e7c57f32e2457595b31d20709837c5b7727741be
[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 "ctf-trace.h"
56 #include "index.h"
57 #include "utils.h"
58 #include "lttng-relayd.h"
59 #include "live.h"
60 #include "health-relayd.h"
61
62 /* command line options */
63 char *opt_output_path;
64 static int opt_daemon;
65 static struct lttng_uri *control_uri;
66 static struct lttng_uri *data_uri;
67 static struct lttng_uri *live_uri;
68
69 const char *progname;
70
71 /*
72 * Quit pipe for all threads. This permits a single cancellation point
73 * for all threads when receiving an event on the pipe.
74 */
75 static int thread_quit_pipe[2] = { -1, -1 };
76
77 /*
78 * This pipe is used to inform the worker thread that a command is queued and
79 * ready to be processed.
80 */
81 static int relay_cmd_pipe[2] = { -1, -1 };
82
83 /* Shared between threads */
84 static int dispatch_thread_exit;
85
86 static pthread_t listener_thread;
87 static pthread_t dispatcher_thread;
88 static pthread_t worker_thread;
89
90 static uint64_t last_relay_stream_id;
91 static uint64_t last_relay_session_id;
92
93 /*
94 * Relay command queue.
95 *
96 * The relay_thread_listener and relay_thread_dispatcher communicate with this
97 * queue.
98 */
99 static struct relay_cmd_queue relay_cmd_queue;
100
101 /* buffer allocated at startup, used to store the trace data */
102 static char *data_buffer;
103 static unsigned int data_buffer_size;
104
105 /* We need those values for the file/dir creation. */
106 static uid_t relayd_uid;
107 static gid_t relayd_gid;
108
109 /* Global relay stream hash table. */
110 struct lttng_ht *relay_streams_ht;
111
112 /* Global relay viewer stream hash table. */
113 struct lttng_ht *viewer_streams_ht;
114
115 /* Global hash table that stores relay index object. */
116 struct lttng_ht *indexes_ht;
117
118 /* Relayd health monitoring */
119 static struct health_app *health_relayd;
120
121 /*
122 * usage function on stderr
123 */
124 static
125 void usage(void)
126 {
127 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
128 fprintf(stderr, " -h, --help Display this usage.\n");
129 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
130 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
131 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
132 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
133 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
134 }
135
136 static
137 int parse_args(int argc, char **argv)
138 {
139 int c;
140 int ret = 0;
141 char *default_address;
142
143 static struct option long_options[] = {
144 { "control-port", 1, 0, 'C', },
145 { "data-port", 1, 0, 'D', },
146 { "daemonize", 0, 0, 'd', },
147 { "help", 0, 0, 'h', },
148 { "output", 1, 0, 'o', },
149 { "verbose", 0, 0, 'v', },
150 { NULL, 0, 0, 0, },
151 };
152
153 while (1) {
154 int option_index = 0;
155 c = getopt_long(argc, argv, "dhv" "C:D:o:",
156 long_options, &option_index);
157 if (c == -1) {
158 break;
159 }
160
161 switch (c) {
162 case 0:
163 fprintf(stderr, "option %s", long_options[option_index].name);
164 if (optarg) {
165 fprintf(stderr, " with arg %s\n", optarg);
166 }
167 break;
168 case 'C':
169 ret = uri_parse(optarg, &control_uri);
170 if (ret < 0) {
171 ERR("Invalid control URI specified");
172 goto exit;
173 }
174 if (control_uri->port == 0) {
175 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
176 }
177 break;
178 case 'D':
179 ret = uri_parse(optarg, &data_uri);
180 if (ret < 0) {
181 ERR("Invalid data URI specified");
182 goto exit;
183 }
184 if (data_uri->port == 0) {
185 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
186 }
187 break;
188 case 'd':
189 opt_daemon = 1;
190 break;
191 case 'h':
192 usage();
193 exit(EXIT_FAILURE);
194 case 'o':
195 ret = asprintf(&opt_output_path, "%s", optarg);
196 if (ret < 0) {
197 PERROR("asprintf opt_output_path");
198 goto exit;
199 }
200 break;
201 case 'v':
202 /* Verbose level can increase using multiple -v */
203 lttng_opt_verbose += 1;
204 break;
205 default:
206 /* Unknown option or other error.
207 * Error is printed by getopt, just return */
208 ret = -1;
209 goto exit;
210 }
211 }
212
213 /* assign default values */
214 if (control_uri == NULL) {
215 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
216 DEFAULT_NETWORK_CONTROL_PORT);
217 if (ret < 0) {
218 PERROR("asprintf default data address");
219 goto exit;
220 }
221
222 ret = uri_parse(default_address, &control_uri);
223 free(default_address);
224 if (ret < 0) {
225 ERR("Invalid control URI specified");
226 goto exit;
227 }
228 }
229 if (data_uri == NULL) {
230 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
231 DEFAULT_NETWORK_DATA_PORT);
232 if (ret < 0) {
233 PERROR("asprintf default data address");
234 goto exit;
235 }
236
237 ret = uri_parse(default_address, &data_uri);
238 free(default_address);
239 if (ret < 0) {
240 ERR("Invalid data URI specified");
241 goto exit;
242 }
243 }
244 if (live_uri == NULL) {
245 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
246 DEFAULT_NETWORK_VIEWER_PORT);
247 if (ret < 0) {
248 PERROR("asprintf default viewer control address");
249 goto exit;
250 }
251
252 ret = uri_parse(default_address, &live_uri);
253 free(default_address);
254 if (ret < 0) {
255 ERR("Invalid viewer control URI specified");
256 goto exit;
257 }
258 }
259
260 exit:
261 return ret;
262 }
263
264 /*
265 * Cleanup the daemon
266 */
267 static
268 void cleanup(void)
269 {
270 DBG("Cleaning up");
271
272 /* free the dynamically allocated opt_output_path */
273 free(opt_output_path);
274
275 /* Close thread quit pipes */
276 utils_close_pipe(thread_quit_pipe);
277
278 uri_free(control_uri);
279 uri_free(data_uri);
280 }
281
282 /*
283 * Write to writable pipe used to notify a thread.
284 */
285 static
286 int notify_thread_pipe(int wpipe)
287 {
288 int ret;
289
290 do {
291 ret = write(wpipe, "!", 1);
292 } while (ret < 0 && errno == EINTR);
293 if (ret < 0 || ret != 1) {
294 PERROR("write poll pipe");
295 }
296
297 return ret;
298 }
299
300 /*
301 * Stop all threads by closing the thread quit pipe.
302 */
303 static
304 void stop_threads(void)
305 {
306 int ret;
307
308 /* Stopping all threads */
309 DBG("Terminating all threads");
310 ret = notify_thread_pipe(thread_quit_pipe[1]);
311 if (ret < 0) {
312 ERR("write error on thread quit pipe");
313 }
314
315 /* Dispatch thread */
316 CMM_STORE_SHARED(dispatch_thread_exit, 1);
317 futex_nto1_wake(&relay_cmd_queue.futex);
318 }
319
320 /*
321 * Signal handler for the daemon
322 *
323 * Simply stop all worker threads, leaving main() return gracefully after
324 * joining all threads and calling cleanup().
325 */
326 static
327 void sighandler(int sig)
328 {
329 switch (sig) {
330 case SIGPIPE:
331 DBG("SIGPIPE caught");
332 return;
333 case SIGINT:
334 DBG("SIGINT caught");
335 stop_threads();
336 break;
337 case SIGTERM:
338 DBG("SIGTERM caught");
339 stop_threads();
340 break;
341 default:
342 break;
343 }
344 }
345
346 /*
347 * Setup signal handler for :
348 * SIGINT, SIGTERM, SIGPIPE
349 */
350 static
351 int set_signal_handler(void)
352 {
353 int ret = 0;
354 struct sigaction sa;
355 sigset_t sigset;
356
357 if ((ret = sigemptyset(&sigset)) < 0) {
358 PERROR("sigemptyset");
359 return ret;
360 }
361
362 sa.sa_handler = sighandler;
363 sa.sa_mask = sigset;
364 sa.sa_flags = 0;
365 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
366 PERROR("sigaction");
367 return ret;
368 }
369
370 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
371 PERROR("sigaction");
372 return ret;
373 }
374
375 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
376 PERROR("sigaction");
377 return ret;
378 }
379
380 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
381
382 return ret;
383 }
384
385 /*
386 * Init thread quit pipe.
387 *
388 * Return -1 on error or 0 if all pipes are created.
389 */
390 static
391 int init_thread_quit_pipe(void)
392 {
393 int ret;
394
395 ret = utils_create_pipe_cloexec(thread_quit_pipe);
396
397 return ret;
398 }
399
400 /*
401 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
402 */
403 static
404 int create_thread_poll_set(struct lttng_poll_event *events, int size)
405 {
406 int ret;
407
408 if (events == NULL || size == 0) {
409 ret = -1;
410 goto error;
411 }
412
413 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
414 if (ret < 0) {
415 goto error;
416 }
417
418 /* Add quit pipe */
419 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
420 if (ret < 0) {
421 goto error;
422 }
423
424 return 0;
425
426 error:
427 return ret;
428 }
429
430 /*
431 * Check if the thread quit pipe was triggered.
