Fix: sessiond: double socket close on allocation failure
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <paths.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <inttypes.h>
31 #include <sys/mman.h>
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37 #include <sys/wait.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40 #include <ctype.h>
41
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/compat/getenv.h>
45 #include <common/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/session-config.h>
52 #include <common/dynamic-buffer.h>
53 #include <lttng/event-internal.h>
54
55 #include "lttng-sessiond.h"
56 #include "buffer-registry.h"
57 #include "channel.h"
58 #include "cmd.h"
59 #include "consumer.h"
60 #include "context.h"
61 #include "event.h"
62 #include "kernel.h"
63 #include "kernel-consumer.h"
64 #include "shm.h"
65 #include "ust-ctl.h"
66 #include "ust-consumer.h"
67 #include "utils.h"
68 #include "fd-limit.h"
69 #include "health-sessiond.h"
70 #include "testpoint.h"
71 #include "notify-apps.h"
72 #include "agent-thread.h"
73 #include "save.h"
74 #include "notification-thread.h"
75 #include "notification-thread-commands.h"
76 #include "rotation-thread.h"
77 #include "agent.h"
78 #include "ht-cleanup.h"
79 #include "sessiond-config.h"
80 #include "timer.h"
81 #include "thread.h"
82 #include "client.h"
83 #include "dispatch.h"
84 #include "register.h"
85 #include "manage-apps.h"
86 #include "manage-kernel.h"
87
88 static const char *help_msg =
89 #ifdef LTTNG_EMBED_HELP
90 #include <lttng-sessiond.8.h>
91 #else
92 NULL
93 #endif
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 { NULL, 0, 0, 0 }
134 };
135
136 /* Command line options to ignore from configuration file */
137 static const char *config_ignore_options[] = { "help", "version", "config" };
138
139 /*
140 * This pipe is used to inform the thread managing application communication
141 * that a command is queued and ready to be processed.
142 */
143 static int apps_cmd_pipe[2] = { -1, -1 };
144 static int apps_cmd_notify_pipe[2] = { -1, -1 };
145
146 /*
147 * UST registration command queue. This queue is tied with a futex and uses a N
148 * wakers / 1 waiter implemented and detailed in futex.c/.h
149 *
150 * The thread_registration_apps and thread_dispatch_ust_registration uses this
151 * queue along with the wait/wake scheme. The thread_manage_apps receives down
152 * the line new application socket and monitors it for any I/O error or clean
153 * close that triggers an unregistration of the application.
154 */
155 static struct ust_cmd_queue ust_cmd_queue;
156
157 /*
158 * Section name to look for in the daemon configuration file.
159 */
160 static const char * const config_section_name = "sessiond";
161
162 /* Am I root or not. Set to 1 if the daemon is running as root */
163 static int is_root;
164
165 /*
166 * Stop all threads by closing the thread quit pipe.
167 */
168 static void stop_threads(void)
169 {
170 int ret;
171
172 /* Stopping all threads */
173 DBG("Terminating all threads");
174 ret = sessiond_notify_quit_pipe();
175 if (ret < 0) {
176 ERR("write error on thread quit pipe");
177 }
178 }
179
180 /*
181 * Close every consumer sockets.
182 */
183 static void close_consumer_sockets(void)
184 {
185 int ret;
186
187 if (kconsumer_data.err_sock >= 0) {
188 ret = close(kconsumer_data.err_sock);
189 if (ret < 0) {
190 PERROR("kernel consumer err_sock close");
191 }
192 }
193 if (ustconsumer32_data.err_sock >= 0) {
194 ret = close(ustconsumer32_data.err_sock);
195 if (ret < 0) {
196 PERROR("UST consumerd32 err_sock close");
197 }
198 }
199 if (ustconsumer64_data.err_sock >= 0) {
200 ret = close(ustconsumer64_data.err_sock);
201 if (ret < 0) {
202 PERROR("UST consumerd64 err_sock close");
203 }
204 }
205 if (kconsumer_data.cmd_sock >= 0) {
206 ret = close(kconsumer_data.cmd_sock);
207 if (ret < 0) {
208 PERROR("kernel consumer cmd_sock close");
209 }
210 }
211 if (ustconsumer32_data.cmd_sock >= 0) {
212 ret = close(ustconsumer32_data.cmd_sock);
213 if (ret < 0) {
214 PERROR("UST consumerd32 cmd_sock close");
215 }
216 }
217 if (ustconsumer64_data.cmd_sock >= 0) {
218 ret = close(ustconsumer64_data.cmd_sock);
219 if (ret < 0) {
220 PERROR("UST consumerd64 cmd_sock close");
221 }
222 }
223 if (kconsumer_data.channel_monitor_pipe >= 0) {
224 ret = close(kconsumer_data.channel_monitor_pipe);
225 if (ret < 0) {
226 PERROR("kernel consumer channel monitor pipe close");
227 }
228 }
229 if (ustconsumer32_data.channel_monitor_pipe >= 0) {
230 ret = close(ustconsumer32_data.channel_monitor_pipe);
231 if (ret < 0) {
232 PERROR("UST consumerd32 channel monitor pipe close");
233 }
234 }
235 if (ustconsumer64_data.channel_monitor_pipe >= 0) {
236 ret = close(ustconsumer64_data.channel_monitor_pipe);
237 if (ret < 0) {
238 PERROR("UST consumerd64 channel monitor pipe close");
239 }
240 }
241 }
242
243 /*
244 * Wait on consumer process termination.
245 *
246 * Need to be called with the consumer data lock held or from a context
247 * ensuring no concurrent access to data (e.g: cleanup).
248 */
249 static void wait_consumer(struct consumer_data *consumer_data)
250 {
251 pid_t ret;
252 int status;
253
254 if (consumer_data->pid <= 0) {
255 return;
256 }
257
258 DBG("Waiting for complete teardown of consumerd (PID: %d)",
259 consumer_data->pid);
260 ret = waitpid(consumer_data->pid, &status, 0);
261 if (ret == -1) {
262 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
263 } else if (!WIFEXITED(status)) {
264 ERR("consumerd termination with error: %d",
265 WEXITSTATUS(ret));
266 }
267 consumer_data->pid = 0;
268 }
269
270 /*
271 * Cleanup the session daemon's data structures.
272 */
273 static void sessiond_cleanup(void)
274 {
275 int ret;
276 struct ltt_session_list *session_list = session_get_list();
277
278 DBG("Cleanup sessiond");
279
280 /*
281 * Close the thread quit pipe. It has already done its job,
282 * since we are now called.
