new connection */
1897 if (pollfd == relay_cmd_pipe[0]) {
1898 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1899 ERR("Relay pipe error");
1900 goto error;
1901 } else if (revents & LPOLLIN) {
1902 DBG("Relay command received");
1903 ret = relay_add_connection(relay_cmd_pipe[0],
1904 &events, relay_connections_ht);
1905 if (ret < 0) {
1906 goto error;
1907 }
1908 }
1909 } else if (revents) {
1910 rcu_read_lock();
1911 lttng_ht_lookup(relay_connections_ht,
1912 (void *)((unsigned long) pollfd),
1913 &iter);
1914 node = lttng_ht_iter_get_node_ulong(&iter);
1915 if (node == NULL) {
1916 DBG2("Relay sock %d not found", pollfd);
1917 rcu_read_unlock();
1918 goto error;
1919 }
1920 relay_connection = caa_container_of(node,
1921 struct relay_command, sock_n);
1922
1923 if (revents & (LPOLLERR)) {
1924 ERR("POLL ERROR");
1925 relay_cleanup_poll_connection(&events, pollfd);
1926 relay_del_connection(relay_connections_ht,
1927 streams_ht, &iter,
1928 relay_connection);
1929 if (last_seen_data_fd == pollfd) {
1930 last_seen_data_fd = last_notdel_data_fd;
1931 }
1932 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
1933 DBG("Socket %d hung up", pollfd);
1934 relay_cleanup_poll_connection(&events, pollfd);
1935 relay_del_connection(relay_connections_ht,
1936 streams_ht, &iter,
1937 relay_connection);
1938 if (last_seen_data_fd == pollfd) {
1939 last_seen_data_fd = last_notdel_data_fd;
1940 }
1941 } else if (revents & LPOLLIN) {
1942 /* control socket */
1943 if (relay_connection->type == RELAY_CONTROL) {
1944 ret = relay_connection->sock->ops->recvmsg(
1945 relay_connection->sock, &recv_hdr,
1946 sizeof(struct lttcomm_relayd_hdr), 0);
1947 /* connection closed */
1948 if (ret <= 0) {
1949 relay_cleanup_poll_connection(&events, pollfd);
1950 relay_del_connection(relay_connections_ht,
1951 streams_ht, &iter,
1952 relay_connection);
1953 DBG("Control connection closed with %d", pollfd);
1954 } else {
1955 if (relay_connection->session) {
1956 DBG2("Relay worker receiving data for session : %" PRIu64,
1957 relay_connection->session->id);
1958 }
1959 ret = relay_process_control(&recv_hdr,
1960 relay_connection,
1961 streams_ht);
1962 if (ret < 0) {
1963 /* Clear the session on error. */
1964 relay_cleanup_poll_connection(&events, pollfd);
1965 relay_del_connection(relay_connections_ht,
1966 streams_ht, &iter,
1967 relay_connection);
1968 DBG("Connection closed with %d", pollfd);
1969 }
1970 seen_control = 1;
1971 }
1972 } else {
1973 /*
1974 * Flag the last seen data fd not deleted. It will be
1975 * used as the last seen fd if any fd gets deleted in
1976 * this first loop.
1977 */
1978 last_notdel_data_fd = pollfd;
1979 }
1980 }
1981 rcu_read_unlock();
1982 }
1983 }
1984
1985 /*
1986 * The last loop handled a control request, go back to poll to make
1987 * sure we prioritise the control socket.
1988 */
1989 if (seen_control) {
1990 continue;
1991 }
1992
1993 if (last_seen_data_fd >= 0) {
1994 for (i = 0; i < nb_fd; i++) {
1995 int pollfd = LTTNG_POLL_GETFD(&events, i);
1996 if (last_seen_data_fd == pollfd) {
1997 idx = i;
1998 break;
1999 }
2000 }
2001 }
2002
2003 /* Process data connection. */
2004 for (i = idx + 1; i < nb_fd; i++) {
2005 /* Fetch the poll data. */
2006 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2007 int pollfd = LTTNG_POLL_GETFD(&events, i);
2008
2009 /* Skip the command pipe. It's handled in the first loop. */
2010 if (pollfd == relay_cmd_pipe[0]) {
2011 continue;
2012 }
2013
2014 if (revents) {
2015 rcu_read_lock();
2016 lttng_ht_lookup(relay_connections_ht,
2017 (void *)((unsigned long) pollfd),
2018 &iter);
2019 node = lttng_ht_iter_get_node_ulong(&iter);
2020 if (node == NULL) {
2021 /* Skip it. Might be removed before. */
2022 rcu_read_unlock();
2023 continue;
2024 }
2025 relay_connection = caa_container_of(node,
2026 struct relay_command, sock_n);
2027
2028 if (revents & LPOLLIN) {
2029 if (relay_connection->type != RELAY_DATA) {
2030 continue;
2031 }
2032
2033 ret = relay_process_data(relay_connection, streams_ht);
2034 /* connection closed */
2035 if (ret < 0) {
2036 relay_cleanup_poll_connection(&events, pollfd);
2037 relay_del_connection(relay_connections_ht,
2038 streams_ht, &iter,
2039 relay_connection);
2040 DBG("Data connection closed with %d", pollfd);
2041 /*
2042 * Every goto restart call sets the last seen fd where
2043 * here we don't really care since we gracefully
2044 * continue the loop after the connection is deleted.
