Fix: Initialize global agent_apps_ht_by_sock on session daemon launch
[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", required_argument, 0, 'c' },
137 { "apps-sock", required_argument, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
144 { "consumerd32-path", required_argument, 0, '\0' },
145 { "consumerd32-libdir", required_argument, 0, '\0' },
146 { "consumerd64-path", required_argument, 0, '\0' },
147 { "consumerd64-libdir", required_argument, 0, '\0' },
148 { "daemonize", no_argument, 0, 'd' },
149 { "background", no_argument, 0, 'b' },
150 { "sig-parent", no_argument, 0, 'S' },
151 { "help", no_argument, 0, 'h' },
152 { "group", required_argument, 0, 'g' },
153 { "version", no_argument, 0, 'V' },
154 { "quiet", no_argument, 0, 'q' },
155 { "verbose", no_argument, 0, 'v' },
156 { "verbose-consumer", no_argument, 0, '\0' },
157 { "no-kernel", no_argument, 0, '\0' },
158 { "pidfile", required_argument, 0, 'p' },
159 { "agent-tcp-port", required_argument, 0, '\0' },
160 { "config", required_argument, 0, 'f' },
161 { "load", required_argument, 0, 'l' },
162 { "kmod-probes", required_argument, 0, '\0' },
163 { "extra-kmod-probes", required_argument, 0, '\0' },
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("Cleaning up all agent apps");
680 agent_app_ht_clean();
681
682 DBG("Closing all UST sockets");
683 ust_app_clean_list();
684 buffer_reg_destroy_registries();
685
686 if (is_root && !opt_no_kernel) {
687 DBG2("Closing kernel fd");
688 if (kernel_tracer_fd >= 0) {
689 ret = close(kernel_tracer_fd);
690 if (ret) {
691 PERROR("close");
692 }
693 }
694 DBG("Unloading kernel modules");
695 modprobe_remove_lttng_all();
696 free(syscall_table);
697 }
698
699 close_consumer_sockets();
700
701 if (load_info) {
702 load_session_destroy_data(load_info);
703 free(load_info);
704 }
705
706 /*
707 * Cleanup lock file by deleting it and finaly closing it which will
708 * release the file system lock.
709 */
710 if (lockfile_fd >= 0) {
711 char lockfile_path[PATH_MAX];
712
713 ret = generate_lock_file_path(lockfile_path,
714 sizeof(lockfile_path));
715 if (ret > 0) {
716 ret = remove(lockfile_path);
717 if (ret < 0) {
718 PERROR("remove lock file");
719 }
720 ret = close(lockfile_fd);
721 if (ret < 0) {
722 PERROR("close lock file");
723 }
724 }
725 }
726
727 /*
728 * We do NOT rmdir rundir because there are other processes
729 * using it, for instance lttng-relayd, which can start in
730 * parallel with this teardown.
731 */
732
733 free(rundir);
734 }
735
736 /*
737 * Cleanup the daemon's option data structures.
738 */
739 static void sessiond_cleanup_options(void)
740 {
741 DBG("Cleaning up options");
742
743 /*
744 * If the override option is set, the pointer points to a *non* const
745 * thus freeing it even though the variable type is set to const.
746 */
747 if (tracing_group_name_override) {
748 free((void *) tracing_group_name);
749 }
750 if (consumerd32_bin_override) {
751 free((void *) consumerd32_bin);
752 }
753 if (consumerd64_bin_override) {
754 free((void *) consumerd64_bin);
755 }
756 if (consumerd32_libdir_override) {
757 free((void *) consumerd32_libdir);
758 }
759 if (consumerd64_libdir_override) {
760 free((void *) consumerd64_libdir);
761 }
762
763 free(opt_pidfile);
764 free(opt_load_session_path);
765 free(kmod_probes_list);
766 free(kmod_extra_probes_list);
767
768 /* <fun> */
769 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
770 "Matthew, BEET driven development works!%c[%dm",
771 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
772 /* </fun> */
773 }
774
775 /*
776 * Send data on a unix socket using the liblttsessiondcomm API.
777 *
778 * Return lttcomm error code.
779 */
780 static int send_unix_sock(int sock, void *buf, size_t len)
781 {
782 /* Check valid length */
783 if (len == 0) {
784 return -1;
785 }
786
787 return lttcomm_send_unix_sock(sock, buf, len);
788 }
789
790 /*
791 * Free memory of a command context structure.
792 */
793 static void clean_command_ctx(struct command_ctx **cmd_ctx)
794 {
795 DBG("Clean command context structure");
796 if (*cmd_ctx) {
797 if ((*cmd_ctx)->llm) {
798 free((*cmd_ctx)->llm);
799 }
800 if ((*cmd_ctx)->lsm) {
801 free((*cmd_ctx)->lsm);
802 }
803 free(*cmd_ctx);
804 *cmd_ctx = NULL;
805 }
806 }
807
808 /*
809 * Notify UST applications using the shm mmap futex.
810 */
811 static int notify_ust_apps(int active)
812 {
813 char *wait_shm_mmap;
814
815 DBG("Notifying applications of session daemon state: %d", active);
816
817 /* See shm.c for this call implying mmap, shm and futex calls */
818 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
819 if (wait_shm_mmap == NULL) {
820 goto error;
821 }
822
823 /* Wake waiting process */
824 futex_wait_update((int32_t *) wait_shm_mmap, active);
825
826 /* Apps notified successfully */
827 return 0;
828
829 error:
830 return -1;
831 }
832
833 /*
834 * Setup the outgoing data buffer for the response (llm) by allocating the
835 * right amount of memory and copying the original information from the lsm
836 * structure.
837 *
838 * Return total size of the buffer pointed by buf.
839 */
840 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
841 {
842 int ret, buf_size;
843
844 buf_size = size;
845
846 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
847 if (cmd_ctx->llm == NULL) {
848 PERROR("zmalloc");
849 ret = -ENOMEM;
850 goto error;
851 }
852
853 /* Copy common data */
854 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
855 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
856
857 cmd_ctx->llm->data_size = size;
858 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
859
860 return buf_size;
861
862 error:
863 return ret;
864 }
865
866 /*
867 * Update the kernel poll set of all channel fd available over all tracing
868 * session. Add the wakeup pipe at the end of the set.
869 */
870 static int update_kernel_poll(struct lttng_poll_event *events)
871 {
872 int ret;
873 struct ltt_session *session;
874 struct ltt_kernel_channel *channel;
875
876 DBG("Updating kernel poll set");
877
878 session_lock_list();
879 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
880 session_lock(session);
881 if (session->kernel_session == NULL) {
882 session_unlock(session);
883 continue;
884 }
885
886 cds_list_for_each_entry(channel,
887 &session->kernel_session->channel_list.head, list) {
888 /* Add channel fd to the kernel poll set */
889 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
890 if (ret < 0) {
891 session_unlock(session);
892 goto error;
893 }
894 DBG("Channel fd %d added to kernel set", channel->fd);
895 }
896 session_unlock(session);
897 }
898 session_unlock_list();
899
900 return 0;
901
902 error:
903 session_unlock_list();
904 return -1;
905 }
906
907 /*
908 * Find the channel fd from 'fd' over all tracing session. When found, check
909 * for new channel stream and send those stream fds to the kernel consumer.
910 *
911 * Useful for CPU hotplug feature.
912 */
913 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
914 {
915 int ret = 0;
916 struct ltt_session *session;
917 struct ltt_kernel_session *ksess;
918 struct ltt_kernel_channel *channel;
919
920 DBG("Updating kernel streams for channel fd %d", fd);
921
922 session_lock_list();
923 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
924 session_lock(session);
925 if (session->kernel_session == NULL) {
926 session_unlock(session);
927 continue;
928 }
929 ksess = session->kernel_session;
930
931 cds_list_for_each_entry(channel,
932 &ksess->channel_list.head, list) {
933 struct lttng_ht_iter iter;
934 struct consumer_socket *socket;
935
936 if (channel->fd != fd) {
937 continue;
938 }
939 DBG("Channel found, updating kernel streams");
940 ret = kernel_open_channel_stream(channel);
941 if (ret < 0) {
942 goto error;
943 }
944 /* Update the stream global counter */
945 ksess->stream_count_global += ret;
946
947 /*
948 * Have we already sent fds to the consumer? If yes, it
949 * means that tracing is started so it is safe to send
950 * our updated stream fds.
951 */
952 if (ksess->consumer_fds_sent != 1
953 || ksess->consumer == NULL) {
954 ret = -1;
955 goto error;
956 }
957
958 rcu_read_lock();
959 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
960 &iter.iter, socket, node.node) {
961 pthread_mutex_lock(socket->lock);
962 ret = kernel_consumer_send_channel_stream(socket,
963 channel, ksess,
964 session->output_traces ? 1 : 0);
965 pthread_mutex_unlock(socket->lock);
966 if (ret < 0) {
967 rcu_read_unlock();
968 goto error;
969 }
970 }
971 rcu_read_unlock();
972 }
973 session_unlock(session);
974 }
975 session_unlock_list();
976 return ret;
977
978 error:
979 session_unlock(session);
980 session_unlock_list();
981 return ret;
982 }
983
984 /*
985 * For each tracing session, update newly registered apps. The session list
986 * lock MUST be acquired before calling this.
987 */
988 static void update_ust_app(int app_sock)
989 {
990 struct ltt_session *sess, *stmp;
991
992 /* Consumer is in an ERROR state. Stop any application update. */
993 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
994 /* Stop the update process since the consumer is dead. */
995 return;
996 }
997
998 /* For all tracing session(s) */
999 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1000 struct ust_app *app;
1001
1002 session_lock(sess);
1003 if (!sess->ust_session) {
1004 goto unlock_session;
1005 }
1006
1007 rcu_read_lock();
1008 assert(app_sock >= 0);
1009 app = ust_app_find_by_sock(app_sock);
1010 if (app == NULL) {
1011 /*
1012 * Application can be unregistered before so
1013 * this is possible hence simply stopping the
1014 * update.
1015 */
1016 DBG3("UST app update failed to find app sock %d",
1017 app_sock);
1018 goto unlock_rcu;
1019 }
1020 ust_app_global_update(sess->ust_session, app);
1021 unlock_rcu:
1022 rcu_read_unlock();
1023 unlock_session:
1024 session_unlock(sess);
1025 }
1026 }
1027
1028 /*
1029 * This thread manage event coming from the kernel.
1030 *
1031 * Features supported in this thread:
1032 * -) CPU Hotplug
1033 */
1034 static void *thread_manage_kernel(void *data)
1035 {
1036 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1037 uint32_t revents, nb_fd;
1038 char tmp;
1039 struct lttng_poll_event events;
1040
1041 DBG("[thread] Thread manage kernel started");
1042
1043 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1044
1045 /*
1046 * This first step of the while is to clean this structure which could free
1047 * non NULL pointers so initialize it before the loop.
1048 */
1049 lttng_poll_init(&events);
1050
1051 if (testpoint(sessiond_thread_manage_kernel)) {
1052 goto error_testpoint;
1053 }
1054
1055 health_code_update();
1056
1057 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1058 goto error_testpoint;
1059 }
1060
1061 while (1) {
1062 health_code_update();
1063
1064 if (update_poll_flag == 1) {
1065 /* Clean events object. We are about to populate it again. */
1066 lttng_poll_clean(&events);
1067
1068 ret = sessiond_set_thread_pollset(&events, 2);
1069 if (ret < 0) {
1070 goto error_poll_create;
1071 }
1072
1073 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1074 if (ret < 0) {
1075 goto error;
1076 }
1077
1078 /* This will add the available kernel channel if any. */
1079 ret = update_kernel_poll(&events);
1080 if (ret < 0) {
1081 goto error;
1082 }
1083 update_poll_flag = 0;
1084 }
1085
1086 DBG("Thread kernel polling");
1087
1088 /* Poll infinite value of time */
1089 restart:
1090 health_poll_entry();
1091 ret = lttng_poll_wait(&events, -1);
1092 DBG("Thread kernel return from poll on %d fds",
1093 LTTNG_POLL_GETNB(&events));
1094 health_poll_exit();
1095 if (ret < 0) {
1096 /*
1097 * Restart interrupted system call.
1098 */
1099 if (errno == EINTR) {
1100 goto restart;
1101 }
1102 goto error;
1103 } else if (ret == 0) {
1104 /* Should not happen since timeout is infinite */
1105 ERR("Return value of poll is 0 with an infinite timeout.\n"
1106 "This should not have happened! Continuing...");
1107 continue;
1108 }
1109
1110 nb_fd = ret;
1111
1112 for (i = 0; i < nb_fd; i++) {
1113 /* Fetch once the poll data */
1114 revents = LTTNG_POLL_GETEV(&events, i);
1115 pollfd = LTTNG_POLL_GETFD(&events, i);
1116
1117 health_code_update();
1118
1119 if (!revents) {
1120 /* No activity for this FD (poll implementation). */
1121 continue;
1122 }
1123
1124 /* Thread quit pipe has been closed. Killing thread. */
1125 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1126 if (ret) {
1127 err = 0;
1128 goto exit;
1129 }
1130
1131 /* Check for data on kernel pipe */
1132 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1133 (void) lttng_read(kernel_poll_pipe[0],
1134 &tmp, 1);
1135 /*
1136 * Ret value is useless here, if this pipe gets any actions an
1137 * update is required anyway.
1138 */
1139 update_poll_flag = 1;
1140 continue;
1141 } else {
1142 /*
1143 * New CPU detected by the kernel. Adding kernel stream to
1144 * kernel session and updating the kernel consumer
1145 */
1146 if (revents & LPOLLIN) {
1147 ret = update_kernel_stream(&kconsumer_data, pollfd);
1148 if (ret < 0) {
1149 continue;
1150 }
1151 break;
1152 /*
1153 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1154 * and unregister kernel stream at this point.
1155 */
1156 }
1157 }
1158 }
1159 }
1160
1161 exit:
1162 error:
1163 lttng_poll_clean(&events);
1164 error_poll_create:
1165 error_testpoint:
1166 utils_close_pipe(kernel_poll_pipe);
1167 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1168 if (err) {
1169 health_error();
1170 ERR("Health error occurred in %s", __func__);
1171 WARN("Kernel thread died unexpectedly. "
1172 "Kernel tracing can continue but CPU hotplug is disabled.");
1173 }
1174 health_unregister(health_sessiond);
1175 DBG("Kernel thread dying");
1176 return NULL;
1177 }
1178
1179 /*
1180 * Signal pthread condition of the consumer data that the thread.
1181 */
1182 static void signal_consumer_condition(struct consumer_data *data, int state)
1183 {
1184 pthread_mutex_lock(&data->cond_mutex);
1185
1186 /*
1187 * The state is set before signaling. It can be any value, it's the waiter
1188 * job to correctly interpret this condition variable associated to the
1189 * consumer pthread_cond.
1190 *
1191 * A value of 0 means that the corresponding thread of the consumer data
1192 * was not started. 1 indicates that the thread has started and is ready
1193 * for action. A negative value means that there was an error during the
1194 * thread bootstrap.
1195 */
1196 data->consumer_thread_is_ready = state;
1197 (void) pthread_cond_signal(&data->cond);
1198
1199 pthread_mutex_unlock(&data->cond_mutex);
1200 }
1201
1202 /*
1203 * This thread manage the consumer error sent back to the session daemon.
1204 */
1205 static void *thread_manage_consumer(void *data)
1206 {
1207 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1208 uint32_t revents, nb_fd;
1209 enum lttcomm_return_code code;
1210 struct lttng_poll_event events;
1211 struct consumer_data *consumer_data = data;
1212
1213 DBG("[thread] Manage consumer started");
1214
1215 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1216
1217 health_code_update();
1218
1219 /*
1220 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1221 * metadata_sock. Nothing more will be added to this poll set.
1222 */
1223 ret = sessiond_set_thread_pollset(&events, 3);
1224 if (ret < 0) {
1225 goto error_poll;
1226 }
1227
1228 /*
1229 * The error socket here is already in a listening state which was done
1230 * just before spawning this thread to avoid a race between the consumer
1231 * daemon exec trying to connect and the listen() call.
1232 */
1233 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1234 if (ret < 0) {
1235 goto error;
1236 }
1237
1238 health_code_update();
1239
1240 /* Infinite blocking call, waiting for transmission */
1241 restart:
1242 health_poll_entry();
1243
1244 if (testpoint(sessiond_thread_manage_consumer)) {
1245 goto error;
1246 }
1247
1248 ret = lttng_poll_wait(&events, -1);
1249 health_poll_exit();
1250 if (ret < 0) {
1251 /*
1252 * Restart interrupted system call.
1253 */
1254 if (errno == EINTR) {
1255 goto restart;
1256 }
1257 goto error;
1258 }
1259
1260 nb_fd = ret;
1261
1262 for (i = 0; i < nb_fd; i++) {
1263 /* Fetch once the poll data */
1264 revents = LTTNG_POLL_GETEV(&events, i);
1265 pollfd = LTTNG_POLL_GETFD(&events, i);
1266
1267 health_code_update();
1268
1269 if (!revents) {
1270 /* No activity for this FD (poll implementation). */
1271 continue;
1272 }
1273
1274 /* Thread quit pipe has been closed. Killing thread. */
1275 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1276 if (ret) {
1277 err = 0;
1278 goto exit;
1279 }
1280
1281 /* Event on the registration socket */
1282 if (pollfd == consumer_data->err_sock) {
1283 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1284 ERR("consumer err socket poll error");
1285 goto error;
1286 }
1287 }
1288 }
1289
1290 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1291 if (sock < 0) {
1292 goto error;
1293 }
1294
1295 /*
1296 * Set the CLOEXEC flag. Return code is useless because either way, the
1297 * show must go on.
