Remove ht-cleanup thread
[lttng-tools.git] / src / bin / lttng-sessiond / main.cpp
1 /*
2 * Copyright (C) 2011 EfficiOS Inc.
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/path.h>
41 #include <common/daemonize.h>
42 #include <common/config/session-config.h>
43 #include <common/ini-config/ini-config.h>
44 #include <common/dynamic-buffer.h>
45 #include <lttng/event-internal.h>
46
47 #include "lttng-sessiond.h"
48 #include "buffer-registry.h"
49 #include "channel.h"
50 #include "cmd.h"
51 #include "consumer.h"
52 #include "context.h"
53 #include "event.h"
54 #include "event-notifier-error-accounting.h"
55 #include "kernel.h"
56 #include "kernel-consumer.h"
57 #include "lttng-ust-ctl.h"
58 #include "ust-consumer.h"
59 #include "utils.h"
60 #include "fd-limit.h"
61 #include "health-sessiond.h"
62 #include "testpoint.h"
63 #include "notify-apps.h"
64 #include "agent-thread.h"
65 #include "save.h"
66 #include "notification-thread.h"
67 #include "notification-thread-commands.h"
68 #include "rotation-thread.h"
69 #include "agent.h"
70 #include "sessiond-config.h"
71 #include "timer.h"
72 #include "thread.h"
73 #include "client.h"
74 #include "dispatch.h"
75 #include "register.h"
76 #include "manage-apps.h"
77 #include "manage-kernel.h"
78 #include "modprobe.h"
79 #include "ust-sigbus.h"
80
81 static const char *help_msg =
82 #ifdef LTTNG_EMBED_HELP
83 #include <lttng-sessiond.8.h>
84 #else
85 NULL
86 #endif
87 ;
88
89 #define EVENT_NOTIFIER_ERROR_COUNTER_NUMBER_OF_BUCKET_MAX 65535
90 #define EVENT_NOTIFIER_ERROR_BUFFER_SIZE_BASE_OPTION_STR \
91 "event-notifier-error-buffer-size"
92 #define EVENT_NOTIFIER_ERROR_BUFFER_SIZE_KERNEL_OPTION_STR \
93 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_BASE_OPTION_STR "-kernel"
94 #define EVENT_NOTIFIER_ERROR_BUFFER_SIZE_USERSPACE_OPTION_STR \
95 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_BASE_OPTION_STR "-userspace"
96
97
98 const char *progname;
99 static int lockfile_fd = -1;
100 static int opt_print_version;
101
102 /* Set to 1 when a SIGUSR1 signal is received. */
103 static int recv_child_signal;
104
105 /* Command line options */
106 static const struct option long_options[] = {
107 { "client-sock", required_argument, 0, 'c' },
108 { "apps-sock", required_argument, 0, 'a' },
109 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
110 { "kconsumerd-err-sock", required_argument, 0, '\0' },
111 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
112 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
113 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
114 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
115 { "consumerd32-path", required_argument, 0, '\0' },
116 { "consumerd32-libdir", required_argument, 0, '\0' },
117 { "consumerd64-path", required_argument, 0, '\0' },
118 { "consumerd64-libdir", required_argument, 0, '\0' },
119 { "daemonize", no_argument, 0, 'd' },
120 { "background", no_argument, 0, 'b' },
121 { "sig-parent", no_argument, 0, 'S' },
122 { "help", no_argument, 0, 'h' },
123 { "group", required_argument, 0, 'g' },
124 { "version", no_argument, 0, 'V' },
125 { "quiet", no_argument, 0, 'q' },
126 { "verbose", no_argument, 0, 'v' },
127 { "verbose-consumer", no_argument, 0, '\0' },
128 { "no-kernel", no_argument, 0, '\0' },
129 { "pidfile", required_argument, 0, 'p' },
130 { "agent-tcp-port", required_argument, 0, '\0' },
131 { "config", required_argument, 0, 'f' },
132 { "load", required_argument, 0, 'l' },
133 { "kmod-probes", required_argument, 0, '\0' },
134 { "extra-kmod-probes", required_argument, 0, '\0' },
135 { EVENT_NOTIFIER_ERROR_BUFFER_SIZE_KERNEL_OPTION_STR, required_argument, 0, '\0' },
136 { EVENT_NOTIFIER_ERROR_BUFFER_SIZE_USERSPACE_OPTION_STR, required_argument, 0, '\0' },
137 { NULL, 0, 0, 0 }
138 };
139
140 /* Command line options to ignore from configuration file */
141 static const char *config_ignore_options[] = { "help", "version", "config" };
142
143 /*
144 * This pipe is used to inform the thread managing application communication
145 * that a command is queued and ready to be processed.
146 */
147 static int apps_cmd_pipe[2] = { -1, -1 };
148 static int apps_cmd_notify_pipe[2] = { -1, -1 };
149
150 /*
151 * UST registration command queue. This queue is tied with a futex and uses a N
152 * wakers / 1 waiter implemented and detailed in futex.c/.h
153 *
154 * The thread_registration_apps and thread_dispatch_ust_registration uses this
155 * queue along with the wait/wake scheme. The thread_manage_apps receives down
156 * the line new application socket and monitors it for any I/O error or clean
157 * close that triggers an unregistration of the application.
158 */
159 static struct ust_cmd_queue ust_cmd_queue;
160
161 /*
162 * Section name to look for in the daemon configuration file.
163 */
164 static const char * const config_section_name = "sessiond";
165
166 /* Am I root or not. Set to 1 if the daemon is running as root */
167 static int is_root;
168
169 /*
170 * Stop all threads by closing the thread quit pipe.
171 */
172 static void stop_threads(void)
173 {
174 int ret;
175
176 /* Stopping all threads */
177 DBG("Terminating all threads");
178 ret = sessiond_notify_quit_pipe();
179 if (ret < 0) {
180 ERR("write error on thread quit pipe");
181 }
182 }
183
184 /*
185 * Close every consumer sockets.
186 */
187 static void close_consumer_sockets(void)
188 {
189 int ret;
190
191 if (the_kconsumer_data.err_sock >= 0) {
192 ret = close(the_kconsumer_data.err_sock);
193 if (ret < 0) {
194 PERROR("kernel consumer err_sock close");
195 }
196 }
197 if (the_ustconsumer32_data.err_sock >= 0) {
198 ret = close(the_ustconsumer32_data.err_sock);
199 if (ret < 0) {
200 PERROR("UST consumerd32 err_sock close");
201 }
202 }
203 if (the_ustconsumer64_data.err_sock >= 0) {
204 ret = close(the_ustconsumer64_data.err_sock);
205 if (ret < 0) {
206 PERROR("UST consumerd64 err_sock close");
207 }
208 }
209 if (the_kconsumer_data.cmd_sock >= 0) {
210 ret = close(the_kconsumer_data.cmd_sock);
211 if (ret < 0) {
212 PERROR("kernel consumer cmd_sock close");
213 }
214 }
215 if (the_ustconsumer32_data.cmd_sock >= 0) {
216 ret = close(the_ustconsumer32_data.cmd_sock);
217 if (ret < 0) {
218 PERROR("UST consumerd32 cmd_sock close");
219 }
220 }
221 if (the_ustconsumer64_data.cmd_sock >= 0) {
222 ret = close(the_ustconsumer64_data.cmd_sock);
223 if (ret < 0) {
224 PERROR("UST consumerd64 cmd_sock close");
225 }
226 }
227 if (the_kconsumer_data.channel_monitor_pipe >= 0) {
228 ret = close(the_kconsumer_data.channel_monitor_pipe);
229 if (ret < 0) {
230 PERROR("kernel consumer channel monitor pipe close");
231 }
232 }
233 if (the_ustconsumer32_data.channel_monitor_pipe >= 0) {
234 ret = close(the_ustconsumer32_data.channel_monitor_pipe);
235 if (ret < 0) {
236 PERROR("UST consumerd32 channel monitor pipe close");
237 }
238 }
239 if (the_ustconsumer64_data.channel_monitor_pipe >= 0) {
240 ret = close(the_ustconsumer64_data.channel_monitor_pipe);
241 if (ret < 0) {
242 PERROR("UST consumerd64 channel monitor pipe close");
243 }
244 }
245 }
246
247 /*
248 * Wait on consumer process termination.
