Fix: metadata stream should be always flagged as ready
[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 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
72
73 /*
74 * Quit pipe for all threads. This permits a single cancellation point
75 * for all threads when receiving an event on the pipe.
76 */
77 static int thread_quit_pipe[2] = { -1, -1 };
78
79 /*
80 * This pipe is used to inform the worker thread that a command is queued and
81 * ready to be processed.
82 */
83 static int relay_cmd_pipe[2] = { -1, -1 };
84
85 /* Shared between threads */
86 static int dispatch_thread_exit;
87
88 static pthread_t listener_thread;
89 static pthread_t dispatcher_thread;
90 static pthread_t worker_thread;
91 static pthread_t health_thread;
92
93 static uint64_t last_relay_stream_id;
94 static uint64_t last_relay_session_id;
95
96 /*
97 * Relay command queue.
98 *
99 * The relay_thread_listener and relay_thread_dispatcher communicate with this
100 * queue.
101 */
102 static struct relay_cmd_queue relay_cmd_queue;
103
104 /* buffer allocated at startup, used to store the trace data */
105 static char *data_buffer;
106 static unsigned int data_buffer_size;
107
108 /* We need those values for the file/dir creation. */
109 static uid_t relayd_uid;
110 static gid_t relayd_gid;
111
112 /* Global relay stream hash table. */
113 struct lttng_ht *relay_streams_ht;
114
115 /* Global relay viewer stream hash table. */
116 struct lttng_ht *viewer_streams_ht;
117
118 /* Global hash table that stores relay index object. */
119 struct lttng_ht *indexes_ht;
120
121 /* Relayd health monitoring */
122 struct health_app *health_relayd;
123
124 /*
125 * usage function on stderr
126 */
127 static
128 void usage(void)
129 {
130 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
131 fprintf(stderr, " -h, --help Display this usage.\n");
132 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
133 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
134 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
135 fprintf(stderr, " -L, --live-port URL Live view port listening.\n");
136 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
137 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
138 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
139 }
140
141 static
142 int parse_args(int argc, char **argv)
143 {
144 int c;
145 int ret = 0;
146 char *default_address;
147
148 static struct option long_options[] = {
149 { "control-port", 1, 0, 'C', },
150 { "data-port", 1, 0, 'D', },
151 { "daemonize", 0, 0, 'd', },
152 { "group", 1, 0, 'g', },
153 { "help", 0, 0, 'h', },
154 { "output", 1, 0, 'o', },
155 { "verbose", 0, 0, 'v', },
156 { NULL, 0, 0, 0, },
157 };
158
159 while (1) {
160 int option_index = 0;
161 c = getopt_long(argc, argv, "dhv" "C:D:L:o:g:",
162 long_options, &option_index);
163 if (c == -1) {
164 break;
165 }
166
167 switch (c) {
168 case 0:
169 fprintf(stderr, "option %s", long_options[option_index].name);
170 if (optarg) {
171 fprintf(stderr, " with arg %s\n", optarg);
172 }
173 break;
174 case 'C':
175 ret = uri_parse(optarg, &control_uri);
176 if (ret < 0) {
177 ERR("Invalid control URI specified");
178 goto exit;
179 }
180 if (control_uri->port == 0) {
181 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
182 }
183 break;
184 case 'D':
185 ret = uri_parse(optarg, &data_uri);
186 if (ret < 0) {
187 ERR("Invalid data URI specified");
188 goto exit;
189 }
190 if (data_uri->port == 0) {
191 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
192 }
193 break;
194 case 'L':
195 ret = uri_parse(optarg, &live_uri);
196 if (ret < 0) {
197 ERR("Invalid live URI specified");
198 goto exit;
199 }
200 if (live_uri->port == 0) {
201 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
202 }
203 break;
204 case 'd':
205 opt_daemon = 1;
206 break;
207 case 'g':
208 tracing_group_name = optarg;
209 break;
210 case 'h':
211 usage();
212 exit(EXIT_FAILURE);
213 case 'o':
214 ret = asprintf(&opt_output_path, "%s", optarg);
215 if (ret < 0) {
216 PERROR("asprintf opt_output_path");
217 goto exit;
218 }
219 break;
220 case 'v':
221 /* Verbose level can increase using multiple -v */
222 lttng_opt_verbose += 1;
223 break;
224 default:
225 /* Unknown option or other error.
226 * Error is printed by getopt, just return */
227 ret = -1;
228 goto exit;
229 }
230 }
231
232 /* assign default values */
233 if (control_uri == NULL) {
234 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
235 DEFAULT_NETWORK_CONTROL_PORT);
236 if (ret < 0) {
237 PERROR("asprintf default data address");
238 goto exit;
239 }
240
241 ret = uri_parse(default_address, &control_uri);
242 free(default_address);
243 if (ret < 0) {
244 ERR("Invalid control URI specified");
245 goto exit;
246 }
247 }
248 if (data_uri == NULL) {
249 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
250 DEFAULT_NETWORK_DATA_PORT);
251 if (ret < 0) {
252 PERROR("asprintf default data address");
253 goto exit;
254 }
255
256 ret = uri_parse(default_address, &data_uri);
257 free(default_address);
258 if (ret < 0) {
259 ERR("Invalid data URI specified");
260 goto exit;
261 }
262 }
263 if (live_uri == NULL) {
264 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
265 DEFAULT_NETWORK_VIEWER_PORT);
266 if (ret < 0) {
267 PERROR("asprintf default viewer control address");
268 goto exit;
269 }
270
271 ret = uri_parse(default_address, &live_uri);
272 free(default_address);
273 if (ret < 0) {
274 ERR("Invalid viewer control URI specified");
275 goto exit;
276 }
277 }
278
279 exit:
280 return ret;
281 }
282
283 /*
284 * Cleanup the daemon
285 */
286 static
287 void cleanup(void)
288 {
289 DBG("Cleaning up");
290
291 /* free the dynamically allocated opt_output_path */
292 free(opt_output_path);
293
294 /* Close thread quit pipes */
295 utils_close_pipe(thread_quit_pipe);
296
297 uri_free(control_uri);
298 uri_free(data_uri);
299 /* Live URI is freed in the live thread. */
300 }
301
302 /*
303 * Write to writable pipe used to notify a thread.
304 */
305 static
306 int notify_thread_pipe(int wpipe)
307 {
308 ssize_t ret;
309
310 ret = lttng_write(wpipe, "!", 1);
311 if (ret < 1) {
312 PERROR("write poll pipe");
313 }
314
315 return ret;
316 }
317
318 static void notify_health_quit_pipe(int *pipe)
319 {
320 ssize_t ret;
321
322 ret = lttng_write(pipe[1], "4", 1);
323 if (ret < 1) {
324 PERROR("write relay health quit");
325 }
326 }
327
328 /*
329 * Stop all threads by closing the thread quit pipe.
330 */
331 static
332 void stop_threads(void)
333 {
334 int ret;
335
336 /* Stopping all threads */
337 DBG("Terminating all threads");
338 ret = notify_thread_pipe(thread_quit_pipe[1]);
339 if (ret < 0) {
340 ERR("write error on thread quit pipe");
341 }
342
343 notify_health_quit_pipe(health_quit_pipe);
344
345 /* Dispatch thread */
346 CMM_STORE_SHARED(dispatch_thread_exit, 1);
347 futex_nto1_wake(&relay_cmd_queue.futex);
348 }
349
350 /*
351 * Signal handler for the daemon
352 *
353 * Simply stop all worker threads, leaving main() return gracefully after
354 * joining all threads and calling cleanup().
355 */
356 static
357 void sighandler(int sig)
358 {
359 switch (sig) {
360 case SIGPIPE:
361 DBG("SIGPIPE caught");
362 return;
363 case SIGINT:
364 DBG("SIGINT caught");
365 stop_threads();
366 break;
367 case SIGTERM:
368 DBG("SIGTERM caught");
369 stop_threads();
370 break;
371 default:
372 break;
373 }
374 }
375
376 /*
377 * Setup signal handler for :
378 * SIGINT, SIGTERM, SIGPIPE
379 */
380 static
381 int set_signal_handler(void)
382 {
383 int ret = 0;
384 struct sigaction sa;
385 sigset_t sigset;
386
387 if ((ret = sigemptyset(&sigset)) < 0) {
388 PERROR("sigemptyset");
389 return ret;
390 }
391
392 sa.sa_handler = sighandler;
393 sa.sa_mask = sigset;
394 sa.sa_flags = 0;
395 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
396 PERROR("sigaction");
397 return ret;
398 }
399
400 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
401 PERROR("sigaction");
402 return ret;
403 }
404
405 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
406 PERROR("sigaction");
407 return ret;
408 }
409
410 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
411
412 return ret;
413 }
414
415 /*
416 * Init thread quit pipe.
417 *
418 * Return -1 on error or 0 if all pipes are created.
419 */
420 static
421 int init_thread_quit_pipe(void)
422 {
423 int ret;
424
425 ret = utils_create_pipe_cloexec(thread_quit_pipe);
426
427 return ret;
428 }
429
430 /*
431 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
432 */
433 static
434 int create_thread_poll_set(struct lttng_poll_event *events, int size)
435 {
436 int ret;
437
438 if (events == NULL || size == 0) {
439 ret = -1;
440 goto error;
441 }
442
443 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
444 if (ret < 0) {
445 goto error;
446 }
447
448 /* Add quit pipe */
449 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
450 if (ret < 0) {
451 goto error;
452 }
453
454 return 0;
455
456 error:
457 return ret;
458 }
459
460 /*
461 * Check if the thread quit pipe was triggered.
462 *
463 * Return 1 if it was triggered else 0;
464 */
465 static
466 int check_thread_quit_pipe(int fd, uint32_t events)
467 {
468 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
469 return 1;
470 }
471
472 return 0;
473 }
474
475 /*
476 * Create and init socket from uri.
