projects / lttng-tools.git / blob
? search:


10db7979852623cbe3611eb0114fde8eba5a654c
[lttng-tools.git] / src / bin / lttng-relayd / main.cpp
1 /*
2 * Copyright (C) 2012 Julien Desfossez <jdesfossez@efficios.com>
3 * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
4 * Copyright (C) 2013 Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 * Copyright (C) 2015 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
6 *
7 * SPDX-License-Identifier: GPL-2.0-only
8 *
9 */
10
11 #define _LGPL_SOURCE
12 #include "backward-compatibility-group-by.hpp"
13 #include "cmd.hpp"
14 #include "connection.hpp"
15 #include "ctf-trace.hpp"
16 #include "health-relayd.hpp"
17 #include "index.hpp"
18 #include "live.hpp"
19 #include "lttng-relayd.hpp"
20 #include "session.hpp"
21 #include "sessiond-trace-chunks.hpp"
22 #include "stream.hpp"
23 #include "tcp_keep_alive.hpp"
24 #include "testpoint.hpp"
25 #include "tracefile-array.hpp"
26 #include "utils.hpp"
27 #include "version.hpp"
28 #include "viewer-stream.hpp"
29
30 #include <common/align.hpp>
31 #include <common/buffer-view.hpp>
32 #include <common/common.hpp>
33 #include <common/compat/endian.hpp>
34 #include <common/compat/getenv.hpp>
35 #include <common/compat/poll.hpp>
36 #include <common/compat/socket.hpp>
37 #include <common/daemonize.hpp>
38 #include <common/defaults.hpp>
39 #include <common/dynamic-buffer.hpp>
40 #include <common/fd-tracker/fd-tracker.hpp>
41 #include <common/fd-tracker/utils.hpp>
42 #include <common/futex.hpp>
43 #include <common/ini-config/ini-config.hpp>
44 #include <common/path.hpp>
45 #include <common/sessiond-comm/inet.hpp>
46 #include <common/sessiond-comm/relayd.hpp>
47 #include <common/sessiond-comm/sessiond-comm.hpp>
48 #include <common/string-utils/format.hpp>
49 #include <common/uri.hpp>
50 #include <common/utils.hpp>
51
52 #include <lttng/lttng.h>
53
54 #include <algorithm>
55 #include <ctype.h>
56 #include <fcntl.h>
57 #include <getopt.h>
58 #include <grp.h>
59 #include <inttypes.h>
60 #include <limits.h>
61 #include <pthread.h>
62 #include <signal.h>
63 #include <stdio.h>
64 #include <stdlib.h>
65 #include <string.h>
66 #include <strings.h>
67 #include <sys/mman.h>
68 #include <sys/mount.h>
69 #include <sys/resource.h>
70 #include <sys/socket.h>
71 #include <sys/stat.h>
72 #include <sys/types.h>
73 #include <sys/wait.h>
74 #include <unistd.h>
75 #include <urcu/futex.h>
76 #include <urcu/rculist.h>
77 #include <urcu/uatomic.h>
78
79 static const char *help_msg =
80 #ifdef LTTNG_EMBED_HELP
81 #include <lttng-relayd.8.h>
82 #else
83 NULL
84 #endif
85 ;
86
87 enum relay_connection_status {
88 RELAY_CONNECTION_STATUS_OK,
89 /* An error occurred while processing an event on the connection. */
90 RELAY_CONNECTION_STATUS_ERROR,
91 /* Connection closed/shutdown cleanly. */
92 RELAY_CONNECTION_STATUS_CLOSED,
93 };
94
95 /* command line options */
96 char *opt_output_path, *opt_working_directory;
97 static int opt_daemon, opt_background, opt_print_version, opt_allow_clear = 1;
98 enum relay_group_output_by opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_UNKNOWN;
99
100 /* Argument variables */
101 int lttng_opt_quiet; /* not static in error.h */
102 int lttng_opt_verbose; /* not static in error.h */
103 int lttng_opt_mi; /* not static in error.h */
104
105 /*
106 * We need to wait for listener and live listener threads, as well as
107 * health check thread, before being ready to signal readiness.
108 */
109 #define NR_LTTNG_RELAY_READY 3
110 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
111
112 /* Size of receive buffer. */
113 #define RECV_DATA_BUFFER_SIZE 65536
114
115 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
116 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
117
118 static struct lttng_uri *control_uri;
119 static struct lttng_uri *data_uri;
120 static struct lttng_uri *live_uri;
121
122 const char *progname;
123
124 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
125 static int tracing_group_name_override;
126
127 const char *const config_section_name = "relayd";
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 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
159 static unsigned int lttng_opt_fd_pool_size = -1;
160
161 /* Global relay stream hash table. */
162 struct lttng_ht *relay_streams_ht;
163
164 /* Global relay viewer stream hash table. */
165 struct lttng_ht *viewer_streams_ht;
166
167 /* Global relay sessions hash table. */
168 struct lttng_ht *sessions_ht;
169
170 /* Relayd health monitoring */
171 struct health_app *health_relayd;
172
173 struct sessiond_trace_chunk_registry *sessiond_trace_chunk_registry;
174
175 /* Global fd tracker. */
176 struct fd_tracker *the_fd_tracker;
177
178 static struct option long_options[] = {
179 {
180 "control-port",
181 1,
182 0,
183 'C',
184 },
185 {
186 "data-port",
187 1,
188 0,
189 'D',
190 },
191 {
192 "live-port",
193 1,
194 0,
195 'L',
196 },
197 {
198 "daemonize",
199 0,
200 0,
201 'd',
202 },
203 {
204 "background",
205 0,
206 0,
207 'b',
208 },
209 {
210 "group",
211 1,
212 0,
213 'g',
214 },
215 {
216 "fd-pool-size",
217 1,
218 0,
219 '\0',
220 },
221 {
222 "help",
223 0,
224 0,
225 'h',
226 },
227 {
228 "output",
229 1,
230 0,
231 'o',
232 },
233 {
234 "verbose",
235 0,
236 0,
237 'v',
238 },
239 { "config", 1, 0, 'f' },
240 { "version", 0, 0, 'V' },
241 {
242 "working-directory",
243 1,
244 0,
245 'w',
246 },
247 {
248 "group-output-by-session",
249 0,
250 0,
251 's',
252 },
253 {
254 "group-output-by-host",
255 0,
256 0,
257 'p',
258 },
259 { "disallow-clear", 0, 0, 'x' },
260 {
261 NULL,
262 0,
263 0,
264 0,
265 },
266 };
267
268 static const char *config_ignore_options[] = { "help", "config", "version" };
269
270 static void print_version(void)
271 {
272 fprintf(stdout, "%s\n", VERSION);
273 }
274
275 static void relayd_config_log(void)
276 {
277 DBG("LTTng-relayd " VERSION " - " VERSION_NAME "%s%s",
278 GIT_VERSION[0] == '\0' ? "" : " - " GIT_VERSION,
279 EXTRA_VERSION_NAME[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME);
280 if (EXTRA_VERSION_DESCRIPTION[0] != '\0') {
281 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION "\n");
282 }
283 if (EXTRA_VERSION_PATCHES[0] != '\0') {
284 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES "\n");
285 }
286 }
287
288 /*
289 * Take an option from the getopt output and set it in the right variable to be
290 * used later.
291 *
292 * Return 0 on success else a negative value.
293 */
294 static int set_option(int opt, const char *arg, const char *optname)
295 {
296 int ret;
297
298 switch (opt) {
299 case 0:
300 if (!strcmp(optname, "fd-pool-size")) {
301 unsigned long v;
302
303 errno = 0;
304 v = strtoul(arg, NULL, 0);
305 if (errno != 0 || !isdigit((unsigned char) arg[0])) {
306 ERR("Wrong value in --fd-pool-size parameter: %s", arg);
307 ret = -1;
308 goto end;
309 }
310 if (v >= UINT_MAX) {
311 ERR("File descriptor cap overflow in --fd-pool-size parameter: %s",
312 arg);
313 ret = -1;
314 goto end;
315 }
316 lttng_opt_fd_pool_size = (unsigned int) v;
317 } else {
318 fprintf(stderr, "unknown option %s", optname);
319 if (arg) {
320 fprintf(stderr, " with arg %s\n", arg);
321 }
322 }
323 break;
324 case 'C':
325 if (lttng_is_setuid_setgid()) {
326 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
327 "-C, --control-port");
328 } else {
329 ret = uri_parse(arg, &control_uri);
330 if (ret < 0) {
331 ERR("Invalid control URI specified");
332 goto end;
333 }
334 if (control_uri->port == 0) {
335 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
336 }
337 }
338 break;
339 case 'D':
340 if (lttng_is_setuid_setgid()) {
341 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
342 "-D, -data-port");
343 } else {
344 ret = uri_parse(arg, &data_uri);
345 if (ret < 0) {
346 ERR("Invalid data URI specified");
347 goto end;
348 }
349 if (data_uri->port == 0) {
350 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
351 }
352 }
353 break;
354 case 'L':
355 if (lttng_is_setuid_setgid()) {
356 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
357 "-L, -live-port");
358 } else {
359 ret = uri_parse(arg, &live_uri);
360 if (ret < 0) {
361 ERR("Invalid live URI specified");
362 goto end;
363 }
364 if (live_uri->port == 0) {
365 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
366 }
367 }
368 break;
369 case 'd':
370 opt_daemon = 1;
371 break;
372 case 'b':
373 opt_background = 1;
374 break;
375 case 'g':
376 if (lttng_is_setuid_setgid()) {
377 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
378 "-g, --group");
379 } else {
380 tracing_group_name = strdup(arg);
381 if (tracing_group_name == NULL) {
382 ret = -errno;
383 PERROR("strdup");
384 goto end;
385 }
386 tracing_group_name_override = 1;
387 }
388 break;
389 case 'h':
390 ret = utils_show_help(8, "lttng-relayd", help_msg);
391 if (ret) {
392 ERR("Cannot show --help for `lttng-relayd`");
393 perror("exec");
394 }
395 exit(EXIT_FAILURE);
396 case 'V':
397 opt_print_version = 1;
398 break;
399 case 'o':
400 if (lttng_is_setuid_setgid()) {
401 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
402 "-o, --output");
403 } else {
404 ret = asprintf(&opt_output_path, "%s", arg);
405 if (ret < 0) {
406 ret = -errno;
407 PERROR("asprintf opt_output_path");
408 goto end;
409 }
410 }
411 break;
412 case 'w':
413 if (lttng_is_setuid_setgid()) {
414 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
415 "-w, --working-directory");
416 } else {
417 ret = asprintf(&opt_working_directory, "%s", arg);
418 if (ret < 0) {
419 ret = -errno;
420 PERROR("asprintf opt_working_directory");
421 goto end;
422 }
423 }
424 break;
425
426 case 'v':
427 /* Verbose level can increase using multiple -v */
428 if (arg) {
429 lttng_opt_verbose = config_parse_value(arg);
430 } else {
431 /* Only 3 level of verbosity (-vvv). */
432 if (lttng_opt_verbose < 3) {
433 lttng_opt_verbose += 1;
434 }
435 }
436 break;
437 case 's':
438 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
439 ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
440 exit(EXIT_FAILURE);
441 }
442 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_SESSION;
443 break;
444 case 'p':
445 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
446 ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
447 exit(EXIT_FAILURE);
448 }
449 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
450 break;
451 case 'x':
452 /* Disallow clear */
453 opt_allow_clear = 0;
454 break;
455 default:
456 /* Unknown option or other error.
457 * Error is printed by getopt, just return */
458 ret = -1;
459 goto end;
460 }
461
462 /* All good. */
463 ret = 0;
464
465 end:
466 return ret;
467 }
468
469 /*
470 * config_entry_handler_cb used to handle options read from a config file.
471 * See config_entry_handler_cb comment in common/config/session-config.h for the
472 * return value conventions.
473 */
474 static int config_entry_handler(const struct config_entry *entry,
475 void *unused __attribute__((unused)))
476 {
477 int ret = 0, i;
478
479 if (!entry || !entry->name || !entry->value) {
480 ret = -EINVAL;
481 goto end;
482 }
483
484 /* Check if the option is to be ignored */
485 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
486 if (!strcmp(entry->name, config_ignore_options[i])) {
487 goto end;
488 }
489 }
490
491 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
492 /* Ignore if entry name is not fully matched. */
493 if (strcmp(entry->name, long_options[i].name)) {
494 continue;
495 }
496
497 /*
498 * If the option takes no argument on the command line,
499 * we have to check if the value is "true". We support
500 * non-zero numeric values, true, on and yes.
501 */
502 if (!long_options[i].has_arg) {
503 ret = config_parse_value(entry->value);
504 if (ret <= 0) {
505 if (ret) {
506 WARN("Invalid configuration value \"%s\" for option %s",
507 entry->value,
508 entry->name);
509 }
510 /* False, skip boolean config option. */
511 goto end;
512 }
513 }
514
515 ret = set_option(long_options[i].val, entry->value, entry->name);
516 goto end;
517 }
518
519 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
520
521 end:
522 return ret;
523 }
524
525 static int parse_env_options(void)
526 {
527 int ret = 0;
528 char *value = NULL;
529
530 value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV);
531 if (value) {
532 opt_working_directory = strdup(value);
533 if (!opt_working_directory) {
534 ERR("Failed to allocate working directory string (\"%s\")", value);
535 ret = -1;
536 }
537 }
538 return ret;
539 }
540
541 static int set_fd_pool_size(void)
542 {
543 int ret = 0;
544 struct rlimit rlimit;
545
546 ret = getrlimit(RLIMIT_NOFILE, &rlimit);
547 if (ret) {
548 PERROR("Failed to get file descriptor limit");
549 ret = -1;
550 goto end;
551 }
552
553 DBG("File descriptor count limits are %" PRIu64 " (soft) and %" PRIu64 " (hard)",
554 (uint64_t) rlimit.rlim_cur,
555 (uint64_t) rlimit.rlim_max);
556 if (lttng_opt_fd_pool_size == -1) {
557 /* Use default value (soft limit - reserve). */
558 if (rlimit.rlim_cur < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
559 ERR("The process' file number limit is too low (%" PRIu64
560 "). The process' file number limit must be set to at least %i.",
561 (uint64_t) rlimit.rlim_cur,
562 DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
563 ret = -1;
564 goto end;
565 }
566 lttng_opt_fd_pool_size = rlimit.rlim_cur - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
567 goto end;
568 }
569
570 if (lttng_opt_fd_pool_size < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
571 ERR("File descriptor pool size must be set to at least %d",
572 DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
573 ret = -1;
574 goto end;
575 }
576
577 if (lttng_opt_fd_pool_size > rlimit.rlim_cur) {
578 ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64
579 ").",
580 lttng_opt_fd_pool_size,
581 (uint64_t) rlimit.rlim_cur);
582 ret = -1;
583 goto end;
584 }
585
586 DBG("File descriptor pool size argument (%u) adjusted to %u to accommodates transient fd uses",
587 lttng_opt_fd_pool_size,
588 lttng_opt_fd_pool_size - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE);
589 lttng_opt_fd_pool_size -= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
590 end:
591 return ret;
592 }
593
594 static int set_options(int argc, char **argv)
595 {
596 int c, ret = 0, option_index = 0, retval = 0;
597 int orig_optopt = optopt, orig_optind = optind;
598 char *default_address, *optstring;
599 char *config_path = NULL;
600
601 optstring = utils_generate_optstring(long_options,
602 sizeof(long_options) / sizeof(struct option));
603 if (!optstring) {
604 retval = -ENOMEM;
605 goto exit;
606 }
607
608 /* Check for the --config option */
609
610 while ((c = getopt_long(argc, argv, optstring, long_options, &option_index)) != -1) {
611 if (c == '?') {
612 retval = -EINVAL;
613 goto exit;
614 } else if (c != 'f') {
615 continue;
616 }
617
618 if (lttng_is_setuid_setgid()) {
619 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
620 "-f, --config");
621 } else {
622 free(config_path);
623 config_path = utils_expand_path(optarg);
624 if (!config_path) {
625 ERR("Failed to resolve path: %s", optarg);
626 }
627 }
628 }
629
630 ret = config_get_section_entries(
631 config_path, config_section_name, config_entry_handler, NULL);
632 if (ret) {
633 if (ret > 0) {
634 ERR("Invalid configuration option at line %i", ret);
635 }
636 retval = -1;
637 goto exit;
638 }
639
640 /* Reset getopt's global state */
641 optopt = orig_optopt;
642 optind = orig_optind;
643 while (1) {
644 c = getopt_long(argc, argv, optstring, long_options, &option_index);
645 if (c == -1) {
646 break;
647 }
648
649 ret = set_option(c, optarg, long_options[option_index].name);
650 if (ret < 0) {
651 retval = -1;
652 goto exit;
653 }
654 }
655
656 /* assign default values */
657 if (control_uri == NULL) {
658 ret = asprintf(&default_address,
659 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
660 DEFAULT_NETWORK_CONTROL_PORT);
661 if (ret < 0) {
662 PERROR("asprintf default data address");
663 retval = -1;
664 goto exit;
665 }
666
667 ret = uri_parse(default_address, &control_uri);
668 free(default_address);
669 if (ret < 0) {
670 ERR("Invalid control URI specified");
671 retval = -1;
672 goto exit;
673 }
674 }
675 if (data_uri == NULL) {
676 ret = asprintf(&default_address,
677 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
678 DEFAULT_NETWORK_DATA_PORT);
679 if (ret < 0) {
680 PERROR("asprintf default data address");
681 retval = -1;
682 goto exit;
683 }
684
685 ret = uri_parse(default_address, &data_uri);
686 free(default_address);
687 if (ret < 0) {
688 ERR("Invalid data URI specified");
689 retval = -1;
690 goto exit;
691 }
692 }
693 if (live_uri == NULL) {
694 ret = asprintf(&default_address,
695 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
696 DEFAULT_NETWORK_VIEWER_PORT);
697 if (ret < 0) {
698 PERROR("asprintf default viewer control address");
699 retval = -1;
700 goto exit;
701 }
702
703 ret = uri_parse(default_address, &live_uri);
704 free(default_address);
705 if (ret < 0) {
706 ERR("Invalid viewer control URI specified");
707 retval = -1;
708 goto exit;
709 }
710 }
711 ret = set_fd_pool_size();
712 if (ret) {
713 retval = -1;
714 goto exit;
715 }
716
717 if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
718 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
719 }
720 if (opt_allow_clear) {
721 /* Check if env variable exists. */
722 const char *value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
723 if (value) {
724 ret = config_parse_value(value);
725 if (ret < 0) {
726 ERR("Invalid value for %s specified",
727 DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
728 retval = -1;
729 goto exit;
730 }
731 opt_allow_clear = !ret;
732 }
733 }
734
735 exit:
736 free(config_path);
737 free(optstring);
738 return retval;
739 }
740
741 static void print_global_objects(void)
742 {
743 print_viewer_streams();
744 print_relay_streams();
745 print_sessions();
746 }
747
748 static int noop_close(void *data __attribute__((unused)), int *fds __attribute__((unused)))
749 {
750 return 0;
751 }
752
753 static void untrack_stdio(void)
754 {
755 int fds[] = { fileno(stdout), fileno(stderr) };
756
757 /*
758 * noop_close is used since we don't really want to close
759 * the stdio output fds; we merely want to stop tracking them.