249 *
250 * Need to be called with the consumer data lock held or from a context
251 * ensuring no concurrent access to data (e.g: cleanup).
252 */
253 static void wait_consumer(struct consumer_data *consumer_data)
254 {
255 pid_t ret;
256 int status;
257
258 if (consumer_data->pid <= 0) {
259 return;
260 }
261
262 DBG("Waiting for complete teardown of consumerd (PID: %d)",
263 consumer_data->pid);
264 ret = waitpid(consumer_data->pid, &status, 0);
265 if (ret == -1) {
266 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
267 } else if (!WIFEXITED(status)) {
268 ERR("consumerd termination with error: %d",
269 WEXITSTATUS(ret));
270 }
271 consumer_data->pid = 0;
272 }
273
274 /*
275 * Cleanup the session daemon's data structures.
276 */
277 static void sessiond_cleanup(void)
278 {
279 int ret;
280 struct ltt_session_list *session_list = session_get_list();
281
282 DBG("Cleanup sessiond");
283
284 /*
285 * Close the thread quit pipe. It has already done its job,
286 * since we are now called.
287 */
288 sessiond_close_quit_pipe();
289 utils_close_pipe(apps_cmd_pipe);
290 utils_close_pipe(apps_cmd_notify_pipe);
291 utils_close_pipe(the_kernel_poll_pipe);
292
293 ret = remove(the_config.pid_file_path.value);
294 if (ret < 0) {
295 PERROR("remove pidfile %s", the_config.pid_file_path.value);
296 }
297
298 DBG("Removing sessiond and consumerd content of directory %s",
299 the_config.rundir.value);
300
301 /* sessiond */
302 DBG("Removing %s", the_config.pid_file_path.value);
303 (void) unlink(the_config.pid_file_path.value);
304
305 DBG("Removing %s", the_config.agent_port_file_path.value);
306 (void) unlink(the_config.agent_port_file_path.value);
307
308 /* kconsumerd */
309 DBG("Removing %s", the_kconsumer_data.err_unix_sock_path);
310 (void) unlink(the_kconsumer_data.err_unix_sock_path);
311
312 DBG("Removing directory %s", the_config.kconsumerd_path.value);
313 (void) rmdir(the_config.kconsumerd_path.value);
314
315 /* ust consumerd 32 */
316 DBG("Removing %s", the_config.consumerd32_err_unix_sock_path.value);
317 (void) unlink(the_config.consumerd32_err_unix_sock_path.value);
318
319 DBG("Removing directory %s", the_config.consumerd32_path.value);
320 (void) rmdir(the_config.consumerd32_path.value);
321
322 /* ust consumerd 64 */
323 DBG("Removing %s", the_config.consumerd64_err_unix_sock_path.value);
324 (void) unlink(the_config.consumerd64_err_unix_sock_path.value);
325
326 DBG("Removing directory %s", the_config.consumerd64_path.value);
327 (void) rmdir(the_config.consumerd64_path.value);
328
329 pthread_mutex_destroy(&session_list->lock);
330
331 DBG("Cleaning up all per-event notifier domain agents");
332 agent_by_event_notifier_domain_ht_destroy();
333
334 DBG("Cleaning up all agent apps");
335 agent_app_ht_clean();
336 DBG("Closing all UST sockets");
337 ust_app_clean_list();
338 buffer_reg_destroy_registries();
339
340 close_consumer_sockets();
341
342 wait_consumer(&the_kconsumer_data);
343 wait_consumer(&the_ustconsumer64_data);
344 wait_consumer(&the_ustconsumer32_data);
345
346 if (is_root && !the_config.no_kernel) {
347 cleanup_kernel_tracer();
348 }
349
350 /*
351 * We do NOT rmdir rundir because there are other processes
352 * using it, for instance lttng-relayd, which can start in
353 * parallel with this teardown.
354 */
355 }
356
357 /*
358 * Cleanup the daemon's option data structures.
359 */
360 static void sessiond_cleanup_options(void)
361 {
362 DBG("Cleaning up options");
363
364 sessiond_config_fini(&the_config);
365
366 run_as_destroy_worker();
367 }
368
369 static int string_match(const char *str1, const char *str2)
370 {
371 return (str1 && str2) && !strcmp(str1, str2);
372 }
373
374 /*
375 * Take an option from the getopt output and set it in the right variable to be
376 * used later.
377 *
378 * Return 0 on success else a negative value.
379 */
380 static int set_option(int opt, const char *arg, const char *optname)
381 {
382 int ret = 0;
383
384 if (string_match(optname, "client-sock") || opt == 'c') {
385 if (!arg || *arg == '\0') {
386 ret = -EINVAL;
387 goto end;
388 }
389 if (lttng_is_setuid_setgid()) {
390 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
391 "-c, --client-sock");
392 } else {
393 config_string_set(&the_config.client_unix_sock_path,
394 strdup(arg));
395 if (!the_config.client_unix_sock_path.value) {
396 ret = -ENOMEM;
397 PERROR("strdup");
398 }
399 }
400 } else if (string_match(optname, "apps-sock") || opt == 'a') {
401 if (!arg || *arg == '\0') {
402 ret = -EINVAL;
403 goto end;
404 }
405 if (lttng_is_setuid_setgid()) {
406 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
407 "-a, --apps-sock");
408 } else {
409 config_string_set(&the_config.apps_unix_sock_path,
410 strdup(arg));
411 if (!the_config.apps_unix_sock_path.value) {
412 ret = -ENOMEM;
413 PERROR("strdup");
414 }
415 }
416 } else if (string_match(optname, "daemonize") || opt == 'd') {
417 the_config.daemonize = true;
418 } else if (string_match(optname, "background") || opt == 'b') {
419 the_config.background = true;
420 } else if (string_match(optname, "group") || opt == 'g') {
421 if (!arg || *arg == '\0') {
422 ret = -EINVAL;
423 goto end;
424 }
425 if (lttng_is_setuid_setgid()) {
426 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
427 "-g, --group");
428 } else {
429 config_string_set(&the_config.tracing_group_name,
430 strdup(arg));
431 if (!the_config.tracing_group_name.value) {
432 ret = -ENOMEM;
433 PERROR("strdup");
434 }
435 }
436 } else if (string_match(optname, "help") || opt == 'h') {
437 ret = utils_show_help(8, "lttng-sessiond", help_msg);
438 if (ret) {
439 ERR("Cannot show --help for `lttng-sessiond`");
440 perror("exec");
441 }
442 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
443 } else if (string_match(optname, "version") || opt == 'V') {
444 opt_print_version = 1;
445 } else if (string_match(optname, "sig-parent") || opt == 'S') {
446 the_config.sig_parent = true;
447 } else if (string_match(optname, "kconsumerd-err-sock")) {
448 if (!arg || *arg == '\0') {
449 ret = -EINVAL;
450 goto end;
451 }
452 if (lttng_is_setuid_setgid()) {
453 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
454 "--kconsumerd-err-sock");
455 } else {
456 config_string_set(
457 &the_config.kconsumerd_err_unix_sock_path,
458 strdup(arg));
459 if (!the_config.kconsumerd_err_unix_sock_path.value) {
460 ret = -ENOMEM;
461 PERROR("strdup");
462 }
463 }
464 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
465 if (!arg || *arg == '\0') {
466 ret = -EINVAL;
467 goto end;
468 }
469 if (lttng_is_setuid_setgid()) {
470 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
471 "--kconsumerd-cmd-sock");
472 } else {
473 config_string_set(
474 &the_config.kconsumerd_cmd_unix_sock_path,
475 strdup(arg));
476 if (!the_config.kconsumerd_cmd_unix_sock_path.value) {
477 ret = -ENOMEM;
478 PERROR("strdup");
479 }
480 }
481 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
482 if (!arg || *arg == '\0') {
483 ret = -EINVAL;
484 goto end;
485 }
486 if (lttng_is_setuid_setgid()) {
487 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
488 "--ustconsumerd64-err-sock");
489 } else {
490 config_string_set(
491 &the_config.consumerd64_err_unix_sock_path,
492 strdup(arg));
493 if (!the_config.consumerd64_err_unix_sock_path.value) {
494 ret = -ENOMEM;
495 PERROR("strdup");
496 }
497 }
498 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
499 if (!arg || *arg == '\0') {
500 ret = -EINVAL;
501 goto end;
502 }
503 if (lttng_is_setuid_setgid()) {
504 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
505 "--ustconsumerd64-cmd-sock");
506 } else {
507 config_string_set(
508 &the_config.consumerd64_cmd_unix_sock_path,
509 strdup(arg));
510 if (!the_config.consumerd64_cmd_unix_sock_path.value) {
511 ret = -ENOMEM;
