4d53ba9c886f80dfa96dd5261d612e603e7ce601
[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/daemonize.h>
48 #include <common/futex.h>
49 #include <common/sessiond-comm/sessiond-comm.h>
50 #include <common/sessiond-comm/inet.h>
51 #include <common/sessiond-comm/relayd.h>
52 #include <common/uri.h>
53 #include <common/utils.h>
54
55 #include "cmd.h"
56 #include "ctf-trace.h"
57 #include "index.h"
58 #include "utils.h"
59 #include "lttng-relayd.h"
60 #include "live.h"
61 #include "health-relayd.h"
62
63 /* command line options */
64 char *opt_output_path;
65 static int opt_daemon, opt_background;
66
67 /*
68 * We need to wait for listener and live listener threads, as well as
69 * health check thread, before being ready to signal readiness.
70 */
71 #define NR_LTTNG_RELAY_READY 3
72 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
73 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
74 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
75
76 static struct lttng_uri *control_uri;
77 static struct lttng_uri *data_uri;
78 static struct lttng_uri *live_uri;
79
80 const char *progname;
81
82 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
83
84 /*
85 * Quit pipe for all threads. This permits a single cancellation point
86 * for all threads when receiving an event on the pipe.
87 */
88 int thread_quit_pipe[2] = { -1, -1 };
89
90 /*
91 * This pipe is used to inform the worker thread that a command is queued and
92 * ready to be processed.
93 */
94 static int relay_cmd_pipe[2] = { -1, -1 };
95
96 /* Shared between threads */
97 static int dispatch_thread_exit;
98
99 static pthread_t listener_thread;
100 static pthread_t dispatcher_thread;
101 static pthread_t worker_thread;
102 static pthread_t health_thread;
103
104 static uint64_t last_relay_stream_id;
105 static uint64_t last_relay_session_id;
106
107 /*
108 * Relay command queue.
109 *
110 * The relay_thread_listener and relay_thread_dispatcher communicate with this
111 * queue.
112 */
113 static struct relay_cmd_queue relay_cmd_queue;
114
115 /* buffer allocated at startup, used to store the trace data */
116 static char *data_buffer;
117 static unsigned int data_buffer_size;
118
119 /* We need those values for the file/dir creation. */
120 static uid_t relayd_uid;
121 static gid_t relayd_gid;
122
123 /* Global relay stream hash table. */
124 struct lttng_ht *relay_streams_ht;
125
126 /* Global relay viewer stream hash table. */
127 struct lttng_ht *viewer_streams_ht;
128
129 /* Global hash table that stores relay index object. */
130 struct lttng_ht *indexes_ht;
131
132 /* Relayd health monitoring */
133 struct health_app *health_relayd;
134
135 /*
136 * usage function on stderr
137 */
138 static
139 void usage(void)
140 {
141 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
142 fprintf(stderr, " -h, --help Display this usage.\n");
143 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
144 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
145 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
146 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
147 fprintf(stderr, " -L, --live-port URL Live view port listening.\n");
148 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
149 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
150 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
151 }
152
153 static
154 int parse_args(int argc, char **argv)
155 {
156 int c;
157 int ret = 0;
158 char *default_address;
159
160 static struct option long_options[] = {
161 { "control-port", 1, 0, 'C', },
162 { "data-port", 1, 0, 'D', },
163 { "daemonize", 0, 0, 'd', },
164 { "group", 1, 0, 'g', },
165 { "help", 0, 0, 'h', },
166 { "output", 1, 0, 'o', },
167 { "verbose", 0, 0, 'v', },
168 { NULL, 0, 0, 0, },
169 };
170
171 while (1) {
172 int option_index = 0;
173 c = getopt_long(argc, argv, "dhv" "C:D:L:o:g:",
174 long_options, &option_index);
175 if (c == -1) {
176 break;
177 }
178
179 switch (c) {
180 case 0:
181 fprintf(stderr, "option %s", long_options[option_index].name);
182 if (optarg) {
183 fprintf(stderr, " with arg %s\n", optarg);
184 }
185 break;
186 case 'C':
187 ret = uri_parse(optarg, &control_uri);
188 if (ret < 0) {
189 ERR("Invalid control URI specified");
190 goto exit;
191 }
192 if (control_uri->port == 0) {
193 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
194 }
195 break;
196 case 'D':
197 ret = uri_parse(optarg, &data_uri);
198 if (ret < 0) {
199 ERR("Invalid data URI specified");
200 goto exit;
201 }
202 if (data_uri->port == 0) {
203 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
204 }
205 break;
206 case 'L':
207 ret = uri_parse(optarg, &live_uri);
208 if (ret < 0) {
209 ERR("Invalid live URI specified");
210 goto exit;
211 }
212 if (live_uri->port == 0) {
213 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
214 }
215 break;
216 case 'd':
217 opt_daemon = 1;
218 break;
219 case 'g':
220 tracing_group_name = optarg;
221 break;
222 case 'h':
223 usage();
224 exit(EXIT_FAILURE);
225 case 'o':
226 ret = asprintf(&opt_output_path, "%s", optarg);
227 if (ret < 0) {
228 PERROR("asprintf opt_output_path");
229 goto exit;
230 }
231 break;
232 case 'v':
233 /* Verbose level can increase using multiple -v */
234 lttng_opt_verbose += 1;
235 break;
236 default:
237 /* Unknown option or other error.
238 * Error is printed by getopt, just return */
239 ret = -1;
240 goto exit;
241 }
242 }
243
244 /* assign default values */
245 if (control_uri == NULL) {
246 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
247 DEFAULT_NETWORK_CONTROL_PORT);
248 if (ret < 0) {
249 PERROR("asprintf default data address");
250 goto exit;
251 }
252
253 ret = uri_parse(default_address, &control_uri);
254 free(default_address);
255 if (ret < 0) {
256 ERR("Invalid control URI specified");
257 goto exit;
258 }
259 }
260 if (data_uri == NULL) {
261 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
262 DEFAULT_NETWORK_DATA_PORT);
263 if (ret < 0) {
264 PERROR("asprintf default data address");
265 goto exit;
266 }
267
268 ret = uri_parse(default_address, &data_uri);
269 free(default_address);
270 if (ret < 0) {
271 ERR("Invalid data URI specified");
272 goto exit;
273 }
274 }
275 if (live_uri == NULL) {
276 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
277 DEFAULT_NETWORK_VIEWER_PORT);
278 if (ret < 0) {
279 PERROR("asprintf default viewer control address");
280 goto exit;
281 }
282
283 ret = uri_parse(default_address, &live_uri);
284 free(default_address);
285 if (ret < 0) {
286 ERR("Invalid viewer control URI specified");
287 goto exit;
288 }
289 }
290
291 exit:
292 return ret;
293 }
294
295 /*
296 * Cleanup the daemon
297 */
298 static
299 void cleanup(void)
300 {
301 DBG("Cleaning up");
302
303 /* free the dynamically allocated opt_output_path */
304 free(opt_output_path);
305
306 /* Close thread quit pipes */
307 utils_close_pipe(thread_quit_pipe);
308
309 uri_free(control_uri);
310 uri_free(data_uri);
311 /* Live URI is freed in the live thread. */
312 }
313
314 /*
315 * Write to writable pipe used to notify a thread.
316 */
317 static
318 int notify_thread_pipe(int wpipe)
319 {
320 ssize_t ret;
321
322 ret = lttng_write(wpipe, "!", 1);
323 if (ret < 1) {
324 PERROR("write poll pipe");
325 }
326
327 return ret;
328 }
329
330 static void notify_health_quit_pipe(int *pipe)
331 {
332 ssize_t ret;
333
334 ret = lttng_write(pipe[1], "4", 1);
335 if (ret < 1) {
336 PERROR("write relay health quit");
337 }
338 }
339
340 /*
341 * Stop all threads by closing the thread quit pipe.
342 */
343 static
344 void stop_threads(void)
345 {
346 int ret;
347
348 /* Stopping all threads */
349 DBG("Terminating all threads");
350 ret = notify_thread_pipe(thread_quit_pipe[1]);
351 if (ret < 0) {
352 ERR("write error on thread quit pipe");
353 }
354
355 notify_health_quit_pipe(health_quit_pipe);
356
357 /* Dispatch thread */
358 CMM_STORE_SHARED(dispatch_thread_exit, 1);
359 futex_nto1_wake(&relay_cmd_queue.futex);
360 }
361
362 /*
363 * Signal handler for the daemon
364 *
365 * Simply stop all worker threads, leaving main() return gracefully after
366 * joining all threads and calling cleanup().
367 */
368 static
369 void sighandler(int sig)
370 {
371 switch (sig) {
372 case SIGPIPE:
373 DBG("SIGPIPE caught");
374 return;
375 case SIGINT:
376 DBG("SIGINT caught");
377 stop_threads();
378 break;
379 case SIGTERM:
380 DBG("SIGTERM caught");
381 stop_threads();
382 break;
383 case SIGUSR1:
384 CMM_STORE_SHARED(recv_child_signal, 1);
385 break;
386 default:
387 break;
388 }
389 }
390
391 /*
392 * Setup signal handler for :
393 * SIGINT, SIGTERM, SIGPIPE
394 */
395 static
396 int set_signal_handler(void)
397 {
398 int ret = 0;
399 struct sigaction sa;
400 sigset_t sigset;
401
402 if ((ret = sigemptyset(&sigset)) < 0) {
403 PERROR("sigemptyset");
404 return ret;
405 }
406
407 sa.sa_handler = sighandler;
408 sa.sa_mask = sigset;
409 sa.sa_flags = 0;
410 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
411 PERROR("sigaction");
412 return ret;
413 }
414
415 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
416 PERROR("sigaction");
417 return ret;
418 }
419
420 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
421 PERROR("sigaction");
422 return ret;
423 }
424
425 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
426 PERROR("sigaction");
427 return ret;
428 }
429
430 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
431
432 return ret;
433 }
434
435 void lttng_relay_notify_ready(void)
436 {
437 /* Notify the parent of the fork() process that we are ready. */
438 if (opt_daemon || opt_background) {
439 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
440 kill(child_ppid, SIGUSR1);
441 }
442 }
443 }
444
445 /*
446 * Init thread quit pipe.
447 *
448 * Return -1 on error or 0 if all pipes are created.
