Fix: take index seq number into account for rotation pending check
[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 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2 only,
9 * as published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * more details.
15 *
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19 */
20
21 #define _LGPL_SOURCE
22 #include <getopt.h>
23 #include <grp.h>
24 #include <limits.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <sys/mman.h>
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
34 #include <sys/stat.h>
35 #include <sys/types.h>
36 #include <sys/wait.h>
37 #include <inttypes.h>
38 #include <urcu/futex.h>
39 #include <urcu/uatomic.h>
40 #include <unistd.h>
41 #include <fcntl.h>
42
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/compat/endian.h>
48 #include <common/compat/getenv.h>
49 #include <common/defaults.h>
50 #include <common/daemonize.h>
51 #include <common/futex.h>
52 #include <common/sessiond-comm/sessiond-comm.h>
53 #include <common/sessiond-comm/inet.h>
54 #include <common/sessiond-comm/relayd.h>
55 #include <common/uri.h>
56 #include <common/utils.h>
57 #include <common/align.h>
58 #include <common/config/session-config.h>
59 #include <common/dynamic-buffer.h>
60 #include <common/buffer-view.h>
61 #include <urcu/rculist.h>
62
63 #include "cmd.h"
64 #include "ctf-trace.h"
65 #include "index.h"
66 #include "utils.h"
67 #include "lttng-relayd.h"
68 #include "live.h"
69 #include "health-relayd.h"
70 #include "testpoint.h"
71 #include "viewer-stream.h"
72 #include "session.h"
73 #include "stream.h"
74 #include "connection.h"
75 #include "tracefile-array.h"
76 #include "tcp_keep_alive.h"
77
78 static const char *help_msg =
79 #ifdef LTTNG_EMBED_HELP
80 #include <lttng-relayd.8.h>
81 #else
82 NULL
83 #endif
84 ;
85
86 enum relay_connection_status {
87 RELAY_CONNECTION_STATUS_OK,
88 /* An error occured while processing an event on the connection. */
89 RELAY_CONNECTION_STATUS_ERROR,
90 /* Connection closed/shutdown cleanly. */
91 RELAY_CONNECTION_STATUS_CLOSED,
92 };
93
94 /* command line options */
95 char *opt_output_path;
96 static int opt_daemon, opt_background;
97
98 /*
99 * We need to wait for listener and live listener threads, as well as
100 * health check thread, before being ready to signal readiness.
101 */
102 #define NR_LTTNG_RELAY_READY 3
103 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
104
105 /* Size of receive buffer. */
106 #define RECV_DATA_BUFFER_SIZE 65536
107 #define FILE_COPY_BUFFER_SIZE 65536
108
109 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
110 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
111
112 static struct lttng_uri *control_uri;
113 static struct lttng_uri *data_uri;
114 static struct lttng_uri *live_uri;
115
116 const char *progname;
117
118 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
119 static int tracing_group_name_override;
120
121 const char * const config_section_name = "relayd";
122
123 /*
124 * Quit pipe for all threads. This permits a single cancellation point
125 * for all threads when receiving an event on the pipe.
126 */
127 int thread_quit_pipe[2] = { -1, -1 };
128
129 /*
130 * This pipe is used to inform the worker thread that a command is queued and
131 * ready to be processed.
132 */
133 static int relay_conn_pipe[2] = { -1, -1 };
134
135 /* Shared between threads */
136 static int dispatch_thread_exit;
137
138 static pthread_t listener_thread;
139 static pthread_t dispatcher_thread;
140 static pthread_t worker_thread;
141 static pthread_t health_thread;
142
143 /*
144 * last_relay_stream_id_lock protects last_relay_stream_id increment
145 * atomicity on 32-bit architectures.
146 */
147 static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
148 static uint64_t last_relay_stream_id;
149
150 /*
151 * Relay command queue.
152 *
153 * The relay_thread_listener and relay_thread_dispatcher communicate with this
154 * queue.
155 */
156 static struct relay_conn_queue relay_conn_queue;
157
158 /* Global relay stream hash table. */
159 struct lttng_ht *relay_streams_ht;
160
161 /* Global relay viewer stream hash table. */
162 struct lttng_ht *viewer_streams_ht;
163
164 /* Global relay sessions hash table. */
165 struct lttng_ht *sessions_ht;
166
167 /* Relayd health monitoring */
168 struct health_app *health_relayd;
169
170 static struct option long_options[] = {
171 { "control-port", 1, 0, 'C', },
172 { "data-port", 1, 0, 'D', },
173 { "live-port", 1, 0, 'L', },
174 { "daemonize", 0, 0, 'd', },
175 { "background", 0, 0, 'b', },
176 { "group", 1, 0, 'g', },
177 { "help", 0, 0, 'h', },
178 { "output", 1, 0, 'o', },
179 { "verbose", 0, 0, 'v', },
180 { "config", 1, 0, 'f' },
181 { "version", 0, 0, 'V' },
182 { NULL, 0, 0, 0, },
183 };
184
185 static const char *config_ignore_options[] = { "help", "config", "version" };
186
187 /*
188 * Take an option from the getopt output and set it in the right variable to be
189 * used later.
190 *
191 * Return 0 on success else a negative value.
192 */
193 static int set_option(int opt, const char *arg, const char *optname)
194 {
195 int ret;
196
197 switch (opt) {
198 case 0:
199 fprintf(stderr, "option %s", optname);
200 if (arg) {
201 fprintf(stderr, " with arg %s\n", arg);
202 }
203 break;
204 case 'C':
205 if (lttng_is_setuid_setgid()) {
206 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
207 "-C, --control-port");
208 } else {
209 ret = uri_parse(arg, &control_uri);
210 if (ret < 0) {
211 ERR("Invalid control URI specified");
212 goto end;
213 }
214 if (control_uri->port == 0) {
215 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
216 }
217 }
218 break;
219 case 'D':
220 if (lttng_is_setuid_setgid()) {
221 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
222 "-D, -data-port");
223 } else {
224 ret = uri_parse(arg, &data_uri);
225 if (ret < 0) {
226 ERR("Invalid data URI specified");
227 goto end;
228 }
229 if (data_uri->port == 0) {
230 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
231 }
232 }
233 break;
234 case 'L':
235 if (lttng_is_setuid_setgid()) {
236 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
237 "-L, -live-port");
238 } else {
239 ret = uri_parse(arg, &live_uri);
240 if (ret < 0) {
241 ERR("Invalid live URI specified");
242 goto end;
243 }
244 if (live_uri->port == 0) {
245 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
246 }
247 }
248 break;
249 case 'd':
250 opt_daemon = 1;
251 break;
252 case 'b':
253 opt_background = 1;
254 break;
255 case 'g':
256 if (lttng_is_setuid_setgid()) {
257 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
258 "-g, --group");
259 } else {
260 tracing_group_name = strdup(arg);
261 if (tracing_group_name == NULL) {
262 ret = -errno;
263 PERROR("strdup");
264 goto end;
265 }
266 tracing_group_name_override = 1;
267 }
268 break;
269 case 'h':
270 ret = utils_show_help(8, "lttng-relayd", help_msg);
271 if (ret) {
272 ERR("Cannot show --help for `lttng-relayd`");
273 perror("exec");
274 }
275 exit(EXIT_FAILURE);
276 case 'V':
277 fprintf(stdout, "%s\n", VERSION);
278 exit(EXIT_SUCCESS);
279 case 'o':
280 if (lttng_is_setuid_setgid()) {
281 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
282 "-o, --output");
283 } else {
284 ret = asprintf(&opt_output_path, "%s", arg);
285 if (ret < 0) {
286 ret = -errno;
287 PERROR("asprintf opt_output_path");
288 goto end;
289 }
290 }
291 break;
292 case 'v':
293 /* Verbose level can increase using multiple -v */
294 if (arg) {
295 lttng_opt_verbose = config_parse_value(arg);
296 } else {
297 /* Only 3 level of verbosity (-vvv). */
298 if (lttng_opt_verbose < 3) {
299 lttng_opt_verbose += 1;
300 }
301 }
302 break;
303 default:
304 /* Unknown option or other error.
305 * Error is printed by getopt, just return */
306 ret = -1;
307 goto end;
308 }
309
310 /* All good. */
311 ret = 0;
312
313 end:
314 return ret;
315 }
316
317 /*
318 * config_entry_handler_cb used to handle options read from a config file.
319 * See config_entry_handler_cb comment in common/config/session-config.h for the
320 * return value conventions.
321 */
322 static int config_entry_handler(const struct config_entry *entry, void *unused)
323 {
324 int ret = 0, i;
325
326 if (!entry || !entry->name || !entry->value) {
327 ret = -EINVAL;
328 goto end;
329 }
330
331 /* Check if the option is to be ignored */
332 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
333 if (!strcmp(entry->name, config_ignore_options[i])) {
334 goto end;
335 }
336 }
337
338 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
339 /* Ignore if entry name is not fully matched. */
340 if (strcmp(entry->name, long_options[i].name)) {
341 continue;
342 }
343
344 /*
345 * If the option takes no argument on the command line,
346 * we have to check if the value is "true". We support
347 * non-zero numeric values, true, on and yes.
348 */
349 if (!long_options[i].has_arg) {
350 ret = config_parse_value(entry->value);
351 if (ret <= 0) {
352 if (ret) {
353 WARN("Invalid configuration value \"%s\" for option %s",
354 entry->value, entry->name);
355 }
356 /* False, skip boolean config option. */
357 goto end;
358 }
359 }
360
361 ret = set_option(long_options[i].val, entry->value, entry->name);
362 goto end;
363 }
364
365 WARN("Unrecognized option \"%s\" in daemon configuration file.",
366 entry->name);
367
368 end:
369 return ret;
370 }
371
372 static int set_options(int argc, char **argv)
373 {
374 int c, ret = 0, option_index = 0, retval = 0;
375 int orig_optopt = optopt, orig_optind = optind;
376 char *default_address, *optstring;
377 const char *config_path = NULL;
378
379 optstring = utils_generate_optstring(long_options,
380 sizeof(long_options) / sizeof(struct option));
381 if (!optstring) {
382 retval = -ENOMEM;
383 goto exit;
384 }
385
386 /* Check for the --config option */
387
388 while ((c = getopt_long(argc, argv, optstring, long_options,
389 &option_index)) != -1) {
390 if (c == '?') {
391 retval = -EINVAL;
392 goto exit;
393 } else if (c != 'f') {
394 continue;
395 }
396
397 if (lttng_is_setuid_setgid()) {
398 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
399 "-f, --config");
400 } else {
401 config_path = utils_expand_path(optarg);
402 if (!config_path) {
403 ERR("Failed to resolve path: %s", optarg);
404 }
405 }
406 }
407
408 ret = config_get_section_entries(config_path, config_section_name,
409 config_entry_handler, NULL);
410 if (ret) {
411 if (ret > 0) {
412 ERR("Invalid configuration option at line %i", ret);
413 }
414 retval = -1;
415 goto exit;
416 }
417
418 /* Reset getopt's global state */
419 optopt = orig_optopt;
420 optind = orig_optind;
421 while (1) {
422 c = getopt_long(argc, argv, optstring, long_options, &option_index);
423 if (c == -1) {
424 break;
425 }
426
427 ret = set_option(c, optarg, long_options[option_index].name);
428 if (ret < 0) {
429 retval = -1;
430 goto exit;
431 }
432 }
433
434 /* assign default values */
435 if (control_uri == NULL) {
436 ret = asprintf(&default_address,
437 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
438 DEFAULT_NETWORK_CONTROL_PORT);
439 if (ret < 0) {
440 PERROR("asprintf default data address");
441 retval = -1;
442 goto exit;
443 }
444
445 ret = uri_parse(default_address, &control_uri);
446 free(default_address);
447 if (ret < 0) {
448 ERR("Invalid control URI specified");
449 retval = -1;
450 goto exit;
451 }
452 }
453 if (data_uri == NULL) {
454 ret = asprintf(&default_address,
455 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
456 DEFAULT_NETWORK_DATA_PORT);
457 if (ret < 0) {
458 PERROR("asprintf default data address");
459 retval = -1;
460 goto exit;
461 }
462
463 ret = uri_parse(default_address, &data_uri);
464 free(default_address);
465 if (ret < 0) {
466 ERR("Invalid data URI specified");
467 retval = -1;
468 goto exit;
469 }
470 }
471 if (live_uri == NULL) {
472 ret = asprintf(&default_address,
473 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
474 DEFAULT_NETWORK_VIEWER_PORT);
475 if (ret < 0) {
476 PERROR("asprintf default viewer control address");
477 retval = -1;
478 goto exit;
479 }
480
481 ret = uri_parse(default_address, &live_uri);
482 free(default_address);
483 if (ret < 0) {
484 ERR("Invalid viewer control URI specified");
485 retval = -1;
486 goto exit;
487 }
488 }
489
490 exit:
491 free(optstring);
492 return retval;
493 }
494
495 static void print_global_objects(void)
496 {
497 rcu_register_thread();
498
499 print_viewer_streams();
500 print_relay_streams();
501 print_sessions();
502
503 rcu_unregister_thread();
504 }
505
506 /*
507 * Cleanup the daemon
508 */
509 static void relayd_cleanup(void)
510 {
511 print_global_objects();
512
513 DBG("Cleaning up");
514
515 if (viewer_streams_ht)
516 lttng_ht_destroy(viewer_streams_ht);
517 if (relay_streams_ht)
518 lttng_ht_destroy(relay_streams_ht);
519 if (sessions_ht)
520 lttng_ht_destroy(sessions_ht);
521
522 /* free the dynamically allocated opt_output_path */
523 free(opt_output_path);
524
525 /* Close thread quit pipes */
526 utils_close_pipe(thread_quit_pipe);
527
528 uri_free(control_uri);
529 uri_free(data_uri);
530 /* Live URI is freed in the live thread. */
531
532 if (tracing_group_name_override) {
533 free((void *) tracing_group_name);
534 }
535 }
536
537 /*
538 * Write to writable pipe used to notify a thread.
