Relayd add_stream command handle tracefile rotation
[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/sessiond-comm/relayd.h>
51 #include <common/uri.h>
52 #include <common/utils.h>
53
54 #include "cmd.h"
55 #include "utils.h"
56 #include "lttng-relayd.h"
57
58 /* command line options */
59 char *opt_output_path;
60 static int opt_daemon;
61 static struct lttng_uri *control_uri;
62 static struct lttng_uri *data_uri;
63
64 const char *progname;
65 static int is_root; /* Set to 1 if the daemon is running as root */
66
67 /*
68 * Quit pipe for all threads. This permits a single cancellation point
69 * for all threads when receiving an event on the pipe.
70 */
71 static int thread_quit_pipe[2] = { -1, -1 };
72
73 /*
74 * This pipe is used to inform the worker thread that a command is queued and
75 * ready to be processed.
76 */
77 static int relay_cmd_pipe[2] = { -1, -1 };
78
79 /* Shared between threads */
80 static int dispatch_thread_exit;
81
82 static pthread_t listener_thread;
83 static pthread_t dispatcher_thread;
84 static pthread_t worker_thread;
85
86 static uint64_t last_relay_stream_id;
87 static uint64_t last_relay_session_id;
88
89 /*
90 * Relay command queue.
91 *
92 * The relay_thread_listener and relay_thread_dispatcher communicate with this
93 * queue.
94 */
95 static struct relay_cmd_queue relay_cmd_queue;
96
97 /* buffer allocated at startup, used to store the trace data */
98 static char *data_buffer;
99 static unsigned int data_buffer_size;
100
101 /*
102 * usage function on stderr
103 */
104 static
105 void usage(void)
106 {
107 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
108 fprintf(stderr, " -h, --help Display this usage.\n");
109 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
110 fprintf(stderr, " -C, --control-port Control port listening (URI)\n");
111 fprintf(stderr, " -D, --data-port Data port listening (URI)\n");
112 fprintf(stderr, " -o, --output Output path for traces (PATH)\n");
113 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
114 }
115
116 static
117 int parse_args(int argc, char **argv)
118 {
119 int c;
120 int ret = 0;
121 char *default_address;
122
123 static struct option long_options[] = {
124 { "control-port", 1, 0, 'C', },
125 { "data-port", 1, 0, 'D', },
126 { "daemonize", 0, 0, 'd', },
127 { "help", 0, 0, 'h', },
128 { "output", 1, 0, 'o', },
129 { "verbose", 0, 0, 'v', },
130 { NULL, 0, 0, 0, },
131 };
132
133 while (1) {
134 int option_index = 0;
135 c = getopt_long(argc, argv, "dhv" "C:D:o:",
136 long_options, &option_index);
137 if (c == -1) {
138 break;
139 }
140
141 switch (c) {
142 case 0:
143 fprintf(stderr, "option %s", long_options[option_index].name);
144 if (optarg) {
145 fprintf(stderr, " with arg %s\n", optarg);
146 }
147 break;
148 case 'C':
149 ret = uri_parse(optarg, &control_uri);
150 if (ret < 0) {
151 ERR("Invalid control URI specified");
152 goto exit;
153 }
154 if (control_uri->port == 0) {
155 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
156 }
157 break;
158 case 'D':
159 ret = uri_parse(optarg, &data_uri);
160 if (ret < 0) {
161 ERR("Invalid data URI specified");
162 goto exit;
163 }
164 if (data_uri->port == 0) {
165 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
166 }
167 break;
168 case 'd':
169 opt_daemon = 1;
170 break;
171 case 'h':
172 usage();
173 exit(EXIT_FAILURE);
174 case 'o':
175 ret = asprintf(&opt_output_path, "%s", optarg);
176 if (ret < 0) {
177 PERROR("asprintf opt_output_path");
178 goto exit;
179 }
180 break;
181 case 'v':
182 /* Verbose level can increase using multiple -v */
183 lttng_opt_verbose += 1;
184 break;
185 default:
186 /* Unknown option or other error.
187 * Error is printed by getopt, just return */
188 ret = -1;
189 goto exit;
190 }
191 }
192
193 /* assign default values */
194 if (control_uri == NULL) {
195 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
196 DEFAULT_NETWORK_CONTROL_PORT);
197 if (ret < 0) {
198 PERROR("asprintf default data address");
199 goto exit;
200 }
201
202 ret = uri_parse(default_address, &control_uri);
203 free(default_address);
204 if (ret < 0) {
205 ERR("Invalid control URI specified");
206 goto exit;
207 }
208 }
209 if (data_uri == NULL) {
210 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
211 DEFAULT_NETWORK_DATA_PORT);
212 if (ret < 0) {
213 PERROR("asprintf default data address");
214 goto exit;
215 }
216
217 ret = uri_parse(default_address, &data_uri);
218 free(default_address);
219 if (ret < 0) {
220 ERR("Invalid data URI specified");
221 goto exit;
222 }
223 }
224
225 exit:
226 return ret;
227 }
228
229 /*
230 * Cleanup the daemon
231 */
232 static
233 void cleanup(void)
234 {
235 DBG("Cleaning up");
236
237 /* free the dynamically allocated opt_output_path */
238 free(opt_output_path);
239
240 /* Close thread quit pipes */
241 utils_close_pipe(thread_quit_pipe);
242
243 uri_free(control_uri);
244 uri_free(data_uri);
245 }
246
247 /*
248 * Write to writable pipe used to notify a thread.
249 */
250 static
251 int notify_thread_pipe(int wpipe)
252 {
253 int ret;
254
255 do {
256 ret = write(wpipe, "!", 1);
257 } while (ret < 0 && errno == EINTR);
258 if (ret < 0 || ret != 1) {
259 PERROR("write poll pipe");
260 }
261
262 return ret;
263 }
264
265 /*
266 * Stop all threads by closing the thread quit pipe.
267 */
268 static
269 void stop_threads(void)
270 {
271 int ret;
272
273 /* Stopping all threads */
274 DBG("Terminating all threads");
275 ret = notify_thread_pipe(thread_quit_pipe[1]);
276 if (ret < 0) {
277 ERR("write error on thread quit pipe");
278 }
279
280 /* Dispatch thread */
281 CMM_STORE_SHARED(dispatch_thread_exit, 1);
282 futex_nto1_wake(&relay_cmd_queue.futex);
283 }
284
285 /*
286 * Signal handler for the daemon
287 *
288 * Simply stop all worker threads, leaving main() return gracefully after
289 * joining all threads and calling cleanup().
290 */
291 static
292 void sighandler(int sig)
293 {
294 switch (sig) {
295 case SIGPIPE:
296 DBG("SIGPIPE caught");
297 return;
298 case SIGINT:
299 DBG("SIGINT caught");
300 stop_threads();
301 break;
302 case SIGTERM:
303 DBG("SIGTERM caught");
304 stop_threads();
305 break;
306 default:
307 break;
308 }
309 }
310
311 /*
312 * Setup signal handler for :
313 * SIGINT, SIGTERM, SIGPIPE
314 */
315 static
316 int set_signal_handler(void)
317 {
318 int ret = 0;
319 struct sigaction sa;
320 sigset_t sigset;
321
322 if ((ret = sigemptyset(&sigset)) < 0) {
323 PERROR("sigemptyset");
324 return ret;
325 }
326
327 sa.sa_handler = sighandler;
328 sa.sa_mask = sigset;
329 sa.sa_flags = 0;
330 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
331 PERROR("sigaction");
332 return ret;
333 }
334
335 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
336 PERROR("sigaction");
337 return ret;
338 }
339
340 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
341 PERROR("sigaction");
342 return ret;
343 }
344
345 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
346
347 return ret;
348 }
349
350 /*
351 * Init thread quit pipe.
