Fix: join live threads after joining others
[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, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
136 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
137 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
138 }
139
140 static
141 int parse_args(int argc, char **argv)
142 {
143 int c;
144 int ret = 0;
145 char *default_address;
146
147 static struct option long_options[] = {
148 { "control-port", 1, 0, 'C', },
149 { "data-port", 1, 0, 'D', },
150 { "daemonize", 0, 0, 'd', },
151 { "group", 1, 0, 'g', },
152 { "help", 0, 0, 'h', },
153 { "output", 1, 0, 'o', },
154 { "verbose", 0, 0, 'v', },
155 { NULL, 0, 0, 0, },
156 };
157
158 while (1) {
159 int option_index = 0;
160 c = getopt_long(argc, argv, "dhv" "C:D:o:g:",
161 long_options, &option_index);
162 if (c == -1) {
163 break;
164 }
165
166 switch (c) {
167 case 0:
168 fprintf(stderr, "option %s", long_options[option_index].name);
169 if (optarg) {
170 fprintf(stderr, " with arg %s\n", optarg);
171 }
172 break;
173 case 'C':
174 ret = uri_parse(optarg, &control_uri);
175 if (ret < 0) {
176 ERR("Invalid control URI specified");
177 goto exit;
178 }
179 if (control_uri->port == 0) {
180 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
181 }
182 break;
183 case 'D':
184 ret = uri_parse(optarg, &data_uri);
185 if (ret < 0) {
186 ERR("Invalid data URI specified");
187 goto exit;
188 }
189 if (data_uri->port == 0) {
190 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
191 }
192 break;
193 case 'd':
194 opt_daemon = 1;
195 break;
196 case 'g':
197 tracing_group_name = optarg;
198 break;
199 case 'h':
200 usage();
201 exit(EXIT_FAILURE);
202 case 'o':
203 ret = asprintf(&opt_output_path, "%s", optarg);
204 if (ret < 0) {
205 PERROR("asprintf opt_output_path");
206 goto exit;
207 }
208 break;
209 case 'v':
210 /* Verbose level can increase using multiple -v */
211 lttng_opt_verbose += 1;
212 break;
213 default:
214 /* Unknown option or other error.
215 * Error is printed by getopt, just return */
216 ret = -1;
217 goto exit;
218 }
219 }
220
221 /* assign default values */
222 if (control_uri == NULL) {
223 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
224 DEFAULT_NETWORK_CONTROL_PORT);
225 if (ret < 0) {
226 PERROR("asprintf default data address");
227 goto exit;
228 }
229
230 ret = uri_parse(default_address, &control_uri);
231 free(default_address);
232 if (ret < 0) {
233 ERR("Invalid control URI specified");
234 goto exit;
235 }
236 }
237 if (data_uri == NULL) {
238 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
239 DEFAULT_NETWORK_DATA_PORT);
240 if (ret < 0) {
241 PERROR("asprintf default data address");
242 goto exit;
243 }
244
245 ret = uri_parse(default_address, &data_uri);
246 free(default_address);
247 if (ret < 0) {
248 ERR("Invalid data URI specified");
249 goto exit;
250 }
251 }
252 if (live_uri == NULL) {
253 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
254 DEFAULT_NETWORK_VIEWER_PORT);
255 if (ret < 0) {
256 PERROR("asprintf default viewer control address");
257 goto exit;
258 }
259
260 ret = uri_parse(default_address, &live_uri);
261 free(default_address);
262 if (ret < 0) {
263 ERR("Invalid viewer control URI specified");
264 goto exit;
265 }
266 }
267
268 exit:
269 return ret;
270 }
271
272 /*
273 * Cleanup the daemon
274 */
275 static
276 void cleanup(void)
277 {
278 DBG("Cleaning up");
279
280 /* free the dynamically allocated opt_output_path */
281 free(opt_output_path);
282
283 /* Close thread quit pipes */
284 utils_close_pipe(thread_quit_pipe);
285
286 uri_free(control_uri);
287 uri_free(data_uri);
288 }
289
290 /*
291 * Write to writable pipe used to notify a thread.
292 */
293 static
294 int notify_thread_pipe(int wpipe)
295 {
296 int ret;
297
298 do {
299 ret = write(wpipe, "!", 1);
300 } while (ret < 0 && errno == EINTR);
301 if (ret < 0 || ret != 1) {
302 PERROR("write poll pipe");
303 }
304
305 return ret;
306 }
307
308 static void notify_health_quit_pipe(int *pipe)
309 {
310 int ret;
311
312 do {
313 ret = write(pipe[1], "4", 1);
314 } while (ret < 0 && errno == EINTR);
315 if (ret < 0 || ret != 1) {
316 PERROR("write relay health quit");
317 }
318 }
319
320 /*
321 * Stop all threads by closing the thread quit pipe.
322 */
323 static
324 void stop_threads(void)
325 {
326 int ret;
327
328 /* Stopping all threads */
329 DBG("Terminating all threads");
330 ret = notify_thread_pipe(thread_quit_pipe[1]);
331 if (ret < 0) {
332 ERR("write error on thread quit pipe");
333 }
334
335 notify_health_quit_pipe(health_quit_pipe);
336
337 /* Dispatch thread */
338 CMM_STORE_SHARED(dispatch_thread_exit, 1);
339 futex_nto1_wake(&relay_cmd_queue.futex);
340 }
341
342 /*
343 * Signal handler for the daemon
344 *
345 * Simply stop all worker threads, leaving main() return gracefully after
346 * joining all threads and calling cleanup().
347 */
348 static
349 void sighandler(int sig)
350 {
351 switch (sig) {
352 case SIGPIPE:
353 DBG("SIGPIPE caught");
354 return;
355 case SIGINT:
356 DBG("SIGINT caught");
357 stop_threads();
358 break;
359 case SIGTERM:
360 DBG("SIGTERM caught");
361 stop_threads();
362 break;
363 default:
364 break;
365 }
366 }
367
368 /*
369 * Setup signal handler for :
370 * SIGINT, SIGTERM, SIGPIPE
371 */
372 static
373 int set_signal_handler(void)
374 {
375 int ret = 0;
376 struct sigaction sa;
377 sigset_t sigset;
378
379 if ((ret = sigemptyset(&sigset)) < 0) {
380 PERROR("sigemptyset");
381 return ret;
382 }
383
384 sa.sa_handler = sighandler;
385 sa.sa_mask = sigset;
386 sa.sa_flags = 0;
387 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
388 PERROR("sigaction");
389 return ret;
390 }
391
392 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
393 PERROR("sigaction");
394 return ret;
395 }
396
397 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
398 PERROR("sigaction");
399 return ret;
400 }
401
402 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
403
404 return ret;
405 }
406
407 /*
408 * Init thread quit pipe.
409 *
410 * Return -1 on error or 0 if all pipes are created.
411 */
412 static
413 int init_thread_quit_pipe(void)
414 {
415 int ret;
416
417 ret = utils_create_pipe_cloexec(thread_quit_pipe);
418
419 return ret;
420 }
421
422 /*
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
424 */
425 static
426 int create_thread_poll_set(struct lttng_poll_event *events, int size)
427 {
428 int ret;
429
430 if (events == NULL || size == 0) {
431 ret = -1;
432 goto error;
433 }
434
435 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
436 if (ret < 0) {
437 goto error;
438 }
439
440 /* Add quit pipe */
441 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
442 if (ret < 0) {
443 goto error;
444 }
445
446 return 0;
447
448 error:
449 return ret;
450 }
451
452 /*
453 * Check if the thread quit pipe was triggered.
454 *
455 * Return 1 if it was triggered else 0;
456 */
457 static
458 int check_thread_quit_pipe(int fd, uint32_t events)
459 {
460 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
461 return 1;
462 }
463
464 return 0;
465 }
466
467 /*
468 * Create and init socket from uri.
469 */
470 static
471 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
472 {
473 int ret;
474 struct lttcomm_sock *sock = NULL;
475
476 sock = lttcomm_alloc_sock_from_uri(uri);
477 if (sock == NULL) {
478 ERR("Allocating socket");
479 goto error;
480 }
481
482 ret = lttcomm_create_sock(sock);
483 if (ret < 0) {
484 goto error;
485 }
486 DBG("Listening on sock %d", sock->fd);
487
488 ret = sock->ops->bind(sock);
489 if (ret < 0) {
490 goto error;
491 }
492
493 ret = sock->ops->listen(sock, -1);
494 if (ret < 0) {
495 goto error;
496
497 }
498
499 return sock;
500
501 error:
502 if (sock) {
503 lttcomm_destroy_sock(sock);
504 }
505 return NULL;
506 }
507
508 /*
509 * Return nonzero if stream needs to be closed.
510 */
511 static
512 int close_stream_check(struct relay_stream *stream)
513 {
514
515 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
516 /*
517 * We are about to close the stream so set the data pending flag to 1
518 * which will make the end data pending command skip the stream which
519 * is now closed and ready. Note that after proceeding to a file close,
520 * the written file is ready for reading.
