Fix: action executor: deadlock on session_list_lock
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
3 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2013 Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * SPDX-License-Identifier: GPL-2.0-only
7 *
8 */
9
10 #define _LGPL_SOURCE
11 #include <getopt.h>
12 #include <grp.h>
13 #include <limits.h>
14 #include <paths.h>
15 #include <pthread.h>
16 #include <signal.h>
17 #include <stdio.h>
18 #include <stdlib.h>
19 #include <string.h>
20 #include <inttypes.h>
21 #include <sys/mman.h>
22 #include <sys/mount.h>
23 #include <sys/resource.h>
24 #include <sys/socket.h>
25 #include <sys/stat.h>
26 #include <sys/types.h>
27 #include <sys/wait.h>
28 #include <urcu/uatomic.h>
29 #include <unistd.h>
30 #include <ctype.h>
31
32 #include <common/common.h>
33 #include <common/compat/socket.h>
34 #include <common/compat/getenv.h>
35 #include <common/defaults.h>
36 #include <common/kernel-consumer/kernel-consumer.h>
37 #include <common/futex.h>
38 #include <common/relayd/relayd.h>
39 #include <common/utils.h>
40 #include <common/daemonize.h>
41 #include <common/config/session-config.h>
42 #include <common/dynamic-buffer.h>
43 #include <lttng/event-internal.h>
44
45 #include "lttng-sessiond.h"
46 #include "buffer-registry.h"
47 #include "channel.h"
48 #include "cmd.h"
49 #include "consumer.h"
50 #include "context.h"
51 #include "event.h"
52 #include "event-notifier-error-accounting.h"
53 #include "kernel.h"
54 #include "kernel-consumer.h"
55 #include "lttng-ust-ctl.h"
56 #include "ust-consumer.h"
57 #include "utils.h"
58 #include "fd-limit.h"
59 #include "health-sessiond.h"
60 #include "testpoint.h"
61 #include "notify-apps.h"
62 #include "agent-thread.h"
63 #include "save.h"
64 #include "notification-thread.h"
65 #include "notification-thread-commands.h"
66 #include "rotation-thread.h"
67 #include "agent.h"
68 #include "ht-cleanup.h"
69 #include "sessiond-config.h"
70 #include "timer.h"
71 #include "thread.h"
72 #include "client.h"
73 #include "dispatch.h"
74 #include "register.h"
75 #include "manage-apps.h"
76 #include "manage-kernel.h"
77 #include "modprobe.h"
78
79 static const char *help_msg =
80 #ifdef LTTNG_EMBED_HELP
81 #include <lttng-sessiond.8.h>
82 #else
83 NULL
84 #endif
85 ;
86
87 #define EVENT_NOTIFIER_ERROR_COUNTER_NUMBER_OF_BUCKET_MAX 65535
88 #define EVENT_NOTIFIER_ERROR_BUFFER_SIZE_BASE_OPTION_STR \
89 "event-notifier-error-buffer-size"
90 #define EVENT_NOTIFIER_ERROR_BUFFER_SIZE_KERNEL_OPTION_STR \
91 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_BASE_OPTION_STR "-kernel"
92 #define EVENT_NOTIFIER_ERROR_BUFFER_SIZE_USERSPACE_OPTION_STR \
93 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_BASE_OPTION_STR "-userspace"
94
95
96 const char *progname;
97 static int lockfile_fd = -1;
98 static int opt_print_version;
99
100 /* Set to 1 when a SIGUSR1 signal is received. */
101 static int recv_child_signal;
102
103 /* Command line options */
104 static const struct option long_options[] = {
105 { "client-sock", required_argument, 0, 'c' },
106 { "apps-sock", required_argument, 0, 'a' },
107 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
108 { "kconsumerd-err-sock", required_argument, 0, '\0' },
109 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
110 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
111 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
112 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
113 { "consumerd32-path", required_argument, 0, '\0' },
114 { "consumerd32-libdir", required_argument, 0, '\0' },
115 { "consumerd64-path", required_argument, 0, '\0' },
116 { "consumerd64-libdir", required_argument, 0, '\0' },
117 { "daemonize", no_argument, 0, 'd' },
118 { "background", no_argument, 0, 'b' },
119 { "sig-parent", no_argument, 0, 'S' },
120 { "help", no_argument, 0, 'h' },
121 { "group", required_argument, 0, 'g' },
122 { "version", no_argument, 0, 'V' },
123 { "quiet", no_argument, 0, 'q' },
124 { "verbose", no_argument, 0, 'v' },
125 { "verbose-consumer", no_argument, 0, '\0' },
126 { "no-kernel", no_argument, 0, '\0' },
127 { "pidfile", required_argument, 0, 'p' },
128 { "agent-tcp-port", required_argument, 0, '\0' },
129 { "config", required_argument, 0, 'f' },
130 { "load", required_argument, 0, 'l' },
131 { "kmod-probes", required_argument, 0, '\0' },
132 { "extra-kmod-probes", required_argument, 0, '\0' },
133 { EVENT_NOTIFIER_ERROR_BUFFER_SIZE_KERNEL_OPTION_STR, required_argument, 0, '\0' },
134 { EVENT_NOTIFIER_ERROR_BUFFER_SIZE_USERSPACE_OPTION_STR, required_argument, 0, '\0' },
135 { NULL, 0, 0, 0 }
136 };
137
138 /* Command line options to ignore from configuration file */
139 static const char *config_ignore_options[] = { "help", "version", "config" };
140
141 /*
142 * This pipe is used to inform the thread managing application communication
143 * that a command is queued and ready to be processed.
144 */
145 static int apps_cmd_pipe[2] = { -1, -1 };
146 static int apps_cmd_notify_pipe[2] = { -1, -1 };
147
148 /*
149 * UST registration command queue. This queue is tied with a futex and uses a N
150 * wakers / 1 waiter implemented and detailed in futex.c/.h
151 *
152 * The thread_registration_apps and thread_dispatch_ust_registration uses this
153 * queue along with the wait/wake scheme. The thread_manage_apps receives down
154 * the line new application socket and monitors it for any I/O error or clean
155 * close that triggers an unregistration of the application.
156 */
157 static struct ust_cmd_queue ust_cmd_queue;
158
159 /*
160 * Section name to look for in the daemon configuration file.
161 */
162 static const char * const config_section_name = "sessiond";
163
164 /* Am I root or not. Set to 1 if the daemon is running as root */
165 static int is_root;
166
167 /*
168 * Stop all threads by closing the thread quit pipe.
169 */
170 static void stop_threads(void)
171 {
172 int ret;
173
174 /* Stopping all threads */
175 DBG("Terminating all threads");
176 ret = sessiond_notify_quit_pipe();
177 if (ret < 0) {
178 ERR("write error on thread quit pipe");
179 }
180 }
181
182 /*
183 * Close every consumer sockets.
184 */
185 static void close_consumer_sockets(void)
186 {
187 int ret;
188
189 if (the_kconsumer_data.err_sock >= 0) {
190 ret = close(the_kconsumer_data.err_sock);
191 if (ret < 0) {
192 PERROR("kernel consumer err_sock close");
193 }
194 }
195 if (the_ustconsumer32_data.err_sock >= 0) {
196 ret = close(the_ustconsumer32_data.err_sock);
197 if (ret < 0) {
198 PERROR("UST consumerd32 err_sock close");
199 }
200 }
201 if (the_ustconsumer64_data.err_sock >= 0) {
202 ret = close(the_ustconsumer64_data.err_sock);
203 if (ret < 0) {
204 PERROR("UST consumerd64 err_sock close");
205 }
206 }
207 if (the_kconsumer_data.cmd_sock >= 0) {
208 ret = close(the_kconsumer_data.cmd_sock);
209 if (ret < 0) {
210 PERROR("kernel consumer cmd_sock close");
211 }
212 }
213 if (the_ustconsumer32_data.cmd_sock >= 0) {
214 ret = close(the_ustconsumer32_data.cmd_sock);
215 if (ret < 0) {
216 PERROR("UST consumerd32 cmd_sock close");
217 }
218 }
219 if (the_ustconsumer64_data.cmd_sock >= 0) {
220 ret = close(the_ustconsumer64_data.cmd_sock);
221 if (ret < 0) {
222 PERROR("UST consumerd64 cmd_sock close");
223 }
224 }
225 if (the_kconsumer_data.channel_monitor_pipe >= 0) {
226 ret = close(the_kconsumer_data.channel_monitor_pipe);
227 if (ret < 0) {
228 PERROR("kernel consumer channel monitor pipe close");
229 }
230 }
231 if (the_ustconsumer32_data.channel_monitor_pipe >= 0) {
232 ret = close(the_ustconsumer32_data.channel_monitor_pipe);
233 if (ret < 0) {
234 PERROR("UST consumerd32 channel monitor pipe close");
235 }
236 }
237 if (the_ustconsumer64_data.channel_monitor_pipe >= 0) {
238 ret = close(the_ustconsumer64_data.channel_monitor_pipe);
239 if (ret < 0) {
240 PERROR("UST consumerd64 channel monitor pipe close");
241 }
242 }
243 }
244
245 /*
246 * Wait on consumer process termination.