539 */
540 static int notify_thread_pipe(int wpipe)
541 {
542 ssize_t ret;
543
544 ret = lttng_write(wpipe, "!", 1);
545 if (ret < 1) {
546 PERROR("write poll pipe");
547 goto end;
548 }
549 ret = 0;
550 end:
551 return ret;
552 }
553
554 static int notify_health_quit_pipe(int *pipe)
555 {
556 ssize_t ret;
557
558 ret = lttng_write(pipe[1], "4", 1);
559 if (ret < 1) {
560 PERROR("write relay health quit");
561 goto end;
562 }
563 ret = 0;
564 end:
565 return ret;
566 }
567
568 /*
569 * Stop all relayd and relayd-live threads.
570 */
571 int lttng_relay_stop_threads(void)
572 {
573 int retval = 0;
574
575 /* Stopping all threads */
576 DBG("Terminating all threads");
577 if (notify_thread_pipe(thread_quit_pipe[1])) {
578 ERR("write error on thread quit pipe");
579 retval = -1;
580 }
581
582 if (notify_health_quit_pipe(health_quit_pipe)) {
583 ERR("write error on health quit pipe");
584 }
585
586 /* Dispatch thread */
587 CMM_STORE_SHARED(dispatch_thread_exit, 1);
588 futex_nto1_wake(&relay_conn_queue.futex);
589
590 if (relayd_live_stop()) {
591 ERR("Error stopping live threads");
592 retval = -1;
593 }
594 return retval;
595 }
596
597 /*
598 * Signal handler for the daemon
599 *
600 * Simply stop all worker threads, leaving main() return gracefully after
601 * joining all threads and calling cleanup().
602 */
603 static void sighandler(int sig)
604 {
605 switch (sig) {
606 case SIGINT:
607 DBG("SIGINT caught");
608 if (lttng_relay_stop_threads()) {
609 ERR("Error stopping threads");
610 }
611 break;
612 case SIGTERM:
613 DBG("SIGTERM caught");
614 if (lttng_relay_stop_threads()) {
615 ERR("Error stopping threads");
616 }
617 break;
618 case SIGUSR1:
619 CMM_STORE_SHARED(recv_child_signal, 1);
620 break;
621 default:
622 break;
623 }
624 }
625
626 /*
627 * Setup signal handler for :
628 * SIGINT, SIGTERM, SIGPIPE
629 */
630 static int set_signal_handler(void)
631 {
632 int ret = 0;
633 struct sigaction sa;
634 sigset_t sigset;
635
636 if ((ret = sigemptyset(&sigset)) < 0) {
637 PERROR("sigemptyset");
638 return ret;
639 }
640
641 sa.sa_mask = sigset;
642 sa.sa_flags = 0;
643
644 sa.sa_handler = sighandler;
645 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
646 PERROR("sigaction");
647 return ret;
648 }
649
650 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
651 PERROR("sigaction");
652 return ret;
653 }
654
655 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
656 PERROR("sigaction");
657 return ret;
658 }
659
660 sa.sa_handler = SIG_IGN;
661 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
662 PERROR("sigaction");
663 return ret;
664 }
665
666 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
667
668 return ret;
669 }
670
671 void lttng_relay_notify_ready(void)
672 {
673 /* Notify the parent of the fork() process that we are ready. */
674 if (opt_daemon || opt_background) {
675 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
676 kill(child_ppid, SIGUSR1);
677 }
678 }
679 }
680
681 /*
682 * Init thread quit pipe.
683 *
684 * Return -1 on error or 0 if all pipes are created.
685 */
686 static int init_thread_quit_pipe(void)
687 {
688 int ret;
689
690 ret = utils_create_pipe_cloexec(thread_quit_pipe);
691
692 return ret;
693 }
694
695 /*
696 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
697 */
698 static int create_thread_poll_set(struct lttng_poll_event *events, int size)
699 {
700 int ret;
701
702 if (events == NULL || size == 0) {
703 ret = -1;
704 goto error;
705 }
706
707 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
708 if (ret < 0) {
709 goto error;
710 }
711
712 /* Add quit pipe */
713 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
714 if (ret < 0) {
715 goto error;
716 }
717
718 return 0;
719
720 error:
721 return ret;
722 }
723
724 /*
725 * Check if the thread quit pipe was triggered.
726 *
727 * Return 1 if it was triggered else 0;
728 */
729 static int check_thread_quit_pipe(int fd, uint32_t events)
730 {
731 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
732 return 1;
733 }
734
735 return 0;
736 }
737
738 /*
739 * Create and init socket from uri.
740 */
741 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri)
742 {
743 int ret;
744 struct lttcomm_sock *sock = NULL;
745
746 sock = lttcomm_alloc_sock_from_uri(uri);
747 if (sock == NULL) {
748 ERR("Allocating socket");
749 goto error;
750 }
751
752 ret = lttcomm_create_sock(sock);
753 if (ret < 0) {
754 goto error;
755 }
756 DBG("Listening on sock %d", sock->fd);
757
758 ret = sock->ops->bind(sock);
759 if (ret < 0) {
760 PERROR("Failed to bind socket");
761 goto error;
762 }
763
764 ret = sock->ops->listen(sock, -1);
765 if (ret < 0) {
766 goto error;
767
768 }
769
770 return sock;
771
772 error:
773 if (sock) {
774 lttcomm_destroy_sock(sock);
775 }
776 return NULL;
777 }
778
779 /*
780 * This thread manages the listening for new connections on the network
781 */
782 static void *relay_thread_listener(void *data)
783 {
784 int i, ret, pollfd, err = -1;
785 uint32_t revents, nb_fd;
786 struct lttng_poll_event events;
787 struct lttcomm_sock *control_sock, *data_sock;
788
789 DBG("[thread] Relay listener started");
790
791 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
792
793 health_code_update();
794
795 control_sock = relay_socket_create(control_uri);
796 if (!control_sock) {
797 goto error_sock_control;
798 }
799
800 data_sock = relay_socket_create(data_uri);
801 if (!data_sock) {
802 goto error_sock_relay;
803 }
804
805 /*
806 * Pass 3 as size here for the thread quit pipe, control and
807 * data socket.
808 */
809 ret = create_thread_poll_set(&events, 3);
810 if (ret < 0) {
811 goto error_create_poll;
812 }
813
814 /* Add the control socket */
815 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
816 if (ret < 0) {
817 goto error_poll_add;
818 }
819
820 /* Add the data socket */
821 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
822 if (ret < 0) {
823 goto error_poll_add;
824 }
825
826 lttng_relay_notify_ready();
827
828 if (testpoint(relayd_thread_listener)) {
829 goto error_testpoint;
830 }
831
832 while (1) {
833 health_code_update();
834
835 DBG("Listener accepting connections");
836
837 restart:
838 health_poll_entry();
839 ret = lttng_poll_wait(&events, -1);
840 health_poll_exit();
841 if (ret < 0) {
842 /*
843 * Restart interrupted system call.
844 */
845 if (errno == EINTR) {
846 goto restart;
847 }
848 goto error;
849 }
850
851 nb_fd = ret;
852
853 DBG("Relay new connection received");
854 for (i = 0; i < nb_fd; i++) {
855 health_code_update();
856
857 /* Fetch once the poll data */
858 revents = LTTNG_POLL_GETEV(&events, i);
859 pollfd = LTTNG_POLL_GETFD(&events, i);
860
861 if (!revents) {
862 /*
863 * No activity for this FD (poll
864 * implementation).
865 */
866 continue;
867 }
868
869 /* Thread quit pipe has been closed. Killing thread. */
870 ret = check_thread_quit_pipe(pollfd, revents);
871 if (ret) {
872 err = 0;
873 goto exit;
874 }
875
876 if (revents & LPOLLIN) {
877 /*
878 * A new connection is requested, therefore a
879 * sessiond/consumerd connection is allocated in
880 * this thread, enqueued to a global queue and
881 * dequeued (and freed) in the worker thread.
882 */
883 int val = 1;
884 struct relay_connection *new_conn;
885 struct lttcomm_sock *newsock;
886 enum connection_type type;
887
888 if (pollfd == data_sock->fd) {
889 type = RELAY_DATA;
890 newsock = data_sock->ops->accept(data_sock);
891 DBG("Relay data connection accepted, socket %d",
892 newsock->fd);
893 } else {
894 assert(pollfd == control_sock->fd);
895 type = RELAY_CONTROL;
896 newsock = control_sock->ops->accept(control_sock);
897 DBG("Relay control connection accepted, socket %d",
898 newsock->fd);
899 }
900 if (!newsock) {
901 PERROR("accepting sock");
902 goto error;
903 }
904
905 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
906 sizeof(val));
907 if (ret < 0) {
908 PERROR("setsockopt inet");
909 lttcomm_destroy_sock(newsock);
910 goto error;
911 }
912
913 ret = socket_apply_keep_alive_config(newsock->fd);
914 if (ret < 0) {
915 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
916 newsock->fd);
917 lttcomm_destroy_sock(newsock);
918 goto error;
919 }
920
921 new_conn = connection_create(newsock, type);
922 if (!new_conn) {
923 lttcomm_destroy_sock(newsock);
924 goto error;
925 }
926
927 /* Enqueue request for the dispatcher thread. */
928 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
929 &new_conn->qnode);
930
931 /*
932 * Wake the dispatch queue futex.
933 * Implicit memory barrier with the
934 * exchange in cds_wfcq_enqueue.