283 */
284 sessiond_close_quit_pipe();
285 utils_close_pipe(apps_cmd_pipe);
286 utils_close_pipe(apps_cmd_notify_pipe);
287 utils_close_pipe(kernel_poll_pipe);
288
289 ret = remove(config.pid_file_path.value);
290 if (ret < 0) {
291 PERROR("remove pidfile %s", config.pid_file_path.value);
292 }
293
294 DBG("Removing sessiond and consumerd content of directory %s",
295 config.rundir.value);
296
297 /* sessiond */
298 DBG("Removing %s", config.pid_file_path.value);
299 (void) unlink(config.pid_file_path.value);
300
301 DBG("Removing %s", config.agent_port_file_path.value);
302 (void) unlink(config.agent_port_file_path.value);
303
304 /* kconsumerd */
305 DBG("Removing %s", kconsumer_data.err_unix_sock_path);
306 (void) unlink(kconsumer_data.err_unix_sock_path);
307
308 DBG("Removing directory %s", config.kconsumerd_path.value);
309 (void) rmdir(config.kconsumerd_path.value);
310
311 /* ust consumerd 32 */
312 DBG("Removing %s", config.consumerd32_err_unix_sock_path.value);
313 (void) unlink(config.consumerd32_err_unix_sock_path.value);
314
315 DBG("Removing directory %s", config.consumerd32_path.value);
316 (void) rmdir(config.consumerd32_path.value);
317
318 /* ust consumerd 64 */
319 DBG("Removing %s", config.consumerd64_err_unix_sock_path.value);
320 (void) unlink(config.consumerd64_err_unix_sock_path.value);
321
322 DBG("Removing directory %s", config.consumerd64_path.value);
323 (void) rmdir(config.consumerd64_path.value);
324
325 pthread_mutex_destroy(&session_list->lock);
326
327 DBG("Cleaning up all agent apps");
328 agent_app_ht_clean();
329 DBG("Closing all UST sockets");
330 ust_app_clean_list();
331 buffer_reg_destroy_registries();
332
333 close_consumer_sockets();
334
335 wait_consumer(&kconsumer_data);
336 wait_consumer(&ustconsumer64_data);
337 wait_consumer(&ustconsumer32_data);
338
339 if (is_root && !config.no_kernel) {
340 cleanup_kernel_tracer();
341 }
342
343 /*
344 * We do NOT rmdir rundir because there are other processes
345 * using it, for instance lttng-relayd, which can start in
346 * parallel with this teardown.
347 */
348 }
349
350 /*
351 * Cleanup the daemon's option data structures.
352 */
353 static void sessiond_cleanup_options(void)
354 {
355 DBG("Cleaning up options");
356
357 sessiond_config_fini(&config);
358
359 run_as_destroy_worker();
360 }
361
362 static int string_match(const char *str1, const char *str2)
363 {
364 return (str1 && str2) && !strcmp(str1, str2);
365 }
366
367 /*
368 * Take an option from the getopt output and set it in the right variable to be
369 * used later.
370 *
371 * Return 0 on success else a negative value.
372 */
373 static int set_option(int opt, const char *arg, const char *optname)
374 {
375 int ret = 0;
376
377 if (string_match(optname, "client-sock") || opt == 'c') {
378 if (!arg || *arg == '\0') {
379 ret = -EINVAL;
380 goto end;
381 }
382 if (lttng_is_setuid_setgid()) {
383 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
384 "-c, --client-sock");
385 } else {
386 config_string_set(&config.client_unix_sock_path,
387 strdup(arg));
388 if (!config.client_unix_sock_path.value) {
389 ret = -ENOMEM;
390 PERROR("strdup");
391 }
392 }
393 } else if (string_match(optname, "apps-sock") || opt == 'a') {
394 if (!arg || *arg == '\0') {
395 ret = -EINVAL;
396 goto end;
397 }
398 if (lttng_is_setuid_setgid()) {
399 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
400 "-a, --apps-sock");
401 } else {
402 config_string_set(&config.apps_unix_sock_path,
403 strdup(arg));
404 if (!config.apps_unix_sock_path.value) {
405 ret = -ENOMEM;
406 PERROR("strdup");
407 }
408 }
409 } else if (string_match(optname, "daemonize") || opt == 'd') {
410 config.daemonize = true;
411 } else if (string_match(optname, "background") || opt == 'b') {
412 config.background = true;
413 } else if (string_match(optname, "group") || opt == 'g') {
414 if (!arg || *arg == '\0') {
415 ret = -EINVAL;
416 goto end;
417 }
418 if (lttng_is_setuid_setgid()) {
419 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
420 "-g, --group");
421 } else {
422 config_string_set(&config.tracing_group_name,
423 strdup(arg));
424 if (!config.tracing_group_name.value) {
425 ret = -ENOMEM;
426 PERROR("strdup");
427 }
428 }
429 } else if (string_match(optname, "help") || opt == 'h') {
430 ret = utils_show_help(8, "lttng-sessiond", help_msg);
431 if (ret) {
432 ERR("Cannot show --help for `lttng-sessiond`");
433 perror("exec");
434 }
435 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
436 } else if (string_match(optname, "version") || opt == 'V') {
437 opt_print_version = 1;
438 } else if (string_match(optname, "sig-parent") || opt == 'S') {
439 config.sig_parent = true;
440 } else if (string_match(optname, "kconsumerd-err-sock")) {
441 if (!arg || *arg == '\0') {
442 ret = -EINVAL;
443 goto end;
444 }
445 if (lttng_is_setuid_setgid()) {
446 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
447 "--kconsumerd-err-sock");
448 } else {
449 config_string_set(&config.kconsumerd_err_unix_sock_path,
450 strdup(arg));
451 if (!config.kconsumerd_err_unix_sock_path.value) {
452 ret = -ENOMEM;
453 PERROR("strdup");
454 }
455 }
456 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
457 if (!arg || *arg == '\0') {
458 ret = -EINVAL;
459 goto end;
460 }
461 if (lttng_is_setuid_setgid()) {
462 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
463 "--kconsumerd-cmd-sock");
464 } else {
465 config_string_set(&config.kconsumerd_cmd_unix_sock_path,
466 strdup(arg));
467 if (!config.kconsumerd_cmd_unix_sock_path.value) {
468 ret = -ENOMEM;
469 PERROR("strdup");
470 }
471 }
472 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
473 if (!arg || *arg == '\0') {
474 ret = -EINVAL;
475 goto end;
476 }
477 if (lttng_is_setuid_setgid()) {
478 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
479 "--ustconsumerd64-err-sock");
480 } else {
481 config_string_set(&config.consumerd64_err_unix_sock_path,
482 strdup(arg));
483 if (!config.consumerd64_err_unix_sock_path.value) {
484 ret = -ENOMEM;
485 PERROR("strdup");
486 }
487 }
488 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
489 if (!arg || *arg == '\0') {
