cf6e01ee0ffd1409269ffc9cfc0f2dc662261d04
[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 const char *stream_path)
1544 {
1545 int ret;
1546 uint32_t major, minor;
1547
1548 /* Put ref on previous index_file. */
1549 if (stream->index_file) {
1550 lttng_index_file_put(stream->index_file);
1551 stream->index_file = NULL;
1552 }
1553 major = stream->trace->session->major;
1554 minor = stream->trace->session->minor;
1555 stream->index_file = lttng_index_file_create(stream_path,
1556 stream->channel_name,
1557 -1, -1, stream->tracefile_size,
1558 tracefile_array_get_file_index_head(stream->tfa),
1559 lttng_to_index_major(major, minor),
1560 lttng_to_index_minor(major, minor));
1561 if (!stream->index_file) {
1562 ret = -1;
1563 goto end;
1564 }
1565
1566 ret = 0;
1567
1568 end:
1569 return ret;
1570 }
1571
1572 static
1573 int do_rotate_stream(struct relay_stream *stream)
1574 {
1575 int ret;
1576
1577 /* Perform the stream rotation. */
1578 ret = utils_rotate_stream_file(stream->path_name,
1579 stream->channel_name, stream->tracefile_size,
1580 stream->tracefile_count, -1,
1581 -1, stream->stream_fd->fd,
1582 NULL, &stream->stream_fd->fd);
1583 if (ret < 0) {
1584 ERR("Rotating stream output file");
1585 goto end;
1586 }
1587 stream->tracefile_size_current = 0;
1588
1589 /* Rotate also the index if the stream is not a metadata stream. */
1590 if (!stream->is_metadata) {
1591 ret = create_rotate_index_file(stream, stream->path_name);
1592 if (ret < 0) {
1593 ERR("Failed to rotate index file");
1594 goto end;
1595 }
1596 }
1597
1598 stream->rotate_at_seq_num = -1ULL;
1599 stream->pos_after_last_complete_data_index = 0;
1600
1601 end:
1602 return ret;
1603 }
1604
1605 /*
1606 * If too much data has been written in a tracefile before we received the
1607 * rotation command, we have to move the excess data to the new tracefile and
1608 * perform the rotation. This can happen because the control and data
1609 * connections are separate, the indexes as well as the commands arrive from
1610 * the control connection and we have no control over the order so we could be
1611 * in a situation where too much data has been received on the data connection
1612 * before the rotation command on the control connection arrives. We don't need
1613 * to update the index because its order is guaranteed with the rotation
1614 * command message.
1615 */
1616 static
1617 int rotate_truncate_stream(struct relay_stream *stream)
1618 {
1619 int ret, new_fd;
1620 off_t lseek_ret;
1621 uint64_t diff, pos = 0;
1622 char buf[FILE_COPY_BUFFER_SIZE];
1623
1624 assert(!stream->is_metadata);
1625
1626 assert(stream->tracefile_size_current >
1627 stream->pos_after_last_complete_data_index);
1628 diff = stream->tracefile_size_current -
1629 stream->pos_after_last_complete_data_index;
1630
1631 /* Create the new tracefile. */
1632 new_fd = utils_create_stream_file(stream->path_name,
1633 stream->channel_name,
1634 stream->tracefile_size, stream->tracefile_count,
1635 /* uid */ -1, /* gid */ -1, /* suffix */ NULL);
1636 if (new_fd < 0) {
1637 ERR("Failed to create new stream file at path %s for channel %s",
1638 stream->path_name, stream->channel_name);
1639 ret = -1;
1640 goto end;
1641 }
1642
1643 /*
1644 * Rewind the current tracefile to the position at which the rotation
1645 * should have occured.
1646 */
1647 lseek_ret = lseek(stream->stream_fd->fd,
1648 stream->pos_after_last_complete_data_index, SEEK_SET);
1649 if (lseek_ret < 0) {
1650 PERROR("seek truncate stream");
1651 ret = -1;
1652 goto end;
1653 }
1654
1655 /* Move data from the old file to the new file. */
1656 while (pos < diff) {
1657 uint64_t count, bytes_left;
1658 ssize_t io_ret;
1659
1660 bytes_left = diff - pos;
1661 count = bytes_left > sizeof(buf) ? sizeof(buf) : bytes_left;
1662 assert(count <= SIZE_MAX);
1663
1664 io_ret = lttng_read(stream->stream_fd->fd, buf, count);
1665 if (io_ret < (ssize_t) count) {
1666 char error_string[256];
1667
1668 snprintf(error_string, sizeof(error_string),
1669 "Failed to read %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi",
1670 count, stream->stream_fd->fd, io_ret);
1671 if (io_ret == -1) {
1672 PERROR("%s", error_string);
1673 } else {
1674 ERR("%s", error_string);
1675 }
1676 ret = -1;
1677 goto end;
1678 }
1679
1680 io_ret = lttng_write(new_fd, buf, count);
1681 if (io_ret < (ssize_t) count) {
1682 char error_string[256];
1683
1684 snprintf(error_string, sizeof(error_string),
1685 "Failed to write %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi",
1686 count, new_fd, io_ret);
1687 if (io_ret == -1) {
1688 PERROR("%s", error_string);
1689 } else {
1690 ERR("%s", error_string);
1691 }
1692 ret = -1;
1693 goto end;
1694 }
1695
1696 pos += count;
1697 }
1698
1699 /* Truncate the file to get rid of the excess data. */
1700 ret = ftruncate(stream->stream_fd->fd,
1701 stream->pos_after_last_complete_data_index);
1702 if (ret) {
1703 PERROR("ftruncate");
1704 goto end;
1705 }
1706
1707 ret = close(stream->stream_fd->fd);
1708 if (ret < 0) {
1709 PERROR("Closing tracefile");
1710 goto end;
1711 }
1712
1713 ret = create_rotate_index_file(stream, stream->path_name);
1714 if (ret < 0) {
1715 ERR("Rotate stream index file");
1716 goto end;
1717 }
1718
1719 /*
1720 * Update the offset and FD of all the eventual indexes created by the
1721 * data connection before the rotation command arrived.
1722 */
1723 ret = relay_index_switch_all_files(stream);
1724 if (ret < 0) {
1725 ERR("Failed to rotate index file");
1726 goto end;
1727 }
1728
1729 stream->stream_fd->fd = new_fd;
1730 stream->tracefile_size_current = diff;
1731 stream->pos_after_last_complete_data_index = 0;
1732 stream->rotate_at_seq_num = -1ULL;
1733
1734 ret = 0;
1735
1736 end:
1737 return ret;
1738 }
1739
1740 /*
1741 * Check if a stream should perform a rotation (for session rotation).
1742 * Must be called with the stream lock held.
1743 *
1744 * Return 0 on success, a negative value on error.
1745 */
1746 static
1747 int try_rotate_stream(struct relay_stream *stream)
1748 {
1749 int ret = 0;
1750 uint64_t trace_seq;
1751
1752 /* No rotation expected. */
1753 if (stream->rotate_at_seq_num == -1ULL) {
1754 goto end;
1755 }
1756
1757 trace_seq = min(stream->prev_data_seq, stream->prev_index_seq);
1758 if (stream->prev_data_seq == -1ULL || stream->prev_index_seq == -1ULL ||
1759 trace_seq < stream->rotate_at_seq_num) {
1760 DBG("Stream %" PRIu64 " not yet ready for rotation (rotate_at_seq_num = %" PRIu64 ", prev_data_seq = %" PRIu64 ", prev_index_seq = %" PRIu64 ")",
1761 stream->stream_handle,
1762 stream->rotate_at_seq_num,
1763 stream->prev_data_seq,
1764 stream->prev_index_seq);
1765 goto end;
1766 } else if (stream->prev_data_seq > stream->rotate_at_seq_num) {
1767 /*
1768 * prev_data_seq is checked here since indexes and rotation
1769 * commands are serialized with respect to each other.
1770 */
1771 DBG("Rotation after too much data has been written in tracefile "
1772 "for stream %" PRIu64 ", need to truncate before "
1773 "rotating", stream->stream_handle);
1774 ret = rotate_truncate_stream(stream);
1775 if (ret) {
1776 ERR("Failed to truncate stream");
1777 goto end;
1778 }
1779 } else {
1780 if (trace_seq != stream->rotate_at_seq_num) {
1781 /*
1782 * Unexpected, protocol error/bug.
1783 * It could mean that we received a rotation position
1784 * that is in the past.
1785 */
1786 ERR("Stream %" PRIu64 " is in an inconsistent state (rotate_at_seq_num = %" PRIu64 ", prev_data_seq = %" PRIu64 ", prev_index_seq = %" PRIu64 ")",
1787 stream->stream_handle,
1788 stream->rotate_at_seq_num,
1789 stream->prev_data_seq,
1790 stream->prev_index_seq);
1791 ret = -1;
1792 goto end;
1793 }
1794 DBG("Stream %" PRIu64 " ready for rotation",
1795 stream->stream_handle);
1796 ret = do_rotate_stream(stream);
1797 }
1798
1799 end:
1800 return ret;
1801 }
1802
1803 /*
1804 * relay_recv_metadata: receive the metadata for the session.
