Use lttng_read/lttng_write wrappers
[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 ssize_t ret;
297
298 ret = lttng_write(wpipe, "!", 1);
299 if (ret < 1) {
300 PERROR("write poll pipe");
301 }
302
303 return ret;
304 }
305
306 static void notify_health_quit_pipe(int *pipe)
307 {
308 ssize_t ret;
309
310 ret = lttng_write(pipe[1], "4", 1);
311 if (ret < 1) {
312 PERROR("write relay health quit");
313 }
314 }
315
316 /*
317 * Stop all threads by closing the thread quit pipe.
318 */
319 static
320 void stop_threads(void)
321 {
322 int ret;
323
324 /* Stopping all threads */
325 DBG("Terminating all threads");
326 ret = notify_thread_pipe(thread_quit_pipe[1]);
327 if (ret < 0) {
328 ERR("write error on thread quit pipe");
329 }
330
331 notify_health_quit_pipe(health_quit_pipe);
332
333 /* Dispatch thread */
334 CMM_STORE_SHARED(dispatch_thread_exit, 1);
335 futex_nto1_wake(&relay_cmd_queue.futex);
336 }
337
338 /*
339 * Signal handler for the daemon
340 *
341 * Simply stop all worker threads, leaving main() return gracefully after
342 * joining all threads and calling cleanup().
343 */
344 static
345 void sighandler(int sig)
346 {
347 switch (sig) {
348 case SIGPIPE:
349 DBG("SIGPIPE caught");
350 return;
351 case SIGINT:
352 DBG("SIGINT caught");
353 stop_threads();
354 break;
355 case SIGTERM:
356 DBG("SIGTERM caught");
357 stop_threads();
358 break;
359 default:
360 break;
361 }
362 }
363
364 /*
365 * Setup signal handler for :
366 * SIGINT, SIGTERM, SIGPIPE
367 */
368 static
369 int set_signal_handler(void)
370 {
371 int ret = 0;
372 struct sigaction sa;
373 sigset_t sigset;
374
375 if ((ret = sigemptyset(&sigset)) < 0) {
376 PERROR("sigemptyset");
377 return ret;
378 }
379
380 sa.sa_handler = sighandler;
381 sa.sa_mask = sigset;
382 sa.sa_flags = 0;
383 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
384 PERROR("sigaction");
385 return ret;
386 }
387
388 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
389 PERROR("sigaction");
390 return ret;
391 }
392
393 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
394 PERROR("sigaction");
395 return ret;
396 }
397
398 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
399
400 return ret;
401 }
402
403 /*
404 * Init thread quit pipe.
405 *
406 * Return -1 on error or 0 if all pipes are created.
407 */
408 static
409 int init_thread_quit_pipe(void)
410 {
411 int ret;
412
413 ret = utils_create_pipe_cloexec(thread_quit_pipe);
414
415 return ret;
416 }
417
418 /*
419 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
420 */
421 static
422 int create_thread_poll_set(struct lttng_poll_event *events, int size)
423 {
424 int ret;
425
426 if (events == NULL || size == 0) {
427 ret = -1;
428 goto error;
429 }
430
431 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
432 if (ret < 0) {
433 goto error;
434 }
435
436 /* Add quit pipe */
437 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
438 if (ret < 0) {
439 goto error;
440 }
441
442 return 0;
443
444 error:
445 return ret;
446 }
447
448 /*
449 * Check if the thread quit pipe was triggered.
450 *
451 * Return 1 if it was triggered else 0;
452 */
453 static
454 int check_thread_quit_pipe(int fd, uint32_t events)
455 {
456 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
457 return 1;
458 }
459
460 return 0;
461 }
462
463 /*
464 * Create and init socket from uri.
465 */
466 static
467 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
468 {
469 int ret;
470 struct lttcomm_sock *sock = NULL;
471
472 sock = lttcomm_alloc_sock_from_uri(uri);
473 if (sock == NULL) {
474 ERR("Allocating socket");
475 goto error;
476 }
477
478 ret = lttcomm_create_sock(sock);
479 if (ret < 0) {
480 goto error;
481 }
482 DBG("Listening on sock %d", sock->fd);
483
484 ret = sock->ops->bind(sock);
485 if (ret < 0) {
486 goto error;
487 }
488
489 ret = sock->ops->listen(sock, -1);
490 if (ret < 0) {
491 goto error;
492
493 }
494
495 return sock;
496
497 error:
498 if (sock) {
499 lttcomm_destroy_sock(sock);
500 }
501 return NULL;
502 }
503
504 /*
505 * Return nonzero if stream needs to be closed.
506 */
507 static
508 int close_stream_check(struct relay_stream *stream)
509 {
510
511 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
512 /*
513 * We are about to close the stream so set the data pending flag to 1
514 * which will make the end data pending command skip the stream which
515 * is now closed and ready. Note that after proceeding to a file close,
516 * the written file is ready for reading.
517 */
518 stream->data_pending_check_done = 1;
519 return 1;
520 }
521 return 0;
522 }
523
524 /*
525 * This thread manages the listening for new connections on the network
526 */
527 static
528 void *relay_thread_listener(void *data)
529 {
530 int i, ret, pollfd, err = -1;
531 int val = 1;
532 uint32_t revents, nb_fd;
533 struct lttng_poll_event events;
534 struct lttcomm_sock *control_sock, *data_sock;
535
536 DBG("[thread] Relay listener started");
537
538 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
539
540 health_code_update();
541
542 control_sock = relay_init_sock(control_uri);
543 if (!control_sock) {
544 goto error_sock_control;
545 }
546
547 data_sock = relay_init_sock(data_uri);
548 if (!data_sock) {
549 goto error_sock_relay;
550 }
551
552 /*
553 * Pass 3 as size here for the thread quit pipe, control and data socket.
554 */
555 ret = create_thread_poll_set(&events, 3);
556 if (ret < 0) {
557 goto error_create_poll;
558 }
559
560 /* Add the control socket */
561 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
562 if (ret < 0) {
563 goto error_poll_add;
564 }
565
566 /* Add the data socket */
567 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
568 if (ret < 0) {
569 goto error_poll_add;
570 }
571
572 while (1) {
573 health_code_update();
574
575 DBG("Listener accepting connections");
576
577 restart:
578 health_poll_entry();
579 ret = lttng_poll_wait(&events, -1);
580 health_poll_exit();
581 if (ret < 0) {
582 /*
583 * Restart interrupted system call.
584 */
585 if (errno == EINTR) {
586 goto restart;
587 }
588 goto error;
589 }
590
591 nb_fd = ret;
592
593 DBG("Relay new connection received");
594 for (i = 0; i < nb_fd; i++) {
595 health_code_update();
596
597 /* Fetch once the poll data */
598 revents = LTTNG_POLL_GETEV(&events, i);
599 pollfd = LTTNG_POLL_GETFD(&events, i);
600
601 /* Thread quit pipe has been closed. Killing thread. */
602 ret = check_thread_quit_pipe(pollfd, revents);
603 if (ret) {
604 err = 0;
605 goto exit;
606 }
607
608 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
609 ERR("socket poll error");
610 goto error;
611 } else if (revents & LPOLLIN) {
612 /*
613 * Get allocated in this thread,
614 * enqueued to a global queue, dequeued
615 * and freed in the worker thread.
616 */
617 struct relay_command *relay_cmd;
618 struct lttcomm_sock *newsock;
619
620 relay_cmd = zmalloc(sizeof(struct relay_command));
621 if (relay_cmd == NULL) {
622 PERROR("relay command zmalloc");
623 goto error;
624 }
625
626 if (pollfd == data_sock->fd) {
627 newsock = data_sock->ops->accept(data_sock);
628 if (!newsock) {
629 PERROR("accepting data sock");
630 free(relay_cmd);
631 goto error;
632 }
633 relay_cmd->type = RELAY_DATA;
634 DBG("Relay data connection accepted, socket %d", newsock->fd);
635 } else {
636 assert(pollfd == control_sock->fd);
637 newsock = control_sock->ops->accept(control_sock);
638 if (!newsock) {
639 PERROR("accepting control sock");
640 free(relay_cmd);
641 goto error;
642 }
643 relay_cmd->type = RELAY_CONTROL;
644 DBG("Relay control connection accepted, socket %d", newsock->fd);
645 }
646 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
647 &val, sizeof(int));
648 if (ret < 0) {
649 PERROR("setsockopt inet");
650 lttcomm_destroy_sock(newsock);
651 free(relay_cmd);
652 goto error;
653 }
654 relay_cmd->sock = newsock;
655 /*
656 * Lock free enqueue the request.
