Fix concurrency issues while overwriting tracefiles in live
[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 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
853 vstream->total_index_received = stream->total_index_received;
854 vstream->close_write_flag = 1;
855 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
856 }
857
858 /* Cleanup index of that stream. */
859 relay_index_destroy_by_stream_id(stream->stream_handle);
860
861 iter.iter.node = &stream->stream_n.node;
862 delret = lttng_ht_del(relay_streams_ht, &iter);
863 assert(!delret);
864 iter.iter.node = &stream->ctf_trace_node.node;
865 delret = lttng_ht_del(stream->ctf_traces_ht, &iter);
866 assert(!delret);
867 call_rcu(&stream->rcu_node, deferred_free_stream);
868 DBG("Closed tracefile %d from close stream", stream->fd);
869 }
870
871 /*
872 * relay_delete_session: Free all memory associated with a session and
873 * close all the FDs
874 */
875 static
876 void relay_delete_session(struct relay_command *cmd,
877 struct lttng_ht *sessions_ht)
878 {
879 struct lttng_ht_iter iter;
880 struct lttng_ht_node_ulong *node;
881 struct relay_stream *stream;
882 int ret;
883
884 if (!cmd->session) {
885 return;
886 }
887
888 DBG("Relay deleting session %" PRIu64, cmd->session->id);
889
890 rcu_read_lock();
891 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
892 node = lttng_ht_iter_get_node_ulong(&iter);
893 if (!node) {
894 continue;
895 }
896 stream = caa_container_of(node, struct relay_stream, stream_n);
897 if (stream->session == cmd->session) {
898 destroy_stream(stream);
899 cmd->session->stream_count--;
900 assert(cmd->session->stream_count >= 0);
901 }
902 }
903
904 /* Make this session not visible anymore. */
905 iter.iter.node = &cmd->session->session_n.node;
906 ret = lttng_ht_del(sessions_ht, &iter);
907 assert(!ret);
908 call_rcu(&cmd->session->rcu_node, deferred_free_session);
909 rcu_read_unlock();
910 }
911
912 /*
913 * Copy index data from the control port to a given index object.
914 */
915 static void copy_index_control_data(struct relay_index *index,
916 struct lttcomm_relayd_index *data)
917 {
918 assert(index);
919 assert(data);
920
921 /*
922 * The index on disk is encoded in big endian, so we don't need to convert
923 * the data received on the network. The data_offset value is NEVER
924 * modified here and is updated by the data thread.
925 */
926 index->index_data.packet_size = data->packet_size;
927 index->index_data.content_size = data->content_size;
928 index->index_data.timestamp_begin = data->timestamp_begin;
929 index->index_data.timestamp_end = data->timestamp_end;
930 index->index_data.events_discarded = data->events_discarded;
931 index->index_data.stream_id = data->stream_id;
932 }
933
934 /*
935 * Handle the RELAYD_CREATE_SESSION command.
936 *
937 * On success, send back the session id or else return a negative value.
938 */
939 static
940 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
941 struct relay_command *cmd,
942 struct lttng_ht *sessions_ht)
943 {
944 int ret = 0, send_ret;
945 struct relay_session *session;
946 struct lttcomm_relayd_status_session reply;
947
948 assert(recv_hdr);
949 assert(cmd);
950
951 memset(&reply, 0, sizeof(reply));
952
953 session = zmalloc(sizeof(struct relay_session));
954 if (session == NULL) {
955 PERROR("relay session zmalloc");
956 ret = -1;
957 goto error;
958 }
959
960 session->id = ++last_relay_session_id;
961 session->sock = cmd->sock;
962 session->minor = cmd->minor;
963 session->major = cmd->major;
964 cmd->session = session;
965
966 reply.session_id = htobe64(session->id);
967
968 switch (cmd->minor) {
969 case 4: /* LTTng sessiond 2.4 */
970 default:
971 ret = cmd_create_session_2_4(cmd, session);
972 break;
973 }
974
975 lttng_ht_node_init_ulong(&session->session_n,
976 (unsigned long) session->id);
977 lttng_ht_add_unique_ulong(sessions_ht,
978 &session->session_n);
979
980 DBG("Created session %" PRIu64, session->id);
981
982 error:
983 if (ret < 0) {
984 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
985 } else {
986 reply.ret_code = htobe32(LTTNG_OK);
987 }
988
989 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
990 if (send_ret < 0) {
991 ERR("Relayd sending session id");
992 ret = send_ret;
993 }
994
995 return ret;
996 }
997
998 /*
999 * relay_add_stream: allocate a new stream for a session
1000 */
1001 static
1002 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
1003 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1004 {
1005 struct relay_session *session = cmd->session;
1006 struct relay_stream *stream = NULL;
1007 struct lttcomm_relayd_status_stream reply;
1008 int ret, send_ret;
1009
1010 if (!session || cmd->version_check_done == 0) {
1011 ERR("Trying to add a stream before version check");
1012 ret = -1;
1013 goto end_no_session;
1014 }
1015
1016 stream = zmalloc(sizeof(struct relay_stream));
1017 if (stream == NULL) {
1018 PERROR("relay stream zmalloc");
1019 ret = -1;
1020 goto end_no_session;
1021 }
1022
1023 switch (cmd->minor) {
1024 case 1: /* LTTng sessiond 2.1 */
1025 ret = cmd_recv_stream_2_1(cmd, stream);
1026 break;
1027 case 2: /* LTTng sessiond 2.2 */
1028 default:
1029 ret = cmd_recv_stream_2_2(cmd, stream);
1030 break;
1031 }
1032 if (ret < 0) {
1033 goto err_free_stream;
1034 }
1035
1036 rcu_read_lock();
1037 stream->stream_handle = ++last_relay_stream_id;
1038 stream->prev_seq = -1ULL;
1039 stream->session = session;
1040 stream->index_fd = -1;
1041 stream->read_index_fd = -1;
1042 stream->ctf_trace = NULL;
1043 pthread_mutex_init(&stream->lock, NULL);
1044
1045 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
1046 if (ret < 0) {
1047 ERR("relay creating output directory");
1048 goto end;
1049 }
1050
1051 /*
1052 * No need to use run_as API here because whatever we receives, the relayd
1053 * uses its own credentials for the stream files.
