projects / lttng-tools.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Make viewer streams HT global
[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 /* Global relay viewer stream hash table. */
115 struct lttng_ht *viewer_streams_ht;
116
117 /*
118 * usage function on stderr
119 */
120 static
121 void usage(void)
122 {
123 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
124 fprintf(stderr, " -h, --help Display this usage.\n");
125 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
126 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
127 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
128 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
129 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
130 }
131
132 static
133 int parse_args(int argc, char **argv)
134 {
135 int c;
136 int ret = 0;
137 char *default_address;
138
139 static struct option long_options[] = {
140 { "control-port", 1, 0, 'C', },
141 { "data-port", 1, 0, 'D', },
142 { "daemonize", 0, 0, 'd', },
143 { "help", 0, 0, 'h', },
144 { "output", 1, 0, 'o', },
145 { "verbose", 0, 0, 'v', },
146 { NULL, 0, 0, 0, },
147 };
148
149 while (1) {
150 int option_index = 0;
151 c = getopt_long(argc, argv, "dhv" "C:D:o:",
152 long_options, &option_index);
153 if (c == -1) {
154 break;
155 }
156
157 switch (c) {
158 case 0:
159 fprintf(stderr, "option %s", long_options[option_index].name);
160 if (optarg) {
161 fprintf(stderr, " with arg %s\n", optarg);
162 }
163 break;
164 case 'C':
165 ret = uri_parse(optarg, &control_uri);
166 if (ret < 0) {
167 ERR("Invalid control URI specified");
168 goto exit;
169 }
170 if (control_uri->port == 0) {
171 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
172 }
173 break;
174 case 'D':
175 ret = uri_parse(optarg, &data_uri);
176 if (ret < 0) {
177 ERR("Invalid data URI specified");
178 goto exit;
179 }
180 if (data_uri->port == 0) {
181 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
182 }
183 break;
184 case 'd':
185 opt_daemon = 1;
186 break;
187 case 'h':
188 usage();
189 exit(EXIT_FAILURE);
190 case 'o':
191 ret = asprintf(&opt_output_path, "%s", optarg);
192 if (ret < 0) {
193 PERROR("asprintf opt_output_path");
194 goto exit;
195 }
196 break;
197 case 'v':
198 /* Verbose level can increase using multiple -v */
199 lttng_opt_verbose += 1;
200 break;
201 default:
202 /* Unknown option or other error.
203 * Error is printed by getopt, just return */
204 ret = -1;
205 goto exit;
206 }
207 }
208
209 /* assign default values */
210 if (control_uri == NULL) {
211 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
212 DEFAULT_NETWORK_CONTROL_PORT);
213 if (ret < 0) {
214 PERROR("asprintf default data address");
215 goto exit;
216 }
217
218 ret = uri_parse(default_address, &control_uri);
219 free(default_address);
220 if (ret < 0) {
221 ERR("Invalid control URI specified");
222 goto exit;
223 }
224 }
225 if (data_uri == NULL) {
226 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
227 DEFAULT_NETWORK_DATA_PORT);
228 if (ret < 0) {
229 PERROR("asprintf default data address");
230 goto exit;
231 }
232
233 ret = uri_parse(default_address, &data_uri);
234 free(default_address);
235 if (ret < 0) {
236 ERR("Invalid data URI specified");
237 goto exit;
238 }
239 }
240 if (live_uri == NULL) {
241 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
242 DEFAULT_NETWORK_VIEWER_PORT);
243 if (ret < 0) {
244 PERROR("asprintf default viewer control address");
245 goto exit;
246 }
247
248 ret = uri_parse(default_address, &live_uri);
249 free(default_address);
250 if (ret < 0) {
251 ERR("Invalid viewer control URI specified");
252 goto exit;
253 }
254 }
255
256 exit:
257 return ret;
258 }
259
260 /*
261 * Cleanup the daemon
262 */
263 static
264 void cleanup(void)
265 {
266 DBG("Cleaning up");
267
268 /* free the dynamically allocated opt_output_path */
269 free(opt_output_path);
270
271 /* Close thread quit pipes */
272 utils_close_pipe(thread_quit_pipe);
273
274 uri_free(control_uri);
275 uri_free(data_uri);
276 }
277
278 /*
279 * Write to writable pipe used to notify a thread.
280 */
281 static
282 int notify_thread_pipe(int wpipe)
283 {
284 int ret;
285
286 do {
287 ret = write(wpipe, "!", 1);
288 } while (ret < 0 && errno == EINTR);
289 if (ret < 0 || ret != 1) {
290 PERROR("write poll pipe");
291 }
292
293 return ret;
294 }
295
296 /*
297 * Stop all threads by closing the thread quit pipe.
298 */
299 static
300 void stop_threads(void)
301 {
302 int ret;
303
304 /* Stopping all threads */
305 DBG("Terminating all threads");
306 ret = notify_thread_pipe(thread_quit_pipe[1]);
307 if (ret < 0) {
308 ERR("write error on thread quit pipe");
309 }
310
311 /* Dispatch thread */
312 CMM_STORE_SHARED(dispatch_thread_exit, 1);
313 futex_nto1_wake(&relay_cmd_queue.futex);
314 }
315
316 /*
317 * Signal handler for the daemon
318 *
319 * Simply stop all worker threads, leaving main() return gracefully after
320 * joining all threads and calling cleanup().
321 */
322 static
323 void sighandler(int sig)
324 {
325 switch (sig) {
326 case SIGPIPE:
327 DBG("SIGPIPE caught");
328 return;
329 case SIGINT:
330 DBG("SIGINT caught");
331 stop_threads();
332 break;
333 case SIGTERM:
334 DBG("SIGTERM caught");
335 stop_threads();
336 break;
337 default:
338 break;
339 }
340 }
341
342 /*
343 * Setup signal handler for :
344 * SIGINT, SIGTERM, SIGPIPE
345 */
346 static
347 int set_signal_handler(void)
348 {
349 int ret = 0;
350 struct sigaction sa;
351 sigset_t sigset;
352
353 if ((ret = sigemptyset(&sigset)) < 0) {
354 PERROR("sigemptyset");
355 return ret;
356 }
357
358 sa.sa_handler = sighandler;
359 sa.sa_mask = sigset;
360 sa.sa_flags = 0;
361 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
362 PERROR("sigaction");
363 return ret;
364 }
365
366 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
367 PERROR("sigaction");
368 return ret;
369 }
370
371 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
372 PERROR("sigaction");
373 return ret;
374 }
375
376 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
377
378 return ret;
379 }
380
381 /*
382 * Init thread quit pipe.
383 *
384 * Return -1 on error or 0 if all pipes are created.
385 */
386 static
387 int init_thread_quit_pipe(void)
388 {
389 int ret;
390
391 ret = utils_create_pipe_cloexec(thread_quit_pipe);
392
393 return ret;
394 }
395
396 /*
397 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
398 */
399 static
400 int create_thread_poll_set(struct lttng_poll_event *events, int size)
401 {
402 int ret;
403
404 if (events == NULL || size == 0) {
405 ret = -1;
406 goto error;
407 }
408
409 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
410 if (ret < 0) {
411 goto error;
412 }
413
414 /* Add quit pipe */
415 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
416 if (ret < 0) {
417 goto error;
418 }
419
420 return 0;
421
422 error:
423 return ret;
424 }
425
426 /*
427 * Check if the thread quit pipe was triggered.
428 *
429 * Return 1 if it was triggered else 0;
430 */
431 static
432 int check_thread_quit_pipe(int fd, uint32_t events)
433 {
434 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
435 return 1;
436 }
437
438 return 0;
439 }
440
441 /*
442 * Create and init socket from uri.