352 *
353 * Return -1 on error or 0 if all pipes are created.
354 */
355 static
356 int init_thread_quit_pipe(void)
357 {
358 int ret;
359
360 ret = utils_create_pipe_cloexec(thread_quit_pipe);
361
362 return ret;
363 }
364
365 /*
366 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
367 */
368 static
369 int create_thread_poll_set(struct lttng_poll_event *events, int size)
370 {
371 int ret;
372
373 if (events == NULL || size == 0) {
374 ret = -1;
375 goto error;
376 }
377
378 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
379 if (ret < 0) {
380 goto error;
381 }
382
383 /* Add quit pipe */
384 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
385 if (ret < 0) {
386 goto error;
387 }
388
389 return 0;
390
391 error:
392 return ret;
393 }
394
395 /*
396 * Check if the thread quit pipe was triggered.
397 *
398 * Return 1 if it was triggered else 0;
399 */
400 static
401 int check_thread_quit_pipe(int fd, uint32_t events)
402 {
403 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
404 return 1;
405 }
406
407 return 0;
408 }
409
410 /*
411 * Create and init socket from uri.
412 */
413 static
414 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
415 {
416 int ret;
417 struct lttcomm_sock *sock = NULL;
418
419 sock = lttcomm_alloc_sock_from_uri(uri);
420 if (sock == NULL) {
421 ERR("Allocating socket");
422 goto error;
423 }
424
425 ret = lttcomm_create_sock(sock);
426 if (ret < 0) {
427 goto error;
428 }
429 DBG("Listening on sock %d", sock->fd);
430
431 ret = sock->ops->bind(sock);
432 if (ret < 0) {
433 goto error;
434 }
435
436 ret = sock->ops->listen(sock, -1);
437 if (ret < 0) {
438 goto error;
439
440 }
441
442 return sock;
443
444 error:
445 if (sock) {
446 lttcomm_destroy_sock(sock);
447 }
448 return NULL;
449 }
450
451 /*
452 * Return nonzero if stream needs to be closed.
453 */
454 static
455 int close_stream_check(struct relay_stream *stream)
456 {
457
458 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
459 /*
460 * We are about to close the stream so set the data pending flag to 1
461 * which will make the end data pending command skip the stream which
462 * is now closed and ready. Note that after proceeding to a file close,
463 * the written file is ready for reading.
464 */
465 stream->data_pending_check_done = 1;
466 return 1;
467 }
468 return 0;
469 }
470
471 /*
472 * This thread manages the listening for new connections on the network
473 */
474 static
475 void *relay_thread_listener(void *data)
476 {
477 int i, ret, pollfd, err = -1;
478 int val = 1;
479 uint32_t revents, nb_fd;
480 struct lttng_poll_event events;
481 struct lttcomm_sock *control_sock, *data_sock;
482
483 DBG("[thread] Relay listener started");
484
485 control_sock = relay_init_sock(control_uri);
486 if (!control_sock) {
487 goto error_sock_control;
488 }
489
490 data_sock = relay_init_sock(data_uri);
491 if (!data_sock) {
492 goto error_sock_relay;
493 }
494
495 /*
496 * Pass 3 as size here for the thread quit pipe, control and data socket.
497 */
498 ret = create_thread_poll_set(&events, 3);
499 if (ret < 0) {
500 goto error_create_poll;
501 }
502
503 /* Add the control socket */
504 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
505 if (ret < 0) {
506 goto error_poll_add;
507 }
508
509 /* Add the data socket */
510 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
511 if (ret < 0) {
512 goto error_poll_add;
513 }
514
515 while (1) {
516 DBG("Listener accepting connections");
517
518 restart:
519 ret = lttng_poll_wait(&events, -1);
520 if (ret < 0) {
521 /*
522 * Restart interrupted system call.
523 */
524 if (errno == EINTR) {
525 goto restart;
526 }
527 goto error;
528 }
529
530 nb_fd = ret;
531
532 DBG("Relay new connection received");
533 for (i = 0; i < nb_fd; i++) {
534 /* Fetch once the poll data */
535 revents = LTTNG_POLL_GETEV(&events, i);
536 pollfd = LTTNG_POLL_GETFD(&events, i);
537
538 /* Thread quit pipe has been closed. Killing thread. */
539 ret = check_thread_quit_pipe(pollfd, revents);
540 if (ret) {
541 err = 0;
542 goto exit;
543 }
544
545 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
546 ERR("socket poll error");
547 goto error;
548 } else if (revents & LPOLLIN) {
549 /*
550 * Get allocated in this thread,
551 * enqueued to a global queue, dequeued
552 * and freed in the worker thread.
553 */
554 struct relay_command *relay_cmd;
555 struct lttcomm_sock *newsock;
556
557 relay_cmd = zmalloc(sizeof(struct relay_command));
558 if (relay_cmd == NULL) {
559 PERROR("relay command zmalloc");
560 goto error;
561 }
562
563 if (pollfd == data_sock->fd) {
564 newsock = data_sock->ops->accept(data_sock);
565 if (!newsock) {
566 PERROR("accepting data sock");
567 free(relay_cmd);
568 goto error;
569 }
570 relay_cmd->type = RELAY_DATA;
571 DBG("Relay data connection accepted, socket %d", newsock->fd);
572 } else {
573 assert(pollfd == control_sock->fd);
574 newsock = control_sock->ops->accept(control_sock);
575 if (!newsock) {
576 PERROR("accepting control sock");
577 free(relay_cmd);
578 goto error;
579 }
580 relay_cmd->type = RELAY_CONTROL;
581 DBG("Relay control connection accepted, socket %d", newsock->fd);
582 }
583 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
584 &val, sizeof(int));
585 if (ret < 0) {
586 PERROR("setsockopt inet");
587 lttcomm_destroy_sock(newsock);
588 free(relay_cmd);
589 goto error;
590 }
591 relay_cmd->sock = newsock;
592 /*
593 * Lock free enqueue the request.
594 */
595 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
596
597 /*
598 * Wake the dispatch queue futex. Implicit memory
599 * barrier with the exchange in cds_wfq_enqueue.
600 */
601 futex_nto1_wake(&relay_cmd_queue.futex);
602 }
603 }
604 }
605
606 exit:
607 error:
608 error_poll_add:
609 lttng_poll_clean(&events);
610 error_create_poll:
611 if (data_sock->fd >= 0) {
612 ret = data_sock->ops->close(data_sock);
613 if (ret) {
614 PERROR("close");
615 }
616 }
617 lttcomm_destroy_sock(data_sock);
618 error_sock_relay:
619 if (control_sock->fd >= 0) {
620 ret = control_sock->ops->close(control_sock);
621 if (ret) {
622 PERROR("close");
623 }
624 }
625 lttcomm_destroy_sock(control_sock);
626 error_sock_control:
627 if (err) {
628 DBG("Thread exited with error");
629 }
630 DBG("Relay listener thread cleanup complete");
631 stop_threads();
632 return NULL;
633 }
634
635 /*
636 * This thread manages the dispatching of the requests to worker threads
637 */
638 static
639 void *relay_thread_dispatcher(void *data)
640 {
641 int ret;
642 struct cds_wfq_node *node;
643 struct relay_command *relay_cmd = NULL;
644
645 DBG("[thread] Relay dispatcher started");
646
647 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
648 /* Atomically prepare the queue futex */
649 futex_nto1_prepare(&relay_cmd_queue.futex);
650
651 do {
652 /* Dequeue commands */
653 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
654 if (node == NULL) {
655 DBG("Woken up but nothing in the relay command queue");
656 /* Continue thread execution */
657 break;
658 }
659
660 relay_cmd = caa_container_of(node, struct relay_command, node);
661 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
662
663 /*
664 * Inform worker thread of the new request. This
665 * call is blocking so we can be assured that the data will be read
666 * at some point in time or wait to the end of the world :)
667 */
668 do {
669 ret = write(relay_cmd_pipe[1], relay_cmd,
670 sizeof(struct relay_command));
671 } while (ret < 0 && errno == EINTR);
672 free(relay_cmd);
673 if (ret < 0 || ret != sizeof(struct relay_command)) {
674 PERROR("write cmd pipe");
675 goto error;
676 }
677 } while (node != NULL);
678
679 /* Futex wait on queue. Blocking call on futex() */
680 futex_nto1_wait(&relay_cmd_queue.futex);
681 }
682
683 error:
684 DBG("Dispatch thread dying");
685 stop_threads();
686 return NULL;
687 }
688
689 /*
690 * Get stream from stream id.
