projects / lttng-tools.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Typo: 'Accomodate' -> 'Accommodate'
[lttng-tools.git] / src / bin / lttng-relayd / main.c
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 <getopt.h>
13 #include <grp.h>
14 #include <limits.h>
15 #include <pthread.h>
16 #include <signal.h>
17 #include <stdio.h>
18 #include <stdlib.h>
19 #include <string.h>
20 #include <sys/mman.h>
21 #include <sys/mount.h>
22 #include <sys/resource.h>
23 #include <sys/socket.h>
24 #include <sys/stat.h>
25 #include <sys/types.h>
26 #include <sys/wait.h>
27 #include <sys/resource.h>
28 #include <inttypes.h>
29 #include <urcu/futex.h>
30 #include <urcu/uatomic.h>
31 #include <urcu/rculist.h>
32 #include <unistd.h>
33 #include <fcntl.h>
34 #include <strings.h>
35 #include <ctype.h>
36
37 #include <lttng/lttng.h>
38 #include <common/common.h>
39 #include <common/compat/poll.h>
40 #include <common/compat/socket.h>
41 #include <common/compat/endian.h>
42 #include <common/compat/getenv.h>
43 #include <common/defaults.h>
44 #include <common/daemonize.h>
45 #include <common/futex.h>
46 #include <common/sessiond-comm/sessiond-comm.h>
47 #include <common/sessiond-comm/inet.h>
48 #include <common/sessiond-comm/relayd.h>
49 #include <common/uri.h>
50 #include <common/utils.h>
51 #include <common/align.h>
52 #include <common/config/session-config.h>
53 #include <common/dynamic-buffer.h>
54 #include <common/buffer-view.h>
55 #include <common/string-utils/format.h>
56 #include <common/fd-tracker/fd-tracker.h>
57 #include <common/fd-tracker/utils.h>
58
59 #include "backward-compatibility-group-by.h"
60 #include "cmd.h"
61 #include "connection.h"
62 #include "ctf-trace.h"
63 #include "health-relayd.h"
64 #include "index.h"
65 #include "live.h"
66 #include "lttng-relayd.h"
67 #include "session.h"
68 #include "sessiond-trace-chunks.h"
69 #include "stream.h"
70 #include "tcp_keep_alive.h"
71 #include "testpoint.h"
72 #include "tracefile-array.h"
73 #include "utils.h"
74 #include "version.h"
75 #include "viewer-stream.h"
76
77 static const char *help_msg =
78 #ifdef LTTNG_EMBED_HELP
79 #include <lttng-relayd.8.h>
80 #else
81 NULL
82 #endif
83 ;
84
85 enum relay_connection_status {
86 RELAY_CONNECTION_STATUS_OK,
87 /* An error occurred while processing an event on the connection. */
88 RELAY_CONNECTION_STATUS_ERROR,
89 /* Connection closed/shutdown cleanly. */
90 RELAY_CONNECTION_STATUS_CLOSED,
91 };
92
93 /* command line options */
94 char *opt_output_path, *opt_working_directory;
95 static int opt_daemon, opt_background, opt_print_version, opt_allow_clear = 1;
96 enum relay_group_output_by opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_UNKNOWN;
97
98 /*
99 * We need to wait for listener and live listener threads, as well as
100 * health check thread, before being ready to signal readiness.
101 */
102 #define NR_LTTNG_RELAY_READY 3
103 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
104
105 /* Size of receive buffer. */
106 #define RECV_DATA_BUFFER_SIZE 65536
107
108 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
109 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
110
111 static struct lttng_uri *control_uri;
112 static struct lttng_uri *data_uri;
113 static struct lttng_uri *live_uri;
114
115 const char *progname;
116
117 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
118 static int tracing_group_name_override;
119
120 const char * const config_section_name = "relayd";
121
122 /*
123 * Quit pipe for all threads. This permits a single cancellation point
124 * for all threads when receiving an event on the pipe.
125 */
126 int thread_quit_pipe[2] = { -1, -1 };
127
128 /*
129 * This pipe is used to inform the worker thread that a command is queued and
130 * ready to be processed.
131 */
132 static int relay_conn_pipe[2] = { -1, -1 };
133
134 /* Shared between threads */
135 static int dispatch_thread_exit;
136
137 static pthread_t listener_thread;
138 static pthread_t dispatcher_thread;
139 static pthread_t worker_thread;
140 static pthread_t health_thread;
141
142 /*
143 * last_relay_stream_id_lock protects last_relay_stream_id increment
144 * atomicity on 32-bit architectures.
145 */
146 static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
147 static uint64_t last_relay_stream_id;
148
149 /*
150 * Relay command queue.
151 *
152 * The relay_thread_listener and relay_thread_dispatcher communicate with this
153 * queue.
154 */
155 static struct relay_conn_queue relay_conn_queue;
156
157 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
158 static unsigned int lttng_opt_fd_pool_size = -1;
159
160 /* Global relay stream hash table. */
161 struct lttng_ht *relay_streams_ht;
162
163 /* Global relay viewer stream hash table. */
164 struct lttng_ht *viewer_streams_ht;
165
166 /* Global relay sessions hash table. */
167 struct lttng_ht *sessions_ht;
168
169 /* Relayd health monitoring */
170 struct health_app *health_relayd;
171
172 struct sessiond_trace_chunk_registry *sessiond_trace_chunk_registry;
173
174 /* Global fd tracker. */
175 struct fd_tracker *the_fd_tracker;
176
177 static struct option long_options[] = {
178 { "control-port", 1, 0, 'C', },
179 { "data-port", 1, 0, 'D', },
180 { "live-port", 1, 0, 'L', },
181 { "daemonize", 0, 0, 'd', },
182 { "background", 0, 0, 'b', },
183 { "group", 1, 0, 'g', },
184 { "fd-pool-size", 1, 0, '\0', },
185 { "help", 0, 0, 'h', },
186 { "output", 1, 0, 'o', },
187 { "verbose", 0, 0, 'v', },
188 { "config", 1, 0, 'f' },
189 { "version", 0, 0, 'V' },
190 { "working-directory", 1, 0, 'w', },
191 { "group-output-by-session", 0, 0, 's', },
192 { "group-output-by-host", 0, 0, 'p', },
193 { "disallow-clear", 0, 0, 'x' },
194 { NULL, 0, 0, 0, },
195 };
196
197 static const char *config_ignore_options[] = { "help", "config", "version" };
198
199 static void print_version(void) {
200 fprintf(stdout, "%s\n", VERSION);
201 }
202
203 static void relayd_config_log(void)
204 {
205 DBG("LTTng-relayd " VERSION " - " VERSION_NAME "%s%s",
206 GIT_VERSION[0] == '\0' ? "" : " - " GIT_VERSION,
207 EXTRA_VERSION_NAME[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME);
208 if (EXTRA_VERSION_DESCRIPTION[0] != '\0') {
209 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION "\n");
210 }
211 if (EXTRA_VERSION_PATCHES[0] != '\0') {
212 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES "\n");
213 }
214 }
215
216 /*
217 * Take an option from the getopt output and set it in the right variable to be
218 * used later.
219 *
220 * Return 0 on success else a negative value.
221 */
222 static int set_option(int opt, const char *arg, const char *optname)
223 {
224 int ret;
225
226 switch (opt) {
227 case 0:
228 if (!strcmp(optname, "fd-pool-size")) {
229 unsigned long v;
230
231 errno = 0;
232 v = strtoul(arg, NULL, 0);
233 if (errno != 0 || !isdigit((unsigned char) arg[0])) {
234 ERR("Wrong value in --fd-pool-size parameter: %s", arg);
235 ret = -1;
236 goto end;
237 }
238 if (v >= UINT_MAX) {
239 ERR("File descriptor cap overflow in --fd-pool-size parameter: %s", arg);
240 ret = -1;
241 goto end;
242 }
243 lttng_opt_fd_pool_size = (unsigned int) v;
244 } else {
245 fprintf(stderr, "unknown option %s", optname);
246 if (arg) {
247 fprintf(stderr, " with arg %s\n", arg);
248 }
249 }
250 break;
251 case 'C':
252 if (lttng_is_setuid_setgid()) {
253 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
254 "-C, --control-port");
255 } else {
256 ret = uri_parse(arg, &control_uri);
257 if (ret < 0) {
258 ERR("Invalid control URI specified");
259 goto end;
260 }
261 if (control_uri->port == 0) {
262 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
263 }
264 }
265 break;
266 case 'D':
267 if (lttng_is_setuid_setgid()) {
268 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
269 "-D, -data-port");
270 } else {
271 ret = uri_parse(arg, &data_uri);
272 if (ret < 0) {
273 ERR("Invalid data URI specified");
274 goto end;
275 }
276 if (data_uri->port == 0) {
277 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
278 }
279 }
280 break;
281 case 'L':
282 if (lttng_is_setuid_setgid()) {
283 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
284 "-L, -live-port");
285 } else {
286 ret = uri_parse(arg, &live_uri);
287 if (ret < 0) {
288 ERR("Invalid live URI specified");
289 goto end;
290 }
291 if (live_uri->port == 0) {
292 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
293 }
294 }
295 break;
296 case 'd':
297 opt_daemon = 1;
298 break;
299 case 'b':
300 opt_background = 1;
301 break;
302 case 'g':
303 if (lttng_is_setuid_setgid()) {
304 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
305 "-g, --group");
306 } else {
307 tracing_group_name = strdup(arg);
308 if (tracing_group_name == NULL) {
309 ret = -errno;
310 PERROR("strdup");
311 goto end;
312 }
313 tracing_group_name_override = 1;
314 }
315 break;
316 case 'h':
317 ret = utils_show_help(8, "lttng-relayd", help_msg);
318 if (ret) {
319 ERR("Cannot show --help for `lttng-relayd`");
320 perror("exec");
321 }
322 exit(EXIT_FAILURE);
323 case 'V':
324 opt_print_version = 1;
325 break;
326 case 'o':
327 if (lttng_is_setuid_setgid()) {
328 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
329 "-o, --output");
330 } else {
331 ret = asprintf(&opt_output_path, "%s", arg);
332 if (ret < 0) {
333 ret = -errno;
334 PERROR("asprintf opt_output_path");
335 goto end;
336 }
337 }
338 break;
339 case 'w':
340 if (lttng_is_setuid_setgid()) {
341 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
342 "-w, --working-directory");
343 } else {
344 ret = asprintf(&opt_working_directory, "%s", arg);
345 if (ret < 0) {
346 ret = -errno;
347 PERROR("asprintf opt_working_directory");
348 goto end;
349 }
350 }
351 break;
352
353 case 'v':
354 /* Verbose level can increase using multiple -v */
355 if (arg) {
356 lttng_opt_verbose = config_parse_value(arg);
357 } else {
358 /* Only 3 level of verbosity (-vvv). */
359 if (lttng_opt_verbose < 3) {
360 lttng_opt_verbose += 1;
361 }
362 }
363 break;
364 case 's':
365 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
366 ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
367 exit(EXIT_FAILURE);
368 }
369 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_SESSION;
370 break;
371 case 'p':
372 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
373 ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
374 exit(EXIT_FAILURE);
375 }
376 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
377 break;
378 case 'x':
379 /* Disallow clear */
380 opt_allow_clear = 0;
381 break;
382 default:
383 /* Unknown option or other error.
384 * Error is printed by getopt, just return */
385 ret = -1;
386 goto end;
387 }
388
389 /* All good. */
390 ret = 0;
391
392 end:
393 return ret;
394 }
395
396 /*
397 * config_entry_handler_cb used to handle options read from a config file.
398 * See config_entry_handler_cb comment in common/config/session-config.h for the
399 * return value conventions.
400 */
401 static int config_entry_handler(const struct config_entry *entry, void *unused)
402 {
403 int ret = 0, i;
404
405 if (!entry || !entry->name || !entry->value) {
406 ret = -EINVAL;
407 goto end;
408 }
409
410 /* Check if the option is to be ignored */
411 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
412 if (!strcmp(entry->name, config_ignore_options[i])) {
413 goto end;
414 }
415 }
416
417 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
418 /* Ignore if entry name is not fully matched. */
419 if (strcmp(entry->name, long_options[i].name)) {
420 continue;
421 }
422
423 /*
424 * If the option takes no argument on the command line,
425 * we have to check if the value is "true". We support
426 * non-zero numeric values, true, on and yes.
427 */
428 if (!long_options[i].has_arg) {
429 ret = config_parse_value(entry->value);
430 if (ret <= 0) {
431 if (ret) {
432 WARN("Invalid configuration value \"%s\" for option %s",
433 entry->value, entry->name);
434 }
435 /* False, skip boolean config option. */
436 goto end;
437 }
438 }
439
440 ret = set_option(long_options[i].val, entry->value, entry->name);
441 goto end;
442 }
443
444 WARN("Unrecognized option \"%s\" in daemon configuration file.",
445 entry->name);
446
447 end:
448 return ret;
449 }
450
451 static int parse_env_options(void)
452 {
453 int ret = 0;
454 char *value = NULL;
455
456 value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV);
457 if (value) {
458 opt_working_directory = strdup(value);
459 if (!opt_working_directory) {
460 ERR("Failed to allocate working directory string (\"%s\")",
461 value);
462 ret = -1;
463 }
464 }
465 return ret;
466 }
467
468 static int set_fd_pool_size(void)
469 {
470 int ret = 0;
471 struct rlimit rlimit;
472
473 ret = getrlimit(RLIMIT_NOFILE, &rlimit);
474 if (ret) {
475 PERROR("Failed to get file descriptor limit");
476 ret = -1;
477 goto end;
478 }
479
480 DBG("File descriptor count limits are %" PRIu64 " (soft) and %" PRIu64 " (hard)",
481 (uint64_t) rlimit.rlim_cur,
482 (uint64_t) rlimit.rlim_max);
483 if (lttng_opt_fd_pool_size == -1) {
484 /* Use default value (soft limit - reserve). */
485 if (rlimit.rlim_cur < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
486 ERR("The process' file number limit is too low (%" PRIu64 "). The process' file number limit must be set to at least %i.",
487 (uint64_t) rlimit.rlim_cur, DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
488 ret = -1;
489 goto end;
490 }
491 lttng_opt_fd_pool_size = rlimit.rlim_cur -
492 DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
493 goto end;
494 }
495
496 if (lttng_opt_fd_pool_size < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
497 ERR("File descriptor pool size must be set to at least %d",
498 DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
499 ret = -1;
500 goto end;
501 }
502
503 if (lttng_opt_fd_pool_size > rlimit.rlim_cur) {
504 ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64 ").",
505 lttng_opt_fd_pool_size, (uint64_t) rlimit.rlim_cur);
506 ret = -1;
507 goto end;
508 }
509
510 DBG("File descriptor pool size argument (%u) adjusted to %u to accommodates transient fd uses",
511 lttng_opt_fd_pool_size,
512 lttng_opt_fd_pool_size - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE);
513 lttng_opt_fd_pool_size -= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
514 end:
515 return ret;
516 }
517
518 static int set_options(int argc, char **argv)
519 {
520 int c, ret = 0, option_index = 0, retval = 0;
521 int orig_optopt = optopt, orig_optind = optind;
522 char *default_address, *optstring;
523 const char *config_path = NULL;
524
525 optstring = utils_generate_optstring(long_options,
526 sizeof(long_options) / sizeof(struct option));
527 if (!optstring) {
528 retval = -ENOMEM;
529 goto exit;
530 }
531
532 /* Check for the --config option */
533
534 while ((c = getopt_long(argc, argv, optstring, long_options,
535 &option_index)) != -1) {
536 if (c == '?') {
537 retval = -EINVAL;
538 goto exit;
539 } else if (c != 'f') {
540 continue;
541 }
542
543 if (lttng_is_setuid_setgid()) {
544 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
545 "-f, --config");
546 } else {
547 config_path = utils_expand_path(optarg);
548 if (!config_path) {
549 ERR("Failed to resolve path: %s", optarg);
550 }
551 }
552 }
553
554 ret = config_get_section_entries(config_path, config_section_name,
555 config_entry_handler, NULL);
556 if (ret) {
557 if (ret > 0) {
558 ERR("Invalid configuration option at line %i", ret);
559 }
560 retval = -1;
561 goto exit;
562 }
563
564 /* Reset getopt's global state */
565 optopt = orig_optopt;
566 optind = orig_optind;
567 while (1) {
568 c = getopt_long(argc, argv, optstring, long_options, &option_index);
569 if (c == -1) {
570 break;
571 }
572
573 ret = set_option(c, optarg, long_options[option_index].name);
574 if (ret < 0) {
575 retval = -1;
576 goto exit;
577 }
578 }
579
580 /* assign default values */
581 if (control_uri == NULL) {
582 ret = asprintf(&default_address,
583 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
584 DEFAULT_NETWORK_CONTROL_PORT);
585 if (ret < 0) {
586 PERROR("asprintf default data address");
587 retval = -1;
588 goto exit;
589 }
590
591 ret = uri_parse(default_address, &control_uri);
592 free(default_address);
593 if (ret < 0) {
594 ERR("Invalid control URI specified");
595 retval = -1;
596 goto exit;
597 }
598 }
599 if (data_uri == NULL) {
600 ret = asprintf(&default_address,
601 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
602 DEFAULT_NETWORK_DATA_PORT);
603 if (ret < 0) {
604 PERROR("asprintf default data address");
605 retval = -1;
606 goto exit;
607 }
608
609 ret = uri_parse(default_address, &data_uri);
610 free(default_address);
611 if (ret < 0) {
612 ERR("Invalid data URI specified");
613 retval = -1;
614 goto exit;
615 }
616 }
617 if (live_uri == NULL) {
618 ret = asprintf(&default_address,
619 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
620 DEFAULT_NETWORK_VIEWER_PORT);
621 if (ret < 0) {
622 PERROR("asprintf default viewer control address");
623 retval = -1;
624 goto exit;
625 }
626
627 ret = uri_parse(default_address, &live_uri);
628 free(default_address);
629 if (ret < 0) {
630 ERR("Invalid viewer control URI specified");
631 retval = -1;
632 goto exit;
633 }
634 }
635 ret = set_fd_pool_size();
636 if (ret) {
637 retval = -1;
638 goto exit;
639 }
640
641 if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
642 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
643 }
644 if (opt_allow_clear) {
645 /* Check if env variable exists. */
646 const char *value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
647 if (value) {
648 ret = config_parse_value(value);
649 if (ret < 0) {
650 ERR("Invalid value for %s specified", DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
651 retval = -1;
652 goto exit;
653 }
654 opt_allow_clear = !ret;
655 }
656 }
657
658 exit:
659 free(optstring);
660 return retval;
661 }
662
663 static void print_global_objects(void)
664 {
665 print_viewer_streams();
666 print_relay_streams();
667 print_sessions();
668 }
669
670 static int noop_close(void *data, int *fds)
671 {
672 return 0;
673 }
674
675 static void untrack_stdio(void)
676 {
677 int fds[] = { fileno(stdout), fileno(stderr) };
678
679 /*
680 * noop_close is used since we don't really want to close
681 * the stdio output fds; we merely want to stop tracking them.