443 */
444 static
445 struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
446 {
447 int ret;
448 struct lttcomm_sock *sock = NULL;
449
450 sock = lttcomm_alloc_sock_from_uri(uri);
451 if (sock == NULL) {
452 ERR("Allocating socket");
453 goto error;
454 }
455
456 ret = lttcomm_create_sock(sock);
457 if (ret < 0) {
458 goto error;
459 }
460 DBG("Listening on sock %d", sock->fd);
461
462 ret = sock->ops->bind(sock);
463 if (ret < 0) {
464 goto error;
465 }
466
467 ret = sock->ops->listen(sock, -1);
468 if (ret < 0) {
469 goto error;
470
471 }
472
473 return sock;
474
475 error:
476 if (sock) {
477 lttcomm_destroy_sock(sock);
478 }
479 return NULL;
480 }
481
482 /*
483 * Return nonzero if stream needs to be closed.
484 */
485 static
486 int close_stream_check(struct relay_stream *stream)
487 {
488
489 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
490 /*
491 * We are about to close the stream so set the data pending flag to 1
492 * which will make the end data pending command skip the stream which
493 * is now closed and ready. Note that after proceeding to a file close,
494 * the written file is ready for reading.
495 */
496 stream->data_pending_check_done = 1;
497 return 1;
498 }
499 return 0;
500 }
501
502 /*
503 * This thread manages the listening for new connections on the network
504 */
505 static
506 void *relay_thread_listener(void *data)
507 {
508 int i, ret, pollfd, err = -1;
509 int val = 1;
510 uint32_t revents, nb_fd;
511 struct lttng_poll_event events;
512 struct lttcomm_sock *control_sock, *data_sock;
513
514 DBG("[thread] Relay listener started");
515
516 control_sock = relay_init_sock(control_uri);
517 if (!control_sock) {
518 goto error_sock_control;
519 }
520
521 data_sock = relay_init_sock(data_uri);
522 if (!data_sock) {
523 goto error_sock_relay;
524 }
525
526 /*
527 * Pass 3 as size here for the thread quit pipe, control and data socket.
528 */
529 ret = create_thread_poll_set(&events, 3);
530 if (ret < 0) {
531 goto error_create_poll;
532 }
533
534 /* Add the control socket */
535 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
536 if (ret < 0) {
537 goto error_poll_add;
538 }
539
540 /* Add the data socket */
541 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
542 if (ret < 0) {
543 goto error_poll_add;
544 }
545
546 while (1) {
547 DBG("Listener accepting connections");
548
549 restart:
550 ret = lttng_poll_wait(&events, -1);
551 if (ret < 0) {
552 /*
553 * Restart interrupted system call.
554 */
555 if (errno == EINTR) {
556 goto restart;
557 }
558 goto error;
559 }
560
561 nb_fd = ret;
562
563 DBG("Relay new connection received");
564 for (i = 0; i < nb_fd; i++) {
565 /* Fetch once the poll data */
566 revents = LTTNG_POLL_GETEV(&events, i);
567 pollfd = LTTNG_POLL_GETFD(&events, i);
568
569 /* Thread quit pipe has been closed. Killing thread. */
570 ret = check_thread_quit_pipe(pollfd, revents);
571 if (ret) {
572 err = 0;
573 goto exit;
574 }
575
576 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
577 ERR("socket poll error");
578 goto error;
579 } else if (revents & LPOLLIN) {
580 /*
581 * Get allocated in this thread,
582 * enqueued to a global queue, dequeued
583 * and freed in the worker thread.
584 */
585 struct relay_command *relay_cmd;
586 struct lttcomm_sock *newsock;
587
588 relay_cmd = zmalloc(sizeof(struct relay_command));
589 if (relay_cmd == NULL) {
590 PERROR("relay command zmalloc");
591 goto error;
592 }
593
594 if (pollfd == data_sock->fd) {
595 newsock = data_sock->ops->accept(data_sock);
596 if (!newsock) {
597 PERROR("accepting data sock");
598 free(relay_cmd);
599 goto error;
600 }
601 relay_cmd->type = RELAY_DATA;
602 DBG("Relay data connection accepted, socket %d", newsock->fd);
603 } else {
604 assert(pollfd == control_sock->fd);
605 newsock = control_sock->ops->accept(control_sock);
606 if (!newsock) {
607 PERROR("accepting control sock");
608 free(relay_cmd);
609 goto error;
610 }
611 relay_cmd->type = RELAY_CONTROL;
612 DBG("Relay control connection accepted, socket %d", newsock->fd);
613 }
614 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
615 &val, sizeof(int));
616 if (ret < 0) {
617 PERROR("setsockopt inet");
618 lttcomm_destroy_sock(newsock);
619 free(relay_cmd);
620 goto error;
621 }
622 relay_cmd->sock = newsock;
623 /*
624 * Lock free enqueue the request.
625 */
626 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
627
628 /*
629 * Wake the dispatch queue futex. Implicit memory
630 * barrier with the exchange in cds_wfq_enqueue.
631 */
632 futex_nto1_wake(&relay_cmd_queue.futex);
633 }
634 }
635 }
636
637 exit:
638 error:
639 error_poll_add:
640 lttng_poll_clean(&events);
641 error_create_poll:
642 if (data_sock->fd >= 0) {
643 ret = data_sock->ops->close(data_sock);
644 if (ret) {
645 PERROR("close");
646 }
647 }
648 lttcomm_destroy_sock(data_sock);
649 error_sock_relay:
650 if (control_sock->fd >= 0) {
651 ret = control_sock->ops->close(control_sock);
652 if (ret) {
653 PERROR("close");
654 }
655 }
656 lttcomm_destroy_sock(control_sock);
657 error_sock_control:
658 if (err) {
659 DBG("Thread exited with error");
660 }
661 DBG("Relay listener thread cleanup complete");
662 stop_threads();
663 return NULL;
664 }
665
666 /*
667 * This thread manages the dispatching of the requests to worker threads
668 */
669 static
670 void *relay_thread_dispatcher(void *data)
671 {
672 int ret;
673 struct cds_wfq_node *node;
674 struct relay_command *relay_cmd = NULL;
675
676 DBG("[thread] Relay dispatcher started");
677
678 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
679 /* Atomically prepare the queue futex */
680 futex_nto1_prepare(&relay_cmd_queue.futex);
681
682 do {
683 /* Dequeue commands */
684 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
685 if (node == NULL) {
686 DBG("Woken up but nothing in the relay command queue");
687 /* Continue thread execution */
688 break;
689 }
690
691 relay_cmd = caa_container_of(node, struct relay_command, node);
692 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
693
694 /*
695 * Inform worker thread of the new request. This
696 * call is blocking so we can be assured that the data will be read
697 * at some point in time or wait to the end of the world :)
698 */
699 do {
700 ret = write(relay_cmd_pipe[1], relay_cmd,
701 sizeof(struct relay_command));
702 } while (ret < 0 && errno == EINTR);
703 free(relay_cmd);
704 if (ret < 0 || ret != sizeof(struct relay_command)) {
705 PERROR("write cmd pipe");
706 goto error;
707 }
708 } while (node != NULL);
709
710 /* Futex wait on queue. Blocking call on futex() */
711 futex_nto1_wait(&relay_cmd_queue.futex);
712 }
713
714 error:
715 DBG("Dispatch thread dying");
716 stop_threads();
717 return NULL;
718 }
719
720 /*
721 * Get stream from stream id.
722 * Need to be called with RCU read-side lock held.