1298 */
1299 (void) utils_set_fd_cloexec(sock);
1300
1301 health_code_update();
1302
1303 DBG2("Receiving code from consumer err_sock");
1304
1305 /* Getting status code from kconsumerd */
1306 ret = lttcomm_recv_unix_sock(sock, &code,
1307 sizeof(enum lttcomm_return_code));
1308 if (ret <= 0) {
1309 goto error;
1310 }
1311
1312 health_code_update();
1313 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1314 /* Connect both socket, command and metadata. */
1315 consumer_data->cmd_sock =
1316 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1317 consumer_data->metadata_fd =
1318 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1319 if (consumer_data->cmd_sock < 0
1320 || consumer_data->metadata_fd < 0) {
1321 PERROR("consumer connect cmd socket");
1322 /* On error, signal condition and quit. */
1323 signal_consumer_condition(consumer_data, -1);
1324 goto error;
1325 }
1326 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1327 /* Create metadata socket lock. */
1328 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1329 if (consumer_data->metadata_sock.lock == NULL) {
1330 PERROR("zmalloc pthread mutex");
1331 ret = -1;
1332 goto error;
1333 }
1334 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1335
1336 signal_consumer_condition(consumer_data, 1);
1337 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1338 DBG("Consumer metadata socket ready (fd: %d)",
1339 consumer_data->metadata_fd);
1340 } else {
1341 ERR("consumer error when waiting for SOCK_READY : %s",
1342 lttcomm_get_readable_code(-code));
1343 goto error;
1344 }
1345
1346 /* Remove the consumerd error sock since we've established a connexion */
1347 ret = lttng_poll_del(&events, consumer_data->err_sock);
1348 if (ret < 0) {
1349 goto error;
1350 }
1351
1352 /* Add new accepted error socket. */
1353 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1354 if (ret < 0) {
1355 goto error;
1356 }
1357
1358 /* Add metadata socket that is successfully connected. */
1359 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1360 LPOLLIN | LPOLLRDHUP);
1361 if (ret < 0) {
1362 goto error;
1363 }
1364
1365 health_code_update();
1366
1367 /* Infinite blocking call, waiting for transmission */
1368 restart_poll:
1369 while (1) {
1370 health_code_update();
1371
1372 /* Exit the thread because the thread quit pipe has been triggered. */
1373 if (should_quit) {
1374 /* Not a health error. */
1375 err = 0;
1376 goto exit;
1377 }
1378
1379 health_poll_entry();
1380 ret = lttng_poll_wait(&events, -1);
1381 health_poll_exit();
1382 if (ret < 0) {
1383 /*
1384 * Restart interrupted system call.
1385 */
1386 if (errno == EINTR) {
1387 goto restart_poll;
1388 }
1389 goto error;
1390 }
1391
1392 nb_fd = ret;
1393
1394 for (i = 0; i < nb_fd; i++) {
1395 /* Fetch once the poll data */
1396 revents = LTTNG_POLL_GETEV(&events, i);
1397 pollfd = LTTNG_POLL_GETFD(&events, i);
1398
1399 health_code_update();
1400
1401 if (!revents) {
1402 /* No activity for this FD (poll implementation). */
1403 continue;
1404 }
1405
1406 /*
1407 * Thread quit pipe has been triggered, flag that we should stop
1408 * but continue the current loop to handle potential data from
1409 * consumer.
1410 */
1411 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1412
1413 if (pollfd == sock) {
1414 /* Event on the consumerd socket */
1415 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1416 ERR("consumer err socket second poll error");
1417 goto error;
1418 }
1419 health_code_update();
1420 /* Wait for any kconsumerd error */
1421 ret = lttcomm_recv_unix_sock(sock, &code,
1422 sizeof(enum lttcomm_return_code));
1423 if (ret <= 0) {
1424 ERR("consumer closed the command socket");
1425 goto error;
1426 }
1427
1428 ERR("consumer return code : %s",
1429 lttcomm_get_readable_code(-code));
1430
1431 goto exit;
1432 } else if (pollfd == consumer_data->metadata_fd) {
1433 /* UST metadata requests */
1434 ret = ust_consumer_metadata_request(
1435 &consumer_data->metadata_sock);
1436 if (ret < 0) {
1437 ERR("Handling metadata request");
1438 goto error;
1439 }
1440 }
1441 /* No need for an else branch all FDs are tested prior. */
1442 }
1443 health_code_update();
1444 }
1445
1446 exit:
1447 error:
1448 /*
1449 * We lock here because we are about to close the sockets and some other
1450 * thread might be using them so get exclusive access which will abort all
1451 * other consumer command by other threads.
1452 */
1453 pthread_mutex_lock(&consumer_data->lock);
1454
1455 /* Immediately set the consumerd state to stopped */
1456 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1457 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1458 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1459 consumer_data->type == LTTNG_CONSUMER32_UST) {
1460 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1461 } else {
1462 /* Code flow error... */
1463 assert(0);
1464 }
1465
1466 if (consumer_data->err_sock >= 0) {
1467 ret = close(consumer_data->err_sock);
1468 if (ret) {
1469 PERROR("close");
1470 }
1471 consumer_data->err_sock = -1;
1472 }
1473 if (consumer_data->cmd_sock >= 0) {
1474 ret = close(consumer_data->cmd_sock);
1475 if (ret) {
1476 PERROR("close");
1477 }
1478 consumer_data->cmd_sock = -1;
1479 }
1480 if (consumer_data->metadata_sock.fd_ptr &&
1481 *consumer_data->metadata_sock.fd_ptr >= 0) {
1482 ret = close(*consumer_data->metadata_sock.fd_ptr);
1483 if (ret) {
1484 PERROR("close");
1485 }
1486 }
1487 if (sock >= 0) {
1488 ret = close(sock);
1489 if (ret) {
1490 PERROR("close");
1491 }
1492 }
1493
1494 unlink(consumer_data->err_unix_sock_path);
1495 unlink(consumer_data->cmd_unix_sock_path);
1496 consumer_data->pid = 0;
1497 pthread_mutex_unlock(&consumer_data->lock);
1498
1499 /* Cleanup metadata socket mutex. */
1500 if (consumer_data->metadata_sock.lock) {
1501 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1502 free(consumer_data->metadata_sock.lock);
1503 }
1504 lttng_poll_clean(&events);
1505 error_poll:
1506 if (err) {
1507 health_error();
1508 ERR("Health error occurred in %s", __func__);
1509 }
1510 health_unregister(health_sessiond);
1511 DBG("consumer thread cleanup completed");
1512
1513 return NULL;
1514 }
1515
1516 /*
1517 * This thread manage application communication.
1518 */
1519 static void *thread_manage_apps(void *data)
1520 {
1521 int i, ret, pollfd, err = -1;
1522 ssize_t size_ret;
1523 uint32_t revents, nb_fd;
1524 struct lttng_poll_event events;
1525
1526 DBG("[thread] Manage application started");
1527
1528 rcu_register_thread();
1529 rcu_thread_online();
1530
1531 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1532
1533 if (testpoint(sessiond_thread_manage_apps)) {
1534 goto error_testpoint;
1535 }
1536
1537 health_code_update();
1538
1539 ret = sessiond_set_thread_pollset(&events, 2);
1540 if (ret < 0) {
1541 goto error_poll_create;
1542 }
1543
1544 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1545 if (ret < 0) {
1546 goto error;
1547 }
1548
1549 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1550 goto error;
1551 }
1552
1553 health_code_update();
1554
1555 while (1) {
1556 DBG("Apps thread polling");
1557
1558 /* Inifinite blocking call, waiting for transmission */
1559 restart:
1560 health_poll_entry();
1561 ret = lttng_poll_wait(&events, -1);
1562 DBG("Apps thread return from poll on %d fds",
1563 LTTNG_POLL_GETNB(&events));
1564 health_poll_exit();
1565 if (ret < 0) {
1566 /*
1567 * Restart interrupted system call.
1568 */
1569 if (errno == EINTR) {
1570 goto restart;
1571 }
1572 goto error;
1573 }
1574
1575 nb_fd = ret;
1576
1577 for (i = 0; i < nb_fd; i++) {
1578 /* Fetch once the poll data */
1579 revents = LTTNG_POLL_GETEV(&events, i);
1580 pollfd = LTTNG_POLL_GETFD(&events, i);
1581
1582 health_code_update();
1583
1584 if (!revents) {
1585 /* No activity for this FD (poll implementation). */
1586 continue;
1587 }
1588
1589 /* Thread quit pipe has been closed. Killing thread. */
1590 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1591 if (ret) {
1592 err = 0;
1593 goto exit;
1594 }
1595
1596 /* Inspect the apps cmd pipe */
1597 if (pollfd == apps_cmd_pipe[0]) {
1598 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1599 ERR("Apps command pipe error");
1600 goto error;
1601 } else if (revents & LPOLLIN) {
1602 int sock;
1603
1604 /* Empty pipe */
1605 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1606 if (size_ret < sizeof(sock)) {
1607 PERROR("read apps cmd pipe");
1608 goto error;
1609 }
1610
1611 health_code_update();
1612
1613 /*
1614 * We only monitor the error events of the socket. This
1615 * thread does not handle any incoming data from UST
1616 * (POLLIN).
1617 */
1618 ret = lttng_poll_add(&events, sock,
1619 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1620 if (ret < 0) {
1621 goto error;
1622 }
1623
1624 DBG("Apps with sock %d added to poll set", sock);
1625 }
1626 } else {
1627 /*
1628 * At this point, we know that a registered application made
1629 * the event at poll_wait.
1630 */
1631 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1632 /* Removing from the poll set */
1633 ret = lttng_poll_del(&events, pollfd);
1634 if (ret < 0) {
1635 goto error;
1636 }
1637
1638 /* Socket closed on remote end. */
1639 ust_app_unregister(pollfd);
1640 }
1641 }
1642
1643 health_code_update();
1644 }
1645 }
1646
1647 exit:
1648 error:
1649 lttng_poll_clean(&events);
1650 error_poll_create:
1651 error_testpoint:
1652 utils_close_pipe(apps_cmd_pipe);
1653 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1654
1655 /*
1656 * We don't clean the UST app hash table here since already registered
1657 * applications can still be controlled so let them be until the session
1658 * daemon dies or the applications stop.
1659 */
1660
1661 if (err) {
1662 health_error();
1663 ERR("Health error occurred in %s", __func__);
1664 }
1665 health_unregister(health_sessiond);
1666 DBG("Application communication apps thread cleanup complete");
1667 rcu_thread_offline();
1668 rcu_unregister_thread();
1669 return NULL;
1670 }
1671
1672 /*
1673 * Send a socket to a thread This is called from the dispatch UST registration
1674 * thread once all sockets are set for the application.
1675 *
1676 * The sock value can be invalid, we don't really care, the thread will handle
1677 * it and make the necessary cleanup if so.
1678 *
1679 * On success, return 0 else a negative value being the errno message of the
1680 * write().
1681 */
1682 static int send_socket_to_thread(int fd, int sock)
1683 {
1684 ssize_t ret;
1685
1686 /*
1687 * It's possible that the FD is set as invalid with -1 concurrently just
1688 * before calling this function being a shutdown state of the thread.
1689 */
1690 if (fd < 0) {
1691 ret = -EBADF;
1692 goto error;
1693 }
1694
1695 ret = lttng_write(fd, &sock, sizeof(sock));
1696 if (ret < sizeof(sock)) {
1697 PERROR("write apps pipe %d", fd);
1698 if (ret < 0) {
1699 ret = -errno;
1700 }
1701 goto error;
1702 }
1703
1704 /* All good. Don't send back the write positive ret value. */
1705 ret = 0;
1706 error:
1707 return (int) ret;
1708 }
1709
1710 /*
1711 * Sanitize the wait queue of the dispatch registration thread meaning removing
1712 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1713 * notify socket is never received.
1714 */
1715 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1716 {
1717 int ret, nb_fd = 0, i;
1718 unsigned int fd_added = 0;
1719 struct lttng_poll_event events;
1720 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1721
1722 assert(wait_queue);
1723
1724 lttng_poll_init(&events);
1725
1726 /* Just skip everything for an empty queue. */
1727 if (!wait_queue->count) {
1728 goto end;
1729 }
1730
1731 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1732 if (ret < 0) {
1733 goto error_create;
1734 }
1735
1736 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1737 &wait_queue->head, head) {
1738 assert(wait_node->app);
1739 ret = lttng_poll_add(&events, wait_node->app->sock,
1740 LPOLLHUP | LPOLLERR);
1741 if (ret < 0) {
1742 goto error;
1743 }
1744
1745 fd_added = 1;
1746 }
1747
1748 if (!fd_added) {
1749 goto end;
1750 }
1751
1752 /*
1753 * Poll but don't block so we can quickly identify the faulty events and
1754 * clean them afterwards from the wait queue.
1755 */
1756 ret = lttng_poll_wait(&events, 0);
1757 if (ret < 0) {
1758 goto error;
1759 }
1760 nb_fd = ret;
1761
1762 for (i = 0; i < nb_fd; i++) {
1763 /* Get faulty FD. */
1764 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1765 int pollfd = LTTNG_POLL_GETFD(&events, i);
1766
1767 if (!revents) {
1768 /* No activity for this FD (poll implementation). */
1769 continue;
1770 }
1771
1772 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1773 &wait_queue->head, head) {
1774 if (pollfd == wait_node->app->sock &&
1775 (revents & (LPOLLHUP | LPOLLERR))) {
1776 cds_list_del(&wait_node->head);
1777 wait_queue->count--;
1778 ust_app_destroy(wait_node->app);
1779 free(wait_node);
1780 break;
1781 }
1782 }
1783 }
1784
1785 if (nb_fd > 0) {
1786 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1787 }
1788
1789 end:
1790 lttng_poll_clean(&events);
1791 return;
1792
1793 error:
1794 lttng_poll_clean(&events);
1795 error_create:
1796 ERR("Unable to sanitize wait queue");
1797 return;
1798 }
1799
1800 /*
1801 * Dispatch request from the registration threads to the application
1802 * communication thread.
1803 */
1804 static void *thread_dispatch_ust_registration(void *data)
1805 {
1806 int ret, err = -1;
1807 struct cds_wfcq_node *node;
1808 struct ust_command *ust_cmd = NULL;
1809 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1810 struct ust_reg_wait_queue wait_queue = {
1811 .count = 0,
1812 };
1813
1814 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1815
1816 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1817 goto error_testpoint;
1818 }
1819
1820 health_code_update();
1821
1822 CDS_INIT_LIST_HEAD(&wait_queue.head);
1823
1824 DBG("[thread] Dispatch UST command started");
1825
1826 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1827 health_code_update();
1828
1829 /* Atomically prepare the queue futex */
1830 futex_nto1_prepare(&ust_cmd_queue.futex);
1831
1832 do {
1833 struct ust_app *app = NULL;
1834 ust_cmd = NULL;
1835
1836 /*
1837 * Make sure we don't have node(s) that have hung up before receiving
1838 * the notify socket. This is to clean the list in order to avoid
1839 * memory leaks from notify socket that are never seen.
1840 */
1841 sanitize_wait_queue(&wait_queue);
1842
1843 health_code_update();
1844 /* Dequeue command for registration */
1845 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1846 if (node == NULL) {
1847 DBG("Woken up but nothing in the UST command queue");
1848 /* Continue thread execution */
1849 break;
1850 }
1851
1852 ust_cmd = caa_container_of(node, struct ust_command, node);
1853
1854 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1855 " gid:%d sock:%d name:%s (version %d.%d)",
1856 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1857 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1858 ust_cmd->sock, ust_cmd->reg_msg.name,
1859 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1860
1861 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1862 wait_node = zmalloc(sizeof(*wait_node));
1863 if (!wait_node) {
1864 PERROR("zmalloc wait_node dispatch");
1865 ret = close(ust_cmd->sock);
1866 if (ret < 0) {
1867 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1868 }
1869 lttng_fd_put(LTTNG_FD_APPS, 1);
1870 free(ust_cmd);
1871 goto error;
1872 }
1873 CDS_INIT_LIST_HEAD(&wait_node->head);
1874
1875 /* Create application object if socket is CMD. */
1876 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1877 ust_cmd->sock);
1878 if (!wait_node->app) {
1879 ret = close(ust_cmd->sock);
1880 if (ret < 0) {
1881 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1882 }
1883 lttng_fd_put(LTTNG_FD_APPS, 1);
1884 free(wait_node);
1885 free(ust_cmd);
1886 continue;
1887 }
1888 /*
1889 * Add application to the wait queue so we can set the notify
1890 * socket before putting this object in the global ht.
1891 */
1892 cds_list_add(&wait_node->head, &wait_queue.head);
1893 wait_queue.count++;
1894
1895 free(ust_cmd);
1896 /*
1897 * We have to continue here since we don't have the notify
1898 * socket and the application MUST be added to the hash table
1899 * only at that moment.
1900 */
1901 continue;
1902 } else {
1903 /*
1904 * Look for the application in the local wait queue and set the
1905 * notify socket if found.
1906 */
1907 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1908 &wait_queue.head, head) {
1909 health_code_update();
1910 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1911 wait_node->app->notify_sock = ust_cmd->sock;
1912 cds_list_del(&wait_node->head);
1913 wait_queue.count--;
1914 app = wait_node->app;
1915 free(wait_node);
1916 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1917 break;
1918 }
1919 }
1920
1921 /*
1922 * With no application at this stage the received socket is
1923 * basically useless so close it before we free the cmd data
1924 * structure for good.
1925 */
1926 if (!app) {
1927 ret = close(ust_cmd->sock);
1928 if (ret < 0) {
1929 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1930 }
1931 lttng_fd_put(LTTNG_FD_APPS, 1);
1932 }
1933 free(ust_cmd);
1934 }
1935
1936 if (app) {
1937 /*
1938 * @session_lock_list
1939 *
1940 * Lock the global session list so from the register up to the
1941 * registration done message, no thread can see the application
1942 * and change its state.
1943 */
1944 session_lock_list();
1945 rcu_read_lock();
1946
1947 /*
1948 * Add application to the global hash table. This needs to be
1949 * done before the update to the UST registry can locate the
1950 * application.