449 */
450 static
451 int init_thread_quit_pipe(void)
452 {
453 int ret;
454
455 ret = utils_create_pipe_cloexec(thread_quit_pipe);
456
457 return ret;
458 }
459
460 /*
461 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
462 */
463 static
464 int create_thread_poll_set(struct lttng_poll_event *events, int size)
465 {
466 int ret;
467
468 if (events == NULL || size == 0) {
469 ret = -1;
470 goto error;
471 }
472
473 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
474 if (ret < 0) {
475 goto error;
476 }
477
478 /* Add quit pipe */
479 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
480 if (ret < 0) {
481 goto error;
482 }
483
484 return 0;
485
486 error:
487 return ret;
488 }
489
490 /*
491 * Check if the thread quit pipe was triggered.
492 *
493 * Return 1 if it was triggered else 0;
494 */
495 static
496 int check_thread_quit_pipe(int fd, uint32_t events)
497 {
498 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
499 return 1;
500 }
501
502 return 0;
503 }
504
505 /*
506 * Create and init socket from uri.
507 */
508 static
509 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
510 {
511 int ret;
512 struct lttcomm_sock *sock = NULL;
513
514 sock = lttcomm_alloc_sock_from_uri(uri);
515 if (sock == NULL) {
516 ERR("Allocating socket");
517 goto error;
518 }
519
520 ret = lttcomm_create_sock(sock);
521 if (ret < 0) {
522 goto error;
523 }
524 DBG("Listening on sock %d", sock->fd);
525
526 ret = sock->ops->bind(sock);
527 if (ret < 0) {
528 goto error;
529 }
530
531 ret = sock->ops->listen(sock, -1);
532 if (ret < 0) {
533 goto error;
534
535 }
536
537 return sock;
538
539 error:
540 if (sock) {
541 lttcomm_destroy_sock(sock);
542 }
543 return NULL;
544 }
545
546 /*
547 * Return nonzero if stream needs to be closed.
548 */
549 static
550 int close_stream_check(struct relay_stream *stream)
551 {
552
553 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
554 /*
555 * We are about to close the stream so set the data pending flag to 1
556 * which will make the end data pending command skip the stream which
557 * is now closed and ready. Note that after proceeding to a file close,
558 * the written file is ready for reading.
559 */
560 stream->data_pending_check_done = 1;
561 return 1;
562 }
563 return 0;
564 }
565
566 /*
567 * This thread manages the listening for new connections on the network
568 */
569 static
570 void *relay_thread_listener(void *data)
571 {
572 int i, ret, pollfd, err = -1;
573 int val = 1;
574 uint32_t revents, nb_fd;
575 struct lttng_poll_event events;
576 struct lttcomm_sock *control_sock, *data_sock;
577
578 DBG("[thread] Relay listener started");
579
580 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
581
582 health_code_update();
583
584 control_sock = relay_init_sock(control_uri);
585 if (!control_sock) {
586 goto error_sock_control;
587 }
588
589 data_sock = relay_init_sock(data_uri);
590 if (!data_sock) {
591 goto error_sock_relay;
592 }
593
594 /*
595 * Pass 3 as size here for the thread quit pipe, control and data socket.
596 */
597 ret = create_thread_poll_set(&events, 3);
598 if (ret < 0) {
599 goto error_create_poll;
600 }
601
602 /* Add the control socket */
603 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
604 if (ret < 0) {
605 goto error_poll_add;
606 }
607
608 /* Add the data socket */
609 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
610 if (ret < 0) {
611 goto error_poll_add;
612 }
613
614 lttng_relay_notify_ready();
615
616 while (1) {
617 health_code_update();
618
619 DBG("Listener accepting connections");
620
621 restart:
622 health_poll_entry();
623 ret = lttng_poll_wait(&events, -1);
624 health_poll_exit();
625 if (ret < 0) {
626 /*
627 * Restart interrupted system call.
628 */
629 if (errno == EINTR) {
630 goto restart;
631 }
632 goto error;
633 }
634
635 nb_fd = ret;
636
637 DBG("Relay new connection received");
638 for (i = 0; i < nb_fd; i++) {
639 health_code_update();
640
641 /* Fetch once the poll data */
642 revents = LTTNG_POLL_GETEV(&events, i);
643 pollfd = LTTNG_POLL_GETFD(&events, i);
644
645 /* Thread quit pipe has been closed. Killing thread. */
646 ret = check_thread_quit_pipe(pollfd, revents);
647 if (ret) {
648 err = 0;
649 goto exit;
650 }
651
652 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
653 ERR("socket poll error");
654 goto error;
655 } else if (revents & LPOLLIN) {
656 /*
657 * Get allocated in this thread,
658 * enqueued to a global queue, dequeued
659 * and freed in the worker thread.
660 */
661 struct relay_command *relay_cmd;
662 struct lttcomm_sock *newsock;
663
664 relay_cmd = zmalloc(sizeof(struct relay_command));
665 if (relay_cmd == NULL) {
666 PERROR("relay command zmalloc");
667 goto error;
668 }
669
670 if (pollfd == data_sock->fd) {
671 newsock = data_sock->ops->accept(data_sock);
672 if (!newsock) {
673 PERROR("accepting data sock");
674 free(relay_cmd);
675 goto error;
676 }
677 relay_cmd->type = RELAY_DATA;
678 DBG("Relay data connection accepted, socket %d", newsock->fd);
679 } else {
680 assert(pollfd == control_sock->fd);
681 newsock = control_sock->ops->accept(control_sock);
682 if (!newsock) {
683 PERROR("accepting control sock");
684 free(relay_cmd);
685 goto error;
686 }
687 relay_cmd->type = RELAY_CONTROL;
688 DBG("Relay control connection accepted, socket %d", newsock->fd);
689 }
690 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
691 &val, sizeof(int));
692 if (ret < 0) {
693 PERROR("setsockopt inet");
694 lttcomm_destroy_sock(newsock);
695 free(relay_cmd);
696 goto error;
697 }
698 relay_cmd->sock = newsock;
699 /*
700 * Lock free enqueue the request.
701 */
702 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
703
704 /*
705 * Wake the dispatch queue futex. Implicit memory
706 * barrier with the exchange in cds_wfq_enqueue.
707 */
708 futex_nto1_wake(&relay_cmd_queue.futex);
709 }
710 }
711 }
712
713 exit:
714 error:
715 error_poll_add:
716 lttng_poll_clean(&events);
717 error_create_poll:
718 if (data_sock->fd >= 0) {
719 ret = data_sock->ops->close(data_sock);
720 if (ret) {
721 PERROR("close");
722 }
723 }
724 lttcomm_destroy_sock(data_sock);
725 error_sock_relay:
726 if (control_sock->fd >= 0) {
727 ret = control_sock->ops->close(control_sock);
728 if (ret) {
729 PERROR("close");
730 }
731 }
732 lttcomm_destroy_sock(control_sock);
733 error_sock_control:
734 if (err) {
735 health_error();
736 ERR("Health error occurred in %s", __func__);
737 }
738 health_unregister(health_relayd);
739 DBG("Relay listener thread cleanup complete");
740 stop_threads();
741 return NULL;
742 }
743
744 /*
745 * This thread manages the dispatching of the requests to worker threads
746 */
747 static
748 void *relay_thread_dispatcher(void *data)
749 {
750 int err = -1;
751 ssize_t ret;
752 struct cds_wfq_node *node;
753 struct relay_command *relay_cmd = NULL;
754
755 DBG("[thread] Relay dispatcher started");
756
757 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
758
759 health_code_update();
760
761 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
762 health_code_update();
763
764 /* Atomically prepare the queue futex */
765 futex_nto1_prepare(&relay_cmd_queue.futex);
766
767 do {
768 health_code_update();
769
770 /* Dequeue commands */
771 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
772 if (node == NULL) {
773 DBG("Woken up but nothing in the relay command queue");
774 /* Continue thread execution */
775 break;
776 }
777
778 relay_cmd = caa_container_of(node, struct relay_command, node);
779 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
780
781 /*
782 * Inform worker thread of the new request. This
783 * call is blocking so we can be assured that the data will be read
784 * at some point in time or wait to the end of the world :)
785 */
786 ret = lttng_write(relay_cmd_pipe[1], relay_cmd,
787 sizeof(struct relay_command));
788 free(relay_cmd);
789 if (ret < sizeof(struct relay_command)) {
790 PERROR("write cmd pipe");
791 goto error;
792 }
793 } while (node != NULL);
794
795 /* Futex wait on queue. Blocking call on futex() */
796 health_poll_entry();
797 futex_nto1_wait(&relay_cmd_queue.futex);
798 health_poll_exit();
799 }
800
801 /* Normal exit, no error */
802 err = 0;
803
804 error:
805 if (err) {
806 health_error();
807 ERR("Health error occurred in %s", __func__);
808 }
809 health_unregister(health_relayd);
810 DBG("Dispatch thread dying");
811 stop_threads();
812 return NULL;
813 }
814
815 /*
816 * Get stream from stream id.
817 * Need to be called with RCU read-side lock held.
818 */
819 struct relay_stream *relay_stream_find_by_id(uint64_t stream_id)
820 {
821 struct lttng_ht_node_ulong *node;
822 struct lttng_ht_iter iter;
823 struct relay_stream *ret;
824
825 lttng_ht_lookup(relay_streams_ht,
826 (void *)((unsigned long) stream_id),
827 &iter);
828 node = lttng_ht_iter_get_node_ulong(&iter);
829 if (node == NULL) {
830 DBG("Relay stream %" PRIu64 " not found", stream_id);
831 ret = NULL;
832 goto end;
833 }
834
835 ret = caa_container_of(node, struct relay_stream, stream_n);
836
837 end:
838 return ret;
839 }
840
841 static
842 void deferred_free_stream(struct rcu_head *head)
843 {
844 struct relay_stream *stream =
845 caa_container_of(head, struct relay_stream, rcu_node);
846
847 free(stream->path_name);
848 free(stream->channel_name);
849 free(stream);
850 }
851
852 static
853 void deferred_free_session(struct rcu_head *head)
854 {
855 struct relay_session *session =
856 caa_container_of(head, struct relay_session, rcu_node);
857 free(session);
858 }
859
860 /*
861 * Close a given stream. The stream is freed using a call RCU.
862 *
863 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
864 * BEFORE the stream lock MUST be acquired.