512 PERROR("strdup");
513 }
514 }
515 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
516 if (!arg || *arg == '\0') {
517 ret = -EINVAL;
518 goto end;
519 }
520 if (lttng_is_setuid_setgid()) {
521 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
522 "--ustconsumerd32-err-sock");
523 } else {
524 config_string_set(
525 &the_config.consumerd32_err_unix_sock_path,
526 strdup(arg));
527 if (!the_config.consumerd32_err_unix_sock_path.value) {
528 ret = -ENOMEM;
529 PERROR("strdup");
530 }
531 }
532 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
533 if (!arg || *arg == '\0') {
534 ret = -EINVAL;
535 goto end;
536 }
537 if (lttng_is_setuid_setgid()) {
538 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
539 "--ustconsumerd32-cmd-sock");
540 } else {
541 config_string_set(
542 &the_config.consumerd32_cmd_unix_sock_path,
543 strdup(arg));
544 if (!the_config.consumerd32_cmd_unix_sock_path.value) {
545 ret = -ENOMEM;
546 PERROR("strdup");
547 }
548 }
549 } else if (string_match(optname, "no-kernel")) {
550 the_config.no_kernel = true;
551 } else if (string_match(optname, "quiet") || opt == 'q') {
552 the_config.quiet = true;
553 } else if (string_match(optname, "verbose") || opt == 'v') {
554 /* Verbose level can increase using multiple -v */
555 if (arg) {
556 /* Value obtained from config file */
557 the_config.verbose = config_parse_value(arg);
558 } else {
559 /* -v used on command line */
560 the_config.verbose++;
561 }
562 /* Clamp value to [0, 3] */
563 the_config.verbose = the_config.verbose < 0 ?
564 0 :
565 (the_config.verbose <= 3 ? the_config.verbose :
566 3);
567 } else if (string_match(optname, "verbose-consumer")) {
568 if (arg) {
569 the_config.verbose_consumer = config_parse_value(arg);
570 } else {
571 the_config.verbose_consumer++;
572 }
573 } else if (string_match(optname, "consumerd32-path")) {
574 if (!arg || *arg == '\0') {
575 ret = -EINVAL;
576 goto end;
577 }
578 if (lttng_is_setuid_setgid()) {
579 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
580 "--consumerd32-path");
581 } else {
582 config_string_set(&the_config.consumerd32_bin_path,
583 strdup(arg));
584 if (!the_config.consumerd32_bin_path.value) {
585 PERROR("strdup");
586 ret = -ENOMEM;
587 }
588 }
589 } else if (string_match(optname, "consumerd32-libdir")) {
590 if (!arg || *arg == '\0') {
591 ret = -EINVAL;
592 goto end;
593 }
594 if (lttng_is_setuid_setgid()) {
595 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
596 "--consumerd32-libdir");
597 } else {
598 config_string_set(&the_config.consumerd32_lib_dir,
599 strdup(arg));
600 if (!the_config.consumerd32_lib_dir.value) {
601 PERROR("strdup");
602 ret = -ENOMEM;
603 }
604 }
605 } else if (string_match(optname, "consumerd64-path")) {
606 if (!arg || *arg == '\0') {
607 ret = -EINVAL;
608 goto end;
609 }
610 if (lttng_is_setuid_setgid()) {
611 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
612 "--consumerd64-path");
613 } else {
614 config_string_set(&the_config.consumerd64_bin_path,
615 strdup(arg));
616 if (!the_config.consumerd64_bin_path.value) {
617 PERROR("strdup");
618 ret = -ENOMEM;
619 }
620 }
621 } else if (string_match(optname, "consumerd64-libdir")) {
622 if (!arg || *arg == '\0') {
623 ret = -EINVAL;
624 goto end;
625 }
626 if (lttng_is_setuid_setgid()) {
627 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
628 "--consumerd64-libdir");
629 } else {
630 config_string_set(&the_config.consumerd64_lib_dir,
631 strdup(arg));
632 if (!the_config.consumerd64_lib_dir.value) {
633 PERROR("strdup");
634 ret = -ENOMEM;
635 }
636 }
637 } else if (string_match(optname, "pidfile") || opt == 'p') {
638 if (!arg || *arg == '\0') {
639 ret = -EINVAL;
640 goto end;
641 }
642 if (lttng_is_setuid_setgid()) {
643 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
644 "-p, --pidfile");
645 } else {
646 config_string_set(
647 &the_config.pid_file_path, strdup(arg));
648 if (!the_config.pid_file_path.value) {
649 PERROR("strdup");
650 ret = -ENOMEM;
651 }
652 }
653 } else if (string_match(optname, "agent-tcp-port")) {
654 if (!arg || *arg == '\0') {
655 ret = -EINVAL;
656 goto end;
657 }
658 if (lttng_is_setuid_setgid()) {
659 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
660 "--agent-tcp-port");
661 } else {
662 unsigned long v;
663
664 errno = 0;
665 v = strtoul(arg, NULL, 0);
666 if (errno != 0 || !isdigit(arg[0])) {
667 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
668 return -1;
669 }
670 if (v == 0 || v >= 65535) {
671 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
672 return -1;
673 }
674 the_config.agent_tcp_port.begin =
675 the_config.agent_tcp_port.end = (int) v;
676 DBG3("Agent TCP port set to non default: %i", (int) v);
677 }
678 } else if (string_match(optname, "load") || opt == 'l') {
679 if (!arg || *arg == '\0') {
680 ret = -EINVAL;
681 goto end;
682 }
683 if (lttng_is_setuid_setgid()) {
684 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
685 "-l, --load");
686 } else {
687 config_string_set(&the_config.load_session_path,
688 strdup(arg));
689 if (!the_config.load_session_path.value) {
690 PERROR("strdup");
691 ret = -ENOMEM;
692 }
693 }
694 } else if (string_match(optname, "kmod-probes")) {
695 if (!arg || *arg == '\0') {
696 ret = -EINVAL;
697 goto end;
698 }
699 if (lttng_is_setuid_setgid()) {
700 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
701 "--kmod-probes");
702 } else {
703 config_string_set(&the_config.kmod_probes_list,
704 strdup(arg));
705 if (!the_config.kmod_probes_list.value) {
706 PERROR("strdup");
707 ret = -ENOMEM;
708 }
709 }
710 } else if (string_match(optname, "extra-kmod-probes")) {
711 if (!arg || *arg == '\0') {
712 ret = -EINVAL;
713 goto end;
714 }
715 if (lttng_is_setuid_setgid()) {
716 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
717 "--extra-kmod-probes");
718 } else {
719 config_string_set(&the_config.kmod_extra_probes_list,
720 strdup(arg));
721 if (!the_config.kmod_extra_probes_list.value) {
722 PERROR("strdup");
723 ret = -ENOMEM;
724 }
725 }
726 } else if (string_match(optname, EVENT_NOTIFIER_ERROR_BUFFER_SIZE_KERNEL_OPTION_STR)) {
727 unsigned long v;
728
729 errno = 0;
730 v = strtoul(arg, NULL, 0);
731 if (errno != 0 || !isdigit(arg[0])) {
732 ERR("Wrong value in --%s parameter: %s",
733 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_KERNEL_OPTION_STR, arg);
734 return -1;
735 }
736 if (v == 0 || v >= EVENT_NOTIFIER_ERROR_COUNTER_NUMBER_OF_BUCKET_MAX) {
737 ERR("Value out of range for --%s parameter: %s",
738 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_KERNEL_OPTION_STR, arg);
739 return -1;
740 }
741 the_config.event_notifier_buffer_size_kernel = (int) v;
742 DBG3("Number of event notifier error buffer kernel size to non default: %i",
743 the_config.event_notifier_buffer_size_kernel);
744 goto end;
745 } else if (string_match(optname, EVENT_NOTIFIER_ERROR_BUFFER_SIZE_USERSPACE_OPTION_STR)) {
746 unsigned long v;
747
748 errno = 0;
749 v = strtoul(arg, NULL, 0);
750 if (errno != 0 || !isdigit(arg[0])) {
751 ERR("Wrong value in --%s parameter: %s",
752 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_USERSPACE_OPTION_STR, arg);
753 return -1;
754 }
755 if (v == 0 || v >= EVENT_NOTIFIER_ERROR_COUNTER_NUMBER_OF_BUCKET_MAX) {
756 ERR("Value out of range for --%s parameter: %s",
757 EVENT_NOTIFIER_ERROR_BUFFER_SIZE_USERSPACE_OPTION_STR, arg);
758 return -1;
759 }
760 the_config.event_notifier_buffer_size_userspace = (int) v;
761 DBG3("Number of event notifier error buffer userspace size to non default: %i",
762 the_config.event_notifier_buffer_size_userspace);
763 goto end;
764 } else if (string_match(optname, "config") || opt == 'f') {
765 /* This is handled in set_options() thus silent skip. */
766 goto end;
767 } else {
768 /* Unknown option or other error.