521 */
522 stream->data_pending_check_done = 1;
523 return 1;
524 }
525 return 0;
526 }
527
528 /*
529 * This thread manages the listening for new connections on the network
530 */
531 static
532 void *relay_thread_listener(void *data)
533 {
534 int i, ret, pollfd, err = -1;
535 int val = 1;
536 uint32_t revents, nb_fd;
537 struct lttng_poll_event events;
538 struct lttcomm_sock *control_sock, *data_sock;
539
540 DBG("[thread] Relay listener started");
541
542 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
543
544 health_code_update();
545
546 control_sock = relay_init_sock(control_uri);
547 if (!control_sock) {
548 goto error_sock_control;
549 }
550
551 data_sock = relay_init_sock(data_uri);
552 if (!data_sock) {
553 goto error_sock_relay;
554 }
555
556 /*
557 * Pass 3 as size here for the thread quit pipe, control and data socket.
558 */
559 ret = create_thread_poll_set(&events, 3);
560 if (ret < 0) {
561 goto error_create_poll;
562 }
563
564 /* Add the control socket */
565 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
566 if (ret < 0) {
567 goto error_poll_add;
568 }
569
570 /* Add the data socket */
571 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
572 if (ret < 0) {
573 goto error_poll_add;
574 }
575
576 while (1) {
577 health_code_update();
578
579 DBG("Listener accepting connections");
580
581 restart:
582 health_poll_entry();
583 ret = lttng_poll_wait(&events, -1);
584 health_poll_exit();
585 if (ret < 0) {
586 /*
587 * Restart interrupted system call.
588 */
589 if (errno == EINTR) {
590 goto restart;
591 }
592 goto error;
593 }
594
595 nb_fd = ret;
596
597 DBG("Relay new connection received");
598 for (i = 0; i < nb_fd; i++) {
599 health_code_update();
600
601 /* Fetch once the poll data */
602 revents = LTTNG_POLL_GETEV(&events, i);
603 pollfd = LTTNG_POLL_GETFD(&events, i);
604
605 /* Thread quit pipe has been closed. Killing thread. */
606 ret = check_thread_quit_pipe(pollfd, revents);
607 if (ret) {
608 err = 0;
609 goto exit;
610 }
611
612 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
613 ERR("socket poll error");
614 goto error;
615 } else if (revents & LPOLLIN) {
616 /*
617 * Get allocated in this thread,
618 * enqueued to a global queue, dequeued
619 * and freed in the worker thread.
620 */
621 struct relay_command *relay_cmd;
622 struct lttcomm_sock *newsock;
623
624 relay_cmd = zmalloc(sizeof(struct relay_command));
625 if (relay_cmd == NULL) {
626 PERROR("relay command zmalloc");
627 goto error;
628 }
629
630 if (pollfd == data_sock->fd) {
631 newsock = data_sock->ops->accept(data_sock);
632 if (!newsock) {
633 PERROR("accepting data sock");
634 free(relay_cmd);
635 goto error;
636 }
637 relay_cmd->type = RELAY_DATA;
638 DBG("Relay data connection accepted, socket %d", newsock->fd);
639 } else {
640 assert(pollfd == control_sock->fd);
641 newsock = control_sock->ops->accept(control_sock);
642 if (!newsock) {
643 PERROR("accepting control sock");
644 free(relay_cmd);
645 goto error;
646 }
647 relay_cmd->type = RELAY_CONTROL;
648 DBG("Relay control connection accepted, socket %d", newsock->fd);
649 }
650 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
651 &val, sizeof(int));
652 if (ret < 0) {
653 PERROR("setsockopt inet");
654 lttcomm_destroy_sock(newsock);
655 free(relay_cmd);
656 goto error;
657 }
658 relay_cmd->sock = newsock;
659 /*
660 * Lock free enqueue the request.
661 */
662 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
663
664 /*
665 * Wake the dispatch queue futex. Implicit memory
666 * barrier with the exchange in cds_wfq_enqueue.
667 */
668 futex_nto1_wake(&relay_cmd_queue.futex);
669 }
670 }
671 }
672
673 exit:
674 error:
675 error_poll_add:
676 lttng_poll_clean(&events);
677 error_create_poll:
678 if (data_sock->fd >= 0) {
679 ret = data_sock->ops->close(data_sock);
680 if (ret) {
681 PERROR("close");
682 }
683 }
684 lttcomm_destroy_sock(data_sock);
685 error_sock_relay:
686 if (control_sock->fd >= 0) {
687 ret = control_sock->ops->close(control_sock);
688 if (ret) {
689 PERROR("close");
690 }
691 }
692 lttcomm_destroy_sock(control_sock);
693 error_sock_control:
694 if (err) {
695 health_error();
696 ERR("Health error occurred in %s", __func__);
697 }
698 health_unregister(health_relayd);
699 DBG("Relay listener thread cleanup complete");
700 stop_threads();
701 return NULL;
702 }
703
704 /*
705 * This thread manages the dispatching of the requests to worker threads
706 */
707 static
708 void *relay_thread_dispatcher(void *data)
709 {
710 int ret, err = -1;
711 struct cds_wfq_node *node;
712 struct relay_command *relay_cmd = NULL;
713
714 DBG("[thread] Relay dispatcher started");
715
716 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
717
718 health_code_update();
719
720 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
721 health_code_update();
722
723 /* Atomically prepare the queue futex */
724 futex_nto1_prepare(&relay_cmd_queue.futex);
725
726 do {
727 health_code_update();
728
729 /* Dequeue commands */
730 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
731 if (node == NULL) {
732 DBG("Woken up but nothing in the relay command queue");
733 /* Continue thread execution */
734 break;
735 }
736
737 relay_cmd = caa_container_of(node, struct relay_command, node);
738 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
739
740 /*
741 * Inform worker thread of the new request. This
742 * call is blocking so we can be assured that the data will be read
743 * at some point in time or wait to the end of the world :)
744 */
745 do {
746 ret = write(relay_cmd_pipe[1], relay_cmd,
747 sizeof(struct relay_command));
748 } while (ret < 0 && errno == EINTR);
749 free(relay_cmd);
750 if (ret < 0 || ret != sizeof(struct relay_command)) {
751 PERROR("write cmd pipe");
752 goto error;
753 }
754 } while (node != NULL);
755
756 /* Futex wait on queue. Blocking call on futex() */
757 health_poll_entry();
758 futex_nto1_wait(&relay_cmd_queue.futex);
759 health_poll_exit();
760 }
761
762 /* Normal exit, no error */
763 err = 0;
764
765 error:
766 if (err) {
767 health_error();
768 ERR("Health error occurred in %s", __func__);
769 }
770 health_unregister(health_relayd);
771 DBG("Dispatch thread dying");
772 stop_threads();
773 return NULL;
774 }
775
776 /*
777 * Get stream from stream id.
778 * Need to be called with RCU read-side lock held.
779 */
780 struct relay_stream *relay_stream_find_by_id(uint64_t stream_id)
781 {
782 struct lttng_ht_node_ulong *node;
783 struct lttng_ht_iter iter;
784 struct relay_stream *ret;
785
786 lttng_ht_lookup(relay_streams_ht,
787 (void *)((unsigned long) stream_id),
788 &iter);
789 node = lttng_ht_iter_get_node_ulong(&iter);
790 if (node == NULL) {
791 DBG("Relay stream %" PRIu64 " not found", stream_id);
792 ret = NULL;
793 goto end;
794 }
795
796 ret = caa_container_of(node, struct relay_stream, stream_n);
797
798 end:
799 return ret;
800 }
801
802 static
803 void deferred_free_stream(struct rcu_head *head)
804 {
805 struct relay_stream *stream =
806 caa_container_of(head, struct relay_stream, rcu_node);
807
808 ctf_trace_try_destroy(stream->ctf_trace);
809
810 free(stream->path_name);
811 free(stream->channel_name);
812 free(stream);
813 }
814
815 static
816 void deferred_free_session(struct rcu_head *head)
817 {
818 struct relay_session *session =
819 caa_container_of(head, struct relay_session, rcu_node);
820 free(session);
821 }
822
823 /*
824 * Close a given stream. The stream is freed using a call RCU.
825 *
826 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
827 * BEFORE the stream lock MUST be acquired.
828 */
829 static void destroy_stream(struct relay_stream *stream,
830 struct lttng_ht *ctf_traces_ht)
831 {
832 int delret;
833 struct relay_viewer_stream *vstream;
834 struct lttng_ht_iter iter;
835
836 assert(stream);
837
838 delret = close(stream->fd);
839 if (delret < 0) {
840 PERROR("close stream");
841 }
842
843 if (stream->index_fd >= 0) {
844 delret = close(stream->index_fd);
845 if (delret < 0) {
846 PERROR("close stream index_fd");
847 }
848 }
849
850 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
851 if (vstream) {
852 /*
853 * Set the last good value into the viewer stream. This is done
854 * right before the stream gets deleted from the hash table. The
855 * lookup failure on the live thread side of a stream indicates
856 * that the viewer stream index received value should be used.