1805 */
1806 static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1807 struct relay_connection *conn,
1808 const struct lttng_buffer_view *payload)
1809 {
1810 int ret = 0;
1811 ssize_t size_ret;
1812 struct relay_session *session = conn->session;
1813 struct lttcomm_relayd_metadata_payload metadata_payload_header;
1814 struct relay_stream *metadata_stream;
1815 uint64_t metadata_payload_size;
1816
1817 if (!session) {
1818 ERR("Metadata sent before version check");
1819 ret = -1;
1820 goto end;
1821 }
1822
1823 if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1824 ERR("Incorrect data size");
1825 ret = -1;
1826 goto end;
1827 }
1828 metadata_payload_size = recv_hdr->data_size -
1829 sizeof(struct lttcomm_relayd_metadata_payload);
1830
1831 memcpy(&metadata_payload_header, payload->data,
1832 sizeof(metadata_payload_header));
1833 metadata_payload_header.stream_id = be64toh(
1834 metadata_payload_header.stream_id);
1835 metadata_payload_header.padding_size = be32toh(
1836 metadata_payload_header.padding_size);
1837
1838 metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
1839 if (!metadata_stream) {
1840 ret = -1;
1841 goto end;
1842 }
1843
1844 pthread_mutex_lock(&metadata_stream->lock);
1845
1846 size_ret = lttng_write(metadata_stream->stream_fd->fd,
1847 payload->data + sizeof(metadata_payload_header),
1848 metadata_payload_size);
1849 if (size_ret < metadata_payload_size) {
1850 ERR("Relay error writing metadata on file");
1851 ret = -1;
1852 goto end_put;
1853 }
1854
1855 size_ret = write_padding_to_file(metadata_stream->stream_fd->fd,
1856 metadata_payload_header.padding_size);
1857 if (size_ret < (int64_t) metadata_payload_header.padding_size) {
1858 ret = -1;
1859 goto end_put;
1860 }
1861
1862 metadata_stream->metadata_received +=
1863 metadata_payload_size + metadata_payload_header.padding_size;
1864 DBG2("Relay metadata written. Updated metadata_received %" PRIu64,
1865 metadata_stream->metadata_received);
1866
1867 ret = try_rotate_stream(metadata_stream);
1868 if (ret < 0) {
1869 goto end_put;
1870 }
1871
1872 end_put:
1873 pthread_mutex_unlock(&metadata_stream->lock);
1874 stream_put(metadata_stream);
1875 end:
1876 return ret;
1877 }
1878
1879 /*
1880 * relay_send_version: send relayd version number
1881 */
1882 static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr,
1883 struct relay_connection *conn,
1884 const struct lttng_buffer_view *payload)
1885 {
1886 int ret;
1887 ssize_t send_ret;
1888 struct lttcomm_relayd_version reply, msg;
1889 bool compatible = true;
1890
1891 conn->version_check_done = true;
1892
1893 /* Get version from the other side. */
1894 if (payload->size < sizeof(msg)) {
1895 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1896 sizeof(msg), payload->size);
1897 ret = -1;
1898 goto end;
1899 }
1900
1901 memcpy(&msg, payload->data, sizeof(msg));
1902 msg.major = be32toh(msg.major);
1903 msg.minor = be32toh(msg.minor);
1904
1905 memset(&reply, 0, sizeof(reply));
1906 reply.major = RELAYD_VERSION_COMM_MAJOR;
1907 reply.minor = RELAYD_VERSION_COMM_MINOR;
1908
1909 /* Major versions must be the same */
1910 if (reply.major != msg.major) {
1911 DBG("Incompatible major versions (%u vs %u), deleting session",
1912 reply.major, msg.major);
1913 compatible = false;
1914 }
1915
1916 conn->major = reply.major;
1917 /* We adapt to the lowest compatible version */
1918 if (reply.minor <= msg.minor) {
1919 conn->minor = reply.minor;
1920 } else {
1921 conn->minor = msg.minor;
1922 }
1923
1924 reply.major = htobe32(reply.major);
1925 reply.minor = htobe32(reply.minor);
1926 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1927 sizeof(reply), 0);
1928 if (send_ret < (ssize_t) sizeof(reply)) {
1929 ERR("Failed to send \"send version\" command reply (ret = %zd)",
1930 send_ret);
1931 ret = -1;
1932 goto end;
1933 } else {
1934 ret = 0;
1935 }
1936
1937 if (!compatible) {
1938 ret = -1;
1939 goto end;
1940 }
1941
1942 DBG("Version check done using protocol %u.%u", conn->major,
1943 conn->minor);
1944
1945 end:
1946 return ret;
1947 }
1948
1949 /*
1950 * Check for data pending for a given stream id from the session daemon.
1951 */
1952 static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
1953 struct relay_connection *conn,
1954 const struct lttng_buffer_view *payload)
1955 {
1956 struct relay_session *session = conn->session;
1957 struct lttcomm_relayd_data_pending msg;
1958 struct lttcomm_relayd_generic_reply reply;
1959 struct relay_stream *stream;
1960 ssize_t send_ret;
1961 int ret;
1962 uint64_t stream_seq;
1963
1964 DBG("Data pending command received");
1965
1966 if (!session || !conn->version_check_done) {
1967 ERR("Trying to check for data before version check");
1968 ret = -1;
1969 goto end_no_session;
1970 }
1971
1972 if (payload->size < sizeof(msg)) {
1973 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
1974 sizeof(msg), payload->size);
1975 ret = -1;
1976 goto end_no_session;
1977 }
1978 memcpy(&msg, payload->data, sizeof(msg));
1979 msg.stream_id = be64toh(msg.stream_id);
1980 msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
1981
1982 stream = stream_get_by_id(msg.stream_id);
1983 if (stream == NULL) {
1984 ret = -1;
1985 goto end;
1986 }
1987
1988 pthread_mutex_lock(&stream->lock);
1989
1990 if (session_streams_have_index(session)) {
1991 /*
1992 * Ensure that both the index and stream data have been
1993 * flushed up to the requested point.
1994 */
1995 stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
1996 } else {
1997 stream_seq = stream->prev_data_seq;
1998 }
1999 DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
2000 ", prev_index_seq %" PRIu64
2001 ", and last_seq %" PRIu64, msg.stream_id,
2002 stream->prev_data_seq, stream->prev_index_seq,
2003 msg.last_net_seq_num);
2004
2005 /* Avoid wrapping issue */
2006 if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
2007 /* Data has in fact been written and is NOT pending */
2008 ret = 0;
2009 } else {
2010 /* Data still being streamed thus pending */
2011 ret = 1;
2012 }
2013
2014 stream->data_pending_check_done = true;
2015 pthread_mutex_unlock(&stream->lock);
2016
2017 stream_put(stream);
2018 end:
2019
2020 memset(&reply, 0, sizeof(reply));
2021 reply.ret_code = htobe32(ret);
2022 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2023 if (send_ret < (ssize_t) sizeof(reply)) {
2024 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2025 send_ret);
2026 ret = -1;
2027 }
2028
2029 end_no_session:
2030 return ret;
2031 }
2032
2033 /*
2034 * Wait for the control socket to reach a quiescent state.
2035 *
2036 * Note that for now, when receiving this command from the session
2037 * daemon, this means that every subsequent commands or data received on
2038 * the control socket has been handled. So, this is why we simply return
2039 * OK here.
2040 */
2041 static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr,
2042 struct relay_connection *conn,
2043 const struct lttng_buffer_view *payload)
2044 {
2045 int ret;
2046 ssize_t send_ret;
2047 struct relay_stream *stream;
2048 struct lttcomm_relayd_quiescent_control msg;
2049 struct lttcomm_relayd_generic_reply reply;
2050
2051 DBG("Checking quiescent state on control socket");
2052
2053 if (!conn->session || !conn->version_check_done) {
2054 ERR("Trying to check for data before version check");
2055 ret = -1;
2056 goto end_no_session;
2057 }
2058
2059 if (payload->size < sizeof(msg)) {
2060 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2061 sizeof(msg), payload->size);
2062 ret = -1;
2063 goto end_no_session;
2064 }
2065 memcpy(&msg, payload->data, sizeof(msg));
2066 msg.stream_id = be64toh(msg.stream_id);
2067
2068 stream = stream_get_by_id(msg.stream_id);
2069 if (!stream) {
2070 goto reply;
2071 }
2072 pthread_mutex_lock(&stream->lock);
2073 stream->data_pending_check_done = true;
2074 pthread_mutex_unlock(&stream->lock);
2075
2076 DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
2077 stream_put(stream);
2078 reply:
2079 memset(&reply, 0, sizeof(reply));
2080 reply.ret_code = htobe32(LTTNG_OK);
2081 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2082 if (send_ret < (ssize_t) sizeof(reply)) {
2083 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2084 send_ret);
2085 ret = -1;
2086 } else {
2087 ret = 0;
2088 }
2089
2090 end_no_session:
2091 return ret;
2092 }
2093
2094 /*
2095 * Initialize a data pending command. This means that a consumer is about
2096 * to ask for data pending for each stream it holds. Simply iterate over
2097 * all streams of a session and set the data_pending_check_done flag.