682 */
683 (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker,
684 fds, 2, noop_close, NULL);
685 }
686
687 /*
688 * Cleanup the daemon
689 */
690 static void relayd_cleanup(void)
691 {
692 print_global_objects();
693
694 DBG("Cleaning up");
695
696 if (viewer_streams_ht)
697 lttng_ht_destroy(viewer_streams_ht);
698 if (relay_streams_ht)
699 lttng_ht_destroy(relay_streams_ht);
700 if (sessions_ht)
701 lttng_ht_destroy(sessions_ht);
702
703 free(opt_output_path);
704 free(opt_working_directory);
705
706 if (health_relayd) {
707 health_app_destroy(health_relayd);
708 }
709 /* Close thread quit pipes */
710 if (health_quit_pipe[0] != -1) {
711 (void) fd_tracker_util_pipe_close(
712 the_fd_tracker, health_quit_pipe);
713 }
714 if (thread_quit_pipe[0] != -1) {
715 (void) fd_tracker_util_pipe_close(
716 the_fd_tracker, thread_quit_pipe);
717 }
718 if (sessiond_trace_chunk_registry) {
719 sessiond_trace_chunk_registry_destroy(
720 sessiond_trace_chunk_registry);
721 }
722 if (the_fd_tracker) {
723 untrack_stdio();
724 /*
725 * fd_tracker_destroy() will log the contents of the fd-tracker
726 * if a leak is detected.
727 */
728 fd_tracker_destroy(the_fd_tracker);
729 }
730
731 uri_free(control_uri);
732 uri_free(data_uri);
733 /* Live URI is freed in the live thread. */
734
735 if (tracing_group_name_override) {
736 free((void *) tracing_group_name);
737 }
738 }
739
740 /*
741 * Write to writable pipe used to notify a thread.
742 */
743 static int notify_thread_pipe(int wpipe)
744 {
745 ssize_t ret;
746
747 ret = lttng_write(wpipe, "!", 1);
748 if (ret < 1) {
749 PERROR("write poll pipe");
750 goto end;
751 }
752 ret = 0;
753 end:
754 return ret;
755 }
756
757 static int notify_health_quit_pipe(int *pipe)
758 {
759 ssize_t ret;
760
761 ret = lttng_write(pipe[1], "4", 1);
762 if (ret < 1) {
763 PERROR("write relay health quit");
764 goto end;
765 }
766 ret = 0;
767 end:
768 return ret;
769 }
770
771 /*
772 * Stop all relayd and relayd-live threads.
773 */
774 int lttng_relay_stop_threads(void)
775 {
776 int retval = 0;
777
778 /* Stopping all threads */
779 DBG("Terminating all threads");
780 if (notify_thread_pipe(thread_quit_pipe[1])) {
781 ERR("write error on thread quit pipe");
782 retval = -1;
783 }
784
785 if (notify_health_quit_pipe(health_quit_pipe)) {
786 ERR("write error on health quit pipe");
787 }
788
789 /* Dispatch thread */
790 CMM_STORE_SHARED(dispatch_thread_exit, 1);
791 futex_nto1_wake(&relay_conn_queue.futex);
792
793 if (relayd_live_stop()) {
794 ERR("Error stopping live threads");
795 retval = -1;
796 }
797 return retval;
798 }
799
800 /*
801 * Signal handler for the daemon
802 *
803 * Simply stop all worker threads, leaving main() return gracefully after
804 * joining all threads and calling cleanup().
805 */
806 static void sighandler(int sig)
807 {
808 switch (sig) {
809 case SIGINT:
810 DBG("SIGINT caught");
811 if (lttng_relay_stop_threads()) {
812 ERR("Error stopping threads");
813 }
814 break;
815 case SIGTERM:
816 DBG("SIGTERM caught");
817 if (lttng_relay_stop_threads()) {
818 ERR("Error stopping threads");
819 }
820 break;
821 case SIGUSR1:
822 CMM_STORE_SHARED(recv_child_signal, 1);
823 break;
824 default:
825 break;
826 }
827 }
828
829 /*
830 * Setup signal handler for :
831 * SIGINT, SIGTERM, SIGPIPE
832 */
833 static int set_signal_handler(void)
834 {
835 int ret = 0;
836 struct sigaction sa;
837 sigset_t sigset;
838
839 if ((ret = sigemptyset(&sigset)) < 0) {
840 PERROR("sigemptyset");
841 return ret;
842 }
843
844 sa.sa_mask = sigset;
845 sa.sa_flags = 0;
846
847 sa.sa_handler = sighandler;
848 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
849 PERROR("sigaction");
850 return ret;
851 }
852
853 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
854 PERROR("sigaction");
855 return ret;
856 }
857
858 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
859 PERROR("sigaction");
860 return ret;
861 }
862
863 sa.sa_handler = SIG_IGN;
864 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
865 PERROR("sigaction");
866 return ret;
867 }
868
869 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
870
871 return ret;
872 }
873
874 void lttng_relay_notify_ready(void)
875 {
876 /* Notify the parent of the fork() process that we are ready. */
877 if (opt_daemon || opt_background) {
878 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
879 kill(child_ppid, SIGUSR1);
880 }
881 }
882 }
883
884 /*
885 * Init thread quit pipe.
886 *
887 * Return -1 on error or 0 if all pipes are created.
888 */
889 static int init_thread_quit_pipe(void)
890 {
891 return fd_tracker_util_pipe_open_cloexec(
892 the_fd_tracker, "Quit pipe", thread_quit_pipe);
893 }
894
895 /*
896 * Init health quit pipe.
897 *
898 * Return -1 on error or 0 if all pipes are created.
899 */
900 static int init_health_quit_pipe(void)
901 {
902 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
903 "Health quit pipe", health_quit_pipe);
904 }
905
906 /*
907 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
908 */
909 static int create_named_thread_poll_set(struct lttng_poll_event *events,
910 int size, const char *name)
911 {
912 int ret;
913
914 if (events == NULL || size == 0) {
915 ret = -1;
916 goto error;
917 }
918
919 ret = fd_tracker_util_poll_create(the_fd_tracker,
920 name, events, 1, LTTNG_CLOEXEC);
921 if (ret) {
922 PERROR("Failed to create \"%s\" poll file descriptor", name);
923 goto error;
924 }
925
926 /* Add quit pipe */
927 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
928 if (ret < 0) {
929 goto error;
930 }
931
932 return 0;
933
934 error:
935 return ret;
936 }
937
938 /*
939 * Check if the thread quit pipe was triggered.
940 *
941 * Return 1 if it was triggered else 0;
942 */
943 static int check_thread_quit_pipe(int fd, uint32_t events)
944 {
945 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
946 return 1;
947 }
948
949 return 0;
950 }
951
952 static int create_sock(void *data, int *out_fd)
953 {
954 int ret;
955 struct lttcomm_sock *sock = data;
956
957 ret = lttcomm_create_sock(sock);
958 if (ret < 0) {
959 goto end;
960 }
961
962 *out_fd = sock->fd;
963 end:
964 return ret;
965 }
966
967 static int close_sock(void *data, int *in_fd)
968 {
969 struct lttcomm_sock *sock = data;
970
971 return sock->ops->close(sock);
972 }
973
974 static int accept_sock(void *data, int *out_fd)
975 {
976 int ret = 0;
977 /* Socks is an array of in_sock, out_sock. */
978 struct lttcomm_sock **socks = data;
979 struct lttcomm_sock *in_sock = socks[0];
980
981 socks[1] = in_sock->ops->accept(in_sock);
982 if (!socks[1]) {
983 ret = -1;
984 goto end;
985 }
986 *out_fd = socks[1]->fd;
987 end:
988 return ret;
989 }
990
991 /*
992 * Create and init socket from uri.
993 */
994 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri,
995 const char *name)
996 {
997 int ret, sock_fd;
998 struct lttcomm_sock *sock = NULL;
999 char uri_str[PATH_MAX];
1000 char *formated_name = NULL;
1001
1002 sock = lttcomm_alloc_sock_from_uri(uri);
1003 if (sock == NULL) {
1004 ERR("Allocating socket");
1005 goto error;
1006 }
1007
1008 /*
1009 * Don't fail to create the socket if the name can't be built as it is
1010 * only used for debugging purposes.
1011 */
1012 ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
1013 uri_str[sizeof(uri_str) - 1] = '\0';
1014 if (ret >= 0) {
1015 ret = asprintf(&formated_name, "%s socket @ %s", name,
1016 uri_str);
1017 if (ret < 0) {
1018 formated_name = NULL;
1019 }
1020 }
1021
1022 ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &sock_fd,
1023 (const char **) (formated_name ? &formated_name : NULL),
1024 1, create_sock, sock);
1025 if (ret) {
1026 PERROR("Failed to open \"%s\" relay socket",
1027 formated_name ?: "Unknown");
1028 goto error;
1029 }
1030 DBG("Listening on %s socket %d", name, sock->fd);
1031
1032 ret = sock->ops->bind(sock);
1033 if (ret < 0) {
1034 PERROR("Failed to bind socket");
1035 goto error;
1036 }
1037
1038 ret = sock->ops->listen(sock, -1);
1039 if (ret < 0) {
1040 goto error;
1041
1042 }
1043
1044 free(formated_name);
1045 return sock;
1046
1047 error:
1048 if (sock) {
1049 lttcomm_destroy_sock(sock);
1050 }
1051 free(formated_name);
1052 return NULL;
1053 }
1054
1055 static
1056 struct lttcomm_sock *accept_relayd_sock(struct lttcomm_sock *listening_sock,
1057 const char *name)
1058 {
1059 int out_fd, ret;
1060 struct lttcomm_sock *socks[2] = { listening_sock, NULL };
1061 struct lttcomm_sock *new_sock = NULL;
1062
1063 ret = fd_tracker_open_unsuspendable_fd(
1064 the_fd_tracker, &out_fd,
1065 (const char **) &name,
1066 1, accept_sock, &socks);
1067 if (ret) {
1068 goto end;
1069 }
1070 new_sock = socks[1];
1071 DBG("%s accepted, socket %d", name, new_sock->fd);
1072 end:
1073 return new_sock;
1074 }
1075
1076 /*
1077 * This thread manages the listening for new connections on the network
1078 */
1079 static void *relay_thread_listener(void *data)
1080 {
1081 int i, ret, pollfd, err = -1;
1082 uint32_t revents, nb_fd;
1083 struct lttng_poll_event events;
1084 struct lttcomm_sock *control_sock, *data_sock;
1085
1086 DBG("[thread] Relay listener started");
1087
1088 rcu_register_thread();
1089 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
1090
1091 health_code_update();
1092
1093 control_sock = relay_socket_create(control_uri, "Control listener");
1094 if (!control_sock) {
1095 goto error_sock_control;
1096 }
1097
1098 data_sock = relay_socket_create(data_uri, "Data listener");
1099 if (!data_sock) {
1100 goto error_sock_relay;
1101 }
1102
1103 /*
1104 * Pass 3 as size here for the thread quit pipe, control and
1105 * data socket.
1106 */
1107 ret = create_named_thread_poll_set(&events, 3, "Listener thread epoll");
1108 if (ret < 0) {
1109 goto error_create_poll;
1110 }
1111
1112 /* Add the control socket */
1113 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
1114 if (ret < 0) {
1115 goto error_poll_add;
1116 }
1117
1118 /* Add the data socket */
1119 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
1120 if (ret < 0) {
1121 goto error_poll_add;
1122 }
1123
1124 lttng_relay_notify_ready();
1125
1126 if (testpoint(relayd_thread_listener)) {
1127 goto error_testpoint;
1128 }
1129
1130 while (1) {
1131 health_code_update();
1132
1133 DBG("Listener accepting connections");
1134
1135 restart:
1136 health_poll_entry();
1137 ret = lttng_poll_wait(&events, -1);
1138 health_poll_exit();
1139 if (ret < 0) {
1140 /*
1141 * Restart interrupted system call.
1142 */
1143 if (errno == EINTR) {
1144 goto restart;
1145 }
1146 goto error;
1147 }
1148
1149 nb_fd = ret;
1150
1151 DBG("Relay new connection received");
1152 for (i = 0; i < nb_fd; i++) {
1153 health_code_update();
1154
1155 /* Fetch once the poll data */
1156 revents = LTTNG_POLL_GETEV(&events, i);
1157 pollfd = LTTNG_POLL_GETFD(&events, i);
1158
1159 /* Thread quit pipe has been closed. Killing thread. */
1160 ret = check_thread_quit_pipe(pollfd, revents);
1161 if (ret) {
1162 err = 0;
1163 goto exit;
1164 }
1165
1166 if (revents & LPOLLIN) {
1167 /*
1168 * A new connection is requested, therefore a
1169 * sessiond/consumerd connection is allocated in
1170 * this thread, enqueued to a global queue and
1171 * dequeued (and freed) in the worker thread.
1172 */
1173 int val = 1;
1174 struct relay_connection *new_conn;
1175 struct lttcomm_sock *newsock = NULL;
1176 enum connection_type type;
1177
1178 if (pollfd == data_sock->fd) {
1179 type = RELAY_DATA;
1180 newsock = accept_relayd_sock(data_sock,
1181 "Data socket to relayd");
1182 } else {
1183 assert(pollfd == control_sock->fd);
1184 type = RELAY_CONTROL;
1185 newsock = accept_relayd_sock(control_sock,
1186 "Control socket to relayd");
1187 }
1188 if (!newsock) {
1189 PERROR("accepting sock");
1190 goto error;
1191 }
1192
1193 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
1194 sizeof(val));
1195 if (ret < 0) {
1196 PERROR("setsockopt inet");
1197 lttcomm_destroy_sock(newsock);
1198 goto error;
1199 }
1200
1201 ret = socket_apply_keep_alive_config(newsock->fd);
1202 if (ret < 0) {
1203 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1204 newsock->fd);
1205 lttcomm_destroy_sock(newsock);
1206 goto error;
1207 }
1208
1209 new_conn = connection_create(newsock, type);
1210 if (!new_conn) {
1211 lttcomm_destroy_sock(newsock);
1212 goto error;
1213 }
1214
1215 /* Enqueue request for the dispatcher thread. */
1216 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
1217 &new_conn->qnode);
1218
1219 /*
1220 * Wake the dispatch queue futex.
1221 * Implicit memory barrier with the
1222 * exchange in cds_wfcq_enqueue.