432 *
433 * Return 1 if it was triggered else 0;
434 */
435 static
436 int check_thread_quit_pipe(int fd, uint32_t events)
437 {
438 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
439 return 1;
440 }
441
442 return 0;
443 }
444
445 /*
446 * Create and init socket from uri.
447 */
448 static
449 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
450 {
451 int ret;
452 struct lttcomm_sock *sock = NULL;
453
454 sock = lttcomm_alloc_sock_from_uri(uri);
455 if (sock == NULL) {
456 ERR("Allocating socket");
457 goto error;
458 }
459
460 ret = lttcomm_create_sock(sock);
461 if (ret < 0) {
462 goto error;
463 }
464 DBG("Listening on sock %d", sock->fd);
465
466 ret = sock->ops->bind(sock);
467 if (ret < 0) {
468 goto error;
469 }
470
471 ret = sock->ops->listen(sock, -1);
472 if (ret < 0) {
473 goto error;
474
475 }
476
477 return sock;
478
479 error:
480 if (sock) {
481 lttcomm_destroy_sock(sock);
482 }
483 return NULL;
484 }
485
486 /*
487 * Return nonzero if stream needs to be closed.
488 */
489 static
490 int close_stream_check(struct relay_stream *stream)
491 {
492
493 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
494 /*
495 * We are about to close the stream so set the data pending flag to 1
496 * which will make the end data pending command skip the stream which
497 * is now closed and ready. Note that after proceeding to a file close,
498 * the written file is ready for reading.
499 */
500 stream->data_pending_check_done = 1;
501 return 1;
502 }
503 return 0;
504 }
505
506 /*
507 * This thread manages the listening for new connections on the network
508 */
509 static
510 void *relay_thread_listener(void *data)
511 {
512 int i, ret, pollfd, err = -1;
513 int val = 1;
514 uint32_t revents, nb_fd;
515 struct lttng_poll_event events;
516 struct lttcomm_sock *control_sock, *data_sock;
517
518 DBG("[thread] Relay listener started");
519
520 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
521
522 health_code_update();
523
524 control_sock = relay_init_sock(control_uri);
525 if (!control_sock) {
526 goto error_sock_control;
527 }
528
529 data_sock = relay_init_sock(data_uri);
530 if (!data_sock) {
531 goto error_sock_relay;
532 }
533
534 /*
535 * Pass 3 as size here for the thread quit pipe, control and data socket.
536 */
537 ret = create_thread_poll_set(&events, 3);
538 if (ret < 0) {
539 goto error_create_poll;
540 }
541
542 /* Add the control socket */
543 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
544 if (ret < 0) {
545 goto error_poll_add;
546 }
547
548 /* Add the data socket */
549 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
550 if (ret < 0) {
551 goto error_poll_add;
552 }
553
554 while (1) {
555 health_code_update();
556
557 DBG("Listener accepting connections");
558
559 restart:
560 health_poll_entry();
561 ret = lttng_poll_wait(&events, -1);
562 health_poll_exit();
563 if (ret < 0) {
564 /*
565 * Restart interrupted system call.
566 */
567 if (errno == EINTR) {
568 goto restart;
569 }
570 goto error;
571 }
572
573 nb_fd = ret;
574
575 DBG("Relay new connection received");
576 for (i = 0; i < nb_fd; i++) {
577 health_code_update();
578
579 /* Fetch once the poll data */
580 revents = LTTNG_POLL_GETEV(&events, i);
581 pollfd = LTTNG_POLL_GETFD(&events, i);
582
583 /* Thread quit pipe has been closed. Killing thread. */
584 ret = check_thread_quit_pipe(pollfd, revents);
585 if (ret) {
586 err = 0;
587 goto exit;
588 }
589
590 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
591 ERR("socket poll error");
592 goto error;
593 } else if (revents & LPOLLIN) {
594 /*
595 * Get allocated in this thread,
596 * enqueued to a global queue, dequeued
597 * and freed in the worker thread.
598 */
599 struct relay_command *relay_cmd;
600 struct lttcomm_sock *newsock;
601
602 relay_cmd = zmalloc(sizeof(struct relay_command));
603 if (relay_cmd == NULL) {
604 PERROR("relay command zmalloc");
605 goto error;
606 }
607
608 if (pollfd == data_sock->fd) {
609 newsock = data_sock->ops->accept(data_sock);
610 if (!newsock) {
611 PERROR("accepting data sock");
612 free(relay_cmd);
613 goto error;
614 }
615 relay_cmd->type = RELAY_DATA;
616 DBG("Relay data connection accepted, socket %d", newsock->fd);
617 } else {
618 assert(pollfd == control_sock->fd);
619 newsock = control_sock->ops->accept(control_sock);
620 if (!newsock) {
621 PERROR("accepting control sock");
622 free(relay_cmd);
623 goto error;
624 }
625 relay_cmd->type = RELAY_CONTROL;
626 DBG("Relay control connection accepted, socket %d", newsock->fd);
627 }
628 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
629 &val, sizeof(int));
630 if (ret < 0) {
631 PERROR("setsockopt inet");
632 lttcomm_destroy_sock(newsock);
633 free(relay_cmd);
634 goto error;
635 }
636 relay_cmd->sock = newsock;
637 /*
638 * Lock free enqueue the request.
639 */
640 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
641
642 /*
643 * Wake the dispatch queue futex. Implicit memory
644 * barrier with the exchange in cds_wfq_enqueue.
645 */
646 futex_nto1_wake(&relay_cmd_queue.futex);
647 }
648 }
649 }
650
651 exit:
652 error:
653 error_poll_add:
654 lttng_poll_clean(&events);
655 error_create_poll:
656 if (data_sock->fd >= 0) {
657 ret = data_sock->ops->close(data_sock);
658 if (ret) {
659 PERROR("close");
660 }
661 }
662 lttcomm_destroy_sock(data_sock);
663 error_sock_relay:
664 if (control_sock->fd >= 0) {
665 ret = control_sock->ops->close(control_sock);
666 if (ret) {
667 PERROR("close");
668 }
669 }
670 lttcomm_destroy_sock(control_sock);
671 error_sock_control:
672 if (err) {
673 health_error();
674 ERR("Health error occurred in %s", __func__);
675 }
676 health_unregister(health_relayd);
677 DBG("Relay listener thread cleanup complete");
678 stop_threads();
679 return NULL;
680 }
681
682 /*
683 * This thread manages the dispatching of the requests to worker threads
684 */
685 static
686 void *relay_thread_dispatcher(void *data)
687 {
688 int ret, err = -1;
689 struct cds_wfq_node *node;
690 struct relay_command *relay_cmd = NULL;
691
692 DBG("[thread] Relay dispatcher started");
693
694 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
695
696 health_code_update();
697
698 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
699 health_code_update();
700
701 /* Atomically prepare the queue futex */
702 futex_nto1_prepare(&relay_cmd_queue.futex);
703
704 do {
705 health_code_update();
706
707 /* Dequeue commands */
708 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
709 if (node == NULL) {
710 DBG("Woken up but nothing in the relay command queue");
711 /* Continue thread execution */
712 break;
713 }
714
715 relay_cmd = caa_container_of(node, struct relay_command, node);
716 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
717
718 /*
719 * Inform worker thread of the new request. This
720 * call is blocking so we can be assured that the data will be read
721 * at some point in time or wait to the end of the world :)
722 */
723 do {
724 ret = write(relay_cmd_pipe[1], relay_cmd,
725 sizeof(struct relay_command));
726 } while (ret < 0 && errno == EINTR);
727 free(relay_cmd);
728 if (ret < 0 || ret != sizeof(struct relay_command)) {
729 PERROR("write cmd pipe");
730 goto error;
731 }
732 } while (node != NULL);
733
734 /* Futex wait on queue. Blocking call on futex() */
735 health_poll_entry();
736 futex_nto1_wait(&relay_cmd_queue.futex);
737 health_poll_exit();
738 }
739
740 /* Normal exit, no error */
741 err = 0;
742
743 error:
744 if (err) {
745 health_error();
746 ERR("Health error occurred in %s", __func__);
747 }
748 health_unregister(health_relayd);
749 DBG("Dispatch thread dying");
750 stop_threads();
751 return NULL;
752 }
753
754 /*
755 * Get stream from stream id.
756 * Need to be called with RCU read-side lock held.
757 */
758 struct relay_stream *relay_stream_find_by_id(uint64_t stream_id)
759 {
760 struct lttng_ht_node_ulong *node;
761 struct lttng_ht_iter iter;
762 struct relay_stream *ret;
763
764 lttng_ht_lookup(relay_streams_ht,
765 (void *)((unsigned long) stream_id),
766 &iter);
767 node = lttng_ht_iter_get_node_ulong(&iter);
768 if (node == NULL) {
769 DBG("Relay stream %" PRIu64 " not found", stream_id);
770 ret = NULL;
771 goto end;
772 }
773
774 ret = caa_container_of(node, struct relay_stream, stream_n);
775
776 end:
777 return ret;
778 }
779
780 static
781 void deferred_free_stream(struct rcu_head *head)
782 {
783 struct relay_stream *stream =
784 caa_container_of(head, struct relay_stream, rcu_node);
785
786 ctf_trace_try_destroy(stream->ctf_trace);
787
788 free(stream->path_name);
789 free(stream->channel_name);
790 free(stream);
791 }
792
793 static
794 void deferred_free_session(struct rcu_head *head)
795 {
796 struct relay_session *session =
797 caa_container_of(head, struct relay_session, rcu_node);
798 free(session);
799 }
800
801 /*
802 * Close a given stream. The stream is freed using a call RCU.