2098 *
2099 * This command returns to the client a LTTNG_OK code.
2100 */
2101 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2102 struct relay_connection *conn,
2103 const struct lttng_buffer_view *payload)
2104 {
2105 int ret;
2106 ssize_t send_ret;
2107 struct lttng_ht_iter iter;
2108 struct lttcomm_relayd_begin_data_pending msg;
2109 struct lttcomm_relayd_generic_reply reply;
2110 struct relay_stream *stream;
2111
2112 assert(recv_hdr);
2113 assert(conn);
2114
2115 DBG("Init streams for data pending");
2116
2117 if (!conn->session || !conn->version_check_done) {
2118 ERR("Trying to check for data before version check");
2119 ret = -1;
2120 goto end_no_session;
2121 }
2122
2123 if (payload->size < sizeof(msg)) {
2124 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2125 sizeof(msg), payload->size);
2126 ret = -1;
2127 goto end_no_session;
2128 }
2129 memcpy(&msg, payload->data, sizeof(msg));
2130 msg.session_id = be64toh(msg.session_id);
2131
2132 /*
2133 * Iterate over all streams to set the begin data pending flag.
2134 * For now, the streams are indexed by stream handle so we have
2135 * to iterate over all streams to find the one associated with
2136 * the right session_id.
2137 */
2138 rcu_read_lock();
2139 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2140 node.node) {
2141 if (!stream_get(stream)) {
2142 continue;
2143 }
2144 if (stream->trace->session->id == msg.session_id) {
2145 pthread_mutex_lock(&stream->lock);
2146 stream->data_pending_check_done = false;
2147 pthread_mutex_unlock(&stream->lock);
2148 DBG("Set begin data pending flag to stream %" PRIu64,
2149 stream->stream_handle);
2150 }
2151 stream_put(stream);
2152 }
2153 rcu_read_unlock();
2154
2155 memset(&reply, 0, sizeof(reply));
2156 /* All good, send back reply. */
2157 reply.ret_code = htobe32(LTTNG_OK);
2158
2159 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2160 if (send_ret < (ssize_t) sizeof(reply)) {
2161 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2162 send_ret);
2163 ret = -1;
2164 } else {
2165 ret = 0;
2166 }
2167
2168 end_no_session:
2169 return ret;
2170 }
2171
2172 /*
2173 * End data pending command. This will check, for a given session id, if
2174 * each stream associated with it has its data_pending_check_done flag
2175 * set. If not, this means that the client lost track of the stream but
2176 * the data is still being streamed on our side. In this case, we inform
2177 * the client that data is in flight.
2178 *
2179 * Return to the client if there is data in flight or not with a ret_code.
2180 */
2181 static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2182 struct relay_connection *conn,
2183 const struct lttng_buffer_view *payload)
2184 {
2185 int ret;
2186 ssize_t send_ret;
2187 struct lttng_ht_iter iter;
2188 struct lttcomm_relayd_end_data_pending msg;
2189 struct lttcomm_relayd_generic_reply reply;
2190 struct relay_stream *stream;
2191 uint32_t is_data_inflight = 0;
2192
2193 DBG("End data pending command");
2194
2195 if (!conn->session || !conn->version_check_done) {
2196 ERR("Trying to check for data before version check");
2197 ret = -1;
2198 goto end_no_session;
2199 }
2200
2201 if (payload->size < sizeof(msg)) {
2202 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2203 sizeof(msg), payload->size);
2204 ret = -1;
2205 goto end_no_session;
2206 }
2207 memcpy(&msg, payload->data, sizeof(msg));
2208 msg.session_id = be64toh(msg.session_id);
2209
2210 /*
2211 * Iterate over all streams to see if the begin data pending
2212 * flag is set.
2213 */
2214 rcu_read_lock();
2215 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2216 node.node) {
2217 if (!stream_get(stream)) {
2218 continue;
2219 }
2220 if (stream->trace->session->id != msg.session_id) {
2221 stream_put(stream);
2222 continue;
2223 }
2224 pthread_mutex_lock(&stream->lock);
2225 if (!stream->data_pending_check_done) {
2226 uint64_t stream_seq;
2227
2228 if (session_streams_have_index(conn->session)) {
2229 /*
2230 * Ensure that both the index and stream data have been
2231 * flushed up to the requested point.
2232 */
2233 stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
2234 } else {
2235 stream_seq = stream->prev_data_seq;
2236 }
2237 if (!stream->closed || !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
2238 is_data_inflight = 1;
2239 DBG("Data is still in flight for stream %" PRIu64,
2240 stream->stream_handle);
2241 pthread_mutex_unlock(&stream->lock);
2242 stream_put(stream);
2243 break;
2244 }
2245 }
2246 pthread_mutex_unlock(&stream->lock);
2247 stream_put(stream);
2248 }
2249 rcu_read_unlock();
2250
2251 memset(&reply, 0, sizeof(reply));
2252 /* All good, send back reply. */
2253 reply.ret_code = htobe32(is_data_inflight);
2254
2255 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2256 if (send_ret < (ssize_t) sizeof(reply)) {
2257 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2258 send_ret);
2259 ret = -1;
2260 } else {
2261 ret = 0;
2262 }
2263
2264 end_no_session:
2265 return ret;
2266 }
2267
2268 /*
2269 * Receive an index for a specific stream.
2270 *
2271 * Return 0 on success else a negative value.
2272 */
2273 static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr,
2274 struct relay_connection *conn,
2275 const struct lttng_buffer_view *payload)
2276 {
2277 int ret;
2278 ssize_t send_ret;
2279 struct relay_session *session = conn->session;
2280 struct lttcomm_relayd_index index_info;
2281 struct relay_index *index;
2282 struct lttcomm_relayd_generic_reply reply;
2283 struct relay_stream *stream;
2284 size_t msg_len;
2285
2286 assert(conn);
2287
2288 DBG("Relay receiving index");
2289
2290 if (!session || !conn->version_check_done) {
2291 ERR("Trying to close a stream before version check");
2292 ret = -1;
2293 goto end_no_session;
2294 }
2295
2296 msg_len = lttcomm_relayd_index_len(
2297 lttng_to_index_major(conn->major, conn->minor),
2298 lttng_to_index_minor(conn->major, conn->minor));
2299 if (payload->size < msg_len) {
2300 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2301 msg_len, payload->size);
2302 ret = -1;
2303 goto end_no_session;
2304 }
2305 memcpy(&index_info, payload->data, msg_len);
2306 index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2307 index_info.net_seq_num = be64toh(index_info.net_seq_num);
2308 index_info.packet_size = be64toh(index_info.packet_size);
2309 index_info.content_size = be64toh(index_info.content_size);
2310 index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2311 index_info.timestamp_end = be64toh(index_info.timestamp_end);
2312 index_info.events_discarded = be64toh(index_info.events_discarded);
2313 index_info.stream_id = be64toh(index_info.stream_id);
2314
2315 if (conn->minor >= 8) {
2316 index_info.stream_instance_id =
2317 be64toh(index_info.stream_instance_id);
2318 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2319 }
2320
2321 stream = stream_get_by_id(index_info.relay_stream_id);
2322 if (!stream) {
2323 ERR("stream_get_by_id not found");
2324 ret = -1;
2325 goto end;
2326 }
2327 pthread_mutex_lock(&stream->lock);
2328
2329 /* Live beacon handling */
2330 if (index_info.packet_size == 0) {
2331 DBG("Received live beacon for stream %" PRIu64,
2332 stream->stream_handle);
2333
2334 /*
2335 * Only flag a stream inactive when it has already
2336 * received data and no indexes are in flight.
2337 */
2338 if (stream->index_received_seqcount > 0
2339 && stream->indexes_in_flight == 0) {
2340 stream->beacon_ts_end = index_info.timestamp_end;
2341 }
2342 ret = 0;
2343 goto end_stream_put;
2344 } else {
2345 stream->beacon_ts_end = -1ULL;
2346 }
2347
2348 if (stream->ctf_stream_id == -1ULL) {
2349 stream->ctf_stream_id = index_info.stream_id;
2350 }
2351 index = relay_index_get_by_id_or_create(stream, index_info.net_seq_num);
2352 if (!index) {
2353 ret = -1;
2354 ERR("relay_index_get_by_id_or_create index NULL");
2355 goto end_stream_put;
2356 }
2357 if (set_index_control_data(index, &index_info, conn)) {
2358 ERR("set_index_control_data error");
2359 relay_index_put(index);
2360 ret = -1;
2361 goto end_stream_put;
2362 }
2363 ret = relay_index_try_flush(index);
2364 if (ret == 0) {
2365 tracefile_array_commit_seq(stream->tfa);
2366 stream->index_received_seqcount++;
2367 stream->pos_after_last_complete_data_index += index->total_size;
2368 stream->prev_index_seq = index_info.net_seq_num;
2369 } else if (ret > 0) {
2370 /* no flush. */
2371 ret = 0;
2372 } else {
2373 /*
2374 * ret < 0
2375 *
2376 * relay_index_try_flush is responsible for the self-reference
2377 * put of the index object on error.