857 */
858 vstream->total_index_received = stream->total_index_received;
859 }
860
861 /* Cleanup index of that stream. */
862 relay_index_destroy_by_stream_id(stream->stream_handle);
863
864 iter.iter.node = &stream->stream_n.node;
865 delret = lttng_ht_del(relay_streams_ht, &iter);
866 assert(!delret);
867 iter.iter.node = &stream->ctf_trace_node.node;
868 delret = lttng_ht_del(ctf_traces_ht, &iter);
869 assert(!delret);
870 call_rcu(&stream->rcu_node, deferred_free_stream);
871 DBG("Closed tracefile %d from close stream", stream->fd);
872 }
873
874 /*
875 * relay_delete_session: Free all memory associated with a session and
876 * close all the FDs
877 */
878 static
879 void relay_delete_session(struct relay_command *cmd,
880 struct lttng_ht *sessions_ht)
881 {
882 struct lttng_ht_iter iter;
883 struct lttng_ht_node_ulong *node;
884 struct relay_stream *stream;
885 int ret;
886
887 if (!cmd->session) {
888 return;
889 }
890
891 DBG("Relay deleting session %" PRIu64, cmd->session->id);
892
893 rcu_read_lock();
894 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
895 node = lttng_ht_iter_get_node_ulong(&iter);
896 if (!node) {
897 continue;
898 }
899 stream = caa_container_of(node, struct relay_stream, stream_n);
900 if (stream->session == cmd->session) {
901 destroy_stream(stream, cmd->ctf_traces_ht);
902 }
903 }
904
905 /* Make this session not visible anymore. */
906 iter.iter.node = &cmd->session->session_n.node;
907 ret = lttng_ht_del(sessions_ht, &iter);
908 assert(!ret);
909 call_rcu(&cmd->session->rcu_node, deferred_free_session);
910 rcu_read_unlock();
911 }
912
913 /*
914 * Copy index data from the control port to a given index object.
915 */
916 static void copy_index_control_data(struct relay_index *index,
917 struct lttcomm_relayd_index *data)
918 {
919 assert(index);
920 assert(data);
921
922 /*
923 * The index on disk is encoded in big endian, so we don't need to convert
924 * the data received on the network. The data_offset value is NEVER
925 * modified here and is updated by the data thread.
926 */
927 index->index_data.packet_size = data->packet_size;
928 index->index_data.content_size = data->content_size;
929 index->index_data.timestamp_begin = data->timestamp_begin;
930 index->index_data.timestamp_end = data->timestamp_end;
931 index->index_data.events_discarded = data->events_discarded;
932 index->index_data.stream_id = data->stream_id;
933 }
934
935 /*
936 * Handle the RELAYD_CREATE_SESSION command.
937 *
938 * On success, send back the session id or else return a negative value.
939 */
940 static
941 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
942 struct relay_command *cmd,
943 struct lttng_ht *sessions_ht)
944 {
945 int ret = 0, send_ret;
946 struct relay_session *session;
947 struct lttcomm_relayd_status_session reply;
948
949 assert(recv_hdr);
950 assert(cmd);
951
952 memset(&reply, 0, sizeof(reply));
953
954 session = zmalloc(sizeof(struct relay_session));
955 if (session == NULL) {
956 PERROR("relay session zmalloc");
957 ret = -1;
958 goto error;
959 }
960
961 session->id = ++last_relay_session_id;
962 session->sock = cmd->sock;
963 session->minor = cmd->minor;
964 session->major = cmd->major;
965 cmd->session = session;
966
967 reply.session_id = htobe64(session->id);
968
969 switch (cmd->minor) {
970 case 4: /* LTTng sessiond 2.4 */
971 default:
972 ret = cmd_create_session_2_4(cmd, session);
973 break;
974 }
975
976 lttng_ht_node_init_ulong(&session->session_n,
977 (unsigned long) session->id);
978 lttng_ht_add_unique_ulong(sessions_ht,
979 &session->session_n);
980
981 DBG("Created session %" PRIu64, session->id);
982
983 error:
984 if (ret < 0) {
985 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
986 } else {
987 reply.ret_code = htobe32(LTTNG_OK);
988 }
989
990 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
991 if (send_ret < 0) {
992 ERR("Relayd sending session id");
993 ret = send_ret;
994 }
995
996 return ret;
997 }
998
999 /*
1000 * relay_add_stream: allocate a new stream for a session
1001 */
1002 static
1003 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1004 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1005 {
1006 struct relay_session *session = cmd->session;
1007 struct relay_stream *stream = NULL;
1008 struct lttcomm_relayd_status_stream reply;
1009 int ret, send_ret;
1010
1011 if (!session || cmd->version_check_done == 0) {
1012 ERR("Trying to add a stream before version check");
1013 ret = -1;
1014 goto end_no_session;
1015 }
1016
1017 stream = zmalloc(sizeof(struct relay_stream));
1018 if (stream == NULL) {
1019 PERROR("relay stream zmalloc");
1020 ret = -1;
1021 goto end_no_session;
1022 }
1023
1024 switch (cmd->minor) {
1025 case 1: /* LTTng sessiond 2.1 */
1026 ret = cmd_recv_stream_2_1(cmd, stream);
1027 break;
1028 case 2: /* LTTng sessiond 2.2 */
1029 default:
1030 ret = cmd_recv_stream_2_2(cmd, stream);
1031 break;
1032 }
1033 if (ret < 0) {
1034 goto err_free_stream;
1035 }
1036
1037 rcu_read_lock();
1038 stream->stream_handle = ++last_relay_stream_id;
1039 stream->prev_seq = -1ULL;
1040 stream->session = session;
1041 stream->index_fd = -1;
1042 stream->read_index_fd = -1;
1043 stream->ctf_trace = NULL;
1044 pthread_mutex_init(&stream->lock, NULL);
1045
1046 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
1047 if (ret < 0) {
1048 ERR("relay creating output directory");
1049 goto end;
1050 }
1051
1052 /*
1053 * No need to use run_as API here because whatever we receives, the relayd
1054 * uses its own credentials for the stream files.
1055 */
1056 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1057 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1058 if (ret < 0) {
1059 ERR("Create output file");
1060 goto end;
1061 }
1062 stream->fd = ret;
1063 if (stream->tracefile_size) {
1064 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1065 } else {
1066 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1067 }
1068
1069 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1070 stream->metadata_flag = 1;
1071 /*
1072 * When we receive a new metadata stream, we create a new
1073 * ctf_trace and we assign this ctf_trace to all streams with
1074 * the same path.
1075 *
1076 * If later on we receive a new stream for the same ctf_trace,
1077 * we copy the information from the first hit in the HT to the
1078 * new stream.
1079 */
1080 stream->ctf_trace = ctf_trace_create();
1081 if (!stream->ctf_trace) {
1082 ret = -1;
1083 goto end;
1084 }
1085 stream->ctf_trace->refcount++;
1086 stream->ctf_trace->metadata_stream = stream;
1087 }
1088 ctf_trace_assign(cmd->ctf_traces_ht, stream);
1089
1090 lttng_ht_node_init_ulong(&stream->stream_n,
1091 (unsigned long) stream->stream_handle);
1092 lttng_ht_add_unique_ulong(relay_streams_ht,
1093 &stream->stream_n);
1094
1095 lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name);
1096 lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node);
1097
1098 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1099 stream->stream_handle);
1100
1101 end:
1102 reply.handle = htobe64(stream->stream_handle);
1103 /* send the session id to the client or a negative return code on error */
1104 if (ret < 0) {
1105 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1106 /* stream was not properly added to the ht, so free it */
1107 free(stream);
1108 } else {
1109 reply.ret_code = htobe32(LTTNG_OK);
1110 }
1111
1112 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1113 sizeof(struct lttcomm_relayd_status_stream), 0);
1114 if (send_ret < 0) {
1115 ERR("Relay sending stream id");
1116 ret = send_ret;
1117 }
1118 rcu_read_unlock();
1119
1120 end_no_session:
1121 return ret;
1122
1123 err_free_stream:
1124 free(stream->path_name);
1125 free(stream->channel_name);
1126 free(stream);
1127 return ret;
1128 }
1129
1130 /*
1131 * relay_close_stream: close a specific stream
1132 */
1133 static
1134 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1135 struct relay_command *cmd)
1136 {
1137 int ret, send_ret;
1138 struct relay_session *session = cmd->session;
1139 struct lttcomm_relayd_close_stream stream_info;
1140 struct lttcomm_relayd_generic_reply reply;
1141 struct relay_stream *stream;
1142
1143 DBG("Close stream received");
1144
1145 if (!session || cmd->version_check_done == 0) {
1146 ERR("Trying to close a stream before version check");
1147 ret = -1;
1148 goto end_no_session;
1149 }
1150
1151 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1152 sizeof(struct lttcomm_relayd_close_stream), 0);
1153 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1154 if (ret == 0) {
1155 /* Orderly shutdown. Not necessary to print an error. */
1156 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1157 } else {
1158 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1159 }
1160 ret = -1;
1161 goto end_no_session;
1162 }
1163
1164 rcu_read_lock();
1165 stream = relay_stream_find_by_id(be64toh(stream_info.stream_id));
1166 if (!stream) {
1167 ret = -1;
1168 goto end_unlock;
1169 }
1170
1171 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1172 stream->close_flag = 1;
1173
1174 if (close_stream_check(stream)) {
1175 destroy_stream(stream, cmd->ctf_traces_ht);
1176 }
1177
1178 end_unlock:
1179 rcu_read_unlock();
1180
1181 if (ret < 0) {
1182 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1183 } else {
1184 reply.ret_code = htobe32(LTTNG_OK);
1185 }
1186 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1187 sizeof(struct lttcomm_relayd_generic_reply), 0);
1188 if (send_ret < 0) {
1189 ERR("Relay sending stream id");
1190 ret = send_ret;
1191 }
1192
1193 end_no_session:
1194 return ret;
1195 }
1196
1197 /*
1198 * relay_unknown_command: send -1 if received unknown command
1199 */
1200 static
1201 void relay_unknown_command(struct relay_command *cmd)
1202 {
1203 struct lttcomm_relayd_generic_reply reply;
1204 int ret;
1205
1206 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1207 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1208 sizeof(struct lttcomm_relayd_generic_reply), 0);
1209 if (ret < 0) {
1210 ERR("Relay sending unknown command");
1211 }
1212 }
1213
1214 /*
1215 * relay_start: send an acknowledgment to the client to tell if we are
1216 * ready to receive data. We are ready if a session is established.
