Test for new metadata at each packet
[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
61 /* command line options */
62 char *opt_output_path;
63 static int opt_daemon;
64 static struct lttng_uri *control_uri;
65 static struct lttng_uri *data_uri;
66 static struct lttng_uri *live_uri;
67
68 const char *progname;
69
70 /*
71 * Quit pipe for all threads. This permits a single cancellation point
72 * for all threads when receiving an event on the pipe.
73 */
74 static int thread_quit_pipe[2] = { -1, -1 };
75
76 /*
77 * This pipe is used to inform the worker thread that a command is queued and
78 * ready to be processed.
79 */
80 static int relay_cmd_pipe[2] = { -1, -1 };
81
82 /* Shared between threads */
83 static int dispatch_thread_exit;
84
85 static pthread_t listener_thread;
86 static pthread_t dispatcher_thread;
87 static pthread_t worker_thread;
88
89 static uint64_t last_relay_stream_id;
90 static uint64_t last_relay_session_id;
91
92 /*
93 * Relay command queue.
94 *
95 * The relay_thread_listener and relay_thread_dispatcher communicate with this
96 * queue.
97 */
98 static struct relay_cmd_queue relay_cmd_queue;
99
100 /* buffer allocated at startup, used to store the trace data */
101 static char *data_buffer;
102 static unsigned int data_buffer_size;
103
104 /* Global hash table that stores relay index object. */
105 static struct lttng_ht *indexes_ht;
106
107 /* We need those values for the file/dir creation. */
108 static uid_t relayd_uid;
109 static gid_t relayd_gid;
110
111 /* Global relay stream hash table. */
112 struct lttng_ht *relay_streams_ht;
113
114 /*
115 * usage function on stderr
116 */
117 static
118 void usage(void)
119 {
120 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
121 fprintf(stderr, " -h, --help Display this usage.\n");
122 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
123 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
124 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
125 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
126 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
127 }
128
129 static
130 int parse_args(int argc, char **argv)
131 {
132 int c;
133 int ret = 0;
134 char *default_address;
135
136 static struct option long_options[] = {
137 { "control-port", 1, 0, 'C', },
138 { "data-port", 1, 0, 'D', },
139 { "daemonize", 0, 0, 'd', },
140 { "help", 0, 0, 'h', },
141 { "output", 1, 0, 'o', },
142 { "verbose", 0, 0, 'v', },
143 { NULL, 0, 0, 0, },
144 };
145
146 while (1) {
147 int option_index = 0;
148 c = getopt_long(argc, argv, "dhv" "C:D:o:",
149 long_options, &option_index);
150 if (c == -1) {
151 break;
152 }
153
154 switch (c) {
155 case 0:
156 fprintf(stderr, "option %s", long_options[option_index].name);
157 if (optarg) {
158 fprintf(stderr, " with arg %s\n", optarg);
159 }
160 break;
161 case 'C':
162 ret = uri_parse(optarg, &control_uri);
163 if (ret < 0) {
164 ERR("Invalid control URI specified");
165 goto exit;
166 }
167 if (control_uri->port == 0) {
168 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
169 }
170 break;
171 case 'D':
172 ret = uri_parse(optarg, &data_uri);
173 if (ret < 0) {
174 ERR("Invalid data URI specified");
175 goto exit;
176 }
177 if (data_uri->port == 0) {
178 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
179 }
180 break;
181 case 'd':
182 opt_daemon = 1;
183 break;
184 case 'h':
185 usage();
186 exit(EXIT_FAILURE);
187 case 'o':
188 ret = asprintf(&opt_output_path, "%s", optarg);
189 if (ret < 0) {
190 PERROR("asprintf opt_output_path");
191 goto exit;
192 }
193 break;
194 case 'v':
195 /* Verbose level can increase using multiple -v */
196 lttng_opt_verbose += 1;
197 break;
198 default:
199 /* Unknown option or other error.
200 * Error is printed by getopt, just return */
201 ret = -1;
202 goto exit;
203 }
204 }
205
206 /* assign default values */
207 if (control_uri == NULL) {
208 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
209 DEFAULT_NETWORK_CONTROL_PORT);
210 if (ret < 0) {
211 PERROR("asprintf default data address");
212 goto exit;
213 }
214
215 ret = uri_parse(default_address, &control_uri);
216 free(default_address);
217 if (ret < 0) {
218 ERR("Invalid control URI specified");
219 goto exit;
220 }
221 }
222 if (data_uri == NULL) {
223 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
224 DEFAULT_NETWORK_DATA_PORT);
225 if (ret < 0) {
226 PERROR("asprintf default data address");
227 goto exit;
228 }
229
230 ret = uri_parse(default_address, &data_uri);
231 free(default_address);
232 if (ret < 0) {
233 ERR("Invalid data URI specified");
234 goto exit;
235 }
236 }
237 if (live_uri == NULL) {
238 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
239 DEFAULT_NETWORK_VIEWER_PORT);
240 if (ret < 0) {
241 PERROR("asprintf default viewer control address");
242 goto exit;
243 }
244
245 ret = uri_parse(default_address, &live_uri);
246 free(default_address);
247 if (ret < 0) {
248 ERR("Invalid viewer control URI specified");
249 goto exit;
250 }
251 }
252
253 exit:
254 return ret;
255 }
256
257 /*
258 * Cleanup the daemon
259 */
260 static
261 void cleanup(void)
262 {
263 DBG("Cleaning up");
264
265 /* free the dynamically allocated opt_output_path */
266 free(opt_output_path);
267
268 /* Close thread quit pipes */
269 utils_close_pipe(thread_quit_pipe);
270
271 uri_free(control_uri);
272 uri_free(data_uri);
273 }
274
275 /*
276 * Write to writable pipe used to notify a thread.
277 */
278 static
279 int notify_thread_pipe(int wpipe)
280 {
281 int ret;
282
283 do {
284 ret = write(wpipe, "!", 1);
285 } while (ret < 0 && errno == EINTR);
286 if (ret < 0 || ret != 1) {
287 PERROR("write poll pipe");
288 }
289
290 return ret;
291 }
292
293 /*
294 * Stop all threads by closing the thread quit pipe.
295 */
296 static
297 void stop_threads(void)
298 {
299 int ret;
300
301 /* Stopping all threads */
302 DBG("Terminating all threads");
303 ret = notify_thread_pipe(thread_quit_pipe[1]);
304 if (ret < 0) {
305 ERR("write error on thread quit pipe");
306 }
307
308 /* Dispatch thread */
309 CMM_STORE_SHARED(dispatch_thread_exit, 1);
310 futex_nto1_wake(&relay_cmd_queue.futex);
311 }
312
313 /*
314 * Signal handler for the daemon
315 *
316 * Simply stop all worker threads, leaving main() return gracefully after
317 * joining all threads and calling cleanup().
318 */
319 static
320 void sighandler(int sig)
321 {
322 switch (sig) {
323 case SIGPIPE:
324 DBG("SIGPIPE caught");
325 return;
326 case SIGINT:
327 DBG("SIGINT caught");
328 stop_threads();
329 break;
330 case SIGTERM:
331 DBG("SIGTERM caught");
332 stop_threads();
333 break;
334 default:
335 break;
336 }
337 }
338
339 /*
340 * Setup signal handler for :
341 * SIGINT, SIGTERM, SIGPIPE
342 */
343 static
344 int set_signal_handler(void)
345 {
346 int ret = 0;
347 struct sigaction sa;
348 sigset_t sigset;
349
350 if ((ret = sigemptyset(&sigset)) < 0) {
351 PERROR("sigemptyset");
352 return ret;
353 }
354
355 sa.sa_handler = sighandler;
356 sa.sa_mask = sigset;
357 sa.sa_flags = 0;
358 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
359 PERROR("sigaction");
360 return ret;
361 }
362
363 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
364 PERROR("sigaction");
365 return ret;
366 }
367
368 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
369 PERROR("sigaction");
370 return ret;
371 }
372
373 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
374
375 return ret;
376 }
377
378 /*
379 * Init thread quit pipe.
380 *
381 * Return -1 on error or 0 if all pipes are created.
382 */
383 static
384 int init_thread_quit_pipe(void)
385 {
386 int ret;
387
388 ret = utils_create_pipe_cloexec(thread_quit_pipe);
389
390 return ret;
391 }
392
393 /*
394 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
395 */
396 static
397 int create_thread_poll_set(struct lttng_poll_event *events, int size)
398 {
399 int ret;
400
401 if (events == NULL || size == 0) {
402 ret = -1;
403 goto error;
404 }
405
406 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
407 if (ret < 0) {
408 goto error;
409 }
410
411 /* Add quit pipe */
412 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
413 if (ret < 0) {
414 goto error;
415 }
416
417 return 0;
418
419 error:
420 return ret;
421 }
422
423 /*
424 * Check if the thread quit pipe was triggered.
425 *
426 * Return 1 if it was triggered else 0;
427 */
428 static
429 int check_thread_quit_pipe(int fd, uint32_t events)
430 {
431 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
432 return 1;
433 }
434
435 return 0;
436 }
437
438 /*
439 * Create and init socket from uri.