2378 */
2379 ERR("relay_index_try_flush error %d", ret);
2380 ret = -1;
2381 }
2382
2383 end_stream_put:
2384 pthread_mutex_unlock(&stream->lock);
2385 stream_put(stream);
2386
2387 end:
2388
2389 memset(&reply, 0, sizeof(reply));
2390 if (ret < 0) {
2391 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2392 } else {
2393 reply.ret_code = htobe32(LTTNG_OK);
2394 }
2395 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2396 if (send_ret < (ssize_t) sizeof(reply)) {
2397 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2398 ret = -1;
2399 }
2400
2401 end_no_session:
2402 return ret;
2403 }
2404
2405 /*
2406 * Receive the streams_sent message.
2407 *
2408 * Return 0 on success else a negative value.
2409 */
2410 static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr,
2411 struct relay_connection *conn,
2412 const struct lttng_buffer_view *payload)
2413 {
2414 int ret;
2415 ssize_t send_ret;
2416 struct lttcomm_relayd_generic_reply reply;
2417
2418 assert(conn);
2419
2420 DBG("Relay receiving streams_sent");
2421
2422 if (!conn->session || !conn->version_check_done) {
2423 ERR("Trying to close a stream before version check");
2424 ret = -1;
2425 goto end_no_session;
2426 }
2427
2428 /*
2429 * Publish every pending stream in the connection recv list which are
2430 * now ready to be used by the viewer.
2431 */
2432 publish_connection_local_streams(conn);
2433
2434 memset(&reply, 0, sizeof(reply));
2435 reply.ret_code = htobe32(LTTNG_OK);
2436 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2437 if (send_ret < (ssize_t) sizeof(reply)) {
2438 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2439 send_ret);
2440 ret = -1;
2441 } else {
2442 /* Success. */
2443 ret = 0;
2444 }
2445
2446 end_no_session:
2447 return ret;
2448 }
2449
2450 /*
2451 * relay_rotate_session_stream: rotate a stream to a new tracefile for the session
2452 * rotation feature (not the tracefile rotation feature).
2453 */
2454 static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr *recv_hdr,
2455 struct relay_connection *conn,
2456 const struct lttng_buffer_view *payload)
2457 {
2458 int ret;
2459 ssize_t send_ret;
2460 struct relay_session *session = conn->session;
2461 struct lttcomm_relayd_rotate_stream stream_info;
2462 struct lttcomm_relayd_generic_reply reply;
2463 struct relay_stream *stream;
2464 size_t header_len;
2465 size_t path_len;
2466 struct lttng_buffer_view new_path_view;
2467
2468 DBG("Rotate stream received");
2469
2470 if (!session || !conn->version_check_done) {
2471 ERR("Trying to rotate a stream before version check");
2472 ret = -1;
2473 goto end_no_reply;
2474 }
2475
2476 if (session->major == 2 && session->minor < 11) {
2477 ERR("Unsupported feature before 2.11");
2478 ret = -1;
2479 goto end_no_reply;
2480 }
2481
2482 header_len = sizeof(struct lttcomm_relayd_rotate_stream);
2483
2484 if (payload->size < header_len) {
2485 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2486 header_len, payload->size);
2487 ret = -1;
2488 goto end_no_reply;
2489 }
2490
2491 memcpy(&stream_info, payload->data, header_len);
2492
2493 /* Convert to host */
2494 stream_info.pathname_length = be32toh(stream_info.pathname_length);
2495 stream_info.stream_id = be64toh(stream_info.stream_id);
2496 stream_info.new_chunk_id = be64toh(stream_info.new_chunk_id);
2497 stream_info.rotate_at_seq_num = be64toh(stream_info.rotate_at_seq_num);
2498
2499 path_len = stream_info.pathname_length;
2500 if (payload->size < header_len + path_len) {
2501 ERR("Unexpected payload size in \"relay_rotate_session_stream\" including path: expected >= %zu bytes, got %zu bytes",
2502 header_len + path_len, payload->size);
2503 ret = -1;
2504 goto end_no_reply;
2505 }
2506
2507 /* Ensure it fits in local filename length. */
2508 if (path_len >= LTTNG_PATH_MAX) {
2509 ret = -ENAMETOOLONG;
2510 ERR("Length of relay_rotate_session_stream command's path name (%zu bytes) exceeds the maximal allowed length of %i bytes",
2511 path_len, LTTNG_PATH_MAX);
2512 goto end;
2513 }
2514
2515 new_path_view = lttng_buffer_view_from_view(payload, header_len,
2516 stream_info.pathname_length);
2517
2518 stream = stream_get_by_id(stream_info.stream_id);
2519 if (!stream) {
2520 ret = -1;
2521 goto end;
2522 }
2523
2524 pthread_mutex_lock(&stream->lock);
2525
2526 /*
2527 * Update the trace path (just the folder, the stream name does not
2528 * change).
2529 */
2530 free(stream->prev_path_name);
2531 stream->prev_path_name = stream->path_name;
2532 stream->path_name = create_output_path(new_path_view.data);
2533 if (!stream->path_name) {
2534 ERR("Failed to create a new output path");
2535 ret = -1;
2536 goto end_stream_unlock;
2537 }
2538 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG,
2539 -1, -1);
2540 if (ret < 0) {
2541 ERR("relay creating output directory");
2542 ret = -1;
2543 goto end_stream_unlock;
2544 }
2545
2546 assert(stream->current_chunk_id.is_set);
2547 stream->current_chunk_id.value = stream_info.new_chunk_id;
2548
2549 if (stream->is_metadata) {
2550 /*
2551 * The metadata stream is sent only over the control connection
2552 * so we know we have all the data to perform the stream
2553 * rotation.
2554 */
2555 ret = do_rotate_stream(stream);
2556 } else {
2557 stream->rotate_at_seq_num = stream_info.rotate_at_seq_num;
2558 ret = try_rotate_stream(stream);
2559 }
2560 if (ret < 0) {
2561 goto end_stream_unlock;
2562 }
2563
2564 end_stream_unlock:
2565 pthread_mutex_unlock(&stream->lock);
2566 stream_put(stream);
2567 end:
2568 memset(&reply, 0, sizeof(reply));
2569 if (ret < 0) {
2570 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2571 } else {
2572 reply.ret_code = htobe32(LTTNG_OK);
2573 }
2574 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2575 sizeof(struct lttcomm_relayd_generic_reply), 0);
2576 if (send_ret < (ssize_t) sizeof(reply)) {
2577 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2578 send_ret);
2579 ret = -1;
2580 }
2581
2582 end_no_reply:
2583 return ret;
2584 }
2585
2586 /*
2587 * relay_mkdir: Create a folder on the disk.
2588 */
2589 static int relay_mkdir(const struct lttcomm_relayd_hdr *recv_hdr,
2590 struct relay_connection *conn,
2591 const struct lttng_buffer_view *payload)
2592 {
2593 int ret;
2594 struct relay_session *session = conn->session;
2595 struct lttcomm_relayd_mkdir path_info_header;
2596 struct lttcomm_relayd_generic_reply reply;
2597 char *path = NULL;
2598 size_t header_len;
2599 ssize_t send_ret;
2600 struct lttng_buffer_view path_view;
2601
2602 if (!session || !conn->version_check_done) {
2603 ERR("Trying to create a directory before version check");
2604 ret = -1;
2605 goto end_no_session;
2606 }
2607
2608 if (session->major == 2 && session->minor < 11) {
2609 /*
2610 * This client is not supposed to use this command since
2611 * it predates its introduction.