1223 */
1224 futex_nto1_wake(&relay_conn_queue.futex);
1225 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1226 ERR("socket poll error");
1227 goto error;
1228 } else {
1229 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1230 goto error;
1231 }
1232 }
1233 }
1234
1235 exit:
1236 error:
1237 error_poll_add:
1238 error_testpoint:
1239 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
1240 error_create_poll:
1241 if (data_sock->fd >= 0) {
1242 int data_sock_fd = data_sock->fd;
1243
1244 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
1245 &data_sock_fd, 1, close_sock,
1246 data_sock);
1247 if (ret) {
1248 PERROR("Failed to close the data listener socket file descriptor");
1249 }
1250 data_sock->fd = -1;
1251 }
1252 lttcomm_destroy_sock(data_sock);
1253 error_sock_relay:
1254 if (control_sock->fd >= 0) {
1255 int control_sock_fd = control_sock->fd;
1256
1257 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
1258 &control_sock_fd, 1, close_sock,
1259 control_sock);
1260 if (ret) {
1261 PERROR("Failed to close the control listener socket file descriptor");
1262 }
1263 control_sock->fd = -1;
1264 }
1265 lttcomm_destroy_sock(control_sock);
1266 error_sock_control:
1267 if (err) {
1268 health_error();
1269 ERR("Health error occurred in %s", __func__);
1270 }
1271 health_unregister(health_relayd);
1272 rcu_unregister_thread();
1273 DBG("Relay listener thread cleanup complete");
1274 lttng_relay_stop_threads();
1275 return NULL;
1276 }
1277
1278 /*
1279 * This thread manages the dispatching of the requests to worker threads
1280 */
1281 static void *relay_thread_dispatcher(void *data)
1282 {
1283 int err = -1;
1284 ssize_t ret;
1285 struct cds_wfcq_node *node;
1286 struct relay_connection *new_conn = NULL;
1287
1288 DBG("[thread] Relay dispatcher started");
1289
1290 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
1291
1292 if (testpoint(relayd_thread_dispatcher)) {
1293 goto error_testpoint;
1294 }
1295
1296 health_code_update();
1297
1298 for (;;) {
1299 health_code_update();
1300
1301 /* Atomically prepare the queue futex */
1302 futex_nto1_prepare(&relay_conn_queue.futex);
1303
1304 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1305 break;
1306 }
1307
1308 do {
1309 health_code_update();
1310
1311 /* Dequeue commands */
1312 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1313 &relay_conn_queue.tail);
1314 if (node == NULL) {
1315 DBG("Woken up but nothing in the relay command queue");
1316 /* Continue thread execution */
1317 break;
1318 }
1319 new_conn = caa_container_of(node, struct relay_connection, qnode);
1320
1321 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1322
1323 /*
1324 * Inform worker thread of the new request. This
1325 * call is blocking so we can be assured that
1326 * the data will be read at some point in time
1327 * or wait to the end of the world :)
1328 */
1329 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1330 if (ret < 0) {
1331 PERROR("write connection pipe");
1332 connection_put(new_conn);
1333 goto error;
1334 }
1335 } while (node != NULL);
1336
1337 /* Futex wait on queue. Blocking call on futex() */
1338 health_poll_entry();
1339 futex_nto1_wait(&relay_conn_queue.futex);
1340 health_poll_exit();
1341 }
1342
1343 /* Normal exit, no error */
1344 err = 0;
1345
1346 error:
1347 error_testpoint:
1348 if (err) {
1349 health_error();
1350 ERR("Health error occurred in %s", __func__);
1351 }
1352 health_unregister(health_relayd);
1353 DBG("Dispatch thread dying");
1354 lttng_relay_stop_threads();
1355 return NULL;
1356 }
1357
1358 static bool session_streams_have_index(const struct relay_session *session)
1359 {
1360 return session->minor >= 4 && !session->snapshot;
1361 }
1362
1363 /*
1364 * Handle the RELAYD_CREATE_SESSION command.
1365 *
1366 * On success, send back the session id or else return a negative value.
1367 */
1368 static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr,
1369 struct relay_connection *conn,
1370 const struct lttng_buffer_view *payload)
1371 {
1372 int ret = 0;
1373 ssize_t send_ret;
1374 struct relay_session *session = NULL;
1375 struct lttcomm_relayd_create_session_reply_2_11 reply = {};
1376 char session_name[LTTNG_NAME_MAX] = {};
1377 char hostname[LTTNG_HOST_NAME_MAX] = {};
1378 uint32_t live_timer = 0;
1379 bool snapshot = false;
1380 bool session_name_contains_creation_timestamp = false;
1381 /* Left nil for peers < 2.11. */
1382 char base_path[LTTNG_PATH_MAX] = {};
1383 lttng_uuid sessiond_uuid = {};
1384 LTTNG_OPTIONAL(uint64_t) id_sessiond = {};
1385 LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
1386 LTTNG_OPTIONAL(time_t) creation_time = {};
1387 struct lttng_dynamic_buffer reply_payload;
1388
1389 lttng_dynamic_buffer_init(&reply_payload);
1390
1391 if (conn->minor < 4) {
1392 /* From 2.1 to 2.3 */
1393 ret = 0;
1394 } else if (conn->minor >= 4 && conn->minor < 11) {
1395 /* From 2.4 to 2.10 */
1396 ret = cmd_create_session_2_4(payload, session_name,
1397 hostname, &live_timer, &snapshot);
1398 } else {
1399 bool has_current_chunk;
1400 uint64_t current_chunk_id_value;
1401 time_t creation_time_value;
1402 uint64_t id_sessiond_value;
1403
1404 /* From 2.11 to ... */
1405 ret = cmd_create_session_2_11(payload, session_name, hostname,
1406 base_path, &live_timer, &snapshot, &id_sessiond_value,
1407 sessiond_uuid, &has_current_chunk,
1408 &current_chunk_id_value, &creation_time_value,
1409 &session_name_contains_creation_timestamp);
1410 if (lttng_uuid_is_nil(sessiond_uuid)) {
1411 /* The nil UUID is reserved for pre-2.11 clients. */
1412 ERR("Illegal nil UUID announced by peer in create session command");
1413 ret = -1;
1414 goto send_reply;
1415 }
1416 LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value);
1417 LTTNG_OPTIONAL_SET(&creation_time, creation_time_value);
1418 if (has_current_chunk) {
1419 LTTNG_OPTIONAL_SET(&current_chunk_id,
1420 current_chunk_id_value);
1421 }
1422 }
1423
1424 if (ret < 0) {
1425 goto send_reply;
1426 }
1427
1428 session = session_create(session_name, hostname, base_path, live_timer,
1429 snapshot, sessiond_uuid,
1430 id_sessiond.is_set ? &id_sessiond.value : NULL,
1431 current_chunk_id.is_set ? &current_chunk_id.value : NULL,
1432 creation_time.is_set ? &creation_time.value : NULL,
1433 conn->major, conn->minor,
1434 session_name_contains_creation_timestamp);
1435 if (!session) {
1436 ret = -1;
1437 goto send_reply;
1438 }
1439 assert(!conn->session);
1440 conn->session = session;
1441 DBG("Created session %" PRIu64, session->id);
1442
1443 reply.generic.session_id = htobe64(session->id);
1444
1445 send_reply:
1446 if (ret < 0) {
1447 reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL);
1448 } else {
1449 reply.generic.ret_code = htobe32(LTTNG_OK);
1450 }
1451
1452 if (conn->minor < 11) {
1453 /* From 2.1 to 2.10 */
1454 ret = lttng_dynamic_buffer_append(&reply_payload,
1455 &reply.generic, sizeof(reply.generic));
1456 if (ret) {
1457 ERR("Failed to append \"create session\" command reply header to payload buffer");
1458 ret = -1;
1459 goto end;
1460 }
1461 } else {
1462 const uint32_t output_path_length =
1463 session ? strlen(session->output_path) + 1 : 0;
1464
1465 reply.output_path_length = htobe32(output_path_length);
1466 ret = lttng_dynamic_buffer_append(
1467 &reply_payload, &reply, sizeof(reply));
1468 if (ret) {
1469 ERR("Failed to append \"create session\" command reply header to payload buffer");
1470 goto end;
1471 }
1472
1473 if (output_path_length) {
1474 ret = lttng_dynamic_buffer_append(&reply_payload,
1475 session->output_path,
1476 output_path_length);
1477 if (ret) {
1478 ERR("Failed to append \"create session\" command reply path to payload buffer");
1479 goto end;
1480 }
1481 }
1482 }
1483
1484 send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data,
1485 reply_payload.size, 0);
1486 if (send_ret < (ssize_t) reply_payload.size) {
1487 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1488 reply_payload.size, send_ret);
1489 ret = -1;
1490 }
1491 end:
1492 if (ret < 0 && session) {
1493 session_put(session);
1494 }
1495 lttng_dynamic_buffer_reset(&reply_payload);
1496 return ret;
1497 }
1498
1499 /*
1500 * When we have received all the streams and the metadata for a channel,
1501 * we make them visible to the viewer threads.
1502 */
1503 static void publish_connection_local_streams(struct relay_connection *conn)
1504 {
1505 struct relay_stream *stream;
1506 struct relay_session *session = conn->session;
1507
1508 /*
1509 * We publish all streams belonging to a session atomically wrt
1510 * session lock.
1511 */
1512 pthread_mutex_lock(&session->lock);
1513 rcu_read_lock();
1514 cds_list_for_each_entry_rcu(stream, &session->recv_list,
1515 recv_node) {
1516 stream_publish(stream);
1517 }
1518 rcu_read_unlock();
1519
1520 /*
1521 * Inform the viewer that there are new streams in the session.
1522 */
1523 if (session->viewer_attached) {
1524 uatomic_set(&session->new_streams, 1);
1525 }
1526 pthread_mutex_unlock(&session->lock);
1527 }
1528
1529 static int conform_channel_path(char *channel_path)
1530 {
1531 int ret = 0;
1532
1533 if (strstr("../", channel_path)) {
1534 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1535 channel_path);
1536 ret = -1;
1537 goto end;
1538 }
1539
1540 if (*channel_path == '/') {
1541 const size_t len = strlen(channel_path);
1542
1543 /*
1544 * Channel paths from peers prior to 2.11 are expressed as an
1545 * absolute path that is, in reality, relative to the relay
1546 * daemon's output directory. Remove the leading slash so it
1547 * is correctly interpreted as a relative path later on.
1548 *
1549 * len (and not len - 1) is used to copy the trailing NULL.
1550 */
1551 bcopy(channel_path + 1, channel_path, len);
1552 }
1553 end:
1554 return ret;
1555 }
1556
1557 /*
1558 * relay_add_stream: allocate a new stream for a session
1559 */
1560 static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1561 struct relay_connection *conn,
1562 const struct lttng_buffer_view *payload)
1563 {
1564 int ret;
1565 ssize_t send_ret;
1566 struct relay_session *session = conn->session;
1567 struct relay_stream *stream = NULL;
1568 struct lttcomm_relayd_status_stream reply;
1569 struct ctf_trace *trace = NULL;
1570 uint64_t stream_handle = -1ULL;
1571 char *path_name = NULL, *channel_name = NULL;
1572 uint64_t tracefile_size = 0, tracefile_count = 0;
1573 LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {};
1574
1575 if (!session || !conn->version_check_done) {
1576 ERR("Trying to add a stream before version check");
1577 ret = -1;
1578 goto end_no_session;
1579 }
1580
1581 if (session->minor == 1) {
1582 /* For 2.1 */
1583 ret = cmd_recv_stream_2_1(payload, &path_name,
1584 &channel_name);
1585 } else if (session->minor > 1 && session->minor < 11) {
1586 /* From 2.2 to 2.10 */
1587 ret = cmd_recv_stream_2_2(payload, &path_name,
1588 &channel_name, &tracefile_size, &tracefile_count);
1589 } else {
1590 /* From 2.11 to ... */
1591 ret = cmd_recv_stream_2_11(payload, &path_name,
1592 &channel_name, &tracefile_size, &tracefile_count,
1593 &stream_chunk_id.value);
1594 stream_chunk_id.is_set = true;
1595 }
1596
1597 if (ret < 0) {
1598 goto send_reply;
1599 }
1600
1601 if (conform_channel_path(path_name)) {
1602 goto send_reply;
1603 }
1604
1605 /*
1606 * Backward compatibility for --group-output-by-session.
1607 * Prior to lttng 2.11, the complete path is passed by the stream.
1608 * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1609 * >=2.11 the chunk is responsible for the output path. When dealing
1610 * with producer < 2.11 the chunk output_path is the root output path
1611 * and the stream carries the complete path (path_name).
1612 * To support --group-output-by-session with older producer (<2.11), we
1613 * need to craft the path based on the stream path.
1614 */
1615 if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_SESSION) {
1616 if (conn->minor < 4) {
1617 /*
1618 * From 2.1 to 2.3, the session_name is not passed on
1619 * the RELAYD_CREATE_SESSION command. The session name
1620 * is necessary to detect the presence of a base_path
1621 * inside the stream path. Without it we cannot perform
1622 * a valid group-output-by-session transformation.
1623 */
1624 WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1625 " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1626 session->id, path_name);
1627 } else if (conn->minor >= 4 && conn->minor < 11) {
1628 char *group_by_session_path_name;
1629
1630 assert(session->session_name[0] != '\0');
1631
1632 group_by_session_path_name =
1633 backward_compat_group_by_session(
1634 path_name,
1635 session->session_name);
1636 if (!group_by_session_path_name) {
1637 ERR("Failed to apply group by session to stream of session %" PRIu64,
1638 session->id);
1639 goto send_reply;
1640 }
1641
1642 DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1643 path_name, group_by_session_path_name);
1644
1645 free(path_name);
1646 path_name = group_by_session_path_name;
1647 }
1648 }
1649
1650 trace = ctf_trace_get_by_path_or_create(session, path_name);
1651 if (!trace) {
1652 goto send_reply;
1653 }
1654
1655 /* This stream here has one reference on the trace. */
1656 pthread_mutex_lock(&last_relay_stream_id_lock);
1657 stream_handle = ++last_relay_stream_id;
1658 pthread_mutex_unlock(&last_relay_stream_id_lock);
1659
1660 /* We pass ownership of path_name and channel_name. */
1661 stream = stream_create(trace, stream_handle, path_name,
1662 channel_name, tracefile_size, tracefile_count);
1663 path_name = NULL;
1664 channel_name = NULL;
1665
1666 /*
1667 * Streams are the owners of their trace. Reference to trace is
1668 * kept within stream_create().
1669 */
1670 ctf_trace_put(trace);
1671
1672 send_reply:
1673 memset(&reply, 0, sizeof(reply));
1674 reply.handle = htobe64(stream_handle);
1675 if (!stream) {
1676 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1677 } else {
1678 reply.ret_code = htobe32(LTTNG_OK);
1679 }
1680
1681 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1682 sizeof(struct lttcomm_relayd_status_stream), 0);
1683 if (send_ret < (ssize_t) sizeof(reply)) {
1684 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1685 send_ret);
1686 ret = -1;
1687 }
1688
1689 end_no_session:
1690 free(path_name);
1691 free(channel_name);
1692 return ret;
1693 }
1694
1695 /*
1696 * relay_close_stream: close a specific stream
1697 */
1698 static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1699 struct relay_connection *conn,
1700 const struct lttng_buffer_view *payload)
1701 {
1702 int ret;
1703 ssize_t send_ret;
1704 struct relay_session *session = conn->session;
1705 struct lttcomm_relayd_close_stream stream_info;
1706 struct lttcomm_relayd_generic_reply reply;
1707 struct relay_stream *stream;
1708
1709 DBG("Close stream received");
1710
1711 if (!session || !conn->version_check_done) {
1712 ERR("Trying to close a stream before version check");
1713 ret = -1;
1714 goto end_no_session;
1715 }
1716
1717 if (payload->size < sizeof(stream_info)) {
1718 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1719 sizeof(stream_info), payload->size);
1720 ret = -1;
1721 goto end_no_session;
1722 }
1723 memcpy(&stream_info, payload->data, sizeof(stream_info));
1724 stream_info.stream_id = be64toh(stream_info.stream_id);
1725 stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1726
1727 stream = stream_get_by_id(stream_info.stream_id);
1728 if (!stream) {
1729 ret = -1;
1730 goto end;
1731 }
1732
1733 /*
1734 * Set last_net_seq_num before the close flag. Required by data
1735 * pending check.
1736 */
1737 pthread_mutex_lock(&stream->lock);
1738 stream->last_net_seq_num = stream_info.last_net_seq_num;
1739 pthread_mutex_unlock(&stream->lock);
1740
1741 /*
1742 * This is one of the conditions which may trigger a stream close
1743 * with the others being:
1744 * 1) A close command is received for a stream
1745 * 2) The control connection owning the stream is closed
1746 * 3) We have received all of the stream's data _after_ a close
1747 * request.
1748 */
1749 try_stream_close(stream);
1750 if (stream->is_metadata) {
1751 struct relay_viewer_stream *vstream;
1752
1753 vstream = viewer_stream_get_by_id(stream->stream_handle);
1754 if (vstream) {
1755 if (stream->no_new_metadata_notified) {
1756 /*
1757 * Since all the metadata has been sent to the
1758 * viewer and that we have a request to close
1759 * its stream, we can safely teardown the
1760 * corresponding metadata viewer stream.
