Fix: setuid/setgid daemons should not get sensitive env. var./args
[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 _GNU_SOURCE
21 #define _LGPL_SOURCE
22 #include <getopt.h>
23 #include <grp.h>
24 #include <limits.h>
25 #include <paths.h>
26 #include <pthread.h>
27 #include <signal.h>
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <string.h>
31 #include <inttypes.h>
32 #include <sys/mman.h>
33 #include <sys/mount.h>
34 #include <sys/resource.h>
35 #include <sys/socket.h>
36 #include <sys/stat.h>
37 #include <sys/types.h>
38 #include <sys/wait.h>
39 #include <urcu/uatomic.h>
40 #include <unistd.h>
41 #include <config.h>
42
43 #include <common/common.h>
44 #include <common/compat/socket.h>
45 #include <common/compat/getenv.h>
46 #include <common/defaults.h>
47 #include <common/kernel-consumer/kernel-consumer.h>
48 #include <common/futex.h>
49 #include <common/relayd/relayd.h>
50 #include <common/utils.h>
51 #include <common/daemonize.h>
52 #include <common/config/config.h>
53
54 #include "lttng-sessiond.h"
55 #include "buffer-registry.h"
56 #include "channel.h"
57 #include "cmd.h"
58 #include "consumer.h"
59 #include "context.h"
60 #include "event.h"
61 #include "kernel.h"
62 #include "kernel-consumer.h"
63 #include "modprobe.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 "ust-thread.h"
72 #include "agent-thread.h"
73 #include "save.h"
74 #include "load-session-thread.h"
75 #include "syscall.h"
76
77 #define CONSUMERD_FILE "lttng-consumerd"
78
79 const char *progname;
80 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
81 static int tracing_group_name_override;
82 static char *opt_pidfile;
83 static int opt_sig_parent;
84 static int opt_verbose_consumer;
85 static int opt_daemon, opt_background;
86 static int opt_no_kernel;
87 static char *opt_load_session_path;
88 static pid_t ppid; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
90 static char *rundir;
91 static int lockfile_fd = -1;
92
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal;
95
96 /*
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
99 */
100 static struct consumer_data kconsumer_data = {
101 .type = LTTNG_CONSUMER_KERNEL,
102 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
103 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
104 .err_sock = -1,
105 .cmd_sock = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer64_data = {
112 .type = LTTNG_CONSUMER64_UST,
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
118 .lock = PTHREAD_MUTEX_INITIALIZER,
119 .cond = PTHREAD_COND_INITIALIZER,
120 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
121 };
122 static struct consumer_data ustconsumer32_data = {
123 .type = LTTNG_CONSUMER32_UST,
124 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
125 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
126 .err_sock = -1,
127 .cmd_sock = -1,
128 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
129 .lock = PTHREAD_MUTEX_INITIALIZER,
130 .cond = PTHREAD_COND_INITIALIZER,
131 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
132 };
133
134 /* Command line options */
135 static const struct option long_options[] = {
136 { "client-sock", 1, 0, 'c' },
137 { "apps-sock", 1, 0, 'a' },
138 { "kconsumerd-cmd-sock", 1, 0, 'C' },
139 { "kconsumerd-err-sock", 1, 0, 'E' },
140 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
141 { "ustconsumerd32-err-sock", 1, 0, 'H' },
142 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
143 { "ustconsumerd64-err-sock", 1, 0, 'F' },
144 { "consumerd32-path", 1, 0, 'u' },
145 { "consumerd32-libdir", 1, 0, 'U' },
146 { "consumerd64-path", 1, 0, 't' },
147 { "consumerd64-libdir", 1, 0, 'T' },
148 { "daemonize", 0, 0, 'd' },
149 { "background", 0, 0, 'b' },
150 { "sig-parent", 0, 0, 'S' },
151 { "help", 0, 0, 'h' },
152 { "group", 1, 0, 'g' },
153 { "version", 0, 0, 'V' },
154 { "quiet", 0, 0, 'q' },
155 { "verbose", 0, 0, 'v' },
156 { "verbose-consumer", 0, 0, 'Z' },
157 { "no-kernel", 0, 0, 'N' },
158 { "pidfile", 1, 0, 'p' },
159 { "agent-tcp-port", 1, 0, 'J' },
160 { "config", 1, 0, 'f' },
161 { "load", 1, 0, 'l' },
162 { "kmod-probes", 1, 0, 'P' },
163 { "extra-kmod-probes", 1, 0, 'e' },
164 { NULL, 0, 0, 0 }
165 };
166
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options[] = { "help", "version", "config" };
169
170 /* Shared between threads */
171 static int dispatch_thread_exit;
172
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path[PATH_MAX];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path[PATH_MAX];
177 /* global wait shm path for UST */
178 static char wait_shm_path[PATH_MAX];
179 /* Global health check unix path */
180 static char health_unix_sock_path[PATH_MAX];
181
182 /* Sockets and FDs */
183 static int client_sock = -1;
184 static int apps_sock = -1;
185 int kernel_tracer_fd = -1;
186 static int kernel_poll_pipe[2] = { -1, -1 };
187
188 /*
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
191 */
192 static int thread_quit_pipe[2] = { -1, -1 };
193 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
194
195 /*
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
198 */
199 static int apps_cmd_pipe[2] = { -1, -1 };
200
201 int apps_cmd_notify_pipe[2] = { -1, -1 };
202
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread;
205 static pthread_t apps_notify_thread;
206 static pthread_t reg_apps_thread;
207 static pthread_t client_thread;
208 static pthread_t kernel_thread;
209 static pthread_t dispatch_thread;
210 static pthread_t health_thread;
211 static pthread_t ht_cleanup_thread;
212 static pthread_t agent_reg_thread;
213 static pthread_t load_session_thread;
214
215 /*
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
218 *
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
223 */
224 static struct ust_cmd_queue ust_cmd_queue;
225
226 /*
227 * Pointer initialized before thread creation.
228 *
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
232 *
233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
234 * session_lock_list and session_unlock_list for lock acquisition.
235 */
236 static struct ltt_session_list *session_list_ptr;
237
238 int ust_consumerd64_fd = -1;
239 int ust_consumerd32_fd = -1;
240
241 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
242 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
243 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
244 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
245 static int consumerd32_bin_override;
246 static int consumerd64_bin_override;
247 static int consumerd32_libdir_override;
248 static int consumerd64_libdir_override;
249
250 static const char *module_proc_lttng = "/proc/lttng";
251
252 /*
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
255 */
256 enum consumerd_state {
257 CONSUMER_STARTED = 1,
258 CONSUMER_STOPPED = 2,
259 CONSUMER_ERROR = 3,
260 };
261
262 /*
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
267 *
268 * The following example shows a possible race condition of this scheme:
269 *
270 * consumer thread error happens
271 * client cmd arrives
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
275 * ...
276 *
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
282 */
283 static enum consumerd_state ust_consumerd_state;
284 static enum consumerd_state kernel_consumerd_state;
285
286 /*
287 * Socket timeout for receiving and sending in seconds.
288 */
289 static int app_socket_timeout;
290
291 /* Set in main() with the current page size. */
292 long page_size;
293
294 /* Application health monitoring */
295 struct health_app *health_sessiond;
296
297 /* Agent TCP port for registration. Used by the agent thread. */
298 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
299
300 /* Am I root or not. */
301 int is_root; /* Set to 1 if the daemon is running as root */
302
303 const char * const config_section_name = "sessiond";
304
305 /* Load session thread information to operate. */
306 struct load_session_thread_data *load_info;
307
308 /*
309 * Whether sessiond is ready for commands/health check requests.
310 * NR_LTTNG_SESSIOND_READY must match the number of calls to
311 * sessiond_notify_ready().
312 */
313 #define NR_LTTNG_SESSIOND_READY 3
314 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
315
316 /* Notify parents that we are ready for cmd and health check */
317 LTTNG_HIDDEN
318 void sessiond_notify_ready(void)
319 {
320 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
321 /*
322 * Notify parent pid that we are ready to accept command
323 * for client side. This ppid is the one from the
324 * external process that spawned us.
325 */
326 if (opt_sig_parent) {
327 kill(ppid, SIGUSR1);
328 }
329
330 /*
331 * Notify the parent of the fork() process that we are
332 * ready.
333 */
334 if (opt_daemon || opt_background) {
335 kill(child_ppid, SIGUSR1);
336 }
337 }
338 }
339
340 static
341 void setup_consumerd_path(void)
342 {
343 const char *bin, *libdir;
344
345 /*
346 * Allow INSTALL_BIN_PATH to be used as a target path for the
347 * native architecture size consumer if CONFIG_CONSUMER*_PATH
348 * has not been defined.
349 */
350 #if (CAA_BITS_PER_LONG == 32)
351 if (!consumerd32_bin[0]) {
352 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
353 }
354 if (!consumerd32_libdir[0]) {
355 consumerd32_libdir = INSTALL_LIB_PATH;
356 }
357 #elif (CAA_BITS_PER_LONG == 64)
358 if (!consumerd64_bin[0]) {
359 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
360 }
361 if (!consumerd64_libdir[0]) {
362 consumerd64_libdir = INSTALL_LIB_PATH;
363 }
364 #else
365 #error "Unknown bitness"
366 #endif
367
368 /*
369 * runtime env. var. overrides the build default.
370 */
371 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
372 if (bin) {
373 consumerd32_bin = bin;
374 }
375 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
376 if (bin) {
377 consumerd64_bin = bin;
378 }
379 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
380 if (libdir) {
381 consumerd32_libdir = libdir;
382 }
383 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
384 if (libdir) {
385 consumerd64_libdir = libdir;
386 }
387 }
388
389 static
390 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
391 int *a_pipe)
392 {
393 int ret;
394
395 assert(events);
396
397 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
398 if (ret < 0) {
399 goto error;
400 }
401
402 /* Add quit pipe */
403 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
404 if (ret < 0) {
405 goto error;
406 }
407
408 return 0;
409
410 error:
411 return ret;
412 }
413
414 /*
415 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
416 */
417 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
418 {
419 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
420 }
421
422 /*
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
424 */
425 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
426 size_t size)
427 {
428 return __sessiond_set_thread_pollset(events, size,
429 ht_cleanup_quit_pipe);
430 }
431
432 static
433 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
434 {
435 if (fd == a_pipe && (events & LPOLLIN)) {
436 return 1;
437 }
438 return 0;
439 }
440
441 /*
442 * Check if the thread quit pipe was triggered.
443 *
444 * Return 1 if it was triggered else 0;
445 */
446 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
447 {
448 return __sessiond_check_thread_quit_pipe(fd, events,
449 thread_quit_pipe[0]);
450 }
451
452 /*
453 * Check if the ht_cleanup thread quit pipe was triggered.
454 *
455 * Return 1 if it was triggered else 0;
456 */
457 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
458 {
459 return __sessiond_check_thread_quit_pipe(fd, events,
460 ht_cleanup_quit_pipe[0]);
461 }
462
463 /*
464 * Init thread quit pipe.
465 *
466 * Return -1 on error or 0 if all pipes are created.
467 */
468 static int __init_thread_quit_pipe(int *a_pipe)
469 {
470 int ret, i;
471
472 ret = pipe(a_pipe);
473 if (ret < 0) {
474 PERROR("thread quit pipe");
475 goto error;
476 }
477
478 for (i = 0; i < 2; i++) {
479 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
480 if (ret < 0) {
481 PERROR("fcntl");
482 goto error;
483 }
484 }
485
486 error:
487 return ret;
488 }
489
490 static int init_thread_quit_pipe(void)
491 {
492 return __init_thread_quit_pipe(thread_quit_pipe);
493 }
494
495 static int init_ht_cleanup_quit_pipe(void)
496 {
497 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
498 }
499
500 /*
501 * Stop all threads by closing the thread quit pipe.
502 */
503 static void stop_threads(void)
504 {
505 int ret;
506
507 /* Stopping all threads */
508 DBG("Terminating all threads");
509 ret = notify_thread_pipe(thread_quit_pipe[1]);
510 if (ret < 0) {
511 ERR("write error on thread quit pipe");
512 }
513
514 /* Dispatch thread */
515 CMM_STORE_SHARED(dispatch_thread_exit, 1);
516 futex_nto1_wake(&ust_cmd_queue.futex);
517 }
518
519 /*
520 * Close every consumer sockets.
521 */
522 static void close_consumer_sockets(void)
523 {
524 int ret;
525
526 if (kconsumer_data.err_sock >= 0) {
527 ret = close(kconsumer_data.err_sock);
528 if (ret < 0) {
529 PERROR("kernel consumer err_sock close");
530 }
531 }
532 if (ustconsumer32_data.err_sock >= 0) {
533 ret = close(ustconsumer32_data.err_sock);
534 if (ret < 0) {
535 PERROR("UST consumerd32 err_sock close");
536 }
537 }
538 if (ustconsumer64_data.err_sock >= 0) {
539 ret = close(ustconsumer64_data.err_sock);
540 if (ret < 0) {
541 PERROR("UST consumerd64 err_sock close");
542 }
543 }
544 if (kconsumer_data.cmd_sock >= 0) {
545 ret = close(kconsumer_data.cmd_sock);
546 if (ret < 0) {
547 PERROR("kernel consumer cmd_sock close");
548 }
549 }
550 if (ustconsumer32_data.cmd_sock >= 0) {
551 ret = close(ustconsumer32_data.cmd_sock);
552 if (ret < 0) {
553 PERROR("UST consumerd32 cmd_sock close");
554 }
555 }
556 if (ustconsumer64_data.cmd_sock >= 0) {
557 ret = close(ustconsumer64_data.cmd_sock);
558 if (ret < 0) {
559 PERROR("UST consumerd64 cmd_sock close");
560 }
561 }
562 }
563
564 /*
565 * Generate the full lock file path using the rundir.
566 *
567 * Return the snprintf() return value thus a negative value is an error.
568 */
569 static int generate_lock_file_path(char *path, size_t len)
570 {
571 int ret;
572
573 assert(path);
574 assert(rundir);
575
576 /* Build lockfile path from rundir. */
577 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
578 if (ret < 0) {
579 PERROR("snprintf lockfile path");
580 }
581
582 return ret;
583 }
584
585 /*
586 * Cleanup the session daemon's data structures.
587 */
588 static void sessiond_cleanup(void)
589 {
590 int ret;
591 struct ltt_session *sess, *stmp;
592 char path[PATH_MAX];
593
594 DBG("Cleanup sessiond");
595
596 /*
597 * Close the thread quit pipe. It has already done its job,
598 * since we are now called.
599 */
600 utils_close_pipe(thread_quit_pipe);
601
602 /*
603 * If opt_pidfile is undefined, the default file will be wiped when
604 * removing the rundir.
605 */
606 if (opt_pidfile) {
607 ret = remove(opt_pidfile);
608 if (ret < 0) {
609 PERROR("remove pidfile %s", opt_pidfile);
610 }
611 }
612
613 DBG("Removing sessiond and consumerd content of directory %s", rundir);
614
615 /* sessiond */
616 snprintf(path, PATH_MAX,
617 "%s/%s",
618 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
619 DBG("Removing %s", path);
620 (void) unlink(path);
621
622 snprintf(path, PATH_MAX, "%s/%s", rundir,
623 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
624 DBG("Removing %s", path);
625 (void) unlink(path);
626
627 /* kconsumerd */
628 snprintf(path, PATH_MAX,
629 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
630 rundir);
631 DBG("Removing %s", path);
632 (void) unlink(path);
633
634 snprintf(path, PATH_MAX,
635 DEFAULT_KCONSUMERD_PATH,
636 rundir);
637 DBG("Removing directory %s", path);
638 (void) rmdir(path);
639
640 /* ust consumerd 32 */
641 snprintf(path, PATH_MAX,
642 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
643 rundir);
644 DBG("Removing %s", path);
645 (void) unlink(path);
646
647 snprintf(path, PATH_MAX,
648 DEFAULT_USTCONSUMERD32_PATH,
649 rundir);
650 DBG("Removing directory %s", path);
651 (void) rmdir(path);
652
653 /* ust consumerd 64 */
654 snprintf(path, PATH_MAX,
655 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
656 rundir);
657 DBG("Removing %s", path);
658 (void) unlink(path);
659
660 snprintf(path, PATH_MAX,
661 DEFAULT_USTCONSUMERD64_PATH,
662 rundir);
663 DBG("Removing directory %s", path);
664 (void) rmdir(path);
665
666 DBG("Cleaning up all sessions");
667
668 /* Destroy session list mutex */
669 if (session_list_ptr != NULL) {
670 pthread_mutex_destroy(&session_list_ptr->lock);
671
672 /* Cleanup ALL session */
673 cds_list_for_each_entry_safe(sess, stmp,
674 &session_list_ptr->head, list) {
675 cmd_destroy_session(sess, kernel_poll_pipe[1]);
676 }
677 }
678
679 DBG("Closing all UST sockets");
680 ust_app_clean_list();
681 buffer_reg_destroy_registries();
682
683 if (is_root && !opt_no_kernel) {
684 DBG2("Closing kernel fd");
685 if (kernel_tracer_fd >= 0) {
686 ret = close(kernel_tracer_fd);
687 if (ret) {
688 PERROR("close");
689 }
690 }
691 DBG("Unloading kernel modules");
692 modprobe_remove_lttng_all();
693 free(syscall_table);
694 }
695
696 close_consumer_sockets();
697
698 if (load_info) {
699 load_session_destroy_data(load_info);
700 free(load_info);
701 }
702
703 /*
704 * Cleanup lock file by deleting it and finaly closing it which will
705 * release the file system lock.
706 */
707 if (lockfile_fd >= 0) {
708 char lockfile_path[PATH_MAX];
709
710 ret = generate_lock_file_path(lockfile_path,
711 sizeof(lockfile_path));
712 if (ret > 0) {
713 ret = remove(lockfile_path);
714 if (ret < 0) {
715 PERROR("remove lock file");
716 }
717 ret = close(lockfile_fd);
718 if (ret < 0) {
719 PERROR("close lock file");
720 }
721 }
722 }
723
724 /*
725 * We do NOT rmdir rundir because there are other processes
726 * using it, for instance lttng-relayd, which can start in
727 * parallel with this teardown.
728 */
729
730 free(rundir);
731 }
732
733 /*
734 * Cleanup the daemon's option data structures.
735 */
736 static void sessiond_cleanup_options(void)
737 {
738 DBG("Cleaning up options");
739
740 /*
741 * If the override option is set, the pointer points to a *non* const
742 * thus freeing it even though the variable type is set to const.
743 */
744 if (tracing_group_name_override) {
745 free((void *) tracing_group_name);
746 }
747 if (consumerd32_bin_override) {
748 free((void *) consumerd32_bin);
749 }
750 if (consumerd64_bin_override) {
751 free((void *) consumerd64_bin);
752 }
753 if (consumerd32_libdir_override) {
754 free((void *) consumerd32_libdir);
755 }
756 if (consumerd64_libdir_override) {
757 free((void *) consumerd64_libdir);
758 }
759
760 free(opt_pidfile);
761 free(opt_load_session_path);
762 free(kmod_probes_list);
763 free(kmod_extra_probes_list);
764
765 /* <fun> */
766 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
767 "Matthew, BEET driven development works!%c[%dm",
768 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
769 /* </fun> */
770 }
771
772 /*
773 * Send data on a unix socket using the liblttsessiondcomm API.
774 *
775 * Return lttcomm error code.
776 */
777 static int send_unix_sock(int sock, void *buf, size_t len)
778 {
779 /* Check valid length */
780 if (len == 0) {
781 return -1;
782 }
783
784 return lttcomm_send_unix_sock(sock, buf, len);
785 }
786
787 /*
788 * Free memory of a command context structure.
789 */
790 static void clean_command_ctx(struct command_ctx **cmd_ctx)
791 {
792 DBG("Clean command context structure");
793 if (*cmd_ctx) {
794 if ((*cmd_ctx)->llm) {
795 free((*cmd_ctx)->llm);
796 }
797 if ((*cmd_ctx)->lsm) {
798 free((*cmd_ctx)->lsm);
799 }
800 free(*cmd_ctx);
801 *cmd_ctx = NULL;
802 }
803 }
804
805 /*
806 * Notify UST applications using the shm mmap futex.