723 */
724 struct relay_stream *relay_stream_find_by_id(uint64_t stream_id)
725 {
726 struct lttng_ht_node_ulong *node;
727 struct lttng_ht_iter iter;
728 struct relay_stream *ret;
729
730 lttng_ht_lookup(relay_streams_ht,
731 (void *)((unsigned long) stream_id),
732 &iter);
733 node = lttng_ht_iter_get_node_ulong(&iter);
734 if (node == NULL) {
735 DBG("Relay stream %" PRIu64 " not found", stream_id);
736 ret = NULL;
737 goto end;
738 }
739
740 ret = caa_container_of(node, struct relay_stream, stream_n);
741
742 end:
743 return ret;
744 }
745
746 static
747 void deferred_free_stream(struct rcu_head *head)
748 {
749 struct relay_stream *stream =
750 caa_container_of(head, struct relay_stream, rcu_node);
751
752 ctf_trace_try_destroy(stream->ctf_trace);
753
754 free(stream->path_name);
755 free(stream->channel_name);
756 free(stream);
757 }
758
759 static
760 void deferred_free_session(struct rcu_head *head)
761 {
762 struct relay_session *session =
763 caa_container_of(head, struct relay_session, rcu_node);
764 free(session);
765 }
766
767 static void close_stream(struct relay_stream *stream,
768 struct lttng_ht *ctf_traces_ht)
769 {
770 int delret;
771 struct relay_viewer_stream *vstream;
772 struct lttng_ht_iter iter;
773
774 assert(stream);
775
776 delret = close(stream->fd);
777 if (delret < 0) {
778 PERROR("close stream");
779 }
780
781 if (stream->index_fd >= 0) {
782 delret = close(stream->index_fd);
783 if (delret < 0) {
784 PERROR("close stream index_fd");
785 }
786 }
787
788 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
789 if (vstream) {
790 /*
791 * Set the last good value into the viewer stream. This is done
792 * right before the stream gets deleted from the hash table. The
793 * lookup failure on the live thread side of a stream indicates
794 * that the viewer stream index received value should be used.
795 */
796 vstream->total_index_received = stream->total_index_received;
797 }
798
799 iter.iter.node = &stream->stream_n.node;
800 delret = lttng_ht_del(relay_streams_ht, &iter);
801 assert(!delret);
802 iter.iter.node = &stream->ctf_trace_node.node;
803 delret = lttng_ht_del(ctf_traces_ht, &iter);
804 assert(!delret);
805 call_rcu(&stream->rcu_node, deferred_free_stream);
806 DBG("Closed tracefile %d from close stream", stream->fd);
807 }
808
809 /*
810 * relay_delete_session: Free all memory associated with a session and
811 * close all the FDs
812 */
813 static
814 void relay_delete_session(struct relay_command *cmd,
815 struct lttng_ht *sessions_ht)
816 {
817 struct lttng_ht_iter iter;
818 struct lttng_ht_node_ulong *node;
819 struct relay_stream *stream;
820 int ret;
821
822 if (!cmd->session) {
823 return;
824 }
825
826 DBG("Relay deleting session %" PRIu64, cmd->session->id);
827
828 rcu_read_lock();
829 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
830 node = lttng_ht_iter_get_node_ulong(&iter);
831 if (node) {
832 stream = caa_container_of(node,
833 struct relay_stream, stream_n);
834 if (stream->session == cmd->session) {
835 ret = close(stream->fd);
836 if (ret < 0) {
837 PERROR("close stream fd on delete session");
838 }
839 ret = lttng_ht_del(relay_streams_ht, &iter);
840 assert(!ret);
841 call_rcu(&stream->rcu_node,
842 deferred_free_stream);
843 }
844 /* Cleanup index of that stream. */
845 relay_index_destroy_by_stream_id(stream->stream_handle,
846 indexes_ht);
847 }
848 }
849 iter.iter.node = &cmd->session->session_n.node;
850 ret = lttng_ht_del(sessions_ht, &iter);
851 assert(!ret);
852 call_rcu(&cmd->session->rcu_node,
853 deferred_free_session);
854 rcu_read_unlock();
855 }
856
857 /*
858 * Copy index data from the control port to a given index object.
859 */
860 static void copy_index_control_data(struct relay_index *index,
861 struct lttcomm_relayd_index *data)
862 {
863 assert(index);
864 assert(data);
865
866 /*
867 * The index on disk is encoded in big endian, so we don't need to convert
868 * the data received on the network. The data_offset value is NEVER
869 * modified here and is updated by the data thread.
870 */
871 index->index_data.packet_size = data->packet_size;
872 index->index_data.content_size = data->content_size;
873 index->index_data.timestamp_begin = data->timestamp_begin;
874 index->index_data.timestamp_end = data->timestamp_end;
875 index->index_data.events_discarded = data->events_discarded;
876 index->index_data.stream_id = data->stream_id;
877 }
878
879 /*
880 * Handle the RELAYD_CREATE_SESSION command.
881 *
882 * On success, send back the session id or else return a negative value.
883 */
884 static
885 int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
886 struct relay_command *cmd,
887 struct lttng_ht *sessions_ht)
888 {
889 int ret = 0, send_ret;
890 struct relay_session *session;
891 struct lttcomm_relayd_status_session reply;
892
893 assert(recv_hdr);
894 assert(cmd);
895
896 memset(&reply, 0, sizeof(reply));
897
898 session = zmalloc(sizeof(struct relay_session));
899 if (session == NULL) {
900 PERROR("relay session zmalloc");
901 ret = -1;
902 goto error;
903 }
904
905 session->id = ++last_relay_session_id;
906 session->sock = cmd->sock;
907 cmd->session = session;
908
909 reply.session_id = htobe64(session->id);
910
911 switch (cmd->minor) {
912 case 4: /* LTTng sessiond 2.4 */
913 default:
914 ret = cmd_create_session_2_4(cmd, session);
915 break;
916 }
917
918 lttng_ht_node_init_ulong(&session->session_n,
919 (unsigned long) session->id);
920 lttng_ht_add_unique_ulong(sessions_ht,
921 &session->session_n);
922
923 DBG("Created session %" PRIu64, session->id);
924
925 error:
926 if (ret < 0) {
927 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
928 } else {
929 reply.ret_code = htobe32(LTTNG_OK);
930 }
931
932 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
933 if (send_ret < 0) {
934 ERR("Relayd sending session id");
935 ret = send_ret;
936 }
937
938 return ret;
939 }
940
941 /*
942 * relay_add_stream: allocate a new stream for a session
943 */
944 static
945 int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
946 struct relay_command *cmd, struct lttng_ht *sessions_ht)
947 {
948 struct relay_session *session = cmd->session;
949 struct relay_stream *stream = NULL;
950 struct lttcomm_relayd_status_stream reply;
951 int ret, send_ret;
952
953 if (!session || cmd->version_check_done == 0) {
954 ERR("Trying to add a stream before version check");
955 ret = -1;
956 goto end_no_session;
957 }
958
959 stream = zmalloc(sizeof(struct relay_stream));
960 if (stream == NULL) {
961 PERROR("relay stream zmalloc");
962 ret = -1;
963 goto end_no_session;
964 }
965
966 switch (cmd->minor) {
967 case 1: /* LTTng sessiond 2.1 */
968 ret = cmd_recv_stream_2_1(cmd, stream);
969 break;
970 case 2: /* LTTng sessiond 2.2 */
971 default:
972 ret = cmd_recv_stream_2_2(cmd, stream);
973 break;
974 }
975 if (ret < 0) {
976 goto err_free_stream;
977 }
978
979 rcu_read_lock();
980 stream->stream_handle = ++last_relay_stream_id;
981 stream->prev_seq = -1ULL;
982 stream->session = session;
983 stream->index_fd = -1;
984 stream->read_index_fd = -1;
985 stream->ctf_trace = NULL;
986 pthread_mutex_init(&stream->lock, NULL);
987
988 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
989 if (ret < 0) {
990 ERR("relay creating output directory");
991 goto end;
992 }
993
994 /*
995 * No need to use run_as API here because whatever we receives, the relayd
996 * uses its own credentials for the stream files.
997 */
998 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
999 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
1000 if (ret < 0) {
1001 ERR("Create output file");
1002 goto end;
1003 }
1004 stream->fd = ret;
1005 if (stream->tracefile_size) {
1006 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1007 } else {
1008 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1009 }
1010
1011 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1012 stream->metadata_flag = 1;
1013 /*
1014 * When we receive a new metadata stream, we create a new
1015 * ctf_trace and we assign this ctf_trace to all streams with
1016 * the same path.
1017 *
1018 * If later on we receive a new stream for the same ctf_trace,
1019 * we copy the information from the first hit in the HT to the
1020 * new stream.