1054 */
1055 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1056 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1057 if (ret < 0) {
1058 ERR("Create output file");
1059 goto end;
1060 }
1061 stream->fd = ret;
1062 if (stream->tracefile_size) {
1063 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1064 } else {
1065 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1066 }
1067
1068 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1069 stream->metadata_flag = 1;
1070 /*
1071 * When we receive a new metadata stream, we create a new
1072 * ctf_trace and we assign this ctf_trace to all streams with
1073 * the same path.
1074 *
1075 * If later on we receive a new stream for the same ctf_trace,
1076 * we copy the information from the first hit in the HT to the
1077 * new stream.
1078 */
1079 stream->ctf_trace = ctf_trace_create();
1080 if (!stream->ctf_trace) {
1081 ret = -1;
1082 goto end;
1083 }
1084 stream->ctf_trace->refcount++;
1085 stream->ctf_trace->metadata_stream = stream;
1086 }
1087 ctf_trace_assign(cmd->ctf_traces_ht, stream);
1088 stream->ctf_traces_ht = cmd->ctf_traces_ht;
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 session->stream_count++;
1098
1099 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1100 stream->stream_handle);
1101
1102 end:
1103 reply.handle = htobe64(stream->stream_handle);
1104 /* send the session id to the client or a negative return code on error */
1105 if (ret < 0) {
1106 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1107 /* stream was not properly added to the ht, so free it */
1108 free(stream);
1109 } else {
1110 reply.ret_code = htobe32(LTTNG_OK);
1111 }
1112
1113 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1114 sizeof(struct lttcomm_relayd_status_stream), 0);
1115 if (send_ret < 0) {
1116 ERR("Relay sending stream id");
1117 ret = send_ret;
1118 }
1119 rcu_read_unlock();
1120
1121 end_no_session:
1122 return ret;
1123
1124 err_free_stream:
1125 free(stream->path_name);
1126 free(stream->channel_name);
1127 free(stream);
1128 return ret;
1129 }
1130
1131 /*
1132 * relay_close_stream: close a specific stream
1133 */
1134 static
1135 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1136 struct relay_command *cmd)
1137 {
1138 int ret, send_ret;
1139 struct relay_session *session = cmd->session;
1140 struct lttcomm_relayd_close_stream stream_info;
1141 struct lttcomm_relayd_generic_reply reply;
1142 struct relay_stream *stream;
1143
1144 DBG("Close stream received");
1145
1146 if (!session || cmd->version_check_done == 0) {
1147 ERR("Trying to close a stream before version check");
1148 ret = -1;
1149 goto end_no_session;
1150 }
1151
1152 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1153 sizeof(struct lttcomm_relayd_close_stream), 0);
1154 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1155 if (ret == 0) {
1156 /* Orderly shutdown. Not necessary to print an error. */
1157 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1158 } else {
1159 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1160 }
1161 ret = -1;
1162 goto end_no_session;
1163 }
1164
1165 rcu_read_lock();
1166 stream = relay_stream_find_by_id(be64toh(stream_info.stream_id));
1167 if (!stream) {
1168 ret = -1;
1169 goto end_unlock;
1170 }
1171
1172 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1173 stream->close_flag = 1;
1174 session->stream_count--;
1175 assert(session->stream_count >= 0);
1176
1177 if (close_stream_check(stream)) {
1178 destroy_stream(stream);
1179 }
1180
1181 end_unlock:
1182 rcu_read_unlock();
1183
1184 if (ret < 0) {
1185 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1186 } else {
1187 reply.ret_code = htobe32(LTTNG_OK);
1188 }
1189 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1190 sizeof(struct lttcomm_relayd_generic_reply), 0);
1191 if (send_ret < 0) {
1192 ERR("Relay sending stream id");
1193 ret = send_ret;
1194 }
1195
1196 end_no_session:
1197 return ret;
1198 }
1199
1200 /*
1201 * relay_unknown_command: send -1 if received unknown command
1202 */
1203 static
1204 void relay_unknown_command(struct relay_command *cmd)
1205 {
1206 struct lttcomm_relayd_generic_reply reply;
1207 int ret;
1208
1209 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1210 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1211 sizeof(struct lttcomm_relayd_generic_reply), 0);
1212 if (ret < 0) {
1213 ERR("Relay sending unknown command");
1214 }
1215 }
1216
1217 /*
1218 * relay_start: send an acknowledgment to the client to tell if we are
1219 * ready to receive data. We are ready if a session is established.