807 */
808 static int notify_ust_apps(int active)
809 {
810 char *wait_shm_mmap;
811
812 DBG("Notifying applications of session daemon state: %d", active);
813
814 /* See shm.c for this call implying mmap, shm and futex calls */
815 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
816 if (wait_shm_mmap == NULL) {
817 goto error;
818 }
819
820 /* Wake waiting process */
821 futex_wait_update((int32_t *) wait_shm_mmap, active);
822
823 /* Apps notified successfully */
824 return 0;
825
826 error:
827 return -1;
828 }
829
830 /*
831 * Setup the outgoing data buffer for the response (llm) by allocating the
832 * right amount of memory and copying the original information from the lsm
833 * structure.
834 *
835 * Return total size of the buffer pointed by buf.
836 */
837 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
838 {
839 int ret, buf_size;
840
841 buf_size = size;
842
843 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
844 if (cmd_ctx->llm == NULL) {
845 PERROR("zmalloc");
846 ret = -ENOMEM;
847 goto error;
848 }
849
850 /* Copy common data */
851 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
852 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
853
854 cmd_ctx->llm->data_size = size;
855 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
856
857 return buf_size;
858
859 error:
860 return ret;
861 }
862
863 /*
864 * Update the kernel poll set of all channel fd available over all tracing
865 * session. Add the wakeup pipe at the end of the set.
866 */
867 static int update_kernel_poll(struct lttng_poll_event *events)
868 {
869 int ret;
870 struct ltt_session *session;
871 struct ltt_kernel_channel *channel;
872
873 DBG("Updating kernel poll set");
874
875 session_lock_list();
876 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
877 session_lock(session);
878 if (session->kernel_session == NULL) {
879 session_unlock(session);
880 continue;
881 }
882
883 cds_list_for_each_entry(channel,
884 &session->kernel_session->channel_list.head, list) {
885 /* Add channel fd to the kernel poll set */
886 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
887 if (ret < 0) {
888 session_unlock(session);
889 goto error;
890 }
891 DBG("Channel fd %d added to kernel set", channel->fd);
892 }
893 session_unlock(session);
894 }
895 session_unlock_list();
896
897 return 0;
898
899 error:
900 session_unlock_list();
901 return -1;
902 }
903
904 /*
905 * Find the channel fd from 'fd' over all tracing session. When found, check
906 * for new channel stream and send those stream fds to the kernel consumer.
907 *
908 * Useful for CPU hotplug feature.
909 */
910 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
911 {
912 int ret = 0;
913 struct ltt_session *session;
914 struct ltt_kernel_session *ksess;
915 struct ltt_kernel_channel *channel;
916
917 DBG("Updating kernel streams for channel fd %d", fd);
918
919 session_lock_list();
920 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
921 session_lock(session);
922 if (session->kernel_session == NULL) {
923 session_unlock(session);
924 continue;
925 }
926 ksess = session->kernel_session;
927
928 cds_list_for_each_entry(channel,
929 &ksess->channel_list.head, list) {
930 struct lttng_ht_iter iter;
931 struct consumer_socket *socket;
932
933 if (channel->fd != fd) {
934 continue;
935 }
936 DBG("Channel found, updating kernel streams");
937 ret = kernel_open_channel_stream(channel);
938 if (ret < 0) {
939 goto error;
940 }
941 /* Update the stream global counter */
942 ksess->stream_count_global += ret;
943
944 /*
945 * Have we already sent fds to the consumer? If yes, it
946 * means that tracing is started so it is safe to send
947 * our updated stream fds.
948 */
949 if (ksess->consumer_fds_sent != 1
950 || ksess->consumer == NULL) {
951 ret = -1;
952 goto error;
953 }
954
955 rcu_read_lock();
956 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
957 &iter.iter, socket, node.node) {
958 pthread_mutex_lock(socket->lock);
959 ret = kernel_consumer_send_channel_stream(socket,
960 channel, ksess,
961 session->output_traces ? 1 : 0);
962 pthread_mutex_unlock(socket->lock);
963 if (ret < 0) {
964 rcu_read_unlock();
965 goto error;
966 }
967 }
968 rcu_read_unlock();
969 }
970 session_unlock(session);
971 }
972 session_unlock_list();
973 return ret;
974
975 error:
976 session_unlock(session);
977 session_unlock_list();
978 return ret;
979 }
980
981 /*
982 * For each tracing session, update newly registered apps. The session list
983 * lock MUST be acquired before calling this.
984 */
985 static void update_ust_app(int app_sock)
986 {
987 struct ltt_session *sess, *stmp;
988
989 /* Consumer is in an ERROR state. Stop any application update. */
990 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
991 /* Stop the update process since the consumer is dead. */
992 return;
993 }
994
995 /* For all tracing session(s) */
996 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
997 session_lock(sess);
998 if (sess->ust_session) {
999 ust_app_global_update(sess->ust_session, app_sock);
1000 }
1001 session_unlock(sess);
1002 }
1003 }
1004
1005 /*
1006 * This thread manage event coming from the kernel.
1007 *
1008 * Features supported in this thread:
1009 * -) CPU Hotplug
1010 */
1011 static void *thread_manage_kernel(void *data)
1012 {
1013 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1014 uint32_t revents, nb_fd;
1015 char tmp;
1016 struct lttng_poll_event events;
1017
1018 DBG("[thread] Thread manage kernel started");
1019
1020 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1021
1022 /*
1023 * This first step of the while is to clean this structure which could free
1024 * non NULL pointers so initialize it before the loop.
1025 */
1026 lttng_poll_init(&events);
1027
1028 if (testpoint(sessiond_thread_manage_kernel)) {
1029 goto error_testpoint;
1030 }
1031
1032 health_code_update();
1033
1034 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1035 goto error_testpoint;
1036 }
1037
1038 while (1) {
1039 health_code_update();
1040
1041 if (update_poll_flag == 1) {
1042 /* Clean events object. We are about to populate it again. */
1043 lttng_poll_clean(&events);
1044
1045 ret = sessiond_set_thread_pollset(&events, 2);
1046 if (ret < 0) {
1047 goto error_poll_create;
1048 }
1049
1050 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1051 if (ret < 0) {
1052 goto error;
1053 }
1054
1055 /* This will add the available kernel channel if any. */
1056 ret = update_kernel_poll(&events);
1057 if (ret < 0) {
1058 goto error;
1059 }
1060 update_poll_flag = 0;
1061 }
1062
1063 DBG("Thread kernel polling");
1064
1065 /* Poll infinite value of time */
1066 restart:
1067 health_poll_entry();
1068 ret = lttng_poll_wait(&events, -1);
1069 DBG("Thread kernel return from poll on %d fds",
1070 LTTNG_POLL_GETNB(&events));
1071 health_poll_exit();
1072 if (ret < 0) {
1073 /*
1074 * Restart interrupted system call.
1075 */
1076 if (errno == EINTR) {
1077 goto restart;
1078 }
1079 goto error;
1080 } else if (ret == 0) {
1081 /* Should not happen since timeout is infinite */
1082 ERR("Return value of poll is 0 with an infinite timeout.\n"
1083 "This should not have happened! Continuing...");
1084 continue;
1085 }
1086
1087 nb_fd = ret;
1088
1089 for (i = 0; i < nb_fd; i++) {
1090 /* Fetch once the poll data */
1091 revents = LTTNG_POLL_GETEV(&events, i);
1092 pollfd = LTTNG_POLL_GETFD(&events, i);
1093
1094 health_code_update();
1095
1096 if (!revents) {
1097 /* No activity for this FD (poll implementation). */
1098 continue;
1099 }
1100
1101 /* Thread quit pipe has been closed. Killing thread. */
1102 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1103 if (ret) {
1104 err = 0;
1105 goto exit;
1106 }
1107
1108 /* Check for data on kernel pipe */
1109 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1110 (void) lttng_read(kernel_poll_pipe[0],
1111 &tmp, 1);
1112 /*
1113 * Ret value is useless here, if this pipe gets any actions an
1114 * update is required anyway.
1115 */
1116 update_poll_flag = 1;
1117 continue;
1118 } else {
1119 /*
1120 * New CPU detected by the kernel. Adding kernel stream to
1121 * kernel session and updating the kernel consumer
1122 */
1123 if (revents & LPOLLIN) {
1124 ret = update_kernel_stream(&kconsumer_data, pollfd);
1125 if (ret < 0) {
1126 continue;
1127 }
1128 break;
1129 /*
1130 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1131 * and unregister kernel stream at this point.
1132 */
1133 }
1134 }
1135 }
1136 }
1137
1138 exit:
1139 error:
1140 lttng_poll_clean(&events);
1141 error_poll_create:
1142 error_testpoint:
1143 utils_close_pipe(kernel_poll_pipe);
1144 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1145 if (err) {
1146 health_error();
1147 ERR("Health error occurred in %s", __func__);
1148 WARN("Kernel thread died unexpectedly. "
1149 "Kernel tracing can continue but CPU hotplug is disabled.");
1150 }
1151 health_unregister(health_sessiond);
1152 DBG("Kernel thread dying");
1153 return NULL;
1154 }
1155
1156 /*
1157 * Signal pthread condition of the consumer data that the thread.
1158 */
1159 static void signal_consumer_condition(struct consumer_data *data, int state)
1160 {
1161 pthread_mutex_lock(&data->cond_mutex);
1162
1163 /*
1164 * The state is set before signaling. It can be any value, it's the waiter
1165 * job to correctly interpret this condition variable associated to the
1166 * consumer pthread_cond.
1167 *
1168 * A value of 0 means that the corresponding thread of the consumer data
1169 * was not started. 1 indicates that the thread has started and is ready
1170 * for action. A negative value means that there was an error during the
1171 * thread bootstrap.
1172 */
1173 data->consumer_thread_is_ready = state;
1174 (void) pthread_cond_signal(&data->cond);
1175
1176 pthread_mutex_unlock(&data->cond_mutex);
1177 }
1178
1179 /*
1180 * This thread manage the consumer error sent back to the session daemon.
1181 */
1182 static void *thread_manage_consumer(void *data)
1183 {
1184 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1185 uint32_t revents, nb_fd;
1186 enum lttcomm_return_code code;
1187 struct lttng_poll_event events;
1188 struct consumer_data *consumer_data = data;
1189
1190 DBG("[thread] Manage consumer started");
1191
1192 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1193
1194 health_code_update();
1195
1196 /*
1197 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1198 * metadata_sock. Nothing more will be added to this poll set.
1199 */
1200 ret = sessiond_set_thread_pollset(&events, 3);
1201 if (ret < 0) {
1202 goto error_poll;
1203 }
1204
1205 /*
1206 * The error socket here is already in a listening state which was done
1207 * just before spawning this thread to avoid a race between the consumer
1208 * daemon exec trying to connect and the listen() call.
1209 */
1210 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1211 if (ret < 0) {
1212 goto error;
1213 }
1214
1215 health_code_update();
1216
1217 /* Infinite blocking call, waiting for transmission */
1218 restart:
1219 health_poll_entry();
1220
1221 if (testpoint(sessiond_thread_manage_consumer)) {
1222 goto error;
1223 }
1224
1225 ret = lttng_poll_wait(&events, -1);
1226 health_poll_exit();
1227 if (ret < 0) {
1228 /*
1229 * Restart interrupted system call.
1230 */
1231 if (errno == EINTR) {
1232 goto restart;
1233 }
1234 goto error;
1235 }
1236
1237 nb_fd = ret;
1238
1239 for (i = 0; i < nb_fd; i++) {
1240 /* Fetch once the poll data */
1241 revents = LTTNG_POLL_GETEV(&events, i);
1242 pollfd = LTTNG_POLL_GETFD(&events, i);
1243
1244 health_code_update();
1245
1246 if (!revents) {
1247 /* No activity for this FD (poll implementation). */
1248 continue;
1249 }
1250
1251 /* Thread quit pipe has been closed. Killing thread. */
1252 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1253 if (ret) {
1254 err = 0;
1255 goto exit;
1256 }
1257
1258 /* Event on the registration socket */
1259 if (pollfd == consumer_data->err_sock) {
1260 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1261 ERR("consumer err socket poll error");
1262 goto error;
1263 }
1264 }
1265 }
1266
1267 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1268 if (sock < 0) {
1269 goto error;
1270 }
1271
1272 /*
1273 * Set the CLOEXEC flag. Return code is useless because either way, the
1274 * show must go on.
1275 */
1276 (void) utils_set_fd_cloexec(sock);
1277
1278 health_code_update();
1279
1280 DBG2("Receiving code from consumer err_sock");
1281
1282 /* Getting status code from kconsumerd */
1283 ret = lttcomm_recv_unix_sock(sock, &code,
1284 sizeof(enum lttcomm_return_code));
1285 if (ret <= 0) {
1286 goto error;
1287 }
1288
1289 health_code_update();
1290 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1291 /* Connect both socket, command and metadata. */
1292 consumer_data->cmd_sock =
1293 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1294 consumer_data->metadata_fd =
1295 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1296 if (consumer_data->cmd_sock < 0
1297 || consumer_data->metadata_fd < 0) {
1298 PERROR("consumer connect cmd socket");
1299 /* On error, signal condition and quit. */
1300 signal_consumer_condition(consumer_data, -1);
1301 goto error;
1302 }
1303 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1304 /* Create metadata socket lock. */
1305 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1306 if (consumer_data->metadata_sock.lock == NULL) {
1307 PERROR("zmalloc pthread mutex");
1308 ret = -1;
1309 goto error;
1310 }
1311 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1312
1313 signal_consumer_condition(consumer_data, 1);
1314 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1315 DBG("Consumer metadata socket ready (fd: %d)",
1316 consumer_data->metadata_fd);
1317 } else {
1318 ERR("consumer error when waiting for SOCK_READY : %s",
1319 lttcomm_get_readable_code(-code));
1320 goto error;
1321 }
1322
1323 /* Remove the consumerd error sock since we've established a connexion */
1324 ret = lttng_poll_del(&events, consumer_data->err_sock);
1325 if (ret < 0) {
1326 goto error;
1327 }
1328
1329 /* Add new accepted error socket. */
1330 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1331 if (ret < 0) {
1332 goto error;
1333 }
1334
1335 /* Add metadata socket that is successfully connected. */
1336 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1337 LPOLLIN | LPOLLRDHUP);
1338 if (ret < 0) {
1339 goto error;
1340 }
1341
1342 health_code_update();
1343
1344 /* Infinite blocking call, waiting for transmission */
1345 restart_poll:
1346 while (1) {
1347 health_code_update();
1348
1349 /* Exit the thread because the thread quit pipe has been triggered. */
1350 if (should_quit) {
1351 /* Not a health error. */
1352 err = 0;
1353 goto exit;
1354 }
1355
1356 health_poll_entry();
1357 ret = lttng_poll_wait(&events, -1);
1358 health_poll_exit();
1359 if (ret < 0) {
1360 /*
1361 * Restart interrupted system call.
1362 */
1363 if (errno == EINTR) {
1364 goto restart_poll;
1365 }
1366 goto error;
1367 }
1368
1369 nb_fd = ret;
1370
1371 for (i = 0; i < nb_fd; i++) {
1372 /* Fetch once the poll data */
1373 revents = LTTNG_POLL_GETEV(&events, i);
1374 pollfd = LTTNG_POLL_GETFD(&events, i);
1375
1376 health_code_update();
1377
1378 if (!revents) {
1379 /* No activity for this FD (poll implementation). */
1380 continue;
1381 }
1382
1383 /*
1384 * Thread quit pipe has been triggered, flag that we should stop
1385 * but continue the current loop to handle potential data from
1386 * consumer.
1387 */
1388 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1389
1390 if (pollfd == sock) {
1391 /* Event on the consumerd socket */
1392 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1393 ERR("consumer err socket second poll error");
1394 goto error;
1395 }
1396 health_code_update();
1397 /* Wait for any kconsumerd error */
1398 ret = lttcomm_recv_unix_sock(sock, &code,
1399 sizeof(enum lttcomm_return_code));
1400 if (ret <= 0) {
1401 ERR("consumer closed the command socket");
1402 goto error;
1403 }
1404
1405 ERR("consumer return code : %s",
1406 lttcomm_get_readable_code(-code));
1407
1408 goto exit;
1409 } else if (pollfd == consumer_data->metadata_fd) {
1410 /* UST metadata requests */
1411 ret = ust_consumer_metadata_request(
1412 &consumer_data->metadata_sock);
1413 if (ret < 0) {
1414 ERR("Handling metadata request");
1415 goto error;
1416 }
1417 }
1418 /* No need for an else branch all FDs are tested prior. */
1419 }
1420 health_code_update();
1421 }
1422
1423 exit:
1424 error:
1425 /*
1426 * We lock here because we are about to close the sockets and some other
1427 * thread might be using them so get exclusive access which will abort all
1428 * other consumer command by other threads.
1429 */
1430 pthread_mutex_lock(&consumer_data->lock);
1431
1432 /* Immediately set the consumerd state to stopped */
1433 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1434 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1435 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1436 consumer_data->type == LTTNG_CONSUMER32_UST) {
1437 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1438 } else {
1439 /* Code flow error... */
1440 assert(0);
1441 }
1442
1443 if (consumer_data->err_sock >= 0) {
1444 ret = close(consumer_data->err_sock);
1445 if (ret) {
1446 PERROR("close");
1447 }
1448 consumer_data->err_sock = -1;
1449 }
1450 if (consumer_data->cmd_sock >= 0) {
1451 ret = close(consumer_data->cmd_sock);
1452 if (ret) {
1453 PERROR("close");
1454 }
1455 consumer_data->cmd_sock = -1;
1456 }
1457 if (consumer_data->metadata_sock.fd_ptr &&
1458 *consumer_data->metadata_sock.fd_ptr >= 0) {
1459 ret = close(*consumer_data->metadata_sock.fd_ptr);
1460 if (ret) {
1461 PERROR("close");
1462 }
1463 }
1464 if (sock >= 0) {
1465 ret = close(sock);
1466 if (ret) {
1467 PERROR("close");
1468 }
1469 }
1470
1471 unlink(consumer_data->err_unix_sock_path);
1472 unlink(consumer_data->cmd_unix_sock_path);
1473 consumer_data->pid = 0;
1474 pthread_mutex_unlock(&consumer_data->lock);
1475
1476 /* Cleanup metadata socket mutex. */
1477 if (consumer_data->metadata_sock.lock) {
1478 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1479 free(consumer_data->metadata_sock.lock);
1480 }
1481 lttng_poll_clean(&events);
1482 error_poll:
1483 if (err) {
1484 health_error();
1485 ERR("Health error occurred in %s", __func__);
1486 }
1487 health_unregister(health_sessiond);
1488 DBG("consumer thread cleanup completed");
1489
1490 return NULL;
1491 }
1492
1493 /*
1494 * This thread manage application communication.
1495 */
1496 static void *thread_manage_apps(void *data)
1497 {
1498 int i, ret, pollfd, err = -1;
1499 ssize_t size_ret;
1500 uint32_t revents, nb_fd;
1501 struct lttng_poll_event events;
1502
1503 DBG("[thread] Manage application started");
1504
1505 rcu_register_thread();
1506 rcu_thread_online();
1507
1508 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1509
1510 if (testpoint(sessiond_thread_manage_apps)) {
1511 goto error_testpoint;
1512 }
1513
1514 health_code_update();
1515
1516 ret = sessiond_set_thread_pollset(&events, 2);
1517 if (ret < 0) {
1518 goto error_poll_create;
1519 }
1520
1521 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1522 if (ret < 0) {
1523 goto error;
1524 }
1525
1526 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1527 goto error;
1528 }
1529
1530 health_code_update();
1531
1532 while (1) {
1533 DBG("Apps thread polling");
1534
1535 /* Inifinite blocking call, waiting for transmission */
1536 restart:
1537 health_poll_entry();
1538 ret = lttng_poll_wait(&events, -1);
1539 DBG("Apps thread return from poll on %d fds",
1540 LTTNG_POLL_GETNB(&events));
1541 health_poll_exit();
1542 if (ret < 0) {
1543 /*
1544 * Restart interrupted system call.