440 */
441 static
442 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
443 {
444 int ret;
445 struct lttcomm_sock *sock = NULL;
446
447 sock = lttcomm_alloc_sock_from_uri(uri);
448 if (sock == NULL) {
449 ERR("Allocating socket");
450 goto error;
451 }
452
453 ret = lttcomm_create_sock(sock);
454 if (ret < 0) {
455 goto error;
456 }
457 DBG("Listening on sock %d", sock->fd);
458
459 ret = sock->ops->bind(sock);
460 if (ret < 0) {
461 goto error;
462 }
463
464 ret = sock->ops->listen(sock, -1);
465 if (ret < 0) {
466 goto error;
467
468 }
469
470 return sock;
471
472 error:
473 if (sock) {
474 lttcomm_destroy_sock(sock);
475 }
476 return NULL;
477 }
478
479 /*
480 * Return nonzero if stream needs to be closed.
481 */
482 static
483 int close_stream_check(struct relay_stream *stream)
484 {
485
486 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
487 /*
488 * We are about to close the stream so set the data pending flag to 1
489 * which will make the end data pending command skip the stream which
490 * is now closed and ready. Note that after proceeding to a file close,
491 * the written file is ready for reading.
492 */
493 stream->data_pending_check_done = 1;
494 return 1;
495 }
496 return 0;
497 }
498
499 /*
500 * This thread manages the listening for new connections on the network
501 */
502 static
503 void *relay_thread_listener(void *data)
504 {
505 int i, ret, pollfd, err = -1;
506 int val = 1;
507 uint32_t revents, nb_fd;
508 struct lttng_poll_event events;
509 struct lttcomm_sock *control_sock, *data_sock;
510
511 DBG("[thread] Relay listener started");
512
513 control_sock = relay_init_sock(control_uri);
514 if (!control_sock) {
515 goto error_sock_control;
516 }
517
518 data_sock = relay_init_sock(data_uri);
519 if (!data_sock) {
520 goto error_sock_relay;
521 }
522
523 /*
524 * Pass 3 as size here for the thread quit pipe, control and data socket.
525 */
526 ret = create_thread_poll_set(&events, 3);
527 if (ret < 0) {
528 goto error_create_poll;
529 }
530
531 /* Add the control socket */
532 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
533 if (ret < 0) {
534 goto error_poll_add;
535 }
536
537 /* Add the data socket */
538 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
539 if (ret < 0) {
540 goto error_poll_add;
541 }
542
543 while (1) {
544 DBG("Listener accepting connections");
545
546 restart:
547 ret = lttng_poll_wait(&events, -1);
548 if (ret < 0) {
549 /*
550 * Restart interrupted system call.
551 */
552 if (errno == EINTR) {
553 goto restart;
554 }
555 goto error;
556 }
557
558 nb_fd = ret;
559
560 DBG("Relay new connection received");
561 for (i = 0; i < nb_fd; i++) {
562 /* Fetch once the poll data */
563 revents = LTTNG_POLL_GETEV(&events, i);
564 pollfd = LTTNG_POLL_GETFD(&events, i);
565
566 /* Thread quit pipe has been closed. Killing thread. */
567 ret = check_thread_quit_pipe(pollfd, revents);
568 if (ret) {
569 err = 0;
570 goto exit;
571 }
572
573 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
574 ERR("socket poll error");
575 goto error;
576 } else if (revents & LPOLLIN) {
577 /*
578 * Get allocated in this thread,
579 * enqueued to a global queue, dequeued
580 * and freed in the worker thread.
581 */
582 struct relay_command *relay_cmd;
583 struct lttcomm_sock *newsock;
584
585 relay_cmd = zmalloc(sizeof(struct relay_command));
586 if (relay_cmd == NULL) {
587 PERROR("relay command zmalloc");
588 goto error;
589 }
590
591 if (pollfd == data_sock->fd) {
592 newsock = data_sock->ops->accept(data_sock);
593 if (!newsock) {
594 PERROR("accepting data sock");
595 free(relay_cmd);
596 goto error;
597 }
598 relay_cmd->type = RELAY_DATA;
599 DBG("Relay data connection accepted, socket %d", newsock->fd);
600 } else {
601 assert(pollfd == control_sock->fd);
602 newsock = control_sock->ops->accept(control_sock);
603 if (!newsock) {
604 PERROR("accepting control sock");
605 free(relay_cmd);
606 goto error;
607 }
608 relay_cmd->type = RELAY_CONTROL;
609 DBG("Relay control connection accepted, socket %d", newsock->fd);
610 }
611 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
612 &val, sizeof(int));
613 if (ret < 0) {
614 PERROR("setsockopt inet");
615 lttcomm_destroy_sock(newsock);
616 free(relay_cmd);
617 goto error;
618 }
619 relay_cmd->sock = newsock;
620 /*
621 * Lock free enqueue the request.
622 */
623 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
624
625 /*
626 * Wake the dispatch queue futex. Implicit memory
627 * barrier with the exchange in cds_wfq_enqueue.
628 */
629 futex_nto1_wake(&relay_cmd_queue.futex);
630 }
631 }
632 }
633
634 exit:
635 error:
636 error_poll_add:
637 lttng_poll_clean(&events);
638 error_create_poll:
639 if (data_sock->fd >= 0) {
640 ret = data_sock->ops->close(data_sock);
641 if (ret) {
642 PERROR("close");
643 }
644 }
645 lttcomm_destroy_sock(data_sock);
646 error_sock_relay:
647 if (control_sock->fd >= 0) {
648 ret = control_sock->ops->close(control_sock);
649 if (ret) {
650 PERROR("close");
651 }
652 }
653 lttcomm_destroy_sock(control_sock);
654 error_sock_control:
655 if (err) {
656 DBG("Thread exited with error");
657 }
658 DBG("Relay listener thread cleanup complete");
659 stop_threads();
660 return NULL;
661 }
662
663 /*
664 * This thread manages the dispatching of the requests to worker threads
665 */
666 static
667 void *relay_thread_dispatcher(void *data)
668 {
669 int ret;
670 struct cds_wfq_node *node;
671 struct relay_command *relay_cmd = NULL;
672
673 DBG("[thread] Relay dispatcher started");
674
675 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
676 /* Atomically prepare the queue futex */
677 futex_nto1_prepare(&relay_cmd_queue.futex);
678
679 do {
680 /* Dequeue commands */
681 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
682 if (node == NULL) {
683 DBG("Woken up but nothing in the relay command queue");
684 /* Continue thread execution */
685 break;
686 }
687
688 relay_cmd = caa_container_of(node, struct relay_command, node);
689 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
690
691 /*
692 * Inform worker thread of the new request. This
693 * call is blocking so we can be assured that the data will be read
694 * at some point in time or wait to the end of the world :)
695 */
696 do {
697 ret = write(relay_cmd_pipe[1], relay_cmd,
698 sizeof(struct relay_command));
699 } while (ret < 0 && errno == EINTR);
700 free(relay_cmd);
701 if (ret < 0 || ret != sizeof(struct relay_command)) {
702 PERROR("write cmd pipe");
703 goto error;
704 }
705 } while (node != NULL);
706
707 /* Futex wait on queue. Blocking call on futex() */
708 futex_nto1_wait(&relay_cmd_queue.futex);
709 }
710
711 error:
712 DBG("Dispatch thread dying");
713 stop_threads();
714 return NULL;
715 }
716
717 /*
718 * Get stream from stream id.
719 * Need to be called with RCU read-side lock held.
720 */
721 struct relay_stream *relay_stream_find_by_id(uint64_t stream_id)
722 {
723 struct lttng_ht_node_ulong *node;
724 struct lttng_ht_iter iter;
725 struct relay_stream *ret;
726
727 lttng_ht_lookup(relay_streams_ht,
728 (void *)((unsigned long) stream_id),
729 &iter);
730 node = lttng_ht_iter_get_node_ulong(&iter);
731 if (node == NULL) {
732 DBG("Relay stream %" PRIu64 " not found", stream_id);
733 ret = NULL;
734 goto end;
735 }
736
737 ret = caa_container_of(node, struct relay_stream, stream_n);
738
739 end:
740 return ret;
741 }
742
743 static
744 void deferred_free_stream(struct rcu_head *head)
745 {
746 struct relay_stream *stream =
747 caa_container_of(head, struct relay_stream, rcu_node);
748
749 ctf_trace_try_destroy(stream->ctf_trace);
750
751 free(stream->path_name);
752 free(stream->channel_name);
753 free(stream);
754 }
755
756 static
757 void deferred_free_session(struct rcu_head *head)
758 {
759 struct relay_session *session =
760 caa_container_of(head, struct relay_session, rcu_node);
761 free(session);
762 }
763
764 static void close_stream(struct relay_stream *stream,
765 struct lttng_ht *viewer_streams_ht, struct lttng_ht *ctf_traces_ht)
766 {
767 int delret;
768 struct relay_viewer_stream *vstream;
769 struct lttng_ht_iter iter;
770
771 assert(stream);
772 assert(viewer_streams_ht);
773
774 delret = close(stream->fd);
775 if (delret < 0) {
776 PERROR("close stream");
777 }
778
779 if (stream->index_fd >= 0) {
780 delret = close(stream->index_fd);
781 if (delret < 0) {
782 PERROR("close stream index_fd");
783 }
784 }
785
786 vstream = live_find_viewer_stream_by_id(stream->stream_handle,
787 viewer_streams_ht);
788 if (vstream) {
789 /*
790 * Set the last good value into the viewer stream. This is done
791 * right before the stream gets deleted from the hash table. The
792 * lookup failure on the live thread side of a stream indicates
793 * that the viewer stream index received value should be used.