865 */
866 static void destroy_stream(struct relay_stream *stream)
867 {
868 int delret;
869 struct relay_viewer_stream *vstream;
870 struct lttng_ht_iter iter;
871
872 assert(stream);
873
874 delret = close(stream->fd);
875 if (delret < 0) {
876 PERROR("close stream");
877 }
878
879 if (stream->index_fd >= 0) {
880 delret = close(stream->index_fd);
881 if (delret < 0) {
882 PERROR("close stream index_fd");
883 }
884 }
885
886 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
887 if (vstream) {
888 /*
889 * Set the last good value into the viewer stream. This is done
890 * right before the stream gets deleted from the hash table. The
891 * lookup failure on the live thread side of a stream indicates
892 * that the viewer stream index received value should be used.
893 */
894 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
895 vstream->total_index_received = stream->total_index_received;
896 vstream->tracefile_count_last = stream->tracefile_count_current;
897 vstream->close_write_flag = 1;
898 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
899 }
900
901 /* Cleanup index of that stream. */
902 relay_index_destroy_by_stream_id(stream->stream_handle);
903
904 iter.iter.node = &stream->stream_n.node;
905 delret = lttng_ht_del(relay_streams_ht, &iter);
906 assert(!delret);
907 iter.iter.node = &stream->ctf_trace_node.node;
908 delret = lttng_ht_del(stream->ctf_traces_ht, &iter);
909 assert(!delret);
910
911 if (stream->ctf_trace) {
912 ctf_trace_try_destroy(stream->ctf_trace);
913 }
914
915 call_rcu(&stream->rcu_node, deferred_free_stream);
916 DBG("Closed tracefile %d from close stream", stream->fd);
917 }
918
919 /*
920 * relay_delete_session: Free all memory associated with a session and
921 * close all the FDs
922 */
923 static
924 void relay_delete_session(struct relay_command *cmd,
925 struct lttng_ht *sessions_ht)
926 {
927 struct lttng_ht_iter iter;
928 struct lttng_ht_node_ulong *node;
929 struct relay_stream *stream;
930 int ret;
931
932 if (!cmd->session) {
933 return;
934 }
935
936 DBG("Relay deleting session %" PRIu64, cmd->session->id);
937
938 rcu_read_lock();
939 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
940 node = lttng_ht_iter_get_node_ulong(&iter);
941 if (!node) {
942 continue;
943 }
944 stream = caa_container_of(node, struct relay_stream, stream_n);
945 if (stream->session == cmd->session) {
946 destroy_stream(stream);
947 cmd->session->stream_count--;
948 assert(cmd->session->stream_count >= 0);
949 }
950 }
951
952 /* Make this session not visible anymore. */
953 iter.iter.node = &cmd->session->session_n.node;
954 ret = lttng_ht_del(sessions_ht, &iter);
955 assert(!ret);
956 call_rcu(&cmd->session->rcu_node, deferred_free_session);
957 rcu_read_unlock();
958 }
959
960 /*
961 * Copy index data from the control port to a given index object.
962 */
963 static void copy_index_control_data(struct relay_index *index,
964 struct lttcomm_relayd_index *data)
965 {
966 assert(index);
967 assert(data);
968
969 /*
970 * The index on disk is encoded in big endian, so we don't need to convert
971 * the data received on the network. The data_offset value is NEVER
972 * modified here and is updated by the data thread.
973 */
974 index->index_data.packet_size = data->packet_size;
975 index->index_data.content_size = data->content_size;
976 index->index_data.timestamp_begin = data->timestamp_begin;
977 index->index_data.timestamp_end = data->timestamp_end;
978 index->index_data.events_discarded = data->events_discarded;
979 index->index_data.stream_id = data->stream_id;
980 }
981
982 /*
983 * Handle the RELAYD_CREATE_SESSION command.
984 *
985 * On success, send back the session id or else return a negative value.
986 */
987 static
988 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
989 struct relay_command *cmd,
990 struct lttng_ht *sessions_ht)
991 {
992 int ret = 0, send_ret;
993 struct relay_session *session;
994 struct lttcomm_relayd_status_session reply;
995
996 assert(recv_hdr);
997 assert(cmd);
998
999 memset(&reply, 0, sizeof(reply));
1000
1001 session = zmalloc(sizeof(struct relay_session));
1002 if (session == NULL) {
1003 PERROR("relay session zmalloc");
1004 ret = -1;
1005 goto error;
1006 }
1007
1008 session->id = ++last_relay_session_id;
1009 session->sock = cmd->sock;
1010 session->minor = cmd->minor;
1011 session->major = cmd->major;
1012 cmd->session = session;
1013
1014 reply.session_id = htobe64(session->id);
1015
1016 switch (cmd->minor) {
1017 case 4: /* LTTng sessiond 2.4 */
1018 default:
1019 ret = cmd_create_session_2_4(cmd, session);
1020 break;
1021 }
1022
1023 lttng_ht_node_init_ulong(&session->session_n,
1024 (unsigned long) session->id);
1025 lttng_ht_add_unique_ulong(sessions_ht,
1026 &session->session_n);
1027
1028 DBG("Created session %" PRIu64, session->id);
1029
1030 error:
1031 if (ret < 0) {
1032 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1033 } else {
1034 reply.ret_code = htobe32(LTTNG_OK);
1035 }
1036
1037 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1038 if (send_ret < 0) {
1039 ERR("Relayd sending session id");
1040 ret = send_ret;
1041 }
1042
1043 return ret;
1044 }
1045
1046 /*
1047 * When we have received all the streams and the metadata for a channel,
1048 * we make them visible to the viewer threads.
1049 */
1050 static
1051 void set_viewer_ready_flag(struct relay_command *cmd)
1052 {
1053 struct relay_stream_recv_handle *node, *tmp_node;
1054
1055 cds_list_for_each_entry_safe(node, tmp_node, &cmd->recv_head, node) {
1056 struct relay_stream *stream;
1057
1058 rcu_read_lock();
1059 stream = relay_stream_find_by_id(node->id);
1060 if (!stream) {
1061 /*
1062 * Stream is most probably being cleaned up by the data thread thus
1063 * simply continue to the next one.
1064 */
1065 rcu_read_unlock();
1066 continue;
1067 }
1068
1069 stream->viewer_ready = 1;
1070 rcu_read_unlock();
1071
1072 /* Clean stream handle node. */
1073 cds_list_del(&node->node);
1074 free(node);
1075 }
1076
1077 return;
1078 }
1079
1080 /*
1081 * Add a recv handle node to the connection recv list with the given stream
1082 * handle. A new node is allocated thus must be freed when the node is deleted
1083 * from the list.
1084 */
1085 static void queue_stream_handle(uint64_t handle, struct relay_command *cmd)
1086 {
1087 struct relay_stream_recv_handle *node;
1088
1089 assert(cmd);
1090
1091 node = zmalloc(sizeof(*node));
1092 if (!node) {
1093 PERROR("zmalloc queue stream handle");
1094 return;
1095 }
1096
1097 node->id = handle;
1098 cds_list_add(&node->node, &cmd->recv_head);
1099 }
1100
1101 /*
1102 * relay_add_stream: allocate a new stream for a session
1103 */
1104 static
1105 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1106 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1107 {
1108 struct relay_session *session = cmd->session;
1109 struct relay_stream *stream = NULL;
1110 struct lttcomm_relayd_status_stream reply;
1111 int ret, send_ret;
1112
1113 if (!session || cmd->version_check_done == 0) {
1114 ERR("Trying to add a stream before version check");
1115 ret = -1;
1116 goto end_no_session;
1117 }
1118
1119 stream = zmalloc(sizeof(struct relay_stream));
1120 if (stream == NULL) {
1121 PERROR("relay stream zmalloc");
1122 ret = -1;
1123 goto end_no_session;
1124 }
1125
1126 switch (cmd->minor) {
1127 case 1: /* LTTng sessiond 2.1 */
1128 ret = cmd_recv_stream_2_1(cmd, stream);
1129 break;
1130 case 2: /* LTTng sessiond 2.2 */
1131 default:
1132 ret = cmd_recv_stream_2_2(cmd, stream);
1133 break;
1134 }
1135 if (ret < 0) {
1136 goto err_free_stream;
1137 }
1138
1139 rcu_read_lock();
1140 stream->stream_handle = ++last_relay_stream_id;
1141 stream->prev_seq = -1ULL;
1142 stream->session = session;
1143 stream->index_fd = -1;
1144 stream->read_index_fd = -1;
1145 stream->ctf_trace = NULL;
1146 pthread_mutex_init(&stream->lock, NULL);
1147
1148 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
1149 if (ret < 0) {
1150 ERR("relay creating output directory");
1151 goto end;
1152 }
1153
1154 /*
1155 * No need to use run_as API here because whatever we receives, the relayd
1156 * uses its own credentials for the stream files.
1157 */
1158 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1159 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1160 if (ret < 0) {
1161 ERR("Create output file");
1162 goto end;
1163 }
1164 stream->fd = ret;
1165 if (stream->tracefile_size) {
1166 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1167 } else {
1168 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1169 }
1170
1171 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1172 stream->metadata_flag = 1;
1173 /*
1174 * When we receive a new metadata stream, we create a new
1175 * ctf_trace and we assign this ctf_trace to all streams with
1176 * the same path.
1177 *
1178 * If later on we receive a new stream for the same ctf_trace,
1179 * we copy the information from the first hit in the HT to the
1180 * new stream.
1181 */
1182 stream->ctf_trace = ctf_trace_create();
1183 if (!stream->ctf_trace) {
1184 ret = -1;
1185 goto end;
1186 }
1187 stream->ctf_trace->refcount++;
1188 stream->ctf_trace->metadata_stream = stream;
1189 }
1190 ctf_trace_assign(cmd->ctf_traces_ht, stream);
1191 stream->ctf_traces_ht = cmd->ctf_traces_ht;
1192
1193 /*
1194 * Add the stream handle in the recv list of the connection. Once the end
1195 * stream message is received, this list is emptied and streams are set
1196 * with the viewer ready flag.