935 */
936 futex_nto1_wake(&relay_conn_queue.futex);
937 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
938 ERR("socket poll error");
939 goto error;
940 } else {
941 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
942 goto error;
943 }
944 }
945 }
946
947 exit:
948 error:
949 error_poll_add:
950 error_testpoint:
951 lttng_poll_clean(&events);
952 error_create_poll:
953 if (data_sock->fd >= 0) {
954 ret = data_sock->ops->close(data_sock);
955 if (ret) {
956 PERROR("close");
957 }
958 }
959 lttcomm_destroy_sock(data_sock);
960 error_sock_relay:
961 if (control_sock->fd >= 0) {
962 ret = control_sock->ops->close(control_sock);
963 if (ret) {
964 PERROR("close");
965 }
966 }
967 lttcomm_destroy_sock(control_sock);
968 error_sock_control:
969 if (err) {
970 health_error();
971 ERR("Health error occurred in %s", __func__);
972 }
973 health_unregister(health_relayd);
974 DBG("Relay listener thread cleanup complete");
975 lttng_relay_stop_threads();
976 return NULL;
977 }
978
979 /*
980 * This thread manages the dispatching of the requests to worker threads
981 */
982 static void *relay_thread_dispatcher(void *data)
983 {
984 int err = -1;
985 ssize_t ret;
986 struct cds_wfcq_node *node;
987 struct relay_connection *new_conn = NULL;
988
989 DBG("[thread] Relay dispatcher started");
990
991 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
992
993 if (testpoint(relayd_thread_dispatcher)) {
994 goto error_testpoint;
995 }
996
997 health_code_update();
998
999 for (;;) {
1000 health_code_update();
1001
1002 /* Atomically prepare the queue futex */
1003 futex_nto1_prepare(&relay_conn_queue.futex);
1004
1005 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1006 break;
1007 }
1008
1009 do {
1010 health_code_update();
1011
1012 /* Dequeue commands */
1013 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1014 &relay_conn_queue.tail);
1015 if (node == NULL) {
1016 DBG("Woken up but nothing in the relay command queue");
1017 /* Continue thread execution */
1018 break;
1019 }
1020 new_conn = caa_container_of(node, struct relay_connection, qnode);
1021
1022 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1023
1024 /*
1025 * Inform worker thread of the new request. This
1026 * call is blocking so we can be assured that
1027 * the data will be read at some point in time
1028 * or wait to the end of the world :)
1029 */
1030 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1031 if (ret < 0) {
1032 PERROR("write connection pipe");
1033 connection_put(new_conn);
1034 goto error;
1035 }
1036 } while (node != NULL);
1037
1038 /* Futex wait on queue. Blocking call on futex() */
1039 health_poll_entry();
1040 futex_nto1_wait(&relay_conn_queue.futex);
1041 health_poll_exit();
1042 }
1043
1044 /* Normal exit, no error */
1045 err = 0;
1046
1047 error:
1048 error_testpoint:
1049 if (err) {
1050 health_error();
1051 ERR("Health error occurred in %s", __func__);
1052 }
1053 health_unregister(health_relayd);
1054 DBG("Dispatch thread dying");
1055 lttng_relay_stop_threads();
1056 return NULL;
1057 }
1058
1059 /*
1060 * Set index data from the control port to a given index object.
1061 */
1062 static int set_index_control_data(struct relay_index *index,
1063 struct lttcomm_relayd_index *data,
1064 struct relay_connection *conn)
1065 {
1066 struct ctf_packet_index index_data;
1067
1068 /*
1069 * The index on disk is encoded in big endian.
1070 */
1071 index_data.packet_size = htobe64(data->packet_size);
1072 index_data.content_size = htobe64(data->content_size);
1073 index_data.timestamp_begin = htobe64(data->timestamp_begin);
1074 index_data.timestamp_end = htobe64(data->timestamp_end);
1075 index_data.events_discarded = htobe64(data->events_discarded);
1076 index_data.stream_id = htobe64(data->stream_id);
1077
1078 if (conn->minor >= 8) {
1079 index->index_data.stream_instance_id = htobe64(data->stream_instance_id);
1080 index->index_data.packet_seq_num = htobe64(data->packet_seq_num);
1081 }
1082
1083 return relay_index_set_data(index, &index_data);
1084 }
1085
1086 static bool session_streams_have_index(const struct relay_session *session)
1087 {
1088 return session->minor >= 4 && !session->snapshot;
1089 }
1090
1091 /*
1092 * Handle the RELAYD_CREATE_SESSION command.
1093 *
1094 * On success, send back the session id or else return a negative value.
1095 */
1096 static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr,
1097 struct relay_connection *conn,
1098 const struct lttng_buffer_view *payload)
1099 {
1100 int ret = 0;
1101 ssize_t send_ret;
1102 struct relay_session *session;
1103 struct lttcomm_relayd_status_session reply;
1104 char session_name[LTTNG_NAME_MAX];
1105 char hostname[LTTNG_HOST_NAME_MAX];
1106 uint32_t live_timer = 0;
1107 bool snapshot = false;
1108
1109 memset(session_name, 0, LTTNG_NAME_MAX);
1110 memset(hostname, 0, LTTNG_HOST_NAME_MAX);
1111
1112 memset(&reply, 0, sizeof(reply));
1113
1114 if (conn->minor < 4) {
1115 /* From 2.1 to 2.3 */
1116 ret = 0;
1117 } else if (conn->minor >= 4 && conn->minor < 11) {
1118 /* From 2.4 to 2.10 */
1119 ret = cmd_create_session_2_4(payload, session_name,
1120 hostname, &live_timer, &snapshot);
1121 } else {
1122 /* From 2.11 to ... */
1123 ret = cmd_create_session_2_11(payload, session_name,
1124 hostname, &live_timer, &snapshot);
1125 }
1126
1127 if (ret < 0) {
1128 goto send_reply;
1129 }
1130
1131 session = session_create(session_name, hostname, live_timer,
1132 snapshot, conn->major, conn->minor);
1133 if (!session) {
1134 ret = -1;
1135 goto send_reply;
1136 }
1137 assert(!conn->session);
1138 conn->session = session;
1139 DBG("Created session %" PRIu64, session->id);
1140
1141 reply.session_id = htobe64(session->id);
1142
1143 send_reply:
1144 if (ret < 0) {
1145 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1146 } else {
1147 reply.ret_code = htobe32(LTTNG_OK);
1148 }
1149
1150 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1151 if (send_ret < (ssize_t) sizeof(reply)) {
1152 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1153 send_ret);
1154 ret = -1;
1155 }
1156
1157 return ret;
1158 }
1159
1160 /*
1161 * When we have received all the streams and the metadata for a channel,
1162 * we make them visible to the viewer threads.
1163 */
1164 static void publish_connection_local_streams(struct relay_connection *conn)
1165 {
1166 struct relay_stream *stream;
1167 struct relay_session *session = conn->session;
1168
1169 /*
1170 * We publish all streams belonging to a session atomically wrt
1171 * session lock.
1172 */
1173 pthread_mutex_lock(&session->lock);
1174 rcu_read_lock();
1175 cds_list_for_each_entry_rcu(stream, &session->recv_list,
1176 recv_node) {
1177 stream_publish(stream);
1178 }
1179 rcu_read_unlock();
1180
1181 /*
1182 * Inform the viewer that there are new streams in the session.
1183 */
1184 if (session->viewer_attached) {
1185 uatomic_set(&session->new_streams, 1);
1186 }
1187 pthread_mutex_unlock(&session->lock);
1188 }
1189
1190 /*
1191 * relay_add_stream: allocate a new stream for a session
1192 */
1193 static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1194 struct relay_connection *conn,
1195 const struct lttng_buffer_view *payload)
1196 {
1197 int ret;
1198 ssize_t send_ret;
1199 struct relay_session *session = conn->session;
1200 struct relay_stream *stream = NULL;
1201 struct lttcomm_relayd_status_stream reply;
1202 struct ctf_trace *trace = NULL;
1203 uint64_t stream_handle = -1ULL;
1204 char *path_name = NULL, *channel_name = NULL;
1205 uint64_t tracefile_size = 0, tracefile_count = 0;
1206 struct relay_stream_chunk_id stream_chunk_id = { 0 };
1207
1208 if (!session || !conn->version_check_done) {
1209 ERR("Trying to add a stream before version check");
1210 ret = -1;
1211 goto end_no_session;
1212 }
1213
1214 if (session->minor == 1) {
1215 /* For 2.1 */
1216 ret = cmd_recv_stream_2_1(payload, &path_name,
1217 &channel_name);
1218 } else if (session->minor > 1 && session->minor < 11) {
1219 /* From 2.2 to 2.10 */
1220 ret = cmd_recv_stream_2_2(payload, &path_name,
1221 &channel_name, &tracefile_size, &tracefile_count);
1222 } else {
1223 /* From 2.11 to ... */
1224 ret = cmd_recv_stream_2_11(payload, &path_name,
1225 &channel_name, &tracefile_size, &tracefile_count,
1226 &stream_chunk_id.value);
1227 stream_chunk_id.is_set = true;
1228 }
1229
1230 if (ret < 0) {
1231 goto send_reply;
1232 }
1233
1234 trace = ctf_trace_get_by_path_or_create(session, path_name);
1235 if (!trace) {
1236 goto send_reply;
1237 }
1238 /* This stream here has one reference on the trace. */
1239
1240 pthread_mutex_lock(&last_relay_stream_id_lock);
1241 stream_handle = ++last_relay_stream_id;
1242 pthread_mutex_unlock(&last_relay_stream_id_lock);
1243
1244 /* We pass ownership of path_name and channel_name. */
1245 stream = stream_create(trace, stream_handle, path_name,
1246 channel_name, tracefile_size, tracefile_count,
1247 &stream_chunk_id);
1248 path_name = NULL;
1249 channel_name = NULL;
1250
1251 /*
1252 * Streams are the owners of their trace. Reference to trace is
1253 * kept within stream_create().
1254 */
1255 ctf_trace_put(trace);
1256
1257 send_reply:
1258 memset(&reply, 0, sizeof(reply));
1259 reply.handle = htobe64(stream_handle);
1260 if (!stream) {
1261 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1262 } else {
1263 reply.ret_code = htobe32(LTTNG_OK);
1264 }
1265
1266 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1267 sizeof(struct lttcomm_relayd_status_stream), 0);
1268 if (send_ret < (ssize_t) sizeof(reply)) {
1269 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1270 send_ret);
1271 ret = -1;
1272 }
1273
1274 end_no_session:
1275 free(path_name);
1276 free(channel_name);
1277 return ret;
1278 }
1279
1280 /*
1281 * relay_close_stream: close a specific stream
1282 */
1283 static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1284 struct relay_connection *conn,
1285 const struct lttng_buffer_view *payload)
1286 {
1287 int ret;
1288 ssize_t send_ret;
1289 struct relay_session *session = conn->session;
1290 struct lttcomm_relayd_close_stream stream_info;
1291 struct lttcomm_relayd_generic_reply reply;
1292 struct relay_stream *stream;
1293
1294 DBG("Close stream received");
1295
1296 if (!session || !conn->version_check_done) {
1297 ERR("Trying to close a stream before version check");
1298 ret = -1;
1299 goto end_no_session;
1300 }
1301
1302 if (payload->size < sizeof(stream_info)) {
1303 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1304 sizeof(stream_info), payload->size);
1305 ret = -1;
1306 goto end_no_session;
1307 }
1308 memcpy(&stream_info, payload->data, sizeof(stream_info));
1309 stream_info.stream_id = be64toh(stream_info.stream_id);
1310 stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1311
1312 stream = stream_get_by_id(stream_info.stream_id);
1313 if (!stream) {
1314 ret = -1;
1315 goto end;
1316 }
1317
1318 /*
1319 * Set last_net_seq_num before the close flag. Required by data
1320 * pending check.
1321 */
1322 pthread_mutex_lock(&stream->lock);
1323 stream->last_net_seq_num = stream_info.last_net_seq_num;
1324 pthread_mutex_unlock(&stream->lock);
1325
1326 /*
1327 * This is one of the conditions which may trigger a stream close
1328 * with the others being:
1329 * 1) A close command is received for a stream
1330 * 2) The control connection owning the stream is closed
1331 * 3) We have received all of the stream's data _after_ a close
1332 * request.
1333 */
1334 try_stream_close(stream);
1335 if (stream->is_metadata) {
1336 struct relay_viewer_stream *vstream;
1337
1338 vstream = viewer_stream_get_by_id(stream->stream_handle);
1339 if (vstream) {
1340 if (vstream->metadata_sent == stream->metadata_received) {
1341 /*
1342 * Since all the metadata has been sent to the
1343 * viewer and that we have a request to close
1344 * its stream, we can safely teardown the
1345 * corresponding metadata viewer stream.