1217 */
1218 static
1219 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1220 struct relay_command *cmd)
1221 {
1222 int ret = htobe32(LTTNG_OK);
1223 struct lttcomm_relayd_generic_reply reply;
1224 struct relay_session *session = cmd->session;
1225
1226 if (!session) {
1227 DBG("Trying to start the streaming without a session established");
1228 ret = htobe32(LTTNG_ERR_UNK);
1229 }
1230
1231 reply.ret_code = ret;
1232 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1233 sizeof(struct lttcomm_relayd_generic_reply), 0);
1234 if (ret < 0) {
1235 ERR("Relay sending start ack");
1236 }
1237
1238 return ret;
1239 }
1240
1241 /*
1242 * Append padding to the file pointed by the file descriptor fd.
1243 */
1244 static int write_padding_to_file(int fd, uint32_t size)
1245 {
1246 int ret = 0;
1247 char *zeros;
1248
1249 if (size == 0) {
1250 goto end;
1251 }
1252
1253 zeros = zmalloc(size);
1254 if (zeros == NULL) {
1255 PERROR("zmalloc zeros for padding");
1256 ret = -1;
1257 goto end;
1258 }
1259
1260 do {
1261 ret = write(fd, zeros, size);
1262 } while (ret < 0 && errno == EINTR);
1263 if (ret < 0 || ret != size) {
1264 PERROR("write padding to file");
1265 }
1266
1267 free(zeros);
1268
1269 end:
1270 return ret;
1271 }
1272
1273 /*
1274 * relay_recv_metadata: receive the metada for the session.
1275 */
1276 static
1277 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1278 struct relay_command *cmd)
1279 {
1280 int ret = htobe32(LTTNG_OK);
1281 struct relay_session *session = cmd->session;
1282 struct lttcomm_relayd_metadata_payload *metadata_struct;
1283 struct relay_stream *metadata_stream;
1284 uint64_t data_size, payload_size;
1285
1286 if (!session) {
1287 ERR("Metadata sent before version check");
1288 ret = -1;
1289 goto end;
1290 }
1291
1292 data_size = payload_size = be64toh(recv_hdr->data_size);
1293 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1294 ERR("Incorrect data size");
1295 ret = -1;
1296 goto end;
1297 }
1298 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1299
1300 if (data_buffer_size < data_size) {
1301 /* In case the realloc fails, we can free the memory */
1302 char *tmp_data_ptr;
1303
1304 tmp_data_ptr = realloc(data_buffer, data_size);
1305 if (!tmp_data_ptr) {
1306 ERR("Allocating data buffer");
1307 free(data_buffer);
1308 ret = -1;
1309 goto end;
1310 }
1311 data_buffer = tmp_data_ptr;
1312 data_buffer_size = data_size;
1313 }
1314 memset(data_buffer, 0, data_size);
1315 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1316 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1317 if (ret < 0 || ret != data_size) {
1318 if (ret == 0) {
1319 /* Orderly shutdown. Not necessary to print an error. */
1320 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1321 } else {
1322 ERR("Relay didn't receive the whole metadata");
1323 }
1324 ret = -1;
1325 goto end;
1326 }
1327 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1328
1329 rcu_read_lock();
1330 metadata_stream = relay_stream_find_by_id(
1331 be64toh(metadata_struct->stream_id));
1332 if (!metadata_stream) {
1333 ret = -1;
1334 goto end_unlock;
1335 }
1336
1337 do {
1338 ret = write(metadata_stream->fd, metadata_struct->payload,
1339 payload_size);
1340 } while (ret < 0 && errno == EINTR);
1341 if (ret < 0 || ret != payload_size) {
1342 ERR("Relay error writing metadata on file");
1343 ret = -1;
1344 goto end_unlock;
1345 }
1346
1347 ret = write_padding_to_file(metadata_stream->fd,
1348 be32toh(metadata_struct->padding_size));
1349 if (ret < 0) {
1350 goto end_unlock;
1351 }
1352 metadata_stream->ctf_trace->metadata_received +=
1353 payload_size + be32toh(metadata_struct->padding_size);
1354
1355 DBG2("Relay metadata written");
1356
1357 end_unlock:
1358 rcu_read_unlock();
1359 end:
1360 return ret;
1361 }
1362
1363 /*
1364 * relay_send_version: send relayd version number
1365 */
1366 static
1367 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1368 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1369 {
1370 int ret;
1371 struct lttcomm_relayd_version reply, msg;
1372
1373 assert(cmd);
1374
1375 cmd->version_check_done = 1;
1376
1377 /* Get version from the other side. */
1378 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1379 if (ret < 0 || ret != sizeof(msg)) {
1380 if (ret == 0) {
1381 /* Orderly shutdown. Not necessary to print an error. */
1382 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1383 } else {
1384 ERR("Relay failed to receive the version values.");
1385 }
1386 ret = -1;
1387 goto end;
1388 }
1389
1390 reply.major = RELAYD_VERSION_COMM_MAJOR;
1391 reply.minor = RELAYD_VERSION_COMM_MINOR;
1392
1393 /* Major versions must be the same */
1394 if (reply.major != be32toh(msg.major)) {
1395 DBG("Incompatible major versions (%u vs %u), deleting session",
1396 reply.major, be32toh(msg.major));
1397 relay_delete_session(cmd, sessions_ht);
1398 ret = 0;
1399 goto end;
1400 }
1401
1402 cmd->major = reply.major;
1403 /* We adapt to the lowest compatible version */
1404 if (reply.minor <= be32toh(msg.minor)) {
1405 cmd->minor = reply.minor;
1406 } else {
1407 cmd->minor = be32toh(msg.minor);
1408 }
1409
1410 reply.major = htobe32(reply.major);
1411 reply.minor = htobe32(reply.minor);
1412 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1413 sizeof(struct lttcomm_relayd_version), 0);
1414 if (ret < 0) {
1415 ERR("Relay sending version");
1416 }
1417
1418 DBG("Version check done using protocol %u.%u", cmd->major,
1419 cmd->minor);
1420
1421 end:
1422 return ret;
1423 }
1424
1425 /*
1426 * Check for data pending for a given stream id from the session daemon.
1427 */
1428 static
1429 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1430 struct relay_command *cmd)
1431 {
1432 struct relay_session *session = cmd->session;
1433 struct lttcomm_relayd_data_pending msg;
1434 struct lttcomm_relayd_generic_reply reply;
1435 struct relay_stream *stream;
1436 int ret;
1437 uint64_t last_net_seq_num, stream_id;
1438
1439 DBG("Data pending command received");
1440
1441 if (!session || cmd->version_check_done == 0) {
1442 ERR("Trying to check for data before version check");
1443 ret = -1;
1444 goto end_no_session;
1445 }
1446
1447 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1448 if (ret < sizeof(msg)) {
1449 if (ret == 0) {
1450 /* Orderly shutdown. Not necessary to print an error. */
1451 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1452 } else {
1453 ERR("Relay didn't receive valid data_pending struct size : %d",
1454 ret);
1455 }
1456 ret = -1;
1457 goto end_no_session;
1458 }
1459
1460 stream_id = be64toh(msg.stream_id);
1461 last_net_seq_num = be64toh(msg.last_net_seq_num);
1462
1463 rcu_read_lock();
1464 stream = relay_stream_find_by_id(stream_id);
1465 if (stream == NULL) {
1466 ret = -1;
1467 goto end_unlock;
1468 }
1469
1470 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1471 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1472 last_net_seq_num);
1473
1474 /* Avoid wrapping issue */
1475 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1476 /* Data has in fact been written and is NOT pending */
1477 ret = 0;
1478 } else {
1479 /* Data still being streamed thus pending */
1480 ret = 1;
1481 }
1482
1483 /* Pending check is now done. */
1484 stream->data_pending_check_done = 1;
1485
1486 end_unlock:
1487 rcu_read_unlock();
1488
1489 reply.ret_code = htobe32(ret);
1490 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1491 if (ret < 0) {
1492 ERR("Relay data pending ret code failed");
1493 }
1494
1495 end_no_session:
1496 return ret;
1497 }
1498
1499 /*
1500 * Wait for the control socket to reach a quiescent state.
1501 *
1502 * Note that for now, when receiving this command from the session daemon, this
1503 * means that every subsequent commands or data received on the control socket
1504 * has been handled. So, this is why we simply return OK here.