1761 */
1762 viewer_stream_put(vstream);
1763 }
1764 /* Put local reference. */
1765 viewer_stream_put(vstream);
1766 }
1767 }
1768 stream_put(stream);
1769 ret = 0;
1770
1771 end:
1772 memset(&reply, 0, sizeof(reply));
1773 if (ret < 0) {
1774 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1775 } else {
1776 reply.ret_code = htobe32(LTTNG_OK);
1777 }
1778 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1779 sizeof(struct lttcomm_relayd_generic_reply), 0);
1780 if (send_ret < (ssize_t) sizeof(reply)) {
1781 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1782 send_ret);
1783 ret = -1;
1784 }
1785
1786 end_no_session:
1787 return ret;
1788 }
1789
1790 /*
1791 * relay_reset_metadata: reset a metadata stream
1792 */
1793 static
1794 int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1795 struct relay_connection *conn,
1796 const struct lttng_buffer_view *payload)
1797 {
1798 int ret;
1799 ssize_t send_ret;
1800 struct relay_session *session = conn->session;
1801 struct lttcomm_relayd_reset_metadata stream_info;
1802 struct lttcomm_relayd_generic_reply reply;
1803 struct relay_stream *stream;
1804
1805 DBG("Reset metadata received");
1806
1807 if (!session || !conn->version_check_done) {
1808 ERR("Trying to reset a metadata stream before version check");
1809 ret = -1;
1810 goto end_no_session;
1811 }
1812
1813 if (payload->size < sizeof(stream_info)) {
1814 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1815 sizeof(stream_info), payload->size);
1816 ret = -1;
1817 goto end_no_session;
1818 }
1819 memcpy(&stream_info, payload->data, sizeof(stream_info));
1820 stream_info.stream_id = be64toh(stream_info.stream_id);
1821 stream_info.version = be64toh(stream_info.version);
1822
1823 DBG("Update metadata to version %" PRIu64, stream_info.version);
1824
1825 /* Unsupported for live sessions for now. */
1826 if (session->live_timer != 0) {
1827 ret = -1;
1828 goto end;
1829 }
1830
1831 stream = stream_get_by_id(stream_info.stream_id);
1832 if (!stream) {
1833 ret = -1;
1834 goto end;
1835 }
1836 pthread_mutex_lock(&stream->lock);
1837 if (!stream->is_metadata) {
1838 ret = -1;
1839 goto end_unlock;
1840 }
1841
1842 ret = stream_reset_file(stream);
1843 if (ret < 0) {
1844 ERR("Failed to reset metadata stream %" PRIu64
1845 ": stream_path = %s, channel = %s",
1846 stream->stream_handle, stream->path_name,
1847 stream->channel_name);
1848 goto end_unlock;
1849 }
1850 end_unlock:
1851 pthread_mutex_unlock(&stream->lock);
1852 stream_put(stream);
1853
1854 end:
1855 memset(&reply, 0, sizeof(reply));
1856 if (ret < 0) {
1857 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1858 } else {
1859 reply.ret_code = htobe32(LTTNG_OK);
1860 }
1861 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1862 sizeof(struct lttcomm_relayd_generic_reply), 0);
1863 if (send_ret < (ssize_t) sizeof(reply)) {
1864 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1865 send_ret);
1866 ret = -1;
1867 }
1868
1869 end_no_session:
1870 return ret;
1871 }
1872
1873 /*
1874 * relay_unknown_command: send -1 if received unknown command
1875 */
1876 static void relay_unknown_command(struct relay_connection *conn)
1877 {
1878 struct lttcomm_relayd_generic_reply reply;
1879 ssize_t send_ret;
1880
1881 memset(&reply, 0, sizeof(reply));
1882 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1883 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1884 if (send_ret < sizeof(reply)) {
1885 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
1886 }
1887 }
1888
1889 /*
1890 * relay_start: send an acknowledgment to the client to tell if we are
1891 * ready to receive data. We are ready if a session is established.
1892 */
1893 static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr,
1894 struct relay_connection *conn,
1895 const struct lttng_buffer_view *payload)
1896 {
1897 int ret = 0;
1898 ssize_t send_ret;
1899 struct lttcomm_relayd_generic_reply reply;
1900 struct relay_session *session = conn->session;
1901
1902 if (!session) {
1903 DBG("Trying to start the streaming without a session established");
1904 ret = htobe32(LTTNG_ERR_UNK);
1905 }
1906
1907 memset(&reply, 0, sizeof(reply));
1908 reply.ret_code = htobe32(LTTNG_OK);
1909 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1910 sizeof(reply), 0);
1911 if (send_ret < (ssize_t) sizeof(reply)) {
1912 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1913 send_ret);
1914 ret = -1;
1915 }
1916
1917 return ret;
1918 }
1919
1920 /*
1921 * relay_recv_metadata: receive the metadata for the session.
1922 */
1923 static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1924 struct relay_connection *conn,
1925 const struct lttng_buffer_view *payload)
1926 {
1927 int ret = 0;
1928 struct relay_session *session = conn->session;
1929 struct lttcomm_relayd_metadata_payload metadata_payload_header;
1930 struct relay_stream *metadata_stream;
1931 uint64_t metadata_payload_size;
1932 struct lttng_buffer_view packet_view;
1933
1934 if (!session) {
1935 ERR("Metadata sent before version check");
1936 ret = -1;
1937 goto end;
1938 }
1939
1940 if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1941 ERR("Incorrect data size");
1942 ret = -1;
1943 goto end;
1944 }
1945 metadata_payload_size = recv_hdr->data_size -
1946 sizeof(struct lttcomm_relayd_metadata_payload);
1947
1948 memcpy(&metadata_payload_header, payload->data,
1949 sizeof(metadata_payload_header));
1950 metadata_payload_header.stream_id = be64toh(
1951 metadata_payload_header.stream_id);
1952 metadata_payload_header.padding_size = be32toh(
1953 metadata_payload_header.padding_size);
1954
1955 metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
1956 if (!metadata_stream) {
1957 ret = -1;
1958 goto end;
1959 }
1960
1961 packet_view = lttng_buffer_view_from_view(payload,
1962 sizeof(metadata_payload_header), metadata_payload_size);
1963 if (!packet_view.data) {
1964 ERR("Invalid metadata packet length announced by header");
1965 ret = -1;
1966 goto end_put;
1967 }
1968
1969 pthread_mutex_lock(&metadata_stream->lock);
1970 ret = stream_write(metadata_stream, &packet_view,
1971 metadata_payload_header.padding_size);
1972 pthread_mutex_unlock(&metadata_stream->lock);
1973 if (ret){
1974 ret = -1;
1975 goto end_put;
1976 }
1977 end_put:
1978 stream_put(metadata_stream);
1979 end:
1980 return ret;
1981 }
1982
1983 /*
1984 * relay_send_version: send relayd version number
1985 */
1986 static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr,
1987 struct relay_connection *conn,
1988 const struct lttng_buffer_view *payload)
1989 {
1990 int ret;
1991 ssize_t send_ret;
1992 struct lttcomm_relayd_version reply, msg;
1993 bool compatible = true;
1994
1995 conn->version_check_done = true;
1996
1997 /* Get version from the other side. */
1998 if (payload->size < sizeof(msg)) {
1999 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
2000 sizeof(msg), payload->size);
2001 ret = -1;
2002 goto end;
2003 }
2004
2005 memcpy(&msg, payload->data, sizeof(msg));
2006 msg.major = be32toh(msg.major);
2007 msg.minor = be32toh(msg.minor);
2008
2009 memset(&reply, 0, sizeof(reply));
2010 reply.major = RELAYD_VERSION_COMM_MAJOR;
2011 reply.minor = RELAYD_VERSION_COMM_MINOR;
2012
2013 /* Major versions must be the same */
2014 if (reply.major != msg.major) {
2015 DBG("Incompatible major versions (%u vs %u), deleting session",
2016 reply.major, msg.major);
2017 compatible = false;
2018 }
2019
2020 conn->major = reply.major;
2021 /* We adapt to the lowest compatible version */
2022 if (reply.minor <= msg.minor) {
2023 conn->minor = reply.minor;
2024 } else {
2025 conn->minor = msg.minor;
2026 }
2027
2028 reply.major = htobe32(reply.major);
2029 reply.minor = htobe32(reply.minor);
2030 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2031 sizeof(reply), 0);
2032 if (send_ret < (ssize_t) sizeof(reply)) {
2033 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2034 send_ret);
2035 ret = -1;
2036 goto end;
2037 } else {
2038 ret = 0;
2039 }
2040
2041 if (!compatible) {
2042 ret = -1;
2043 goto end;
2044 }
2045
2046 DBG("Version check done using protocol %u.%u", conn->major,
2047 conn->minor);
2048
2049 end:
2050 return ret;
2051 }
2052
2053 /*
2054 * Check for data pending for a given stream id from the session daemon.
2055 */
2056 static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2057 struct relay_connection *conn,
2058 const struct lttng_buffer_view *payload)
2059 {
2060 struct relay_session *session = conn->session;
2061 struct lttcomm_relayd_data_pending msg;
2062 struct lttcomm_relayd_generic_reply reply;
2063 struct relay_stream *stream;
2064 ssize_t send_ret;
2065 int ret;
2066 uint64_t stream_seq;
2067
2068 DBG("Data pending command received");
2069
2070 if (!session || !conn->version_check_done) {
2071 ERR("Trying to check for data before version check");
2072 ret = -1;
2073 goto end_no_session;
2074 }
2075
2076 if (payload->size < sizeof(msg)) {
2077 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2078 sizeof(msg), payload->size);
2079 ret = -1;
2080 goto end_no_session;
2081 }
2082 memcpy(&msg, payload->data, sizeof(msg));
2083 msg.stream_id = be64toh(msg.stream_id);
2084 msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
2085
2086 stream = stream_get_by_id(msg.stream_id);
2087 if (stream == NULL) {
2088 ret = -1;
2089 goto end;
2090 }
2091
2092 pthread_mutex_lock(&stream->lock);
2093
2094 if (session_streams_have_index(session)) {
2095 /*
2096 * Ensure that both the index and stream data have been
2097 * flushed up to the requested point.
2098 */
2099 stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
2100 } else {
2101 stream_seq = stream->prev_data_seq;
2102 }
2103 DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
2104 ", prev_index_seq %" PRIu64
2105 ", and last_seq %" PRIu64, msg.stream_id,
2106 stream->prev_data_seq, stream->prev_index_seq,
2107 msg.last_net_seq_num);
2108
2109 /* Avoid wrapping issue */
2110 if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
2111 /* Data has in fact been written and is NOT pending */
2112 ret = 0;
2113 } else {
2114 /* Data still being streamed thus pending */
2115 ret = 1;
2116 }
2117
2118 stream->data_pending_check_done = true;
2119 pthread_mutex_unlock(&stream->lock);
2120
2121 stream_put(stream);
2122 end:
2123
2124 memset(&reply, 0, sizeof(reply));
2125 reply.ret_code = htobe32(ret);
2126 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2127 if (send_ret < (ssize_t) sizeof(reply)) {
2128 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2129 send_ret);
2130 ret = -1;
2131 }
2132
2133 end_no_session:
2134 return ret;
2135 }
2136
2137 /*
2138 * Wait for the control socket to reach a quiescent state.
2139 *
2140 * Note that for now, when receiving this command from the session
2141 * daemon, this means that every subsequent commands or data received on
2142 * the control socket has been handled. So, this is why we simply return
2143 * OK here.
2144 */
2145 static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr,
2146 struct relay_connection *conn,
2147 const struct lttng_buffer_view *payload)
2148 {
2149 int ret;
2150 ssize_t send_ret;
2151 struct relay_stream *stream;
2152 struct lttcomm_relayd_quiescent_control msg;
2153 struct lttcomm_relayd_generic_reply reply;
2154
2155 DBG("Checking quiescent state on control socket");
2156
2157 if (!conn->session || !conn->version_check_done) {
2158 ERR("Trying to check for data before version check");
2159 ret = -1;
2160 goto end_no_session;
2161 }
2162
2163 if (payload->size < sizeof(msg)) {
2164 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2165 sizeof(msg), payload->size);
2166 ret = -1;
2167 goto end_no_session;
2168 }
2169 memcpy(&msg, payload->data, sizeof(msg));
2170 msg.stream_id = be64toh(msg.stream_id);
2171
2172 stream = stream_get_by_id(msg.stream_id);
2173 if (!stream) {
2174 goto reply;
2175 }
2176 pthread_mutex_lock(&stream->lock);
2177 stream->data_pending_check_done = true;
2178 pthread_mutex_unlock(&stream->lock);
2179
2180 DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
2181 stream_put(stream);
2182 reply:
2183 memset(&reply, 0, sizeof(reply));
2184 reply.ret_code = htobe32(LTTNG_OK);
2185 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2186 if (send_ret < (ssize_t) sizeof(reply)) {
2187 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2188 send_ret);
2189 ret = -1;
2190 } else {
2191 ret = 0;
2192 }
2193
2194 end_no_session:
2195 return ret;
2196 }
2197
2198 /*
2199 * Initialize a data pending command. This means that a consumer is about
2200 * to ask for data pending for each stream it holds. Simply iterate over
2201 * all streams of a session and set the data_pending_check_done flag.
2202 *
2203 * This command returns to the client a LTTNG_OK code.
2204 */
2205 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2206 struct relay_connection *conn,
2207 const struct lttng_buffer_view *payload)
2208 {
2209 int ret;
2210 ssize_t send_ret;
2211 struct lttng_ht_iter iter;
2212 struct lttcomm_relayd_begin_data_pending msg;
2213 struct lttcomm_relayd_generic_reply reply;
2214 struct relay_stream *stream;
2215
2216 assert(recv_hdr);
2217 assert(conn);
2218
2219 DBG("Init streams for data pending");
2220
2221 if (!conn->session || !conn->version_check_done) {
2222 ERR("Trying to check for data before version check");
2223 ret = -1;
2224 goto end_no_session;
2225 }
2226
2227 if (payload->size < sizeof(msg)) {
2228 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2229 sizeof(msg), payload->size);
2230 ret = -1;
2231 goto end_no_session;
2232 }
2233 memcpy(&msg, payload->data, sizeof(msg));
2234 msg.session_id = be64toh(msg.session_id);
2235
2236 /*
2237 * Iterate over all streams to set the begin data pending flag.
2238 * For now, the streams are indexed by stream handle so we have
2239 * to iterate over all streams to find the one associated with
2240 * the right session_id.
2241 */
2242 rcu_read_lock();
2243 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2244 node.node) {
2245 if (!stream_get(stream)) {
2246 continue;
2247 }
2248 if (stream->trace->session->id == msg.session_id) {
2249 pthread_mutex_lock(&stream->lock);
2250 stream->data_pending_check_done = false;
2251 pthread_mutex_unlock(&stream->lock);
2252 DBG("Set begin data pending flag to stream %" PRIu64,
2253 stream->stream_handle);
2254 }
2255 stream_put(stream);
2256 }
2257 rcu_read_unlock();
2258
2259 memset(&reply, 0, sizeof(reply));
2260 /* All good, send back reply. */
2261 reply.ret_code = htobe32(LTTNG_OK);
2262
2263 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2264 if (send_ret < (ssize_t) sizeof(reply)) {
2265 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2266 send_ret);
2267 ret = -1;
2268 } else {
2269 ret = 0;
2270 }
2271
2272 end_no_session:
2273 return ret;
2274 }
2275
2276 /*
2277 * End data pending command. This will check, for a given session id, if
2278 * each stream associated with it has its data_pending_check_done flag
2279 * set. If not, this means that the client lost track of the stream but
2280 * the data is still being streamed on our side. In this case, we inform
2281 * the client that data is in flight.
2282 *
2283 * Return to the client if there is data in flight or not with a ret_code.
2284 */
2285 static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2286 struct relay_connection *conn,
2287 const struct lttng_buffer_view *payload)
2288 {
2289 int ret;
2290 ssize_t send_ret;
2291 struct lttng_ht_iter iter;
2292 struct lttcomm_relayd_end_data_pending msg;
2293 struct lttcomm_relayd_generic_reply reply;
2294 struct relay_stream *stream;
2295 uint32_t is_data_inflight = 0;
2296
2297 DBG("End data pending command");
2298
2299 if (!conn->session || !conn->version_check_done) {
2300 ERR("Trying to check for data before version check");
2301 ret = -1;
2302 goto end_no_session;
2303 }
2304
2305 if (payload->size < sizeof(msg)) {
2306 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2307 sizeof(msg), payload->size);
2308 ret = -1;
2309 goto end_no_session;
2310 }
2311 memcpy(&msg, payload->data, sizeof(msg));
2312 msg.session_id = be64toh(msg.session_id);
2313
2314 /*
2315 * Iterate over all streams to see if the begin data pending
2316 * flag is set.
2317 */
2318 rcu_read_lock();
2319 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2320 node.node) {
2321 if (!stream_get(stream)) {
2322 continue;
2323 }
2324 if (stream->trace->session->id != msg.session_id) {
2325 stream_put(stream);
2326 continue;
2327 }
2328 pthread_mutex_lock(&stream->lock);
2329 if (!stream->data_pending_check_done) {
2330 uint64_t stream_seq;
2331
2332 if (session_streams_have_index(conn->session)) {
2333 /*
2334 * Ensure that both the index and stream data have been
2335 * flushed up to the requested point.
