projects / lttng-tools.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Fix: make indexes HT global for real
[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 /* We need those values for the file/dir creation. */
105 static uid_t relayd_uid;
106 static gid_t relayd_gid;
107
108 /* Global relay stream hash table. */
109 struct lttng_ht *relay_streams_ht;
110
111 /* Global relay viewer stream hash table. */
112 struct lttng_ht *viewer_streams_ht;
113
114 /* Global hash table that stores relay index object. */
115 struct lttng_ht *indexes_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 /*
768 * Close a given stream. The stream is freed using a call RCU.
769 *
770 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
771 * BEFORE the stream lock MUST be acquired.
772 */
773 static void destroy_stream(struct relay_stream *stream,
774 struct lttng_ht *ctf_traces_ht)
775 {
776 int delret;
777 struct relay_viewer_stream *vstream;
778 struct lttng_ht_iter iter;
779
780 assert(stream);
781
782 delret = close(stream->fd);
783 if (delret < 0) {
784 PERROR("close stream");
785 }
786
787 if (stream->index_fd >= 0) {
788 delret = close(stream->index_fd);
789 if (delret < 0) {
790 PERROR("close stream index_fd");
791 }
792 }
793
794 vstream = live_find_viewer_stream_by_id(stream->stream_handle);
795 if (vstream) {
796 /*
797 * Set the last good value into the viewer stream. This is done
798 * right before the stream gets deleted from the hash table. The
799 * lookup failure on the live thread side of a stream indicates
800 * that the viewer stream index received value should be used.
801 */
802 vstream->total_index_received = stream->total_index_received;
803 }
804
805 iter.iter.node = &stream->stream_n.node;
806 delret = lttng_ht_del(relay_streams_ht, &iter);
807 assert(!delret);
808 iter.iter.node = &stream->ctf_trace_node.node;
809 delret = lttng_ht_del(ctf_traces_ht, &iter);
810 assert(!delret);
811 call_rcu(&stream->rcu_node, deferred_free_stream);
812 DBG("Closed tracefile %d from close stream", stream->fd);
813 }
814
815 /*
816 * relay_delete_session: Free all memory associated with a session and
817 * close all the FDs
818 */
819 static
820 void relay_delete_session(struct relay_command *cmd,
821 struct lttng_ht *sessions_ht)
822 {
823 struct lttng_ht_iter iter;
824 struct lttng_ht_node_ulong *node;
825 struct relay_stream *stream;
826 int ret;
827
828 if (!cmd->session) {
829 return;
830 }
831
832 DBG("Relay deleting session %" PRIu64, cmd->session->id);
833
834 rcu_read_lock();
835 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
836 node = lttng_ht_iter_get_node_ulong(&iter);
837 if (!node) {
838 continue;
839 }
840 stream = caa_container_of(node, struct relay_stream, stream_n);
841 if (stream->session == cmd->session) {
842 destroy_stream(stream, cmd->ctf_traces_ht);
843 }
844 /* Cleanup index of that stream. */
845 relay_index_destroy_by_stream_id(stream->stream_handle);
846 }
847
848 /* Make this session not visible anymore. */
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, 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)
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 destroy_stream(stream, 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)
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
1667 DBG("Relay receiving index");
1668
1669 if (!session || cmd->version_check_done == 0) {
1670 ERR("Trying to close a stream before version check");
1671 ret = -1;
1672 goto end_no_session;
1673 }
1674
1675 ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info,
1676 sizeof(index_info), 0);
1677 if (ret < sizeof(index_info)) {
1678 if (ret == 0) {
1679 /* Orderly shutdown. Not necessary to print an error. */
1680 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1681 } else {
1682 ERR("Relay didn't receive valid index struct size : %d", ret);
1683 }
1684 ret = -1;
1685 goto end_no_session;
1686 }
1687
1688 net_seq_num = be64toh(index_info.net_seq_num);
1689
1690 rcu_read_lock();
1691 stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id));
1692 if (!stream) {
1693 ret = -1;
1694 goto end_rcu_unlock;
1695 }
1696
1697 /* Live beacon handling */
1698 if (index_info.packet_size == 0) {
1699 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1700
1701 /*
1702 * Only flag a stream inactive when it has already received data.