247 *
248 * Need to be called with the consumer data lock held or from a context
249 * ensuring no concurrent access to data (e.g: cleanup).
250 */
251 static void wait_consumer(struct consumer_data *consumer_data)
252 {
253 pid_t ret;
254 int status;
255
256 if (consumer_data->pid <= 0) {
257 return;
258 }
259
260 DBG("Waiting for complete teardown of consumerd (PID: %d)",
261 consumer_data->pid);
262 ret = waitpid(consumer_data->pid, &status, 0);
263 if (ret == -1) {
264 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
265 } else if (!WIFEXITED(status)) {
266 ERR("consumerd termination with error: %d",
267 WEXITSTATUS(ret));
268 }
269 consumer_data->pid = 0;
270 }
271
272 /*
273 * Cleanup the session daemon's data structures.
274 */
275 static void sessiond_cleanup(void)
276 {
277 int ret;
278 struct ltt_session_list *session_list = session_get_list();
279
280 DBG("Cleanup sessiond");
281
282 /*
283 * Close the thread quit pipe. It has already done its job,
284 * since we are now called.
285 */
286 sessiond_close_quit_pipe();
287 utils_close_pipe(apps_cmd_pipe);
288 utils_close_pipe(apps_cmd_notify_pipe);
289 utils_close_pipe(the_kernel_poll_pipe);
290
291 ret = remove(the_config.pid_file_path.value);
292 if (ret < 0) {
293 PERROR("remove pidfile %s", the_config.pid_file_path.value);
294 }
295
296 DBG("Removing sessiond and consumerd content of directory %s",
297 the_config.rundir.value);
298
299 /* sessiond */
300 DBG("Removing %s", the_config.pid_file_path.value);
301 (void) unlink(the_config.pid_file_path.value);
302
303 DBG("Removing %s", the_config.agent_port_file_path.value);
304 (void) unlink(the_config.agent_port_file_path.value);
305
306 /* kconsumerd */
307 DBG("Removing %s", the_kconsumer_data.err_unix_sock_path);
308 (void) unlink(the_kconsumer_data.err_unix_sock_path);
309
310 DBG("Removing directory %s", the_config.kconsumerd_path.value);
311 (void) rmdir(the_config.kconsumerd_path.value);
312
313 /* ust consumerd 32 */
314 DBG("Removing %s", the_config.consumerd32_err_unix_sock_path.value);
315 (void) unlink(the_config.consumerd32_err_unix_sock_path.value);
316
317 DBG("Removing directory %s", the_config.consumerd32_path.value);
318 (void) rmdir(the_config.consumerd32_path.value);
319
320 /* ust consumerd 64 */
321 DBG("Removing %s", the_config.consumerd64_err_unix_sock_path.value);
322 (void) unlink(the_config.consumerd64_err_unix_sock_path.value);
323
324 DBG("Removing directory %s", the_config.consumerd64_path.value);
325 (void) rmdir(the_config.consumerd64_path.value);
326
327 pthread_mutex_destroy(&session_list->lock);
328
329 DBG("Cleaning up all per-event notifier domain agents");
330 agent_by_event_notifier_domain_ht_destroy();
331
332 DBG("Cleaning up all agent apps");
333 agent_app_ht_clean();
334 DBG("Closing all UST sockets");
335 ust_app_clean_list();
336 buffer_reg_destroy_registries();
337
338 close_consumer_sockets();
339
340 wait_consumer(&the_kconsumer_data);
341 wait_consumer(&the_ustconsumer64_data);
342 wait_consumer(&the_ustconsumer32_data);
343
344 if (is_root && !the_config.no_kernel) {
345 cleanup_kernel_tracer();
346 }
347
348 /*
349 * We do NOT rmdir rundir because there are other processes
350 * using it, for instance lttng-relayd, which can start in
351 * parallel with this teardown.
352 */
353 }
354
355 /*
356 * Cleanup the daemon's option data structures.
357 */
358 static void sessiond_cleanup_options(void)
359 {
360 DBG("Cleaning up options");
361
362 sessiond_config_fini(&the_config);
363
364 run_as_destroy_worker();
365 }
366
367 static int string_match(const char *str1, const char *str2)
368 {
369 return (str1 && str2) && !strcmp(str1, str2);
370 }
371
372 /*
373 * Take an option from the getopt output and set it in the right variable to be
374 * used later.
375 *
376 * Return 0 on success else a negative value.
377 */
378 static int set_option(int opt, const char *arg, const char *optname)
379 {
380 int ret = 0;
381
382 if (string_match(optname, "client-sock") || opt == 'c') {
383 if (!arg || *arg == '\0') {
384 ret = -EINVAL;
385 goto end;
386 }
387 if (lttng_is_setuid_setgid()) {
388 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
389 "-c, --client-sock");
390 } else {
391 config_string_set(&the_config.client_unix_sock_path,
392 strdup(arg));
393 if (!the_config.client_unix_sock_path.value) {
394 ret = -ENOMEM;
395 PERROR("strdup");
396 }
397 }
398 } else if (string_match(optname, "apps-sock") || opt == 'a') {
399 if (!arg || *arg == '\0') {
400 ret = -EINVAL;
401 goto end;
402 }
403 if (lttng_is_setuid_setgid()) {
404 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
405 "-a, --apps-sock");
406 } else {
407 config_string_set(&the_config.apps_unix_sock_path,
408 strdup(arg));
409 if (!the_config.apps_unix_sock_path.value) {
410 ret = -ENOMEM;
411 PERROR("strdup");
412 }
413 }
414 } else if (string_match(optname, "daemonize") || opt == 'd') {
415 the_config.daemonize = true;
416 } else if (string_match(optname, "background") || opt == 'b') {
417 the_config.background = true;
418 } else if (string_match(optname, "group") || opt == 'g') {
419 if (!arg || *arg == '\0') {
420 ret = -EINVAL;
421 goto end;
422 }
423 if (lttng_is_setuid_setgid()) {
424 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
425 "-g, --group");
426 } else {
427 config_string_set(&the_config.tracing_group_name,
428 strdup(arg));
429 if (!the_config.tracing_group_name.value) {
430 ret = -ENOMEM;
431 PERROR("strdup");
432 }
433 }
434 } else if (string_match(optname, "help") || opt == 'h') {
435 ret = utils_show_help(8, "lttng-sessiond", help_msg);
436 if (ret) {
437 ERR("Cannot show --help for `lttng-sessiond`");
438 perror("exec");
439 }
440 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
441 } else if (string_match(optname, "version") || opt == 'V') {
442 opt_print_version = 1;
443 } else if (string_match(optname, "sig-parent") || opt == 'S') {
444 the_config.sig_parent = true;
445 } else if (string_match(optname, "kconsumerd-err-sock")) {
446 if (!arg || *arg == '\0') {
447 ret = -EINVAL;
448 goto end;
449 }
450 if (lttng_is_setuid_setgid()) {
451 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
452 "--kconsumerd-err-sock");
453 } else {
454 config_string_set(
455 &the_config.kconsumerd_err_unix_sock_path,
456 strdup(arg));
457 if (!the_config.kconsumerd_err_unix_sock_path.value) {
458 ret = -ENOMEM;
459 PERROR("strdup");
460 }
461 }
462 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
463 if (!arg || *arg == '\0') {
464 ret = -EINVAL;
465 goto end;
466 }
467 if (lttng_is_setuid_setgid()) {
468 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
469 "--kconsumerd-cmd-sock");
470 } else {
471 config_string_set(
472 &the_config.kconsumerd_cmd_unix_sock_path,
473 strdup(arg));
474 if (!the_config.kconsumerd_cmd_unix_sock_path.value) {
475 ret = -ENOMEM;
476 PERROR("strdup");
477 }
478 }
479 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
480 if (!arg || *arg == '\0') {
481 ret = -EINVAL;
482 goto end;
483 }
484 if (lttng_is_setuid_setgid()) {
485 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
486 "--ustconsumerd64-err-sock");
487 } else {
488 config_string_set(
489 &the_config.consumerd64_err_unix_sock_path,
490 strdup(arg));
491 if (!the_config.consumerd64_err_unix_sock_path.value) {
492 ret = -ENOMEM;
493 PERROR("strdup");
494 }
495 }
496 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
497 if (!arg || *arg == '\0') {
498 ret = -EINVAL;
499 goto end;
500 }
501 if (lttng_is_setuid_setgid()) {
502 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
503 "--ustconsumerd64-cmd-sock");
504 } else {
505 config_string_set(
506 &the_config.consumerd64_cmd_unix_sock_path,
507 strdup(arg));
508 if (!the_config.consumerd64_cmd_unix_sock_path.value) {
509 ret = -ENOMEM;
510 PERROR("strdup");