490 ret = -EINVAL;
491 goto end;
492 }
493 if (lttng_is_setuid_setgid()) {
494 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
495 "--ustconsumerd64-cmd-sock");
496 } else {
497 config_string_set(&config.consumerd64_cmd_unix_sock_path,
498 strdup(arg));
499 if (!config.consumerd64_cmd_unix_sock_path.value) {
500 ret = -ENOMEM;
501 PERROR("strdup");
502 }
503 }
504 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
505 if (!arg || *arg == '\0') {
506 ret = -EINVAL;
507 goto end;
508 }
509 if (lttng_is_setuid_setgid()) {
510 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
511 "--ustconsumerd32-err-sock");
512 } else {
513 config_string_set(&config.consumerd32_err_unix_sock_path,
514 strdup(arg));
515 if (!config.consumerd32_err_unix_sock_path.value) {
516 ret = -ENOMEM;
517 PERROR("strdup");
518 }
519 }
520 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
521 if (!arg || *arg == '\0') {
522 ret = -EINVAL;
523 goto end;
524 }
525 if (lttng_is_setuid_setgid()) {
526 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
527 "--ustconsumerd32-cmd-sock");
528 } else {
529 config_string_set(&config.consumerd32_cmd_unix_sock_path,
530 strdup(arg));
531 if (!config.consumerd32_cmd_unix_sock_path.value) {
532 ret = -ENOMEM;
533 PERROR("strdup");
534 }
535 }
536 } else if (string_match(optname, "no-kernel")) {
537 config.no_kernel = true;
538 } else if (string_match(optname, "quiet") || opt == 'q') {
539 config.quiet = true;
540 } else if (string_match(optname, "verbose") || opt == 'v') {
541 /* Verbose level can increase using multiple -v */
542 if (arg) {
543 /* Value obtained from config file */
544 config.verbose = config_parse_value(arg);
545 } else {
546 /* -v used on command line */
547 config.verbose++;
548 }
549 /* Clamp value to [0, 3] */
550 config.verbose = config.verbose < 0 ? 0 :
551 (config.verbose <= 3 ? config.verbose : 3);
552 } else if (string_match(optname, "verbose-consumer")) {
553 if (arg) {
554 config.verbose_consumer = config_parse_value(arg);
555 } else {
556 config.verbose_consumer++;
557 }
558 } else if (string_match(optname, "consumerd32-path")) {
559 if (!arg || *arg == '\0') {
560 ret = -EINVAL;
561 goto end;
562 }
563 if (lttng_is_setuid_setgid()) {
564 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
565 "--consumerd32-path");
566 } else {
567 config_string_set(&config.consumerd32_bin_path,
568 strdup(arg));
569 if (!config.consumerd32_bin_path.value) {
570 PERROR("strdup");
571 ret = -ENOMEM;
572 }
573 }
574 } else if (string_match(optname, "consumerd32-libdir")) {
575 if (!arg || *arg == '\0') {
576 ret = -EINVAL;
577 goto end;
578 }
579 if (lttng_is_setuid_setgid()) {
580 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
581 "--consumerd32-libdir");
582 } else {
583 config_string_set(&config.consumerd32_lib_dir,
584 strdup(arg));
585 if (!config.consumerd32_lib_dir.value) {
586 PERROR("strdup");
587 ret = -ENOMEM;
588 }
589 }
590 } else if (string_match(optname, "consumerd64-path")) {
591 if (!arg || *arg == '\0') {
592 ret = -EINVAL;
593 goto end;
594 }
595 if (lttng_is_setuid_setgid()) {
596 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
597 "--consumerd64-path");
598 } else {
599 config_string_set(&config.consumerd64_bin_path,
600 strdup(arg));
601 if (!config.consumerd64_bin_path.value) {
602 PERROR("strdup");
603 ret = -ENOMEM;
604 }
605 }
606 } else if (string_match(optname, "consumerd64-libdir")) {
607 if (!arg || *arg == '\0') {
608 ret = -EINVAL;
609 goto end;
610 }
611 if (lttng_is_setuid_setgid()) {
612 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
613 "--consumerd64-libdir");
614 } else {
615 config_string_set(&config.consumerd64_lib_dir,
616 strdup(arg));
617 if (!config.consumerd64_lib_dir.value) {
618 PERROR("strdup");
619 ret = -ENOMEM;
620 }
621 }
622 } else if (string_match(optname, "pidfile") || opt == 'p') {
623 if (!arg || *arg == '\0') {
624 ret = -EINVAL;
625 goto end;
626 }
627 if (lttng_is_setuid_setgid()) {
628 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
629 "-p, --pidfile");
630 } else {
631 config_string_set(&config.pid_file_path, strdup(arg));
632 if (!config.pid_file_path.value) {
633 PERROR("strdup");
634 ret = -ENOMEM;
635 }
636 }
637 } else if (string_match(optname, "agent-tcp-port")) {
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 "--agent-tcp-port");
645 } else {
646 unsigned long v;
647
648 errno = 0;
649 v = strtoul(arg, NULL, 0);
650 if (errno != 0 || !isdigit(arg[0])) {
651 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
652 return -1;
653 }
654 if (v == 0 || v >= 65535) {
655 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
656 return -1;
657 }
658 config.agent_tcp_port.begin = config.agent_tcp_port.end = (int) v;
659 DBG3("Agent TCP port set to non default: %i", (int) v);
660 }
661 } else if (string_match(optname, "load") || opt == 'l') {
662 if (!arg || *arg == '\0') {
663 ret = -EINVAL;
664 goto end;
665 }
666 if (lttng_is_setuid_setgid()) {
667 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
668 "-l, --load");
669 } else {
670 config_string_set(&config.load_session_path, strdup(arg));
671 if (!config.load_session_path.value) {
672 PERROR("strdup");
673 ret = -ENOMEM;
674 }
675 }
676 } else if (string_match(optname, "kmod-probes")) {
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 "--kmod-probes");
684 } else {
685 config_string_set(&config.kmod_probes_list, strdup(arg));
686 if (!config.kmod_probes_list.value) {
687 PERROR("strdup");
688 ret = -ENOMEM;
689 }
690 }
691 } else if (string_match(optname, "extra-kmod-probes")) {
692 if (!arg || *arg == '\0') {
693 ret = -EINVAL;
694 goto end;
695 }
696 if (lttng_is_setuid_setgid()) {
697 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
698 "--extra-kmod-probes");
699 } else {
700 config_string_set(&config.kmod_extra_probes_list,
701 strdup(arg));
702 if (!config.kmod_extra_probes_list.value) {
703 PERROR("strdup");
704 ret = -ENOMEM;
705 }
706 }
707 } else if (string_match(optname, "config") || opt == 'f') {
708 /* This is handled in set_options() thus silent skip. */
709 goto end;
710 } else {
711 /* Unknown option or other error.