1951 */
1952 ust_app_add(app);
1953
1954 /* Set app version. This call will print an error if needed. */
1955 (void) ust_app_version(app);
1956
1957 /* Send notify socket through the notify pipe. */
1958 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1959 app->notify_sock);
1960 if (ret < 0) {
1961 rcu_read_unlock();
1962 session_unlock_list();
1963 /*
1964 * No notify thread, stop the UST tracing. However, this is
1965 * not an internal error of the this thread thus setting
1966 * the health error code to a normal exit.
1967 */
1968 err = 0;
1969 goto error;
1970 }
1971
1972 /*
1973 * Update newly registered application with the tracing
1974 * registry info already enabled information.
1975 */
1976 update_ust_app(app->sock);
1977
1978 /*
1979 * Don't care about return value. Let the manage apps threads
1980 * handle app unregistration upon socket close.
1981 */
1982 (void) ust_app_register_done(app->sock);
1983
1984 /*
1985 * Even if the application socket has been closed, send the app
1986 * to the thread and unregistration will take place at that
1987 * place.
1988 */
1989 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1990 if (ret < 0) {
1991 rcu_read_unlock();
1992 session_unlock_list();
1993 /*
1994 * No apps. thread, stop the UST tracing. However, this is
1995 * not an internal error of the this thread thus setting
1996 * the health error code to a normal exit.
1997 */
1998 err = 0;
1999 goto error;
2000 }
2001
2002 rcu_read_unlock();
2003 session_unlock_list();
2004 }
2005 } while (node != NULL);
2006
2007 health_poll_entry();
2008 /* Futex wait on queue. Blocking call on futex() */
2009 futex_nto1_wait(&ust_cmd_queue.futex);
2010 health_poll_exit();
2011 }
2012 /* Normal exit, no error */
2013 err = 0;
2014
2015 error:
2016 /* Clean up wait queue. */
2017 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2018 &wait_queue.head, head) {
2019 cds_list_del(&wait_node->head);
2020 wait_queue.count--;
2021 free(wait_node);
2022 }
2023
2024 error_testpoint:
2025 DBG("Dispatch thread dying");
2026 if (err) {
2027 health_error();
2028 ERR("Health error occurred in %s", __func__);
2029 }
2030 health_unregister(health_sessiond);
2031 return NULL;
2032 }
2033
2034 /*
2035 * This thread manage application registration.
2036 */
2037 static void *thread_registration_apps(void *data)
2038 {
2039 int sock = -1, i, ret, pollfd, err = -1;
2040 uint32_t revents, nb_fd;
2041 struct lttng_poll_event events;
2042 /*
2043 * Get allocated in this thread, enqueued to a global queue, dequeued and
2044 * freed in the manage apps thread.
2045 */
2046 struct ust_command *ust_cmd = NULL;
2047
2048 DBG("[thread] Manage application registration started");
2049
2050 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2051
2052 if (testpoint(sessiond_thread_registration_apps)) {
2053 goto error_testpoint;
2054 }
2055
2056 ret = lttcomm_listen_unix_sock(apps_sock);
2057 if (ret < 0) {
2058 goto error_listen;
2059 }
2060
2061 /*
2062 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2063 * more will be added to this poll set.
2064 */
2065 ret = sessiond_set_thread_pollset(&events, 2);
2066 if (ret < 0) {
2067 goto error_create_poll;
2068 }
2069
2070 /* Add the application registration socket */
2071 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2072 if (ret < 0) {
2073 goto error_poll_add;
2074 }
2075
2076 /* Notify all applications to register */
2077 ret = notify_ust_apps(1);
2078 if (ret < 0) {
2079 ERR("Failed to notify applications or create the wait shared memory.\n"
2080 "Execution continues but there might be problem for already\n"
2081 "running applications that wishes to register.");
2082 }
2083
2084 while (1) {
2085 DBG("Accepting application registration");
2086
2087 /* Inifinite blocking call, waiting for transmission */
2088 restart:
2089 health_poll_entry();
2090 ret = lttng_poll_wait(&events, -1);
2091 health_poll_exit();
2092 if (ret < 0) {
2093 /*
2094 * Restart interrupted system call.
2095 */
2096 if (errno == EINTR) {
2097 goto restart;
2098 }
2099 goto error;
2100 }
2101
2102 nb_fd = ret;
2103
2104 for (i = 0; i < nb_fd; i++) {
2105 health_code_update();
2106
2107 /* Fetch once the poll data */
2108 revents = LTTNG_POLL_GETEV(&events, i);
2109 pollfd = LTTNG_POLL_GETFD(&events, i);
2110
2111 if (!revents) {
2112 /* No activity for this FD (poll implementation). */
2113 continue;
2114 }
2115
2116 /* Thread quit pipe has been closed. Killing thread. */
2117 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2118 if (ret) {
2119 err = 0;
2120 goto exit;
2121 }
2122
2123 /* Event on the registration socket */
2124 if (pollfd == apps_sock) {
2125 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2126 ERR("Register apps socket poll error");
2127 goto error;
2128 } else if (revents & LPOLLIN) {
2129 sock = lttcomm_accept_unix_sock(apps_sock);
2130 if (sock < 0) {
2131 goto error;
2132 }
2133
2134 /*
2135 * Set socket timeout for both receiving and ending.
2136 * app_socket_timeout is in seconds, whereas
2137 * lttcomm_setsockopt_rcv_timeout and
2138 * lttcomm_setsockopt_snd_timeout expect msec as
2139 * parameter.
2140 */
2141 (void) lttcomm_setsockopt_rcv_timeout(sock,
2142 app_socket_timeout * 1000);
2143 (void) lttcomm_setsockopt_snd_timeout(sock,
2144 app_socket_timeout * 1000);
2145
2146 /*
2147 * Set the CLOEXEC flag. Return code is useless because
2148 * either way, the show must go on.
2149 */
2150 (void) utils_set_fd_cloexec(sock);
2151
2152 /* Create UST registration command for enqueuing */
2153 ust_cmd = zmalloc(sizeof(struct ust_command));
2154 if (ust_cmd == NULL) {
2155 PERROR("ust command zmalloc");
2156 goto error;
2157 }
2158
2159 /*
2160 * Using message-based transmissions to ensure we don't
2161 * have to deal with partially received messages.
2162 */
2163 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2164 if (ret < 0) {
2165 ERR("Exhausted file descriptors allowed for applications.");
2166 free(ust_cmd);
2167 ret = close(sock);
2168 if (ret) {
2169 PERROR("close");
2170 }
2171 sock = -1;
2172 continue;
2173 }
2174
2175 health_code_update();
2176 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2177 if (ret < 0) {
2178 free(ust_cmd);
2179 /* Close socket of the application. */
2180 ret = close(sock);
2181 if (ret) {
2182 PERROR("close");
2183 }
2184 lttng_fd_put(LTTNG_FD_APPS, 1);
2185 sock = -1;
2186 continue;
2187 }
2188 health_code_update();
2189
2190 ust_cmd->sock = sock;
2191 sock = -1;
2192
2193 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2194 " gid:%d sock:%d name:%s (version %d.%d)",
2195 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2196 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2197 ust_cmd->sock, ust_cmd->reg_msg.name,
2198 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2199
2200 /*
2201 * Lock free enqueue the registration request. The red pill
2202 * has been taken! This apps will be part of the *system*.
2203 */
2204 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2205
2206 /*
2207 * Wake the registration queue futex. Implicit memory
2208 * barrier with the exchange in cds_wfcq_enqueue.
2209 */
2210 futex_nto1_wake(&ust_cmd_queue.futex);
2211 }
2212 }
2213 }
2214 }
2215
2216 exit:
2217 error:
2218 /* Notify that the registration thread is gone */
2219 notify_ust_apps(0);
2220
2221 if (apps_sock >= 0) {
2222 ret = close(apps_sock);
2223 if (ret) {
2224 PERROR("close");
2225 }
2226 }
2227 if (sock >= 0) {
2228 ret = close(sock);
2229 if (ret) {
2230 PERROR("close");
2231 }
2232 lttng_fd_put(LTTNG_FD_APPS, 1);
2233 }
2234 unlink(apps_unix_sock_path);
2235
2236 error_poll_add:
2237 lttng_poll_clean(&events);
2238 error_listen:
2239 error_create_poll:
2240 error_testpoint:
2241 DBG("UST Registration thread cleanup complete");
2242 if (err) {
2243 health_error();
2244 ERR("Health error occurred in %s", __func__);
2245 }
2246 health_unregister(health_sessiond);
2247
2248 return NULL;
2249 }
2250
2251 /*
2252 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2253 * exec or it will fails.
2254 */
2255 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2256 {
2257 int ret, clock_ret;
2258 struct timespec timeout;
2259
2260 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2261 consumer_data->consumer_thread_is_ready = 0;
2262
2263 /* Setup pthread condition */
2264 ret = pthread_condattr_init(&consumer_data->condattr);
2265 if (ret) {
2266 errno = ret;
2267 PERROR("pthread_condattr_init consumer data");
2268 goto error;
2269 }
2270
2271 /*
2272 * Set the monotonic clock in order to make sure we DO NOT jump in time
2273 * between the clock_gettime() call and the timedwait call. See bug #324
2274 * for a more details and how we noticed it.
2275 */
2276 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2277 if (ret) {
2278 errno = ret;
2279 PERROR("pthread_condattr_setclock consumer data");
2280 goto error;
2281 }
2282
2283 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2284 if (ret) {
2285 errno = ret;
2286 PERROR("pthread_cond_init consumer data");
2287 goto error;
2288 }
2289
2290 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2291 consumer_data);
2292 if (ret) {
2293 errno = ret;
2294 PERROR("pthread_create consumer");
2295 ret = -1;
2296 goto error;
2297 }
2298
2299 /* We are about to wait on a pthread condition */
2300 pthread_mutex_lock(&consumer_data->cond_mutex);
2301
2302 /* Get time for sem_timedwait absolute timeout */
2303 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2304 /*
2305 * Set the timeout for the condition timed wait even if the clock gettime
2306 * call fails since we might loop on that call and we want to avoid to
2307 * increment the timeout too many times.
2308 */
2309 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2310
2311 /*
2312 * The following loop COULD be skipped in some conditions so this is why we
2313 * set ret to 0 in order to make sure at least one round of the loop is
2314 * done.
2315 */
2316 ret = 0;
2317
2318 /*
2319 * Loop until the condition is reached or when a timeout is reached. Note
2320 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2321 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2322 * possible. This loop does not take any chances and works with both of
2323 * them.
2324 */
2325 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2326 if (clock_ret < 0) {
2327 PERROR("clock_gettime spawn consumer");
2328 /* Infinite wait for the consumerd thread to be ready */
2329 ret = pthread_cond_wait(&consumer_data->cond,
2330 &consumer_data->cond_mutex);
2331 } else {
2332 ret = pthread_cond_timedwait(&consumer_data->cond,
2333 &consumer_data->cond_mutex, &timeout);
2334 }
2335 }
2336
2337 /* Release the pthread condition */
2338 pthread_mutex_unlock(&consumer_data->cond_mutex);
2339
2340 if (ret != 0) {
2341 errno = ret;
2342 if (ret == ETIMEDOUT) {
2343 int pth_ret;
2344
2345 /*
2346 * Call has timed out so we kill the kconsumerd_thread and return
2347 * an error.
2348 */
2349 ERR("Condition timed out. The consumer thread was never ready."
2350 " Killing it");
2351 pth_ret = pthread_cancel(consumer_data->thread);
2352 if (pth_ret < 0) {
2353 PERROR("pthread_cancel consumer thread");
2354 }
2355 } else {
2356 PERROR("pthread_cond_wait failed consumer thread");
2357 }
2358 /* Caller is expecting a negative value on failure. */
2359 ret = -1;
2360 goto error;
2361 }
2362
2363 pthread_mutex_lock(&consumer_data->pid_mutex);
2364 if (consumer_data->pid == 0) {
2365 ERR("Consumerd did not start");
2366 pthread_mutex_unlock(&consumer_data->pid_mutex);
2367 goto error;
2368 }
2369 pthread_mutex_unlock(&consumer_data->pid_mutex);
2370
2371 return 0;
2372
2373 error:
2374 return ret;
2375 }
2376
2377 /*
2378 * Join consumer thread
2379 */
2380 static int join_consumer_thread(struct consumer_data *consumer_data)
2381 {
2382 void *status;
2383
2384 /* Consumer pid must be a real one. */
2385 if (consumer_data->pid > 0) {
2386 int ret;
2387 ret = kill(consumer_data->pid, SIGTERM);
2388 if (ret) {
2389 PERROR("Error killing consumer daemon");
2390 return ret;
2391 }
2392 return pthread_join(consumer_data->thread, &status);
2393 } else {
2394 return 0;
2395 }
2396 }
2397
2398 /*
2399 * Fork and exec a consumer daemon (consumerd).
2400 *
2401 * Return pid if successful else -1.
2402 */
2403 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2404 {
2405 int ret;
2406 pid_t pid;
2407 const char *consumer_to_use;
2408 const char *verbosity;
2409 struct stat st;
2410
2411 DBG("Spawning consumerd");
2412
2413 pid = fork();
2414 if (pid == 0) {
2415 /*
2416 * Exec consumerd.
2417 */
2418 if (opt_verbose_consumer) {
2419 verbosity = "--verbose";
2420 } else if (lttng_opt_quiet) {
2421 verbosity = "--quiet";
2422 } else {
2423 verbosity = "";
2424 }
2425
2426 switch (consumer_data->type) {
2427 case LTTNG_CONSUMER_KERNEL:
2428 /*
2429 * Find out which consumerd to execute. We will first try the
2430 * 64-bit path, then the sessiond's installation directory, and
2431 * fallback on the 32-bit one,
2432 */
2433 DBG3("Looking for a kernel consumer at these locations:");
2434 DBG3(" 1) %s", consumerd64_bin);
2435 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2436 DBG3(" 3) %s", consumerd32_bin);
2437 if (stat(consumerd64_bin, &st) == 0) {
2438 DBG3("Found location #1");
2439 consumer_to_use = consumerd64_bin;
2440 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2441 DBG3("Found location #2");
2442 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2443 } else if (stat(consumerd32_bin, &st) == 0) {
2444 DBG3("Found location #3");
2445 consumer_to_use = consumerd32_bin;
2446 } else {
2447 DBG("Could not find any valid consumerd executable");
2448 ret = -EINVAL;
2449 break;
2450 }
2451 DBG("Using kernel consumer at: %s", consumer_to_use);
2452 ret = execl(consumer_to_use,
2453 "lttng-consumerd", verbosity, "-k",
2454 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2455 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2456 "--group", tracing_group_name,
2457 NULL);
2458 break;
2459 case LTTNG_CONSUMER64_UST:
2460 {
2461 char *tmpnew = NULL;
2462
2463 if (consumerd64_libdir[0] != '\0') {
2464 char *tmp;
2465 size_t tmplen;
2466
2467 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2468 if (!tmp) {
2469 tmp = "";
2470 }
2471 tmplen = strlen("LD_LIBRARY_PATH=")
2472 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2473 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2474 if (!tmpnew) {
2475 ret = -ENOMEM;
2476 goto error;
2477 }
2478 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2479 strcat(tmpnew, consumerd64_libdir);
2480 if (tmp[0] != '\0') {
2481 strcat(tmpnew, ":");
2482 strcat(tmpnew, tmp);
2483 }
2484 ret = putenv(tmpnew);
2485 if (ret) {
2486 ret = -errno;
2487 free(tmpnew);
2488 goto error;
2489 }
2490 }
2491 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2492 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2493 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2494 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2495 "--group", tracing_group_name,
2496 NULL);
2497 if (consumerd64_libdir[0] != '\0') {
2498 free(tmpnew);
2499 }
2500 break;
2501 }
2502 case LTTNG_CONSUMER32_UST:
2503 {
2504 char *tmpnew = NULL;
2505
2506 if (consumerd32_libdir[0] != '\0') {
2507 char *tmp;
2508 size_t tmplen;
2509
2510 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2511 if (!tmp) {
2512 tmp = "";
2513 }
2514 tmplen = strlen("LD_LIBRARY_PATH=")
2515 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2516 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2517 if (!tmpnew) {
2518 ret = -ENOMEM;
2519 goto error;
2520 }
2521 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2522 strcat(tmpnew, consumerd32_libdir);
2523 if (tmp[0] != '\0') {
2524 strcat(tmpnew, ":");
2525 strcat(tmpnew, tmp);
2526 }
2527 ret = putenv(tmpnew);
2528 if (ret) {
2529 ret = -errno;
2530 free(tmpnew);
2531 goto error;
2532 }
2533 }
2534 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2535 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2536 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2537 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2538 "--group", tracing_group_name,
2539 NULL);
2540 if (consumerd32_libdir[0] != '\0') {
2541 free(tmpnew);
2542 }
2543 break;
2544 }
2545 default:
2546 PERROR("unknown consumer type");
2547 exit(EXIT_FAILURE);
2548 }
2549 if (errno != 0) {
2550 PERROR("Consumer execl()");
2551 }
2552 /* Reaching this point, we got a failure on our execl(). */
2553 exit(EXIT_FAILURE);
2554 } else if (pid > 0) {
2555 ret = pid;
2556 } else {
2557 PERROR("start consumer fork");
2558 ret = -errno;
2559 }
2560 error:
2561 return ret;
2562 }
2563
2564 /*
2565 * Spawn the consumerd daemon and session daemon thread.
2566 */
2567 static int start_consumerd(struct consumer_data *consumer_data)
2568 {
2569 int ret;
2570
2571 /*
2572 * Set the listen() state on the socket since there is a possible race
2573 * between the exec() of the consumer daemon and this call if place in the
2574 * consumer thread. See bug #366 for more details.