1346 */
1347 viewer_stream_put(vstream);
1348 }
1349 /* Put local reference. */
1350 viewer_stream_put(vstream);
1351 }
1352 }
1353 stream_put(stream);
1354 ret = 0;
1355
1356 end:
1357 memset(&reply, 0, sizeof(reply));
1358 if (ret < 0) {
1359 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1360 } else {
1361 reply.ret_code = htobe32(LTTNG_OK);
1362 }
1363 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1364 sizeof(struct lttcomm_relayd_generic_reply), 0);
1365 if (send_ret < (ssize_t) sizeof(reply)) {
1366 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1367 send_ret);
1368 ret = -1;
1369 }
1370
1371 end_no_session:
1372 return ret;
1373 }
1374
1375 /*
1376 * relay_reset_metadata: reset a metadata stream
1377 */
1378 static
1379 int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1380 struct relay_connection *conn,
1381 const struct lttng_buffer_view *payload)
1382 {
1383 int ret;
1384 ssize_t send_ret;
1385 struct relay_session *session = conn->session;
1386 struct lttcomm_relayd_reset_metadata stream_info;
1387 struct lttcomm_relayd_generic_reply reply;
1388 struct relay_stream *stream;
1389
1390 DBG("Reset metadata received");
1391
1392 if (!session || !conn->version_check_done) {
1393 ERR("Trying to reset a metadata stream before version check");
1394 ret = -1;
1395 goto end_no_session;
1396 }
1397
1398 if (payload->size < sizeof(stream_info)) {
1399 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1400 sizeof(stream_info), payload->size);
1401 ret = -1;
1402 goto end_no_session;
1403 }
1404 memcpy(&stream_info, payload->data, sizeof(stream_info));
1405 stream_info.stream_id = be64toh(stream_info.stream_id);
1406 stream_info.version = be64toh(stream_info.version);
1407
1408 DBG("Update metadata to version %" PRIu64, stream_info.version);
1409
1410 /* Unsupported for live sessions for now. */
1411 if (session->live_timer != 0) {
1412 ret = -1;
1413 goto end;
1414 }
1415
1416 stream = stream_get_by_id(stream_info.stream_id);
1417 if (!stream) {
1418 ret = -1;
1419 goto end;
1420 }
1421 pthread_mutex_lock(&stream->lock);
1422 if (!stream->is_metadata) {
1423 ret = -1;
1424 goto end_unlock;
1425 }
1426
1427 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
1428 0, 0, -1, -1, stream->stream_fd->fd, NULL,
1429 &stream->stream_fd->fd);
1430 if (ret < 0) {
1431 ERR("Failed to rotate metadata file %s of channel %s",
1432 stream->path_name, stream->channel_name);
1433 goto end_unlock;
1434 }
1435
1436 end_unlock:
1437 pthread_mutex_unlock(&stream->lock);
1438 stream_put(stream);
1439
1440 end:
1441 memset(&reply, 0, sizeof(reply));
1442 if (ret < 0) {
1443 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1444 } else {
1445 reply.ret_code = htobe32(LTTNG_OK);
1446 }
1447 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1448 sizeof(struct lttcomm_relayd_generic_reply), 0);
1449 if (send_ret < (ssize_t) sizeof(reply)) {
1450 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1451 send_ret);
1452 ret = -1;
1453 }
1454
1455 end_no_session:
1456 return ret;
1457 }
1458
1459 /*
1460 * relay_unknown_command: send -1 if received unknown command
1461 */
1462 static void relay_unknown_command(struct relay_connection *conn)
1463 {
1464 struct lttcomm_relayd_generic_reply reply;
1465 ssize_t send_ret;
1466
1467 memset(&reply, 0, sizeof(reply));
1468 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1469 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1470 if (send_ret < sizeof(reply)) {
1471 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
1472 }
1473 }
1474
1475 /*
1476 * relay_start: send an acknowledgment to the client to tell if we are
1477 * ready to receive data. We are ready if a session is established.
1478 */
1479 static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr,
1480 struct relay_connection *conn,
1481 const struct lttng_buffer_view *payload)
1482 {
1483 int ret = 0;
1484 ssize_t send_ret;
1485 struct lttcomm_relayd_generic_reply reply;
1486 struct relay_session *session = conn->session;
1487
1488 if (!session) {
1489 DBG("Trying to start the streaming without a session established");
1490 ret = htobe32(LTTNG_ERR_UNK);
1491 }
1492
1493 memset(&reply, 0, sizeof(reply));
1494 reply.ret_code = htobe32(LTTNG_OK);
1495 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1496 sizeof(reply), 0);
1497 if (send_ret < (ssize_t) sizeof(reply)) {
1498 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1499 send_ret);
1500 ret = -1;
1501 }
1502
1503 return ret;
1504 }
1505
1506 /*
1507 * Append padding to the file pointed by the file descriptor fd.
1508 */
1509 static int write_padding_to_file(int fd, uint32_t size)
1510 {
1511 ssize_t ret = 0;
1512 char *zeros;
1513
1514 if (size == 0) {
1515 goto end;
1516 }
1517
1518 zeros = zmalloc(size);
1519 if (zeros == NULL) {
1520 PERROR("zmalloc zeros for padding");
1521 ret = -1;
1522 goto end;
1523 }
1524
1525 ret = lttng_write(fd, zeros, size);
1526 if (ret < size) {
1527 PERROR("write padding to file");
1528 }
1529
1530 free(zeros);
1531
1532 end:
1533 return ret;
1534 }
1535
1536 /*
1537 * Close the current index file if it is open, and create a new one.
1538 *
1539 * Return 0 on success, -1 on error.
1540 */
1541 static
1542 int create_rotate_index_file(struct relay_stream *stream)
1543 {
1544 int ret;
1545 uint32_t major, minor;
1546
1547 /* Put ref on previous index_file. */
1548 if (stream->index_file) {
1549 lttng_index_file_put(stream->index_file);
1550 stream->index_file = NULL;
1551 }
1552 major = stream->trace->session->major;
1553 minor = stream->trace->session->minor;
1554 stream->index_file = lttng_index_file_create(stream->path_name,
1555 stream->channel_name,
1556 -1, -1, stream->tracefile_size,
1557 tracefile_array_get_file_index_head(stream->tfa),
1558 lttng_to_index_major(major, minor),
1559 lttng_to_index_minor(major, minor));
1560 if (!stream->index_file) {
1561 ret = -1;
1562 goto end;
1563 }
1564
1565 ret = 0;
1566
1567 end:
1568 return ret;
1569 }
1570
1571 static
1572 int do_rotate_stream(struct relay_stream *stream)
1573 {
1574 int ret;
1575
1576 /* Perform the stream rotation. */
1577 ret = utils_rotate_stream_file(stream->path_name,
1578 stream->channel_name, stream->tracefile_size,
1579 stream->tracefile_count, -1,
1580 -1, stream->stream_fd->fd,
1581 NULL, &stream->stream_fd->fd);
1582 if (ret < 0) {
1583 ERR("Rotating stream output file");
1584 goto end;
1585 }
1586 stream->tracefile_size_current = 0;
1587
1588 /* Rotate also the index if the stream is not a metadata stream. */
1589 if (!stream->is_metadata) {
1590 ret = create_rotate_index_file(stream);
1591 if (ret < 0) {
1592 ERR("Failed to rotate index file");
1593 goto end;
1594 }
1595 }
1596
1597 stream->rotate_at_seq_num = -1ULL;
1598 stream->pos_after_last_complete_data_index = 0;
1599
1600 end:
1601 return ret;
1602 }
1603
1604 /*
1605 * If too much data has been written in a tracefile before we received the
1606 * rotation command, we have to move the excess data to the new tracefile and
1607 * perform the rotation. This can happen because the control and data
1608 * connections are separate, the indexes as well as the commands arrive from
1609 * the control connection and we have no control over the order so we could be
1610 * in a situation where too much data has been received on the data connection
1611 * before the rotation command on the control connection arrives. We don't need
1612 * to update the index because its order is guaranteed with the rotation
1613 * command message.
1614 */
1615 static
1616 int rotate_truncate_stream(struct relay_stream *stream)
1617 {
1618 int ret, new_fd;
1619 off_t lseek_ret;
1620 uint64_t diff, pos = 0;
1621 char buf[FILE_COPY_BUFFER_SIZE];
1622
1623 assert(!stream->is_metadata);
1624
1625 assert(stream->tracefile_size_current >
1626 stream->pos_after_last_complete_data_index);
1627 diff = stream->tracefile_size_current -
1628 stream->pos_after_last_complete_data_index;
1629
1630 /* Create the new tracefile. */
1631 new_fd = utils_create_stream_file(stream->path_name,
1632 stream->channel_name,
1633 stream->tracefile_size, stream->tracefile_count,
1634 /* uid */ -1, /* gid */ -1, /* suffix */ NULL);
1635 if (new_fd < 0) {
1636 ERR("Failed to create new stream file at path %s for channel %s",
1637 stream->path_name, stream->channel_name);
1638 ret = -1;
1639 goto end;
1640 }
1641
1642 /*
1643 * Rewind the current tracefile to the position at which the rotation
1644 * should have occured.
1645 */
1646 lseek_ret = lseek(stream->stream_fd->fd,
1647 stream->pos_after_last_complete_data_index, SEEK_SET);
1648 if (lseek_ret < 0) {
1649 PERROR("seek truncate stream");
1650 ret = -1;
1651 goto end;
1652 }
1653
1654 /* Move data from the old file to the new file. */
1655 while (pos < diff) {
1656 uint64_t count, bytes_left;
1657 ssize_t io_ret;
1658
1659 bytes_left = diff - pos;
1660 count = bytes_left > sizeof(buf) ? sizeof(buf) : bytes_left;
1661 assert(count <= SIZE_MAX);
1662
1663 io_ret = lttng_read(stream->stream_fd->fd, buf, count);
1664 if (io_ret < (ssize_t) count) {
1665 char error_string[256];
1666
1667 snprintf(error_string, sizeof(error_string),
1668 "Failed to read %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi",
1669 count, stream->stream_fd->fd, io_ret);
1670 if (io_ret == -1) {
1671 PERROR("%s", error_string);
1672 } else {
1673 ERR("%s", error_string);
1674 }
1675 ret = -1;
1676 goto end;
1677 }
1678
1679 io_ret = lttng_write(new_fd, buf, count);
1680 if (io_ret < (ssize_t) count) {
1681 char error_string[256];
1682
1683 snprintf(error_string, sizeof(error_string),
1684 "Failed to write %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi",
1685 count, new_fd, io_ret);
1686 if (io_ret == -1) {
1687 PERROR("%s", error_string);
1688 } else {
1689 ERR("%s", error_string);
1690 }
1691 ret = -1;
1692 goto end;
1693 }
1694
1695 pos += count;
1696 }
1697
1698 /* Truncate the file to get rid of the excess data. */
1699 ret = ftruncate(stream->stream_fd->fd,
1700 stream->pos_after_last_complete_data_index);
1701 if (ret) {
1702 PERROR("ftruncate");
1703 goto end;
1704 }
1705
1706 ret = close(stream->stream_fd->fd);
1707 if (ret < 0) {
1708 PERROR("Closing tracefile");
1709 goto end;
1710 }
1711
1712 ret = create_rotate_index_file(stream);
1713 if (ret < 0) {
1714 ERR("Rotate stream index file");
1715 goto end;
1716 }
1717
1718 /*
1719 * Update the offset and FD of all the eventual indexes created by the
1720 * data connection before the rotation command arrived.
1721 */
1722 ret = relay_index_switch_all_files(stream);
1723 if (ret < 0) {
1724 ERR("Failed to rotate index file");
1725 goto end;
1726 }
1727
1728 stream->stream_fd->fd = new_fd;
1729 stream->tracefile_size_current = diff;
1730 stream->pos_after_last_complete_data_index = 0;
1731 stream->rotate_at_seq_num = -1ULL;
1732
1733 ret = 0;
1734
1735 end:
1736 return ret;
1737 }
1738
1739 /*
1740 * Check if a stream should perform a rotation (for session rotation).
1741 * Must be called with the stream lock held.
1742 *
1743 * Return 0 on success, a negative value on error.