1505 */
1506 static
1507 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1508 struct relay_command *cmd)
1509 {
1510 int ret;
1511 uint64_t stream_id;
1512 struct relay_stream *stream;
1513 struct lttng_ht_iter iter;
1514 struct lttcomm_relayd_quiescent_control msg;
1515 struct lttcomm_relayd_generic_reply reply;
1516
1517 DBG("Checking quiescent state on control socket");
1518
1519 if (!cmd->session || cmd->version_check_done == 0) {
1520 ERR("Trying to check for data before version check");
1521 ret = -1;
1522 goto end_no_session;
1523 }
1524
1525 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1526 if (ret < sizeof(msg)) {
1527 if (ret == 0) {
1528 /* Orderly shutdown. Not necessary to print an error. */
1529 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1530 } else {
1531 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1532 ret);
1533 }
1534 ret = -1;
1535 goto end_no_session;
1536 }
1537
1538 stream_id = be64toh(msg.stream_id);
1539
1540 rcu_read_lock();
1541 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1542 stream_n.node) {
1543 if (stream->stream_handle == stream_id) {
1544 stream->data_pending_check_done = 1;
1545 DBG("Relay quiescent control pending flag set to %" PRIu64,
1546 stream_id);
1547 break;
1548 }
1549 }
1550 rcu_read_unlock();
1551
1552 reply.ret_code = htobe32(LTTNG_OK);
1553 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1554 if (ret < 0) {
1555 ERR("Relay data quiescent control ret code failed");
1556 }
1557
1558 end_no_session:
1559 return ret;
1560 }
1561
1562 /*
1563 * Initialize a data pending command. This means that a client is about to ask
1564 * for data pending for each stream he/she holds. Simply iterate over all
1565 * streams of a session and set the data_pending_check_done flag.
1566 *
1567 * This command returns to the client a LTTNG_OK code.
1568 */
1569 static
1570 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1571 struct relay_command *cmd)
1572 {
1573 int ret;
1574 struct lttng_ht_iter iter;
1575 struct lttcomm_relayd_begin_data_pending msg;
1576 struct lttcomm_relayd_generic_reply reply;
1577 struct relay_stream *stream;
1578 uint64_t session_id;
1579
1580 assert(recv_hdr);
1581 assert(cmd);
1582
1583 DBG("Init streams for data pending");
1584
1585 if (!cmd->session || cmd->version_check_done == 0) {
1586 ERR("Trying to check for data before version check");
1587 ret = -1;
1588 goto end_no_session;
1589 }
1590
1591 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1592 if (ret < sizeof(msg)) {
1593 if (ret == 0) {
1594 /* Orderly shutdown. Not necessary to print an error. */
1595 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1596 } else {
1597 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1598 ret);
1599 }
1600 ret = -1;
1601 goto end_no_session;
1602 }
1603
1604 session_id = be64toh(msg.session_id);
1605
1606 /*
1607 * Iterate over all streams to set the begin data pending flag. For now, the
1608 * streams are indexed by stream handle so we have to iterate over all
1609 * streams to find the one associated with the right session_id.
1610 */
1611 rcu_read_lock();
1612 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1613 stream_n.node) {
1614 if (stream->session->id == session_id) {
1615 stream->data_pending_check_done = 0;
1616 DBG("Set begin data pending flag to stream %" PRIu64,
1617 stream->stream_handle);
1618 }
1619 }
1620 rcu_read_unlock();
1621
1622 /* All good, send back reply. */
1623 reply.ret_code = htobe32(LTTNG_OK);
1624
1625 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1626 if (ret < 0) {
1627 ERR("Relay begin data pending send reply failed");
1628 }
1629
1630 end_no_session:
1631 return ret;
1632 }
1633
1634 /*
1635 * End data pending command. This will check, for a given session id, if each
1636 * stream associated with it has its data_pending_check_done flag set. If not,
1637 * this means that the client lost track of the stream but the data is still
1638 * being streamed on our side. In this case, we inform the client that data is
1639 * inflight.
1640 *
1641 * Return to the client if there is data in flight or not with a ret_code.
1642 */
1643 static
1644 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1645 struct relay_command *cmd)
1646 {
1647 int ret;
1648 struct lttng_ht_iter iter;
1649 struct lttcomm_relayd_end_data_pending msg;
1650 struct lttcomm_relayd_generic_reply reply;
1651 struct relay_stream *stream;
1652 uint64_t session_id;
1653 uint32_t is_data_inflight = 0;
1654
1655 assert(recv_hdr);
1656 assert(cmd);
1657
1658 DBG("End data pending command");
1659
1660 if (!cmd->session || cmd->version_check_done == 0) {
1661 ERR("Trying to check for data before version check");
1662 ret = -1;
1663 goto end_no_session;
1664 }
1665
1666 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1667 if (ret < sizeof(msg)) {
1668 if (ret == 0) {
1669 /* Orderly shutdown. Not necessary to print an error. */
1670 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1671 } else {
1672 ERR("Relay didn't receive valid end data_pending struct size: %d",
1673 ret);
1674 }
1675 ret = -1;
1676 goto end_no_session;
1677 }
1678
1679 session_id = be64toh(msg.session_id);
1680
1681 /* Iterate over all streams to see if the begin data pending flag is set. */
1682 rcu_read_lock();
1683 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1684 stream_n.node) {
1685 if (stream->session->id == session_id &&
1686 !stream->data_pending_check_done) {
1687 is_data_inflight = 1;
1688 DBG("Data is still in flight for stream %" PRIu64,
1689 stream->stream_handle);
1690 break;
1691 }
1692 }
1693 rcu_read_unlock();
1694
1695 /* All good, send back reply. */
1696 reply.ret_code = htobe32(is_data_inflight);
1697
1698 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1699 if (ret < 0) {
1700 ERR("Relay end data pending send reply failed");
1701 }
1702
1703 end_no_session:
1704 return ret;
1705 }
1706
1707 /*
1708 * Receive an index for a specific stream.
1709 *
1710 * Return 0 on success else a negative value.
1711 */
1712 static
1713 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1714 struct relay_command *cmd)
1715 {
1716 int ret, send_ret, index_created = 0;
1717 struct relay_session *session = cmd->session;
1718 struct lttcomm_relayd_index index_info;
1719 struct relay_index *index, *wr_index = NULL;
1720 struct lttcomm_relayd_generic_reply reply;
1721 struct relay_stream *stream;
1722 uint64_t net_seq_num;
1723
1724 assert(cmd);
1725
1726 DBG("Relay receiving index");
1727
1728 if (!session || cmd->version_check_done == 0) {
1729 ERR("Trying to close a stream before version check");
1730 ret = -1;
1731 goto end_no_session;
1732 }
1733
1734 ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info,
1735 sizeof(index_info), 0);
1736 if (ret < sizeof(index_info)) {
1737 if (ret == 0) {
1738 /* Orderly shutdown. Not necessary to print an error. */
1739 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1740 } else {
1741 ERR("Relay didn't receive valid index struct size : %d", ret);
1742 }
1743 ret = -1;
1744 goto end_no_session;
1745 }
1746
1747 net_seq_num = be64toh(index_info.net_seq_num);
1748
1749 rcu_read_lock();
1750 stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id));
1751 if (!stream) {
1752 ret = -1;
1753 goto end_rcu_unlock;
1754 }
1755
1756 /* Live beacon handling */
1757 if (index_info.packet_size == 0) {
1758 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1759
1760 /*
1761 * Only flag a stream inactive when it has already received data.
1762 */
1763 if (stream->total_index_received > 0) {
1764 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1765 }
1766 ret = 0;
1767 goto end_rcu_unlock;
1768 } else {
1769 stream->beacon_ts_end = -1ULL;
1770 }
1771
1772 index = relay_index_find(stream->stream_handle, net_seq_num);
1773 if (!index) {
1774 /* A successful creation will add the object to the HT. */
1775 index = relay_index_create(stream->stream_handle, net_seq_num);
1776 if (!index) {
1777 goto end_rcu_unlock;
1778 }
1779 index_created = 1;
1780 }
1781
1782 copy_index_control_data(index, &index_info);
1783
1784 if (index_created) {
1785 /*
1786 * Try to add the relay index object to the hash table. If an object
1787 * already exist, destroy back the index created, set the data in this
1788 * object and write it on disk.