691 * Need to be called with RCU read-side lock held.
692 */
693 static
694 struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id,
695 struct lttng_ht *streams_ht)
696 {
697 struct lttng_ht_node_ulong *node;
698 struct lttng_ht_iter iter;
699 struct relay_stream *ret;
700
701 lttng_ht_lookup(streams_ht,
702 (void *)((unsigned long) stream_id),
703 &iter);
704 node = lttng_ht_iter_get_node_ulong(&iter);
705 if (node == NULL) {
706 DBG("Relay stream %" PRIu64 " not found", stream_id);
707 ret = NULL;
708 goto end;
709 }
710
711 ret = caa_container_of(node, struct relay_stream, stream_n);
712
713 end:
714 return ret;
715 }
716
717 static
718 void deferred_free_stream(struct rcu_head *head)
719 {
720 struct relay_stream *stream =
721 caa_container_of(head, struct relay_stream, rcu_node);
722 free(stream->path_name);
723 free(stream->channel_name);
724 free(stream);
725 }
726
727 /*
728 * relay_delete_session: Free all memory associated with a session and
729 * close all the FDs
730 */
731 static
732 void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht)
733 {
734 struct lttng_ht_iter iter;
735 struct lttng_ht_node_ulong *node;
736 struct relay_stream *stream;
737 int ret;
738
739 if (!cmd->session) {
740 return;
741 }
742
743 DBG("Relay deleting session %" PRIu64, cmd->session->id);
744
745 rcu_read_lock();
746 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, node, node) {
747 node = lttng_ht_iter_get_node_ulong(&iter);
748 if (node) {
749 stream = caa_container_of(node,
750 struct relay_stream, stream_n);
751 if (stream->session == cmd->session) {
752 ret = close(stream->fd);
753 if (ret < 0) {
754 PERROR("close stream fd on delete session");
755 }
756 ret = lttng_ht_del(streams_ht, &iter);
757 assert(!ret);
758 call_rcu(&stream->rcu_node,
759 deferred_free_stream);
760 }
761 }
762 }
763 rcu_read_unlock();
764
765 free(cmd->session);
766 }
767
768 /*
769 * Handle the RELAYD_CREATE_SESSION command.
770 *
771 * On success, send back the session id or else return a negative value.
772 */
773 static
774 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
775 struct relay_command *cmd)
776 {
777 int ret = 0, send_ret;
778 struct relay_session *session;
779 struct lttcomm_relayd_status_session reply;
780
781 assert(recv_hdr);
782 assert(cmd);
783
784 memset(&reply, 0, sizeof(reply));
785
786 session = zmalloc(sizeof(struct relay_session));
787 if (session == NULL) {
788 PERROR("relay session zmalloc");
789 ret = -1;
790 goto error;
791 }
792
793 session->id = ++last_relay_session_id;
794 session->sock = cmd->sock;
795 cmd->session = session;
796
797 reply.session_id = htobe64(session->id);
798
799 DBG("Created session %" PRIu64, session->id);
800
801 error:
802 if (ret < 0) {
803 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
804 } else {
805 reply.ret_code = htobe32(LTTNG_OK);
806 }
807
808 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
809 if (send_ret < 0) {
810 ERR("Relayd sending session id");
811 ret = send_ret;
812 }
813
814 return ret;
815 }
816
817 /*
818 * relay_add_stream: allocate a new stream for a session
819 */
820 static
821 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
822 struct relay_command *cmd, struct lttng_ht *streams_ht)
823 {
824 struct relay_session *session = cmd->session;
825 struct relay_stream *stream = NULL;
826 struct lttcomm_relayd_status_stream reply;
827 int ret, send_ret;
828
829 if (!session || cmd->version_check_done == 0) {
830 ERR("Trying to add a stream before version check");
831 ret = -1;
832 goto end_no_session;
833 }
834
835 stream = zmalloc(sizeof(struct relay_stream));
836 if (stream == NULL) {
837 PERROR("relay stream zmalloc");
838 ret = -1;
839 goto end_no_session;
840 }
841
842 switch (cmd->minor) {
843 case 1: /* LTTng sessiond 2.1 */
844 ret = cmd_recv_stream_2_1(cmd, stream);
845 break;
846 case 2: /* LTTng sessiond 2.2 */
847 default:
848 ret = cmd_recv_stream_2_2(cmd, stream);
849 break;
850 }
851 if (ret < 0) {
852 goto err_free_stream;
853 }
854
855 rcu_read_lock();
856 stream->stream_handle = ++last_relay_stream_id;
857 stream->prev_seq = -1ULL;
858 stream->session = session;
859
860 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
861 if (ret < 0) {
862 ERR("relay creating output directory");
863 goto end;
864 }
865
866 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
867 stream->tracefile_size, 0, getuid(), getgid());
868 if (ret < 0) {
869 ERR("Create output file");
870 goto end;
871 }
872 stream->fd = ret;
873 if (stream->tracefile_size) {
874 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
875 } else {
876 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
877 }
878
879 lttng_ht_node_init_ulong(&stream->stream_n,
880 (unsigned long) stream->stream_handle);
881 lttng_ht_add_unique_ulong(streams_ht,
882 &stream->stream_n);
883
884 DBG("Relay new stream added %s", stream->channel_name);
885
886 end:
887 reply.handle = htobe64(stream->stream_handle);
888 /* send the session id to the client or a negative return code on error */
889 if (ret < 0) {
890 reply.ret_code = htobe32(LTTNG_ERR_UNK);
891 /* stream was not properly added to the ht, so free it */
892 free(stream);
893 } else {
894 reply.ret_code = htobe32(LTTNG_OK);
895 }
896
897 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
898 sizeof(struct lttcomm_relayd_status_stream), 0);
899 if (send_ret < 0) {
900 ERR("Relay sending stream id");
901 ret = send_ret;
902 }
903 rcu_read_unlock();
904
905 end_no_session:
906 return ret;
907
908 err_free_stream:
909 free(stream->path_name);
910 free(stream->channel_name);
911 free(stream);
912 return ret;
913 }
914
915 /*
916 * relay_close_stream: close a specific stream
917 */
918 static
919 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
920 struct relay_command *cmd, struct lttng_ht *streams_ht)
921 {
922 struct relay_session *session = cmd->session;
923 struct lttcomm_relayd_close_stream stream_info;
924 struct lttcomm_relayd_generic_reply reply;
925 struct relay_stream *stream;
926 int ret, send_ret;
927 struct lttng_ht_iter iter;
928
929 DBG("Close stream received");
930
931 if (!session || cmd->version_check_done == 0) {
932 ERR("Trying to close a stream before version check");
933 ret = -1;
934 goto end_no_session;
935 }
936
937 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
938 sizeof(struct lttcomm_relayd_close_stream), 0);
939 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
940 if (ret == 0) {
941 /* Orderly shutdown. Not necessary to print an error. */
942 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
943 } else {
944 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
945 }
946 ret = -1;
947 goto end_no_session;
948 }
949
950 rcu_read_lock();
951 stream = relay_stream_from_stream_id(be64toh(stream_info.stream_id),
952 streams_ht);
953 if (!stream) {
954 ret = -1;
955 goto end_unlock;
956 }
957
958 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
959 stream->close_flag = 1;
960
961 if (close_stream_check(stream)) {