1545 */
1546 if (errno == EINTR) {
1547 goto restart;
1548 }
1549 goto error;
1550 }
1551
1552 nb_fd = ret;
1553
1554 for (i = 0; i < nb_fd; i++) {
1555 /* Fetch once the poll data */
1556 revents = LTTNG_POLL_GETEV(&events, i);
1557 pollfd = LTTNG_POLL_GETFD(&events, i);
1558
1559 health_code_update();
1560
1561 if (!revents) {
1562 /* No activity for this FD (poll implementation). */
1563 continue;
1564 }
1565
1566 /* Thread quit pipe has been closed. Killing thread. */
1567 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1568 if (ret) {
1569 err = 0;
1570 goto exit;
1571 }
1572
1573 /* Inspect the apps cmd pipe */
1574 if (pollfd == apps_cmd_pipe[0]) {
1575 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1576 ERR("Apps command pipe error");
1577 goto error;
1578 } else if (revents & LPOLLIN) {
1579 int sock;
1580
1581 /* Empty pipe */
1582 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1583 if (size_ret < sizeof(sock)) {
1584 PERROR("read apps cmd pipe");
1585 goto error;
1586 }
1587
1588 health_code_update();
1589
1590 /*
1591 * We only monitor the error events of the socket. This
1592 * thread does not handle any incoming data from UST
1593 * (POLLIN).
1594 */
1595 ret = lttng_poll_add(&events, sock,
1596 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1597 if (ret < 0) {
1598 goto error;
1599 }
1600
1601 DBG("Apps with sock %d added to poll set", sock);
1602 }
1603 } else {
1604 /*
1605 * At this point, we know that a registered application made
1606 * the event at poll_wait.
1607 */
1608 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1609 /* Removing from the poll set */
1610 ret = lttng_poll_del(&events, pollfd);
1611 if (ret < 0) {
1612 goto error;
1613 }
1614
1615 /* Socket closed on remote end. */
1616 ust_app_unregister(pollfd);
1617 }
1618 }
1619
1620 health_code_update();
1621 }
1622 }
1623
1624 exit:
1625 error:
1626 lttng_poll_clean(&events);
1627 error_poll_create:
1628 error_testpoint:
1629 utils_close_pipe(apps_cmd_pipe);
1630 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1631
1632 /*
1633 * We don't clean the UST app hash table here since already registered
1634 * applications can still be controlled so let them be until the session
1635 * daemon dies or the applications stop.
1636 */
1637
1638 if (err) {
1639 health_error();
1640 ERR("Health error occurred in %s", __func__);
1641 }
1642 health_unregister(health_sessiond);
1643 DBG("Application communication apps thread cleanup complete");
1644 rcu_thread_offline();
1645 rcu_unregister_thread();
1646 return NULL;
1647 }
1648
1649 /*
1650 * Send a socket to a thread This is called from the dispatch UST registration
1651 * thread once all sockets are set for the application.
1652 *
1653 * The sock value can be invalid, we don't really care, the thread will handle
1654 * it and make the necessary cleanup if so.
1655 *
1656 * On success, return 0 else a negative value being the errno message of the
1657 * write().
1658 */
1659 static int send_socket_to_thread(int fd, int sock)
1660 {
1661 ssize_t ret;
1662
1663 /*
1664 * It's possible that the FD is set as invalid with -1 concurrently just
1665 * before calling this function being a shutdown state of the thread.
1666 */
1667 if (fd < 0) {
1668 ret = -EBADF;
1669 goto error;
1670 }
1671
1672 ret = lttng_write(fd, &sock, sizeof(sock));
1673 if (ret < sizeof(sock)) {
1674 PERROR("write apps pipe %d", fd);
1675 if (ret < 0) {
1676 ret = -errno;
1677 }
1678 goto error;
1679 }
1680
1681 /* All good. Don't send back the write positive ret value. */
1682 ret = 0;
1683 error:
1684 return (int) ret;
1685 }
1686
1687 /*
1688 * Sanitize the wait queue of the dispatch registration thread meaning removing
1689 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1690 * notify socket is never received.
1691 */
1692 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1693 {
1694 int ret, nb_fd = 0, i;
1695 unsigned int fd_added = 0;
1696 struct lttng_poll_event events;
1697 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1698
1699 assert(wait_queue);
1700
1701 lttng_poll_init(&events);
1702
1703 /* Just skip everything for an empty queue. */
1704 if (!wait_queue->count) {
1705 goto end;
1706 }
1707
1708 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1709 if (ret < 0) {
1710 goto error_create;
1711 }
1712
1713 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1714 &wait_queue->head, head) {
1715 assert(wait_node->app);
1716 ret = lttng_poll_add(&events, wait_node->app->sock,
1717 LPOLLHUP | LPOLLERR);
1718 if (ret < 0) {
1719 goto error;
1720 }
1721
1722 fd_added = 1;
1723 }
1724
1725 if (!fd_added) {
1726 goto end;
1727 }
1728
1729 /*
1730 * Poll but don't block so we can quickly identify the faulty events and
1731 * clean them afterwards from the wait queue.
1732 */
1733 ret = lttng_poll_wait(&events, 0);
1734 if (ret < 0) {
1735 goto error;
1736 }
1737 nb_fd = ret;
1738
1739 for (i = 0; i < nb_fd; i++) {
1740 /* Get faulty FD. */
1741 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1742 int pollfd = LTTNG_POLL_GETFD(&events, i);
1743
1744 if (!revents) {
1745 /* No activity for this FD (poll implementation). */
1746 continue;
1747 }
1748
1749 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1750 &wait_queue->head, head) {
1751 if (pollfd == wait_node->app->sock &&
1752 (revents & (LPOLLHUP | LPOLLERR))) {
1753 cds_list_del(&wait_node->head);
1754 wait_queue->count--;
1755 ust_app_destroy(wait_node->app);
1756 free(wait_node);
1757 break;
1758 }
1759 }
1760 }
1761
1762 if (nb_fd > 0) {
1763 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1764 }
1765
1766 end:
1767 lttng_poll_clean(&events);
1768 return;
1769
1770 error:
1771 lttng_poll_clean(&events);
1772 error_create:
1773 ERR("Unable to sanitize wait queue");
1774 return;
1775 }
1776
1777 /*
1778 * Dispatch request from the registration threads to the application
1779 * communication thread.
1780 */
1781 static void *thread_dispatch_ust_registration(void *data)
1782 {
1783 int ret, err = -1;
1784 struct cds_wfcq_node *node;
1785 struct ust_command *ust_cmd = NULL;
1786 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1787 struct ust_reg_wait_queue wait_queue = {
1788 .count = 0,
1789 };
1790
1791 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1792
1793 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1794 goto error_testpoint;
1795 }
1796
1797 health_code_update();
1798
1799 CDS_INIT_LIST_HEAD(&wait_queue.head);
1800
1801 DBG("[thread] Dispatch UST command started");
1802
1803 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1804 health_code_update();
1805
1806 /* Atomically prepare the queue futex */
1807 futex_nto1_prepare(&ust_cmd_queue.futex);
1808
1809 do {
1810 struct ust_app *app = NULL;
1811 ust_cmd = NULL;
1812
1813 /*
1814 * Make sure we don't have node(s) that have hung up before receiving
1815 * the notify socket. This is to clean the list in order to avoid
1816 * memory leaks from notify socket that are never seen.
1817 */
1818 sanitize_wait_queue(&wait_queue);
1819
1820 health_code_update();
1821 /* Dequeue command for registration */
1822 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1823 if (node == NULL) {
1824 DBG("Woken up but nothing in the UST command queue");
1825 /* Continue thread execution */
1826 break;
1827 }
1828
1829 ust_cmd = caa_container_of(node, struct ust_command, node);
1830
1831 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1832 " gid:%d sock:%d name:%s (version %d.%d)",
1833 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1834 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1835 ust_cmd->sock, ust_cmd->reg_msg.name,
1836 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1837
1838 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1839 wait_node = zmalloc(sizeof(*wait_node));
1840 if (!wait_node) {
1841 PERROR("zmalloc wait_node dispatch");
1842 ret = close(ust_cmd->sock);
1843 if (ret < 0) {
1844 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1845 }
1846 lttng_fd_put(LTTNG_FD_APPS, 1);
1847 free(ust_cmd);
1848 goto error;
1849 }
1850 CDS_INIT_LIST_HEAD(&wait_node->head);
1851
1852 /* Create application object if socket is CMD. */
1853 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1854 ust_cmd->sock);
1855 if (!wait_node->app) {
1856 ret = close(ust_cmd->sock);
1857 if (ret < 0) {
1858 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1859 }
1860 lttng_fd_put(LTTNG_FD_APPS, 1);
1861 free(wait_node);
1862 free(ust_cmd);
1863 continue;
1864 }
1865 /*
1866 * Add application to the wait queue so we can set the notify
1867 * socket before putting this object in the global ht.
1868 */
1869 cds_list_add(&wait_node->head, &wait_queue.head);
1870 wait_queue.count++;
1871
1872 free(ust_cmd);
1873 /*
1874 * We have to continue here since we don't have the notify
1875 * socket and the application MUST be added to the hash table
1876 * only at that moment.
1877 */
1878 continue;
1879 } else {
1880 /*
1881 * Look for the application in the local wait queue and set the
1882 * notify socket if found.
1883 */
1884 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1885 &wait_queue.head, head) {
1886 health_code_update();
1887 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1888 wait_node->app->notify_sock = ust_cmd->sock;
1889 cds_list_del(&wait_node->head);
1890 wait_queue.count--;
1891 app = wait_node->app;
1892 free(wait_node);
1893 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1894 break;
1895 }
1896 }
1897
1898 /*
1899 * With no application at this stage the received socket is
1900 * basically useless so close it before we free the cmd data
1901 * structure for good.
1902 */
1903 if (!app) {
1904 ret = close(ust_cmd->sock);
1905 if (ret < 0) {
1906 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1907 }
1908 lttng_fd_put(LTTNG_FD_APPS, 1);
1909 }
1910 free(ust_cmd);
1911 }
1912
1913 if (app) {
1914 /*
1915 * @session_lock_list
1916 *
1917 * Lock the global session list so from the register up to the
1918 * registration done message, no thread can see the application
1919 * and change its state.
1920 */
1921 session_lock_list();
1922 rcu_read_lock();
1923
1924 /*
1925 * Add application to the global hash table. This needs to be
1926 * done before the update to the UST registry can locate the
1927 * application.
1928 */
1929 ust_app_add(app);
1930
1931 /* Set app version. This call will print an error if needed. */
1932 (void) ust_app_version(app);
1933
1934 /* Send notify socket through the notify pipe. */
1935 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1936 app->notify_sock);
1937 if (ret < 0) {
1938 rcu_read_unlock();
1939 session_unlock_list();
1940 /*
1941 * No notify thread, stop the UST tracing. However, this is
1942 * not an internal error of the this thread thus setting
1943 * the health error code to a normal exit.
1944 */
1945 err = 0;
1946 goto error;
1947 }
1948
1949 /*
1950 * Update newly registered application with the tracing
1951 * registry info already enabled information.
1952 */
1953 update_ust_app(app->sock);
1954
1955 /*
1956 * Don't care about return value. Let the manage apps threads
1957 * handle app unregistration upon socket close.
1958 */
1959 (void) ust_app_register_done(app->sock);
1960
1961 /*
1962 * Even if the application socket has been closed, send the app
1963 * to the thread and unregistration will take place at that
1964 * place.
1965 */
1966 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1967 if (ret < 0) {
1968 rcu_read_unlock();
1969 session_unlock_list();
1970 /*
1971 * No apps. thread, stop the UST tracing. However, this is
1972 * not an internal error of the this thread thus setting
1973 * the health error code to a normal exit.
1974 */
1975 err = 0;
1976 goto error;
1977 }
1978
1979 rcu_read_unlock();
1980 session_unlock_list();
1981 }
1982 } while (node != NULL);
1983
1984 health_poll_entry();
1985 /* Futex wait on queue. Blocking call on futex() */
1986 futex_nto1_wait(&ust_cmd_queue.futex);
1987 health_poll_exit();
1988 }
1989 /* Normal exit, no error */
1990 err = 0;
1991
1992 error:
1993 /* Clean up wait queue. */
1994 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1995 &wait_queue.head, head) {
1996 cds_list_del(&wait_node->head);
1997 wait_queue.count--;
1998 free(wait_node);
1999 }
2000
2001 error_testpoint:
2002 DBG("Dispatch thread dying");
2003 if (err) {
2004 health_error();
2005 ERR("Health error occurred in %s", __func__);
2006 }
2007 health_unregister(health_sessiond);
2008 return NULL;
2009 }
2010
2011 /*
2012 * This thread manage application registration.
2013 */
2014 static void *thread_registration_apps(void *data)
2015 {
2016 int sock = -1, i, ret, pollfd, err = -1;
2017 uint32_t revents, nb_fd;
2018 struct lttng_poll_event events;
2019 /*
2020 * Get allocated in this thread, enqueued to a global queue, dequeued and
2021 * freed in the manage apps thread.
2022 */
2023 struct ust_command *ust_cmd = NULL;
2024
2025 DBG("[thread] Manage application registration started");
2026
2027 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2028
2029 if (testpoint(sessiond_thread_registration_apps)) {
2030 goto error_testpoint;
2031 }
2032
2033 ret = lttcomm_listen_unix_sock(apps_sock);
2034 if (ret < 0) {
2035 goto error_listen;
2036 }
2037
2038 /*
2039 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2040 * more will be added to this poll set.
2041 */
2042 ret = sessiond_set_thread_pollset(&events, 2);
2043 if (ret < 0) {
2044 goto error_create_poll;
2045 }
2046
2047 /* Add the application registration socket */
2048 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2049 if (ret < 0) {
2050 goto error_poll_add;
2051 }
2052
2053 /* Notify all applications to register */
2054 ret = notify_ust_apps(1);
2055 if (ret < 0) {
2056 ERR("Failed to notify applications or create the wait shared memory.\n"
2057 "Execution continues but there might be problem for already\n"
2058 "running applications that wishes to register.");
2059 }
2060
2061 while (1) {
2062 DBG("Accepting application registration");
2063
2064 /* Inifinite blocking call, waiting for transmission */
2065 restart:
2066 health_poll_entry();
2067 ret = lttng_poll_wait(&events, -1);
2068 health_poll_exit();
2069 if (ret < 0) {
2070 /*
2071 * Restart interrupted system call.
2072 */
2073 if (errno == EINTR) {
2074 goto restart;
2075 }
2076 goto error;
2077 }
2078
2079 nb_fd = ret;
2080
2081 for (i = 0; i < nb_fd; i++) {
2082 health_code_update();
2083
2084 /* Fetch once the poll data */
2085 revents = LTTNG_POLL_GETEV(&events, i);
2086 pollfd = LTTNG_POLL_GETFD(&events, i);
2087
2088 if (!revents) {
2089 /* No activity for this FD (poll implementation). */
2090 continue;
2091 }
2092
2093 /* Thread quit pipe has been closed. Killing thread. */
2094 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2095 if (ret) {
2096 err = 0;
2097 goto exit;
2098 }
2099
2100 /* Event on the registration socket */
2101 if (pollfd == apps_sock) {
2102 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2103 ERR("Register apps socket poll error");
2104 goto error;
2105 } else if (revents & LPOLLIN) {
2106 sock = lttcomm_accept_unix_sock(apps_sock);
2107 if (sock < 0) {
2108 goto error;
2109 }
2110
2111 /*
2112 * Set socket timeout for both receiving and ending.
2113 * app_socket_timeout is in seconds, whereas
2114 * lttcomm_setsockopt_rcv_timeout and
2115 * lttcomm_setsockopt_snd_timeout expect msec as
2116 * parameter.
2117 */
2118 (void) lttcomm_setsockopt_rcv_timeout(sock,
2119 app_socket_timeout * 1000);
2120 (void) lttcomm_setsockopt_snd_timeout(sock,
2121 app_socket_timeout * 1000);
2122
2123 /*
2124 * Set the CLOEXEC flag. Return code is useless because
2125 * either way, the show must go on.
2126 */
2127 (void) utils_set_fd_cloexec(sock);
2128
2129 /* Create UST registration command for enqueuing */
2130 ust_cmd = zmalloc(sizeof(struct ust_command));
2131 if (ust_cmd == NULL) {
2132 PERROR("ust command zmalloc");
2133 goto error;
2134 }
2135
2136 /*
2137 * Using message-based transmissions to ensure we don't
2138 * have to deal with partially received messages.
2139 */
2140 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2141 if (ret < 0) {
2142 ERR("Exhausted file descriptors allowed for applications.");
2143 free(ust_cmd);
2144 ret = close(sock);
2145 if (ret) {
2146 PERROR("close");
2147 }
2148 sock = -1;
2149 continue;
2150 }
2151
2152 health_code_update();
2153 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2154 if (ret < 0) {
2155 free(ust_cmd);
2156 /* Close socket of the application. */
2157 ret = close(sock);
2158 if (ret) {
2159 PERROR("close");
2160 }
2161 lttng_fd_put(LTTNG_FD_APPS, 1);
2162 sock = -1;
2163 continue;
2164 }
2165 health_code_update();
2166
2167 ust_cmd->sock = sock;
2168 sock = -1;
2169
2170 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2171 " gid:%d sock:%d name:%s (version %d.%d)",
2172 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2173 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2174 ust_cmd->sock, ust_cmd->reg_msg.name,
2175 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2176
2177 /*
2178 * Lock free enqueue the registration request. The red pill
2179 * has been taken! This apps will be part of the *system*.
2180 */
2181 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2182
2183 /*
2184 * Wake the registration queue futex. Implicit memory
2185 * barrier with the exchange in cds_wfcq_enqueue.
2186 */
2187 futex_nto1_wake(&ust_cmd_queue.futex);
2188 }
2189 }
2190 }
2191 }
2192
2193 exit:
2194 error:
2195 /* Notify that the registration thread is gone */
2196 notify_ust_apps(0);
2197
2198 if (apps_sock >= 0) {
2199 ret = close(apps_sock);
2200 if (ret) {
2201 PERROR("close");
2202 }
2203 }
2204 if (sock >= 0) {
2205 ret = close(sock);
2206 if (ret) {
2207 PERROR("close");
2208 }
2209 lttng_fd_put(LTTNG_FD_APPS, 1);
2210 }
2211 unlink(apps_unix_sock_path);
2212
2213 error_poll_add:
2214 lttng_poll_clean(&events);
2215 error_listen:
2216 error_create_poll:
2217 error_testpoint:
2218 DBG("UST Registration thread cleanup complete");
2219 if (err) {
2220 health_error();
2221 ERR("Health error occurred in %s", __func__);
2222 }
2223 health_unregister(health_sessiond);
2224
2225 return NULL;
2226 }
2227
2228 /*
2229 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2230 * exec or it will fails.
2231 */
2232 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2233 {
2234 int ret, clock_ret;
2235 struct timespec timeout;
2236
2237 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2238 consumer_data->consumer_thread_is_ready = 0;
2239
2240 /* Setup pthread condition */
2241 ret = pthread_condattr_init(&consumer_data->condattr);
2242 if (ret) {
2243 errno = ret;
2244 PERROR("pthread_condattr_init consumer data");
2245 goto error;
2246 }
2247
2248 /*
2249 * Set the monotonic clock in order to make sure we DO NOT jump in time
2250 * between the clock_gettime() call and the timedwait call. See bug #324
2251 * for a more details and how we noticed it.
2252 */
2253 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2254 if (ret) {
2255 errno = ret;
2256 PERROR("pthread_condattr_setclock consumer data");
2257 goto error;
2258 }
2259
2260 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2261 if (ret) {
2262 errno = ret;
2263 PERROR("pthread_cond_init consumer data");
2264 goto error;
2265 }
2266
2267 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2268 consumer_data);
2269 if (ret) {
2270 errno = ret;
2271 PERROR("pthread_create consumer");
2272 ret = -1;
2273 goto error;
2274 }
2275
2276 /* We are about to wait on a pthread condition */
2277 pthread_mutex_lock(&consumer_data->cond_mutex);
2278
2279 /* Get time for sem_timedwait absolute timeout */
2280 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2281 /*
2282 * Set the timeout for the condition timed wait even if the clock gettime
2283 * call fails since we might loop on that call and we want to avoid to
2284 * increment the timeout too many times.