1703 */
1704 if (stream->total_index_received > 0) {
1705 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1706 }
1707 ret = 0;
1708 goto end_rcu_unlock;
1709 } else {
1710 stream->beacon_ts_end = -1ULL;
1711 }
1712
1713 index = relay_index_find(stream->stream_handle, net_seq_num);
1714 if (!index) {
1715 /* A successful creation will add the object to the HT. */
1716 index = relay_index_create(stream->stream_handle, net_seq_num);
1717 if (!index) {
1718 goto end_rcu_unlock;
1719 }
1720 index_created = 1;
1721 }
1722
1723 copy_index_control_data(index, &index_info);
1724
1725 if (index_created) {
1726 /*
1727 * Try to add the relay index object to the hash table. If an object
1728 * already exist, destroy back the index created, set the data in this
1729 * object and write it on disk.
1730 */
1731 relay_index_add(index, &wr_index);
1732 if (wr_index) {
1733 copy_index_control_data(wr_index, &index_info);
1734 free(index);
1735 }
1736 } else {
1737 /* The index already exists so write it on disk. */
1738 wr_index = index;
1739 }
1740
1741 /* Do we have a writable ready index to write on disk. */
1742 if (wr_index) {
1743 /* Starting at 2.4, create the index file if none available. */
1744 if (cmd->minor >= 4 && stream->index_fd < 0) {
1745 ret = index_create_file(stream->path_name, stream->channel_name,
1746 relayd_uid, relayd_gid, stream->tracefile_size,
1747 stream->tracefile_count_current);
1748 if (ret < 0) {
1749 goto end_rcu_unlock;
1750 }
1751 stream->index_fd = ret;
1752 }
1753
1754 ret = relay_index_write(wr_index->fd, wr_index);
1755 if (ret < 0) {
1756 goto end_rcu_unlock;
1757 }
1758 stream->total_index_received++;
1759 }
1760
1761 end_rcu_unlock:
1762 rcu_read_unlock();
1763
1764 if (ret < 0) {
1765 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1766 } else {
1767 reply.ret_code = htobe32(LTTNG_OK);
1768 }
1769 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1770 if (send_ret < 0) {
1771 ERR("Relay sending close index id reply");
1772 ret = send_ret;
1773 }
1774
1775 end_no_session:
1776 return ret;
1777 }
1778
1779 /*
1780 * Process the commands received on the control socket
1781 */
1782 static
1783 int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1784 struct relay_command *cmd, struct relay_local_data *ctx)
1785 {
1786 int ret = 0;
1787
1788 switch (be32toh(recv_hdr->cmd)) {
1789 case RELAYD_CREATE_SESSION:
1790 ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht);
1791 break;
1792 case RELAYD_ADD_STREAM:
1793 ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht);
1794 break;
1795 case RELAYD_START_DATA:
1796 ret = relay_start(recv_hdr, cmd);
1797 break;
1798 case RELAYD_SEND_METADATA:
1799 ret = relay_recv_metadata(recv_hdr, cmd);
1800 break;
1801 case RELAYD_VERSION:
1802 ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht);
1803 break;
1804 case RELAYD_CLOSE_STREAM:
1805 ret = relay_close_stream(recv_hdr, cmd);
1806 break;
1807 case RELAYD_DATA_PENDING:
1808 ret = relay_data_pending(recv_hdr, cmd);
1809 break;
1810 case RELAYD_QUIESCENT_CONTROL:
1811 ret = relay_quiescent_control(recv_hdr, cmd);
1812 break;
1813 case RELAYD_BEGIN_DATA_PENDING:
1814 ret = relay_begin_data_pending(recv_hdr, cmd);
1815 break;
1816 case RELAYD_END_DATA_PENDING:
1817 ret = relay_end_data_pending(recv_hdr, cmd);
1818 break;
1819 case RELAYD_SEND_INDEX:
1820 ret = relay_recv_index(recv_hdr, cmd);
1821 break;
1822 case RELAYD_UPDATE_SYNC_INFO:
1823 default:
1824 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1825 relay_unknown_command(cmd);
1826 ret = -1;
1827 goto end;
1828 }
1829
1830 end:
1831 return ret;
1832 }
1833
1834 /*
1835 * relay_process_data: Process the data received on the data socket
1836 */
1837 static