511 }
512 }
513 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
514 if (!arg || *arg == '\0') {
515 ret = -EINVAL;
516 goto end;
517 }
518 if (lttng_is_setuid_setgid()) {
519 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
520 "--ustconsumerd32-err-sock");
521 } else {
522 config_string_set(
523 &the_config.consumerd32_err_unix_sock_path,
524 strdup(arg));
525 if (!the_config.consumerd32_err_unix_sock_path.value) {
526 ret = -ENOMEM;
527 PERROR("strdup");
528 }
529 }
530 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
531 if (!arg || *arg == '\0') {
532 ret = -EINVAL;
533 goto end;
534 }
535 if (lttng_is_setuid_setgid()) {
536 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
537 "--ustconsumerd32-cmd-sock");
538 } else {
539 config_string_set(
540 &the_config.consumerd32_cmd_unix_sock_path,
541 strdup(arg));
542 if (!the_config.consumerd32_cmd_unix_sock_path.value) {
543 ret = -ENOMEM;
544 PERROR("strdup");
545 }
546 }
547 } else if (string_match(optname, "no-kernel")) {
548 the_config.no_kernel = true;
549 } else if (string_match(optname, "quiet") || opt == 'q') {
550 the_config.quiet = true;
551 } else if (string_match(optname, "verbose") || opt == 'v') {
552 /* Verbose level can increase using multiple -v */
553 if (arg) {
554 /* Value obtained from config file */
555 the_config.verbose = config_parse_value(arg);
556 } else {
557 /* -v used on command line */
558 the_config.verbose++;
559 }
560 /* Clamp value to [0, 3] */
561 the_config.verbose = the_config.verbose < 0 ?
562 0 :
563 (the_config.verbose <= 3 ? the_config.verbose :
564 3);
565 } else if (string_match(optname, "verbose-consumer")) {
566 if (arg) {
567 the_config.verbose_consumer = config_parse_value(arg);
568 } else {
569 the_config.verbose_consumer++;
570 }
571 } else if (string_match(optname, "consumerd32-path")) {
572 if (!arg || *arg == '\0') {
573 ret = -EINVAL;
574 goto end;
575 }
576 if (lttng_is_setuid_setgid()) {
577 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
578 "--consumerd32-path");
579 } else {
580 config_string_set(&the_config.consumerd32_bin_path,
581 strdup(arg));
582 if (!the_config.consumerd32_bin_path.value) {
583 PERROR("strdup");
584 ret = -ENOMEM;
585 }
586 }
587 } else if (string_match(optname, "consumerd32-libdir")) {
588 if (!arg || *arg == '\0') {
589 ret = -EINVAL;
590 goto end;
591 }
592 if (lttng_is_setuid_setgid()) {
593 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
594 "--consumerd32-libdir");
595 } else {
596 config_string_set(&the_config.consumerd32_lib_dir,
597 strdup(arg));
598 if (!the_config.consumerd32_lib_dir.value) {
599 PERROR("strdup");
600 ret = -ENOMEM;
601 }
602 }
603 } else if (string_match(optname, "consumerd64-path")) {
604 if (!arg || *arg == '\0') {
605 ret = -EINVAL;
606 goto end;
607 }
608 if (lttng_is_setuid_setgid()) {
609 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
610 "--consumerd64-path");
611 } else {
612 config_string_set(&the_config.consumerd64_bin_path,
613 strdup(arg));
614 if (!the_config.consumerd64_bin_path.value) {
615 PERROR("strdup");
616 ret = -ENOMEM;
617 }
618 }
619 } else if (string_match(optname, "consumerd64-libdir")) {
620 if (!arg || *arg == '\0') {
621 ret = -EINVAL;
622 goto end;
623 }
624 if (lttng_is_setuid_setgid()) {
625 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
626 "--consumerd64-libdir");
627 } else {
628 config_string_set(&the_config.consumerd64_lib_dir,
629 strdup(arg));
630 if (!the_config.consumerd64_lib_dir.value) {
631 PERROR("strdup");
632 ret = -ENOMEM;
633 }
634 }
635 } else if (string_match(optname, "pidfile") || opt == 'p') {
636 if (!arg || *arg == '\0') {
637 ret = -EINVAL;
638 goto end;
639 }
640 if (lttng_is_setuid_setgid()) {
641 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
642 "-p, --pidfile");
643 } else {
644 config_string_set(
645 &the_config.pid_file_path, strdup(arg));
646 if (!the_config.pid_file_path.value) {
647 PERROR("strdup");
648 ret = -ENOMEM;
649 }
650 }
651 } else if (string_match(optname, "agent-tcp-port")) {
652 if (!arg || *arg == '\0') {
653 ret = -EINVAL;
654 goto end;
655 }
656 if (lttng_is_setuid_setgid()) {
657 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
658 "--agent-tcp-port");
659 } else {
660 unsigned long v;
661
662 errno = 0;
663 v = strtoul(arg, NULL, 0);
664 if (errno != 0 || !isdigit(arg[0])) {
665 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
666 return -1;
667 }
668 if (v == 0 || v >= 65535) {
669 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
670 return -1;
671 }
672 the_config.agent_tcp_port.begin =
673 the_config.agent_tcp_port.end = (int) v;
674 DBG3("Agent TCP port set to non default: %i", (int) v);
675 }
676 } else if (string_match(optname, "load") || opt == 'l') {
677 if (!arg || *arg == '\0') {
678 ret = -EINVAL;
679 goto end;
680 }
681 if (lttng_is_setuid_setgid()) {
682 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
683 "-l, --load");
684 } else {
685 config_string_set(&the_config.load_session_path,
686 strdup(arg));
687 if (!the_config.load_session_path.value) {
688 PERROR("strdup");
689 ret = -ENOMEM;
690 }
691 }
692 } else if (string_match(optname, "kmod-probes")) {
693 if (!arg || *arg == '\0') {
694 ret = -EINVAL;
695 goto end;
696 }
697 if (lttng_is_setuid_setgid()) {
698 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
699 "--kmod-probes");
700 } else {
701 config_string_set(&the_config.kmod_probes_list,
702 strdup(arg));
703 if (!the_config.kmod_probes_list.value) {
704 PERROR("strdup");
705 ret = -ENOMEM;
706 }
707 }
708 } else if (string_match(optname, "extra-kmod-probes")) {
709 if (!arg || *arg == '\0') {
710 ret = -EINVAL;
711 goto end;
712 }
713 if (lttng_is_setuid_setgid()) {
714 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
715 "--extra-kmod-probes");
716 } else {
717 config_string_set(&the_config.kmod_extra_probes_list,
718 strdup(arg));
719 if (!the_config.kmod_extra_probes_list.value) {
720 PERROR("strdup");
721 ret = -ENOMEM;
722 }
723 }
724 } else if (string_match(optname, EVENT_NOTIFIER_ERROR_BUFFER_SIZE_KERNEL_OPTION_STR)) {
725 unsigned long v;
726
727 errno = 0;
728 v = strtoul(arg, NULL, 0);
729 if (errno != 0 || !isdigit(arg[0])) {
730 ERR("Wrong value in --%s parameter: %s",
731 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_KERNEL_OPTION_STR, arg);
732 return -1;
733 }
734 if (v == 0 || v >= EVENT_NOTIFIER_ERROR_COUNTER_NUMBER_OF_BUCKET_MAX) {
735 ERR("Value out of range for --%s parameter: %s",
736 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_KERNEL_OPTION_STR, arg);
737 return -1;
738 }
739 the_config.event_notifier_buffer_size_kernel = (int) v;
740 DBG3("Number of event notifier error buffer kernel size to non default: %i",
741 the_config.event_notifier_buffer_size_kernel);
742 goto end;
743 } else if (string_match(optname, EVENT_NOTIFIER_ERROR_BUFFER_SIZE_USERSPACE_OPTION_STR)) {
744 unsigned long v;
745
746 errno = 0;
747 v = strtoul(arg, NULL, 0);
748 if (errno != 0 || !isdigit(arg[0])) {
749 ERR("Wrong value in --%s parameter: %s",
750 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_USERSPACE_OPTION_STR, arg);
751 return -1;
752 }
753 if (v == 0 || v >= EVENT_NOTIFIER_ERROR_COUNTER_NUMBER_OF_BUCKET_MAX) {
754 ERR("Value out of range for --%s parameter: %s",
755 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_USERSPACE_OPTION_STR, arg);
756 return -1;
757 }
758 the_config.event_notifier_buffer_size_userspace = (int) v;
759 DBG3("Number of event notifier error buffer userspace size to non default: %i",
760 the_config.event_notifier_buffer_size_userspace);
761 goto end;
762 } else if (string_match(optname, "config") || opt == 'f') {
763 /* This is handled in set_options() thus silent skip. */
764 goto end;
765 } else {
766 /* Unknown option or other error.