769 * Error is printed by getopt, just return */
770 ret = -1;
771 }
772
773 end:
774 if (ret == -EINVAL) {
775 const char *opt_name = "unknown";
776 int i;
777
778 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
779 i++) {
780 if (opt == long_options[i].val) {
781 opt_name = long_options[i].name;
782 break;
783 }
784 }
785
786 WARN("Invalid argument provided for option \"%s\", using default value.",
787 opt_name);
788 }
789
790 return ret;
791 }
792
793 /*
794 * config_entry_handler_cb used to handle options read from a config file.
795 * See config_entry_handler_cb comment in common/config/session-config.h for the
796 * return value conventions.
797 */
798 static int config_entry_handler(const struct config_entry *entry, void *unused)
799 {
800 int ret = 0, i;
801
802 if (!entry || !entry->name || !entry->value) {
803 ret = -EINVAL;
804 goto end;
805 }
806
807 /* Check if the option is to be ignored */
808 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
809 if (!strcmp(entry->name, config_ignore_options[i])) {
810 goto end;
811 }
812 }
813
814 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
815 i++) {
816
817 /* Ignore if not fully matched. */
818 if (strcmp(entry->name, long_options[i].name)) {
819 continue;
820 }
821
822 /*
823 * If the option takes no argument on the command line, we have to
824 * check if the value is "true". We support non-zero numeric values,
825 * true, on and yes.
826 */
827 if (!long_options[i].has_arg) {
828 ret = config_parse_value(entry->value);
829 if (ret <= 0) {
830 if (ret) {
831 WARN("Invalid configuration value \"%s\" for option %s",
832 entry->value, entry->name);
833 }
834 /* False, skip boolean config option. */
835 goto end;
836 }
837 }
838
839 ret = set_option(long_options[i].val, entry->value, entry->name);
840 goto end;
841 }
842
843 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
844
845 end:
846 return ret;
847 }
848
849 static void print_version(void) {
850 fprintf(stdout, "%s\n", VERSION);
851 }
852
853 /*
854 * daemon configuration loading and argument parsing
855 */
856 static int set_options(int argc, char **argv)
857 {
858 int ret = 0, c = 0, option_index = 0;
859 int orig_optopt = optopt, orig_optind = optind;
860 char *optstring;
861 char *config_path = NULL;
862
863 optstring = utils_generate_optstring(long_options,
864 sizeof(long_options) / sizeof(struct option));
865 if (!optstring) {
866 ret = -ENOMEM;
867 goto end;
868 }
869
870 /* Check for the --config option */
871 while ((c = getopt_long(argc, argv, optstring, long_options,
872 &option_index)) != -1) {
873 if (c == '?') {
874 ret = -EINVAL;
875 goto end;
876 } else if (c != 'f') {
877 /* if not equal to --config option. */
878 continue;
879 }
880
881 if (lttng_is_setuid_setgid()) {
882 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
883 "-f, --config");
884 } else {
885 free(config_path);
886 config_path = utils_expand_path(optarg);
887 if (!config_path) {
888 ERR("Failed to resolve path: %s", optarg);
889 }
890 }
891 }
892
893 ret = config_get_section_entries(config_path, config_section_name,
894 config_entry_handler, NULL);
895 if (ret) {
896 if (ret > 0) {
897 ERR("Invalid configuration option at line %i", ret);
898 ret = -1;
899 }
900 goto end;
901 }
902
903 /* Reset getopt's global state */
904 optopt = orig_optopt;
905 optind = orig_optind;
906 while (1) {
907 option_index = -1;
908 /*
909 * getopt_long() will not set option_index if it encounters a
910 * short option.
911 */
912 c = getopt_long(argc, argv, optstring, long_options,
913 &option_index);
914 if (c == -1) {
915 break;
916 }
917
918 /*
919 * Pass NULL as the long option name if popt left the index
920 * unset.
921 */
922 ret = set_option(c, optarg,
923 option_index < 0 ? NULL :
924 long_options[option_index].name);
925 if (ret < 0) {
926 break;
927 }
928 }
929
930 end:
931 free(config_path);
932 free(optstring);
933 return ret;
934 }
935
936 /*
937 * Create lockfile using the rundir and return its fd.
938 */
939 static int create_lockfile(void)
940 {
941 return utils_create_lock_file(the_config.lock_file_path.value);
942 }
943
944 /*
945 * Check if the global socket is available, and if a daemon is answering at the
946 * other side. If yes, error is returned.
947 *
948 * Also attempts to create and hold the lock file.
949 */
950 static int check_existing_daemon(void)
951 {
952 int ret = 0;
953
954 /* Is there anybody out there ? */
955 if (lttng_session_daemon_alive()) {
956 ret = -EEXIST;
957 goto end;
958 }
959
960 lockfile_fd = create_lockfile();
961 if (lockfile_fd < 0) {
962 ret = -EEXIST;
963 goto end;
964 }
965 end:
966 return ret;
967 }
968
969 static void sessiond_cleanup_lock_file(void)
970 {
971 int ret;
972
973 /*
974 * Cleanup lock file by deleting it and finaly closing it which will
975 * release the file system lock.
976 */
977 if (lockfile_fd >= 0) {
978 ret = remove(the_config.lock_file_path.value);
979 if (ret < 0) {
980 PERROR("remove lock file");
981 }
982 ret = close(lockfile_fd);
983 if (ret < 0) {
984 PERROR("close lock file");
985 }
986 }
987 }
988
989 /*
990 * Set the tracing group gid onto the client socket.
991 *
992 * Race window between mkdir and chown is OK because we are going from more
993 * permissive (root.root) to less permissive (root.tracing).
994 */
995 static int set_permissions(char *rundir)
996 {
997 int ret;
998 gid_t gid;
999
1000 ret = utils_get_group_id(
1001 the_config.tracing_group_name.value, true, &gid);
1002 if (ret) {
1003 /* Default to root group. */
1004 gid = 0;
1005 }
1006
1007 /* Set lttng run dir */
1008 ret = chown(rundir, 0, gid);
1009 if (ret < 0) {
1010 ERR("Unable to set group on %s", rundir);
1011 PERROR("chown");
1012 }
1013
1014 /*
1015 * Ensure all applications and tracing group can search the run
1016 * dir. Allow everyone to read the directory, since it does not
1017 * buy us anything to hide its content.