1197 */
1198 if (stream->metadata_flag) {
1199 stream->viewer_ready = 1;
1200 } else {
1201 queue_stream_handle(stream->stream_handle, cmd);
1202 }
1203
1204 lttng_ht_node_init_ulong(&stream->stream_n,
1205 (unsigned long) stream->stream_handle);
1206 lttng_ht_add_unique_ulong(relay_streams_ht,
1207 &stream->stream_n);
1208
1209 lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name);
1210 lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node);
1211 session->stream_count++;
1212
1213 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1214 stream->stream_handle);
1215
1216 end:
1217 reply.handle = htobe64(stream->stream_handle);
1218 /* send the session id to the client or a negative return code on error */
1219 if (ret < 0) {
1220 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1221 /* stream was not properly added to the ht, so free it */
1222 free(stream);
1223 } else {
1224 reply.ret_code = htobe32(LTTNG_OK);
1225 }
1226
1227 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1228 sizeof(struct lttcomm_relayd_status_stream), 0);
1229 if (send_ret < 0) {
1230 ERR("Relay sending stream id");
1231 ret = send_ret;
1232 }
1233 rcu_read_unlock();
1234
1235 end_no_session:
1236 return ret;
1237
1238 err_free_stream:
1239 free(stream->path_name);
1240 free(stream->channel_name);
1241 free(stream);
1242 return ret;
1243 }
1244
1245 /*
1246 * relay_close_stream: close a specific stream
1247 */
1248 static
1249 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1250 struct relay_command *cmd)
1251 {
1252 int ret, send_ret;
1253 struct relay_session *session = cmd->session;
1254 struct lttcomm_relayd_close_stream stream_info;
1255 struct lttcomm_relayd_generic_reply reply;
1256 struct relay_stream *stream;
1257
1258 DBG("Close stream received");
1259
1260 if (!session || cmd->version_check_done == 0) {
1261 ERR("Trying to close a stream before version check");
1262 ret = -1;
1263 goto end_no_session;
1264 }
1265
1266 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1267 sizeof(struct lttcomm_relayd_close_stream), 0);
1268 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1269 if (ret == 0) {
1270 /* Orderly shutdown. Not necessary to print an error. */
1271 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1272 } else {
1273 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1274 }
1275 ret = -1;
1276 goto end_no_session;
1277 }
1278
1279 rcu_read_lock();
1280 stream = relay_stream_find_by_id(be64toh(stream_info.stream_id));
1281 if (!stream) {
1282 ret = -1;
1283 goto end_unlock;
1284 }
1285
1286 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1287 stream->close_flag = 1;
1288 session->stream_count--;
1289 assert(session->stream_count >= 0);
1290
1291 if (close_stream_check(stream)) {
1292 destroy_stream(stream);
1293 }
1294
1295 end_unlock:
1296 rcu_read_unlock();
1297
1298 if (ret < 0) {
1299 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1300 } else {
1301 reply.ret_code = htobe32(LTTNG_OK);
1302 }
1303 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1304 sizeof(struct lttcomm_relayd_generic_reply), 0);
1305 if (send_ret < 0) {
1306 ERR("Relay sending stream id");
1307 ret = send_ret;
1308 }
1309
1310 end_no_session:
1311 return ret;
1312 }
1313
1314 /*
1315 * relay_unknown_command: send -1 if received unknown command
1316 */
1317 static
1318 void relay_unknown_command(struct relay_command *cmd)
1319 {
1320 struct lttcomm_relayd_generic_reply reply;
1321 int ret;
1322
1323 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1324 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1325 sizeof(struct lttcomm_relayd_generic_reply), 0);
1326 if (ret < 0) {
1327 ERR("Relay sending unknown command");
1328 }
1329 }
1330
1331 /*
1332 * relay_start: send an acknowledgment to the client to tell if we are
1333 * ready to receive data. We are ready if a session is established.
1334 */
1335 static
1336 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1337 struct relay_command *cmd)
1338 {
1339 int ret = htobe32(LTTNG_OK);
1340 struct lttcomm_relayd_generic_reply reply;
1341 struct relay_session *session = cmd->session;
1342
1343 if (!session) {
1344 DBG("Trying to start the streaming without a session established");
1345 ret = htobe32(LTTNG_ERR_UNK);
1346 }
1347
1348 reply.ret_code = ret;
1349 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1350 sizeof(struct lttcomm_relayd_generic_reply), 0);
1351 if (ret < 0) {
1352 ERR("Relay sending start ack");
1353 }
1354
1355 return ret;
1356 }
1357
1358 /*
1359 * Append padding to the file pointed by the file descriptor fd.
1360 */
1361 static int write_padding_to_file(int fd, uint32_t size)
1362 {
1363 ssize_t ret = 0;
1364 char *zeros;
1365
1366 if (size == 0) {
1367 goto end;
1368 }
1369
1370 zeros = zmalloc(size);
1371 if (zeros == NULL) {
1372 PERROR("zmalloc zeros for padding");
1373 ret = -1;
1374 goto end;
1375 }
1376
1377 ret = lttng_write(fd, zeros, size);
1378 if (ret < size) {
1379 PERROR("write padding to file");
1380 }
1381
1382 free(zeros);
1383
1384 end:
1385 return ret;
1386 }
1387
1388 /*
1389 * relay_recv_metadata: receive the metada for the session.
1390 */
1391 static
1392 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1393 struct relay_command *cmd)
1394 {
1395 int ret = htobe32(LTTNG_OK);
1396 ssize_t size_ret;
1397 struct relay_session *session = cmd->session;
1398 struct lttcomm_relayd_metadata_payload *metadata_struct;
1399 struct relay_stream *metadata_stream;
1400 uint64_t data_size, payload_size;
1401
1402 if (!session) {
1403 ERR("Metadata sent before version check");
1404 ret = -1;
1405 goto end;
1406 }
1407
1408 data_size = payload_size = be64toh(recv_hdr->data_size);
1409 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1410 ERR("Incorrect data size");
1411 ret = -1;
1412 goto end;
1413 }
1414 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1415
1416 if (data_buffer_size < data_size) {
1417 /* In case the realloc fails, we can free the memory */
1418 char *tmp_data_ptr;
1419
1420 tmp_data_ptr = realloc(data_buffer, data_size);
1421 if (!tmp_data_ptr) {
1422 ERR("Allocating data buffer");
1423 free(data_buffer);
1424 ret = -1;
1425 goto end;
1426 }
1427 data_buffer = tmp_data_ptr;
1428 data_buffer_size = data_size;
1429 }
1430 memset(data_buffer, 0, data_size);
1431 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1432 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1433 if (ret < 0 || ret != data_size) {
1434 if (ret == 0) {
1435 /* Orderly shutdown. Not necessary to print an error. */
1436 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1437 } else {
1438 ERR("Relay didn't receive the whole metadata");
1439 }
1440 ret = -1;
1441 goto end;
1442 }
1443 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1444
1445 rcu_read_lock();
1446 metadata_stream = relay_stream_find_by_id(
1447 be64toh(metadata_struct->stream_id));
1448 if (!metadata_stream) {
1449 ret = -1;
1450 goto end_unlock;
1451 }
1452
1453 size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
1454 payload_size);
1455 if (size_ret < payload_size) {
1456 ERR("Relay error writing metadata on file");
1457 ret = -1;
1458 goto end_unlock;
1459 }
1460
1461 ret = write_padding_to_file(metadata_stream->fd,
1462 be32toh(metadata_struct->padding_size));
1463 if (ret < 0) {
1464 goto end_unlock;
1465 }
1466 metadata_stream->ctf_trace->metadata_received +=
1467 payload_size + be32toh(metadata_struct->padding_size);
1468
1469 DBG2("Relay metadata written");
1470
1471 end_unlock:
1472 rcu_read_unlock();
1473 end:
1474 return ret;
1475 }
1476
1477 /*
1478 * relay_send_version: send relayd version number
1479 */
1480 static
1481 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1482 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1483 {
1484 int ret;
1485 struct lttcomm_relayd_version reply, msg;
1486
1487 assert(cmd);
1488
1489 cmd->version_check_done = 1;
1490
1491 /* Get version from the other side. */
1492 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1493 if (ret < 0 || ret != sizeof(msg)) {
1494 if (ret == 0) {
1495 /* Orderly shutdown. Not necessary to print an error. */
1496 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1497 } else {
1498 ERR("Relay failed to receive the version values.");
1499 }
1500 ret = -1;
1501 goto end;
1502 }
1503
1504 reply.major = RELAYD_VERSION_COMM_MAJOR;
1505 reply.minor = RELAYD_VERSION_COMM_MINOR;
1506
1507 /* Major versions must be the same */
1508 if (reply.major != be32toh(msg.major)) {
1509 DBG("Incompatible major versions (%u vs %u), deleting session",
1510 reply.major, be32toh(msg.major));
1511 relay_delete_session(cmd, sessions_ht);
1512 ret = 0;
1513 goto end;
1514 }
1515
1516 cmd->major = reply.major;
1517 /* We adapt to the lowest compatible version */
1518 if (reply.minor <= be32toh(msg.minor)) {
1519 cmd->minor = reply.minor;
1520 } else {
1521 cmd->minor = be32toh(msg.minor);
1522 }
1523
1524 reply.major = htobe32(reply.major);
1525 reply.minor = htobe32(reply.minor);
1526 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1527 sizeof(struct lttcomm_relayd_version), 0);
1528 if (ret < 0) {
1529 ERR("Relay sending version");
1530 }
1531
1532 DBG("Version check done using protocol %u.%u", cmd->major,
1533 cmd->minor);
1534
1535 end:
1536 return ret;
1537 }
1538
1539 /*
1540 * Check for data pending for a given stream id from the session daemon.
1541 */
1542 static
1543 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1544 struct relay_command *cmd)
1545 {
1546 struct relay_session *session = cmd->session;
1547 struct lttcomm_relayd_data_pending msg;
1548 struct lttcomm_relayd_generic_reply reply;
1549 struct relay_stream *stream;
1550 int ret;
1551 uint64_t last_net_seq_num, stream_id;
1552
1553 DBG("Data pending command received");
1554
1555 if (!session || cmd->version_check_done == 0) {
1556 ERR("Trying to check for data before version check");
1557 ret = -1;
1558 goto end_no_session;
1559 }
1560
1561 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1562 if (ret < sizeof(msg)) {
1563 if (ret == 0) {
1564 /* Orderly shutdown. Not necessary to print an error. */
1565 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1566 } else {
1567 ERR("Relay didn't receive valid data_pending struct size : %d",
1568 ret);
1569 }
1570 ret = -1;
1571 goto end_no_session;
1572 }
1573
1574 stream_id = be64toh(msg.stream_id);
1575 last_net_seq_num = be64toh(msg.last_net_seq_num);
1576
1577 rcu_read_lock();
1578 stream = relay_stream_find_by_id(stream_id);
1579 if (stream == NULL) {
1580 ret = -1;
1581 goto end_unlock;
1582 }
1583
1584 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1585 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1586 last_net_seq_num);
1587
1588 /* Avoid wrapping issue */
1589 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1590 /* Data has in fact been written and is NOT pending */
1591 ret = 0;
1592 } else {
1593 /* Data still being streamed thus pending */
1594 ret = 1;
1595 }
1596
1597 /* Pending check is now done. */
1598 stream->data_pending_check_done = 1;
1599
1600 end_unlock:
1601 rcu_read_unlock();
1602
1603 reply.ret_code = htobe32(ret);
1604 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1605 if (ret < 0) {
1606 ERR("Relay data pending ret code failed");
1607 }
1608
1609 end_no_session:
1610 return ret;
1611 }
1612
1613 /*
1614 * Wait for the control socket to reach a quiescent state.