1220 */
1221 static
1222 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1223 struct relay_command *cmd)
1224 {
1225 int ret = htobe32(LTTNG_OK);
1226 struct lttcomm_relayd_generic_reply reply;
1227 struct relay_session *session = cmd->session;
1228
1229 if (!session) {
1230 DBG("Trying to start the streaming without a session established");
1231 ret = htobe32(LTTNG_ERR_UNK);
1232 }
1233
1234 reply.ret_code = ret;
1235 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1236 sizeof(struct lttcomm_relayd_generic_reply), 0);
1237 if (ret < 0) {
1238 ERR("Relay sending start ack");
1239 }
1240
1241 return ret;
1242 }
1243
1244 /*
1245 * Append padding to the file pointed by the file descriptor fd.
1246 */
1247 static int write_padding_to_file(int fd, uint32_t size)
1248 {
1249 ssize_t ret = 0;
1250 char *zeros;
1251
1252 if (size == 0) {
1253 goto end;
1254 }
1255
1256 zeros = zmalloc(size);
1257 if (zeros == NULL) {
1258 PERROR("zmalloc zeros for padding");
1259 ret = -1;
1260 goto end;
1261 }
1262
1263 ret = lttng_write(fd, zeros, size);
1264 if (ret < size) {
1265 PERROR("write padding to file");
1266 }
1267
1268 free(zeros);
1269
1270 end:
1271 return ret;
1272 }
1273
1274 /*
1275 * relay_recv_metadata: receive the metada for the session.
1276 */
1277 static
1278 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1279 struct relay_command *cmd)
1280 {
1281 int ret = htobe32(LTTNG_OK);
1282 ssize_t size_ret;
1283 struct relay_session *session = cmd->session;
1284 struct lttcomm_relayd_metadata_payload *metadata_struct;
1285 struct relay_stream *metadata_stream;
1286 uint64_t data_size, payload_size;
1287
1288 if (!session) {
1289 ERR("Metadata sent before version check");
1290 ret = -1;
1291 goto end;
1292 }
1293
1294 data_size = payload_size = be64toh(recv_hdr->data_size);
1295 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1296 ERR("Incorrect data size");
1297 ret = -1;
1298 goto end;
1299 }
1300 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1301
1302 if (data_buffer_size < data_size) {
1303 /* In case the realloc fails, we can free the memory */
1304 char *tmp_data_ptr;
1305
1306 tmp_data_ptr = realloc(data_buffer, data_size);
1307 if (!tmp_data_ptr) {
1308 ERR("Allocating data buffer");
1309 free(data_buffer);
1310 ret = -1;
1311 goto end;
1312 }
1313 data_buffer = tmp_data_ptr;
1314 data_buffer_size = data_size;
1315 }
1316 memset(data_buffer, 0, data_size);
1317 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1318 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1319 if (ret < 0 || ret != data_size) {
1320 if (ret == 0) {
1321 /* Orderly shutdown. Not necessary to print an error. */
1322 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1323 } else {
1324 ERR("Relay didn't receive the whole metadata");
1325 }
1326 ret = -1;
1327 goto end;
1328 }
1329 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1330
1331 rcu_read_lock();
1332 metadata_stream = relay_stream_find_by_id(
1333 be64toh(metadata_struct->stream_id));
1334 if (!metadata_stream) {
1335 ret = -1;
1336 goto end_unlock;
1337 }
1338
1339 size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
1340 payload_size);
1341 if (size_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 ssize_t size_ret;
1983 struct relay_stream *stream;
1984 struct lttcomm_relayd_data_hdr data_hdr;
1985 uint64_t stream_id;
1986 uint64_t net_seq_num;
1987 uint32_t data_size;
1988
1989 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1990 sizeof(struct lttcomm_relayd_data_hdr), 0);