803 *
804 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
805 * BEFORE the stream lock MUST be acquired.
806 */
807 static void destroy_stream(struct relay_stream *stream,
808 struct lttng_ht *ctf_traces_ht)
809 {
810 int delret;
811 struct relay_viewer_stream *vstream;
812 struct lttng_ht_iter iter;
813
814 assert(stream);
815
816 delret = close(stream->fd);
817 if (delret < 0) {
818 PERROR("close stream");
819 }
820
821 if (stream->index_fd >= 0) {
822 delret = close(stream->index_fd);
823 if (delret < 0) {
824 PERROR("close stream index_fd");
825 }
826 }
827
828 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
829 if (vstream) {
830 /*
831 * Set the last good value into the viewer stream. This is done
832 * right before the stream gets deleted from the hash table. The
833 * lookup failure on the live thread side of a stream indicates
834 * that the viewer stream index received value should be used.
835 */
836 vstream->total_index_received = stream->total_index_received;
837 }
838
839 /* Cleanup index of that stream. */
840 relay_index_destroy_by_stream_id(stream->stream_handle);
841
842 iter.iter.node = &stream->stream_n.node;
843 delret = lttng_ht_del(relay_streams_ht, &iter);
844 assert(!delret);
845 iter.iter.node = &stream->ctf_trace_node.node;
846 delret = lttng_ht_del(ctf_traces_ht, &iter);
847 assert(!delret);
848 call_rcu(&stream->rcu_node, deferred_free_stream);
849 DBG("Closed tracefile %d from close stream", stream->fd);
850 }
851
852 /*
853 * relay_delete_session: Free all memory associated with a session and
854 * close all the FDs
855 */
856 static
857 void relay_delete_session(struct relay_command *cmd,
858 struct lttng_ht *sessions_ht)
859 {
860 struct lttng_ht_iter iter;
861 struct lttng_ht_node_ulong *node;
862 struct relay_stream *stream;
863 int ret;
864
865 if (!cmd->session) {
866 return;
867 }
868
869 DBG("Relay deleting session %" PRIu64, cmd->session->id);
870
871 rcu_read_lock();
872 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
873 node = lttng_ht_iter_get_node_ulong(&iter);
874 if (!node) {
875 continue;
876 }
877 stream = caa_container_of(node, struct relay_stream, stream_n);
878 if (stream->session == cmd->session) {
879 destroy_stream(stream, cmd->ctf_traces_ht);
880 }
881 }
882
883 /* Make this session not visible anymore. */
884 iter.iter.node = &cmd->session->session_n.node;
885 ret = lttng_ht_del(sessions_ht, &iter);
886 assert(!ret);
887 call_rcu(&cmd->session->rcu_node, deferred_free_session);
888 rcu_read_unlock();
889 }
890
891 /*
892 * Copy index data from the control port to a given index object.
893 */
894 static void copy_index_control_data(struct relay_index *index,
895 struct lttcomm_relayd_index *data)
896 {
897 assert(index);
898 assert(data);
899
900 /*
901 * The index on disk is encoded in big endian, so we don't need to convert
902 * the data received on the network. The data_offset value is NEVER
903 * modified here and is updated by the data thread.
904 */
905 index->index_data.packet_size = data->packet_size;
906 index->index_data.content_size = data->content_size;
907 index->index_data.timestamp_begin = data->timestamp_begin;
908 index->index_data.timestamp_end = data->timestamp_end;
909 index->index_data.events_discarded = data->events_discarded;
910 index->index_data.stream_id = data->stream_id;
911 }
912
913 /*
914 * Handle the RELAYD_CREATE_SESSION command.
915 *
916 * On success, send back the session id or else return a negative value.
917 */
918 static
919 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
920 struct relay_command *cmd,
921 struct lttng_ht *sessions_ht)
922 {
923 int ret = 0, send_ret;
924 struct relay_session *session;
925 struct lttcomm_relayd_status_session reply;
926
927 assert(recv_hdr);
928 assert(cmd);
929
930 memset(&reply, 0, sizeof(reply));
931
932 session = zmalloc(sizeof(struct relay_session));
933 if (session == NULL) {
934 PERROR("relay session zmalloc");
935 ret = -1;
936 goto error;
937 }
938
939 session->id = ++last_relay_session_id;
940 session->sock = cmd->sock;
941 session->minor = cmd->minor;
942 session->major = cmd->major;
943 cmd->session = session;
944
945 reply.session_id = htobe64(session->id);
946
947 switch (cmd->minor) {
948 case 4: /* LTTng sessiond 2.4 */
949 default:
950 ret = cmd_create_session_2_4(cmd, session);
951 break;
952 }
953
954 lttng_ht_node_init_ulong(&session->session_n,
955 (unsigned long) session->id);
956 lttng_ht_add_unique_ulong(sessions_ht,
957 &session->session_n);
958
959 DBG("Created session %" PRIu64, session->id);
960
961 error:
962 if (ret < 0) {
963 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
964 } else {
965 reply.ret_code = htobe32(LTTNG_OK);
966 }
967
968 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
969 if (send_ret < 0) {
970 ERR("Relayd sending session id");
971 ret = send_ret;
972 }
973
974 return ret;
975 }
976
977 /*
978 * relay_add_stream: allocate a new stream for a session
979 */
980 static
981 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
982 struct relay_command *cmd, struct lttng_ht *sessions_ht)
983 {
984 struct relay_session *session = cmd->session;
985 struct relay_stream *stream = NULL;
986 struct lttcomm_relayd_status_stream reply;
987 int ret, send_ret;
988
989 if (!session || cmd->version_check_done == 0) {
990 ERR("Trying to add a stream before version check");
991 ret = -1;
992 goto end_no_session;
993 }
994
995 stream = zmalloc(sizeof(struct relay_stream));
996 if (stream == NULL) {
997 PERROR("relay stream zmalloc");
998 ret = -1;
999 goto end_no_session;
1000 }
1001
1002 switch (cmd->minor) {
1003 case 1: /* LTTng sessiond 2.1 */
1004 ret = cmd_recv_stream_2_1(cmd, stream);
1005 break;
1006 case 2: /* LTTng sessiond 2.2 */
1007 default:
1008 ret = cmd_recv_stream_2_2(cmd, stream);
1009 break;
1010 }
1011 if (ret < 0) {
1012 goto err_free_stream;
1013 }
1014
1015 rcu_read_lock();
1016 stream->stream_handle = ++last_relay_stream_id;
1017 stream->prev_seq = -1ULL;
1018 stream->session = session;
1019 stream->index_fd = -1;
1020 stream->read_index_fd = -1;
1021 stream->ctf_trace = NULL;
1022 pthread_mutex_init(&stream->lock, NULL);
1023
1024 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
1025 if (ret < 0) {
1026 ERR("relay creating output directory");
1027 goto end;
1028 }
1029
1030 /*
1031 * No need to use run_as API here because whatever we receives, the relayd
1032 * uses its own credentials for the stream files.
1033 */
1034 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1035 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1036 if (ret < 0) {
1037 ERR("Create output file");
1038 goto end;
1039 }
1040 stream->fd = ret;
1041 if (stream->tracefile_size) {
1042 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1043 } else {
1044 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1045 }
1046
1047 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1048 stream->metadata_flag = 1;
1049 /*
1050 * When we receive a new metadata stream, we create a new
1051 * ctf_trace and we assign this ctf_trace to all streams with
1052 * the same path.
1053 *
1054 * If later on we receive a new stream for the same ctf_trace,
1055 * we copy the information from the first hit in the HT to the
1056 * new stream.