1021 */
1022 stream->ctf_trace = ctf_trace_create();
1023 if (!stream->ctf_trace) {
1024 ret = -1;
1025 goto end;
1026 }
1027 stream->ctf_trace->refcount++;
1028 stream->ctf_trace->metadata_stream = stream;
1029 }
1030 ctf_trace_assign(cmd->ctf_traces_ht, stream);
1031
1032 lttng_ht_node_init_ulong(&stream->stream_n,
1033 (unsigned long) stream->stream_handle);
1034 lttng_ht_add_unique_ulong(relay_streams_ht,
1035 &stream->stream_n);
1036
1037 lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name);
1038 lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node);
1039
1040 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1041 stream->stream_handle);
1042
1043 end:
1044 reply.handle = htobe64(stream->stream_handle);
1045 /* send the session id to the client or a negative return code on error */
1046 if (ret < 0) {
1047 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1048 /* stream was not properly added to the ht, so free it */
1049 free(stream);
1050 } else {
1051 reply.ret_code = htobe32(LTTNG_OK);
1052 }
1053
1054 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1055 sizeof(struct lttcomm_relayd_status_stream), 0);
1056 if (send_ret < 0) {
1057 ERR("Relay sending stream id");
1058 ret = send_ret;
1059 }
1060 rcu_read_unlock();
1061
1062 end_no_session:
1063 return ret;
1064
1065 err_free_stream:
1066 free(stream->path_name);
1067 free(stream->channel_name);
1068 free(stream);
1069 return ret;
1070 }
1071
1072 /*
1073 * relay_close_stream: close a specific stream
1074 */
1075 static
1076 int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1077 struct relay_command *cmd)
1078 {
1079 int ret, send_ret;
1080 struct relay_session *session = cmd->session;
1081 struct lttcomm_relayd_close_stream stream_info;
1082 struct lttcomm_relayd_generic_reply reply;
1083 struct relay_stream *stream;
1084
1085 DBG("Close stream received");
1086
1087 if (!session || cmd->version_check_done == 0) {
1088 ERR("Trying to close a stream before version check");
1089 ret = -1;
1090 goto end_no_session;
1091 }
1092
1093 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1094 sizeof(struct lttcomm_relayd_close_stream), 0);
1095 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1096 if (ret == 0) {
1097 /* Orderly shutdown. Not necessary to print an error. */
1098 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1099 } else {
1100 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1101 }
1102 ret = -1;
1103 goto end_no_session;
1104 }
1105
1106 rcu_read_lock();
1107 stream = relay_stream_find_by_id(be64toh(stream_info.stream_id));
1108 if (!stream) {
1109 ret = -1;
1110 goto end_unlock;
1111 }
1112
1113 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1114 stream->close_flag = 1;
1115
1116 if (close_stream_check(stream)) {
1117 close_stream(stream, cmd->ctf_traces_ht);
1118 }
1119
1120 end_unlock:
1121 rcu_read_unlock();
1122
1123 if (ret < 0) {
1124 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1125 } else {
1126 reply.ret_code = htobe32(LTTNG_OK);
1127 }
1128 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1129 sizeof(struct lttcomm_relayd_generic_reply), 0);
1130 if (send_ret < 0) {
1131 ERR("Relay sending stream id");
1132 ret = send_ret;
1133 }
1134
1135 end_no_session:
1136 return ret;
1137 }
1138
1139 /*
1140 * relay_unknown_command: send -1 if received unknown command
1141 */
1142 static
1143 void relay_unknown_command(struct relay_command *cmd)
1144 {
1145 struct lttcomm_relayd_generic_reply reply;
1146 int ret;
1147
1148 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1149 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1150 sizeof(struct lttcomm_relayd_generic_reply), 0);
1151 if (ret < 0) {
1152 ERR("Relay sending unknown command");
1153 }
1154 }
1155
1156 /*
1157 * relay_start: send an acknowledgment to the client to tell if we are
1158 * ready to receive data. We are ready if a session is established.
1159 */
1160 static
1161 int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1162 struct relay_command *cmd)
1163 {
1164 int ret = htobe32(LTTNG_OK);
1165 struct lttcomm_relayd_generic_reply reply;
1166 struct relay_session *session = cmd->session;
1167
1168 if (!session) {
1169 DBG("Trying to start the streaming without a session established");
1170 ret = htobe32(LTTNG_ERR_UNK);
1171 }
1172
1173 reply.ret_code = ret;
1174 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1175 sizeof(struct lttcomm_relayd_generic_reply), 0);
1176 if (ret < 0) {
1177 ERR("Relay sending start ack");
1178 }
1179
1180 return ret;
1181 }
1182
1183 /*
1184 * Append padding to the file pointed by the file descriptor fd.
1185 */
1186 static int write_padding_to_file(int fd, uint32_t size)
1187 {
1188 int ret = 0;
1189 char *zeros;
1190
1191 if (size == 0) {
1192 goto end;
1193 }
1194
1195 zeros = zmalloc(size);
1196 if (zeros == NULL) {
1197 PERROR("zmalloc zeros for padding");
1198 ret = -1;
1199 goto end;
1200 }
1201
1202 do {
1203 ret = write(fd, zeros, size);
1204 } while (ret < 0 && errno == EINTR);
1205 if (ret < 0 || ret != size) {
1206 PERROR("write padding to file");
1207 }
1208
1209 free(zeros);
1210
1211 end:
1212 return ret;
1213 }
1214
1215 /*
1216 * relay_recv_metadata: receive the metada for the session.
1217 */
1218 static
1219 int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1220 struct relay_command *cmd)
1221 {
1222 int ret = htobe32(LTTNG_OK);
1223 struct relay_session *session = cmd->session;
1224 struct lttcomm_relayd_metadata_payload *metadata_struct;
1225 struct relay_stream *metadata_stream;
1226 uint64_t data_size, payload_size;
1227
1228 if (!session) {
1229 ERR("Metadata sent before version check");
1230 ret = -1;
1231 goto end;
1232 }
1233
1234 data_size = payload_size = be64toh(recv_hdr->data_size);
1235 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1236 ERR("Incorrect data size");
1237 ret = -1;
1238 goto end;
1239 }
1240 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1241
1242 if (data_buffer_size < data_size) {
1243 /* In case the realloc fails, we can free the memory */
1244 char *tmp_data_ptr;
1245
1246 tmp_data_ptr = realloc(data_buffer, data_size);
1247 if (!tmp_data_ptr) {
1248 ERR("Allocating data buffer");
1249 free(data_buffer);
1250 ret = -1;
1251 goto end;
1252 }
1253 data_buffer = tmp_data_ptr;
1254 data_buffer_size = data_size;
1255 }
1256 memset(data_buffer, 0, data_size);
1257 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
1258 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1259 if (ret < 0 || ret != data_size) {
1260 if (ret == 0) {
1261 /* Orderly shutdown. Not necessary to print an error. */
1262 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1263 } else {
1264 ERR("Relay didn't receive the whole metadata");
1265 }
1266 ret = -1;
1267 goto end;
1268 }
1269 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
1270
1271 rcu_read_lock();
1272 metadata_stream = relay_stream_find_by_id(
1273 be64toh(metadata_struct->stream_id));
1274 if (!metadata_stream) {
1275 ret = -1;
1276 goto end_unlock;
1277 }
1278
1279 do {
1280 ret = write(metadata_stream->fd, metadata_struct->payload,
1281 payload_size);
1282 } while (ret < 0 && errno == EINTR);
1283 if (ret < 0 || ret != payload_size) {
1284 ERR("Relay error writing metadata on file");
1285 ret = -1;
1286 goto end_unlock;
1287 }
1288
1289 ret = write_padding_to_file(metadata_stream->fd,
1290 be32toh(metadata_struct->padding_size));
1291 if (ret < 0) {
1292 goto end_unlock;
1293 }
1294 metadata_stream->ctf_trace->metadata_received +=
1295 payload_size + be32toh(metadata_struct->padding_size);
1296
1297 DBG2("Relay metadata written");
1298
1299 end_unlock:
1300 rcu_read_unlock();
1301 end:
1302 return ret;
1303 }
1304
1305 /*
1306 * relay_send_version: send relayd version number
1307 */
1308 static
1309 int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1310 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1311 {
1312 int ret;
1313 struct lttcomm_relayd_version reply, msg;
1314
1315 assert(cmd);
1316
1317 cmd->version_check_done = 1;
1318
1319 /* Get version from the other side. */
1320 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1321 if (ret < 0 || ret != sizeof(msg)) {
1322 if (ret == 0) {
1323 /* Orderly shutdown. Not necessary to print an error. */
1324 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1325 } else {
1326 ERR("Relay failed to receive the version values.");
1327 }
1328 ret = -1;
1329 goto end;
1330 }
1331
1332 reply.major = RELAYD_VERSION_COMM_MAJOR;
1333 reply.minor = RELAYD_VERSION_COMM_MINOR;
1334
1335 /* Major versions must be the same */
1336 if (reply.major != be32toh(msg.major)) {
1337 DBG("Incompatible major versions (%u vs %u), deleting session",
1338 reply.major, be32toh(msg.major));
1339 relay_delete_session(cmd, sessions_ht);
1340 ret = 0;
1341 goto end;
1342 }
1343
1344 cmd->major = reply.major;
1345 /* We adapt to the lowest compatible version */
1346 if (reply.minor <= be32toh(msg.minor)) {
1347 cmd->minor = reply.minor;
1348 } else {
1349 cmd->minor = be32toh(msg.minor);
1350 }
1351
1352 reply.major = htobe32(reply.major);
1353 reply.minor = htobe32(reply.minor);
1354 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1355 sizeof(struct lttcomm_relayd_version), 0);
1356 if (ret < 0) {
1357 ERR("Relay sending version");
1358 }
1359
1360 DBG("Version check done using protocol %u.%u", cmd->major,
1361 cmd->minor);
1362
1363 end:
1364 return ret;
1365 }
1366
1367 /*
1368 * Check for data pending for a given stream id from the session daemon.