1991 if (ret <= 0) {
1992 if (ret == 0) {
1993 /* Orderly shutdown. Not necessary to print an error. */
1994 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1995 } else {
1996 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
1997 }
1998 ret = -1;
1999 goto end;
2000 }
2001
2002 stream_id = be64toh(data_hdr.stream_id);
2003
2004 rcu_read_lock();
2005 stream = relay_stream_find_by_id(stream_id);
2006 if (!stream) {
2007 ret = -1;
2008 goto end_rcu_unlock;
2009 }
2010
2011 data_size = be32toh(data_hdr.data_size);
2012 if (data_buffer_size < data_size) {
2013 char *tmp_data_ptr;
2014
2015 tmp_data_ptr = realloc(data_buffer, data_size);
2016 if (!tmp_data_ptr) {
2017 ERR("Allocating data buffer");
2018 free(data_buffer);
2019 ret = -1;
2020 goto end_rcu_unlock;
2021 }
2022 data_buffer = tmp_data_ptr;
2023 data_buffer_size = data_size;
2024 }
2025 memset(data_buffer, 0, data_size);
2026
2027 net_seq_num = be64toh(data_hdr.net_seq_num);
2028
2029 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
2030 data_size, stream_id, net_seq_num);
2031 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
2032 if (ret <= 0) {
2033 if (ret == 0) {
2034 /* Orderly shutdown. Not necessary to print an error. */
2035 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
2036 }
2037 ret = -1;
2038 goto end_rcu_unlock;
2039 }
2040
2041 /* Check if a rotation is needed. */
2042 if (stream->tracefile_size > 0 &&
2043 (stream->tracefile_size_current + data_size) >
2044 stream->tracefile_size) {
2045 struct relay_viewer_stream *vstream;
2046 uint64_t new_id;
2047
2048 new_id = (stream->tracefile_count_current + 1) %
2049 stream->tracefile_count;
2050 /*
2051 * When we wrap-around back to 0, we start overwriting old
2052 * trace data.
2053 */
2054 if (!stream->tracefile_overwrite && new_id == 0) {
2055 stream->tracefile_overwrite = 1;
2056 }
2057 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
2058 if (stream->tracefile_overwrite) {
2059 stream->oldest_tracefile_id =
2060 (stream->oldest_tracefile_id + 1) %
2061 stream->tracefile_count;
2062 }
2063 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
2064 if (vstream) {
2065 /*
2066 * The viewer is reading a file about to be
2067 * overwritten. Close the FDs it is
2068 * currently using and let it handle the fault.
2069 */
2070 if (vstream->tracefile_count_current == new_id) {
2071 pthread_mutex_lock(&vstream->overwrite_lock);
2072 vstream->abort_flag = 1;
2073 pthread_mutex_unlock(&vstream->overwrite_lock);
2074 DBG("Streaming side setting abort_flag on stream %s_%lu\n",
2075 stream->channel_name, new_id);
2076 } else if (vstream->tracefile_count_current ==
2077 stream->tracefile_count_current) {
2078 /*
2079 * The reader and writer were in the
2080 * same trace file, inform the viewer
2081 * that no new index will ever be added
2082 * to this file.
2083 */
2084 vstream->close_write_flag = 1;
2085 }
2086 }
2087 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2088 stream->tracefile_size, stream->tracefile_count,
2089 relayd_uid, relayd_gid, stream->fd,
2090 &(stream->tracefile_count_current), &stream->fd);
2091 stream->total_index_received = 0;
2092 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
2093 if (ret < 0) {
2094 ERR("Rotating stream output file");
2095 goto end_rcu_unlock;
2096 }
2097 /* Reset current size because we just perform a stream rotation. */
2098 stream->tracefile_size_current = 0;
2099 rotate_index = 1;
2100 }
2101
2102 /*
2103 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2104 * index are NOT supported.