2575 */
2576 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2577 if (ret < 0) {
2578 goto error;
2579 }
2580
2581 pthread_mutex_lock(&consumer_data->pid_mutex);
2582 if (consumer_data->pid != 0) {
2583 pthread_mutex_unlock(&consumer_data->pid_mutex);
2584 goto end;
2585 }
2586
2587 ret = spawn_consumerd(consumer_data);
2588 if (ret < 0) {
2589 ERR("Spawning consumerd failed");
2590 pthread_mutex_unlock(&consumer_data->pid_mutex);
2591 goto error;
2592 }
2593
2594 /* Setting up the consumer_data pid */
2595 consumer_data->pid = ret;
2596 DBG2("Consumer pid %d", consumer_data->pid);
2597 pthread_mutex_unlock(&consumer_data->pid_mutex);
2598
2599 DBG2("Spawning consumer control thread");
2600 ret = spawn_consumer_thread(consumer_data);
2601 if (ret < 0) {
2602 ERR("Fatal error spawning consumer control thread");
2603 goto error;
2604 }
2605
2606 end:
2607 return 0;
2608
2609 error:
2610 /* Cleanup already created sockets on error. */
2611 if (consumer_data->err_sock >= 0) {
2612 int err;
2613
2614 err = close(consumer_data->err_sock);
2615 if (err < 0) {
2616 PERROR("close consumer data error socket");
2617 }
2618 }
2619 return ret;
2620 }
2621
2622 /*
2623 * Setup necessary data for kernel tracer action.
2624 */
2625 static int init_kernel_tracer(void)
2626 {
2627 int ret;
2628
2629 /* Modprobe lttng kernel modules */
2630 ret = modprobe_lttng_control();
2631 if (ret < 0) {
2632 goto error;
2633 }
2634
2635 /* Open debugfs lttng */
2636 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2637 if (kernel_tracer_fd < 0) {
2638 DBG("Failed to open %s", module_proc_lttng);
2639 ret = -1;
2640 goto error_open;
2641 }
2642
2643 /* Validate kernel version */
2644 ret = kernel_validate_version(kernel_tracer_fd);
2645 if (ret < 0) {
2646 goto error_version;
2647 }
2648
2649 ret = modprobe_lttng_data();
2650 if (ret < 0) {
2651 goto error_modules;
2652 }
2653
2654 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2655 return 0;
2656
2657 error_version:
2658 modprobe_remove_lttng_control();
2659 ret = close(kernel_tracer_fd);
2660 if (ret) {
2661 PERROR("close");
2662 }
2663 kernel_tracer_fd = -1;
2664 return LTTNG_ERR_KERN_VERSION;
2665
2666 error_modules:
2667 ret = close(kernel_tracer_fd);
2668 if (ret) {
2669 PERROR("close");
2670 }
2671
2672 error_open:
2673 modprobe_remove_lttng_control();
2674
2675 error:
2676 WARN("No kernel tracer available");
2677 kernel_tracer_fd = -1;
2678 if (!is_root) {
2679 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2680 } else {
2681 return LTTNG_ERR_KERN_NA;
2682 }
2683 }
2684
2685
2686 /*
2687 * Copy consumer output from the tracing session to the domain session. The
2688 * function also applies the right modification on a per domain basis for the
2689 * trace files destination directory.
2690 *
2691 * Should *NOT* be called with RCU read-side lock held.
2692 */
2693 static int copy_session_consumer(int domain, struct ltt_session *session)
2694 {
2695 int ret;
2696 const char *dir_name;
2697 struct consumer_output *consumer;
2698
2699 assert(session);
2700 assert(session->consumer);
2701
2702 switch (domain) {
2703 case LTTNG_DOMAIN_KERNEL:
2704 DBG3("Copying tracing session consumer output in kernel session");
2705 /*
2706 * XXX: We should audit the session creation and what this function
2707 * does "extra" in order to avoid a destroy since this function is used
2708 * in the domain session creation (kernel and ust) only. Same for UST
2709 * domain.
2710 */
2711 if (session->kernel_session->consumer) {
2712 consumer_destroy_output(session->kernel_session->consumer);
2713 }
2714 session->kernel_session->consumer =
2715 consumer_copy_output(session->consumer);
2716 /* Ease our life a bit for the next part */
2717 consumer = session->kernel_session->consumer;
2718 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2719 break;
2720 case LTTNG_DOMAIN_JUL:
2721 case LTTNG_DOMAIN_LOG4J:
2722 case LTTNG_DOMAIN_PYTHON:
2723 case LTTNG_DOMAIN_UST:
2724 DBG3("Copying tracing session consumer output in UST session");
2725 if (session->ust_session->consumer) {
2726 consumer_destroy_output(session->ust_session->consumer);
2727 }
2728 session->ust_session->consumer =
2729 consumer_copy_output(session->consumer);
2730 /* Ease our life a bit for the next part */
2731 consumer = session->ust_session->consumer;
2732 dir_name = DEFAULT_UST_TRACE_DIR;
2733 break;
2734 default:
2735 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2736 goto error;
2737 }
2738
2739 /* Append correct directory to subdir */
2740 strncat(consumer->subdir, dir_name,
2741 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2742 DBG3("Copy session consumer subdir %s", consumer->subdir);
2743
2744 ret = LTTNG_OK;
2745
2746 error:
2747 return ret;
2748 }
2749
2750 /*
2751 * Create an UST session and add it to the session ust list.
2752 *
2753 * Should *NOT* be called with RCU read-side lock held.
2754 */
2755 static int create_ust_session(struct ltt_session *session,
2756 struct lttng_domain *domain)
2757 {
2758 int ret;
2759 struct ltt_ust_session *lus = NULL;
2760
2761 assert(session);
2762 assert(domain);
2763 assert(session->consumer);
2764
2765 switch (domain->type) {
2766 case LTTNG_DOMAIN_JUL:
2767 case LTTNG_DOMAIN_LOG4J:
2768 case LTTNG_DOMAIN_PYTHON:
2769 case LTTNG_DOMAIN_UST:
2770 break;
2771 default:
2772 ERR("Unknown UST domain on create session %d", domain->type);
2773 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2774 goto error;
2775 }
2776
2777 DBG("Creating UST session");
2778
2779 lus = trace_ust_create_session(session->id);
2780 if (lus == NULL) {
2781 ret = LTTNG_ERR_UST_SESS_FAIL;
2782 goto error;
2783 }
2784
2785 lus->uid = session->uid;
2786 lus->gid = session->gid;
2787 lus->output_traces = session->output_traces;
2788 lus->snapshot_mode = session->snapshot_mode;
2789 lus->live_timer_interval = session->live_timer;
2790 session->ust_session = lus;
2791 if (session->shm_path[0]) {
2792 strncpy(lus->root_shm_path, session->shm_path,
2793 sizeof(lus->root_shm_path));
2794 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2795 strncpy(lus->shm_path, session->shm_path,
2796 sizeof(lus->shm_path));
2797 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2798 strncat(lus->shm_path, "/ust",
2799 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2800 }
2801 /* Copy session output to the newly created UST session */
2802 ret = copy_session_consumer(domain->type, session);
2803 if (ret != LTTNG_OK) {
2804 goto error;
2805 }
2806
2807 return LTTNG_OK;
2808
2809 error:
2810 free(lus);
2811 session->ust_session = NULL;
2812 return ret;
2813 }
2814
2815 /*
2816 * Create a kernel tracer session then create the default channel.
2817 */
2818 static int create_kernel_session(struct ltt_session *session)
2819 {
2820 int ret;
2821
2822 DBG("Creating kernel session");
2823
2824 ret = kernel_create_session(session, kernel_tracer_fd);
2825 if (ret < 0) {
2826 ret = LTTNG_ERR_KERN_SESS_FAIL;
2827 goto error;
2828 }
2829
2830 /* Code flow safety */
2831 assert(session->kernel_session);
2832
2833 /* Copy session output to the newly created Kernel session */
2834 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2835 if (ret != LTTNG_OK) {
2836 goto error;
2837 }
2838
2839 /* Create directory(ies) on local filesystem. */
2840 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2841 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2842 ret = run_as_mkdir_recursive(
2843 session->kernel_session->consumer->dst.trace_path,
2844 S_IRWXU | S_IRWXG, session->uid, session->gid);
2845 if (ret < 0) {
2846 if (ret != -EEXIST) {
2847 ERR("Trace directory creation error");
2848 goto error;
2849 }
2850 }
2851 }
2852
2853 session->kernel_session->uid = session->uid;
2854 session->kernel_session->gid = session->gid;
2855 session->kernel_session->output_traces = session->output_traces;
2856 session->kernel_session->snapshot_mode = session->snapshot_mode;
2857
2858 return LTTNG_OK;
2859
2860 error:
2861 trace_kernel_destroy_session(session->kernel_session);
2862 session->kernel_session = NULL;
2863 return ret;
2864 }
2865
2866 /*
2867 * Count number of session permitted by uid/gid.
2868 */
2869 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2870 {
2871 unsigned int i = 0;
2872 struct ltt_session *session;
2873
2874 DBG("Counting number of available session for UID %d GID %d",
2875 uid, gid);
2876 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2877 /*
2878 * Only list the sessions the user can control.
2879 */
2880 if (!session_access_ok(session, uid, gid)) {
2881 continue;
2882 }
2883 i++;
2884 }
2885 return i;
2886 }
2887
2888 /*
2889 * Process the command requested by the lttng client within the command
2890 * context structure. This function make sure that the return structure (llm)
2891 * is set and ready for transmission before returning.
2892 *
2893 * Return any error encountered or 0 for success.
2894 *
2895 * "sock" is only used for special-case var. len data.
2896 *
2897 * Should *NOT* be called with RCU read-side lock held.
2898 */
2899 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2900 int *sock_error)
2901 {
2902 int ret = LTTNG_OK;
2903 int need_tracing_session = 1;
2904 int need_domain;
2905
2906 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2907
2908 *sock_error = 0;
2909
2910 switch (cmd_ctx->lsm->cmd_type) {
2911 case LTTNG_CREATE_SESSION:
2912 case LTTNG_CREATE_SESSION_SNAPSHOT:
2913 case LTTNG_CREATE_SESSION_LIVE:
2914 case LTTNG_DESTROY_SESSION:
2915 case LTTNG_LIST_SESSIONS:
2916 case LTTNG_LIST_DOMAINS:
2917 case LTTNG_START_TRACE:
2918 case LTTNG_STOP_TRACE:
2919 case LTTNG_DATA_PENDING:
2920 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2921 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2922 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2923 case LTTNG_SNAPSHOT_RECORD:
2924 case LTTNG_SAVE_SESSION:
2925 case LTTNG_SET_SESSION_SHM_PATH:
2926 need_domain = 0;
2927 break;
2928 default:
2929 need_domain = 1;
2930 }
2931
2932 if (opt_no_kernel && need_domain
2933 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2934 if (!is_root) {
2935 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2936 } else {
2937 ret = LTTNG_ERR_KERN_NA;
2938 }
2939 goto error;
2940 }
2941
2942 /* Deny register consumer if we already have a spawned consumer. */
2943 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2944 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2945 if (kconsumer_data.pid > 0) {
2946 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2947 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2948 goto error;
2949 }
2950 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2951 }
2952
2953 /*
2954 * Check for command that don't needs to allocate a returned payload. We do
2955 * this here so we don't have to make the call for no payload at each
2956 * command.
2957 */
2958 switch(cmd_ctx->lsm->cmd_type) {
2959 case LTTNG_LIST_SESSIONS:
2960 case LTTNG_LIST_TRACEPOINTS:
2961 case LTTNG_LIST_TRACEPOINT_FIELDS:
2962 case LTTNG_LIST_DOMAINS:
2963 case LTTNG_LIST_CHANNELS:
2964 case LTTNG_LIST_EVENTS:
2965 case LTTNG_LIST_SYSCALLS:
2966 case LTTNG_LIST_TRACKER_PIDS:
2967 break;
2968 default:
2969 /* Setup lttng message with no payload */
2970 ret = setup_lttng_msg(cmd_ctx, 0);
2971 if (ret < 0) {
2972 /* This label does not try to unlock the session */
2973 goto init_setup_error;
2974 }
2975 }
2976
2977 /* Commands that DO NOT need a session. */
2978 switch (cmd_ctx->lsm->cmd_type) {
2979 case LTTNG_CREATE_SESSION:
2980 case LTTNG_CREATE_SESSION_SNAPSHOT:
2981 case LTTNG_CREATE_SESSION_LIVE:
2982 case LTTNG_CALIBRATE:
2983 case LTTNG_LIST_SESSIONS:
2984 case LTTNG_LIST_TRACEPOINTS:
2985 case LTTNG_LIST_SYSCALLS:
2986 case LTTNG_LIST_TRACEPOINT_FIELDS:
2987 case LTTNG_SAVE_SESSION:
2988 need_tracing_session = 0;
2989 break;
2990 default:
2991 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2992 /*
2993 * We keep the session list lock across _all_ commands
2994 * for now, because the per-session lock does not
2995 * handle teardown properly.
2996 */
2997 session_lock_list();
2998 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2999 if (cmd_ctx->session == NULL) {
3000 ret = LTTNG_ERR_SESS_NOT_FOUND;
3001 goto error;
3002 } else {
3003 /* Acquire lock for the session */
3004 session_lock(cmd_ctx->session);
3005 }
3006 break;
3007 }
3008
3009 /*
3010 * Commands that need a valid session but should NOT create one if none
3011 * exists. Instead of creating one and destroying it when the command is
3012 * handled, process that right before so we save some round trip in useless
3013 * code path.
3014 */
3015 switch (cmd_ctx->lsm->cmd_type) {
3016 case LTTNG_DISABLE_CHANNEL:
3017 case LTTNG_DISABLE_EVENT:
3018 switch (cmd_ctx->lsm->domain.type) {
3019 case LTTNG_DOMAIN_KERNEL:
3020 if (!cmd_ctx->session->kernel_session) {
3021 ret = LTTNG_ERR_NO_CHANNEL;
3022 goto error;
3023 }
3024 break;
3025 case LTTNG_DOMAIN_JUL:
3026 case LTTNG_DOMAIN_LOG4J:
3027 case LTTNG_DOMAIN_PYTHON:
3028 case LTTNG_DOMAIN_UST:
3029 if (!cmd_ctx->session->ust_session) {
3030 ret = LTTNG_ERR_NO_CHANNEL;
3031 goto error;
3032 }
3033 break;
3034 default:
3035 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3036 goto error;
3037 }
3038 default:
3039 break;
3040 }
3041
3042 if (!need_domain) {
3043 goto skip_domain;
3044 }
3045
3046 /*
3047 * Check domain type for specific "pre-action".
3048 */
3049 switch (cmd_ctx->lsm->domain.type) {
3050 case LTTNG_DOMAIN_KERNEL:
3051 if (!is_root) {
3052 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3053 goto error;
3054 }
3055
3056 /* Kernel tracer check */
3057 if (kernel_tracer_fd == -1) {
3058 /* Basically, load kernel tracer modules */
3059 ret = init_kernel_tracer();
3060 if (ret != 0) {
3061 goto error;
3062 }
3063 }
3064
3065 /* Consumer is in an ERROR state. Report back to client */
3066 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3067 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3068 goto error;
3069 }
3070
3071 /* Need a session for kernel command */
3072 if (need_tracing_session) {
3073 if (cmd_ctx->session->kernel_session == NULL) {
3074 ret = create_kernel_session(cmd_ctx->session);
3075 if (ret < 0) {
3076 ret = LTTNG_ERR_KERN_SESS_FAIL;
3077 goto error;
3078 }
3079 }
3080
3081 /* Start the kernel consumer daemon */
3082 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3083 if (kconsumer_data.pid == 0 &&
3084 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3085 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3086 ret = start_consumerd(&kconsumer_data);
3087 if (ret < 0) {
3088 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3089 goto error;
3090 }
3091 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3092 } else {
3093 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3094 }
3095
3096 /*
3097 * The consumer was just spawned so we need to add the socket to
3098 * the consumer output of the session if exist.
3099 */
3100 ret = consumer_create_socket(&kconsumer_data,
3101 cmd_ctx->session->kernel_session->consumer);
3102 if (ret < 0) {
3103 goto error;
3104 }
3105 }
3106
3107 break;
3108 case LTTNG_DOMAIN_JUL:
3109 case LTTNG_DOMAIN_LOG4J:
3110 case LTTNG_DOMAIN_PYTHON:
3111 case LTTNG_DOMAIN_UST:
3112 {
3113 if (!ust_app_supported()) {
3114 ret = LTTNG_ERR_NO_UST;
3115 goto error;
3116 }
3117 /* Consumer is in an ERROR state. Report back to client */
3118 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3119 ret = LTTNG_ERR_NO_USTCONSUMERD;
3120 goto error;
3121 }
3122
3123 if (need_tracing_session) {
3124 /* Create UST session if none exist. */
3125 if (cmd_ctx->session->ust_session == NULL) {
3126 ret = create_ust_session(cmd_ctx->session,
3127 &cmd_ctx->lsm->domain);
3128 if (ret != LTTNG_OK) {
3129 goto error;
3130 }
3131 }
3132
3133 /* Start the UST consumer daemons */
3134 /* 64-bit */
3135 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3136 if (consumerd64_bin[0] != '\0' &&
3137 ustconsumer64_data.pid == 0 &&
3138 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3139 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3140 ret = start_consumerd(&ustconsumer64_data);
3141 if (ret < 0) {
3142 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3143 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3144 goto error;
3145 }
3146
3147 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3148 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3149 } else {
3150 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3151 }
3152
3153 /*
3154 * Setup socket for consumer 64 bit. No need for atomic access
3155 * since it was set above and can ONLY be set in this thread.