794 */
795 vstream->total_index_received = stream->total_index_received;
796 }
797
798 iter.iter.node = &stream->stream_n.node;
799 delret = lttng_ht_del(relay_streams_ht, &iter);
800 assert(!delret);
801 iter.iter.node = &stream->ctf_trace_node.node;
802 delret = lttng_ht_del(ctf_traces_ht, &iter);
803 assert(!delret);
804 call_rcu(&stream->rcu_node, deferred_free_stream);
805 DBG("Closed tracefile %d from close stream", stream->fd);
806 }
807
808 /*
809 * relay_delete_session: Free all memory associated with a session and
810 * close all the FDs
811 */
812 static
813 void relay_delete_session(struct relay_command *cmd,
814 struct lttng_ht *sessions_ht)
815 {
816 struct lttng_ht_iter iter;
817 struct lttng_ht_node_ulong *node;
818 struct relay_stream *stream;
819 int ret;
820
821 if (!cmd->session) {
822 return;
823 }
824
825 DBG("Relay deleting session %" PRIu64, cmd->session->id);
826
827 rcu_read_lock();
828 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
829 node = lttng_ht_iter_get_node_ulong(&iter);
830 if (node) {
831 stream = caa_container_of(node,
832 struct relay_stream, stream_n);
833 if (stream->session == cmd->session) {
834 ret = close(stream->fd);
835 if (ret < 0) {
836 PERROR("close stream fd on delete session");
837 }
838 ret = lttng_ht_del(relay_streams_ht, &iter);
839 assert(!ret);
840 call_rcu(&stream->rcu_node,
841 deferred_free_stream);
842 }
843 /* Cleanup index of that stream. */
844 relay_index_destroy_by_stream_id(stream->stream_handle,
845 indexes_ht);
846 }
847 }
848 iter.iter.node = &cmd->session->session_n.node;
849 ret = lttng_ht_del(sessions_ht, &iter);
850 assert(!ret);
851 call_rcu(&cmd->session->rcu_node,
852 deferred_free_session);
853 rcu_read_unlock();
854 }
855
856 /*
857 * Copy index data from the control port to a given index object.
858 */
859 static void copy_index_control_data(struct relay_index *index,
860 struct lttcomm_relayd_index *data)
861 {
862 assert(index);
863 assert(data);
864
865 /*
866 * The index on disk is encoded in big endian, so we don't need to convert
867 * the data received on the network. The data_offset value is NEVER
868 * modified here and is updated by the data thread.
869 */
870 index->index_data.packet_size = data->packet_size;
871 index->index_data.content_size = data->content_size;
872 index->index_data.timestamp_begin = data->timestamp_begin;
873 index->index_data.timestamp_end = data->timestamp_end;
874 index->index_data.events_discarded = data->events_discarded;
875 index->index_data.stream_id = data->stream_id;
876 }
877
878 /*
879 * Handle the RELAYD_CREATE_SESSION command.
880 *
881 * On success, send back the session id or else return a negative value.
882 */
883 static
884 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
885 struct relay_command *cmd,
886 struct lttng_ht *sessions_ht)
887 {
888 int ret = 0, send_ret;
889 struct relay_session *session;
890 struct lttcomm_relayd_status_session reply;
891
892 assert(recv_hdr);
893 assert(cmd);
894
895 memset(&reply, 0, sizeof(reply));
896
897 session = zmalloc(sizeof(struct relay_session));
898 if (session == NULL) {
899 PERROR("relay session zmalloc");
900 ret = -1;
901 goto error;
902 }
903
904 session->id = ++last_relay_session_id;
905 session->sock = cmd->sock;
906 cmd->session = session;
907
908 reply.session_id = htobe64(session->id);
909
910 switch (cmd->minor) {
911 case 4: /* LTTng sessiond 2.4 */
912 default:
913 ret = cmd_create_session_2_4(cmd, session);
914 break;
915 }
916
917 lttng_ht_node_init_ulong(&session->session_n,
918 (unsigned long) session->id);
919 lttng_ht_add_unique_ulong(sessions_ht,
920 &session->session_n);
921
922 DBG("Created session %" PRIu64, session->id);
923
924 error:
925 if (ret < 0) {
926 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
927 } else {
928 reply.ret_code = htobe32(LTTNG_OK);
929 }
930
931 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
932 if (send_ret < 0) {
933 ERR("Relayd sending session id");
934 ret = send_ret;
935 }
936
937 return ret;
938 }
939
940 /*
941 * relay_add_stream: allocate a new stream for a session
942 */
943 static
944 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
945 struct relay_command *cmd, struct lttng_ht *sessions_ht)
946 {
947 struct relay_session *session = cmd->session;
948 struct relay_stream *stream = NULL;
949 struct lttcomm_relayd_status_stream reply;
950 int ret, send_ret;
951
952 if (!session || cmd->version_check_done == 0) {
953 ERR("Trying to add a stream before version check");
954 ret = -1;
955 goto end_no_session;
956 }
957
958 stream = zmalloc(sizeof(struct relay_stream));
959 if (stream == NULL) {
960 PERROR("relay stream zmalloc");
961 ret = -1;
962 goto end_no_session;
963 }
964
965 switch (cmd->minor) {
966 case 1: /* LTTng sessiond 2.1 */
967 ret = cmd_recv_stream_2_1(cmd, stream);
968 break;
969 case 2: /* LTTng sessiond 2.2 */
970 default:
971 ret = cmd_recv_stream_2_2(cmd, stream);
972 break;
973 }
974 if (ret < 0) {
975 goto err_free_stream;
976 }
977
978 rcu_read_lock();
979 stream->stream_handle = ++last_relay_stream_id;
980 stream->prev_seq = -1ULL;
981 stream->session = session;
982 stream->index_fd = -1;
983 stream->read_index_fd = -1;
984 stream->ctf_trace = NULL;
985 pthread_mutex_init(&stream->lock, NULL);
986
987 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
988 if (ret < 0) {
989 ERR("relay creating output directory");
990 goto end;
991 }
992
993 /*
994 * No need to use run_as API here because whatever we receives, the relayd
995 * uses its own credentials for the stream files.
996 */
997 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
998 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
999 if (ret < 0) {
1000 ERR("Create output file");
1001 goto end;
1002 }
1003 stream->fd = ret;
1004 if (stream->tracefile_size) {
1005 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1006 } else {
1007 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1008 }
1009
1010 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1011 stream->metadata_flag = 1;
1012 /*
1013 * When we receive a new metadata stream, we create a new
1014 * ctf_trace and we assign this ctf_trace to all streams with
1015 * the same path.
1016 *
1017 * If later on we receive a new stream for the same ctf_trace,
1018 * we copy the information from the first hit in the HT to the
1019 * new stream.
1020 */
1021 stream->ctf_trace = ctf_trace_create();
1022 if (!stream->ctf_trace) {
1023 ret = -1;
1024 goto end;
1025 }
1026 stream->ctf_trace->refcount++;
1027 stream->ctf_trace->metadata_stream = stream;
1028 }
1029 ctf_trace_assign(cmd->ctf_traces_ht, stream);
1030
1031 lttng_ht_node_init_ulong(&stream->stream_n,
1032 (unsigned long) stream->stream_handle);
1033 lttng_ht_add_unique_ulong(relay_streams_ht,
1034 &stream->stream_n);
1035
1036 lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name);
1037 lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node);
1038
1039 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1040 stream->stream_handle);
1041
1042 end:
1043 reply.handle = htobe64(stream->stream_handle);
1044 /* send the session id to the client or a negative return code on error */
1045 if (ret < 0) {
1046 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1047 /* stream was not properly added to the ht, so free it */
1048 free(stream);
1049 } else {
1050 reply.ret_code = htobe32(LTTNG_OK);
1051 }
1052
1053 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1054 sizeof(struct lttcomm_relayd_status_stream), 0);
1055 if (send_ret < 0) {
1056 ERR("Relay sending stream id");
1057 ret = send_ret;
1058 }
1059 rcu_read_unlock();
1060
1061 end_no_session:
1062 return ret;
1063
1064 err_free_stream:
1065 free(stream->path_name);
1066 free(stream->channel_name);
1067 free(stream);
1068 return ret;
1069 }
1070
1071 /*
1072 * relay_close_stream: close a specific stream
1073 */
1074 static
1075 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1076 struct relay_command *cmd, struct lttng_ht *viewer_streams_ht)
1077 {
1078 int ret, send_ret;
1079 struct relay_session *session = cmd->session;
1080 struct lttcomm_relayd_close_stream stream_info;
1081 struct lttcomm_relayd_generic_reply reply;
1082 struct relay_stream *stream;
1083
1084 DBG("Close stream received");
1085
1086 if (!session || cmd->version_check_done == 0) {
1087 ERR("Trying to close a stream before version check");
1088 ret = -1;
1089 goto end_no_session;
1090 }
1091
1092 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1093 sizeof(struct lttcomm_relayd_close_stream), 0);
1094 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1095 if (ret == 0) {
1096 /* Orderly shutdown. Not necessary to print an error. */
1097 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1098 } else {
1099 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1100 }
1101 ret = -1;
1102 goto end_no_session;
1103 }
1104
1105 rcu_read_lock();
1106 stream = relay_stream_find_by_id(be64toh(stream_info.stream_id));
1107 if (!stream) {
1108 ret = -1;
1109 goto end_unlock;
1110 }
1111
1112 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1113 stream->close_flag = 1;
1114
1115 if (close_stream_check(stream)) {
1116 close_stream(stream, viewer_streams_ht, cmd->ctf_traces_ht);
1117 }
1118
1119 end_unlock:
1120 rcu_read_unlock();
1121
1122 if (ret < 0) {
1123 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1124 } else {
1125 reply.ret_code = htobe32(LTTNG_OK);
1126 }
1127 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1128 sizeof(struct lttcomm_relayd_generic_reply), 0);
1129 if (send_ret < 0) {
1130 ERR("Relay sending stream id");
1131 ret = send_ret;
1132 }
1133
1134 end_no_session:
1135 return ret;
1136 }
1137
1138 /*
1139 * relay_unknown_command: send -1 if received unknown command
1140 */
1141 static
1142 void relay_unknown_command(struct relay_command *cmd)
1143 {
1144 struct lttcomm_relayd_generic_reply reply;
1145 int ret;
1146
1147 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1148 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1149 sizeof(struct lttcomm_relayd_generic_reply), 0);
1150 if (ret < 0) {
1151 ERR("Relay sending unknown command");
1152 }
1153 }
1154
1155 /*
1156 * relay_start: send an acknowledgment to the client to tell if we are
1157 * ready to receive data. We are ready if a session is established.