2612 */
2613 ERR("relay_mkdir command is unsupported before LTTng 2.11");
2614 ret = -1;
2615 goto end_no_session;
2616 }
2617
2618 header_len = sizeof(path_info_header);
2619 if (payload->size < header_len) {
2620 ERR("Unexpected payload size in \"relay_mkdir\": expected >= %zu bytes, got %zu bytes",
2621 header_len, payload->size);
2622 ret = -1;
2623 goto end_no_session;
2624 }
2625
2626 memcpy(&path_info_header, payload->data, header_len);
2627
2628 path_info_header.length = be32toh(path_info_header.length);
2629
2630 if (payload->size < header_len + path_info_header.length) {
2631 ERR("Unexpected payload size in \"relay_mkdir\" including path: expected >= %zu bytes, got %zu bytes",
2632 header_len + path_info_header.length, payload->size);
2633 ret = -1;
2634 goto end_no_session;
2635 }
2636
2637 /* Ensure that it fits in local path length. */
2638 if (path_info_header.length >= LTTNG_PATH_MAX) {
2639 ret = -ENAMETOOLONG;
2640 ERR("Path name argument of mkdir command (%" PRIu32 " bytes) exceeds the maximal length allowed (%d bytes)",
2641 path_info_header.length, LTTNG_PATH_MAX);
2642 goto end;
2643 }
2644
2645 path_view = lttng_buffer_view_from_view(payload, header_len,
2646 path_info_header.length);
2647
2648 path = create_output_path(path_view.data);
2649 if (!path) {
2650 ERR("Failed to create output path");
2651 ret = -1;
2652 goto end;
2653 }
2654
2655 DBG("MKDIR command has path \"%s\", changed to \"%s\"", path_view.data, path);
2656 ret = utils_mkdir_recursive(path, S_IRWXU | S_IRWXG, -1, -1);
2657 if (ret < 0) {
2658 ERR("relay creating output directory");
2659 goto end;
2660 }
2661
2662 ret = 0;
2663
2664 end:
2665 memset(&reply, 0, sizeof(reply));
2666 if (ret < 0) {
2667 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2668 } else {
2669 reply.ret_code = htobe32(LTTNG_OK);
2670 }
2671 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2672 if (send_ret < (ssize_t) sizeof(reply)) {
2673 ERR("Failed to send \"mkdir\" command reply (ret = %zd)", send_ret);
2674 ret = -1;
2675 }
2676
2677 end_no_session:
2678 free(path);
2679 return ret;
2680 }
2681
2682 static int validate_rotate_rename_path_length(const char *path_type,
2683 uint32_t path_length)
2684 {
2685 int ret = 0;
2686
2687 if (path_length > LTTNG_PATH_MAX) {
2688 ret = -ENAMETOOLONG;
2689 ERR("rotate rename \"%s\" path name length (%" PRIu32 " bytes) exceeds the allowed size of %i bytes",
2690 path_type, path_length, LTTNG_PATH_MAX);
2691 } else if (path_length == 0) {
2692 ret = -EINVAL;
2693 ERR("rotate rename \"%s\" path name has an illegal length of 0", path_type);
2694 }
2695 return ret;
2696 }
2697
2698 /*
2699 * relay_rotate_rename: rename the trace folder after a rotation is
2700 * completed. We are not closing any fd here, just moving the folder, so it
2701 * works even if data is still in-flight.
2702 */
2703 static int relay_rotate_rename(const struct lttcomm_relayd_hdr *recv_hdr,
2704 struct relay_connection *conn,
2705 const struct lttng_buffer_view *payload)
2706 {
2707 int ret;
2708 ssize_t send_ret;
2709 struct relay_session *session = conn->session;
2710 struct lttcomm_relayd_generic_reply reply;
2711 struct lttcomm_relayd_rotate_rename header;
2712 size_t header_len;
2713 size_t received_paths_size;
2714 char *complete_old_path = NULL, *complete_new_path = NULL;
2715 struct lttng_buffer_view old_path_view;
2716 struct lttng_buffer_view new_path_view;
2717
2718 if (!session || !conn->version_check_done) {
2719 ERR("Trying to rename a trace folder before version check");
2720 ret = -1;
2721 goto end_no_reply;
2722 }
2723
2724 if (session->major == 2 && session->minor < 11) {
2725 ERR("relay_rotate_rename command is unsupported before LTTng 2.11");
2726 ret = -1;
2727 goto end_no_reply;
2728 }
2729
2730 header_len = sizeof(header);
2731 if (payload->size < header_len) {
2732 ERR("Unexpected payload size in \"relay_rotate_rename\": expected >= %zu bytes, got %zu bytes",
2733 header_len, payload->size);
2734 ret = -1;
2735 goto end_no_reply;
2736 }
2737
2738 memcpy(&header, payload->data, header_len);
2739
2740 header.old_path_length = be32toh(header.old_path_length);
2741 header.new_path_length = be32toh(header.new_path_length);
2742 received_paths_size = header.old_path_length + header.new_path_length;
2743
2744 if (payload->size < header_len + received_paths_size) {
2745 ERR("Unexpected payload size in \"relay_rotate_rename\" including paths: expected >= %zu bytes, got %zu bytes",
2746 header_len, payload->size);
2747 ret = -1;
2748 goto end_no_reply;
2749 }
2750
2751 /* Ensure the paths don't exceed their allowed size. */
2752 ret = validate_rotate_rename_path_length("old", header.old_path_length);
2753 if (ret) {
2754 goto end;
2755 }
2756 ret = validate_rotate_rename_path_length("new", header.new_path_length);
2757 if (ret) {
2758 goto end;
2759 }
2760
2761 old_path_view = lttng_buffer_view_from_view(payload, header_len,
2762 header.old_path_length);
2763 new_path_view = lttng_buffer_view_from_view(payload,
2764 header_len + header.old_path_length,
2765 header.new_path_length);
2766
2767 /* Validate that both paths received are NULL terminated. */
2768 if (old_path_view.data[old_path_view.size - 1] != '\0') {
2769 ERR("relay_rotate_rename command's \"old\" path is invalid (not NULL terminated)");
2770 ret = -1;
2771 goto end;
2772 }
2773 if (new_path_view.data[new_path_view.size - 1] != '\0') {
2774 ERR("relay_rotate_rename command's \"new\" path is invalid (not NULL terminated)");
2775 ret = -1;
2776 goto end;
2777 }
2778
2779 DBG("ROTATE_RENAME command has argument old path = \"%s\", new_path = \"%s\"",
2780 old_path_view.data, new_path_view.data);
2781 complete_old_path = create_output_path(old_path_view.data);
2782 if (!complete_old_path) {
2783 ERR("Failed to build old output path in rotate_rename command");
2784 ret = -1;
2785 goto end;
2786 }
2787
2788 complete_new_path = create_output_path(new_path_view.data);
2789 if (!complete_new_path) {
2790 ERR("Failed to build new output path in rotate_rename command");
2791 ret = -1;
2792 goto end;
2793 }
2794 DBG("Expanded ROTATE_RENAME arguments to old path = \"%s\", new_path = \"%s\"",
2795 complete_old_path, complete_new_path);
2796
2797 ret = utils_mkdir_recursive(complete_new_path, S_IRWXU | S_IRWXG,
2798 -1, -1);
2799 if (ret < 0) {
2800 ERR("Failed to mkdir() rotate_rename's \"new\" output directory at \"%s\"",
2801 complete_new_path);
2802 goto end;
2803 }
2804
2805 /*
2806 * If a domain has not yet created its channel, the domain-specific
2807 * folder might not exist, but this is not an error.
2808 */
2809 ret = rename(complete_old_path, complete_new_path);
2810 if (ret < 0 && errno != ENOENT) {
2811 PERROR("Renaming chunk in rotate_rename command from \"%s\" to \"%s\"",
2812 complete_old_path, complete_new_path);
2813 goto end;
2814 }
2815 ret = 0;
2816
2817 end:
2818 memset(&reply, 0, sizeof(reply));
2819 if (ret < 0) {
2820 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2821 } else {
2822 reply.ret_code = htobe32(LTTNG_OK);
2823 }
2824 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2825 sizeof(reply), 0);
2826 if (send_ret < sizeof(reply)) {
2827 ERR("Failed to send \"rotate rename\" command reply (ret = %zd)",
2828 send_ret);
2829 ret = -1;
2830 }
2831
2832 end_no_reply:
2833 free(complete_old_path);
2834 free(complete_new_path);
2835 return ret;
2836 }
2837
2838 /*
2839 * Check if all the streams in the session have completed the last rotation.
2840 * The chunk_id value is used to distinguish the cases where a stream was
2841 * closed on the consumerd before the rotation started but it still active on
2842 * the relayd, and the case where a stream appeared on the consumerd/relayd
2843 * after the last rotation started (in that case, it is already writing in the
2844 * new chunk folder).
2845 */
2846 static
2847 int relay_rotate_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2848 struct relay_connection *conn,
2849 const struct lttng_buffer_view *payload)
2850 {
2851 struct relay_session *session = conn->session;
2852 struct lttcomm_relayd_rotate_pending msg;
2853 struct lttcomm_relayd_rotate_pending_reply reply;
2854 struct lttng_ht_iter iter;
2855 struct relay_stream *stream;
2856 int ret = 0;
2857 ssize_t send_ret;
2858 uint64_t chunk_id;
2859 bool rotate_pending = false;
2860
2861 DBG("Rotate pending command received");
2862
2863 if (!session || !conn->version_check_done) {
2864 ERR("Trying to check for data before version check");
2865 ret = -1;
2866 goto end_no_reply;
2867 }
2868
2869 if (session->major == 2 && session->minor < 11) {
2870 ERR("Unsupported feature before 2.11");
2871 ret = -1;
2872 goto end_no_reply;
2873 }
2874
2875 if (payload->size < sizeof(msg)) {
2876 ERR("Unexpected payload size in \"relay_rotate_pending\": expected >= %zu bytes, got %zu bytes",
2877 sizeof(msg), payload->size);
2878 ret = -1;
2879 goto end_no_reply;
2880 }
2881
2882 memcpy(&msg, payload->data, sizeof(msg));
2883
2884 chunk_id = be64toh(msg.chunk_id);
2885
2886 DBG("Evaluating rotate pending for session \"%s\" and chunk id %" PRIu64,
2887 session->session_name, chunk_id);
2888
2889 /*
2890 * Iterate over all the streams in the session and check if they are
2891 * still waiting for data to perform their rotation.