767 * Error is printed by getopt, just return */
768 ret = -1;
769 }
770
771 end:
772 if (ret == -EINVAL) {
773 const char *opt_name = "unknown";
774 int i;
775
776 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
777 i++) {
778 if (opt == long_options[i].val) {
779 opt_name = long_options[i].name;
780 break;
781 }
782 }
783
784 WARN("Invalid argument provided for option \"%s\", using default value.",
785 opt_name);
786 }
787
788 return ret;
789 }
790
791 /*
792 * config_entry_handler_cb used to handle options read from a config file.
793 * See config_entry_handler_cb comment in common/config/session-config.h for the
794 * return value conventions.
795 */
796 static int config_entry_handler(const struct config_entry *entry, void *unused)
797 {
798 int ret = 0, i;
799
800 if (!entry || !entry->name || !entry->value) {
801 ret = -EINVAL;
802 goto end;
803 }
804
805 /* Check if the option is to be ignored */
806 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
807 if (!strcmp(entry->name, config_ignore_options[i])) {
808 goto end;
809 }
810 }
811
812 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
813 i++) {
814
815 /* Ignore if not fully matched. */
816 if (strcmp(entry->name, long_options[i].name)) {
817 continue;
818 }
819
820 /*
821 * If the option takes no argument on the command line, we have to
822 * check if the value is "true". We support non-zero numeric values,
823 * true, on and yes.
824 */
825 if (!long_options[i].has_arg) {
826 ret = config_parse_value(entry->value);
827 if (ret <= 0) {
828 if (ret) {
829 WARN("Invalid configuration value \"%s\" for option %s",
830 entry->value, entry->name);
831 }
832 /* False, skip boolean config option. */
833 goto end;
834 }
835 }
836
837 ret = set_option(long_options[i].val, entry->value, entry->name);
838 goto end;
839 }
840
841 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
842
843 end:
844 return ret;
845 }
846
847 static void print_version(void) {
848 fprintf(stdout, "%s\n", VERSION);
849 }
850
851 /*
852 * daemon configuration loading and argument parsing
853 */
854 static int set_options(int argc, char **argv)
855 {
856 int ret = 0, c = 0, option_index = 0;
857 int orig_optopt = optopt, orig_optind = optind;
858 char *optstring;
859 char *config_path = NULL;
860
861 optstring = utils_generate_optstring(long_options,
862 sizeof(long_options) / sizeof(struct option));
863 if (!optstring) {
864 ret = -ENOMEM;
865 goto end;
866 }
867
868 /* Check for the --config option */
869 while ((c = getopt_long(argc, argv, optstring, long_options,
870 &option_index)) != -1) {
871 if (c == '?') {
872 ret = -EINVAL;
873 goto end;
874 } else if (c != 'f') {
875 /* if not equal to --config option. */
876 continue;
877 }
878
879 if (lttng_is_setuid_setgid()) {
880 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
881 "-f, --config");
882 } else {
883 free(config_path);
884 config_path = utils_expand_path(optarg);
885 if (!config_path) {
886 ERR("Failed to resolve path: %s", optarg);
887 }
888 }
889 }
890
891 ret = config_get_section_entries(config_path, config_section_name,
892 config_entry_handler, NULL);
893 if (ret) {
894 if (ret > 0) {
895 ERR("Invalid configuration option at line %i", ret);
896 ret = -1;
897 }
898 goto end;
899 }
900
901 /* Reset getopt's global state */
902 optopt = orig_optopt;
903 optind = orig_optind;
904 while (1) {
905 option_index = -1;
906 /*
907 * getopt_long() will not set option_index if it encounters a
908 * short option.
909 */
910 c = getopt_long(argc, argv, optstring, long_options,
911 &option_index);
912 if (c == -1) {
913 break;
914 }
915
916 /*
917 * Pass NULL as the long option name if popt left the index
918 * unset.
919 */
920 ret = set_option(c, optarg,
921 option_index < 0 ? NULL :
922 long_options[option_index].name);
923 if (ret < 0) {
924 break;
925 }
926 }
927
928 end:
929 free(config_path);
930 free(optstring);
931 return ret;
932 }
933
934 /*
935 * Create lockfile using the rundir and return its fd.
936 */
937 static int create_lockfile(void)
938 {
939 return utils_create_lock_file(the_config.lock_file_path.value);
940 }
941
942 /*
943 * Check if the global socket is available, and if a daemon is answering at the
944 * other side. If yes, error is returned.
945 *
946 * Also attempts to create and hold the lock file.
947 */
948 static int check_existing_daemon(void)
949 {
950 int ret = 0;
951
952 /* Is there anybody out there ? */
953 if (lttng_session_daemon_alive()) {
954 ret = -EEXIST;
955 goto end;
956 }
957
958 lockfile_fd = create_lockfile();
959 if (lockfile_fd < 0) {
960 ret = -EEXIST;
961 goto end;
962 }
963 end:
964 return ret;
965 }
966
967 static void sessiond_cleanup_lock_file(void)
968 {
969 int ret;
970
971 /*
972 * Cleanup lock file by deleting it and finaly closing it which will
973 * release the file system lock.
974 */
975 if (lockfile_fd >= 0) {
976 ret = remove(the_config.lock_file_path.value);
977 if (ret < 0) {
978 PERROR("remove lock file");
979 }
980 ret = close(lockfile_fd);
981 if (ret < 0) {
982 PERROR("close lock file");
983 }
984 }
985 }
986
987 /*
988 * Set the tracing group gid onto the client socket.
989 *
990 * Race window between mkdir and chown is OK because we are going from more
991 * permissive (root.root) to less permissive (root.tracing).
992 */
993 static int set_permissions(char *rundir)
994 {
995 int ret;
996 gid_t gid;
997
998 ret = utils_get_group_id(
999 the_config.tracing_group_name.value, true, &gid);
1000 if (ret) {
1001 /* Default to root group. */
1002 gid = 0;
1003 }
1004
1005 /* Set lttng run dir */
1006 ret = chown(rundir, 0, gid);
1007 if (ret < 0) {
1008 ERR("Unable to set group on %s", rundir);
1009 PERROR("chown");
1010 }
1011
1012 /*
1013 * Ensure all applications and tracing group can search the run
1014 * dir. Allow everyone to read the directory, since it does not
1015 * buy us anything to hide its content.