477 */
478 static
479 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
480 {
481 int ret;
482 struct lttcomm_sock *sock = NULL;
483
484 sock = lttcomm_alloc_sock_from_uri(uri);
485 if (sock == NULL) {
486 ERR("Allocating socket");
487 goto error;
488 }
489
490 ret = lttcomm_create_sock(sock);
491 if (ret < 0) {
492 goto error;
493 }
494 DBG("Listening on sock %d", sock->fd);
495
496 ret = sock->ops->bind(sock);
497 if (ret < 0) {
498 goto error;
499 }
500
501 ret = sock->ops->listen(sock, -1);
502 if (ret < 0) {
503 goto error;
504
505 }
506
507 return sock;
508
509 error:
510 if (sock) {
511 lttcomm_destroy_sock(sock);
512 }
513 return NULL;
514 }
515
516 /*
517 * Return nonzero if stream needs to be closed.
518 */
519 static
520 int close_stream_check(struct relay_stream *stream)
521 {
522
523 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
524 /*
525 * We are about to close the stream so set the data pending flag to 1
526 * which will make the end data pending command skip the stream which
527 * is now closed and ready. Note that after proceeding to a file close,
528 * the written file is ready for reading.
529 */
530 stream->data_pending_check_done = 1;
531 return 1;
532 }
533 return 0;
534 }
535
536 /*
537 * This thread manages the listening for new connections on the network
538 */
539 static
540 void *relay_thread_listener(void *data)
541 {
542 int i, ret, pollfd, err = -1;
543 int val = 1;
544 uint32_t revents, nb_fd;
545 struct lttng_poll_event events;
546 struct lttcomm_sock *control_sock, *data_sock;
547
548 DBG("[thread] Relay listener started");
549
550 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
551
552 health_code_update();
553
554 control_sock = relay_init_sock(control_uri);
555 if (!control_sock) {
556 goto error_sock_control;
557 }
558
559 data_sock = relay_init_sock(data_uri);
560 if (!data_sock) {
561 goto error_sock_relay;
562 }
563
564 /*
565 * Pass 3 as size here for the thread quit pipe, control and data socket.
566 */
567 ret = create_thread_poll_set(&events, 3);
568 if (ret < 0) {
569 goto error_create_poll;
570 }
571
572 /* Add the control socket */
573 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
574 if (ret < 0) {
575 goto error_poll_add;
576 }
577
578 /* Add the data socket */
579 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
580 if (ret < 0) {
581 goto error_poll_add;
582 }
583
584 while (1) {
585 health_code_update();
586
587 DBG("Listener accepting connections");
588
589 restart:
590 health_poll_entry();
591 ret = lttng_poll_wait(&events, -1);
592 health_poll_exit();
593 if (ret < 0) {
594 /*
595 * Restart interrupted system call.
596 */
597 if (errno == EINTR) {
598 goto restart;
599 }
600 goto error;
601 }
602
603 nb_fd = ret;
604
605 DBG("Relay new connection received");
606 for (i = 0; i < nb_fd; i++) {
607 health_code_update();
608
609 /* Fetch once the poll data */
610 revents = LTTNG_POLL_GETEV(&events, i);
611 pollfd = LTTNG_POLL_GETFD(&events, i);
612
613 /* Thread quit pipe has been closed. Killing thread. */
614 ret = check_thread_quit_pipe(pollfd, revents);
615 if (ret) {
616 err = 0;
617 goto exit;
618 }
619
620 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
621 ERR("socket poll error");
622 goto error;
623 } else if (revents & LPOLLIN) {
624 /*
625 * Get allocated in this thread,
626 * enqueued to a global queue, dequeued
627 * and freed in the worker thread.
628 */
629 struct relay_command *relay_cmd;
630 struct lttcomm_sock *newsock;
631
632 relay_cmd = zmalloc(sizeof(struct relay_command));
633 if (relay_cmd == NULL) {
634 PERROR("relay command zmalloc");
635 goto error;
636 }
637
638 if (pollfd == data_sock->fd) {
639 newsock = data_sock->ops->accept(data_sock);
640 if (!newsock) {
641 PERROR("accepting data sock");
642 free(relay_cmd);
643 goto error;
644 }
645 relay_cmd->type = RELAY_DATA;
646 DBG("Relay data connection accepted, socket %d", newsock->fd);
647 } else {
648 assert(pollfd == control_sock->fd);
649 newsock = control_sock->ops->accept(control_sock);
650 if (!newsock) {
651 PERROR("accepting control sock");
652 free(relay_cmd);
653 goto error;
654 }
655 relay_cmd->type = RELAY_CONTROL;
656 DBG("Relay control connection accepted, socket %d", newsock->fd);
657 }
658 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
659 &val, sizeof(int));
660 if (ret < 0) {
661 PERROR("setsockopt inet");
662 lttcomm_destroy_sock(newsock);
663 free(relay_cmd);
664 goto error;
665 }
666 relay_cmd->sock = newsock;
667 /*
668 * Lock free enqueue the request.
669 */
670 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
671
672 /*
673 * Wake the dispatch queue futex. Implicit memory
674 * barrier with the exchange in cds_wfq_enqueue.
675 */
676 futex_nto1_wake(&relay_cmd_queue.futex);
677 }
678 }
679 }
680
681 exit:
682 error:
683 error_poll_add:
684 lttng_poll_clean(&events);
685 error_create_poll:
686 if (data_sock->fd >= 0) {
687 ret = data_sock->ops->close(data_sock);
688 if (ret) {
689 PERROR("close");
690 }
691 }
692 lttcomm_destroy_sock(data_sock);
693 error_sock_relay:
694 if (control_sock->fd >= 0) {
695 ret = control_sock->ops->close(control_sock);
696 if (ret) {
697 PERROR("close");
698 }
699 }
700 lttcomm_destroy_sock(control_sock);
701 error_sock_control:
702 if (err) {
703 health_error();
704 ERR("Health error occurred in %s", __func__);
705 }
706 health_unregister(health_relayd);
707 DBG("Relay listener thread cleanup complete");
708 stop_threads();
709 return NULL;
710 }
711
712 /*
713 * This thread manages the dispatching of the requests to worker threads
714 */
715 static
716 void *relay_thread_dispatcher(void *data)
717 {
718 int err = -1;
719 ssize_t ret;
720 struct cds_wfq_node *node;
721 struct relay_command *relay_cmd = NULL;
722
723 DBG("[thread] Relay dispatcher started");
724
725 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
726
727 health_code_update();
728
729 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
730 health_code_update();
731
732 /* Atomically prepare the queue futex */
733 futex_nto1_prepare(&relay_cmd_queue.futex);
734
735 do {
736 health_code_update();
737
738 /* Dequeue commands */
739 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
740 if (node == NULL) {
741 DBG("Woken up but nothing in the relay command queue");
742 /* Continue thread execution */
743 break;
744 }
745
746 relay_cmd = caa_container_of(node, struct relay_command, node);
747 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
748
749 /*
750 * Inform worker thread of the new request. This
751 * call is blocking so we can be assured that the data will be read
752 * at some point in time or wait to the end of the world :)
753 */
754 ret = lttng_write(relay_cmd_pipe[1], relay_cmd,
755 sizeof(struct relay_command));
756 free(relay_cmd);
757 if (ret < sizeof(struct relay_command)) {
758 PERROR("write cmd pipe");
759 goto error;
760 }
761 } while (node != NULL);
762
763 /* Futex wait on queue. Blocking call on futex() */
764 health_poll_entry();
765 futex_nto1_wait(&relay_cmd_queue.futex);
766 health_poll_exit();
767 }
768
769 /* Normal exit, no error */
770 err = 0;
771
772 error:
773 if (err) {
774 health_error();
775 ERR("Health error occurred in %s", __func__);
776 }
777 health_unregister(health_relayd);
778 DBG("Dispatch thread dying");
779 stop_threads();
780 return NULL;
781 }
782
783 /*
784 * Get stream from stream id.
785 * Need to be called with RCU read-side lock held.
786 */
787 struct relay_stream *relay_stream_find_by_id(uint64_t stream_id)
788 {
789 struct lttng_ht_node_ulong *node;
790 struct lttng_ht_iter iter;
791 struct relay_stream *ret;
792
793 lttng_ht_lookup(relay_streams_ht,
794 (void *)((unsigned long) stream_id),
795 &iter);
796 node = lttng_ht_iter_get_node_ulong(&iter);
797 if (node == NULL) {
798 DBG("Relay stream %" PRIu64 " not found", stream_id);
799 ret = NULL;
800 goto end;
801 }
802
803 ret = caa_container_of(node, struct relay_stream, stream_n);
804
805 end:
806 return ret;
807 }
808
809 static
810 void deferred_free_stream(struct rcu_head *head)
811 {
812 struct relay_stream *stream =
813 caa_container_of(head, struct relay_stream, rcu_node);
814
815 free(stream->path_name);
816 free(stream->channel_name);
817 free(stream);
818 }
819
820 static
821 void deferred_free_session(struct rcu_head *head)
822 {
823 struct relay_session *session =
824 caa_container_of(head, struct relay_session, rcu_node);
825 free(session);
826 }
827
828 /*
829 * Close a given stream. The stream is freed using a call RCU.
830 *
831 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
832 * BEFORE the stream lock MUST be acquired.
833 */
834 static void destroy_stream(struct relay_stream *stream)
835 {
836 int delret;
837 struct relay_viewer_stream *vstream;
838 struct lttng_ht_iter iter;
839
840 assert(stream);
841
842 delret = close(stream->fd);
843 if (delret < 0) {
844 PERROR("close stream");
845 }
846
847 if (stream->index_fd >= 0) {
848 delret = close(stream->index_fd);
849 if (delret < 0) {
850 PERROR("close stream index_fd");
851 }
852 }
853
854 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
855 if (vstream) {
856 /*
857 * Set the last good value into the viewer stream. This is done
858 * right before the stream gets deleted from the hash table. The
859 * lookup failure on the live thread side of a stream indicates
860 * that the viewer stream index received value should be used.