2105 */
2106 if (stream->session->minor >= 4 && !stream->session->snapshot) {
2107 ret = handle_index_data(stream, net_seq_num, rotate_index);
2108 if (ret < 0) {
2109 goto end_rcu_unlock;
2110 }
2111 }
2112
2113 /* Write data to stream output fd. */
2114 size_ret = lttng_write(stream->fd, data_buffer, data_size);
2115 if (size_ret < data_size) {
2116 ERR("Relay error writing data to file");
2117 ret = -1;
2118 goto end_rcu_unlock;
2119 }
2120
2121 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2122 ret, stream->stream_handle);
2123
2124 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2125 if (ret < 0) {
2126 goto end_rcu_unlock;
2127 }
2128 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2129
2130 stream->prev_seq = net_seq_num;
2131
2132 /* Check if we need to close the FD */
2133 if (close_stream_check(stream)) {
2134 destroy_stream(stream);
2135 }
2136
2137 end_rcu_unlock:
2138 rcu_read_unlock();
2139 end:
2140 return ret;
2141 }
2142
2143 static
2144 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
2145 {
2146 int ret;
2147
2148 lttng_poll_del(events, pollfd);
2149
2150 ret = close(pollfd);
2151 if (ret < 0) {
2152 ERR("Closing pollfd %d", pollfd);
2153 }
2154 }
2155
2156 static
2157 int relay_add_connection(int fd, struct lttng_poll_event *events,
2158 struct lttng_ht *relay_connections_ht)
2159 {
2160 struct relay_command *relay_connection;
2161 ssize_t ret;
2162
2163 relay_connection = zmalloc(sizeof(struct relay_command));
2164 if (relay_connection == NULL) {
2165 PERROR("Relay command zmalloc");
2166 goto error;
2167 }
2168 ret = lttng_read(fd, relay_connection, sizeof(struct relay_command));
2169 if (ret < sizeof(struct relay_command)) {
2170 PERROR("read relay cmd pipe");
2171 goto error_read;
2172 }
2173
2174 /*
2175 * Only used by the control side and the reference is copied inside each
2176 * stream from that connection. Thus a destroy HT must be done after every
2177 * stream has been destroyed.
2178 */
2179 if (relay_connection->type == RELAY_CONTROL) {
2180 relay_connection->ctf_traces_ht = lttng_ht_new(0,
2181 LTTNG_HT_TYPE_STRING);
2182 if (!relay_connection->ctf_traces_ht) {
2183 goto error_read;
2184 }
2185 }
2186
2187 lttng_ht_node_init_ulong(&relay_connection->sock_n,
2188 (unsigned long) relay_connection->sock->fd);
2189 rcu_read_lock();
2190 lttng_ht_add_unique_ulong(relay_connections_ht,
2191 &relay_connection->sock_n);
2192 rcu_read_unlock();
2193 return lttng_poll_add(events,
2194 relay_connection->sock->fd,
2195 LPOLLIN | LPOLLRDHUP);
2196
2197 error_read:
2198 free(relay_connection);
2199 error:
2200 return -1;
2201 }
2202
2203 static
2204 void deferred_free_connection(struct rcu_head *head)
2205 {
2206 struct relay_command *relay_connection =
2207 caa_container_of(head, struct relay_command, rcu_node);
2208
2209 lttcomm_destroy_sock(relay_connection->sock);
2210 free(relay_connection);
2211 }
2212
2213 static
2214 void relay_del_connection(struct lttng_ht *relay_connections_ht,
2215 struct lttng_ht_iter *iter, struct relay_command *relay_connection,
2216 struct lttng_ht *sessions_ht)
2217 {
2218 int ret;
2219
2220 ret = lttng_ht_del(relay_connections_ht, iter);
2221 assert(!ret);
2222
2223 if (relay_connection->type == RELAY_CONTROL) {
2224 relay_delete_session(relay_connection, sessions_ht);
2225 lttng_ht_destroy(relay_connection->ctf_traces_ht);
2226 }
2227
2228 call_rcu(&relay_connection->rcu_node, deferred_free_connection);
2229 }
2230
2231 /*
2232 * This thread does the actual work
2233 */
2234 static
2235 void *relay_thread_worker(void *data)
2236 {
2237 int ret, err = -1, last_seen_data_fd = -1;
2238 uint32_t nb_fd;
2239 struct relay_command *relay_connection;
2240 struct lttng_poll_event events;
2241 struct lttng_ht *relay_connections_ht;
2242 struct lttng_ht_node_ulong *node;
2243 struct lttng_ht_iter iter;
2244 struct lttcomm_relayd_hdr recv_hdr;
2245 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2246 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2247
2248 DBG("[thread] Relay worker started");
2249
2250 rcu_register_thread();
2251
2252 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2253
2254 health_code_update();
2255
2256 /* table of connections indexed on socket */
2257 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2258 if (!relay_connections_ht) {
2259 goto relay_connections_ht_error;
2260 }
2261
2262 /* Tables of received indexes indexed by index handle and net_seq_num. */
2263 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2264 if (!indexes_ht) {
2265 goto indexes_ht_error;
2266 }
2267
2268 ret = create_thread_poll_set(&events, 2);
2269 if (ret < 0) {
2270 goto error_poll_create;
2271 }
2272
2273 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
2274 if (ret < 0) {
2275 goto error;
2276 }
2277
2278 restart:
2279 while (1) {
2280 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2281
2282 health_code_update();
2283
2284 /* Infinite blocking call, waiting for transmission */
2285 DBG3("Relayd worker thread polling...");
2286 health_poll_entry();
2287 ret = lttng_poll_wait(&events, -1);
2288 health_poll_exit();
2289 if (ret < 0) {
2290 /*
2291 * Restart interrupted system call.