1838 int relay_process_data(struct relay_command *cmd)
1839 {
1840 int ret = 0, rotate_index = 0, index_created = 0;
1841 struct relay_stream *stream;
1842 struct relay_index *index, *wr_index = NULL;
1843 struct lttcomm_relayd_data_hdr data_hdr;
1844 uint64_t stream_id, data_offset;
1845 uint64_t net_seq_num;
1846 uint32_t data_size;
1847
1848 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1849 sizeof(struct lttcomm_relayd_data_hdr), 0);
1850 if (ret <= 0) {
1851 if (ret == 0) {
1852 /* Orderly shutdown. Not necessary to print an error. */
1853 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1854 } else {
1855 ERR("Unable to receive data header on sock %d", cmd->sock->fd);
1856 }
1857 ret = -1;
1858 goto end;
1859 }
1860
1861 stream_id = be64toh(data_hdr.stream_id);
1862
1863 rcu_read_lock();
1864 stream = relay_stream_find_by_id(stream_id);
1865 if (!stream) {
1866 ret = -1;
1867 goto end_rcu_unlock;
1868 }
1869
1870 data_size = be32toh(data_hdr.data_size);
1871 if (data_buffer_size < data_size) {
1872 char *tmp_data_ptr;
1873
1874 tmp_data_ptr = realloc(data_buffer, data_size);
1875 if (!tmp_data_ptr) {
1876 ERR("Allocating data buffer");
1877 free(data_buffer);
1878 ret = -1;
1879 goto end_rcu_unlock;
1880 }
1881 data_buffer = tmp_data_ptr;
1882 data_buffer_size = data_size;
1883 }
1884 memset(data_buffer, 0, data_size);
1885
1886 net_seq_num = be64toh(data_hdr.net_seq_num);
1887
1888 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
1889 data_size, stream_id, net_seq_num);
1890 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
1891 if (ret <= 0) {
1892 if (ret == 0) {
1893 /* Orderly shutdown. Not necessary to print an error. */
1894 DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
1895 }
1896 ret = -1;
1897 goto end_rcu_unlock;
1898 }
1899
1900 /* Check if a rotation is needed. */
1901 if (stream->tracefile_size > 0 &&
1902 (stream->tracefile_size_current + data_size) >
1903 stream->tracefile_size) {
1904 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
1905 stream->tracefile_size, stream->tracefile_count,
1906 relayd_uid, relayd_gid, stream->fd,
1907 &(stream->tracefile_count_current), &stream->fd);
1908 if (ret < 0) {
1909 ERR("Rotating stream output file");
1910 goto end_rcu_unlock;
1911 }
1912 /* Reset current size because we just perform a stream rotation. */
1913 stream->tracefile_size_current = 0;
1914 rotate_index = 1;
1915 }
1916
1917 /* Get data offset because we are about to update the index. */
1918 data_offset = htobe64(stream->tracefile_size_current);
1919
1920 /*
1921 * Lookup for an existing index for that stream id/sequence number. If on
1922 * exists, the control thread already received the data for it thus we need
1923 * to write it on disk.
1924 */
1925 index = relay_index_find(stream_id, net_seq_num);
1926 if (!index) {
1927 /* A successful creation will add the object to the HT. */
1928 index = relay_index_create(stream->stream_handle, net_seq_num);
1929 if (!index) {
1930 goto end_rcu_unlock;
1931 }
1932 index_created = 1;
1933 }
1934
1935 if (rotate_index || stream->index_fd < 0) {
1936 index->to_close_fd = stream->index_fd;
1937 ret = index_create_file(stream->path_name, stream->channel_name,
1938 relayd_uid, relayd_gid, stream->tracefile_size,
1939 stream->tracefile_count_current);
1940 if (ret < 0) {
1941 /* This will close the stream's index fd if one. */
1942 relay_index_free_safe(index);
1943 goto end_rcu_unlock;
1944 }
1945 stream->index_fd = ret;
1946 }
1947 index->fd = stream->index_fd;
1948 index->index_data.offset = data_offset;
1949
1950 if (index_created) {
1951 /*
1952 * Try to add the relay index object to the hash table. If an object
1953 * already exist, destroy back the index created and set the data.