2045 */
2046 } else {
2047 /* Keep last seen port. */
2048 last_seen_data_fd = pollfd;
2049 rcu_read_unlock();
2050 goto restart;
2051 }
2052 }
2053 rcu_read_unlock();
2054 }
2055 }
2056 last_seen_data_fd = -1;
2057 }
2058
2059 exit:
2060 error:
2061 lttng_poll_clean(&events);
2062
2063 /* empty the hash table and free the memory */
2064 rcu_read_lock();
2065 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2066 node = lttng_ht_iter_get_node_ulong(&iter);
2067 if (node) {
2068 relay_connection = caa_container_of(node,
2069 struct relay_command, sock_n);
2070 relay_del_connection(relay_connections_ht,
2071 streams_ht, &iter,
2072 relay_connection);
2073 }
2074 }
2075 rcu_read_unlock();
2076 error_poll_create:
2077 lttng_ht_destroy(streams_ht);
2078 streams_ht_error:
2079 lttng_ht_destroy(relay_connections_ht);
2080 relay_connections_ht_error:
2081 /* Close relay cmd pipes */
2082 utils_close_pipe(relay_cmd_pipe);
2083 if (err) {
2084 DBG("Thread exited with error");
2085 }
2086 DBG("Worker thread cleanup complete");
2087 free(data_buffer);
2088 stop_threads();
2089 rcu_unregister_thread();
2090 return NULL;
2091 }
2092
2093 /*
2094 * Create the relay command pipe to wake thread_manage_apps.
2095 * Closed in cleanup().
2096 */
2097 static int create_relay_cmd_pipe(void)
2098 {
2099 int ret;
2100
2101 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2102
2103 return ret;
2104 }
2105
2106 /*
2107 * main
2108 */
2109 int main(int argc, char **argv)
2110 {
2111 int ret = 0;
2112 void *status;
2113
2114 /* Create thread quit pipe */
2115 if ((ret = init_thread_quit_pipe()) < 0) {
2116 goto error;
2117 }
2118
2119 /* Parse arguments */
2120 progname = argv[0];
2121 if ((ret = parse_args(argc, argv)) < 0) {
2122 goto exit;
2123 }
2124
2125 if ((ret = set_signal_handler()) < 0) {
2126 goto exit;
2127 }
2128
2129 /* Try to create directory if -o, --output is specified. */
2130 if (opt_output_path) {
2131 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2132 if (ret < 0) {
2133 ERR("Unable to create %s", opt_output_path);
2134 goto exit;
2135 }
2136 }
2137
2138 /* Daemonize */
2139 if (opt_daemon) {
2140 ret = daemon(0, 0);
2141 if (ret < 0) {
2142 PERROR("daemon");
2143 goto exit;
2144 }
2145 }
2146
2147 /* Check if daemon is UID = 0 */
2148 is_root = !getuid();
2149
2150 if (!is_root) {
2151 if (control_uri->port < 1024 || data_uri->port < 1024) {
2152 ERR("Need to be root to use ports < 1024");
2153 ret = -1;
2154 goto exit;
2155 }
2156 }
2157
2158 /* Setup the thread apps communication pipe. */
2159 if ((ret = create_relay_cmd_pipe()) < 0) {
2160 goto exit;
2161 }
2162
2163 /* Init relay command queue. */
2164 cds_wfq_init(&relay_cmd_queue.queue);
2165
2166 /* Set up max poll set size */
2167 lttng_poll_set_max_size();
2168
2169 /* Setup the dispatcher thread */
2170 ret = pthread_create(&dispatcher_thread, NULL,
2171 relay_thread_dispatcher, (void *) NULL);
2172 if (ret != 0) {
2173 PERROR("pthread_create dispatcher");
2174 goto exit_dispatcher;
2175 }
2176
2177 /* Setup the worker thread */
2178 ret = pthread_create(&worker_thread, NULL,
2179 relay_thread_worker, (void *) NULL);
2180 if (ret != 0) {
2181 PERROR("pthread_create worker");
2182 goto exit_worker;
2183 }
2184
2185 /* Setup the listener thread */
2186 ret = pthread_create(&listener_thread, NULL,
2187 relay_thread_listener, (void *) NULL);
2188 if (ret != 0) {
2189 PERROR("pthread_create listener");
2190 goto exit_listener;
2191 }
2192
2193 exit_listener:
2194 ret = pthread_join(listener_thread, &status);
2195 if (ret != 0) {
2196 PERROR("pthread_join");
2197 goto error; /* join error, exit without cleanup */
2198 }
2199
2200 exit_worker:
2201 ret = pthread_join(worker_thread, &status);
2202 if (ret != 0) {
2203 PERROR("pthread_join");
2204 goto error; /* join error, exit without cleanup */
2205 }
2206
2207 exit_dispatcher:
2208 ret = pthread_join(dispatcher_thread, &status);
2209 if (ret != 0) {
2210 PERROR("pthread_join");
2211 goto error; /* join error, exit without cleanup */
2212 }
2213
2214 exit:
2215 cleanup();
2216 if (!ret) {
2217 exit(EXIT_SUCCESS);
2218 }
2219
2220 error:
2221 exit(EXIT_FAILURE);
2222 }
This page took 0.131866 seconds and 4 git commands to generate.