1057 */
1058 stream->ctf_trace = ctf_trace_create();
1059 if (!stream->ctf_trace) {
1060 ret = -1;
1061 goto end;
1062 }
1063 stream->ctf_trace->refcount++;
1064 stream->ctf_trace->metadata_stream = stream;
1065 }
1066 ctf_trace_assign(cmd->ctf_traces_ht, stream);
1067
1068 lttng_ht_node_init_ulong(&stream->stream_n,
1069 (unsigned long) stream->stream_handle);
1070 lttng_ht_add_unique_ulong(relay_streams_ht,
1071 &stream->stream_n);
1072
1073 lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name);
1074 lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node);
1075
1076 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1077 stream->stream_handle);
1078
1079 end:
1080 reply.handle = htobe64(stream->stream_handle);
1081 /* send the session id to the client or a negative return code on error */
1082 if (ret < 0) {
1083 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1084 /* stream was not properly added to the ht, so free it */
1085 free(stream);
1086 } else {
1087 reply.ret_code = htobe32(LTTNG_OK);
1088 }
1089
1090 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1091 sizeof(struct lttcomm_relayd_status_stream), 0);
1092 if (send_ret < 0) {
1093 ERR("Relay sending stream id");
1094 ret = send_ret;
1095 }
1096 rcu_read_unlock();
1097
1098 end_no_session:
1099 return ret;
1100
1101 err_free_stream:
1102 free(stream->path_name);
1103 free(stream->channel_name);
1104 free(stream);
1105 return ret;
1106 }
1107
1108 /*
1109 * relay_close_stream: close a specific stream
1110 */
1111 static
1112 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1113 struct relay_command *cmd)
1114 {
1115 int ret, send_ret;
1116 struct relay_session *session = cmd->session;
1117 struct lttcomm_relayd_close_stream stream_info;
1118 struct lttcomm_relayd_generic_reply reply;
1119 struct relay_stream *stream;
1120
1121 DBG("Close stream received");
1122
1123 if (!session || cmd->version_check_done == 0) {
1124 ERR("Trying to close a stream before version check");
1125 ret = -1;
1126 goto end_no_session;
1127 }
1128
1129 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1130 sizeof(struct lttcomm_relayd_close_stream), 0);
1131 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1132 if (ret == 0) {
1133 /* Orderly shutdown. Not necessary to print an error. */
1134 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1135 } else {
1136 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1137 }
1138 ret = -1;
1139 goto end_no_session;
1140 }
1141
1142 rcu_read_lock();
1143 stream = relay_stream_find_by_id(be64toh(stream_info.stream_id));
1144 if (!stream) {
1145 ret = -1;
1146 goto end_unlock;
1147 }
1148
1149 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1150 stream->close_flag = 1;
1151
1152 if (close_stream_check(stream)) {
1153 destroy_stream(stream, cmd->ctf_traces_ht);
1154 }
1155
1156 end_unlock:
1157 rcu_read_unlock();
1158
1159 if (ret < 0) {
1160 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1161 } else {
1162 reply.ret_code = htobe32(LTTNG_OK);
1163 }
1164 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1165 sizeof(struct lttcomm_relayd_generic_reply), 0);
1166 if (send_ret < 0) {
1167 ERR("Relay sending stream id");
1168 ret = send_ret;
1169 }
1170
1171 end_no_session:
1172 return ret;
1173 }
1174
1175 /*
1176 * relay_unknown_command: send -1 if received unknown command
1177 */
1178 static
1179 void relay_unknown_command(struct relay_command *cmd)
1180 {
1181 struct lttcomm_relayd_generic_reply reply;
1182 int ret;
1183
1184 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1185 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1186 sizeof(struct lttcomm_relayd_generic_reply), 0);
1187 if (ret < 0) {
1188 ERR("Relay sending unknown command");
1189 }
1190 }
1191
1192 /*
1193 * relay_start: send an acknowledgment to the client to tell if we are
1194 * ready to receive data. We are ready if a session is established.
1195 */
1196 static
1197 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1198 struct relay_command *cmd)
1199 {
1200 int ret = htobe32(LTTNG_OK);
1201 struct lttcomm_relayd_generic_reply reply;
1202 struct relay_session *session = cmd->session;
1203
1204 if (!session) {
1205 DBG("Trying to start the streaming without a session established");
1206 ret = htobe32(LTTNG_ERR_UNK);
1207 }
1208
1209 reply.ret_code = ret;
1210 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1211 sizeof(struct lttcomm_relayd_generic_reply), 0);
1212 if (ret < 0) {
1213 ERR("Relay sending start ack");
1214 }
1215
1216 return ret;
1217 }
1218
1219 /*
1220 * Append padding to the file pointed by the file descriptor fd.
1221 */
1222 static int write_padding_to_file(int fd, uint32_t size)
1223 {
1224 int ret = 0;
1225 char *zeros;
1226
1227 if (size == 0) {
1228 goto end;
1229 }
1230
1231 zeros = zmalloc(size);
1232 if (zeros == NULL) {
1233 PERROR("zmalloc zeros for padding");
1234 ret = -1;
1235 goto end;
1236 }
1237
1238 do {
1239 ret = write(fd, zeros, size);
1240 } while (ret < 0 && errno == EINTR);
1241 if (ret < 0 || ret != size) {
1242 PERROR("write padding to file");
1243 }
1244
1245 free(zeros);
1246
1247 end:
1248 return ret;
1249 }
1250
1251 /*
1252 * relay_recv_metadata: receive the metada for the session.
1253 */
1254 static
1255 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1256 struct relay_command *cmd)
1257 {
1258 int ret = htobe32(LTTNG_OK);
1259 struct relay_session *session = cmd->session;
1260 struct lttcomm_relayd_metadata_payload *metadata_struct;
1261 struct relay_stream *metadata_stream;
1262 uint64_t data_size, payload_size;
1263
1264 if (!session) {
1265 ERR("Metadata sent before version check");
1266 ret = -1;
1267 goto end;
1268 }
1269
1270 data_size = payload_size = be64toh(recv_hdr->data_size);
1271 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1272 ERR("Incorrect data size");
1273 ret = -1;
1274 goto end;
1275 }
1276 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1277
1278 if (data_buffer_size < data_size) {
1279 /* In case the realloc fails, we can free the memory */
1280 char *tmp_data_ptr;
1281
1282 tmp_data_ptr = realloc(data_buffer, data_size);
1283 if (!tmp_data_ptr) {
1284 ERR("Allocating data buffer");
1285 free(data_buffer);
1286 ret = -1;
1287 goto end;
1288 }
1289 data_buffer = tmp_data_ptr;
1290 data_buffer_size = data_size;
1291 }
1292 memset(data_buffer, 0, data_size);
1293 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1294 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1295 if (ret < 0 || ret != data_size) {
1296 if (ret == 0) {
1297 /* Orderly shutdown. Not necessary to print an error. */
1298 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1299 } else {
1300 ERR("Relay didn't receive the whole metadata");
1301 }
1302 ret = -1;
1303 goto end;
1304 }
1305 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1306
1307 rcu_read_lock();
1308 metadata_stream = relay_stream_find_by_id(
1309 be64toh(metadata_struct->stream_id));
1310 if (!metadata_stream) {
1311 ret = -1;
1312 goto end_unlock;
1313 }
1314
1315 do {
1316 ret = write(metadata_stream->fd, metadata_struct->payload,
1317 payload_size);
1318 } while (ret < 0 && errno == EINTR);
1319 if (ret < 0 || ret != payload_size) {
1320 ERR("Relay error writing metadata on file");
1321 ret = -1;
1322 goto end_unlock;
1323 }
1324
1325 ret = write_padding_to_file(metadata_stream->fd,
1326 be32toh(metadata_struct->padding_size));
1327 if (ret < 0) {
1328 goto end_unlock;
1329 }
1330 metadata_stream->ctf_trace->metadata_received +=
1331 payload_size + be32toh(metadata_struct->padding_size);
1332
1333 DBG2("Relay metadata written");
1334
1335 end_unlock:
1336 rcu_read_unlock();
1337 end:
1338 return ret;
1339 }
1340
1341 /*
1342 * relay_send_version: send relayd version number
1343 */
1344 static
1345 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1346 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1347 {
1348 int ret;
1349 struct lttcomm_relayd_version reply, msg;
1350
1351 assert(cmd);
1352
1353 cmd->version_check_done = 1;
1354
1355 /* Get version from the other side. */
1356 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1357 if (ret < 0 || ret != sizeof(msg)) {
1358 if (ret == 0) {
1359 /* Orderly shutdown. Not necessary to print an error. */
1360 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1361 } else {
1362 ERR("Relay failed to receive the version values.");
1363 }
1364 ret = -1;
1365 goto end;
1366 }
1367
1368 reply.major = RELAYD_VERSION_COMM_MAJOR;
1369 reply.minor = RELAYD_VERSION_COMM_MINOR;
1370
1371 /* Major versions must be the same */
1372 if (reply.major != be32toh(msg.major)) {
1373 DBG("Incompatible major versions (%u vs %u), deleting session",
1374 reply.major, be32toh(msg.major));
1375 relay_delete_session(cmd, sessions_ht);
1376 ret = 0;
1377 goto end;
1378 }
1379
1380 cmd->major = reply.major;
1381 /* We adapt to the lowest compatible version */
1382 if (reply.minor <= be32toh(msg.minor)) {
1383 cmd->minor = reply.minor;
1384 } else {
1385 cmd->minor = be32toh(msg.minor);
1386 }
1387
1388 reply.major = htobe32(reply.major);
1389 reply.minor = htobe32(reply.minor);
1390 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1391 sizeof(struct lttcomm_relayd_version), 0);
1392 if (ret < 0) {
1393 ERR("Relay sending version");
1394 }
1395
1396 DBG("Version check done using protocol %u.%u", cmd->major,
1397 cmd->minor);
1398
1399 end:
1400 return ret;
1401 }
1402
1403 /*
1404 * Check for data pending for a given stream id from the session daemon.