2285 */
2286 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2287
2288 /*
2289 * The following loop COULD be skipped in some conditions so this is why we
2290 * set ret to 0 in order to make sure at least one round of the loop is
2291 * done.
2292 */
2293 ret = 0;
2294
2295 /*
2296 * Loop until the condition is reached or when a timeout is reached. Note
2297 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2298 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2299 * possible. This loop does not take any chances and works with both of
2300 * them.
2301 */
2302 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2303 if (clock_ret < 0) {
2304 PERROR("clock_gettime spawn consumer");
2305 /* Infinite wait for the consumerd thread to be ready */
2306 ret = pthread_cond_wait(&consumer_data->cond,
2307 &consumer_data->cond_mutex);
2308 } else {
2309 ret = pthread_cond_timedwait(&consumer_data->cond,
2310 &consumer_data->cond_mutex, &timeout);
2311 }
2312 }
2313
2314 /* Release the pthread condition */
2315 pthread_mutex_unlock(&consumer_data->cond_mutex);
2316
2317 if (ret != 0) {
2318 errno = ret;
2319 if (ret == ETIMEDOUT) {
2320 int pth_ret;
2321
2322 /*
2323 * Call has timed out so we kill the kconsumerd_thread and return
2324 * an error.
2325 */
2326 ERR("Condition timed out. The consumer thread was never ready."
2327 " Killing it");
2328 pth_ret = pthread_cancel(consumer_data->thread);
2329 if (pth_ret < 0) {
2330 PERROR("pthread_cancel consumer thread");
2331 }
2332 } else {
2333 PERROR("pthread_cond_wait failed consumer thread");
2334 }
2335 /* Caller is expecting a negative value on failure. */
2336 ret = -1;
2337 goto error;
2338 }
2339
2340 pthread_mutex_lock(&consumer_data->pid_mutex);
2341 if (consumer_data->pid == 0) {
2342 ERR("Consumerd did not start");
2343 pthread_mutex_unlock(&consumer_data->pid_mutex);
2344 goto error;
2345 }
2346 pthread_mutex_unlock(&consumer_data->pid_mutex);
2347
2348 return 0;
2349
2350 error:
2351 return ret;
2352 }
2353
2354 /*
2355 * Join consumer thread
2356 */
2357 static int join_consumer_thread(struct consumer_data *consumer_data)
2358 {
2359 void *status;
2360
2361 /* Consumer pid must be a real one. */
2362 if (consumer_data->pid > 0) {
2363 int ret;
2364 ret = kill(consumer_data->pid, SIGTERM);
2365 if (ret) {
2366 PERROR("Error killing consumer daemon");
2367 return ret;
2368 }
2369 return pthread_join(consumer_data->thread, &status);
2370 } else {
2371 return 0;
2372 }
2373 }
2374
2375 /*
2376 * Fork and exec a consumer daemon (consumerd).
2377 *
2378 * Return pid if successful else -1.
2379 */
2380 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2381 {
2382 int ret;
2383 pid_t pid;
2384 const char *consumer_to_use;
2385 const char *verbosity;
2386 struct stat st;
2387
2388 DBG("Spawning consumerd");
2389
2390 pid = fork();
2391 if (pid == 0) {
2392 /*
2393 * Exec consumerd.
2394 */
2395 if (opt_verbose_consumer) {
2396 verbosity = "--verbose";
2397 } else if (lttng_opt_quiet) {
2398 verbosity = "--quiet";
2399 } else {
2400 verbosity = "";
2401 }
2402
2403 switch (consumer_data->type) {
2404 case LTTNG_CONSUMER_KERNEL:
2405 /*
2406 * Find out which consumerd to execute. We will first try the
2407 * 64-bit path, then the sessiond's installation directory, and
2408 * fallback on the 32-bit one,
2409 */
2410 DBG3("Looking for a kernel consumer at these locations:");
2411 DBG3(" 1) %s", consumerd64_bin);
2412 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2413 DBG3(" 3) %s", consumerd32_bin);
2414 if (stat(consumerd64_bin, &st) == 0) {
2415 DBG3("Found location #1");
2416 consumer_to_use = consumerd64_bin;
2417 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2418 DBG3("Found location #2");
2419 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2420 } else if (stat(consumerd32_bin, &st) == 0) {
2421 DBG3("Found location #3");
2422 consumer_to_use = consumerd32_bin;
2423 } else {
2424 DBG("Could not find any valid consumerd executable");
2425 ret = -EINVAL;
2426 break;
2427 }
2428 DBG("Using kernel consumer at: %s", consumer_to_use);
2429 ret = execl(consumer_to_use,
2430 "lttng-consumerd", verbosity, "-k",
2431 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2432 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2433 "--group", tracing_group_name,
2434 NULL);
2435 break;
2436 case LTTNG_CONSUMER64_UST:
2437 {
2438 char *tmpnew = NULL;
2439
2440 if (consumerd64_libdir[0] != '\0') {
2441 char *tmp;
2442 size_t tmplen;
2443
2444 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2445 if (!tmp) {
2446 tmp = "";
2447 }
2448 tmplen = strlen("LD_LIBRARY_PATH=")
2449 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2450 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2451 if (!tmpnew) {
2452 ret = -ENOMEM;
2453 goto error;
2454 }
2455 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2456 strcat(tmpnew, consumerd64_libdir);
2457 if (tmp[0] != '\0') {
2458 strcat(tmpnew, ":");
2459 strcat(tmpnew, tmp);
2460 }
2461 ret = putenv(tmpnew);
2462 if (ret) {
2463 ret = -errno;
2464 free(tmpnew);
2465 goto error;
2466 }
2467 }
2468 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2469 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2470 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2471 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2472 "--group", tracing_group_name,
2473 NULL);
2474 if (consumerd64_libdir[0] != '\0') {
2475 free(tmpnew);
2476 }
2477 break;
2478 }
2479 case LTTNG_CONSUMER32_UST:
2480 {
2481 char *tmpnew = NULL;
2482
2483 if (consumerd32_libdir[0] != '\0') {
2484 char *tmp;
2485 size_t tmplen;
2486
2487 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2488 if (!tmp) {
2489 tmp = "";
2490 }
2491 tmplen = strlen("LD_LIBRARY_PATH=")
2492 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2493 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2494 if (!tmpnew) {
2495 ret = -ENOMEM;
2496 goto error;
2497 }
2498 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2499 strcat(tmpnew, consumerd32_libdir);
2500 if (tmp[0] != '\0') {
2501 strcat(tmpnew, ":");
2502 strcat(tmpnew, tmp);
2503 }
2504 ret = putenv(tmpnew);
2505 if (ret) {
2506 ret = -errno;
2507 free(tmpnew);
2508 goto error;
2509 }
2510 }
2511 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2512 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2513 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2514 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2515 "--group", tracing_group_name,
2516 NULL);
2517 if (consumerd32_libdir[0] != '\0') {
2518 free(tmpnew);
2519 }
2520 break;
2521 }
2522 default:
2523 PERROR("unknown consumer type");
2524 exit(EXIT_FAILURE);
2525 }
2526 if (errno != 0) {
2527 PERROR("Consumer execl()");
2528 }
2529 /* Reaching this point, we got a failure on our execl(). */
2530 exit(EXIT_FAILURE);
2531 } else if (pid > 0) {
2532 ret = pid;
2533 } else {
2534 PERROR("start consumer fork");
2535 ret = -errno;
2536 }
2537 error:
2538 return ret;
2539 }
2540
2541 /*
2542 * Spawn the consumerd daemon and session daemon thread.
2543 */
2544 static int start_consumerd(struct consumer_data *consumer_data)
2545 {
2546 int ret;
2547
2548 /*
2549 * Set the listen() state on the socket since there is a possible race
2550 * between the exec() of the consumer daemon and this call if place in the
2551 * consumer thread. See bug #366 for more details.
2552 */
2553 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2554 if (ret < 0) {
2555 goto error;
2556 }
2557
2558 pthread_mutex_lock(&consumer_data->pid_mutex);
2559 if (consumer_data->pid != 0) {
2560 pthread_mutex_unlock(&consumer_data->pid_mutex);
2561 goto end;
2562 }
2563
2564 ret = spawn_consumerd(consumer_data);
2565 if (ret < 0) {
2566 ERR("Spawning consumerd failed");
2567 pthread_mutex_unlock(&consumer_data->pid_mutex);
2568 goto error;
2569 }
2570
2571 /* Setting up the consumer_data pid */
2572 consumer_data->pid = ret;
2573 DBG2("Consumer pid %d", consumer_data->pid);
2574 pthread_mutex_unlock(&consumer_data->pid_mutex);
2575
2576 DBG2("Spawning consumer control thread");
2577 ret = spawn_consumer_thread(consumer_data);
2578 if (ret < 0) {
2579 ERR("Fatal error spawning consumer control thread");
2580 goto error;
2581 }
2582
2583 end:
2584 return 0;
2585
2586 error:
2587 /* Cleanup already created sockets on error. */
2588 if (consumer_data->err_sock >= 0) {
2589 int err;
2590
2591 err = close(consumer_data->err_sock);
2592 if (err < 0) {
2593 PERROR("close consumer data error socket");
2594 }
2595 }
2596 return ret;
2597 }
2598
2599 /*
2600 * Setup necessary data for kernel tracer action.
2601 */
2602 static int init_kernel_tracer(void)
2603 {
2604 int ret;
2605
2606 /* Modprobe lttng kernel modules */
2607 ret = modprobe_lttng_control();
2608 if (ret < 0) {
2609 goto error;
2610 }
2611
2612 /* Open debugfs lttng */
2613 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2614 if (kernel_tracer_fd < 0) {
2615 DBG("Failed to open %s", module_proc_lttng);
2616 ret = -1;
2617 goto error_open;
2618 }
2619
2620 /* Validate kernel version */
2621 ret = kernel_validate_version(kernel_tracer_fd);
2622 if (ret < 0) {
2623 goto error_version;
2624 }
2625
2626 ret = modprobe_lttng_data();
2627 if (ret < 0) {
2628 goto error_modules;
2629 }
2630
2631 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2632 return 0;
2633
2634 error_version:
2635 modprobe_remove_lttng_control();
2636 ret = close(kernel_tracer_fd);
2637 if (ret) {
2638 PERROR("close");
2639 }
2640 kernel_tracer_fd = -1;
2641 return LTTNG_ERR_KERN_VERSION;
2642
2643 error_modules:
2644 ret = close(kernel_tracer_fd);
2645 if (ret) {
2646 PERROR("close");
2647 }
2648
2649 error_open:
2650 modprobe_remove_lttng_control();
2651
2652 error:
2653 WARN("No kernel tracer available");
2654 kernel_tracer_fd = -1;
2655 if (!is_root) {
2656 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2657 } else {
2658 return LTTNG_ERR_KERN_NA;
2659 }
2660 }
2661
2662
2663 /*
2664 * Copy consumer output from the tracing session to the domain session. The
2665 * function also applies the right modification on a per domain basis for the
2666 * trace files destination directory.
2667 *
2668 * Should *NOT* be called with RCU read-side lock held.
2669 */
2670 static int copy_session_consumer(int domain, struct ltt_session *session)
2671 {
2672 int ret;
2673 const char *dir_name;
2674 struct consumer_output *consumer;
2675
2676 assert(session);
2677 assert(session->consumer);
2678
2679 switch (domain) {
2680 case LTTNG_DOMAIN_KERNEL:
2681 DBG3("Copying tracing session consumer output in kernel session");
2682 /*
2683 * XXX: We should audit the session creation and what this function
2684 * does "extra" in order to avoid a destroy since this function is used
2685 * in the domain session creation (kernel and ust) only. Same for UST
2686 * domain.
2687 */
2688 if (session->kernel_session->consumer) {
2689 consumer_destroy_output(session->kernel_session->consumer);
2690 }
2691 session->kernel_session->consumer =
2692 consumer_copy_output(session->consumer);
2693 /* Ease our life a bit for the next part */
2694 consumer = session->kernel_session->consumer;
2695 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2696 break;
2697 case LTTNG_DOMAIN_JUL:
2698 case LTTNG_DOMAIN_LOG4J:
2699 case LTTNG_DOMAIN_PYTHON:
2700 case LTTNG_DOMAIN_UST:
2701 DBG3("Copying tracing session consumer output in UST session");
2702 if (session->ust_session->consumer) {
2703 consumer_destroy_output(session->ust_session->consumer);
2704 }
2705 session->ust_session->consumer =
2706 consumer_copy_output(session->consumer);
2707 /* Ease our life a bit for the next part */
2708 consumer = session->ust_session->consumer;
2709 dir_name = DEFAULT_UST_TRACE_DIR;
2710 break;
2711 default:
2712 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2713 goto error;
2714 }
2715
2716 /* Append correct directory to subdir */
2717 strncat(consumer->subdir, dir_name,
2718 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2719 DBG3("Copy session consumer subdir %s", consumer->subdir);
2720
2721 ret = LTTNG_OK;
2722
2723 error:
2724 return ret;
2725 }
2726
2727 /*
2728 * Create an UST session and add it to the session ust list.
2729 *
2730 * Should *NOT* be called with RCU read-side lock held.
2731 */
2732 static int create_ust_session(struct ltt_session *session,
2733 struct lttng_domain *domain)
2734 {
2735 int ret;
2736 struct ltt_ust_session *lus = NULL;
2737
2738 assert(session);
2739 assert(domain);
2740 assert(session->consumer);
2741
2742 switch (domain->type) {
2743 case LTTNG_DOMAIN_JUL:
2744 case LTTNG_DOMAIN_LOG4J:
2745 case LTTNG_DOMAIN_PYTHON:
2746 case LTTNG_DOMAIN_UST:
2747 break;
2748 default:
2749 ERR("Unknown UST domain on create session %d", domain->type);
2750 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2751 goto error;
2752 }
2753
2754 DBG("Creating UST session");
2755
2756 lus = trace_ust_create_session(session->id);
2757 if (lus == NULL) {
2758 ret = LTTNG_ERR_UST_SESS_FAIL;
2759 goto error;
2760 }
2761
2762 lus->uid = session->uid;
2763 lus->gid = session->gid;
2764 lus->output_traces = session->output_traces;
2765 lus->snapshot_mode = session->snapshot_mode;
2766 lus->live_timer_interval = session->live_timer;
2767 session->ust_session = lus;
2768
2769 /* Copy session output to the newly created UST session */
2770 ret = copy_session_consumer(domain->type, session);
2771 if (ret != LTTNG_OK) {
2772 goto error;
2773 }
2774
2775 return LTTNG_OK;
2776
2777 error:
2778 free(lus);
2779 session->ust_session = NULL;
2780 return ret;
2781 }
2782
2783 /*
2784 * Create a kernel tracer session then create the default channel.
2785 */
2786 static int create_kernel_session(struct ltt_session *session)
2787 {
2788 int ret;
2789
2790 DBG("Creating kernel session");
2791
2792 ret = kernel_create_session(session, kernel_tracer_fd);
2793 if (ret < 0) {
2794 ret = LTTNG_ERR_KERN_SESS_FAIL;
2795 goto error;
2796 }
2797
2798 /* Code flow safety */
2799 assert(session->kernel_session);
2800
2801 /* Copy session output to the newly created Kernel session */
2802 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2803 if (ret != LTTNG_OK) {
2804 goto error;
2805 }
2806
2807 /* Create directory(ies) on local filesystem. */
2808 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2809 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2810 ret = run_as_mkdir_recursive(
2811 session->kernel_session->consumer->dst.trace_path,
2812 S_IRWXU | S_IRWXG, session->uid, session->gid);
2813 if (ret < 0) {
2814 if (ret != -EEXIST) {
2815 ERR("Trace directory creation error");
2816 goto error;
2817 }
2818 }
2819 }
2820
2821 session->kernel_session->uid = session->uid;
2822 session->kernel_session->gid = session->gid;
2823 session->kernel_session->output_traces = session->output_traces;
2824 session->kernel_session->snapshot_mode = session->snapshot_mode;
2825
2826 return LTTNG_OK;
2827
2828 error:
2829 trace_kernel_destroy_session(session->kernel_session);
2830 session->kernel_session = NULL;
2831 return ret;
2832 }
2833
2834 /*
2835 * Count number of session permitted by uid/gid.
2836 */
2837 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2838 {
2839 unsigned int i = 0;
2840 struct ltt_session *session;
2841
2842 DBG("Counting number of available session for UID %d GID %d",
2843 uid, gid);
2844 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2845 /*
2846 * Only list the sessions the user can control.
2847 */
2848 if (!session_access_ok(session, uid, gid)) {
2849 continue;
2850 }
2851 i++;
2852 }
2853 return i;
2854 }
2855
2856 /*
2857 * Process the command requested by the lttng client within the command
2858 * context structure. This function make sure that the return structure (llm)
2859 * is set and ready for transmission before returning.
2860 *
2861 * Return any error encountered or 0 for success.
2862 *
2863 * "sock" is only used for special-case var. len data.
2864 *
2865 * Should *NOT* be called with RCU read-side lock held.
2866 */
2867 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2868 int *sock_error)
2869 {
2870 int ret = LTTNG_OK;
2871 int need_tracing_session = 1;
2872 int need_domain;
2873
2874 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2875
2876 *sock_error = 0;
2877
2878 switch (cmd_ctx->lsm->cmd_type) {
2879 case LTTNG_CREATE_SESSION:
2880 case LTTNG_CREATE_SESSION_SNAPSHOT:
2881 case LTTNG_CREATE_SESSION_LIVE:
2882 case LTTNG_DESTROY_SESSION:
2883 case LTTNG_LIST_SESSIONS:
2884 case LTTNG_LIST_DOMAINS:
2885 case LTTNG_START_TRACE:
2886 case LTTNG_STOP_TRACE:
2887 case LTTNG_DATA_PENDING:
2888 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2889 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2890 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2891 case LTTNG_SNAPSHOT_RECORD:
2892 case LTTNG_SAVE_SESSION:
2893 need_domain = 0;
2894 break;
2895 default:
2896 need_domain = 1;
2897 }
2898
2899 if (opt_no_kernel && need_domain
2900 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2901 if (!is_root) {
2902 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2903 } else {
2904 ret = LTTNG_ERR_KERN_NA;
2905 }
2906 goto error;
2907 }
2908
2909 /* Deny register consumer if we already have a spawned consumer. */
2910 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2911 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2912 if (kconsumer_data.pid > 0) {
2913 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2914 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2915 goto error;
2916 }
2917 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2918 }
2919
2920 /*
2921 * Check for command that don't needs to allocate a returned payload. We do
2922 * this here so we don't have to make the call for no payload at each
2923 * command.
2924 */
2925 switch(cmd_ctx->lsm->cmd_type) {
2926 case LTTNG_LIST_SESSIONS:
2927 case LTTNG_LIST_TRACEPOINTS:
2928 case LTTNG_LIST_TRACEPOINT_FIELDS:
2929 case LTTNG_LIST_DOMAINS:
2930 case LTTNG_LIST_CHANNELS:
2931 case LTTNG_LIST_EVENTS:
2932 case LTTNG_LIST_SYSCALLS:
2933 break;
2934 default:
2935 /* Setup lttng message with no payload */
2936 ret = setup_lttng_msg(cmd_ctx, 0);
2937 if (ret < 0) {
2938 /* This label does not try to unlock the session */
2939 goto init_setup_error;
2940 }
2941 }
2942
2943 /* Commands that DO NOT need a session. */
2944 switch (cmd_ctx->lsm->cmd_type) {
2945 case LTTNG_CREATE_SESSION:
2946 case LTTNG_CREATE_SESSION_SNAPSHOT:
2947 case LTTNG_CREATE_SESSION_LIVE:
2948 case LTTNG_CALIBRATE:
2949 case LTTNG_LIST_SESSIONS:
2950 case LTTNG_LIST_TRACEPOINTS:
2951 case LTTNG_LIST_SYSCALLS:
2952 case LTTNG_LIST_TRACEPOINT_FIELDS:
2953 case LTTNG_SAVE_SESSION:
2954 need_tracing_session = 0;
2955 break;
2956 default:
2957 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2958 /*
2959 * We keep the session list lock across _all_ commands
2960 * for now, because the per-session lock does not
2961 * handle teardown properly.