1954 */
1955 relay_index_add(index, &wr_index);
1956 if (wr_index) {
1957 /* Copy back data from the created index. */
1958 wr_index->fd = index->fd;
1959 wr_index->to_close_fd = index->to_close_fd;
1960 wr_index->index_data.offset = data_offset;
1961 free(index);
1962 }
1963 } else {
1964 /* The index already exists so write it on disk. */
1965 wr_index = index;
1966 }
1967
1968 /* Do we have a writable ready index to write on disk. */
1969 if (wr_index) {
1970 /* Starting at 2.4, create the index file if none available. */
1971 if (cmd->minor >= 4 && stream->index_fd < 0) {
1972 ret = index_create_file(stream->path_name, stream->channel_name,
1973 relayd_uid, relayd_gid, stream->tracefile_size,
1974 stream->tracefile_count_current);
1975 if (ret < 0) {
1976 goto end_rcu_unlock;
1977 }
1978 stream->index_fd = ret;
1979 }
1980
1981 ret = relay_index_write(wr_index->fd, wr_index);
1982 if (ret < 0) {
1983 goto end_rcu_unlock;
1984 }
1985 stream->total_index_received++;
1986 }
1987
1988 do {
1989 ret = write(stream->fd, data_buffer, data_size);
1990 } while (ret < 0 && errno == EINTR);
1991 if (ret < 0 || ret != data_size) {
1992 ERR("Relay error writing data to file");
1993 ret = -1;
1994 goto end_rcu_unlock;
1995 }
1996
1997 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
1998 ret, stream->stream_handle);
1999
2000 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2001 if (ret < 0) {
2002 goto end_rcu_unlock;
2003 }
2004 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
2005
2006 stream->prev_seq = net_seq_num;
2007
2008 /* Check if we need to close the FD */
2009 if (close_stream_check(stream)) {
2010 destroy_stream(stream, cmd->ctf_traces_ht);
2011 }
2012
2013 end_rcu_unlock:
2014 rcu_read_unlock();
2015 end:
2016 return ret;
2017 }
2018
2019 static
2020 void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
2021 {
2022 int ret;
2023
2024 lttng_poll_del(events, pollfd);
2025
2026 ret = close(pollfd);
2027 if (ret < 0) {
2028 ERR("Closing pollfd %d", pollfd);
2029 }
2030 }
2031
2032 static
2033 int relay_add_connection(int fd, struct lttng_poll_event *events,
2034 struct lttng_ht *relay_connections_ht)
2035 {
2036 struct relay_command *relay_connection;
2037 int ret;
2038
2039 relay_connection = zmalloc(sizeof(struct relay_command));
2040 if (relay_connection == NULL) {
2041 PERROR("Relay command zmalloc");
2042 goto error;
2043 }
2044 do {
2045 ret = read(fd, relay_connection, sizeof(struct relay_command));
2046 } while (ret < 0 && errno == EINTR);
2047 if (ret < 0 || ret < sizeof(struct relay_command)) {
2048 PERROR("read relay cmd pipe");
2049 goto error_read;
2050 }
2051
2052 relay_connection->ctf_traces_ht = lttng_ht_new(0, LTTNG_HT_TYPE_STRING);
2053 if (!relay_connection->ctf_traces_ht) {
2054 goto error_read;
2055 }
2056
2057 lttng_ht_node_init_ulong(&relay_connection->sock_n,
2058 (unsigned long) relay_connection->sock->fd);
2059 rcu_read_lock();