760 */
761 (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker, fds, 2, noop_close, NULL);
762 }
763
764 /*
765 * Cleanup the daemon
766 */
767 static void relayd_cleanup(void)
768 {
769 print_global_objects();
770
771 DBG("Cleaning up");
772
773 if (viewer_streams_ht)
774 lttng_ht_destroy(viewer_streams_ht);
775 if (relay_streams_ht)
776 lttng_ht_destroy(relay_streams_ht);
777 if (sessions_ht)
778 lttng_ht_destroy(sessions_ht);
779
780 free(opt_output_path);
781 free(opt_working_directory);
782
783 if (health_relayd) {
784 health_app_destroy(health_relayd);
785 }
786 /* Close thread quit pipes */
787 if (health_quit_pipe[0] != -1) {
788 (void) fd_tracker_util_pipe_close(the_fd_tracker, health_quit_pipe);
789 }
790 relayd_close_thread_quit_pipe();
791 if (sessiond_trace_chunk_registry) {
792 sessiond_trace_chunk_registry_destroy(sessiond_trace_chunk_registry);
793 }
794 if (the_fd_tracker) {
795 untrack_stdio();
796 /*
797 * fd_tracker_destroy() will log the contents of the fd-tracker
798 * if a leak is detected.
799 */
800 fd_tracker_destroy(the_fd_tracker);
801 }
802
803 uri_free(control_uri);
804 uri_free(data_uri);
805 /* Live URI is freed in the live thread. */
806
807 if (tracing_group_name_override) {
808 free((void *) tracing_group_name);
809 }
810 }
811
812 static int notify_health_quit_pipe(int *pipe)
813 {
814 ssize_t ret;
815
816 ret = lttng_write(pipe[1], "4", 1);
817 if (ret < 1) {
818 PERROR("write relay health quit");
819 goto end;
820 }
821 ret = 0;
822 end:
823 return ret;
824 }
825
826 /*
827 * Stop all relayd and relayd-live threads.
828 */
829 int lttng_relay_stop_threads(void)
830 {
831 int retval = 0;
832
833 /* Stopping all threads */
834 DBG("Terminating all threads");
835 if (relayd_notify_thread_quit_pipe()) {
836 ERR("write error on thread quit pipe");
837 retval = -1;
838 }
839
840 if (notify_health_quit_pipe(health_quit_pipe)) {
841 ERR("write error on health quit pipe");
842 }
843
844 /* Dispatch thread */
845 CMM_STORE_SHARED(dispatch_thread_exit, 1);
846 futex_nto1_wake(&relay_conn_queue.futex);
847
848 if (relayd_live_stop()) {
849 ERR("Error stopping live threads");
850 retval = -1;
851 }
852 return retval;
853 }
854
855 /*
856 * Signal handler for the daemon
857 *
858 * Simply stop all worker threads, leaving main() return gracefully after
859 * joining all threads and calling cleanup().
860 */
861 static void sighandler(int sig)
862 {
863 switch (sig) {
864 case SIGINT:
865 DBG("SIGINT caught");
866 if (lttng_relay_stop_threads()) {
867 ERR("Error stopping threads");
868 }
869 break;
870 case SIGTERM:
871 DBG("SIGTERM caught");
872 if (lttng_relay_stop_threads()) {
873 ERR("Error stopping threads");
874 }
875 break;
876 case SIGUSR1:
877 CMM_STORE_SHARED(recv_child_signal, 1);
878 break;
879 default:
880 break;
881 }
882 }
883
884 /*
885 * Setup signal handler for :
886 * SIGINT, SIGTERM, SIGPIPE
887 */
888 static int set_signal_handler(void)
889 {
890 int ret = 0;
891 struct sigaction sa;
892 sigset_t sigset;
893
894 if ((ret = sigemptyset(&sigset)) < 0) {
895 PERROR("sigemptyset");
896 return ret;
897 }
898
899 sa.sa_mask = sigset;
900 sa.sa_flags = 0;
901
902 sa.sa_handler = sighandler;
903 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
904 PERROR("sigaction");
905 return ret;
906 }
907
908 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
909 PERROR("sigaction");
910 return ret;
911 }
912
913 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
914 PERROR("sigaction");
915 return ret;
916 }
917
918 sa.sa_handler = SIG_IGN;
919 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
920 PERROR("sigaction");
921 return ret;
922 }
923
924 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
925
926 return ret;
927 }
928
929 void lttng_relay_notify_ready(void)
930 {
931 /* Notify the parent of the fork() process that we are ready. */
932 if (opt_daemon || opt_background) {
933 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
934 kill(child_ppid, SIGUSR1);
935 }
936 }
937 }
938
939 /*
940 * Init health quit pipe.
941 *
942 * Return -1 on error or 0 if all pipes are created.
943 */
944 static int init_health_quit_pipe(void)
945 {
946 return fd_tracker_util_pipe_open_cloexec(
947 the_fd_tracker, "Health quit pipe", health_quit_pipe);
948 }
949
950 static int create_sock(void *data, int *out_fd)
951 {
952 int ret;
953 struct lttcomm_sock *sock = (lttcomm_sock *) data;
954
955 ret = lttcomm_create_sock(sock);
956 if (ret < 0) {
957 goto end;
958 }
959
960 *out_fd = sock->fd;
961 end:
962 return ret;
963 }
964
965 static int close_sock(void *data, int *in_fd __attribute__((unused)))
966 {
967 struct lttcomm_sock *sock = (lttcomm_sock *) data;
968
969 return sock->ops->close(sock);
970 }
971
972 static int accept_sock(void *data, int *out_fd)
973 {
974 int ret = 0;
975 /* Socks is an array of in_sock, out_sock. */
976 struct lttcomm_sock **socks = (lttcomm_sock **) data;
977 struct lttcomm_sock *in_sock = socks[0];
978
979 socks[1] = in_sock->ops->accept(in_sock);
980 if (!socks[1]) {
981 ret = -1;
982 goto end;
983 }
984 *out_fd = socks[1]->fd;
985 end:
986 return ret;
987 }
988
989 /*
990 * Create and init socket from uri.
991 */
992 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri, const char *name)
993 {
994 int ret, sock_fd;
995 struct lttcomm_sock *sock = NULL;
996 char uri_str[PATH_MAX];
997 char *formated_name = NULL;
998
999 sock = lttcomm_alloc_sock_from_uri(uri);
1000 if (sock == NULL) {
1001 ERR("Allocating socket");
1002 goto error;
1003 }
1004
1005 /*
1006 * Don't fail to create the socket if the name can't be built as it is
1007 * only used for debugging purposes.
1008 */
1009 ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
1010 uri_str[sizeof(uri_str) - 1] = '\0';
1011 if (ret >= 0) {
1012 ret = asprintf(&formated_name, "%s socket @ %s", name, uri_str);
1013 if (ret < 0) {
1014 formated_name = NULL;
1015 }
1016 }
1017
1018 ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker,
1019 &sock_fd,
1020 (const char **) (formated_name ? &formated_name :
1021 NULL),
1022 1,
1023 create_sock,
1024 sock);
1025 if (ret) {
1026 PERROR("Failed to open \"%s\" relay socket", formated_name ?: "Unknown");
1027 goto error;
1028 }
1029 DBG("Listening on %s socket %d", name, sock->fd);
1030
1031 ret = sock->ops->bind(sock);
1032 if (ret < 0) {
1033 PERROR("Failed to bind socket");
1034 goto error;
1035 }
1036
1037 ret = sock->ops->listen(sock, -1);
1038 if (ret < 0) {
1039 goto error;
1040 }
1041
1042 free(formated_name);
1043 return sock;
1044
1045 error:
1046 if (sock) {
1047 lttcomm_destroy_sock(sock);
1048 }
1049 free(formated_name);
1050 return NULL;
1051 }
1052
1053 static struct lttcomm_sock *accept_relayd_sock(struct lttcomm_sock *listening_sock,
1054 const char *name)
1055 {
1056 int out_fd, ret;
1057 struct lttcomm_sock *socks[2] = { listening_sock, NULL };
1058 struct lttcomm_sock *new_sock = NULL;
1059
1060 ret = fd_tracker_open_unsuspendable_fd(
1061 the_fd_tracker, &out_fd, (const char **) &name, 1, accept_sock, &socks);
1062 if (ret) {
1063 goto end;
1064 }
1065 new_sock = socks[1];
1066 DBG("%s accepted, socket %d", name, new_sock->fd);
1067 end:
1068 return new_sock;
1069 }
1070
1071 /*
1072 * This thread manages the listening for new connections on the network
1073 */
1074 static void *relay_thread_listener(void *data __attribute__((unused)))
1075 {
1076 int i, ret, err = -1;
1077 uint32_t nb_fd;
1078 struct lttng_poll_event events;
1079 struct lttcomm_sock *control_sock, *data_sock;
1080
1081 DBG("[thread] Relay listener started");
1082
1083 rcu_register_thread();
1084 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
1085
1086 health_code_update();
1087
1088 control_sock = relay_socket_create(control_uri, "Control listener");
1089 if (!control_sock) {
1090 goto error_sock_control;
1091 }
1092
1093 data_sock = relay_socket_create(data_uri, "Data listener");
1094 if (!data_sock) {
1095 goto error_sock_relay;
1096 }
1097
1098 /*
1099 * Pass 3 as size here for the thread quit pipe, control and
1100 * data socket.
1101 */
1102 ret = create_named_thread_poll_set(&events, 3, "Listener thread epoll");
1103 if (ret < 0) {
1104 goto error_create_poll;
1105 }
1106
1107 /* Add the control socket */
1108 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
1109 if (ret < 0) {
1110 goto error_poll_add;
1111 }
1112
1113 /* Add the data socket */
1114 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
1115 if (ret < 0) {
1116 goto error_poll_add;
1117 }
1118
1119 lttng_relay_notify_ready();
1120
1121 if (testpoint(relayd_thread_listener)) {
1122 goto error_testpoint;
1123 }
1124
1125 while (1) {
1126 health_code_update();
1127
1128 DBG("Listener accepting connections");
1129
1130 restart:
1131 health_poll_entry();
1132 ret = lttng_poll_wait(&events, -1);
1133 health_poll_exit();
1134 if (ret < 0) {
1135 /*
1136 * Restart interrupted system call.
1137 */
1138 if (errno == EINTR) {
1139 goto restart;
1140 }
1141 goto error;
1142 }
1143
1144 nb_fd = ret;
1145
1146 DBG("Relay new connection received");
1147 for (i = 0; i < nb_fd; i++) {
1148 /* Fetch once the poll data */
1149 const auto revents = LTTNG_POLL_GETEV(&events, i);
1150 const auto pollfd = LTTNG_POLL_GETFD(&events, i);
1151
1152 health_code_update();
1153
1154 /* Activity on thread quit pipe, exiting. */
1155 if (relayd_is_thread_quit_pipe(pollfd)) {
1156 DBG("Activity on thread quit pipe");
1157 err = 0;
1158 goto exit;
1159 }
1160
1161 if (revents & LPOLLIN) {
1162 /*
1163 * A new connection is requested, therefore a
1164 * sessiond/consumerd connection is allocated in
1165 * this thread, enqueued to a global queue and
1166 * dequeued (and freed) in the worker thread.
1167 */
1168 int val = 1;
1169 struct relay_connection *new_conn;
1170 struct lttcomm_sock *newsock = NULL;
1171 enum connection_type type;
1172
1173 if (pollfd == data_sock->fd) {
1174 type = RELAY_DATA;
1175 newsock = accept_relayd_sock(data_sock,
1176 "Data socket to relayd");
1177 } else {
1178 LTTNG_ASSERT(pollfd == control_sock->fd);
1179 type = RELAY_CONTROL;
1180 newsock = accept_relayd_sock(control_sock,
1181 "Control socket to relayd");
1182 }
1183 if (!newsock) {
1184 PERROR("accepting sock");
1185 goto error;
1186 }
1187
1188 ret = setsockopt(
1189 newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val, sizeof(val));
1190 if (ret < 0) {
1191 PERROR("setsockopt inet");
1192 lttcomm_destroy_sock(newsock);
1193 goto error;
1194 }
1195
1196 ret = socket_apply_keep_alive_config(newsock->fd);
1197 if (ret < 0) {
1198 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1199 newsock->fd);
1200 lttcomm_destroy_sock(newsock);
1201 goto error;
1202 }
1203
1204 new_conn = connection_create(newsock, type);
1205 if (!new_conn) {
1206 lttcomm_destroy_sock(newsock);
1207 goto error;
1208 }
1209
1210 /* Enqueue request for the dispatcher thread. */
1211 cds_wfcq_head_ptr_t head;
1212 head.h = &relay_conn_queue.head;
1213 cds_wfcq_enqueue(head, &relay_conn_queue.tail, &new_conn->qnode);
1214
1215 /*
1216 * Wake the dispatch queue futex.
1217 * Implicit memory barrier with the
1218 * exchange in cds_wfcq_enqueue.
1219 */
1220 futex_nto1_wake(&relay_conn_queue.futex);
1221 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1222 ERR("socket poll error");
1223 goto error;
1224 } else {
1225 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1226 goto error;
1227 }
1228 }
1229 }
1230
1231 exit:
1232 error:
1233 error_poll_add:
1234 error_testpoint:
1235 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
1236 error_create_poll:
1237 if (data_sock->fd >= 0) {
1238 int data_sock_fd = data_sock->fd;
1239
1240 ret = fd_tracker_close_unsuspendable_fd(
1241 the_fd_tracker, &data_sock_fd, 1, close_sock, data_sock);
1242 if (ret) {
1243 PERROR("Failed to close the data listener socket file descriptor");
1244 }
1245 data_sock->fd = -1;
1246 }
1247 lttcomm_destroy_sock(data_sock);
1248 error_sock_relay:
1249 if (control_sock->fd >= 0) {
1250 int control_sock_fd = control_sock->fd;
1251
1252 ret = fd_tracker_close_unsuspendable_fd(
1253 the_fd_tracker, &control_sock_fd, 1, close_sock, control_sock);
1254 if (ret) {
1255 PERROR("Failed to close the control listener socket file descriptor");
1256 }
1257 control_sock->fd = -1;
1258 }
1259 lttcomm_destroy_sock(control_sock);
1260 error_sock_control:
1261 if (err) {
1262 health_error();
1263 ERR("Health error occurred in %s", __func__);
1264 }
1265 health_unregister(health_relayd);
1266 rcu_unregister_thread();
1267 DBG("Relay listener thread cleanup complete");
1268 lttng_relay_stop_threads();
1269 return NULL;
1270 }
1271
1272 /*
1273 * This thread manages the dispatching of the requests to worker threads
1274 */
1275 static void *relay_thread_dispatcher(void *data __attribute__((unused)))
1276 {
1277 int err = -1;
1278 ssize_t ret;
1279 struct cds_wfcq_node *node;
1280 struct relay_connection *new_conn = NULL;
1281
1282 DBG("[thread] Relay dispatcher started");
1283
1284 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
1285
1286 if (testpoint(relayd_thread_dispatcher)) {
1287 goto error_testpoint;
1288 }
1289
1290 health_code_update();
1291
1292 for (;;) {
1293 health_code_update();
1294
1295 /* Atomically prepare the queue futex */
1296 futex_nto1_prepare(&relay_conn_queue.futex);
1297
1298 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1299 break;
1300 }
1301
1302 do {
1303 health_code_update();
1304
1305 /* Dequeue commands */
1306 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1307 &relay_conn_queue.tail);
1308 if (node == NULL) {
1309 DBG("Woken up but nothing in the relay command queue");
1310 /* Continue thread execution */
1311 break;
1312 }
1313 new_conn = lttng::utils::container_of(node, &relay_connection::qnode);
1314
1315 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1316
1317 /*
1318 * Inform worker thread of the new request. This
1319 * call is blocking so we can be assured that
1320 * the data will be read at some point in time
1321 * or wait to the end of the world :)
1322 */
1323 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1324 if (ret < 0) {
1325 PERROR("write connection pipe");
1326 connection_put(new_conn);
1327 goto error;
1328 }
1329 } while (node != NULL);
1330
1331 /* Futex wait on queue. Blocking call on futex() */
1332 health_poll_entry();
1333 futex_nto1_wait(&relay_conn_queue.futex);
1334 health_poll_exit();
1335 }
1336
1337 /* Normal exit, no error */
1338 err = 0;
1339
1340 error:
1341 error_testpoint:
1342 if (err) {
1343 health_error();
1344 ERR("Health error occurred in %s", __func__);
1345 }
1346 health_unregister(health_relayd);
1347 DBG("Dispatch thread dying");
1348 lttng_relay_stop_threads();
1349 return NULL;
1350 }
1351
1352 static bool session_streams_have_index(const struct relay_session *session)
1353 {
1354 return session->minor >= 4 && !session->snapshot;
1355 }
1356
1357 /*
1358 * Handle the RELAYD_CREATE_SESSION command.