2962 */
2963 session_lock_list();
2964 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2965 if (cmd_ctx->session == NULL) {
2966 ret = LTTNG_ERR_SESS_NOT_FOUND;
2967 goto error;
2968 } else {
2969 /* Acquire lock for the session */
2970 session_lock(cmd_ctx->session);
2971 }
2972 break;
2973 }
2974
2975 /*
2976 * Commands that need a valid session but should NOT create one if none
2977 * exists. Instead of creating one and destroying it when the command is
2978 * handled, process that right before so we save some round trip in useless
2979 * code path.
2980 */
2981 switch (cmd_ctx->lsm->cmd_type) {
2982 case LTTNG_DISABLE_CHANNEL:
2983 case LTTNG_DISABLE_EVENT:
2984 switch (cmd_ctx->lsm->domain.type) {
2985 case LTTNG_DOMAIN_KERNEL:
2986 if (!cmd_ctx->session->kernel_session) {
2987 ret = LTTNG_ERR_NO_CHANNEL;
2988 goto error;
2989 }
2990 break;
2991 case LTTNG_DOMAIN_JUL:
2992 case LTTNG_DOMAIN_LOG4J:
2993 case LTTNG_DOMAIN_PYTHON:
2994 case LTTNG_DOMAIN_UST:
2995 if (!cmd_ctx->session->ust_session) {
2996 ret = LTTNG_ERR_NO_CHANNEL;
2997 goto error;
2998 }
2999 break;
3000 default:
3001 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3002 goto error;
3003 }
3004 default:
3005 break;
3006 }
3007
3008 if (!need_domain) {
3009 goto skip_domain;
3010 }
3011
3012 /*
3013 * Check domain type for specific "pre-action".
3014 */
3015 switch (cmd_ctx->lsm->domain.type) {
3016 case LTTNG_DOMAIN_KERNEL:
3017 if (!is_root) {
3018 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3019 goto error;
3020 }
3021
3022 /* Kernel tracer check */
3023 if (kernel_tracer_fd == -1) {
3024 /* Basically, load kernel tracer modules */
3025 ret = init_kernel_tracer();
3026 if (ret != 0) {
3027 goto error;
3028 }
3029 }
3030
3031 /* Consumer is in an ERROR state. Report back to client */
3032 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3033 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3034 goto error;
3035 }
3036
3037 /* Need a session for kernel command */
3038 if (need_tracing_session) {
3039 if (cmd_ctx->session->kernel_session == NULL) {
3040 ret = create_kernel_session(cmd_ctx->session);
3041 if (ret < 0) {
3042 ret = LTTNG_ERR_KERN_SESS_FAIL;
3043 goto error;
3044 }
3045 }
3046
3047 /* Start the kernel consumer daemon */
3048 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3049 if (kconsumer_data.pid == 0 &&
3050 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3051 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3052 ret = start_consumerd(&kconsumer_data);
3053 if (ret < 0) {
3054 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3055 goto error;
3056 }
3057 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3058 } else {
3059 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3060 }
3061
3062 /*
3063 * The consumer was just spawned so we need to add the socket to
3064 * the consumer output of the session if exist.
3065 */
3066 ret = consumer_create_socket(&kconsumer_data,
3067 cmd_ctx->session->kernel_session->consumer);
3068 if (ret < 0) {
3069 goto error;
3070 }
3071 }
3072
3073 break;
3074 case LTTNG_DOMAIN_JUL:
3075 case LTTNG_DOMAIN_LOG4J:
3076 case LTTNG_DOMAIN_PYTHON:
3077 case LTTNG_DOMAIN_UST:
3078 {
3079 if (!ust_app_supported()) {
3080 ret = LTTNG_ERR_NO_UST;
3081 goto error;
3082 }
3083 /* Consumer is in an ERROR state. Report back to client */
3084 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3085 ret = LTTNG_ERR_NO_USTCONSUMERD;
3086 goto error;
3087 }
3088
3089 if (need_tracing_session) {
3090 /* Create UST session if none exist. */
3091 if (cmd_ctx->session->ust_session == NULL) {
3092 ret = create_ust_session(cmd_ctx->session,
3093 &cmd_ctx->lsm->domain);
3094 if (ret != LTTNG_OK) {
3095 goto error;
3096 }
3097 }
3098
3099 /* Start the UST consumer daemons */
3100 /* 64-bit */
3101 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3102 if (consumerd64_bin[0] != '\0' &&
3103 ustconsumer64_data.pid == 0 &&
3104 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3105 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3106 ret = start_consumerd(&ustconsumer64_data);
3107 if (ret < 0) {
3108 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3109 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3110 goto error;
3111 }
3112
3113 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3114 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3115 } else {
3116 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3117 }
3118
3119 /*
3120 * Setup socket for consumer 64 bit. No need for atomic access
3121 * since it was set above and can ONLY be set in this thread.
3122 */
3123 ret = consumer_create_socket(&ustconsumer64_data,
3124 cmd_ctx->session->ust_session->consumer);
3125 if (ret < 0) {
3126 goto error;
3127 }
3128
3129 /* 32-bit */
3130 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3131 if (consumerd32_bin[0] != '\0' &&
3132 ustconsumer32_data.pid == 0 &&
3133 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3134 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3135 ret = start_consumerd(&ustconsumer32_data);
3136 if (ret < 0) {
3137 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3138 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3139 goto error;
3140 }
3141
3142 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3143 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3144 } else {
3145 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3146 }
3147
3148 /*
3149 * Setup socket for consumer 64 bit. No need for atomic access
3150 * since it was set above and can ONLY be set in this thread.
3151 */
3152 ret = consumer_create_socket(&ustconsumer32_data,
3153 cmd_ctx->session->ust_session->consumer);
3154 if (ret < 0) {
3155 goto error;
3156 }
3157 }
3158 break;
3159 }
3160 default:
3161 break;
3162 }
3163 skip_domain:
3164
3165 /* Validate consumer daemon state when start/stop trace command */
3166 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3167 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3168 switch (cmd_ctx->lsm->domain.type) {
3169 case LTTNG_DOMAIN_JUL:
3170 case LTTNG_DOMAIN_LOG4J:
3171 case LTTNG_DOMAIN_PYTHON:
3172 case LTTNG_DOMAIN_UST:
3173 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3174 ret = LTTNG_ERR_NO_USTCONSUMERD;
3175 goto error;
3176 }
3177 break;
3178 case LTTNG_DOMAIN_KERNEL:
3179 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3180 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3181 goto error;
3182 }
3183 break;
3184 }
3185 }
3186
3187 /*
3188 * Check that the UID or GID match that of the tracing session.
3189 * The root user can interact with all sessions.
3190 */
3191 if (need_tracing_session) {
3192 if (!session_access_ok(cmd_ctx->session,
3193 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3194 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3195 ret = LTTNG_ERR_EPERM;
3196 goto error;
3197 }
3198 }
3199
3200 /*
3201 * Send relayd information to consumer as soon as we have a domain and a
3202 * session defined.
3203 */
3204 if (cmd_ctx->session && need_domain) {
3205 /*
3206 * Setup relayd if not done yet. If the relayd information was already
3207 * sent to the consumer, this call will gracefully return.
3208 */
3209 ret = cmd_setup_relayd(cmd_ctx->session);
3210 if (ret != LTTNG_OK) {
3211 goto error;
3212 }
3213 }
3214
3215 /* Process by command type */
3216 switch (cmd_ctx->lsm->cmd_type) {
3217 case LTTNG_ADD_CONTEXT:
3218 {
3219 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3220 cmd_ctx->lsm->u.context.channel_name,
3221 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3222 break;
3223 }
3224 case LTTNG_DISABLE_CHANNEL:
3225 {
3226 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3227 cmd_ctx->lsm->u.disable.channel_name);
3228 break;
3229 }
3230 case LTTNG_DISABLE_EVENT:
3231 {
3232 /* FIXME: passing packed structure to non-packed pointer */
3233 /* TODO: handle filter */
3234 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3235 cmd_ctx->lsm->u.disable.channel_name,
3236 &cmd_ctx->lsm->u.disable.event);
3237 break;
3238 }
3239 case LTTNG_ENABLE_CHANNEL:
3240 {
3241 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3242 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3243 break;
3244 }
3245 case LTTNG_ENABLE_EVENT:
3246 {
3247 struct lttng_event_exclusion *exclusion = NULL;
3248 struct lttng_filter_bytecode *bytecode = NULL;
3249 char *filter_expression = NULL;
3250
3251 /* Handle exclusion events and receive it from the client. */
3252 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3253 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3254
3255 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3256 (count * LTTNG_SYMBOL_NAME_LEN));
3257 if (!exclusion) {
3258 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3259 goto error;
3260 }
3261
3262 DBG("Receiving var len exclusion event list from client ...");
3263 exclusion->count = count;
3264 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3265 count * LTTNG_SYMBOL_NAME_LEN);
3266 if (ret <= 0) {
3267 DBG("Nothing recv() from client var len data... continuing");
3268 *sock_error = 1;
3269 free(exclusion);
3270 ret = LTTNG_ERR_EXCLUSION_INVAL;
3271 goto error;
3272 }
3273 }
3274
3275 /* Get filter expression from client. */
3276 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3277 size_t expression_len =
3278 cmd_ctx->lsm->u.enable.expression_len;
3279
3280 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3281 ret = LTTNG_ERR_FILTER_INVAL;
3282 free(exclusion);
3283 goto error;
3284 }
3285
3286 filter_expression = zmalloc(expression_len);
3287 if (!filter_expression) {
3288 free(exclusion);
3289 ret = LTTNG_ERR_FILTER_NOMEM;
3290 goto error;
3291 }
3292
3293 /* Receive var. len. data */
3294 DBG("Receiving var len filter's expression from client ...");
3295 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3296 expression_len);
3297 if (ret <= 0) {
3298 DBG("Nothing recv() from client car len data... continuing");
3299 *sock_error = 1;
3300 free(filter_expression);
3301 free(exclusion);
3302 ret = LTTNG_ERR_FILTER_INVAL;
3303 goto error;
3304 }
3305 }
3306
3307 /* Handle filter and get bytecode from client. */
3308 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3309 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3310
3311 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3312 ret = LTTNG_ERR_FILTER_INVAL;
3313 free(filter_expression);
3314 free(exclusion);
3315 goto error;
3316 }
3317
3318 bytecode = zmalloc(bytecode_len);
3319 if (!bytecode) {
3320 free(filter_expression);
3321 free(exclusion);
3322 ret = LTTNG_ERR_FILTER_NOMEM;
3323 goto error;
3324 }
3325
3326 /* Receive var. len. data */
3327 DBG("Receiving var len filter's bytecode from client ...");
3328 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3329 if (ret <= 0) {
3330 DBG("Nothing recv() from client car len data... continuing");
3331 *sock_error = 1;
3332 free(filter_expression);
3333 free(bytecode);
3334 free(exclusion);
3335 ret = LTTNG_ERR_FILTER_INVAL;
3336 goto error;
3337 }
3338
3339 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3340 free(filter_expression);
3341 free(bytecode);
3342 free(exclusion);
3343 ret = LTTNG_ERR_FILTER_INVAL;
3344 goto error;
3345 }
3346 }
3347
3348 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3349 cmd_ctx->lsm->u.enable.channel_name,
3350 &cmd_ctx->lsm->u.enable.event,
3351 filter_expression, bytecode, exclusion,
3352 kernel_poll_pipe[1]);
3353 break;
3354 }
3355 case LTTNG_LIST_TRACEPOINTS:
3356 {
3357 struct lttng_event *events;
3358 ssize_t nb_events;
3359
3360 session_lock_list();
3361 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3362 session_unlock_list();
3363 if (nb_events < 0) {
3364 /* Return value is a negative lttng_error_code. */
3365 ret = -nb_events;
3366 goto error;
3367 }
3368
3369 /*
3370 * Setup lttng message with payload size set to the event list size in
3371 * bytes and then copy list into the llm payload.
3372 */
3373 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3374 if (ret < 0) {
3375 free(events);
3376 goto setup_error;
3377 }
3378
3379 /* Copy event list into message payload */
3380 memcpy(cmd_ctx->llm->payload, events,
3381 sizeof(struct lttng_event) * nb_events);
3382
3383 free(events);
3384
3385 ret = LTTNG_OK;
3386 break;
3387 }
3388 case LTTNG_LIST_TRACEPOINT_FIELDS:
3389 {
3390 struct lttng_event_field *fields;
3391 ssize_t nb_fields;
3392
3393 session_lock_list();
3394 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3395 &fields);
3396 session_unlock_list();
3397 if (nb_fields < 0) {
3398 /* Return value is a negative lttng_error_code. */
3399 ret = -nb_fields;
3400 goto error;
3401 }
3402
3403 /*
3404 * Setup lttng message with payload size set to the event list size in
3405 * bytes and then copy list into the llm payload.
3406 */
3407 ret = setup_lttng_msg(cmd_ctx,
3408 sizeof(struct lttng_event_field) * nb_fields);
3409 if (ret < 0) {
3410 free(fields);
3411 goto setup_error;
3412 }
3413
3414 /* Copy event list into message payload */
3415 memcpy(cmd_ctx->llm->payload, fields,
3416 sizeof(struct lttng_event_field) * nb_fields);
3417
3418 free(fields);
3419
3420 ret = LTTNG_OK;
3421 break;
3422 }
3423 case LTTNG_LIST_SYSCALLS:
3424 {
3425 struct lttng_event *events;
3426 ssize_t nb_events;
3427
3428 nb_events = cmd_list_syscalls(&events);
3429 if (nb_events < 0) {
3430 /* Return value is a negative lttng_error_code. */
3431 ret = -nb_events;
3432 goto error;
3433 }
3434
3435 /*
3436 * Setup lttng message with payload size set to the event list size in
3437 * bytes and then copy list into the llm payload.
3438 */
3439 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3440 if (ret < 0) {
3441 free(events);
3442 goto setup_error;
3443 }
3444
3445 /* Copy event list into message payload */
3446 memcpy(cmd_ctx->llm->payload, events,
3447 sizeof(struct lttng_event) * nb_events);
3448
3449 free(events);
3450
3451 ret = LTTNG_OK;
3452 break;
3453 }
3454 case LTTNG_SET_CONSUMER_URI:
3455 {
3456 size_t nb_uri, len;
3457 struct lttng_uri *uris;
3458
3459 nb_uri = cmd_ctx->lsm->u.uri.size;
3460 len = nb_uri * sizeof(struct lttng_uri);
3461
3462 if (nb_uri == 0) {
3463 ret = LTTNG_ERR_INVALID;
3464 goto error;
3465 }
3466
3467 uris = zmalloc(len);
3468 if (uris == NULL) {
3469 ret = LTTNG_ERR_FATAL;
3470 goto error;
3471 }
3472
3473 /* Receive variable len data */
3474 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3475 ret = lttcomm_recv_unix_sock(sock, uris, len);
3476 if (ret <= 0) {
3477 DBG("No URIs received from client... continuing");
3478 *sock_error = 1;
3479 ret = LTTNG_ERR_SESSION_FAIL;
3480 free(uris);
3481 goto error;
3482 }
3483
3484 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3485 free(uris);
3486 if (ret != LTTNG_OK) {
3487 goto error;
3488 }
3489
3490
3491 break;
3492 }
3493 case LTTNG_START_TRACE:
3494 {
3495 ret = cmd_start_trace(cmd_ctx->session);
3496 break;
3497 }
3498 case LTTNG_STOP_TRACE:
3499 {
3500 ret = cmd_stop_trace(cmd_ctx->session);
3501 break;
3502 }
3503 case LTTNG_CREATE_SESSION:
3504 {
3505 size_t nb_uri, len;
3506 struct lttng_uri *uris = NULL;
3507
3508 nb_uri = cmd_ctx->lsm->u.uri.size;
3509 len = nb_uri * sizeof(struct lttng_uri);
3510
3511 if (nb_uri > 0) {
3512 uris = zmalloc(len);
3513 if (uris == NULL) {
3514 ret = LTTNG_ERR_FATAL;
3515 goto error;
3516 }
3517
3518 /* Receive variable len data */
3519 DBG("Waiting for %zu URIs from client ...", nb_uri);
3520 ret = lttcomm_recv_unix_sock(sock, uris, len);
3521 if (ret <= 0) {
3522 DBG("No URIs received from client... continuing");
3523 *sock_error = 1;
3524 ret = LTTNG_ERR_SESSION_FAIL;
3525 free(uris);
3526 goto error;
3527 }
3528
3529 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3530 DBG("Creating session with ONE network URI is a bad call");
3531 ret = LTTNG_ERR_SESSION_FAIL;
3532 free(uris);
3533 goto error;
3534 }
3535 }
3536
3537 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3538 &cmd_ctx->creds, 0);
3539
3540 free(uris);
3541
3542 break;
3543 }
3544 case LTTNG_DESTROY_SESSION:
3545 {
3546 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3547
3548 /* Set session to NULL so we do not unlock it after free. */
3549 cmd_ctx->session = NULL;
3550 break;
3551 }
3552 case LTTNG_LIST_DOMAINS:
3553 {
3554 ssize_t nb_dom;
3555 struct lttng_domain *domains = NULL;
3556
3557 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3558 if (nb_dom < 0) {
3559 /* Return value is a negative lttng_error_code. */
3560 ret = -nb_dom;
3561 goto error;
3562 }
3563
3564 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3565 if (ret < 0) {
3566 free(domains);
3567 goto setup_error;
3568 }
3569
3570 /* Copy event list into message payload */
3571 memcpy(cmd_ctx->llm->payload, domains,
3572 nb_dom * sizeof(struct lttng_domain));
3573
3574 free(domains);
3575
3576 ret = LTTNG_OK;
3577 break;
3578 }
3579 case LTTNG_LIST_CHANNELS:
3580 {
3581 int nb_chan;
3582 struct lttng_channel *channels = NULL;
3583
3584 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3585 cmd_ctx->session, &channels);
3586 if (nb_chan < 0) {
3587 /* Return value is a negative lttng_error_code. */
3588 ret = -nb_chan;
3589 goto error;
3590 }
3591
3592 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3593 if (ret < 0) {
3594 free(channels);
3595 goto setup_error;
3596 }
3597
3598 /* Copy event list into message payload */
3599 memcpy(cmd_ctx->llm->payload, channels,
3600 nb_chan * sizeof(struct lttng_channel));
3601
3602 free(channels);
3603
3604 ret = LTTNG_OK;
3605 break;
3606 }
3607 case LTTNG_LIST_EVENTS:
3608 {
3609 ssize_t nb_event;
3610 struct lttng_event *events = NULL;
3611
3612 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3613 cmd_ctx->lsm->u.list.channel_name, &events);
3614 if (nb_event < 0) {
3615 /* Return value is a negative lttng_error_code. */
3616 ret = -nb_event;
3617 goto error;
3618 }
3619
3620 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3621 if (ret < 0) {
3622 free(events);
3623 goto setup_error;
3624 }
3625
3626 /* Copy event list into message payload */
3627 memcpy(cmd_ctx->llm->payload, events,
3628 nb_event * sizeof(struct lttng_event));
3629
3630 free(events);
3631
3632 ret = LTTNG_OK;
3633 break;
3634 }
3635 case LTTNG_LIST_SESSIONS:
3636 {
3637 unsigned int nr_sessions;
3638
3639 session_lock_list();
3640 nr_sessions = lttng_sessions_count(
3641 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3642 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3643
3644 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3645 if (ret < 0) {
3646 session_unlock_list();
3647 goto setup_error;
3648 }
3649
3650 /* Filled the session array */
3651 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3652 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3653 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3654
3655 session_unlock_list();
3656
3657 ret = LTTNG_OK;
3658 break;
3659 }
3660 case LTTNG_CALIBRATE:
3661 {
3662 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3663 &cmd_ctx->lsm->u.calibrate);
3664 break;
3665 }
3666 case LTTNG_REGISTER_CONSUMER:
3667 {
3668 struct consumer_data *cdata;
3669
3670 switch (cmd_ctx->lsm->domain.type) {
3671 case LTTNG_DOMAIN_KERNEL:
3672 cdata = &kconsumer_data;
3673 break;
3674 default:
3675 ret = LTTNG_ERR_UND;
3676 goto error;
3677 }
3678
3679 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3680 cmd_ctx->lsm->u.reg.path, cdata);
3681 break;
3682 }
3683 case LTTNG_DATA_PENDING:
3684 {
3685 ret = cmd_data_pending(cmd_ctx->session);
3686 break;
3687 }
3688 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3689 {
3690 struct lttcomm_lttng_output_id reply;
3691
3692 ret = cmd_snapshot_add_output(cmd_ctx->session,
3693 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3694 if (ret != LTTNG_OK) {
3695 goto error;
3696 }
3697
3698 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3699 if (ret < 0) {
3700 goto setup_error;
3701 }
3702
3703 /* Copy output list into message payload */
3704 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3705 ret = LTTNG_OK;
3706 break;
3707 }
3708 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3709 {
3710 ret = cmd_snapshot_del_output(cmd_ctx->session,
3711 &cmd_ctx->lsm->u.snapshot_output.output);
3712 break;
3713 }
3714 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3715 {
3716 ssize_t nb_output;
3717 struct lttng_snapshot_output *outputs = NULL;
3718
3719 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3720 if (nb_output < 0) {
3721 ret = -nb_output;
3722 goto error;
3723 }
3724
3725 ret = setup_lttng_msg(cmd_ctx,
3726 nb_output * sizeof(struct lttng_snapshot_output));
3727 if (ret < 0) {
3728 free(outputs);
3729 goto setup_error;
3730 }
3731
3732 if (outputs) {
3733 /* Copy output list into message payload */
3734 memcpy(cmd_ctx->llm->payload, outputs,
3735 nb_output * sizeof(struct lttng_snapshot_output));
3736 free(outputs);
3737 }
3738
3739 ret = LTTNG_OK;
3740 break;
3741 }
3742 case LTTNG_SNAPSHOT_RECORD:
3743 {
3744 ret = cmd_snapshot_record(cmd_ctx->session,
3745 &cmd_ctx->lsm->u.snapshot_record.output,
3746 cmd_ctx->lsm->u.snapshot_record.wait);
3747 break;
3748 }
3749 case LTTNG_CREATE_SESSION_SNAPSHOT:
3750 {
3751 size_t nb_uri, len;
3752 struct lttng_uri *uris = NULL;
3753
3754 nb_uri = cmd_ctx->lsm->u.uri.size;
3755 len = nb_uri * sizeof(struct lttng_uri);
3756
3757 if (nb_uri > 0) {
3758 uris = zmalloc(len);
3759 if (uris == NULL) {
3760 ret = LTTNG_ERR_FATAL;
3761 goto error;
3762 }
3763
3764 /* Receive variable len data */
3765 DBG("Waiting for %zu URIs from client ...", nb_uri);
3766 ret = lttcomm_recv_unix_sock(sock, uris, len);
3767 if (ret <= 0) {
3768 DBG("No URIs received from client... continuing");
3769 *sock_error = 1;
3770 ret = LTTNG_ERR_SESSION_FAIL;
3771 free(uris);
3772 goto error;
3773 }
3774
3775 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3776 DBG("Creating session with ONE network URI is a bad call");
3777 ret = LTTNG_ERR_SESSION_FAIL;
3778 free(uris);
3779 goto error;
3780 }
3781 }
3782
3783 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3784 nb_uri, &cmd_ctx->creds);
3785 free(uris);
3786 break;
3787 }
3788 case LTTNG_CREATE_SESSION_LIVE:
3789 {
3790 size_t nb_uri, len;
3791 struct lttng_uri *uris = NULL;
3792
3793 nb_uri = cmd_ctx->lsm->u.uri.size;
3794 len = nb_uri * sizeof(struct lttng_uri);
3795
3796 if (nb_uri > 0) {
3797 uris = zmalloc(len);
3798 if (uris == NULL) {
3799 ret = LTTNG_ERR_FATAL;
3800 goto error;
3801 }
3802
3803 /* Receive variable len data */
3804 DBG("Waiting for %zu URIs from client ...", nb_uri);
3805 ret = lttcomm_recv_unix_sock(sock, uris, len);
3806 if (ret <= 0) {
3807 DBG("No URIs received from client... continuing");
3808 *sock_error = 1;
3809 ret = LTTNG_ERR_SESSION_FAIL;
3810 free(uris);
3811 goto error;
3812 }
3813
3814 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3815 DBG("Creating session with ONE network URI is a bad call");
3816 ret = LTTNG_ERR_SESSION_FAIL;
3817 free(uris);
3818 goto error;
3819 }
3820 }
3821
3822 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3823 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3824 free(uris);
3825 break;
3826 }
3827 case LTTNG_SAVE_SESSION:
3828 {
3829 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3830 &cmd_ctx->creds);
3831 break;
3832 }
3833 default:
3834 ret = LTTNG_ERR_UND;
3835 break;
3836 }
3837
3838 error:
3839 if (cmd_ctx->llm == NULL) {
3840 DBG("Missing llm structure. Allocating one.");
3841 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3842 goto setup_error;
3843 }
3844 }
3845 /* Set return code */
3846 cmd_ctx->llm->ret_code = ret;
3847 setup_error:
3848 if (cmd_ctx->session) {
3849 session_unlock(cmd_ctx->session);
3850 }
3851 if (need_tracing_session) {
3852 session_unlock_list();
3853 }
3854 init_setup_error:
3855 return ret;
3856 }
3857
3858 /*
3859 * Thread managing health check socket.