2292 */
2293 if (errno == EINTR) {
2294 goto restart;
2295 }
2296 goto error;
2297 }
2298
2299 nb_fd = ret;
2300
2301 /*
2302 * Process control. The control connection is prioritised so we don't
2303 * starve it with high throughout put tracing data on the data
2304 * connection.
2305 */
2306 for (i = 0; i < nb_fd; i++) {
2307 /* Fetch once the poll data */
2308 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2309 int pollfd = LTTNG_POLL_GETFD(&events, i);
2310
2311 health_code_update();
2312
2313 /* Thread quit pipe has been closed. Killing thread. */
2314 ret = check_thread_quit_pipe(pollfd, revents);
2315 if (ret) {
2316 err = 0;
2317 goto exit;
2318 }
2319
2320 /* Inspect the relay cmd pipe for new connection */
2321 if (pollfd == relay_cmd_pipe[0]) {
2322 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2323 ERR("Relay pipe error");
2324 goto error;
2325 } else if (revents & LPOLLIN) {
2326 DBG("Relay command received");
2327 ret = relay_add_connection(relay_cmd_pipe[0],
2328 &events, relay_connections_ht);
2329 if (ret < 0) {
2330 goto error;
2331 }
2332 }
2333 } else if (revents) {
2334 rcu_read_lock();
2335 lttng_ht_lookup(relay_connections_ht,
2336 (void *)((unsigned long) pollfd),
2337 &iter);
2338 node = lttng_ht_iter_get_node_ulong(&iter);
2339 if (node == NULL) {
2340 DBG2("Relay sock %d not found", pollfd);
2341 rcu_read_unlock();
2342 goto error;
2343 }
2344 relay_connection = caa_container_of(node,
2345 struct relay_command, sock_n);
2346
2347 if (revents & (LPOLLERR)) {
2348 ERR("POLL ERROR");
2349 relay_cleanup_poll_connection(&events, pollfd);
2350 relay_del_connection(relay_connections_ht,
2351 &iter, relay_connection, sessions_ht);
2352 if (last_seen_data_fd == pollfd) {
2353 last_seen_data_fd = last_notdel_data_fd;
2354 }
2355 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2356 DBG("Socket %d hung up", pollfd);
2357 relay_cleanup_poll_connection(&events, pollfd);
2358 relay_del_connection(relay_connections_ht,
2359 &iter, relay_connection, sessions_ht);
2360 if (last_seen_data_fd == pollfd) {
2361 last_seen_data_fd = last_notdel_data_fd;
2362 }
2363 } else if (revents & LPOLLIN) {
2364 /* control socket */
2365 if (relay_connection->type == RELAY_CONTROL) {
2366 ret = relay_connection->sock->ops->recvmsg(
2367 relay_connection->sock, &recv_hdr,
2368 sizeof(struct lttcomm_relayd_hdr), 0);
2369 /* connection closed */
2370 if (ret <= 0) {
2371 relay_cleanup_poll_connection(&events, pollfd);
2372 relay_del_connection(relay_connections_ht,
2373 &iter, relay_connection, sessions_ht);
2374 DBG("Control connection closed with %d", pollfd);
2375 } else {
2376 if (relay_connection->session) {
2377 DBG2("Relay worker receiving data for session : %" PRIu64,
2378 relay_connection->session->id);
2379 }
2380 ret = relay_process_control(&recv_hdr,
2381 relay_connection, relay_ctx);
2382 if (ret < 0) {
2383 /* Clear the session on error. */
2384 relay_cleanup_poll_connection(&events, pollfd);
2385 relay_del_connection(relay_connections_ht,
2386 &iter, relay_connection, sessions_ht);
2387 DBG("Connection closed with %d", pollfd);
2388 }
2389 seen_control = 1;
2390 }
2391 } else {
2392 /*
2393 * Flag the last seen data fd not deleted. It will be
2394 * used as the last seen fd if any fd gets deleted in
2395 * this first loop.
2396 */
2397 last_notdel_data_fd = pollfd;
2398 }
2399 }
2400 rcu_read_unlock();
2401 }
2402 }
2403
2404 /*
2405 * The last loop handled a control request, go back to poll to make
2406 * sure we prioritise the control socket.