861 */
862 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
863 vstream->total_index_received = stream->total_index_received;
864 vstream->tracefile_count_last = stream->tracefile_count_current;
865 vstream->close_write_flag = 1;
866 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
867 }
868
869 /* Cleanup index of that stream. */
870 relay_index_destroy_by_stream_id(stream->stream_handle);
871
872 iter.iter.node = &stream->stream_n.node;
873 delret = lttng_ht_del(relay_streams_ht, &iter);
874 assert(!delret);
875 iter.iter.node = &stream->ctf_trace_node.node;
876 delret = lttng_ht_del(stream->ctf_traces_ht, &iter);
877 assert(!delret);
878
879 if (stream->ctf_trace) {
880 ctf_trace_try_destroy(stream->ctf_trace);
881 }
882
883 call_rcu(&stream->rcu_node, deferred_free_stream);
884 DBG("Closed tracefile %d from close stream", stream->fd);
885 }
886
887 /*
888 * relay_delete_session: Free all memory associated with a session and
889 * close all the FDs
890 */
891 static
892 void relay_delete_session(struct relay_command *cmd,
893 struct lttng_ht *sessions_ht)
894 {
895 struct lttng_ht_iter iter;
896 struct lttng_ht_node_ulong *node;
897 struct relay_stream *stream;
898 int ret;
899
900 if (!cmd->session) {
901 return;
902 }
903
904 DBG("Relay deleting session %" PRIu64, cmd->session->id);
905
906 rcu_read_lock();
907 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
908 node = lttng_ht_iter_get_node_ulong(&iter);
909 if (!node) {
910 continue;
911 }
912 stream = caa_container_of(node, struct relay_stream, stream_n);
913 if (stream->session == cmd->session) {
914 destroy_stream(stream);
915 cmd->session->stream_count--;
916 assert(cmd->session->stream_count >= 0);
917 }
918 }
919
920 /* Make this session not visible anymore. */
921 iter.iter.node = &cmd->session->session_n.node;
922 ret = lttng_ht_del(sessions_ht, &iter);
923 assert(!ret);
924 call_rcu(&cmd->session->rcu_node, deferred_free_session);
925 rcu_read_unlock();
926 }
927
928 /*
929 * Copy index data from the control port to a given index object.
930 */
931 static void copy_index_control_data(struct relay_index *index,
932 struct lttcomm_relayd_index *data)
933 {
934 assert(index);
935 assert(data);
936
937 /*
938 * The index on disk is encoded in big endian, so we don't need to convert
939 * the data received on the network. The data_offset value is NEVER
940 * modified here and is updated by the data thread.
941 */
942 index->index_data.packet_size = data->packet_size;
943 index->index_data.content_size = data->content_size;
944 index->index_data.timestamp_begin = data->timestamp_begin;
945 index->index_data.timestamp_end = data->timestamp_end;
946 index->index_data.events_discarded = data->events_discarded;
947 index->index_data.stream_id = data->stream_id;
948 }
949
950 /*
951 * Handle the RELAYD_CREATE_SESSION command.
952 *
953 * On success, send back the session id or else return a negative value.
954 */
955 static
956 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
957 struct relay_command *cmd,
958 struct lttng_ht *sessions_ht)
959 {
960 int ret = 0, send_ret;
961 struct relay_session *session;
962 struct lttcomm_relayd_status_session reply;
963
964 assert(recv_hdr);
965 assert(cmd);
966
967 memset(&reply, 0, sizeof(reply));
968
969 session = zmalloc(sizeof(struct relay_session));
970 if (session == NULL) {
971 PERROR("relay session zmalloc");
972 ret = -1;
973 goto error;
974 }
975
976 session->id = ++last_relay_session_id;
977 session->sock = cmd->sock;
978 session->minor = cmd->minor;
979 session->major = cmd->major;
980 cmd->session = session;
981
982 reply.session_id = htobe64(session->id);
983
984 switch (cmd->minor) {
985 case 4: /* LTTng sessiond 2.4 */
986 default:
987 ret = cmd_create_session_2_4(cmd, session);
988 break;
989 }
990
991 lttng_ht_node_init_ulong(&session->session_n,
992 (unsigned long) session->id);
993 lttng_ht_add_unique_ulong(sessions_ht,
994 &session->session_n);
995
996 DBG("Created session %" PRIu64, session->id);
997
998 error:
999 if (ret < 0) {
1000 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1001 } else {
1002 reply.ret_code = htobe32(LTTNG_OK);
1003 }
1004
1005 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1006 if (send_ret < 0) {
1007 ERR("Relayd sending session id");
1008 ret = send_ret;
1009 }
1010
1011 return ret;
1012 }
1013
1014 /*
1015 * When we have received all the streams and the metadata for a channel,
1016 * we make them visible to the viewer threads.
1017 */
1018 static
1019 void set_viewer_ready_flag(struct relay_command *cmd)
1020 {
1021 struct relay_stream_recv_handle *node, *tmp_node;
1022
1023 cds_list_for_each_entry_safe(node, tmp_node, &cmd->recv_head, node) {
1024 struct relay_stream *stream;
1025
1026 rcu_read_lock();
1027 stream = relay_stream_find_by_id(node->id);
1028 if (!stream) {
1029 /*
1030 * Stream is most probably being cleaned up by the data thread thus
1031 * simply continue to the next one.
1032 */
1033 rcu_read_unlock();
1034 continue;
1035 }
1036
1037 stream->viewer_ready = 1;
1038 rcu_read_unlock();
1039
1040 /* Clean stream handle node. */
1041 cds_list_del(&node->node);
1042 free(node);
1043 }
1044
1045 return;
1046 }
1047
1048 /*
1049 * Add a recv handle node to the connection recv list with the given stream
1050 * handle. A new node is allocated thus must be freed when the node is deleted
1051 * from the list.
1052 */
1053 static void queue_stream_handle(uint64_t handle, struct relay_command *cmd)
1054 {
1055 struct relay_stream_recv_handle *node;
1056
1057 assert(cmd);
1058
1059 node = zmalloc(sizeof(*node));
1060 if (!node) {
1061 PERROR("zmalloc queue stream handle");
1062 return;
1063 }
1064
1065 node->id = handle;
1066 cds_list_add(&node->node, &cmd->recv_head);
1067 }
1068
1069 /*
1070 * relay_add_stream: allocate a new stream for a session
1071 */
1072 static
1073 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1074 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1075 {
1076 struct relay_session *session = cmd->session;
1077 struct relay_stream *stream = NULL;
1078 struct lttcomm_relayd_status_stream reply;
1079 int ret, send_ret;
1080
1081 if (!session || cmd->version_check_done == 0) {
1082 ERR("Trying to add a stream before version check");
1083 ret = -1;
1084 goto end_no_session;
1085 }
1086
1087 stream = zmalloc(sizeof(struct relay_stream));
1088 if (stream == NULL) {
1089 PERROR("relay stream zmalloc");
1090 ret = -1;
1091 goto end_no_session;
1092 }
1093
1094 switch (cmd->minor) {
1095 case 1: /* LTTng sessiond 2.1 */
1096 ret = cmd_recv_stream_2_1(cmd, stream);
1097 break;
1098 case 2: /* LTTng sessiond 2.2 */
1099 default:
1100 ret = cmd_recv_stream_2_2(cmd, stream);
1101 break;
1102 }
1103 if (ret < 0) {
1104 goto err_free_stream;
1105 }
1106
1107 rcu_read_lock();
1108 stream->stream_handle = ++last_relay_stream_id;
1109 stream->prev_seq = -1ULL;
1110 stream->session = session;
1111 stream->index_fd = -1;
1112 stream->read_index_fd = -1;
1113 stream->ctf_trace = NULL;
1114 pthread_mutex_init(&stream->lock, NULL);
1115
1116 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
1117 if (ret < 0) {
1118 ERR("relay creating output directory");
1119 goto end;
1120 }
1121
1122 /*
1123 * No need to use run_as API here because whatever we receives, the relayd
1124 * uses its own credentials for the stream files.
1125 */
1126 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1127 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1128 if (ret < 0) {
1129 ERR("Create output file");
1130 goto end;
1131 }
1132 stream->fd = ret;
1133 if (stream->tracefile_size) {
1134 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1135 } else {
1136 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1137 }
1138
1139 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1140 stream->metadata_flag = 1;
1141 /*
1142 * When we receive a new metadata stream, we create a new
1143 * ctf_trace and we assign this ctf_trace to all streams with
1144 * the same path.
1145 *
1146 * If later on we receive a new stream for the same ctf_trace,
1147 * we copy the information from the first hit in the HT to the
1148 * new stream.
1149 */
1150 stream->ctf_trace = ctf_trace_create();
1151 if (!stream->ctf_trace) {
1152 ret = -1;
1153 goto end;
1154 }
1155 stream->ctf_trace->refcount++;
1156 stream->ctf_trace->metadata_stream = stream;
1157 }
1158 ctf_trace_assign(cmd->ctf_traces_ht, stream);
1159 stream->ctf_traces_ht = cmd->ctf_traces_ht;
1160
1161 /*
1162 * Add the stream handle in the recv list of the connection. Once the end
1163 * stream message is received, this list is emptied and streams are set
1164 * with the viewer ready flag.