2336 */
2337 stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
2338 } else {
2339 stream_seq = stream->prev_data_seq;
2340 }
2341 if (!stream->closed || !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
2342 is_data_inflight = 1;
2343 DBG("Data is still in flight for stream %" PRIu64,
2344 stream->stream_handle);
2345 pthread_mutex_unlock(&stream->lock);
2346 stream_put(stream);
2347 break;
2348 }
2349 }
2350 pthread_mutex_unlock(&stream->lock);
2351 stream_put(stream);
2352 }
2353 rcu_read_unlock();
2354
2355 memset(&reply, 0, sizeof(reply));
2356 /* All good, send back reply. */
2357 reply.ret_code = htobe32(is_data_inflight);
2358
2359 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2360 if (send_ret < (ssize_t) sizeof(reply)) {
2361 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2362 send_ret);
2363 ret = -1;
2364 } else {
2365 ret = 0;
2366 }
2367
2368 end_no_session:
2369 return ret;
2370 }
2371
2372 /*
2373 * Receive an index for a specific stream.
2374 *
2375 * Return 0 on success else a negative value.
2376 */
2377 static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr,
2378 struct relay_connection *conn,
2379 const struct lttng_buffer_view *payload)
2380 {
2381 int ret;
2382 ssize_t send_ret;
2383 struct relay_session *session = conn->session;
2384 struct lttcomm_relayd_index index_info;
2385 struct lttcomm_relayd_generic_reply reply;
2386 struct relay_stream *stream;
2387 size_t msg_len;
2388
2389 assert(conn);
2390
2391 DBG("Relay receiving index");
2392
2393 if (!session || !conn->version_check_done) {
2394 ERR("Trying to close a stream before version check");
2395 ret = -1;
2396 goto end_no_session;
2397 }
2398
2399 msg_len = lttcomm_relayd_index_len(
2400 lttng_to_index_major(conn->major, conn->minor),
2401 lttng_to_index_minor(conn->major, conn->minor));
2402 if (payload->size < msg_len) {
2403 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2404 msg_len, payload->size);
2405 ret = -1;
2406 goto end_no_session;
2407 }
2408 memcpy(&index_info, payload->data, msg_len);
2409 index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2410 index_info.net_seq_num = be64toh(index_info.net_seq_num);
2411 index_info.packet_size = be64toh(index_info.packet_size);
2412 index_info.content_size = be64toh(index_info.content_size);
2413 index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2414 index_info.timestamp_end = be64toh(index_info.timestamp_end);
2415 index_info.events_discarded = be64toh(index_info.events_discarded);
2416 index_info.stream_id = be64toh(index_info.stream_id);
2417
2418 if (conn->minor >= 8) {
2419 index_info.stream_instance_id =
2420 be64toh(index_info.stream_instance_id);
2421 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2422 } else {
2423 index_info.stream_instance_id = -1ULL;
2424 index_info.packet_seq_num = -1ULL;
2425 }
2426
2427 stream = stream_get_by_id(index_info.relay_stream_id);
2428 if (!stream) {
2429 ERR("stream_get_by_id not found");
2430 ret = -1;
2431 goto end;
2432 }
2433
2434 pthread_mutex_lock(&stream->lock);
2435 ret = stream_add_index(stream, &index_info);
2436 pthread_mutex_unlock(&stream->lock);
2437 if (ret) {
2438 goto end_stream_put;
2439 }
2440
2441 end_stream_put:
2442 stream_put(stream);
2443 end:
2444 memset(&reply, 0, sizeof(reply));
2445 if (ret < 0) {
2446 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2447 } else {
2448 reply.ret_code = htobe32(LTTNG_OK);
2449 }
2450 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2451 if (send_ret < (ssize_t) sizeof(reply)) {
2452 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2453 ret = -1;
2454 }
2455
2456 end_no_session:
2457 return ret;
2458 }
2459
2460 /*
2461 * Receive the streams_sent message.
2462 *
2463 * Return 0 on success else a negative value.
2464 */
2465 static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr,
2466 struct relay_connection *conn,
2467 const struct lttng_buffer_view *payload)
2468 {
2469 int ret;
2470 ssize_t send_ret;
2471 struct lttcomm_relayd_generic_reply reply;
2472
2473 assert(conn);
2474
2475 DBG("Relay receiving streams_sent");
2476
2477 if (!conn->session || !conn->version_check_done) {
2478 ERR("Trying to close a stream before version check");
2479 ret = -1;
2480 goto end_no_session;
2481 }
2482
2483 /*
2484 * Publish every pending stream in the connection recv list which are
2485 * now ready to be used by the viewer.
2486 */
2487 publish_connection_local_streams(conn);
2488
2489 memset(&reply, 0, sizeof(reply));
2490 reply.ret_code = htobe32(LTTNG_OK);
2491 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2492 if (send_ret < (ssize_t) sizeof(reply)) {
2493 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2494 send_ret);
2495 ret = -1;
2496 } else {
2497 /* Success. */
2498 ret = 0;
2499 }
2500
2501 end_no_session:
2502 return ret;
2503 }
2504
2505 /*
2506 * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2507 * session rotation feature (not the tracefile rotation feature).
2508 */
2509 static int relay_rotate_session_streams(
2510 const struct lttcomm_relayd_hdr *recv_hdr,
2511 struct relay_connection *conn,
2512 const struct lttng_buffer_view *payload)
2513 {
2514 int ret = 0;
2515 uint32_t i;
2516 ssize_t send_ret;
2517 enum lttng_error_code reply_code = LTTNG_ERR_UNK;
2518 struct relay_session *session = conn->session;
2519 struct lttcomm_relayd_rotate_streams rotate_streams;
2520 struct lttcomm_relayd_generic_reply reply = {};
2521 struct relay_stream *stream = NULL;
2522 const size_t header_len = sizeof(struct lttcomm_relayd_rotate_streams);
2523 struct lttng_trace_chunk *next_trace_chunk = NULL;
2524 struct lttng_buffer_view stream_positions;
2525 char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
2526 const char *chunk_id_str = "none";
2527
2528 if (!session || !conn->version_check_done) {
2529 ERR("Trying to rotate a stream before version check");
2530 ret = -1;
2531 goto end_no_reply;
2532 }
2533
2534 if (session->major == 2 && session->minor < 11) {
2535 ERR("Unsupported feature before 2.11");
2536 ret = -1;
2537 goto end_no_reply;
2538 }
2539
2540 if (payload->size < header_len) {
2541 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2542 header_len, payload->size);
2543 ret = -1;
2544 goto end_no_reply;
2545 }
2546
2547 memcpy(&rotate_streams, payload->data, header_len);
2548
2549 /* Convert header to host endianness. */
2550 rotate_streams = (typeof(rotate_streams)) {
2551 .stream_count = be32toh(rotate_streams.stream_count),
2552 .new_chunk_id = (typeof(rotate_streams.new_chunk_id)) {
2553 .is_set = !!rotate_streams.new_chunk_id.is_set,
2554 .value = be64toh(rotate_streams.new_chunk_id.value),
2555 }
2556 };
2557
2558 if (rotate_streams.new_chunk_id.is_set) {
2559 /*
2560 * Retrieve the trace chunk the stream must transition to. As
2561 * per the protocol, this chunk should have been created
2562 * before this command is received.
2563 */
2564 next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
2565 sessiond_trace_chunk_registry,
2566 session->sessiond_uuid, session->id,
2567 rotate_streams.new_chunk_id.value);
2568 if (!next_trace_chunk) {
2569 char uuid_str[LTTNG_UUID_STR_LEN];
2570
2571 lttng_uuid_to_str(session->sessiond_uuid, uuid_str);
2572 ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2573 ", trace_chunk_id = %" PRIu64,
2574 uuid_str, session->id,
2575 rotate_streams.new_chunk_id.value);
2576 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2577 ret = -1;
2578 goto end;
2579 }
2580
2581 ret = snprintf(chunk_id_buf, sizeof(chunk_id_buf), "%" PRIu64,
2582 rotate_streams.new_chunk_id.value);
2583 if (ret < 0 || ret >= sizeof(chunk_id_buf)) {
2584 chunk_id_str = "formatting error";
2585 } else {
2586 chunk_id_str = chunk_id_buf;
2587 }
2588 }
2589
2590 DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
2591 rotate_streams.stream_count, session->session_name,
2592 chunk_id_str);
2593
2594 stream_positions = lttng_buffer_view_from_view(payload,
2595 sizeof(rotate_streams), -1);
2596 if (!stream_positions.data ||
2597 stream_positions.size <
2598 (rotate_streams.stream_count *
2599 sizeof(struct lttcomm_relayd_stream_rotation_position))) {
2600 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2601 ret = -1;
2602 goto end;
2603 }
2604
2605 for (i = 0; i < rotate_streams.stream_count; i++) {
2606 struct lttcomm_relayd_stream_rotation_position *position_comm =
2607 &((typeof(position_comm)) stream_positions.data)[i];
2608 const struct lttcomm_relayd_stream_rotation_position pos = {
2609 .stream_id = be64toh(position_comm->stream_id),
2610 .rotate_at_seq_num = be64toh(
2611 position_comm->rotate_at_seq_num),
2612 };
2613
2614 stream = stream_get_by_id(pos.stream_id);
2615 if (!stream) {
2616 reply_code = LTTNG_ERR_INVALID;
2617 ret = -1;
2618 goto end;
2619 }
2620
2621 pthread_mutex_lock(&stream->lock);
2622 ret = stream_set_pending_rotation(stream, next_trace_chunk,
2623 pos.rotate_at_seq_num);
2624 pthread_mutex_unlock(&stream->lock);
2625 if (ret) {
2626 reply_code = LTTNG_ERR_FILE_CREATION_ERROR;
2627 goto end;
2628 }
2629
2630 stream_put(stream);
2631 stream = NULL;
2632 }
2633
2634 reply_code = LTTNG_OK;
2635 ret = 0;
2636 end:
2637 if (stream) {
2638 stream_put(stream);
2639 }
2640
2641 reply.ret_code = htobe32((uint32_t) reply_code);
2642 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2643 sizeof(struct lttcomm_relayd_generic_reply), 0);
2644 if (send_ret < (ssize_t) sizeof(reply)) {
2645 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2646 send_ret);
2647 ret = -1;
2648 }
2649 end_no_reply:
2650 lttng_trace_chunk_put(next_trace_chunk);
2651 return ret;
2652 }
2653
2654
2655
2656 /*
2657 * relay_create_trace_chunk: create a new trace chunk
2658 */
2659 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
2660 struct relay_connection *conn,
2661 const struct lttng_buffer_view *payload)
2662 {
2663 int ret = 0;
2664 ssize_t send_ret;
2665 struct relay_session *session = conn->session;
2666 struct lttcomm_relayd_create_trace_chunk *msg;
2667 struct lttcomm_relayd_generic_reply reply = {};
2668 struct lttng_buffer_view header_view;
2669 struct lttng_buffer_view chunk_name_view;
2670 struct lttng_trace_chunk *chunk = NULL, *published_chunk = NULL;
2671 enum lttng_error_code reply_code = LTTNG_OK;
2672 enum lttng_trace_chunk_status chunk_status;
2673 const char *new_path;
2674
2675 if (!session || !conn->version_check_done) {
2676 ERR("Trying to create a trace chunk before version check");
2677 ret = -1;
2678 goto end_no_reply;
2679 }
2680
2681 if (session->major == 2 && session->minor < 11) {
2682 ERR("Chunk creation command is unsupported before 2.11");
2683 ret = -1;
2684 goto end_no_reply;
2685 }
2686
2687 header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2688 if (!header_view.data) {
2689 ERR("Failed to receive payload of chunk creation command");
2690 ret = -1;
2691 goto end_no_reply;
2692 }
2693
2694 /* Convert to host endianness. */
2695 msg = (typeof(msg)) header_view.data;
2696 msg->chunk_id = be64toh(msg->chunk_id);
2697 msg->creation_timestamp = be64toh(msg->creation_timestamp);
2698 msg->override_name_length = be32toh(msg->override_name_length);
2699
2700 if (session->current_trace_chunk &&
2701 !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2702 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2703 DEFAULT_CHUNK_TMP_OLD_DIRECTORY);
2704 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2705 ERR("Failed to rename old chunk");
2706 ret = -1;
2707 reply_code = LTTNG_ERR_UNK;
2708 goto end;
2709 }
2710 }
2711 session->ongoing_rotation = true;
2712 if (!session->current_trace_chunk) {
2713 if (!session->has_rotated) {
2714 new_path = "";
2715 } else {
2716 new_path = NULL;
2717 }
2718 } else {
2719 new_path = DEFAULT_CHUNK_TMP_NEW_DIRECTORY;
2720 }
2721 chunk = lttng_trace_chunk_create(
2722 msg->chunk_id, msg->creation_timestamp, new_path);
2723 if (!chunk) {
2724 ERR("Failed to create trace chunk in trace chunk creation command");
2725 ret = -1;
2726 reply_code = LTTNG_ERR_NOMEM;
2727 goto end;
2728 }
2729 lttng_trace_chunk_set_fd_tracker(chunk, the_fd_tracker);
2730
2731 if (msg->override_name_length) {
2732 const char *name;
2733
2734 chunk_name_view = lttng_buffer_view_from_view(payload,
2735 sizeof(*msg),
2736 msg->override_name_length);
2737 name = chunk_name_view.data;
2738 if (!name || name[msg->override_name_length - 1]) {
2739 ERR("Failed to receive payload of chunk creation command");
2740 ret = -1;
2741 reply_code = LTTNG_ERR_INVALID;
2742 goto end;
2743 }
2744
2745 chunk_status = lttng_trace_chunk_override_name(
2746 chunk, chunk_name_view.data);
2747 switch (chunk_status) {
2748 case LTTNG_TRACE_CHUNK_STATUS_OK:
2749 break;
2750 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
2751 ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2752 reply_code = LTTNG_ERR_INVALID;
2753 ret = -1;
2754 goto end;
2755 default:
2756 ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2757 reply_code = LTTNG_ERR_UNK;
2758 ret = -1;
2759 goto end;
2760 }
2761 }
2762
2763 chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk);
2764 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2765 reply_code = LTTNG_ERR_UNK;
2766 ret = -1;
2767 goto end;
2768 }
2769
2770 assert(conn->session->output_directory);
2771 chunk_status = lttng_trace_chunk_set_as_owner(chunk,
2772 conn->session->output_directory);
2773 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2774 reply_code = LTTNG_ERR_UNK;
2775 ret = -1;
2776 goto end;
2777 }
2778
2779 published_chunk = sessiond_trace_chunk_registry_publish_chunk(
2780 sessiond_trace_chunk_registry,
2781 conn->session->sessiond_uuid,
2782 conn->session->id,
2783 chunk);
2784 if (!published_chunk) {
2785 char uuid_str[LTTNG_UUID_STR_LEN];
2786
2787 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2788 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
2789 uuid_str,
2790 conn->session->id,
2791 msg->chunk_id);
2792 ret = -1;
2793 reply_code = LTTNG_ERR_NOMEM;
2794 goto end;
2795 }
2796
2797 pthread_mutex_lock(&conn->session->lock);
2798 if (conn->session->pending_closure_trace_chunk) {
2799 /*
2800 * Invalid; this means a second create_trace_chunk command was
2801 * received before a close_trace_chunk.