712 * Error is printed by getopt, just return */
713 ret = -1;
714 }
715
716 end:
717 if (ret == -EINVAL) {
718 const char *opt_name = "unknown";
719 int i;
720
721 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
722 i++) {
723 if (opt == long_options[i].val) {
724 opt_name = long_options[i].name;
725 break;
726 }
727 }
728
729 WARN("Invalid argument provided for option \"%s\", using default value.",
730 opt_name);
731 }
732
733 return ret;
734 }
735
736 /*
737 * config_entry_handler_cb used to handle options read from a config file.
738 * See config_entry_handler_cb comment in common/config/session-config.h for the
739 * return value conventions.
740 */
741 static int config_entry_handler(const struct config_entry *entry, void *unused)
742 {
743 int ret = 0, i;
744
745 if (!entry || !entry->name || !entry->value) {
746 ret = -EINVAL;
747 goto end;
748 }
749
750 /* Check if the option is to be ignored */
751 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
752 if (!strcmp(entry->name, config_ignore_options[i])) {
753 goto end;
754 }
755 }
756
757 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
758 i++) {
759
760 /* Ignore if not fully matched. */
761 if (strcmp(entry->name, long_options[i].name)) {
762 continue;
763 }
764
765 /*
766 * If the option takes no argument on the command line, we have to
767 * check if the value is "true". We support non-zero numeric values,
768 * true, on and yes.
769 */
770 if (!long_options[i].has_arg) {
771 ret = config_parse_value(entry->value);
772 if (ret <= 0) {
773 if (ret) {
774 WARN("Invalid configuration value \"%s\" for option %s",
775 entry->value, entry->name);
776 }
777 /* False, skip boolean config option. */
778 goto end;
779 }
780 }
781
782 ret = set_option(long_options[i].val, entry->value, entry->name);
783 goto end;
784 }
785
786 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
787
788 end:
789 return ret;
790 }
791
792 static void print_version(void) {
793 fprintf(stdout, "%s\n", VERSION);
794 }
795
796 /*
797 * daemon configuration loading and argument parsing
798 */
799 static int set_options(int argc, char **argv)
800 {
801 int ret = 0, c = 0, option_index = 0;
802 int orig_optopt = optopt, orig_optind = optind;
803 char *optstring;
804 const char *config_path = NULL;
805
806 optstring = utils_generate_optstring(long_options,
807 sizeof(long_options) / sizeof(struct option));
808 if (!optstring) {
809 ret = -ENOMEM;
810 goto end;
811 }
812
813 /* Check for the --config option */
814 while ((c = getopt_long(argc, argv, optstring, long_options,
815 &option_index)) != -1) {
816 if (c == '?') {
817 ret = -EINVAL;
818 goto end;
819 } else if (c != 'f') {
820 /* if not equal to --config option. */
821 continue;
822 }
823
824 if (lttng_is_setuid_setgid()) {
825 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
826 "-f, --config");
827 } else {
828 config_path = utils_expand_path(optarg);
829 if (!config_path) {
830 ERR("Failed to resolve path: %s", optarg);
831 }
832 }
833 }
834
835 ret = config_get_section_entries(config_path, config_section_name,
836 config_entry_handler, NULL);
837 if (ret) {
838 if (ret > 0) {
839 ERR("Invalid configuration option at line %i", ret);
840 ret = -1;
841 }
842 goto end;
843 }
844
845 /* Reset getopt's global state */
846 optopt = orig_optopt;
847 optind = orig_optind;
848 while (1) {
849 option_index = -1;
850 /*
851 * getopt_long() will not set option_index if it encounters a
852 * short option.
853 */
854 c = getopt_long(argc, argv, optstring, long_options,
855 &option_index);
856 if (c == -1) {
857 break;
858 }
859
860 /*
861 * Pass NULL as the long option name if popt left the index
862 * unset.
863 */
864 ret = set_option(c, optarg,
865 option_index < 0 ? NULL :
866 long_options[option_index].name);
867 if (ret < 0) {
868 break;
869 }
870 }
871
872 end:
873 free(optstring);
874 return ret;
875 }
876
877 /*
878 * Create lockfile using the rundir and return its fd.
879 */
880 static int create_lockfile(void)
881 {
882 return utils_create_lock_file(config.lock_file_path.value);
883 }
884
885 /*
886 * Check if the global socket is available, and if a daemon is answering at the
887 * other side. If yes, error is returned.
888 *
889 * Also attempts to create and hold the lock file.
890 */
891 static int check_existing_daemon(void)
892 {
893 int ret = 0;
894
895 /* Is there anybody out there ? */
896 if (lttng_session_daemon_alive()) {
897 ret = -EEXIST;
898 goto end;
899 }
900
901 lockfile_fd = create_lockfile();
902 if (lockfile_fd < 0) {
903 ret = -EEXIST;
904 goto end;
905 }
906 end:
907 return ret;
908 }
909
910 static void sessiond_cleanup_lock_file(void)
911 {
912 int ret;
913
914 /*
915 * Cleanup lock file by deleting it and finaly closing it which will
916 * release the file system lock.
917 */
918 if (lockfile_fd >= 0) {
919 ret = remove(config.lock_file_path.value);
920 if (ret < 0) {
921 PERROR("remove lock file");
922 }
923 ret = close(lockfile_fd);
924 if (ret < 0) {
925 PERROR("close lock file");
926 }
927 }
928 }
929
930 /*
931 * Set the tracing group gid onto the client socket.
932 *
933 * Race window between mkdir and chown is OK because we are going from more
934 * permissive (root.root) to less permissive (root.tracing).
935 */
936 static int set_permissions(char *rundir)
937 {
938 int ret;
939 gid_t gid;
940
941 ret = utils_get_group_id(config.tracing_group_name.value, true, &gid);
942 if (ret) {
943 /* Default to root group. */
944 gid = 0;
945 }
946
947 /* Set lttng run dir */
948 ret = chown(rundir, 0, gid);
949 if (ret < 0) {
950 ERR("Unable to set group on %s", rundir);
951 PERROR("chown");
952 }
953
954 /*
955 * Ensure all applications and tracing group can search the run
956 * dir. Allow everyone to read the directory, since it does not
957 * buy us anything to hide its content.