2060 lttng_ht_add_unique_ulong(relay_connections_ht,
2061 &relay_connection->sock_n);
2062 rcu_read_unlock();
2063 return lttng_poll_add(events,
2064 relay_connection->sock->fd,
2065 LPOLLIN | LPOLLRDHUP);
2066
2067 error_read:
2068 free(relay_connection);
2069 error:
2070 return -1;
2071 }
2072
2073 static
2074 void deferred_free_connection(struct rcu_head *head)
2075 {
2076 struct relay_command *relay_connection =
2077 caa_container_of(head, struct relay_command, rcu_node);
2078
2079 lttng_ht_destroy(relay_connection->ctf_traces_ht);
2080 lttcomm_destroy_sock(relay_connection->sock);
2081 free(relay_connection);
2082 }
2083
2084 static
2085 void relay_del_connection(struct lttng_ht *relay_connections_ht,
2086 struct lttng_ht_iter *iter, struct relay_command *relay_connection,
2087 struct lttng_ht *sessions_ht)
2088 {
2089 int ret;
2090
2091 ret = lttng_ht_del(relay_connections_ht, iter);
2092 assert(!ret);
2093 if (relay_connection->type == RELAY_CONTROL) {
2094 relay_delete_session(relay_connection, sessions_ht);
2095 }
2096
2097 call_rcu(&relay_connection->rcu_node,
2098 deferred_free_connection);
2099 }
2100
2101 /*
2102 * This thread does the actual work
2103 */
2104 static
2105 void *relay_thread_worker(void *data)
2106 {
2107 int ret, err = -1, last_seen_data_fd = -1;
2108 uint32_t nb_fd;
2109 struct relay_command *relay_connection;
2110 struct lttng_poll_event events;
2111 struct lttng_ht *relay_connections_ht;
2112 struct lttng_ht_node_ulong *node;
2113 struct lttng_ht_iter iter;
2114 struct lttcomm_relayd_hdr recv_hdr;
2115 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2116 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
2117
2118 DBG("[thread] Relay worker started");
2119
2120 rcu_register_thread();
2121
2122 /* table of connections indexed on socket */
2123 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2124 if (!relay_connections_ht) {
2125 goto relay_connections_ht_error;
2126 }
2127
2128 /* Tables of received indexes indexed by index handle and net_seq_num. */
2129 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2130 if (!indexes_ht) {
2131 goto indexes_ht_error;
2132 }
2133
2134 ret = create_thread_poll_set(&events, 2);
2135 if (ret < 0) {
2136 goto error_poll_create;
2137 }
2138
2139 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
2140 if (ret < 0) {
2141 goto error;
2142 }
2143
2144 restart:
2145 while (1) {
2146 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2147
2148 /* Infinite blocking call, waiting for transmission */
2149 DBG3("Relayd worker thread polling...");
2150 ret = lttng_poll_wait(&events, -1);
2151 if (ret < 0) {
2152 /*
2153 * Restart interrupted system call.
2154 */
2155 if (errno == EINTR) {
2156 goto restart;
2157 }
2158 goto error;
2159 }
2160
2161 nb_fd = ret;
2162
2163 /*
2164 * Process control. The control connection is prioritised so we don't
2165 * starve it with high throughout put tracing data on the data
2166 * connection.