1369 */
1370 static
1371 int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1372 struct relay_command *cmd)
1373 {
1374 struct relay_session *session = cmd->session;
1375 struct lttcomm_relayd_data_pending msg;
1376 struct lttcomm_relayd_generic_reply reply;
1377 struct relay_stream *stream;
1378 int ret;
1379 uint64_t last_net_seq_num, stream_id;
1380
1381 DBG("Data pending command received");
1382
1383 if (!session || cmd->version_check_done == 0) {
1384 ERR("Trying to check for data before version check");
1385 ret = -1;
1386 goto end_no_session;
1387 }
1388
1389 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1390 if (ret < sizeof(msg)) {
1391 if (ret == 0) {
1392 /* Orderly shutdown. Not necessary to print an error. */
1393 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1394 } else {
1395 ERR("Relay didn't receive valid data_pending struct size : %d",
1396 ret);
1397 }
1398 ret = -1;
1399 goto end_no_session;
1400 }
1401
1402 stream_id = be64toh(msg.stream_id);
1403 last_net_seq_num = be64toh(msg.last_net_seq_num);
1404
1405 rcu_read_lock();
1406 stream = relay_stream_find_by_id(stream_id);
1407 if (stream == NULL) {
1408 ret = -1;
1409 goto end_unlock;
1410 }
1411
1412 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
1413 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1414 last_net_seq_num);
1415
1416 /* Avoid wrapping issue */
1417 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
1418 /* Data has in fact been written and is NOT pending */
1419 ret = 0;
1420 } else {
1421 /* Data still being streamed thus pending */
1422 ret = 1;
1423 }
1424
1425 /* Pending check is now done. */
1426 stream->data_pending_check_done = 1;
1427
1428 end_unlock:
1429 rcu_read_unlock();
1430
1431 reply.ret_code = htobe32(ret);
1432 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1433 if (ret < 0) {
1434 ERR("Relay data pending ret code failed");
1435 }
1436
1437 end_no_session:
1438 return ret;
1439 }
1440
1441 /*
1442 * Wait for the control socket to reach a quiescent state.
1443 *
1444 * Note that for now, when receiving this command from the session daemon, this
1445 * means that every subsequent commands or data received on the control socket
1446 * has been handled. So, this is why we simply return OK here.
1447 */
1448 static
1449 int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1450 struct relay_command *cmd)
1451 {
1452 int ret;
1453 uint64_t stream_id;
1454 struct relay_stream *stream;
1455 struct lttng_ht_iter iter;
1456 struct lttcomm_relayd_quiescent_control msg;
1457 struct lttcomm_relayd_generic_reply reply;
1458
1459 DBG("Checking quiescent state on control socket");
1460
1461 if (!cmd->session || cmd->version_check_done == 0) {
1462 ERR("Trying to check for data before version check");
1463 ret = -1;
1464 goto end_no_session;
1465 }
1466
1467 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1468 if (ret < sizeof(msg)) {
1469 if (ret == 0) {
1470 /* Orderly shutdown. Not necessary to print an error. */
1471 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1472 } else {
1473 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1474 ret);
1475 }
1476 ret = -1;
1477 goto end_no_session;
1478 }
1479
1480 stream_id = be64toh(msg.stream_id);
1481
1482 rcu_read_lock();
1483 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1484 stream_n.node) {
1485 if (stream->stream_handle == stream_id) {
1486 stream->data_pending_check_done = 1;
1487 DBG("Relay quiescent control pending flag set to %" PRIu64,
1488 stream_id);
1489 break;
1490 }
1491 }
1492 rcu_read_unlock();
1493
1494 reply.ret_code = htobe32(LTTNG_OK);
1495 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1496 if (ret < 0) {
1497 ERR("Relay data quiescent control ret code failed");
1498 }
1499
1500 end_no_session:
1501 return ret;
1502 }
1503
1504 /*
1505 * Initialize a data pending command. This means that a client is about to ask
1506 * for data pending for each stream he/she holds. Simply iterate over all
1507 * streams of a session and set the data_pending_check_done flag.
1508 *
1509 * This command returns to the client a LTTNG_OK code.
1510 */
1511 static
1512 int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1513 struct relay_command *cmd)
1514 {
1515 int ret;
1516 struct lttng_ht_iter iter;
1517 struct lttcomm_relayd_begin_data_pending msg;
1518 struct lttcomm_relayd_generic_reply reply;
1519 struct relay_stream *stream;
1520 uint64_t session_id;
1521
1522 assert(recv_hdr);
1523 assert(cmd);
1524
1525 DBG("Init streams for data pending");
1526
1527 if (!cmd->session || cmd->version_check_done == 0) {
1528 ERR("Trying to check for data before version check");
1529 ret = -1;
1530 goto end_no_session;
1531 }
1532
1533 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1534 if (ret < sizeof(msg)) {
1535 if (ret == 0) {
1536 /* Orderly shutdown. Not necessary to print an error. */
1537 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1538 } else {
1539 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1540 ret);
1541 }
1542 ret = -1;
1543 goto end_no_session;
1544 }
1545
1546 session_id = be64toh(msg.session_id);
1547
1548 /*
1549 * Iterate over all streams to set the begin data pending flag. For now, the
1550 * streams are indexed by stream handle so we have to iterate over all
1551 * streams to find the one associated with the right session_id.
1552 */
1553 rcu_read_lock();
1554 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1555 stream_n.node) {
1556 if (stream->session->id == session_id) {
1557 stream->data_pending_check_done = 0;
1558 DBG("Set begin data pending flag to stream %" PRIu64,
1559 stream->stream_handle);
1560 }
1561 }
1562 rcu_read_unlock();
1563
1564 /* All good, send back reply. */
1565 reply.ret_code = htobe32(LTTNG_OK);
1566
1567 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1568 if (ret < 0) {
1569 ERR("Relay begin data pending send reply failed");
1570 }
1571
1572 end_no_session:
1573 return ret;
1574 }
1575
1576 /*
1577 * End data pending command. This will check, for a given session id, if each
1578 * stream associated with it has its data_pending_check_done flag set. If not,
1579 * this means that the client lost track of the stream but the data is still
1580 * being streamed on our side. In this case, we inform the client that data is
1581 * inflight.