1405 */
1406 static
1407 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1408 struct relay_command *cmd)
1409 {
1410 struct relay_session *session = cmd->session;
1411 struct lttcomm_relayd_data_pending msg;
1412 struct lttcomm_relayd_generic_reply reply;
1413 struct relay_stream *stream;
1414 int ret;
1415 uint64_t last_net_seq_num, stream_id;
1416
1417 DBG("Data pending command received");
1418
1419 if (!session || cmd->version_check_done == 0) {
1420 ERR("Trying to check for data before version check");
1421 ret = -1;
1422 goto end_no_session;
1423 }
1424
1425 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1426 if (ret < sizeof(msg)) {
1427 if (ret == 0) {
1428 /* Orderly shutdown. Not necessary to print an error. */
1429 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1430 } else {
1431 ERR("Relay didn't receive valid data_pending struct size : %d",
1432 ret);
1433 }
1434 ret = -1;
1435 goto end_no_session;
1436 }
1437
1438 stream_id = be64toh(msg.stream_id);
1439 last_net_seq_num = be64toh(msg.last_net_seq_num);
1440
1441 rcu_read_lock();
1442 stream = relay_stream_find_by_id(stream_id);
1443 if (stream == NULL) {
1444 ret = -1;
1445 goto end_unlock;
1446 }
1447
1448 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1449 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1450 last_net_seq_num);
1451
1452 /* Avoid wrapping issue */
1453 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1454 /* Data has in fact been written and is NOT pending */
1455 ret = 0;
1456 } else {
1457 /* Data still being streamed thus pending */
1458 ret = 1;
1459 }
1460
1461 /* Pending check is now done. */
1462 stream->data_pending_check_done = 1;
1463
1464 end_unlock:
1465 rcu_read_unlock();
1466
1467 reply.ret_code = htobe32(ret);
1468 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1469 if (ret < 0) {
1470 ERR("Relay data pending ret code failed");
1471 }
1472
1473 end_no_session:
1474 return ret;
1475 }
1476
1477 /*
1478 * Wait for the control socket to reach a quiescent state.
1479 *
1480 * Note that for now, when receiving this command from the session daemon, this
1481 * means that every subsequent commands or data received on the control socket
1482 * has been handled. So, this is why we simply return OK here.
1483 */
1484 static
1485 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1486 struct relay_command *cmd)
1487 {
1488 int ret;
1489 uint64_t stream_id;
1490 struct relay_stream *stream;
1491 struct lttng_ht_iter iter;
1492 struct lttcomm_relayd_quiescent_control msg;
1493 struct lttcomm_relayd_generic_reply reply;
1494
1495 DBG("Checking quiescent state on control socket");
1496
1497 if (!cmd->session || cmd->version_check_done == 0) {
1498 ERR("Trying to check for data before version check");
1499 ret = -1;
1500 goto end_no_session;
1501 }
1502
1503 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1504 if (ret < sizeof(msg)) {
1505 if (ret == 0) {
1506 /* Orderly shutdown. Not necessary to print an error. */
1507 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1508 } else {
1509 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1510 ret);
1511 }
1512 ret = -1;
1513 goto end_no_session;
1514 }
1515
1516 stream_id = be64toh(msg.stream_id);
1517
1518 rcu_read_lock();
1519 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1520 stream_n.node) {
1521 if (stream->stream_handle == stream_id) {
1522 stream->data_pending_check_done = 1;
1523 DBG("Relay quiescent control pending flag set to %" PRIu64,
1524 stream_id);
1525 break;
1526 }
1527 }
1528 rcu_read_unlock();
1529
1530 reply.ret_code = htobe32(LTTNG_OK);
1531 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1532 if (ret < 0) {
1533 ERR("Relay data quiescent control ret code failed");
1534 }
1535
1536 end_no_session:
1537 return ret;
1538 }
1539
1540 /*
1541 * Initialize a data pending command. This means that a client is about to ask
1542 * for data pending for each stream he/she holds. Simply iterate over all
1543 * streams of a session and set the data_pending_check_done flag.
1544 *
1545 * This command returns to the client a LTTNG_OK code.
1546 */
1547 static
1548 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1549 struct relay_command *cmd)
1550 {
1551 int ret;
1552 struct lttng_ht_iter iter;
1553 struct lttcomm_relayd_begin_data_pending msg;
1554 struct lttcomm_relayd_generic_reply reply;
1555 struct relay_stream *stream;
1556 uint64_t session_id;
1557
1558 assert(recv_hdr);
1559 assert(cmd);
1560
1561 DBG("Init streams for data pending");
1562
1563 if (!cmd->session || cmd->version_check_done == 0) {
1564 ERR("Trying to check for data before version check");
1565 ret = -1;
1566 goto end_no_session;
1567 }
1568
1569 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1570 if (ret < sizeof(msg)) {
1571 if (ret == 0) {
1572 /* Orderly shutdown. Not necessary to print an error. */
1573 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1574 } else {
1575 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1576 ret);
1577 }
1578 ret = -1;
1579 goto end_no_session;
1580 }
1581
1582 session_id = be64toh(msg.session_id);
1583
1584 /*
1585 * Iterate over all streams to set the begin data pending flag. For now, the
1586 * streams are indexed by stream handle so we have to iterate over all
1587 * streams to find the one associated with the right session_id.
1588 */
1589 rcu_read_lock();
1590 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1591 stream_n.node) {
1592 if (stream->session->id == session_id) {
1593 stream->data_pending_check_done = 0;
1594 DBG("Set begin data pending flag to stream %" PRIu64,
1595 stream->stream_handle);
1596 }
1597 }
1598 rcu_read_unlock();
1599
1600 /* All good, send back reply. */
1601 reply.ret_code = htobe32(LTTNG_OK);
1602
1603 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1604 if (ret < 0) {
1605 ERR("Relay begin data pending send reply failed");
1606 }
1607
1608 end_no_session:
1609 return ret;
1610 }
1611
1612 /*
1613 * End data pending command. This will check, for a given session id, if each
1614 * stream associated with it has its data_pending_check_done flag set. If not,
1615 * this means that the client lost track of the stream but the data is still
1616 * being streamed on our side. In this case, we inform the client that data is
1617 * inflight.
1618 *
1619 * Return to the client if there is data in flight or not with a ret_code.
1620 */
1621 static
1622 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1623 struct relay_command *cmd)
1624 {
1625 int ret;
1626 struct lttng_ht_iter iter;
1627 struct lttcomm_relayd_end_data_pending msg;
1628 struct lttcomm_relayd_generic_reply reply;
1629 struct relay_stream *stream;
1630 uint64_t session_id;
1631 uint32_t is_data_inflight = 0;
1632
1633 assert(recv_hdr);
1634 assert(cmd);
1635
1636 DBG("End data pending command");
1637
1638 if (!cmd->session || cmd->version_check_done == 0) {
1639 ERR("Trying to check for data before version check");
1640 ret = -1;
1641 goto end_no_session;
1642 }
1643
1644 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1645 if (ret < sizeof(msg)) {
1646 if (ret == 0) {
1647 /* Orderly shutdown. Not necessary to print an error. */
1648 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1649 } else {
1650 ERR("Relay didn't receive valid end data_pending struct size: %d",
1651 ret);
1652 }
1653 ret = -1;
1654 goto end_no_session;
1655 }
1656
1657 session_id = be64toh(msg.session_id);
1658
1659 /* Iterate over all streams to see if the begin data pending flag is set. */
1660 rcu_read_lock();
1661 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1662 stream_n.node) {
1663 if (stream->session->id == session_id &&
1664 !stream->data_pending_check_done) {
1665 is_data_inflight = 1;
1666 DBG("Data is still in flight for stream %" PRIu64,
1667 stream->stream_handle);
1668 break;
1669 }
1670 }
1671 rcu_read_unlock();
1672
1673 /* All good, send back reply. */
1674 reply.ret_code = htobe32(is_data_inflight);
1675
1676 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1677 if (ret < 0) {
1678 ERR("Relay end data pending send reply failed");
1679 }
1680
1681 end_no_session:
1682 return ret;
1683 }
1684
1685 /*
1686 * Receive an index for a specific stream.
1687 *
1688 * Return 0 on success else a negative value.
1689 */
1690 static
1691 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1692 struct relay_command *cmd)
1693 {
1694 int ret, send_ret, index_created = 0;
1695 struct relay_session *session = cmd->session;
1696 struct lttcomm_relayd_index index_info;
1697 struct relay_index *index, *wr_index = NULL;
1698 struct lttcomm_relayd_generic_reply reply;
1699 struct relay_stream *stream;
1700 uint64_t net_seq_num;
1701
1702 assert(cmd);
1703
1704 DBG("Relay receiving index");
1705
1706 if (!session || cmd->version_check_done == 0) {
1707 ERR("Trying to close a stream before version check");
1708 ret = -1;
1709 goto end_no_session;
1710 }
1711
1712 ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info,
1713 sizeof(index_info), 0);
1714 if (ret < sizeof(index_info)) {
1715 if (ret == 0) {
1716 /* Orderly shutdown. Not necessary to print an error. */
1717 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1718 } else {
1719 ERR("Relay didn't receive valid index struct size : %d", ret);
1720 }
1721 ret = -1;
1722 goto end_no_session;
1723 }
1724
1725 net_seq_num = be64toh(index_info.net_seq_num);
1726
1727 rcu_read_lock();
1728 stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id));
1729 if (!stream) {
1730 ret = -1;
1731 goto end_rcu_unlock;
1732 }
1733
1734 /* Live beacon handling */
1735 if (index_info.packet_size == 0) {
1736 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1737
1738 /*
1739 * Only flag a stream inactive when it has already received data.
1740 */
1741 if (stream->total_index_received > 0) {
1742 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1743 }
1744 ret = 0;
1745 goto end_rcu_unlock;
1746 } else {
1747 stream->beacon_ts_end = -1ULL;
1748 }
1749
1750 index = relay_index_find(stream->stream_handle, net_seq_num);
1751 if (!index) {
1752 /* A successful creation will add the object to the HT. */
1753 index = relay_index_create(stream->stream_handle, net_seq_num);
1754 if (!index) {
1755 goto end_rcu_unlock;
1756 }
1757 index_created = 1;
1758 }
1759
1760 copy_index_control_data(index, &index_info);
1761
1762 if (index_created) {
1763 /*
1764 * Try to add the relay index object to the hash table. If an object
1765 * already exist, destroy back the index created, set the data in this
1766 * object and write it on disk.