657 */
658 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
659
660 /*
661 * Wake the dispatch queue futex. Implicit memory
662 * barrier with the exchange in cds_wfq_enqueue.
663 */
664 futex_nto1_wake(&relay_cmd_queue.futex);
665 }
666 }
667 }
668
669 exit:
670 error:
671 error_poll_add:
672 lttng_poll_clean(&events);
673 error_create_poll:
674 if (data_sock->fd >= 0) {
675 ret = data_sock->ops->close(data_sock);
676 if (ret) {
677 PERROR("close");
678 }
679 }
680 lttcomm_destroy_sock(data_sock);
681 error_sock_relay:
682 if (control_sock->fd >= 0) {
683 ret = control_sock->ops->close(control_sock);
684 if (ret) {
685 PERROR("close");
686 }
687 }
688 lttcomm_destroy_sock(control_sock);
689 error_sock_control:
690 if (err) {
691 health_error();
692 ERR("Health error occurred in %s", __func__);
693 }
694 health_unregister(health_relayd);
695 DBG("Relay listener thread cleanup complete");
696 stop_threads();
697 return NULL;
698 }
699
700 /*
701 * This thread manages the dispatching of the requests to worker threads
702 */
703 static
704 void *relay_thread_dispatcher(void *data)
705 {
706 int err = -1;
707 ssize_t ret;
708 struct cds_wfq_node *node;
709 struct relay_command *relay_cmd = NULL;
710
711 DBG("[thread] Relay dispatcher started");
712
713 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
714
715 health_code_update();
716
717 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
718 health_code_update();
719
720 /* Atomically prepare the queue futex */
721 futex_nto1_prepare(&relay_cmd_queue.futex);
722
723 do {
724 health_code_update();
725
726 /* Dequeue commands */
727 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
728 if (node == NULL) {
729 DBG("Woken up but nothing in the relay command queue");
730 /* Continue thread execution */
731 break;
732 }
733
734 relay_cmd = caa_container_of(node, struct relay_command, node);
735 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
736
737 /*
738 * Inform worker thread of the new request. This
739 * call is blocking so we can be assured that the data will be read
740 * at some point in time or wait to the end of the world :)
741 */
742 ret = lttng_write(relay_cmd_pipe[1], relay_cmd,
743 sizeof(struct relay_command));
744 free(relay_cmd);
745 if (ret < sizeof(struct relay_command)) {
746 PERROR("write cmd pipe");
747 goto error;
748 }
749 } while (node != NULL);
750
751 /* Futex wait on queue. Blocking call on futex() */
752 health_poll_entry();
753 futex_nto1_wait(&relay_cmd_queue.futex);
754 health_poll_exit();
755 }
756
757 /* Normal exit, no error */
758 err = 0;
759
760 error:
761 if (err) {
762 health_error();
763 ERR("Health error occurred in %s", __func__);
764 }
765 health_unregister(health_relayd);
766 DBG("Dispatch thread dying");
767 stop_threads();
768 return NULL;
769 }
770
771 /*
772 * Get stream from stream id.
773 * Need to be called with RCU read-side lock held.
774 */
775 struct relay_stream *relay_stream_find_by_id(uint64_t stream_id)
776 {
777 struct lttng_ht_node_ulong *node;
778 struct lttng_ht_iter iter;
779 struct relay_stream *ret;
780
781 lttng_ht_lookup(relay_streams_ht,
782 (void *)((unsigned long) stream_id),
783 &iter);
784 node = lttng_ht_iter_get_node_ulong(&iter);
785 if (node == NULL) {
786 DBG("Relay stream %" PRIu64 " not found", stream_id);
787 ret = NULL;
788 goto end;
789 }
790
791 ret = caa_container_of(node, struct relay_stream, stream_n);
792
793 end:
794 return ret;
795 }
796
797 static
798 void deferred_free_stream(struct rcu_head *head)
799 {
800 struct relay_stream *stream =
801 caa_container_of(head, struct relay_stream, rcu_node);
802
803 ctf_trace_try_destroy(stream->ctf_trace);
804
805 free(stream->path_name);
806 free(stream->channel_name);
807 free(stream);
808 }
809
810 static
811 void deferred_free_session(struct rcu_head *head)
812 {
813 struct relay_session *session =
814 caa_container_of(head, struct relay_session, rcu_node);
815 free(session);
816 }
817
818 /*
819 * Close a given stream. The stream is freed using a call RCU.
820 *
821 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
822 * BEFORE the stream lock MUST be acquired.
823 */
824 static void destroy_stream(struct relay_stream *stream)
825 {
826 int delret;
827 struct relay_viewer_stream *vstream;
828 struct lttng_ht_iter iter;
829
830 assert(stream);
831
832 delret = close(stream->fd);
833 if (delret < 0) {
834 PERROR("close stream");
835 }
836
837 if (stream->index_fd >= 0) {
838 delret = close(stream->index_fd);
839 if (delret < 0) {
840 PERROR("close stream index_fd");
841 }
842 }
843
844 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
845 if (vstream) {
846 /*
847 * Set the last good value into the viewer stream. This is done
848 * right before the stream gets deleted from the hash table. The
849 * lookup failure on the live thread side of a stream indicates
850 * that the viewer stream index received value should be used.
851 */
852 vstream->total_index_received = stream->total_index_received;
853 }
854
855 /* Cleanup index of that stream. */
856 relay_index_destroy_by_stream_id(stream->stream_handle);
857
858 iter.iter.node = &stream->stream_n.node;
859 delret = lttng_ht_del(relay_streams_ht, &iter);
860 assert(!delret);
861 iter.iter.node = &stream->ctf_trace_node.node;
862 delret = lttng_ht_del(stream->ctf_traces_ht, &iter);
863 assert(!delret);
864 call_rcu(&stream->rcu_node, deferred_free_stream);
865 DBG("Closed tracefile %d from close stream", stream->fd);
866 }
867
868 /*
869 * relay_delete_session: Free all memory associated with a session and
870 * close all the FDs
871 */
872 static
873 void relay_delete_session(struct relay_command *cmd,
874 struct lttng_ht *sessions_ht)
875 {
876 struct lttng_ht_iter iter;
877 struct lttng_ht_node_ulong *node;
878 struct relay_stream *stream;
879 int ret;
880
881 if (!cmd->session) {
882 return;
883 }
884
885 DBG("Relay deleting session %" PRIu64, cmd->session->id);
886
887 rcu_read_lock();
888 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
889 node = lttng_ht_iter_get_node_ulong(&iter);
890 if (!node) {
891 continue;
892 }
893 stream = caa_container_of(node, struct relay_stream, stream_n);
894 if (stream->session == cmd->session) {
895 destroy_stream(stream);
896 cmd->session->stream_count--;
897 assert(cmd->session->stream_count >= 0);
898 }
899 }
900
901 /* Make this session not visible anymore. */
902 iter.iter.node = &cmd->session->session_n.node;
903 ret = lttng_ht_del(sessions_ht, &iter);
904 assert(!ret);
905 call_rcu(&cmd->session->rcu_node, deferred_free_session);
906 rcu_read_unlock();
907 }
908
909 /*
910 * Copy index data from the control port to a given index object.
911 */
912 static void copy_index_control_data(struct relay_index *index,
913 struct lttcomm_relayd_index *data)
914 {
915 assert(index);
916 assert(data);
917
918 /*
919 * The index on disk is encoded in big endian, so we don't need to convert
920 * the data received on the network. The data_offset value is NEVER
921 * modified here and is updated by the data thread.
922 */
923 index->index_data.packet_size = data->packet_size;
924 index->index_data.content_size = data->content_size;
925 index->index_data.timestamp_begin = data->timestamp_begin;
926 index->index_data.timestamp_end = data->timestamp_end;
927 index->index_data.events_discarded = data->events_discarded;
928 index->index_data.stream_id = data->stream_id;
929 }
930
931 /*
932 * Handle the RELAYD_CREATE_SESSION command.
933 *
934 * On success, send back the session id or else return a negative value.