2892 */
2893 rcu_read_lock();
2894 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2895 node.node) {
2896 if (!stream_get(stream)) {
2897 continue;
2898 }
2899 if (stream->trace->session != session) {
2900 stream_put(stream);
2901 continue;
2902 }
2903 pthread_mutex_lock(&stream->lock);
2904 if (stream->rotate_at_seq_num != -1ULL) {
2905 /* We have not yet performed the rotation. */
2906 rotate_pending = true;
2907 DBG("Stream %" PRIu64 " is still rotating",
2908 stream->stream_handle);
2909 } else if (stream->current_chunk_id.value < chunk_id) {
2910 /*
2911 * Stream closed on the consumer but still active on the
2912 * relay.
2913 */
2914 rotate_pending = true;
2915 DBG("Stream %" PRIu64 " did not exist on the consumer "
2916 "when the last rotation started, but is"
2917 "still waiting for data before getting"
2918 "closed",
2919 stream->stream_handle);
2920 }
2921 pthread_mutex_unlock(&stream->lock);
2922 stream_put(stream);
2923 if (rotate_pending) {
2924 goto send_reply;
2925 }
2926 }
2927
2928 send_reply:
2929 rcu_read_unlock();
2930 memset(&reply, 0, sizeof(reply));
2931 reply.generic.ret_code = htobe32((uint32_t) LTTNG_OK);
2932 reply.is_pending = (uint8_t) !!rotate_pending;
2933 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2934 sizeof(reply), 0);
2935 if (send_ret < (ssize_t) sizeof(reply)) {
2936 ERR("Failed to send \"rotate pending\" command reply (ret = %zd)",
2937 send_ret);
2938 ret = -1;
2939 }
2940
2941 end_no_reply:
2942 return ret;
2943 }
2944
2945 #define DBG_CMD(cmd_name, conn) \
2946 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
2947
2948 static int relay_process_control_command(struct relay_connection *conn,
2949 const struct lttcomm_relayd_hdr *header,
2950 const struct lttng_buffer_view *payload)
2951 {
2952 int ret = 0;
2953
2954 switch (header->cmd) {
2955 case RELAYD_CREATE_SESSION:
2956 DBG_CMD("RELAYD_CREATE_SESSION", conn);
2957 ret = relay_create_session(header, conn, payload);
2958 break;
2959 case RELAYD_ADD_STREAM:
2960 DBG_CMD("RELAYD_ADD_STREAM", conn);
2961 ret = relay_add_stream(header, conn, payload);
2962 break;
2963 case RELAYD_START_DATA:
2964 DBG_CMD("RELAYD_START_DATA", conn);
2965 ret = relay_start(header, conn, payload);
2966 break;
2967 case RELAYD_SEND_METADATA:
2968 DBG_CMD("RELAYD_SEND_METADATA", conn);
2969 ret = relay_recv_metadata(header, conn, payload);
2970 break;
2971 case RELAYD_VERSION:
2972 DBG_CMD("RELAYD_VERSION", conn);
2973 ret = relay_send_version(header, conn, payload);
2974 break;
2975 case RELAYD_CLOSE_STREAM:
2976 DBG_CMD("RELAYD_CLOSE_STREAM", conn);
2977 ret = relay_close_stream(header, conn, payload);
2978 break;
2979 case RELAYD_DATA_PENDING:
2980 DBG_CMD("RELAYD_DATA_PENDING", conn);
2981 ret = relay_data_pending(header, conn, payload);
2982 break;
2983 case RELAYD_QUIESCENT_CONTROL:
2984 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn);
2985 ret = relay_quiescent_control(header, conn, payload);
2986 break;
2987 case RELAYD_BEGIN_DATA_PENDING:
2988 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn);
2989 ret = relay_begin_data_pending(header, conn, payload);
2990 break;
2991 case RELAYD_END_DATA_PENDING:
2992 DBG_CMD("RELAYD_END_DATA_PENDING", conn);
2993 ret = relay_end_data_pending(header, conn, payload);
2994 break;
2995 case RELAYD_SEND_INDEX:
2996 DBG_CMD("RELAYD_SEND_INDEX", conn);
2997 ret = relay_recv_index(header, conn, payload);
2998 break;
2999 case RELAYD_STREAMS_SENT:
3000 DBG_CMD("RELAYD_STREAMS_SENT", conn);
3001 ret = relay_streams_sent(header, conn, payload);
3002 break;
3003 case RELAYD_RESET_METADATA:
3004 DBG_CMD("RELAYD_RESET_METADATA", conn);
3005 ret = relay_reset_metadata(header, conn, payload);
3006 break;
3007 case RELAYD_ROTATE_STREAM:
3008 DBG_CMD("RELAYD_ROTATE_STREAM", conn);
3009 ret = relay_rotate_session_stream(header, conn, payload);
3010 break;
3011 case RELAYD_ROTATE_RENAME:
3012 DBG_CMD("RELAYD_ROTATE_RENAME", conn);
3013 ret = relay_rotate_rename(header, conn, payload);
3014 break;
3015 case RELAYD_ROTATE_PENDING:
3016 DBG_CMD("RELAYD_ROTATE_PENDING", conn);
3017 ret = relay_rotate_pending(header, conn, payload);
3018 break;
3019 case RELAYD_MKDIR:
3020 DBG_CMD("RELAYD_MKDIR", conn);
3021 ret = relay_mkdir(header, conn, payload);
3022 break;
3023 case RELAYD_UPDATE_SYNC_INFO:
3024 default:
3025 ERR("Received unknown command (%u)", header->cmd);
3026 relay_unknown_command(conn);
3027 ret = -1;
3028 goto end;
3029 }
3030
3031 end:
3032 return ret;
3033 }
3034
3035 static enum relay_connection_status relay_process_control_receive_payload(
3036 struct relay_connection *conn)
3037 {
3038 int ret = 0;
3039 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3040 struct lttng_dynamic_buffer *reception_buffer =
3041 &conn->protocol.ctrl.reception_buffer;
3042 struct ctrl_connection_state_receive_payload *state =
3043 &conn->protocol.ctrl.state.receive_payload;
3044 struct lttng_buffer_view payload_view;
3045
3046 if (state->left_to_receive == 0) {
3047 /* Short-circuit for payload-less commands. */
3048 goto reception_complete;
3049 }
3050
3051 ret = conn->sock->ops->recvmsg(conn->sock,
3052 reception_buffer->data + state->received,
3053 state->left_to_receive, MSG_DONTWAIT);
3054 if (ret < 0) {
3055 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3056 PERROR("Unable to receive command payload on sock %d",
3057 conn->sock->fd);
3058 status = RELAY_CONNECTION_STATUS_ERROR;
3059 }
3060 goto end;
3061 } else if (ret == 0) {
3062 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3063 status = RELAY_CONNECTION_STATUS_CLOSED;
3064 goto end;
3065 }
3066
3067 assert(ret > 0);
3068 assert(ret <= state->left_to_receive);
3069
3070 state->left_to_receive -= ret;
3071 state->received += ret;
3072
3073 if (state->left_to_receive > 0) {
3074 /*
3075 * Can't transition to the protocol's next state, wait to
3076 * receive the rest of the header.
3077 */
3078 DBG3("Partial reception of control connection protocol payload (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3079 state->received, state->left_to_receive,
3080 conn->sock->fd);
3081 goto end;
3082 }
3083
3084 reception_complete:
3085 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3086 conn->sock->fd, state->received);
3087 /*
3088 * The payload required to process the command has been received.
3089 * A view to the reception buffer is forwarded to the various
3090 * commands and the state of the control is reset on success.
3091 *
3092 * Commands are responsible for sending their reply to the peer.
3093 */
3094 payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer,
3095 0, -1);
3096 ret = relay_process_control_command(conn,
3097 &state->header, &payload_view);
3098 if (ret < 0) {
3099 status = RELAY_CONNECTION_STATUS_ERROR;
3100 goto end;
3101 }
3102
3103 ret = connection_reset_protocol_state(conn);
3104 if (ret) {
3105 status = RELAY_CONNECTION_STATUS_ERROR;
3106 }
3107 end:
3108 return status;
3109 }
3110
3111 static enum relay_connection_status relay_process_control_receive_header(
3112 struct relay_connection *conn)
3113 {
3114 int ret = 0;
3115 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3116 struct lttcomm_relayd_hdr header;
3117 struct lttng_dynamic_buffer *reception_buffer =
3118 &conn->protocol.ctrl.reception_buffer;
3119 struct ctrl_connection_state_receive_header *state =
3120 &conn->protocol.ctrl.state.receive_header;
3121
3122 assert(state->left_to_receive != 0);
3123
3124 ret = conn->sock->ops->recvmsg(conn->sock,
3125 reception_buffer->data + state->received,
3126 state->left_to_receive, MSG_DONTWAIT);
3127 if (ret < 0) {
3128 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3129 PERROR("Unable to receive control command header on sock %d",
3130 conn->sock->fd);
3131 status = RELAY_CONNECTION_STATUS_ERROR;
3132 }
3133 goto end;
3134 } else if (ret == 0) {
3135 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3136 status = RELAY_CONNECTION_STATUS_CLOSED;
3137 goto end;
3138 }
3139
3140 assert(ret > 0);
3141 assert(ret <= state->left_to_receive);
3142
3143 state->left_to_receive -= ret;
3144 state->received += ret;
3145
3146 if (state->left_to_receive > 0) {
3147 /*
3148 * Can't transition to the protocol's next state, wait to
3149 * receive the rest of the header.