2802 */
2803 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2804 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2805 ret = -1;
2806 goto end_unlock_session;
2807 }
2808 conn->session->pending_closure_trace_chunk =
2809 conn->session->current_trace_chunk;
2810 conn->session->current_trace_chunk = published_chunk;
2811 published_chunk = NULL;
2812 if (!conn->session->pending_closure_trace_chunk) {
2813 session->ongoing_rotation = false;
2814 }
2815 end_unlock_session:
2816 pthread_mutex_unlock(&conn->session->lock);
2817 end:
2818 reply.ret_code = htobe32((uint32_t) reply_code);
2819 send_ret = conn->sock->ops->sendmsg(conn->sock,
2820 &reply,
2821 sizeof(struct lttcomm_relayd_generic_reply),
2822 0);
2823 if (send_ret < (ssize_t) sizeof(reply)) {
2824 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2825 send_ret);
2826 ret = -1;
2827 }
2828 end_no_reply:
2829 lttng_trace_chunk_put(chunk);
2830 lttng_trace_chunk_put(published_chunk);
2831 return ret;
2832 }
2833
2834 /*
2835 * relay_close_trace_chunk: close a trace chunk
2836 */
2837 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
2838 struct relay_connection *conn,
2839 const struct lttng_buffer_view *payload)
2840 {
2841 int ret = 0, buf_ret;
2842 ssize_t send_ret;
2843 struct relay_session *session = conn->session;
2844 struct lttcomm_relayd_close_trace_chunk *msg;
2845 struct lttcomm_relayd_close_trace_chunk_reply reply = {};
2846 struct lttng_buffer_view header_view;
2847 struct lttng_trace_chunk *chunk = NULL;
2848 enum lttng_error_code reply_code = LTTNG_OK;
2849 enum lttng_trace_chunk_status chunk_status;
2850 uint64_t chunk_id;
2851 LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {};
2852 time_t close_timestamp;
2853 char closed_trace_chunk_path[LTTNG_PATH_MAX];
2854 size_t path_length = 0;
2855 const char *chunk_name = NULL;
2856 struct lttng_dynamic_buffer reply_payload;
2857 const char *new_path;
2858
2859 lttng_dynamic_buffer_init(&reply_payload);
2860
2861 if (!session || !conn->version_check_done) {
2862 ERR("Trying to close a trace chunk before version check");
2863 ret = -1;
2864 goto end_no_reply;
2865 }
2866
2867 if (session->major == 2 && session->minor < 11) {
2868 ERR("Chunk close command is unsupported before 2.11");
2869 ret = -1;
2870 goto end_no_reply;
2871 }
2872
2873 header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2874 if (!header_view.data) {
2875 ERR("Failed to receive payload of chunk close command");
2876 ret = -1;
2877 goto end_no_reply;
2878 }
2879
2880 /* Convert to host endianness. */
2881 msg = (typeof(msg)) header_view.data;
2882 chunk_id = be64toh(msg->chunk_id);
2883 close_timestamp = (time_t) be64toh(msg->close_timestamp);
2884 close_command = (typeof(close_command)){
2885 .value = be32toh(msg->close_command.value),
2886 .is_set = msg->close_command.is_set,
2887 };
2888
2889 chunk = sessiond_trace_chunk_registry_get_chunk(
2890 sessiond_trace_chunk_registry,
2891 conn->session->sessiond_uuid,
2892 conn->session->id,
2893 chunk_id);
2894 if (!chunk) {
2895 char uuid_str[LTTNG_UUID_STR_LEN];
2896
2897 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2898 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
2899 uuid_str,
2900 conn->session->id,
2901 msg->chunk_id);
2902 ret = -1;
2903 reply_code = LTTNG_ERR_NOMEM;
2904 goto end;
2905 }
2906
2907 pthread_mutex_lock(&session->lock);
2908 if (close_command.is_set &&
2909 close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) {
2910 /*
2911 * Clear command. It is a protocol error to ask for a
2912 * clear on a relay which does not allow it. Querying
2913 * the configuration allows figuring out whether
2914 * clearing is allowed before doing the clear.
2915 */
2916 if (!opt_allow_clear) {
2917 ret = -1;
2918 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2919 goto end_unlock_session;
2920 }
2921 }
2922 if (session->pending_closure_trace_chunk &&
2923 session->pending_closure_trace_chunk != chunk) {
2924 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
2925 session->session_name);
2926 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2927 ret = -1;
2928 goto end_unlock_session;
2929 }
2930
2931 if (session->current_trace_chunk && session->current_trace_chunk != chunk &&
2932 !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2933 if (close_command.is_set &&
2934 close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE &&
2935 !session->has_rotated) {
2936 /* New chunk stays in session output directory. */
2937 new_path = "";
2938 } else {
2939 /* Use chunk name for new chunk. */
2940 new_path = NULL;
2941 }
2942 /* Rename new chunk path. */
2943 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2944 new_path);
2945 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2946 ret = -1;
2947 goto end_unlock_session;
2948 }
2949 session->ongoing_rotation = false;
2950 }
2951 if ((!close_command.is_set ||
2952 close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION) &&
2953 !lttng_trace_chunk_get_name_overridden(chunk)) {
2954 const char *old_path;
2955
2956 if (!session->has_rotated) {
2957 old_path = "";
2958 } else {
2959 old_path = NULL;
2960 }
2961 /* We need to move back the .tmp_old_chunk to its rightful place. */
2962 chunk_status = lttng_trace_chunk_rename_path(chunk, old_path);
2963 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2964 ret = -1;
2965 goto end_unlock_session;
2966 }
2967 }
2968 chunk_status = lttng_trace_chunk_set_close_timestamp(
2969 chunk, close_timestamp);
2970 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2971 ERR("Failed to set trace chunk close timestamp");
2972 ret = -1;
2973 reply_code = LTTNG_ERR_UNK;
2974 goto end_unlock_session;
2975 }
2976
2977 if (close_command.is_set) {
2978 chunk_status = lttng_trace_chunk_set_close_command(
2979 chunk, close_command.value);
2980 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2981 ret = -1;
2982 reply_code = LTTNG_ERR_INVALID;
2983 goto end_unlock_session;
2984 }
2985 }
2986 chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, NULL);
2987 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2988 ERR("Failed to get chunk name");
2989 ret = -1;
2990 reply_code = LTTNG_ERR_UNK;
2991 goto end_unlock_session;
2992 }
2993 if (!session->has_rotated && !session->snapshot) {
2994 ret = lttng_strncpy(closed_trace_chunk_path,
2995 session->output_path,
2996 sizeof(closed_trace_chunk_path));
2997 if (ret) {
2998 ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
2999 strlen(session->output_path),
3000 sizeof(closed_trace_chunk_path));
3001 reply_code = LTTNG_ERR_NOMEM;
3002 ret = -1;
3003 goto end_unlock_session;
3004 }
3005 } else {
3006 if (session->snapshot) {
3007 ret = snprintf(closed_trace_chunk_path,
3008 sizeof(closed_trace_chunk_path),
3009 "%s/%s", session->output_path,
3010 chunk_name);
3011 } else {
3012 ret = snprintf(closed_trace_chunk_path,
3013 sizeof(closed_trace_chunk_path),
3014 "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY
3015 "/%s",
3016 session->output_path, chunk_name);
3017 }
3018 if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) {
3019 ERR("Failed to format closed trace chunk resulting path");
3020 reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM;
3021 ret = -1;
3022 goto end_unlock_session;
3023 }
3024 }
3025 if (close_command.is_set &&
3026 close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) {
3027 session->has_rotated = true;
3028 }
3029 DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
3030 path_length = strlen(closed_trace_chunk_path) + 1;
3031 if (path_length > UINT32_MAX) {
3032 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3033 ret = -1;
3034 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3035 goto end_unlock_session;
3036 }
3037
3038 if (session->current_trace_chunk == chunk) {
3039 /*
3040 * After a trace chunk close command, no new streams
3041 * referencing the chunk may be created. Hence, on the
3042 * event that no new trace chunk have been created for
3043 * the session, the reference to the current trace chunk
3044 * is released in order to allow it to be reclaimed when
3045 * the last stream releases its reference to it.
3046 */
3047 lttng_trace_chunk_put(session->current_trace_chunk);
3048 session->current_trace_chunk = NULL;
3049 }
3050 lttng_trace_chunk_put(session->pending_closure_trace_chunk);
3051 session->pending_closure_trace_chunk = NULL;
3052 end_unlock_session:
3053 pthread_mutex_unlock(&session->lock);
3054
3055 end:
3056 reply.generic.ret_code = htobe32((uint32_t) reply_code);
3057 reply.path_length = htobe32((uint32_t) path_length);
3058 buf_ret = lttng_dynamic_buffer_append(
3059 &reply_payload, &reply, sizeof(reply));
3060 if (buf_ret) {
3061 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3062 goto end_no_reply;
3063 }
3064
3065 if (reply_code == LTTNG_OK) {
3066 buf_ret = lttng_dynamic_buffer_append(&reply_payload,
3067 closed_trace_chunk_path, path_length);
3068 if (buf_ret) {
3069 ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3070 goto end_no_reply;
3071 }
3072 }
3073
3074 send_ret = conn->sock->ops->sendmsg(conn->sock,
3075 reply_payload.data,
3076 reply_payload.size,
3077 0);
3078 if (send_ret < reply_payload.size) {
3079 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3080 reply_payload.size, send_ret);
3081 ret = -1;
3082 goto end_no_reply;
3083 }
3084 end_no_reply:
3085 lttng_trace_chunk_put(chunk);
3086 lttng_dynamic_buffer_reset(&reply_payload);
3087 return ret;
3088 }
3089
3090 /*
3091 * relay_trace_chunk_exists: check if a trace chunk exists
3092 */
3093 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr,
3094 struct relay_connection *conn,
3095 const struct lttng_buffer_view *payload)
3096 {
3097 int ret = 0;
3098 ssize_t send_ret;
3099 struct relay_session *session = conn->session;
3100 struct lttcomm_relayd_trace_chunk_exists *msg;
3101 struct lttcomm_relayd_trace_chunk_exists_reply reply = {};
3102 struct lttng_buffer_view header_view;
3103 uint64_t chunk_id;
3104 bool chunk_exists;
3105
3106 if (!session || !conn->version_check_done) {
3107 ERR("Trying to close a trace chunk before version check");
3108 ret = -1;
3109 goto end_no_reply;
3110 }
3111
3112 if (session->major == 2 && session->minor < 11) {
3113 ERR("Chunk close command is unsupported before 2.11");
3114 ret = -1;
3115 goto end_no_reply;
3116 }
3117
3118 header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3119 if (!header_view.data) {
3120 ERR("Failed to receive payload of chunk close command");
3121 ret = -1;
3122 goto end_no_reply;
3123 }
3124
3125 /* Convert to host endianness. */
3126 msg = (typeof(msg)) header_view.data;
3127 chunk_id = be64toh(msg->chunk_id);
3128
3129 ret = sessiond_trace_chunk_registry_chunk_exists(
3130 sessiond_trace_chunk_registry,
3131 conn->session->sessiond_uuid,
3132 conn->session->id,
3133 chunk_id, &chunk_exists);
3134 /*
3135 * If ret is not 0, send the reply and report the error to the caller.
3136 * It is a protocol (or internal) error and the session/connection
3137 * should be torn down.
3138 */
3139 reply = (typeof(reply)){
3140 .generic.ret_code = htobe32((uint32_t)
3141 (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)),
3142 .trace_chunk_exists = ret == 0 ? chunk_exists : 0,
3143 };
3144 send_ret = conn->sock->ops->sendmsg(
3145 conn->sock, &reply, sizeof(reply), 0);
3146 if (send_ret < (ssize_t) sizeof(reply)) {
3147 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
3148 send_ret);
3149 ret = -1;
3150 }
3151 end_no_reply:
3152 return ret;
3153 }
3154
3155 /*
3156 * relay_get_configuration: query whether feature is available
3157 */
3158 static int relay_get_configuration(const struct lttcomm_relayd_hdr *recv_hdr,
3159 struct relay_connection *conn,
3160 const struct lttng_buffer_view *payload)
3161 {
3162 int ret = 0;
3163 ssize_t send_ret;
3164 struct lttcomm_relayd_get_configuration *msg;
3165 struct lttcomm_relayd_get_configuration_reply reply = {};
3166 struct lttng_buffer_view header_view;
3167 uint64_t query_flags = 0;
3168 uint64_t result_flags = 0;
3169
3170 header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3171 if (!header_view.data) {
3172 ERR("Failed to receive payload of chunk close command");
3173 ret = -1;
3174 goto end_no_reply;
3175 }
3176
3177 /* Convert to host endianness. */
3178 msg = (typeof(msg)) header_view.data;
3179 query_flags = be64toh(msg->query_flags);
3180
3181 if (query_flags) {
3182 ret = LTTNG_ERR_INVALID_PROTOCOL;
3183 goto reply;
3184 }
3185 if (opt_allow_clear) {
3186 result_flags |= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED;
3187 }
3188 ret = 0;
3189 reply:
3190 reply = (typeof(reply)){
3191 .generic.ret_code = htobe32((uint32_t)
3192 (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)),
3193 .relayd_configuration_flags = htobe64(result_flags),
3194 };
3195 send_ret = conn->sock->ops->sendmsg(
3196 conn->sock, &reply, sizeof(reply), 0);
3197 if (send_ret < (ssize_t) sizeof(reply)) {
3198 ERR("Failed to send \"get configuration\" command reply (ret = %zd)",
3199 send_ret);
3200 ret = -1;
3201 }
3202 end_no_reply:
3203 return ret;
3204 }
3205
3206 #define DBG_CMD(cmd_name, conn) \
3207 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
3208
3209 static int relay_process_control_command(struct relay_connection *conn,
3210 const struct lttcomm_relayd_hdr *header,
3211 const struct lttng_buffer_view *payload)
3212 {
3213 int ret = 0;
3214
3215 switch (header->cmd) {
3216 case RELAYD_CREATE_SESSION:
3217 DBG_CMD("RELAYD_CREATE_SESSION", conn);
3218 ret = relay_create_session(header, conn, payload);
3219 break;
3220 case RELAYD_ADD_STREAM:
3221 DBG_CMD("RELAYD_ADD_STREAM", conn);
3222 ret = relay_add_stream(header, conn, payload);
3223 break;
3224 case RELAYD_START_DATA:
3225 DBG_CMD("RELAYD_START_DATA", conn);
3226 ret = relay_start(header, conn, payload);
3227 break;
3228 case RELAYD_SEND_METADATA:
3229 DBG_CMD("RELAYD_SEND_METADATA", conn);
3230 ret = relay_recv_metadata(header, conn, payload);
3231 break;
3232 case RELAYD_VERSION:
3233 DBG_CMD("RELAYD_VERSION", conn);
3234 ret = relay_send_version(header, conn, payload);
3235 break;
3236 case RELAYD_CLOSE_STREAM:
3237 DBG_CMD("RELAYD_CLOSE_STREAM", conn);
3238 ret = relay_close_stream(header, conn, payload);
3239 break;
3240 case RELAYD_DATA_PENDING:
3241 DBG_CMD("RELAYD_DATA_PENDING", conn);
3242 ret = relay_data_pending(header, conn, payload);
3243 break;
3244 case RELAYD_QUIESCENT_CONTROL:
3245 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn);
3246 ret = relay_quiescent_control(header, conn, payload);
3247 break;
3248 case RELAYD_BEGIN_DATA_PENDING:
3249 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn);
3250 ret = relay_begin_data_pending(header, conn, payload);
3251 break;
3252 case RELAYD_END_DATA_PENDING:
3253 DBG_CMD("RELAYD_END_DATA_PENDING", conn);
3254 ret = relay_end_data_pending(header, conn, payload);
3255 break;
3256 case RELAYD_SEND_INDEX:
3257 DBG_CMD("RELAYD_SEND_INDEX", conn);
3258 ret = relay_recv_index(header, conn, payload);
3259 break;
3260 case RELAYD_STREAMS_SENT:
3261 DBG_CMD("RELAYD_STREAMS_SENT", conn);
3262 ret = relay_streams_sent(header, conn, payload);
3263 break;
3264 case RELAYD_RESET_METADATA:
3265 DBG_CMD("RELAYD_RESET_METADATA", conn);
3266 ret = relay_reset_metadata(header, conn, payload);
3267 break;
3268 case RELAYD_ROTATE_STREAMS:
3269 DBG_CMD("RELAYD_ROTATE_STREAMS", conn);
3270 ret = relay_rotate_session_streams(header, conn, payload);
3271 break;
3272 case RELAYD_CREATE_TRACE_CHUNK:
3273 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn);
3274 ret = relay_create_trace_chunk(header, conn, payload);
3275 break;
3276 case RELAYD_CLOSE_TRACE_CHUNK:
3277 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn);
3278 ret = relay_close_trace_chunk(header, conn, payload);
3279 break;
3280 case RELAYD_TRACE_CHUNK_EXISTS:
3281 DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn);
3282 ret = relay_trace_chunk_exists(header, conn, payload);
3283 break;
3284 case RELAYD_GET_CONFIGURATION:
3285 DBG_CMD("RELAYD_GET_CONFIGURATION", conn);
3286 ret = relay_get_configuration(header, conn, payload);
3287 break;
3288 case RELAYD_UPDATE_SYNC_INFO:
3289 default:
3290 ERR("Received unknown command (%u)", header->cmd);
3291 relay_unknown_command(conn);
3292 ret = -1;
3293 goto end;
3294 }
3295
3296 end:
3297 return ret;
3298 }
3299
3300 static enum relay_connection_status relay_process_control_receive_payload(
3301 struct relay_connection *conn)
3302 {
3303 int ret = 0;
3304 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3305 struct lttng_dynamic_buffer *reception_buffer =
3306 &conn->protocol.ctrl.reception_buffer;
3307 struct ctrl_connection_state_receive_payload *state =
3308 &conn->protocol.ctrl.state.receive_payload;
3309 struct lttng_buffer_view payload_view;
3310
3311 if (state->left_to_receive == 0) {
3312 /* Short-circuit for payload-less commands. */
3313 goto reception_complete;
3314 }
3315
3316 ret = conn->sock->ops->recvmsg(conn->sock,
3317 reception_buffer->data + state->received,
3318 state->left_to_receive, MSG_DONTWAIT);
3319 if (ret < 0) {
3320 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3321 PERROR("Unable to receive command payload on sock %d",
3322 conn->sock->fd);
3323 status = RELAY_CONNECTION_STATUS_ERROR;
3324 }
3325 goto end;
3326 } else if (ret == 0) {
3327 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3328 status = RELAY_CONNECTION_STATUS_CLOSED;
3329 goto end;
3330 }
3331
3332 assert(ret > 0);
3333 assert(ret <= state->left_to_receive);
3334
3335 state->left_to_receive -= ret;
3336 state->received += ret;
3337
3338 if (state->left_to_receive > 0) {
3339 /*
3340 * Can't transition to the protocol's next state, wait to
3341 * receive the rest of the header.