1359 *
1360 * On success, send back the session id or else return a negative value.
1361 */
1362 static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1363 struct relay_connection *conn,
1364 const struct lttng_buffer_view *payload)
1365 {
1366 int ret = 0;
1367 ssize_t send_ret;
1368 struct relay_session *session = NULL;
1369 struct lttcomm_relayd_create_session_reply_2_11 reply = {};
1370 char session_name[LTTNG_NAME_MAX] = {};
1371 char hostname[LTTNG_HOST_NAME_MAX] = {};
1372 uint32_t live_timer = 0;
1373 bool snapshot = false;
1374 bool session_name_contains_creation_timestamp = false;
1375 /* Left nil for peers < 2.11. */
1376 char base_path[LTTNG_PATH_MAX] = {};
1377 lttng_uuid sessiond_uuid = {};
1378 LTTNG_OPTIONAL(uint64_t) id_sessiond = {};
1379 LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
1380 LTTNG_OPTIONAL(time_t) creation_time = {};
1381 struct lttng_dynamic_buffer reply_payload;
1382
1383 lttng_dynamic_buffer_init(&reply_payload);
1384
1385 if (conn->minor < 4) {
1386 /* From 2.1 to 2.3 */
1387 ret = 0;
1388 } else if (conn->minor >= 4 && conn->minor < 11) {
1389 /* From 2.4 to 2.10 */
1390 ret = cmd_create_session_2_4(
1391 payload, session_name, hostname, &live_timer, &snapshot);
1392 } else {
1393 bool has_current_chunk;
1394 uint64_t current_chunk_id_value;
1395 time_t creation_time_value;
1396 uint64_t id_sessiond_value;
1397
1398 /* From 2.11 to ... */
1399 ret = cmd_create_session_2_11(payload,
1400 session_name,
1401 hostname,
1402 base_path,
1403 &live_timer,
1404 &snapshot,
1405 &id_sessiond_value,
1406 sessiond_uuid,
1407 &has_current_chunk,
1408 &current_chunk_id_value,
1409 &creation_time_value,
1410 &session_name_contains_creation_timestamp);
1411 if (lttng_uuid_is_nil(sessiond_uuid)) {
1412 /* The nil UUID is reserved for pre-2.11 clients. */
1413 ERR("Illegal nil UUID announced by peer in create session command");
1414 ret = -1;
1415 goto send_reply;
1416 }
1417 LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value);
1418 LTTNG_OPTIONAL_SET(&creation_time, creation_time_value);
1419 if (has_current_chunk) {
1420 LTTNG_OPTIONAL_SET(&current_chunk_id, current_chunk_id_value);
1421 }
1422 }
1423
1424 if (ret < 0) {
1425 goto send_reply;
1426 }
1427
1428 session = session_create(session_name,
1429 hostname,
1430 base_path,
1431 live_timer,
1432 snapshot,
1433 sessiond_uuid,
1434 id_sessiond.is_set ? &id_sessiond.value : NULL,
1435 current_chunk_id.is_set ? &current_chunk_id.value : NULL,
1436 creation_time.is_set ? &creation_time.value : NULL,
1437 conn->major,
1438 conn->minor,
1439 session_name_contains_creation_timestamp);
1440 if (!session) {
1441 ret = -1;
1442 goto send_reply;
1443 }
1444 LTTNG_ASSERT(!conn->session);
1445 conn->session = session;
1446 DBG("Created session %" PRIu64, session->id);
1447
1448 reply.generic.session_id = htobe64(session->id);
1449
1450 send_reply:
1451 if (ret < 0) {
1452 reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL);
1453 } else {
1454 reply.generic.ret_code = htobe32(LTTNG_OK);
1455 }
1456
1457 if (conn->minor < 11) {
1458 /* From 2.1 to 2.10 */
1459 ret = lttng_dynamic_buffer_append(
1460 &reply_payload, &reply.generic, sizeof(reply.generic));
1461 if (ret) {
1462 ERR("Failed to append \"create session\" command reply header to payload buffer");
1463 ret = -1;
1464 goto end;
1465 }
1466 } else {
1467 const uint32_t output_path_length = session ? strlen(session->output_path) + 1 : 0;
1468
1469 reply.output_path_length = htobe32(output_path_length);
1470 ret = lttng_dynamic_buffer_append(&reply_payload, &reply, sizeof(reply));
1471 if (ret) {
1472 ERR("Failed to append \"create session\" command reply header to payload buffer");
1473 goto end;
1474 }
1475
1476 if (output_path_length) {
1477 ret = lttng_dynamic_buffer_append(
1478 &reply_payload, session->output_path, output_path_length);
1479 if (ret) {
1480 ERR("Failed to append \"create session\" command reply path to payload buffer");
1481 goto end;
1482 }
1483 }
1484 }
1485
1486 send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, reply_payload.size, 0);
1487 if (send_ret < (ssize_t) reply_payload.size) {
1488 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1489 reply_payload.size,
1490 send_ret);
1491 ret = -1;
1492 }
1493 end:
1494 if (ret < 0 && session) {
1495 session_put(session);
1496 }
1497 lttng_dynamic_buffer_reset(&reply_payload);
1498 return ret;
1499 }
1500
1501 /*
1502 * When we have received all the streams and the metadata for a channel,
1503 * we make them visible to the viewer threads.
1504 */
1505 static void publish_connection_local_streams(struct relay_connection *conn)
1506 {
1507 struct relay_stream *stream;
1508 struct relay_session *session = conn->session;
1509
1510 /*
1511 * We publish all streams belonging to a session atomically wrt
1512 * session lock.
1513 */
1514 pthread_mutex_lock(&session->lock);
1515 rcu_read_lock();
1516 cds_list_for_each_entry_rcu(stream, &session->recv_list, recv_node)
1517 {
1518 stream_publish(stream);
1519 }
1520 rcu_read_unlock();
1521
1522 /*
1523 * Inform the viewer that there are new streams in the session.
1524 */
1525 if (session->viewer_attached) {
1526 uatomic_set(&session->new_streams, 1);
1527 }
1528 pthread_mutex_unlock(&session->lock);
1529 }
1530
1531 static int conform_channel_path(char *channel_path)
1532 {
1533 int ret = 0;
1534
1535 if (strstr("../", channel_path)) {
1536 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1537 channel_path);
1538 ret = -1;
1539 goto end;
1540 }
1541
1542 if (*channel_path == '/') {
1543 const size_t len = strlen(channel_path);
1544
1545 /*
1546 * Channel paths from peers prior to 2.11 are expressed as an
1547 * absolute path that is, in reality, relative to the relay
1548 * daemon's output directory. Remove the leading slash so it
1549 * is correctly interpreted as a relative path later on.
1550 *
1551 * len (and not len - 1) is used to copy the trailing NULL.
1552 */
1553 bcopy(channel_path + 1, channel_path, len);
1554 }
1555 end:
1556 return ret;
1557 }
1558
1559 /*
1560 * relay_add_stream: allocate a new stream for a session
1561 */
1562 static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1563 struct relay_connection *conn,
1564 const struct lttng_buffer_view *payload)
1565 {
1566 int ret;
1567 ssize_t send_ret;
1568 struct relay_session *session = conn->session;
1569 struct relay_stream *stream = NULL;
1570 struct lttcomm_relayd_status_stream reply;
1571 struct ctf_trace *trace = NULL;
1572 uint64_t stream_handle = -1ULL;
1573 char *path_name = NULL, *channel_name = NULL;
1574 uint64_t tracefile_size = 0, tracefile_count = 0;
1575 LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {};
1576
1577 if (!session || !conn->version_check_done) {
1578 ERR("Trying to add a stream before version check");
1579 ret = -1;
1580 goto end_no_session;
1581 }
1582
1583 if (session->minor == 1) {
1584 /* For 2.1 */
1585 ret = cmd_recv_stream_2_1(payload, &path_name, &channel_name);
1586 } else if (session->minor > 1 && session->minor < 11) {
1587 /* From 2.2 to 2.10 */
1588 ret = cmd_recv_stream_2_2(
1589 payload, &path_name, &channel_name, &tracefile_size, &tracefile_count);
1590 } else {
1591 /* From 2.11 to ... */
1592 ret = cmd_recv_stream_2_11(payload,
1593 &path_name,
1594 &channel_name,
1595 &tracefile_size,
1596 &tracefile_count,
1597 &stream_chunk_id.value);
1598 stream_chunk_id.is_set = true;
1599 }
1600
1601 if (ret < 0) {
1602 goto send_reply;
1603 }
1604
1605 if (conform_channel_path(path_name)) {
1606 goto send_reply;
1607 }
1608
1609 /*
1610 * Backward compatibility for --group-output-by-session.
1611 * Prior to lttng 2.11, the complete path is passed by the stream.
1612 * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1613 * >=2.11 the chunk is responsible for the output path. When dealing
1614 * with producer < 2.11 the chunk output_path is the root output path
1615 * and the stream carries the complete path (path_name).
1616 * To support --group-output-by-session with older producer (<2.11), we
1617 * need to craft the path based on the stream path.
1618 */
1619 if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_SESSION) {
1620 if (conn->minor < 4) {
1621 /*
1622 * From 2.1 to 2.3, the session_name is not passed on
1623 * the RELAYD_CREATE_SESSION command. The session name
1624 * is necessary to detect the presence of a base_path
1625 * inside the stream path. Without it we cannot perform
1626 * a valid group-output-by-session transformation.
1627 */
1628 WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1629 " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1630 session->id,
1631 path_name);
1632 } else if (conn->minor >= 4 && conn->minor < 11) {
1633 char *group_by_session_path_name;
1634
1635 LTTNG_ASSERT(session->session_name[0] != '\0');
1636
1637 group_by_session_path_name = backward_compat_group_by_session(
1638 path_name, session->session_name, session->creation_time.value);
1639 if (!group_by_session_path_name) {
1640 ERR("Failed to apply group by session to stream of session %" PRIu64,
1641 session->id);
1642 goto send_reply;
1643 }
1644
1645 DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1646 path_name,
1647 group_by_session_path_name);
1648
1649 free(path_name);
1650 path_name = group_by_session_path_name;
1651 }
1652 }
1653
1654 trace = ctf_trace_get_by_path_or_create(session, path_name);
1655 if (!trace) {
1656 goto send_reply;
1657 }
1658
1659 /* This stream here has one reference on the trace. */
1660 pthread_mutex_lock(&last_relay_stream_id_lock);
1661 stream_handle = ++last_relay_stream_id;
1662 pthread_mutex_unlock(&last_relay_stream_id_lock);
1663
1664 /* We pass ownership of path_name and channel_name. */
1665 stream = stream_create(
1666 trace, stream_handle, path_name, channel_name, tracefile_size, tracefile_count);
1667 path_name = NULL;
1668 channel_name = NULL;
1669
1670 /*
1671 * Streams are the owners of their trace. Reference to trace is
1672 * kept within stream_create().
1673 */
1674 ctf_trace_put(trace);
1675
1676 send_reply:
1677 memset(&reply, 0, sizeof(reply));
1678 reply.handle = htobe64(stream_handle);
1679 if (!stream) {
1680 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1681 } else {
1682 reply.ret_code = htobe32(LTTNG_OK);
1683 }
1684
1685 send_ret = conn->sock->ops->sendmsg(
1686 conn->sock, &reply, sizeof(struct lttcomm_relayd_status_stream), 0);
1687 if (send_ret < (ssize_t) sizeof(reply)) {
1688 ERR("Failed to send \"add stream\" command reply (ret = %zd)", send_ret);
1689 ret = -1;
1690 }
1691
1692 end_no_session:
1693 free(path_name);
1694 free(channel_name);
1695 return ret;
1696 }
1697
1698 /*
1699 * relay_close_stream: close a specific stream
1700 */
1701 static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1702 struct relay_connection *conn,
1703 const struct lttng_buffer_view *payload)
1704 {
1705 int ret;
1706 ssize_t send_ret;
1707 struct relay_session *session = conn->session;
1708 struct lttcomm_relayd_close_stream stream_info;
1709 struct lttcomm_relayd_generic_reply reply;
1710 struct relay_stream *stream;
1711
1712 DBG("Close stream received");
1713
1714 if (!session || !conn->version_check_done) {
1715 ERR("Trying to close a stream before version check");
1716 ret = -1;
1717 goto end_no_session;
1718 }
1719
1720 if (payload->size < sizeof(stream_info)) {
1721 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1722 sizeof(stream_info),
1723 payload->size);
1724 ret = -1;
1725 goto end_no_session;
1726 }
1727 memcpy(&stream_info, payload->data, sizeof(stream_info));
1728 stream_info.stream_id = be64toh(stream_info.stream_id);
1729 stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1730
1731 stream = stream_get_by_id(stream_info.stream_id);
1732 if (!stream) {
1733 ret = -1;
1734 goto end;
1735 }
1736
1737 /*
1738 * Set last_net_seq_num before the close flag. Required by data
1739 * pending check.
1740 */
1741 pthread_mutex_lock(&stream->lock);
1742 stream->last_net_seq_num = stream_info.last_net_seq_num;
1743 pthread_mutex_unlock(&stream->lock);
1744
1745 /*
1746 * This is one of the conditions which may trigger a stream close
1747 * with the others being:
1748 * 1) A close command is received for a stream
1749 * 2) The control connection owning the stream is closed
1750 * 3) We have received all of the stream's data _after_ a close
1751 * request.
1752 */
1753 try_stream_close(stream);
1754 stream_put(stream);
1755 ret = 0;
1756
1757 end:
1758 memset(&reply, 0, sizeof(reply));
1759 if (ret < 0) {
1760 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1761 } else {
1762 reply.ret_code = htobe32(LTTNG_OK);
1763 }
1764 send_ret = conn->sock->ops->sendmsg(
1765 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
1766 if (send_ret < (ssize_t) sizeof(reply)) {
1767 ERR("Failed to send \"close stream\" command reply (ret = %zd)", send_ret);
1768 ret = -1;
1769 }
1770
1771 end_no_session:
1772 return ret;
1773 }
1774
1775 /*
1776 * relay_reset_metadata: reset a metadata stream
1777 */
1778 static int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1779 struct relay_connection *conn,
1780 const struct lttng_buffer_view *payload)
1781 {
1782 int ret;
1783 ssize_t send_ret;
1784 struct relay_session *session = conn->session;
1785 struct lttcomm_relayd_reset_metadata stream_info;
1786 struct lttcomm_relayd_generic_reply reply;
1787 struct relay_stream *stream;
1788
1789 DBG("Reset metadata received");
1790
1791 if (!session || !conn->version_check_done) {
1792 ERR("Trying to reset a metadata stream before version check");
1793 ret = -1;
1794 goto end_no_session;
1795 }
1796
1797 if (payload->size < sizeof(stream_info)) {
1798 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1799 sizeof(stream_info),
1800 payload->size);
1801 ret = -1;
1802 goto end_no_session;
1803 }
1804 memcpy(&stream_info, payload->data, sizeof(stream_info));
1805 stream_info.stream_id = be64toh(stream_info.stream_id);
1806 stream_info.version = be64toh(stream_info.version);
1807
1808 DBG("Update metadata to version %" PRIu64, stream_info.version);
1809
1810 /* Unsupported for live sessions for now. */
1811 if (session->live_timer != 0) {
1812 ret = -1;
1813 goto end;
1814 }
1815
1816 stream = stream_get_by_id(stream_info.stream_id);
1817 if (!stream) {
1818 ret = -1;
1819 goto end;
1820 }
1821 pthread_mutex_lock(&stream->lock);
1822 if (!stream->is_metadata) {
1823 ret = -1;
1824 goto end_unlock;
1825 }
1826
1827 ret = stream_reset_file(stream);
1828 if (ret < 0) {
1829 ERR("Failed to reset metadata stream %" PRIu64 ": stream_path = %s, channel = %s",
1830 stream->stream_handle,
1831 stream->path_name,
1832 stream->channel_name);
1833 goto end_unlock;
1834 }
1835 end_unlock:
1836 pthread_mutex_unlock(&stream->lock);
1837 stream_put(stream);
1838
1839 end:
1840 memset(&reply, 0, sizeof(reply));
1841 if (ret < 0) {
1842 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1843 } else {
1844 reply.ret_code = htobe32(LTTNG_OK);
1845 }
1846 send_ret = conn->sock->ops->sendmsg(
1847 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
1848 if (send_ret < (ssize_t) sizeof(reply)) {
1849 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)", send_ret);
1850 ret = -1;
1851 }
1852
1853 end_no_session:
1854 return ret;
1855 }
1856
1857 /*
1858 * relay_unknown_command: send -1 if received unknown command
1859 */
1860 static void relay_unknown_command(struct relay_connection *conn)
1861 {
1862 struct lttcomm_relayd_generic_reply reply;
1863 ssize_t send_ret;
1864
1865 memset(&reply, 0, sizeof(reply));
1866 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1867 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1868 if (send_ret < sizeof(reply)) {
1869 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
1870 }
1871 }
1872
1873 /*
1874 * relay_start: send an acknowledgment to the client to tell if we are
1875 * ready to receive data. We are ready if a session is established.