2167 */
2168 for (i = 0; i < nb_fd; i++) {
2169 /* Fetch once the poll data */
2170 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2171 int pollfd = LTTNG_POLL_GETFD(&events, i);
2172
2173 /* Thread quit pipe has been closed. Killing thread. */
2174 ret = check_thread_quit_pipe(pollfd, revents);
2175 if (ret) {
2176 err = 0;
2177 goto exit;
2178 }
2179
2180 /* Inspect the relay cmd pipe for new connection */
2181 if (pollfd == relay_cmd_pipe[0]) {
2182 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2183 ERR("Relay pipe error");
2184 goto error;
2185 } else if (revents & LPOLLIN) {
2186 DBG("Relay command received");
2187 ret = relay_add_connection(relay_cmd_pipe[0],
2188 &events, relay_connections_ht);
2189 if (ret < 0) {
2190 goto error;
2191 }
2192 }
2193 } else if (revents) {
2194 rcu_read_lock();
2195 lttng_ht_lookup(relay_connections_ht,
2196 (void *)((unsigned long) pollfd),
2197 &iter);
2198 node = lttng_ht_iter_get_node_ulong(&iter);
2199 if (node == NULL) {
2200 DBG2("Relay sock %d not found", pollfd);
2201 rcu_read_unlock();
2202 goto error;
2203 }
2204 relay_connection = caa_container_of(node,
2205 struct relay_command, sock_n);
2206
2207 if (revents & (LPOLLERR)) {
2208 ERR("POLL ERROR");
2209 relay_cleanup_poll_connection(&events, pollfd);
2210 relay_del_connection(relay_connections_ht,
2211 &iter, relay_connection, sessions_ht);
2212 if (last_seen_data_fd == pollfd) {
2213 last_seen_data_fd = last_notdel_data_fd;
2214 }
2215 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
2216 DBG("Socket %d hung up", pollfd);
2217 relay_cleanup_poll_connection(&events, pollfd);
2218 relay_del_connection(relay_connections_ht,
2219 &iter, relay_connection, sessions_ht);
2220 if (last_seen_data_fd == pollfd) {
2221 last_seen_data_fd = last_notdel_data_fd;
2222 }
2223 } else if (revents & LPOLLIN) {
2224 /* control socket */
2225 if (relay_connection->type == RELAY_CONTROL) {
2226 ret = relay_connection->sock->ops->recvmsg(
2227 relay_connection->sock, &recv_hdr,
2228 sizeof(struct lttcomm_relayd_hdr), 0);
2229 /* connection closed */
2230 if (ret <= 0) {
2231 relay_cleanup_poll_connection(&events, pollfd);
2232 relay_del_connection(relay_connections_ht,
2233 &iter, relay_connection, sessions_ht);
2234 DBG("Control connection closed with %d", pollfd);
2235 } else {
2236 if (relay_connection->session) {
2237 DBG2("Relay worker receiving data for session : %" PRIu64,
2238 relay_connection->session->id);
2239 }
2240 ret = relay_process_control(&recv_hdr,
2241 relay_connection, relay_ctx);
2242 if (ret < 0) {
2243 /* Clear the session on error. */
2244 relay_cleanup_poll_connection(&events, pollfd);
2245 relay_del_connection(relay_connections_ht,
2246 &iter, relay_connection, sessions_ht);
2247 DBG("Connection closed with %d", pollfd);
2248 }
2249 seen_control = 1;
2250 }
2251 } else {
2252 /*
2253 * Flag the last seen data fd not deleted. It will be
2254 * used as the last seen fd if any fd gets deleted in
2255 * this first loop.
2256 */
2257 last_notdel_data_fd = pollfd;
2258 }
2259 }
2260 rcu_read_unlock();
2261 }
2262 }
2263
2264 /*
2265 * The last loop handled a control request, go back to poll to make
2266 * sure we prioritise the control socket.
2267 */
2268 if (seen_control) {
2269 continue;
2270 }
2271
2272 if (last_seen_data_fd >= 0) {
2273 for (i = 0; i < nb_fd; i++) {
2274 int pollfd = LTTNG_POLL_GETFD(&events, i);
2275 if (last_seen_data_fd == pollfd) {
2276 idx = i;
2277 break;
2278 }
2279 }
2280 }
2281
2282 /* Process data connection. */
2283 for (i = idx + 1; i < nb_fd; i++) {
2284 /* Fetch the poll data. */
2285 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2286 int pollfd = LTTNG_POLL_GETFD(&events, i);
2287
2288 /* Skip the command pipe. It's handled in the first loop. */
2289 if (pollfd == relay_cmd_pipe[0]) {
2290 continue;
2291 }
2292
2293 if (revents) {
2294 rcu_read_lock();
2295 lttng_ht_lookup(relay_connections_ht,
2296 (void *)((unsigned long) pollfd),
2297 &iter);
2298 node = lttng_ht_iter_get_node_ulong(&iter);
2299 if (node == NULL) {
2300 /* Skip it. Might be removed before. */
2301 rcu_read_unlock();
2302 continue;
2303 }
2304 relay_connection = caa_container_of(node,
2305 struct relay_command, sock_n);
2306
2307 if (revents & LPOLLIN) {
2308 if (relay_connection->type != RELAY_DATA) {
2309 continue;
2310 }
2311
2312 ret = relay_process_data(relay_connection);
2313 /* connection closed */
2314 if (ret < 0) {
2315 relay_cleanup_poll_connection(&events, pollfd);
2316 relay_del_connection(relay_connections_ht,
2317 &iter, relay_connection, sessions_ht);
2318 DBG("Data connection closed with %d", pollfd);
2319 /*
2320 * Every goto restart call sets the last seen fd where
2321 * here we don't really care since we gracefully
2322 * continue the loop after the connection is deleted.