1016 */
1017 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
1018 if (ret < 0) {
1019 ERR("Unable to set permissions on %s", rundir);
1020 PERROR("chmod");
1021 }
1022
1023 /* lttng client socket path */
1024 ret = chown(the_config.client_unix_sock_path.value, 0, gid);
1025 if (ret < 0) {
1026 ERR("Unable to set group on %s",
1027 the_config.client_unix_sock_path.value);
1028 PERROR("chown");
1029 }
1030
1031 /* kconsumer error socket path */
1032 ret = chown(the_kconsumer_data.err_unix_sock_path, 0, 0);
1033 if (ret < 0) {
1034 ERR("Unable to set group on %s",
1035 the_kconsumer_data.err_unix_sock_path);
1036 PERROR("chown");
1037 }
1038
1039 /* 64-bit ustconsumer error socket path */
1040 ret = chown(the_ustconsumer64_data.err_unix_sock_path, 0, 0);
1041 if (ret < 0) {
1042 ERR("Unable to set group on %s",
1043 the_ustconsumer64_data.err_unix_sock_path);
1044 PERROR("chown");
1045 }
1046
1047 /* 32-bit ustconsumer compat32 error socket path */
1048 ret = chown(the_ustconsumer32_data.err_unix_sock_path, 0, 0);
1049 if (ret < 0) {
1050 ERR("Unable to set group on %s",
1051 the_ustconsumer32_data.err_unix_sock_path);
1052 PERROR("chown");
1053 }
1054
1055 DBG("All permissions are set");
1056
1057 return ret;
1058 }
1059
1060 /*
1061 * Create the lttng run directory needed for all global sockets and pipe.
1062 */
1063 static int create_lttng_rundir(void)
1064 {
1065 int ret;
1066
1067 DBG3("Creating LTTng run directory: %s", the_config.rundir.value);
1068
1069 ret = mkdir(the_config.rundir.value, S_IRWXU);
1070 if (ret < 0) {
1071 if (errno != EEXIST) {
1072 ERR("Unable to create %s", the_config.rundir.value);
1073 goto error;
1074 } else {
1075 ret = 0;
1076 }
1077 }
1078
1079 error:
1080 return ret;
1081 }
1082
1083 /*
1084 * Setup sockets and directory needed by the consumerds' communication with the
1085 * session daemon.
1086 */
1087 static int set_consumer_sockets(struct consumer_data *consumer_data)
1088 {
1089 int ret;
1090 char *path = NULL;
1091
1092 switch (consumer_data->type) {
1093 case LTTNG_CONSUMER_KERNEL:
1094 path = the_config.kconsumerd_path.value;
1095 break;
1096 case LTTNG_CONSUMER64_UST:
1097 path = the_config.consumerd64_path.value;
1098 break;
1099 case LTTNG_CONSUMER32_UST:
1100 path = the_config.consumerd32_path.value;
1101 break;
1102 default:
1103 ERR("Consumer type unknown");
1104 ret = -EINVAL;
1105 goto error;
1106 }
1107 assert(path);
1108
1109 DBG2("Creating consumer directory: %s", path);
1110
1111 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
1112 if (ret < 0 && errno != EEXIST) {
1113 PERROR("mkdir");
1114 ERR("Failed to create %s", path);
1115 goto error;
1116 }
1117 if (is_root) {
1118 gid_t gid;
1119
1120 ret = utils_get_group_id(the_config.tracing_group_name.value,
1121 true, &gid);
1122 if (ret) {
1123 /* Default to root group. */
1124 gid = 0;
1125 }
1126
1127 ret = chown(path, 0, gid);
1128 if (ret < 0) {
1129 ERR("Unable to set group on %s", path);
1130 PERROR("chown");
1131 goto error;
1132 }
1133 }
1134
1135 /* Create the consumerd error unix socket */
1136 consumer_data->err_sock =
1137 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
1138 if (consumer_data->err_sock < 0) {
1139 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
1140 ret = -1;
1141 goto error;
1142 }
1143
1144 /*
1145 * Set the CLOEXEC flag. Return code is useless because either way, the
1146 * show must go on.
1147 */
1148 ret = utils_set_fd_cloexec(consumer_data->err_sock);
1149 if (ret < 0) {
1150 PERROR("utils_set_fd_cloexec");
1151 /* continue anyway */
1152 }
1153
1154 /* File permission MUST be 660 */
1155 ret = chmod(consumer_data->err_unix_sock_path,
1156 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
1157 if (ret < 0) {
1158 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
1159 PERROR("chmod");
1160 goto error;
1161 }
1162
1163 error:
1164 return ret;
1165 }
1166
1167 /*
1168 * Signal handler for the daemon
1169 *
1170 * Simply stop all worker threads, leaving main() return gracefully after
1171 * joining all threads and calling cleanup().
1172 */
1173 static void sighandler(int sig)
1174 {
1175 switch (sig) {
1176 case SIGINT:
1177 DBG("SIGINT caught");
1178 stop_threads();
1179 break;
1180 case SIGTERM:
1181 DBG("SIGTERM caught");
1182 stop_threads();
1183 break;
1184 case SIGUSR1:
1185 CMM_STORE_SHARED(recv_child_signal, 1);
1186 break;
1187 default:
1188 break;
1189 }
1190 }
1191
1192 /*
1193 * Setup signal handler for :
1194 * SIGINT, SIGTERM, SIGPIPE
1195 */
1196 static int set_signal_handler(void)
1197 {
1198 int ret = 0;
1199 struct sigaction sa;
1200 sigset_t sigset;
1201
1202 if ((ret = sigemptyset(&sigset)) < 0) {
1203 PERROR("sigemptyset");
1204 return ret;
1205 }
1206
1207 sa.sa_mask = sigset;
1208 sa.sa_flags = 0;
1209
1210 sa.sa_handler = sighandler;
1211 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
1212 PERROR("sigaction");
1213 return ret;
1214 }
1215
1216 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
1217 PERROR("sigaction");
1218 return ret;
1219 }
1220
1221 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
1222 PERROR("sigaction");
1223 return ret;
1224 }
1225
1226 sa.sa_handler = SIG_IGN;
1227 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
1228 PERROR("sigaction");
1229 return ret;
1230 }
1231
1232 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
1233
1234 return ret;
1235 }
1236
1237 /*
1238 * Set open files limit to unlimited. This daemon can open a large number of
1239 * file descriptors in order to consume multiple kernel traces.
1240 */
1241 static void set_ulimit(void)
1242 {
1243 int ret;
1244 struct rlimit lim;
1245
1246 /* The kernel does not allow an infinite limit for open files */
1247 lim.rlim_cur = 65535;
1248 lim.rlim_max = 65535;
1249
1250 ret = setrlimit(RLIMIT_NOFILE, &lim);
1251 if (ret < 0) {
1252 PERROR("failed to set open files limit");
1253 }
1254 }
1255
1256 static int write_pidfile(void)
1257 {
1258 return utils_create_pid_file(getpid(), the_config.pid_file_path.value);
1259 }
1260
1261 static int set_clock_plugin_env(void)
1262 {
1263 int ret = 0;
1264 char *env_value = NULL;
1265
1266 if (!the_config.lttng_ust_clock_plugin.value) {
1267 goto end;
1268 }
1269
1270 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
1271 the_config.lttng_ust_clock_plugin.value);
1272 if (ret < 0) {
1273 PERROR("asprintf");
1274 goto end;
1275 }
1276
1277 ret = putenv(env_value);
1278 if (ret) {
1279 free(env_value);
1280 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
1281 goto end;
1282 }
1283
1284 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
1285 the_config.lttng_ust_clock_plugin.value);
1286 end:
1287 return ret;
1288 }
1289
1290 static void destroy_all_sessions_and_wait(void)
1291 {
1292 struct ltt_session *session, *tmp;
1293 struct ltt_session_list *session_list;
1294
1295 session_list = session_get_list();
1296 DBG("Initiating destruction of all sessions");
1297
1298 if (!session_list) {
1299 return;
1300 }
1301
1302 session_lock_list();
1303 /* Initiate the destruction of all sessions. */
1304 cds_list_for_each_entry_safe(session, tmp,
1305 &session_list->head, list) {
1306 if (!session_get(session)) {
1307 continue;
1308 }
1309
1310 session_lock(session);
1311 if (session->destroyed) {
1312 goto unlock_session;
1313 }
1314 (void) cmd_stop_trace(session);
1315 (void) cmd_destroy_session(
1316 session, the_notification_thread_handle, NULL);
1317 unlock_session:
1318 session_unlock(session);
1319 session_put(session);
1320 }
1321 session_unlock_list();
1322
1323 /* Wait for the destruction of all sessions to complete. */
1324 DBG("Waiting for the destruction of all sessions to complete");
1325 session_list_wait_empty();
1326 DBG("Destruction of all sessions completed");
1327 }
1328
1329 static void unregister_all_triggers(void)
1330 {
1331 enum lttng_error_code ret_code;
1332 enum lttng_trigger_status trigger_status;
1333 struct lttng_triggers *triggers = NULL;
1334 unsigned int trigger_count, i;
1335 const struct lttng_credentials creds = {
1336 .uid = LTTNG_OPTIONAL_INIT_VALUE(0),
1337 };
1338
1339 DBG("Unregistering all triggers");
1340
1341 /*
1342 * List all triggers as "root" since we wish to unregister all triggers.