1018 */
1019 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
1020 if (ret < 0) {
1021 ERR("Unable to set permissions on %s", rundir);
1022 PERROR("chmod");
1023 }
1024
1025 /* lttng client socket path */
1026 ret = chown(the_config.client_unix_sock_path.value, 0, gid);
1027 if (ret < 0) {
1028 ERR("Unable to set group on %s",
1029 the_config.client_unix_sock_path.value);
1030 PERROR("chown");
1031 }
1032
1033 /* kconsumer error socket path */
1034 ret = chown(the_kconsumer_data.err_unix_sock_path, 0, 0);
1035 if (ret < 0) {
1036 ERR("Unable to set group on %s",
1037 the_kconsumer_data.err_unix_sock_path);
1038 PERROR("chown");
1039 }
1040
1041 /* 64-bit ustconsumer error socket path */
1042 ret = chown(the_ustconsumer64_data.err_unix_sock_path, 0, 0);
1043 if (ret < 0) {
1044 ERR("Unable to set group on %s",
1045 the_ustconsumer64_data.err_unix_sock_path);
1046 PERROR("chown");
1047 }
1048
1049 /* 32-bit ustconsumer compat32 error socket path */
1050 ret = chown(the_ustconsumer32_data.err_unix_sock_path, 0, 0);
1051 if (ret < 0) {
1052 ERR("Unable to set group on %s",
1053 the_ustconsumer32_data.err_unix_sock_path);
1054 PERROR("chown");
1055 }
1056
1057 DBG("All permissions are set");
1058
1059 return ret;
1060 }
1061
1062 /*
1063 * Create the lttng run directory needed for all global sockets and pipe.
1064 */
1065 static int create_lttng_rundir(void)
1066 {
1067 int ret;
1068
1069 DBG3("Creating LTTng run directory: %s", the_config.rundir.value);
1070
1071 ret = mkdir(the_config.rundir.value, S_IRWXU);
1072 if (ret < 0) {
1073 if (errno != EEXIST) {
1074 ERR("Unable to create %s", the_config.rundir.value);
1075 goto error;
1076 } else {
1077 ret = 0;
1078 }
1079 }
1080
1081 error:
1082 return ret;
1083 }
1084
1085 /*
1086 * Setup sockets and directory needed by the consumerds' communication with the
1087 * session daemon.
1088 */
1089 static int set_consumer_sockets(struct consumer_data *consumer_data)
1090 {
1091 int ret;
1092 char *path = NULL;
1093
1094 switch (consumer_data->type) {
1095 case LTTNG_CONSUMER_KERNEL:
1096 path = the_config.kconsumerd_path.value;
1097 break;
1098 case LTTNG_CONSUMER64_UST:
1099 path = the_config.consumerd64_path.value;
1100 break;
1101 case LTTNG_CONSUMER32_UST:
1102 path = the_config.consumerd32_path.value;
1103 break;
1104 default:
1105 ERR("Consumer type unknown");
1106 ret = -EINVAL;
1107 goto error;
1108 }
1109 LTTNG_ASSERT(path);
1110
1111 DBG2("Creating consumer directory: %s", path);
1112
1113 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
1114 if (ret < 0 && errno != EEXIST) {
1115 PERROR("mkdir");
1116 ERR("Failed to create %s", path);
1117 goto error;
1118 }
1119 if (is_root) {
1120 gid_t gid;
1121
1122 ret = utils_get_group_id(the_config.tracing_group_name.value,
1123 true, &gid);
1124 if (ret) {
1125 /* Default to root group. */
1126 gid = 0;
1127 }
1128
1129 ret = chown(path, 0, gid);
1130 if (ret < 0) {
1131 ERR("Unable to set group on %s", path);
1132 PERROR("chown");
1133 goto error;
1134 }
1135 }
1136
1137 /* Create the consumerd error unix socket */
1138 consumer_data->err_sock =
1139 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
1140 if (consumer_data->err_sock < 0) {
1141 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
1142 ret = -1;
1143 goto error;
1144 }
1145
1146 /*
1147 * Set the CLOEXEC flag. Return code is useless because either way, the
1148 * show must go on.
1149 */
1150 ret = utils_set_fd_cloexec(consumer_data->err_sock);
1151 if (ret < 0) {
1152 PERROR("utils_set_fd_cloexec");
1153 /* continue anyway */
1154 }
1155
1156 /* File permission MUST be 660 */
1157 ret = chmod(consumer_data->err_unix_sock_path,
1158 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
1159 if (ret < 0) {
1160 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
1161 PERROR("chmod");
1162 goto error;
1163 }
1164
1165 error:
1166 return ret;
1167 }
1168
1169 /*
1170 * Signal handler for the daemon
1171 *
1172 * Simply stop all worker threads, leaving main() return gracefully after
1173 * joining all threads and calling cleanup().
1174 */
1175 static void sighandler(int sig, siginfo_t *siginfo, void *arg)
1176 {
1177 switch (sig) {
1178 case SIGINT:
1179 DBG("SIGINT caught");
1180 stop_threads();
1181 break;
1182 case SIGTERM:
1183 DBG("SIGTERM caught");
1184 stop_threads();
1185 break;
1186 case SIGUSR1:
1187 CMM_STORE_SHARED(recv_child_signal, 1);
1188 break;
1189 case SIGBUS:
1190 {
1191 int write_ret;
1192 const char msg[] = "Received SIGBUS, aborting program.\n";
1193
1194 lttng_ust_handle_sigbus(siginfo->si_addr);
1195 /*
1196 * If ustctl did not catch this signal (triggering a
1197 * siglongjmp), abort the program. Otherwise, the execution
1198 * will resume from the ust-ctl call which caused this error.
1199 *
1200 * The return value is ignored since the program aborts anyhow.
1201 */
1202 write_ret = write(STDERR_FILENO, msg, sizeof(msg));
1203 (void) write_ret;
1204 abort();
1205 }
1206 default:
1207 break;
1208 }
1209 }
1210
1211 /*
1212 * Setup signal handler for :
1213 * SIGINT, SIGTERM, SIGPIPE
1214 */
1215 static int set_signal_handler(void)
1216 {
1217 int ret = 0;
1218 struct sigaction sa;
1219 sigset_t sigset;
1220
1221 if ((ret = sigemptyset(&sigset)) < 0) {
1222 PERROR("sigemptyset");
1223 return ret;
1224 }
1225
1226 sa.sa_mask = sigset;
1227 sa.sa_flags = SA_SIGINFO;
1228
1229 sa.sa_sigaction = sighandler;
1230 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
1231 PERROR("sigaction");
1232 return ret;
1233 }
1234
1235 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
1236 PERROR("sigaction");
1237 return ret;
1238 }
1239
1240 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
1241 PERROR("sigaction");
1242 return ret;
1243 }
1244
1245 if ((ret = sigaction(SIGBUS, &sa, NULL)) < 0) {
1246 PERROR("sigaction");
1247 return ret;
1248 }
1249
1250 sa.sa_flags = 0;
1251 sa.sa_handler = SIG_IGN;
1252 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
1253 PERROR("sigaction");
1254 return ret;
1255 }
1256
1257 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE, SIGINT, and SIGBUS");
1258
1259 return ret;
1260 }
1261
1262 /*
1263 * Set open files limit to unlimited. This daemon can open a large number of
1264 * file descriptors in order to consume multiple kernel traces.