935 */
936 static
937 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
938 struct relay_command *cmd,
939 struct lttng_ht *sessions_ht)
940 {
941 int ret = 0, send_ret;
942 struct relay_session *session;
943 struct lttcomm_relayd_status_session reply;
944
945 assert(recv_hdr);
946 assert(cmd);
947
948 memset(&reply, 0, sizeof(reply));
949
950 session = zmalloc(sizeof(struct relay_session));
951 if (session == NULL) {
952 PERROR("relay session zmalloc");
953 ret = -1;
954 goto error;
955 }
956
957 session->id = ++last_relay_session_id;
958 session->sock = cmd->sock;
959 session->minor = cmd->minor;
960 session->major = cmd->major;
961 cmd->session = session;
962
963 reply.session_id = htobe64(session->id);
964
965 switch (cmd->minor) {
966 case 4: /* LTTng sessiond 2.4 */
967 default:
968 ret = cmd_create_session_2_4(cmd, session);
969 break;
970 }
971
972 lttng_ht_node_init_ulong(&session->session_n,
973 (unsigned long) session->id);
974 lttng_ht_add_unique_ulong(sessions_ht,
975 &session->session_n);
976
977 DBG("Created session %" PRIu64, session->id);
978
979 error:
980 if (ret < 0) {
981 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
982 } else {
983 reply.ret_code = htobe32(LTTNG_OK);
984 }
985
986 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
987 if (send_ret < 0) {
988 ERR("Relayd sending session id");
989 ret = send_ret;
990 }
991
992 return ret;
993 }
994
995 /*
996 * relay_add_stream: allocate a new stream for a session
997 */
998 static
999 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1000 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1001 {
1002 struct relay_session *session = cmd->session;
1003 struct relay_stream *stream = NULL;
1004 struct lttcomm_relayd_status_stream reply;
1005 int ret, send_ret;
1006
1007 if (!session || cmd->version_check_done == 0) {
1008 ERR("Trying to add a stream before version check");
1009 ret = -1;
1010 goto end_no_session;
1011 }
1012
1013 stream = zmalloc(sizeof(struct relay_stream));
1014 if (stream == NULL) {
1015 PERROR("relay stream zmalloc");
1016 ret = -1;
1017 goto end_no_session;
1018 }
1019
1020 switch (cmd->minor) {
1021 case 1: /* LTTng sessiond 2.1 */
1022 ret = cmd_recv_stream_2_1(cmd, stream);
1023 break;
1024 case 2: /* LTTng sessiond 2.2 */
1025 default:
1026 ret = cmd_recv_stream_2_2(cmd, stream);
1027 break;
1028 }
1029 if (ret < 0) {
1030 goto err_free_stream;
1031 }
1032
1033 rcu_read_lock();
1034 stream->stream_handle = ++last_relay_stream_id;
1035 stream->prev_seq = -1ULL;
1036 stream->session = session;
1037 stream->index_fd = -1;
1038 stream->read_index_fd = -1;
1039 stream->ctf_trace = NULL;
1040 pthread_mutex_init(&stream->lock, NULL);
1041
1042 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
1043 if (ret < 0) {
1044 ERR("relay creating output directory");
1045 goto end;
1046 }
1047
1048 /*
1049 * No need to use run_as API here because whatever we receives, the relayd
1050 * uses its own credentials for the stream files.
1051 */
1052 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1053 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1054 if (ret < 0) {
1055 ERR("Create output file");
1056 goto end;
1057 }
1058 stream->fd = ret;
1059 if (stream->tracefile_size) {
1060 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1061 } else {
1062 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1063 }
1064
1065 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1066 stream->metadata_flag = 1;
1067 /*
1068 * When we receive a new metadata stream, we create a new
1069 * ctf_trace and we assign this ctf_trace to all streams with
1070 * the same path.
1071 *
1072 * If later on we receive a new stream for the same ctf_trace,
1073 * we copy the information from the first hit in the HT to the
1074 * new stream.
1075 */
1076 stream->ctf_trace = ctf_trace_create();
1077 if (!stream->ctf_trace) {
1078 ret = -1;
1079 goto end;
1080 }
1081 stream->ctf_trace->refcount++;
1082 stream->ctf_trace->metadata_stream = stream;
1083 }
1084 ctf_trace_assign(cmd->ctf_traces_ht, stream);
1085 stream->ctf_traces_ht = cmd->ctf_traces_ht;
1086
1087 lttng_ht_node_init_ulong(&stream->stream_n,
1088 (unsigned long) stream->stream_handle);
1089 lttng_ht_add_unique_ulong(relay_streams_ht,
1090 &stream->stream_n);
1091
1092 lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name);
1093 lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node);
1094 session->stream_count++;
1095
1096 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1097 stream->stream_handle);
1098
1099 end:
1100 reply.handle = htobe64(stream->stream_handle);
1101 /* send the session id to the client or a negative return code on error */
1102 if (ret < 0) {
1103 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1104 /* stream was not properly added to the ht, so free it */
1105 free(stream);
1106 } else {
1107 reply.ret_code = htobe32(LTTNG_OK);
1108 }
1109
1110 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1111 sizeof(struct lttcomm_relayd_status_stream), 0);
1112 if (send_ret < 0) {
1113 ERR("Relay sending stream id");
1114 ret = send_ret;
1115 }
1116 rcu_read_unlock();
1117
1118 end_no_session:
1119 return ret;
1120
1121 err_free_stream:
1122 free(stream->path_name);
1123 free(stream->channel_name);
1124 free(stream);
1125 return ret;
1126 }
1127
1128 /*
1129 * relay_close_stream: close a specific stream
1130 */
1131 static
1132 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1133 struct relay_command *cmd)
1134 {
1135 int ret, send_ret;
1136 struct relay_session *session = cmd->session;
1137 struct lttcomm_relayd_close_stream stream_info;
1138 struct lttcomm_relayd_generic_reply reply;
1139 struct relay_stream *stream;
1140
1141 DBG("Close stream received");
1142
1143 if (!session || cmd->version_check_done == 0) {
1144 ERR("Trying to close a stream before version check");
1145 ret = -1;
1146 goto end_no_session;
1147 }
1148
1149 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1150 sizeof(struct lttcomm_relayd_close_stream), 0);
1151 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1152 if (ret == 0) {
1153 /* Orderly shutdown. Not necessary to print an error. */
1154 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1155 } else {
1156 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1157 }
1158 ret = -1;
1159 goto end_no_session;
1160 }
1161
1162 rcu_read_lock();
1163 stream = relay_stream_find_by_id(be64toh(stream_info.stream_id));
1164 if (!stream) {
1165 ret = -1;
1166 goto end_unlock;
1167 }
1168
1169 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1170 stream->close_flag = 1;
1171 session->stream_count--;
1172 assert(session->stream_count >= 0);
1173
1174 if (close_stream_check(stream)) {
1175 destroy_stream(stream);
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 ssize_t 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 ret = lttng_write(fd, zeros, size);
1261 if (ret < size) {
1262 PERROR("write padding to file");
1263 }
1264
1265 free(zeros);
1266
1267 end:
1268 return ret;
1269 }
1270
1271 /*
1272 * relay_recv_metadata: receive the metada for the session.