962 int delret;
963
964 delret = close(stream->fd);
965 if (delret < 0) {
966 PERROR("close stream");
967 }
968 iter.iter.node = &stream->stream_n.node;
969 delret = lttng_ht_del(streams_ht, &iter);
970 assert(!delret);
971 call_rcu(&stream->rcu_node,
972 deferred_free_stream);
973 DBG("Closed tracefile %d from close stream", stream->fd);
974 }
975
976 end_unlock:
977 rcu_read_unlock();
978
979 if (ret < 0) {
980 reply.ret_code = htobe32(LTTNG_ERR_UNK);
981 } else {
982 reply.ret_code = htobe32(LTTNG_OK);
983 }
984 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
985 sizeof(struct lttcomm_relayd_generic_reply), 0);
986 if (send_ret < 0) {
987 ERR("Relay sending stream id");
988 ret = send_ret;
989 }
990
991 end_no_session:
992 return ret;
993 }
994
995 /*
996 * relay_unknown_command: send -1 if received unknown command
997 */
998 static
999 void relay_unknown_command(struct relay_command *cmd)
1000 {
1001 struct lttcomm_relayd_generic_reply reply;
1002 int ret;
1003
1004 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1005 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1006 sizeof(struct lttcomm_relayd_generic_reply), 0);
1007 if (ret < 0) {
1008 ERR("Relay sending unknown command");
1009 }
1010 }
1011
1012 /*
1013 * relay_start: send an acknowledgment to the client to tell if we are
1014 * ready to receive data. We are ready if a session is established.
1015 */
1016 static
1017 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1018 struct relay_command *cmd)
1019 {
1020 int ret = htobe32(LTTNG_OK);
1021 struct lttcomm_relayd_generic_reply reply;
1022 struct relay_session *session = cmd->session;
1023
1024 if (!session) {
1025 DBG("Trying to start the streaming without a session established");
1026 ret = htobe32(LTTNG_ERR_UNK);
1027 }
1028
1029 reply.ret_code = ret;
1030 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1031 sizeof(struct lttcomm_relayd_generic_reply), 0);
1032 if (ret < 0) {
1033 ERR("Relay sending start ack");
1034 }
1035
1036 return ret;
1037 }
1038
1039 /*
1040 * Append padding to the file pointed by the file descriptor fd.
1041 */
1042 static int write_padding_to_file(int fd, uint32_t size)
1043 {
1044 int ret = 0;
1045 char *zeros;
1046
1047 if (size == 0) {
1048 goto end;
1049 }
1050
1051 zeros = zmalloc(size);
1052 if (zeros == NULL) {
1053 PERROR("zmalloc zeros for padding");
1054 ret = -1;
1055 goto end;
1056 }
1057
1058 do {
1059 ret = write(fd, zeros, size);
1060 } while (ret < 0 && errno == EINTR);
1061 if (ret < 0 || ret != size) {
1062 PERROR("write padding to file");
1063 }
1064
1065 free(zeros);
1066
1067 end:
1068 return ret;
1069 }
1070
1071 /*
1072 * relay_recv_metadata: receive the metada for the session.
1073 */
1074 static
1075 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1076 struct relay_command *cmd, struct lttng_ht *streams_ht)
1077 {
1078 int ret = htobe32(LTTNG_OK);
1079 struct relay_session *session = cmd->session;
1080 struct lttcomm_relayd_metadata_payload *metadata_struct;
1081 struct relay_stream *metadata_stream;
1082 uint64_t data_size, payload_size;
1083
1084 if (!session) {
1085 ERR("Metadata sent before version check");
1086 ret = -1;
1087 goto end;
1088 }
1089
1090 data_size = payload_size = be64toh(recv_hdr->data_size);
1091 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1092 ERR("Incorrect data size");
1093 ret = -1;
1094 goto end;
1095 }
1096 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1097
1098 if (data_buffer_size < data_size) {
1099 /* In case the realloc fails, we can free the memory */
1100 char *tmp_data_ptr;
1101
1102 tmp_data_ptr = realloc(data_buffer, data_size);
1103 if (!tmp_data_ptr) {
1104 ERR("Allocating data buffer");
1105 free(data_buffer);
1106 ret = -1;
1107 goto end;
1108 }
1109 data_buffer = tmp_data_ptr;
1110 data_buffer_size = data_size;
1111 }
1112 memset(data_buffer, 0, data_size);
1113 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1114 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1115 if (ret < 0 || ret != data_size) {
1116 if (ret == 0) {
1117 /* Orderly shutdown. Not necessary to print an error. */
1118 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1119 } else {
1120 ERR("Relay didn't receive the whole metadata");
1121 }
1122 ret = -1;
1123 goto end;
1124 }
1125 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1126
1127 rcu_read_lock();
1128 metadata_stream = relay_stream_from_stream_id(
1129 be64toh(metadata_struct->stream_id), streams_ht);
1130 if (!metadata_stream) {
1131 ret = -1;
1132 goto end_unlock;
1133 }
1134
1135 do {
1136 ret = write(metadata_stream->fd, metadata_struct->payload,
1137 payload_size);
1138 } while (ret < 0 && errno == EINTR);
1139 if (ret < 0 || ret != payload_size) {
1140 ERR("Relay error writing metadata on file");
1141 ret = -1;
1142 goto end_unlock;
1143 }
1144
1145 ret = write_padding_to_file(metadata_stream->fd,
1146 be32toh(metadata_struct->padding_size));
1147 if (ret < 0) {
1148 goto end_unlock;
1149 }
1150
1151 DBG2("Relay metadata written");
1152
1153 end_unlock:
1154 rcu_read_unlock();
1155 end:
1156 return ret;
1157 }
1158
1159 /*
1160 * relay_send_version: send relayd version number
1161 */
1162 static
1163 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1164 struct relay_command *cmd, struct lttng_ht *streams_ht)
1165 {
1166 int ret;
1167 struct lttcomm_relayd_version reply, msg;
1168
1169 assert(cmd);
1170
1171 cmd->version_check_done = 1;
1172
1173 /* Get version from the other side. */
1174 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1175 if (ret < 0 || ret != sizeof(msg)) {
1176 if (ret == 0) {
1177 /* Orderly shutdown. Not necessary to print an error. */
1178 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1179 } else {
1180 ERR("Relay failed to receive the version values.");
1181 }
1182 ret = -1;
1183 goto end;
1184 }
1185
1186 ret = sscanf(VERSION, "%10u.%10u", &reply.major, &reply.minor);
1187 if (ret < 2) {
1188 ERR("Error in scanning version");
1189 ret = -1;
1190 goto end;
1191 }
1192
1193 /* Major versions must be the same */
1194 if (reply.major != be32toh(msg.major)) {
1195 DBG("Incompatible major versions (%u vs %u), deleting session",
1196 reply.major, be32toh(msg.major));
1197 relay_delete_session(cmd, streams_ht);
1198 ret = 0;
1199 goto end;
1200 }
1201
1202 cmd->major = reply.major;
1203 /* We adapt to the lowest compatible version */
1204 if (reply.minor <= be32toh(msg.minor)) {
1205 cmd->minor = reply.minor;
1206 } else {
1207 cmd->minor = be32toh(msg.minor);
1208 }
1209
1210 reply.major = htobe32(reply.major);
1211 reply.minor = htobe32(reply.minor);
1212 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1213 sizeof(struct lttcomm_relayd_version), 0);
1214 if (ret < 0) {
1215 ERR("Relay sending version");
1216 }
1217
1218 DBG("Version check done using protocol %u.%u", cmd->major,
1219 cmd->minor);
1220
1221 end:
1222 return ret;
1223 }
1224
1225 /*
1226 * Check for data pending for a given stream id from the session daemon.