2323 */
2324 } else {
2325 /* Keep last seen port. */
2326 last_seen_data_fd = pollfd;
2327 rcu_read_unlock();
2328 goto restart;
2329 }
2330 }
2331 rcu_read_unlock();
2332 }
2333 }
2334 last_seen_data_fd = -1;
2335 }
2336
2337 exit:
2338 error:
2339 lttng_poll_clean(&events);
2340
2341 /* empty the hash table and free the memory */
2342 rcu_read_lock();
2343 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
2344 node = lttng_ht_iter_get_node_ulong(&iter);
2345 if (node) {
2346 relay_connection = caa_container_of(node,
2347 struct relay_command, sock_n);
2348 relay_del_connection(relay_connections_ht,
2349 &iter, relay_connection, sessions_ht);
2350 }
2351 }
2352 error_poll_create:
2353 {
2354 struct relay_index *index;
2355 cds_lfht_for_each_entry(indexes_ht->ht, &iter.iter, index, index_n.node) {
2356 relay_index_delete(index);
2357 }
2358 lttng_ht_destroy(indexes_ht);
2359 }
2360 rcu_read_unlock();
2361 indexes_ht_error:
2362 lttng_ht_destroy(relay_connections_ht);
2363 relay_connections_ht_error:
2364 /* Close relay cmd pipes */
2365 utils_close_pipe(relay_cmd_pipe);
2366 if (err) {
2367 DBG("Thread exited with error");
2368 }
2369 DBG("Worker thread cleanup complete");
2370 free(data_buffer);
2371 stop_threads();
2372 rcu_unregister_thread();
2373 return NULL;
2374 }
2375
2376 /*
2377 * Create the relay command pipe to wake thread_manage_apps.
2378 * Closed in cleanup().
2379 */
2380 static int create_relay_cmd_pipe(void)
2381 {
2382 int ret;
2383
2384 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
2385
2386 return ret;
2387 }
2388
2389 /*
2390 * main
2391 */
2392 int main(int argc, char **argv)
2393 {
2394 int ret = 0;
2395 void *status;
2396 struct relay_local_data *relay_ctx;
2397
2398 /* Create thread quit pipe */
2399 if ((ret = init_thread_quit_pipe()) < 0) {
2400 goto error;
2401 }
2402
2403 /* Parse arguments */
2404 progname = argv[0];
2405 if ((ret = parse_args(argc, argv)) < 0) {
2406 goto exit;
2407 }
2408
2409 if ((ret = set_signal_handler()) < 0) {
2410 goto exit;
2411 }
2412
2413 /* Try to create directory if -o, --output is specified. */
2414 if (opt_output_path) {
2415 if (*opt_output_path != '/') {
2416 ERR("Please specify an absolute path for -o, --output PATH");
2417 goto exit;
2418 }
2419
2420 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2421 if (ret < 0) {
2422 ERR("Unable to create %s", opt_output_path);
2423 goto exit;
2424 }
2425 }
2426
2427 /* Daemonize */
2428 if (opt_daemon) {
2429 ret = daemon(0, 0);
2430 if (ret < 0) {
2431 PERROR("daemon");
2432 goto exit;
2433 }
2434 }
2435
2436 /* We need those values for the file/dir creation. */
2437 relayd_uid = getuid();
2438 relayd_gid = getgid();
2439
2440 /* Check if daemon is UID = 0 */
2441 if (relayd_uid == 0) {
2442 if (control_uri->port < 1024 || data_uri->port < 1024) {
2443 ERR("Need to be root to use ports < 1024");
2444 ret = -1;
2445 goto exit;
2446 }
2447 }
2448