958 */
959 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
960 if (ret < 0) {
961 ERR("Unable to set permissions on %s", rundir);
962 PERROR("chmod");
963 }
964
965 /* lttng client socket path */
966 ret = chown(config.client_unix_sock_path.value, 0, gid);
967 if (ret < 0) {
968 ERR("Unable to set group on %s", config.client_unix_sock_path.value);
969 PERROR("chown");
970 }
971
972 /* kconsumer error socket path */
973 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
974 if (ret < 0) {
975 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
976 PERROR("chown");
977 }
978
979 /* 64-bit ustconsumer error socket path */
980 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
981 if (ret < 0) {
982 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
983 PERROR("chown");
984 }
985
986 /* 32-bit ustconsumer compat32 error socket path */
987 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
988 if (ret < 0) {
989 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
990 PERROR("chown");
991 }
992
993 DBG("All permissions are set");
994
995 return ret;
996 }
997
998 /*
999 * Create the lttng run directory needed for all global sockets and pipe.
1000 */
1001 static int create_lttng_rundir(void)
1002 {
1003 int ret;
1004
1005 DBG3("Creating LTTng run directory: %s", config.rundir.value);
1006
1007 ret = mkdir(config.rundir.value, S_IRWXU);
1008 if (ret < 0) {
1009 if (errno != EEXIST) {
1010 ERR("Unable to create %s", config.rundir.value);
1011 goto error;
1012 } else {
1013 ret = 0;
1014 }
1015 }
1016
1017 error:
1018 return ret;
1019 }
1020
1021 /*
1022 * Setup sockets and directory needed by the consumerds' communication with the
1023 * session daemon.
1024 */
1025 static int set_consumer_sockets(struct consumer_data *consumer_data)
1026 {
1027 int ret;
1028 char *path = NULL;
1029
1030 switch (consumer_data->type) {
1031 case LTTNG_CONSUMER_KERNEL:
1032 path = config.kconsumerd_path.value;
1033 break;
1034 case LTTNG_CONSUMER64_UST:
1035 path = config.consumerd64_path.value;
1036 break;
1037 case LTTNG_CONSUMER32_UST:
1038 path = config.consumerd32_path.value;
1039 break;
1040 default:
1041 ERR("Consumer type unknown");
1042 ret = -EINVAL;
1043 goto error;
1044 }
1045 assert(path);
1046
1047 DBG2("Creating consumer directory: %s", path);
1048
1049 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
1050 if (ret < 0 && errno != EEXIST) {
1051 PERROR("mkdir");
1052 ERR("Failed to create %s", path);
1053 goto error;
1054 }
1055 if (is_root) {
1056 gid_t gid;
1057
1058 ret = utils_get_group_id(config.tracing_group_name.value, true,
1059 &gid);
1060 if (ret) {
1061 /* Default to root group. */
1062 gid = 0;
1063 }
1064
1065 ret = chown(path, 0, gid);
1066 if (ret < 0) {
1067 ERR("Unable to set group on %s", path);
1068 PERROR("chown");
1069 goto error;
1070 }
1071 }
1072
1073 /* Create the consumerd error unix socket */
1074 consumer_data->err_sock =
1075 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
1076 if (consumer_data->err_sock < 0) {
1077 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
1078 ret = -1;
1079 goto error;
1080 }
1081
1082 /*
1083 * Set the CLOEXEC flag. Return code is useless because either way, the
1084 * show must go on.
1085 */
1086 ret = utils_set_fd_cloexec(consumer_data->err_sock);
1087 if (ret < 0) {
1088 PERROR("utils_set_fd_cloexec");
1089 /* continue anyway */
1090 }
1091
1092 /* File permission MUST be 660 */
1093 ret = chmod(consumer_data->err_unix_sock_path,
1094 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
1095 if (ret < 0) {
1096 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
1097 PERROR("chmod");
1098 goto error;
1099 }
1100
1101 error:
1102 return ret;
1103 }
1104
1105 /*
1106 * Signal handler for the daemon
1107 *
1108 * Simply stop all worker threads, leaving main() return gracefully after
1109 * joining all threads and calling cleanup().
1110 */
1111 static void sighandler(int sig)
1112 {
1113 switch (sig) {
1114 case SIGINT:
1115 DBG("SIGINT caught");
1116 stop_threads();
1117 break;
1118 case SIGTERM:
1119 DBG("SIGTERM caught");
1120 stop_threads();
1121 break;
1122 case SIGUSR1:
1123 CMM_STORE_SHARED(recv_child_signal, 1);
1124 break;
1125 default:
1126 break;
1127 }
1128 }
1129
1130 /*
1131 * Setup signal handler for :
1132 * SIGINT, SIGTERM, SIGPIPE
1133 */
1134 static int set_signal_handler(void)
1135 {
1136 int ret = 0;
1137 struct sigaction sa;
1138 sigset_t sigset;
1139
1140 if ((ret = sigemptyset(&sigset)) < 0) {
1141 PERROR("sigemptyset");
1142 return ret;
1143 }
1144
1145 sa.sa_mask = sigset;
1146 sa.sa_flags = 0;
1147
1148 sa.sa_handler = sighandler;
1149 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
1150 PERROR("sigaction");
1151 return ret;
1152 }
1153
1154 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
1155 PERROR("sigaction");
1156 return ret;
1157 }
1158
1159 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
1160 PERROR("sigaction");
1161 return ret;
1162 }
1163
1164 sa.sa_handler = SIG_IGN;
1165 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
1166 PERROR("sigaction");
1167 return ret;
1168 }
1169
1170 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
1171
1172 return ret;
1173 }
1174
1175 /*
1176 * Set open files limit to unlimited. This daemon can open a large number of
1177 * file descriptors in order to consume multiple kernel traces.