1227 */
1228 static
1229 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1230 struct relay_command *cmd, struct lttng_ht *streams_ht)
1231 {
1232 struct relay_session *session = cmd->session;
1233 struct lttcomm_relayd_data_pending msg;
1234 struct lttcomm_relayd_generic_reply reply;
1235 struct relay_stream *stream;
1236 int ret;
1237 uint64_t last_net_seq_num, stream_id;
1238
1239 DBG("Data pending command received");
1240
1241 if (!session || cmd->version_check_done == 0) {
1242 ERR("Trying to check for data before version check");
1243 ret = -1;
1244 goto end_no_session;
1245 }
1246
1247 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1248 if (ret < sizeof(msg)) {
1249 if (ret == 0) {
1250 /* Orderly shutdown. Not necessary to print an error. */
1251 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1252 } else {
1253 ERR("Relay didn't receive valid data_pending struct size : %d",
1254 ret);
1255 }
1256 ret = -1;
1257 goto end_no_session;
1258 }
1259
1260 stream_id = be64toh(msg.stream_id);
1261 last_net_seq_num = be64toh(msg.last_net_seq_num);
1262
1263 rcu_read_lock();
1264 stream = relay_stream_from_stream_id(stream_id, streams_ht);
1265 if (stream == NULL) {
1266 ret = -1;
1267 goto end_unlock;
1268 }
1269
1270 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1271 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1272 last_net_seq_num);
1273
1274 /* Avoid wrapping issue */
1275 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1276 /* Data has in fact been written and is NOT pending */
1277 ret = 0;
1278 } else {
1279 /* Data still being streamed thus pending */
1280 ret = 1;
1281 }
1282
1283 /* Pending check is now done. */
1284 stream->data_pending_check_done = 1;
1285
1286 end_unlock:
1287 rcu_read_unlock();
1288
1289 reply.ret_code = htobe32(ret);
1290 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1291 if (ret < 0) {
1292 ERR("Relay data pending ret code failed");
1293 }
1294
1295 end_no_session:
1296 return ret;
1297 }
1298
1299 /*
1300 * Wait for the control socket to reach a quiescent state.
1301 *
1302 * Note that for now, when receiving this command from the session daemon, this
1303 * means that every subsequent commands or data received on the control socket
1304 * has been handled. So, this is why we simply return OK here.
1305 */
1306 static
1307 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1308 struct relay_command *cmd, struct lttng_ht *streams_ht)
1309 {
1310 int ret;
1311 uint64_t stream_id;
1312 struct relay_stream *stream;
1313 struct lttng_ht_iter iter;
1314 struct lttcomm_relayd_quiescent_control msg;
1315 struct lttcomm_relayd_generic_reply reply;
1316
1317 DBG("Checking quiescent state on control socket");
1318
1319 if (!cmd->session || cmd->version_check_done == 0) {
1320 ERR("Trying to check for data before version check");
1321 ret = -1;
1322 goto end_no_session;
1323 }
1324
1325 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1326 if (ret < sizeof(msg)) {
1327 if (ret == 0) {
1328 /* Orderly shutdown. Not necessary to print an error. */
1329 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1330 } else {
1331 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1332 ret);
1333 }
1334 ret = -1;
1335 goto end_no_session;
1336 }
1337
1338 stream_id = be64toh(msg.stream_id);
1339
1340 rcu_read_lock();
1341 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) {
1342 if (stream->stream_handle == stream_id) {
1343 stream->data_pending_check_done = 1;
1344 DBG("Relay quiescent control pending flag set to %" PRIu64,
1345 stream_id);
1346 break;
1347 }
1348 }
1349 rcu_read_unlock();
1350
1351 reply.ret_code = htobe32(LTTNG_OK);
1352 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1353 if (ret < 0) {
1354 ERR("Relay data quiescent control ret code failed");
1355 }
1356
1357 end_no_session:
1358 return ret;
1359 }
1360
1361 /*
1362 * Initialize a data pending command. This means that a client is about to ask
1363 * for data pending for each stream he/she holds. Simply iterate over all
1364 * streams of a session and set the data_pending_check_done flag.
1365 *
1366 * This command returns to the client a LTTNG_OK code.
1367 */
1368 static
1369 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1370 struct relay_command *cmd, struct lttng_ht *streams_ht)
1371 {
1372 int ret;
1373 struct lttng_ht_iter iter;
1374 struct lttcomm_relayd_begin_data_pending msg;
1375 struct lttcomm_relayd_generic_reply reply;
1376 struct relay_stream *stream;
1377 uint64_t session_id;
1378
1379 assert(recv_hdr);
1380 assert(cmd);
1381 assert(streams_ht);
1382
1383 DBG("Init streams for data pending");
1384
1385 if (!cmd->session || cmd->version_check_done == 0) {
1386 ERR("Trying to check for data before version check");
1387 ret = -1;
1388 goto end_no_session;
1389 }
1390
1391 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1392 if (ret < sizeof(msg)) {
1393 if (ret == 0) {
1394 /* Orderly shutdown. Not necessary to print an error. */
1395 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1396 } else {
1397 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1398 ret);
1399 }
1400 ret = -1;
1401 goto end_no_session;
1402 }
1403
1404 session_id = be64toh(msg.session_id);
1405
1406 /*
1407 * Iterate over all streams to set the begin data pending flag. For now, the
1408 * streams are indexed by stream handle so we have to iterate over all
1409 * streams to find the one associated with the right session_id.
1410 */
1411 rcu_read_lock();
1412 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) {
1413 if (stream->session->id == session_id) {
1414 stream->data_pending_check_done = 0;
1415 DBG("Set begin data pending flag to stream %" PRIu64,
1416 stream->stream_handle);
1417 }
1418 }
1419 rcu_read_unlock();
1420
1421 /* All good, send back reply. */
1422 reply.ret_code = htobe32(LTTNG_OK);
1423
1424 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1425 if (ret < 0) {
1426 ERR("Relay begin data pending send reply failed");
1427 }
1428
1429 end_no_session:
1430 return ret;
1431 }
1432
1433 /*
1434 * End data pending command. This will check, for a given session id, if each
1435 * stream associated with it has its data_pending_check_done flag set. If not,
1436 * this means that the client lost track of the stream but the data is still
1437 * being streamed on our side. In this case, we inform the client that data is
1438 * inflight.
1439 *
1440 * Return to the client if there is data in flight or not with a ret_code.