2407 */
2408 if (seen_control) {
2409 continue;
2410 }
2411
2412 if (last_seen_data_fd >= 0) {
2413 for (i = 0; i < nb_fd; i++) {
2414 int pollfd = LTTNG_POLL_GETFD(&events, i);
2415
2416 health_code_update();
2417
2418 if (last_seen_data_fd == pollfd) {
2419 idx = i;
2420 break;
2421 }
2422 }
2423 }
2424
2425 /* Process data connection. */
2426 for (i = idx + 1; i < nb_fd; i++) {
2427 /* Fetch the poll data. */
2428 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2429 int pollfd = LTTNG_POLL_GETFD(&events, i);
2430
2431 health_code_update();
2432
2433 /* Skip the command pipe. It's handled in the first loop. */
2434 if (pollfd == relay_cmd_pipe[0]) {
2435 continue;
2436 }
2437
2438 if (revents) {
2439 rcu_read_lock();
2440 lttng_ht_lookup(relay_connections_ht,
2441 (void *)((unsigned long) pollfd),
2442 &iter);
2443 node = lttng_ht_iter_get_node_ulong(&iter);
2444 if (node == NULL) {
2445 /* Skip it. Might be removed before. */
2446 rcu_read_unlock();
2447 continue;
2448 }
2449 relay_connection = caa_container_of(node,
2450 struct relay_command, sock_n);
2451
2452 if (revents & LPOLLIN) {
2453 if (relay_connection->type != RELAY_DATA) {
2454 continue;
2455 }
2456
2457 ret = relay_process_data(relay_connection);
2458 /* connection closed */
2459 if (ret < 0) {
2460 relay_cleanup_poll_connection(&events, pollfd);
2461 relay_del_connection(relay_connections_ht,
2462 &iter, relay_connection, sessions_ht);
2463 DBG("Data connection closed with %d", pollfd);
2464 /*
2465 * Every goto restart call sets the last seen fd where
2466 * here we don't really care since we gracefully
2467 * continue the loop after the connection is deleted.
2468 */
2469 } else {
2470 /* Keep last seen port. */
2471 last_seen_data_fd = pollfd;
2472 rcu_read_unlock();
2473 goto restart;
2474 }
2475 }
2476 rcu_read_unlock();
2477 }
2478 }
2479 last_seen_data_fd = -1;
2480 }
2481
2482 /* Normal exit, no error */
2483 ret = 0;
2484
2485 exit:
2486 error:
2487 lttng_poll_clean(&events);
2488
2489 /* empty the hash table and free the memory */
2490 rcu_read_lock();
2491 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2492 health_code_update();
2493
2494 node = lttng_ht_iter_get_node_ulong(&iter);
2495 if (node) {
2496 relay_connection = caa_container_of(node,
2497 struct relay_command, sock_n);
2498 relay_del_connection(relay_connections_ht,
2499 &iter, relay_connection, sessions_ht);
2500 }
2501 }
2502 rcu_read_unlock();
2503 error_poll_create:
2504 lttng_ht_destroy(indexes_ht);
2505 indexes_ht_error:
2506 lttng_ht_destroy(relay_connections_ht);
2507 relay_connections_ht_error:
2508 /* Close relay cmd pipes */
2509 utils_close_pipe(relay_cmd_pipe);
2510 if (err) {
2511 DBG("Thread exited with error");
2512 }
2513 DBG("Worker thread cleanup complete");
2514 free(data_buffer);
2515 if (err) {
2516 health_error();
2517 ERR("Health error occurred in %s", __func__);
2518 }
2519 health_unregister(health_relayd);
2520 rcu_unregister_thread();
2521 stop_threads();
2522 return NULL;
2523 }
2524
2525 /*
2526 * Create the relay command pipe to wake thread_manage_apps.
2527 * Closed in cleanup().
2528 */
2529 static int create_relay_cmd_pipe(void)
2530 {
2531 int ret;
2532
2533 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2534
2535 return ret;
2536 }
2537
2538 /*
2539 * main
2540 */
2541 int main(int argc, char **argv)
2542 {
2543 int ret = 0;
2544 void *status;
2545 struct relay_local_data *relay_ctx;
2546
2547 /* Create thread quit pipe */
2548 if ((ret = init_thread_quit_pipe()) < 0) {
2549 goto error;
2550 }
2551
2552 /* Parse arguments */
2553 progname = argv[0];
2554 if ((ret = parse_args(argc, argv)) < 0) {
2555 goto exit;
2556 }
2557
2558 if ((ret = set_signal_handler()) < 0) {
2559 goto exit;
2560 }
2561
2562 /* Try to create directory if -o, --output is specified. */
2563 if (opt_output_path) {
2564 if (*opt_output_path != '/') {
2565 ERR("Please specify an absolute path for -o, --output PATH");
2566 goto exit;
2567 }
2568
2569 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2570 if (ret < 0) {
2571 ERR("Unable to create %s", opt_output_path);
2572 goto exit;
2573 }
2574 }
2575
2576 /* Daemonize */
2577 if (opt_daemon) {
2578 ret = daemon(0, 0);
2579 if (ret < 0) {
2580 PERROR("daemon");
2581 goto exit;
2582 }
2583 }
2584