1273 */
1274 static
1275 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1276 struct relay_command *cmd)
1277 {
1278 int ret = htobe32(LTTNG_OK);
1279 ssize_t size_ret;
1280 struct relay_session *session = cmd->session;
1281 struct lttcomm_relayd_metadata_payload *metadata_struct;
1282 struct relay_stream *metadata_stream;
1283 uint64_t data_size, payload_size;
1284
1285 if (!session) {
1286 ERR("Metadata sent before version check");
1287 ret = -1;
1288 goto end;
1289 }
1290
1291 data_size = payload_size = be64toh(recv_hdr->data_size);
1292 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1293 ERR("Incorrect data size");
1294 ret = -1;
1295 goto end;
1296 }
1297 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1298
1299 if (data_buffer_size < data_size) {
1300 /* In case the realloc fails, we can free the memory */
1301 char *tmp_data_ptr;
1302
1303 tmp_data_ptr = realloc(data_buffer, data_size);
1304 if (!tmp_data_ptr) {
1305 ERR("Allocating data buffer");
1306 free(data_buffer);
1307 ret = -1;
1308 goto end;
1309 }
1310 data_buffer = tmp_data_ptr;
1311 data_buffer_size = data_size;
1312 }
1313 memset(data_buffer, 0, data_size);
1314 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1315 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1316 if (ret < 0 || ret != data_size) {
1317 if (ret == 0) {
1318 /* Orderly shutdown. Not necessary to print an error. */
1319 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1320 } else {
1321 ERR("Relay didn't receive the whole metadata");
1322 }
1323 ret = -1;
1324 goto end;
1325 }
1326 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1327
1328 rcu_read_lock();
1329 metadata_stream = relay_stream_find_by_id(
1330 be64toh(metadata_struct->stream_id));
1331 if (!metadata_stream) {
1332 ret = -1;
1333 goto end_unlock;
1334 }
1335
1336 size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
1337 payload_size);
1338 if (size_ret < payload_size) {
1339 ERR("Relay error writing metadata on file");
1340 ret = -1;
1341 goto end_unlock;
1342 }
1343
1344 ret = write_padding_to_file(metadata_stream->fd,
1345 be32toh(metadata_struct->padding_size));
1346 if (ret < 0) {
1347 goto end_unlock;
1348 }
1349 metadata_stream->ctf_trace->metadata_received +=
1350 payload_size + be32toh(metadata_struct->padding_size);
1351
1352 DBG2("Relay metadata written");
1353
1354 end_unlock:
1355 rcu_read_unlock();
1356 end:
1357 return ret;
1358 }
1359
1360 /*
1361 * relay_send_version: send relayd version number
1362 */
1363 static
1364 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1365 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1366 {
1367 int ret;
1368 struct lttcomm_relayd_version reply, msg;
1369
1370 assert(cmd);
1371
1372 cmd->version_check_done = 1;
1373
1374 /* Get version from the other side. */
1375 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1376 if (ret < 0 || ret != sizeof(msg)) {
1377 if (ret == 0) {
1378 /* Orderly shutdown. Not necessary to print an error. */
1379 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1380 } else {
1381 ERR("Relay failed to receive the version values.");
1382 }
1383 ret = -1;
1384 goto end;
1385 }
1386
1387 reply.major = RELAYD_VERSION_COMM_MAJOR;
1388 reply.minor = RELAYD_VERSION_COMM_MINOR;
1389
1390 /* Major versions must be the same */
1391 if (reply.major != be32toh(msg.major)) {
1392 DBG("Incompatible major versions (%u vs %u), deleting session",
1393 reply.major, be32toh(msg.major));
1394 relay_delete_session(cmd, sessions_ht);
1395 ret = 0;
1396 goto end;
1397 }
1398
1399 cmd->major = reply.major;
1400 /* We adapt to the lowest compatible version */
1401 if (reply.minor <= be32toh(msg.minor)) {
1402 cmd->minor = reply.minor;
1403 } else {
1404 cmd->minor = be32toh(msg.minor);
1405 }
1406
1407 reply.major = htobe32(reply.major);
1408 reply.minor = htobe32(reply.minor);
1409 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1410 sizeof(struct lttcomm_relayd_version), 0);
1411 if (ret < 0) {
1412 ERR("Relay sending version");
1413 }
1414
1415 DBG("Version check done using protocol %u.%u", cmd->major,
1416 cmd->minor);
1417
1418 end:
1419 return ret;
1420 }
1421
1422 /*
1423 * Check for data pending for a given stream id from the session daemon.
1424 */
1425 static
1426 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1427 struct relay_command *cmd)
1428 {
1429 struct relay_session *session = cmd->session;
1430 struct lttcomm_relayd_data_pending msg;
1431 struct lttcomm_relayd_generic_reply reply;
1432 struct relay_stream *stream;
1433 int ret;
1434 uint64_t last_net_seq_num, stream_id;
1435
1436 DBG("Data pending command received");
1437
1438 if (!session || cmd->version_check_done == 0) {
1439 ERR("Trying to check for data before version check");
1440 ret = -1;
1441 goto end_no_session;
1442 }
1443
1444 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1445 if (ret < sizeof(msg)) {
1446 if (ret == 0) {
1447 /* Orderly shutdown. Not necessary to print an error. */
1448 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1449 } else {
1450 ERR("Relay didn't receive valid data_pending struct size : %d",
1451 ret);
1452 }
1453 ret = -1;
1454 goto end_no_session;
1455 }
1456
1457 stream_id = be64toh(msg.stream_id);
1458 last_net_seq_num = be64toh(msg.last_net_seq_num);
1459
1460 rcu_read_lock();
1461 stream = relay_stream_find_by_id(stream_id);
1462 if (stream == NULL) {
1463 ret = -1;
1464 goto end_unlock;
1465 }
1466
1467 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1468 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1469 last_net_seq_num);
1470
1471 /* Avoid wrapping issue */
1472 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1473 /* Data has in fact been written and is NOT pending */
1474 ret = 0;
1475 } else {
1476 /* Data still being streamed thus pending */
1477 ret = 1;
1478 }
1479
1480 /* Pending check is now done. */
1481 stream->data_pending_check_done = 1;
1482
1483 end_unlock:
1484 rcu_read_unlock();
1485
1486 reply.ret_code = htobe32(ret);
1487 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1488 if (ret < 0) {
1489 ERR("Relay data pending ret code failed");
1490 }
1491
1492 end_no_session:
1493 return ret;
1494 }
1495
1496 /*
1497 * Wait for the control socket to reach a quiescent state.
1498 *
1499 * Note that for now, when receiving this command from the session daemon, this
1500 * means that every subsequent commands or data received on the control socket
1501 * has been handled. So, this is why we simply return OK here.
1502 */
1503 static
1504 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1505 struct relay_command *cmd)
1506 {
1507 int ret;
1508 uint64_t stream_id;
1509 struct relay_stream *stream;
1510 struct lttng_ht_iter iter;
1511 struct lttcomm_relayd_quiescent_control msg;
1512 struct lttcomm_relayd_generic_reply reply;
1513
1514 DBG("Checking quiescent state on control socket");
1515
1516 if (!cmd->session || cmd->version_check_done == 0) {
1517 ERR("Trying to check for data before version check");
1518 ret = -1;
1519 goto end_no_session;
1520 }
1521
1522 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1523 if (ret < sizeof(msg)) {
1524 if (ret == 0) {
1525 /* Orderly shutdown. Not necessary to print an error. */
1526 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1527 } else {
1528 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1529 ret);
1530 }
1531 ret = -1;
1532 goto end_no_session;
1533 }
1534
1535 stream_id = be64toh(msg.stream_id);
1536
1537 rcu_read_lock();
1538 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1539 stream_n.node) {
1540 if (stream->stream_handle == stream_id) {
1541 stream->data_pending_check_done = 1;
1542 DBG("Relay quiescent control pending flag set to %" PRIu64,
1543 stream_id);
1544 break;
1545 }
1546 }
1547 rcu_read_unlock();
1548
1549 reply.ret_code = htobe32(LTTNG_OK);
1550 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1551 if (ret < 0) {
1552 ERR("Relay data quiescent control ret code failed");
1553 }
1554
1555 end_no_session:
1556 return ret;
1557 }
1558
1559 /*
1560 * Initialize a data pending command. This means that a client is about to ask
1561 * for data pending for each stream he/she holds. Simply iterate over all
1562 * streams of a session and set the data_pending_check_done flag.
1563 *
1564 * This command returns to the client a LTTNG_OK code.
1565 */
1566 static
1567 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1568 struct relay_command *cmd)
1569 {
1570 int ret;
1571 struct lttng_ht_iter iter;
1572 struct lttcomm_relayd_begin_data_pending msg;
1573 struct lttcomm_relayd_generic_reply reply;
1574 struct relay_stream *stream;
1575 uint64_t session_id;
1576
1577 assert(recv_hdr);
1578 assert(cmd);
1579
1580 DBG("Init streams for data pending");
1581
1582 if (!cmd->session || cmd->version_check_done == 0) {
1583 ERR("Trying to check for data before version check");
1584 ret = -1;
1585 goto end_no_session;
1586 }
1587
1588 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1589 if (ret < sizeof(msg)) {
1590 if (ret == 0) {
1591 /* Orderly shutdown. Not necessary to print an error. */
1592 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1593 } else {
1594 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1595 ret);
1596 }
1597 ret = -1;
1598 goto end_no_session;
1599 }
1600
1601 session_id = be64toh(msg.session_id);
1602
1603 /*
1604 * Iterate over all streams to set the begin data pending flag. For now, the
1605 * streams are indexed by stream handle so we have to iterate over all
1606 * streams to find the one associated with the right session_id.