3860 */
3861 static void *thread_manage_health(void *data)
3862 {
3863 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3864 uint32_t revents, nb_fd;
3865 struct lttng_poll_event events;
3866 struct health_comm_msg msg;
3867 struct health_comm_reply reply;
3868
3869 DBG("[thread] Manage health check started");
3870
3871 rcu_register_thread();
3872
3873 /* We might hit an error path before this is created. */
3874 lttng_poll_init(&events);
3875
3876 /* Create unix socket */
3877 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3878 if (sock < 0) {
3879 ERR("Unable to create health check Unix socket");
3880 ret = -1;
3881 goto error;
3882 }
3883
3884 if (is_root) {
3885 /* lttng health client socket path permissions */
3886 ret = chown(health_unix_sock_path, 0,
3887 utils_get_group_id(tracing_group_name));
3888 if (ret < 0) {
3889 ERR("Unable to set group on %s", health_unix_sock_path);
3890 PERROR("chown");
3891 ret = -1;
3892 goto error;
3893 }
3894
3895 ret = chmod(health_unix_sock_path,
3896 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3897 if (ret < 0) {
3898 ERR("Unable to set permissions on %s", health_unix_sock_path);
3899 PERROR("chmod");
3900 ret = -1;
3901 goto error;
3902 }
3903 }
3904
3905 /*
3906 * Set the CLOEXEC flag. Return code is useless because either way, the
3907 * show must go on.
3908 */
3909 (void) utils_set_fd_cloexec(sock);
3910
3911 ret = lttcomm_listen_unix_sock(sock);
3912 if (ret < 0) {
3913 goto error;
3914 }
3915
3916 /*
3917 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3918 * more will be added to this poll set.
3919 */
3920 ret = sessiond_set_thread_pollset(&events, 2);
3921 if (ret < 0) {
3922 goto error;
3923 }
3924
3925 /* Add the application registration socket */
3926 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3927 if (ret < 0) {
3928 goto error;
3929 }
3930
3931 sessiond_notify_ready();
3932
3933 while (1) {
3934 DBG("Health check ready");
3935
3936 /* Inifinite blocking call, waiting for transmission */
3937 restart:
3938 ret = lttng_poll_wait(&events, -1);
3939 if (ret < 0) {
3940 /*
3941 * Restart interrupted system call.
3942 */
3943 if (errno == EINTR) {
3944 goto restart;
3945 }
3946 goto error;
3947 }
3948
3949 nb_fd = ret;
3950
3951 for (i = 0; i < nb_fd; i++) {
3952 /* Fetch once the poll data */
3953 revents = LTTNG_POLL_GETEV(&events, i);
3954 pollfd = LTTNG_POLL_GETFD(&events, i);
3955
3956 if (!revents) {
3957 /* No activity for this FD (poll implementation). */
3958 continue;
3959 }
3960
3961 /* Thread quit pipe has been closed. Killing thread. */
3962 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3963 if (ret) {
3964 err = 0;
3965 goto exit;
3966 }
3967
3968 /* Event on the registration socket */
3969 if (pollfd == sock) {
3970 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3971 ERR("Health socket poll error");
3972 goto error;
3973 }
3974 }
3975 }
3976
3977 new_sock = lttcomm_accept_unix_sock(sock);
3978 if (new_sock < 0) {
3979 goto error;
3980 }
3981
3982 /*
3983 * Set the CLOEXEC flag. Return code is useless because either way, the
3984 * show must go on.
3985 */
3986 (void) utils_set_fd_cloexec(new_sock);
3987
3988 DBG("Receiving data from client for health...");
3989 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3990 if (ret <= 0) {
3991 DBG("Nothing recv() from client... continuing");
3992 ret = close(new_sock);
3993 if (ret) {
3994 PERROR("close");
3995 }
3996 new_sock = -1;
3997 continue;
3998 }
3999
4000 rcu_thread_online();
4001
4002 memset(&reply, 0, sizeof(reply));
4003 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4004 /*
4005 * health_check_state returns 0 if health is
4006 * bad.
4007 */
4008 if (!health_check_state(health_sessiond, i)) {
4009 reply.ret_code |= 1ULL << i;
4010 }
4011 }
4012
4013 DBG2("Health check return value %" PRIx64, reply.ret_code);
4014
4015 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4016 if (ret < 0) {
4017 ERR("Failed to send health data back to client");
4018 }
4019
4020 /* End of transmission */
4021 ret = close(new_sock);
4022 if (ret) {
4023 PERROR("close");
4024 }
4025 new_sock = -1;
4026 }
4027
4028 exit:
4029 error:
4030 if (err) {
4031 ERR("Health error occurred in %s", __func__);
4032 }
4033 DBG("Health check thread dying");
4034 unlink(health_unix_sock_path);
4035 if (sock >= 0) {
4036 ret = close(sock);
4037 if (ret) {
4038 PERROR("close");
4039 }
4040 }
4041
4042 lttng_poll_clean(&events);
4043
4044 rcu_unregister_thread();
4045 return NULL;
4046 }
4047
4048 /*
4049 * This thread manage all clients request using the unix client socket for
4050 * communication.
4051 */
4052 static void *thread_manage_clients(void *data)
4053 {
4054 int sock = -1, ret, i, pollfd, err = -1;
4055 int sock_error;
4056 uint32_t revents, nb_fd;
4057 struct command_ctx *cmd_ctx = NULL;
4058 struct lttng_poll_event events;
4059
4060 DBG("[thread] Manage client started");
4061
4062 rcu_register_thread();
4063
4064 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4065
4066 health_code_update();
4067
4068 ret = lttcomm_listen_unix_sock(client_sock);
4069 if (ret < 0) {
4070 goto error_listen;
4071 }
4072
4073 /*
4074 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4075 * more will be added to this poll set.
4076 */
4077 ret = sessiond_set_thread_pollset(&events, 2);
4078 if (ret < 0) {
4079 goto error_create_poll;
4080 }
4081
4082 /* Add the application registration socket */
4083 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4084 if (ret < 0) {
4085 goto error;
4086 }
4087
4088 sessiond_notify_ready();
4089 ret = sem_post(&load_info->message_thread_ready);
4090 if (ret) {
4091 PERROR("sem_post message_thread_ready");
4092 goto error;
4093 }
4094
4095 /* This testpoint is after we signal readiness to the parent. */
4096 if (testpoint(sessiond_thread_manage_clients)) {
4097 goto error;
4098 }
4099
4100 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4101 goto error;
4102 }
4103
4104 health_code_update();
4105
4106 while (1) {
4107 DBG("Accepting client command ...");
4108
4109 /* Inifinite blocking call, waiting for transmission */
4110 restart:
4111 health_poll_entry();
4112 ret = lttng_poll_wait(&events, -1);
4113 health_poll_exit();
4114 if (ret < 0) {
4115 /*
4116 * Restart interrupted system call.
4117 */
4118 if (errno == EINTR) {
4119 goto restart;
4120 }
4121 goto error;
4122 }
4123
4124 nb_fd = ret;
4125
4126 for (i = 0; i < nb_fd; i++) {
4127 /* Fetch once the poll data */
4128 revents = LTTNG_POLL_GETEV(&events, i);
4129 pollfd = LTTNG_POLL_GETFD(&events, i);
4130
4131 health_code_update();
4132
4133 if (!revents) {
4134 /* No activity for this FD (poll implementation). */
4135 continue;
4136 }
4137
4138 /* Thread quit pipe has been closed. Killing thread. */
4139 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4140 if (ret) {
4141 err = 0;
4142 goto exit;
4143 }
4144
4145 /* Event on the registration socket */
4146 if (pollfd == client_sock) {
4147 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4148 ERR("Client socket poll error");
4149 goto error;
4150 }
4151 }
4152 }
4153
4154 DBG("Wait for client response");
4155
4156 health_code_update();
4157
4158 sock = lttcomm_accept_unix_sock(client_sock);
4159 if (sock < 0) {
4160 goto error;
4161 }
4162
4163 /*
4164 * Set the CLOEXEC flag. Return code is useless because either way, the
4165 * show must go on.
4166 */
4167 (void) utils_set_fd_cloexec(sock);
4168
4169 /* Set socket option for credentials retrieval */
4170 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4171 if (ret < 0) {
4172 goto error;
4173 }
4174
4175 /* Allocate context command to process the client request */
4176 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4177 if (cmd_ctx == NULL) {
4178 PERROR("zmalloc cmd_ctx");
4179 goto error;
4180 }
4181
4182 /* Allocate data buffer for reception */
4183 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4184 if (cmd_ctx->lsm == NULL) {
4185 PERROR("zmalloc cmd_ctx->lsm");
4186 goto error;
4187 }
4188
4189 cmd_ctx->llm = NULL;
4190 cmd_ctx->session = NULL;
4191
4192 health_code_update();
4193
4194 /*
4195 * Data is received from the lttng client. The struct
4196 * lttcomm_session_msg (lsm) contains the command and data request of
4197 * the client.
4198 */
4199 DBG("Receiving data from client ...");
4200 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4201 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4202 if (ret <= 0) {
4203 DBG("Nothing recv() from client... continuing");
4204 ret = close(sock);
4205 if (ret) {
4206 PERROR("close");
4207 }
4208 sock = -1;
4209 clean_command_ctx(&cmd_ctx);
4210 continue;
4211 }
4212
4213 health_code_update();
4214
4215 // TODO: Validate cmd_ctx including sanity check for
4216 // security purpose.
4217
4218 rcu_thread_online();
4219 /*
4220 * This function dispatch the work to the kernel or userspace tracer
4221 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4222 * informations for the client. The command context struct contains
4223 * everything this function may needs.
4224 */
4225 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4226 rcu_thread_offline();
4227 if (ret < 0) {
4228 ret = close(sock);
4229 if (ret) {
4230 PERROR("close");
4231 }
4232 sock = -1;
4233 /*
4234 * TODO: Inform client somehow of the fatal error. At
4235 * this point, ret < 0 means that a zmalloc failed
4236 * (ENOMEM). Error detected but still accept
4237 * command, unless a socket error has been
4238 * detected.
4239 */
4240 clean_command_ctx(&cmd_ctx);
4241 continue;
4242 }
4243
4244 health_code_update();
4245
4246 DBG("Sending response (size: %d, retcode: %s)",
4247 cmd_ctx->lttng_msg_size,
4248 lttng_strerror(-cmd_ctx->llm->ret_code));
4249 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4250 if (ret < 0) {
4251 ERR("Failed to send data back to client");
4252 }
4253
4254 /* End of transmission */
4255 ret = close(sock);
4256 if (ret) {
4257 PERROR("close");
4258 }
4259 sock = -1;
4260
4261 clean_command_ctx(&cmd_ctx);
4262
4263 health_code_update();
4264 }
4265
4266 exit:
4267 error:
4268 if (sock >= 0) {
4269 ret = close(sock);
4270 if (ret) {
4271 PERROR("close");
4272 }
4273 }
4274
4275 lttng_poll_clean(&events);
4276 clean_command_ctx(&cmd_ctx);
4277
4278 error_listen:
4279 error_create_poll:
4280 unlink(client_unix_sock_path);
4281 if (client_sock >= 0) {
4282 ret = close(client_sock);
4283 if (ret) {
4284 PERROR("close");
4285 }
4286 }
4287
4288 if (err) {
4289 health_error();
4290 ERR("Health error occurred in %s", __func__);
4291 }
4292
4293 health_unregister(health_sessiond);
4294
4295 DBG("Client thread dying");
4296
4297 rcu_unregister_thread();
4298
4299 /*
4300 * Since we are creating the consumer threads, we own them, so we need
4301 * to join them before our thread exits.
4302 */
4303 ret = join_consumer_thread(&kconsumer_data);
4304 if (ret) {
4305 errno = ret;
4306 PERROR("join_consumer");
4307 }
4308
4309 ret = join_consumer_thread(&ustconsumer32_data);
4310 if (ret) {
4311 errno = ret;
4312 PERROR("join_consumer ust32");
4313 }
4314
4315 ret = join_consumer_thread(&ustconsumer64_data);
4316 if (ret) {
4317 errno = ret;
4318 PERROR("join_consumer ust64");
4319 }
4320 return NULL;
4321 }
4322
4323
4324 /*
4325 * usage function on stderr
4326 */
4327 static void usage(void)
4328 {
4329 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4330 fprintf(stderr, " -h, --help Display this usage.\n");
4331 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4332 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4333 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4334 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4335 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4336 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4337 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4338 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4339 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4340 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4341 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4342 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4343 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4344 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4345 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4346 fprintf(stderr, " -V, --version Show version number.\n");
4347 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4348 fprintf(stderr, " -q, --quiet No output at all.\n");
4349 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4350 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4351 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4352 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4353 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4354 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4355 fprintf(stderr, " -l --load PATH Load session configuration\n");
4356 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4357 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4358 }
4359
4360 /*
4361 * Take an option from the getopt output and set it in the right variable to be
4362 * used later.
4363 *
4364 * Return 0 on success else a negative value.
4365 */
4366 static int set_option(int opt, const char *arg, const char *optname)
4367 {
4368 int ret = 0;
4369
4370 if (arg && arg[0] == '\0') {
4371 /*
4372 * This only happens if the value is read from daemon config
4373 * file. This means the option requires an argument and the
4374 * configuration file contains a line such as:
4375 * my_option =
4376 */
4377 ret = -EINVAL;
4378 goto end;
4379 }
4380
4381 switch (opt) {
4382 case 0:
4383 fprintf(stderr, "option %s", optname);
4384 if (arg) {
4385 fprintf(stderr, " with arg %s\n", arg);
4386 }
4387 break;
4388 case 'c':
4389 if (lttng_is_setuid_setgid()) {
4390 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4391 "-c, --client-sock");
4392 } else {
4393 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4394 }
4395 break;
4396 case 'a':
4397 if (lttng_is_setuid_setgid()) {
4398 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4399 "-a, --apps-sock");
4400 } else {
4401 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4402 }
4403 break;
4404 case 'd':
4405 opt_daemon = 1;
4406 break;
4407 case 'b':
4408 opt_background = 1;
4409 break;
4410 case 'g':
4411 if (lttng_is_setuid_setgid()) {
4412 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4413 "-g, --group");
4414 } else {
4415 /*
4416 * If the override option is set, the pointer points to a
4417 * *non* const thus freeing it even though the variable type is
4418 * set to const.