1265 */
1266 static void set_ulimit(void)
1267 {
1268 int ret;
1269 struct rlimit lim;
1270
1271 /* The kernel does not allow an infinite limit for open files */
1272 lim.rlim_cur = 65535;
1273 lim.rlim_max = 65535;
1274
1275 ret = setrlimit(RLIMIT_NOFILE, &lim);
1276 if (ret < 0) {
1277 PERROR("failed to set open files limit");
1278 }
1279 }
1280
1281 static int write_pidfile(void)
1282 {
1283 return utils_create_pid_file(getpid(), the_config.pid_file_path.value);
1284 }
1285
1286 static int set_clock_plugin_env(void)
1287 {
1288 int ret = 0;
1289 char *env_value = NULL;
1290
1291 if (!the_config.lttng_ust_clock_plugin.value) {
1292 goto end;
1293 }
1294
1295 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
1296 the_config.lttng_ust_clock_plugin.value);
1297 if (ret < 0) {
1298 PERROR("asprintf");
1299 goto end;
1300 }
1301
1302 ret = putenv(env_value);
1303 if (ret) {
1304 free(env_value);
1305 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
1306 goto end;
1307 }
1308
1309 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
1310 the_config.lttng_ust_clock_plugin.value);
1311 end:
1312 return ret;
1313 }
1314
1315 static void destroy_all_sessions_and_wait(void)
1316 {
1317 struct ltt_session *session, *tmp;
1318 struct ltt_session_list *session_list;
1319
1320 session_list = session_get_list();
1321 DBG("Initiating destruction of all sessions");
1322
1323 if (!session_list) {
1324 return;
1325 }
1326
1327 session_lock_list();
1328 /* Initiate the destruction of all sessions. */
1329 cds_list_for_each_entry_safe(session, tmp,
1330 &session_list->head, list) {
1331 if (!session_get(session)) {
1332 continue;
1333 }
1334
1335 session_lock(session);
1336 if (session->destroyed) {
1337 goto unlock_session;
1338 }
1339 (void) cmd_stop_trace(session);
1340 (void) cmd_destroy_session(
1341 session, the_notification_thread_handle, NULL);
1342 unlock_session:
1343 session_unlock(session);
1344 session_put(session);
1345 }
1346 session_unlock_list();
1347
1348 /* Wait for the destruction of all sessions to complete. */
1349 DBG("Waiting for the destruction of all sessions to complete");
1350 session_list_wait_empty();
1351 DBG("Destruction of all sessions completed");
1352 }
1353
1354 static void unregister_all_triggers(void)
1355 {
1356 enum lttng_error_code ret_code;
1357 enum lttng_trigger_status trigger_status;
1358 struct lttng_triggers *triggers = NULL;
1359 unsigned int trigger_count, i;
1360 const struct lttng_credentials creds = {
1361 .uid = LTTNG_OPTIONAL_INIT_VALUE(0),
1362 };
1363
1364 DBG("Unregistering all triggers");
1365
1366 /*
1367 * List all triggers as "root" since we wish to unregister all triggers.
1368 */
1369 ret_code = notification_thread_command_list_triggers(
1370 the_notification_thread_handle, creds.uid.value,
1371 &triggers);
1372 if (ret_code != LTTNG_OK) {
1373 ERR("Failed to list triggers while unregistering all triggers");
1374 goto end;
1375 }
1376
1377 trigger_status = lttng_triggers_get_count(triggers, &trigger_count);
1378 LTTNG_ASSERT(trigger_status == LTTNG_TRIGGER_STATUS_OK);
1379
1380 for (i = 0; i < trigger_count; i++) {
1381 uid_t trigger_owner;
1382 const char *trigger_name;
1383 const struct lttng_trigger *trigger =
1384 lttng_triggers_get_at_index(triggers, i);
1385
1386 LTTNG_ASSERT(trigger);
1387
1388 trigger_status = lttng_trigger_get_owner_uid(
1389 trigger, &trigger_owner);
1390 LTTNG_ASSERT(trigger_status == LTTNG_TRIGGER_STATUS_OK);
1391
1392 trigger_status = lttng_trigger_get_name(trigger, &trigger_name);
1393 trigger_name = trigger_status == LTTNG_TRIGGER_STATUS_OK ?
1394 trigger_name : "(anonymous)";
1395
1396 DBG("Unregistering trigger: trigger owner uid = %d, trigger name = '%s'",
1397 (int) trigger_owner, trigger_name);
1398
1399 ret_code = cmd_unregister_trigger(&creds, trigger,
1400 the_notification_thread_handle);
1401 if (ret_code != LTTNG_OK) {
1402 ERR("Failed to unregister trigger: trigger owner uid = %d, trigger name = '%s', error: '%s'",
1403 (int) trigger_owner, trigger_name,
1404 lttng_strerror(-ret_code));
1405 /* Continue to unregister the remaining triggers. */
1406 }
1407 }
1408 end:
1409 lttng_triggers_destroy(triggers);
1410 }
1411
1412 static int run_as_worker_post_fork_cleanup(void *data)
1413 {
1414 struct sessiond_config *sessiond_config = (struct sessiond_config *) data;
1415
1416 sessiond_config_fini(sessiond_config);
1417 return 0;
1418 }
1419
1420 static int launch_run_as_worker(const char *procname)
1421 {
1422 /*
1423 * Clean-up before forking the run-as worker. Any dynamically
1424 * allocated memory of which the worker is not aware will
1425 * be leaked as the process forks a run-as worker (and performs
1426 * no exec*()). The same would apply to any opened fd.
1427 */
1428 return run_as_create_worker(
1429 procname, run_as_worker_post_fork_cleanup, &the_config);
1430 }
1431
1432 static void sessiond_uuid_log(void)
1433 {
1434 char uuid_str[LTTNG_UUID_STR_LEN];
1435
1436 lttng_uuid_to_str(the_sessiond_uuid, uuid_str);
1437 DBG("Starting lttng-sessiond {%s}", uuid_str);
1438 }
1439
1440 /*
1441 * main
1442 */
1443 int main(int argc, char **argv)
1444 {
1445 int ret = 0, retval = 0;
1446 const char *env_app_timeout;
1447 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
1448 *ust64_channel_monitor_pipe = NULL,
1449 *kernel_channel_monitor_pipe = NULL;
1450 struct timer_thread_parameters timer_thread_parameters;
1451 /* Rotation thread handle. */
1452 struct rotation_thread_handle *rotation_thread_handle = NULL;
1453 /* Queue of rotation jobs populated by the sessiond-timer. */
1454 struct rotation_thread_timer_queue *rotation_timer_queue = NULL;
1455 struct lttng_thread *client_thread = NULL;
1456 struct lttng_thread *notification_thread = NULL;
1457 struct lttng_thread *register_apps_thread = NULL;
1458 enum event_notifier_error_accounting_status event_notifier_error_accounting_status;
1459
1460 logger_set_thread_name("Main", false);
1461 init_kernel_workarounds();
1462
1463 rcu_register_thread();
1464
1465 if (set_signal_handler()) {
1466 retval = -1;
1467 goto exit_set_signal_handler;
1468 }
1469
1470 if (timer_signal_init()) {
1471 retval = -1;
1472 goto exit_set_signal_handler;
1473 }
1474
1475 the_page_size = sysconf(_SC_PAGE_SIZE);
1476 if (the_page_size < 0) {
1477 PERROR("sysconf _SC_PAGE_SIZE");
1478 the_page_size = LONG_MAX;
1479 WARN("Fallback page size to %ld", the_page_size);
1480 }
1481
1482 ret = sessiond_config_init(&the_config);
1483 if (ret) {
1484 retval = -1;
1485 goto exit_set_signal_handler;
1486 }
1487
1488 /*
1489 * Init config from environment variables.
1490 * Command line option override env configuration per-doc. Do env first.
1491 */
1492 sessiond_config_apply_env_config(&the_config);
1493
1494 /*
1495 * Parse arguments and load the daemon configuration file.
1496 *
1497 * We have an exit_options exit path to free memory reserved by
1498 * set_options.
1499 */
1500 progname = argv[0];
1501 if (set_options(argc, argv)) {
1502 retval = -1;
1503 goto exit_options;
1504 }
1505
1506 /*
1507 * Resolve all paths received as arguments, configuration option, or
1508 * through environment variable as absolute paths. This is necessary
1509 * since daemonizing causes the sessiond's current working directory
1510 * to '/'.