1158 */
1159 static
1160 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1161 struct relay_command *cmd)
1162 {
1163 int ret = htobe32(LTTNG_OK);
1164 struct lttcomm_relayd_generic_reply reply;
1165 struct relay_session *session = cmd->session;
1166
1167 if (!session) {
1168 DBG("Trying to start the streaming without a session established");
1169 ret = htobe32(LTTNG_ERR_UNK);
1170 }
1171
1172 reply.ret_code = ret;
1173 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1174 sizeof(struct lttcomm_relayd_generic_reply), 0);
1175 if (ret < 0) {
1176 ERR("Relay sending start ack");
1177 }
1178
1179 return ret;
1180 }
1181
1182 /*
1183 * Append padding to the file pointed by the file descriptor fd.
1184 */
1185 static int write_padding_to_file(int fd, uint32_t size)
1186 {
1187 int ret = 0;
1188 char *zeros;
1189
1190 if (size == 0) {
1191 goto end;
1192 }
1193
1194 zeros = zmalloc(size);
1195 if (zeros == NULL) {
1196 PERROR("zmalloc zeros for padding");
1197 ret = -1;
1198 goto end;
1199 }
1200
1201 do {
1202 ret = write(fd, zeros, size);
1203 } while (ret < 0 && errno == EINTR);
1204 if (ret < 0 || ret != size) {
1205 PERROR("write padding to file");
1206 }
1207
1208 free(zeros);
1209
1210 end:
1211 return ret;
1212 }
1213
1214 /*
1215 * relay_recv_metadata: receive the metada for the session.
1216 */
1217 static
1218 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1219 struct relay_command *cmd)
1220 {
1221 int ret = htobe32(LTTNG_OK);
1222 struct relay_session *session = cmd->session;
1223 struct lttcomm_relayd_metadata_payload *metadata_struct;
1224 struct relay_stream *metadata_stream;
1225 uint64_t data_size, payload_size;
1226
1227 if (!session) {
1228 ERR("Metadata sent before version check");
1229 ret = -1;
1230 goto end;
1231 }
1232
1233 data_size = payload_size = be64toh(recv_hdr->data_size);
1234 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1235 ERR("Incorrect data size");
1236 ret = -1;
1237 goto end;
1238 }
1239 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1240
1241 if (data_buffer_size < data_size) {
1242 /* In case the realloc fails, we can free the memory */
1243 char *tmp_data_ptr;
1244
1245 tmp_data_ptr = realloc(data_buffer, data_size);
1246 if (!tmp_data_ptr) {
1247 ERR("Allocating data buffer");
1248 free(data_buffer);
1249 ret = -1;
1250 goto end;
1251 }
1252 data_buffer = tmp_data_ptr;
1253 data_buffer_size = data_size;
1254 }
1255 memset(data_buffer, 0, data_size);
1256 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1257 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1258 if (ret < 0 || ret != data_size) {
1259 if (ret == 0) {
1260 /* Orderly shutdown. Not necessary to print an error. */
1261 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1262 } else {
1263 ERR("Relay didn't receive the whole metadata");
1264 }
1265 ret = -1;
1266 goto end;
1267 }
1268 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1269
1270 rcu_read_lock();
1271 metadata_stream = relay_stream_find_by_id(
1272 be64toh(metadata_struct->stream_id));
1273 if (!metadata_stream) {
1274 ret = -1;
1275 goto end_unlock;
1276 }
1277
1278 do {
1279 ret = write(metadata_stream->fd, metadata_struct->payload,
1280 payload_size);
1281 } while (ret < 0 && errno == EINTR);
1282 if (ret < 0 || ret != payload_size) {
1283 ERR("Relay error writing metadata on file");
1284 ret = -1;
1285 goto end_unlock;
1286 }
1287
1288 ret = write_padding_to_file(metadata_stream->fd,
1289 be32toh(metadata_struct->padding_size));
1290 if (ret < 0) {
1291 goto end_unlock;
1292 }
1293 metadata_stream->ctf_trace->metadata_received +=
1294 payload_size + be32toh(metadata_struct->padding_size);
1295
1296 DBG2("Relay metadata written");
1297
1298 end_unlock:
1299 rcu_read_unlock();
1300 end:
1301 return ret;
1302 }
1303
1304 /*
1305 * relay_send_version: send relayd version number
1306 */
1307 static
1308 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1309 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1310 {
1311 int ret;
1312 struct lttcomm_relayd_version reply, msg;
1313
1314 assert(cmd);
1315
1316 cmd->version_check_done = 1;
1317
1318 /* Get version from the other side. */
1319 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1320 if (ret < 0 || ret != sizeof(msg)) {
1321 if (ret == 0) {
1322 /* Orderly shutdown. Not necessary to print an error. */
1323 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1324 } else {
1325 ERR("Relay failed to receive the version values.");
1326 }
1327 ret = -1;
1328 goto end;
1329 }
1330
1331 reply.major = RELAYD_VERSION_COMM_MAJOR;
1332 reply.minor = RELAYD_VERSION_COMM_MINOR;
1333
1334 /* Major versions must be the same */
1335 if (reply.major != be32toh(msg.major)) {
1336 DBG("Incompatible major versions (%u vs %u), deleting session",
1337 reply.major, be32toh(msg.major));
1338 relay_delete_session(cmd, sessions_ht);
1339 ret = 0;
1340 goto end;
1341 }
1342
1343 cmd->major = reply.major;
1344 /* We adapt to the lowest compatible version */
1345 if (reply.minor <= be32toh(msg.minor)) {
1346 cmd->minor = reply.minor;
1347 } else {
1348 cmd->minor = be32toh(msg.minor);
1349 }
1350
1351 reply.major = htobe32(reply.major);
1352 reply.minor = htobe32(reply.minor);
1353 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1354 sizeof(struct lttcomm_relayd_version), 0);
1355 if (ret < 0) {
1356 ERR("Relay sending version");
1357 }
1358
1359 DBG("Version check done using protocol %u.%u", cmd->major,
1360 cmd->minor);
1361
1362 end:
1363 return ret;
1364 }
1365
1366 /*
1367 * Check for data pending for a given stream id from the session daemon.
1368 */
1369 static
1370 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1371 struct relay_command *cmd)
1372 {
1373 struct relay_session *session = cmd->session;
1374 struct lttcomm_relayd_data_pending msg;
1375 struct lttcomm_relayd_generic_reply reply;
1376 struct relay_stream *stream;
1377 int ret;
1378 uint64_t last_net_seq_num, stream_id;
1379
1380 DBG("Data pending command received");
1381
1382 if (!session || cmd->version_check_done == 0) {
1383 ERR("Trying to check for data before version check");
1384 ret = -1;
1385 goto end_no_session;
1386 }
1387
1388 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1389 if (ret < sizeof(msg)) {
1390 if (ret == 0) {
1391 /* Orderly shutdown. Not necessary to print an error. */
1392 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1393 } else {
1394 ERR("Relay didn't receive valid data_pending struct size : %d",
1395 ret);
1396 }
1397 ret = -1;
1398 goto end_no_session;
1399 }
1400
1401 stream_id = be64toh(msg.stream_id);
1402 last_net_seq_num = be64toh(msg.last_net_seq_num);
1403
1404 rcu_read_lock();
1405 stream = relay_stream_find_by_id(stream_id);
1406 if (stream == NULL) {
1407 ret = -1;
1408 goto end_unlock;
1409 }
1410
1411 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1412 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1413 last_net_seq_num);
1414
1415 /* Avoid wrapping issue */
1416 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1417 /* Data has in fact been written and is NOT pending */
1418 ret = 0;
1419 } else {
1420 /* Data still being streamed thus pending */
1421 ret = 1;
1422 }
1423
1424 /* Pending check is now done. */
1425 stream->data_pending_check_done = 1;
1426
1427 end_unlock:
1428 rcu_read_unlock();
1429
1430 reply.ret_code = htobe32(ret);
1431 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1432 if (ret < 0) {
1433 ERR("Relay data pending ret code failed");
1434 }
1435
1436 end_no_session:
1437 return ret;
1438 }
1439
1440 /*
1441 * Wait for the control socket to reach a quiescent state.