4419 */
4420 if (tracing_group_name_override) {
4421 free((void *) tracing_group_name);
4422 }
4423 tracing_group_name = strdup(arg);
4424 if (!tracing_group_name) {
4425 PERROR("strdup");
4426 ret = -ENOMEM;
4427 }
4428 tracing_group_name_override = 1;
4429 }
4430 break;
4431 case 'h':
4432 usage();
4433 exit(EXIT_SUCCESS);
4434 case 'V':
4435 fprintf(stdout, "%s\n", VERSION);
4436 exit(EXIT_SUCCESS);
4437 case 'S':
4438 opt_sig_parent = 1;
4439 break;
4440 case 'E':
4441 if (lttng_is_setuid_setgid()) {
4442 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4443 "--kconsumerd-err-sock");
4444 } else {
4445 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4446 }
4447 break;
4448 case 'C':
4449 if (lttng_is_setuid_setgid()) {
4450 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4451 "--kconsumerd-cmd-sock");
4452 } else {
4453 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4454 }
4455 break;
4456 case 'F':
4457 if (lttng_is_setuid_setgid()) {
4458 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4459 "--ustconsumerd64-err-sock");
4460 } else {
4461 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4462 }
4463 break;
4464 case 'D':
4465 if (lttng_is_setuid_setgid()) {
4466 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4467 "--ustconsumerd64-cmd-sock");
4468 } else {
4469 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4470 }
4471 break;
4472 case 'H':
4473 if (lttng_is_setuid_setgid()) {
4474 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4475 "--ustconsumerd32-err-sock");
4476 } else {
4477 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4478 }
4479 break;
4480 case 'G':
4481 if (lttng_is_setuid_setgid()) {
4482 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4483 "--ustconsumerd32-cmd-sock");
4484 } else {
4485 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4486 }
4487 break;
4488 case 'N':
4489 opt_no_kernel = 1;
4490 break;
4491 case 'q':
4492 lttng_opt_quiet = 1;
4493 break;
4494 case 'v':
4495 /* Verbose level can increase using multiple -v */
4496 if (arg) {
4497 /* Value obtained from config file */
4498 lttng_opt_verbose = config_parse_value(arg);
4499 } else {
4500 /* -v used on command line */
4501 lttng_opt_verbose++;
4502 }
4503 /* Clamp value to [0, 3] */
4504 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4505 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4506 break;
4507 case 'Z':
4508 if (arg) {
4509 opt_verbose_consumer = config_parse_value(arg);
4510 } else {
4511 opt_verbose_consumer += 1;
4512 }
4513 break;
4514 case 'u':
4515 if (lttng_is_setuid_setgid()) {
4516 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4517 "--consumerd32-path");
4518 } else {
4519 if (consumerd32_bin_override) {
4520 free((void *) consumerd32_bin);
4521 }
4522 consumerd32_bin = strdup(arg);
4523 if (!consumerd32_bin) {
4524 PERROR("strdup");
4525 ret = -ENOMEM;
4526 }
4527 consumerd32_bin_override = 1;
4528 }
4529 break;
4530 case 'U':
4531 if (lttng_is_setuid_setgid()) {
4532 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4533 "--consumerd32-libdir");
4534 } else {
4535 if (consumerd32_libdir_override) {
4536 free((void *) consumerd32_libdir);
4537 }
4538 consumerd32_libdir = strdup(arg);
4539 if (!consumerd32_libdir) {
4540 PERROR("strdup");
4541 ret = -ENOMEM;
4542 }
4543 consumerd32_libdir_override = 1;
4544 }
4545 break;
4546 case 't':
4547 if (lttng_is_setuid_setgid()) {
4548 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4549 "--consumerd64-path");
4550 } else {
4551 if (consumerd64_bin_override) {
4552 free((void *) consumerd64_bin);
4553 }
4554 consumerd64_bin = strdup(arg);
4555 if (!consumerd64_bin) {
4556 PERROR("strdup");
4557 ret = -ENOMEM;
4558 }
4559 consumerd64_bin_override = 1;
4560 }
4561 break;
4562 case 'T':
4563 if (lttng_is_setuid_setgid()) {
4564 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4565 "--consumerd64-libdir");
4566 } else {
4567 if (consumerd64_libdir_override) {
4568 free((void *) consumerd64_libdir);
4569 }
4570 consumerd64_libdir = strdup(arg);
4571 if (!consumerd64_libdir) {
4572 PERROR("strdup");
4573 ret = -ENOMEM;
4574 }
4575 consumerd64_libdir_override = 1;
4576 }
4577 break;
4578 case 'p':
4579 if (lttng_is_setuid_setgid()) {
4580 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4581 "-p, --pidfile");
4582 } else {
4583 free(opt_pidfile);
4584 opt_pidfile = strdup(arg);
4585 if (!opt_pidfile) {
4586 PERROR("strdup");
4587 ret = -ENOMEM;
4588 }
4589 }
4590 break;
4591 case 'J': /* Agent TCP port. */
4592 {
4593 if (lttng_is_setuid_setgid()) {
4594 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4595 "--agent-tcp-port");
4596 } else {
4597 unsigned long v;
4598
4599 errno = 0;
4600 v = strtoul(arg, NULL, 0);
4601 if (errno != 0 || !isdigit(arg[0])) {
4602 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4603 return -1;
4604 }
4605 if (v == 0 || v >= 65535) {
4606 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4607 return -1;
4608 }
4609 agent_tcp_port = (uint32_t) v;
4610 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4611 }
4612 break;
4613 }
4614 case 'l':
4615 if (lttng_is_setuid_setgid()) {
4616 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4617 "-l, --load");
4618 } else {
4619 free(opt_load_session_path);
4620 opt_load_session_path = strdup(arg);
4621 if (!opt_load_session_path) {
4622 PERROR("strdup");
4623 ret = -ENOMEM;
4624 }
4625 }
4626 break;
4627 case 'P': /* probe modules list */
4628 if (lttng_is_setuid_setgid()) {
4629 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4630 "--kmod-probes");
4631 } else {
4632 free(kmod_probes_list);
4633 kmod_probes_list = strdup(arg);
4634 if (!kmod_probes_list) {
4635 PERROR("strdup");
4636 ret = -ENOMEM;
4637 }
4638 }
4639 break;
4640 case 'e':
4641 if (lttng_is_setuid_setgid()) {
4642 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4643 "--extra-kmod-probes");
4644 } else {
4645 free(kmod_extra_probes_list);
4646 kmod_extra_probes_list = strdup(arg);
4647 if (!kmod_extra_probes_list) {
4648 PERROR("strdup");
4649 ret = -ENOMEM;
4650 }
4651 }
4652 break;
4653 case 'f':
4654 /* This is handled in set_options() thus silent break. */
4655 break;
4656 default:
4657 /* Unknown option or other error.
4658 * Error is printed by getopt, just return */
4659 ret = -1;
4660 }
4661
4662 end:
4663 if (ret == -EINVAL) {
4664 const char *opt_name = "unknown";
4665 int i;
4666
4667 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4668 i++) {
4669 if (opt == long_options[i].val) {
4670 opt_name = long_options[i].name;
4671 break;
4672 }
4673 }
4674
4675 WARN("Invalid argument provided for option \"%s\", using default value.",
4676 opt_name);
4677 }
4678
4679 return ret;
4680 }
4681
4682 /*
4683 * config_entry_handler_cb used to handle options read from a config file.
4684 * See config_entry_handler_cb comment in common/config/config.h for the
4685 * return value conventions.
4686 */
4687 static int config_entry_handler(const struct config_entry *entry, void *unused)
4688 {
4689 int ret = 0, i;
4690
4691 if (!entry || !entry->name || !entry->value) {
4692 ret = -EINVAL;
4693 goto end;
4694 }
4695
4696 /* Check if the option is to be ignored */
4697 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4698 if (!strcmp(entry->name, config_ignore_options[i])) {
4699 goto end;
4700 }
4701 }
4702
4703 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4704 i++) {
4705
4706 /* Ignore if not fully matched. */
4707 if (strcmp(entry->name, long_options[i].name)) {
4708 continue;
4709 }
4710
4711 /*
4712 * If the option takes no argument on the command line, we have to
4713 * check if the value is "true". We support non-zero numeric values,
4714 * true, on and yes.
4715 */
4716 if (!long_options[i].has_arg) {
4717 ret = config_parse_value(entry->value);
4718 if (ret <= 0) {
4719 if (ret) {
4720 WARN("Invalid configuration value \"%s\" for option %s",
4721 entry->value, entry->name);
4722 }
4723 /* False, skip boolean config option. */
4724 goto end;
4725 }
4726 }
4727
4728 ret = set_option(long_options[i].val, entry->value, entry->name);
4729 goto end;
4730 }
4731
4732 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4733
4734 end:
4735 return ret;
4736 }
4737
4738 /*
4739 * daemon configuration loading and argument parsing
4740 */
4741 static int set_options(int argc, char **argv)
4742 {
4743 int ret = 0, c = 0, option_index = 0;
4744 int orig_optopt = optopt, orig_optind = optind;
4745 char *optstring;
4746 const char *config_path = NULL;
4747
4748 optstring = utils_generate_optstring(long_options,
4749 sizeof(long_options) / sizeof(struct option));
4750 if (!optstring) {
4751 ret = -ENOMEM;
4752 goto end;
4753 }
4754
4755 /* Check for the --config option */
4756 while ((c = getopt_long(argc, argv, optstring, long_options,
4757 &option_index)) != -1) {
4758 if (c == '?') {
4759 ret = -EINVAL;
4760 goto end;
4761 } else if (c != 'f') {
4762 /* if not equal to --config option. */
4763 continue;
4764 }
4765
4766 if (lttng_is_setuid_setgid()) {
4767 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4768 "-f, --config");
4769 } else {
4770 config_path = utils_expand_path(optarg);
4771 if (!config_path) {
4772 ERR("Failed to resolve path: %s", optarg);
4773 }
4774 }
4775 }
4776
4777 ret = config_get_section_entries(config_path, config_section_name,
4778 config_entry_handler, NULL);
4779 if (ret) {
4780 if (ret > 0) {
4781 ERR("Invalid configuration option at line %i", ret);
4782 ret = -1;
4783 }
4784 goto end;
4785 }
4786
4787 /* Reset getopt's global state */
4788 optopt = orig_optopt;
4789 optind = orig_optind;
4790 while (1) {
4791 c = getopt_long(argc, argv, optstring, long_options, &option_index);
4792 if (c == -1) {
4793 break;
4794 }
4795
4796 ret = set_option(c, optarg, long_options[option_index].name);
4797 if (ret < 0) {
4798 break;
4799 }
4800 }
4801
4802 end:
4803 free(optstring);
4804 return ret;
4805 }
4806
4807 /*
4808 * Creates the two needed socket by the daemon.
4809 * apps_sock - The communication socket for all UST apps.
4810 * client_sock - The communication of the cli tool (lttng).
4811 */
4812 static int init_daemon_socket(void)
4813 {
4814 int ret = 0;
4815 mode_t old_umask;
4816
4817 old_umask = umask(0);
4818
4819 /* Create client tool unix socket */
4820 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4821 if (client_sock < 0) {
4822 ERR("Create unix sock failed: %s", client_unix_sock_path);
4823 ret = -1;
4824 goto end;
4825 }
4826
4827 /* Set the cloexec flag */
4828 ret = utils_set_fd_cloexec(client_sock);
4829 if (ret < 0) {
4830 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4831 "Continuing but note that the consumer daemon will have a "
4832 "reference to this socket on exec()", client_sock);
4833 }
4834
4835 /* File permission MUST be 660 */
4836 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4837 if (ret < 0) {
4838 ERR("Set file permissions failed: %s", client_unix_sock_path);
4839 PERROR("chmod");
4840 goto end;
4841 }
4842
4843 /* Create the application unix socket */
4844 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4845 if (apps_sock < 0) {
4846 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4847 ret = -1;
4848 goto end;
4849 }
4850
4851 /* Set the cloexec flag */
4852 ret = utils_set_fd_cloexec(apps_sock);
4853 if (ret < 0) {
4854 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4855 "Continuing but note that the consumer daemon will have a "
4856 "reference to this socket on exec()", apps_sock);
4857 }
4858
4859 /* File permission MUST be 666 */
4860 ret = chmod(apps_unix_sock_path,
4861 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4862 if (ret < 0) {
4863 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4864 PERROR("chmod");
4865 goto end;
4866 }
4867
4868 DBG3("Session daemon client socket %d and application socket %d created",
4869 client_sock, apps_sock);
4870
4871 end:
4872 umask(old_umask);
4873 return ret;
4874 }
4875
4876 /*
4877 * Check if the global socket is available, and if a daemon is answering at the
4878 * other side. If yes, error is returned.
4879 */
4880 static int check_existing_daemon(void)
4881 {
4882 /* Is there anybody out there ? */
4883 if (lttng_session_daemon_alive()) {
4884 return -EEXIST;
4885 }
4886
4887 return 0;
4888 }
4889
4890 /*
4891 * Set the tracing group gid onto the client socket.
4892 *
4893 * Race window between mkdir and chown is OK because we are going from more
4894 * permissive (root.root) to less permissive (root.tracing).
4895 */
4896 static int set_permissions(char *rundir)
4897 {
4898 int ret;
4899 gid_t gid;
4900
4901 gid = utils_get_group_id(tracing_group_name);
4902
4903 /* Set lttng run dir */
4904 ret = chown(rundir, 0, gid);
4905 if (ret < 0) {
4906 ERR("Unable to set group on %s", rundir);
4907 PERROR("chown");
4908 }
4909
4910 /*
4911 * Ensure all applications and tracing group can search the run
4912 * dir. Allow everyone to read the directory, since it does not
4913 * buy us anything to hide its content.
4914 */
4915 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4916 if (ret < 0) {
4917 ERR("Unable to set permissions on %s", rundir);
4918 PERROR("chmod");
4919 }
4920
4921 /* lttng client socket path */
4922 ret = chown(client_unix_sock_path, 0, gid);
4923 if (ret < 0) {
4924 ERR("Unable to set group on %s", client_unix_sock_path);
4925 PERROR("chown");
4926 }
4927
4928 /* kconsumer error socket path */
4929 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4930 if (ret < 0) {
4931 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4932 PERROR("chown");
4933 }
4934
4935 /* 64-bit ustconsumer error socket path */
4936 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4937 if (ret < 0) {
4938 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4939 PERROR("chown");
4940 }
4941
4942 /* 32-bit ustconsumer compat32 error socket path */
4943 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4944 if (ret < 0) {
4945 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4946 PERROR("chown");
4947 }
4948
4949 DBG("All permissions are set");
4950
4951 return ret;
4952 }
4953
4954 /*
4955 * Create the lttng run directory needed for all global sockets and pipe.
4956 */
4957 static int create_lttng_rundir(const char *rundir)
4958 {
4959 int ret;
4960
4961 DBG3("Creating LTTng run directory: %s", rundir);
4962
4963 ret = mkdir(rundir, S_IRWXU);
4964 if (ret < 0) {
4965 if (errno != EEXIST) {
4966 ERR("Unable to create %s", rundir);
4967 goto error;
4968 } else {
4969 ret = 0;
4970 }
4971 }
4972
4973 error:
4974 return ret;
4975 }
4976
4977 /*
4978 * Setup sockets and directory needed by the kconsumerd communication with the
4979 * session daemon.
4980 */
4981 static int set_consumer_sockets(struct consumer_data *consumer_data,
4982 const char *rundir)
4983 {
4984 int ret;
4985 char path[PATH_MAX];
4986
4987 switch (consumer_data->type) {
4988 case LTTNG_CONSUMER_KERNEL:
4989 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4990 break;
4991 case LTTNG_CONSUMER64_UST:
4992 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4993 break;
4994 case LTTNG_CONSUMER32_UST:
4995 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4996 break;
4997 default:
4998 ERR("Consumer type unknown");
4999 ret = -EINVAL;
5000 goto error;
5001 }
5002
5003 DBG2("Creating consumer directory: %s", path);
5004
5005 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5006 if (ret < 0) {
5007 if (errno != EEXIST) {
5008 PERROR("mkdir");
5009 ERR("Failed to create %s", path);
5010 goto error;
5011 }
5012 ret = -1;
5013 }
5014 if (is_root) {
5015 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5016 if (ret < 0) {
5017 ERR("Unable to set group on %s", path);
5018 PERROR("chown");
5019 goto error;
5020 }
5021 }
5022
5023 /* Create the kconsumerd error unix socket */
5024 consumer_data->err_sock =
5025 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5026 if (consumer_data->err_sock < 0) {
5027 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5028 ret = -1;
5029 goto error;
5030 }
5031
5032 /*
5033 * Set the CLOEXEC flag. Return code is useless because either way, the
5034 * show must go on.
5035 */
5036 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5037 if (ret < 0) {
5038 PERROR("utils_set_fd_cloexec");
5039 /* continue anyway */
5040 }
5041
5042 /* File permission MUST be 660 */
5043 ret = chmod(consumer_data->err_unix_sock_path,
5044 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5045 if (ret < 0) {
5046 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5047 PERROR("chmod");
5048 goto error;
5049 }
5050
5051 error:
5052 return ret;
5053 }
5054
5055 /*
5056 * Signal handler for the daemon
5057 *
5058 * Simply stop all worker threads, leaving main() return gracefully after
5059 * joining all threads and calling cleanup().
5060 */
5061 static void sighandler(int sig)
5062 {
5063 switch (sig) {
5064 case SIGPIPE:
5065 DBG("SIGPIPE caught");
5066 return;
5067 case SIGINT:
5068 DBG("SIGINT caught");
5069 stop_threads();
5070 break;
5071 case SIGTERM:
5072 DBG("SIGTERM caught");
5073 stop_threads();
5074 break;
5075 case SIGUSR1:
5076 CMM_STORE_SHARED(recv_child_signal, 1);
5077 break;
5078 default:
5079 break;
5080 }
5081 }
5082
5083 /*
5084 * Setup signal handler for :
5085 * SIGINT, SIGTERM, SIGPIPE
5086 */
5087 static int set_signal_handler(void)
5088 {
5089 int ret = 0;
5090 struct sigaction sa;
5091 sigset_t sigset;
5092
5093 if ((ret = sigemptyset(&sigset)) < 0) {
5094 PERROR("sigemptyset");
5095 return ret;
5096 }
5097
5098 sa.sa_handler = sighandler;
5099 sa.sa_mask = sigset;
5100 sa.sa_flags = 0;
5101 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5102 PERROR("sigaction");
5103 return ret;
5104 }
5105
5106 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5107 PERROR("sigaction");
5108 return ret;
5109 }
5110
5111 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5112 PERROR("sigaction");
5113 return ret;
5114 }
5115
5116 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5117 PERROR("sigaction");
5118 return ret;
5119 }
5120
5121 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5122
5123 return ret;
5124 }
5125
5126 /*
5127 * Set open files limit to unlimited. This daemon can open a large number of
5128 * file descriptors in order to consumer multiple kernel traces.
5129 */
5130 static void set_ulimit(void)
5131 {
5132 int ret;
5133 struct rlimit lim;
5134
5135 /* The kernel does not allowed an infinite limit for open files */
5136 lim.rlim_cur = 65535;
5137 lim.rlim_max = 65535;
5138
5139 ret = setrlimit(RLIMIT_NOFILE, &lim);
5140 if (ret < 0) {
5141 PERROR("failed to set open files limit");
5142 }
5143 }
5144
5145 /*
5146 * Write pidfile using the rundir and opt_pidfile.
5147 */
5148 static int write_pidfile(void)
5149 {
5150 int ret;
5151 char pidfile_path[PATH_MAX];
5152
5153 assert(rundir);
5154
5155 if (opt_pidfile) {
5156 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5157 } else {
5158 /* Build pidfile path from rundir and opt_pidfile. */
5159 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5160 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5161 if (ret < 0) {
5162 PERROR("snprintf pidfile path");
5163 goto error;
5164 }
5165 }
5166
5167 /*
5168 * Create pid file in rundir.
5169 */
5170 ret = utils_create_pid_file(getpid(), pidfile_path);
5171 error:
5172 return ret;
5173 }
5174
5175 /*
5176 * Create lockfile using the rundir and return its fd.
5177 */
5178 static int create_lockfile(void)
5179 {
5180 int ret;
5181 char lockfile_path[PATH_MAX];
5182
5183 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5184 if (ret < 0) {
5185 goto error;
5186 }
5187
5188 ret = utils_create_lock_file(lockfile_path);
5189 error:
5190 return ret;
5191 }
5192
5193 /*
5194 * Write agent TCP port using the rundir.