1165 */
1166 if (stream->metadata_flag) {
1167 stream->viewer_ready = 1;
1168 } else {
1169 queue_stream_handle(stream->stream_handle, cmd);
1170 }
1171
1172 lttng_ht_node_init_ulong(&stream->stream_n,
1173 (unsigned long) stream->stream_handle);
1174 lttng_ht_add_unique_ulong(relay_streams_ht,
1175 &stream->stream_n);
1176
1177 lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name);
1178 lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node);
1179 session->stream_count++;
1180
1181 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1182 stream->stream_handle);
1183
1184 end:
1185 reply.handle = htobe64(stream->stream_handle);
1186 /* send the session id to the client or a negative return code on error */
1187 if (ret < 0) {
1188 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1189 /* stream was not properly added to the ht, so free it */
1190 free(stream);
1191 } else {
1192 reply.ret_code = htobe32(LTTNG_OK);
1193 }
1194
1195 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1196 sizeof(struct lttcomm_relayd_status_stream), 0);
1197 if (send_ret < 0) {
1198 ERR("Relay sending stream id");
1199 ret = send_ret;
1200 }
1201 rcu_read_unlock();
1202
1203 end_no_session:
1204 return ret;
1205
1206 err_free_stream:
1207 free(stream->path_name);
1208 free(stream->channel_name);
1209 free(stream);
1210 return ret;
1211 }
1212
1213 /*
1214 * relay_close_stream: close a specific stream
1215 */
1216 static
1217 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1218 struct relay_command *cmd)
1219 {
1220 int ret, send_ret;
1221 struct relay_session *session = cmd->session;
1222 struct lttcomm_relayd_close_stream stream_info;
1223 struct lttcomm_relayd_generic_reply reply;
1224 struct relay_stream *stream;
1225
1226 DBG("Close stream received");
1227
1228 if (!session || cmd->version_check_done == 0) {
1229 ERR("Trying to close a stream before version check");
1230 ret = -1;
1231 goto end_no_session;
1232 }
1233
1234 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1235 sizeof(struct lttcomm_relayd_close_stream), 0);
1236 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1237 if (ret == 0) {
1238 /* Orderly shutdown. Not necessary to print an error. */
1239 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1240 } else {
1241 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1242 }
1243 ret = -1;
1244 goto end_no_session;
1245 }
1246
1247 rcu_read_lock();
1248 stream = relay_stream_find_by_id(be64toh(stream_info.stream_id));
1249 if (!stream) {
1250 ret = -1;
1251 goto end_unlock;
1252 }
1253
1254 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1255 stream->close_flag = 1;
1256 session->stream_count--;
1257 assert(session->stream_count >= 0);
1258
1259 if (close_stream_check(stream)) {
1260 destroy_stream(stream);
1261 }
1262
1263 end_unlock:
1264 rcu_read_unlock();
1265
1266 if (ret < 0) {
1267 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1268 } else {
1269 reply.ret_code = htobe32(LTTNG_OK);
1270 }
1271 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1272 sizeof(struct lttcomm_relayd_generic_reply), 0);
1273 if (send_ret < 0) {
1274 ERR("Relay sending stream id");
1275 ret = send_ret;
1276 }
1277
1278 end_no_session:
1279 return ret;
1280 }
1281
1282 /*
1283 * relay_unknown_command: send -1 if received unknown command
1284 */
1285 static
1286 void relay_unknown_command(struct relay_command *cmd)
1287 {
1288 struct lttcomm_relayd_generic_reply reply;
1289 int ret;
1290
1291 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1292 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1293 sizeof(struct lttcomm_relayd_generic_reply), 0);
1294 if (ret < 0) {
1295 ERR("Relay sending unknown command");
1296 }
1297 }
1298
1299 /*
1300 * relay_start: send an acknowledgment to the client to tell if we are
1301 * ready to receive data. We are ready if a session is established.
1302 */
1303 static
1304 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1305 struct relay_command *cmd)
1306 {
1307 int ret = htobe32(LTTNG_OK);
1308 struct lttcomm_relayd_generic_reply reply;
1309 struct relay_session *session = cmd->session;
1310
1311 if (!session) {
1312 DBG("Trying to start the streaming without a session established");
1313 ret = htobe32(LTTNG_ERR_UNK);
1314 }
1315
1316 reply.ret_code = ret;
1317 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1318 sizeof(struct lttcomm_relayd_generic_reply), 0);
1319 if (ret < 0) {
1320 ERR("Relay sending start ack");
1321 }
1322
1323 return ret;
1324 }
1325
1326 /*
1327 * Append padding to the file pointed by the file descriptor fd.
1328 */
1329 static int write_padding_to_file(int fd, uint32_t size)
1330 {
1331 ssize_t ret = 0;
1332 char *zeros;
1333
1334 if (size == 0) {
1335 goto end;
1336 }
1337
1338 zeros = zmalloc(size);
1339 if (zeros == NULL) {
1340 PERROR("zmalloc zeros for padding");
1341 ret = -1;
1342 goto end;
1343 }
1344
1345 ret = lttng_write(fd, zeros, size);
1346 if (ret < size) {
1347 PERROR("write padding to file");
1348 }
1349
1350 free(zeros);
1351
1352 end:
1353 return ret;
1354 }
1355
1356 /*
1357 * relay_recv_metadata: receive the metada for the session.
1358 */
1359 static
1360 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1361 struct relay_command *cmd)
1362 {
1363 int ret = htobe32(LTTNG_OK);
1364 ssize_t size_ret;
1365 struct relay_session *session = cmd->session;
1366 struct lttcomm_relayd_metadata_payload *metadata_struct;
1367 struct relay_stream *metadata_stream;
1368 uint64_t data_size, payload_size;
1369
1370 if (!session) {
1371 ERR("Metadata sent before version check");
1372 ret = -1;
1373 goto end;
1374 }
1375
1376 data_size = payload_size = be64toh(recv_hdr->data_size);
1377 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1378 ERR("Incorrect data size");
1379 ret = -1;
1380 goto end;
1381 }
1382 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1383
1384 if (data_buffer_size < data_size) {
1385 /* In case the realloc fails, we can free the memory */
1386 char *tmp_data_ptr;
1387
1388 tmp_data_ptr = realloc(data_buffer, data_size);
1389 if (!tmp_data_ptr) {
1390 ERR("Allocating data buffer");
1391 free(data_buffer);
1392 ret = -1;
1393 goto end;
1394 }
1395 data_buffer = tmp_data_ptr;
1396 data_buffer_size = data_size;
1397 }
1398 memset(data_buffer, 0, data_size);
1399 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1400 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1401 if (ret < 0 || ret != data_size) {
1402 if (ret == 0) {
1403 /* Orderly shutdown. Not necessary to print an error. */
1404 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1405 } else {
1406 ERR("Relay didn't receive the whole metadata");
1407 }
1408 ret = -1;
1409 goto end;
1410 }
1411 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1412
1413 rcu_read_lock();
1414 metadata_stream = relay_stream_find_by_id(
1415 be64toh(metadata_struct->stream_id));
1416 if (!metadata_stream) {
1417 ret = -1;
1418 goto end_unlock;
1419 }
1420
1421 size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
1422 payload_size);
1423 if (size_ret < payload_size) {
1424 ERR("Relay error writing metadata on file");
1425 ret = -1;
1426 goto end_unlock;
1427 }
1428
1429 ret = write_padding_to_file(metadata_stream->fd,
1430 be32toh(metadata_struct->padding_size));
1431 if (ret < 0) {
1432 goto end_unlock;
1433 }
1434 metadata_stream->ctf_trace->metadata_received +=
1435 payload_size + be32toh(metadata_struct->padding_size);
1436
1437 DBG2("Relay metadata written");
1438
1439 end_unlock:
1440 rcu_read_unlock();
1441 end:
1442 return ret;
1443 }
1444
1445 /*
1446 * relay_send_version: send relayd version number
1447 */
1448 static
1449 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1450 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1451 {
1452 int ret;
1453 struct lttcomm_relayd_version reply, msg;
1454
1455 assert(cmd);
1456
1457 cmd->version_check_done = 1;
1458
1459 /* Get version from the other side. */
1460 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1461 if (ret < 0 || ret != sizeof(msg)) {
1462 if (ret == 0) {
1463 /* Orderly shutdown. Not necessary to print an error. */
1464 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1465 } else {
1466 ERR("Relay failed to receive the version values.");
1467 }
1468 ret = -1;
1469 goto end;
1470 }
1471
1472 reply.major = RELAYD_VERSION_COMM_MAJOR;
1473 reply.minor = RELAYD_VERSION_COMM_MINOR;
1474
1475 /* Major versions must be the same */
1476 if (reply.major != be32toh(msg.major)) {
1477 DBG("Incompatible major versions (%u vs %u), deleting session",
1478 reply.major, be32toh(msg.major));
1479 relay_delete_session(cmd, sessions_ht);
1480 ret = 0;
1481 goto end;
1482 }
1483
1484 cmd->major = reply.major;
1485 /* We adapt to the lowest compatible version */
1486 if (reply.minor <= be32toh(msg.minor)) {
1487 cmd->minor = reply.minor;
1488 } else {
1489 cmd->minor = be32toh(msg.minor);
1490 }
1491
1492 reply.major = htobe32(reply.major);
1493 reply.minor = htobe32(reply.minor);
1494 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1495 sizeof(struct lttcomm_relayd_version), 0);
1496 if (ret < 0) {
1497 ERR("Relay sending version");
1498 }
1499
1500 DBG("Version check done using protocol %u.%u", cmd->major,
1501 cmd->minor);
1502
1503 end:
1504 return ret;
1505 }
1506
1507 /*
1508 * Check for data pending for a given stream id from the session daemon.
1509 */
1510 static
1511 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1512 struct relay_command *cmd)
1513 {
1514 struct relay_session *session = cmd->session;
1515 struct lttcomm_relayd_data_pending msg;
1516 struct lttcomm_relayd_generic_reply reply;
1517 struct relay_stream *stream;
1518 int ret;
1519 uint64_t last_net_seq_num, stream_id;
1520
1521 DBG("Data pending command received");
1522
1523 if (!session || cmd->version_check_done == 0) {
1524 ERR("Trying to check for data before version check");
1525 ret = -1;
1526 goto end_no_session;
1527 }
1528
1529 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1530 if (ret < sizeof(msg)) {
1531 if (ret == 0) {
1532 /* Orderly shutdown. Not necessary to print an error. */
1533 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1534 } else {
1535 ERR("Relay didn't receive valid data_pending struct size : %d",
1536 ret);
1537 }
1538 ret = -1;
1539 goto end_no_session;
1540 }
1541
1542 stream_id = be64toh(msg.stream_id);
1543 last_net_seq_num = be64toh(msg.last_net_seq_num);
1544
1545 rcu_read_lock();
1546 stream = relay_stream_find_by_id(stream_id);
1547 if (stream == NULL) {
1548 ret = -1;
1549 goto end_unlock;
1550 }
1551
1552 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1553 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1554 last_net_seq_num);
1555
1556 /* Avoid wrapping issue */
1557 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1558 /* Data has in fact been written and is NOT pending */
1559 ret = 0;
1560 } else {
1561 /* Data still being streamed thus pending */
1562 ret = 1;
1563 }
1564
1565 /* Pending check is now done. */
1566 stream->data_pending_check_done = 1;
1567
1568 end_unlock:
1569 rcu_read_unlock();
1570
1571 reply.ret_code = htobe32(ret);
1572 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1573 if (ret < 0) {
1574 ERR("Relay data pending ret code failed");
1575 }
1576
1577 end_no_session:
1578 return ret;
1579 }
1580
1581 /*
1582 * Wait for the control socket to reach a quiescent state.