1442 *
1443 * Note that for now, when receiving this command from the session daemon, this
1444 * means that every subsequent commands or data received on the control socket
1445 * has been handled. So, this is why we simply return OK here.
1446 */
1447 static
1448 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1449 struct relay_command *cmd)
1450 {
1451 int ret;
1452 uint64_t stream_id;
1453 struct relay_stream *stream;
1454 struct lttng_ht_iter iter;
1455 struct lttcomm_relayd_quiescent_control msg;
1456 struct lttcomm_relayd_generic_reply reply;
1457
1458 DBG("Checking quiescent state on control socket");
1459
1460 if (!cmd->session || cmd->version_check_done == 0) {
1461 ERR("Trying to check for data before version check");
1462 ret = -1;
1463 goto end_no_session;
1464 }
1465
1466 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1467 if (ret < sizeof(msg)) {
1468 if (ret == 0) {
1469 /* Orderly shutdown. Not necessary to print an error. */
1470 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1471 } else {
1472 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1473 ret);
1474 }
1475 ret = -1;
1476 goto end_no_session;
1477 }
1478
1479 stream_id = be64toh(msg.stream_id);
1480
1481 rcu_read_lock();
1482 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1483 stream_n.node) {
1484 if (stream->stream_handle == stream_id) {
1485 stream->data_pending_check_done = 1;
1486 DBG("Relay quiescent control pending flag set to %" PRIu64,
1487 stream_id);
1488 break;
1489 }
1490 }
1491 rcu_read_unlock();
1492
1493 reply.ret_code = htobe32(LTTNG_OK);
1494 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1495 if (ret < 0) {
1496 ERR("Relay data quiescent control ret code failed");
1497 }
1498
1499 end_no_session:
1500 return ret;
1501 }
1502
1503 /*
1504 * Initialize a data pending command. This means that a client is about to ask
1505 * for data pending for each stream he/she holds. Simply iterate over all
1506 * streams of a session and set the data_pending_check_done flag.
1507 *
1508 * This command returns to the client a LTTNG_OK code.
1509 */
1510 static
1511 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1512 struct relay_command *cmd)
1513 {
1514 int ret;
1515 struct lttng_ht_iter iter;
1516 struct lttcomm_relayd_begin_data_pending msg;
1517 struct lttcomm_relayd_generic_reply reply;
1518 struct relay_stream *stream;
1519 uint64_t session_id;
1520
1521 assert(recv_hdr);
1522 assert(cmd);
1523
1524 DBG("Init streams for data pending");
1525
1526 if (!cmd->session || cmd->version_check_done == 0) {
1527 ERR("Trying to check for data before version check");
1528 ret = -1;
1529 goto end_no_session;
1530 }
1531
1532 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1533 if (ret < sizeof(msg)) {
1534 if (ret == 0) {
1535 /* Orderly shutdown. Not necessary to print an error. */
1536 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1537 } else {
1538 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1539 ret);
1540 }
1541 ret = -1;
1542 goto end_no_session;
1543 }
1544
1545 session_id = be64toh(msg.session_id);
1546
1547 /*
1548 * Iterate over all streams to set the begin data pending flag. For now, the
1549 * streams are indexed by stream handle so we have to iterate over all
1550 * streams to find the one associated with the right session_id.
1551 */
1552 rcu_read_lock();
1553 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1554 stream_n.node) {
1555 if (stream->session->id == session_id) {
1556 stream->data_pending_check_done = 0;
1557 DBG("Set begin data pending flag to stream %" PRIu64,
1558 stream->stream_handle);
1559 }
1560 }
1561 rcu_read_unlock();
1562
1563 /* All good, send back reply. */
1564 reply.ret_code = htobe32(LTTNG_OK);
1565
1566 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1567 if (ret < 0) {
1568 ERR("Relay begin data pending send reply failed");
1569 }
1570
1571 end_no_session:
1572 return ret;
1573 }
1574
1575 /*
1576 * End data pending command. This will check, for a given session id, if each
1577 * stream associated with it has its data_pending_check_done flag set. If not,
1578 * this means that the client lost track of the stream but the data is still
1579 * being streamed on our side. In this case, we inform the client that data is
1580 * inflight.
1581 *
1582 * Return to the client if there is data in flight or not with a ret_code.
1583 */
1584 static
1585 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1586 struct relay_command *cmd)
1587 {
1588 int ret;
1589 struct lttng_ht_iter iter;
1590 struct lttcomm_relayd_end_data_pending msg;
1591 struct lttcomm_relayd_generic_reply reply;
1592 struct relay_stream *stream;
1593 uint64_t session_id;
1594 uint32_t is_data_inflight = 0;
1595
1596 assert(recv_hdr);
1597 assert(cmd);
1598
1599 DBG("End data pending command");
1600
1601 if (!cmd->session || cmd->version_check_done == 0) {
1602 ERR("Trying to check for data before version check");
1603 ret = -1;
1604 goto end_no_session;
1605 }
1606
1607 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1608 if (ret < sizeof(msg)) {
1609 if (ret == 0) {
1610 /* Orderly shutdown. Not necessary to print an error. */
1611 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1612 } else {
1613 ERR("Relay didn't receive valid end data_pending struct size: %d",
1614 ret);
1615 }
1616 ret = -1;
1617 goto end_no_session;
1618 }
1619
1620 session_id = be64toh(msg.session_id);
1621
1622 /* Iterate over all streams to see if the begin data pending flag is set. */
1623 rcu_read_lock();
1624 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1625 stream_n.node) {
1626 if (stream->session->id == session_id &&
1627 !stream->data_pending_check_done) {
1628 is_data_inflight = 1;
1629 DBG("Data is still in flight for stream %" PRIu64,
1630 stream->stream_handle);
1631 break;
1632 }
1633 }
1634 rcu_read_unlock();
1635
1636 /* All good, send back reply. */
1637 reply.ret_code = htobe32(is_data_inflight);
1638
1639 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1640 if (ret < 0) {
1641 ERR("Relay end data pending send reply failed");
1642 }
1643
1644 end_no_session:
1645 return ret;
1646 }
1647
1648 /*
1649 * Receive an index for a specific stream.
1650 *
1651 * Return 0 on success else a negative value.
1652 */
1653 static
1654 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1655 struct relay_command *cmd, struct lttng_ht *indexes_ht)
1656 {
1657 int ret, send_ret, index_created = 0;
1658 struct relay_session *session = cmd->session;
1659 struct lttcomm_relayd_index index_info;
1660 struct relay_index *index, *wr_index = NULL;
1661 struct lttcomm_relayd_generic_reply reply;
1662 struct relay_stream *stream;
1663 uint64_t net_seq_num;
1664
1665 assert(cmd);
1666 assert(indexes_ht);
1667
1668 DBG("Relay receiving index");
1669
1670 if (!session || cmd->version_check_done == 0) {
1671 ERR("Trying to close a stream before version check");
1672 ret = -1;
1673 goto end_no_session;
1674 }
1675
1676 ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info,
1677 sizeof(index_info), 0);
1678 if (ret < sizeof(index_info)) {
1679 if (ret == 0) {
1680 /* Orderly shutdown. Not necessary to print an error. */
1681 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1682 } else {
1683 ERR("Relay didn't receive valid index struct size : %d", ret);
1684 }
1685 ret = -1;
1686 goto end_no_session;
1687 }
1688
1689 net_seq_num = be64toh(index_info.net_seq_num);
1690
1691 rcu_read_lock();
1692 stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id));
1693 if (!stream) {
1694 ret = -1;
1695 goto end_rcu_unlock;
1696 }
1697
1698 /* Live beacon handling */
1699 if (index_info.packet_size == 0) {
1700 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1701
1702 /*
1703 * Only flag a stream inactive when it has already received data.
1704 */
1705 if (stream->total_index_received > 0) {
1706 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1707 }
1708 ret = 0;
1709 goto end_rcu_unlock;
1710 } else {
1711 stream->beacon_ts_end = -1ULL;
1712 }
1713
1714 index = relay_index_find(stream->stream_handle, net_seq_num, indexes_ht);
1715 if (!index) {
1716 /* A successful creation will add the object to the HT. */
1717 index = relay_index_create(stream->stream_handle, net_seq_num);
1718 if (!index) {
1719 goto end_rcu_unlock;
1720 }
1721 index_created = 1;
1722 }
1723
1724 copy_index_control_data(index, &index_info);
1725
1726 if (index_created) {
1727 /*
1728 * Try to add the relay index object to the hash table. If an object
1729 * already exist, destroy back the index created, set the data in this
1730 * object and write it on disk.