1767 */
1768 relay_index_add(index, &wr_index);
1769 if (wr_index) {
1770 copy_index_control_data(wr_index, &index_info);
1771 free(index);
1772 }
1773 } else {
1774 /* The index already exists so write it on disk. */
1775 wr_index = index;
1776 }
1777
1778 /* Do we have a writable ready index to write on disk. */
1779 if (wr_index) {
1780 /* Starting at 2.4, create the index file if none available. */
1781 if (cmd->minor >= 4 && stream->index_fd < 0) {
1782 ret = index_create_file(stream->path_name, stream->channel_name,
1783 relayd_uid, relayd_gid, stream->tracefile_size,
1784 stream->tracefile_count_current);
1785 if (ret < 0) {
1786 goto end_rcu_unlock;
1787 }
1788 stream->index_fd = ret;
1789 }
1790
1791 ret = relay_index_write(wr_index->fd, wr_index);
1792 if (ret < 0) {
1793 goto end_rcu_unlock;
1794 }
1795 stream->total_index_received++;
1796 }
1797
1798 end_rcu_unlock:
1799 rcu_read_unlock();
1800
1801 if (ret < 0) {
1802 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1803 } else {
1804 reply.ret_code = htobe32(LTTNG_OK);
1805 }
1806 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1807 if (send_ret < 0) {
1808 ERR("Relay sending close index id reply");
1809 ret = send_ret;
1810 }
1811
1812 end_no_session:
1813 return ret;
1814 }
1815
1816 /*
1817 * Process the commands received on the control socket
1818 */
1819 static
1820 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1821 struct relay_command *cmd, struct relay_local_data *ctx)
1822 {
1823 int ret = 0;
1824
1825 switch (be32toh(recv_hdr->cmd)) {
1826 case RELAYD_CREATE_SESSION:
1827 ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht);
1828 break;
1829 case RELAYD_ADD_STREAM:
1830 ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht);
1831 break;
1832 case RELAYD_START_DATA:
1833 ret = relay_start(recv_hdr, cmd);
1834 break;
1835 case RELAYD_SEND_METADATA:
1836 ret = relay_recv_metadata(recv_hdr, cmd);
1837 break;
1838 case RELAYD_VERSION:
1839 ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht);
1840 break;
1841 case RELAYD_CLOSE_STREAM:
1842 ret = relay_close_stream(recv_hdr, cmd);
1843 break;
1844 case RELAYD_DATA_PENDING:
1845 ret = relay_data_pending(recv_hdr, cmd);
1846 break;
1847 case RELAYD_QUIESCENT_CONTROL:
1848 ret = relay_quiescent_control(recv_hdr, cmd);
1849 break;
1850 case RELAYD_BEGIN_DATA_PENDING:
1851 ret = relay_begin_data_pending(recv_hdr, cmd);
1852 break;
1853 case RELAYD_END_DATA_PENDING:
1854 ret = relay_end_data_pending(recv_hdr, cmd);
1855 break;
1856 case RELAYD_SEND_INDEX:
1857 ret = relay_recv_index(recv_hdr, cmd);
1858 break;
1859 case RELAYD_UPDATE_SYNC_INFO:
1860 default:
1861 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1862 relay_unknown_command(cmd);
1863 ret = -1;
1864 goto end;
1865 }
1866
1867 end:
1868 return ret;
1869 }
1870
1871 /*
1872 * Handle index for a data stream.
1873 *
1874 * RCU read side lock MUST be acquired.
1875 *
1876 * Return 0 on success else a negative value.
1877 */
1878 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
1879 int rotate_index)
1880 {
1881 int ret = 0, index_created = 0;
1882 uint64_t stream_id, data_offset;
1883 struct relay_index *index, *wr_index = NULL;
1884
1885 assert(stream);
1886
1887 stream_id = stream->stream_handle;
1888 /* Get data offset because we are about to update the index. */
1889 data_offset = htobe64(stream->tracefile_size_current);
1890
1891 /*
1892 * Lookup for an existing index for that stream id/sequence number. If on
1893 * exists, the control thread already received the data for it thus we need
1894 * to write it on disk.
1895 */
1896 index = relay_index_find(stream_id, net_seq_num);
1897 if (!index) {
1898 /* A successful creation will add the object to the HT. */
1899 index = relay_index_create(stream_id, net_seq_num);
1900 if (!index) {
1901 ret = -1;
1902 goto error;
1903 }
1904 index_created = 1;
1905 }
1906
1907 if (rotate_index || stream->index_fd < 0) {
1908 index->to_close_fd = stream->index_fd;
1909 ret = index_create_file(stream->path_name, stream->channel_name,
1910 relayd_uid, relayd_gid, stream->tracefile_size,
1911 stream->tracefile_count_current);
1912 if (ret < 0) {
1913 /* This will close the stream's index fd if one. */
1914 relay_index_free_safe(index);
1915 goto error;
1916 }
1917 stream->index_fd = ret;
1918 }
1919 index->fd = stream->index_fd;
1920 index->index_data.offset = data_offset;
1921
1922 if (index_created) {
1923 /*
1924 * Try to add the relay index object to the hash table. If an object
1925 * already exist, destroy back the index created and set the data.
1926 */
1927 relay_index_add(index, &wr_index);
1928 if (wr_index) {
1929 /* Copy back data from the created index. */
1930 wr_index->fd = index->fd;
1931 wr_index->to_close_fd = index->to_close_fd;
1932 wr_index->index_data.offset = data_offset;
1933 free(index);
1934 }
1935 } else {
1936 /* The index already exists so write it on disk. */
1937 wr_index = index;
1938 }
1939
1940 /* Do we have a writable ready index to write on disk. */
1941 if (wr_index) {
1942 ret = relay_index_write(wr_index->fd, wr_index);
1943 if (ret < 0) {
1944 goto error;
1945 }
1946 stream->total_index_received++;
1947 }
1948
1949 error:
1950 return ret;
1951 }
1952
1953 /*
1954 * relay_process_data: Process the data received on the data socket
1955 */
1956 static
1957 int relay_process_data(struct relay_command *cmd)
1958 {
1959 int ret = 0, rotate_index = 0;
1960 struct relay_stream *stream;
1961 struct lttcomm_relayd_data_hdr data_hdr;
1962 uint64_t stream_id;
1963 uint64_t net_seq_num;
1964 uint32_t data_size;
1965
1966 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1967 sizeof(struct lttcomm_relayd_data_hdr), 0);
1968 if (ret <= 0) {
1969 if (ret == 0) {
1970 /* Orderly shutdown. Not necessary to print an error. */
1971 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1972 } else {
1973 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
1974 }
1975 ret = -1;
1976 goto end;
1977 }
1978
1979 stream_id = be64toh(data_hdr.stream_id);
1980
1981 rcu_read_lock();
1982 stream = relay_stream_find_by_id(stream_id);
1983 if (!stream) {
1984 ret = -1;
1985 goto end_rcu_unlock;
1986 }
1987
1988 data_size = be32toh(data_hdr.data_size);
1989 if (data_buffer_size < data_size) {
1990 char *tmp_data_ptr;
1991
1992 tmp_data_ptr = realloc(data_buffer, data_size);
1993 if (!tmp_data_ptr) {
1994 ERR("Allocating data buffer");
1995 free(data_buffer);
1996 ret = -1;
1997 goto end_rcu_unlock;
1998 }
1999 data_buffer = tmp_data_ptr;
2000 data_buffer_size = data_size;
2001 }
2002 memset(data_buffer, 0, data_size);
2003
2004 net_seq_num = be64toh(data_hdr.net_seq_num);
2005
2006 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2007 data_size, stream_id, net_seq_num);
2008 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
2009 if (ret <= 0) {
2010 if (ret == 0) {
2011 /* Orderly shutdown. Not necessary to print an error. */
2012 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2013 }
2014 ret = -1;
2015 goto end_rcu_unlock;
2016 }
2017
2018 /* Check if a rotation is needed. */
2019 if (stream->tracefile_size > 0 &&
2020 (stream->tracefile_size_current + data_size) >
2021 stream->tracefile_size) {
2022 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2023 stream->tracefile_size, stream->tracefile_count,
2024 relayd_uid, relayd_gid, stream->fd,
2025 &(stream->tracefile_count_current), &stream->fd);
2026 if (ret < 0) {
2027 ERR("Rotating stream output file");
2028 goto end_rcu_unlock;
2029 }
2030 /* Reset current size because we just perform a stream rotation. */
2031 stream->tracefile_size_current = 0;
2032 rotate_index = 1;
2033 }
2034
2035 /*
2036 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2037 * index are NOT supported.