1876 */
1877 static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1878 struct relay_connection *conn,
1879 const struct lttng_buffer_view *payload __attribute__((unused)))
1880 {
1881 int ret = 0;
1882 ssize_t send_ret;
1883 struct lttcomm_relayd_generic_reply reply;
1884 struct relay_session *session = conn->session;
1885
1886 if (!session) {
1887 DBG("Trying to start the streaming without a session established");
1888 ret = htobe32(LTTNG_ERR_UNK);
1889 }
1890
1891 memset(&reply, 0, sizeof(reply));
1892 reply.ret_code = htobe32(LTTNG_OK);
1893 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1894 if (send_ret < (ssize_t) sizeof(reply)) {
1895 ERR("Failed to send \"relay_start\" command reply (ret = %zd)", send_ret);
1896 ret = -1;
1897 }
1898
1899 return ret;
1900 }
1901
1902 /*
1903 * relay_recv_metadata: receive the metadata for the session.
1904 */
1905 static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1906 struct relay_connection *conn,
1907 const struct lttng_buffer_view *payload)
1908 {
1909 int ret = 0;
1910 struct relay_session *session = conn->session;
1911 struct lttcomm_relayd_metadata_payload metadata_payload_header;
1912 struct relay_stream *metadata_stream;
1913 uint64_t metadata_payload_size;
1914 struct lttng_buffer_view packet_view;
1915
1916 if (!session) {
1917 ERR("Metadata sent before version check");
1918 ret = -1;
1919 goto end;
1920 }
1921
1922 if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1923 ERR("Incorrect data size");
1924 ret = -1;
1925 goto end;
1926 }
1927 metadata_payload_size =
1928 recv_hdr->data_size - sizeof(struct lttcomm_relayd_metadata_payload);
1929
1930 memcpy(&metadata_payload_header, payload->data, sizeof(metadata_payload_header));
1931 metadata_payload_header.stream_id = be64toh(metadata_payload_header.stream_id);
1932 metadata_payload_header.padding_size = be32toh(metadata_payload_header.padding_size);
1933
1934 metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
1935 if (!metadata_stream) {
1936 ret = -1;
1937 goto end;
1938 }
1939
1940 packet_view = lttng_buffer_view_from_view(
1941 payload, sizeof(metadata_payload_header), metadata_payload_size);
1942 if (!lttng_buffer_view_is_valid(&packet_view)) {
1943 ERR("Invalid metadata packet length announced by header");
1944 ret = -1;
1945 goto end_put;
1946 }
1947
1948 pthread_mutex_lock(&metadata_stream->lock);
1949 ret = stream_write(metadata_stream, &packet_view, metadata_payload_header.padding_size);
1950 pthread_mutex_unlock(&metadata_stream->lock);
1951 if (ret) {
1952 ret = -1;
1953 goto end_put;
1954 }
1955 end_put:
1956 stream_put(metadata_stream);
1957 end:
1958 return ret;
1959 }
1960
1961 /*
1962 * relay_send_version: send relayd version number
1963 */
1964 static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1965 struct relay_connection *conn,
1966 const struct lttng_buffer_view *payload)
1967 {
1968 int ret;
1969 ssize_t send_ret;
1970 struct lttcomm_relayd_version reply, msg;
1971 bool compatible = true;
1972
1973 conn->version_check_done = true;
1974
1975 /* Get version from the other side. */
1976 if (payload->size < sizeof(msg)) {
1977 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1978 sizeof(msg),
1979 payload->size);
1980 ret = -1;
1981 goto end;
1982 }
1983
1984 memcpy(&msg, payload->data, sizeof(msg));
1985 msg.major = be32toh(msg.major);
1986 msg.minor = be32toh(msg.minor);
1987
1988 memset(&reply, 0, sizeof(reply));
1989 reply.major = RELAYD_VERSION_COMM_MAJOR;
1990 reply.minor = RELAYD_VERSION_COMM_MINOR;
1991
1992 /* Major versions must be the same */
1993 if (reply.major != msg.major) {
1994 DBG("Incompatible major versions (%u vs %u), deleting session",
1995 reply.major,
1996 msg.major);
1997 compatible = false;
1998 }
1999
2000 conn->major = reply.major;
2001 /* We adapt to the lowest compatible version */
2002 if (reply.minor <= msg.minor) {
2003 conn->minor = reply.minor;
2004 } else {
2005 conn->minor = msg.minor;
2006 }
2007
2008 reply.major = htobe32(reply.major);
2009 reply.minor = htobe32(reply.minor);
2010 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2011 if (send_ret < (ssize_t) sizeof(reply)) {
2012 ERR("Failed to send \"send version\" command reply (ret = %zd)", send_ret);
2013 ret = -1;
2014 goto end;
2015 } else {
2016 ret = 0;
2017 }
2018
2019 if (!compatible) {
2020 ret = -1;
2021 goto end;
2022 }
2023
2024 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
2025
2026 end:
2027 return ret;
2028 }
2029
2030 /*
2031 * Check for data pending for a given stream id from the session daemon.
2032 */
2033 static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2034 struct relay_connection *conn,
2035 const struct lttng_buffer_view *payload)
2036 {
2037 struct relay_session *session = conn->session;
2038 struct lttcomm_relayd_data_pending msg;
2039 struct lttcomm_relayd_generic_reply reply;
2040 struct relay_stream *stream;
2041 ssize_t send_ret;
2042 int ret;
2043 uint64_t stream_seq;
2044
2045 DBG("Data pending command received");
2046
2047 if (!session || !conn->version_check_done) {
2048 ERR("Trying to check for data before version check");
2049 ret = -1;
2050 goto end_no_session;
2051 }
2052
2053 if (payload->size < sizeof(msg)) {
2054 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2055 sizeof(msg),
2056 payload->size);
2057 ret = -1;
2058 goto end_no_session;
2059 }
2060 memcpy(&msg, payload->data, sizeof(msg));
2061 msg.stream_id = be64toh(msg.stream_id);
2062 msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
2063
2064 stream = stream_get_by_id(msg.stream_id);
2065 if (stream == NULL) {
2066 ret = -1;
2067 goto end;
2068 }
2069
2070 pthread_mutex_lock(&stream->lock);
2071
2072 if (session_streams_have_index(session)) {
2073 /*
2074 * Ensure that both the index and stream data have been
2075 * flushed up to the requested point.
2076 */
2077 stream_seq = std::min(stream->prev_data_seq, stream->prev_index_seq);
2078 } else {
2079 stream_seq = stream->prev_data_seq;
2080 }
2081 DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
2082 ", prev_index_seq %" PRIu64 ", and last_seq %" PRIu64,
2083 msg.stream_id,
2084 stream->prev_data_seq,
2085 stream->prev_index_seq,
2086 msg.last_net_seq_num);
2087
2088 /* Avoid wrapping issue */
2089 if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
2090 /* Data has in fact been written and is NOT pending */
2091 ret = 0;
2092 } else {
2093 /* Data still being streamed thus pending */
2094 ret = 1;
2095 }
2096
2097 stream->data_pending_check_done = true;
2098 pthread_mutex_unlock(&stream->lock);
2099
2100 stream_put(stream);
2101 end:
2102
2103 memset(&reply, 0, sizeof(reply));
2104 reply.ret_code = htobe32(ret);
2105 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2106 if (send_ret < (ssize_t) sizeof(reply)) {
2107 ERR("Failed to send \"data pending\" command reply (ret = %zd)", send_ret);
2108 ret = -1;
2109 }
2110
2111 end_no_session:
2112 return ret;
2113 }
2114
2115 /*
2116 * Wait for the control socket to reach a quiescent state.
2117 *
2118 * Note that for now, when receiving this command from the session
2119 * daemon, this means that every subsequent commands or data received on
2120 * the control socket has been handled. So, this is why we simply return
2121 * OK here.
2122 */
2123 static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr
2124 __attribute__((unused)),
2125 struct relay_connection *conn,
2126 const struct lttng_buffer_view *payload)
2127 {
2128 int ret;
2129 ssize_t send_ret;
2130 struct relay_stream *stream;
2131 struct lttcomm_relayd_quiescent_control msg;
2132 struct lttcomm_relayd_generic_reply reply;
2133
2134 DBG("Checking quiescent state on control socket");
2135
2136 if (!conn->session || !conn->version_check_done) {
2137 ERR("Trying to check for data before version check");
2138 ret = -1;
2139 goto end_no_session;
2140 }
2141
2142 if (payload->size < sizeof(msg)) {
2143 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2144 sizeof(msg),
2145 payload->size);
2146 ret = -1;
2147 goto end_no_session;
2148 }
2149 memcpy(&msg, payload->data, sizeof(msg));
2150 msg.stream_id = be64toh(msg.stream_id);
2151
2152 stream = stream_get_by_id(msg.stream_id);
2153 if (!stream) {
2154 goto reply;
2155 }
2156 pthread_mutex_lock(&stream->lock);
2157 stream->data_pending_check_done = true;
2158 pthread_mutex_unlock(&stream->lock);
2159
2160 DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
2161 stream_put(stream);
2162 reply:
2163 memset(&reply, 0, sizeof(reply));
2164 reply.ret_code = htobe32(LTTNG_OK);
2165 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2166 if (send_ret < (ssize_t) sizeof(reply)) {
2167 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)", send_ret);
2168 ret = -1;
2169 } else {
2170 ret = 0;
2171 }
2172
2173 end_no_session:
2174 return ret;
2175 }
2176
2177 /*
2178 * Initialize a data pending command. This means that a consumer is about
2179 * to ask for data pending for each stream it holds. Simply iterate over
2180 * all streams of a session and set the data_pending_check_done flag.
2181 *
2182 * This command returns to the client a LTTNG_OK code.
2183 */
2184 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2185 struct relay_connection *conn,
2186 const struct lttng_buffer_view *payload)
2187 {
2188 int ret;
2189 ssize_t send_ret;
2190 struct lttng_ht_iter iter;
2191 struct lttcomm_relayd_begin_data_pending msg;
2192 struct lttcomm_relayd_generic_reply reply;
2193 struct relay_stream *stream;
2194
2195 LTTNG_ASSERT(recv_hdr);
2196 LTTNG_ASSERT(conn);
2197
2198 DBG("Init streams for data pending");
2199
2200 if (!conn->session || !conn->version_check_done) {
2201 ERR("Trying to check for data before version check");
2202 ret = -1;
2203 goto end_no_session;
2204 }
2205
2206 if (payload->size < sizeof(msg)) {
2207 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2208 sizeof(msg),
2209 payload->size);
2210 ret = -1;
2211 goto end_no_session;
2212 }
2213 memcpy(&msg, payload->data, sizeof(msg));
2214 msg.session_id = be64toh(msg.session_id);
2215
2216 /*
2217 * Iterate over all streams to set the begin data pending flag.
2218 * For now, the streams are indexed by stream handle so we have
2219 * to iterate over all streams to find the one associated with
2220 * the right session_id.
2221 */
2222 rcu_read_lock();
2223 cds_lfht_for_each_entry (relay_streams_ht->ht, &iter.iter, stream, node.node) {
2224 if (!stream_get(stream)) {
2225 continue;
2226 }
2227 if (stream->trace->session->id == msg.session_id) {
2228 pthread_mutex_lock(&stream->lock);
2229 stream->data_pending_check_done = false;
2230 pthread_mutex_unlock(&stream->lock);
2231 DBG("Set begin data pending flag to stream %" PRIu64,
2232 stream->stream_handle);
2233 }
2234 stream_put(stream);
2235 }
2236 rcu_read_unlock();
2237
2238 memset(&reply, 0, sizeof(reply));
2239 /* All good, send back reply. */
2240 reply.ret_code = htobe32(LTTNG_OK);
2241
2242 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2243 if (send_ret < (ssize_t) sizeof(reply)) {
2244 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)", send_ret);
2245 ret = -1;
2246 } else {
2247 ret = 0;
2248 }
2249
2250 end_no_session:
2251 return ret;
2252 }
2253
2254 /*
2255 * End data pending command. This will check, for a given session id, if
2256 * each stream associated with it has its data_pending_check_done flag
2257 * set. If not, this means that the client lost track of the stream but
2258 * the data is still being streamed on our side. In this case, we inform
2259 * the client that data is in flight.
2260 *
2261 * Return to the client if there is data in flight or not with a ret_code.
2262 */
2263 static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2264 struct relay_connection *conn,
2265 const struct lttng_buffer_view *payload)
2266 {
2267 int ret;
2268 ssize_t send_ret;
2269 struct lttng_ht_iter iter;
2270 struct lttcomm_relayd_end_data_pending msg;
2271 struct lttcomm_relayd_generic_reply reply;
2272 struct relay_stream *stream;
2273 uint32_t is_data_inflight = 0;
2274
2275 DBG("End data pending command");
2276
2277 if (!conn->session || !conn->version_check_done) {
2278 ERR("Trying to check for data before version check");
2279 ret = -1;
2280 goto end_no_session;
2281 }
2282
2283 if (payload->size < sizeof(msg)) {
2284 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2285 sizeof(msg),
2286 payload->size);
2287 ret = -1;
2288 goto end_no_session;
2289 }
2290 memcpy(&msg, payload->data, sizeof(msg));
2291 msg.session_id = be64toh(msg.session_id);
2292
2293 /*
2294 * Iterate over all streams to see if the begin data pending
2295 * flag is set.
2296 */
2297 rcu_read_lock();
2298 cds_lfht_for_each_entry (relay_streams_ht->ht, &iter.iter, stream, node.node) {
2299 if (!stream_get(stream)) {
2300 continue;
2301 }
2302 if (stream->trace->session->id != msg.session_id) {
2303 stream_put(stream);
2304 continue;
2305 }
2306 pthread_mutex_lock(&stream->lock);
2307 if (!stream->data_pending_check_done) {
2308 uint64_t stream_seq;
2309
2310 if (session_streams_have_index(conn->session)) {
2311 /*
2312 * Ensure that both the index and stream data have been
2313 * flushed up to the requested point.
2314 */
2315 stream_seq =
2316 std::min(stream->prev_data_seq, stream->prev_index_seq);
2317 } else {
2318 stream_seq = stream->prev_data_seq;
2319 }
2320 if (!stream->closed ||
2321 !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
2322 is_data_inflight = 1;
2323 DBG("Data is still in flight for stream %" PRIu64,
2324 stream->stream_handle);
2325 pthread_mutex_unlock(&stream->lock);
2326 stream_put(stream);
2327 break;
2328 }
2329 }
2330 pthread_mutex_unlock(&stream->lock);
2331 stream_put(stream);
2332 }
2333 rcu_read_unlock();
2334
2335 memset(&reply, 0, sizeof(reply));
2336 /* All good, send back reply. */
2337 reply.ret_code = htobe32(is_data_inflight);
2338
2339 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2340 if (send_ret < (ssize_t) sizeof(reply)) {
2341 ERR("Failed to send \"end data pending\" command reply (ret = %zd)", send_ret);
2342 ret = -1;
2343 } else {
2344 ret = 0;
2345 }
2346
2347 end_no_session:
2348 return ret;
2349 }
2350
2351 /*
2352 * Receive an index for a specific stream.
2353 *
2354 * Return 0 on success else a negative value.
2355 */
2356 static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2357 struct relay_connection *conn,
2358 const struct lttng_buffer_view *payload)
2359 {
2360 int ret;
2361 ssize_t send_ret;
2362 struct relay_session *session = conn->session;
2363 struct lttcomm_relayd_index index_info;
2364 struct lttcomm_relayd_generic_reply reply;
2365 struct relay_stream *stream;
2366 size_t msg_len;
2367
2368 LTTNG_ASSERT(conn);
2369
2370 DBG("Relay receiving index");
2371
2372 if (!session || !conn->version_check_done) {
2373 ERR("Trying to close a stream before version check");
2374 ret = -1;
2375 goto end_no_session;
2376 }
2377
2378 msg_len = lttcomm_relayd_index_len(lttng_to_index_major(conn->major, conn->minor),
2379 lttng_to_index_minor(conn->major, conn->minor));
2380 if (payload->size < msg_len) {
2381 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2382 msg_len,
2383 payload->size);
2384 ret = -1;
2385 goto end_no_session;
2386 }
2387 memcpy(&index_info, payload->data, msg_len);
2388 index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2389 index_info.net_seq_num = be64toh(index_info.net_seq_num);
2390 index_info.packet_size = be64toh(index_info.packet_size);
2391 index_info.content_size = be64toh(index_info.content_size);
2392 index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2393 index_info.timestamp_end = be64toh(index_info.timestamp_end);
2394 index_info.events_discarded = be64toh(index_info.events_discarded);
2395 index_info.stream_id = be64toh(index_info.stream_id);
2396
2397 if (conn->minor >= 8) {
2398 index_info.stream_instance_id = be64toh(index_info.stream_instance_id);
2399 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2400 } else {
2401 index_info.stream_instance_id = -1ULL;
2402 index_info.packet_seq_num = -1ULL;
2403 }
2404
2405 stream = stream_get_by_id(index_info.relay_stream_id);
2406 if (!stream) {
2407 ERR("stream_get_by_id not found");
2408 ret = -1;
2409 goto end;
2410 }
2411
2412 pthread_mutex_lock(&stream->lock);
2413 ret = stream_add_index(stream, &index_info);
2414 pthread_mutex_unlock(&stream->lock);
2415 if (ret) {
2416 goto end_stream_put;
2417 }
2418
2419 end_stream_put:
2420 stream_put(stream);
2421 end:
2422 memset(&reply, 0, sizeof(reply));
2423 if (ret < 0) {
2424 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2425 } else {
2426 reply.ret_code = htobe32(LTTNG_OK);
2427 }
2428 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2429 if (send_ret < (ssize_t) sizeof(reply)) {
2430 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2431 ret = -1;
2432 }
2433
2434 end_no_session:
2435 return ret;
2436 }
2437
2438 /*
2439 * Receive the streams_sent message.