1607 */
1608 rcu_read_lock();
1609 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1610 stream_n.node) {
1611 if (stream->session->id == session_id) {
1612 stream->data_pending_check_done = 0;
1613 DBG("Set begin data pending flag to stream %" PRIu64,
1614 stream->stream_handle);
1615 }
1616 }
1617 rcu_read_unlock();
1618
1619 /* All good, send back reply. */
1620 reply.ret_code = htobe32(LTTNG_OK);
1621
1622 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1623 if (ret < 0) {
1624 ERR("Relay begin data pending send reply failed");
1625 }
1626
1627 end_no_session:
1628 return ret;
1629 }
1630
1631 /*
1632 * End data pending command. This will check, for a given session id, if each
1633 * stream associated with it has its data_pending_check_done flag set. If not,
1634 * this means that the client lost track of the stream but the data is still
1635 * being streamed on our side. In this case, we inform the client that data is
1636 * inflight.
1637 *
1638 * Return to the client if there is data in flight or not with a ret_code.
1639 */
1640 static
1641 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1642 struct relay_command *cmd)
1643 {
1644 int ret;
1645 struct lttng_ht_iter iter;
1646 struct lttcomm_relayd_end_data_pending msg;
1647 struct lttcomm_relayd_generic_reply reply;
1648 struct relay_stream *stream;
1649 uint64_t session_id;
1650 uint32_t is_data_inflight = 0;
1651
1652 assert(recv_hdr);
1653 assert(cmd);
1654
1655 DBG("End data pending command");
1656
1657 if (!cmd->session || cmd->version_check_done == 0) {
1658 ERR("Trying to check for data before version check");
1659 ret = -1;
1660 goto end_no_session;
1661 }
1662
1663 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1664 if (ret < sizeof(msg)) {
1665 if (ret == 0) {
1666 /* Orderly shutdown. Not necessary to print an error. */
1667 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1668 } else {
1669 ERR("Relay didn't receive valid end data_pending struct size: %d",
1670 ret);
1671 }
1672 ret = -1;
1673 goto end_no_session;
1674 }
1675
1676 session_id = be64toh(msg.session_id);
1677
1678 /* Iterate over all streams to see if the begin data pending flag is set. */
1679 rcu_read_lock();
1680 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1681 stream_n.node) {
1682 if (stream->session->id == session_id &&
1683 !stream->data_pending_check_done) {
1684 is_data_inflight = 1;
1685 DBG("Data is still in flight for stream %" PRIu64,
1686 stream->stream_handle);
1687 break;
1688 }
1689 }
1690 rcu_read_unlock();
1691
1692 /* All good, send back reply. */
1693 reply.ret_code = htobe32(is_data_inflight);
1694
1695 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1696 if (ret < 0) {
1697 ERR("Relay end data pending send reply failed");
1698 }
1699
1700 end_no_session:
1701 return ret;
1702 }
1703
1704 /*
1705 * Receive an index for a specific stream.
1706 *
1707 * Return 0 on success else a negative value.
1708 */
1709 static
1710 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1711 struct relay_command *cmd)
1712 {
1713 int ret, send_ret, index_created = 0;
1714 struct relay_session *session = cmd->session;
1715 struct lttcomm_relayd_index index_info;
1716 struct relay_index *index, *wr_index = NULL;
1717 struct lttcomm_relayd_generic_reply reply;
1718 struct relay_stream *stream;
1719 uint64_t net_seq_num;
1720
1721 assert(cmd);
1722
1723 DBG("Relay receiving index");
1724
1725 if (!session || cmd->version_check_done == 0) {
1726 ERR("Trying to close a stream before version check");
1727 ret = -1;
1728 goto end_no_session;
1729 }
1730
1731 ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info,
1732 sizeof(index_info), 0);
1733 if (ret < sizeof(index_info)) {
1734 if (ret == 0) {
1735 /* Orderly shutdown. Not necessary to print an error. */
1736 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1737 } else {
1738 ERR("Relay didn't receive valid index struct size : %d", ret);
1739 }
1740 ret = -1;
1741 goto end_no_session;
1742 }
1743
1744 net_seq_num = be64toh(index_info.net_seq_num);
1745
1746 rcu_read_lock();
1747 stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id));
1748 if (!stream) {
1749 ret = -1;
1750 goto end_rcu_unlock;
1751 }
1752
1753 /* Live beacon handling */
1754 if (index_info.packet_size == 0) {
1755 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1756
1757 /*
1758 * Only flag a stream inactive when it has already received data.
1759 */
1760 if (stream->total_index_received > 0) {
1761 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1762 }
1763 ret = 0;
1764 goto end_rcu_unlock;
1765 } else {
1766 stream->beacon_ts_end = -1ULL;
1767 }
1768
1769 index = relay_index_find(stream->stream_handle, net_seq_num);
1770 if (!index) {
1771 /* A successful creation will add the object to the HT. */
1772 index = relay_index_create(stream->stream_handle, net_seq_num);
1773 if (!index) {
1774 goto end_rcu_unlock;
1775 }
1776 index_created = 1;
1777 }
1778
1779 copy_index_control_data(index, &index_info);
1780
1781 if (index_created) {
1782 /*
1783 * Try to add the relay index object to the hash table. If an object
1784 * already exist, destroy back the index created, set the data in this
1785 * object and write it on disk.
1786 */
1787 relay_index_add(index, &wr_index);
1788 if (wr_index) {
1789 copy_index_control_data(wr_index, &index_info);
1790 free(index);
1791 }
1792 } else {
1793 /* The index already exists so write it on disk. */
1794 wr_index = index;
1795 }
1796
1797 /* Do we have a writable ready index to write on disk. */
1798 if (wr_index) {
1799 /* Starting at 2.4, create the index file if none available. */
1800 if (cmd->minor >= 4 && stream->index_fd < 0) {
1801 ret = index_create_file(stream->path_name, stream->channel_name,
1802 relayd_uid, relayd_gid, stream->tracefile_size,
1803 stream->tracefile_count_current);
1804 if (ret < 0) {
1805 goto end_rcu_unlock;
1806 }
1807 stream->index_fd = ret;
1808 }
1809
1810 ret = relay_index_write(wr_index->fd, wr_index);
1811 if (ret < 0) {
1812 goto end_rcu_unlock;
1813 }
1814 stream->total_index_received++;
1815 }
1816
1817 end_rcu_unlock:
1818 rcu_read_unlock();
1819
1820 if (ret < 0) {
1821 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1822 } else {
1823 reply.ret_code = htobe32(LTTNG_OK);
1824 }
1825 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1826 if (send_ret < 0) {
1827 ERR("Relay sending close index id reply");
1828 ret = send_ret;
1829 }
1830
1831 end_no_session:
1832 return ret;
1833 }
1834
1835 /*
1836 * Process the commands received on the control socket
1837 */
1838 static
1839 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1840 struct relay_command *cmd, struct relay_local_data *ctx)
1841 {
1842 int ret = 0;
1843
1844 switch (be32toh(recv_hdr->cmd)) {
1845 case RELAYD_CREATE_SESSION:
1846 ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht);
1847 break;
1848 case RELAYD_ADD_STREAM:
1849 ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht);
1850 break;
1851 case RELAYD_START_DATA:
1852 ret = relay_start(recv_hdr, cmd);
1853 break;
1854 case RELAYD_SEND_METADATA:
1855 ret = relay_recv_metadata(recv_hdr, cmd);
1856 break;
1857 case RELAYD_VERSION:
1858 ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht);
1859 break;
1860 case RELAYD_CLOSE_STREAM:
1861 ret = relay_close_stream(recv_hdr, cmd);
1862 break;
1863 case RELAYD_DATA_PENDING:
1864 ret = relay_data_pending(recv_hdr, cmd);
1865 break;
1866 case RELAYD_QUIESCENT_CONTROL:
1867 ret = relay_quiescent_control(recv_hdr, cmd);
1868 break;
1869 case RELAYD_BEGIN_DATA_PENDING:
1870 ret = relay_begin_data_pending(recv_hdr, cmd);
1871 break;
1872 case RELAYD_END_DATA_PENDING:
1873 ret = relay_end_data_pending(recv_hdr, cmd);
1874 break;
1875 case RELAYD_SEND_INDEX:
1876 ret = relay_recv_index(recv_hdr, cmd);
1877 break;
1878 case RELAYD_UPDATE_SYNC_INFO:
1879 default:
1880 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1881 relay_unknown_command(cmd);
1882 ret = -1;
1883 goto end;
1884 }
1885
1886 end:
1887 return ret;
1888 }
1889
1890 /*
1891 * Handle index for a data stream.
1892 *
1893 * RCU read side lock MUST be acquired.