1615 *
1616 * Note that for now, when receiving this command from the session daemon, this
1617 * means that every subsequent commands or data received on the control socket
1618 * has been handled. So, this is why we simply return OK here.
1619 */
1620 static
1621 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1622 struct relay_command *cmd)
1623 {
1624 int ret;
1625 uint64_t stream_id;
1626 struct relay_stream *stream;
1627 struct lttng_ht_iter iter;
1628 struct lttcomm_relayd_quiescent_control msg;
1629 struct lttcomm_relayd_generic_reply reply;
1630
1631 DBG("Checking quiescent state on control socket");
1632
1633 if (!cmd->session || cmd->version_check_done == 0) {
1634 ERR("Trying to check for data before version check");
1635 ret = -1;
1636 goto end_no_session;
1637 }
1638
1639 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1640 if (ret < sizeof(msg)) {
1641 if (ret == 0) {
1642 /* Orderly shutdown. Not necessary to print an error. */
1643 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1644 } else {
1645 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1646 ret);
1647 }
1648 ret = -1;
1649 goto end_no_session;
1650 }
1651
1652 stream_id = be64toh(msg.stream_id);
1653
1654 rcu_read_lock();
1655 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1656 stream_n.node) {
1657 if (stream->stream_handle == stream_id) {
1658 stream->data_pending_check_done = 1;
1659 DBG("Relay quiescent control pending flag set to %" PRIu64,
1660 stream_id);
1661 break;
1662 }
1663 }
1664 rcu_read_unlock();
1665
1666 reply.ret_code = htobe32(LTTNG_OK);
1667 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1668 if (ret < 0) {
1669 ERR("Relay data quiescent control ret code failed");
1670 }
1671
1672 end_no_session:
1673 return ret;
1674 }
1675
1676 /*
1677 * Initialize a data pending command. This means that a client is about to ask
1678 * for data pending for each stream he/she holds. Simply iterate over all
1679 * streams of a session and set the data_pending_check_done flag.
1680 *
1681 * This command returns to the client a LTTNG_OK code.
1682 */
1683 static
1684 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1685 struct relay_command *cmd)
1686 {
1687 int ret;
1688 struct lttng_ht_iter iter;
1689 struct lttcomm_relayd_begin_data_pending msg;
1690 struct lttcomm_relayd_generic_reply reply;
1691 struct relay_stream *stream;
1692 uint64_t session_id;
1693
1694 assert(recv_hdr);
1695 assert(cmd);
1696
1697 DBG("Init streams for data pending");
1698
1699 if (!cmd->session || cmd->version_check_done == 0) {
1700 ERR("Trying to check for data before version check");
1701 ret = -1;
1702 goto end_no_session;
1703 }
1704
1705 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1706 if (ret < sizeof(msg)) {
1707 if (ret == 0) {
1708 /* Orderly shutdown. Not necessary to print an error. */
1709 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1710 } else {
1711 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1712 ret);
1713 }
1714 ret = -1;
1715 goto end_no_session;
1716 }
1717
1718 session_id = be64toh(msg.session_id);
1719
1720 /*
1721 * Iterate over all streams to set the begin data pending flag. For now, the
1722 * streams are indexed by stream handle so we have to iterate over all
1723 * streams to find the one associated with the right session_id.
1724 */
1725 rcu_read_lock();
1726 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1727 stream_n.node) {
1728 if (stream->session->id == session_id) {
1729 stream->data_pending_check_done = 0;
1730 DBG("Set begin data pending flag to stream %" PRIu64,
1731 stream->stream_handle);
1732 }
1733 }
1734 rcu_read_unlock();
1735
1736 /* All good, send back reply. */
1737 reply.ret_code = htobe32(LTTNG_OK);
1738
1739 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1740 if (ret < 0) {
1741 ERR("Relay begin data pending send reply failed");
1742 }
1743
1744 end_no_session:
1745 return ret;
1746 }
1747
1748 /*
1749 * End data pending command. This will check, for a given session id, if each
1750 * stream associated with it has its data_pending_check_done flag set. If not,
1751 * this means that the client lost track of the stream but the data is still
1752 * being streamed on our side. In this case, we inform the client that data is
1753 * inflight.
1754 *
1755 * Return to the client if there is data in flight or not with a ret_code.
1756 */
1757 static
1758 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1759 struct relay_command *cmd)
1760 {
1761 int ret;
1762 struct lttng_ht_iter iter;
1763 struct lttcomm_relayd_end_data_pending msg;
1764 struct lttcomm_relayd_generic_reply reply;
1765 struct relay_stream *stream;
1766 uint64_t session_id;
1767 uint32_t is_data_inflight = 0;
1768
1769 assert(recv_hdr);
1770 assert(cmd);
1771
1772 DBG("End data pending command");
1773
1774 if (!cmd->session || cmd->version_check_done == 0) {
1775 ERR("Trying to check for data before version check");
1776 ret = -1;
1777 goto end_no_session;
1778 }
1779
1780 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1781 if (ret < sizeof(msg)) {
1782 if (ret == 0) {
1783 /* Orderly shutdown. Not necessary to print an error. */
1784 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1785 } else {
1786 ERR("Relay didn't receive valid end data_pending struct size: %d",
1787 ret);
1788 }
1789 ret = -1;
1790 goto end_no_session;
1791 }
1792
1793 session_id = be64toh(msg.session_id);
1794
1795 /* Iterate over all streams to see if the begin data pending flag is set. */
1796 rcu_read_lock();
1797 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1798 stream_n.node) {
1799 if (stream->session->id == session_id &&
1800 !stream->data_pending_check_done) {
1801 is_data_inflight = 1;
1802 DBG("Data is still in flight for stream %" PRIu64,
1803 stream->stream_handle);
1804 break;
1805 }
1806 }
1807 rcu_read_unlock();
1808
1809 /* All good, send back reply. */
1810 reply.ret_code = htobe32(is_data_inflight);
1811
1812 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1813 if (ret < 0) {
1814 ERR("Relay end data pending send reply failed");
1815 }
1816
1817 end_no_session:
1818 return ret;
1819 }
1820
1821 /*
1822 * Receive an index for a specific stream.
1823 *
1824 * Return 0 on success else a negative value.
1825 */
1826 static
1827 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1828 struct relay_command *cmd)
1829 {
1830 int ret, send_ret, index_created = 0;
1831 struct relay_session *session = cmd->session;
1832 struct lttcomm_relayd_index index_info;
1833 struct relay_index *index, *wr_index = NULL;
1834 struct lttcomm_relayd_generic_reply reply;
1835 struct relay_stream *stream;
1836 uint64_t net_seq_num;
1837
1838 assert(cmd);
1839
1840 DBG("Relay receiving index");
1841
1842 if (!session || cmd->version_check_done == 0) {
1843 ERR("Trying to close a stream before version check");
1844 ret = -1;
1845 goto end_no_session;
1846 }
1847
1848 ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info,
1849 sizeof(index_info), 0);
1850 if (ret < sizeof(index_info)) {
1851 if (ret == 0) {
1852 /* Orderly shutdown. Not necessary to print an error. */
1853 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1854 } else {
1855 ERR("Relay didn't receive valid index struct size : %d", ret);
1856 }
1857 ret = -1;
1858 goto end_no_session;
1859 }
1860
1861 net_seq_num = be64toh(index_info.net_seq_num);
1862
1863 rcu_read_lock();
1864 stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id));
1865 if (!stream) {
1866 ret = -1;
1867 goto end_rcu_unlock;
1868 }
1869
1870 /* Live beacon handling */
1871 if (index_info.packet_size == 0) {
1872 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1873
1874 /*
1875 * Only flag a stream inactive when it has already received data.
1876 */
1877 if (stream->total_index_received > 0) {
1878 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1879 }
1880 ret = 0;
1881 goto end_rcu_unlock;
1882 } else {
1883 stream->beacon_ts_end = -1ULL;
1884 }
1885
1886 index = relay_index_find(stream->stream_handle, net_seq_num);
1887 if (!index) {
1888 /* A successful creation will add the object to the HT. */
1889 index = relay_index_create(stream->stream_handle, net_seq_num);
1890 if (!index) {
1891 goto end_rcu_unlock;
1892 }
1893 index_created = 1;
1894 }
1895
1896 copy_index_control_data(index, &index_info);
1897
1898 if (index_created) {
1899 /*
1900 * Try to add the relay index object to the hash table. If an object
1901 * already exist, destroy back the index created, set the data in this
1902 * object and write it on disk.
1903 */
1904 relay_index_add(index, &wr_index);
1905 if (wr_index) {
1906 copy_index_control_data(wr_index, &index_info);
1907 free(index);
1908 }
1909 } else {
1910 /* The index already exists so write it on disk. */
1911 wr_index = index;
1912 }
1913
1914 /* Do we have a writable ready index to write on disk. */
1915 if (wr_index) {
1916 /* Starting at 2.4, create the index file if none available. */
1917 if (cmd->minor >= 4 && stream->index_fd < 0) {
1918 ret = index_create_file(stream->path_name, stream->channel_name,
1919 relayd_uid, relayd_gid, stream->tracefile_size,
1920 stream->tracefile_count_current);
1921 if (ret < 0) {
1922 goto end_rcu_unlock;
1923 }
1924 stream->index_fd = ret;
1925 }
1926
1927 ret = relay_index_write(wr_index->fd, wr_index);
1928 if (ret < 0) {
1929 goto end_rcu_unlock;
1930 }
1931 stream->total_index_received++;
1932 }
1933
1934 end_rcu_unlock:
1935 rcu_read_unlock();
1936
1937 if (ret < 0) {
1938 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1939 } else {
1940 reply.ret_code = htobe32(LTTNG_OK);
1941 }
1942 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1943 if (send_ret < 0) {
1944 ERR("Relay sending close index id reply");
1945 ret = send_ret;
1946 }
1947
1948 end_no_session:
1949 return ret;
1950 }
1951
1952 /*
1953 * Receive the streams_sent message.