2440 *
2441 * Return 0 on success else a negative value.
2442 */
2443 static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2444 struct relay_connection *conn,
2445 const struct lttng_buffer_view *payload __attribute__((unused)))
2446 {
2447 int ret;
2448 ssize_t send_ret;
2449 struct lttcomm_relayd_generic_reply reply;
2450
2451 LTTNG_ASSERT(conn);
2452
2453 DBG("Relay receiving streams_sent");
2454
2455 if (!conn->session || !conn->version_check_done) {
2456 ERR("Trying to close a stream before version check");
2457 ret = -1;
2458 goto end_no_session;
2459 }
2460
2461 /*
2462 * Publish every pending stream in the connection recv list which are
2463 * now ready to be used by the viewer.
2464 */
2465 publish_connection_local_streams(conn);
2466
2467 memset(&reply, 0, sizeof(reply));
2468 reply.ret_code = htobe32(LTTNG_OK);
2469 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2470 if (send_ret < (ssize_t) sizeof(reply)) {
2471 ERR("Failed to send \"streams sent\" command reply (ret = %zd)", send_ret);
2472 ret = -1;
2473 } else {
2474 /* Success. */
2475 ret = 0;
2476 }
2477
2478 end_no_session:
2479 return ret;
2480 }
2481
2482 static ssize_t
2483 relay_unpack_rotate_streams_header(const struct lttng_buffer_view *payload,
2484 struct lttcomm_relayd_rotate_streams *_rotate_streams)
2485 {
2486 struct lttcomm_relayd_rotate_streams rotate_streams;
2487 /*
2488 * Set to the smallest version (packed) of `lttcomm_relayd_rotate_streams`.
2489 * This is the smallest version of this structure, but it can be larger;
2490 * this variable is updated once the proper size of the structure is known.
2491 *
2492 * See comment at the declaration of this structure for more information.
2493 */
2494 ssize_t header_len = sizeof(struct lttcomm_relayd_rotate_streams_packed);
2495 size_t expected_payload_size_no_padding, expected_payload_size_3_bytes_padding,
2496 expected_payload_size_7_bytes_padding;
2497
2498 if (payload->size < header_len) {
2499 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2500 header_len,
2501 payload->size);
2502 goto error;
2503 }
2504
2505 /*
2506 * Some versions incorrectly omitted the LTTNG_PACKED annotation on the
2507 * `new_chunk_id` optional field of struct lttcomm_relayd_rotate_streams.
2508 *
2509 * We start by "unpacking" `stream_count` to figure out the padding length
2510 * emited by our peer.
2511 */
2512 {
2513 decltype(rotate_streams.stream_count) stream_count;
2514
2515 memcpy(&stream_count, payload->data, sizeof(stream_count));
2516 rotate_streams.stream_count = be32toh(stream_count);
2517 }
2518
2519 rotate_streams.new_chunk_id = LTTNG_OPTIONAL_INIT_UNSET;
2520
2521 /*
2522 * Payload size expected given the possible padding lengths in
2523 * `struct lttcomm_relayd_rotate_streams`.
2524 */
2525 expected_payload_size_no_padding =
2526 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2527 sizeof(lttcomm_relayd_rotate_streams_packed);
2528 expected_payload_size_3_bytes_padding =
2529 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2530 sizeof(lttcomm_relayd_rotate_streams_3_bytes_padding);
2531 expected_payload_size_7_bytes_padding =
2532 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2533 sizeof(lttcomm_relayd_rotate_streams_7_bytes_padding);
2534
2535 if (payload->size == expected_payload_size_no_padding) {
2536 struct lttcomm_relayd_rotate_streams_packed packed_rotate_streams;
2537
2538 /*
2539 * This handles cases where someone might build with
2540 * -fpack-struct or any other toolchain that wouldn't produce
2541 * padding to align `value`.
2542 */
2543 DBG("Received `struct lttcomm_relayd_rotate_streams` with no padding");
2544
2545 header_len = sizeof(packed_rotate_streams);
2546 memcpy(&packed_rotate_streams, payload->data, header_len);
2547
2548 /* Unpack the packed structure to the natively-packed version. */
2549 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2550 .is_set = !!packed_rotate_streams.new_chunk_id.is_set,
2551 .value = be64toh(packed_rotate_streams.new_chunk_id.value),
2552 };
2553 _rotate_streams->stream_count = be32toh(packed_rotate_streams.stream_count);
2554 } else if (payload->size == expected_payload_size_3_bytes_padding) {
2555 struct lttcomm_relayd_rotate_streams_3_bytes_padding padded_rotate_streams;
2556
2557 DBG("Received `struct lttcomm_relayd_rotate_streams` with 3 bytes of padding (4-byte aligned peer)");
2558
2559 header_len = sizeof(padded_rotate_streams);
2560 memcpy(&padded_rotate_streams, payload->data, header_len);
2561
2562 /* Unpack the 3-byte padded structure to the natively-packed version. */
2563 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2564 .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2565 .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2566 };
2567 _rotate_streams->stream_count = be32toh(padded_rotate_streams.stream_count);
2568 } else if (payload->size == expected_payload_size_7_bytes_padding) {
2569 struct lttcomm_relayd_rotate_streams_7_bytes_padding padded_rotate_streams;
2570
2571 DBG("Received `struct lttcomm_relayd_rotate_streams` with 7 bytes of padding (8-byte aligned peer)");
2572
2573 header_len = sizeof(padded_rotate_streams);
2574 memcpy(&padded_rotate_streams, payload->data, header_len);
2575
2576 /* Unpack the 7-byte padded structure to the natively-packed version. */
2577 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2578 .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2579 .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2580 };
2581 _rotate_streams->stream_count = be32toh(padded_rotate_streams.stream_count);
2582
2583 header_len = sizeof(padded_rotate_streams);
2584 } else {
2585 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected %zu, %zu or %zu bytes, got %zu bytes",
2586 expected_payload_size_no_padding,
2587 expected_payload_size_3_bytes_padding,
2588 expected_payload_size_7_bytes_padding,
2589 payload->size);
2590 goto error;
2591 }
2592
2593 return header_len;
2594 error:
2595 return -1;
2596 }
2597
2598 /*
2599 * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2600 * session rotation feature (not the tracefile rotation feature).
2601 */
2602 static int relay_rotate_session_streams(const struct lttcomm_relayd_hdr *recv_hdr
2603 __attribute__((unused)),
2604 struct relay_connection *conn,
2605 const struct lttng_buffer_view *payload)
2606 {
2607 int ret = 0;
2608 uint32_t i;
2609 ssize_t send_ret;
2610 enum lttng_error_code reply_code = LTTNG_ERR_UNK;
2611 struct relay_session *session = conn->session;
2612 struct lttcomm_relayd_rotate_streams rotate_streams;
2613 struct lttcomm_relayd_generic_reply reply = {};
2614 struct relay_stream *stream = NULL;
2615 struct lttng_trace_chunk *next_trace_chunk = NULL;
2616 struct lttng_buffer_view stream_positions;
2617 char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
2618 const char *chunk_id_str = "none";
2619 ssize_t header_len;
2620
2621 if (!session || !conn->version_check_done) {
2622 ERR("Trying to rotate a stream before version check");
2623 ret = -1;
2624 goto end_no_reply;
2625 }
2626
2627 if (session->major == 2 && session->minor < 11) {
2628 ERR("Unsupported feature before 2.11");
2629 ret = -1;
2630 goto end_no_reply;
2631 }
2632
2633 header_len = relay_unpack_rotate_streams_header(payload, &rotate_streams);
2634 if (header_len < 0) {
2635 ret = -1;
2636 goto end_no_reply;
2637 }
2638
2639 if (rotate_streams.new_chunk_id.is_set) {
2640 /*
2641 * Retrieve the trace chunk the stream must transition to. As
2642 * per the protocol, this chunk should have been created
2643 * before this command is received.
2644 */
2645 next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
2646 sessiond_trace_chunk_registry,
2647 session->sessiond_uuid,
2648 conn->session->id_sessiond.is_set ? conn->session->id_sessiond.value :
2649 conn->session->id,
2650 rotate_streams.new_chunk_id.value);
2651 if (!next_trace_chunk) {
2652 char uuid_str[LTTNG_UUID_STR_LEN];
2653
2654 lttng_uuid_to_str(session->sessiond_uuid, uuid_str);
2655 ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2656 ", trace_chunk_id = %" PRIu64,
2657 uuid_str,
2658 session->id,
2659 rotate_streams.new_chunk_id.value);
2660 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2661 ret = -1;
2662 goto end;
2663 }
2664
2665 ret = snprintf(chunk_id_buf,
2666 sizeof(chunk_id_buf),
2667 "%" PRIu64,
2668 rotate_streams.new_chunk_id.value);
2669 if (ret < 0 || ret >= sizeof(chunk_id_buf)) {
2670 chunk_id_str = "formatting error";
2671 } else {
2672 chunk_id_str = chunk_id_buf;
2673 }
2674 }
2675
2676 DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
2677 rotate_streams.stream_count,
2678 session->session_name,
2679 chunk_id_str);
2680
2681 stream_positions = lttng_buffer_view_from_view(payload, header_len, -1);
2682 if (!stream_positions.data ||
2683 stream_positions.size < (rotate_streams.stream_count *
2684 sizeof(struct lttcomm_relayd_stream_rotation_position))) {
2685 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2686 ret = -1;
2687 goto end;
2688 }
2689
2690 for (i = 0; i < rotate_streams.stream_count; i++) {
2691 struct lttcomm_relayd_stream_rotation_position *position_comm =
2692 &((typeof(position_comm)) stream_positions.data)[i];
2693 const struct lttcomm_relayd_stream_rotation_position pos = {
2694 .stream_id = be64toh(position_comm->stream_id),
2695 .rotate_at_seq_num = be64toh(position_comm->rotate_at_seq_num),
2696 };
2697
2698 stream = stream_get_by_id(pos.stream_id);
2699 if (!stream) {
2700 reply_code = LTTNG_ERR_INVALID;
2701 ret = -1;
2702 goto end;
2703 }
2704
2705 pthread_mutex_lock(&stream->lock);
2706 ret = stream_set_pending_rotation(stream, next_trace_chunk, pos.rotate_at_seq_num);
2707 pthread_mutex_unlock(&stream->lock);
2708 if (ret) {
2709 reply_code = LTTNG_ERR_FILE_CREATION_ERROR;
2710 goto end;
2711 }
2712
2713 stream_put(stream);
2714 stream = NULL;
2715 }
2716
2717 reply_code = LTTNG_OK;
2718 ret = 0;
2719 end:
2720 if (stream) {
2721 stream_put(stream);
2722 }
2723
2724 reply.ret_code = htobe32((uint32_t) reply_code);
2725 send_ret = conn->sock->ops->sendmsg(
2726 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
2727 if (send_ret < (ssize_t) sizeof(reply)) {
2728 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)", send_ret);
2729 ret = -1;
2730 }
2731 end_no_reply:
2732 lttng_trace_chunk_put(next_trace_chunk);
2733 return ret;
2734 }
2735
2736 /*
2737 * relay_create_trace_chunk: create a new trace chunk
2738 */
2739 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr
2740 __attribute__((unused)),
2741 struct relay_connection *conn,
2742 const struct lttng_buffer_view *payload)
2743 {
2744 int ret = 0;
2745 ssize_t send_ret;
2746 struct relay_session *session = conn->session;
2747 struct lttcomm_relayd_create_trace_chunk *msg;
2748 struct lttcomm_relayd_generic_reply reply = {};
2749 struct lttng_buffer_view header_view;
2750 struct lttng_trace_chunk *chunk = NULL, *published_chunk = NULL;
2751 enum lttng_error_code reply_code = LTTNG_OK;
2752 enum lttng_trace_chunk_status chunk_status;
2753 const char *new_path;
2754
2755 if (!session || !conn->version_check_done) {
2756 ERR("Trying to create a trace chunk before version check");
2757 ret = -1;
2758 goto end_no_reply;
2759 }
2760
2761 if (session->major == 2 && session->minor < 11) {
2762 ERR("Chunk creation command is unsupported before 2.11");
2763 ret = -1;
2764 goto end_no_reply;
2765 }
2766
2767 header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2768 if (!lttng_buffer_view_is_valid(&header_view)) {
2769 ERR("Failed to receive payload of chunk creation command");
2770 ret = -1;
2771 goto end_no_reply;
2772 }
2773
2774 /* Convert to host endianness. */
2775 msg = (typeof(msg)) header_view.data;
2776 msg->chunk_id = be64toh(msg->chunk_id);
2777 msg->creation_timestamp = be64toh(msg->creation_timestamp);
2778 msg->override_name_length = be32toh(msg->override_name_length);
2779
2780 pthread_mutex_lock(&conn->session->lock);
2781 session->ongoing_rotation = true;
2782 if (session->current_trace_chunk &&
2783 !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2784 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2785 DEFAULT_CHUNK_TMP_OLD_DIRECTORY);
2786 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2787 ERR("Failed to rename old chunk");
2788 ret = -1;
2789 reply_code = LTTNG_ERR_UNK;
2790 goto end;
2791 }
2792 }
2793 if (!session->current_trace_chunk) {
2794 if (!session->has_rotated) {
2795 new_path = "";
2796 } else {
2797 new_path = NULL;
2798 }
2799 } else {
2800 new_path = DEFAULT_CHUNK_TMP_NEW_DIRECTORY;
2801 }
2802 chunk = lttng_trace_chunk_create(msg->chunk_id, msg->creation_timestamp, new_path);
2803 if (!chunk) {
2804 ERR("Failed to create trace chunk in trace chunk creation command");
2805 ret = -1;
2806 reply_code = LTTNG_ERR_NOMEM;
2807 goto end;
2808 }
2809 lttng_trace_chunk_set_fd_tracker(chunk, the_fd_tracker);
2810
2811 if (msg->override_name_length) {
2812 const char *name;
2813 const struct lttng_buffer_view chunk_name_view = lttng_buffer_view_from_view(
2814 payload, sizeof(*msg), msg->override_name_length);
2815
2816 if (!lttng_buffer_view_is_valid(&chunk_name_view)) {
2817 ERR("Invalid payload of chunk creation command (protocol error): buffer too short for expected name length");
2818 ret = -1;
2819 reply_code = LTTNG_ERR_INVALID;
2820 goto end;
2821 }
2822
2823 name = chunk_name_view.data;
2824 if (name[msg->override_name_length - 1]) {
2825 ERR("Invalid payload of chunk creation command (protocol error): name is not null-terminated");
2826 ret = -1;
2827 reply_code = LTTNG_ERR_INVALID;
2828 goto end;
2829 }
2830
2831 chunk_status = lttng_trace_chunk_override_name(chunk, chunk_name_view.data);
2832 switch (chunk_status) {
2833 case LTTNG_TRACE_CHUNK_STATUS_OK:
2834 break;
2835 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
2836 ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2837 reply_code = LTTNG_ERR_INVALID;
2838 ret = -1;
2839 goto end;
2840 default:
2841 ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2842 reply_code = LTTNG_ERR_UNK;
2843 ret = -1;
2844 goto end;
2845 }
2846 }
2847
2848 chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk);
2849 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2850 reply_code = LTTNG_ERR_UNK;
2851 ret = -1;
2852 goto end;
2853 }
2854
2855 LTTNG_ASSERT(conn->session->output_directory);
2856 chunk_status = lttng_trace_chunk_set_as_owner(chunk, conn->session->output_directory);
2857 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2858 reply_code = LTTNG_ERR_UNK;
2859 ret = -1;
2860 goto end;
2861 }
2862
2863 published_chunk = sessiond_trace_chunk_registry_publish_chunk(
2864 sessiond_trace_chunk_registry,
2865 conn->session->sessiond_uuid,
2866 conn->session->id_sessiond.is_set ? conn->session->id_sessiond.value :
2867 conn->session->id,
2868 chunk);
2869 if (!published_chunk) {
2870 char uuid_str[LTTNG_UUID_STR_LEN];
2871
2872 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2873 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
2874 ", chunk_id = %" PRIu64,
2875 uuid_str,
2876 conn->session->id,
2877 msg->chunk_id);
2878 ret = -1;
2879 reply_code = LTTNG_ERR_NOMEM;
2880 goto end;
2881 }
2882
2883 if (conn->session->pending_closure_trace_chunk) {
2884 /*
2885 * Invalid; this means a second create_trace_chunk command was
2886 * received before a close_trace_chunk.