3156 */
3157 ret = consumer_create_socket(&ustconsumer64_data,
3158 cmd_ctx->session->ust_session->consumer);
3159 if (ret < 0) {
3160 goto error;
3161 }
3162
3163 /* 32-bit */
3164 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3165 if (consumerd32_bin[0] != '\0' &&
3166 ustconsumer32_data.pid == 0 &&
3167 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3168 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3169 ret = start_consumerd(&ustconsumer32_data);
3170 if (ret < 0) {
3171 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3172 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3173 goto error;
3174 }
3175
3176 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3177 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3178 } else {
3179 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3180 }
3181
3182 /*
3183 * Setup socket for consumer 64 bit. No need for atomic access
3184 * since it was set above and can ONLY be set in this thread.
3185 */
3186 ret = consumer_create_socket(&ustconsumer32_data,
3187 cmd_ctx->session->ust_session->consumer);
3188 if (ret < 0) {
3189 goto error;
3190 }
3191 }
3192 break;
3193 }
3194 default:
3195 break;
3196 }
3197 skip_domain:
3198
3199 /* Validate consumer daemon state when start/stop trace command */
3200 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3201 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3202 switch (cmd_ctx->lsm->domain.type) {
3203 case LTTNG_DOMAIN_JUL:
3204 case LTTNG_DOMAIN_LOG4J:
3205 case LTTNG_DOMAIN_PYTHON:
3206 case LTTNG_DOMAIN_UST:
3207 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3208 ret = LTTNG_ERR_NO_USTCONSUMERD;
3209 goto error;
3210 }
3211 break;
3212 case LTTNG_DOMAIN_KERNEL:
3213 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3214 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3215 goto error;
3216 }
3217 break;
3218 }
3219 }
3220
3221 /*
3222 * Check that the UID or GID match that of the tracing session.
3223 * The root user can interact with all sessions.
3224 */
3225 if (need_tracing_session) {
3226 if (!session_access_ok(cmd_ctx->session,
3227 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3228 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3229 ret = LTTNG_ERR_EPERM;
3230 goto error;
3231 }
3232 }
3233
3234 /*
3235 * Send relayd information to consumer as soon as we have a domain and a
3236 * session defined.
3237 */
3238 if (cmd_ctx->session && need_domain) {
3239 /*
3240 * Setup relayd if not done yet. If the relayd information was already
3241 * sent to the consumer, this call will gracefully return.
3242 */
3243 ret = cmd_setup_relayd(cmd_ctx->session);
3244 if (ret != LTTNG_OK) {
3245 goto error;
3246 }
3247 }
3248
3249 /* Process by command type */
3250 switch (cmd_ctx->lsm->cmd_type) {
3251 case LTTNG_ADD_CONTEXT:
3252 {
3253 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3254 cmd_ctx->lsm->u.context.channel_name,
3255 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3256 break;
3257 }
3258 case LTTNG_DISABLE_CHANNEL:
3259 {
3260 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3261 cmd_ctx->lsm->u.disable.channel_name);
3262 break;
3263 }
3264 case LTTNG_DISABLE_EVENT:
3265 {
3266 /* FIXME: passing packed structure to non-packed pointer */
3267 /* TODO: handle filter */
3268 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3269 cmd_ctx->lsm->u.disable.channel_name,
3270 &cmd_ctx->lsm->u.disable.event);
3271 break;
3272 }
3273 case LTTNG_ENABLE_CHANNEL:
3274 {
3275 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3276 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3277 break;
3278 }
3279 case LTTNG_TRACK_PID:
3280 {
3281 ret = cmd_track_pid(cmd_ctx->session,
3282 cmd_ctx->lsm->domain.type,
3283 cmd_ctx->lsm->u.pid_tracker.pid);
3284 break;
3285 }
3286 case LTTNG_UNTRACK_PID:
3287 {
3288 ret = cmd_untrack_pid(cmd_ctx->session,
3289 cmd_ctx->lsm->domain.type,
3290 cmd_ctx->lsm->u.pid_tracker.pid);
3291 break;
3292 }
3293 case LTTNG_ENABLE_EVENT:
3294 {
3295 struct lttng_event_exclusion *exclusion = NULL;
3296 struct lttng_filter_bytecode *bytecode = NULL;
3297 char *filter_expression = NULL;
3298
3299 /* Handle exclusion events and receive it from the client. */
3300 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3301 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3302
3303 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3304 (count * LTTNG_SYMBOL_NAME_LEN));
3305 if (!exclusion) {
3306 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3307 goto error;
3308 }
3309
3310 DBG("Receiving var len exclusion event list from client ...");
3311 exclusion->count = count;
3312 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3313 count * LTTNG_SYMBOL_NAME_LEN);
3314 if (ret <= 0) {
3315 DBG("Nothing recv() from client var len data... continuing");
3316 *sock_error = 1;
3317 free(exclusion);
3318 ret = LTTNG_ERR_EXCLUSION_INVAL;
3319 goto error;
3320 }
3321 }
3322
3323 /* Get filter expression from client. */
3324 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3325 size_t expression_len =
3326 cmd_ctx->lsm->u.enable.expression_len;
3327
3328 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3329 ret = LTTNG_ERR_FILTER_INVAL;
3330 free(exclusion);
3331 goto error;
3332 }
3333
3334 filter_expression = zmalloc(expression_len);
3335 if (!filter_expression) {
3336 free(exclusion);
3337 ret = LTTNG_ERR_FILTER_NOMEM;
3338 goto error;
3339 }
3340
3341 /* Receive var. len. data */
3342 DBG("Receiving var len filter's expression from client ...");
3343 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3344 expression_len);
3345 if (ret <= 0) {
3346 DBG("Nothing recv() from client car len data... continuing");
3347 *sock_error = 1;
3348 free(filter_expression);
3349 free(exclusion);
3350 ret = LTTNG_ERR_FILTER_INVAL;
3351 goto error;
3352 }
3353 }
3354
3355 /* Handle filter and get bytecode from client. */
3356 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3357 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3358
3359 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3360 ret = LTTNG_ERR_FILTER_INVAL;
3361 free(filter_expression);
3362 free(exclusion);
3363 goto error;
3364 }
3365
3366 bytecode = zmalloc(bytecode_len);
3367 if (!bytecode) {
3368 free(filter_expression);
3369 free(exclusion);
3370 ret = LTTNG_ERR_FILTER_NOMEM;
3371 goto error;
3372 }
3373
3374 /* Receive var. len. data */
3375 DBG("Receiving var len filter's bytecode from client ...");
3376 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3377 if (ret <= 0) {
3378 DBG("Nothing recv() from client car len data... continuing");
3379 *sock_error = 1;
3380 free(filter_expression);
3381 free(bytecode);
3382 free(exclusion);
3383 ret = LTTNG_ERR_FILTER_INVAL;
3384 goto error;
3385 }
3386
3387 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3388 free(filter_expression);
3389 free(bytecode);
3390 free(exclusion);
3391 ret = LTTNG_ERR_FILTER_INVAL;
3392 goto error;
3393 }
3394 }
3395
3396 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3397 cmd_ctx->lsm->u.enable.channel_name,
3398 &cmd_ctx->lsm->u.enable.event,
3399 filter_expression, bytecode, exclusion,
3400 kernel_poll_pipe[1]);
3401 break;
3402 }
3403 case LTTNG_LIST_TRACEPOINTS:
3404 {
3405 struct lttng_event *events;
3406 ssize_t nb_events;
3407
3408 session_lock_list();
3409 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3410 session_unlock_list();
3411 if (nb_events < 0) {
3412 /* Return value is a negative lttng_error_code. */
3413 ret = -nb_events;
3414 goto error;
3415 }
3416
3417 /*
3418 * Setup lttng message with payload size set to the event list size in
3419 * bytes and then copy list into the llm payload.
3420 */
3421 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3422 if (ret < 0) {
3423 free(events);
3424 goto setup_error;
3425 }
3426
3427 /* Copy event list into message payload */
3428 memcpy(cmd_ctx->llm->payload, events,
3429 sizeof(struct lttng_event) * nb_events);
3430
3431 free(events);
3432
3433 ret = LTTNG_OK;
3434 break;
3435 }
3436 case LTTNG_LIST_TRACEPOINT_FIELDS:
3437 {
3438 struct lttng_event_field *fields;
3439 ssize_t nb_fields;
3440
3441 session_lock_list();
3442 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3443 &fields);
3444 session_unlock_list();
3445 if (nb_fields < 0) {
3446 /* Return value is a negative lttng_error_code. */
3447 ret = -nb_fields;
3448 goto error;
3449 }
3450
3451 /*
3452 * Setup lttng message with payload size set to the event list size in
3453 * bytes and then copy list into the llm payload.
3454 */
3455 ret = setup_lttng_msg(cmd_ctx,
3456 sizeof(struct lttng_event_field) * nb_fields);
3457 if (ret < 0) {
3458 free(fields);
3459 goto setup_error;
3460 }
3461
3462 /* Copy event list into message payload */
3463 memcpy(cmd_ctx->llm->payload, fields,
3464 sizeof(struct lttng_event_field) * nb_fields);
3465
3466 free(fields);
3467
3468 ret = LTTNG_OK;
3469 break;
3470 }
3471 case LTTNG_LIST_SYSCALLS:
3472 {
3473 struct lttng_event *events;
3474 ssize_t nb_events;
3475
3476 nb_events = cmd_list_syscalls(&events);
3477 if (nb_events < 0) {
3478 /* Return value is a negative lttng_error_code. */
3479 ret = -nb_events;
3480 goto error;
3481 }
3482
3483 /*
3484 * Setup lttng message with payload size set to the event list size in
3485 * bytes and then copy list into the llm payload.
3486 */
3487 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3488 if (ret < 0) {
3489 free(events);
3490 goto setup_error;
3491 }
3492
3493 /* Copy event list into message payload */
3494 memcpy(cmd_ctx->llm->payload, events,
3495 sizeof(struct lttng_event) * nb_events);
3496
3497 free(events);
3498
3499 ret = LTTNG_OK;
3500 break;
3501 }
3502 case LTTNG_LIST_TRACKER_PIDS:
3503 {
3504 int32_t *pids = NULL;
3505 ssize_t nr_pids;
3506
3507 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3508 cmd_ctx->lsm->domain.type, &pids);
3509 if (nr_pids < 0) {
3510 /* Return value is a negative lttng_error_code. */
3511 ret = -nr_pids;
3512 goto error;
3513 }
3514
3515 /*
3516 * Setup lttng message with payload size set to the event list size in
3517 * bytes and then copy list into the llm payload.
3518 */
3519 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3520 if (ret < 0) {
3521 free(pids);
3522 goto setup_error;
3523 }
3524
3525 /* Copy event list into message payload */
3526 memcpy(cmd_ctx->llm->payload, pids,
3527 sizeof(int) * nr_pids);
3528
3529 free(pids);
3530
3531 ret = LTTNG_OK;
3532 break;
3533 }
3534 case LTTNG_SET_CONSUMER_URI:
3535 {
3536 size_t nb_uri, len;
3537 struct lttng_uri *uris;
3538
3539 nb_uri = cmd_ctx->lsm->u.uri.size;
3540 len = nb_uri * sizeof(struct lttng_uri);
3541
3542 if (nb_uri == 0) {
3543 ret = LTTNG_ERR_INVALID;
3544 goto error;
3545 }
3546
3547 uris = zmalloc(len);
3548 if (uris == NULL) {
3549 ret = LTTNG_ERR_FATAL;
3550 goto error;
3551 }
3552
3553 /* Receive variable len data */
3554 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3555 ret = lttcomm_recv_unix_sock(sock, uris, len);
3556 if (ret <= 0) {
3557 DBG("No URIs received from client... continuing");
3558 *sock_error = 1;
3559 ret = LTTNG_ERR_SESSION_FAIL;
3560 free(uris);
3561 goto error;
3562 }
3563
3564 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3565 free(uris);
3566 if (ret != LTTNG_OK) {
3567 goto error;
3568 }
3569
3570
3571 break;
3572 }
3573 case LTTNG_START_TRACE:
3574 {
3575 ret = cmd_start_trace(cmd_ctx->session);
3576 break;
3577 }
3578 case LTTNG_STOP_TRACE:
3579 {
3580 ret = cmd_stop_trace(cmd_ctx->session);
3581 break;
3582 }
3583 case LTTNG_CREATE_SESSION:
3584 {
3585 size_t nb_uri, len;
3586 struct lttng_uri *uris = NULL;
3587
3588 nb_uri = cmd_ctx->lsm->u.uri.size;
3589 len = nb_uri * sizeof(struct lttng_uri);
3590
3591 if (nb_uri > 0) {
3592 uris = zmalloc(len);
3593 if (uris == NULL) {
3594 ret = LTTNG_ERR_FATAL;
3595 goto error;
3596 }
3597
3598 /* Receive variable len data */
3599 DBG("Waiting for %zu URIs from client ...", nb_uri);
3600 ret = lttcomm_recv_unix_sock(sock, uris, len);
3601 if (ret <= 0) {
3602 DBG("No URIs received from client... continuing");
3603 *sock_error = 1;
3604 ret = LTTNG_ERR_SESSION_FAIL;
3605 free(uris);
3606 goto error;
3607 }
3608
3609 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3610 DBG("Creating session with ONE network URI is a bad call");
3611 ret = LTTNG_ERR_SESSION_FAIL;
3612 free(uris);
3613 goto error;
3614 }
3615 }
3616
3617 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3618 &cmd_ctx->creds, 0);
3619
3620 free(uris);
3621
3622 break;
3623 }
3624 case LTTNG_DESTROY_SESSION:
3625 {
3626 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3627
3628 /* Set session to NULL so we do not unlock it after free. */
3629 cmd_ctx->session = NULL;
3630 break;
3631 }
3632 case LTTNG_LIST_DOMAINS:
3633 {
3634 ssize_t nb_dom;
3635 struct lttng_domain *domains = NULL;
3636
3637 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3638 if (nb_dom < 0) {
3639 /* Return value is a negative lttng_error_code. */
3640 ret = -nb_dom;
3641 goto error;
3642 }
3643
3644 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3645 if (ret < 0) {
3646 free(domains);
3647 goto setup_error;
3648 }
3649
3650 /* Copy event list into message payload */
3651 memcpy(cmd_ctx->llm->payload, domains,
3652 nb_dom * sizeof(struct lttng_domain));
3653
3654 free(domains);
3655
3656 ret = LTTNG_OK;
3657 break;
3658 }
3659 case LTTNG_LIST_CHANNELS:
3660 {
3661 int nb_chan;
3662 struct lttng_channel *channels = NULL;
3663
3664 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3665 cmd_ctx->session, &channels);
3666 if (nb_chan < 0) {
3667 /* Return value is a negative lttng_error_code. */
3668 ret = -nb_chan;
3669 goto error;
3670 }
3671
3672 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3673 if (ret < 0) {
3674 free(channels);
3675 goto setup_error;
3676 }
3677
3678 /* Copy event list into message payload */
3679 memcpy(cmd_ctx->llm->payload, channels,
3680 nb_chan * sizeof(struct lttng_channel));
3681
3682 free(channels);
3683
3684 ret = LTTNG_OK;
3685 break;
3686 }
3687 case LTTNG_LIST_EVENTS:
3688 {
3689 ssize_t nb_event;
3690 struct lttng_event *events = NULL;
3691
3692 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3693 cmd_ctx->lsm->u.list.channel_name, &events);
3694 if (nb_event < 0) {
3695 /* Return value is a negative lttng_error_code. */
3696 ret = -nb_event;
3697 goto error;
3698 }
3699
3700 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3701 if (ret < 0) {
3702 free(events);
3703 goto setup_error;
3704 }
3705
3706 /* Copy event list into message payload */
3707 memcpy(cmd_ctx->llm->payload, events,
3708 nb_event * sizeof(struct lttng_event));
3709
3710 free(events);
3711
3712 ret = LTTNG_OK;
3713 break;
3714 }
3715 case LTTNG_LIST_SESSIONS:
3716 {
3717 unsigned int nr_sessions;
3718
3719 session_lock_list();
3720 nr_sessions = lttng_sessions_count(
3721 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3722 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3723
3724 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3725 if (ret < 0) {
3726 session_unlock_list();
3727 goto setup_error;
3728 }
3729
3730 /* Filled the session array */
3731 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3732 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3733 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3734
3735 session_unlock_list();
3736
3737 ret = LTTNG_OK;
3738 break;
3739 }
3740 case LTTNG_CALIBRATE:
3741 {
3742 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3743 &cmd_ctx->lsm->u.calibrate);
3744 break;
3745 }
3746 case LTTNG_REGISTER_CONSUMER:
3747 {
3748 struct consumer_data *cdata;
3749
3750 switch (cmd_ctx->lsm->domain.type) {
3751 case LTTNG_DOMAIN_KERNEL:
3752 cdata = &kconsumer_data;
3753 break;
3754 default:
3755 ret = LTTNG_ERR_UND;
3756 goto error;
3757 }
3758
3759 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3760 cmd_ctx->lsm->u.reg.path, cdata);
3761 break;
3762 }
3763 case LTTNG_DATA_PENDING:
3764 {
3765 ret = cmd_data_pending(cmd_ctx->session);
3766 break;
3767 }
3768 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3769 {
3770 struct lttcomm_lttng_output_id reply;
3771
3772 ret = cmd_snapshot_add_output(cmd_ctx->session,
3773 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3774 if (ret != LTTNG_OK) {
3775 goto error;
3776 }
3777
3778 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3779 if (ret < 0) {
3780 goto setup_error;
3781 }
3782
3783 /* Copy output list into message payload */
3784 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3785 ret = LTTNG_OK;
3786 break;
3787 }
3788 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3789 {
3790 ret = cmd_snapshot_del_output(cmd_ctx->session,
3791 &cmd_ctx->lsm->u.snapshot_output.output);
3792 break;
3793 }
3794 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3795 {
3796 ssize_t nb_output;
3797 struct lttng_snapshot_output *outputs = NULL;
3798
3799 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3800 if (nb_output < 0) {
3801 ret = -nb_output;
3802 goto error;
3803 }
3804
3805 ret = setup_lttng_msg(cmd_ctx,
3806 nb_output * sizeof(struct lttng_snapshot_output));
3807 if (ret < 0) {
3808 free(outputs);
3809 goto setup_error;
3810 }
3811
3812 if (outputs) {
3813 /* Copy output list into message payload */
3814 memcpy(cmd_ctx->llm->payload, outputs,
3815 nb_output * sizeof(struct lttng_snapshot_output));
3816 free(outputs);
3817 }
3818
3819 ret = LTTNG_OK;
3820 break;
3821 }
3822 case LTTNG_SNAPSHOT_RECORD:
3823 {
3824 ret = cmd_snapshot_record(cmd_ctx->session,
3825 &cmd_ctx->lsm->u.snapshot_record.output,
3826 cmd_ctx->lsm->u.snapshot_record.wait);
3827 break;
3828 }
3829 case LTTNG_CREATE_SESSION_SNAPSHOT:
3830 {
3831 size_t nb_uri, len;
3832 struct lttng_uri *uris = NULL;
3833
3834 nb_uri = cmd_ctx->lsm->u.uri.size;
3835 len = nb_uri * sizeof(struct lttng_uri);
3836
3837 if (nb_uri > 0) {
3838 uris = zmalloc(len);
3839 if (uris == NULL) {
3840 ret = LTTNG_ERR_FATAL;
3841 goto error;
3842 }
3843
3844 /* Receive variable len data */
3845 DBG("Waiting for %zu URIs from client ...", nb_uri);
3846 ret = lttcomm_recv_unix_sock(sock, uris, len);
3847 if (ret <= 0) {
3848 DBG("No URIs received from client... continuing");
3849 *sock_error = 1;
3850 ret = LTTNG_ERR_SESSION_FAIL;
3851 free(uris);
3852 goto error;
3853 }
3854
3855 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3856 DBG("Creating session with ONE network URI is a bad call");
3857 ret = LTTNG_ERR_SESSION_FAIL;
3858 free(uris);
3859 goto error;
3860 }
3861 }
3862
3863 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3864 nb_uri, &cmd_ctx->creds);
3865 free(uris);
3866 break;
3867 }
3868 case LTTNG_CREATE_SESSION_LIVE:
3869 {
3870 size_t nb_uri, len;
3871 struct lttng_uri *uris = NULL;
3872
3873 nb_uri = cmd_ctx->lsm->u.uri.size;
3874 len = nb_uri * sizeof(struct lttng_uri);
3875
3876 if (nb_uri > 0) {
3877 uris = zmalloc(len);
3878 if (uris == NULL) {
3879 ret = LTTNG_ERR_FATAL;
3880 goto error;
3881 }
3882
3883 /* Receive variable len data */
3884 DBG("Waiting for %zu URIs from client ...", nb_uri);
3885 ret = lttcomm_recv_unix_sock(sock, uris, len);
3886 if (ret <= 0) {
3887 DBG("No URIs received from client... continuing");
3888 *sock_error = 1;
3889 ret = LTTNG_ERR_SESSION_FAIL;
3890 free(uris);
3891 goto error;
3892 }
3893
3894 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3895 DBG("Creating session with ONE network URI is a bad call");
3896 ret = LTTNG_ERR_SESSION_FAIL;
3897 free(uris);
3898 goto error;
3899 }
3900 }
3901
3902 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3903 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3904 free(uris);
3905 break;
3906 }
3907 case LTTNG_SAVE_SESSION:
3908 {
3909 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3910 &cmd_ctx->creds);
3911 break;
3912 }
3913 case LTTNG_SET_SESSION_SHM_PATH:
3914 {
3915 ret = cmd_set_session_shm_path(cmd_ctx->session,
3916 cmd_ctx->lsm->u.set_shm_path.shm_path);
3917 break;
3918 }
3919 default:
3920 ret = LTTNG_ERR_UND;
3921 break;
3922 }
3923
3924 error:
3925 if (cmd_ctx->llm == NULL) {
3926 DBG("Missing llm structure. Allocating one.");
3927 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3928 goto setup_error;
3929 }
3930 }
3931 /* Set return code */
3932 cmd_ctx->llm->ret_code = ret;
3933 setup_error:
3934 if (cmd_ctx->session) {
3935 session_unlock(cmd_ctx->session);
3936 }
3937 if (need_tracing_session) {
3938 session_unlock_list();
3939 }
3940 init_setup_error:
3941 return ret;
3942 }
3943
3944 /*
3945 * Thread managing health check socket.