1744 */
1745 static
1746 int try_rotate_stream(struct relay_stream *stream)
1747 {
1748 int ret = 0;
1749 uint64_t trace_seq;
1750
1751 /* No rotation expected. */
1752 if (stream->rotate_at_seq_num == -1ULL) {
1753 goto end;
1754 }
1755
1756 trace_seq = min(stream->prev_seq, stream->prev_index_seq);
1757 if (stream->prev_seq == -1ULL || stream->prev_index_seq == -1ULL ||
1758 trace_seq < stream->rotate_at_seq_num) {
1759 DBG("Stream %" PRIu64 " not yet ready for rotation (rotate_at_seq_num = %" PRIu64 ", prev_seq = %" PRIu64 ", prev_index_seq = %" PRIu64 ")",
1760 stream->stream_handle,
1761 stream->rotate_at_seq_num,
1762 stream->prev_seq,
1763 stream->prev_index_seq);
1764 goto end;
1765 } else if (stream->prev_seq > stream->rotate_at_seq_num) {
1766 /*
1767 * prev_seq is checked here since indexes and rotation
1768 * commands are serialized with respect to each other.
1769 */
1770 DBG("Rotation after too much data has been written in tracefile "
1771 "for stream %" PRIu64 ", need to truncate before "
1772 "rotating", stream->stream_handle);
1773 ret = rotate_truncate_stream(stream);
1774 if (ret) {
1775 ERR("Failed to truncate stream");
1776 goto end;
1777 }
1778 } else {
1779 if (trace_seq != stream->rotate_at_seq_num) {
1780 /*
1781 * Unexpected, protocol error/bug.
1782 * It could mean that we received a rotation position
1783 * that is in the past.
1784 */
1785 ERR("Stream %" PRIu64 " is in an inconsistent state (rotate_at_seq_num = %" PRIu64 ", prev_seq = %" PRIu64 ", prev_index_seq = %" PRIu64 ")",
1786 stream->stream_handle,
1787 stream->rotate_at_seq_num,
1788 stream->prev_seq,
1789 stream->prev_index_seq);
1790 ret = -1;
1791 goto end;
1792 }
1793 DBG("Stream %" PRIu64 " ready for rotation",
1794 stream->stream_handle);
1795 ret = do_rotate_stream(stream);
1796 }
1797
1798 end:
1799 return ret;
1800 }
1801
1802 /*
1803 * relay_recv_metadata: receive the metadata for the session.
1804 */
1805 static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1806 struct relay_connection *conn,
1807 const struct lttng_buffer_view *payload)
1808 {
1809 int ret = 0;
1810 ssize_t size_ret;
1811 struct relay_session *session = conn->session;
1812 struct lttcomm_relayd_metadata_payload metadata_payload_header;
1813 struct relay_stream *metadata_stream;
1814 uint64_t metadata_payload_size;
1815
1816 if (!session) {
1817 ERR("Metadata sent before version check");
1818 ret = -1;
1819 goto end;
1820 }
1821
1822 if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1823 ERR("Incorrect data size");
1824 ret = -1;
1825 goto end;
1826 }
1827 metadata_payload_size = recv_hdr->data_size -
1828 sizeof(struct lttcomm_relayd_metadata_payload);
1829
1830 memcpy(&metadata_payload_header, payload->data,
1831 sizeof(metadata_payload_header));
1832 metadata_payload_header.stream_id = be64toh(
1833 metadata_payload_header.stream_id);
1834 metadata_payload_header.padding_size = be32toh(
1835 metadata_payload_header.padding_size);
1836
1837 metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
1838 if (!metadata_stream) {
1839 ret = -1;
1840 goto end;
1841 }
1842
1843 pthread_mutex_lock(&metadata_stream->lock);
1844
1845 size_ret = lttng_write(metadata_stream->stream_fd->fd,
1846 payload->data + sizeof(metadata_payload_header),
1847 metadata_payload_size);
1848 if (size_ret < metadata_payload_size) {
1849 ERR("Relay error writing metadata on file");
1850 ret = -1;
1851 goto end_put;
1852 }
1853
1854 size_ret = write_padding_to_file(metadata_stream->stream_fd->fd,
1855 metadata_payload_header.padding_size);
1856 if (size_ret < (int64_t) metadata_payload_header.padding_size) {
1857 ret = -1;
1858 goto end_put;
1859 }
1860
1861 metadata_stream->metadata_received +=
1862 metadata_payload_size + metadata_payload_header.padding_size;
1863 DBG2("Relay metadata written. Updated metadata_received %" PRIu64,
1864 metadata_stream->metadata_received);
1865
1866 ret = try_rotate_stream(metadata_stream);
1867 if (ret < 0) {
1868 goto end_put;
1869 }
1870
1871 end_put:
1872 pthread_mutex_unlock(&metadata_stream->lock);
1873 stream_put(metadata_stream);
1874 end:
1875 return ret;
1876 }
1877
1878 /*
1879 * relay_send_version: send relayd version number
1880 */
1881 static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr,
1882 struct relay_connection *conn,
1883 const struct lttng_buffer_view *payload)
1884 {
1885 int ret;
1886 ssize_t send_ret;
1887 struct lttcomm_relayd_version reply, msg;
1888 bool compatible = true;
1889
1890 conn->version_check_done = true;
1891
1892 /* Get version from the other side. */
1893 if (payload->size < sizeof(msg)) {
1894 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1895 sizeof(msg), payload->size);
1896 ret = -1;
1897 goto end;
1898 }
1899
1900 memcpy(&msg, payload->data, sizeof(msg));
1901 msg.major = be32toh(msg.major);
1902 msg.minor = be32toh(msg.minor);
1903
1904 memset(&reply, 0, sizeof(reply));
1905 reply.major = RELAYD_VERSION_COMM_MAJOR;
1906 reply.minor = RELAYD_VERSION_COMM_MINOR;
1907
1908 /* Major versions must be the same */
1909 if (reply.major != msg.major) {
1910 DBG("Incompatible major versions (%u vs %u), deleting session",
1911 reply.major, msg.major);
1912 compatible = false;
1913 }
1914
1915 conn->major = reply.major;
1916 /* We adapt to the lowest compatible version */
1917 if (reply.minor <= msg.minor) {
1918 conn->minor = reply.minor;
1919 } else {
1920 conn->minor = msg.minor;
1921 }
1922
1923 reply.major = htobe32(reply.major);
1924 reply.minor = htobe32(reply.minor);
1925 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1926 sizeof(reply), 0);
1927 if (send_ret < (ssize_t) sizeof(reply)) {
1928 ERR("Failed to send \"send version\" command reply (ret = %zd)",
1929 send_ret);
1930 ret = -1;
1931 goto end;
1932 } else {
1933 ret = 0;
1934 }
1935
1936 if (!compatible) {
1937 ret = -1;
1938 goto end;
1939 }
1940
1941 DBG("Version check done using protocol %u.%u", conn->major,
1942 conn->minor);
1943
1944 end:
1945 return ret;
1946 }
1947
1948 /*
1949 * Check for data pending for a given stream id from the session daemon.
1950 */
1951 static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
1952 struct relay_connection *conn,
1953 const struct lttng_buffer_view *payload)
1954 {
1955 struct relay_session *session = conn->session;
1956 struct lttcomm_relayd_data_pending msg;
1957 struct lttcomm_relayd_generic_reply reply;
1958 struct relay_stream *stream;
1959 ssize_t send_ret;
1960 int ret;
1961 uint64_t stream_seq;
1962
1963 DBG("Data pending command received");
1964
1965 if (!session || !conn->version_check_done) {
1966 ERR("Trying to check for data before version check");
1967 ret = -1;
1968 goto end_no_session;
1969 }
1970
1971 if (payload->size < sizeof(msg)) {
1972 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
1973 sizeof(msg), payload->size);
1974 ret = -1;
1975 goto end_no_session;
1976 }
1977 memcpy(&msg, payload->data, sizeof(msg));
1978 msg.stream_id = be64toh(msg.stream_id);
1979 msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
1980
1981 stream = stream_get_by_id(msg.stream_id);
1982 if (stream == NULL) {
1983 ret = -1;
1984 goto end;
1985 }
1986
1987 pthread_mutex_lock(&stream->lock);
1988
1989 if (session_streams_have_index(session)) {
1990 /*
1991 * Ensure that both the index and stream data have been
1992 * flushed up to the requested point.
1993 */
1994 stream_seq = min(stream->prev_seq, stream->prev_index_seq);
1995 } else {
1996 stream_seq = stream->prev_seq;
1997 }
1998 DBG("Data pending for stream id %" PRIu64 ": prev_seq %" PRIu64
1999 ", prev_index_seq %" PRIu64
2000 ", and last_seq %" PRIu64, msg.stream_id,
2001 stream->prev_seq, stream->prev_index_seq,
2002 msg.last_net_seq_num);
2003
2004 /* Avoid wrapping issue */
2005 if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
2006 /* Data has in fact been written and is NOT pending */
2007 ret = 0;
2008 } else {
2009 /* Data still being streamed thus pending */
2010 ret = 1;
2011 }
2012
2013 stream->data_pending_check_done = true;
2014 pthread_mutex_unlock(&stream->lock);
2015
2016 stream_put(stream);
2017 end:
2018
2019 memset(&reply, 0, sizeof(reply));
2020 reply.ret_code = htobe32(ret);
2021 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2022 if (send_ret < (ssize_t) sizeof(reply)) {
2023 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2024 send_ret);
2025 ret = -1;
2026 }
2027
2028 end_no_session:
2029 return ret;
2030 }
2031
2032 /*
2033 * Wait for the control socket to reach a quiescent state.
2034 *
2035 * Note that for now, when receiving this command from the session
2036 * daemon, this means that every subsequent commands or data received on
2037 * the control socket has been handled. So, this is why we simply return
2038 * OK here.
2039 */
2040 static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr,
2041 struct relay_connection *conn,
2042 const struct lttng_buffer_view *payload)
2043 {
2044 int ret;
2045 ssize_t send_ret;
2046 struct relay_stream *stream;
2047 struct lttcomm_relayd_quiescent_control msg;
2048 struct lttcomm_relayd_generic_reply reply;
2049
2050 DBG("Checking quiescent state on control socket");
2051
2052 if (!conn->session || !conn->version_check_done) {
2053 ERR("Trying to check for data before version check");
2054 ret = -1;
2055 goto end_no_session;
2056 }
2057
2058 if (payload->size < sizeof(msg)) {
2059 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2060 sizeof(msg), payload->size);
2061 ret = -1;
2062 goto end_no_session;
2063 }
2064 memcpy(&msg, payload->data, sizeof(msg));
2065 msg.stream_id = be64toh(msg.stream_id);
2066
2067 stream = stream_get_by_id(msg.stream_id);
2068 if (!stream) {
2069 goto reply;
2070 }
2071 pthread_mutex_lock(&stream->lock);
2072 stream->data_pending_check_done = true;
2073 pthread_mutex_unlock(&stream->lock);
2074
2075 DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
2076 stream_put(stream);
2077 reply:
2078 memset(&reply, 0, sizeof(reply));
2079 reply.ret_code = htobe32(LTTNG_OK);
2080 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2081 if (send_ret < (ssize_t) sizeof(reply)) {
2082 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2083 send_ret);
2084 ret = -1;
2085 } else {
2086 ret = 0;
2087 }
2088
2089 end_no_session:
2090 return ret;
2091 }
2092
2093 /*
2094 * Initialize a data pending command. This means that a consumer is about
2095 * to ask for data pending for each stream it holds. Simply iterate over
2096 * all streams of a session and set the data_pending_check_done flag.
2097 *
2098 * This command returns to the client a LTTNG_OK code.
2099 */
2100 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2101 struct relay_connection *conn,
2102 const struct lttng_buffer_view *payload)
2103 {
2104 int ret;
2105 ssize_t send_ret;
2106 struct lttng_ht_iter iter;
2107 struct lttcomm_relayd_begin_data_pending msg;
2108 struct lttcomm_relayd_generic_reply reply;
2109 struct relay_stream *stream;
2110
2111 assert(recv_hdr);
2112 assert(conn);
2113
2114 DBG("Init streams for data pending");
2115
2116 if (!conn->session || !conn->version_check_done) {
2117 ERR("Trying to check for data before version check");
2118 ret = -1;
2119 goto end_no_session;
2120 }
2121
2122 if (payload->size < sizeof(msg)) {
2123 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2124 sizeof(msg), payload->size);
2125 ret = -1;
2126 goto end_no_session;
2127 }
2128 memcpy(&msg, payload->data, sizeof(msg));
2129 msg.session_id = be64toh(msg.session_id);
2130
2131 /*
2132 * Iterate over all streams to set the begin data pending flag.