1343 */
1344 ret_code = notification_thread_command_list_triggers(
1345 the_notification_thread_handle, creds.uid.value,
1346 &triggers);
1347 if (ret_code != LTTNG_OK) {
1348 ERR("Failed to list triggers while unregistering all triggers");
1349 goto end;
1350 }
1351
1352 trigger_status = lttng_triggers_get_count(triggers, &trigger_count);
1353 assert(trigger_status == LTTNG_TRIGGER_STATUS_OK);
1354
1355 for (i = 0; i < trigger_count; i++) {
1356 uid_t trigger_owner;
1357 const char *trigger_name;
1358 const struct lttng_trigger *trigger =
1359 lttng_triggers_get_at_index(triggers, i);
1360
1361 assert(trigger);
1362
1363 trigger_status = lttng_trigger_get_owner_uid(
1364 trigger, &trigger_owner);
1365 assert(trigger_status == LTTNG_TRIGGER_STATUS_OK);
1366
1367 trigger_status = lttng_trigger_get_name(trigger, &trigger_name);
1368 trigger_name = trigger_status == LTTNG_TRIGGER_STATUS_OK ?
1369 trigger_name : "(anonymous)";
1370
1371 DBG("Unregistering trigger: trigger owner uid = %d, trigger name = '%s'",
1372 (int) trigger_owner, trigger_name);
1373
1374 ret_code = cmd_unregister_trigger(&creds, trigger,
1375 the_notification_thread_handle);
1376 if (ret_code != LTTNG_OK) {
1377 ERR("Failed to unregister trigger: trigger owner uid = %d, trigger name = '%s', error: '%s'",
1378 (int) trigger_owner, trigger_name,
1379 lttng_strerror(-ret_code));
1380 /* Continue to unregister the remaining triggers. */
1381 }
1382 }
1383 end:
1384 lttng_triggers_destroy(triggers);
1385 }
1386
1387 static int run_as_worker_post_fork_cleanup(void *data)
1388 {
1389 struct sessiond_config *sessiond_config = data;
1390
1391 sessiond_config_fini(sessiond_config);
1392 return 0;
1393 }
1394
1395 static int launch_run_as_worker(const char *procname)
1396 {
1397 /*
1398 * Clean-up before forking the run-as worker. Any dynamically
1399 * allocated memory of which the worker is not aware will
1400 * be leaked as the process forks a run-as worker (and performs
1401 * no exec*()). The same would apply to any opened fd.
1402 */
1403 return run_as_create_worker(
1404 procname, run_as_worker_post_fork_cleanup, &the_config);
1405 }
1406
1407 static void sessiond_uuid_log(void)
1408 {
1409 char uuid_str[LTTNG_UUID_STR_LEN];
1410
1411 lttng_uuid_to_str(the_sessiond_uuid, uuid_str);
1412 DBG("Starting lttng-sessiond {%s}", uuid_str);
1413 }
1414
1415 /*
1416 * main
1417 */
1418 int main(int argc, char **argv)
1419 {
1420 int ret = 0, retval = 0;
1421 const char *env_app_timeout;
1422 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
1423 *ust64_channel_monitor_pipe = NULL,
1424 *kernel_channel_monitor_pipe = NULL;
1425 struct lttng_thread *ht_cleanup_thread = NULL;
1426 struct timer_thread_parameters timer_thread_parameters;
1427 /* Rotation thread handle. */
1428 struct rotation_thread_handle *rotation_thread_handle = NULL;
1429 /* Queue of rotation jobs populated by the sessiond-timer. */
1430 struct rotation_thread_timer_queue *rotation_timer_queue = NULL;
1431 struct lttng_thread *client_thread = NULL;
1432 struct lttng_thread *notification_thread = NULL;
1433 struct lttng_thread *register_apps_thread = NULL;
1434 enum event_notifier_error_accounting_status event_notifier_error_accounting_status;
1435
1436 logger_set_thread_name("Main", false);
1437 init_kernel_workarounds();
1438
1439 rcu_register_thread();
1440
1441 if (set_signal_handler()) {
1442 retval = -1;
1443 goto exit_set_signal_handler;
1444 }
1445
1446 if (timer_signal_init()) {
1447 retval = -1;
1448 goto exit_set_signal_handler;
1449 }
1450
1451 the_page_size = sysconf(_SC_PAGESIZE);
1452 if (the_page_size < 0) {
1453 PERROR("sysconf _SC_PAGESIZE");
1454 the_page_size = LONG_MAX;
1455 WARN("Fallback page size to %ld", the_page_size);
1456 }
1457
1458 ret = sessiond_config_init(&the_config);
1459 if (ret) {
1460 retval = -1;
1461 goto exit_set_signal_handler;
1462 }
1463
1464 /*
1465 * Init config from environment variables.
1466 * Command line option override env configuration per-doc. Do env first.
1467 */
1468 sessiond_config_apply_env_config(&the_config);
1469
1470 /*
1471 * Parse arguments and load the daemon configuration file.
1472 *
1473 * We have an exit_options exit path to free memory reserved by
1474 * set_options. This is needed because the rest of sessiond_cleanup()
1475 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
1476 * depends on set_options.
1477 */
1478 progname = argv[0];
1479 if (set_options(argc, argv)) {
1480 retval = -1;
1481 goto exit_options;
1482 }
1483
1484 /*
1485 * Resolve all paths received as arguments, configuration option, or
1486 * through environment variable as absolute paths. This is necessary
1487 * since daemonizing causes the sessiond's current working directory
1488 * to '/'.
1489 */
1490 ret = sessiond_config_resolve_paths(&the_config);
1491 if (ret) {
1492 goto exit_options;
1493 }
1494
1495 /* Apply config. */
1496 lttng_opt_verbose = the_config.verbose;
1497 lttng_opt_quiet = the_config.quiet;
1498 the_kconsumer_data.err_unix_sock_path =
1499 the_config.kconsumerd_err_unix_sock_path.value;
1500 the_kconsumer_data.cmd_unix_sock_path =
1501 the_config.kconsumerd_cmd_unix_sock_path.value;
1502 the_ustconsumer32_data.err_unix_sock_path =
1503 the_config.consumerd32_err_unix_sock_path.value;
1504 the_ustconsumer32_data.cmd_unix_sock_path =
1505 the_config.consumerd32_cmd_unix_sock_path.value;
1506 the_ustconsumer64_data.err_unix_sock_path =
1507 the_config.consumerd64_err_unix_sock_path.value;
1508 the_ustconsumer64_data.cmd_unix_sock_path =
1509 the_config.consumerd64_cmd_unix_sock_path.value;
1510 set_clock_plugin_env();
1511
1512 sessiond_config_log(&the_config);
1513 sessiond_uuid_log();
1514
1515 if (opt_print_version) {
1516 print_version();
1517 retval = 0;
1518 goto exit_options;
1519 }
1520
1521 if (create_lttng_rundir()) {
1522 retval = -1;
1523 goto exit_options;
1524 }
1525
1526 /* Abort launch if a session daemon is already running. */
1527 if (check_existing_daemon()) {
1528 ERR("A session daemon is already running.");
1529 retval = -1;
1530 goto exit_options;
1531 }
1532
1533 /* Daemonize */
1534 if (the_config.daemonize || the_config.background) {
1535 int i;
1536
1537 ret = lttng_daemonize(&the_child_ppid, &recv_child_signal,
1538 !the_config.background);
1539 if (ret < 0) {
1540 retval = -1;
1541 goto exit_options;
1542 }
1543
1544 /*
1545 * We are in the child. Make sure all other file descriptors are
1546 * closed, in case we are called with more opened file
1547 * descriptors than the standard ones and the lock file.