1441 */
1442 static
1443 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1444 struct relay_command *cmd, struct lttng_ht *streams_ht)
1445 {
1446 int ret;
1447 struct lttng_ht_iter iter;
1448 struct lttcomm_relayd_end_data_pending msg;
1449 struct lttcomm_relayd_generic_reply reply;
1450 struct relay_stream *stream;
1451 uint64_t session_id;
1452 uint32_t is_data_inflight = 0;
1453
1454 assert(recv_hdr);
1455 assert(cmd);
1456 assert(streams_ht);
1457
1458 DBG("End data pending command");
1459
1460 if (!cmd->session || cmd->version_check_done == 0) {
1461 ERR("Trying to check for data before version check");
1462 ret = -1;
1463 goto end_no_session;
1464 }
1465
1466 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1467 if (ret < sizeof(msg)) {
1468 if (ret == 0) {
1469 /* Orderly shutdown. Not necessary to print an error. */
1470 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1471 } else {
1472 ERR("Relay didn't receive valid end data_pending struct size: %d",
1473 ret);
1474 }
1475 ret = -1;
1476 goto end_no_session;
1477 }
1478
1479 session_id = be64toh(msg.session_id);
1480
1481 /* Iterate over all streams to see if the begin data pending flag is set. */
1482 rcu_read_lock();
1483 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) {
1484 if (stream->session->id == session_id &&
1485 !stream->data_pending_check_done) {
1486 is_data_inflight = 1;
1487 DBG("Data is still in flight for stream %" PRIu64,
1488 stream->stream_handle);
1489 break;
1490 }
1491 }
1492 rcu_read_unlock();
1493
1494 /* All good, send back reply. */
1495 reply.ret_code = htobe32(is_data_inflight);
1496
1497 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1498 if (ret < 0) {
1499 ERR("Relay end data pending send reply failed");
1500 }
1501
1502 end_no_session:
1503 return ret;
1504 }
1505
1506 /*
1507 * relay_process_control: Process the commands received on the control socket
1508 */
1509 static
1510 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1511 struct relay_command *cmd, struct lttng_ht *streams_ht)
1512 {
1513 int ret = 0;
1514
1515 switch (be32toh(recv_hdr->cmd)) {
1516 case RELAYD_CREATE_SESSION:
1517 ret = relay_create_session(recv_hdr, cmd);
1518 break;
1519 case RELAYD_ADD_STREAM:
1520 ret = relay_add_stream(recv_hdr, cmd, streams_ht);
1521 break;
1522 case RELAYD_START_DATA:
1523 ret = relay_start(recv_hdr, cmd);
1524 break;
1525 case RELAYD_SEND_METADATA:
1526 ret = relay_recv_metadata(recv_hdr, cmd, streams_ht);
1527 break;
1528 case RELAYD_VERSION:
1529 ret = relay_send_version(recv_hdr, cmd, streams_ht);
1530 break;
1531 case RELAYD_CLOSE_STREAM:
1532 ret = relay_close_stream(recv_hdr, cmd, streams_ht);
1533 break;
1534 case RELAYD_DATA_PENDING:
1535 ret = relay_data_pending(recv_hdr, cmd, streams_ht);
1536 break;
1537 case RELAYD_QUIESCENT_CONTROL:
1538 ret = relay_quiescent_control(recv_hdr, cmd, streams_ht);
1539 break;
1540 case RELAYD_BEGIN_DATA_PENDING:
1541 ret = relay_begin_data_pending(recv_hdr, cmd, streams_ht);
1542 break;
1543 case RELAYD_END_DATA_PENDING:
1544 ret = relay_end_data_pending(recv_hdr, cmd, streams_ht);
1545 break;
1546 case RELAYD_UPDATE_SYNC_INFO:
1547 default:
1548 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1549 relay_unknown_command(cmd);
1550 ret = -1;
1551 goto end;
1552 }
1553
1554 end:
1555 return ret;
1556 }
1557
1558 /*
1559 * relay_process_data: Process the data received on the data socket
1560 */
1561 static
1562 int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht)
1563 {
1564 int ret = 0;
1565 struct relay_stream *stream;
1566 struct lttcomm_relayd_data_hdr data_hdr;
1567 uint64_t stream_id;
1568 uint64_t net_seq_num;
1569 uint32_t data_size;
1570
1571 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1572 sizeof(struct lttcomm_relayd_data_hdr), 0);
1573 if (ret <= 0) {
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("Unable to receive data header on sock %d", cmd->sock->fd);
1579 }
1580 ret = -1;
1581 goto end;
1582 }
1583
1584 stream_id = be64toh(data_hdr.stream_id);
1585
1586 rcu_read_lock();
1587 stream = relay_stream_from_stream_id(stream_id, streams_ht);
1588 if (!stream) {
1589 ret = -1;
1590 goto end_unlock;
1591 }
1592
1593 data_size = be32toh(data_hdr.data_size);
1594 if (data_buffer_size < data_size) {
1595 char *tmp_data_ptr;
1596
1597 tmp_data_ptr = realloc(data_buffer, data_size);
1598 if (!tmp_data_ptr) {
1599 ERR("Allocating data buffer");
1600 free(data_buffer);
1601 ret = -1;
1602 goto end_unlock;
1603 }
1604 data_buffer = tmp_data_ptr;
1605 data_buffer_size = data_size;
1606 }
1607 memset(data_buffer, 0, data_size);
1608
1609 net_seq_num = be64toh(data_hdr.net_seq_num);
1610
1611 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
1612 data_size, stream_id, net_seq_num);
1613 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1614 if (ret <= 0) {
1615 if (ret == 0) {
1616 /* Orderly shutdown. Not necessary to print an error. */
1617 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1618 }
1619 ret = -1;
1620 goto end_unlock;
1621 }
1622
1623 if (stream->tracefile_size > 0 &&
1624 (stream->tracefile_size_current + data_size) >
1625 stream->tracefile_size) {
1626 ret = utils_rotate_stream_file(stream->path_name,
1627 stream->channel_name, stream->tracefile_size,
1628 stream->tracefile_count, getuid(), getgid(),
1629 stream->fd, &(stream->tracefile_count_current));
1630 if (ret < 0) {
1631 ERR("Rotating output file");
1632 goto end;
1633 }
1634 stream->fd = ret;
1635 }
1636 stream->tracefile_size_current += data_size;
1637 do {
1638 ret = write(stream->fd, data_buffer, data_size);
1639 } while (ret < 0 && errno == EINTR);
1640 if (ret < 0 || ret != data_size) {
1641 ERR("Relay error writing data to file");
1642 ret = -1;
1643 goto end_unlock;
1644 }
1645
1646 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
1647 ret, stream->stream_handle);
1648
1649 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
1650 if (ret < 0) {
1651 goto end_unlock;
1652 }
1653
1654 stream->prev_seq = net_seq_num;
1655
1656 /* Check if we need to close the FD */
1657 if (close_stream_check(stream)) {
1658 int cret;
1659 struct lttng_ht_iter iter;
1660
1661 cret = close(stream->fd);
1662 if (cret < 0) {
1663 PERROR("close stream process data");
1664 }
1665 iter.iter.node = &stream->stream_n.node;
1666 ret = lttng_ht_del(streams_ht, &iter);
1667 assert(!ret);
1668 call_rcu(&stream->rcu_node,
1669 deferred_free_stream);
1670 DBG("Closed tracefile %d after recv data", stream->fd);
1671 }
1672
1673 end_unlock:
1674 rcu_read_unlock();
1675 end:
1676 return ret;
1677 }
1678
1679 static
1680 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
1681 {
1682 int ret;
1683