1582 *
1583 * Return to the client if there is data in flight or not with a ret_code.
1584 */
1585 static
1586 int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
1587 struct relay_command *cmd)
1588 {
1589 int ret;
1590 struct lttng_ht_iter iter;
1591 struct lttcomm_relayd_end_data_pending msg;
1592 struct lttcomm_relayd_generic_reply reply;
1593 struct relay_stream *stream;
1594 uint64_t session_id;
1595 uint32_t is_data_inflight = 0;
1596
1597 assert(recv_hdr);
1598 assert(cmd);
1599
1600 DBG("End data pending command");
1601
1602 if (!cmd->session || cmd->version_check_done == 0) {
1603 ERR("Trying to check for data before version check");
1604 ret = -1;
1605 goto end_no_session;
1606 }
1607
1608 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
1609 if (ret < sizeof(msg)) {
1610 if (ret == 0) {
1611 /* Orderly shutdown. Not necessary to print an error. */
1612 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1613 } else {
1614 ERR("Relay didn't receive valid end data_pending struct size: %d",
1615 ret);
1616 }
1617 ret = -1;
1618 goto end_no_session;
1619 }
1620
1621 session_id = be64toh(msg.session_id);
1622
1623 /* Iterate over all streams to see if the begin data pending flag is set. */
1624 rcu_read_lock();
1625 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
1626 stream_n.node) {
1627 if (stream->session->id == session_id &&
1628 !stream->data_pending_check_done) {
1629 is_data_inflight = 1;
1630 DBG("Data is still in flight for stream %" PRIu64,
1631 stream->stream_handle);
1632 break;
1633 }
1634 }
1635 rcu_read_unlock();
1636
1637 /* All good, send back reply. */
1638 reply.ret_code = htobe32(is_data_inflight);
1639
1640 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1641 if (ret < 0) {
1642 ERR("Relay end data pending send reply failed");
1643 }
1644
1645 end_no_session:
1646 return ret;
1647 }
1648
1649 /*
1650 * Receive an index for a specific stream.
1651 *
1652 * Return 0 on success else a negative value.
1653 */
1654 static
1655 int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
1656 struct relay_command *cmd, struct lttng_ht *indexes_ht)
1657 {
1658 int ret, send_ret, index_created = 0;
1659 struct relay_session *session = cmd->session;
1660 struct lttcomm_relayd_index index_info;
1661 struct relay_index *index, *wr_index = NULL;
1662 struct lttcomm_relayd_generic_reply reply;
1663 struct relay_stream *stream;
1664 uint64_t net_seq_num;
1665
1666 assert(cmd);
1667 assert(indexes_ht);
1668
1669 DBG("Relay receiving index");
1670
1671 if (!session || cmd->version_check_done == 0) {
1672 ERR("Trying to close a stream before version check");
1673 ret = -1;
1674 goto end_no_session;
1675 }
1676
1677 ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info,
1678 sizeof(index_info), 0);
1679 if (ret < sizeof(index_info)) {
1680 if (ret == 0) {
1681 /* Orderly shutdown. Not necessary to print an error. */
1682 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1683 } else {
1684 ERR("Relay didn't receive valid index struct size : %d", ret);
1685 }
1686 ret = -1;
1687 goto end_no_session;
1688 }
1689
1690 net_seq_num = be64toh(index_info.net_seq_num);
1691
1692 rcu_read_lock();
1693 stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id));
1694 if (!stream) {
1695 ret = -1;
1696 goto end_rcu_unlock;
1697 }
1698
1699 /* Live beacon handling */
1700 if (index_info.packet_size == 0) {
1701 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1702
1703 /*
1704 * Only flag a stream inactive when it has already received data.
1705 */
1706 if (stream->total_index_received > 0) {
1707 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1708 }
1709 ret = 0;
1710 goto end_rcu_unlock;
1711 } else {
1712 stream->beacon_ts_end = -1ULL;
1713 }
1714
1715 index = relay_index_find(stream->stream_handle, net_seq_num, indexes_ht);
1716 if (!index) {
1717 /* A successful creation will add the object to the HT. */
1718 index = relay_index_create(stream->stream_handle, net_seq_num);
1719 if (!index) {
1720 goto end_rcu_unlock;
1721 }
1722 index_created = 1;
1723 }
1724
1725 copy_index_control_data(index, &index_info);
1726
1727 if (index_created) {
1728 /*
1729 * Try to add the relay index object to the hash table. If an object
1730 * already exist, destroy back the index created, set the data in this
1731 * object and write it on disk.
1732 */
1733 relay_index_add(index, indexes_ht, &wr_index);
1734 if (wr_index) {
1735 copy_index_control_data(wr_index, &index_info);
1736 free(index);
1737 }
1738 } else {
1739 /* The index already exists so write it on disk. */
1740 wr_index = index;
1741 }
1742
1743 /* Do we have a writable ready index to write on disk. */
1744 if (wr_index) {
1745 /* Starting at 2.4, create the index file if none available. */
1746 if (cmd->minor >= 4 && stream->index_fd < 0) {
1747 ret = index_create_file(stream->path_name, stream->channel_name,
1748 relayd_uid, relayd_gid, stream->tracefile_size,
1749 stream->tracefile_count_current);
1750 if (ret < 0) {
1751 goto end_rcu_unlock;
1752 }
1753 stream->index_fd = ret;
1754 }
1755
1756 ret = relay_index_write(wr_index->fd, wr_index, indexes_ht);
1757 if (ret < 0) {
1758 goto end_rcu_unlock;
1759 }
1760 stream->total_index_received++;
1761 }
1762
1763 end_rcu_unlock:
1764 rcu_read_unlock();
1765
1766 if (ret < 0) {
1767 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1768 } else {
1769 reply.ret_code = htobe32(LTTNG_OK);
1770 }
1771 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1772 if (send_ret < 0) {
1773 ERR("Relay sending close index id reply");
1774 ret = send_ret;
1775 }
1776
1777 end_no_session:
1778 return ret;
1779 }
1780
1781 /*
1782 * Process the commands received on the control socket
1783 */
1784 static
1785 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1786 struct relay_command *cmd, struct relay_local_data *ctx)
1787 {
1788 int ret = 0;
1789
1790 switch (be32toh(recv_hdr->cmd)) {
1791 case RELAYD_CREATE_SESSION:
1792 ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht);
1793 break;
1794 case RELAYD_ADD_STREAM:
1795 ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht);
1796 break;
1797 case RELAYD_START_DATA:
1798 ret = relay_start(recv_hdr, cmd);
1799 break;
1800 case RELAYD_SEND_METADATA:
1801 ret = relay_recv_metadata(recv_hdr, cmd);
1802 break;
1803 case RELAYD_VERSION:
1804 ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht);
1805 break;
1806 case RELAYD_CLOSE_STREAM:
1807 ret = relay_close_stream(recv_hdr, cmd);
1808 break;
1809 case RELAYD_DATA_PENDING:
1810 ret = relay_data_pending(recv_hdr, cmd);
1811 break;
1812 case RELAYD_QUIESCENT_CONTROL:
1813 ret = relay_quiescent_control(recv_hdr, cmd);
1814 break;
1815 case RELAYD_BEGIN_DATA_PENDING:
1816 ret = relay_begin_data_pending(recv_hdr, cmd);
1817 break;
1818 case RELAYD_END_DATA_PENDING:
1819 ret = relay_end_data_pending(recv_hdr, cmd);
1820 break;
1821 case RELAYD_SEND_INDEX:
1822 ret = relay_recv_index(recv_hdr, cmd, indexes_ht);
1823 break;
1824 case RELAYD_UPDATE_SYNC_INFO:
1825 default:
1826 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1827 relay_unknown_command(cmd);
1828 ret = -1;
1829 goto end;
1830 }
1831
1832 end:
1833 return ret;
1834 }
1835
1836 /*
1837 * relay_process_data: Process the data received on the data socket
1838 */
1839 static
1840 int relay_process_data(struct relay_command *cmd,
1841 struct lttng_ht *indexes_ht)
1842 {
1843 int ret = 0, rotate_index = 0, index_created = 0;
1844 struct relay_stream *stream;
1845 struct relay_index *index, *wr_index = NULL;
1846 struct lttcomm_relayd_data_hdr data_hdr;
1847 uint64_t stream_id, data_offset;
1848 uint64_t net_seq_num;
1849 uint32_t data_size;
1850
1851 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1852 sizeof(struct lttcomm_relayd_data_hdr), 0);
1853 if (ret <= 0) {
1854 if (ret == 0) {
1855 /* Orderly shutdown. Not necessary to print an error. */
1856 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1857 } else {
1858 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
1859 }
1860 ret = -1;
1861 goto end;
1862 }
1863
1864 stream_id = be64toh(data_hdr.stream_id);
1865
1866 rcu_read_lock();
1867 stream = relay_stream_find_by_id(stream_id);
1868 if (!stream) {
1869 ret = -1;
1870 goto end_rcu_unlock;
1871 }
1872
1873 data_size = be32toh(data_hdr.data_size);
1874 if (data_buffer_size < data_size) {