1511 */
1512 ret = sessiond_config_resolve_paths(&the_config);
1513 if (ret) {
1514 goto exit_options;
1515 }
1516
1517 /* Apply config. */
1518 lttng_opt_verbose = the_config.verbose;
1519 lttng_opt_quiet = the_config.quiet;
1520 the_kconsumer_data.err_unix_sock_path =
1521 the_config.kconsumerd_err_unix_sock_path.value;
1522 the_kconsumer_data.cmd_unix_sock_path =
1523 the_config.kconsumerd_cmd_unix_sock_path.value;
1524 the_ustconsumer32_data.err_unix_sock_path =
1525 the_config.consumerd32_err_unix_sock_path.value;
1526 the_ustconsumer32_data.cmd_unix_sock_path =
1527 the_config.consumerd32_cmd_unix_sock_path.value;
1528 the_ustconsumer64_data.err_unix_sock_path =
1529 the_config.consumerd64_err_unix_sock_path.value;
1530 the_ustconsumer64_data.cmd_unix_sock_path =
1531 the_config.consumerd64_cmd_unix_sock_path.value;
1532 set_clock_plugin_env();
1533
1534 sessiond_config_log(&the_config);
1535 sessiond_uuid_log();
1536
1537 if (opt_print_version) {
1538 print_version();
1539 retval = 0;
1540 goto exit_options;
1541 }
1542
1543 if (create_lttng_rundir()) {
1544 retval = -1;
1545 goto exit_options;
1546 }
1547
1548 /* Abort launch if a session daemon is already running. */
1549 if (check_existing_daemon()) {
1550 ERR("A session daemon is already running.");
1551 retval = -1;
1552 goto exit_options;
1553 }
1554
1555 /* Daemonize */
1556 if (the_config.daemonize || the_config.background) {
1557 int i;
1558
1559 ret = lttng_daemonize(&the_child_ppid, &recv_child_signal,
1560 !the_config.background);
1561 if (ret < 0) {
1562 retval = -1;
1563 goto exit_options;
1564 }
1565
1566 /*
1567 * We are in the child. Make sure all other file descriptors are
1568 * closed, in case we are called with more opened file
1569 * descriptors than the standard ones and the lock file.
1570 */
1571 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
1572 if (i == lockfile_fd) {
1573 continue;
1574 }
1575 (void) close(i);
1576 }
1577 }
1578
1579 if (launch_run_as_worker(argv[0]) < 0) {
1580 goto exit_create_run_as_worker_cleanup;
1581 }
1582
1583 /*
1584 * Starting from here, we can create threads. This needs to be after
1585 * lttng_daemonize due to RCU.
1586 */
1587
1588 /*
1589 * Initialize the health check subsystem. This call should set the
1590 * appropriate time values.
1591 */
1592 the_health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
1593 if (!the_health_sessiond) {
1594 PERROR("health_app_create error");
1595 retval = -1;
1596 goto stop_threads;
1597 }
1598
1599 /* Create thread quit pipe */
1600 if (sessiond_init_thread_quit_pipe()) {
1601 retval = -1;
1602 goto stop_threads;
1603 }
1604
1605 /* Check if daemon is UID = 0 */
1606 is_root = !getuid();
1607 if (is_root) {
1608 /* Create global run dir with root access */
1609
1610 kernel_channel_monitor_pipe = lttng_pipe_open(0);
1611 if (!kernel_channel_monitor_pipe) {
1612 ERR("Failed to create kernel consumer channel monitor pipe");
1613 retval = -1;
1614 goto stop_threads;
1615 }
1616 the_kconsumer_data.channel_monitor_pipe =
1617 lttng_pipe_release_writefd(
1618 kernel_channel_monitor_pipe);
1619 if (the_kconsumer_data.channel_monitor_pipe < 0) {
1620 retval = -1;
1621 goto stop_threads;
1622 }
1623 }
1624
1625 /* Set consumer initial state */
1626 the_kernel_consumerd_state = CONSUMER_STOPPED;
1627 the_ust_consumerd_state = CONSUMER_STOPPED;
1628
1629 ust32_channel_monitor_pipe = lttng_pipe_open(0);
1630 if (!ust32_channel_monitor_pipe) {
1631 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
1632 retval = -1;
1633 goto stop_threads;
1634 }
1635 the_ustconsumer32_data.channel_monitor_pipe =
1636 lttng_pipe_release_writefd(ust32_channel_monitor_pipe);
1637 if (the_ustconsumer32_data.channel_monitor_pipe < 0) {
1638 retval = -1;
1639 goto stop_threads;
1640 }
1641
1642 /*
1643 * The rotation_thread_timer_queue structure is shared between the
1644 * sessiond timer thread and the rotation thread. The main thread keeps
1645 * its ownership and destroys it when both threads have been joined.
1646 */
1647 rotation_timer_queue = rotation_thread_timer_queue_create();
1648 if (!rotation_timer_queue) {
1649 retval = -1;
1650 goto stop_threads;
1651 }
1652 timer_thread_parameters.rotation_thread_job_queue =
1653 rotation_timer_queue;
1654
1655 ust64_channel_monitor_pipe = lttng_pipe_open(0);
1656 if (!ust64_channel_monitor_pipe) {
1657 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
1658 retval = -1;
1659 goto stop_threads;
1660 }
1661 the_ustconsumer64_data.channel_monitor_pipe =
1662 lttng_pipe_release_writefd(ust64_channel_monitor_pipe);
1663 if (the_ustconsumer64_data.channel_monitor_pipe < 0) {
1664 retval = -1;
1665 goto stop_threads;
1666 }
1667
1668 /*
1669 * Init UST app hash table. Alloc hash table before this point since
1670 * cleanup() can get called after that point.
1671 */
1672 if (ust_app_ht_alloc()) {
1673 ERR("Failed to allocate UST app hash table");
1674 retval = -1;
1675 goto stop_threads;
1676 }
1677
1678 event_notifier_error_accounting_status = event_notifier_error_accounting_init(
1679 the_config.event_notifier_buffer_size_kernel,
1680 the_config.event_notifier_buffer_size_userspace);
1681 if (event_notifier_error_accounting_status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) {
1682 ERR("Failed to initialize event notifier error accounting system");
1683 retval = -1;
1684 goto stop_threads;
1685 }
1686
1687 /*
1688 * Initialize agent app hash table. We allocate the hash table here
1689 * since cleanup() can get called after this point.
1690 */
1691 if (agent_app_ht_alloc()) {
1692 ERR("Failed to allocate Agent app hash table");
1693 retval = -1;
1694 goto stop_threads;
1695 }
1696
1697 if (agent_by_event_notifier_domain_ht_create()) {
1698 ERR("Failed to allocate per-event notifier domain agent hash table");
1699 retval = -1;
1700 goto stop_threads;
1701 }
1702 /*
1703 * These actions must be executed as root. We do that *after* setting up
1704 * the sockets path because we MUST make the check for another daemon using
1705 * those paths *before* trying to set the kernel consumer sockets and init
1706 * kernel tracer.