1789 */
1790 relay_index_add(index, &wr_index);
1791 if (wr_index) {
1792 copy_index_control_data(wr_index, &index_info);
1793 free(index);
1794 }
1795 } else {
1796 /* The index already exists so write it on disk. */
1797 wr_index = index;
1798 }
1799
1800 /* Do we have a writable ready index to write on disk. */
1801 if (wr_index) {
1802 /* Starting at 2.4, create the index file if none available. */
1803 if (cmd->minor >= 4 && stream->index_fd < 0) {
1804 ret = index_create_file(stream->path_name, stream->channel_name,
1805 relayd_uid, relayd_gid, stream->tracefile_size,
1806 stream->tracefile_count_current);
1807 if (ret < 0) {
1808 goto end_rcu_unlock;
1809 }
1810 stream->index_fd = ret;
1811 }
1812
1813 ret = relay_index_write(wr_index->fd, wr_index);
1814 if (ret < 0) {
1815 goto end_rcu_unlock;
1816 }
1817 stream->total_index_received++;
1818 }
1819
1820 end_rcu_unlock:
1821 rcu_read_unlock();
1822
1823 if (ret < 0) {
1824 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1825 } else {
1826 reply.ret_code = htobe32(LTTNG_OK);
1827 }
1828 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1829 if (send_ret < 0) {
1830 ERR("Relay sending close index id reply");
1831 ret = send_ret;
1832 }
1833
1834 end_no_session:
1835 return ret;
1836 }
1837
1838 /*
1839 * Process the commands received on the control socket
1840 */
1841 static
1842 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1843 struct relay_command *cmd, struct relay_local_data *ctx)
1844 {
1845 int ret = 0;
1846
1847 switch (be32toh(recv_hdr->cmd)) {
1848 case RELAYD_CREATE_SESSION:
1849 ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht);
1850 break;
1851 case RELAYD_ADD_STREAM:
1852 ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht);
1853 break;
1854 case RELAYD_START_DATA:
1855 ret = relay_start(recv_hdr, cmd);
1856 break;
1857 case RELAYD_SEND_METADATA:
1858 ret = relay_recv_metadata(recv_hdr, cmd);
1859 break;
1860 case RELAYD_VERSION:
1861 ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht);
1862 break;
1863 case RELAYD_CLOSE_STREAM:
1864 ret = relay_close_stream(recv_hdr, cmd);
1865 break;
1866 case RELAYD_DATA_PENDING:
1867 ret = relay_data_pending(recv_hdr, cmd);
1868 break;
1869 case RELAYD_QUIESCENT_CONTROL:
1870 ret = relay_quiescent_control(recv_hdr, cmd);
1871 break;
1872 case RELAYD_BEGIN_DATA_PENDING:
1873 ret = relay_begin_data_pending(recv_hdr, cmd);
1874 break;
1875 case RELAYD_END_DATA_PENDING:
1876 ret = relay_end_data_pending(recv_hdr, cmd);
1877 break;
1878 case RELAYD_SEND_INDEX:
1879 ret = relay_recv_index(recv_hdr, cmd);
1880 break;
1881 case RELAYD_UPDATE_SYNC_INFO:
1882 default:
1883 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1884 relay_unknown_command(cmd);
1885 ret = -1;
1886 goto end;
1887 }
1888
1889 end:
1890 return ret;
1891 }
1892
1893 /*
1894 * Handle index for a data stream.
1895 *
1896 * RCU read side lock MUST be acquired.
1897 *
1898 * Return 0 on success else a negative value.
1899 */
1900 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
1901 int rotate_index)
1902 {
1903 int ret = 0, index_created = 0;
1904 uint64_t stream_id, data_offset;
1905 struct relay_index *index, *wr_index = NULL;
1906
1907 assert(stream);
1908
1909 stream_id = stream->stream_handle;
1910 /* Get data offset because we are about to update the index. */
1911 data_offset = htobe64(stream->tracefile_size_current);
1912
1913 /*
1914 * Lookup for an existing index for that stream id/sequence number. If on
1915 * exists, the control thread already received the data for it thus we need
1916 * to write it on disk.
1917 */
1918 index = relay_index_find(stream_id, net_seq_num);
1919 if (!index) {
1920 /* A successful creation will add the object to the HT. */
1921 index = relay_index_create(stream_id, net_seq_num);
1922 if (!index) {
1923 ret = -1;
1924 goto error;
1925 }
1926 index_created = 1;
1927 }
1928
1929 if (rotate_index || stream->index_fd < 0) {
1930 index->to_close_fd = stream->index_fd;
1931 ret = index_create_file(stream->path_name, stream->channel_name,
1932 relayd_uid, relayd_gid, stream->tracefile_size,
1933 stream->tracefile_count_current);
1934 if (ret < 0) {
1935 /* This will close the stream's index fd if one. */
1936 relay_index_free_safe(index);
1937 goto error;
1938 }
1939 stream->index_fd = ret;
1940 }
1941 index->fd = stream->index_fd;
1942 index->index_data.offset = data_offset;
1943
1944 if (index_created) {
1945 /*
1946 * Try to add the relay index object to the hash table. If an object
1947 * already exist, destroy back the index created and set the data.
1948 */
1949 relay_index_add(index, &wr_index);
1950 if (wr_index) {
1951 /* Copy back data from the created index. */
1952 wr_index->fd = index->fd;
1953 wr_index->to_close_fd = index->to_close_fd;
1954 wr_index->index_data.offset = data_offset;
1955 free(index);
1956 }
1957 } else {
1958 /* The index already exists so write it on disk. */
1959 wr_index = index;
1960 }
1961
1962 /* Do we have a writable ready index to write on disk. */
1963 if (wr_index) {
1964 ret = relay_index_write(wr_index->fd, wr_index);
1965 if (ret < 0) {
1966 goto error;
1967 }
1968 stream->total_index_received++;
1969 }
1970
1971 error:
1972 return ret;
1973 }
1974
1975 /*
1976 * relay_process_data: Process the data received on the data socket
1977 */
1978 static
1979 int relay_process_data(struct relay_command *cmd)
1980 {
1981 int ret = 0, rotate_index = 0;
1982 struct relay_stream *stream;
1983 struct lttcomm_relayd_data_hdr data_hdr;
1984 uint64_t stream_id;
1985 uint64_t net_seq_num;
1986 uint32_t data_size;
1987
1988 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1989 sizeof(struct lttcomm_relayd_data_hdr), 0);
1990 if (ret <= 0) {
1991 if (ret == 0) {
1992 /* Orderly shutdown. Not necessary to print an error. */
1993 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1994 } else {
1995 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
1996 }
1997 ret = -1;
1998 goto end;
1999 }
2000
2001 stream_id = be64toh(data_hdr.stream_id);
2002
2003 rcu_read_lock();
2004 stream = relay_stream_find_by_id(stream_id);
2005 if (!stream) {
2006 ret = -1;
2007 goto end_rcu_unlock;
2008 }
2009
2010 data_size = be32toh(data_hdr.data_size);
2011 if (data_buffer_size < data_size) {
2012 char *tmp_data_ptr;
2013
2014 tmp_data_ptr = realloc(data_buffer, data_size);
2015 if (!tmp_data_ptr) {
2016 ERR("Allocating data buffer");
2017 free(data_buffer);
2018 ret = -1;
2019 goto end_rcu_unlock;
2020 }
2021 data_buffer = tmp_data_ptr;
2022 data_buffer_size = data_size;
2023 }
2024 memset(data_buffer, 0, data_size);
2025
2026 net_seq_num = be64toh(data_hdr.net_seq_num);
2027
2028 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2029 data_size, stream_id, net_seq_num);
2030 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
2031 if (ret <= 0) {
2032 if (ret == 0) {
2033 /* Orderly shutdown. Not necessary to print an error. */
2034 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2035 }
2036 ret = -1;
2037 goto end_rcu_unlock;
2038 }
2039
2040 /* Check if a rotation is needed. */
2041 if (stream->tracefile_size > 0 &&
2042 (stream->tracefile_size_current + data_size) >
2043 stream->tracefile_size) {
2044 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2045 stream->tracefile_size, stream->tracefile_count,
2046 relayd_uid, relayd_gid, stream->fd,
2047 &(stream->tracefile_count_current), &stream->fd);
2048 if (ret < 0) {
2049 ERR("Rotating stream output file");
2050 goto end_rcu_unlock;
2051 }
2052 /* Reset current size because we just perform a stream rotation. */
2053 stream->tracefile_size_current = 0;
2054 rotate_index = 1;
2055 }
2056
2057 /*
2058 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2059 * index are NOT supported.