1894 *
1895 * Return 0 on success else a negative value.
1896 */
1897 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
1898 int rotate_index)
1899 {
1900 int ret = 0, index_created = 0;
1901 uint64_t stream_id, data_offset;
1902 struct relay_index *index, *wr_index = NULL;
1903
1904 assert(stream);
1905
1906 stream_id = stream->stream_handle;
1907 /* Get data offset because we are about to update the index. */
1908 data_offset = htobe64(stream->tracefile_size_current);
1909
1910 /*
1911 * Lookup for an existing index for that stream id/sequence number. If on
1912 * exists, the control thread already received the data for it thus we need
1913 * to write it on disk.
1914 */
1915 index = relay_index_find(stream_id, net_seq_num);
1916 if (!index) {
1917 /* A successful creation will add the object to the HT. */
1918 index = relay_index_create(stream_id, net_seq_num);
1919 if (!index) {
1920 ret = -1;
1921 goto error;
1922 }
1923 index_created = 1;
1924 }
1925
1926 if (rotate_index || stream->index_fd < 0) {
1927 index->to_close_fd = stream->index_fd;
1928 ret = index_create_file(stream->path_name, stream->channel_name,
1929 relayd_uid, relayd_gid, stream->tracefile_size,
1930 stream->tracefile_count_current);
1931 if (ret < 0) {
1932 /* This will close the stream's index fd if one. */
1933 relay_index_free_safe(index);
1934 goto error;
1935 }
1936 stream->index_fd = ret;
1937 }
1938 index->fd = stream->index_fd;
1939 index->index_data.offset = data_offset;
1940
1941 if (index_created) {
1942 /*
1943 * Try to add the relay index object to the hash table. If an object
1944 * already exist, destroy back the index created and set the data.
1945 */
1946 relay_index_add(index, &wr_index);
1947 if (wr_index) {
1948 /* Copy back data from the created index. */
1949 wr_index->fd = index->fd;
1950 wr_index->to_close_fd = index->to_close_fd;
1951 wr_index->index_data.offset = data_offset;
1952 free(index);
1953 }
1954 } else {
1955 /* The index already exists so write it on disk. */
1956 wr_index = index;
1957 }
1958
1959 /* Do we have a writable ready index to write on disk. */
1960 if (wr_index) {
1961 ret = relay_index_write(wr_index->fd, wr_index);
1962 if (ret < 0) {
1963 goto error;
1964 }
1965 stream->total_index_received++;
1966 }
1967
1968 error:
1969 return ret;
1970 }
1971
1972 /*
1973 * relay_process_data: Process the data received on the data socket
1974 */
1975 static
1976 int relay_process_data(struct relay_command *cmd)
1977 {
1978 int ret = 0, rotate_index = 0;
1979 ssize_t size_ret;
1980 struct relay_stream *stream;
1981 struct lttcomm_relayd_data_hdr data_hdr;
1982 uint64_t stream_id;
1983 uint64_t net_seq_num;
1984 uint32_t data_size;
1985
1986 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1987 sizeof(struct lttcomm_relayd_data_hdr), 0);
1988 if (ret <= 0) {
1989 if (ret == 0) {
1990 /* Orderly shutdown. Not necessary to print an error. */
1991 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1992 } else {
1993 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
1994 }
1995 ret = -1;
1996 goto end;
1997 }
1998
1999 stream_id = be64toh(data_hdr.stream_id);
2000
2001 rcu_read_lock();
2002 stream = relay_stream_find_by_id(stream_id);
2003 if (!stream) {
2004 ret = -1;
2005 goto end_rcu_unlock;
2006 }
2007
2008 data_size = be32toh(data_hdr.data_size);
2009 if (data_buffer_size < data_size) {
2010 char *tmp_data_ptr;
2011
2012 tmp_data_ptr = realloc(data_buffer, data_size);
2013 if (!tmp_data_ptr) {
2014 ERR("Allocating data buffer");
2015 free(data_buffer);
2016 ret = -1;
2017 goto end_rcu_unlock;
2018 }
2019 data_buffer = tmp_data_ptr;
2020 data_buffer_size = data_size;
2021 }
2022 memset(data_buffer, 0, data_size);
2023
2024 net_seq_num = be64toh(data_hdr.net_seq_num);
2025
2026 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2027 data_size, stream_id, net_seq_num);
2028 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
2029 if (ret <= 0) {
2030 if (ret == 0) {
2031 /* Orderly shutdown. Not necessary to print an error. */
2032 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2033 }
2034 ret = -1;
2035 goto end_rcu_unlock;
2036 }
2037
2038 /* Check if a rotation is needed. */
2039 if (stream->tracefile_size > 0 &&
2040 (stream->tracefile_size_current + data_size) >
2041 stream->tracefile_size) {
2042 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2043 stream->tracefile_size, stream->tracefile_count,
2044 relayd_uid, relayd_gid, stream->fd,
2045 &(stream->tracefile_count_current), &stream->fd);
2046 if (ret < 0) {
2047 ERR("Rotating stream output file");
2048 goto end_rcu_unlock;
2049 }
2050 /* Reset current size because we just perform a stream rotation. */
2051 stream->tracefile_size_current = 0;
2052 rotate_index = 1;
2053 }
2054
2055 /*
2056 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2057 * index are NOT supported.
2058 */
2059 if (stream->session->minor >= 4 && !stream->session->snapshot) {
2060 ret = handle_index_data(stream, net_seq_num, rotate_index);
2061 if (ret < 0) {
2062 goto end_rcu_unlock;
2063 }
2064 }
2065
2066 /* Write data to stream output fd. */
2067 size_ret = lttng_write(stream->fd, data_buffer, data_size);
2068 if (size_ret < data_size) {
2069 ERR("Relay error writing data to file");
2070 ret = -1;
2071 goto end_rcu_unlock;
2072 }
2073
2074 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2075 ret, stream->stream_handle);
2076
2077 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2078 if (ret < 0) {
2079 goto end_rcu_unlock;
2080 }
2081 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2082
2083 stream->prev_seq = net_seq_num;
2084
2085 /* Check if we need to close the FD */
2086 if (close_stream_check(stream)) {
2087 destroy_stream(stream);
2088 }
2089
2090 end_rcu_unlock:
2091 rcu_read_unlock();
2092 end:
2093 return ret;
2094 }
2095
2096 static
2097 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
2098 {
2099 int ret;
2100
2101 lttng_poll_del(events, pollfd);
2102
2103 ret = close(pollfd);
2104 if (ret < 0) {
2105 ERR("Closing pollfd %d", pollfd);
2106 }
2107 }
2108
2109 static
2110 int relay_add_connection(int fd, struct lttng_poll_event *events,
2111 struct lttng_ht *relay_connections_ht)
2112 {
2113 struct relay_command *relay_connection;
2114 ssize_t ret;
2115
2116 relay_connection = zmalloc(sizeof(struct relay_command));
2117 if (relay_connection == NULL) {
2118 PERROR("Relay command zmalloc");
2119 goto error;
2120 }
2121 ret = lttng_read(fd, relay_connection, sizeof(struct relay_command));
2122 if (ret < sizeof(struct relay_command)) {
2123 PERROR("read relay cmd pipe");
2124 goto error_read;
2125 }
2126
2127 /*
2128 * Only used by the control side and the reference is copied inside each
2129 * stream from that connection. Thus a destroy HT must be done after every
2130 * stream has been destroyed.