1875 char *tmp_data_ptr;
1876
1877 tmp_data_ptr = realloc(data_buffer, data_size);
1878 if (!tmp_data_ptr) {
1879 ERR("Allocating data buffer");
1880 free(data_buffer);
1881 ret = -1;
1882 goto end_rcu_unlock;
1883 }
1884 data_buffer = tmp_data_ptr;
1885 data_buffer_size = data_size;
1886 }
1887 memset(data_buffer, 0, data_size);
1888
1889 net_seq_num = be64toh(data_hdr.net_seq_num);
1890
1891 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
1892 data_size, stream_id, net_seq_num);
1893 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1894 if (ret <= 0) {
1895 if (ret == 0) {
1896 /* Orderly shutdown. Not necessary to print an error. */
1897 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1898 }
1899 ret = -1;
1900 goto end_rcu_unlock;
1901 }
1902
1903 /* Check if a rotation is needed. */
1904 if (stream->tracefile_size > 0 &&
1905 (stream->tracefile_size_current + data_size) >
1906 stream->tracefile_size) {
1907 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
1908 stream->tracefile_size, stream->tracefile_count,
1909 relayd_uid, relayd_gid, stream->fd,
1910 &(stream->tracefile_count_current), &stream->fd);
1911 if (ret < 0) {
1912 ERR("Rotating stream output file");
1913 goto end_rcu_unlock;
1914 }
1915 /* Reset current size because we just perform a stream rotation. */
1916 stream->tracefile_size_current = 0;
1917 rotate_index = 1;
1918 }
1919
1920 /* Get data offset because we are about to update the index. */
1921 data_offset = htobe64(stream->tracefile_size_current);
1922
1923 /*
1924 * Lookup for an existing index for that stream id/sequence number. If on
1925 * exists, the control thread already received the data for it thus we need
1926 * to write it on disk.
1927 */
1928 index = relay_index_find(stream_id, net_seq_num, indexes_ht);
1929 if (!index) {
1930 /* A successful creation will add the object to the HT. */
1931 index = relay_index_create(stream->stream_handle, net_seq_num);
1932 if (!index) {
1933 goto end_rcu_unlock;
1934 }
1935 index_created = 1;
1936 }
1937
1938 if (rotate_index || stream->index_fd < 0) {
1939 index->to_close_fd = stream->index_fd;
1940 ret = index_create_file(stream->path_name, stream->channel_name,
1941 relayd_uid, relayd_gid, stream->tracefile_size,
1942 stream->tracefile_count_current);
1943 if (ret < 0) {
1944 /* This will close the stream's index fd if one. */
1945 relay_index_free_safe(index);
1946 goto end_rcu_unlock;
1947 }
1948 stream->index_fd = ret;
1949 }
1950 index->fd = stream->index_fd;
1951 index->index_data.offset = data_offset;
1952
1953 if (index_created) {
1954 /*
1955 * Try to add the relay index object to the hash table. If an object
1956 * already exist, destroy back the index created and set the data.
1957 */
1958 relay_index_add(index, indexes_ht, &wr_index);
1959 if (wr_index) {
1960 /* Copy back data from the created index. */
1961 wr_index->fd = index->fd;
1962 wr_index->to_close_fd = index->to_close_fd;
1963 wr_index->index_data.offset = data_offset;
1964 free(index);
1965 }
1966 } else {
1967 /* The index already exists so write it on disk. */
1968 wr_index = index;
1969 }
1970
1971 /* Do we have a writable ready index to write on disk. */
1972 if (wr_index) {
1973 /* Starting at 2.4, create the index file if none available. */
1974 if (cmd->minor >= 4 && stream->index_fd < 0) {
1975 ret = index_create_file(stream->path_name, stream->channel_name,
1976 relayd_uid, relayd_gid, stream->tracefile_size,
1977 stream->tracefile_count_current);
1978 if (ret < 0) {
1979 goto end_rcu_unlock;
1980 }
1981 stream->index_fd = ret;
1982 }
1983
1984 ret = relay_index_write(wr_index->fd, wr_index, indexes_ht);
1985 if (ret < 0) {
1986 goto end_rcu_unlock;
1987 }
1988 stream->total_index_received++;
1989 }
1990
1991 do {
1992 ret = write(stream->fd, data_buffer, data_size);
1993 } while (ret < 0 && errno == EINTR);
1994 if (ret < 0 || ret != data_size) {
1995 ERR("Relay error writing data to file");
1996 ret = -1;
1997 goto end_rcu_unlock;
1998 }
1999
2000 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2001 ret, stream->stream_handle);
2002
2003 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2004 if (ret < 0) {
2005 goto end_rcu_unlock;
2006 }
2007 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2008
2009 stream->prev_seq = net_seq_num;
2010
2011 /* Check if we need to close the FD */
2012 if (close_stream_check(stream)) {
2013 close_stream(stream, cmd->ctf_traces_ht);
2014 }
2015
2016 end_rcu_unlock:
2017 rcu_read_unlock();
2018 end:
2019 return ret;
2020 }
2021
2022 static
2023 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
2024 {
2025 int ret;
2026
2027 lttng_poll_del(events, pollfd);
2028
2029 ret = close(pollfd);
2030 if (ret < 0) {
2031 ERR("Closing pollfd %d", pollfd);
2032 }
2033 }
2034
2035 static
2036 int relay_add_connection(int fd, struct lttng_poll_event *events,
2037 struct lttng_ht *relay_connections_ht)
2038 {
2039 struct relay_command *relay_connection;
2040 int ret;
2041
2042 relay_connection = zmalloc(sizeof(struct relay_command));
2043 if (relay_connection == NULL) {
2044 PERROR("Relay command zmalloc");
2045 goto error;
2046 }
2047 do {
2048 ret = read(fd, relay_connection, sizeof(struct relay_command));
2049 } while (ret < 0 && errno == EINTR);
2050 if (ret < 0 || ret < sizeof(struct relay_command)) {
2051 PERROR("read relay cmd pipe");
2052 goto error_read;
2053 }
2054
2055 relay_connection->ctf_traces_ht = lttng_ht_new(0, LTTNG_HT_TYPE_STRING);
2056 if (!relay_connection->ctf_traces_ht) {
2057 goto error_read;
2058 }
2059
2060 lttng_ht_node_init_ulong(&relay_connection->sock_n,
2061 (unsigned long) relay_connection->sock->fd);
2062 rcu_read_lock();
2063 lttng_ht_add_unique_ulong(relay_connections_ht,
2064 &relay_connection->sock_n);
2065 rcu_read_unlock();
2066 return lttng_poll_add(events,
2067 relay_connection->sock->fd,
2068 LPOLLIN | LPOLLRDHUP);
2069
2070 error_read:
2071 free(relay_connection);
2072 error:
2073 return -1;
2074 }
2075
2076 static
2077 void deferred_free_connection(struct rcu_head *head)
2078 {
2079 struct relay_command *relay_connection =
2080 caa_container_of(head, struct relay_command, rcu_node);
2081
2082 lttng_ht_destroy(relay_connection->ctf_traces_ht);
2083 lttcomm_destroy_sock(relay_connection->sock);
2084 free(relay_connection);
2085 }
2086
2087 static
2088 void relay_del_connection(struct lttng_ht *relay_connections_ht,
2089 struct lttng_ht_iter *iter, struct relay_command *relay_connection,
2090 struct lttng_ht *sessions_ht)
2091 {
2092 int ret;
2093
2094 ret = lttng_ht_del(relay_connections_ht, iter);
2095 assert(!ret);
2096 if (relay_connection->type == RELAY_CONTROL) {
2097 relay_delete_session(relay_connection, sessions_ht);
2098 }
2099
2100 call_rcu(&relay_connection->rcu_node,
2101 deferred_free_connection);
2102 }
2103
2104 /*
2105 * This thread does the actual work
2106 */
2107 static
2108 void *relay_thread_worker(void *data)
2109 {
2110 int ret, err = -1, last_seen_data_fd = -1;
2111 uint32_t nb_fd;
2112 struct relay_command *relay_connection;
2113 struct lttng_poll_event events;
2114 struct lttng_ht *relay_connections_ht;
2115 struct lttng_ht_node_ulong *node;
2116 struct lttng_ht_iter iter;
2117 struct lttcomm_relayd_hdr recv_hdr;
2118 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2119 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2120
2121 DBG("[thread] Relay worker started");
2122
2123 rcu_register_thread();
2124
2125 /* table of connections indexed on socket */
2126 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2127 if (!relay_connections_ht) {
2128 goto relay_connections_ht_error;
2129 }
2130
2131 /* Tables of received indexes indexed by index handle and net_seq_num. */
2132 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2133 if (!indexes_ht) {
2134 goto indexes_ht_error;
2135 }
2136
2137 ret = create_thread_poll_set(&events, 2);
2138 if (ret < 0) {
2139 goto error_poll_create;
2140 }
2141
2142 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
2143 if (ret < 0) {
2144 goto error;
2145 }
2146
2147 restart:
2148 while (1) {
2149 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2150
2151 /* Infinite blocking call, waiting for transmission */
2152 DBG3("Relayd worker thread polling...");
2153 ret = lttng_poll_wait(&events, -1);
2154 if (ret < 0) {
2155 /*
2156 * Restart interrupted system call.