2060 */
2061 if (stream->session->minor >= 4 && !stream->session->snapshot) {
2062 ret = handle_index_data(stream, net_seq_num, rotate_index);
2063 if (ret < 0) {
2064 goto end_rcu_unlock;
2065 }
2066 }
2067
2068 /* Write data to stream output fd. */
2069 do {
2070 ret = write(stream->fd, data_buffer, data_size);
2071 } while (ret < 0 && errno == EINTR);
2072 if (ret < 0 || ret != data_size) {
2073 ERR("Relay error writing data to file");
2074 ret = -1;
2075 goto end_rcu_unlock;
2076 }
2077
2078 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2079 ret, stream->stream_handle);
2080
2081 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2082 if (ret < 0) {
2083 goto end_rcu_unlock;
2084 }
2085 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2086
2087 stream->prev_seq = net_seq_num;
2088
2089 /* Check if we need to close the FD */
2090 if (close_stream_check(stream)) {
2091 destroy_stream(stream, cmd->ctf_traces_ht);
2092 }
2093
2094 end_rcu_unlock:
2095 rcu_read_unlock();
2096 end:
2097 return ret;
2098 }
2099
2100 static
2101 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
2102 {
2103 int ret;
2104
2105 lttng_poll_del(events, pollfd);
2106
2107 ret = close(pollfd);
2108 if (ret < 0) {
2109 ERR("Closing pollfd %d", pollfd);
2110 }
2111 }
2112
2113 static
2114 int relay_add_connection(int fd, struct lttng_poll_event *events,
2115 struct lttng_ht *relay_connections_ht)
2116 {
2117 struct relay_command *relay_connection;
2118 int ret;
2119
2120 relay_connection = zmalloc(sizeof(struct relay_command));
2121 if (relay_connection == NULL) {
2122 PERROR("Relay command zmalloc");
2123 goto error;
2124 }
2125 do {
2126 ret = read(fd, relay_connection, sizeof(struct relay_command));
2127 } while (ret < 0 && errno == EINTR);
2128 if (ret < 0 || ret < sizeof(struct relay_command)) {
2129 PERROR("read relay cmd pipe");
2130 goto error_read;
2131 }
2132
2133 relay_connection->ctf_traces_ht = lttng_ht_new(0, LTTNG_HT_TYPE_STRING);
2134 if (!relay_connection->ctf_traces_ht) {
2135 goto error_read;
2136 }
2137
2138 lttng_ht_node_init_ulong(&relay_connection->sock_n,
2139 (unsigned long) relay_connection->sock->fd);
2140 rcu_read_lock();
2141 lttng_ht_add_unique_ulong(relay_connections_ht,
2142 &relay_connection->sock_n);
2143 rcu_read_unlock();
2144 return lttng_poll_add(events,
2145 relay_connection->sock->fd,
2146 LPOLLIN | LPOLLRDHUP);
2147
2148 error_read:
2149 free(relay_connection);
2150 error:
2151 return -1;
2152 }
2153
2154 static
2155 void deferred_free_connection(struct rcu_head *head)
2156 {
2157 struct relay_command *relay_connection =
2158 caa_container_of(head, struct relay_command, rcu_node);
2159
2160 lttng_ht_destroy(relay_connection->ctf_traces_ht);
2161 lttcomm_destroy_sock(relay_connection->sock);
2162 free(relay_connection);
2163 }
2164
2165 static
2166 void relay_del_connection(struct lttng_ht *relay_connections_ht,
2167 struct lttng_ht_iter *iter, struct relay_command *relay_connection,
2168 struct lttng_ht *sessions_ht)
2169 {
2170 int ret;
2171
2172 ret = lttng_ht_del(relay_connections_ht, iter);
2173 assert(!ret);
2174 if (relay_connection->type == RELAY_CONTROL) {
2175 relay_delete_session(relay_connection, sessions_ht);
2176 }
2177
2178 call_rcu(&relay_connection->rcu_node,
2179 deferred_free_connection);
2180 }
2181
2182 /*
2183 * This thread does the actual work
2184 */
2185 static
2186 void *relay_thread_worker(void *data)
2187 {
2188 int ret, err = -1, last_seen_data_fd = -1;
2189 uint32_t nb_fd;
2190 struct relay_command *relay_connection;
2191 struct lttng_poll_event events;
2192 struct lttng_ht *relay_connections_ht;
2193 struct lttng_ht_node_ulong *node;
2194 struct lttng_ht_iter iter;
2195 struct lttcomm_relayd_hdr recv_hdr;
2196 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2197 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2198
2199 DBG("[thread] Relay worker started");
2200
2201 rcu_register_thread();
2202
2203 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2204
2205 health_code_update();
2206
2207 /* table of connections indexed on socket */
2208 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2209 if (!relay_connections_ht) {
2210 goto relay_connections_ht_error;
2211 }
2212
2213 /* Tables of received indexes indexed by index handle and net_seq_num. */
2214 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2215 if (!indexes_ht) {
2216 goto indexes_ht_error;
2217 }
2218
2219 ret = create_thread_poll_set(&events, 2);
2220 if (ret < 0) {
2221 goto error_poll_create;
2222 }
2223
2224 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
2225 if (ret < 0) {
2226 goto error;
2227 }
2228
2229 restart:
2230 while (1) {
2231 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2232
2233 health_code_update();
2234
2235 /* Infinite blocking call, waiting for transmission */
2236 DBG3("Relayd worker thread polling...");
2237 health_poll_entry();
2238 ret = lttng_poll_wait(&events, -1);
2239 health_poll_exit();
2240 if (ret < 0) {
2241 /*
2242 * Restart interrupted system call.
2243 */
2244 if (errno == EINTR) {
2245 goto restart;
2246 }
2247 goto error;
2248 }
2249
2250 nb_fd = ret;
2251
2252 /*
2253 * Process control. The control connection is prioritised so we don't
2254 * starve it with high throughout put tracing data on the data
2255 * connection.
2256 */
2257 for (i = 0; i < nb_fd; i++) {
2258 /* Fetch once the poll data */
2259 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2260 int pollfd = LTTNG_POLL_GETFD(&events, i);
2261
2262 health_code_update();
2263
2264 /* Thread quit pipe has been closed. Killing thread. */
2265 ret = check_thread_quit_pipe(pollfd, revents);
2266 if (ret) {
2267 err = 0;
2268 goto exit;
2269 }
2270
2271 /* Inspect the relay cmd pipe for new connection */
2272 if (pollfd == relay_cmd_pipe[0]) {
2273 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2274 ERR("Relay pipe error");
2275 goto error;
2276 } else if (revents & LPOLLIN) {
2277 DBG("Relay command received");
2278 ret = relay_add_connection(relay_cmd_pipe[0],
2279 &events, relay_connections_ht);
2280 if (ret < 0) {
2281 goto error;
2282 }
2283 }
2284 } else if (revents) {
2285 rcu_read_lock();
2286 lttng_ht_lookup(relay_connections_ht,
2287 (void *)((unsigned long) pollfd),
2288 &iter);
2289 node = lttng_ht_iter_get_node_ulong(&iter);
2290 if (node == NULL) {
2291 DBG2("Relay sock %d not found", pollfd);
2292 rcu_read_unlock();
2293 goto error;
2294 }
2295 relay_connection = caa_container_of(node,
2296 struct relay_command, sock_n);
2297
2298 if (revents & (LPOLLERR)) {
2299 ERR("POLL ERROR");
2300 relay_cleanup_poll_connection(&events, pollfd);
2301 relay_del_connection(relay_connections_ht,
2302 &iter, relay_connection, sessions_ht);
2303 if (last_seen_data_fd == pollfd) {
2304 last_seen_data_fd = last_notdel_data_fd;
2305 }
2306 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2307 DBG("Socket %d hung up", pollfd);
2308 relay_cleanup_poll_connection(&events, pollfd);
2309 relay_del_connection(relay_connections_ht,
2310 &iter, relay_connection, sessions_ht);
2311 if (last_seen_data_fd == pollfd) {
2312 last_seen_data_fd = last_notdel_data_fd;
2313 }
2314 } else if (revents & LPOLLIN) {
2315 /* control socket */
2316 if (relay_connection->type == RELAY_CONTROL) {
2317 ret = relay_connection->sock->ops->recvmsg(
2318 relay_connection->sock, &recv_hdr,
2319 sizeof(struct lttcomm_relayd_hdr), 0);
2320 /* connection closed */
2321 if (ret <= 0) {
2322 relay_cleanup_poll_connection(&events, pollfd);
2323 relay_del_connection(relay_connections_ht,
2324 &iter, relay_connection, sessions_ht);
2325 DBG("Control connection closed with %d", pollfd);
2326 } else {
2327 if (relay_connection->session) {
2328 DBG2("Relay worker receiving data for session : %" PRIu64,
2329 relay_connection->session->id);
2330 }
2331 ret = relay_process_control(&recv_hdr,
2332 relay_connection, relay_ctx);
2333 if (ret < 0) {
2334 /* Clear the session on error. */
2335 relay_cleanup_poll_connection(&events, pollfd);
2336 relay_del_connection(relay_connections_ht,
2337 &iter, relay_connection, sessions_ht);
2338 DBG("Connection closed with %d", pollfd);
2339 }
2340 seen_control = 1;
2341 }
2342 } else {
2343 /*
2344 * Flag the last seen data fd not deleted. It will be
2345 * used as the last seen fd if any fd gets deleted in
2346 * this first loop.
2347 */
2348 last_notdel_data_fd = pollfd;
2349 }
2350 }
2351 rcu_read_unlock();
2352 }
2353 }
2354
2355 /*
2356 * The last loop handled a control request, go back to poll to make
2357 * sure we prioritise the control socket.
2358 */
2359 if (seen_control) {
2360 continue;
2361 }
2362
2363 if (last_seen_data_fd >= 0) {
2364 for (i = 0; i < nb_fd; i++) {
2365 int pollfd = LTTNG_POLL_GETFD(&events, i);
2366
2367 health_code_update();
2368
2369 if (last_seen_data_fd == pollfd) {
2370 idx = i;
2371 break;
2372 }
2373 }
2374 }
2375
2376 /* Process data connection. */
2377 for (i = idx + 1; i < nb_fd; i++) {
2378 /* Fetch the poll data. */
2379 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2380 int pollfd = LTTNG_POLL_GETFD(&events, i);
2381
2382 health_code_update();
2383
2384 /* Skip the command pipe. It's handled in the first loop. */
2385 if (pollfd == relay_cmd_pipe[0]) {
2386 continue;
2387 }
2388
2389 if (revents) {
2390 rcu_read_lock();
2391 lttng_ht_lookup(relay_connections_ht,
2392 (void *)((unsigned long) pollfd),
2393 &iter);
2394 node = lttng_ht_iter_get_node_ulong(&iter);
2395 if (node == NULL) {
2396 /* Skip it. Might be removed before. */
2397 rcu_read_unlock();
2398 continue;
2399 }
2400 relay_connection = caa_container_of(node,
2401 struct relay_command, sock_n);
2402
2403 if (revents & LPOLLIN) {
2404 if (relay_connection->type != RELAY_DATA) {
2405 continue;
2406 }
2407
2408 ret = relay_process_data(relay_connection);
2409 /* connection closed */
2410 if (ret < 0) {
2411 relay_cleanup_poll_connection(&events, pollfd);
2412 relay_del_connection(relay_connections_ht,
2413 &iter, relay_connection, sessions_ht);
2414 DBG("Data connection closed with %d", pollfd);
2415 /*
2416 * Every goto restart call sets the last seen fd where
2417 * here we don't really care since we gracefully
2418 * continue the loop after the connection is deleted.