2038 */
2039 if (stream->session->minor >= 4 && !stream->session->snapshot) {
2040 ret = handle_index_data(stream, net_seq_num, rotate_index);
2041 if (ret < 0) {
2042 goto end_rcu_unlock;
2043 }
2044 }
2045
2046 /* Write data to stream output fd. */
2047 do {
2048 ret = write(stream->fd, data_buffer, data_size);
2049 } while (ret < 0 && errno == EINTR);
2050 if (ret < 0 || ret != data_size) {
2051 ERR("Relay error writing data to file");
2052 ret = -1;
2053 goto end_rcu_unlock;
2054 }
2055
2056 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2057 ret, stream->stream_handle);
2058
2059 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2060 if (ret < 0) {
2061 goto end_rcu_unlock;
2062 }
2063 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2064
2065 stream->prev_seq = net_seq_num;
2066
2067 /* Check if we need to close the FD */
2068 if (close_stream_check(stream)) {
2069 destroy_stream(stream, cmd->ctf_traces_ht);
2070 }
2071
2072 end_rcu_unlock:
2073 rcu_read_unlock();
2074 end:
2075 return ret;
2076 }
2077
2078 static
2079 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
2080 {
2081 int ret;
2082
2083 lttng_poll_del(events, pollfd);
2084
2085 ret = close(pollfd);
2086 if (ret < 0) {
2087 ERR("Closing pollfd %d", pollfd);
2088 }
2089 }
2090
2091 static
2092 int relay_add_connection(int fd, struct lttng_poll_event *events,
2093 struct lttng_ht *relay_connections_ht)
2094 {
2095 struct relay_command *relay_connection;
2096 int ret;
2097
2098 relay_connection = zmalloc(sizeof(struct relay_command));
2099 if (relay_connection == NULL) {
2100 PERROR("Relay command zmalloc");
2101 goto error;
2102 }
2103 do {
2104 ret = read(fd, relay_connection, sizeof(struct relay_command));
2105 } while (ret < 0 && errno == EINTR);
2106 if (ret < 0 || ret < sizeof(struct relay_command)) {
2107 PERROR("read relay cmd pipe");
2108 goto error_read;
2109 }
2110
2111 relay_connection->ctf_traces_ht = lttng_ht_new(0, LTTNG_HT_TYPE_STRING);
2112 if (!relay_connection->ctf_traces_ht) {
2113 goto error_read;
2114 }
2115
2116 lttng_ht_node_init_ulong(&relay_connection->sock_n,
2117 (unsigned long) relay_connection->sock->fd);
2118 rcu_read_lock();
2119 lttng_ht_add_unique_ulong(relay_connections_ht,
2120 &relay_connection->sock_n);
2121 rcu_read_unlock();
2122 return lttng_poll_add(events,
2123 relay_connection->sock->fd,
2124 LPOLLIN | LPOLLRDHUP);
2125
2126 error_read:
2127 free(relay_connection);
2128 error:
2129 return -1;
2130 }
2131
2132 static
2133 void deferred_free_connection(struct rcu_head *head)
2134 {
2135 struct relay_command *relay_connection =
2136 caa_container_of(head, struct relay_command, rcu_node);
2137
2138 lttng_ht_destroy(relay_connection->ctf_traces_ht);
2139 lttcomm_destroy_sock(relay_connection->sock);
2140 free(relay_connection);
2141 }
2142
2143 static
2144 void relay_del_connection(struct lttng_ht *relay_connections_ht,
2145 struct lttng_ht_iter *iter, struct relay_command *relay_connection,
2146 struct lttng_ht *sessions_ht)
2147 {
2148 int ret;
2149
2150 ret = lttng_ht_del(relay_connections_ht, iter);
2151 assert(!ret);
2152 if (relay_connection->type == RELAY_CONTROL) {
2153 relay_delete_session(relay_connection, sessions_ht);
2154 }
2155
2156 call_rcu(&relay_connection->rcu_node,
2157 deferred_free_connection);
2158 }
2159
2160 /*
2161 * This thread does the actual work
2162 */
2163 static
2164 void *relay_thread_worker(void *data)
2165 {
2166 int ret, err = -1, last_seen_data_fd = -1;
2167 uint32_t nb_fd;
2168 struct relay_command *relay_connection;
2169 struct lttng_poll_event events;
2170 struct lttng_ht *relay_connections_ht;
2171 struct lttng_ht_node_ulong *node;
2172 struct lttng_ht_iter iter;
2173 struct lttcomm_relayd_hdr recv_hdr;
2174 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2175 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2176
2177 DBG("[thread] Relay worker started");
2178
2179 rcu_register_thread();
2180
2181 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2182
2183 health_code_update();
2184
2185 /* table of connections indexed on socket */
2186 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2187 if (!relay_connections_ht) {
2188 goto relay_connections_ht_error;
2189 }
2190
2191 /* Tables of received indexes indexed by index handle and net_seq_num. */
2192 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2193 if (!indexes_ht) {
2194 goto indexes_ht_error;
2195 }
2196
2197 ret = create_thread_poll_set(&events, 2);
2198 if (ret < 0) {
2199 goto error_poll_create;
2200 }
2201
2202 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
2203 if (ret < 0) {
2204 goto error;
2205 }
2206
2207 restart:
2208 while (1) {
2209 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2210
2211 health_code_update();
2212
2213 /* Infinite blocking call, waiting for transmission */
2214 DBG3("Relayd worker thread polling...");
2215 health_poll_entry();
2216 ret = lttng_poll_wait(&events, -1);
2217 health_poll_exit();
2218 if (ret < 0) {
2219 /*
2220 * Restart interrupted system call.
2221 */
2222 if (errno == EINTR) {
2223 goto restart;
2224 }
2225 goto error;
2226 }
2227
2228 nb_fd = ret;
2229
2230 /*
2231 * Process control. The control connection is prioritised so we don't
2232 * starve it with high throughout put tracing data on the data
2233 * connection.
2234 */
2235 for (i = 0; i < nb_fd; i++) {
2236 /* Fetch once the poll data */
2237 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2238 int pollfd = LTTNG_POLL_GETFD(&events, i);
2239
2240 health_code_update();
2241
2242 /* Thread quit pipe has been closed. Killing thread. */
2243 ret = check_thread_quit_pipe(pollfd, revents);
2244 if (ret) {
2245 err = 0;
2246 goto exit;
2247 }
2248
2249 /* Inspect the relay cmd pipe for new connection */
2250 if (pollfd == relay_cmd_pipe[0]) {
2251 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2252 ERR("Relay pipe error");
2253 goto error;
2254 } else if (revents & LPOLLIN) {
2255 DBG("Relay command received");
2256 ret = relay_add_connection(relay_cmd_pipe[0],
2257 &events, relay_connections_ht);
2258 if (ret < 0) {
2259 goto error;
2260 }
2261 }
2262 } else if (revents) {
2263 rcu_read_lock();
2264 lttng_ht_lookup(relay_connections_ht,
2265 (void *)((unsigned long) pollfd),
2266 &iter);
2267 node = lttng_ht_iter_get_node_ulong(&iter);
2268 if (node == NULL) {
2269 DBG2("Relay sock %d not found", pollfd);
2270 rcu_read_unlock();
2271 goto error;
2272 }
2273 relay_connection = caa_container_of(node,
2274 struct relay_command, sock_n);
2275
2276 if (revents & (LPOLLERR)) {
2277 ERR("POLL ERROR");
2278 relay_cleanup_poll_connection(&events, pollfd);
2279 relay_del_connection(relay_connections_ht,
2280 &iter, relay_connection, sessions_ht);
2281 if (last_seen_data_fd == pollfd) {
2282 last_seen_data_fd = last_notdel_data_fd;
2283 }
2284 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2285 DBG("Socket %d hung up", pollfd);
2286 relay_cleanup_poll_connection(&events, pollfd);
2287 relay_del_connection(relay_connections_ht,
2288 &iter, relay_connection, sessions_ht);
2289 if (last_seen_data_fd == pollfd) {
2290 last_seen_data_fd = last_notdel_data_fd;
2291 }
2292 } else if (revents & LPOLLIN) {
2293 /* control socket */
2294 if (relay_connection->type == RELAY_CONTROL) {
2295 ret = relay_connection->sock->ops->recvmsg(
2296 relay_connection->sock, &recv_hdr,
2297 sizeof(struct lttcomm_relayd_hdr), 0);
2298 /* connection closed */
2299 if (ret <= 0) {
2300 relay_cleanup_poll_connection(&events, pollfd);
2301 relay_del_connection(relay_connections_ht,
2302 &iter, relay_connection, sessions_ht);
2303 DBG("Control connection closed with %d", pollfd);
2304 } else {
2305 if (relay_connection->session) {
2306 DBG2("Relay worker receiving data for session : %" PRIu64,
2307 relay_connection->session->id);
2308 }
2309 ret = relay_process_control(&recv_hdr,
2310 relay_connection, relay_ctx);
2311 if (ret < 0) {
2312 /* Clear the session on error. */
2313 relay_cleanup_poll_connection(&events, pollfd);
2314 relay_del_connection(relay_connections_ht,
2315 &iter, relay_connection, sessions_ht);
2316 DBG("Connection closed with %d", pollfd);
2317 }
2318 seen_control = 1;
2319 }
2320 } else {
2321 /*
2322 * Flag the last seen data fd not deleted. It will be
2323 * used as the last seen fd if any fd gets deleted in
2324 * this first loop.
2325 */
2326 last_notdel_data_fd = pollfd;
2327 }
2328 }
2329 rcu_read_unlock();
2330 }
2331 }
2332
2333 /*
2334 * The last loop handled a control request, go back to poll to make
2335 * sure we prioritise the control socket.