2887 */
2888 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2889 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2890 ret = -1;
2891 goto end;
2892 }
2893 conn->session->pending_closure_trace_chunk = conn->session->current_trace_chunk;
2894 conn->session->current_trace_chunk = published_chunk;
2895 published_chunk = NULL;
2896 if (!conn->session->pending_closure_trace_chunk) {
2897 session->ongoing_rotation = false;
2898 }
2899 end:
2900 pthread_mutex_unlock(&conn->session->lock);
2901 reply.ret_code = htobe32((uint32_t) reply_code);
2902 send_ret = conn->sock->ops->sendmsg(
2903 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
2904 if (send_ret < (ssize_t) sizeof(reply)) {
2905 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", send_ret);
2906 ret = -1;
2907 }
2908 end_no_reply:
2909 lttng_trace_chunk_put(chunk);
2910 lttng_trace_chunk_put(published_chunk);
2911 return ret;
2912 }
2913
2914 /*
2915 * relay_close_trace_chunk: close a trace chunk
2916 */
2917 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr
2918 __attribute__((unused)),
2919 struct relay_connection *conn,
2920 const struct lttng_buffer_view *payload)
2921 {
2922 int ret = 0, buf_ret;
2923 ssize_t send_ret;
2924 struct relay_session *session = conn->session;
2925 struct lttcomm_relayd_close_trace_chunk *msg;
2926 struct lttcomm_relayd_close_trace_chunk_reply reply = {};
2927 struct lttng_buffer_view header_view;
2928 struct lttng_trace_chunk *chunk = NULL;
2929 enum lttng_error_code reply_code = LTTNG_OK;
2930 enum lttng_trace_chunk_status chunk_status;
2931 uint64_t chunk_id;
2932 LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {};
2933 time_t close_timestamp;
2934 char closed_trace_chunk_path[LTTNG_PATH_MAX];
2935 size_t path_length = 0;
2936 const char *chunk_name = NULL;
2937 struct lttng_dynamic_buffer reply_payload;
2938 const char *new_path;
2939
2940 lttng_dynamic_buffer_init(&reply_payload);
2941
2942 if (!session || !conn->version_check_done) {
2943 ERR("Trying to close a trace chunk before version check");
2944 ret = -1;
2945 goto end_no_reply;
2946 }
2947
2948 if (session->major == 2 && session->minor < 11) {
2949 ERR("Chunk close command is unsupported before 2.11");
2950 ret = -1;
2951 goto end_no_reply;
2952 }
2953
2954 header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2955 if (!lttng_buffer_view_is_valid(&header_view)) {
2956 ERR("Failed to receive payload of chunk close command");
2957 ret = -1;
2958 goto end_no_reply;
2959 }
2960
2961 /* Convert to host endianness. */
2962 msg = (typeof(msg)) header_view.data;
2963 chunk_id = be64toh(msg->chunk_id);
2964 close_timestamp = (time_t) be64toh(msg->close_timestamp);
2965 close_command.value = (lttng_trace_chunk_command_type) be32toh(msg->close_command.value);
2966 close_command.is_set = msg->close_command.is_set;
2967
2968 chunk = sessiond_trace_chunk_registry_get_chunk(sessiond_trace_chunk_registry,
2969 conn->session->sessiond_uuid,
2970 conn->session->id_sessiond.is_set ?
2971 conn->session->id_sessiond.value :
2972 conn->session->id,
2973 chunk_id);
2974 if (!chunk) {
2975 char uuid_str[LTTNG_UUID_STR_LEN];
2976
2977 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2978 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
2979 ", chunk_id = %" PRIu64,
2980 uuid_str,
2981 conn->session->id,
2982 msg->chunk_id);
2983 ret = -1;
2984 reply_code = LTTNG_ERR_NOMEM;
2985 goto end;
2986 }
2987
2988 pthread_mutex_lock(&session->lock);
2989 if (close_command.is_set && close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) {
2990 /*
2991 * Clear command. It is a protocol error to ask for a
2992 * clear on a relay which does not allow it. Querying
2993 * the configuration allows figuring out whether
2994 * clearing is allowed before doing the clear.
2995 */
2996 if (!opt_allow_clear) {
2997 ret = -1;
2998 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2999 goto end_unlock_session;
3000 }
3001 }
3002 if (session->pending_closure_trace_chunk && session->pending_closure_trace_chunk != chunk) {
3003 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
3004 session->session_name);
3005 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3006 ret = -1;
3007 goto end_unlock_session;
3008 }
3009
3010 if (session->current_trace_chunk && session->current_trace_chunk != chunk &&
3011 !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
3012 if (close_command.is_set &&
3013 close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE &&
3014 !session->has_rotated) {
3015 /* New chunk stays in session output directory. */
3016 new_path = "";
3017 } else {
3018 /* Use chunk name for new chunk. */
3019 new_path = NULL;
3020 }
3021 /* Rename new chunk path. */
3022 chunk_status =
3023 lttng_trace_chunk_rename_path(session->current_trace_chunk, new_path);
3024 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3025 ret = -1;
3026 goto end_unlock_session;
3027 }
3028 session->ongoing_rotation = false;
3029 }
3030 if ((!close_command.is_set ||
3031 close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION) &&
3032 !lttng_trace_chunk_get_name_overridden(chunk)) {
3033 const char *old_path;
3034
3035 if (!session->has_rotated) {
3036 old_path = "";
3037 } else {
3038 old_path = NULL;
3039 }
3040 /* We need to move back the .tmp_old_chunk to its rightful place. */
3041 chunk_status = lttng_trace_chunk_rename_path(chunk, old_path);
3042 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3043 ret = -1;
3044 goto end_unlock_session;
3045 }
3046 }
3047 chunk_status = lttng_trace_chunk_set_close_timestamp(chunk, close_timestamp);
3048 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3049 ERR("Failed to set trace chunk close timestamp");
3050 ret = -1;
3051 reply_code = LTTNG_ERR_UNK;
3052 goto end_unlock_session;
3053 }
3054
3055 if (close_command.is_set) {
3056 chunk_status = lttng_trace_chunk_set_close_command(chunk, close_command.value);
3057 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3058 ret = -1;
3059 reply_code = LTTNG_ERR_INVALID;
3060 goto end_unlock_session;
3061 }
3062 }
3063 chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, NULL);
3064 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3065 ERR("Failed to get chunk name");
3066 ret = -1;
3067 reply_code = LTTNG_ERR_UNK;
3068 goto end_unlock_session;
3069 }
3070 if (!session->has_rotated && !session->snapshot) {
3071 ret = lttng_strncpy(closed_trace_chunk_path,
3072 session->output_path,
3073 sizeof(closed_trace_chunk_path));
3074 if (ret) {
3075 ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
3076 strlen(session->output_path),
3077 sizeof(closed_trace_chunk_path));
3078 reply_code = LTTNG_ERR_NOMEM;
3079 ret = -1;
3080 goto end_unlock_session;
3081 }
3082 } else {
3083 if (session->snapshot) {
3084 ret = snprintf(closed_trace_chunk_path,
3085 sizeof(closed_trace_chunk_path),
3086 "%s/%s",
3087 session->output_path,
3088 chunk_name);
3089 } else {
3090 ret = snprintf(closed_trace_chunk_path,
3091 sizeof(closed_trace_chunk_path),
3092 "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY "/%s",
3093 session->output_path,
3094 chunk_name);
3095 }
3096 if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) {
3097 ERR("Failed to format closed trace chunk resulting path");
3098 reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM;
3099 ret = -1;
3100 goto end_unlock_session;
3101 }
3102 }
3103 if (close_command.is_set &&
3104 close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) {
3105 session->has_rotated = true;
3106 }
3107 DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
3108 path_length = strlen(closed_trace_chunk_path) + 1;
3109 if (path_length > UINT32_MAX) {
3110 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3111 ret = -1;
3112 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3113 goto end_unlock_session;
3114 }
3115
3116 if (session->current_trace_chunk == chunk) {
3117 /*
3118 * After a trace chunk close command, no new streams
3119 * referencing the chunk may be created. Hence, on the
3120 * event that no new trace chunk have been created for
3121 * the session, the reference to the current trace chunk
3122 * is released in order to allow it to be reclaimed when
3123 * the last stream releases its reference to it.
3124 */
3125 lttng_trace_chunk_put(session->current_trace_chunk);
3126 session->current_trace_chunk = NULL;
3127 }
3128 lttng_trace_chunk_put(session->pending_closure_trace_chunk);
3129 session->pending_closure_trace_chunk = NULL;
3130 end_unlock_session:
3131 pthread_mutex_unlock(&session->lock);
3132
3133 end:
3134 reply.generic.ret_code = htobe32((uint32_t) reply_code);
3135 reply.path_length = htobe32((uint32_t) path_length);
3136 buf_ret = lttng_dynamic_buffer_append(&reply_payload, &reply, sizeof(reply));
3137 if (buf_ret) {
3138 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3139 goto end_no_reply;
3140 }
3141
3142 if (reply_code == LTTNG_OK) {
3143 buf_ret = lttng_dynamic_buffer_append(
3144 &reply_payload, closed_trace_chunk_path, path_length);
3145 if (buf_ret) {
3146 ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3147 goto end_no_reply;
3148 }
3149 }
3150
3151 send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, reply_payload.size, 0);
3152 if (send_ret < reply_payload.size) {
3153 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3154 reply_payload.size,
3155 send_ret);
3156 ret = -1;
3157 goto end_no_reply;
3158 }
3159 end_no_reply:
3160 lttng_trace_chunk_put(chunk);
3161 lttng_dynamic_buffer_reset(&reply_payload);
3162 return ret;
3163 }
3164
3165 /*
3166 * relay_trace_chunk_exists: check if a trace chunk exists
3167 */
3168 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr
3169 __attribute__((unused)),
3170 struct relay_connection *conn,
3171 const struct lttng_buffer_view *payload)
3172 {
3173 int ret = 0;
3174 ssize_t send_ret;
3175 struct relay_session *session = conn->session;
3176 struct lttcomm_relayd_trace_chunk_exists *msg;
3177 struct lttcomm_relayd_trace_chunk_exists_reply reply = {};
3178 struct lttng_buffer_view header_view;
3179 uint64_t chunk_id;
3180 bool chunk_exists;
3181
3182 if (!session || !conn->version_check_done) {
3183 ERR("Trying to check for the presence of a trace chunk before version check");
3184 ret = -1;
3185 goto end_no_reply;
3186 }
3187
3188 if (session->major == 2 && session->minor < 11) {
3189 ERR("Chunk exists command is unsupported before 2.11");
3190 ret = -1;
3191 goto end_no_reply;
3192 }
3193
3194 header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3195 if (!lttng_buffer_view_is_valid(&header_view)) {
3196 ERR("Failed to receive payload of chunk exists command");
3197 ret = -1;
3198 goto end_no_reply;
3199 }
3200
3201 /* Convert to host endianness. */
3202 msg = (typeof(msg)) header_view.data;
3203 chunk_id = be64toh(msg->chunk_id);
3204
3205 ret = sessiond_trace_chunk_registry_chunk_exists(sessiond_trace_chunk_registry,
3206 conn->session->sessiond_uuid,
3207 conn->session->id,
3208 chunk_id,
3209 &chunk_exists);
3210 /*
3211 * If ret is not 0, send the reply and report the error to the caller.
3212 * It is a protocol (or internal) error and the session/connection
3213 * should be torn down.
3214 */
3215 reply.generic.ret_code =
3216 htobe32((uint32_t) (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL));
3217 reply.trace_chunk_exists = ret == 0 ? chunk_exists : 0;
3218
3219 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
3220 if (send_ret < (ssize_t) sizeof(reply)) {
3221 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", send_ret);
3222 ret = -1;
3223 }
3224 end_no_reply:
3225 return ret;
3226 }
3227
3228 /*
3229 * relay_get_configuration: query whether feature is available
3230 */
3231 static int relay_get_configuration(const struct lttcomm_relayd_hdr *recv_hdr
3232 __attribute__((unused)),
3233 struct relay_connection *conn,
3234 const struct lttng_buffer_view *payload)
3235 {
3236 int ret = 0;
3237 ssize_t send_ret;
3238 struct lttcomm_relayd_get_configuration *msg;
3239 struct lttcomm_relayd_get_configuration_reply reply = {};
3240 struct lttng_buffer_view header_view;
3241 uint64_t query_flags = 0;
3242 uint64_t result_flags = 0;
3243
3244 header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3245 if (!lttng_buffer_view_is_valid(&header_view)) {
3246 ERR("Failed to receive payload of chunk close command");
3247 ret = -1;
3248 goto end_no_reply;
3249 }
3250
3251 /* Convert to host endianness. */
3252 msg = (typeof(msg)) header_view.data;
3253 query_flags = be64toh(msg->query_flags);
3254
3255 if (query_flags) {
3256 ret = LTTNG_ERR_INVALID_PROTOCOL;
3257 goto reply;
3258 }
3259 if (opt_allow_clear) {
3260 result_flags |= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED;
3261 }
3262 ret = 0;
3263 reply:
3264 reply.generic.ret_code =
3265 htobe32((uint32_t) (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL));
3266 reply.relayd_configuration_flags = htobe64(result_flags);
3267
3268 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
3269 if (send_ret < (ssize_t) sizeof(reply)) {
3270 ERR("Failed to send \"get configuration\" command reply (ret = %zd)", send_ret);
3271 ret = -1;
3272 }
3273 end_no_reply:
3274 return ret;
3275 }
3276
3277 static int relay_process_control_command(struct relay_connection *conn,
3278 const struct lttcomm_relayd_hdr *header,
3279 const struct lttng_buffer_view *payload)
3280 {
3281 int ret = 0;
3282
3283 DBG3("Processing \"%s\" command for socket %i",
3284 lttcomm_relayd_command_str((lttcomm_relayd_command) header->cmd),
3285 conn->sock->fd);
3286 switch (header->cmd) {
3287 case RELAYD_CREATE_SESSION:
3288 ret = relay_create_session(header, conn, payload);
3289 break;
3290 case RELAYD_ADD_STREAM:
3291 ret = relay_add_stream(header, conn, payload);
3292 break;
3293 case RELAYD_START_DATA:
3294 ret = relay_start(header, conn, payload);
3295 break;
3296 case RELAYD_SEND_METADATA:
3297 ret = relay_recv_metadata(header, conn, payload);
3298 break;
3299 case RELAYD_VERSION:
3300 ret = relay_send_version(header, conn, payload);
3301 break;
3302 case RELAYD_CLOSE_STREAM:
3303 ret = relay_close_stream(header, conn, payload);
3304 break;
3305 case RELAYD_DATA_PENDING:
3306 ret = relay_data_pending(header, conn, payload);
3307 break;
3308 case RELAYD_QUIESCENT_CONTROL:
3309 ret = relay_quiescent_control(header, conn, payload);
3310 break;
3311 case RELAYD_BEGIN_DATA_PENDING:
3312 ret = relay_begin_data_pending(header, conn, payload);
3313 break;
3314 case RELAYD_END_DATA_PENDING:
3315 ret = relay_end_data_pending(header, conn, payload);
3316 break;
3317 case RELAYD_SEND_INDEX:
3318 ret = relay_recv_index(header, conn, payload);
3319 break;
3320 case RELAYD_STREAMS_SENT:
3321 ret = relay_streams_sent(header, conn, payload);
3322 break;
3323 case RELAYD_RESET_METADATA:
3324 ret = relay_reset_metadata(header, conn, payload);
3325 break;
3326 case RELAYD_ROTATE_STREAMS:
3327 ret = relay_rotate_session_streams(header, conn, payload);
3328 break;
3329 case RELAYD_CREATE_TRACE_CHUNK:
3330 ret = relay_create_trace_chunk(header, conn, payload);
3331 break;
3332 case RELAYD_CLOSE_TRACE_CHUNK:
3333 ret = relay_close_trace_chunk(header, conn, payload);
3334 break;
3335 case RELAYD_TRACE_CHUNK_EXISTS:
3336 ret = relay_trace_chunk_exists(header, conn, payload);
3337 break;
3338 case RELAYD_GET_CONFIGURATION:
3339 ret = relay_get_configuration(header, conn, payload);
3340 break;
3341 case RELAYD_UPDATE_SYNC_INFO:
3342 default:
3343 ERR("Received unknown command (%u)", header->cmd);
3344 relay_unknown_command(conn);
3345 ret = -1;
3346 goto end;
3347 }
3348
3349 end:
3350 return ret;
3351 }
3352
3353 static enum relay_connection_status
3354 relay_process_control_receive_payload(struct relay_connection *conn)
3355 {
3356 int ret = 0;
3357 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3358 struct lttng_dynamic_buffer *reception_buffer = &conn->protocol.ctrl.reception_buffer;
3359 struct ctrl_connection_state_receive_payload *state =
3360 &conn->protocol.ctrl.state.receive_payload;
3361 struct lttng_buffer_view payload_view;
3362
3363 if (state->left_to_receive == 0) {
3364 /* Short-circuit for payload-less commands. */
3365 goto reception_complete;
3366 }
3367
3368 ret = conn->sock->ops->recvmsg(conn->sock,
3369 reception_buffer->data + state->received,
3370 state->left_to_receive,
3371 MSG_DONTWAIT);
3372 if (ret < 0) {
3373 DIAGNOSTIC_PUSH
3374 DIAGNOSTIC_IGNORE_LOGICAL_OP
3375 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3376 DIAGNOSTIC_POP
3377 PERROR("Unable to receive command payload on sock %d", conn->sock->fd);
3378 status = RELAY_CONNECTION_STATUS_ERROR;
3379 }
3380 goto end;
3381 } else if (ret == 0) {
3382 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3383 status = RELAY_CONNECTION_STATUS_CLOSED;
3384 goto end;
3385 }
3386
3387 LTTNG_ASSERT(ret > 0);
3388 LTTNG_ASSERT(ret <= state->left_to_receive);
3389
3390 state->left_to_receive -= ret;
3391 state->received += ret;
3392
3393 if (state->left_to_receive > 0) {
3394 /*
3395 * Can't transition to the protocol's next state, wait to
3396 * receive the rest of the header.