3150 */
3151 DBG3("Partial reception of control connection protocol header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3152 state->received, state->left_to_receive,
3153 conn->sock->fd);
3154 goto end;
3155 }
3156
3157 /* Transition to next state: receiving the command's payload. */
3158 conn->protocol.ctrl.state_id =
3159 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3160 memcpy(&header, reception_buffer->data, sizeof(header));
3161 header.circuit_id = be64toh(header.circuit_id);
3162 header.data_size = be64toh(header.data_size);
3163 header.cmd = be32toh(header.cmd);
3164 header.cmd_version = be32toh(header.cmd_version);
3165 memcpy(&conn->protocol.ctrl.state.receive_payload.header,
3166 &header, sizeof(header));
3167
3168 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32 ", cmd_version = %" PRIu32 ", payload size = %" PRIu64 " bytes",
3169 conn->sock->fd, header.cmd, header.cmd_version,
3170 header.data_size);
3171
3172 if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
3173 ERR("Command header indicates a payload (%" PRIu64 " bytes) that exceeds the maximal payload size allowed on a control connection.",
3174 header.data_size);
3175 status = RELAY_CONNECTION_STATUS_ERROR;
3176 goto end;
3177 }
3178
3179 conn->protocol.ctrl.state.receive_payload.left_to_receive =
3180 header.data_size;
3181 conn->protocol.ctrl.state.receive_payload.received = 0;
3182 ret = lttng_dynamic_buffer_set_size(reception_buffer,
3183 header.data_size);
3184 if (ret) {
3185 status = RELAY_CONNECTION_STATUS_ERROR;
3186 goto end;
3187 }
3188
3189 if (header.data_size == 0) {
3190 /*
3191 * Manually invoke the next state as the poll loop
3192 * will not wake-up to allow us to proceed further.
3193 */
3194 status = relay_process_control_receive_payload(conn);
3195 }
3196 end:
3197 return status;
3198 }
3199
3200 /*
3201 * Process the commands received on the control socket
3202 */
3203 static enum relay_connection_status relay_process_control(
3204 struct relay_connection *conn)
3205 {
3206 enum relay_connection_status status;
3207
3208 switch (conn->protocol.ctrl.state_id) {
3209 case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
3210 status = relay_process_control_receive_header(conn);
3211 break;
3212 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
3213 status = relay_process_control_receive_payload(conn);
3214 break;
3215 default:
3216 ERR("Unknown control connection protocol state encountered.");
3217 abort();
3218 }
3219
3220 return status;
3221 }
3222
3223 /*
3224 * Handle index for a data stream.
3225 *
3226 * Called with the stream lock held.
3227 *
3228 * Return 0 on success else a negative value.
3229 */
3230 static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
3231 bool rotate_index, bool *flushed, uint64_t total_size)
3232 {
3233 int ret = 0;
3234 uint64_t data_offset;
3235 struct relay_index *index;
3236
3237 /* Get data offset because we are about to update the index. */
3238 data_offset = htobe64(stream->tracefile_size_current);
3239
3240 DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64,
3241 stream->stream_handle, net_seq_num, stream->tracefile_size_current);
3242
3243 /*
3244 * Lookup for an existing index for that stream id/sequence
3245 * number. If it exists, the control thread has already received the
3246 * data for it, thus we need to write it to disk.
3247 */
3248 index = relay_index_get_by_id_or_create(stream, net_seq_num);
3249 if (!index) {
3250 ret = -1;
3251 goto end;
3252 }
3253
3254 if (rotate_index || !stream->index_file) {
3255 const char *stream_path;
3256
3257 /*
3258 * The data connection creates the stream's first index file.
3259 *
3260 * This can happen _after_ a ROTATE_STREAM command. In
3261 * other words, the data of the first packet of this stream
3262 * can be received after a ROTATE_STREAM command.
3263 *
3264 * The ROTATE_STREAM command changes the stream's path_name
3265 * to point to the "next" chunk. If a rotation is pending for
3266 * this stream, as indicated by "rotate_at_seq_num != -1ULL",
3267 * it means that we are still receiving data that belongs in the
3268 * stream's former path.
3269 *
3270 * In this very specific case, we must ensure that the index
3271 * file is created in the streams's former path,
3272 * "prev_path_name".
3273 *
3274 * All other rotations beyond the first one are not affected
3275 * by this problem since the actual rotation operation creates
3276 * the new chunk's index file.
3277 */
3278 stream_path = stream->rotate_at_seq_num == -1ULL ?
3279 stream->path_name:
3280 stream->prev_path_name;
3281
3282 ret = create_rotate_index_file(stream, stream_path);
3283 if (ret < 0) {
3284 ERR("Failed to rotate index");
3285 /* Put self-ref for this index due to error. */
3286 relay_index_put(index);
3287 index = NULL;
3288 goto end;
3289 }
3290 }
3291
3292 if (relay_index_set_file(index, stream->index_file, data_offset)) {
3293 ret = -1;
3294 /* Put self-ref for this index due to error. */
3295 relay_index_put(index);
3296 index = NULL;
3297 goto end;
3298 }
3299
3300 ret = relay_index_try_flush(index);
3301 if (ret == 0) {
3302 tracefile_array_commit_seq(stream->tfa);
3303 stream->index_received_seqcount++;
3304 *flushed = true;
3305 } else if (ret > 0) {
3306 index->total_size = total_size;
3307 /* No flush. */
3308 ret = 0;
3309 } else {
3310 /*
3311 * ret < 0
3312 *
3313 * relay_index_try_flush is responsible for the self-reference
3314 * put of the index object on error.
3315 */
3316 ERR("relay_index_try_flush error %d", ret);
3317 ret = -1;
3318 }
3319 end:
3320 return ret;
3321 }
3322
3323 static enum relay_connection_status relay_process_data_receive_header(
3324 struct relay_connection *conn)
3325 {
3326 int ret;
3327 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3328 struct data_connection_state_receive_header *state =
3329 &conn->protocol.data.state.receive_header;
3330 struct lttcomm_relayd_data_hdr header;
3331 struct relay_stream *stream;
3332
3333 assert(state->left_to_receive != 0);
3334
3335 ret = conn->sock->ops->recvmsg(conn->sock,
3336 state->header_reception_buffer + state->received,
3337 state->left_to_receive, MSG_DONTWAIT);
3338 if (ret < 0) {
3339 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3340 PERROR("Unable to receive data header on sock %d", conn->sock->fd);
3341 status = RELAY_CONNECTION_STATUS_ERROR;
3342 }
3343 goto end;
3344 } else if (ret == 0) {
3345 /* Orderly shutdown. Not necessary to print an error. */
3346 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3347 status = RELAY_CONNECTION_STATUS_CLOSED;
3348 goto end;
3349 }
3350
3351 assert(ret > 0);
3352 assert(ret <= state->left_to_receive);
3353
3354 state->left_to_receive -= ret;
3355 state->received += ret;
3356
3357 if (state->left_to_receive > 0) {
3358 /*
3359 * Can't transition to the protocol's next state, wait to
3360 * receive the rest of the header.
3361 */
3362 DBG3("Partial reception of data connection header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3363 state->received, state->left_to_receive,
3364 conn->sock->fd);
3365 goto end;
3366 }
3367
3368 /* Transition to next state: receiving the payload. */
3369 conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
3370
3371 memcpy(&header, state->header_reception_buffer, sizeof(header));
3372 header.circuit_id = be64toh(header.circuit_id);
3373 header.stream_id = be64toh(header.stream_id);
3374 header.data_size = be32toh(header.data_size);
3375 header.net_seq_num = be64toh(header.net_seq_num);
3376 header.padding_size = be32toh(header.padding_size);
3377 memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
3378
3379 conn->protocol.data.state.receive_payload.left_to_receive =
3380 header.data_size;
3381 conn->protocol.data.state.receive_payload.received = 0;
3382 conn->protocol.data.state.receive_payload.rotate_index = false;
3383
3384 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64 ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64 ", padding_size = %" PRIu32,
3385 conn->sock->fd, header.circuit_id,
3386 header.stream_id, header.data_size,
3387 header.net_seq_num, header.padding_size);
3388
3389 stream = stream_get_by_id(header.stream_id);
3390 if (!stream) {
3391 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
3392 header.stream_id);
3393 /* Protocol error. */
3394 status = RELAY_CONNECTION_STATUS_ERROR;
3395 goto end;
3396 }
3397
3398 pthread_mutex_lock(&stream->lock);
3399
3400 /* Check if a rotation is needed. */
3401 if (stream->tracefile_size > 0 &&
3402 (stream->tracefile_size_current + header.data_size) >
3403 stream->tracefile_size) {
3404 uint64_t old_id, new_id;
3405
3406 old_id = tracefile_array_get_file_index_head(stream->tfa);
3407 tracefile_array_file_rotate(stream->tfa);
3408
3409 /* new_id is updated by utils_rotate_stream_file. */
3410 new_id = old_id;
3411
3412 ret = utils_rotate_stream_file(stream->path_name,
3413 stream->channel_name, stream->tracefile_size,
3414 stream->tracefile_count, -1,
3415 -1, stream->stream_fd->fd,
3416 &new_id, &stream->stream_fd->fd);
3417 if (ret < 0) {
3418 ERR("Failed to rotate stream output file");
3419 status = RELAY_CONNECTION_STATUS_ERROR;
3420 goto end_stream_unlock;
3421 }
3422
3423 /*
3424 * Reset current size because we just performed a stream
3425 * rotation.