3342 */
3343 DBG3("Partial reception of control connection protocol payload (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3344 state->received, state->left_to_receive,
3345 conn->sock->fd);
3346 goto end;
3347 }
3348
3349 reception_complete:
3350 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3351 conn->sock->fd, state->received);
3352 /*
3353 * The payload required to process the command has been received.
3354 * A view to the reception buffer is forwarded to the various
3355 * commands and the state of the control is reset on success.
3356 *
3357 * Commands are responsible for sending their reply to the peer.
3358 */
3359 payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer,
3360 0, -1);
3361 ret = relay_process_control_command(conn,
3362 &state->header, &payload_view);
3363 if (ret < 0) {
3364 status = RELAY_CONNECTION_STATUS_ERROR;
3365 goto end;
3366 }
3367
3368 ret = connection_reset_protocol_state(conn);
3369 if (ret) {
3370 status = RELAY_CONNECTION_STATUS_ERROR;
3371 }
3372 end:
3373 return status;
3374 }
3375
3376 static enum relay_connection_status relay_process_control_receive_header(
3377 struct relay_connection *conn)
3378 {
3379 int ret = 0;
3380 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3381 struct lttcomm_relayd_hdr header;
3382 struct lttng_dynamic_buffer *reception_buffer =
3383 &conn->protocol.ctrl.reception_buffer;
3384 struct ctrl_connection_state_receive_header *state =
3385 &conn->protocol.ctrl.state.receive_header;
3386
3387 assert(state->left_to_receive != 0);
3388
3389 ret = conn->sock->ops->recvmsg(conn->sock,
3390 reception_buffer->data + state->received,
3391 state->left_to_receive, MSG_DONTWAIT);
3392 if (ret < 0) {
3393 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3394 PERROR("Unable to receive control command header on sock %d",
3395 conn->sock->fd);
3396 status = RELAY_CONNECTION_STATUS_ERROR;
3397 }
3398 goto end;
3399 } else if (ret == 0) {
3400 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3401 status = RELAY_CONNECTION_STATUS_CLOSED;
3402 goto end;
3403 }
3404
3405 assert(ret > 0);
3406 assert(ret <= state->left_to_receive);
3407
3408 state->left_to_receive -= ret;
3409 state->received += ret;
3410
3411 if (state->left_to_receive > 0) {
3412 /*
3413 * Can't transition to the protocol's next state, wait to
3414 * receive the rest of the header.
3415 */
3416 DBG3("Partial reception of control connection protocol header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3417 state->received, state->left_to_receive,
3418 conn->sock->fd);
3419 goto end;
3420 }
3421
3422 /* Transition to next state: receiving the command's payload. */
3423 conn->protocol.ctrl.state_id =
3424 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3425 memcpy(&header, reception_buffer->data, sizeof(header));
3426 header.circuit_id = be64toh(header.circuit_id);
3427 header.data_size = be64toh(header.data_size);
3428 header.cmd = be32toh(header.cmd);
3429 header.cmd_version = be32toh(header.cmd_version);
3430 memcpy(&conn->protocol.ctrl.state.receive_payload.header,
3431 &header, sizeof(header));
3432
3433 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32 ", cmd_version = %" PRIu32 ", payload size = %" PRIu64 " bytes",
3434 conn->sock->fd, header.cmd, header.cmd_version,
3435 header.data_size);
3436
3437 if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
3438 ERR("Command header indicates a payload (%" PRIu64 " bytes) that exceeds the maximal payload size allowed on a control connection.",
3439 header.data_size);
3440 status = RELAY_CONNECTION_STATUS_ERROR;
3441 goto end;
3442 }
3443
3444 conn->protocol.ctrl.state.receive_payload.left_to_receive =
3445 header.data_size;
3446 conn->protocol.ctrl.state.receive_payload.received = 0;
3447 ret = lttng_dynamic_buffer_set_size(reception_buffer,
3448 header.data_size);
3449 if (ret) {
3450 status = RELAY_CONNECTION_STATUS_ERROR;
3451 goto end;
3452 }
3453
3454 if (header.data_size == 0) {
3455 /*
3456 * Manually invoke the next state as the poll loop
3457 * will not wake-up to allow us to proceed further.
3458 */
3459 status = relay_process_control_receive_payload(conn);
3460 }
3461 end:
3462 return status;
3463 }
3464
3465 /*
3466 * Process the commands received on the control socket
3467 */
3468 static enum relay_connection_status relay_process_control(
3469 struct relay_connection *conn)
3470 {
3471 enum relay_connection_status status;
3472
3473 switch (conn->protocol.ctrl.state_id) {
3474 case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
3475 status = relay_process_control_receive_header(conn);
3476 break;
3477 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
3478 status = relay_process_control_receive_payload(conn);
3479 break;
3480 default:
3481 ERR("Unknown control connection protocol state encountered.");
3482 abort();
3483 }
3484
3485 return status;
3486 }
3487
3488 static enum relay_connection_status relay_process_data_receive_header(
3489 struct relay_connection *conn)
3490 {
3491 int ret;
3492 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3493 struct data_connection_state_receive_header *state =
3494 &conn->protocol.data.state.receive_header;
3495 struct lttcomm_relayd_data_hdr header;
3496 struct relay_stream *stream;
3497
3498 assert(state->left_to_receive != 0);
3499
3500 ret = conn->sock->ops->recvmsg(conn->sock,
3501 state->header_reception_buffer + state->received,
3502 state->left_to_receive, MSG_DONTWAIT);
3503 if (ret < 0) {
3504 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3505 PERROR("Unable to receive data header on sock %d", conn->sock->fd);
3506 status = RELAY_CONNECTION_STATUS_ERROR;
3507 }
3508 goto end;
3509 } else if (ret == 0) {
3510 /* Orderly shutdown. Not necessary to print an error. */
3511 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3512 status = RELAY_CONNECTION_STATUS_CLOSED;
3513 goto end;
3514 }
3515
3516 assert(ret > 0);
3517 assert(ret <= state->left_to_receive);
3518
3519 state->left_to_receive -= ret;
3520 state->received += ret;
3521
3522 if (state->left_to_receive > 0) {
3523 /*
3524 * Can't transition to the protocol's next state, wait to
3525 * receive the rest of the header.
3526 */
3527 DBG3("Partial reception of data connection header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3528 state->received, state->left_to_receive,
3529 conn->sock->fd);
3530 goto end;
3531 }
3532
3533 /* Transition to next state: receiving the payload. */
3534 conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
3535
3536 memcpy(&header, state->header_reception_buffer, sizeof(header));
3537 header.circuit_id = be64toh(header.circuit_id);
3538 header.stream_id = be64toh(header.stream_id);
3539 header.data_size = be32toh(header.data_size);
3540 header.net_seq_num = be64toh(header.net_seq_num);
3541 header.padding_size = be32toh(header.padding_size);
3542 memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
3543
3544 conn->protocol.data.state.receive_payload.left_to_receive =
3545 header.data_size;
3546 conn->protocol.data.state.receive_payload.received = 0;
3547 conn->protocol.data.state.receive_payload.rotate_index = false;
3548
3549 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64 ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64 ", padding_size = %" PRIu32,
3550 conn->sock->fd, header.circuit_id,
3551 header.stream_id, header.data_size,
3552 header.net_seq_num, header.padding_size);
3553
3554 stream = stream_get_by_id(header.stream_id);
3555 if (!stream) {
3556 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
3557 header.stream_id);
3558 /* Protocol error. */
3559 status = RELAY_CONNECTION_STATUS_ERROR;
3560 goto end;
3561 }
3562
3563 pthread_mutex_lock(&stream->lock);
3564 /* Prepare stream for the reception of a new packet. */
3565 ret = stream_init_packet(stream, header.data_size,
3566 &conn->protocol.data.state.receive_payload.rotate_index);
3567 pthread_mutex_unlock(&stream->lock);
3568 if (ret) {
3569 ERR("Failed to rotate stream output file");
3570 status = RELAY_CONNECTION_STATUS_ERROR;
3571 goto end_stream_unlock;
3572 }
3573
3574 end_stream_unlock:
3575 stream_put(stream);
3576 end:
3577 return status;
3578 }
3579
3580 static enum relay_connection_status relay_process_data_receive_payload(
3581 struct relay_connection *conn)
3582 {
3583 int ret;
3584 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3585 struct relay_stream *stream;
3586 struct data_connection_state_receive_payload *state =
3587 &conn->protocol.data.state.receive_payload;
3588 const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
3589 char data_buffer[chunk_size];
3590 bool partial_recv = false;
3591 bool new_stream = false, close_requested = false, index_flushed = false;
3592 uint64_t left_to_receive = state->left_to_receive;
3593 struct relay_session *session;
3594
3595 DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64" bytes received, %" PRIu64 " bytes left to receive",
3596 state->header.stream_id, state->header.net_seq_num,
3597 state->received, left_to_receive);
3598
3599 stream = stream_get_by_id(state->header.stream_id);
3600 if (!stream) {
3601 /* Protocol error. */
3602 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
3603 state->header.stream_id);
3604 status = RELAY_CONNECTION_STATUS_ERROR;
3605 goto end;
3606 }
3607
3608 pthread_mutex_lock(&stream->lock);
3609 session = stream->trace->session;
3610 if (!conn->session) {
3611 ret = connection_set_session(conn, session);
3612 if (ret) {
3613 status = RELAY_CONNECTION_STATUS_ERROR;
3614 goto end_stream_unlock;
3615 }
3616 }
3617
3618 /*
3619 * The size of the "chunk" received on any iteration is bounded by:
3620 * - the data left to receive,
3621 * - the data immediately available on the socket,
3622 * - the on-stack data buffer
3623 */
3624 while (left_to_receive > 0 && !partial_recv) {
3625 size_t recv_size = min(left_to_receive, chunk_size);
3626 struct lttng_buffer_view packet_chunk;
3627
3628 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer,
3629 recv_size, MSG_DONTWAIT);
3630 if (ret < 0) {
3631 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3632 PERROR("Socket %d error", conn->sock->fd);
3633 status = RELAY_CONNECTION_STATUS_ERROR;
3634 }
3635 goto end_stream_unlock;
3636 } else if (ret == 0) {
3637 /* No more data ready to be consumed on socket. */
3638 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
3639 state->header.stream_id);
3640 status = RELAY_CONNECTION_STATUS_CLOSED;
3641 break;
3642 } else if (ret < (int) recv_size) {
3643 /*
3644 * All the data available on the socket has been
3645 * consumed.
3646 */
3647 partial_recv = true;
3648 recv_size = ret;
3649 }
3650
3651 packet_chunk = lttng_buffer_view_init(data_buffer,
3652 0, recv_size);
3653 assert(packet_chunk.data);
3654
3655 ret = stream_write(stream, &packet_chunk, 0);
3656 if (ret) {
3657 ERR("Relay error writing data to file");
3658 status = RELAY_CONNECTION_STATUS_ERROR;
3659 goto end_stream_unlock;
3660 }
3661
3662 left_to_receive -= recv_size;
3663 state->received += recv_size;
3664 state->left_to_receive = left_to_receive;
3665 }
3666
3667 if (state->left_to_receive > 0) {
3668 /*
3669 * Did not receive all the data expected, wait for more data to
3670 * become available on the socket.
3671 */
3672 DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64 " bytes received, %" PRIu64 " bytes left to receive",
3673 state->header.stream_id, state->received,
3674 state->left_to_receive);
3675 goto end_stream_unlock;
3676 }
3677
3678 ret = stream_write(stream, NULL, state->header.padding_size);
3679 if (ret) {
3680 status = RELAY_CONNECTION_STATUS_ERROR;
3681 goto end_stream_unlock;
3682 }
3683
3684 if (session_streams_have_index(session)) {
3685 ret = stream_update_index(stream, state->header.net_seq_num,
3686 state->rotate_index, &index_flushed,
3687 state->header.data_size + state->header.padding_size);
3688 if (ret < 0) {
3689 ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
3690 stream->stream_handle,
3691 state->header.net_seq_num, ret);
3692 status = RELAY_CONNECTION_STATUS_ERROR;
3693 goto end_stream_unlock;
3694 }
3695 }
3696
3697 if (stream->prev_data_seq == -1ULL) {
3698 new_stream = true;
3699 }
3700
3701 ret = stream_complete_packet(stream, state->header.data_size +
3702 state->header.padding_size, state->header.net_seq_num,
3703 index_flushed);
3704 if (ret) {
3705 status = RELAY_CONNECTION_STATUS_ERROR;
3706 goto end_stream_unlock;
3707 }
3708
3709 /*
3710 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3711 * contents of *state which are aliased (union) to the same location as
3712 * the new state. Don't use it beyond this point.
3713 */
3714 connection_reset_protocol_state(conn);
3715 state = NULL;
3716
3717 end_stream_unlock:
3718 close_requested = stream->close_requested;
3719 pthread_mutex_unlock(&stream->lock);
3720 if (close_requested && left_to_receive == 0) {
3721 try_stream_close(stream);
3722 }
3723
3724 if (new_stream) {
3725 pthread_mutex_lock(&session->lock);
3726 uatomic_set(&session->new_streams, 1);
3727 pthread_mutex_unlock(&session->lock);
3728 }
3729
3730 stream_put(stream);
3731 end:
3732 return status;
3733 }
3734
3735 /*
3736 * relay_process_data: Process the data received on the data socket
3737 */
3738 static enum relay_connection_status relay_process_data(
3739 struct relay_connection *conn)
3740 {
3741 enum relay_connection_status status;
3742
3743 switch (conn->protocol.data.state_id) {
3744 case DATA_CONNECTION_STATE_RECEIVE_HEADER:
3745 status = relay_process_data_receive_header(conn);
3746 break;
3747 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
3748 status = relay_process_data_receive_payload(conn);
3749 break;
3750 default:
3751 ERR("Unexpected data connection communication state.");
3752 abort();
3753 }
3754
3755 return status;
3756 }
3757
3758 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
3759 {
3760 int ret;
3761
3762 (void) lttng_poll_del(events, pollfd);
3763
3764 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, &pollfd, 1,
3765 fd_tracker_util_close_fd, NULL);
3766 if (ret < 0) {
3767 ERR("Closing pollfd %d", pollfd);
3768 }
3769 }
3770
3771 static void relay_thread_close_connection(struct lttng_poll_event *events,
3772 int pollfd, struct relay_connection *conn)
3773 {
3774 const char *type_str;
3775
3776 switch (conn->type) {
3777 case RELAY_DATA:
3778 type_str = "Data";
3779 break;
3780 case RELAY_CONTROL:
3781 type_str = "Control";
3782 break;
3783 case RELAY_VIEWER_COMMAND:
3784 type_str = "Viewer Command";
3785 break;
3786 case RELAY_VIEWER_NOTIFICATION:
3787 type_str = "Viewer Notification";
3788 break;
3789 default:
3790 type_str = "Unknown";
3791 }
3792 cleanup_connection_pollfd(events, pollfd);
3793 connection_put(conn);
3794 DBG("%s connection closed with %d", type_str, pollfd);
3795 }
3796
3797 /*
3798 * This thread does the actual work
3799 */
3800 static void *relay_thread_worker(void *data)
3801 {
3802 int ret, err = -1, last_seen_data_fd = -1;
3803 uint32_t nb_fd;
3804 struct lttng_poll_event events;
3805 struct lttng_ht *relay_connections_ht;
3806 struct lttng_ht_iter iter;
3807 struct relay_connection *destroy_conn = NULL;
3808
3809 DBG("[thread] Relay worker started");
3810
3811 rcu_register_thread();
3812
3813 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
3814
3815 if (testpoint(relayd_thread_worker)) {
3816 goto error_testpoint;
3817 }
3818
3819 health_code_update();
3820
3821 /* table of connections indexed on socket */
3822 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
3823 if (!relay_connections_ht) {
3824 goto relay_connections_ht_error;
3825 }
3826
3827 ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll");
3828 if (ret < 0) {
3829 goto error_poll_create;
3830 }
3831
3832 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
3833 if (ret < 0) {
3834 goto error;
3835 }
3836
3837 restart:
3838 while (1) {
3839 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
3840
3841 health_code_update();
3842
3843 /* Infinite blocking call, waiting for transmission */
3844 DBG3("Relayd worker thread polling...");
3845 health_poll_entry();
3846 ret = lttng_poll_wait(&events, -1);
3847 health_poll_exit();
3848 if (ret < 0) {
3849 /*
3850 * Restart interrupted system call.