1954 *
1955 * Return 0 on success else a negative value.
1956 */
1957 static
1958 int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
1959 struct relay_command *cmd)
1960 {
1961 int ret, send_ret;
1962 struct lttcomm_relayd_generic_reply reply;
1963
1964 assert(cmd);
1965
1966 DBG("Relay receiving streams_sent");
1967
1968 if (!cmd->session || cmd->version_check_done == 0) {
1969 ERR("Trying to close a stream before version check");
1970 ret = -1;
1971 goto end_no_session;
1972 }
1973
1974 /*
1975 * Flag every pending stream in the connection recv list that they are
1976 * ready to be used by the viewer.
1977 */
1978 set_viewer_ready_flag(cmd);
1979
1980 reply.ret_code = htobe32(LTTNG_OK);
1981 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1982 if (send_ret < 0) {
1983 ERR("Relay sending sent_stream reply");
1984 ret = send_ret;
1985 } else {
1986 /* Success. */
1987 ret = 0;
1988 }
1989
1990 end_no_session:
1991 return ret;
1992 }
1993
1994 /*
1995 * Process the commands received on the control socket
1996 */
1997 static
1998 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1999 struct relay_command *cmd, struct relay_local_data *ctx)
2000 {
2001 int ret = 0;
2002
2003 switch (be32toh(recv_hdr->cmd)) {
2004 case RELAYD_CREATE_SESSION:
2005 ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht);
2006 break;
2007 case RELAYD_ADD_STREAM:
2008 ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht);
2009 break;
2010 case RELAYD_START_DATA:
2011 ret = relay_start(recv_hdr, cmd);
2012 break;
2013 case RELAYD_SEND_METADATA:
2014 ret = relay_recv_metadata(recv_hdr, cmd);
2015 break;
2016 case RELAYD_VERSION:
2017 ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht);
2018 break;
2019 case RELAYD_CLOSE_STREAM:
2020 ret = relay_close_stream(recv_hdr, cmd);
2021 break;
2022 case RELAYD_DATA_PENDING:
2023 ret = relay_data_pending(recv_hdr, cmd);
2024 break;
2025 case RELAYD_QUIESCENT_CONTROL:
2026 ret = relay_quiescent_control(recv_hdr, cmd);
2027 break;
2028 case RELAYD_BEGIN_DATA_PENDING:
2029 ret = relay_begin_data_pending(recv_hdr, cmd);
2030 break;
2031 case RELAYD_END_DATA_PENDING:
2032 ret = relay_end_data_pending(recv_hdr, cmd);
2033 break;
2034 case RELAYD_SEND_INDEX:
2035 ret = relay_recv_index(recv_hdr, cmd);
2036 break;
2037 case RELAYD_STREAMS_SENT:
2038 ret = relay_streams_sent(recv_hdr, cmd);
2039 break;
2040 case RELAYD_UPDATE_SYNC_INFO:
2041 default:
2042 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
2043 relay_unknown_command(cmd);
2044 ret = -1;
2045 goto end;
2046 }
2047
2048 end:
2049 return ret;
2050 }
2051
2052 /*
2053 * Handle index for a data stream.
2054 *
2055 * RCU read side lock MUST be acquired.
2056 *
2057 * Return 0 on success else a negative value.
2058 */
2059 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2060 int rotate_index)
2061 {
2062 int ret = 0, index_created = 0;
2063 uint64_t stream_id, data_offset;
2064 struct relay_index *index, *wr_index = NULL;
2065
2066 assert(stream);
2067
2068 stream_id = stream->stream_handle;
2069 /* Get data offset because we are about to update the index. */
2070 data_offset = htobe64(stream->tracefile_size_current);
2071
2072 /*
2073 * Lookup for an existing index for that stream id/sequence number. If on
2074 * exists, the control thread already received the data for it thus we need
2075 * to write it on disk.
2076 */
2077 index = relay_index_find(stream_id, net_seq_num);
2078 if (!index) {
2079 /* A successful creation will add the object to the HT. */
2080 index = relay_index_create(stream_id, net_seq_num);
2081 if (!index) {
2082 ret = -1;
2083 goto error;
2084 }
2085 index_created = 1;
2086 }
2087
2088 if (rotate_index || stream->index_fd < 0) {
2089 index->to_close_fd = stream->index_fd;
2090 ret = index_create_file(stream->path_name, stream->channel_name,
2091 relayd_uid, relayd_gid, stream->tracefile_size,
2092 stream->tracefile_count_current);
2093 if (ret < 0) {
2094 /* This will close the stream's index fd if one. */
2095 relay_index_free_safe(index);
2096 goto error;
2097 }
2098 stream->index_fd = ret;
2099 }
2100 index->fd = stream->index_fd;
2101 index->index_data.offset = data_offset;
2102
2103 if (index_created) {
2104 /*
2105 * Try to add the relay index object to the hash table. If an object
2106 * already exist, destroy back the index created and set the data.
2107 */
2108 relay_index_add(index, &wr_index);
2109 if (wr_index) {
2110 /* Copy back data from the created index. */
2111 wr_index->fd = index->fd;
2112 wr_index->to_close_fd = index->to_close_fd;
2113 wr_index->index_data.offset = data_offset;
2114 free(index);
2115 }
2116 } else {
2117 /* The index already exists so write it on disk. */
2118 wr_index = index;
2119 }
2120
2121 /* Do we have a writable ready index to write on disk. */
2122 if (wr_index) {
2123 ret = relay_index_write(wr_index->fd, wr_index);
2124 if (ret < 0) {
2125 goto error;
2126 }
2127 stream->total_index_received++;
2128 }
2129
2130 error:
2131 return ret;
2132 }
2133
2134 /*
2135 * relay_process_data: Process the data received on the data socket
2136 */
2137 static
2138 int relay_process_data(struct relay_command *cmd)
2139 {
2140 int ret = 0, rotate_index = 0;
2141 ssize_t size_ret;
2142 struct relay_stream *stream;
2143 struct lttcomm_relayd_data_hdr data_hdr;
2144 uint64_t stream_id;
2145 uint64_t net_seq_num;
2146 uint32_t data_size;
2147
2148 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
2149 sizeof(struct lttcomm_relayd_data_hdr), 0);
2150 if (ret <= 0) {
2151 if (ret == 0) {
2152 /* Orderly shutdown. Not necessary to print an error. */
2153 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2154 } else {
2155 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
2156 }
2157 ret = -1;
2158 goto end;
2159 }
2160
2161 stream_id = be64toh(data_hdr.stream_id);
2162
2163 rcu_read_lock();
2164 stream = relay_stream_find_by_id(stream_id);
2165 if (!stream) {
2166 ret = -1;
2167 goto end_rcu_unlock;
2168 }
2169
2170 data_size = be32toh(data_hdr.data_size);
2171 if (data_buffer_size < data_size) {
2172 char *tmp_data_ptr;
2173
2174 tmp_data_ptr = realloc(data_buffer, data_size);
2175 if (!tmp_data_ptr) {
2176 ERR("Allocating data buffer");
2177 free(data_buffer);
2178 ret = -1;
2179 goto end_rcu_unlock;
2180 }
2181 data_buffer = tmp_data_ptr;
2182 data_buffer_size = data_size;
2183 }
2184 memset(data_buffer, 0, data_size);
2185
2186 net_seq_num = be64toh(data_hdr.net_seq_num);
2187
2188 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2189 data_size, stream_id, net_seq_num);
2190 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
2191 if (ret <= 0) {
2192 if (ret == 0) {
2193 /* Orderly shutdown. Not necessary to print an error. */
2194 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2195 }
2196 ret = -1;
2197 goto end_rcu_unlock;
2198 }
2199
2200 /* Check if a rotation is needed. */
2201 if (stream->tracefile_size > 0 &&
2202 (stream->tracefile_size_current + data_size) >
2203 stream->tracefile_size) {
2204 struct relay_viewer_stream *vstream;
2205 uint64_t new_id;
2206
2207 new_id = (stream->tracefile_count_current + 1) %
2208 stream->tracefile_count;
2209 /*
2210 * When we wrap-around back to 0, we start overwriting old
2211 * trace data.
2212 */
2213 if (!stream->tracefile_overwrite && new_id == 0) {
2214 stream->tracefile_overwrite = 1;
2215 }
2216 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
2217 if (stream->tracefile_overwrite) {
2218 stream->oldest_tracefile_id =
2219 (stream->oldest_tracefile_id + 1) %
2220 stream->tracefile_count;
2221 }
2222 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
2223 if (vstream) {
2224 /*
2225 * The viewer is reading a file about to be
2226 * overwritten. Close the FDs it is
2227 * currently using and let it handle the fault.
2228 */
2229 if (vstream->tracefile_count_current == new_id) {
2230 pthread_mutex_lock(&vstream->overwrite_lock);
2231 vstream->abort_flag = 1;
2232 pthread_mutex_unlock(&vstream->overwrite_lock);
2233 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2234 stream->channel_name, new_id);
2235 } else if (vstream->tracefile_count_current ==
2236 stream->tracefile_count_current) {
2237 /*
2238 * The reader and writer were in the
2239 * same trace file, inform the viewer
2240 * that no new index will ever be added
2241 * to this file.
2242 */
2243 vstream->close_write_flag = 1;
2244 }
2245 }
2246 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2247 stream->tracefile_size, stream->tracefile_count,
2248 relayd_uid, relayd_gid, stream->fd,
2249 &(stream->tracefile_count_current), &stream->fd);
2250 stream->total_index_received = 0;
2251 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
2252 if (ret < 0) {
2253 ERR("Rotating stream output file");
2254 goto end_rcu_unlock;
2255 }
2256 /* Reset current size because we just perform a stream rotation. */
2257 stream->tracefile_size_current = 0;
2258 rotate_index = 1;
2259 }
2260
2261 /*
2262 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2263 * index are NOT supported.