1548 */
1549 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
1550 if (i == lockfile_fd) {
1551 continue;
1552 }
1553 (void) close(i);
1554 }
1555 }
1556
1557 if (launch_run_as_worker(argv[0]) < 0) {
1558 goto exit_create_run_as_worker_cleanup;
1559 }
1560
1561 /*
1562 * Starting from here, we can create threads. This needs to be after
1563 * lttng_daemonize due to RCU.
1564 */
1565
1566 /*
1567 * Initialize the health check subsystem. This call should set the
1568 * appropriate time values.
1569 */
1570 the_health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
1571 if (!the_health_sessiond) {
1572 PERROR("health_app_create error");
1573 retval = -1;
1574 goto stop_threads;
1575 }
1576
1577 /* Create thread to clean up RCU hash tables */
1578 ht_cleanup_thread = launch_ht_cleanup_thread();
1579 if (!ht_cleanup_thread) {
1580 retval = -1;
1581 goto stop_threads;
1582 }
1583
1584 /* Create thread quit pipe */
1585 if (sessiond_init_thread_quit_pipe()) {
1586 retval = -1;
1587 goto stop_threads;
1588 }
1589
1590 /* Check if daemon is UID = 0 */
1591 is_root = !getuid();
1592 if (is_root) {
1593 /* Create global run dir with root access */
1594
1595 kernel_channel_monitor_pipe = lttng_pipe_open(0);
1596 if (!kernel_channel_monitor_pipe) {
1597 ERR("Failed to create kernel consumer channel monitor pipe");
1598 retval = -1;
1599 goto stop_threads;
1600 }
1601 the_kconsumer_data.channel_monitor_pipe =
1602 lttng_pipe_release_writefd(
1603 kernel_channel_monitor_pipe);
1604 if (the_kconsumer_data.channel_monitor_pipe < 0) {
1605 retval = -1;
1606 goto stop_threads;
1607 }
1608 }
1609
1610 /* Set consumer initial state */
1611 the_kernel_consumerd_state = CONSUMER_STOPPED;
1612 the_ust_consumerd_state = CONSUMER_STOPPED;
1613
1614 ust32_channel_monitor_pipe = lttng_pipe_open(0);
1615 if (!ust32_channel_monitor_pipe) {
1616 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
1617 retval = -1;
1618 goto stop_threads;
1619 }
1620 the_ustconsumer32_data.channel_monitor_pipe =
1621 lttng_pipe_release_writefd(ust32_channel_monitor_pipe);
1622 if (the_ustconsumer32_data.channel_monitor_pipe < 0) {
1623 retval = -1;
1624 goto stop_threads;
1625 }
1626
1627 /*
1628 * The rotation_thread_timer_queue structure is shared between the
1629 * sessiond timer thread and the rotation thread. The main thread keeps
1630 * its ownership and destroys it when both threads have been joined.
1631 */
1632 rotation_timer_queue = rotation_thread_timer_queue_create();
1633 if (!rotation_timer_queue) {
1634 retval = -1;
1635 goto stop_threads;
1636 }
1637 timer_thread_parameters.rotation_thread_job_queue =
1638 rotation_timer_queue;
1639
1640 ust64_channel_monitor_pipe = lttng_pipe_open(0);
1641 if (!ust64_channel_monitor_pipe) {
1642 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
1643 retval = -1;
1644 goto stop_threads;
1645 }
1646 the_ustconsumer64_data.channel_monitor_pipe =
1647 lttng_pipe_release_writefd(ust64_channel_monitor_pipe);
1648 if (the_ustconsumer64_data.channel_monitor_pipe < 0) {
1649 retval = -1;
1650 goto stop_threads;
1651 }
1652
1653 /*
1654 * Init UST app hash table. Alloc hash table before this point since
1655 * cleanup() can get called after that point.
1656 */
1657 if (ust_app_ht_alloc()) {
1658 ERR("Failed to allocate UST app hash table");
1659 retval = -1;
1660 goto stop_threads;
1661 }
1662
1663 event_notifier_error_accounting_status = event_notifier_error_accounting_init(
1664 the_config.event_notifier_buffer_size_kernel,
1665 the_config.event_notifier_buffer_size_userspace);
1666 if (event_notifier_error_accounting_status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) {
1667 ERR("Failed to initialize event notifier error accounting system");
1668 retval = -1;
1669 goto stop_threads;
1670 }
1671
1672 /*
1673 * Initialize agent app hash table. We allocate the hash table here
1674 * since cleanup() can get called after this point.
1675 */
1676 if (agent_app_ht_alloc()) {
1677 ERR("Failed to allocate Agent app hash table");
1678 retval = -1;
1679 goto stop_threads;
1680 }
1681
1682 if (agent_by_event_notifier_domain_ht_create()) {
1683 ERR("Failed to allocate per-event notifier domain agent hash table");
1684 retval = -1;
1685 goto stop_threads;
1686 }
1687 /*
1688 * These actions must be executed as root. We do that *after* setting up
1689 * the sockets path because we MUST make the check for another daemon using
1690 * those paths *before* trying to set the kernel consumer sockets and init
1691 * kernel tracer.
1692 */
1693 if (is_root) {
1694 if (set_consumer_sockets(&the_kconsumer_data)) {
1695 retval = -1;
1696 goto stop_threads;
1697 }
1698
1699 /* Setup kernel tracer */
1700 if (!the_config.no_kernel) {
1701 init_kernel_tracer();
1702 }
1703
1704 /* Set ulimit for open files */
1705 set_ulimit();
1706 }
1707 /* init lttng_fd tracking must be done after set_ulimit. */
1708 lttng_fd_init();
1709
1710 if (set_consumer_sockets(&the_ustconsumer64_data)) {
1711 retval = -1;
1712 goto stop_threads;
1713 }
1714
1715 if (set_consumer_sockets(&the_ustconsumer32_data)) {
1716 retval = -1;
1717 goto stop_threads;
1718 }
1719
1720 /* Get parent pid if -S, --sig-parent is specified. */
1721 if (the_config.sig_parent) {
1722 the_ppid = getppid();
1723 }
1724
1725 /* Setup the kernel pipe for waking up the kernel thread */
1726 if (is_root && !the_config.no_kernel) {
1727 if (utils_create_pipe_cloexec(the_kernel_poll_pipe)) {
1728 retval = -1;
1729 goto stop_threads;
1730 }
1731 }
1732
1733 /* Setup the thread apps communication pipe. */
1734 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
1735 retval = -1;
1736 goto stop_threads;
1737 }
1738
1739 /* Setup the thread apps notify communication pipe. */
1740 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
1741 retval = -1;
1742 goto stop_threads;
1743 }
1744
1745 /* Initialize global buffer per UID and PID registry. */
1746 buffer_reg_init_uid_registry();
1747 buffer_reg_init_pid_registry();
1748
1749 /* Init UST command queue. */
1750 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1751
1752 cmd_init();
1753
1754 /* Check for the application socket timeout env variable. */
1755 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
1756 if (env_app_timeout) {
1757 the_config.app_socket_timeout = atoi(env_app_timeout);
1758 } else {
1759 the_config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
1760 }
1761
1762 ret = write_pidfile();
1763 if (ret) {
1764 ERR("Error in write_pidfile");
1765 retval = -1;
1766 goto stop_threads;
1767 }
1768
1769 /* Initialize communication library */
1770 lttcomm_init();
1771 /* Initialize TCP timeout values */
1772 lttcomm_inet_init();
1773
1774 /* Create health-check thread. */
1775 if (!launch_health_management_thread()) {
1776 retval = -1;
1777 goto stop_threads;
1778 }
1779
1780 /* notification_thread_data acquires the pipes' read side. */
1781 the_notification_thread_handle = notification_thread_handle_create(
1782 ust32_channel_monitor_pipe, ust64_channel_monitor_pipe,
1783 kernel_channel_monitor_pipe);
1784 if (!the_notification_thread_handle) {
1785 retval = -1;
1786 ERR("Failed to create notification thread shared data");
1787 goto stop_threads;
1788 }
1789
1790 /* Create notification thread. */
1791 notification_thread = launch_notification_thread(