2131 */
2132 if (relay_connection->type == RELAY_CONTROL) {
2133 relay_connection->ctf_traces_ht = lttng_ht_new(0,
2134 LTTNG_HT_TYPE_STRING);
2135 if (!relay_connection->ctf_traces_ht) {
2136 goto error_read;
2137 }
2138 }
2139
2140 lttng_ht_node_init_ulong(&relay_connection->sock_n,
2141 (unsigned long) relay_connection->sock->fd);
2142 rcu_read_lock();
2143 lttng_ht_add_unique_ulong(relay_connections_ht,
2144 &relay_connection->sock_n);
2145 rcu_read_unlock();
2146 return lttng_poll_add(events,
2147 relay_connection->sock->fd,
2148 LPOLLIN | LPOLLRDHUP);
2149
2150 error_read:
2151 free(relay_connection);
2152 error:
2153 return -1;
2154 }
2155
2156 static
2157 void deferred_free_connection(struct rcu_head *head)
2158 {
2159 struct relay_command *relay_connection =
2160 caa_container_of(head, struct relay_command, rcu_node);
2161
2162 lttcomm_destroy_sock(relay_connection->sock);
2163 free(relay_connection);
2164 }
2165
2166 static
2167 void relay_del_connection(struct lttng_ht *relay_connections_ht,
2168 struct lttng_ht_iter *iter, struct relay_command *relay_connection,
2169 struct lttng_ht *sessions_ht)
2170 {
2171 int ret;
2172
2173 ret = lttng_ht_del(relay_connections_ht, iter);
2174 assert(!ret);
2175
2176 if (relay_connection->type == RELAY_CONTROL) {
2177 relay_delete_session(relay_connection, sessions_ht);
2178 lttng_ht_destroy(relay_connection->ctf_traces_ht);
2179 }
2180
2181 call_rcu(&relay_connection->rcu_node, deferred_free_connection);
2182 }
2183
2184 /*
2185 * This thread does the actual work
2186 */
2187 static
2188 void *relay_thread_worker(void *data)
2189 {
2190 int ret, err = -1, last_seen_data_fd = -1;
2191 uint32_t nb_fd;
2192 struct relay_command *relay_connection;
2193 struct lttng_poll_event events;
2194 struct lttng_ht *relay_connections_ht;
2195 struct lttng_ht_node_ulong *node;
2196 struct lttng_ht_iter iter;
2197 struct lttcomm_relayd_hdr recv_hdr;
2198 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2199 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2200
2201 DBG("[thread] Relay worker started");
2202
2203 rcu_register_thread();
2204
2205 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2206
2207 health_code_update();
2208
2209 /* table of connections indexed on socket */
2210 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2211 if (!relay_connections_ht) {
2212 goto relay_connections_ht_error;
2213 }
2214
2215 /* Tables of received indexes indexed by index handle and net_seq_num. */
2216 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2217 if (!indexes_ht) {
2218 goto indexes_ht_error;
2219 }
2220
2221 ret = create_thread_poll_set(&events, 2);
2222 if (ret < 0) {
2223 goto error_poll_create;
2224 }
2225
2226 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
2227 if (ret < 0) {
2228 goto error;
2229 }
2230
2231 restart:
2232 while (1) {
2233 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2234
2235 health_code_update();
2236
2237 /* Infinite blocking call, waiting for transmission */
2238 DBG3("Relayd worker thread polling...");
2239 health_poll_entry();
2240 ret = lttng_poll_wait(&events, -1);
2241 health_poll_exit();
2242 if (ret < 0) {
2243 /*
2244 * Restart interrupted system call.
2245 */
2246 if (errno == EINTR) {
2247 goto restart;
2248 }
2249 goto error;
2250 }
2251
2252 nb_fd = ret;
2253
2254 /*
2255 * Process control. The control connection is prioritised so we don't
2256 * starve it with high throughout put tracing data on the data
2257 * connection.
2258 */
2259 for (i = 0; i < nb_fd; i++) {
2260 /* Fetch once the poll data */
2261 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2262 int pollfd = LTTNG_POLL_GETFD(&events, i);
2263
2264 health_code_update();
2265
2266 /* Thread quit pipe has been closed. Killing thread. */
2267 ret = check_thread_quit_pipe(pollfd, revents);
2268 if (ret) {
2269 err = 0;
2270 goto exit;
2271 }
2272
2273 /* Inspect the relay cmd pipe for new connection */
2274 if (pollfd == relay_cmd_pipe[0]) {
2275 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2276 ERR("Relay pipe error");
2277 goto error;
2278 } else if (revents & LPOLLIN) {
2279 DBG("Relay command received");
2280 ret = relay_add_connection(relay_cmd_pipe[0],
2281 &events, relay_connections_ht);
2282 if (ret < 0) {
2283 goto error;
2284 }
2285 }
2286 } else if (revents) {
2287 rcu_read_lock();
2288 lttng_ht_lookup(relay_connections_ht,
2289 (void *)((unsigned long) pollfd),
2290 &iter);
2291 node = lttng_ht_iter_get_node_ulong(&iter);
2292 if (node == NULL) {
2293 DBG2("Relay sock %d not found", pollfd);
2294 rcu_read_unlock();
2295 goto error;
2296 }
2297 relay_connection = caa_container_of(node,
2298 struct relay_command, sock_n);
2299
2300 if (revents & (LPOLLERR)) {
2301 ERR("POLL ERROR");
2302 relay_cleanup_poll_connection(&events, pollfd);
2303 relay_del_connection(relay_connections_ht,
2304 &iter, relay_connection, sessions_ht);
2305 if (last_seen_data_fd == pollfd) {
2306 last_seen_data_fd = last_notdel_data_fd;
2307 }
2308 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2309 DBG("Socket %d hung up", pollfd);
2310 relay_cleanup_poll_connection(&events, pollfd);
2311 relay_del_connection(relay_connections_ht,
2312 &iter, relay_connection, sessions_ht);
2313 if (last_seen_data_fd == pollfd) {
2314 last_seen_data_fd = last_notdel_data_fd;
2315 }
2316 } else if (revents & LPOLLIN) {
2317 /* control socket */
2318 if (relay_connection->type == RELAY_CONTROL) {
2319 ret = relay_connection->sock->ops->recvmsg(
2320 relay_connection->sock, &recv_hdr,
2321 sizeof(struct lttcomm_relayd_hdr), 0);
2322 /* connection closed */
2323 if (ret <= 0) {
2324 relay_cleanup_poll_connection(&events, pollfd);
2325 relay_del_connection(relay_connections_ht,
2326 &iter, relay_connection, sessions_ht);
2327 DBG("Control connection closed with %d", pollfd);
2328 } else {
2329 if (relay_connection->session) {
2330 DBG2("Relay worker receiving data for session : %" PRIu64,
2331 relay_connection->session->id);
2332 }
2333 ret = relay_process_control(&recv_hdr,
2334 relay_connection, relay_ctx);
2335 if (ret < 0) {
2336 /* Clear the session on error. */
2337 relay_cleanup_poll_connection(&events, pollfd);
2338 relay_del_connection(relay_connections_ht,
2339 &iter, relay_connection, sessions_ht);
2340 DBG("Connection closed with %d", pollfd);
2341 }
2342 seen_control = 1;
2343 }
2344 } else {
2345 /*
2346 * Flag the last seen data fd not deleted. It will be
2347 * used as the last seen fd if any fd gets deleted in
2348 * this first loop.
2349 */
2350 last_notdel_data_fd = pollfd;
2351 }
2352 }
2353 rcu_read_unlock();
2354 }
2355 }
2356
2357 /*
2358 * The last loop handled a control request, go back to poll to make
2359 * sure we prioritise the control socket.
2360 */
2361 if (seen_control) {
2362 continue;
2363 }
2364
2365 if (last_seen_data_fd >= 0) {
2366 for (i = 0; i < nb_fd; i++) {
2367 int pollfd = LTTNG_POLL_GETFD(&events, i);
2368
2369 health_code_update();
2370
2371 if (last_seen_data_fd == pollfd) {
2372 idx = i;
2373 break;
2374 }
2375 }
2376 }
2377
2378 /* Process data connection. */
2379 for (i = idx + 1; i < nb_fd; i++) {
2380 /* Fetch the poll data. */
2381 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2382 int pollfd = LTTNG_POLL_GETFD(&events, i);
2383
2384 health_code_update();
2385
2386 /* Skip the command pipe. It's handled in the first loop. */
2387 if (pollfd == relay_cmd_pipe[0]) {
2388 continue;
2389 }
2390
2391 if (revents) {
2392 rcu_read_lock();
2393 lttng_ht_lookup(relay_connections_ht,
2394 (void *)((unsigned long) pollfd),
2395 &iter);
2396 node = lttng_ht_iter_get_node_ulong(&iter);
2397 if (node == NULL) {
2398 /* Skip it. Might be removed before. */
2399 rcu_read_unlock();
2400 continue;
2401 }
2402 relay_connection = caa_container_of(node,
2403 struct relay_command, sock_n);
2404
2405 if (revents & LPOLLIN) {
2406 if (relay_connection->type != RELAY_DATA) {
2407 continue;
2408 }
2409
2410 ret = relay_process_data(relay_connection);
2411 /* connection closed */
2412 if (ret < 0) {
2413 relay_cleanup_poll_connection(&events, pollfd);
2414 relay_del_connection(relay_connections_ht,
2415 &iter, relay_connection, sessions_ht);
2416 DBG("Data connection closed with %d", pollfd);
2417 /*
2418 * Every goto restart call sets the last seen fd where
2419 * here we don't really care since we gracefully
2420 * continue the loop after the connection is deleted.