2264 */
2265 if (stream->session->minor >= 4 && !stream->session->snapshot) {
2266 ret = handle_index_data(stream, net_seq_num, rotate_index);
2267 if (ret < 0) {
2268 goto end_rcu_unlock;
2269 }
2270 }
2271
2272 /* Write data to stream output fd. */
2273 size_ret = lttng_write(stream->fd, data_buffer, data_size);
2274 if (size_ret < data_size) {
2275 ERR("Relay error writing data to file");
2276 ret = -1;
2277 goto end_rcu_unlock;
2278 }
2279
2280 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2281 ret, stream->stream_handle);
2282
2283 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2284 if (ret < 0) {
2285 goto end_rcu_unlock;
2286 }
2287 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2288
2289 stream->prev_seq = net_seq_num;
2290
2291 /* Check if we need to close the FD */
2292 if (close_stream_check(stream)) {
2293 destroy_stream(stream);
2294 }
2295
2296 end_rcu_unlock:
2297 rcu_read_unlock();
2298 end:
2299 return ret;
2300 }
2301
2302 static
2303 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
2304 {
2305 int ret;
2306
2307 lttng_poll_del(events, pollfd);
2308
2309 ret = close(pollfd);
2310 if (ret < 0) {
2311 ERR("Closing pollfd %d", pollfd);
2312 }
2313 }
2314
2315 static
2316 int relay_add_connection(int fd, struct lttng_poll_event *events,
2317 struct lttng_ht *relay_connections_ht)
2318 {
2319 struct relay_command *relay_connection;
2320 ssize_t ret;
2321
2322 relay_connection = zmalloc(sizeof(struct relay_command));
2323 if (relay_connection == NULL) {
2324 PERROR("Relay command zmalloc");
2325 goto error;
2326 }
2327 ret = lttng_read(fd, relay_connection, sizeof(struct relay_command));
2328 if (ret < sizeof(struct relay_command)) {
2329 PERROR("read relay cmd pipe");
2330 goto error_read;
2331 }
2332 CDS_INIT_LIST_HEAD(&relay_connection->recv_head);
2333
2334 /*
2335 * Only used by the control side and the reference is copied inside each
2336 * stream from that connection. Thus a destroy HT must be done after every
2337 * stream has been destroyed.
2338 */
2339 if (relay_connection->type == RELAY_CONTROL) {
2340 relay_connection->ctf_traces_ht = lttng_ht_new(0,
2341 LTTNG_HT_TYPE_STRING);
2342 if (!relay_connection->ctf_traces_ht) {
2343 goto error_read;
2344 }
2345 }
2346
2347 lttng_ht_node_init_ulong(&relay_connection->sock_n,
2348 (unsigned long) relay_connection->sock->fd);
2349 rcu_read_lock();
2350 lttng_ht_add_unique_ulong(relay_connections_ht,
2351 &relay_connection->sock_n);
2352 rcu_read_unlock();
2353 return lttng_poll_add(events,
2354 relay_connection->sock->fd,
2355 LPOLLIN | LPOLLRDHUP);
2356
2357 error_read:
2358 free(relay_connection);
2359 error:
2360 return -1;
2361 }
2362
2363 static
2364 void deferred_free_connection(struct rcu_head *head)
2365 {
2366 struct relay_command *relay_connection =
2367 caa_container_of(head, struct relay_command, rcu_node);
2368
2369 lttcomm_destroy_sock(relay_connection->sock);
2370 free(relay_connection);
2371 }
2372
2373 static
2374 void relay_del_connection(struct lttng_ht *relay_connections_ht,
2375 struct lttng_ht_iter *iter, struct relay_command *relay_connection,
2376 struct lttng_ht *sessions_ht)
2377 {
2378 int ret;
2379
2380 ret = lttng_ht_del(relay_connections_ht, iter);
2381 assert(!ret);
2382
2383 if (relay_connection->type == RELAY_CONTROL) {
2384 struct relay_stream_recv_handle *node, *tmp_node;
2385
2386 relay_delete_session(relay_connection, sessions_ht);
2387 lttng_ht_destroy(relay_connection->ctf_traces_ht);
2388
2389 /* Clean up recv list. */
2390 cds_list_for_each_entry_safe(node, tmp_node,
2391 &relay_connection->recv_head, node) {
2392 cds_list_del(&node->node);
2393 free(node);
2394 }
2395 }
2396
2397 call_rcu(&relay_connection->rcu_node, deferred_free_connection);
2398 }
2399
2400 /*
2401 * This thread does the actual work
2402 */
2403 static
2404 void *relay_thread_worker(void *data)
2405 {
2406 int ret, err = -1, last_seen_data_fd = -1;
2407 uint32_t nb_fd;
2408 struct relay_command *relay_connection;
2409 struct lttng_poll_event events;
2410 struct lttng_ht *relay_connections_ht;
2411 struct lttng_ht_node_ulong *node;
2412 struct lttng_ht_iter iter;
2413 struct lttcomm_relayd_hdr recv_hdr;
2414 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2415 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2416
2417 DBG("[thread] Relay worker started");
2418
2419 rcu_register_thread();
2420
2421 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2422
2423 health_code_update();
2424
2425 /* table of connections indexed on socket */
2426 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2427 if (!relay_connections_ht) {
2428 goto relay_connections_ht_error;
2429 }
2430
2431 /* Tables of received indexes indexed by index handle and net_seq_num. */
2432 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2433 if (!indexes_ht) {
2434 goto indexes_ht_error;
2435 }
2436
2437 ret = create_thread_poll_set(&events, 2);
2438 if (ret < 0) {
2439 goto error_poll_create;
2440 }
2441
2442 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
2443 if (ret < 0) {
2444 goto error;
2445 }
2446
2447 restart:
2448 while (1) {
2449 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2450
2451 health_code_update();
2452
2453 /* Infinite blocking call, waiting for transmission */
2454 DBG3("Relayd worker thread polling...");
2455 health_poll_entry();
2456 ret = lttng_poll_wait(&events, -1);
2457 health_poll_exit();
2458 if (ret < 0) {
2459 /*
2460 * Restart interrupted system call.
2461 */
2462 if (errno == EINTR) {
2463 goto restart;
2464 }
2465 goto error;
2466 }
2467
2468 nb_fd = ret;
2469
2470 /*
2471 * Process control. The control connection is prioritised so we don't
2472 * starve it with high throughout put tracing data on the data
2473 * connection.
2474 */
2475 for (i = 0; i < nb_fd; i++) {
2476 /* Fetch once the poll data */
2477 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2478 int pollfd = LTTNG_POLL_GETFD(&events, i);
2479
2480 health_code_update();
2481
2482 /* Thread quit pipe has been closed. Killing thread. */
2483 ret = check_thread_quit_pipe(pollfd, revents);
2484 if (ret) {
2485 err = 0;
2486 goto exit;
2487 }
2488
2489 /* Inspect the relay cmd pipe for new connection */
2490 if (pollfd == relay_cmd_pipe[0]) {
2491 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2492 ERR("Relay pipe error");
2493 goto error;
2494 } else if (revents & LPOLLIN) {
2495 DBG("Relay command received");
2496 ret = relay_add_connection(relay_cmd_pipe[0],
2497 &events, relay_connections_ht);
2498 if (ret < 0) {
2499 goto error;
2500 }
2501 }
2502 } else if (revents) {
2503 rcu_read_lock();
2504 lttng_ht_lookup(relay_connections_ht,
2505 (void *)((unsigned long) pollfd),
2506 &iter);
2507 node = lttng_ht_iter_get_node_ulong(&iter);
2508 if (node == NULL) {
2509 DBG2("Relay sock %d not found", pollfd);
2510 rcu_read_unlock();
2511 goto error;
2512 }
2513 relay_connection = caa_container_of(node,
2514 struct relay_command, sock_n);
2515
2516 if (revents & (LPOLLERR)) {
2517 ERR("POLL ERROR");
2518 relay_cleanup_poll_connection(&events, pollfd);
2519 relay_del_connection(relay_connections_ht,
2520 &iter, relay_connection, sessions_ht);
2521 if (last_seen_data_fd == pollfd) {
2522 last_seen_data_fd = last_notdel_data_fd;
2523 }
2524 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2525 DBG("Socket %d hung up", pollfd);
2526 relay_cleanup_poll_connection(&events, pollfd);
2527 relay_del_connection(relay_connections_ht,
2528 &iter, relay_connection, sessions_ht);
2529 if (last_seen_data_fd == pollfd) {
2530 last_seen_data_fd = last_notdel_data_fd;
2531 }
2532 } else if (revents & LPOLLIN) {
2533 /* control socket */
2534 if (relay_connection->type == RELAY_CONTROL) {
2535 ret = relay_connection->sock->ops->recvmsg(
2536 relay_connection->sock, &recv_hdr,
2537 sizeof(struct lttcomm_relayd_hdr), 0);
2538 /* connection closed */
2539 if (ret <= 0) {
2540 relay_cleanup_poll_connection(&events, pollfd);
2541 relay_del_connection(relay_connections_ht,
2542 &iter, relay_connection, sessions_ht);
2543 DBG("Control connection closed with %d", pollfd);
2544 } else {
2545 if (relay_connection->session) {
2546 DBG2("Relay worker receiving data for session : %" PRIu64,
2547 relay_connection->session->id);
2548 }
2549 ret = relay_process_control(&recv_hdr,
2550 relay_connection, relay_ctx);
2551 if (ret < 0) {
2552 /* Clear the session on error. */
2553 relay_cleanup_poll_connection(&events, pollfd);
2554 relay_del_connection(relay_connections_ht,
2555 &iter, relay_connection, sessions_ht);
2556 DBG("Connection closed with %d", pollfd);
2557 }
2558 seen_control = 1;
2559 }
2560 } else {
2561 /*
2562 * Flag the last seen data fd not deleted. It will be
2563 * used as the last seen fd if any fd gets deleted in
2564 * this first loop.
2565 */
2566 last_notdel_data_fd = pollfd;
2567 }
2568 }
2569 rcu_read_unlock();
2570 }
2571 }
2572
2573 /*
2574 * The last loop handled a control request, go back to poll to make
2575 * sure we prioritise the control socket.