2449 /* Setup the thread apps communication pipe. */
2450 if ((ret = create_relay_cmd_pipe()) < 0) {
2451 goto exit;
2452 }
2453
2454 /* Init relay command queue. */
2455 cds_wfq_init(&relay_cmd_queue.queue);
2456
2457 /* Set up max poll set size */
2458 lttng_poll_set_max_size();
2459
2460 /* Initialize communication library */
2461 lttcomm_init();
2462
2463 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2464 if (!relay_ctx) {
2465 PERROR("relay_ctx");
2466 goto exit;
2467 }
2468
2469 /* tables of sessions indexed by session ID */
2470 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2471 if (!relay_ctx->sessions_ht) {
2472 goto exit_relay_ctx_sessions;
2473 }
2474
2475 /* tables of streams indexed by stream ID */
2476 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2477 if (!relay_streams_ht) {
2478 goto exit_relay_ctx_streams;
2479 }
2480
2481 /* tables of streams indexed by stream ID */
2482 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2483 if (!viewer_streams_ht) {
2484 goto exit_relay_ctx_viewer_streams;
2485 }
2486
2487 /* Setup the dispatcher thread */
2488 ret = pthread_create(&dispatcher_thread, NULL,
2489 relay_thread_dispatcher, (void *) NULL);
2490 if (ret != 0) {
2491 PERROR("pthread_create dispatcher");
2492 goto exit_dispatcher;
2493 }
2494
2495 /* Setup the worker thread */
2496 ret = pthread_create(&worker_thread, NULL,
2497 relay_thread_worker, (void *) relay_ctx);
2498 if (ret != 0) {
2499 PERROR("pthread_create worker");
2500 goto exit_worker;
2501 }
2502
2503 /* Setup the listener thread */
2504 ret = pthread_create(&listener_thread, NULL,
2505 relay_thread_listener, (void *) NULL);
2506 if (ret != 0) {
2507 PERROR("pthread_create listener");
2508 goto exit_listener;
2509 }
2510
2511 ret = live_start_threads(live_uri, relay_ctx);
2512 if (ret != 0) {
2513 ERR("Starting live viewer threads");
2514 }
2515
2516 exit_listener:
2517 ret = pthread_join(listener_thread, &status);
2518 if (ret != 0) {
2519 PERROR("pthread_join");
2520 goto error; /* join error, exit without cleanup */
2521 }
2522
2523 exit_worker:
2524 ret = pthread_join(worker_thread, &status);
2525 if (ret != 0) {
2526 PERROR("pthread_join");
2527 goto error; /* join error, exit without cleanup */
2528 }
2529
2530 exit_dispatcher:
2531 ret = pthread_join(dispatcher_thread, &status);
2532 if (ret != 0) {
2533 PERROR("pthread_join");
2534 goto error; /* join error, exit without cleanup */
2535 }
2536 lttng_ht_destroy(viewer_streams_ht);
2537
2538 exit_relay_ctx_viewer_streams:
2539 lttng_ht_destroy(relay_streams_ht);
2540
2541 exit_relay_ctx_streams:
2542 lttng_ht_destroy(relay_ctx->sessions_ht);
2543
2544 exit_relay_ctx_sessions:
2545 free(relay_ctx);
2546
2547 exit:
2548 live_stop_threads();
2549 cleanup();
2550 if (!ret) {
2551 exit(EXIT_SUCCESS);
2552 }
2553
2554 error:
2555 exit(EXIT_FAILURE);
2556 }
LTTng tools - Deliverables
This page took 0.126944 seconds and 6 git commands to generate.