2133 * For now, the streams are indexed by stream handle so we have
2134 * to iterate over all streams to find the one associated with
2135 * the right session_id.
2136 */
2137 rcu_read_lock();
2138 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2139 node.node) {
2140 if (!stream_get(stream)) {
2141 continue;
2142 }
2143 if (stream->trace->session->id == msg.session_id) {
2144 pthread_mutex_lock(&stream->lock);
2145 stream->data_pending_check_done = false;
2146 pthread_mutex_unlock(&stream->lock);
2147 DBG("Set begin data pending flag to stream %" PRIu64,
2148 stream->stream_handle);
2149 }
2150 stream_put(stream);
2151 }
2152 rcu_read_unlock();
2153
2154 memset(&reply, 0, sizeof(reply));
2155 /* All good, send back reply. */
2156 reply.ret_code = htobe32(LTTNG_OK);
2157
2158 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2159 if (send_ret < (ssize_t) sizeof(reply)) {
2160 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2161 send_ret);
2162 ret = -1;
2163 } else {
2164 ret = 0;
2165 }
2166
2167 end_no_session:
2168 return ret;
2169 }
2170
2171 /*
2172 * End data pending command. This will check, for a given session id, if
2173 * each stream associated with it has its data_pending_check_done flag
2174 * set. If not, this means that the client lost track of the stream but
2175 * the data is still being streamed on our side. In this case, we inform
2176 * the client that data is in flight.
2177 *
2178 * Return to the client if there is data in flight or not with a ret_code.
2179 */
2180 static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2181 struct relay_connection *conn,
2182 const struct lttng_buffer_view *payload)
2183 {
2184 int ret;
2185 ssize_t send_ret;
2186 struct lttng_ht_iter iter;
2187 struct lttcomm_relayd_end_data_pending msg;
2188 struct lttcomm_relayd_generic_reply reply;
2189 struct relay_stream *stream;
2190 uint32_t is_data_inflight = 0;
2191
2192 DBG("End data pending command");
2193
2194 if (!conn->session || !conn->version_check_done) {
2195 ERR("Trying to check for data before version check");
2196 ret = -1;
2197 goto end_no_session;
2198 }
2199
2200 if (payload->size < sizeof(msg)) {
2201 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2202 sizeof(msg), payload->size);
2203 ret = -1;
2204 goto end_no_session;
2205 }
2206 memcpy(&msg, payload->data, sizeof(msg));
2207 msg.session_id = be64toh(msg.session_id);
2208
2209 /*
2210 * Iterate over all streams to see if the begin data pending
2211 * flag is set.
2212 */
2213 rcu_read_lock();
2214 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2215 node.node) {
2216 if (!stream_get(stream)) {
2217 continue;
2218 }
2219 if (stream->trace->session->id != msg.session_id) {
2220 stream_put(stream);
2221 continue;
2222 }
2223 pthread_mutex_lock(&stream->lock);
2224 if (!stream->data_pending_check_done) {
2225 uint64_t stream_seq;
2226
2227 if (session_streams_have_index(conn->session)) {
2228 /*
2229 * Ensure that both the index and stream data have been
2230 * flushed up to the requested point.
2231 */
2232 stream_seq = min(stream->prev_seq, stream->prev_index_seq);
2233 } else {
2234 stream_seq = stream->prev_seq;
2235 }
2236 if (!stream->closed || !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
2237 is_data_inflight = 1;
2238 DBG("Data is still in flight for stream %" PRIu64,
2239 stream->stream_handle);
2240 pthread_mutex_unlock(&stream->lock);
2241 stream_put(stream);
2242 break;
2243 }
2244 }
2245 pthread_mutex_unlock(&stream->lock);
2246 stream_put(stream);
2247 }
2248 rcu_read_unlock();
2249
2250 memset(&reply, 0, sizeof(reply));
2251 /* All good, send back reply. */
2252 reply.ret_code = htobe32(is_data_inflight);
2253
2254 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2255 if (send_ret < (ssize_t) sizeof(reply)) {
2256 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2257 send_ret);
2258 ret = -1;
2259 } else {
2260 ret = 0;
2261 }
2262
2263 end_no_session:
2264 return ret;
2265 }
2266
2267 /*
2268 * Receive an index for a specific stream.
2269 *
2270 * Return 0 on success else a negative value.
2271 */
2272 static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr,
2273 struct relay_connection *conn,
2274 const struct lttng_buffer_view *payload)
2275 {
2276 int ret;
2277 ssize_t send_ret;
2278 struct relay_session *session = conn->session;
2279 struct lttcomm_relayd_index index_info;
2280 struct relay_index *index;
2281 struct lttcomm_relayd_generic_reply reply;
2282 struct relay_stream *stream;
2283 size_t msg_len;
2284
2285 assert(conn);
2286
2287 DBG("Relay receiving index");
2288
2289 if (!session || !conn->version_check_done) {
2290 ERR("Trying to close a stream before version check");
2291 ret = -1;
2292 goto end_no_session;
2293 }
2294
2295 msg_len = lttcomm_relayd_index_len(
2296 lttng_to_index_major(conn->major, conn->minor),
2297 lttng_to_index_minor(conn->major, conn->minor));
2298 if (payload->size < msg_len) {
2299 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2300 msg_len, payload->size);
2301 ret = -1;
2302 goto end_no_session;
2303 }
2304 memcpy(&index_info, payload->data, msg_len);
2305 index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2306 index_info.net_seq_num = be64toh(index_info.net_seq_num);
2307 index_info.packet_size = be64toh(index_info.packet_size);
2308 index_info.content_size = be64toh(index_info.content_size);
2309 index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2310 index_info.timestamp_end = be64toh(index_info.timestamp_end);
2311 index_info.events_discarded = be64toh(index_info.events_discarded);
2312 index_info.stream_id = be64toh(index_info.stream_id);
2313
2314 if (conn->minor >= 8) {
2315 index_info.stream_instance_id =
2316 be64toh(index_info.stream_instance_id);
2317 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2318 }
2319
2320 stream = stream_get_by_id(index_info.relay_stream_id);
2321 if (!stream) {
2322 ERR("stream_get_by_id not found");
2323 ret = -1;
2324 goto end;
2325 }
2326 pthread_mutex_lock(&stream->lock);
2327
2328 /* Live beacon handling */
2329 if (index_info.packet_size == 0) {
2330 DBG("Received live beacon for stream %" PRIu64,
2331 stream->stream_handle);
2332
2333 /*
2334 * Only flag a stream inactive when it has already
2335 * received data and no indexes are in flight.
2336 */
2337 if (stream->index_received_seqcount > 0
2338 && stream->indexes_in_flight == 0) {
2339 stream->beacon_ts_end = index_info.timestamp_end;
2340 }
2341 ret = 0;
2342 goto end_stream_put;
2343 } else {
2344 stream->beacon_ts_end = -1ULL;
2345 }
2346
2347 if (stream->ctf_stream_id == -1ULL) {
2348 stream->ctf_stream_id = index_info.stream_id;
2349 }
2350 index = relay_index_get_by_id_or_create(stream, index_info.net_seq_num);
2351 if (!index) {
2352 ret = -1;
2353 ERR("relay_index_get_by_id_or_create index NULL");
2354 goto end_stream_put;
2355 }
2356 if (set_index_control_data(index, &index_info, conn)) {
2357 ERR("set_index_control_data error");
2358 relay_index_put(index);
2359 ret = -1;
2360 goto end_stream_put;
2361 }
2362 ret = relay_index_try_flush(index);
2363 if (ret == 0) {
2364 tracefile_array_commit_seq(stream->tfa);
2365 stream->index_received_seqcount++;
2366 stream->pos_after_last_complete_data_index += index->total_size;
2367 stream->prev_index_seq = index_info.net_seq_num;
2368 } else if (ret > 0) {
2369 /* no flush. */
2370 ret = 0;
2371 } else {
2372 /*
2373 * ret < 0
2374 *
2375 * relay_index_try_flush is responsible for the self-reference
2376 * put of the index object on error.
2377 */
2378 ERR("relay_index_try_flush error %d", ret);
2379 ret = -1;
2380 }
2381
2382 end_stream_put:
2383 pthread_mutex_unlock(&stream->lock);
2384 stream_put(stream);
2385
2386 end:
2387
2388 memset(&reply, 0, sizeof(reply));
2389 if (ret < 0) {
2390 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2391 } else {
2392 reply.ret_code = htobe32(LTTNG_OK);
2393 }
2394 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2395 if (send_ret < (ssize_t) sizeof(reply)) {
2396 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2397 ret = -1;
2398 }
2399
2400 end_no_session:
2401 return ret;
2402 }
2403
2404 /*
2405 * Receive the streams_sent message.
2406 *
2407 * Return 0 on success else a negative value.
2408 */
2409 static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr,
2410 struct relay_connection *conn,
2411 const struct lttng_buffer_view *payload)
2412 {
2413 int ret;
2414 ssize_t send_ret;
2415 struct lttcomm_relayd_generic_reply reply;
2416
2417 assert(conn);
2418
2419 DBG("Relay receiving streams_sent");
2420
2421 if (!conn->session || !conn->version_check_done) {
2422 ERR("Trying to close a stream before version check");
2423 ret = -1;
2424 goto end_no_session;
2425 }
2426
2427 /*
2428 * Publish every pending stream in the connection recv list which are
2429 * now ready to be used by the viewer.
2430 */
2431 publish_connection_local_streams(conn);
2432
2433 memset(&reply, 0, sizeof(reply));
2434 reply.ret_code = htobe32(LTTNG_OK);
2435 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2436 if (send_ret < (ssize_t) sizeof(reply)) {
2437 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2438 send_ret);
2439 ret = -1;
2440 } else {
2441 /* Success. */
2442 ret = 0;
2443 }
2444
2445 end_no_session:
2446 return ret;
2447 }
2448
2449 /*
2450 * relay_rotate_session_stream: rotate a stream to a new tracefile for the session
2451 * rotation feature (not the tracefile rotation feature).
2452 */
2453 static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr *recv_hdr,
2454 struct relay_connection *conn,
2455 const struct lttng_buffer_view *payload)
2456 {
2457 int ret;
2458 ssize_t send_ret;
2459 struct relay_session *session = conn->session;
2460 struct lttcomm_relayd_rotate_stream stream_info;
2461 struct lttcomm_relayd_generic_reply reply;
2462 struct relay_stream *stream;
2463 size_t header_len;
2464 size_t path_len;
2465 struct lttng_buffer_view new_path_view;
2466
2467 DBG("Rotate stream received");
2468
2469 if (!session || !conn->version_check_done) {
2470 ERR("Trying to rotate a stream before version check");
2471 ret = -1;
2472 goto end_no_reply;
2473 }
2474
2475 if (session->major == 2 && session->minor < 11) {
2476 ERR("Unsupported feature before 2.11");
2477 ret = -1;
2478 goto end_no_reply;
2479 }
2480
2481 header_len = sizeof(struct lttcomm_relayd_rotate_stream);
2482
2483 if (payload->size < header_len) {
2484 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2485 header_len, payload->size);
2486 ret = -1;
2487 goto end_no_reply;
2488 }
2489
2490 memcpy(&stream_info, payload->data, header_len);
2491
2492 /* Convert to host */
2493 stream_info.pathname_length = be32toh(stream_info.pathname_length);
2494 stream_info.stream_id = be64toh(stream_info.stream_id);
2495 stream_info.new_chunk_id = be64toh(stream_info.new_chunk_id);
2496 stream_info.rotate_at_seq_num = be64toh(stream_info.rotate_at_seq_num);
2497
2498 path_len = stream_info.pathname_length;
2499 if (payload->size < header_len + path_len) {
2500 ERR("Unexpected payload size in \"relay_rotate_session_stream\" including path: expected >= %zu bytes, got %zu bytes",
2501 header_len + path_len, payload->size);
2502 ret = -1;
2503 goto end_no_reply;
2504 }
2505
2506 /* Ensure it fits in local filename length. */
2507 if (path_len >= LTTNG_PATH_MAX) {
2508 ret = -ENAMETOOLONG;
2509 ERR("Length of relay_rotate_session_stream command's path name (%zu bytes) exceeds the maximal allowed length of %i bytes",
2510 path_len, LTTNG_PATH_MAX);
2511 goto end;
2512 }
2513
2514 new_path_view = lttng_buffer_view_from_view(payload, header_len,
2515 stream_info.pathname_length);
2516
2517 stream = stream_get_by_id(stream_info.stream_id);
2518 if (!stream) {
2519 ret = -1;
2520 goto end;
2521 }
2522
2523 pthread_mutex_lock(&stream->lock);
2524
2525 /*
2526 * Update the trace path (just the folder, the stream name does not
2527 * change).