1731 */
1732 relay_index_add(index, indexes_ht, &wr_index);
1733 if (wr_index) {
1734 copy_index_control_data(wr_index, &index_info);
1735 free(index);
1736 }
1737 } else {
1738 /* The index already exists so write it on disk. */
1739 wr_index = index;
1740 }
1741
1742 /* Do we have a writable ready index to write on disk. */
1743 if (wr_index) {
1744 /* Starting at 2.4, create the index file if none available. */
1745 if (cmd->minor >= 4 && stream->index_fd < 0) {
1746 ret = index_create_file(stream->path_name, stream->channel_name,
1747 relayd_uid, relayd_gid, stream->tracefile_size,
1748 stream->tracefile_count_current);
1749 if (ret < 0) {
1750 goto end_rcu_unlock;
1751 }
1752 stream->index_fd = ret;
1753 }
1754
1755 ret = relay_index_write(wr_index->fd, wr_index, indexes_ht);
1756 if (ret < 0) {
1757 goto end_rcu_unlock;
1758 }
1759 stream->total_index_received++;
1760 }
1761
1762 end_rcu_unlock:
1763 rcu_read_unlock();
1764
1765 if (ret < 0) {
1766 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1767 } else {
1768 reply.ret_code = htobe32(LTTNG_OK);
1769 }
1770 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1771 if (send_ret < 0) {
1772 ERR("Relay sending close index id reply");
1773 ret = send_ret;
1774 }
1775
1776 end_no_session:
1777 return ret;
1778 }
1779
1780 /*
1781 * Process the commands received on the control socket
1782 */
1783 static
1784 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1785 struct relay_command *cmd, struct relay_local_data *ctx)
1786 {
1787 int ret = 0;
1788
1789 switch (be32toh(recv_hdr->cmd)) {
1790 case RELAYD_CREATE_SESSION:
1791 ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht);
1792 break;
1793 case RELAYD_ADD_STREAM:
1794 ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht);
1795 break;
1796 case RELAYD_START_DATA:
1797 ret = relay_start(recv_hdr, cmd);
1798 break;
1799 case RELAYD_SEND_METADATA:
1800 ret = relay_recv_metadata(recv_hdr, cmd);
1801 break;
1802 case RELAYD_VERSION:
1803 ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht);
1804 break;
1805 case RELAYD_CLOSE_STREAM:
1806 ret = relay_close_stream(recv_hdr, cmd, ctx->viewer_streams_ht);
1807 break;
1808 case RELAYD_DATA_PENDING:
1809 ret = relay_data_pending(recv_hdr, cmd);
1810 break;
1811 case RELAYD_QUIESCENT_CONTROL:
1812 ret = relay_quiescent_control(recv_hdr, cmd);
1813 break;
1814 case RELAYD_BEGIN_DATA_PENDING:
1815 ret = relay_begin_data_pending(recv_hdr, cmd);
1816 break;
1817 case RELAYD_END_DATA_PENDING:
1818 ret = relay_end_data_pending(recv_hdr, cmd);
1819 break;
1820 case RELAYD_SEND_INDEX:
1821 ret = relay_recv_index(recv_hdr, cmd, indexes_ht);
1822 break;
1823 case RELAYD_UPDATE_SYNC_INFO:
1824 default:
1825 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1826 relay_unknown_command(cmd);
1827 ret = -1;
1828 goto end;
1829 }
1830
1831 end:
1832 return ret;
1833 }
1834
1835 /*
1836 * relay_process_data: Process the data received on the data socket
1837 */
1838 static
1839 int relay_process_data(struct relay_command *cmd,
1840 struct lttng_ht *indexes_ht, struct lttng_ht *viewer_streams_ht)
1841 {
1842 int ret = 0, rotate_index = 0, index_created = 0;
1843 struct relay_stream *stream;
1844 struct relay_index *index, *wr_index = NULL;
1845 struct lttcomm_relayd_data_hdr data_hdr;
1846 uint64_t stream_id, data_offset;
1847 uint64_t net_seq_num;
1848 uint32_t data_size;
1849
1850 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1851 sizeof(struct lttcomm_relayd_data_hdr), 0);
1852 if (ret <= 0) {
1853 if (ret == 0) {
1854 /* Orderly shutdown. Not necessary to print an error. */
1855 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1856 } else {
1857 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
1858 }
1859 ret = -1;
1860 goto end;
1861 }
1862
1863 stream_id = be64toh(data_hdr.stream_id);
1864
1865 rcu_read_lock();
1866 stream = relay_stream_find_by_id(stream_id);
1867 if (!stream) {
1868 ret = -1;
1869 goto end_rcu_unlock;
1870 }
1871
1872 data_size = be32toh(data_hdr.data_size);
1873 if (data_buffer_size < data_size) {
1874 char *tmp_data_ptr;
1875
1876 tmp_data_ptr = realloc(data_buffer, data_size);
1877 if (!tmp_data_ptr) {
1878 ERR("Allocating data buffer");
1879 free(data_buffer);
1880 ret = -1;
1881 goto end_rcu_unlock;
1882 }
1883 data_buffer = tmp_data_ptr;
1884 data_buffer_size = data_size;
1885 }
1886 memset(data_buffer, 0, data_size);
1887
1888 net_seq_num = be64toh(data_hdr.net_seq_num);
1889
1890 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
1891 data_size, stream_id, net_seq_num);
1892 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1893 if (ret <= 0) {
1894 if (ret == 0) {
1895 /* Orderly shutdown. Not necessary to print an error. */
1896 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1897 }
1898 ret = -1;
1899 goto end_rcu_unlock;
1900 }
1901
1902 /* Check if a rotation is needed. */
1903 if (stream->tracefile_size > 0 &&
1904 (stream->tracefile_size_current + data_size) >
1905 stream->tracefile_size) {
1906 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
1907 stream->tracefile_size, stream->tracefile_count,
1908 relayd_uid, relayd_gid, stream->fd,
1909 &(stream->tracefile_count_current), &stream->fd);
1910 if (ret < 0) {
1911 ERR("Rotating stream output file");
1912 goto end_rcu_unlock;
1913 }
1914 /* Reset current size because we just perform a stream rotation. */
1915 stream->tracefile_size_current = 0;
1916 rotate_index = 1;
1917 }
1918
1919 /* Get data offset because we are about to update the index. */
1920 data_offset = htobe64(stream->tracefile_size_current);
1921
1922 /*
1923 * Lookup for an existing index for that stream id/sequence number. If on
1924 * exists, the control thread already received the data for it thus we need
1925 * to write it on disk.
1926 */
1927 index = relay_index_find(stream_id, net_seq_num, indexes_ht);
1928 if (!index) {
1929 /* A successful creation will add the object to the HT. */
1930 index = relay_index_create(stream->stream_handle, net_seq_num);
1931 if (!index) {
1932 goto end_rcu_unlock;
1933 }
1934 index_created = 1;
1935 }
1936
1937 if (rotate_index || stream->index_fd < 0) {
1938 index->to_close_fd = stream->index_fd;
1939 ret = index_create_file(stream->path_name, stream->channel_name,
1940 relayd_uid, relayd_gid, stream->tracefile_size,
1941 stream->tracefile_count_current);
1942 if (ret < 0) {
1943 /* This will close the stream's index fd if one. */
1944 relay_index_free_safe(index);
1945 goto end_rcu_unlock;
1946 }
1947 stream->index_fd = ret;
1948 }
1949 index->fd = stream->index_fd;
1950 index->index_data.offset = data_offset;
1951
1952 if (index_created) {
1953 /*
1954 * Try to add the relay index object to the hash table. If an object
1955 * already exist, destroy back the index created and set the data.