1178 */
1179 static void set_ulimit(void)
1180 {
1181 int ret;
1182 struct rlimit lim;
1183
1184 /* The kernel does not allow an infinite limit for open files */
1185 lim.rlim_cur = 65535;
1186 lim.rlim_max = 65535;
1187
1188 ret = setrlimit(RLIMIT_NOFILE, &lim);
1189 if (ret < 0) {
1190 PERROR("failed to set open files limit");
1191 }
1192 }
1193
1194 static int write_pidfile(void)
1195 {
1196 return utils_create_pid_file(getpid(), config.pid_file_path.value);
1197 }
1198
1199 static int set_clock_plugin_env(void)
1200 {
1201 int ret = 0;
1202 char *env_value = NULL;
1203
1204 if (!config.lttng_ust_clock_plugin.value) {
1205 goto end;
1206 }
1207
1208 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
1209 config.lttng_ust_clock_plugin.value);
1210 if (ret < 0) {
1211 PERROR("asprintf");
1212 goto end;
1213 }
1214
1215 ret = putenv(env_value);
1216 if (ret) {
1217 free(env_value);
1218 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
1219 goto end;
1220 }
1221
1222 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
1223 config.lttng_ust_clock_plugin.value);
1224 end:
1225 return ret;
1226 }
1227
1228 static void destroy_all_sessions_and_wait(void)
1229 {
1230 struct ltt_session *session, *tmp;
1231 struct ltt_session_list *session_list;
1232
1233 session_list = session_get_list();
1234 DBG("Initiating destruction of all sessions");
1235
1236 if (!session_list) {
1237 return;
1238 }
1239
1240 session_lock_list();
1241 /* Initiate the destruction of all sessions. */
1242 cds_list_for_each_entry_safe(session, tmp,
1243 &session_list->head, list) {
1244 if (!session_get(session)) {
1245 continue;
1246 }
1247
1248 session_lock(session);
1249 if (session->destroyed) {
1250 goto unlock_session;
1251 }
1252 (void) cmd_stop_trace(session);
1253 (void) cmd_destroy_session(session, notification_thread_handle,
1254 NULL);
1255 unlock_session:
1256 session_unlock(session);
1257 session_put(session);
1258 }
1259 session_unlock_list();
1260
1261 /* Wait for the destruction of all sessions to complete. */
1262 DBG("Waiting for the destruction of all sessions to complete");
1263 session_list_wait_empty();
1264 DBG("Destruction of all sessions completed");
1265 }
1266
1267 static int run_as_worker_post_fork_cleanup(void *data)
1268 {
1269 struct sessiond_config *sessiond_config = data;
1270
1271 sessiond_config_fini(sessiond_config);
1272 return 0;
1273 }
1274
1275 static int launch_run_as_worker(const char *procname)
1276 {
1277 /*
1278 * Clean-up before forking the run-as worker. Any dynamically
1279 * allocated memory of which the worker is not aware will
1280 * be leaked as the process forks a run-as worker (and performs
1281 * no exec*()). The same would apply to any opened fd.
1282 */
1283 return run_as_create_worker(procname, run_as_worker_post_fork_cleanup,
1284 &config);
1285 }
1286
1287 static void sessiond_uuid_log(void)
1288 {
1289 char uuid_str[UUID_STR_LEN];
1290
1291 lttng_uuid_to_str(sessiond_uuid, uuid_str);
1292 DBG("Starting lttng-sessiond {%s}", uuid_str);
1293 }
1294
1295 /*
1296 * main
1297 */
1298 int main(int argc, char **argv)
1299 {
1300 int ret = 0, retval = 0;
1301 const char *env_app_timeout;
1302 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
1303 *ust64_channel_monitor_pipe = NULL,
1304 *kernel_channel_monitor_pipe = NULL;
1305 struct lttng_thread *ht_cleanup_thread = NULL;
1306 struct timer_thread_parameters timer_thread_parameters;
1307 /* Rotation thread handle. */
1308 struct rotation_thread_handle *rotation_thread_handle = NULL;
1309 /* Queue of rotation jobs populated by the sessiond-timer. */
1310 struct rotation_thread_timer_queue *rotation_timer_queue = NULL;
1311 struct lttng_thread *client_thread = NULL;
1312 struct lttng_thread *notification_thread = NULL;
1313 struct lttng_thread *register_apps_thread = NULL;
1314
1315 init_kernel_workarounds();
1316
1317 rcu_register_thread();
1318
1319 if (set_signal_handler()) {
1320 retval = -1;
1321 goto exit_set_signal_handler;
1322 }
1323
1324 if (timer_signal_init()) {
1325 retval = -1;
1326 goto exit_set_signal_handler;
1327 }
1328
1329 page_size = sysconf(_SC_PAGESIZE);
1330 if (page_size < 0) {
1331 PERROR("sysconf _SC_PAGESIZE");
1332 page_size = LONG_MAX;
1333 WARN("Fallback page size to %ld", page_size);
1334 }
1335
1336 ret = sessiond_config_init(&config);
1337 if (ret) {
1338 retval = -1;
1339 goto exit_set_signal_handler;
1340 }
1341
1342 /*
1343 * Init config from environment variables.
1344 * Command line option override env configuration per-doc. Do env first.
1345 */
1346 sessiond_config_apply_env_config(&config);
1347
1348 /*
1349 * Parse arguments and load the daemon configuration file.
1350 *
1351 * We have an exit_options exit path to free memory reserved by
1352 * set_options. This is needed because the rest of sessiond_cleanup()
1353 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
1354 * depends on set_options.
1355 */
1356 progname = argv[0];
1357 if (set_options(argc, argv)) {
1358 retval = -1;
1359 goto exit_options;
1360 }
1361
1362 /*
1363 * Resolve all paths received as arguments, configuration option, or
1364 * through environment variable as absolute paths. This is necessary
1365 * since daemonizing causes the sessiond's current working directory
1366 * to '/'.
1367 */
1368 ret = sessiond_config_resolve_paths(&config);
1369 if (ret) {
1370 goto exit_options;
1371 }
1372
1373 /* Apply config. */
1374 lttng_opt_verbose = config.verbose;
1375 lttng_opt_quiet = config.quiet;
1376 kconsumer_data.err_unix_sock_path =
1377 config.kconsumerd_err_unix_sock_path.value;
1378 kconsumer_data.cmd_unix_sock_path =
1379 config.kconsumerd_cmd_unix_sock_path.value;
1380 ustconsumer32_data.err_unix_sock_path =
1381 config.consumerd32_err_unix_sock_path.value;
1382 ustconsumer32_data.cmd_unix_sock_path =
1383 config.consumerd32_cmd_unix_sock_path.value;
1384 ustconsumer64_data.err_unix_sock_path =
1385 config.consumerd64_err_unix_sock_path.value;
1386 ustconsumer64_data.cmd_unix_sock_path =
1387 config.consumerd64_cmd_unix_sock_path.value;
1388 set_clock_plugin_env();
1389
1390 sessiond_config_log(&config);
1391 sessiond_uuid_log();
1392
1393 if (opt_print_version) {
1394 print_version();
1395 retval = 0;
1396 goto exit_options;
1397 }
1398
1399 if (create_lttng_rundir()) {
1400 retval = -1;
1401 goto exit_options;
1402 }
1403
1404 /* Abort launch if a session daemon is already running. */
1405 if (check_existing_daemon()) {
1406 ERR("A session daemon is already running.");
1407 retval = -1;
1408 goto exit_options;
1409 }
1410
1411 /* Daemonize */
1412 if (config.daemonize || config.background) {
1413 int i;
1414
1415 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
1416 !config.background);
1417 if (ret < 0) {
1418 retval = -1;
1419 goto exit_options;
1420 }
1421
1422 /*
1423 * We are in the child. Make sure all other file descriptors are
1424 * closed, in case we are called with more opened file
1425 * descriptors than the standard ones and the lock file.
1426 */
1427 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
1428 if (i == lockfile_fd) {
1429 continue;
1430 }
1431 (void) close(i);
1432 }
1433 }
1434
1435 if (launch_run_as_worker(argv[0]) < 0) {
1436 goto exit_create_run_as_worker_cleanup;
1437 }
1438
1439 /*
1440 * Starting from here, we can create threads. This needs to be after
1441 * lttng_daemonize due to RCU.