3851 */
3852 if (errno == EINTR) {
3853 goto restart;
3854 }
3855 goto error;
3856 }
3857
3858 nb_fd = ret;
3859
3860 /*
3861 * Process control. The control connection is
3862 * prioritized so we don't starve it with high
3863 * throughput tracing data on the data connection.
3864 */
3865 for (i = 0; i < nb_fd; i++) {
3866 /* Fetch once the poll data */
3867 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
3868 int pollfd = LTTNG_POLL_GETFD(&events, i);
3869
3870 health_code_update();
3871
3872 /* Thread quit pipe has been closed. Killing thread. */
3873 ret = check_thread_quit_pipe(pollfd, revents);
3874 if (ret) {
3875 err = 0;
3876 goto exit;
3877 }
3878
3879 /* Inspect the relay conn pipe for new connection */
3880 if (pollfd == relay_conn_pipe[0]) {
3881 if (revents & LPOLLIN) {
3882 struct relay_connection *conn;
3883
3884 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
3885 if (ret < 0) {
3886 goto error;
3887 }
3888 ret = lttng_poll_add(&events,
3889 conn->sock->fd,
3890 LPOLLIN | LPOLLRDHUP);
3891 if (ret) {
3892 ERR("Failed to add new connection file descriptor to poll set");
3893 goto error;
3894 }
3895 connection_ht_add(relay_connections_ht, conn);
3896 DBG("Connection socket %d added", conn->sock->fd);
3897 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3898 ERR("Relay connection pipe error");
3899 goto error;
3900 } else {
3901 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3902 goto error;
3903 }
3904 } else {
3905 struct relay_connection *ctrl_conn;
3906
3907 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
3908 /* If not found, there is a synchronization issue. */
3909 assert(ctrl_conn);
3910
3911 if (ctrl_conn->type == RELAY_DATA) {
3912 if (revents & LPOLLIN) {
3913 /*
3914 * Flag the last seen data fd not deleted. It will be
3915 * used as the last seen fd if any fd gets deleted in
3916 * this first loop.
3917 */
3918 last_notdel_data_fd = pollfd;
3919 }
3920 goto put_ctrl_connection;
3921 }
3922 assert(ctrl_conn->type == RELAY_CONTROL);
3923
3924 if (revents & LPOLLIN) {
3925 enum relay_connection_status status;
3926
3927 status = relay_process_control(ctrl_conn);
3928 if (status != RELAY_CONNECTION_STATUS_OK) {
3929 /*
3930 * On socket error flag the session as aborted to force
3931 * the cleanup of its stream otherwise it can leak
3932 * during the lifetime of the relayd.
3933 *
3934 * This prevents situations in which streams can be
3935 * left opened because an index was received, the
3936 * control connection is closed, and the data
3937 * connection is closed (uncleanly) before the packet's
3938 * data provided.
3939 *
3940 * Since the control connection encountered an error,
3941 * it is okay to be conservative and close the
3942 * session right now as we can't rely on the protocol
3943 * being respected anymore.
3944 */
3945 if (status == RELAY_CONNECTION_STATUS_ERROR) {
3946 session_abort(ctrl_conn->session);
3947 }
3948
3949 /* Clear the connection on error or close. */
3950 relay_thread_close_connection(&events,
3951 pollfd,
3952 ctrl_conn);
3953 }
3954 seen_control = 1;
3955 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3956 relay_thread_close_connection(&events,
3957 pollfd, ctrl_conn);
3958 if (last_seen_data_fd == pollfd) {
3959 last_seen_data_fd = last_notdel_data_fd;
3960 }
3961 } else {
3962 ERR("Unexpected poll events %u for control sock %d",
3963 revents, pollfd);
3964 connection_put(ctrl_conn);
3965 goto error;
3966 }
3967 put_ctrl_connection:
3968 connection_put(ctrl_conn);
3969 }
3970 }
3971
3972 /*
3973 * The last loop handled a control request, go back to poll to make
3974 * sure we prioritise the control socket.
3975 */
3976 if (seen_control) {
3977 continue;
3978 }
3979
3980 if (last_seen_data_fd >= 0) {
3981 for (i = 0; i < nb_fd; i++) {
3982 int pollfd = LTTNG_POLL_GETFD(&events, i);
3983
3984 health_code_update();
3985
3986 if (last_seen_data_fd == pollfd) {
3987 idx = i;
3988 break;
3989 }
3990 }
3991 }
3992
3993 /* Process data connection. */
3994 for (i = idx + 1; i < nb_fd; i++) {
3995 /* Fetch the poll data. */
3996 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
3997 int pollfd = LTTNG_POLL_GETFD(&events, i);
3998 struct relay_connection *data_conn;
3999
4000 health_code_update();
4001
4002 if (!revents) {
4003 /* No activity for this FD (poll implementation). */
4004 continue;
4005 }
4006
4007 /* Skip the command pipe. It's handled in the first loop. */
4008 if (pollfd == relay_conn_pipe[0]) {
4009 continue;
4010 }
4011
4012 data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4013 if (!data_conn) {
4014 /* Skip it. Might be removed before. */
4015 continue;
4016 }
4017 if (data_conn->type == RELAY_CONTROL) {
4018 goto put_data_connection;
4019 }
4020 assert(data_conn->type == RELAY_DATA);
4021
4022 if (revents & LPOLLIN) {
4023 enum relay_connection_status status;
4024
4025 status = relay_process_data(data_conn);
4026 /* Connection closed or error. */
4027 if (status != RELAY_CONNECTION_STATUS_OK) {
4028 /*
4029 * On socket error flag the session as aborted to force
4030 * the cleanup of its stream otherwise it can leak
4031 * during the lifetime of the relayd.
4032 *
4033 * This prevents situations in which streams can be
4034 * left opened because an index was received, the
4035 * control connection is closed, and the data
4036 * connection is closed (uncleanly) before the packet's
4037 * data provided.
4038 *
4039 * Since the data connection encountered an error,
4040 * it is okay to be conservative and close the
4041 * session right now as we can't rely on the protocol
4042 * being respected anymore.
4043 */
4044 if (status == RELAY_CONNECTION_STATUS_ERROR) {
4045 session_abort(data_conn->session);
4046 }
4047 relay_thread_close_connection(&events, pollfd,
4048 data_conn);
4049 /*
4050 * Every goto restart call sets the last seen fd where
4051 * here we don't really care since we gracefully
4052 * continue the loop after the connection is deleted.
4053 */
4054 } else {
4055 /* Keep last seen port. */
4056 last_seen_data_fd = pollfd;
4057 connection_put(data_conn);
4058 goto restart;
4059 }
4060 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4061 relay_thread_close_connection(&events, pollfd,
4062 data_conn);
4063 } else {
4064 ERR("Unknown poll events %u for data sock %d",
4065 revents, pollfd);
4066 }
4067 put_data_connection:
4068 connection_put(data_conn);
4069 }
4070 last_seen_data_fd = -1;
4071 }
4072
4073 /* Normal exit, no error */
4074 ret = 0;
4075
4076 exit:
4077 error:
4078 /* Cleanup remaining connection object. */
4079 rcu_read_lock();
4080 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
4081 destroy_conn,
4082 sock_n.node) {
4083 health_code_update();
4084
4085 session_abort(destroy_conn->session);
4086
4087 /*
4088 * No need to grab another ref, because we own
4089 * destroy_conn.
4090 */
4091 relay_thread_close_connection(&events, destroy_conn->sock->fd,
4092 destroy_conn);
4093 }
4094 rcu_read_unlock();
4095
4096 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
4097 error_poll_create:
4098 lttng_ht_destroy(relay_connections_ht);
4099 relay_connections_ht_error:
4100 /* Close relay conn pipes */
4101 (void) fd_tracker_util_pipe_close(the_fd_tracker,
4102 relay_conn_pipe);
4103 if (err) {
4104 DBG("Thread exited with error");
4105 }
4106 DBG("Worker thread cleanup complete");
4107 error_testpoint:
4108 if (err) {
4109 health_error();
4110 ERR("Health error occurred in %s", __func__);
4111 }
4112 health_unregister(health_relayd);
4113 rcu_unregister_thread();
4114 lttng_relay_stop_threads();
4115 return NULL;
4116 }
4117
4118 /*
4119 * Create the relay command pipe to wake thread_manage_apps.
4120 * Closed in cleanup().
4121 */
4122 static int create_relay_conn_pipe(void)
4123 {
4124 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
4125 "Relayd connection pipe", relay_conn_pipe);
4126 }
4127
4128 static int stdio_open(void *data, int *fds)
4129 {
4130 fds[0] = fileno(stdout);
4131 fds[1] = fileno(stderr);
4132 return 0;
4133 }
4134
4135 static int track_stdio(void)
4136 {
4137 int fds[2];
4138 const char *names[] = { "stdout", "stderr" };
4139
4140 return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds,
4141 names, 2, stdio_open, NULL);
4142 }
4143
4144 /*
4145 * main
4146 */
4147 int main(int argc, char **argv)
4148 {
4149 bool thread_is_rcu_registered = false;
4150 int ret = 0, retval = 0;
4151 void *status;
4152 char *unlinked_file_directory_path = NULL, *output_path = NULL;
4153
4154 /* Parse environment variables */
4155 ret = parse_env_options();
4156 if (ret) {
4157 retval = -1;
4158 goto exit_options;
4159 }
4160
4161 /*
4162 * Parse arguments.
4163 * Command line arguments overwrite environment.
4164 */
4165 progname = argv[0];
4166 if (set_options(argc, argv)) {
4167 retval = -1;
4168 goto exit_options;
4169 }
4170
4171 if (set_signal_handler()) {
4172 retval = -1;
4173 goto exit_options;
4174 }
4175
4176 relayd_config_log();
4177
4178 if (opt_print_version) {
4179 print_version();
4180 retval = 0;
4181 goto exit_options;
4182 }
4183
4184 ret = fclose(stdin);
4185 if (ret) {
4186 PERROR("Failed to close stdin");
4187 goto exit_options;
4188 }
4189
4190 DBG("Clear command %s", opt_allow_clear ? "allowed" : "disallowed");
4191
4192 /* Try to create directory if -o, --output is specified. */
4193 if (opt_output_path) {
4194 if (*opt_output_path != '/') {
4195 ERR("Please specify an absolute path for -o, --output PATH");
4196 retval = -1;
4197 goto exit_options;
4198 }
4199
4200 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG,
4201 -1, -1);
4202 if (ret < 0) {
4203 ERR("Unable to create %s", opt_output_path);
4204 retval = -1;
4205 goto exit_options;
4206 }
4207 }
4208
4209 /* Daemonize */
4210 if (opt_daemon || opt_background) {
4211 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
4212 !opt_background);
4213 if (ret < 0) {
4214 retval = -1;
4215 goto exit_options;
4216 }
4217 }
4218
4219 if (opt_working_directory) {
4220 ret = utils_change_working_directory(opt_working_directory);
4221 if (ret) {
4222 /* All errors are already logged. */
4223 goto exit_options;
4224 }
4225 }
4226
4227 sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create();
4228 if (!sessiond_trace_chunk_registry) {
4229 ERR("Failed to initialize session daemon trace chunk registry");
4230 retval = -1;
4231 goto exit_options;
4232 }
4233
4234 /*
4235 * The RCU thread registration (and use, through the fd-tracker's
4236 * creation) is done after the daemonization to allow us to not
4237 * deal with liburcu's fork() management as the call RCU needs to
4238 * be restored.
4239 */
4240 rcu_register_thread();
4241 thread_is_rcu_registered = true;
4242
4243 output_path = create_output_path("");
4244 if (!output_path) {
4245 ERR("Failed to get output path");
4246 retval = -1;
4247 goto exit_options;
4248 }
4249 ret = asprintf(&unlinked_file_directory_path, "%s/%s", output_path,
4250 DEFAULT_UNLINKED_FILES_DIRECTORY);
4251 free(output_path);
4252 if (ret < 0) {
4253 ERR("Failed to format unlinked file directory path");
4254 retval = -1;
4255 goto exit_options;
4256 }
4257 the_fd_tracker = fd_tracker_create(
4258 unlinked_file_directory_path, lttng_opt_fd_pool_size);
4259 free(unlinked_file_directory_path);
4260 if (!the_fd_tracker) {
4261 retval = -1;
4262 goto exit_options;
4263 }
4264
4265 ret = track_stdio();
4266 if (ret) {
4267 retval = -1;
4268 goto exit_options;
4269 }
4270
4271 /* Initialize thread health monitoring */
4272 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
4273 if (!health_relayd) {
4274 PERROR("health_app_create error");
4275 retval = -1;
4276 goto exit_options;
4277 }
4278
4279 /* Create thread quit pipe */
4280 if (init_thread_quit_pipe()) {
4281 retval = -1;
4282 goto exit_options;
4283 }
4284
4285 /* Setup the thread apps communication pipe. */
4286 if (create_relay_conn_pipe()) {
4287 retval = -1;
4288 goto exit_options;
4289 }
4290
4291 /* Init relay command queue. */
4292 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
4293
4294 /* Initialize communication library */
4295 lttcomm_init();
4296 lttcomm_inet_init();
4297
4298 /* tables of sessions indexed by session ID */
4299 sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4300 if (!sessions_ht) {
4301 retval = -1;
4302 goto exit_options;
4303 }
4304
4305 /* tables of streams indexed by stream ID */
4306 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4307 if (!relay_streams_ht) {
4308 retval = -1;
4309 goto exit_options;
4310 }
4311
4312 /* tables of streams indexed by stream ID */
4313 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4314 if (!viewer_streams_ht) {
4315 retval = -1;
4316 goto exit_options;
4317 }
4318
4319 ret = init_health_quit_pipe();
4320 if (ret) {
4321 retval = -1;
4322 goto exit_options;
4323 }
4324
4325 /* Create thread to manage the client socket */
4326 ret = pthread_create(&health_thread, default_pthread_attr(),
4327 thread_manage_health, (void *) NULL);
4328 if (ret) {
4329 errno = ret;
4330 PERROR("pthread_create health");
4331 retval = -1;
4332 goto exit_options;
4333 }
4334
4335 /* Setup the dispatcher thread */
4336 ret = pthread_create(&dispatcher_thread, default_pthread_attr(),
4337 relay_thread_dispatcher, (void *) NULL);
4338 if (ret) {
4339 errno = ret;
4340 PERROR("pthread_create dispatcher");
4341 retval = -1;
4342 goto exit_dispatcher_thread;
4343 }
4344
4345 /* Setup the worker thread */
4346 ret = pthread_create(&worker_thread, default_pthread_attr(),
4347 relay_thread_worker, NULL);
4348 if (ret) {
4349 errno = ret;
4350 PERROR("pthread_create worker");
4351 retval = -1;
4352 goto exit_worker_thread;
4353 }
4354
4355 /* Setup the listener thread */
4356 ret = pthread_create(&listener_thread, default_pthread_attr(),
4357 relay_thread_listener, (void *) NULL);
4358 if (ret) {
4359 errno = ret;
4360 PERROR("pthread_create listener");
4361 retval = -1;
4362 goto exit_listener_thread;
4363 }
4364
4365 ret = relayd_live_create(live_uri);
4366 if (ret) {
4367 ERR("Starting live viewer threads");
4368 retval = -1;
4369 goto exit_live;
4370 }
4371
4372 /*
4373 * This is where we start awaiting program completion (e.g. through
4374 * signal that asks threads to teardown).
4375 */
4376
4377 ret = relayd_live_join();
4378 if (ret) {
4379 retval = -1;
4380 }
4381 exit_live:
4382
4383 ret = pthread_join(listener_thread, &status);
4384 if (ret) {
4385 errno = ret;
4386 PERROR("pthread_join listener_thread");
4387 retval = -1;
4388 }
4389
4390 exit_listener_thread:
4391 ret = pthread_join(worker_thread, &status);
4392 if (ret) {
4393 errno = ret;
4394 PERROR("pthread_join worker_thread");
4395 retval = -1;
4396 }
4397
4398 exit_worker_thread:
4399 ret = pthread_join(dispatcher_thread, &status);
4400 if (ret) {
4401 errno = ret;
4402 PERROR("pthread_join dispatcher_thread");
4403 retval = -1;
4404 }
4405 exit_dispatcher_thread:
4406
4407 ret = pthread_join(health_thread, &status);
4408 if (ret) {
4409 errno = ret;
4410 PERROR("pthread_join health_thread");
4411 retval = -1;
4412 }
4413 exit_options:
4414 /*
4415 * Wait for all pending call_rcu work to complete before tearing
4416 * down data structures. call_rcu worker may be trying to
4417 * perform lookups in those structures.
4418 */
4419 rcu_barrier();
4420 relayd_cleanup();
4421
4422 /* Ensure all prior call_rcu are done. */
4423 rcu_barrier();
4424
4425 if (thread_is_rcu_registered) {
4426 rcu_unregister_thread();
4427 }
4428
4429 if (!retval) {
4430 exit(EXIT_SUCCESS);
4431 } else {
4432 exit(EXIT_FAILURE);
4433 }
4434 }
LTTng tools - Deliverables
This page took 0.16644 seconds and 6 git commands to generate.