1956 */
1957 relay_index_add(index, indexes_ht, &wr_index);
1958 if (wr_index) {
1959 /* Copy back data from the created index. */
1960 wr_index->fd = index->fd;
1961 wr_index->to_close_fd = index->to_close_fd;
1962 wr_index->index_data.offset = data_offset;
1963 free(index);
1964 }
1965 } else {
1966 /* The index already exists so write it on disk. */
1967 wr_index = index;
1968 }
1969
1970 /* Do we have a writable ready index to write on disk. */
1971 if (wr_index) {
1972 /* Starting at 2.4, create the index file if none available. */
1973 if (cmd->minor >= 4 && stream->index_fd < 0) {
1974 ret = index_create_file(stream->path_name, stream->channel_name,
1975 relayd_uid, relayd_gid, stream->tracefile_size,
1976 stream->tracefile_count_current);
1977 if (ret < 0) {
1978 goto end_rcu_unlock;
1979 }
1980 stream->index_fd = ret;
1981 }
1982
1983 ret = relay_index_write(wr_index->fd, wr_index, indexes_ht);
1984 if (ret < 0) {
1985 goto end_rcu_unlock;
1986 }
1987 stream->total_index_received++;
1988 }
1989
1990 do {
1991 ret = write(stream->fd, data_buffer, data_size);
1992 } while (ret < 0 && errno == EINTR);
1993 if (ret < 0 || ret != data_size) {
1994 ERR("Relay error writing data to file");
1995 ret = -1;
1996 goto end_rcu_unlock;
1997 }
1998
1999 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2000 ret, stream->stream_handle);
2001
2002 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2003 if (ret < 0) {
2004 goto end_rcu_unlock;
2005 }
2006 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2007
2008 stream->prev_seq = net_seq_num;
2009
2010 /* Check if we need to close the FD */
2011 if (close_stream_check(stream)) {
2012 close_stream(stream, viewer_streams_ht, cmd->ctf_traces_ht);
2013 }
2014
2015 end_rcu_unlock:
2016 rcu_read_unlock();
2017 end:
2018 return ret;
2019 }
2020
2021 static
2022 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
2023 {
2024 int ret;
2025
2026 lttng_poll_del(events, pollfd);
2027
2028 ret = close(pollfd);
2029 if (ret < 0) {
2030 ERR("Closing pollfd %d", pollfd);
2031 }
2032 }
2033
2034 static
2035 int relay_add_connection(int fd, struct lttng_poll_event *events,
2036 struct lttng_ht *relay_connections_ht)
2037 {
2038 struct relay_command *relay_connection;
2039 int ret;
2040
2041 relay_connection = zmalloc(sizeof(struct relay_command));
2042 if (relay_connection == NULL) {
2043 PERROR("Relay command zmalloc");
2044 goto error;
2045 }
2046 do {
2047 ret = read(fd, relay_connection, sizeof(struct relay_command));
2048 } while (ret < 0 && errno == EINTR);
2049 if (ret < 0 || ret < sizeof(struct relay_command)) {
2050 PERROR("read relay cmd pipe");
2051 goto error_read;
2052 }
2053
2054 relay_connection->ctf_traces_ht = lttng_ht_new(0, LTTNG_HT_TYPE_STRING);
2055 if (!relay_connection->ctf_traces_ht) {
2056 goto error_read;
2057 }
2058
2059 lttng_ht_node_init_ulong(&relay_connection->sock_n,
2060 (unsigned long) relay_connection->sock->fd);
2061 rcu_read_lock();
2062 lttng_ht_add_unique_ulong(relay_connections_ht,
2063 &relay_connection->sock_n);
2064 rcu_read_unlock();
2065 return lttng_poll_add(events,
2066 relay_connection->sock->fd,
2067 LPOLLIN | LPOLLRDHUP);
2068
2069 error_read:
2070 free(relay_connection);
2071 error:
2072 return -1;
2073 }
2074
2075 static
2076 void deferred_free_connection(struct rcu_head *head)
2077 {
2078 struct relay_command *relay_connection =
2079 caa_container_of(head, struct relay_command, rcu_node);
2080
2081 lttng_ht_destroy(relay_connection->ctf_traces_ht);
2082 lttcomm_destroy_sock(relay_connection->sock);
2083 free(relay_connection);
2084 }
2085
2086 static
2087 void relay_del_connection(struct lttng_ht *relay_connections_ht,
2088 struct lttng_ht_iter *iter, struct relay_command *relay_connection,
2089 struct lttng_ht *sessions_ht)
2090 {
2091 int ret;
2092
2093 ret = lttng_ht_del(relay_connections_ht, iter);
2094 assert(!ret);
2095 if (relay_connection->type == RELAY_CONTROL) {
2096 relay_delete_session(relay_connection, sessions_ht);
2097 }
2098
2099 call_rcu(&relay_connection->rcu_node,
2100 deferred_free_connection);
2101 }
2102
2103 /*
2104 * This thread does the actual work
2105 */
2106 static
2107 void *relay_thread_worker(void *data)
2108 {
2109 int ret, err = -1, last_seen_data_fd = -1;
2110 uint32_t nb_fd;
2111 struct relay_command *relay_connection;
2112 struct lttng_poll_event events;
2113 struct lttng_ht *relay_connections_ht;
2114 struct lttng_ht_node_ulong *node;
2115 struct lttng_ht_iter iter;
2116 struct lttcomm_relayd_hdr recv_hdr;
2117 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2118 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2119
2120 DBG("[thread] Relay worker started");
2121
2122 rcu_register_thread();
2123
2124 /* table of connections indexed on socket */
2125 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2126 if (!relay_connections_ht) {
2127 goto relay_connections_ht_error;
2128 }
2129
2130 /* Tables of received indexes indexed by index handle and net_seq_num. */
2131 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2132 if (!indexes_ht) {
2133 goto indexes_ht_error;
2134 }
2135
2136 ret = create_thread_poll_set(&events, 2);
2137 if (ret < 0) {
2138 goto error_poll_create;
2139 }
2140
2141 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
2142 if (ret < 0) {
2143 goto error;
2144 }
2145
2146 restart:
2147 while (1) {
2148 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2149
2150 /* Infinite blocking call, waiting for transmission */
2151 DBG3("Relayd worker thread polling...");
2152 ret = lttng_poll_wait(&events, -1);
2153 if (ret < 0) {
2154 /*
2155 * Restart interrupted system call.
2156 */
2157 if (errno == EINTR) {
2158 goto restart;
2159 }
2160 goto error;
2161 }
2162
2163 nb_fd = ret;
2164
2165 /*
2166 * Process control. The control connection is prioritised so we don't
2167 * starve it with high throughout put tracing data on the data
2168 * connection.
2169 */
2170 for (i = 0; i < nb_fd; i++) {
2171 /* Fetch once the poll data */
2172 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2173 int pollfd = LTTNG_POLL_GETFD(&events, i);
2174
2175 /* Thread quit pipe has been closed. Killing thread. */
2176 ret = check_thread_quit_pipe(pollfd, revents);
2177 if (ret) {
2178 err = 0;
2179 goto exit;
2180 }
2181
2182 /* Inspect the relay cmd pipe for new connection */
2183 if (pollfd == relay_cmd_pipe[0]) {
2184 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2185 ERR("Relay pipe error");
2186 goto error;
2187 } else if (revents & LPOLLIN) {
2188 DBG("Relay command received");
2189 ret = relay_add_connection(relay_cmd_pipe[0],
2190 &events, relay_connections_ht);
2191 if (ret < 0) {
2192 goto error;
2193 }
2194 }
2195 } else if (revents) {
2196 rcu_read_lock();
2197 lttng_ht_lookup(relay_connections_ht,
2198 (void *)((unsigned long) pollfd),
2199 &iter);
2200 node = lttng_ht_iter_get_node_ulong(&iter);
2201 if (node == NULL) {
2202 DBG2("Relay sock %d not found", pollfd);
2203 rcu_read_unlock();
2204 goto error;
2205 }
2206 relay_connection = caa_container_of(node,
2207 struct relay_command, sock_n);
2208
2209 if (revents & (LPOLLERR)) {
2210 ERR("POLL ERROR");
2211 relay_cleanup_poll_connection(&events, pollfd);
2212 relay_del_connection(relay_connections_ht,
2213 &iter, relay_connection, sessions_ht);
2214 if (last_seen_data_fd == pollfd) {
2215 last_seen_data_fd = last_notdel_data_fd;
2216 }
2217 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2218 DBG("Socket %d hung up", pollfd);
2219 relay_cleanup_poll_connection(&events, pollfd);
2220 relay_del_connection(relay_connections_ht,
2221 &iter, relay_connection, sessions_ht);
2222 if (last_seen_data_fd == pollfd) {
2223 last_seen_data_fd = last_notdel_data_fd;
2224 }
2225 } else if (revents & LPOLLIN) {
2226 /* control socket */
2227 if (relay_connection->type == RELAY_CONTROL) {
2228 ret = relay_connection->sock->ops->recvmsg(
2229 relay_connection->sock, &recv_hdr,
2230 sizeof(struct lttcomm_relayd_hdr), 0);
2231 /* connection closed */
2232 if (ret <= 0) {
2233 relay_cleanup_poll_connection(&events, pollfd);
2234 relay_del_connection(relay_connections_ht,
2235 &iter, relay_connection, sessions_ht);
2236 DBG("Control connection closed with %d", pollfd);
2237 } else {
2238 if (relay_connection->session) {
2239 DBG2("Relay worker receiving data for session : %" PRIu64,
2240 relay_connection->session->id);
2241 }
2242 ret = relay_process_control(&recv_hdr,
2243 relay_connection, relay_ctx);
2244 if (ret < 0) {
2245 /* Clear the session on error. */
2246 relay_cleanup_poll_connection(&events, pollfd);
2247 relay_del_connection(relay_connections_ht,
2248 &iter, relay_connection, sessions_ht);
2249 DBG("Connection closed with %d", pollfd);
2250 }
2251 seen_control = 1;
2252 }
2253 } else {
2254 /*
2255 * Flag the last seen data fd not deleted. It will be
2256 * used as the last seen fd if any fd gets deleted in
2257 * this first loop.
2258 */
2259 last_notdel_data_fd = pollfd;
2260 }
2261 }
2262 rcu_read_unlock();
2263 }
2264 }
2265
2266 /*
2267 * The last loop handled a control request, go back to poll to make
2268 * sure we prioritise the control socket.