1684 lttng_poll_del(events, pollfd);
1685
1686 ret = close(pollfd);
1687 if (ret < 0) {
1688 ERR("Closing pollfd %d", pollfd);
1689 }
1690 }
1691
1692 static
1693 int relay_add_connection(int fd, struct lttng_poll_event *events,
1694 struct lttng_ht *relay_connections_ht)
1695 {
1696 struct relay_command *relay_connection;
1697 int ret;
1698
1699 relay_connection = zmalloc(sizeof(struct relay_command));
1700 if (relay_connection == NULL) {
1701 PERROR("Relay command zmalloc");
1702 goto error;
1703 }
1704 do {
1705 ret = read(fd, relay_connection, sizeof(struct relay_command));
1706 } while (ret < 0 && errno == EINTR);
1707 if (ret < 0 || ret < sizeof(struct relay_command)) {
1708 PERROR("read relay cmd pipe");
1709 goto error_read;
1710 }
1711
1712 lttng_ht_node_init_ulong(&relay_connection->sock_n,
1713 (unsigned long) relay_connection->sock->fd);
1714 rcu_read_lock();
1715 lttng_ht_add_unique_ulong(relay_connections_ht,
1716 &relay_connection->sock_n);
1717 rcu_read_unlock();
1718 return lttng_poll_add(events,
1719 relay_connection->sock->fd,
1720 LPOLLIN | LPOLLRDHUP);
1721
1722 error_read:
1723 free(relay_connection);
1724 error:
1725 return -1;
1726 }
1727
1728 static
1729 void deferred_free_connection(struct rcu_head *head)
1730 {
1731 struct relay_command *relay_connection =
1732 caa_container_of(head, struct relay_command, rcu_node);
1733
1734 lttcomm_destroy_sock(relay_connection->sock);
1735 free(relay_connection);
1736 }
1737
1738 static
1739 void relay_del_connection(struct lttng_ht *relay_connections_ht,
1740 struct lttng_ht *streams_ht, struct lttng_ht_iter *iter,
1741 struct relay_command *relay_connection)
1742 {
1743 int ret;
1744
1745 ret = lttng_ht_del(relay_connections_ht, iter);
1746 assert(!ret);
1747 if (relay_connection->type == RELAY_CONTROL) {
1748 relay_delete_session(relay_connection, streams_ht);
1749 }
1750
1751 call_rcu(&relay_connection->rcu_node,
1752 deferred_free_connection);
1753 }
1754
1755 /*
1756 * This thread does the actual work
1757 */
1758 static
1759 void *relay_thread_worker(void *data)
1760 {
1761 int ret, err = -1, last_seen_data_fd = -1;
1762 uint32_t nb_fd;
1763 struct relay_command *relay_connection;
1764 struct lttng_poll_event events;
1765 struct lttng_ht *relay_connections_ht;
1766 struct lttng_ht_node_ulong *node;
1767 struct lttng_ht_iter iter;
1768 struct lttng_ht *streams_ht;
1769 struct lttcomm_relayd_hdr recv_hdr;
1770
1771 DBG("[thread] Relay worker started");
1772
1773 rcu_register_thread();
1774
1775 /* table of connections indexed on socket */
1776 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1777 if (!relay_connections_ht) {
1778 goto relay_connections_ht_error;
1779 }
1780
1781 /* tables of streams indexed by stream ID */
1782 streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1783 if (!streams_ht) {
1784 goto streams_ht_error;
1785 }
1786
1787 ret = create_thread_poll_set(&events, 2);
1788 if (ret < 0) {
1789 goto error_poll_create;
1790 }
1791
1792 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1793 if (ret < 0) {
1794 goto error;
1795 }
1796
1797 restart:
1798 while (1) {
1799 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
1800
1801 /* Infinite blocking call, waiting for transmission */
1802 DBG3("Relayd worker thread polling...");
1803 ret = lttng_poll_wait(&events, -1);
1804 if (ret < 0) {
1805 /*
1806 * Restart interrupted system call.
1807 */
1808 if (errno == EINTR) {
1809 goto restart;
1810 }
1811 goto error;
1812 }
1813
1814 nb_fd = ret;
1815
1816 /*
1817 * Process control. The control connection is prioritised so we don't
1818 * starve it with high throughout put tracing data on the data
1819 * connection.
1820 */
1821 for (i = 0; i < nb_fd; i++) {
1822 /* Fetch once the poll data */
1823 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1824 int pollfd = LTTNG_POLL_GETFD(&events, i);
1825
1826 /* Thread quit pipe has been closed. Killing thread. */
1827 ret = check_thread_quit_pipe(pollfd, revents);
1828 if (ret) {
1829 err = 0;
1830 goto exit;
1831 }
1832
1833 /* Inspect the relay cmd pipe for new connection */
1834 if (pollfd == relay_cmd_pipe[0]) {
1835 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1836 ERR("Relay pipe error");
1837 goto error;
1838 } else if (revents & LPOLLIN) {
1839 DBG("Relay command received");
1840 ret = relay_add_connection(relay_cmd_pipe[0],
1841 &events, relay_connections_ht);
1842 if (ret < 0) {
1843 goto error;
1844 }
1845 }
1846 } else if (revents) {
1847 rcu_read_lock();
1848 lttng_ht_lookup(relay_connections_ht,
1849 (void *)((unsigned long) pollfd),
1850 &iter);
1851 node = lttng_ht_iter_get_node_ulong(&iter);
1852 if (node == NULL) {
1853 DBG2("Relay sock %d not found", pollfd);
1854 rcu_read_unlock();
1855 goto error;
1856 }
1857 relay_connection = caa_container_of(node,
1858 struct relay_command, sock_n);
1859
1860 if (revents & (LPOLLERR)) {
1861 ERR("POLL ERROR");
1862 relay_cleanup_poll_connection(&events, pollfd);
1863 relay_del_connection(relay_connections_ht,
1864 streams_ht, &iter,
1865 relay_connection);
1866 if (last_seen_data_fd == pollfd) {
1867 last_seen_data_fd = last_notdel_data_fd;
1868 }
1869 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
1870 DBG("Socket %d hung up", pollfd);
1871 relay_cleanup_poll_connection(&events, pollfd);
1872 relay_del_connection(relay_connections_ht,
1873 streams_ht, &iter,
1874 relay_connection);
1875 if (last_seen_data_fd == pollfd) {
1876 last_seen_data_fd = last_notdel_data_fd;
1877 }
1878 } else if (revents & LPOLLIN) {
1879 /* control socket */
1880 if (relay_connection->type == RELAY_CONTROL) {
1881 ret = relay_connection->sock->ops->recvmsg(
1882 relay_connection->sock, &recv_hdr,
1883 sizeof(struct lttcomm_relayd_hdr), 0);
1884 /* connection closed */
1885 if (ret <= 0) {
1886 relay_cleanup_poll_connection(&events, pollfd);
1887 relay_del_connection(relay_connections_ht,
1888 streams_ht, &iter,
1889 relay_connection);
1890 DBG("Control connection closed with %d", pollfd);
1891 } else {
1892 if (relay_connection->session) {
1893 DBG2("Relay worker receiving data for session : %" PRIu64,
1894 relay_connection->session->id);
1895 }
1896 ret = relay_process_control(&recv_hdr,
1897 relay_connection,
1898 streams_ht);
1899 if (ret < 0) {
1900 /* Clear the session on error. */
1901 relay_cleanup_poll_connection(&events, pollfd);
1902 relay_del_connection(relay_connections_ht,
1903 streams_ht, &iter,
1904 relay_connection);
1905 DBG("Connection closed with %d", pollfd);
1906 }
1907 seen_control = 1;
1908 }
1909 } else {
1910 /*
1911 * Flag the last seen data fd not deleted. It will be
1912 * used as the last seen fd if any fd gets deleted in
1913 * this first loop.