1792 the_notification_thread_handle);
1793 if (!notification_thread) {
1794 retval = -1;
1795 goto stop_threads;
1796 }
1797
1798 /* Create timer thread. */
1799 if (!launch_timer_thread(&timer_thread_parameters)) {
1800 retval = -1;
1801 goto stop_threads;
1802 }
1803
1804 /* rotation_thread_data acquires the pipes' read side. */
1805 rotation_thread_handle = rotation_thread_handle_create(
1806 rotation_timer_queue, the_notification_thread_handle);
1807 if (!rotation_thread_handle) {
1808 retval = -1;
1809 ERR("Failed to create rotation thread shared data");
1810 stop_threads();
1811 goto stop_threads;
1812 }
1813
1814 /* Create rotation thread. */
1815 if (!launch_rotation_thread(rotation_thread_handle)) {
1816 retval = -1;
1817 goto stop_threads;
1818 }
1819
1820 /* Create thread to manage the client socket */
1821 client_thread = launch_client_thread();
1822 if (!client_thread) {
1823 retval = -1;
1824 goto stop_threads;
1825 }
1826
1827 /* Set credentials of the client socket and rundir */
1828 if (is_root && set_permissions(the_config.rundir.value)) {
1829 retval = -1;
1830 goto stop_threads;
1831 }
1832
1833 if (!launch_ust_dispatch_thread(&ust_cmd_queue, apps_cmd_pipe[1],
1834 apps_cmd_notify_pipe[1])) {
1835 retval = -1;
1836 goto stop_threads;
1837 }
1838
1839 /* Create thread to manage application registration. */
1840 register_apps_thread = launch_application_registration_thread(
1841 &ust_cmd_queue);
1842 if (!register_apps_thread) {
1843 retval = -1;
1844 goto stop_threads;
1845 }
1846
1847 /* Create thread to manage application socket */
1848 if (!launch_application_management_thread(apps_cmd_pipe[0])) {
1849 retval = -1;
1850 goto stop_threads;
1851 }
1852
1853 /* Create thread to manage application notify socket */
1854 if (!launch_application_notification_thread(apps_cmd_notify_pipe[0])) {
1855 retval = -1;
1856 goto stop_threads;
1857 }
1858
1859 /* Create agent management thread. */
1860 if (!launch_agent_management_thread()) {
1861 retval = -1;
1862 goto stop_threads;
1863 }
1864
1865 /* Don't start this thread if kernel tracing is not requested nor root */
1866 if (is_root && !the_config.no_kernel) {
1867 /* Create kernel thread to manage kernel event */
1868 if (!launch_kernel_management_thread(the_kernel_poll_pipe[0])) {
1869 retval = -1;
1870 goto stop_threads;
1871 }
1872
1873 if (kernel_get_notification_fd() >= 0) {
1874 ret = notification_thread_command_add_tracer_event_source(
1875 the_notification_thread_handle,
1876 kernel_get_notification_fd(),
1877 LTTNG_DOMAIN_KERNEL);
1878 if (ret != LTTNG_OK) {
1879 ERR("Failed to add kernel trigger event source to notification thread");
1880 retval = -1;
1881 goto stop_threads;
1882 }
1883 }
1884 }
1885
1886 /* Load sessions. */
1887 ret = config_load_session(
1888 the_config.load_session_path.value, NULL, 1, 1, NULL);
1889 if (ret) {
1890 ERR("Session load failed: %s", error_get_str(ret));
1891 retval = -1;
1892 goto stop_threads;
1893 }
1894
1895 /* Initialization completed. */
1896 sessiond_signal_parents();
1897
1898 /*
1899 * This is where we start awaiting program completion (e.g. through
1900 * signal that asks threads to teardown).
1901 */
1902
1903 /* Initiate teardown once activity occurs on the quit pipe. */
1904 sessiond_wait_for_quit_pipe(-1);
1905
1906 stop_threads:
1907
1908 /*
1909 * Ensure that the client thread is no longer accepting new commands,
1910 * which could cause new sessions to be created.
1911 */
1912 if (client_thread) {
1913 lttng_thread_shutdown(client_thread);
1914 lttng_thread_put(client_thread);
1915 }
1916
1917 destroy_all_sessions_and_wait();
1918
1919 /*
1920 * At this point no new trigger can be registered (no sessions are
1921 * running/rotating) and clients can't connect to the session daemon
1922 * anymore. Unregister all triggers.
1923 */
1924 unregister_all_triggers();
1925
1926 if (register_apps_thread) {
1927 lttng_thread_shutdown(register_apps_thread);
1928 lttng_thread_put(register_apps_thread);
1929 }
1930 lttng_thread_list_shutdown_orphans();
1931
1932 /*
1933 * Wait for all pending call_rcu work to complete before tearing
1934 * down data structures. call_rcu worker may be trying to
1935 * perform lookups in those structures.
1936 */
1937 rcu_barrier();
1938 /*
1939 * sessiond_cleanup() is called when no other thread is running, except
1940 * the ht_cleanup thread, which is needed to destroy the hash tables.
1941 */
1942 rcu_thread_online();
1943 sessiond_cleanup();
1944
1945 /*
1946 * Wait for all pending call_rcu work to complete before shutting down
1947 * the notification thread. This call_rcu work includes shutting down
1948 * UST apps and event notifier pipes.
1949 */
1950 rcu_barrier();
1951
1952 if (notification_thread) {
1953 lttng_thread_shutdown(notification_thread);
1954 lttng_thread_put(notification_thread);
1955 }
1956
1957 /*
1958 * Error accounting teardown has to be done after the teardown of all
1959 * event notifier pipes to ensure that no tracer may try to use the
1960 * error accounting facilities.
1961 */
1962 event_notifier_error_accounting_fini();
1963
1964 /*
1965 * Unloading the kernel modules needs to be done after all kernel
1966 * ressources have been released. In our case, this includes the
1967 * notification fd, the event notifier group fd, error accounting fd,
1968 * all event and event notifier fds, etc.
1969 *
1970 * In short, at this point, we need to have called close() on all fds
1971 * received from the kernel tracer.
1972 */
1973 if (is_root && !the_config.no_kernel) {
1974 DBG("Unloading kernel modules");
1975 modprobe_remove_lttng_all();
1976 }
1977
1978 /*
1979 * Ensure all prior call_rcu are done. call_rcu callbacks may push
1980 * hash tables to the ht_cleanup thread. Therefore, we ensure that
1981 * the queue is empty before shutting down the clean-up thread.
1982 */
1983 rcu_barrier();
1984
1985 if (ht_cleanup_thread) {
1986 lttng_thread_shutdown(ht_cleanup_thread);
1987 lttng_thread_put(ht_cleanup_thread);
1988 }
1989
1990 rcu_thread_offline();
1991 rcu_unregister_thread();
1992
1993 if (rotation_thread_handle) {
1994 rotation_thread_handle_destroy(rotation_thread_handle);
1995 }
1996
1997 /*
1998 * After the rotation and timer thread have quit, we can safely destroy
1999 * the rotation_timer_queue.
2000 */
2001 rotation_thread_timer_queue_destroy(rotation_timer_queue);
2002 /*
2003 * The teardown of the notification system is performed after the
2004 * session daemon's teardown in order to allow it to be notified
2005 * of the active session and channels at the moment of the teardown.
2006 */
2007 if (the_notification_thread_handle) {
2008 notification_thread_handle_destroy(
2009 the_notification_thread_handle);
2010 }
2011 lttng_pipe_destroy(ust32_channel_monitor_pipe);
2012 lttng_pipe_destroy(ust64_channel_monitor_pipe);
2013 lttng_pipe_destroy(kernel_channel_monitor_pipe);
2014
2015 if (the_health_sessiond) {
2016 health_app_destroy(the_health_sessiond);
2017 }
2018 exit_create_run_as_worker_cleanup:
2019 exit_options:
2020 sessiond_cleanup_lock_file();
2021 sessiond_cleanup_options();
2022
2023 exit_set_signal_handler:
2024 if (!retval) {
2025 exit(EXIT_SUCCESS);
2026 } else {
2027 exit(EXIT_FAILURE);
2028 }
2029 }
This page took 0.123196 seconds and 4 git commands to generate.