1583 *
1584 * Note that for now, when receiving this command from the session daemon, this
1585 * means that every subsequent commands or data received on the control socket
1586 * has been handled. So, this is why we simply return OK here.
1587 */
1588 static
1589 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1590 struct relay_command *cmd)
1591 {
1592 int ret;
1593 uint64_t stream_id;
1594 struct relay_stream *stream;
1595 struct lttng_ht_iter iter;
1596 struct lttcomm_relayd_quiescent_control msg;
1597 struct lttcomm_relayd_generic_reply reply;
1598
1599 DBG("Checking quiescent state on control socket");
1600
1601 if (!cmd->session || cmd->version_check_done == 0) {
1602 ERR("Trying to check for data before version check");
1603 ret = -1;
1604 goto end_no_session;
1605 }
1606
1607 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1608 if (ret < sizeof(msg)) {
1609 if (ret == 0) {
1610 /* Orderly shutdown. Not necessary to print an error. */
1611 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1612 } else {
1613 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1614 ret);
1615 }
1616 ret = -1;
1617 goto end_no_session;
1618 }
1619
1620 stream_id = be64toh(msg.stream_id);
1621
1622 rcu_read_lock();
1623 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1624 stream_n.node) {
1625 if (stream->stream_handle == stream_id) {
1626 stream->data_pending_check_done = 1;
1627 DBG("Relay quiescent control pending flag set to %" PRIu64,
1628 stream_id);
1629 break;
1630 }
1631 }
1632 rcu_read_unlock();
1633
1634 reply.ret_code = htobe32(LTTNG_OK);
1635 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1636 if (ret < 0) {
1637 ERR("Relay data quiescent control ret code failed");
1638 }
1639
1640 end_no_session:
1641 return ret;
1642 }
1643
1644 /*
1645 * Initialize a data pending command. This means that a client is about to ask
1646 * for data pending for each stream he/she holds. Simply iterate over all
1647 * streams of a session and set the data_pending_check_done flag.
1648 *
1649 * This command returns to the client a LTTNG_OK code.
1650 */
1651 static
1652 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1653 struct relay_command *cmd)
1654 {
1655 int ret;
1656 struct lttng_ht_iter iter;
1657 struct lttcomm_relayd_begin_data_pending msg;
1658 struct lttcomm_relayd_generic_reply reply;
1659 struct relay_stream *stream;
1660 uint64_t session_id;
1661
1662 assert(recv_hdr);
1663 assert(cmd);
1664
1665 DBG("Init streams for data pending");
1666
1667 if (!cmd->session || cmd->version_check_done == 0) {
1668 ERR("Trying to check for data before version check");
1669 ret = -1;
1670 goto end_no_session;
1671 }
1672
1673 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1674 if (ret < sizeof(msg)) {
1675 if (ret == 0) {
1676 /* Orderly shutdown. Not necessary to print an error. */
1677 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1678 } else {
1679 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1680 ret);
1681 }
1682 ret = -1;
1683 goto end_no_session;
1684 }
1685
1686 session_id = be64toh(msg.session_id);
1687
1688 /*
1689 * Iterate over all streams to set the begin data pending flag. For now, the
1690 * streams are indexed by stream handle so we have to iterate over all
1691 * streams to find the one associated with the right session_id.
1692 */
1693 rcu_read_lock();
1694 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1695 stream_n.node) {
1696 if (stream->session->id == session_id) {
1697 stream->data_pending_check_done = 0;
1698 DBG("Set begin data pending flag to stream %" PRIu64,
1699 stream->stream_handle);
1700 }
1701 }
1702 rcu_read_unlock();
1703
1704 /* All good, send back reply. */
1705 reply.ret_code = htobe32(LTTNG_OK);
1706
1707 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1708 if (ret < 0) {
1709 ERR("Relay begin data pending send reply failed");
1710 }
1711
1712 end_no_session:
1713 return ret;
1714 }
1715
1716 /*
1717 * End data pending command. This will check, for a given session id, if each
1718 * stream associated with it has its data_pending_check_done flag set. If not,
1719 * this means that the client lost track of the stream but the data is still
1720 * being streamed on our side. In this case, we inform the client that data is
1721 * inflight.
1722 *
1723 * Return to the client if there is data in flight or not with a ret_code.
1724 */
1725 static
1726 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1727 struct relay_command *cmd)
1728 {
1729 int ret;
1730 struct lttng_ht_iter iter;
1731 struct lttcomm_relayd_end_data_pending msg;
1732 struct lttcomm_relayd_generic_reply reply;
1733 struct relay_stream *stream;
1734 uint64_t session_id;
1735 uint32_t is_data_inflight = 0;
1736
1737 assert(recv_hdr);
1738 assert(cmd);
1739
1740 DBG("End data pending command");
1741
1742 if (!cmd->session || cmd->version_check_done == 0) {
1743 ERR("Trying to check for data before version check");
1744 ret = -1;
1745 goto end_no_session;
1746 }
1747
1748 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1749 if (ret < sizeof(msg)) {
1750 if (ret == 0) {
1751 /* Orderly shutdown. Not necessary to print an error. */
1752 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1753 } else {
1754 ERR("Relay didn't receive valid end data_pending struct size: %d",
1755 ret);
1756 }
1757 ret = -1;
1758 goto end_no_session;
1759 }
1760
1761 session_id = be64toh(msg.session_id);
1762
1763 /* Iterate over all streams to see if the begin data pending flag is set. */
1764 rcu_read_lock();
1765 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1766 stream_n.node) {
1767 if (stream->session->id == session_id &&
1768 !stream->data_pending_check_done) {
1769 is_data_inflight = 1;
1770 DBG("Data is still in flight for stream %" PRIu64,
1771 stream->stream_handle);
1772 break;
1773 }
1774 }
1775 rcu_read_unlock();
1776
1777 /* All good, send back reply. */
1778 reply.ret_code = htobe32(is_data_inflight);
1779
1780 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1781 if (ret < 0) {
1782 ERR("Relay end data pending send reply failed");
1783 }
1784
1785 end_no_session:
1786 return ret;
1787 }
1788
1789 /*
1790 * Receive an index for a specific stream.
1791 *
1792 * Return 0 on success else a negative value.
1793 */
1794 static
1795 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1796 struct relay_command *cmd)
1797 {
1798 int ret, send_ret, index_created = 0;
1799 struct relay_session *session = cmd->session;
1800 struct lttcomm_relayd_index index_info;
1801 struct relay_index *index, *wr_index = NULL;
1802 struct lttcomm_relayd_generic_reply reply;
1803 struct relay_stream *stream;
1804 uint64_t net_seq_num;
1805
1806 assert(cmd);
1807
1808 DBG("Relay receiving index");
1809
1810 if (!session || cmd->version_check_done == 0) {
1811 ERR("Trying to close a stream before version check");
1812 ret = -1;
1813 goto end_no_session;
1814 }
1815
1816 ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info,
1817 sizeof(index_info), 0);
1818 if (ret < sizeof(index_info)) {
1819 if (ret == 0) {
1820 /* Orderly shutdown. Not necessary to print an error. */
1821 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1822 } else {
1823 ERR("Relay didn't receive valid index struct size : %d", ret);
1824 }
1825 ret = -1;
1826 goto end_no_session;
1827 }
1828
1829 net_seq_num = be64toh(index_info.net_seq_num);
1830
1831 rcu_read_lock();
1832 stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id));
1833 if (!stream) {
1834 ret = -1;
1835 goto end_rcu_unlock;
1836 }
1837
1838 /* Live beacon handling */
1839 if (index_info.packet_size == 0) {
1840 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1841
1842 /*
1843 * Only flag a stream inactive when it has already received data.
1844 */
1845 if (stream->total_index_received > 0) {
1846 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1847 }
1848 ret = 0;
1849 goto end_rcu_unlock;
1850 } else {
1851 stream->beacon_ts_end = -1ULL;
1852 }
1853
1854 index = relay_index_find(stream->stream_handle, net_seq_num);
1855 if (!index) {
1856 /* A successful creation will add the object to the HT. */
1857 index = relay_index_create(stream->stream_handle, net_seq_num);
1858 if (!index) {
1859 goto end_rcu_unlock;
1860 }
1861 index_created = 1;
1862 }
1863
1864 copy_index_control_data(index, &index_info);
1865
1866 if (index_created) {
1867 /*
1868 * Try to add the relay index object to the hash table. If an object
1869 * already exist, destroy back the index created, set the data in this
1870 * object and write it on disk.
1871 */
1872 relay_index_add(index, &wr_index);
1873 if (wr_index) {
1874 copy_index_control_data(wr_index, &index_info);
1875 free(index);
1876 }
1877 } else {
1878 /* The index already exists so write it on disk. */
1879 wr_index = index;
1880 }
1881
1882 /* Do we have a writable ready index to write on disk. */
1883 if (wr_index) {
1884 /* Starting at 2.4, create the index file if none available. */
1885 if (cmd->minor >= 4 && stream->index_fd < 0) {
1886 ret = index_create_file(stream->path_name, stream->channel_name,
1887 relayd_uid, relayd_gid, stream->tracefile_size,
1888 stream->tracefile_count_current);
1889 if (ret < 0) {
1890 goto end_rcu_unlock;
1891 }
1892 stream->index_fd = ret;
1893 }
1894
1895 ret = relay_index_write(wr_index->fd, wr_index);
1896 if (ret < 0) {
1897 goto end_rcu_unlock;
1898 }
1899 stream->total_index_received++;
1900 }
1901
1902 end_rcu_unlock:
1903 rcu_read_unlock();
1904
1905 if (ret < 0) {
1906 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1907 } else {
1908 reply.ret_code = htobe32(LTTNG_OK);
1909 }
1910 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1911 if (send_ret < 0) {
1912 ERR("Relay sending close index id reply");
1913 ret = send_ret;
1914 }
1915
1916 end_no_session:
1917 return ret;
1918 }
1919
1920 /*
1921 * Receive the streams_sent message.