3397 */
3398 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
3399 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3400 state->received,
3401 state->left_to_receive,
3402 conn->sock->fd);
3403 goto end;
3404 }
3405
3406 reception_complete:
3407 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3408 conn->sock->fd,
3409 state->received);
3410 /*
3411 * The payload required to process the command has been received.
3412 * A view to the reception buffer is forwarded to the various
3413 * commands and the state of the control is reset on success.
3414 *
3415 * Commands are responsible for sending their reply to the peer.
3416 */
3417 payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer, 0, -1);
3418 ret = relay_process_control_command(conn, &state->header, &payload_view);
3419 if (ret < 0) {
3420 status = RELAY_CONNECTION_STATUS_ERROR;
3421 goto end;
3422 }
3423
3424 ret = connection_reset_protocol_state(conn);
3425 if (ret) {
3426 status = RELAY_CONNECTION_STATUS_ERROR;
3427 }
3428 end:
3429 return status;
3430 }
3431
3432 static enum relay_connection_status
3433 relay_process_control_receive_header(struct relay_connection *conn)
3434 {
3435 int ret = 0;
3436 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3437 struct lttcomm_relayd_hdr header;
3438 struct lttng_dynamic_buffer *reception_buffer = &conn->protocol.ctrl.reception_buffer;
3439 struct ctrl_connection_state_receive_header *state =
3440 &conn->protocol.ctrl.state.receive_header;
3441
3442 LTTNG_ASSERT(state->left_to_receive != 0);
3443
3444 ret = conn->sock->ops->recvmsg(conn->sock,
3445 reception_buffer->data + state->received,
3446 state->left_to_receive,
3447 MSG_DONTWAIT);
3448 if (ret < 0) {
3449 DIAGNOSTIC_PUSH
3450 DIAGNOSTIC_IGNORE_LOGICAL_OP
3451 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3452 DIAGNOSTIC_POP
3453 PERROR("Unable to receive control command header on sock %d",
3454 conn->sock->fd);
3455 status = RELAY_CONNECTION_STATUS_ERROR;
3456 }
3457 goto end;
3458 } else if (ret == 0) {
3459 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3460 status = RELAY_CONNECTION_STATUS_CLOSED;
3461 goto end;
3462 }
3463
3464 LTTNG_ASSERT(ret > 0);
3465 LTTNG_ASSERT(ret <= state->left_to_receive);
3466
3467 state->left_to_receive -= ret;
3468 state->received += ret;
3469
3470 if (state->left_to_receive > 0) {
3471 /*
3472 * Can't transition to the protocol's next state, wait to
3473 * receive the rest of the header.
3474 */
3475 DBG3("Partial reception of control connection protocol header (received %" PRIu64
3476 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3477 state->received,
3478 state->left_to_receive,
3479 conn->sock->fd);
3480 goto end;
3481 }
3482
3483 /* Transition to next state: receiving the command's payload. */
3484 conn->protocol.ctrl.state_id = CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3485 memcpy(&header, reception_buffer->data, sizeof(header));
3486 header.circuit_id = be64toh(header.circuit_id);
3487 header.data_size = be64toh(header.data_size);
3488 header.cmd = be32toh(header.cmd);
3489 header.cmd_version = be32toh(header.cmd_version);
3490 memcpy(&conn->protocol.ctrl.state.receive_payload.header, &header, sizeof(header));
3491
3492 DBG("Done receiving control command header: fd = %i, cmd = %s, cmd_version = %" PRIu32
3493 ", payload size = %" PRIu64 " bytes",
3494 conn->sock->fd,
3495 lttcomm_relayd_command_str((enum lttcomm_relayd_command) header.cmd),
3496 header.cmd_version,
3497 header.data_size);
3498
3499 if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
3500 ERR("Command header indicates a payload (%" PRIu64
3501 " bytes) that exceeds the maximal payload size allowed on a control connection.",
3502 header.data_size);
3503 status = RELAY_CONNECTION_STATUS_ERROR;
3504 goto end;
3505 }
3506
3507 conn->protocol.ctrl.state.receive_payload.left_to_receive = header.data_size;
3508 conn->protocol.ctrl.state.receive_payload.received = 0;
3509 ret = lttng_dynamic_buffer_set_size(reception_buffer, header.data_size);
3510 if (ret) {
3511 status = RELAY_CONNECTION_STATUS_ERROR;
3512 goto end;
3513 }
3514
3515 if (header.data_size == 0) {
3516 /*
3517 * Manually invoke the next state as the poll loop
3518 * will not wake-up to allow us to proceed further.
3519 */
3520 status = relay_process_control_receive_payload(conn);
3521 }
3522 end:
3523 return status;
3524 }
3525
3526 /*
3527 * Process the commands received on the control socket
3528 */
3529 static enum relay_connection_status relay_process_control(struct relay_connection *conn)
3530 {
3531 enum relay_connection_status status;
3532
3533 switch (conn->protocol.ctrl.state_id) {
3534 case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
3535 status = relay_process_control_receive_header(conn);
3536 break;
3537 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
3538 status = relay_process_control_receive_payload(conn);
3539 break;
3540 default:
3541 ERR("Unknown control connection protocol state encountered.");
3542 abort();
3543 }
3544
3545 return status;
3546 }
3547
3548 static enum relay_connection_status relay_process_data_receive_header(struct relay_connection *conn)
3549 {
3550 int ret;
3551 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3552 struct data_connection_state_receive_header *state =
3553 &conn->protocol.data.state.receive_header;
3554 struct lttcomm_relayd_data_hdr header;
3555 struct relay_stream *stream;
3556
3557 LTTNG_ASSERT(state->left_to_receive != 0);
3558
3559 ret = conn->sock->ops->recvmsg(conn->sock,
3560 state->header_reception_buffer + state->received,
3561 state->left_to_receive,
3562 MSG_DONTWAIT);
3563 if (ret < 0) {
3564 DIAGNOSTIC_PUSH
3565 DIAGNOSTIC_IGNORE_LOGICAL_OP
3566 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3567 DIAGNOSTIC_POP
3568 PERROR("Unable to receive data header on sock %d", conn->sock->fd);
3569 status = RELAY_CONNECTION_STATUS_ERROR;
3570 }
3571 goto end;
3572 } else if (ret == 0) {
3573 /* Orderly shutdown. Not necessary to print an error. */
3574 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3575 status = RELAY_CONNECTION_STATUS_CLOSED;
3576 goto end;
3577 }
3578
3579 LTTNG_ASSERT(ret > 0);
3580 LTTNG_ASSERT(ret <= state->left_to_receive);
3581
3582 state->left_to_receive -= ret;
3583 state->received += ret;
3584
3585 if (state->left_to_receive > 0) {
3586 /*
3587 * Can't transition to the protocol's next state, wait to
3588 * receive the rest of the header.
3589 */
3590 DBG3("Partial reception of data connection header (received %" PRIu64
3591 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3592 state->received,
3593 state->left_to_receive,
3594 conn->sock->fd);
3595 goto end;
3596 }
3597
3598 /* Transition to next state: receiving the payload. */
3599 conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
3600
3601 memcpy(&header, state->header_reception_buffer, sizeof(header));
3602 header.circuit_id = be64toh(header.circuit_id);
3603 header.stream_id = be64toh(header.stream_id);
3604 header.data_size = be32toh(header.data_size);
3605 header.net_seq_num = be64toh(header.net_seq_num);
3606 header.padding_size = be32toh(header.padding_size);
3607 memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
3608
3609 conn->protocol.data.state.receive_payload.left_to_receive = header.data_size;
3610 conn->protocol.data.state.receive_payload.received = 0;
3611 conn->protocol.data.state.receive_payload.rotate_index = false;
3612
3613 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
3614 ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64
3615 ", padding_size = %" PRIu32,
3616 conn->sock->fd,
3617 header.circuit_id,
3618 header.stream_id,
3619 header.data_size,
3620 header.net_seq_num,
3621 header.padding_size);
3622
3623 stream = stream_get_by_id(header.stream_id);
3624 if (!stream) {
3625 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
3626 header.stream_id);
3627 /* Protocol error. */
3628 status = RELAY_CONNECTION_STATUS_ERROR;
3629 goto end;
3630 }
3631
3632 pthread_mutex_lock(&stream->lock);
3633 /* Prepare stream for the reception of a new packet. */
3634 ret = stream_init_packet(
3635 stream, header.data_size, &conn->protocol.data.state.receive_payload.rotate_index);
3636 pthread_mutex_unlock(&stream->lock);
3637 if (ret) {
3638 ERR("Failed to rotate stream output file");
3639 status = RELAY_CONNECTION_STATUS_ERROR;
3640 goto end_stream_unlock;
3641 }
3642
3643 end_stream_unlock:
3644 stream_put(stream);
3645 end:
3646 return status;
3647 }
3648
3649 static enum relay_connection_status
3650 relay_process_data_receive_payload(struct relay_connection *conn)
3651 {
3652 int ret;
3653 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3654 struct relay_stream *stream;
3655 struct data_connection_state_receive_payload *state =
3656 &conn->protocol.data.state.receive_payload;
3657 const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
3658 char data_buffer[chunk_size];
3659 bool partial_recv = false;
3660 bool new_stream = false, close_requested = false, index_flushed = false;
3661 uint64_t left_to_receive = state->left_to_receive;
3662 struct relay_session *session;
3663
3664 DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64
3665 " bytes received, %" PRIu64 " bytes left to receive",
3666 state->header.stream_id,
3667 state->header.net_seq_num,
3668 state->received,
3669 left_to_receive);
3670
3671 stream = stream_get_by_id(state->header.stream_id);
3672 if (!stream) {
3673 /* Protocol error. */
3674 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
3675 state->header.stream_id);
3676 status = RELAY_CONNECTION_STATUS_ERROR;
3677 goto end;
3678 }
3679
3680 pthread_mutex_lock(&stream->lock);
3681 session = stream->trace->session;
3682 if (!conn->session) {
3683 ret = connection_set_session(conn, session);
3684 if (ret) {
3685 status = RELAY_CONNECTION_STATUS_ERROR;
3686 goto end_stream_unlock;
3687 }
3688 }
3689
3690 /*
3691 * The size of the "chunk" received on any iteration is bounded by:
3692 * - the data left to receive,
3693 * - the data immediately available on the socket,
3694 * - the on-stack data buffer
3695 */
3696 while (left_to_receive > 0 && !partial_recv) {
3697 size_t recv_size = std::min<uint64_t>(left_to_receive, chunk_size);
3698 struct lttng_buffer_view packet_chunk;
3699
3700 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_size, MSG_DONTWAIT);
3701 if (ret < 0) {
3702 DIAGNOSTIC_PUSH
3703 DIAGNOSTIC_IGNORE_LOGICAL_OP
3704 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3705 DIAGNOSTIC_POP
3706 PERROR("Socket %d error", conn->sock->fd);
3707 status = RELAY_CONNECTION_STATUS_ERROR;
3708 }
3709 goto end_stream_unlock;
3710 } else if (ret == 0) {
3711 /* No more data ready to be consumed on socket. */
3712 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
3713 state->header.stream_id);
3714 status = RELAY_CONNECTION_STATUS_CLOSED;
3715 break;
3716 } else if (ret < (int) recv_size) {
3717 /*
3718 * All the data available on the socket has been
3719 * consumed.
3720 */
3721 partial_recv = true;
3722 recv_size = ret;
3723 }
3724
3725 packet_chunk = lttng_buffer_view_init(data_buffer, 0, recv_size);
3726 LTTNG_ASSERT(packet_chunk.data);
3727
3728 ret = stream_write(stream, &packet_chunk, 0);
3729 if (ret) {
3730 ERR("Relay error writing data to file");
3731 status = RELAY_CONNECTION_STATUS_ERROR;
3732 goto end_stream_unlock;
3733 }
3734
3735 left_to_receive -= recv_size;
3736 state->received += recv_size;
3737 state->left_to_receive = left_to_receive;
3738 }
3739
3740 if (state->left_to_receive > 0) {
3741 /*
3742 * Did not receive all the data expected, wait for more data to
3743 * become available on the socket.
3744 */
3745 DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64
3746 " bytes received, %" PRIu64 " bytes left to receive",
3747 state->header.stream_id,
3748 state->received,
3749 state->left_to_receive);
3750 goto end_stream_unlock;
3751 }
3752
3753 ret = stream_write(stream, NULL, state->header.padding_size);
3754 if (ret) {
3755 status = RELAY_CONNECTION_STATUS_ERROR;
3756 goto end_stream_unlock;
3757 }
3758
3759 if (session_streams_have_index(session)) {
3760 ret = stream_update_index(stream,
3761 state->header.net_seq_num,
3762 state->rotate_index,
3763 &index_flushed,
3764 state->header.data_size + state->header.padding_size);
3765 if (ret < 0) {
3766 ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64
3767 " ret %d",
3768 stream->stream_handle,
3769 state->header.net_seq_num,
3770 ret);
3771 status = RELAY_CONNECTION_STATUS_ERROR;
3772 goto end_stream_unlock;
3773 }
3774 }
3775
3776 if (stream->prev_data_seq == -1ULL) {
3777 new_stream = true;
3778 }
3779
3780 ret = stream_complete_packet(stream,
3781 state->header.data_size + state->header.padding_size,
3782 state->header.net_seq_num,
3783 index_flushed);
3784 if (ret) {
3785 status = RELAY_CONNECTION_STATUS_ERROR;
3786 goto end_stream_unlock;
3787 }
3788
3789 /*
3790 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3791 * contents of *state which are aliased (union) to the same location as
3792 * the new state. Don't use it beyond this point.
3793 */
3794 connection_reset_protocol_state(conn);
3795 state = NULL;
3796
3797 end_stream_unlock:
3798 close_requested = stream->close_requested;
3799 pthread_mutex_unlock(&stream->lock);
3800 if (close_requested && left_to_receive == 0) {
3801 try_stream_close(stream);
3802 }
3803
3804 if (new_stream) {
3805 pthread_mutex_lock(&session->lock);
3806 uatomic_set(&session->new_streams, 1);
3807 pthread_mutex_unlock(&session->lock);
3808 }
3809
3810 stream_put(stream);
3811 end:
3812 return status;
3813 }
3814
3815 /*
3816 * relay_process_data: Process the data received on the data socket
3817 */
3818 static enum relay_connection_status relay_process_data(struct relay_connection *conn)
3819 {
3820 enum relay_connection_status status;
3821
3822 switch (conn->protocol.data.state_id) {
3823 case DATA_CONNECTION_STATE_RECEIVE_HEADER:
3824 status = relay_process_data_receive_header(conn);
3825 break;
3826 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
3827 status = relay_process_data_receive_payload(conn);
3828 break;
3829 default:
3830 ERR("Unexpected data connection communication state.");
3831 abort();
3832 }
3833
3834 return status;
3835 }
3836
3837 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
3838 {
3839 int ret;
3840
3841 (void) lttng_poll_del(events, pollfd);
3842
3843 ret = fd_tracker_close_unsuspendable_fd(
3844 the_fd_tracker, &pollfd, 1, fd_tracker_util_close_fd, NULL);
3845 if (ret < 0) {
3846 ERR("Closing pollfd %d", pollfd);
3847 }
3848 }
3849
3850 static void relay_thread_close_connection(struct lttng_poll_event *events,
3851 int pollfd,
3852 struct relay_connection *conn)
3853 {
3854 const char *type_str;
3855
3856 switch (conn->type) {
3857 case RELAY_DATA:
3858 type_str = "Data";
3859 break;
3860 case RELAY_CONTROL:
3861 type_str = "Control";
3862 break;
3863 case RELAY_VIEWER_COMMAND:
3864 type_str = "Viewer Command";
3865 break;
3866 case RELAY_VIEWER_NOTIFICATION:
3867 type_str = "Viewer Notification";
3868 break;
3869 default:
3870 type_str = "Unknown";
3871 }
3872 cleanup_connection_pollfd(events, pollfd);
3873 connection_put(conn);
3874 DBG("%s connection closed with %d", type_str, pollfd);
3875 }
3876
3877 /*
3878 * This thread does the actual work
3879 */
3880 static void *relay_thread_worker(void *data __attribute__((unused)))
3881 {
3882 int ret, err = -1, last_seen_data_fd = -1;
3883 uint32_t nb_fd;
3884 struct lttng_poll_event events;
3885 struct lttng_ht *relay_connections_ht;
3886 struct lttng_ht_iter iter;
3887 struct relay_connection *destroy_conn = NULL;
3888
3889 DBG("[thread] Relay worker started");
3890
3891 rcu_register_thread();
3892
3893 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
3894
3895 if (testpoint(relayd_thread_worker)) {
3896 goto error_testpoint;
3897 }
3898
3899 health_code_update();
3900
3901 /* table of connections indexed on socket */
3902 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
3903 if (!relay_connections_ht) {
3904 goto relay_connections_ht_error;
3905 }
3906
3907 ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll");
3908 if (ret < 0) {
3909 goto error_poll_create;
3910 }
3911
3912 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
3913 if (ret < 0) {
3914 goto error;
3915 }
3916
3917 restart:
3918 while (1) {
3919 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
3920
3921 health_code_update();
3922
3923 /* Infinite blocking call, waiting for transmission */
3924 DBG3("Relayd worker thread polling...");
3925 health_poll_entry();
3926 ret = lttng_poll_wait(&events, -1);
3927 health_poll_exit();
3928 if (ret < 0) {
3929 /*
3930 * Restart interrupted system call.