3946 */
3947 static void *thread_manage_health(void *data)
3948 {
3949 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3950 uint32_t revents, nb_fd;
3951 struct lttng_poll_event events;
3952 struct health_comm_msg msg;
3953 struct health_comm_reply reply;
3954
3955 DBG("[thread] Manage health check started");
3956
3957 rcu_register_thread();
3958
3959 /* We might hit an error path before this is created. */
3960 lttng_poll_init(&events);
3961
3962 /* Create unix socket */
3963 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3964 if (sock < 0) {
3965 ERR("Unable to create health check Unix socket");
3966 ret = -1;
3967 goto error;
3968 }
3969
3970 if (is_root) {
3971 /* lttng health client socket path permissions */
3972 ret = chown(health_unix_sock_path, 0,
3973 utils_get_group_id(tracing_group_name));
3974 if (ret < 0) {
3975 ERR("Unable to set group on %s", health_unix_sock_path);
3976 PERROR("chown");
3977 ret = -1;
3978 goto error;
3979 }
3980
3981 ret = chmod(health_unix_sock_path,
3982 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3983 if (ret < 0) {
3984 ERR("Unable to set permissions on %s", health_unix_sock_path);
3985 PERROR("chmod");
3986 ret = -1;
3987 goto error;
3988 }
3989 }
3990
3991 /*
3992 * Set the CLOEXEC flag. Return code is useless because either way, the
3993 * show must go on.
3994 */
3995 (void) utils_set_fd_cloexec(sock);
3996
3997 ret = lttcomm_listen_unix_sock(sock);
3998 if (ret < 0) {
3999 goto error;
4000 }
4001
4002 /*
4003 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4004 * more will be added to this poll set.
4005 */
4006 ret = sessiond_set_thread_pollset(&events, 2);
4007 if (ret < 0) {
4008 goto error;
4009 }
4010
4011 /* Add the application registration socket */
4012 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4013 if (ret < 0) {
4014 goto error;
4015 }
4016
4017 sessiond_notify_ready();
4018
4019 while (1) {
4020 DBG("Health check ready");
4021
4022 /* Inifinite blocking call, waiting for transmission */
4023 restart:
4024 ret = lttng_poll_wait(&events, -1);
4025 if (ret < 0) {
4026 /*
4027 * Restart interrupted system call.
4028 */
4029 if (errno == EINTR) {
4030 goto restart;
4031 }
4032 goto error;
4033 }
4034
4035 nb_fd = ret;
4036
4037 for (i = 0; i < nb_fd; i++) {
4038 /* Fetch once the poll data */
4039 revents = LTTNG_POLL_GETEV(&events, i);
4040 pollfd = LTTNG_POLL_GETFD(&events, i);
4041
4042 if (!revents) {
4043 /* No activity for this FD (poll implementation). */
4044 continue;
4045 }
4046
4047 /* Thread quit pipe has been closed. Killing thread. */
4048 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4049 if (ret) {
4050 err = 0;
4051 goto exit;
4052 }
4053
4054 /* Event on the registration socket */
4055 if (pollfd == sock) {
4056 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4057 ERR("Health socket poll error");
4058 goto error;
4059 }
4060 }
4061 }
4062
4063 new_sock = lttcomm_accept_unix_sock(sock);
4064 if (new_sock < 0) {
4065 goto error;
4066 }
4067
4068 /*
4069 * Set the CLOEXEC flag. Return code is useless because either way, the
4070 * show must go on.
4071 */
4072 (void) utils_set_fd_cloexec(new_sock);
4073
4074 DBG("Receiving data from client for health...");
4075 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4076 if (ret <= 0) {
4077 DBG("Nothing recv() from client... continuing");
4078 ret = close(new_sock);
4079 if (ret) {
4080 PERROR("close");
4081 }
4082 new_sock = -1;
4083 continue;
4084 }
4085
4086 rcu_thread_online();
4087
4088 memset(&reply, 0, sizeof(reply));
4089 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4090 /*
4091 * health_check_state returns 0 if health is
4092 * bad.
4093 */
4094 if (!health_check_state(health_sessiond, i)) {
4095 reply.ret_code |= 1ULL << i;
4096 }
4097 }
4098
4099 DBG2("Health check return value %" PRIx64, reply.ret_code);
4100
4101 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4102 if (ret < 0) {
4103 ERR("Failed to send health data back to client");
4104 }
4105
4106 /* End of transmission */
4107 ret = close(new_sock);
4108 if (ret) {
4109 PERROR("close");
4110 }
4111 new_sock = -1;
4112 }
4113
4114 exit:
4115 error:
4116 if (err) {
4117 ERR("Health error occurred in %s", __func__);
4118 }
4119 DBG("Health check thread dying");
4120 unlink(health_unix_sock_path);
4121 if (sock >= 0) {
4122 ret = close(sock);
4123 if (ret) {
4124 PERROR("close");
4125 }
4126 }
4127
4128 lttng_poll_clean(&events);
4129
4130 rcu_unregister_thread();
4131 return NULL;
4132 }
4133
4134 /*
4135 * This thread manage all clients request using the unix client socket for
4136 * communication.
4137 */
4138 static void *thread_manage_clients(void *data)
4139 {
4140 int sock = -1, ret, i, pollfd, err = -1;
4141 int sock_error;
4142 uint32_t revents, nb_fd;
4143 struct command_ctx *cmd_ctx = NULL;
4144 struct lttng_poll_event events;
4145
4146 DBG("[thread] Manage client started");
4147
4148 rcu_register_thread();
4149
4150 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4151
4152 health_code_update();
4153
4154 ret = lttcomm_listen_unix_sock(client_sock);
4155 if (ret < 0) {
4156 goto error_listen;
4157 }
4158
4159 /*
4160 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4161 * more will be added to this poll set.
4162 */
4163 ret = sessiond_set_thread_pollset(&events, 2);
4164 if (ret < 0) {
4165 goto error_create_poll;
4166 }
4167
4168 /* Add the application registration socket */
4169 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4170 if (ret < 0) {
4171 goto error;
4172 }
4173
4174 sessiond_notify_ready();
4175 ret = sem_post(&load_info->message_thread_ready);
4176 if (ret) {
4177 PERROR("sem_post message_thread_ready");
4178 goto error;
4179 }
4180
4181 /* This testpoint is after we signal readiness to the parent. */
4182 if (testpoint(sessiond_thread_manage_clients)) {
4183 goto error;
4184 }
4185
4186 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4187 goto error;
4188 }
4189
4190 health_code_update();
4191
4192 while (1) {
4193 DBG("Accepting client command ...");
4194
4195 /* Inifinite blocking call, waiting for transmission */
4196 restart:
4197 health_poll_entry();
4198 ret = lttng_poll_wait(&events, -1);
4199 health_poll_exit();
4200 if (ret < 0) {
4201 /*
4202 * Restart interrupted system call.
4203 */
4204 if (errno == EINTR) {
4205 goto restart;
4206 }
4207 goto error;
4208 }
4209
4210 nb_fd = ret;
4211
4212 for (i = 0; i < nb_fd; i++) {
4213 /* Fetch once the poll data */
4214 revents = LTTNG_POLL_GETEV(&events, i);
4215 pollfd = LTTNG_POLL_GETFD(&events, i);
4216
4217 health_code_update();
4218
4219 if (!revents) {
4220 /* No activity for this FD (poll implementation). */
4221 continue;
4222 }
4223
4224 /* Thread quit pipe has been closed. Killing thread. */
4225 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4226 if (ret) {
4227 err = 0;
4228 goto exit;
4229 }
4230
4231 /* Event on the registration socket */
4232 if (pollfd == client_sock) {
4233 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4234 ERR("Client socket poll error");
4235 goto error;
4236 }
4237 }
4238 }
4239
4240 DBG("Wait for client response");
4241
4242 health_code_update();
4243
4244 sock = lttcomm_accept_unix_sock(client_sock);
4245 if (sock < 0) {
4246 goto error;
4247 }
4248
4249 /*
4250 * Set the CLOEXEC flag. Return code is useless because either way, the
4251 * show must go on.
4252 */
4253 (void) utils_set_fd_cloexec(sock);
4254
4255 /* Set socket option for credentials retrieval */
4256 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4257 if (ret < 0) {
4258 goto error;
4259 }
4260
4261 /* Allocate context command to process the client request */
4262 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4263 if (cmd_ctx == NULL) {
4264 PERROR("zmalloc cmd_ctx");
4265 goto error;
4266 }
4267
4268 /* Allocate data buffer for reception */
4269 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4270 if (cmd_ctx->lsm == NULL) {
4271 PERROR("zmalloc cmd_ctx->lsm");
4272 goto error;
4273 }
4274
4275 cmd_ctx->llm = NULL;
4276 cmd_ctx->session = NULL;
4277
4278 health_code_update();
4279
4280 /*
4281 * Data is received from the lttng client. The struct
4282 * lttcomm_session_msg (lsm) contains the command and data request of
4283 * the client.
4284 */
4285 DBG("Receiving data from client ...");
4286 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4287 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4288 if (ret <= 0) {
4289 DBG("Nothing recv() from client... continuing");
4290 ret = close(sock);
4291 if (ret) {
4292 PERROR("close");
4293 }
4294 sock = -1;
4295 clean_command_ctx(&cmd_ctx);
4296 continue;
4297 }
4298
4299 health_code_update();
4300
4301 // TODO: Validate cmd_ctx including sanity check for
4302 // security purpose.
4303
4304 rcu_thread_online();
4305 /*
4306 * This function dispatch the work to the kernel or userspace tracer
4307 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4308 * informations for the client. The command context struct contains
4309 * everything this function may needs.
4310 */
4311 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4312 rcu_thread_offline();
4313 if (ret < 0) {
4314 ret = close(sock);
4315 if (ret) {
4316 PERROR("close");
4317 }
4318 sock = -1;
4319 /*
4320 * TODO: Inform client somehow of the fatal error. At
4321 * this point, ret < 0 means that a zmalloc failed
4322 * (ENOMEM). Error detected but still accept
4323 * command, unless a socket error has been
4324 * detected.
4325 */
4326 clean_command_ctx(&cmd_ctx);
4327 continue;
4328 }
4329
4330 health_code_update();
4331
4332 DBG("Sending response (size: %d, retcode: %s)",
4333 cmd_ctx->lttng_msg_size,
4334 lttng_strerror(-cmd_ctx->llm->ret_code));
4335 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4336 if (ret < 0) {
4337 ERR("Failed to send data back to client");
4338 }
4339
4340 /* End of transmission */
4341 ret = close(sock);
4342 if (ret) {
4343 PERROR("close");
4344 }
4345 sock = -1;
4346
4347 clean_command_ctx(&cmd_ctx);
4348
4349 health_code_update();
4350 }
4351
4352 exit:
4353 error:
4354 if (sock >= 0) {
4355 ret = close(sock);
4356 if (ret) {
4357 PERROR("close");
4358 }
4359 }
4360
4361 lttng_poll_clean(&events);
4362 clean_command_ctx(&cmd_ctx);
4363
4364 error_listen:
4365 error_create_poll:
4366 unlink(client_unix_sock_path);
4367 if (client_sock >= 0) {
4368 ret = close(client_sock);
4369 if (ret) {
4370 PERROR("close");
4371 }
4372 }
4373
4374 if (err) {
4375 health_error();
4376 ERR("Health error occurred in %s", __func__);
4377 }
4378
4379 health_unregister(health_sessiond);
4380
4381 DBG("Client thread dying");
4382
4383 rcu_unregister_thread();
4384
4385 /*
4386 * Since we are creating the consumer threads, we own them, so we need
4387 * to join them before our thread exits.
4388 */
4389 ret = join_consumer_thread(&kconsumer_data);
4390 if (ret) {
4391 errno = ret;
4392 PERROR("join_consumer");
4393 }
4394
4395 ret = join_consumer_thread(&ustconsumer32_data);
4396 if (ret) {
4397 errno = ret;
4398 PERROR("join_consumer ust32");
4399 }
4400
4401 ret = join_consumer_thread(&ustconsumer64_data);
4402 if (ret) {
4403 errno = ret;
4404 PERROR("join_consumer ust64");
4405 }
4406 return NULL;
4407 }
4408
4409
4410 /*
4411 * usage function on stderr
4412 */
4413 static void usage(void)
4414 {
4415 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4416 fprintf(stderr, " -h, --help Display this usage.\n");
4417 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4418 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4419 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4420 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4421 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4422 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4423 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4424 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4425 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4426 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4427 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4428 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4429 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4430 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4431 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4432 fprintf(stderr, " -V, --version Show version number.\n");
4433 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4434 fprintf(stderr, " -q, --quiet No output at all.\n");
4435 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4436 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4437 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4438 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4439 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4440 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4441 fprintf(stderr, " -l --load PATH Load session configuration\n");
4442 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4443 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4444 }
4445
4446 static int string_match(const char *str1, const char *str2)
4447 {
4448 return (str1 && str2) && !strcmp(str1, str2);
4449 }
4450
4451 /*
4452 * Take an option from the getopt output and set it in the right variable to be
4453 * used later.
4454 *
4455 * Return 0 on success else a negative value.