1914 */
1915 last_notdel_data_fd = pollfd;
1916 }
1917 }
1918 rcu_read_unlock();
1919 }
1920 }
1921
1922 /*
1923 * The last loop handled a control request, go back to poll to make
1924 * sure we prioritise the control socket.
1925 */
1926 if (seen_control) {
1927 continue;
1928 }
1929
1930 if (last_seen_data_fd >= 0) {
1931 for (i = 0; i < nb_fd; i++) {
1932 int pollfd = LTTNG_POLL_GETFD(&events, i);
1933 if (last_seen_data_fd == pollfd) {
1934 idx = i;
1935 break;
1936 }
1937 }
1938 }
1939
1940 /* Process data connection. */
1941 for (i = idx + 1; i < nb_fd; i++) {
1942 /* Fetch the poll data. */
1943 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1944 int pollfd = LTTNG_POLL_GETFD(&events, i);
1945
1946 /* Skip the command pipe. It's handled in the first loop. */
1947 if (pollfd == relay_cmd_pipe[0]) {
1948 continue;
1949 }
1950
1951 if (revents) {
1952 rcu_read_lock();
1953 lttng_ht_lookup(relay_connections_ht,
1954 (void *)((unsigned long) pollfd),
1955 &iter);
1956 node = lttng_ht_iter_get_node_ulong(&iter);
1957 if (node == NULL) {
1958 /* Skip it. Might be removed before. */
1959 rcu_read_unlock();
1960 continue;
1961 }
1962 relay_connection = caa_container_of(node,
1963 struct relay_command, sock_n);
1964
1965 if (revents & LPOLLIN) {
1966 if (relay_connection->type != RELAY_DATA) {
1967 continue;
1968 }
1969
1970 ret = relay_process_data(relay_connection, streams_ht);
1971 /* connection closed */
1972 if (ret < 0) {
1973 relay_cleanup_poll_connection(&events, pollfd);
1974 relay_del_connection(relay_connections_ht,
1975 streams_ht, &iter,
1976 relay_connection);
1977 DBG("Data connection closed with %d", pollfd);
1978 /*
1979 * Every goto restart call sets the last seen fd where
1980 * here we don't really care since we gracefully
1981 * continue the loop after the connection is deleted.
1982 */
1983 } else {
1984 /* Keep last seen port. */
1985 last_seen_data_fd = pollfd;
1986 rcu_read_unlock();
1987 goto restart;
1988 }
1989 }
1990 rcu_read_unlock();
1991 }
1992 }
1993 last_seen_data_fd = -1;
1994 }
1995
1996 exit:
1997 error:
1998 lttng_poll_clean(&events);
1999
2000 /* empty the hash table and free the memory */
2001 rcu_read_lock();
2002 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2003 node = lttng_ht_iter_get_node_ulong(&iter);
2004 if (node) {
2005 relay_connection = caa_container_of(node,
2006 struct relay_command, sock_n);
2007 relay_del_connection(relay_connections_ht,
2008 streams_ht, &iter,
2009 relay_connection);
2010 }
2011 }
2012 rcu_read_unlock();
2013 error_poll_create:
2014 lttng_ht_destroy(streams_ht);
2015 streams_ht_error:
2016 lttng_ht_destroy(relay_connections_ht);
2017 relay_connections_ht_error:
2018 /* Close relay cmd pipes */
2019 utils_close_pipe(relay_cmd_pipe);
2020 if (err) {
2021 DBG("Thread exited with error");
2022 }
2023 DBG("Worker thread cleanup complete");
2024 free(data_buffer);
2025 stop_threads();
2026 rcu_unregister_thread();
2027 return NULL;
2028 }
2029
2030 /*
2031 * Create the relay command pipe to wake thread_manage_apps.
2032 * Closed in cleanup().
2033 */
2034 static int create_relay_cmd_pipe(void)
2035 {
2036 int ret;
2037
2038 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2039
2040 return ret;
2041 }
2042
2043 /*
2044 * main
2045 */
2046 int main(int argc, char **argv)
2047 {
2048 int ret = 0;
2049 void *status;
2050
2051 /* Create thread quit pipe */
2052 if ((ret = init_thread_quit_pipe()) < 0) {
2053 goto error;
2054 }
2055
2056 /* Parse arguments */
2057 progname = argv[0];
2058 if ((ret = parse_args(argc, argv)) < 0) {
2059 goto exit;
2060 }
2061
2062 if ((ret = set_signal_handler()) < 0) {
2063 goto exit;
2064 }
2065
2066 /* Try to create directory if -o, --output is specified. */
2067 if (opt_output_path) {
2068 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2069 if (ret < 0) {
2070 ERR("Unable to create %s", opt_output_path);
2071 goto exit;
2072 }
2073 }
2074
2075 /* Daemonize */
2076 if (opt_daemon) {
2077 ret = daemon(0, 0);
2078 if (ret < 0) {
2079 PERROR("daemon");
2080 goto exit;
2081 }
2082 }
2083
2084 /* Check if daemon is UID = 0 */
2085 is_root = !getuid();
2086
2087 if (!is_root) {
2088 if (control_uri->port < 1024 || data_uri->port < 1024) {
2089 ERR("Need to be root to use ports < 1024");
2090 ret = -1;
2091 goto exit;
2092 }
2093 }
2094
2095 /* Setup the thread apps communication pipe. */
2096 if ((ret = create_relay_cmd_pipe()) < 0) {
2097 goto exit;
2098 }
2099
2100 /* Init relay command queue. */
2101 cds_wfq_init(&relay_cmd_queue.queue);
2102
2103 /* Set up max poll set size */
2104 lttng_poll_set_max_size();
2105
2106 /* Setup the dispatcher thread */
2107 ret = pthread_create(&dispatcher_thread, NULL,
2108 relay_thread_dispatcher, (void *) NULL);
2109 if (ret != 0) {
2110 PERROR("pthread_create dispatcher");
2111 goto exit_dispatcher;
2112 }
2113
2114 /* Setup the worker thread */
2115 ret = pthread_create(&worker_thread, NULL,
2116 relay_thread_worker, (void *) NULL);
2117 if (ret != 0) {
2118 PERROR("pthread_create worker");
2119 goto exit_worker;
2120 }
2121
2122 /* Setup the listener thread */
2123 ret = pthread_create(&listener_thread, NULL,
2124 relay_thread_listener, (void *) NULL);
2125 if (ret != 0) {
2126 PERROR("pthread_create listener");
2127 goto exit_listener;
2128 }
2129
2130 exit_listener:
2131 ret = pthread_join(listener_thread, &status);
2132 if (ret != 0) {
2133 PERROR("pthread_join");
2134 goto error; /* join error, exit without cleanup */
2135 }
2136
2137 exit_worker:
2138 ret = pthread_join(worker_thread, &status);
2139 if (ret != 0) {
2140 PERROR("pthread_join");
2141 goto error; /* join error, exit without cleanup */
2142 }
2143
2144 exit_dispatcher:
2145 ret = pthread_join(dispatcher_thread, &status);
2146 if (ret != 0) {
2147 PERROR("pthread_join");
2148 goto error; /* join error, exit without cleanup */
2149 }
2150
2151 exit:
2152 cleanup();
2153 if (!ret) {
2154 exit(EXIT_SUCCESS);
2155 }
2156
2157 error:
2158 exit(EXIT_FAILURE);
2159 }
This page took 0.107157 seconds and 6 git commands to generate.