2576 */
2577 if (seen_control) {
2578 continue;
2579 }
2580
2581 if (last_seen_data_fd >= 0) {
2582 for (i = 0; i < nb_fd; i++) {
2583 int pollfd = LTTNG_POLL_GETFD(&events, i);
2584
2585 health_code_update();
2586
2587 if (last_seen_data_fd == pollfd) {
2588 idx = i;
2589 break;
2590 }
2591 }
2592 }
2593
2594 /* Process data connection. */
2595 for (i = idx + 1; i < nb_fd; i++) {
2596 /* Fetch the poll data. */
2597 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2598 int pollfd = LTTNG_POLL_GETFD(&events, i);
2599
2600 health_code_update();
2601
2602 /* Skip the command pipe. It's handled in the first loop. */
2603 if (pollfd == relay_cmd_pipe[0]) {
2604 continue;
2605 }
2606
2607 if (revents) {
2608 rcu_read_lock();
2609 lttng_ht_lookup(relay_connections_ht,
2610 (void *)((unsigned long) pollfd),
2611 &iter);
2612 node = lttng_ht_iter_get_node_ulong(&iter);
2613 if (node == NULL) {
2614 /* Skip it. Might be removed before. */
2615 rcu_read_unlock();
2616 continue;
2617 }
2618 relay_connection = caa_container_of(node,
2619 struct relay_command, sock_n);
2620
2621 if (revents & LPOLLIN) {
2622 if (relay_connection->type != RELAY_DATA) {
2623 continue;
2624 }
2625
2626 ret = relay_process_data(relay_connection);
2627 /* connection closed */
2628 if (ret < 0) {
2629 relay_cleanup_poll_connection(&events, pollfd);
2630 relay_del_connection(relay_connections_ht,
2631 &iter, relay_connection, sessions_ht);
2632 DBG("Data connection closed with %d", pollfd);
2633 /*
2634 * Every goto restart call sets the last seen fd where
2635 * here we don't really care since we gracefully
2636 * continue the loop after the connection is deleted.
2637 */
2638 } else {
2639 /* Keep last seen port. */
2640 last_seen_data_fd = pollfd;
2641 rcu_read_unlock();
2642 goto restart;
2643 }
2644 }
2645 rcu_read_unlock();
2646 }
2647 }
2648 last_seen_data_fd = -1;
2649 }
2650
2651 /* Normal exit, no error */
2652 ret = 0;
2653
2654 exit:
2655 error:
2656 lttng_poll_clean(&events);
2657
2658 /* empty the hash table and free the memory */
2659 rcu_read_lock();
2660 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2661 health_code_update();
2662
2663 node = lttng_ht_iter_get_node_ulong(&iter);
2664 if (node) {
2665 relay_connection = caa_container_of(node,
2666 struct relay_command, sock_n);
2667 relay_del_connection(relay_connections_ht,
2668 &iter, relay_connection, sessions_ht);
2669 }
2670 }
2671 rcu_read_unlock();
2672 error_poll_create:
2673 lttng_ht_destroy(indexes_ht);
2674 indexes_ht_error:
2675 lttng_ht_destroy(relay_connections_ht);
2676 relay_connections_ht_error:
2677 /* Close relay cmd pipes */
2678 utils_close_pipe(relay_cmd_pipe);
2679 if (err) {
2680 DBG("Thread exited with error");
2681 }
2682 DBG("Worker thread cleanup complete");
2683 free(data_buffer);
2684 if (err) {
2685 health_error();
2686 ERR("Health error occurred in %s", __func__);
2687 }
2688 health_unregister(health_relayd);
2689 rcu_unregister_thread();
2690 stop_threads();
2691 return NULL;
2692 }
2693
2694 /*
2695 * Create the relay command pipe to wake thread_manage_apps.
2696 * Closed in cleanup().
2697 */
2698 static int create_relay_cmd_pipe(void)
2699 {
2700 int ret;
2701
2702 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2703
2704 return ret;
2705 }
2706
2707 /*
2708 * main
2709 */
2710 int main(int argc, char **argv)
2711 {
2712 int ret = 0;
2713 void *status;
2714 struct relay_local_data *relay_ctx;
2715
2716 /* Parse arguments */
2717 progname = argv[0];
2718 if ((ret = parse_args(argc, argv)) < 0) {
2719 goto exit;
2720 }
2721
2722 if ((ret = set_signal_handler()) < 0) {
2723 goto exit;
2724 }
2725
2726 /* Try to create directory if -o, --output is specified. */
2727 if (opt_output_path) {
2728 if (*opt_output_path != '/') {
2729 ERR("Please specify an absolute path for -o, --output PATH");
2730 goto exit;
2731 }
2732
2733 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2734 if (ret < 0) {
2735 ERR("Unable to create %s", opt_output_path);
2736 goto exit;
2737 }
2738 }
2739
2740 /* Daemonize */
2741 if (opt_daemon || opt_background) {
2742 int i;
2743
2744 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2745 !opt_background);
2746 if (ret < 0) {
2747 goto exit;
2748 }
2749
2750 /*
2751 * We are in the child. Make sure all other file
2752 * descriptors are closed, in case we are called with
2753 * more opened file descriptors than the standard ones.
2754 */
2755 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2756 (void) close(i);
2757 }
2758 }
2759
2760 /* Create thread quit pipe */
2761 if ((ret = init_thread_quit_pipe()) < 0) {
2762 goto error;
2763 }
2764
2765 /* We need those values for the file/dir creation. */
2766 relayd_uid = getuid();
2767 relayd_gid = getgid();
2768
2769 /* Check if daemon is UID = 0 */
2770 if (relayd_uid == 0) {
2771 if (control_uri->port < 1024 || data_uri->port < 1024 ||
2772 live_uri->port < 1024) {
2773 ERR("Need to be root to use ports < 1024");
2774 ret = -1;
2775 goto exit;
2776 }
2777 }
2778
2779 /* Setup the thread apps communication pipe. */
2780 if ((ret = create_relay_cmd_pipe()) < 0) {
2781 goto exit;
2782 }
2783
2784 /* Init relay command queue. */
2785 cds_wfq_init(&relay_cmd_queue.queue);
2786
2787 /* Set up max poll set size */
2788 lttng_poll_set_max_size();
2789
2790 /* Initialize communication library */
2791 lttcomm_init();
2792 lttcomm_inet_init();
2793
2794 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2795 if (!relay_ctx) {
2796 PERROR("relay_ctx");
2797 goto exit;
2798 }
2799
2800 /* tables of sessions indexed by session ID */
2801 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2802 if (!relay_ctx->sessions_ht) {
2803 goto exit_relay_ctx_sessions;
2804 }
2805
2806 /* tables of streams indexed by stream ID */
2807 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2808 if (!relay_streams_ht) {
2809 goto exit_relay_ctx_streams;
2810 }
2811
2812 /* tables of streams indexed by stream ID */
2813 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2814 if (!viewer_streams_ht) {
2815 goto exit_relay_ctx_viewer_streams;
2816 }
2817
2818 /* Initialize thread health monitoring */
2819 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2820 if (!health_relayd) {
2821 PERROR("health_app_create error");
2822 goto exit_health_app_create;
2823 }
2824
2825 ret = utils_create_pipe(health_quit_pipe);
2826 if (ret < 0) {
2827 goto error_health_pipe;
2828 }
2829
2830 /* Create thread to manage the client socket */
2831 ret = pthread_create(&health_thread, NULL,
2832 thread_manage_health, (void *) NULL);
2833 if (ret != 0) {
2834 PERROR("pthread_create health");
2835 goto health_error;
2836 }
2837
2838 /* Setup the dispatcher thread */
2839 ret = pthread_create(&dispatcher_thread, NULL,
2840 relay_thread_dispatcher, (void *) NULL);
2841 if (ret != 0) {
2842 PERROR("pthread_create dispatcher");
2843 goto exit_dispatcher;
2844 }
2845
2846 /* Setup the worker thread */
2847 ret = pthread_create(&worker_thread, NULL,
2848 relay_thread_worker, (void *) relay_ctx);
2849 if (ret != 0) {
2850 PERROR("pthread_create worker");
2851 goto exit_worker;
2852 }
2853
2854 /* Setup the listener thread */
2855 ret = pthread_create(&listener_thread, NULL,
2856 relay_thread_listener, (void *) NULL);
2857 if (ret != 0) {
2858 PERROR("pthread_create listener");
2859 goto exit_listener;
2860 }
2861
2862 ret = live_start_threads(live_uri, relay_ctx);
2863 if (ret != 0) {
2864 ERR("Starting live viewer threads");
2865 goto exit_live;
2866 }
2867
2868 exit_live:
2869 ret = pthread_join(listener_thread, &status);
2870 if (ret != 0) {
2871 PERROR("pthread_join");
2872 goto error; /* join error, exit without cleanup */
2873 }
2874
2875 exit_listener:
2876 ret = pthread_join(worker_thread, &status);
2877 if (ret != 0) {
2878 PERROR("pthread_join");
2879 goto error; /* join error, exit without cleanup */
2880 }
2881
2882 exit_worker:
2883 ret = pthread_join(dispatcher_thread, &status);
2884 if (ret != 0) {
2885 PERROR("pthread_join");
2886 goto error; /* join error, exit without cleanup */
2887 }
2888
2889 exit_dispatcher:
2890 ret = pthread_join(health_thread, &status);
2891 if (ret != 0) {
2892 PERROR("pthread_join health thread");
2893 goto error; /* join error, exit without cleanup */
2894 }
2895
2896 /*
2897 * Stop live threads only after joining other threads.
2898 */
2899 live_stop_threads();
2900
2901 health_error:
2902 utils_close_pipe(health_quit_pipe);
2903
2904 error_health_pipe:
2905 health_app_destroy(health_relayd);
2906
2907 exit_health_app_create:
2908 lttng_ht_destroy(viewer_streams_ht);
2909
2910 exit_relay_ctx_viewer_streams:
2911 lttng_ht_destroy(relay_streams_ht);
2912
2913 exit_relay_ctx_streams:
2914 lttng_ht_destroy(relay_ctx->sessions_ht);
2915
2916 exit_relay_ctx_sessions:
2917 free(relay_ctx);
2918
2919 exit:
2920 cleanup();
2921 if (!ret) {
2922 exit(EXIT_SUCCESS);
2923 }
2924
2925 error:
2926 exit(EXIT_FAILURE);
2927 }
This page took 0.177642 seconds and 3 git commands to generate.