1922 *
1923 * Return 0 on success else a negative value.
1924 */
1925 static
1926 int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
1927 struct relay_command *cmd)
1928 {
1929 int ret, send_ret;
1930 struct lttcomm_relayd_generic_reply reply;
1931
1932 assert(cmd);
1933
1934 DBG("Relay receiving streams_sent");
1935
1936 if (!cmd->session || cmd->version_check_done == 0) {
1937 ERR("Trying to close a stream before version check");
1938 ret = -1;
1939 goto end_no_session;
1940 }
1941
1942 /*
1943 * Flag every pending stream in the connection recv list that they are
1944 * ready to be used by the viewer.
1945 */
1946 set_viewer_ready_flag(cmd);
1947
1948 reply.ret_code = htobe32(LTTNG_OK);
1949 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1950 if (send_ret < 0) {
1951 ERR("Relay sending sent_stream reply");
1952 ret = send_ret;
1953 } else {
1954 /* Success. */
1955 ret = 0;
1956 }
1957
1958 end_no_session:
1959 return ret;
1960 }
1961
1962 /*
1963 * Process the commands received on the control socket
1964 */
1965 static
1966 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1967 struct relay_command *cmd, struct relay_local_data *ctx)
1968 {
1969 int ret = 0;
1970
1971 switch (be32toh(recv_hdr->cmd)) {
1972 case RELAYD_CREATE_SESSION:
1973 ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht);
1974 break;
1975 case RELAYD_ADD_STREAM:
1976 ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht);
1977 break;
1978 case RELAYD_START_DATA:
1979 ret = relay_start(recv_hdr, cmd);
1980 break;
1981 case RELAYD_SEND_METADATA:
1982 ret = relay_recv_metadata(recv_hdr, cmd);
1983 break;
1984 case RELAYD_VERSION:
1985 ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht);
1986 break;
1987 case RELAYD_CLOSE_STREAM:
1988 ret = relay_close_stream(recv_hdr, cmd);
1989 break;
1990 case RELAYD_DATA_PENDING:
1991 ret = relay_data_pending(recv_hdr, cmd);
1992 break;
1993 case RELAYD_QUIESCENT_CONTROL:
1994 ret = relay_quiescent_control(recv_hdr, cmd);
1995 break;
1996 case RELAYD_BEGIN_DATA_PENDING:
1997 ret = relay_begin_data_pending(recv_hdr, cmd);
1998 break;
1999 case RELAYD_END_DATA_PENDING:
2000 ret = relay_end_data_pending(recv_hdr, cmd);
2001 break;
2002 case RELAYD_SEND_INDEX:
2003 ret = relay_recv_index(recv_hdr, cmd);
2004 break;
2005 case RELAYD_STREAMS_SENT:
2006 ret = relay_streams_sent(recv_hdr, cmd);
2007 break;
2008 case RELAYD_UPDATE_SYNC_INFO:
2009 default:
2010 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2011 relay_unknown_command(cmd);
2012 ret = -1;
2013 goto end;
2014 }
2015
2016 end:
2017 return ret;
2018 }
2019
2020 /*
2021 * Handle index for a data stream.
2022 *
2023 * RCU read side lock MUST be acquired.
2024 *
2025 * Return 0 on success else a negative value.
2026 */
2027 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2028 int rotate_index)
2029 {
2030 int ret = 0, index_created = 0;
2031 uint64_t stream_id, data_offset;
2032 struct relay_index *index, *wr_index = NULL;
2033
2034 assert(stream);
2035
2036 stream_id = stream->stream_handle;
2037 /* Get data offset because we are about to update the index. */
2038 data_offset = htobe64(stream->tracefile_size_current);
2039
2040 /*
2041 * Lookup for an existing index for that stream id/sequence number. If on
2042 * exists, the control thread already received the data for it thus we need
2043 * to write it on disk.
2044 */
2045 index = relay_index_find(stream_id, net_seq_num);
2046 if (!index) {
2047 /* A successful creation will add the object to the HT. */
2048 index = relay_index_create(stream_id, net_seq_num);
2049 if (!index) {
2050 ret = -1;
2051 goto error;
2052 }
2053 index_created = 1;
2054 }
2055
2056 if (rotate_index || stream->index_fd < 0) {
2057 index->to_close_fd = stream->index_fd;
2058 ret = index_create_file(stream->path_name, stream->channel_name,
2059 relayd_uid, relayd_gid, stream->tracefile_size,
2060 stream->tracefile_count_current);
2061 if (ret < 0) {
2062 /* This will close the stream's index fd if one. */
2063 relay_index_free_safe(index);
2064 goto error;
2065 }
2066 stream->index_fd = ret;
2067 }
2068 index->fd = stream->index_fd;
2069 index->index_data.offset = data_offset;
2070
2071 if (index_created) {
2072 /*
2073 * Try to add the relay index object to the hash table. If an object
2074 * already exist, destroy back the index created and set the data.
2075 */
2076 relay_index_add(index, &wr_index);
2077 if (wr_index) {
2078 /* Copy back data from the created index. */
2079 wr_index->fd = index->fd;
2080 wr_index->to_close_fd = index->to_close_fd;
2081 wr_index->index_data.offset = data_offset;
2082 free(index);
2083 }
2084 } else {
2085 /* The index already exists so write it on disk. */
2086 wr_index = index;
2087 }
2088
2089 /* Do we have a writable ready index to write on disk. */
2090 if (wr_index) {
2091 ret = relay_index_write(wr_index->fd, wr_index);
2092 if (ret < 0) {
2093 goto error;
2094 }
2095 stream->total_index_received++;
2096 }
2097
2098 error:
2099 return ret;
2100 }
2101
2102 /*
2103 * relay_process_data: Process the data received on the data socket
2104 */
2105 static
2106 int relay_process_data(struct relay_command *cmd)
2107 {
2108 int ret = 0, rotate_index = 0;
2109 ssize_t size_ret;
2110 struct relay_stream *stream;
2111 struct lttcomm_relayd_data_hdr data_hdr;
2112 uint64_t stream_id;
2113 uint64_t net_seq_num;
2114 uint32_t data_size;
2115
2116 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
2117 sizeof(struct lttcomm_relayd_data_hdr), 0);
2118 if (ret <= 0) {
2119 if (ret == 0) {
2120 /* Orderly shutdown. Not necessary to print an error. */
2121 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2122 } else {
2123 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
2124 }
2125 ret = -1;
2126 goto end;
2127 }
2128
2129 stream_id = be64toh(data_hdr.stream_id);
2130
2131 rcu_read_lock();
2132 stream = relay_stream_find_by_id(stream_id);
2133 if (!stream) {
2134 ret = -1;
2135 goto end_rcu_unlock;
2136 }
2137
2138 data_size = be32toh(data_hdr.data_size);
2139 if (data_buffer_size < data_size) {
2140 char *tmp_data_ptr;
2141
2142 tmp_data_ptr = realloc(data_buffer, data_size);
2143 if (!tmp_data_ptr) {
2144 ERR("Allocating data buffer");
2145 free(data_buffer);
2146 ret = -1;
2147 goto end_rcu_unlock;
2148 }
2149 data_buffer = tmp_data_ptr;
2150 data_buffer_size = data_size;
2151 }
2152 memset(data_buffer, 0, data_size);
2153
2154 net_seq_num = be64toh(data_hdr.net_seq_num);
2155
2156 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2157 data_size, stream_id, net_seq_num);
2158 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
2159 if (ret <= 0) {
2160 if (ret == 0) {
2161 /* Orderly shutdown. Not necessary to print an error. */
2162 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2163 }
2164 ret = -1;
2165 goto end_rcu_unlock;
2166 }
2167
2168 /* Check if a rotation is needed. */
2169 if (stream->tracefile_size > 0 &&
2170 (stream->tracefile_size_current + data_size) >
2171 stream->tracefile_size) {
2172 struct relay_viewer_stream *vstream;
2173 uint64_t new_id;
2174
2175 new_id = (stream->tracefile_count_current + 1) %
2176 stream->tracefile_count;
2177 /*
2178 * When we wrap-around back to 0, we start overwriting old
2179 * trace data.
2180 */
2181 if (!stream->tracefile_overwrite && new_id == 0) {
2182 stream->tracefile_overwrite = 1;
2183 }
2184 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
2185 if (stream->tracefile_overwrite) {
2186 stream->oldest_tracefile_id =
2187 (stream->oldest_tracefile_id + 1) %
2188 stream->tracefile_count;
2189 }
2190 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
2191 if (vstream) {
2192 /*
2193 * The viewer is reading a file about to be
2194 * overwritten. Close the FDs it is
2195 * currently using and let it handle the fault.
2196 */
2197 if (vstream->tracefile_count_current == new_id) {
2198 pthread_mutex_lock(&vstream->overwrite_lock);
2199 vstream->abort_flag = 1;
2200 pthread_mutex_unlock(&vstream->overwrite_lock);
2201 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2202 stream->channel_name, new_id);
2203 } else if (vstream->tracefile_count_current ==
2204 stream->tracefile_count_current) {
2205 /*
2206 * The reader and writer were in the
2207 * same trace file, inform the viewer
2208 * that no new index will ever be added
2209 * to this file.
2210 */
2211 vstream->close_write_flag = 1;
2212 }
2213 }
2214 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2215 stream->tracefile_size, stream->tracefile_count,
2216 relayd_uid, relayd_gid, stream->fd,
2217 &(stream->tracefile_count_current), &stream->fd);
2218 stream->total_index_received = 0;
2219 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
2220 if (ret < 0) {
2221 ERR("Rotating stream output file");
2222 goto end_rcu_unlock;
2223 }
2224 /* Reset current size because we just perform a stream rotation. */
2225 stream->tracefile_size_current = 0;
2226 rotate_index = 1;
2227 }
2228
2229 /*
2230 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2231 * index are NOT supported.