4456 */
4457 static int set_option(int opt, const char *arg, const char *optname)
4458 {
4459 int ret = 0;
4460
4461 if (arg && arg[0] == '\0') {
4462 /*
4463 * This only happens if the value is read from daemon config
4464 * file. This means the option requires an argument and the
4465 * configuration file contains a line such as:
4466 * my_option =
4467 */
4468 ret = -EINVAL;
4469 goto end;
4470 }
4471
4472 if (string_match(optname, "client-sock") || opt == 'c') {
4473 if (lttng_is_setuid_setgid()) {
4474 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4475 "-c, --client-sock");
4476 } else {
4477 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4478 }
4479 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4480 if (lttng_is_setuid_setgid()) {
4481 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4482 "-a, --apps-sock");
4483 } else {
4484 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4485 }
4486 } else if (string_match(optname, "daemonize") || opt == 'd') {
4487 opt_daemon = 1;
4488 } else if (string_match(optname, "background") || opt == 'b') {
4489 opt_background = 1;
4490 } else if (string_match(optname, "group") || opt == 'g') {
4491 if (lttng_is_setuid_setgid()) {
4492 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4493 "-g, --group");
4494 } else {
4495 /*
4496 * If the override option is set, the pointer points to a
4497 * *non* const thus freeing it even though the variable type is
4498 * set to const.
4499 */
4500 if (tracing_group_name_override) {
4501 free((void *) tracing_group_name);
4502 }
4503 tracing_group_name = strdup(arg);
4504 if (!tracing_group_name) {
4505 PERROR("strdup");
4506 ret = -ENOMEM;
4507 }
4508 tracing_group_name_override = 1;
4509 }
4510 } else if (string_match(optname, "help") || opt == 'h') {
4511 usage();
4512 exit(EXIT_SUCCESS);
4513 } else if (string_match(optname, "version") || opt == 'V') {
4514 fprintf(stdout, "%s\n", VERSION);
4515 exit(EXIT_SUCCESS);
4516 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4517 opt_sig_parent = 1;
4518 } else if (string_match(optname, "kconsumerd-err-sock")) {
4519 if (lttng_is_setuid_setgid()) {
4520 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4521 "--kconsumerd-err-sock");
4522 } else {
4523 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4524 }
4525 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4526 if (lttng_is_setuid_setgid()) {
4527 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4528 "--kconsumerd-cmd-sock");
4529 } else {
4530 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4531 }
4532 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4533 if (lttng_is_setuid_setgid()) {
4534 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4535 "--ustconsumerd64-err-sock");
4536 } else {
4537 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4538 }
4539 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4540 if (lttng_is_setuid_setgid()) {
4541 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4542 "--ustconsumerd64-cmd-sock");
4543 } else {
4544 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4545 }
4546 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4547 if (lttng_is_setuid_setgid()) {
4548 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4549 "--ustconsumerd32-err-sock");
4550 } else {
4551 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4552 }
4553 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4554 if (lttng_is_setuid_setgid()) {
4555 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4556 "--ustconsumerd32-cmd-sock");
4557 } else {
4558 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4559 }
4560 } else if (string_match(optname, "no-kernel")) {
4561 opt_no_kernel = 1;
4562 } else if (string_match(optname, "quiet") || opt == 'q') {
4563 lttng_opt_quiet = 1;
4564 } else if (string_match(optname, "verbose") || opt == 'v') {
4565 /* Verbose level can increase using multiple -v */
4566 if (arg) {
4567 /* Value obtained from config file */
4568 lttng_opt_verbose = config_parse_value(arg);
4569 } else {
4570 /* -v used on command line */
4571 lttng_opt_verbose++;
4572 }
4573 /* Clamp value to [0, 3] */
4574 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4575 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4576 } else if (string_match(optname, "verbose-consumer")) {
4577 if (arg) {
4578 opt_verbose_consumer = config_parse_value(arg);
4579 } else {
4580 opt_verbose_consumer += 1;
4581 }
4582 } else if (string_match(optname, "consumerd32-path")) {
4583 if (lttng_is_setuid_setgid()) {
4584 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4585 "--consumerd32-path");
4586 } else {
4587 if (consumerd32_bin_override) {
4588 free((void *) consumerd32_bin);
4589 }
4590 consumerd32_bin = strdup(arg);
4591 if (!consumerd32_bin) {
4592 PERROR("strdup");
4593 ret = -ENOMEM;
4594 }
4595 consumerd32_bin_override = 1;
4596 }
4597 } else if (string_match(optname, "consumerd32-libdir")) {
4598 if (lttng_is_setuid_setgid()) {
4599 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4600 "--consumerd32-libdir");
4601 } else {
4602 if (consumerd32_libdir_override) {
4603 free((void *) consumerd32_libdir);
4604 }
4605 consumerd32_libdir = strdup(arg);
4606 if (!consumerd32_libdir) {
4607 PERROR("strdup");
4608 ret = -ENOMEM;
4609 }
4610 consumerd32_libdir_override = 1;
4611 }
4612 } else if (string_match(optname, "consumerd64-path")) {
4613 if (lttng_is_setuid_setgid()) {
4614 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4615 "--consumerd64-path");
4616 } else {
4617 if (consumerd64_bin_override) {
4618 free((void *) consumerd64_bin);
4619 }
4620 consumerd64_bin = strdup(arg);
4621 if (!consumerd64_bin) {
4622 PERROR("strdup");
4623 ret = -ENOMEM;
4624 }
4625 consumerd64_bin_override = 1;
4626 }
4627 } else if (string_match(optname, "consumerd64-libdir")) {
4628 if (lttng_is_setuid_setgid()) {
4629 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4630 "--consumerd64-libdir");
4631 } else {
4632 if (consumerd64_libdir_override) {
4633 free((void *) consumerd64_libdir);
4634 }
4635 consumerd64_libdir = strdup(arg);
4636 if (!consumerd64_libdir) {
4637 PERROR("strdup");
4638 ret = -ENOMEM;
4639 }
4640 consumerd64_libdir_override = 1;
4641 }
4642 } else if (string_match(optname, "pidfile") || opt == 'p') {
4643 if (lttng_is_setuid_setgid()) {
4644 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4645 "-p, --pidfile");
4646 } else {
4647 free(opt_pidfile);
4648 opt_pidfile = strdup(arg);
4649 if (!opt_pidfile) {
4650 PERROR("strdup");
4651 ret = -ENOMEM;
4652 }
4653 }
4654 } else if (string_match(optname, "agent-tcp-port")) {
4655 if (lttng_is_setuid_setgid()) {
4656 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4657 "--agent-tcp-port");
4658 } else {
4659 unsigned long v;
4660
4661 if (!arg) {
4662 ret = -EINVAL;
4663 goto end;
4664 }
4665 errno = 0;
4666 v = strtoul(arg, NULL, 0);
4667 if (errno != 0 || !isdigit(arg[0])) {
4668 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4669 return -1;
4670 }
4671 if (v == 0 || v >= 65535) {
4672 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4673 return -1;
4674 }
4675 agent_tcp_port = (uint32_t) v;
4676 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4677 }
4678 } else if (string_match(optname, "load") || opt == 'l') {
4679 if (lttng_is_setuid_setgid()) {
4680 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4681 "-l, --load");
4682 } else {
4683 free(opt_load_session_path);
4684 opt_load_session_path = strdup(arg);
4685 if (!opt_load_session_path) {
4686 PERROR("strdup");
4687 ret = -ENOMEM;
4688 }
4689 }
4690 } else if (string_match(optname, "kmod-probes")) {
4691 if (lttng_is_setuid_setgid()) {
4692 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4693 "--kmod-probes");
4694 } else {
4695 free(kmod_probes_list);
4696 kmod_probes_list = strdup(arg);
4697 if (!kmod_probes_list) {
4698 PERROR("strdup");
4699 ret = -ENOMEM;
4700 }
4701 }
4702 } else if (string_match(optname, "extra-kmod-probes")) {
4703 if (lttng_is_setuid_setgid()) {
4704 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4705 "--extra-kmod-probes");
4706 } else {
4707 free(kmod_extra_probes_list);
4708 kmod_extra_probes_list = strdup(arg);
4709 if (!kmod_extra_probes_list) {
4710 PERROR("strdup");
4711 ret = -ENOMEM;
4712 }
4713 }
4714 } else if (string_match(optname, "config") || opt == 'f') {
4715 /* This is handled in set_options() thus silent skip. */
4716 goto end;
4717 } else {
4718 /* Unknown option or other error.
4719 * Error is printed by getopt, just return */
4720 ret = -1;
4721 }
4722
4723 end:
4724 if (ret == -EINVAL) {
4725 const char *opt_name = "unknown";
4726 int i;
4727
4728 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4729 i++) {
4730 if (opt == long_options[i].val) {
4731 opt_name = long_options[i].name;
4732 break;
4733 }
4734 }
4735
4736 WARN("Invalid argument provided for option \"%s\", using default value.",
4737 opt_name);
4738 }
4739
4740 return ret;
4741 }
4742
4743 /*
4744 * config_entry_handler_cb used to handle options read from a config file.
4745 * See config_entry_handler_cb comment in common/config/config.h for the
4746 * return value conventions.
4747 */
4748 static int config_entry_handler(const struct config_entry *entry, void *unused)
4749 {
4750 int ret = 0, i;
4751
4752 if (!entry || !entry->name || !entry->value) {
4753 ret = -EINVAL;
4754 goto end;
4755 }
4756
4757 /* Check if the option is to be ignored */
4758 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4759 if (!strcmp(entry->name, config_ignore_options[i])) {
4760 goto end;
4761 }
4762 }
4763
4764 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4765 i++) {
4766
4767 /* Ignore if not fully matched. */
4768 if (strcmp(entry->name, long_options[i].name)) {
4769 continue;
4770 }
4771
4772 /*
4773 * If the option takes no argument on the command line, we have to
4774 * check if the value is "true". We support non-zero numeric values,
4775 * true, on and yes.
4776 */
4777 if (!long_options[i].has_arg) {
4778 ret = config_parse_value(entry->value);
4779 if (ret <= 0) {
4780 if (ret) {
4781 WARN("Invalid configuration value \"%s\" for option %s",
4782 entry->value, entry->name);
4783 }
4784 /* False, skip boolean config option. */
4785 goto end;
4786 }
4787 }
4788
4789 ret = set_option(long_options[i].val, entry->value, entry->name);
4790 goto end;
4791 }
4792
4793 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4794
4795 end:
4796 return ret;
4797 }
4798
4799 /*
4800 * daemon configuration loading and argument parsing
4801 */
4802 static int set_options(int argc, char **argv)
4803 {
4804 int ret = 0, c = 0, option_index = 0;
4805 int orig_optopt = optopt, orig_optind = optind;
4806 char *optstring;
4807 const char *config_path = NULL;
4808
4809 optstring = utils_generate_optstring(long_options,
4810 sizeof(long_options) / sizeof(struct option));
4811 if (!optstring) {
4812 ret = -ENOMEM;
4813 goto end;
4814 }
4815
4816 /* Check for the --config option */
4817 while ((c = getopt_long(argc, argv, optstring, long_options,
4818 &option_index)) != -1) {
4819 if (c == '?') {
4820 ret = -EINVAL;
4821 goto end;
4822 } else if (c != 'f') {
4823 /* if not equal to --config option. */
4824 continue;
4825 }
4826
4827 if (lttng_is_setuid_setgid()) {
4828 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4829 "-f, --config");
4830 } else {
4831 config_path = utils_expand_path(optarg);
4832 if (!config_path) {
4833 ERR("Failed to resolve path: %s", optarg);
4834 }
4835 }
4836 }
4837
4838 ret = config_get_section_entries(config_path, config_section_name,
4839 config_entry_handler, NULL);
4840 if (ret) {
4841 if (ret > 0) {
4842 ERR("Invalid configuration option at line %i", ret);
4843 ret = -1;
4844 }
4845 goto end;
4846 }
4847
4848 /* Reset getopt's global state */
4849 optopt = orig_optopt;
4850 optind = orig_optind;
4851 while (1) {
4852 option_index = -1;
4853 /*
4854 * getopt_long() will not set option_index if it encounters a
4855 * short option.
4856 */
4857 c = getopt_long(argc, argv, optstring, long_options,
4858 &option_index);
4859 if (c == -1) {
4860 break;
4861 }
4862
4863 /*
4864 * Pass NULL as the long option name if popt left the index
4865 * unset.
4866 */
4867 ret = set_option(c, optarg,
4868 option_index < 0 ? NULL :
4869 long_options[option_index].name);
4870 if (ret < 0) {
4871 break;
4872 }
4873 }
4874
4875 end:
4876 free(optstring);
4877 return ret;
4878 }
4879
4880 /*
4881 * Creates the two needed socket by the daemon.
4882 * apps_sock - The communication socket for all UST apps.
4883 * client_sock - The communication of the cli tool (lttng).
4884 */
4885 static int init_daemon_socket(void)
4886 {
4887 int ret = 0;
4888 mode_t old_umask;
4889
4890 old_umask = umask(0);
4891
4892 /* Create client tool unix socket */
4893 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4894 if (client_sock < 0) {
4895 ERR("Create unix sock failed: %s", client_unix_sock_path);
4896 ret = -1;
4897 goto end;
4898 }
4899
4900 /* Set the cloexec flag */
4901 ret = utils_set_fd_cloexec(client_sock);
4902 if (ret < 0) {
4903 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4904 "Continuing but note that the consumer daemon will have a "
4905 "reference to this socket on exec()", client_sock);
4906 }
4907
4908 /* File permission MUST be 660 */
4909 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4910 if (ret < 0) {
4911 ERR("Set file permissions failed: %s", client_unix_sock_path);
4912 PERROR("chmod");
4913 goto end;
4914 }
4915
4916 /* Create the application unix socket */
4917 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4918 if (apps_sock < 0) {
4919 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4920 ret = -1;
4921 goto end;
4922 }
4923
4924 /* Set the cloexec flag */
4925 ret = utils_set_fd_cloexec(apps_sock);
4926 if (ret < 0) {
4927 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4928 "Continuing but note that the consumer daemon will have a "
4929 "reference to this socket on exec()", apps_sock);
4930 }
4931
4932 /* File permission MUST be 666 */
4933 ret = chmod(apps_unix_sock_path,
4934 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4935 if (ret < 0) {
4936 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4937 PERROR("chmod");
4938 goto end;
4939 }
4940
4941 DBG3("Session daemon client socket %d and application socket %d created",
4942 client_sock, apps_sock);
4943
4944 end:
4945 umask(old_umask);
4946 return ret;
4947 }
4948
4949 /*
4950 * Check if the global socket is available, and if a daemon is answering at the
4951 * other side. If yes, error is returned.
4952 */
4953 static int check_existing_daemon(void)
4954 {
4955 /* Is there anybody out there ? */
4956 if (lttng_session_daemon_alive()) {
4957 return -EEXIST;
4958 }
4959
4960 return 0;
4961 }
4962
4963 /*
4964 * Set the tracing group gid onto the client socket.
4965 *
4966 * Race window between mkdir and chown is OK because we are going from more
4967 * permissive (root.root) to less permissive (root.tracing).
4968 */
4969 static int set_permissions(char *rundir)
4970 {
4971 int ret;
4972 gid_t gid;
4973
4974 gid = utils_get_group_id(tracing_group_name);
4975
4976 /* Set lttng run dir */
4977 ret = chown(rundir, 0, gid);
4978 if (ret < 0) {
4979 ERR("Unable to set group on %s", rundir);
4980 PERROR("chown");
4981 }
4982
4983 /*
4984 * Ensure all applications and tracing group can search the run
4985 * dir. Allow everyone to read the directory, since it does not
4986 * buy us anything to hide its content.
4987 */
4988 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4989 if (ret < 0) {
4990 ERR("Unable to set permissions on %s", rundir);
4991 PERROR("chmod");
4992 }
4993
4994 /* lttng client socket path */
4995 ret = chown(client_unix_sock_path, 0, gid);
4996 if (ret < 0) {
4997 ERR("Unable to set group on %s", client_unix_sock_path);
4998 PERROR("chown");
4999 }
5000
5001 /* kconsumer error socket path */
5002 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5003 if (ret < 0) {
5004 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5005 PERROR("chown");
5006 }
5007
5008 /* 64-bit ustconsumer error socket path */
5009 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5010 if (ret < 0) {
5011 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5012 PERROR("chown");
5013 }
5014
5015 /* 32-bit ustconsumer compat32 error socket path */
5016 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5017 if (ret < 0) {
5018 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5019 PERROR("chown");
5020 }
5021
5022 DBG("All permissions are set");
5023
5024 return ret;
5025 }
5026
5027 /*
5028 * Create the lttng run directory needed for all global sockets and pipe.
5029 */
5030 static int create_lttng_rundir(const char *rundir)
5031 {
5032 int ret;
5033
5034 DBG3("Creating LTTng run directory: %s", rundir);
5035
5036 ret = mkdir(rundir, S_IRWXU);
5037 if (ret < 0) {
5038 if (errno != EEXIST) {
5039 ERR("Unable to create %s", rundir);
5040 goto error;
5041 } else {
5042 ret = 0;
5043 }
5044 }
5045
5046 error:
5047 return ret;
5048 }
5049
5050 /*
5051 * Setup sockets and directory needed by the kconsumerd communication with the
5052 * session daemon.
5053 */
5054 static int set_consumer_sockets(struct consumer_data *consumer_data,
5055 const char *rundir)
5056 {
5057 int ret;
5058 char path[PATH_MAX];
5059
5060 switch (consumer_data->type) {
5061 case LTTNG_CONSUMER_KERNEL:
5062 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5063 break;
5064 case LTTNG_CONSUMER64_UST:
5065 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5066 break;
5067 case LTTNG_CONSUMER32_UST:
5068 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5069 break;
5070 default:
5071 ERR("Consumer type unknown");
5072 ret = -EINVAL;
5073 goto error;
5074 }
5075
5076 DBG2("Creating consumer directory: %s", path);
5077
5078 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5079 if (ret < 0) {
5080 if (errno != EEXIST) {
5081 PERROR("mkdir");
5082 ERR("Failed to create %s", path);
5083 goto error;
5084 }
5085 ret = -1;
5086 }
5087 if (is_root) {
5088 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5089 if (ret < 0) {
5090 ERR("Unable to set group on %s", path);
5091 PERROR("chown");
5092 goto error;
5093 }
5094 }
5095
5096 /* Create the kconsumerd error unix socket */
5097 consumer_data->err_sock =
5098 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5099 if (consumer_data->err_sock < 0) {
5100 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5101 ret = -1;
5102 goto error;
5103 }
5104
5105 /*
5106 * Set the CLOEXEC flag. Return code is useless because either way, the
5107 * show must go on.