5195 */
5196 static int write_agent_port(void)
5197 {
5198 int ret;
5199 char path[PATH_MAX];
5200
5201 assert(rundir);
5202
5203 ret = snprintf(path, sizeof(path), "%s/"
5204 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5205 if (ret < 0) {
5206 PERROR("snprintf agent port path");
5207 goto error;
5208 }
5209
5210 /*
5211 * Create TCP agent port file in rundir.
5212 */
5213 ret = utils_create_pid_file(agent_tcp_port, path);
5214
5215 error:
5216 return ret;
5217 }
5218
5219 /*
5220 * main
5221 */
5222 int main(int argc, char **argv)
5223 {
5224 int ret = 0, retval = 0;
5225 void *status;
5226 const char *home_path, *env_app_timeout;
5227
5228 init_kernel_workarounds();
5229
5230 rcu_register_thread();
5231
5232 if (set_signal_handler()) {
5233 retval = -1;
5234 goto exit_set_signal_handler;
5235 }
5236
5237 setup_consumerd_path();
5238
5239 page_size = sysconf(_SC_PAGESIZE);
5240 if (page_size < 0) {
5241 PERROR("sysconf _SC_PAGESIZE");
5242 page_size = LONG_MAX;
5243 WARN("Fallback page size to %ld", page_size);
5244 }
5245
5246 /*
5247 * Parse arguments and load the daemon configuration file.
5248 *
5249 * We have an exit_options exit path to free memory reserved by
5250 * set_options. This is needed because the rest of sessiond_cleanup()
5251 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5252 * depends on set_options.
5253 */
5254 progname = argv[0];
5255 if (set_options(argc, argv)) {
5256 retval = -1;
5257 goto exit_options;
5258 }
5259
5260 /* Daemonize */
5261 if (opt_daemon || opt_background) {
5262 int i;
5263
5264 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5265 !opt_background);
5266 if (ret < 0) {
5267 retval = -1;
5268 goto exit_options;
5269 }
5270
5271 /*
5272 * We are in the child. Make sure all other file descriptors are
5273 * closed, in case we are called with more opened file
5274 * descriptors than the standard ones.
5275 */
5276 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5277 (void) close(i);
5278 }
5279 }
5280
5281 /*
5282 * Starting from here, we can create threads. This needs to be after
5283 * lttng_daemonize due to RCU.
5284 */
5285
5286 /*
5287 * Initialize the health check subsystem. This call should set the
5288 * appropriate time values.
5289 */
5290 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5291 if (!health_sessiond) {
5292 PERROR("health_app_create error");
5293 retval = -1;
5294 goto exit_health_sessiond_cleanup;
5295 }
5296
5297 if (init_ht_cleanup_quit_pipe()) {
5298 retval = -1;
5299 goto exit_ht_cleanup_quit_pipe;
5300 }
5301
5302 /* Setup the thread ht_cleanup communication pipe. */
5303 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5304 retval = -1;
5305 goto exit_ht_cleanup_pipe;
5306 }
5307
5308 /* Set up max poll set size */
5309 if (lttng_poll_set_max_size()) {
5310 retval = -1;
5311 goto exit_set_max_size;
5312 }
5313
5314 /* Create thread to clean up RCU hash tables */
5315 ret = pthread_create(&ht_cleanup_thread, NULL,
5316 thread_ht_cleanup, (void *) NULL);
5317 if (ret) {
5318 errno = ret;
5319 PERROR("pthread_create ht_cleanup");
5320 retval = -1;
5321 goto exit_ht_cleanup;
5322 }
5323
5324 /* Create thread quit pipe */
5325 if (init_thread_quit_pipe()) {
5326 retval = -1;
5327 goto exit_init_data;
5328 }
5329
5330 /* Check if daemon is UID = 0 */
5331 is_root = !getuid();
5332
5333 if (is_root) {
5334 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5335 if (!rundir) {
5336 retval = -1;
5337 goto exit_init_data;
5338 }
5339
5340 /* Create global run dir with root access */
5341 if (create_lttng_rundir(rundir)) {
5342 retval = -1;
5343 goto exit_init_data;
5344 }
5345
5346 if (strlen(apps_unix_sock_path) == 0) {
5347 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5348 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5349 if (ret < 0) {
5350 retval = -1;
5351 goto exit_init_data;
5352 }
5353 }
5354
5355 if (strlen(client_unix_sock_path) == 0) {
5356 ret = snprintf(client_unix_sock_path, PATH_MAX,
5357 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5358 if (ret < 0) {
5359 retval = -1;
5360 goto exit_init_data;
5361 }
5362 }
5363
5364 /* Set global SHM for ust */
5365 if (strlen(wait_shm_path) == 0) {
5366 ret = snprintf(wait_shm_path, PATH_MAX,
5367 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5368 if (ret < 0) {
5369 retval = -1;
5370 goto exit_init_data;
5371 }
5372 }
5373
5374 if (strlen(health_unix_sock_path) == 0) {
5375 ret = snprintf(health_unix_sock_path,
5376 sizeof(health_unix_sock_path),
5377 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5378 if (ret < 0) {
5379 retval = -1;
5380 goto exit_init_data;
5381 }
5382 }
5383
5384 /* Setup kernel consumerd path */
5385 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5386 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5387 if (ret < 0) {
5388 retval = -1;
5389 goto exit_init_data;
5390 }
5391 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5392 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5393 if (ret < 0) {
5394 retval = -1;
5395 goto exit_init_data;
5396 }
5397
5398 DBG2("Kernel consumer err path: %s",
5399 kconsumer_data.err_unix_sock_path);
5400 DBG2("Kernel consumer cmd path: %s",
5401 kconsumer_data.cmd_unix_sock_path);
5402 } else {
5403 home_path = utils_get_home_dir();
5404 if (home_path == NULL) {
5405 /* TODO: Add --socket PATH option */
5406 ERR("Can't get HOME directory for sockets creation.");
5407 retval = -1;
5408 goto exit_init_data;
5409 }
5410
5411 /*
5412 * Create rundir from home path. This will create something like
5413 * $HOME/.lttng
5414 */
5415 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5416 if (ret < 0) {
5417 retval = -1;
5418 goto exit_init_data;
5419 }
5420
5421 if (create_lttng_rundir(rundir)) {
5422 retval = -1;
5423 goto exit_init_data;
5424 }
5425
5426 if (strlen(apps_unix_sock_path) == 0) {
5427 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5428 DEFAULT_HOME_APPS_UNIX_SOCK,
5429 home_path);
5430 if (ret < 0) {
5431 retval = -1;
5432 goto exit_init_data;
5433 }
5434 }
5435
5436 /* Set the cli tool unix socket path */
5437 if (strlen(client_unix_sock_path) == 0) {
5438 ret = snprintf(client_unix_sock_path, PATH_MAX,
5439 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5440 home_path);
5441 if (ret < 0) {
5442 retval = -1;
5443 goto exit_init_data;
5444 }
5445 }
5446
5447 /* Set global SHM for ust */
5448 if (strlen(wait_shm_path) == 0) {
5449 ret = snprintf(wait_shm_path, PATH_MAX,
5450 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5451 getuid());
5452 if (ret < 0) {
5453 retval = -1;
5454 goto exit_init_data;
5455 }
5456 }
5457
5458 /* Set health check Unix path */
5459 if (strlen(health_unix_sock_path) == 0) {
5460 ret = snprintf(health_unix_sock_path,
5461 sizeof(health_unix_sock_path),
5462 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5463 home_path);
5464 if (ret < 0) {
5465 retval = -1;
5466 goto exit_init_data;
5467 }
5468 }
5469 }
5470
5471 lockfile_fd = create_lockfile();
5472 if (lockfile_fd < 0) {
5473 retval = -1;
5474 goto exit_init_data;
5475 }
5476
5477 /* Set consumer initial state */
5478 kernel_consumerd_state = CONSUMER_STOPPED;
5479 ust_consumerd_state = CONSUMER_STOPPED;
5480
5481 DBG("Client socket path %s", client_unix_sock_path);
5482 DBG("Application socket path %s", apps_unix_sock_path);
5483 DBG("Application wait path %s", wait_shm_path);
5484 DBG("LTTng run directory path: %s", rundir);
5485
5486 /* 32 bits consumerd path setup */
5487 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5488 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5489 if (ret < 0) {
5490 PERROR("snprintf 32-bit consumer error socket path");
5491 retval = -1;
5492 goto exit_init_data;
5493 }
5494 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5495 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5496 if (ret < 0) {
5497 PERROR("snprintf 32-bit consumer command socket path");
5498 retval = -1;
5499 goto exit_init_data;
5500 }
5501
5502 DBG2("UST consumer 32 bits err path: %s",
5503 ustconsumer32_data.err_unix_sock_path);
5504 DBG2("UST consumer 32 bits cmd path: %s",
5505 ustconsumer32_data.cmd_unix_sock_path);
5506
5507 /* 64 bits consumerd path setup */
5508 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5509 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5510 if (ret < 0) {
5511 PERROR("snprintf 64-bit consumer error socket path");
5512 retval = -1;
5513 goto exit_init_data;
5514 }
5515 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5516 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5517 if (ret < 0) {
5518 PERROR("snprintf 64-bit consumer command socket path");
5519 retval = -1;
5520 goto exit_init_data;
5521 }
5522
5523 DBG2("UST consumer 64 bits err path: %s",
5524 ustconsumer64_data.err_unix_sock_path);
5525 DBG2("UST consumer 64 bits cmd path: %s",
5526 ustconsumer64_data.cmd_unix_sock_path);
5527
5528 /*
5529 * See if daemon already exist.
5530 */
5531 if (check_existing_daemon()) {
5532 ERR("Already running daemon.\n");
5533 /*
5534 * We do not goto exit because we must not cleanup()
5535 * because a daemon is already running.
5536 */
5537 retval = -1;
5538 goto exit_init_data;
5539 }
5540
5541 /*
5542 * Init UST app hash table. Alloc hash table before this point since
5543 * cleanup() can get called after that point.
5544 */
5545 if (ust_app_ht_alloc()) {
5546 retval = -1;
5547 goto exit_init_data;
5548 }
5549
5550 /* Initialize agent domain subsystem. */
5551 if (agent_setup()) {
5552 /* ENOMEM at this point. */
5553 retval = -1;
5554 goto exit_init_data;
5555 }
5556
5557 /*
5558 * These actions must be executed as root. We do that *after* setting up
5559 * the sockets path because we MUST make the check for another daemon using
5560 * those paths *before* trying to set the kernel consumer sockets and init
5561 * kernel tracer.
5562 */
5563 if (is_root) {
5564 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5565 retval = -1;
5566 goto exit_init_data;
5567 }
5568
5569 /* Setup kernel tracer */
5570 if (!opt_no_kernel) {
5571 init_kernel_tracer();
5572 if (kernel_tracer_fd >= 0) {
5573 ret = syscall_init_table();
5574 if (ret < 0) {
5575 ERR("Unable to populate syscall table. "
5576 "Syscall tracing won't work "
5577 "for this session daemon.");
5578 }
5579 }
5580 }
5581
5582 /* Set ulimit for open files */
5583 set_ulimit();
5584 }
5585 /* init lttng_fd tracking must be done after set_ulimit. */
5586 lttng_fd_init();
5587
5588 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5589 retval = -1;
5590 goto exit_init_data;
5591 }
5592
5593 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5594 retval = -1;
5595 goto exit_init_data;
5596 }
5597
5598 /* Setup the needed unix socket */
5599 if (init_daemon_socket()) {
5600 retval = -1;
5601 goto exit_init_data;
5602 }
5603
5604 /* Set credentials to socket */
5605 if (is_root && set_permissions(rundir)) {
5606 retval = -1;
5607 goto exit_init_data;
5608 }
5609
5610 /* Get parent pid if -S, --sig-parent is specified. */
5611 if (opt_sig_parent) {
5612 ppid = getppid();
5613 }
5614
5615 /* Setup the kernel pipe for waking up the kernel thread */
5616 if (is_root && !opt_no_kernel) {
5617 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5618 retval = -1;
5619 goto exit_init_data;
5620 }
5621 }
5622
5623 /* Setup the thread apps communication pipe. */
5624 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5625 retval = -1;
5626 goto exit_init_data;
5627 }
5628
5629 /* Setup the thread apps notify communication pipe. */
5630 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5631 retval = -1;
5632 goto exit_init_data;
5633 }
5634
5635 /* Initialize global buffer per UID and PID registry. */
5636 buffer_reg_init_uid_registry();
5637 buffer_reg_init_pid_registry();
5638
5639 /* Init UST command queue. */
5640 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5641
5642 /*
5643 * Get session list pointer. This pointer MUST NOT be free'd. This list
5644 * is statically declared in session.c
5645 */
5646 session_list_ptr = session_get_list();
5647
5648 cmd_init();
5649
5650 /* Check for the application socket timeout env variable. */
5651 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5652 if (env_app_timeout) {
5653 app_socket_timeout = atoi(env_app_timeout);
5654 } else {
5655 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5656 }
5657
5658 ret = write_pidfile();
5659 if (ret) {
5660 ERR("Error in write_pidfile");
5661 retval = -1;
5662 goto exit_init_data;
5663 }
5664 ret = write_agent_port();
5665 if (ret) {
5666 ERR("Error in write_agent_port");
5667 retval = -1;
5668 goto exit_init_data;
5669 }
5670
5671 /* Initialize communication library */
5672 lttcomm_init();
5673 /* Initialize TCP timeout values */
5674 lttcomm_inet_init();
5675
5676 if (load_session_init_data(&load_info) < 0) {
5677 retval = -1;
5678 goto exit_init_data;
5679 }
5680 load_info->path = opt_load_session_path;
5681
5682 /* Create health-check thread */
5683 ret = pthread_create(&health_thread, NULL,
5684 thread_manage_health, (void *) NULL);
5685 if (ret) {
5686 errno = ret;
5687 PERROR("pthread_create health");
5688 retval = -1;
5689 goto exit_health;
5690 }
5691
5692 /* Create thread to manage the client socket */
5693 ret = pthread_create(&client_thread, NULL,
5694 thread_manage_clients, (void *) NULL);
5695 if (ret) {
5696 errno = ret;
5697 PERROR("pthread_create clients");
5698 retval = -1;
5699 goto exit_client;
5700 }
5701
5702 /* Create thread to dispatch registration */
5703 ret = pthread_create(&dispatch_thread, NULL,
5704 thread_dispatch_ust_registration, (void *) NULL);
5705 if (ret) {
5706 errno = ret;
5707 PERROR("pthread_create dispatch");
5708 retval = -1;
5709 goto exit_dispatch;
5710 }
5711
5712 /* Create thread to manage application registration. */
5713 ret = pthread_create(&reg_apps_thread, NULL,
5714 thread_registration_apps, (void *) NULL);
5715 if (ret) {
5716 errno = ret;
5717 PERROR("pthread_create registration");
5718 retval = -1;
5719 goto exit_reg_apps;
5720 }
5721
5722 /* Create thread to manage application socket */
5723 ret = pthread_create(&apps_thread, NULL,
5724 thread_manage_apps, (void *) NULL);
5725 if (ret) {
5726 errno = ret;
5727 PERROR("pthread_create apps");
5728 retval = -1;
5729 goto exit_apps;
5730 }
5731
5732 /* Create thread to manage application notify socket */
5733 ret = pthread_create(&apps_notify_thread, NULL,
5734 ust_thread_manage_notify, (void *) NULL);
5735 if (ret) {
5736 errno = ret;
5737 PERROR("pthread_create notify");
5738 retval = -1;
5739 goto exit_apps_notify;
5740 }
5741
5742 /* Create agent registration thread. */
5743 ret = pthread_create(&agent_reg_thread, NULL,
5744 agent_thread_manage_registration, (void *) NULL);
5745 if (ret) {
5746 errno = ret;
5747 PERROR("pthread_create agent");
5748 retval = -1;
5749 goto exit_agent_reg;
5750 }
5751
5752 /* Don't start this thread if kernel tracing is not requested nor root */
5753 if (is_root && !opt_no_kernel) {
5754 /* Create kernel thread to manage kernel event */
5755 ret = pthread_create(&kernel_thread, NULL,
5756 thread_manage_kernel, (void *) NULL);
5757 if (ret) {
5758 errno = ret;
5759 PERROR("pthread_create kernel");
5760 retval = -1;
5761 goto exit_kernel;
5762 }
5763 }
5764
5765 /* Create session loading thread. */
5766 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5767 load_info);
5768 if (ret) {
5769 errno = ret;
5770 PERROR("pthread_create load_session_thread");
5771 retval = -1;
5772 goto exit_load_session;
5773 }
5774
5775 /*
5776 * This is where we start awaiting program completion (e.g. through
5777 * signal that asks threads to teardown).
5778 */
5779
5780 ret = pthread_join(load_session_thread, &status);
5781 if (ret) {
5782 errno = ret;
5783 PERROR("pthread_join load_session_thread");
5784 retval = -1;
5785 }
5786 exit_load_session:
5787
5788 if (is_root && !opt_no_kernel) {
5789 ret = pthread_join(kernel_thread, &status);
5790 if (ret) {
5791 errno = ret;
5792 PERROR("pthread_join");
5793 retval = -1;
5794 }
5795 }
5796 exit_kernel:
5797
5798 ret = pthread_join(agent_reg_thread, &status);
5799 if (ret) {
5800 errno = ret;
5801 PERROR("pthread_join agent");
5802 retval = -1;
5803 }
5804 exit_agent_reg:
5805
5806 ret = pthread_join(apps_notify_thread, &status);
5807 if (ret) {
5808 errno = ret;
5809 PERROR("pthread_join apps notify");
5810 retval = -1;
5811 }
5812 exit_apps_notify:
5813
5814 ret = pthread_join(apps_thread, &status);
5815 if (ret) {
5816 errno = ret;
5817 PERROR("pthread_join apps");
5818 retval = -1;
5819 }
5820 exit_apps:
5821
5822 ret = pthread_join(reg_apps_thread, &status);
5823 if (ret) {
5824 errno = ret;
5825 PERROR("pthread_join");
5826 retval = -1;
5827 }
5828 exit_reg_apps:
5829
5830 ret = pthread_join(dispatch_thread, &status);
5831 if (ret) {
5832 errno = ret;
5833 PERROR("pthread_join");
5834 retval = -1;
5835 }
5836 exit_dispatch:
5837
5838 ret = pthread_join(client_thread, &status);
5839 if (ret) {
5840 errno = ret;
5841 PERROR("pthread_join");
5842 retval = -1;
5843 }
5844 exit_client:
5845
5846 ret = pthread_join(health_thread, &status);
5847 if (ret) {
5848 errno = ret;
5849 PERROR("pthread_join health thread");
5850 retval = -1;
5851 }
5852 exit_health:
5853
5854 exit_init_data:
5855 /*
5856 * sessiond_cleanup() is called when no other thread is running, except
5857 * the ht_cleanup thread, which is needed to destroy the hash tables.
5858 */
5859 rcu_thread_online();
5860 sessiond_cleanup();
5861 rcu_thread_offline();
5862 rcu_unregister_thread();
5863
5864 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
5865 if (ret < 0) {
5866 ERR("write error on ht_cleanup quit pipe");
5867 retval = -1;
5868 }
5869
5870 ret = pthread_join(ht_cleanup_thread, &status);
5871 if (ret) {
5872 errno = ret;
5873 PERROR("pthread_join ht cleanup thread");
5874 retval = -1;
5875 }
5876 exit_ht_cleanup:
5877 exit_set_max_size:
5878
5879 utils_close_pipe(ht_cleanup_pipe);
5880 exit_ht_cleanup_pipe:
5881
5882 /*
5883 * Close the ht_cleanup quit pipe.
5884 */
5885 utils_close_pipe(ht_cleanup_quit_pipe);
5886 exit_ht_cleanup_quit_pipe:
5887
5888 health_app_destroy(health_sessiond);
5889 exit_health_sessiond_cleanup:
5890
5891 exit_options:
5892 sessiond_cleanup_options();
5893
5894 exit_set_signal_handler:
5895 if (!retval) {
5896 exit(EXIT_SUCCESS);
5897 } else {
5898 exit(EXIT_FAILURE);
5899 }
5900 }
This page took 0.187444 seconds and 5 git commands to generate.