2157 */
2158 if (errno == EINTR) {
2159 goto restart;
2160 }
2161 goto error;
2162 }
2163
2164 nb_fd = ret;
2165
2166 /*
2167 * Process control. The control connection is prioritised so we don't
2168 * starve it with high throughout put tracing data on the data
2169 * connection.
2170 */
2171 for (i = 0; i < nb_fd; i++) {
2172 /* Fetch once the poll data */
2173 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2174 int pollfd = LTTNG_POLL_GETFD(&events, i);
2175
2176 /* Thread quit pipe has been closed. Killing thread. */
2177 ret = check_thread_quit_pipe(pollfd, revents);
2178 if (ret) {
2179 err = 0;
2180 goto exit;
2181 }
2182
2183 /* Inspect the relay cmd pipe for new connection */
2184 if (pollfd == relay_cmd_pipe[0]) {
2185 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2186 ERR("Relay pipe error");
2187 goto error;
2188 } else if (revents & LPOLLIN) {
2189 DBG("Relay command received");
2190 ret = relay_add_connection(relay_cmd_pipe[0],
2191 &events, relay_connections_ht);
2192 if (ret < 0) {
2193 goto error;
2194 }
2195 }
2196 } else if (revents) {
2197 rcu_read_lock();
2198 lttng_ht_lookup(relay_connections_ht,
2199 (void *)((unsigned long) pollfd),
2200 &iter);
2201 node = lttng_ht_iter_get_node_ulong(&iter);
2202 if (node == NULL) {
2203 DBG2("Relay sock %d not found", pollfd);
2204 rcu_read_unlock();
2205 goto error;
2206 }
2207 relay_connection = caa_container_of(node,
2208 struct relay_command, sock_n);
2209
2210 if (revents & (LPOLLERR)) {
2211 ERR("POLL ERROR");
2212 relay_cleanup_poll_connection(&events, pollfd);
2213 relay_del_connection(relay_connections_ht,
2214 &iter, relay_connection, sessions_ht);
2215 if (last_seen_data_fd == pollfd) {
2216 last_seen_data_fd = last_notdel_data_fd;
2217 }
2218 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2219 DBG("Socket %d hung up", pollfd);
2220 relay_cleanup_poll_connection(&events, pollfd);
2221 relay_del_connection(relay_connections_ht,
2222 &iter, relay_connection, sessions_ht);
2223 if (last_seen_data_fd == pollfd) {
2224 last_seen_data_fd = last_notdel_data_fd;
2225 }
2226 } else if (revents & LPOLLIN) {
2227 /* control socket */
2228 if (relay_connection->type == RELAY_CONTROL) {
2229 ret = relay_connection->sock->ops->recvmsg(
2230 relay_connection->sock, &recv_hdr,
2231 sizeof(struct lttcomm_relayd_hdr), 0);
2232 /* connection closed */
2233 if (ret <= 0) {
2234 relay_cleanup_poll_connection(&events, pollfd);
2235 relay_del_connection(relay_connections_ht,
2236 &iter, relay_connection, sessions_ht);
2237 DBG("Control connection closed with %d", pollfd);
2238 } else {
2239 if (relay_connection->session) {
2240 DBG2("Relay worker receiving data for session : %" PRIu64,
2241 relay_connection->session->id);
2242 }
2243 ret = relay_process_control(&recv_hdr,
2244 relay_connection, relay_ctx);
2245 if (ret < 0) {
2246 /* Clear the session on error. */
2247 relay_cleanup_poll_connection(&events, pollfd);
2248 relay_del_connection(relay_connections_ht,
2249 &iter, relay_connection, sessions_ht);
2250 DBG("Connection closed with %d", pollfd);
2251 }
2252 seen_control = 1;
2253 }
2254 } else {
2255 /*
2256 * Flag the last seen data fd not deleted. It will be
2257 * used as the last seen fd if any fd gets deleted in
2258 * this first loop.
2259 */
2260 last_notdel_data_fd = pollfd;
2261 }
2262 }
2263 rcu_read_unlock();
2264 }
2265 }
2266
2267 /*
2268 * The last loop handled a control request, go back to poll to make
2269 * sure we prioritise the control socket.
2270 */
2271 if (seen_control) {
2272 continue;
2273 }
2274
2275 if (last_seen_data_fd >= 0) {
2276 for (i = 0; i < nb_fd; i++) {
2277 int pollfd = LTTNG_POLL_GETFD(&events, i);
2278 if (last_seen_data_fd == pollfd) {
2279 idx = i;
2280 break;
2281 }
2282 }
2283 }
2284
2285 /* Process data connection. */
2286 for (i = idx + 1; i < nb_fd; i++) {
2287 /* Fetch the poll data. */
2288 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2289 int pollfd = LTTNG_POLL_GETFD(&events, i);
2290
2291 /* Skip the command pipe. It's handled in the first loop. */
2292 if (pollfd == relay_cmd_pipe[0]) {
2293 continue;
2294 }
2295
2296 if (revents) {
2297 rcu_read_lock();
2298 lttng_ht_lookup(relay_connections_ht,
2299 (void *)((unsigned long) pollfd),
2300 &iter);
2301 node = lttng_ht_iter_get_node_ulong(&iter);
2302 if (node == NULL) {
2303 /* Skip it. Might be removed before. */
2304 rcu_read_unlock();
2305 continue;
2306 }
2307 relay_connection = caa_container_of(node,
2308 struct relay_command, sock_n);
2309
2310 if (revents & LPOLLIN) {
2311 if (relay_connection->type != RELAY_DATA) {
2312 continue;
2313 }
2314
2315 ret = relay_process_data(relay_connection, indexes_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 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2486 if (!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(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 }
LTTng 2.0 tools and control repository
This page took 0.077869 seconds and 5 git commands to generate.