2419 */
2420 } else {
2421 /* Keep last seen port. */
2422 last_seen_data_fd = pollfd;
2423 rcu_read_unlock();
2424 goto restart;
2425 }
2426 }
2427 rcu_read_unlock();
2428 }
2429 }
2430 last_seen_data_fd = -1;
2431 }
2432
2433 /* Normal exit, no error */
2434 ret = 0;
2435
2436 exit:
2437 error:
2438 lttng_poll_clean(&events);
2439
2440 /* empty the hash table and free the memory */
2441 rcu_read_lock();
2442 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2443 health_code_update();
2444
2445 node = lttng_ht_iter_get_node_ulong(&iter);
2446 if (node) {
2447 relay_connection = caa_container_of(node,
2448 struct relay_command, sock_n);
2449 relay_del_connection(relay_connections_ht,
2450 &iter, relay_connection, sessions_ht);
2451 }
2452 }
2453 rcu_read_unlock();
2454 error_poll_create:
2455 lttng_ht_destroy(indexes_ht);
2456 indexes_ht_error:
2457 lttng_ht_destroy(relay_connections_ht);
2458 relay_connections_ht_error:
2459 /* Close relay cmd pipes */
2460 utils_close_pipe(relay_cmd_pipe);
2461 if (err) {
2462 DBG("Thread exited with error");
2463 }
2464 DBG("Worker thread cleanup complete");
2465 free(data_buffer);
2466 if (err) {
2467 health_error();
2468 ERR("Health error occurred in %s", __func__);
2469 }
2470 health_unregister(health_relayd);
2471 rcu_unregister_thread();
2472 stop_threads();
2473 return NULL;
2474 }
2475
2476 /*
2477 * Create the relay command pipe to wake thread_manage_apps.
2478 * Closed in cleanup().
2479 */
2480 static int create_relay_cmd_pipe(void)
2481 {
2482 int ret;
2483
2484 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2485
2486 return ret;
2487 }
2488
2489 /*
2490 * main
2491 */
2492 int main(int argc, char **argv)
2493 {
2494 int ret = 0;
2495 void *status;
2496 struct relay_local_data *relay_ctx;
2497
2498 /* Create thread quit pipe */
2499 if ((ret = init_thread_quit_pipe()) < 0) {
2500 goto error;
2501 }
2502
2503 /* Parse arguments */
2504 progname = argv[0];
2505 if ((ret = parse_args(argc, argv)) < 0) {
2506 goto exit;
2507 }
2508
2509 if ((ret = set_signal_handler()) < 0) {
2510 goto exit;
2511 }
2512
2513 /* Try to create directory if -o, --output is specified. */
2514 if (opt_output_path) {
2515 if (*opt_output_path != '/') {
2516 ERR("Please specify an absolute path for -o, --output PATH");
2517 goto exit;
2518 }
2519
2520 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2521 if (ret < 0) {
2522 ERR("Unable to create %s", opt_output_path);
2523 goto exit;
2524 }
2525 }
2526
2527 /* Daemonize */
2528 if (opt_daemon) {
2529 ret = daemon(0, 0);
2530 if (ret < 0) {
2531 PERROR("daemon");
2532 goto exit;
2533 }
2534 }
2535
2536 /* We need those values for the file/dir creation. */
2537 relayd_uid = getuid();
2538 relayd_gid = getgid();
2539
2540 /* Check if daemon is UID = 0 */
2541 if (relayd_uid == 0) {
2542 if (control_uri->port < 1024 || data_uri->port < 1024) {
2543 ERR("Need to be root to use ports < 1024");
2544 ret = -1;
2545 goto exit;
2546 }
2547 }
2548
2549 /* Setup the thread apps communication pipe. */
2550 if ((ret = create_relay_cmd_pipe()) < 0) {
2551 goto exit;
2552 }
2553
2554 /* Init relay command queue. */
2555 cds_wfq_init(&relay_cmd_queue.queue);
2556
2557 /* Set up max poll set size */
2558 lttng_poll_set_max_size();
2559
2560 /* Initialize communication library */
2561 lttcomm_init();
2562
2563 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2564 if (!relay_ctx) {
2565 PERROR("relay_ctx");
2566 goto exit;
2567 }
2568
2569 /* tables of sessions indexed by session ID */
2570 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2571 if (!relay_ctx->sessions_ht) {
2572 goto exit_relay_ctx_sessions;
2573 }
2574
2575 /* tables of streams indexed by stream ID */
2576 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2577 if (!relay_streams_ht) {
2578 goto exit_relay_ctx_streams;
2579 }
2580
2581 /* tables of streams indexed by stream ID */
2582 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2583 if (!viewer_streams_ht) {
2584 goto exit_relay_ctx_viewer_streams;
2585 }
2586
2587 /* Initialize thread health monitoring */
2588 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2589 if (!health_relayd) {
2590 PERROR("health_app_create error");
2591 goto exit_health_app_create;
2592 }
2593
2594 ret = utils_create_pipe(health_quit_pipe);
2595 if (ret < 0) {
2596 goto error_health_pipe;
2597 }
2598
2599 /* Create thread to manage the client socket */
2600 ret = pthread_create(&health_thread, NULL,
2601 thread_manage_health, (void *) NULL);
2602 if (ret != 0) {
2603 PERROR("pthread_create health");
2604 goto health_error;
2605 }
2606
2607 /* Setup the dispatcher thread */
2608 ret = pthread_create(&dispatcher_thread, NULL,
2609 relay_thread_dispatcher, (void *) NULL);
2610 if (ret != 0) {
2611 PERROR("pthread_create dispatcher");
2612 goto exit_dispatcher;
2613 }
2614
2615 /* Setup the worker thread */
2616 ret = pthread_create(&worker_thread, NULL,
2617 relay_thread_worker, (void *) relay_ctx);
2618 if (ret != 0) {
2619 PERROR("pthread_create worker");
2620 goto exit_worker;
2621 }
2622
2623 /* Setup the listener thread */
2624 ret = pthread_create(&listener_thread, NULL,
2625 relay_thread_listener, (void *) NULL);
2626 if (ret != 0) {
2627 PERROR("pthread_create listener");
2628 goto exit_listener;
2629 }
2630
2631 ret = live_start_threads(live_uri, relay_ctx, thread_quit_pipe);
2632 if (ret != 0) {
2633 ERR("Starting live viewer threads");
2634 goto exit_live;
2635 }
2636
2637 exit_live:
2638 ret = pthread_join(listener_thread, &status);
2639 if (ret != 0) {
2640 PERROR("pthread_join");
2641 goto error; /* join error, exit without cleanup */
2642 }
2643
2644 exit_listener:
2645 ret = pthread_join(worker_thread, &status);
2646 if (ret != 0) {
2647 PERROR("pthread_join");
2648 goto error; /* join error, exit without cleanup */
2649 }
2650
2651 exit_worker:
2652 ret = pthread_join(dispatcher_thread, &status);
2653 if (ret != 0) {
2654 PERROR("pthread_join");
2655 goto error; /* join error, exit without cleanup */
2656 }
2657
2658 exit_dispatcher:
2659 ret = pthread_join(health_thread, &status);
2660 if (ret != 0) {
2661 PERROR("pthread_join health thread");
2662 goto error; /* join error, exit without cleanup */
2663 }
2664
2665 /*
2666 * Stop live threads only after joining other threads.
2667 */
2668 live_stop_threads();
2669
2670 health_error:
2671 utils_close_pipe(health_quit_pipe);
2672
2673 error_health_pipe:
2674 health_app_destroy(health_relayd);
2675
2676 exit_health_app_create:
2677 lttng_ht_destroy(viewer_streams_ht);
2678
2679 exit_relay_ctx_viewer_streams:
2680 lttng_ht_destroy(relay_streams_ht);
2681
2682 exit_relay_ctx_streams:
2683 lttng_ht_destroy(relay_ctx->sessions_ht);
2684
2685 exit_relay_ctx_sessions:
2686 free(relay_ctx);
2687
2688 exit:
2689 cleanup();
2690 if (!ret) {
2691 exit(EXIT_SUCCESS);
2692 }
2693
2694 error:
2695 exit(EXIT_FAILURE);
2696 }
This page took 0.120349 seconds and 5 git commands to generate.