1707 */
1708 if (is_root) {
1709 if (set_consumer_sockets(&the_kconsumer_data)) {
1710 retval = -1;
1711 goto stop_threads;
1712 }
1713
1714 /* Setup kernel tracer */
1715 if (!the_config.no_kernel) {
1716 init_kernel_tracer();
1717 }
1718
1719 /* Set ulimit for open files */
1720 set_ulimit();
1721 }
1722 /* init lttng_fd tracking must be done after set_ulimit. */
1723 lttng_fd_init();
1724
1725 if (set_consumer_sockets(&the_ustconsumer64_data)) {
1726 retval = -1;
1727 goto stop_threads;
1728 }
1729
1730 if (set_consumer_sockets(&the_ustconsumer32_data)) {
1731 retval = -1;
1732 goto stop_threads;
1733 }
1734
1735 /* Get parent pid if -S, --sig-parent is specified. */
1736 if (the_config.sig_parent) {
1737 the_ppid = getppid();
1738 }
1739
1740 /* Setup the kernel pipe for waking up the kernel thread */
1741 if (is_root && !the_config.no_kernel) {
1742 if (utils_create_pipe_cloexec(the_kernel_poll_pipe)) {
1743 retval = -1;
1744 goto stop_threads;
1745 }
1746 }
1747
1748 /* Setup the thread apps communication pipe. */
1749 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
1750 retval = -1;
1751 goto stop_threads;
1752 }
1753
1754 /* Setup the thread apps notify communication pipe. */
1755 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
1756 retval = -1;
1757 goto stop_threads;
1758 }
1759
1760 /* Initialize global buffer per UID and PID registry. */
1761 buffer_reg_init_uid_registry();
1762 buffer_reg_init_pid_registry();
1763
1764 /* Init UST command queue. */
1765 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1766
1767 cmd_init();
1768
1769 /* Check for the application socket timeout env variable. */
1770 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
1771 if (env_app_timeout) {
1772 the_config.app_socket_timeout = atoi(env_app_timeout);
1773 } else {
1774 the_config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
1775 }
1776
1777 ret = write_pidfile();
1778 if (ret) {
1779 ERR("Error in write_pidfile");
1780 retval = -1;
1781 goto stop_threads;
1782 }
1783
1784 /* Initialize communication library */
1785 lttcomm_init();
1786 /* Initialize TCP timeout values */
1787 lttcomm_inet_init();
1788
1789 /* Create health-check thread. */
1790 if (!launch_health_management_thread()) {
1791 retval = -1;
1792 goto stop_threads;
1793 }
1794
1795 /* notification_thread_data acquires the pipes' read side. */
1796 the_notification_thread_handle = notification_thread_handle_create(
1797 ust32_channel_monitor_pipe, ust64_channel_monitor_pipe,
1798 kernel_channel_monitor_pipe);
1799 if (!the_notification_thread_handle) {
1800 retval = -1;
1801 ERR("Failed to create notification thread shared data");
1802 goto stop_threads;
1803 }
1804
1805 /* Create notification thread. */
1806 notification_thread = launch_notification_thread(
1807 the_notification_thread_handle);
1808 if (!notification_thread) {
1809 retval = -1;
1810 goto stop_threads;
1811 }
1812
1813 /* Create timer thread. */
1814 if (!launch_timer_thread(&timer_thread_parameters)) {
1815 retval = -1;
1816 goto stop_threads;
1817 }
1818
1819 /* rotation_thread_data acquires the pipes' read side. */
1820 rotation_thread_handle = rotation_thread_handle_create(
1821 rotation_timer_queue, the_notification_thread_handle);
1822 if (!rotation_thread_handle) {
1823 retval = -1;
1824 ERR("Failed to create rotation thread shared data");
1825 stop_threads();
1826 goto stop_threads;
1827 }
1828
1829 /* Create rotation thread. */
1830 if (!launch_rotation_thread(rotation_thread_handle)) {
1831 retval = -1;
1832 goto stop_threads;
1833 }
1834
1835 /* Create thread to manage the client socket */
1836 client_thread = launch_client_thread();
1837 if (!client_thread) {
1838 retval = -1;
1839 goto stop_threads;
1840 }
1841
1842 /* Set credentials of the client socket and rundir */
1843 if (is_root && set_permissions(the_config.rundir.value)) {
1844 retval = -1;
1845 goto stop_threads;
1846 }
1847
1848 if (!launch_ust_dispatch_thread(&ust_cmd_queue, apps_cmd_pipe[1],
1849 apps_cmd_notify_pipe[1])) {
1850 retval = -1;
1851 goto stop_threads;
1852 }
1853
1854 /* Create thread to manage application registration. */
1855 register_apps_thread = launch_application_registration_thread(
1856 &ust_cmd_queue);
1857 if (!register_apps_thread) {
1858 retval = -1;
1859 goto stop_threads;
1860 }
1861
1862 /* Create thread to manage application socket */
1863 if (!launch_application_management_thread(apps_cmd_pipe[0])) {
1864 retval = -1;
1865 goto stop_threads;
1866 }
1867
1868 /* Create thread to manage application notify socket */
1869 if (!launch_application_notification_thread(apps_cmd_notify_pipe[0])) {
1870 retval = -1;
1871 goto stop_threads;
1872 }
1873
1874 /* Create agent management thread. */
1875 if (!launch_agent_management_thread()) {
1876 retval = -1;
1877 goto stop_threads;
1878 }
1879
1880 /* Don't start this thread if kernel tracing is not requested nor root */
1881 if (is_root && !the_config.no_kernel) {
1882 /* Create kernel thread to manage kernel event */
1883 if (!launch_kernel_management_thread(the_kernel_poll_pipe[0])) {
1884 retval = -1;
1885 goto stop_threads;
1886 }
1887
1888 if (kernel_get_notification_fd() >= 0) {
1889 ret = notification_thread_command_add_tracer_event_source(
1890 the_notification_thread_handle,
1891 kernel_get_notification_fd(),
1892 LTTNG_DOMAIN_KERNEL);
1893 if (ret != LTTNG_OK) {
1894 ERR("Failed to add kernel trigger event source to notification thread");
1895 retval = -1;
1896 goto stop_threads;
1897 }
1898 }
1899 }
1900
1901 /* Load sessions. */
1902 ret = config_load_session(
1903 the_config.load_session_path.value, NULL, 1, 1, NULL);
1904 if (ret) {
1905 ERR("Session load failed: %s", error_get_str(ret));
1906 retval = -1;
1907 goto stop_threads;
1908 }
1909
1910 /* Initialization completed. */
1911 sessiond_signal_parents();
1912
1913 /*
1914 * This is where we start awaiting program completion (e.g. through
1915 * signal that asks threads to teardown).
1916 */
1917
1918 /* Initiate teardown once activity occurs on the quit pipe. */
1919 sessiond_wait_for_quit_pipe(-1);
1920
1921 stop_threads:
1922
1923 /*
1924 * Ensure that the client thread is no longer accepting new commands,
1925 * which could cause new sessions to be created.
1926 */
1927 if (client_thread) {
1928 lttng_thread_shutdown(client_thread);
1929 lttng_thread_put(client_thread);
1930 }
1931
1932 destroy_all_sessions_and_wait();
1933
1934 /*
1935 * At this point no new trigger can be registered (no sessions are
1936 * running/rotating) and clients can't connect to the session daemon
1937 * anymore. Unregister all triggers.
1938 */
1939 unregister_all_triggers();
1940
1941 if (register_apps_thread) {
1942 lttng_thread_shutdown(register_apps_thread);
1943 lttng_thread_put(register_apps_thread);
1944 }
1945 lttng_thread_list_shutdown_orphans();
1946
1947 /*
1948 * Wait for all pending call_rcu work to complete before tearing
1949 * down data structures. call_rcu worker may be trying to
1950 * perform lookups in those structures.
1951 */
1952 rcu_barrier();
1953
1954 rcu_thread_online();
1955 sessiond_cleanup();
1956
1957 /*
1958 * Wait for all pending call_rcu work to complete before shutting down
1959 * the notification thread. This call_rcu work includes shutting down
1960 * UST apps and event notifier pipes.
1961 */
1962 rcu_barrier();
1963
1964 if (notification_thread) {
1965 lttng_thread_shutdown(notification_thread);
1966 lttng_thread_put(notification_thread);
1967 }
1968
1969 /*
1970 * Error accounting teardown has to be done after the teardown of all
1971 * event notifier pipes to ensure that no tracer may try to use the
1972 * error accounting facilities.
1973 */
1974 event_notifier_error_accounting_fini();
1975
1976 /*
1977 * Unloading the kernel modules needs to be done after all kernel
1978 * ressources have been released. In our case, this includes the
1979 * notification fd, the event notifier group fd, error accounting fd,
1980 * all event and event notifier fds, etc.
1981 *
1982 * In short, at this point, we need to have called close() on all fds
1983 * received from the kernel tracer.
1984 */
1985 if (is_root && !the_config.no_kernel) {
1986 DBG("Unloading kernel modules");
1987 modprobe_remove_lttng_all();
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 0.071134 seconds and 4 git commands to generate.