1442 */
1443
1444 /*
1445 * Initialize the health check subsystem. This call should set the
1446 * appropriate time values.
1447 */
1448 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
1449 if (!health_sessiond) {
1450 PERROR("health_app_create error");
1451 retval = -1;
1452 goto stop_threads;
1453 }
1454
1455 /* Create thread to clean up RCU hash tables */
1456 ht_cleanup_thread = launch_ht_cleanup_thread();
1457 if (!ht_cleanup_thread) {
1458 retval = -1;
1459 goto stop_threads;
1460 }
1461
1462 /* Create thread quit pipe */
1463 if (sessiond_init_thread_quit_pipe()) {
1464 retval = -1;
1465 goto stop_threads;
1466 }
1467
1468 /* Check if daemon is UID = 0 */
1469 is_root = !getuid();
1470 if (is_root) {
1471 /* Create global run dir with root access */
1472
1473 kernel_channel_monitor_pipe = lttng_pipe_open(0);
1474 if (!kernel_channel_monitor_pipe) {
1475 ERR("Failed to create kernel consumer channel monitor pipe");
1476 retval = -1;
1477 goto stop_threads;
1478 }
1479 kconsumer_data.channel_monitor_pipe =
1480 lttng_pipe_release_writefd(
1481 kernel_channel_monitor_pipe);
1482 if (kconsumer_data.channel_monitor_pipe < 0) {
1483 retval = -1;
1484 goto stop_threads;
1485 }
1486 }
1487
1488 /* Set consumer initial state */
1489 kernel_consumerd_state = CONSUMER_STOPPED;
1490 ust_consumerd_state = CONSUMER_STOPPED;
1491
1492 ust32_channel_monitor_pipe = lttng_pipe_open(0);
1493 if (!ust32_channel_monitor_pipe) {
1494 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
1495 retval = -1;
1496 goto stop_threads;
1497 }
1498 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
1499 ust32_channel_monitor_pipe);
1500 if (ustconsumer32_data.channel_monitor_pipe < 0) {
1501 retval = -1;
1502 goto stop_threads;
1503 }
1504
1505 /*
1506 * The rotation_thread_timer_queue structure is shared between the
1507 * sessiond timer thread and the rotation thread. The main thread keeps
1508 * its ownership and destroys it when both threads have been joined.
1509 */
1510 rotation_timer_queue = rotation_thread_timer_queue_create();
1511 if (!rotation_timer_queue) {
1512 retval = -1;
1513 goto stop_threads;
1514 }
1515 timer_thread_parameters.rotation_thread_job_queue =
1516 rotation_timer_queue;
1517
1518 ust64_channel_monitor_pipe = lttng_pipe_open(0);
1519 if (!ust64_channel_monitor_pipe) {
1520 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
1521 retval = -1;
1522 goto stop_threads;
1523 }
1524 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
1525 ust64_channel_monitor_pipe);
1526 if (ustconsumer64_data.channel_monitor_pipe < 0) {
1527 retval = -1;
1528 goto stop_threads;
1529 }
1530
1531 /*
1532 * Init UST app hash table. Alloc hash table before this point since
1533 * cleanup() can get called after that point.
1534 */
1535 if (ust_app_ht_alloc()) {
1536 ERR("Failed to allocate UST app hash table");
1537 retval = -1;
1538 goto stop_threads;
1539 }
1540
1541 /*
1542 * Initialize agent app hash table. We allocate the hash table here
1543 * since cleanup() can get called after this point.
1544 */
1545 if (agent_app_ht_alloc()) {
1546 ERR("Failed to allocate Agent app hash table");
1547 retval = -1;
1548 goto stop_threads;
1549 }
1550
1551 /*
1552 * These actions must be executed as root. We do that *after* setting up
1553 * the sockets path because we MUST make the check for another daemon using
1554 * those paths *before* trying to set the kernel consumer sockets and init
1555 * kernel tracer.
1556 */
1557 if (is_root) {
1558 if (set_consumer_sockets(&kconsumer_data)) {
1559 retval = -1;
1560 goto stop_threads;
1561 }
1562
1563 /* Setup kernel tracer */
1564 if (!config.no_kernel) {
1565 init_kernel_tracer();
1566 }
1567
1568 /* Set ulimit for open files */
1569 set_ulimit();
1570 }
1571 /* init lttng_fd tracking must be done after set_ulimit. */
1572 lttng_fd_init();
1573
1574 if (set_consumer_sockets(&ustconsumer64_data)) {
1575 retval = -1;
1576 goto stop_threads;
1577 }
1578
1579 if (set_consumer_sockets(&ustconsumer32_data)) {
1580 retval = -1;
1581 goto stop_threads;
1582 }
1583
1584 /* Get parent pid if -S, --sig-parent is specified. */
1585 if (config.sig_parent) {
1586 ppid = getppid();
1587 }
1588
1589 /* Setup the kernel pipe for waking up the kernel thread */
1590 if (is_root && !config.no_kernel) {
1591 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
1592 retval = -1;
1593 goto stop_threads;
1594 }
1595 }
1596
1597 /* Setup the thread apps communication pipe. */
1598 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
1599 retval = -1;
1600 goto stop_threads;
1601 }
1602
1603 /* Setup the thread apps notify communication pipe. */
1604 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
1605 retval = -1;
1606 goto stop_threads;
1607 }
1608
1609 /* Initialize global buffer per UID and PID registry. */
1610 buffer_reg_init_uid_registry();
1611 buffer_reg_init_pid_registry();
1612
1613 /* Init UST command queue. */
1614 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1615
1616 cmd_init();
1617
1618 /* Check for the application socket timeout env variable. */
1619 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
1620 if (env_app_timeout) {
1621 config.app_socket_timeout = atoi(env_app_timeout);
1622 } else {
1623 config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
1624 }
1625
1626 ret = write_pidfile();
1627 if (ret) {
1628 ERR("Error in write_pidfile");
1629 retval = -1;
1630 goto stop_threads;
1631 }
1632
1633 /* Initialize communication library */
1634 lttcomm_init();
1635 /* Initialize TCP timeout values */
1636 lttcomm_inet_init();
1637
1638 /* Create health-check thread. */
1639 if (!launch_health_management_thread()) {
1640 retval = -1;
1641 goto stop_threads;
1642 }
1643
1644 /* notification_thread_data acquires the pipes' read side. */
1645 notification_thread_handle = notification_thread_handle_create(
1646 ust32_channel_monitor_pipe,
1647 ust64_channel_monitor_pipe,