2421 */
2422 } else {
2423 /* Keep last seen port. */
2424 last_seen_data_fd = pollfd;
2425 rcu_read_unlock();
2426 goto restart;
2427 }
2428 }
2429 rcu_read_unlock();
2430 }
2431 }
2432 last_seen_data_fd = -1;
2433 }
2434
2435 /* Normal exit, no error */
2436 ret = 0;
2437
2438 exit:
2439 error:
2440 lttng_poll_clean(&events);
2441
2442 /* empty the hash table and free the memory */
2443 rcu_read_lock();
2444 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2445 health_code_update();
2446
2447 node = lttng_ht_iter_get_node_ulong(&iter);
2448 if (node) {
2449 relay_connection = caa_container_of(node,
2450 struct relay_command, sock_n);
2451 relay_del_connection(relay_connections_ht,
2452 &iter, relay_connection, sessions_ht);
2453 }
2454 }
2455 rcu_read_unlock();
2456 error_poll_create:
2457 lttng_ht_destroy(indexes_ht);
2458 indexes_ht_error:
2459 lttng_ht_destroy(relay_connections_ht);
2460 relay_connections_ht_error:
2461 /* Close relay cmd pipes */
2462 utils_close_pipe(relay_cmd_pipe);
2463 if (err) {
2464 DBG("Thread exited with error");
2465 }
2466 DBG("Worker thread cleanup complete");
2467 free(data_buffer);
2468 if (err) {
2469 health_error();
2470 ERR("Health error occurred in %s", __func__);
2471 }
2472 health_unregister(health_relayd);
2473 rcu_unregister_thread();
2474 stop_threads();
2475 return NULL;
2476 }
2477
2478 /*
2479 * Create the relay command pipe to wake thread_manage_apps.
2480 * Closed in cleanup().
2481 */
2482 static int create_relay_cmd_pipe(void)
2483 {
2484 int ret;
2485
2486 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2487
2488 return ret;
2489 }
2490
2491 /*
2492 * main
2493 */
2494 int main(int argc, char **argv)
2495 {
2496 int ret = 0;
2497 void *status;
2498 struct relay_local_data *relay_ctx;
2499
2500 /* Create thread quit pipe */
2501 if ((ret = init_thread_quit_pipe()) < 0) {
2502 goto error;
2503 }
2504
2505 /* Parse arguments */
2506 progname = argv[0];
2507 if ((ret = parse_args(argc, argv)) < 0) {
2508 goto exit;
2509 }
2510
2511 if ((ret = set_signal_handler()) < 0) {
2512 goto exit;
2513 }
2514
2515 /* Try to create directory if -o, --output is specified. */
2516 if (opt_output_path) {
2517 if (*opt_output_path != '/') {
2518 ERR("Please specify an absolute path for -o, --output PATH");
2519 goto exit;
2520 }
2521
2522 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2523 if (ret < 0) {
2524 ERR("Unable to create %s", opt_output_path);
2525 goto exit;
2526 }
2527 }
2528
2529 /* Daemonize */
2530 if (opt_daemon) {
2531 ret = daemon(0, 0);
2532 if (ret < 0) {
2533 PERROR("daemon");
2534 goto exit;
2535 }
2536 }
2537
2538 /* We need those values for the file/dir creation. */
2539 relayd_uid = getuid();
2540 relayd_gid = getgid();
2541
2542 /* Check if daemon is UID = 0 */
2543 if (relayd_uid == 0) {
2544 if (control_uri->port < 1024 || data_uri->port < 1024) {
2545 ERR("Need to be root to use ports < 1024");
2546 ret = -1;
2547 goto exit;
2548 }
2549 }
2550
2551 /* Setup the thread apps communication pipe. */
2552 if ((ret = create_relay_cmd_pipe()) < 0) {
2553 goto exit;
2554 }
2555
2556 /* Init relay command queue. */
2557 cds_wfq_init(&relay_cmd_queue.queue);
2558
2559 /* Set up max poll set size */
2560 lttng_poll_set_max_size();
2561
2562 /* Initialize communication library */
2563 lttcomm_init();
2564
2565 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2566 if (!relay_ctx) {
2567 PERROR("relay_ctx");
2568 goto exit;
2569 }
2570
2571 /* tables of sessions indexed by session ID */
2572 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2573 if (!relay_ctx->sessions_ht) {
2574 goto exit_relay_ctx_sessions;
2575 }
2576
2577 /* tables of streams indexed by stream ID */
2578 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2579 if (!relay_streams_ht) {
2580 goto exit_relay_ctx_streams;
2581 }
2582
2583 /* tables of streams indexed by stream ID */
2584 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2585 if (!viewer_streams_ht) {
2586 goto exit_relay_ctx_viewer_streams;
2587 }
2588
2589 /* Initialize thread health monitoring */
2590 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2591 if (!health_relayd) {
2592 PERROR("health_app_create error");
2593 goto exit_health_app_create;
2594 }
2595
2596 ret = utils_create_pipe(health_quit_pipe);
2597 if (ret < 0) {
2598 goto error_health_pipe;
2599 }
2600
2601 /* Create thread to manage the client socket */
2602 ret = pthread_create(&health_thread, NULL,
2603 thread_manage_health, (void *) NULL);
2604 if (ret != 0) {
2605 PERROR("pthread_create health");
2606 goto health_error;
2607 }
2608
2609 /* Setup the dispatcher thread */
2610 ret = pthread_create(&dispatcher_thread, NULL,
2611 relay_thread_dispatcher, (void *) NULL);
2612 if (ret != 0) {
2613 PERROR("pthread_create dispatcher");
2614 goto exit_dispatcher;
2615 }
2616
2617 /* Setup the worker thread */
2618 ret = pthread_create(&worker_thread, NULL,
2619 relay_thread_worker, (void *) relay_ctx);
2620 if (ret != 0) {
2621 PERROR("pthread_create worker");
2622 goto exit_worker;
2623 }
2624
2625 /* Setup the listener thread */
2626 ret = pthread_create(&listener_thread, NULL,
2627 relay_thread_listener, (void *) NULL);
2628 if (ret != 0) {
2629 PERROR("pthread_create listener");
2630 goto exit_listener;
2631 }
2632
2633 ret = live_start_threads(live_uri, relay_ctx, thread_quit_pipe);
2634 if (ret != 0) {
2635 ERR("Starting live viewer threads");
2636 goto exit_live;
2637 }
2638
2639 exit_live:
2640 ret = pthread_join(listener_thread, &status);
2641 if (ret != 0) {
2642 PERROR("pthread_join");
2643 goto error; /* join error, exit without cleanup */
2644 }
2645
2646 exit_listener:
2647 ret = pthread_join(worker_thread, &status);
2648 if (ret != 0) {
2649 PERROR("pthread_join");
2650 goto error; /* join error, exit without cleanup */
2651 }
2652
2653 exit_worker:
2654 ret = pthread_join(dispatcher_thread, &status);
2655 if (ret != 0) {
2656 PERROR("pthread_join");
2657 goto error; /* join error, exit without cleanup */
2658 }
2659
2660 exit_dispatcher:
2661 ret = pthread_join(health_thread, &status);
2662 if (ret != 0) {
2663 PERROR("pthread_join health thread");
2664 goto error; /* join error, exit without cleanup */
2665 }
2666
2667 /*
2668 * Stop live threads only after joining other threads.
2669 */
2670 live_stop_threads();
2671
2672 health_error:
2673 utils_close_pipe(health_quit_pipe);
2674
2675 error_health_pipe:
2676 health_app_destroy(health_relayd);
2677
2678 exit_health_app_create:
2679 lttng_ht_destroy(viewer_streams_ht);
2680
2681 exit_relay_ctx_viewer_streams:
2682 lttng_ht_destroy(relay_streams_ht);
2683
2684 exit_relay_ctx_streams:
2685 lttng_ht_destroy(relay_ctx->sessions_ht);
2686
2687 exit_relay_ctx_sessions:
2688 free(relay_ctx);
2689
2690 exit:
2691 cleanup();
2692 if (!ret) {
2693 exit(EXIT_SUCCESS);
2694 }
2695
2696 error:
2697 exit(EXIT_FAILURE);
2698 }
This page took 0.125447 seconds and 5 git commands to generate.