2269 */
2270 if (seen_control) {
2271 continue;
2272 }
2273
2274 if (last_seen_data_fd >= 0) {
2275 for (i = 0; i < nb_fd; i++) {
2276 int pollfd = LTTNG_POLL_GETFD(&events, i);
2277 if (last_seen_data_fd == pollfd) {
2278 idx = i;
2279 break;
2280 }
2281 }
2282 }
2283
2284 /* Process data connection. */
2285 for (i = idx + 1; i < nb_fd; i++) {
2286 /* Fetch the poll data. */
2287 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2288 int pollfd = LTTNG_POLL_GETFD(&events, i);
2289
2290 /* Skip the command pipe. It's handled in the first loop. */
2291 if (pollfd == relay_cmd_pipe[0]) {
2292 continue;
2293 }
2294
2295 if (revents) {
2296 rcu_read_lock();
2297 lttng_ht_lookup(relay_connections_ht,
2298 (void *)((unsigned long) pollfd),
2299 &iter);
2300 node = lttng_ht_iter_get_node_ulong(&iter);
2301 if (node == NULL) {
2302 /* Skip it. Might be removed before. */
2303 rcu_read_unlock();
2304 continue;
2305 }
2306 relay_connection = caa_container_of(node,
2307 struct relay_command, sock_n);
2308
2309 if (revents & LPOLLIN) {
2310 if (relay_connection->type != RELAY_DATA) {
2311 continue;
2312 }
2313
2314 ret = relay_process_data(relay_connection, indexes_ht,
2315 relay_ctx->viewer_streams_ht);
2316 /* connection closed */
2317 if (ret < 0) {
2318 relay_cleanup_poll_connection(&events, pollfd);
2319 relay_del_connection(relay_connections_ht,
2320 &iter, relay_connection, sessions_ht);
2321 DBG("Data connection closed with %d", pollfd);
2322 /*
2323 * Every goto restart call sets the last seen fd where
2324 * here we don't really care since we gracefully
2325 * continue the loop after the connection is deleted.
2326 */
2327 } else {
2328 /* Keep last seen port. */
2329 last_seen_data_fd = pollfd;
2330 rcu_read_unlock();
2331 goto restart;
2332 }
2333 }
2334 rcu_read_unlock();
2335 }
2336 }
2337 last_seen_data_fd = -1;
2338 }
2339
2340 exit:
2341 error:
2342 lttng_poll_clean(&events);
2343
2344 /* empty the hash table and free the memory */
2345 rcu_read_lock();
2346 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2347 node = lttng_ht_iter_get_node_ulong(&iter);
2348 if (node) {
2349 relay_connection = caa_container_of(node,
2350 struct relay_command, sock_n);
2351 relay_del_connection(relay_connections_ht,
2352 &iter, relay_connection, sessions_ht);
2353 }
2354 }
2355 error_poll_create:
2356 {
2357 struct relay_index *index;
2358 cds_lfht_for_each_entry(indexes_ht->ht, &iter.iter, index, index_n.node) {
2359 relay_index_delete(index, indexes_ht);
2360 }
2361 lttng_ht_destroy(indexes_ht);
2362 }
2363 rcu_read_unlock();
2364 indexes_ht_error:
2365 lttng_ht_destroy(relay_connections_ht);
2366 relay_connections_ht_error:
2367 /* Close relay cmd pipes */
2368 utils_close_pipe(relay_cmd_pipe);
2369 if (err) {
2370 DBG("Thread exited with error");
2371 }
2372 DBG("Worker thread cleanup complete");
2373 free(data_buffer);
2374 stop_threads();
2375 rcu_unregister_thread();
2376 return NULL;
2377 }
2378
2379 /*
2380 * Create the relay command pipe to wake thread_manage_apps.
2381 * Closed in cleanup().
2382 */
2383 static int create_relay_cmd_pipe(void)
2384 {
2385 int ret;
2386
2387 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2388
2389 return ret;
2390 }
2391
2392 /*
2393 * main
2394 */
2395 int main(int argc, char **argv)
2396 {
2397 int ret = 0;
2398 void *status;
2399 struct relay_local_data *relay_ctx;
2400
2401 /* Create thread quit pipe */
2402 if ((ret = init_thread_quit_pipe()) < 0) {
2403 goto error;
2404 }
2405
2406 /* Parse arguments */
2407 progname = argv[0];
2408 if ((ret = parse_args(argc, argv)) < 0) {
2409 goto exit;
2410 }
2411
2412 if ((ret = set_signal_handler()) < 0) {
2413 goto exit;
2414 }
2415
2416 /* Try to create directory if -o, --output is specified. */
2417 if (opt_output_path) {
2418 if (*opt_output_path != '/') {
2419 ERR("Please specify an absolute path for -o, --output PATH");
2420 goto exit;
2421 }
2422
2423 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2424 if (ret < 0) {
2425 ERR("Unable to create %s", opt_output_path);
2426 goto exit;
2427 }
2428 }
2429
2430 /* Daemonize */
2431 if (opt_daemon) {
2432 ret = daemon(0, 0);
2433 if (ret < 0) {
2434 PERROR("daemon");
2435 goto exit;
2436 }
2437 }
2438
2439 /* We need those values for the file/dir creation. */
2440 relayd_uid = getuid();
2441 relayd_gid = getgid();
2442
2443 /* Check if daemon is UID = 0 */
2444 if (relayd_uid == 0) {
2445 if (control_uri->port < 1024 || data_uri->port < 1024) {
2446 ERR("Need to be root to use ports < 1024");
2447 ret = -1;
2448 goto exit;
2449 }
2450 }
2451
2452 /* Setup the thread apps communication pipe. */
2453 if ((ret = create_relay_cmd_pipe()) < 0) {
2454 goto exit;
2455 }
2456
2457 /* Init relay command queue. */
2458 cds_wfq_init(&relay_cmd_queue.queue);
2459
2460 /* Set up max poll set size */
2461 lttng_poll_set_max_size();
2462
2463 /* Initialize communication library */
2464 lttcomm_init();
2465
2466 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2467 if (!relay_ctx) {
2468 PERROR("relay_ctx");
2469 goto exit;
2470 }
2471
2472 /* tables of sessions indexed by session ID */
2473 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2474 if (!relay_ctx->sessions_ht) {
2475 goto exit_relay_ctx_sessions;
2476 }
2477
2478 /* tables of streams indexed by stream ID */
2479 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2480 if (!relay_streams_ht) {
2481 goto exit_relay_ctx_streams;
2482 }
2483
2484 /* tables of streams indexed by stream ID */
2485 relay_ctx->viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2486 if (!relay_ctx->viewer_streams_ht) {
2487 goto exit_relay_ctx_viewer_streams;
2488 }
2489
2490 /* Setup the dispatcher thread */
2491 ret = pthread_create(&dispatcher_thread, NULL,
2492 relay_thread_dispatcher, (void *) NULL);
2493 if (ret != 0) {
2494 PERROR("pthread_create dispatcher");
2495 goto exit_dispatcher;
2496 }
2497
2498 /* Setup the worker thread */
2499 ret = pthread_create(&worker_thread, NULL,
2500 relay_thread_worker, (void *) relay_ctx);
2501 if (ret != 0) {
2502 PERROR("pthread_create worker");
2503 goto exit_worker;
2504 }
2505
2506 /* Setup the listener thread */
2507 ret = pthread_create(&listener_thread, NULL,
2508 relay_thread_listener, (void *) NULL);
2509 if (ret != 0) {
2510 PERROR("pthread_create listener");
2511 goto exit_listener;
2512 }
2513
2514 ret = live_start_threads(live_uri, relay_ctx);
2515 if (ret != 0) {
2516 ERR("Starting live viewer threads");
2517 }
2518
2519 exit_listener:
2520 ret = pthread_join(listener_thread, &status);
2521 if (ret != 0) {
2522 PERROR("pthread_join");
2523 goto error; /* join error, exit without cleanup */
2524 }
2525
2526 exit_worker:
2527 ret = pthread_join(worker_thread, &status);
2528 if (ret != 0) {
2529 PERROR("pthread_join");
2530 goto error; /* join error, exit without cleanup */
2531 }
2532
2533 exit_dispatcher:
2534 ret = pthread_join(dispatcher_thread, &status);
2535 if (ret != 0) {
2536 PERROR("pthread_join");
2537 goto error; /* join error, exit without cleanup */
2538 }
2539 lttng_ht_destroy(relay_ctx->viewer_streams_ht);
2540
2541 exit_relay_ctx_viewer_streams:
2542 lttng_ht_destroy(relay_streams_ht);
2543
2544 exit_relay_ctx_streams:
2545 lttng_ht_destroy(relay_ctx->sessions_ht);
2546
2547 exit_relay_ctx_sessions:
2548 free(relay_ctx);
2549
2550 exit:
2551 live_stop_threads();
2552 cleanup();
2553 if (!ret) {
2554 exit(EXIT_SUCCESS);
2555 }
2556
2557 error:
2558 exit(EXIT_FAILURE);
2559 }
This page took 0.114042 seconds and 6 git commands to generate.