1648 kernel_channel_monitor_pipe);
1649 if (!notification_thread_handle) {
1650 retval = -1;
1651 ERR("Failed to create notification thread shared data");
1652 goto stop_threads;
1653 }
1654
1655 /* Create notification thread. */
1656 notification_thread = launch_notification_thread(
1657 notification_thread_handle);
1658 if (!notification_thread) {
1659 retval = -1;
1660 goto stop_threads;
1661 }
1662
1663 /* Create timer thread. */
1664 if (!launch_timer_thread(&timer_thread_parameters)) {
1665 retval = -1;
1666 goto stop_threads;
1667 }
1668
1669 /* rotation_thread_data acquires the pipes' read side. */
1670 rotation_thread_handle = rotation_thread_handle_create(
1671 rotation_timer_queue,
1672 notification_thread_handle);
1673 if (!rotation_thread_handle) {
1674 retval = -1;
1675 ERR("Failed to create rotation thread shared data");
1676 stop_threads();
1677 goto stop_threads;
1678 }
1679
1680 /* Create rotation thread. */
1681 if (!launch_rotation_thread(rotation_thread_handle)) {
1682 retval = -1;
1683 goto stop_threads;
1684 }
1685
1686 /* Create thread to manage the client socket */
1687 client_thread = launch_client_thread();
1688 if (!client_thread) {
1689 retval = -1;
1690 goto stop_threads;
1691 }
1692
1693 /* Set credentials of the client socket and rundir */
1694 if (is_root && set_permissions(config.rundir.value)) {
1695 retval = -1;
1696 goto stop_threads;
1697 }
1698
1699 if (!launch_ust_dispatch_thread(&ust_cmd_queue, apps_cmd_pipe[1],
1700 apps_cmd_notify_pipe[1])) {
1701 retval = -1;
1702 goto stop_threads;
1703 }
1704
1705 /* Create thread to manage application registration. */
1706 register_apps_thread = launch_application_registration_thread(
1707 &ust_cmd_queue);
1708 if (!register_apps_thread) {
1709 retval = -1;
1710 goto stop_threads;
1711 }
1712
1713 /* Create thread to manage application socket */
1714 if (!launch_application_management_thread(apps_cmd_pipe[0])) {
1715 retval = -1;
1716 goto stop_threads;
1717 }
1718
1719 /* Create thread to manage application notify socket */
1720 if (!launch_application_notification_thread(apps_cmd_notify_pipe[0])) {
1721 retval = -1;
1722 goto stop_threads;
1723 }
1724
1725 /* Create agent management thread. */
1726 if (!launch_agent_management_thread()) {
1727 retval = -1;
1728 goto stop_threads;
1729 }
1730
1731 /* Don't start this thread if kernel tracing is not requested nor root */
1732 if (is_root && !config.no_kernel) {
1733 /* Create kernel thread to manage kernel event */
1734 if (!launch_kernel_management_thread(kernel_poll_pipe[0])) {
1735 retval = -1;
1736 goto stop_threads;
1737 }
1738 }
1739
1740 /* Load sessions. */
1741 ret = config_load_session(config.load_session_path.value,
1742 NULL, 1, 1, NULL);
1743 if (ret) {
1744 ERR("Session load failed: %s", error_get_str(ret));
1745 retval = -1;
1746 goto stop_threads;
1747 }
1748
1749 /* Initialization completed. */
1750 sessiond_signal_parents();
1751
1752 /*
1753 * This is where we start awaiting program completion (e.g. through
1754 * signal that asks threads to teardown).
1755 */
1756
1757 /* Initiate teardown once activity occurs on the quit pipe. */
1758 sessiond_wait_for_quit_pipe(-1);
1759
1760 stop_threads:
1761 /*
1762 * Ensure that the client thread is no longer accepting new commands,
1763 * which could cause new sessions to be created.
1764 */
1765 if (client_thread) {
1766 lttng_thread_shutdown(client_thread);
1767 lttng_thread_put(client_thread);
1768 }
1769
1770 destroy_all_sessions_and_wait();
1771
1772 if (register_apps_thread) {
1773 lttng_thread_shutdown(register_apps_thread);
1774 lttng_thread_put(register_apps_thread);
1775 }
1776 lttng_thread_list_shutdown_orphans();
1777
1778 /*
1779 * Wait for all pending call_rcu work to complete before tearing
1780 * down data structures. call_rcu worker may be trying to
1781 * perform lookups in those structures.
1782 */
1783 rcu_barrier();
1784 /*
1785 * sessiond_cleanup() is called when no other thread is running, except
1786 * the ht_cleanup thread, which is needed to destroy the hash tables.
1787 */
1788 rcu_thread_online();
1789 sessiond_cleanup();
1790
1791 if (notification_thread) {
1792 lttng_thread_shutdown(notification_thread);
1793 lttng_thread_put(notification_thread);
1794 }
1795
1796 /*
1797 * Ensure all prior call_rcu are done. call_rcu callbacks may push
1798 * hash tables to the ht_cleanup thread. Therefore, we ensure that
1799 * the queue is empty before shutting down the clean-up thread.
1800 */
1801 rcu_barrier();
1802
1803 if (ht_cleanup_thread) {
1804 lttng_thread_shutdown(ht_cleanup_thread);
1805 lttng_thread_put(ht_cleanup_thread);
1806 }
1807
1808 rcu_thread_offline();
1809 rcu_unregister_thread();
1810
1811 if (rotation_thread_handle) {
1812 rotation_thread_handle_destroy(rotation_thread_handle);
1813 }
1814
1815 /*
1816 * After the rotation and timer thread have quit, we can safely destroy
1817 * the rotation_timer_queue.
1818 */
1819 rotation_thread_timer_queue_destroy(rotation_timer_queue);
1820 /*
1821 * The teardown of the notification system is performed after the
1822 * session daemon's teardown in order to allow it to be notified
1823 * of the active session and channels at the moment of the teardown.
1824 */
1825 if (notification_thread_handle) {
1826 notification_thread_handle_destroy(notification_thread_handle);
1827 }
1828 lttng_pipe_destroy(ust32_channel_monitor_pipe);
1829 lttng_pipe_destroy(ust64_channel_monitor_pipe);
1830 lttng_pipe_destroy(kernel_channel_monitor_pipe);
1831
1832 if (health_sessiond) {
1833 health_app_destroy(health_sessiond);
1834 }
1835 exit_create_run_as_worker_cleanup:
1836 exit_options:
1837 sessiond_cleanup_lock_file();
1838 sessiond_cleanup_options();
1839
1840 exit_set_signal_handler:
1841 if (!retval) {
1842 exit(EXIT_SUCCESS);
1843 } else {
1844 exit(EXIT_FAILURE);
1845 }
1846 }
This page took 0.068667 seconds and 4 git commands to generate.