2232 */
2233 if (stream->session->minor >= 4 && !stream->session->snapshot) {
2234 ret = handle_index_data(stream, net_seq_num, rotate_index);
2235 if (ret < 0) {
2236 goto end_rcu_unlock;
2237 }
2238 }
2239
2240 /* Write data to stream output fd. */
2241 size_ret = lttng_write(stream->fd, data_buffer, data_size);
2242 if (size_ret < data_size) {
2243 ERR("Relay error writing data to file");
2244 ret = -1;
2245 goto end_rcu_unlock;
2246 }
2247
2248 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2249 ret, stream->stream_handle);
2250
2251 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2252 if (ret < 0) {
2253 goto end_rcu_unlock;
2254 }
2255 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2256
2257 stream->prev_seq = net_seq_num;
2258
2259 /* Check if we need to close the FD */
2260 if (close_stream_check(stream)) {
2261 destroy_stream(stream);
2262 }
2263
2264 end_rcu_unlock:
2265 rcu_read_unlock();
2266 end:
2267 return ret;
2268 }
2269
2270 static
2271 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
2272 {
2273 int ret;
2274
2275 lttng_poll_del(events, pollfd);
2276
2277 ret = close(pollfd);
2278 if (ret < 0) {
2279 ERR("Closing pollfd %d", pollfd);
2280 }
2281 }
2282
2283 static
2284 int relay_add_connection(int fd, struct lttng_poll_event *events,
2285 struct lttng_ht *relay_connections_ht)
2286 {
2287 struct relay_command *relay_connection;
2288 ssize_t ret;
2289
2290 relay_connection = zmalloc(sizeof(struct relay_command));
2291 if (relay_connection == NULL) {
2292 PERROR("Relay command zmalloc");
2293 goto error;
2294 }
2295 ret = lttng_read(fd, relay_connection, sizeof(struct relay_command));
2296 if (ret < sizeof(struct relay_command)) {
2297 PERROR("read relay cmd pipe");
2298 goto error_read;
2299 }
2300 CDS_INIT_LIST_HEAD(&relay_connection->recv_head);
2301
2302 /*
2303 * Only used by the control side and the reference is copied inside each
2304 * stream from that connection. Thus a destroy HT must be done after every
2305 * stream has been destroyed.
2306 */
2307 if (relay_connection->type == RELAY_CONTROL) {
2308 relay_connection->ctf_traces_ht = lttng_ht_new(0,
2309 LTTNG_HT_TYPE_STRING);
2310 if (!relay_connection->ctf_traces_ht) {
2311 goto error_read;
2312 }
2313 }
2314
2315 lttng_ht_node_init_ulong(&relay_connection->sock_n,
2316 (unsigned long) relay_connection->sock->fd);
2317 rcu_read_lock();
2318 lttng_ht_add_unique_ulong(relay_connections_ht,
2319 &relay_connection->sock_n);
2320 rcu_read_unlock();
2321 return lttng_poll_add(events,
2322 relay_connection->sock->fd,
2323 LPOLLIN | LPOLLRDHUP);
2324
2325 error_read:
2326 free(relay_connection);
2327 error:
2328 return -1;
2329 }
2330
2331 static
2332 void deferred_free_connection(struct rcu_head *head)
2333 {
2334 struct relay_command *relay_connection =
2335 caa_container_of(head, struct relay_command, rcu_node);
2336
2337 lttcomm_destroy_sock(relay_connection->sock);
2338 free(relay_connection);
2339 }
2340
2341 static
2342 void relay_del_connection(struct lttng_ht *relay_connections_ht,
2343 struct lttng_ht_iter *iter, struct relay_command *relay_connection,
2344 struct lttng_ht *sessions_ht)
2345 {
2346 int ret;
2347
2348 ret = lttng_ht_del(relay_connections_ht, iter);
2349 assert(!ret);
2350
2351 if (relay_connection->type == RELAY_CONTROL) {
2352 struct relay_stream_recv_handle *node, *tmp_node;
2353
2354 relay_delete_session(relay_connection, sessions_ht);
2355 lttng_ht_destroy(relay_connection->ctf_traces_ht);
2356
2357 /* Clean up recv list. */
2358 cds_list_for_each_entry_safe(node, tmp_node,
2359 &relay_connection->recv_head, node) {
2360 cds_list_del(&node->node);
2361 free(node);
2362 }
2363 }
2364
2365 call_rcu(&relay_connection->rcu_node, deferred_free_connection);
2366 }
2367
2368 /*
2369 * This thread does the actual work
2370 */
2371 static
2372 void *relay_thread_worker(void *data)
2373 {
2374 int ret, err = -1, last_seen_data_fd = -1;
2375 uint32_t nb_fd;
2376 struct relay_command *relay_connection;
2377 struct lttng_poll_event events;
2378 struct lttng_ht *relay_connections_ht;
2379 struct lttng_ht_node_ulong *node;
2380 struct lttng_ht_iter iter;
2381 struct lttcomm_relayd_hdr recv_hdr;
2382 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2383 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2384
2385 DBG("[thread] Relay worker started");
2386
2387 rcu_register_thread();
2388
2389 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2390
2391 health_code_update();
2392
2393 /* table of connections indexed on socket */
2394 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2395 if (!relay_connections_ht) {
2396 goto relay_connections_ht_error;
2397 }
2398
2399 /* Tables of received indexes indexed by index handle and net_seq_num. */
2400 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2401 if (!indexes_ht) {
2402 goto indexes_ht_error;
2403 }
2404
2405 ret = create_thread_poll_set(&events, 2);
2406 if (ret < 0) {
2407 goto error_poll_create;
2408 }
2409
2410 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
2411 if (ret < 0) {
2412 goto error;
2413 }
2414
2415 restart:
2416 while (1) {
2417 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2418
2419 health_code_update();
2420
2421 /* Infinite blocking call, waiting for transmission */
2422 DBG3("Relayd worker thread polling...");
2423 health_poll_entry();
2424 ret = lttng_poll_wait(&events, -1);
2425 health_poll_exit();
2426 if (ret < 0) {
2427 /*
2428 * Restart interrupted system call.
2429 */
2430 if (errno == EINTR) {
2431 goto restart;
2432 }
2433 goto error;
2434 }
2435
2436 nb_fd = ret;
2437
2438 /*
2439 * Process control. The control connection is prioritised so we don't
2440 * starve it with high throughout put tracing data on the data
2441 * connection.
2442 */
2443 for (i = 0; i < nb_fd; i++) {
2444 /* Fetch once the poll data */
2445 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2446 int pollfd = LTTNG_POLL_GETFD(&events, i);
2447
2448 health_code_update();
2449
2450 /* Thread quit pipe has been closed. Killing thread. */
2451 ret = check_thread_quit_pipe(pollfd, revents);
2452 if (ret) {
2453 err = 0;
2454 goto exit;
2455 }
2456
2457 /* Inspect the relay cmd pipe for new connection */
2458 if (pollfd == relay_cmd_pipe[0]) {
2459 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2460 ERR("Relay pipe error");
2461 goto error;
2462 } else if (revents & LPOLLIN) {
2463 DBG("Relay command received");
2464 ret = relay_add_connection(relay_cmd_pipe[0],
2465 &events, relay_connections_ht);
2466 if (ret < 0) {
2467 goto error;
2468 }
2469 }
2470 } else if (revents) {
2471 rcu_read_lock();
2472 lttng_ht_lookup(relay_connections_ht,
2473 (void *)((unsigned long) pollfd),
2474 &iter);
2475 node = lttng_ht_iter_get_node_ulong(&iter);
2476 if (node == NULL) {
2477 DBG2("Relay sock %d not found", pollfd);
2478 rcu_read_unlock();
2479 goto error;
2480 }
2481 relay_connection = caa_container_of(node,
2482 struct relay_command, sock_n);
2483
2484 if (revents & (LPOLLERR)) {
2485 ERR("POLL ERROR");
2486 relay_cleanup_poll_connection(&events, pollfd);
2487 relay_del_connection(relay_connections_ht,
2488 &iter, relay_connection, sessions_ht);
2489 if (last_seen_data_fd == pollfd) {
2490 last_seen_data_fd = last_notdel_data_fd;
2491 }
2492 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2493 DBG("Socket %d hung up", pollfd);
2494 relay_cleanup_poll_connection(&events, pollfd);
2495 relay_del_connection(relay_connections_ht,
2496 &iter, relay_connection, sessions_ht);
2497 if (last_seen_data_fd == pollfd) {
2498 last_seen_data_fd = last_notdel_data_fd;
2499 }
2500 } else if (revents & LPOLLIN) {
2501 /* control socket */
2502 if (relay_connection->type == RELAY_CONTROL) {
2503 ret = relay_connection->sock->ops->recvmsg(
2504 relay_connection->sock, &recv_hdr,
2505 sizeof(struct lttcomm_relayd_hdr), 0);
2506 /* connection closed */
2507 if (ret <= 0) {
2508 relay_cleanup_poll_connection(&events, pollfd);
2509 relay_del_connection(relay_connections_ht,
2510 &iter, relay_connection, sessions_ht);
2511 DBG("Control connection closed with %d", pollfd);
2512 } else {
2513 if (relay_connection->session) {
2514 DBG2("Relay worker receiving data for session : %" PRIu64,
2515 relay_connection->session->id);
2516 }
2517 ret = relay_process_control(&recv_hdr,
2518 relay_connection, relay_ctx);
2519 if (ret < 0) {
2520 /* Clear the session on error. */
2521 relay_cleanup_poll_connection(&events, pollfd);
2522 relay_del_connection(relay_connections_ht,
2523 &iter, relay_connection, sessions_ht);
2524 DBG("Connection closed with %d", pollfd);
2525 }
2526 seen_control = 1;
2527 }
2528 } else {
2529 /*
2530 * Flag the last seen data fd not deleted. It will be
2531 * used as the last seen fd if any fd gets deleted in
2532 * this first loop.
2533 */
2534 last_notdel_data_fd = pollfd;
2535 }
2536 }
2537 rcu_read_unlock();
2538 }
2539 }
2540
2541 /*
2542 * The last loop handled a control request, go back to poll to make
2543 * sure we prioritise the control socket.