3426 */
3427 stream->tracefile_size_current = 0;
3428 conn->protocol.data.state.receive_payload.rotate_index = true;
3429 }
3430
3431 end_stream_unlock:
3432 pthread_mutex_unlock(&stream->lock);
3433 stream_put(stream);
3434 end:
3435 return status;
3436 }
3437
3438 static enum relay_connection_status relay_process_data_receive_payload(
3439 struct relay_connection *conn)
3440 {
3441 int ret;
3442 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3443 struct relay_stream *stream;
3444 struct data_connection_state_receive_payload *state =
3445 &conn->protocol.data.state.receive_payload;
3446 const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
3447 char data_buffer[chunk_size];
3448 bool partial_recv = false;
3449 bool new_stream = false, close_requested = false, index_flushed = false;
3450 uint64_t left_to_receive = state->left_to_receive;
3451 struct relay_session *session;
3452
3453 DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64" bytes received, %" PRIu64 " bytes left to receive",
3454 state->header.stream_id, state->header.net_seq_num,
3455 state->received, left_to_receive);
3456
3457 stream = stream_get_by_id(state->header.stream_id);
3458 if (!stream) {
3459 /* Protocol error. */
3460 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
3461 state->header.stream_id);
3462 status = RELAY_CONNECTION_STATUS_ERROR;
3463 goto end;
3464 }
3465
3466 pthread_mutex_lock(&stream->lock);
3467 session = stream->trace->session;
3468 if (!conn->session) {
3469 ret = connection_set_session(conn, session);
3470 if (ret) {
3471 status = RELAY_CONNECTION_STATUS_ERROR;
3472 goto end_stream_unlock;
3473 }
3474 }
3475
3476 /*
3477 * The size of the "chunk" received on any iteration is bounded by:
3478 * - the data left to receive,
3479 * - the data immediately available on the socket,
3480 * - the on-stack data buffer
3481 */
3482 while (left_to_receive > 0 && !partial_recv) {
3483 ssize_t write_ret;
3484 size_t recv_size = min(left_to_receive, chunk_size);
3485
3486 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer,
3487 recv_size, MSG_DONTWAIT);
3488 if (ret < 0) {
3489 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3490 PERROR("Socket %d error", conn->sock->fd);
3491 status = RELAY_CONNECTION_STATUS_ERROR;
3492 }
3493 goto end_stream_unlock;
3494 } else if (ret == 0) {
3495 /* No more data ready to be consumed on socket. */
3496 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
3497 state->header.stream_id);
3498 status = RELAY_CONNECTION_STATUS_CLOSED;
3499 break;
3500 } else if (ret < (int) recv_size) {
3501 /*
3502 * All the data available on the socket has been
3503 * consumed.
3504 */
3505 partial_recv = true;
3506 }
3507
3508 recv_size = ret;
3509
3510 /* Write data to stream output fd. */
3511 write_ret = lttng_write(stream->stream_fd->fd, data_buffer,
3512 recv_size);
3513 if (write_ret < (ssize_t) recv_size) {
3514 ERR("Relay error writing data to file");
3515 status = RELAY_CONNECTION_STATUS_ERROR;
3516 goto end_stream_unlock;
3517 }
3518
3519 left_to_receive -= recv_size;
3520 state->received += recv_size;
3521 state->left_to_receive = left_to_receive;
3522
3523 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64,
3524 write_ret, stream->stream_handle);
3525 }
3526
3527 if (state->left_to_receive > 0) {
3528 /*
3529 * Did not receive all the data expected, wait for more data to
3530 * become available on the socket.
3531 */
3532 DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64 " bytes received, %" PRIu64 " bytes left to receive",
3533 state->header.stream_id, state->received,
3534 state->left_to_receive);
3535 goto end_stream_unlock;
3536 }
3537
3538 ret = write_padding_to_file(stream->stream_fd->fd,
3539 state->header.padding_size);
3540 if ((int64_t) ret < (int64_t) state->header.padding_size) {
3541 ERR("write_padding_to_file: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
3542 stream->stream_handle,
3543 state->header.net_seq_num, ret);
3544 status = RELAY_CONNECTION_STATUS_ERROR;
3545 goto end_stream_unlock;
3546 }
3547
3548
3549 if (session_streams_have_index(session)) {
3550 ret = handle_index_data(stream, state->header.net_seq_num,
3551 state->rotate_index, &index_flushed, state->header.data_size + state->header.padding_size);
3552 if (ret < 0) {
3553 ERR("handle_index_data: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
3554 stream->stream_handle,
3555 state->header.net_seq_num, ret);
3556 status = RELAY_CONNECTION_STATUS_ERROR;
3557 goto end_stream_unlock;
3558 }
3559 }
3560
3561 stream->tracefile_size_current += state->header.data_size +
3562 state->header.padding_size;
3563
3564 if (stream->prev_data_seq == -1ULL) {
3565 new_stream = true;
3566 }
3567 if (index_flushed) {
3568 stream->pos_after_last_complete_data_index =
3569 stream->tracefile_size_current;
3570 stream->prev_index_seq = state->header.net_seq_num;
3571 }
3572
3573 stream->prev_data_seq = state->header.net_seq_num;
3574
3575 /*
3576 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3577 * contents of *state which are aliased (union) to the same location as
3578 * the new state. Don't use it beyond this point.
3579 */
3580 connection_reset_protocol_state(conn);
3581 state = NULL;
3582
3583 ret = try_rotate_stream(stream);
3584 if (ret < 0) {
3585 status = RELAY_CONNECTION_STATUS_ERROR;
3586 goto end_stream_unlock;
3587 }
3588
3589 end_stream_unlock:
3590 close_requested = stream->close_requested;
3591 pthread_mutex_unlock(&stream->lock);
3592 if (close_requested && left_to_receive == 0) {
3593 try_stream_close(stream);
3594 }
3595
3596 if (new_stream) {
3597 pthread_mutex_lock(&session->lock);
3598 uatomic_set(&session->new_streams, 1);
3599 pthread_mutex_unlock(&session->lock);
3600 }
3601
3602 stream_put(stream);
3603 end:
3604 return status;
3605 }
3606
3607 /*
3608 * relay_process_data: Process the data received on the data socket
3609 */
3610 static enum relay_connection_status relay_process_data(
3611 struct relay_connection *conn)
3612 {
3613 enum relay_connection_status status;
3614
3615 switch (conn->protocol.data.state_id) {
3616 case DATA_CONNECTION_STATE_RECEIVE_HEADER:
3617 status = relay_process_data_receive_header(conn);
3618 break;
3619 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
3620 status = relay_process_data_receive_payload(conn);
3621 break;
3622 default:
3623 ERR("Unexpected data connection communication state.");
3624 abort();
3625 }
3626
3627 return status;
3628 }
3629
3630 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
3631 {
3632 int ret;
3633
3634 (void) lttng_poll_del(events, pollfd);
3635
3636 ret = close(pollfd);
3637 if (ret < 0) {
3638 ERR("Closing pollfd %d", pollfd);
3639 }
3640 }
3641
3642 static void relay_thread_close_connection(struct lttng_poll_event *events,
3643 int pollfd, struct relay_connection *conn)
3644 {
3645 const char *type_str;
3646
3647 switch (conn->type) {
3648 case RELAY_DATA:
3649 type_str = "Data";
3650 break;
3651 case RELAY_CONTROL:
3652 type_str = "Control";
3653 break;
3654 case RELAY_VIEWER_COMMAND:
3655 type_str = "Viewer Command";
3656 break;
3657 case RELAY_VIEWER_NOTIFICATION:
3658 type_str = "Viewer Notification";
3659 break;
3660 default:
3661 type_str = "Unknown";
3662 }
3663 cleanup_connection_pollfd(events, pollfd);
3664 connection_put(conn);
3665 DBG("%s connection closed with %d", type_str, pollfd);
3666 }
3667
3668 /*
3669 * This thread does the actual work
3670 */
3671 static void *relay_thread_worker(void *data)
3672