3931 */
3932 if (errno == EINTR) {
3933 goto restart;
3934 }
3935 goto error;
3936 }
3937
3938 nb_fd = ret;
3939
3940 /*
3941 * Process control. The control connection is
3942 * prioritized so we don't starve it with high
3943 * throughput tracing data on the data connection.
3944 */
3945 for (i = 0; i < nb_fd; i++) {
3946 /* Fetch once the poll data */
3947 const auto revents = LTTNG_POLL_GETEV(&events, i);
3948 const auto pollfd = LTTNG_POLL_GETFD(&events, i);
3949
3950 health_code_update();
3951
3952 /* Activity on thread quit pipe, exiting. */
3953 if (relayd_is_thread_quit_pipe(pollfd)) {
3954 DBG("Activity on thread quit pipe");
3955 err = 0;
3956 goto exit;
3957 }
3958
3959 /* Inspect the relay conn pipe for new connection */
3960 if (pollfd == relay_conn_pipe[0]) {
3961 if (revents & LPOLLIN) {
3962 struct relay_connection *conn;
3963
3964 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
3965 if (ret < 0) {
3966 goto error;
3967 }
3968 ret = lttng_poll_add(
3969 &events, conn->sock->fd, LPOLLIN | LPOLLRDHUP);
3970 if (ret) {
3971 ERR("Failed to add new connection file descriptor to poll set");
3972 goto error;
3973 }
3974 connection_ht_add(relay_connections_ht, conn);
3975 DBG("Connection socket %d added", conn->sock->fd);
3976 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3977 ERR("Relay connection pipe error");
3978 goto error;
3979 } else {
3980 ERR("Unexpected poll events %u for sock %d",
3981 revents,
3982 pollfd);
3983 goto error;
3984 }
3985 } else {
3986 struct relay_connection *ctrl_conn;
3987
3988 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
3989 /* If not found, there is a synchronization issue. */
3990 LTTNG_ASSERT(ctrl_conn);
3991
3992 if (ctrl_conn->type == RELAY_DATA) {
3993 if (revents & LPOLLIN) {
3994 /*
3995 * Flag the last seen data fd not deleted. It will
3996 * be used as the last seen fd if any fd gets
3997 * deleted in this first loop.
3998 */
3999 last_notdel_data_fd = pollfd;
4000 }
4001 goto put_ctrl_connection;
4002 }
4003 LTTNG_ASSERT(ctrl_conn->type == RELAY_CONTROL);
4004
4005 if (revents & LPOLLIN) {
4006 enum relay_connection_status status;
4007
4008 status = relay_process_control(ctrl_conn);
4009 if (status != RELAY_CONNECTION_STATUS_OK) {
4010 /*
4011 * On socket error flag the session as aborted to
4012 * force the cleanup of its stream otherwise it can
4013 * leak during the lifetime of the relayd.
4014 *
4015 * This prevents situations in which streams can be
4016 * left opened because an index was received, the
4017 * control connection is closed, and the data
4018 * connection is closed (uncleanly) before the
4019 * packet's data provided.
4020 *
4021 * Since the control connection encountered an
4022 * error, it is okay to be conservative and close
4023 * the session right now as we can't rely on the
4024 * protocol being respected anymore.
4025 */
4026 if (status == RELAY_CONNECTION_STATUS_ERROR) {
4027 session_abort(ctrl_conn->session);
4028 }
4029
4030 /* Clear the connection on error or close. */
4031 relay_thread_close_connection(
4032 &events, pollfd, ctrl_conn);
4033 }
4034 seen_control = 1;
4035 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4036 relay_thread_close_connection(&events, pollfd, ctrl_conn);
4037 if (last_seen_data_fd == pollfd) {
4038 last_seen_data_fd = last_notdel_data_fd;
4039 }
4040 } else {
4041 ERR("Unexpected poll events %u for control sock %d",
4042 revents,
4043 pollfd);
4044 connection_put(ctrl_conn);
4045 goto error;
4046 }
4047 put_ctrl_connection:
4048 connection_put(ctrl_conn);
4049 }
4050 }
4051
4052 /*
4053 * The last loop handled a control request, go back to poll to make
4054 * sure we prioritise the control socket.
4055 */
4056 if (seen_control) {
4057 continue;
4058 }
4059
4060 if (last_seen_data_fd >= 0) {
4061 for (i = 0; i < nb_fd; i++) {
4062 int pollfd = LTTNG_POLL_GETFD(&events, i);
4063
4064 health_code_update();
4065
4066 if (last_seen_data_fd == pollfd) {
4067 idx = i;
4068 break;
4069 }
4070 }
4071 }
4072
4073 /* Process data connection. */
4074 for (i = idx + 1; i < nb_fd; i++) {
4075 /* Fetch the poll data. */
4076 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
4077 int pollfd = LTTNG_POLL_GETFD(&events, i);
4078 struct relay_connection *data_conn;
4079
4080 health_code_update();
4081
4082 if (!revents) {
4083 /* No activity for this FD (poll implementation). */
4084 continue;
4085 }
4086
4087 /* Skip the command pipe. It's handled in the first loop. */
4088 if (pollfd == relay_conn_pipe[0]) {
4089 continue;
4090 }
4091
4092 data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4093 if (!data_conn) {
4094 /* Skip it. Might be removed before. */
4095 continue;
4096 }
4097 if (data_conn->type == RELAY_CONTROL) {
4098 goto put_data_connection;
4099 }
4100 LTTNG_ASSERT(data_conn->type == RELAY_DATA);
4101
4102 if (revents & LPOLLIN) {
4103 enum relay_connection_status status;
4104
4105 status = relay_process_data(data_conn);
4106 /* Connection closed or error. */
4107 if (status != RELAY_CONNECTION_STATUS_OK) {
4108 /*
4109 * On socket error flag the session as aborted to force
4110 * the cleanup of its stream otherwise it can leak
4111 * during the lifetime of the relayd.
4112 *
4113 * This prevents situations in which streams can be
4114 * left opened because an index was received, the
4115 * control connection is closed, and the data
4116 * connection is closed (uncleanly) before the packet's
4117 * data provided.
4118 *
4119 * Since the data connection encountered an error,
4120 * it is okay to be conservative and close the
4121 * session right now as we can't rely on the protocol
4122 * being respected anymore.
4123 */
4124 if (status == RELAY_CONNECTION_STATUS_ERROR) {
4125 session_abort(data_conn->session);
4126 }
4127 relay_thread_close_connection(&events, pollfd, data_conn);
4128 /*
4129 * Every goto restart call sets the last seen fd where
4130 * here we don't really care since we gracefully
4131 * continue the loop after the connection is deleted.
4132 */
4133 } else {
4134 /* Keep last seen port. */
4135 last_seen_data_fd = pollfd;
4136 connection_put(data_conn);
4137 goto restart;
4138 }
4139 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4140 relay_thread_close_connection(&events, pollfd, data_conn);
4141 } else {
4142 ERR("Unknown poll events %u for data sock %d", revents, pollfd);
4143 }
4144 put_data_connection:
4145 connection_put(data_conn);
4146 }
4147 last_seen_data_fd = -1;
4148 }
4149
4150 /* Normal exit, no error */
4151 ret = 0;
4152
4153 exit:
4154 error:
4155 /* Cleanup remaining connection object. */
4156 rcu_read_lock();
4157 cds_lfht_for_each_entry (relay_connections_ht->ht, &iter.iter, destroy_conn, sock_n.node) {
4158 health_code_update();
4159
4160 session_abort(destroy_conn->session);
4161
4162 /*
4163 * No need to grab another ref, because we own
4164 * destroy_conn.
4165 */
4166 relay_thread_close_connection(&events, destroy_conn->sock->fd, destroy_conn);
4167 }
4168 rcu_read_unlock();
4169
4170 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
4171 error_poll_create:
4172 lttng_ht_destroy(relay_connections_ht);
4173 relay_connections_ht_error:
4174 /* Close relay conn pipes */
4175 (void) fd_tracker_util_pipe_close(the_fd_tracker, relay_conn_pipe);
4176 if (err) {
4177 DBG("Thread exited with error");
4178 }
4179 DBG("Worker thread cleanup complete");
4180 error_testpoint:
4181 if (err) {
4182 health_error();
4183 ERR("Health error occurred in %s", __func__);
4184 }
4185 health_unregister(health_relayd);
4186 rcu_unregister_thread();
4187 lttng_relay_stop_threads();
4188 return NULL;
4189 }
4190
4191 /*
4192 * Create the relay command pipe to wake thread_manage_apps.
4193 * Closed in cleanup().
4194 */
4195 static int create_relay_conn_pipe(void)
4196 {
4197 return fd_tracker_util_pipe_open_cloexec(
4198 the_fd_tracker, "Relayd connection pipe", relay_conn_pipe);
4199 }
4200
4201 static int stdio_open(void *data __attribute__((unused)), int *fds)
4202 {
4203 fds[0] = fileno(stdout);
4204 fds[1] = fileno(stderr);
4205 return 0;
4206 }
4207
4208 static int track_stdio(void)
4209 {
4210 int fds[2];
4211 const char *names[] = { "stdout", "stderr" };
4212
4213 return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds, names, 2, stdio_open, NULL);
4214 }
4215
4216 /*
4217 * main
4218 */
4219 int main(int argc, char **argv)
4220 {
4221 bool thread_is_rcu_registered = false;
4222 int ret = 0, retval = 0;
4223 void *status;
4224 char *unlinked_file_directory_path = NULL, *output_path = NULL;
4225
4226 /* Parse environment variables */
4227 ret = parse_env_options();
4228 if (ret) {
4229 retval = -1;
4230 goto exit_options;
4231 }
4232
4233 /*
4234 * Parse arguments.
4235 * Command line arguments overwrite environment.
4236 */
4237 progname = argv[0];
4238 if (set_options(argc, argv)) {
4239 retval = -1;
4240 goto exit_options;
4241 }
4242
4243 if (set_signal_handler()) {
4244 retval = -1;
4245 goto exit_options;
4246 }
4247
4248 relayd_config_log();
4249
4250 if (opt_print_version) {
4251 print_version();
4252 retval = 0;
4253 goto exit_options;
4254 }
4255
4256 ret = fclose(stdin);
4257 if (ret) {
4258 PERROR("Failed to close stdin");
4259 goto exit_options;
4260 }
4261
4262 DBG("Clear command %s", opt_allow_clear ? "allowed" : "disallowed");
4263
4264 /* Try to create directory if -o, --output is specified. */
4265 if (opt_output_path) {
4266 if (*opt_output_path != '/') {
4267 ERR("Please specify an absolute path for -o, --output PATH");
4268 retval = -1;
4269 goto exit_options;
4270 }
4271
4272 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG, -1, -1);
4273 if (ret < 0) {
4274 ERR("Unable to create %s", opt_output_path);
4275 retval = -1;
4276 goto exit_options;
4277 }
4278 }
4279
4280 /* Daemonize */
4281 if (opt_daemon || opt_background) {
4282 ret = lttng_daemonize(&child_ppid, &recv_child_signal, !opt_background);
4283 if (ret < 0) {
4284 retval = -1;
4285 goto exit_options;
4286 }
4287 }
4288
4289 if (opt_working_directory) {
4290 ret = utils_change_working_directory(opt_working_directory);
4291 if (ret) {
4292 /* All errors are already logged. */
4293 goto exit_options;
4294 }
4295 }
4296
4297 sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create();
4298 if (!sessiond_trace_chunk_registry) {
4299 ERR("Failed to initialize session daemon trace chunk registry");
4300 retval = -1;
4301 goto exit_options;
4302 }
4303
4304 /*
4305 * The RCU thread registration (and use, through the fd-tracker's
4306 * creation) is done after the daemonization to allow us to not
4307 * deal with liburcu's fork() management as the call RCU needs to
4308 * be restored.
4309 */
4310 rcu_register_thread();
4311 thread_is_rcu_registered = true;
4312
4313 output_path = create_output_path("");
4314 if (!output_path) {
4315 ERR("Failed to get output path");
4316 retval = -1;
4317 goto exit_options;
4318 }
4319 ret = asprintf(&unlinked_file_directory_path,
4320 "%s/%s",
4321 output_path,
4322 DEFAULT_UNLINKED_FILES_DIRECTORY);
4323 free(output_path);
4324 if (ret < 0) {
4325 ERR("Failed to format unlinked file directory path");
4326 retval = -1;
4327 goto exit_options;
4328 }
4329 the_fd_tracker = fd_tracker_create(unlinked_file_directory_path, lttng_opt_fd_pool_size);
4330 free(unlinked_file_directory_path);
4331 if (!the_fd_tracker) {
4332 retval = -1;
4333 goto exit_options;
4334 }
4335
4336 ret = track_stdio();
4337 if (ret) {
4338 retval = -1;
4339 goto exit_options;
4340 }
4341
4342 /* Initialize thread health monitoring */
4343 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
4344 if (!health_relayd) {
4345 PERROR("health_app_create error");
4346 retval = -1;
4347 goto exit_options;
4348 }
4349
4350 /* Create thread quit pipe */
4351 if (relayd_init_thread_quit_pipe()) {
4352 retval = -1;
4353 goto exit_options;
4354 }
4355
4356 /* Setup the thread apps communication pipe. */
4357 if (create_relay_conn_pipe()) {
4358 retval = -1;
4359 goto exit_options;
4360 }
4361
4362 /* Init relay command queue. */
4363 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
4364
4365 /* Initialize communication library */
4366 lttcomm_init();
4367 lttcomm_inet_init();
4368
4369 /* tables of sessions indexed by session ID */
4370 sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4371 if (!sessions_ht) {
4372 retval = -1;
4373 goto exit_options;
4374 }
4375
4376 /* tables of streams indexed by stream ID */
4377 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4378 if (!relay_streams_ht) {
4379 retval = -1;
4380 goto exit_options;
4381 }
4382
4383 /* tables of streams indexed by stream ID */
4384 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4385 if (!viewer_streams_ht) {
4386 retval = -1;
4387 goto exit_options;
4388 }
4389
4390 ret = init_health_quit_pipe();
4391 if (ret) {
4392 retval = -1;
4393 goto exit_options;
4394 }
4395
4396 /* Create thread to manage the client socket */
4397 ret = pthread_create(
4398 &health_thread, default_pthread_attr(), thread_manage_health_relayd, (void *) NULL);
4399 if (ret) {
4400 errno = ret;
4401 PERROR("pthread_create health");
4402 retval = -1;
4403 goto exit_options;
4404 }
4405
4406 /* Setup the dispatcher thread */
4407 ret = pthread_create(
4408 &dispatcher_thread, default_pthread_attr(), relay_thread_dispatcher, (void *) NULL);
4409 if (ret) {
4410 errno = ret;
4411 PERROR("pthread_create dispatcher");
4412 retval = -1;
4413 goto exit_dispatcher_thread;
4414 }
4415
4416 /* Setup the worker thread */
4417 ret = pthread_create(&worker_thread, default_pthread_attr(), relay_thread_worker, NULL);
4418 if (ret) {
4419 errno = ret;
4420 PERROR("pthread_create worker");
4421 retval = -1;
4422 goto exit_worker_thread;
4423 }
4424
4425 /* Setup the listener thread */
4426 ret = pthread_create(
4427 &listener_thread, default_pthread_attr(), relay_thread_listener, (void *) NULL);
4428 if (ret) {
4429 errno = ret;
4430 PERROR("pthread_create listener");
4431 retval = -1;
4432 goto exit_listener_thread;
4433 }
4434
4435 ret = relayd_live_create(live_uri);
4436 if (ret) {
4437 ERR("Starting live viewer threads");
4438 retval = -1;
4439 goto exit_live;
4440 }
4441
4442 /*
4443 * This is where we start awaiting program completion (e.g. through
4444 * signal that asks threads to teardown).
4445 */
4446
4447 ret = relayd_live_join();
4448 if (ret) {
4449 retval = -1;
4450 }
4451 exit_live:
4452
4453 ret = pthread_join(listener_thread, &status);
4454 if (ret) {
4455 errno = ret;
4456 PERROR("pthread_join listener_thread");
4457 retval = -1;
4458 }
4459
4460 exit_listener_thread:
4461 ret = pthread_join(worker_thread, &status);
4462 if (ret) {
4463 errno = ret;
4464 PERROR("pthread_join worker_thread");
4465 retval = -1;
4466 }
4467
4468 exit_worker_thread:
4469 ret = pthread_join(dispatcher_thread, &status);
4470 if (ret) {
4471 errno = ret;
4472 PERROR("pthread_join dispatcher_thread");
4473 retval = -1;
4474 }
4475 exit_dispatcher_thread:
4476
4477 ret = pthread_join(health_thread, &status);
4478 if (ret) {
4479 errno = ret;
4480 PERROR("pthread_join health_thread");
4481 retval = -1;
4482 }
4483 exit_options:
4484 /*
4485 * Wait for all pending call_rcu work to complete before tearing
4486 * down data structures. call_rcu worker may be trying to
4487 * perform lookups in those structures.
4488 */
4489 rcu_barrier();
4490 relayd_cleanup();
4491
4492 /* Ensure all prior call_rcu are done. */
4493 rcu_barrier();
4494
4495 if (thread_is_rcu_registered) {
4496 rcu_unregister_thread();
4497 }
4498
4499 if (!retval) {
4500 exit(EXIT_SUCCESS);
4501 } else {
4502 exit(EXIT_FAILURE);
4503 }
4504 }
LTTng 2.0 tools and control repository
This page took 0.452278 seconds and 4 git commands to generate.