2544 */
2545 if (seen_control) {
2546 continue;
2547 }
2548
2549 if (last_seen_data_fd >= 0) {
2550 for (i = 0; i < nb_fd; i++) {
2551 int pollfd = LTTNG_POLL_GETFD(&events, i);
2552
2553 health_code_update();
2554
2555 if (last_seen_data_fd == pollfd) {
2556 idx = i;
2557 break;
2558 }
2559 }
2560 }
2561
2562 /* Process data connection. */
2563 for (i = idx + 1; i < nb_fd; i++) {
2564 /* Fetch the poll data. */
2565 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2566 int pollfd = LTTNG_POLL_GETFD(&events, i);
2567
2568 health_code_update();
2569
2570 /* Skip the command pipe. It's handled in the first loop. */
2571 if (pollfd == relay_cmd_pipe[0]) {
2572 continue;
2573 }
2574
2575 if (revents) {
2576 rcu_read_lock();
2577 lttng_ht_lookup(relay_connections_ht,
2578 (void *)((unsigned long) pollfd),
2579 &iter);
2580 node = lttng_ht_iter_get_node_ulong(&iter);
2581 if (node == NULL) {
2582 /* Skip it. Might be removed before. */
2583 rcu_read_unlock();
2584 continue;
2585 }
2586 relay_connection = caa_container_of(node,
2587 struct relay_command, sock_n);
2588
2589 if (revents & LPOLLIN) {
2590 if (relay_connection->type != RELAY_DATA) {
2591 continue;
2592 }
2593
2594 ret = relay_process_data(relay_connection);
2595 /* connection closed */
2596 if (ret < 0) {
2597 relay_cleanup_poll_connection(&events, pollfd);
2598 relay_del_connection(relay_connections_ht,
2599 &iter, relay_connection, sessions_ht);
2600 DBG("Data connection closed with %d", pollfd);
2601 /*
2602 * Every goto restart call sets the last seen fd where
2603 * here we don't really care since we gracefully
2604 * continue the loop after the connection is deleted.
2605 */
2606 } else {
2607 /* Keep last seen port. */
2608 last_seen_data_fd = pollfd;
2609 rcu_read_unlock();
2610 goto restart;
2611 }
2612 }
2613 rcu_read_unlock();
2614 }
2615 }
2616 last_seen_data_fd = -1;
2617 }
2618
2619 /* Normal exit, no error */
2620 ret = 0;
2621
2622 exit:
2623 error:
2624 lttng_poll_clean(&events);
2625
2626 /* empty the hash table and free the memory */
2627 rcu_read_lock();
2628 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2629 health_code_update();
2630
2631 node = lttng_ht_iter_get_node_ulong(&iter);
2632 if (node) {
2633 relay_connection = caa_container_of(node,
2634 struct relay_command, sock_n);
2635 relay_del_connection(relay_connections_ht,
2636 &iter, relay_connection, sessions_ht);
2637 }
2638 }
2639 rcu_read_unlock();
2640 error_poll_create:
2641 lttng_ht_destroy(indexes_ht);
2642 indexes_ht_error:
2643 lttng_ht_destroy(relay_connections_ht);
2644 relay_connections_ht_error:
2645 /* Close relay cmd pipes */
2646 utils_close_pipe(relay_cmd_pipe);
2647 if (err) {
2648 DBG("Thread exited with error");
2649 }
2650 DBG("Worker thread cleanup complete");
2651 free(data_buffer);
2652 if (err) {
2653 health_error();
2654 ERR("Health error occurred in %s", __func__);
2655 }
2656 health_unregister(health_relayd);
2657 rcu_unregister_thread();
2658 stop_threads();
2659 return NULL;
2660 }
2661
2662 /*
2663 * Create the relay command pipe to wake thread_manage_apps.
2664 * Closed in cleanup().
2665 */
2666 static int create_relay_cmd_pipe(void)
2667 {
2668 int ret;
2669
2670 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2671
2672 return ret;
2673 }
2674
2675 /*
2676 * main
2677 */
2678 int main(int argc, char **argv)
2679 {
2680 int ret = 0;
2681 void *status;
2682 struct relay_local_data *relay_ctx;
2683
2684 /* Create thread quit pipe */
2685 if ((ret = init_thread_quit_pipe()) < 0) {
2686 goto error;
2687 }
2688
2689 /* Parse arguments */
2690 progname = argv[0];
2691 if ((ret = parse_args(argc, argv)) < 0) {
2692 goto exit;
2693 }
2694
2695 if ((ret = set_signal_handler()) < 0) {
2696 goto exit;
2697 }
2698
2699 /* Try to create directory if -o, --output is specified. */
2700 if (opt_output_path) {
2701 if (*opt_output_path != '/') {
2702 ERR("Please specify an absolute path for -o, --output PATH");
2703 goto exit;
2704 }
2705
2706 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2707 if (ret < 0) {
2708 ERR("Unable to create %s", opt_output_path);
2709 goto exit;
2710 }
2711 }
2712
2713 /* Daemonize */
2714 if (opt_daemon) {
2715 ret = daemon(0, 0);
2716 if (ret < 0) {
2717 PERROR("daemon");
2718 goto exit;
2719 }
2720 }
2721
2722 /* We need those values for the file/dir creation. */
2723 relayd_uid = getuid();
2724 relayd_gid = getgid();
2725
2726 /* Check if daemon is UID = 0 */
2727 if (relayd_uid == 0) {
2728 if (control_uri->port < 1024 || data_uri->port < 1024 ||
2729 live_uri->port < 1024) {
2730 ERR("Need to be root to use ports < 1024");
2731 ret = -1;
2732 goto exit;
2733 }
2734 }
2735
2736 /* Setup the thread apps communication pipe. */
2737 if ((ret = create_relay_cmd_pipe()) < 0) {
2738 goto exit;
2739 }
2740
2741 /* Init relay command queue. */
2742 cds_wfq_init(&relay_cmd_queue.queue);
2743
2744 /* Set up max poll set size */
2745 lttng_poll_set_max_size();
2746
2747 /* Initialize communication library */
2748 lttcomm_init();
2749
2750 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2751 if (!relay_ctx) {
2752 PERROR("relay_ctx");
2753 goto exit;
2754 }
2755
2756 /* tables of sessions indexed by session ID */
2757 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2758 if (!relay_ctx->sessions_ht) {
2759 goto exit_relay_ctx_sessions;
2760 }
2761
2762 /* tables of streams indexed by stream ID */
2763 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2764 if (!relay_streams_ht) {
2765 goto exit_relay_ctx_streams;
2766 }
2767
2768 /* tables of streams indexed by stream ID */
2769 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2770 if (!viewer_streams_ht) {
2771 goto exit_relay_ctx_viewer_streams;
2772 }
2773
2774 /* Initialize thread health monitoring */
2775 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2776 if (!health_relayd) {
2777 PERROR("health_app_create error");
2778 goto exit_health_app_create;
2779 }
2780
2781 ret = utils_create_pipe(health_quit_pipe);
2782 if (ret < 0) {
2783 goto error_health_pipe;
2784 }
2785
2786 /* Create thread to manage the client socket */
2787 ret = pthread_create(&health_thread, NULL,
2788 thread_manage_health, (void *) NULL);
2789 if (ret != 0) {
2790 PERROR("pthread_create health");
2791 goto health_error;
2792 }
2793
2794 /* Setup the dispatcher thread */
2795 ret = pthread_create(&dispatcher_thread, NULL,
2796 relay_thread_dispatcher, (void *) NULL);
2797 if (ret != 0) {
2798 PERROR("pthread_create dispatcher");
2799 goto exit_dispatcher;
2800 }
2801
2802 /* Setup the worker thread */
2803 ret = pthread_create(&worker_thread, NULL,
2804 relay_thread_worker, (void *) relay_ctx);
2805 if (ret != 0) {
2806 PERROR("pthread_create worker");
2807 goto exit_worker;
2808 }
2809
2810 /* Setup the listener thread */
2811 ret = pthread_create(&listener_thread, NULL,
2812 relay_thread_listener, (void *) NULL);
2813 if (ret != 0) {
2814 PERROR("pthread_create listener");
2815 goto exit_listener;
2816 }
2817
2818 ret = live_start_threads(live_uri, relay_ctx, thread_quit_pipe);
2819 if (ret != 0) {
2820 ERR("Starting live viewer threads");
2821 goto exit_live;
2822 }
2823
2824 exit_live:
2825 ret = pthread_join(listener_thread, &status);
2826 if (ret != 0) {
2827 PERROR("pthread_join");
2828 goto error; /* join error, exit without cleanup */
2829 }
2830
2831 exit_listener:
2832 ret = pthread_join(worker_thread, &status);
2833 if (ret != 0) {
2834 PERROR("pthread_join");
2835 goto error; /* join error, exit without cleanup */
2836 }
2837
2838 exit_worker:
2839 ret = pthread_join(dispatcher_thread, &status);
2840 if (ret != 0) {
2841 PERROR("pthread_join");
2842 goto error; /* join error, exit without cleanup */
2843 }
2844
2845 exit_dispatcher:
2846 ret = pthread_join(health_thread, &status);
2847 if (ret != 0) {
2848 PERROR("pthread_join health thread");
2849 goto error; /* join error, exit without cleanup */
2850 }
2851
2852 /*
2853 * Stop live threads only after joining other threads.
2854 */
2855 live_stop_threads();
2856
2857 health_error:
2858 utils_close_pipe(health_quit_pipe);
2859
2860 error_health_pipe:
2861 health_app_destroy(health_relayd);
2862
2863 exit_health_app_create:
2864 lttng_ht_destroy(viewer_streams_ht);
2865
2866 exit_relay_ctx_viewer_streams:
2867 lttng_ht_destroy(relay_streams_ht);
2868
2869 exit_relay_ctx_streams:
2870 lttng_ht_destroy(relay_ctx->sessions_ht);
2871
2872 exit_relay_ctx_sessions:
2873 free(relay_ctx);
2874
2875 exit:
2876 cleanup();
2877 if (!ret) {
2878 exit(EXIT_SUCCESS);
2879 }
2880
2881 error:
2882 exit(EXIT_FAILURE);
2883 }
This page took 0.086922 seconds and 5 git commands to generate.