2585 /* We need those values for the file/dir creation. */
2586 relayd_uid = getuid();
2587 relayd_gid = getgid();
2588
2589 /* Check if daemon is UID = 0 */
2590 if (relayd_uid == 0) {
2591 if (control_uri->port < 1024 || data_uri->port < 1024) {
2592 ERR("Need to be root to use ports < 1024");
2593 ret = -1;
2594 goto exit;
2595 }
2596 }
2597
2598 /* Setup the thread apps communication pipe. */
2599 if ((ret = create_relay_cmd_pipe()) < 0) {
2600 goto exit;
2601 }
2602
2603 /* Init relay command queue. */
2604 cds_wfq_init(&relay_cmd_queue.queue);
2605
2606 /* Set up max poll set size */
2607 lttng_poll_set_max_size();
2608
2609 /* Initialize communication library */
2610 lttcomm_init();
2611
2612 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2613 if (!relay_ctx) {
2614 PERROR("relay_ctx");
2615 goto exit;
2616 }
2617
2618 /* tables of sessions indexed by session ID */
2619 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2620 if (!relay_ctx->sessions_ht) {
2621 goto exit_relay_ctx_sessions;
2622 }
2623
2624 /* tables of streams indexed by stream ID */
2625 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2626 if (!relay_streams_ht) {
2627 goto exit_relay_ctx_streams;
2628 }
2629
2630 /* tables of streams indexed by stream ID */
2631 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2632 if (!viewer_streams_ht) {
2633 goto exit_relay_ctx_viewer_streams;
2634 }
2635
2636 /* Initialize thread health monitoring */
2637 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2638 if (!health_relayd) {
2639 PERROR("health_app_create error");
2640 goto exit_health_app_create;
2641 }
2642
2643 ret = utils_create_pipe(health_quit_pipe);
2644 if (ret < 0) {
2645 goto error_health_pipe;
2646 }
2647
2648 /* Create thread to manage the client socket */
2649 ret = pthread_create(&health_thread, NULL,
2650 thread_manage_health, (void *) NULL);
2651 if (ret != 0) {
2652 PERROR("pthread_create health");
2653 goto health_error;
2654 }
2655
2656 /* Setup the dispatcher thread */
2657 ret = pthread_create(&dispatcher_thread, NULL,
2658 relay_thread_dispatcher, (void *) NULL);
2659 if (ret != 0) {
2660 PERROR("pthread_create dispatcher");
2661 goto exit_dispatcher;
2662 }
2663
2664 /* Setup the worker thread */
2665 ret = pthread_create(&worker_thread, NULL,
2666 relay_thread_worker, (void *) relay_ctx);
2667 if (ret != 0) {
2668 PERROR("pthread_create worker");
2669 goto exit_worker;
2670 }
2671
2672 /* Setup the listener thread */
2673 ret = pthread_create(&listener_thread, NULL,
2674 relay_thread_listener, (void *) NULL);
2675 if (ret != 0) {
2676 PERROR("pthread_create listener");
2677 goto exit_listener;
2678 }
2679
2680 ret = live_start_threads(live_uri, relay_ctx, thread_quit_pipe);
2681 if (ret != 0) {
2682 ERR("Starting live viewer threads");
2683 goto exit_live;
2684 }
2685
2686 exit_live:
2687 ret = pthread_join(listener_thread, &status);
2688 if (ret != 0) {
2689 PERROR("pthread_join");
2690 goto error; /* join error, exit without cleanup */
2691 }
2692
2693 exit_listener:
2694 ret = pthread_join(worker_thread, &status);
2695 if (ret != 0) {
2696 PERROR("pthread_join");
2697 goto error; /* join error, exit without cleanup */
2698 }
2699
2700 exit_worker:
2701 ret = pthread_join(dispatcher_thread, &status);
2702 if (ret != 0) {
2703 PERROR("pthread_join");
2704 goto error; /* join error, exit without cleanup */
2705 }
2706
2707 exit_dispatcher:
2708 ret = pthread_join(health_thread, &status);
2709 if (ret != 0) {
2710 PERROR("pthread_join health thread");
2711 goto error; /* join error, exit without cleanup */
2712 }
2713
2714 /*
2715 * Stop live threads only after joining other threads.
2716 */
2717 live_stop_threads();
2718
2719 health_error:
2720 utils_close_pipe(health_quit_pipe);
2721
2722 error_health_pipe:
2723 health_app_destroy(health_relayd);
2724
2725 exit_health_app_create:
2726 lttng_ht_destroy(viewer_streams_ht);
2727
2728 exit_relay_ctx_viewer_streams:
2729 lttng_ht_destroy(relay_streams_ht);
2730
2731 exit_relay_ctx_streams:
2732 lttng_ht_destroy(relay_ctx->sessions_ht);
2733
2734 exit_relay_ctx_sessions:
2735 free(relay_ctx);
2736
2737 exit:
2738 cleanup();
2739 if (!ret) {
2740 exit(EXIT_SUCCESS);
2741 }
2742
2743 error:
2744 exit(EXIT_FAILURE);
2745 }
This page took 0.082194 seconds and 5 git commands to generate.