2336 */
2337 if (seen_control) {
2338 continue;
2339 }
2340
2341 if (last_seen_data_fd >= 0) {
2342 for (i = 0; i < nb_fd; i++) {
2343 int pollfd = LTTNG_POLL_GETFD(&events, i);
2344
2345 health_code_update();
2346
2347 if (last_seen_data_fd == pollfd) {
2348 idx = i;
2349 break;
2350 }
2351 }
2352 }
2353
2354 /* Process data connection. */
2355 for (i = idx + 1; i < nb_fd; i++) {
2356 /* Fetch the poll data. */
2357 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2358 int pollfd = LTTNG_POLL_GETFD(&events, i);
2359
2360 health_code_update();
2361
2362 /* Skip the command pipe. It's handled in the first loop. */
2363 if (pollfd == relay_cmd_pipe[0]) {
2364 continue;
2365 }
2366
2367 if (revents) {
2368 rcu_read_lock();
2369 lttng_ht_lookup(relay_connections_ht,
2370 (void *)((unsigned long) pollfd),
2371 &iter);
2372 node = lttng_ht_iter_get_node_ulong(&iter);
2373 if (node == NULL) {
2374 /* Skip it. Might be removed before. */
2375 rcu_read_unlock();
2376 continue;
2377 }
2378 relay_connection = caa_container_of(node,
2379 struct relay_command, sock_n);
2380
2381 if (revents & LPOLLIN) {
2382 if (relay_connection->type != RELAY_DATA) {
2383 continue;
2384 }
2385
2386 ret = relay_process_data(relay_connection);
2387 /* connection closed */
2388 if (ret < 0) {
2389 relay_cleanup_poll_connection(&events, pollfd);
2390 relay_del_connection(relay_connections_ht,
2391 &iter, relay_connection, sessions_ht);
2392 DBG("Data connection closed with %d", pollfd);
2393 /*
2394 * Every goto restart call sets the last seen fd where
2395 * here we don't really care since we gracefully
2396 * continue the loop after the connection is deleted.
2397 */
2398 } else {
2399 /* Keep last seen port. */
2400 last_seen_data_fd = pollfd;
2401 rcu_read_unlock();
2402 goto restart;
2403 }
2404 }
2405 rcu_read_unlock();
2406 }
2407 }
2408 last_seen_data_fd = -1;
2409 }
2410
2411 /* Normal exit, no error */
2412 ret = 0;
2413
2414 exit:
2415 error:
2416 lttng_poll_clean(&events);
2417
2418 /* empty the hash table and free the memory */
2419 rcu_read_lock();
2420 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2421 health_code_update();
2422
2423 node = lttng_ht_iter_get_node_ulong(&iter);
2424 if (node) {
2425 relay_connection = caa_container_of(node,
2426 struct relay_command, sock_n);
2427 relay_del_connection(relay_connections_ht,
2428 &iter, relay_connection, sessions_ht);
2429 }
2430 }
2431 rcu_read_unlock();
2432 error_poll_create:
2433 lttng_ht_destroy(indexes_ht);
2434 indexes_ht_error:
2435 lttng_ht_destroy(relay_connections_ht);
2436 relay_connections_ht_error:
2437 /* Close relay cmd pipes */
2438 utils_close_pipe(relay_cmd_pipe);
2439 if (err) {
2440 DBG("Thread exited with error");
2441 }
2442 DBG("Worker thread cleanup complete");
2443 free(data_buffer);
2444 if (err) {
2445 health_error();
2446 ERR("Health error occurred in %s", __func__);
2447 }
2448 health_unregister(health_relayd);
2449 rcu_unregister_thread();
2450 stop_threads();
2451 return NULL;
2452 }
2453
2454 /*
2455 * Create the relay command pipe to wake thread_manage_apps.
2456 * Closed in cleanup().
2457 */
2458 static int create_relay_cmd_pipe(void)
2459 {
2460 int ret;
2461
2462 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2463
2464 return ret;
2465 }
2466
2467 /*
2468 * main
2469 */
2470 int main(int argc, char **argv)
2471 {
2472 int ret = 0;
2473 void *status;
2474 struct relay_local_data *relay_ctx;
2475
2476 /* Create thread quit pipe */
2477 if ((ret = init_thread_quit_pipe()) < 0) {
2478 goto error;
2479 }
2480
2481 /* Parse arguments */
2482 progname = argv[0];
2483 if ((ret = parse_args(argc, argv)) < 0) {
2484 goto exit;
2485 }
2486
2487 if ((ret = set_signal_handler()) < 0) {
2488 goto exit;
2489 }
2490
2491 /* Try to create directory if -o, --output is specified. */
2492 if (opt_output_path) {
2493 if (*opt_output_path != '/') {
2494 ERR("Please specify an absolute path for -o, --output PATH");
2495 goto exit;
2496 }
2497
2498 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2499 if (ret < 0) {
2500 ERR("Unable to create %s", opt_output_path);
2501 goto exit;
2502 }
2503 }
2504
2505 /* Daemonize */
2506 if (opt_daemon) {
2507 ret = daemon(0, 0);
2508 if (ret < 0) {
2509 PERROR("daemon");
2510 goto exit;
2511 }
2512 }
2513
2514 /* We need those values for the file/dir creation. */
2515 relayd_uid = getuid();
2516 relayd_gid = getgid();
2517
2518 /* Check if daemon is UID = 0 */
2519 if (relayd_uid == 0) {
2520 if (control_uri->port < 1024 || data_uri->port < 1024) {
2521 ERR("Need to be root to use ports < 1024");
2522 ret = -1;
2523 goto exit;
2524 }
2525 }
2526
2527 /* Setup the thread apps communication pipe. */
2528 if ((ret = create_relay_cmd_pipe()) < 0) {
2529 goto exit;
2530 }
2531
2532 /* Init relay command queue. */
2533 cds_wfq_init(&relay_cmd_queue.queue);
2534
2535 /* Set up max poll set size */
2536 lttng_poll_set_max_size();
2537
2538 /* Initialize communication library */
2539 lttcomm_init();
2540
2541 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2542 if (!relay_ctx) {
2543 PERROR("relay_ctx");
2544 goto exit;
2545 }
2546
2547 /* tables of sessions indexed by session ID */
2548 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2549 if (!relay_ctx->sessions_ht) {
2550 goto exit_relay_ctx_sessions;
2551 }
2552
2553 /* tables of streams indexed by stream ID */
2554 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2555 if (!relay_streams_ht) {
2556 goto exit_relay_ctx_streams;
2557 }
2558
2559 /* tables of streams indexed by stream ID */
2560 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2561 if (!viewer_streams_ht) {
2562 goto exit_relay_ctx_viewer_streams;
2563 }
2564
2565 /* Initialize thread health monitoring */
2566 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2567 if (!health_relayd) {
2568 PERROR("health_app_create error");
2569 goto exit_health_app_create;
2570 }
2571
2572 /* Setup the dispatcher thread */
2573 ret = pthread_create(&dispatcher_thread, NULL,
2574 relay_thread_dispatcher, (void *) NULL);
2575 if (ret != 0) {
2576 PERROR("pthread_create dispatcher");
2577 goto exit_dispatcher;
2578 }
2579
2580 /* Setup the worker thread */
2581 ret = pthread_create(&worker_thread, NULL,
2582 relay_thread_worker, (void *) relay_ctx);
2583 if (ret != 0) {
2584 PERROR("pthread_create worker");
2585 goto exit_worker;
2586 }
2587
2588 /* Setup the listener thread */
2589 ret = pthread_create(&listener_thread, NULL,
2590 relay_thread_listener, (void *) NULL);
2591 if (ret != 0) {
2592 PERROR("pthread_create listener");
2593 goto exit_listener;
2594 }
2595
2596 ret = live_start_threads(live_uri, relay_ctx, thread_quit_pipe);
2597 if (ret != 0) {
2598 ERR("Starting live viewer threads");
2599 goto exit_live;
2600 }
2601
2602 live_stop_threads();
2603
2604 exit_live:
2605 ret = pthread_join(listener_thread, &status);
2606 if (ret != 0) {
2607 PERROR("pthread_join");
2608 goto error; /* join error, exit without cleanup */
2609 }
2610
2611 exit_listener:
2612 ret = pthread_join(worker_thread, &status);
2613 if (ret != 0) {
2614 PERROR("pthread_join");
2615 goto error; /* join error, exit without cleanup */
2616 }
2617
2618 exit_worker:
2619 ret = pthread_join(dispatcher_thread, &status);
2620 if (ret != 0) {
2621 PERROR("pthread_join");
2622 goto error; /* join error, exit without cleanup */
2623 }
2624
2625 exit_dispatcher:
2626 health_app_destroy(health_relayd);
2627
2628 exit_health_app_create:
2629 lttng_ht_destroy(viewer_streams_ht);
2630
2631 exit_relay_ctx_viewer_streams:
2632 lttng_ht_destroy(relay_streams_ht);
2633
2634 exit_relay_ctx_streams:
2635 lttng_ht_destroy(relay_ctx->sessions_ht);
2636
2637 exit_relay_ctx_sessions:
2638 free(relay_ctx);
2639
2640 exit:
2641 cleanup();
2642 if (!ret) {
2643 exit(EXIT_SUCCESS);
2644 }
2645
2646 error:
2647 exit(EXIT_FAILURE);
2648 }
This page took 0.134521 seconds and 3 git commands to generate.