2528 */
2529 free(stream->path_name);
2530 stream->path_name = create_output_path(new_path_view.data);
2531 if (!stream->path_name) {
2532 ERR("Failed to create a new output path");
2533 ret = -1;
2534 goto end_stream_unlock;
2535 }
2536 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG,
2537 -1, -1);
2538 if (ret < 0) {
2539 ERR("relay creating output directory");
2540 ret = -1;
2541 goto end_stream_unlock;
2542 }
2543
2544 assert(stream->current_chunk_id.is_set);
2545 stream->current_chunk_id.value = stream_info.new_chunk_id;
2546
2547 if (stream->is_metadata) {
2548 /*
2549 * The metadata stream is sent only over the control connection
2550 * so we know we have all the data to perform the stream
2551 * rotation.
2552 */
2553 ret = do_rotate_stream(stream);
2554 } else {
2555 stream->rotate_at_seq_num = stream_info.rotate_at_seq_num;
2556 ret = try_rotate_stream(stream);
2557 }
2558 if (ret < 0) {
2559 goto end_stream_unlock;
2560 }
2561
2562 end_stream_unlock:
2563 pthread_mutex_unlock(&stream->lock);
2564 stream_put(stream);
2565 end:
2566 memset(&reply, 0, sizeof(reply));
2567 if (ret < 0) {
2568 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2569 } else {
2570 reply.ret_code = htobe32(LTTNG_OK);
2571 }
2572 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2573 sizeof(struct lttcomm_relayd_generic_reply), 0);
2574 if (send_ret < (ssize_t) sizeof(reply)) {
2575 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2576 send_ret);
2577 ret = -1;
2578 }
2579
2580 end_no_reply:
2581 return ret;
2582 }
2583
2584 /*
2585 * relay_mkdir: Create a folder on the disk.
2586 */
2587 static int relay_mkdir(const struct lttcomm_relayd_hdr *recv_hdr,
2588 struct relay_connection *conn,
2589 const struct lttng_buffer_view *payload)
2590 {
2591 int ret;
2592 struct relay_session *session = conn->session;
2593 struct lttcomm_relayd_mkdir path_info_header;
2594 struct lttcomm_relayd_generic_reply reply;
2595 char *path = NULL;
2596 size_t header_len;
2597 ssize_t send_ret;
2598 struct lttng_buffer_view path_view;
2599
2600 if (!session || !conn->version_check_done) {
2601 ERR("Trying to create a directory before version check");
2602 ret = -1;
2603 goto end_no_session;
2604 }
2605
2606 if (session->major == 2 && session->minor < 11) {
2607 /*
2608 * This client is not supposed to use this command since
2609 * it predates its introduction.
2610 */
2611 ERR("relay_mkdir command is unsupported before LTTng 2.11");
2612 ret = -1;
2613 goto end_no_session;
2614 }
2615
2616 header_len = sizeof(path_info_header);
2617 if (payload->size < header_len) {
2618 ERR("Unexpected payload size in \"relay_mkdir\": expected >= %zu bytes, got %zu bytes",
2619 header_len, payload->size);
2620 ret = -1;
2621 goto end_no_session;
2622 }
2623
2624 memcpy(&path_info_header, payload->data, header_len);
2625
2626 path_info_header.length = be32toh(path_info_header.length);
2627
2628 if (payload->size < header_len + path_info_header.length) {
2629 ERR("Unexpected payload size in \"relay_mkdir\" including path: expected >= %zu bytes, got %zu bytes",
2630 header_len + path_info_header.length, payload->size);
2631 ret = -1;
2632 goto end_no_session;
2633 }
2634
2635 /* Ensure that it fits in local path length. */
2636 if (path_info_header.length >= LTTNG_PATH_MAX) {
2637 ret = -ENAMETOOLONG;
2638 ERR("Path name argument of mkdir command (%" PRIu32 " bytes) exceeds the maximal length allowed (%d bytes)",
2639 path_info_header.length, LTTNG_PATH_MAX);
2640 goto end;
2641 }
2642
2643 path_view = lttng_buffer_view_from_view(payload, header_len,
2644 path_info_header.length);
2645
2646 path = create_output_path(path_view.data);
2647 if (!path) {
2648 ERR("Failed to create output path");
2649 ret = -1;
2650 goto end;
2651 }
2652
2653 ret = utils_mkdir_recursive(path, S_IRWXU | S_IRWXG, -1, -1);
2654 if (ret < 0) {
2655 ERR("relay creating output directory");
2656 goto end;
2657 }
2658
2659 ret = 0;
2660
2661 end:
2662 memset(&reply, 0, sizeof(reply));
2663 if (ret < 0) {
2664 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2665 } else {
2666 reply.ret_code = htobe32(LTTNG_OK);
2667 }
2668 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2669 if (send_ret < (ssize_t) sizeof(reply)) {
2670 ERR("Failed to send \"mkdir\" command reply (ret = %zd)", send_ret);
2671 ret = -1;
2672 }
2673
2674 end_no_session:
2675 free(path);
2676 return ret;
2677 }
2678
2679 static int validate_rotate_rename_path_length(const char *path_type,
2680 uint32_t path_length)
2681 {
2682 int ret = 0;
2683
2684 if (path_length > LTTNG_PATH_MAX) {
2685 ret = -ENAMETOOLONG;
2686 ERR("rotate rename \"%s\" path name length (%" PRIu32 " bytes) exceeds the allowed size of %i bytes",
2687 path_type, path_length, LTTNG_PATH_MAX);
2688 } else if (path_length == 0) {
2689 ret = -EINVAL;
2690 ERR("rotate rename \"%s\" path name has an illegal length of 0", path_type);
2691 }
2692 return ret;
2693 }
2694
2695 /*
2696 * relay_rotate_rename: rename the trace folder after a rotation is
2697 * completed. We are not closing any fd here, just moving the folder, so it
2698 * works even if data is still in-flight.
2699 */
2700 static int relay_rotate_rename(const struct lttcomm_relayd_hdr *recv_hdr,
2701 struct relay_connection *conn,
2702 const struct lttng_buffer_view *payload)
2703 {
2704 int ret;
2705 ssize_t send_ret;
2706 struct relay_session *session = conn->session;
2707 struct lttcomm_relayd_generic_reply reply;
2708 struct lttcomm_relayd_rotate_rename header;
2709 size_t header_len;
2710 size_t received_paths_size;
2711 char *complete_old_path = NULL, *complete_new_path = NULL;
2712 struct lttng_buffer_view old_path_view;
2713 struct lttng_buffer_view new_path_view;
2714
2715 if (!session || !conn->version_check_done) {
2716 ERR("Trying to rename a trace folder before version check");
2717 ret = -1;
2718 goto end_no_reply;
2719 }
2720
2721 if (session->major == 2 && session->minor < 11) {
2722 ERR("relay_rotate_rename command is unsupported before LTTng 2.11");
2723 ret = -1;
2724 goto end_no_reply;
2725 }
2726
2727 header_len = sizeof(header);
2728 if (payload->size < header_len) {
2729 ERR("Unexpected payload size in \"relay_rotate_rename\": expected >= %zu bytes, got %zu bytes",
2730 header_len, payload->size);
2731 ret = -1;
2732 goto end_no_reply;
2733 }
2734
2735 memcpy(&header, payload->data, header_len);
2736
2737 header.old_path_length = be32toh(header.old_path_length);
2738 header.new_path_length = be32toh(header.new_path_length);
2739 received_paths_size = header.old_path_length + header.new_path_length;
2740
2741 if (payload->size < header_len + received_paths_size) {
2742 ERR("Unexpected payload size in \"relay_rotate_rename\" including paths: expected >= %zu bytes, got %zu bytes",
2743 header_len, payload->size);
2744 ret = -1;
2745 goto end_no_reply;
2746 }
2747
2748 /* Ensure the paths don't exceed their allowed size. */
2749 ret = validate_rotate_rename_path_length("old", header.old_path_length);
2750 if (ret) {
2751 goto end;
2752 }
2753 ret = validate_rotate_rename_path_length("new", header.new_path_length);
2754 if (ret) {
2755 goto end;
2756 }
2757
2758 old_path_view = lttng_buffer_view_from_view(payload, header_len,
2759 header.old_path_length);
2760 new_path_view = lttng_buffer_view_from_view(payload,
2761 header_len + header.old_path_length,
2762 header.new_path_length);
2763
2764 /* Validate that both paths received are NULL terminated. */
2765 if (old_path_view.data[old_path_view.size - 1] != '\0') {
2766 ERR("relay_rotate_rename command's \"old\" path is invalid (not NULL terminated)");
2767 ret = -1;
2768 goto end;
2769 }
2770 if (new_path_view.data[new_path_view.size - 1] != '\0') {
2771 ERR("relay_rotate_rename command's \"new\" path is invalid (not NULL terminated)");
2772 ret = -1;
2773 goto end;
2774 }
2775
2776 complete_old_path = create_output_path(old_path_view.data);
2777 if (!complete_old_path) {
2778 ERR("Failed to build old output path in rotate_rename command");
2779 ret = -1;
2780 goto end;
2781 }
2782
2783 complete_new_path = create_output_path(new_path_view.data);
2784 if (!complete_new_path) {
2785 ERR("Failed to build new output path in rotate_rename command");
2786 ret = -1;
2787 goto end;
2788 }
2789
2790 ret = utils_mkdir_recursive(complete_new_path, S_IRWXU | S_IRWXG,
2791 -1, -1);
2792 if (ret < 0) {
2793 ERR("Failed to mkdir() rotate_rename's \"new\" output directory at \"%s\"",
2794 complete_new_path);
2795 goto end;
2796 }
2797
2798 /*
2799 * If a domain has not yet created its channel, the domain-specific
2800 * folder might not exist, but this is not an error.
2801 */
2802 ret = rename(complete_old_path, complete_new_path);
2803 if (ret < 0 && errno != ENOENT) {
2804 PERROR("Renaming chunk in rotate_rename command from \"%s\" to \"%s\"",
2805 complete_old_path, complete_new_path);
2806 goto end;
2807 }
2808 ret = 0;
2809
2810 end:
2811 memset(&reply, 0, sizeof(reply));
2812 if (ret < 0) {
2813 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2814 } else {
2815 reply.ret_code = htobe32(LTTNG_OK);
2816 }
2817 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2818 sizeof(reply), 0);
2819 if (send_ret < sizeof(reply)) {
2820 ERR("Failed to send \"rotate rename\" command reply (ret = %zd)",
2821 send_ret);
2822 ret = -1;
2823 }
2824
2825 end_no_reply:
2826 free(complete_old_path);
2827 free(complete_new_path);
2828 return ret;
2829 }
2830
2831 /*
2832 * Check if all the streams in the session have completed the last rotation.
2833 * The chunk_id value is used to distinguish the cases where a stream was
2834 * closed on the consumerd before the rotation started but it still active on
2835 * the relayd, and the case where a stream appeared on the consumerd/relayd
2836 * after the last rotation started (in that case, it is already writing in the
2837 * new chunk folder).
2838 */
2839 static
2840 int relay_rotate_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2841 struct relay_connection *conn,
2842 const struct lttng_buffer_view *payload)
2843 {
2844 struct relay_session *session = conn->session;
2845 struct lttcomm_relayd_rotate_pending msg;
2846 struct lttcomm_relayd_rotate_pending_reply reply;
2847 struct lttng_ht_iter iter;
2848 struct relay_stream *stream;
2849 int ret = 0;
2850 ssize_t send_ret;
2851 uint64_t chunk_id;
2852 bool rotate_pending = false;
2853
2854 DBG("Rotate pending command received");
2855
2856 if (!session || !conn->version_check_done) {
2857 ERR("Trying to check for data before version check");
2858 ret = -1;
2859 goto end_no_reply;
2860 }
2861
2862 if (session->major == 2 && session->minor < 11) {
2863 ERR("Unsupported feature before 2.11");
2864 ret = -1;
2865 goto end_no_reply;
2866 }
2867
2868 if (payload->size < sizeof(msg)) {
2869 ERR("Unexpected payload size in \"relay_rotate_pending\": expected >= %zu bytes, got %zu bytes",
2870 sizeof(msg), payload->size);
2871 ret = -1;
2872 goto end_no_reply;
2873 }
2874
2875 memcpy(&msg, payload->data, sizeof(msg));
2876
2877 chunk_id = be64toh(msg.chunk_id);
2878
2879 DBG("Evaluating rotate pending for session \"%s\" and chunk id %" PRIu64,
2880 session->session_name, chunk_id);
2881
2882 /*
2883 * Iterate over all the streams in the session and check if they are
2884 * still waiting for data to perform their rotation.
2885 */
2886 rcu_read_lock();
2887 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2888 node.node) {
2889 if (!stream_get(stream)) {
2890 continue;
2891 }
2892 if (stream->trace->session != session) {
2893 stream_put(stream);
2894 continue;
2895 }
2896 pthread_mutex_lock(&stream->lock);
2897 if (stream->rotate_at_seq_num != -1ULL) {
2898 /* We have not yet performed the rotation. */
2899 rotate_pending = true;
2900 DBG("Stream %" PRIu64 " is still rotating",
2901 stream->stream_handle);
2902 } else if (stream->current_chunk_id.value < chunk_id) {
2903 /*
2904 * Stream closed on the consumer but still active on the
2905 * relay.
2906 */
2907 rotate_pending = true;
2908 DBG("Stream %" PRIu64 " did not exist on the consumer "
2909 "when the last rotation started, but is"
2910 "still waiting for data before getting"
2911 "closed",
2912 stream->stream_handle);
2913 }
2914 pthread_mutex_unlock(&stream->lock);
2915 stream_put(stream);
2916 if (rotate_pending) {
2917 goto send_reply;
2918 }
2919 }
2920
2921 send_reply:
2922 rcu_read_unlock();
2923 memset(&reply, 0, sizeof(reply));
2924 reply.generic.ret_code = htobe32((uint32_t) LTTNG_OK);
2925 reply.is_pending = (uint8_t) !!rotate_pending;
2926 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2927 sizeof(reply), 0);
2928 if (send_ret < (ssize_t) sizeof(reply)) {
2929 ERR("Failed to send \"rotate pending\" command reply (ret = %zd)",
2930 send_ret);
2931 ret = -1;
2932 }
2933
2934 end_no_reply:
2935 return ret;
2936 }
2937
2938 #define DBG_CMD(cmd_name, conn) \
2939 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
2940
2941 static int relay_process_control_command(struct relay_connection *conn,
2942 const struct lttcomm_relayd_hdr *header,
2943 const struct lttng_buffer_view *payload)
2944 {
2945 int ret = 0;
2946
2947 switch (header->cmd) {
2948 case RELAYD_CREATE_SESSION:
2949 DBG_CMD("RELAYD_CREATE_SESSION", conn);
2950 ret = relay_create_session(header, conn, payload);
2951 break;
2952 case RELAYD_ADD_STREAM:
2953 DBG_CMD("RELAYD_ADD_STREAM", conn);
2954 ret = relay_add_stream(header, conn, payload);
2955 break;
2956 case RELAYD_START_DATA:
2957 DBG_CMD("RELAYD_START_DATA", conn);
2958 ret = relay_start(header, conn, payload);
2959 break;
2960 case RELAYD_SEND_METADATA:
2961 DBG_CMD("RELAYD_SEND_METADATA", conn);
2962 ret = relay_recv_metadata(header, conn, payload);
2963 break;
2964 case RELAYD_VERSION:
2965 DBG_CMD("RELAYD_VERSION", conn);
2966 ret = relay_send_version(header, conn, payload);
2967 break;
2968 case RELAYD_CLOSE_STREAM:
2969 DBG_CMD("RELAYD_CLOSE_STREAM", conn);
2970 ret = relay_close_stream(header, conn, payload);
2971 break;
2972 case RELAYD_DATA_PENDING:
2973 DBG_CMD("RELAYD_DATA_PENDING", conn);
2974 ret = relay_data_pending(header, conn, payload);
2975 break;
2976 case RELAYD_QUIESCENT_CONTROL:
2977 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn);
2978 ret = relay_quiescent_control(header, conn, payload);
2979 break;
2980 case RELAYD_BEGIN_DATA_PENDING:
2981 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn);
2982 ret = relay_begin_data_pending(header, conn, payload);
2983 break;
2984 case RELAYD_END_DATA_PENDING:
2985 DBG_CMD("RELAYD_END_DATA_PENDING", conn);
2986 ret = relay_end_data_pending(header, conn, payload);
2987 break;
2988 case RELAYD_SEND_INDEX:
2989 DBG_CMD("RELAYD_SEND_INDEX", conn);
2990 ret = relay_recv_index(header, conn, payload);
2991 break;
2992 case RELAYD_STREAMS_SENT:
2993 DBG_CMD("RELAYD_STREAMS_SENT", conn);
2994 ret = relay_streams_sent(header, conn, payload);
2995 break;
2996 case RELAYD_RESET_METADATA:
2997 DBG_CMD("RELAYD_RESET_METADATA", conn);
2998 ret = relay_reset_metadata(header, conn, payload);
2999 break;
3000 case RELAYD_ROTATE_STREAM:
3001 DBG_CMD("RELAYD_ROTATE_STREAM", conn);
3002 ret = relay_rotate_session_stream(header, conn, payload);
3003 break;
3004 case RELAYD_ROTATE_RENAME:
3005 DBG_CMD("RELAYD_ROTATE_RENAME", conn);
3006 ret = relay_rotate_rename(header, conn, payload);
3007 break;
3008 case RELAYD_ROTATE_PENDING:
3009 DBG_CMD("RELAYD_ROTATE_PENDING", conn);
3010 ret = relay_rotate_pending(header, conn, payload);
3011 break;
3012 case RELAYD_MKDIR:
3013 DBG_CMD("RELAYD_MKDIR", conn);
3014 ret = relay_mkdir(header, conn, payload);
3015 break;
3016 case RELAYD_UPDATE_SYNC_INFO:
3017 default:
3018 ERR("Received unknown command (%u)", header->cmd);
3019 relay_unknown_command(conn);
3020 ret = -1;
3021 goto end;
3022 }
3023
3024 end:
3025 return ret;
3026 }
3027
3028 static enum relay_connection_status relay_process_control_receive_payload(
3029 struct relay_connection *conn)
3030 {
3031 int ret = 0;
3032 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3033 struct lttng_dynamic_buffer *reception_buffer =
3034 &conn->protocol.ctrl.reception_buffer;
3035 struct ctrl_connection_state_receive_payload *state =
3036 &conn->protocol.ctrl.state.receive_payload;
3037 struct lttng_buffer_view payload_view;
3038
3039 if (state->left_to_receive == 0) {
3040 /* Short-circuit for payload-less commands. */
3041 goto reception_complete;
3042 }
3043
3044 ret = conn->sock->ops->recvmsg(conn->sock,
3045 reception_buffer->data + state->received,
3046 state->left_to_receive, MSG_DONTWAIT);
3047 if (ret < 0) {
3048 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3049 PERROR("Unable to receive command payload on sock %d",
3050 conn->sock->fd);
3051 status = RELAY_CONNECTION_STATUS_ERROR;
3052 }
3053 goto end;
3054 } else if (ret == 0) {
3055 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3056 status = RELAY_CONNECTION_STATUS_CLOSED;
3057 goto end;
3058 }
3059
3060 assert(ret > 0);
3061 assert(ret <= state->left_to_receive);
3062
3063 state->left_to_receive -= ret;
3064 state->received += ret;
3065
3066 if (state->left_to_receive > 0) {
3067 /*
3068 * Can't transition to the protocol's next state, wait to
3069 * receive the rest of the header.
3070 */
3071 DBG3("Partial reception of control connection protocol payload (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3072 state->received, state->left_to_receive,
3073 conn->sock->fd);
3074 goto end;
3075 }
3076
3077 reception_complete:
3078 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3079 conn->sock->fd, state->received);
3080 /*
3081 * The payload required to process the command has been received.
3082 * A view to the reception buffer is forwarded to the various
3083 * commands and the state of the control is reset on success.
3084 *
3085 * Commands are responsible for sending their reply to the peer.
3086 */
3087 payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer,
3088 0, -1);
3089 ret = relay_process_control_command(conn,
3090 &state->header, &payload_view);
3091 if (ret < 0) {
3092 status = RELAY_CONNECTION_STATUS_ERROR;
3093 goto end;
3094 }
3095
3096 ret = connection_reset_protocol_state(conn);
3097 if (ret) {
3098 status = RELAY_CONNECTION_STATUS_ERROR;
3099 }
3100 end:
3101 return status;
3102 }
3103
3104 static enum relay_connection_status relay_process_control_receive_header(
3105 struct relay_connection *conn)
3106 {
3107 int ret = 0;
3108 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3109 struct lttcomm_relayd_hdr header;
3110 struct lttng_dynamic_buffer *reception_buffer =
3111 &conn->protocol.ctrl.reception_buffer;
3112 struct ctrl_connection_state_receive_header *state =
3113 &conn->protocol.ctrl.state.receive_header;
3114
3115 assert(state->left_to_receive != 0);
3116
3117 ret = conn->sock->ops->recvmsg(conn->sock,
3118 reception_buffer->data + state->received,
3119 state->left_to_receive, MSG_DONTWAIT);
3120 if (ret < 0) {
3121 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3122 PERROR("Unable to receive control command header on sock %d",
3123 conn->sock->fd);
3124 status = RELAY_CONNECTION_STATUS_ERROR;
3125 }
3126 goto end;
3127 } else if (ret == 0) {
3128 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3129 status = RELAY_CONNECTION_STATUS_CLOSED;
3130 goto end;
3131 }
3132
3133 assert(ret > 0);
3134 assert(ret <= state->left_to_receive);
3135
3136 state->left_to_receive -= ret;
3137 state->received += ret;
3138
3139 if (state->left_to_receive > 0) {
3140 /*
3141 * Can't transition to the protocol's next state, wait to
3142 * receive the rest of the header.
3143 */
3144 DBG3("Partial reception of control connection protocol header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3145 state->received, state->left_to_receive,
3146 conn->sock->fd);
3147 goto end;
3148 }
3149
3150 /* Transition to next state: receiving the command's payload. */
3151 conn->protocol.ctrl.state_id =
3152 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3153 memcpy(&header, reception_buffer->data, sizeof(header));
3154 header.circuit_id = be64toh(header.circuit_id);
3155 header.data_size = be64toh(header.data_size);
3156 header.cmd = be32toh(header.cmd);
3157 header.cmd_version = be32toh(header.cmd_version);
3158 memcpy(&conn->protocol.ctrl.state.receive_payload.header,
3159 &header, sizeof(header));
3160
3161 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32 ", cmd_version = %" PRIu32 ", payload size = %" PRIu64 " bytes",
3162 conn->sock->fd, header.cmd, header.cmd_version,
3163 header.data_size);
3164
3165 if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
3166 ERR("Command header indicates a payload (%" PRIu64 " bytes) that exceeds the maximal payload size allowed on a control connection.",
3167 header.data_size);
3168 status = RELAY_CONNECTION_STATUS_ERROR;
3169 goto end;
3170 }
3171
3172 conn->protocol.ctrl.state.receive_payload.left_to_receive =
3173 header.data_size;
3174 conn->protocol.ctrl.state.receive_payload.received = 0;
3175 ret = lttng_dynamic_buffer_set_size(reception_buffer,
3176 header.data_size);
3177 if (ret) {
3178 status = RELAY_CONNECTION_STATUS_ERROR;
3179 goto end;
3180 }
3181