5108 */
5109 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5110 if (ret < 0) {
5111 PERROR("utils_set_fd_cloexec");
5112 /* continue anyway */
5113 }
5114
5115 /* File permission MUST be 660 */
5116 ret = chmod(consumer_data->err_unix_sock_path,
5117 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5118 if (ret < 0) {
5119 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5120 PERROR("chmod");
5121 goto error;
5122 }
5123
5124 error:
5125 return ret;
5126 }
5127
5128 /*
5129 * Signal handler for the daemon
5130 *
5131 * Simply stop all worker threads, leaving main() return gracefully after
5132 * joining all threads and calling cleanup().
5133 */
5134 static void sighandler(int sig)
5135 {
5136 switch (sig) {
5137 case SIGPIPE:
5138 DBG("SIGPIPE caught");
5139 return;
5140 case SIGINT:
5141 DBG("SIGINT caught");
5142 stop_threads();
5143 break;
5144 case SIGTERM:
5145 DBG("SIGTERM caught");
5146 stop_threads();
5147 break;
5148 case SIGUSR1:
5149 CMM_STORE_SHARED(recv_child_signal, 1);
5150 break;
5151 default:
5152 break;
5153 }
5154 }
5155
5156 /*
5157 * Setup signal handler for :
5158 * SIGINT, SIGTERM, SIGPIPE
5159 */
5160 static int set_signal_handler(void)
5161 {
5162 int ret = 0;
5163 struct sigaction sa;
5164 sigset_t sigset;
5165
5166 if ((ret = sigemptyset(&sigset)) < 0) {
5167 PERROR("sigemptyset");
5168 return ret;
5169 }
5170
5171 sa.sa_handler = sighandler;
5172 sa.sa_mask = sigset;
5173 sa.sa_flags = 0;
5174 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5175 PERROR("sigaction");
5176 return ret;
5177 }
5178
5179 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5180 PERROR("sigaction");
5181 return ret;
5182 }
5183
5184 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5185 PERROR("sigaction");
5186 return ret;
5187 }
5188
5189 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5190 PERROR("sigaction");
5191 return ret;
5192 }
5193
5194 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5195
5196 return ret;
5197 }
5198
5199 /*
5200 * Set open files limit to unlimited. This daemon can open a large number of
5201 * file descriptors in order to consumer multiple kernel traces.
5202 */
5203 static void set_ulimit(void)
5204 {
5205 int ret;
5206 struct rlimit lim;
5207
5208 /* The kernel does not allowed an infinite limit for open files */
5209 lim.rlim_cur = 65535;
5210 lim.rlim_max = 65535;
5211
5212 ret = setrlimit(RLIMIT_NOFILE, &lim);
5213 if (ret < 0) {
5214 PERROR("failed to set open files limit");
5215 }
5216 }
5217
5218 /*
5219 * Write pidfile using the rundir and opt_pidfile.
5220 */
5221 static int write_pidfile(void)
5222 {
5223 int ret;
5224 char pidfile_path[PATH_MAX];
5225
5226 assert(rundir);
5227
5228 if (opt_pidfile) {
5229 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5230 } else {
5231 /* Build pidfile path from rundir and opt_pidfile. */
5232 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5233 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5234 if (ret < 0) {
5235 PERROR("snprintf pidfile path");
5236 goto error;
5237 }
5238 }
5239
5240 /*
5241 * Create pid file in rundir.
5242 */
5243 ret = utils_create_pid_file(getpid(), pidfile_path);
5244 error:
5245 return ret;
5246 }
5247
5248 /*
5249 * Create lockfile using the rundir and return its fd.
5250 */
5251 static int create_lockfile(void)
5252 {
5253 int ret;
5254 char lockfile_path[PATH_MAX];
5255
5256 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5257 if (ret < 0) {
5258 goto error;
5259 }
5260
5261 ret = utils_create_lock_file(lockfile_path);
5262 error:
5263 return ret;
5264 }
5265
5266 /*
5267 * Write agent TCP port using the rundir.
5268 */
5269 static int write_agent_port(void)
5270 {
5271 int ret;
5272 char path[PATH_MAX];
5273
5274 assert(rundir);
5275
5276 ret = snprintf(path, sizeof(path), "%s/"
5277 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5278 if (ret < 0) {
5279 PERROR("snprintf agent port path");
5280 goto error;
5281 }
5282
5283 /*
5284 * Create TCP agent port file in rundir.
5285 */
5286 ret = utils_create_pid_file(agent_tcp_port, path);
5287
5288 error:
5289 return ret;
5290 }
5291
5292 /*
5293 * main
5294 */
5295 int main(int argc, char **argv)
5296 {
5297 int ret = 0, retval = 0;
5298 void *status;
5299 const char *home_path, *env_app_timeout;
5300
5301 /* Initialize agent apps ht global variable */
5302 agent_apps_ht_by_sock = NULL;
5303
5304 init_kernel_workarounds();
5305
5306 rcu_register_thread();
5307
5308 if (set_signal_handler()) {
5309 retval = -1;
5310 goto exit_set_signal_handler;
5311 }
5312
5313 setup_consumerd_path();
5314
5315 page_size = sysconf(_SC_PAGESIZE);
5316 if (page_size < 0) {
5317 PERROR("sysconf _SC_PAGESIZE");
5318 page_size = LONG_MAX;
5319 WARN("Fallback page size to %ld", page_size);
5320 }
5321
5322 /*
5323 * Parse arguments and load the daemon configuration file.
5324 *
5325 * We have an exit_options exit path to free memory reserved by
5326 * set_options. This is needed because the rest of sessiond_cleanup()
5327 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5328 * depends on set_options.
5329 */
5330 progname = argv[0];
5331 if (set_options(argc, argv)) {
5332 retval = -1;
5333 goto exit_options;
5334 }
5335
5336 /* Daemonize */
5337 if (opt_daemon || opt_background) {
5338 int i;
5339
5340 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5341 !opt_background);
5342 if (ret < 0) {
5343 retval = -1;
5344 goto exit_options;
5345 }
5346
5347 /*
5348 * We are in the child. Make sure all other file descriptors are
5349 * closed, in case we are called with more opened file
5350 * descriptors than the standard ones.
5351 */
5352 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5353 (void) close(i);
5354 }
5355 }
5356
5357 /*
5358 * Starting from here, we can create threads. This needs to be after
5359 * lttng_daemonize due to RCU.
5360 */
5361
5362 /*
5363 * Initialize the health check subsystem. This call should set the
5364 * appropriate time values.
5365 */
5366 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5367 if (!health_sessiond) {
5368 PERROR("health_app_create error");
5369 retval = -1;
5370 goto exit_health_sessiond_cleanup;
5371 }
5372
5373 if (init_ht_cleanup_quit_pipe()) {
5374 retval = -1;
5375 goto exit_ht_cleanup_quit_pipe;
5376 }
5377
5378 /* Setup the thread ht_cleanup communication pipe. */
5379 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5380 retval = -1;
5381 goto exit_ht_cleanup_pipe;
5382 }
5383
5384 /* Set up max poll set size */
5385 if (lttng_poll_set_max_size()) {
5386 retval = -1;
5387 goto exit_set_max_size;
5388 }
5389
5390 /* Create thread to clean up RCU hash tables */
5391 ret = pthread_create(&ht_cleanup_thread, NULL,
5392 thread_ht_cleanup, (void *) NULL);
5393 if (ret) {
5394 errno = ret;
5395 PERROR("pthread_create ht_cleanup");
5396 retval = -1;
5397 goto exit_ht_cleanup;
5398 }
5399
5400 /* Create thread quit pipe */
5401 if (init_thread_quit_pipe()) {
5402 retval = -1;
5403 goto exit_init_data;
5404 }
5405
5406 /* Check if daemon is UID = 0 */
5407 is_root = !getuid();
5408
5409 if (is_root) {
5410 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5411 if (!rundir) {
5412 retval = -1;
5413 goto exit_init_data;
5414 }
5415
5416 /* Create global run dir with root access */
5417 if (create_lttng_rundir(rundir)) {
5418 retval = -1;
5419 goto exit_init_data;
5420 }
5421
5422 if (strlen(apps_unix_sock_path) == 0) {
5423 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5424 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5425 if (ret < 0) {
5426 retval = -1;
5427 goto exit_init_data;
5428 }
5429 }
5430
5431 if (strlen(client_unix_sock_path) == 0) {
5432 ret = snprintf(client_unix_sock_path, PATH_MAX,
5433 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5434 if (ret < 0) {
5435 retval = -1;
5436 goto exit_init_data;
5437 }
5438 }
5439
5440 /* Set global SHM for ust */
5441 if (strlen(wait_shm_path) == 0) {
5442 ret = snprintf(wait_shm_path, PATH_MAX,
5443 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5444 if (ret < 0) {
5445 retval = -1;
5446 goto exit_init_data;
5447 }
5448 }
5449
5450 if (strlen(health_unix_sock_path) == 0) {
5451 ret = snprintf(health_unix_sock_path,
5452 sizeof(health_unix_sock_path),
5453 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5454 if (ret < 0) {
5455 retval = -1;
5456 goto exit_init_data;
5457 }
5458 }
5459
5460 /* Setup kernel consumerd path */
5461 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5462 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5463 if (ret < 0) {
5464 retval = -1;
5465 goto exit_init_data;
5466 }
5467 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5468 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5469 if (ret < 0) {
5470 retval = -1;
5471 goto exit_init_data;
5472 }
5473
5474 DBG2("Kernel consumer err path: %s",
5475 kconsumer_data.err_unix_sock_path);
5476 DBG2("Kernel consumer cmd path: %s",
5477 kconsumer_data.cmd_unix_sock_path);
5478 } else {
5479 home_path = utils_get_home_dir();
5480 if (home_path == NULL) {
5481 /* TODO: Add --socket PATH option */
5482 ERR("Can't get HOME directory for sockets creation.");
5483 retval = -1;
5484 goto exit_init_data;
5485 }
5486
5487 /*
5488 * Create rundir from home path. This will create something like
5489 * $HOME/.lttng
5490 */
5491 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5492 if (ret < 0) {
5493 retval = -1;
5494 goto exit_init_data;
5495 }
5496
5497 if (create_lttng_rundir(rundir)) {
5498 retval = -1;
5499 goto exit_init_data;
5500 }
5501
5502 if (strlen(apps_unix_sock_path) == 0) {
5503 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5504 DEFAULT_HOME_APPS_UNIX_SOCK,
5505 home_path);
5506 if (ret < 0) {
5507 retval = -1;
5508 goto exit_init_data;
5509 }
5510 }
5511
5512 /* Set the cli tool unix socket path */
5513 if (strlen(client_unix_sock_path) == 0) {
5514 ret = snprintf(client_unix_sock_path, PATH_MAX,
5515 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5516 home_path);
5517 if (ret < 0) {
5518 retval = -1;
5519 goto exit_init_data;
5520 }
5521 }
5522
5523 /* Set global SHM for ust */
5524 if (strlen(wait_shm_path) == 0) {
5525 ret = snprintf(wait_shm_path, PATH_MAX,
5526 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5527 getuid());
5528 if (ret < 0) {
5529 retval = -1;
5530 goto exit_init_data;
5531 }
5532 }
5533
5534 /* Set health check Unix path */
5535 if (strlen(health_unix_sock_path) == 0) {
5536 ret = snprintf(health_unix_sock_path,
5537 sizeof(health_unix_sock_path),
5538 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5539 home_path);
5540 if (ret < 0) {
5541 retval = -1;
5542 goto exit_init_data;
5543 }
5544 }
5545 }
5546
5547 lockfile_fd = create_lockfile();
5548 if (lockfile_fd < 0) {
5549 retval = -1;
5550 goto exit_init_data;
5551 }
5552
5553 /* Set consumer initial state */
5554 kernel_consumerd_state = CONSUMER_STOPPED;
5555 ust_consumerd_state = CONSUMER_STOPPED;
5556
5557 DBG("Client socket path %s", client_unix_sock_path);
5558 DBG("Application socket path %s", apps_unix_sock_path);
5559 DBG("Application wait path %s", wait_shm_path);
5560 DBG("LTTng run directory path: %s", rundir);
5561
5562 /* 32 bits consumerd path setup */
5563 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5564 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5565 if (ret < 0) {
5566 PERROR("snprintf 32-bit consumer error socket path");
5567 retval = -1;
5568 goto exit_init_data;
5569 }
5570 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5571 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5572 if (ret < 0) {
5573 PERROR("snprintf 32-bit consumer command socket path");
5574 retval = -1;
5575 goto exit_init_data;
5576 }
5577
5578 DBG2("UST consumer 32 bits err path: %s",
5579 ustconsumer32_data.err_unix_sock_path);
5580 DBG2("UST consumer 32 bits cmd path: %s",
5581 ustconsumer32_data.cmd_unix_sock_path);
5582
5583 /* 64 bits consumerd path setup */
5584 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5585 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5586 if (ret < 0) {
5587 PERROR("snprintf 64-bit consumer error socket path");
5588 retval = -1;
5589 goto exit_init_data;
5590 }
5591 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5592 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5593 if (ret < 0) {
5594 PERROR("snprintf 64-bit consumer command socket path");
5595 retval = -1;
5596 goto exit_init_data;
5597 }
5598
5599 DBG2("UST consumer 64 bits err path: %s",
5600 ustconsumer64_data.err_unix_sock_path);
5601 DBG2("UST consumer 64 bits cmd path: %s",
5602 ustconsumer64_data.cmd_unix_sock_path);
5603
5604 /*
5605 * See if daemon already exist.
5606 */
5607 if (check_existing_daemon()) {
5608 ERR("Already running daemon.\n");
5609 /*
5610 * We do not goto exit because we must not cleanup()
5611 * because a daemon is already running.
5612 */
5613 retval = -1;
5614 goto exit_init_data;
5615 }
5616
5617 /*
5618 * Init UST app hash table. Alloc hash table before this point since
5619 * cleanup() can get called after that point.
5620 */
5621 if (ust_app_ht_alloc()) {
5622 ERR("Failed to allocate UST app hash table");
5623 retval = -1;
5624 goto exit_init_data;
5625 }
5626
5627 /*
5628 * Initialize agent app hash table. We allocate the hash table here
5629 * since cleanup() can get called after this point.
5630 */
5631 if (agent_app_ht_alloc()) {
5632 ERR("Failed to allocate Agent app hash table");
5633 retval = -1;
5634 goto exit_init_data;
5635 }
5636
5637 /*
5638 * These actions must be executed as root. We do that *after* setting up
5639 * the sockets path because we MUST make the check for another daemon using
5640 * those paths *before* trying to set the kernel consumer sockets and init
5641 * kernel tracer.
5642 */
5643 if (is_root) {
5644 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5645 retval = -1;
5646 goto exit_init_data;
5647 }
5648
5649 /* Setup kernel tracer */
5650 if (!opt_no_kernel) {
5651 init_kernel_tracer();
5652 if (kernel_tracer_fd >= 0) {
5653 ret = syscall_init_table();
5654 if (ret < 0) {
5655 ERR("Unable to populate syscall table. "
5656 "Syscall tracing won't work "
5657 "for this session daemon.");
5658 }
5659 }
5660 }
5661
5662 /* Set ulimit for open files */
5663 set_ulimit();
5664 }
5665 /* init lttng_fd tracking must be done after set_ulimit. */
5666 lttng_fd_init();
5667
5668 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5669 retval = -1;
5670 goto exit_init_data;
5671 }
5672
5673 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5674 retval = -1;
5675 goto exit_init_data;
5676 }
5677
5678 /* Setup the needed unix socket */
5679 if (init_daemon_socket()) {
5680 retval = -1;
5681 goto exit_init_data;
5682 }
5683
5684 /* Set credentials to socket */
5685 if (is_root && set_permissions(rundir)) {
5686 retval = -1;
5687 goto exit_init_data;
5688 }
5689
5690 /* Get parent pid if -S, --sig-parent is specified. */
5691 if (opt_sig_parent) {
5692 ppid = getppid();
5693 }
5694
5695 /* Setup the kernel pipe for waking up the kernel thread */
5696 if (is_root && !opt_no_kernel) {
5697 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5698 retval = -1;
5699 goto exit_init_data;
5700 }
5701 }
5702
5703 /* Setup the thread apps communication pipe. */
5704 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5705 retval = -1;
5706 goto exit_init_data;
5707 }
5708
5709 /* Setup the thread apps notify communication pipe. */
5710 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5711 retval = -1;
5712 goto exit_init_data;
5713 }
5714
5715 /* Initialize global buffer per UID and PID registry. */
5716 buffer_reg_init_uid_registry();
5717 buffer_reg_init_pid_registry();
5718
5719 /* Init UST command queue. */
5720 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5721
5722 /*
5723 * Get session list pointer. This pointer MUST NOT be free'd. This list
5724 * is statically declared in session.c
5725 */
5726 session_list_ptr = session_get_list();
5727
5728 cmd_init();
5729
5730 /* Check for the application socket timeout env variable. */
5731 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5732 if (env_app_timeout) {
5733 app_socket_timeout = atoi(env_app_timeout);
5734 } else {
5735 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5736 }
5737
5738 ret = write_pidfile();
5739 if (ret) {
5740 ERR("Error in write_pidfile");
5741 retval = -1;
5742 goto exit_init_data;
5743 }
5744 ret = write_agent_port();
5745 if (ret) {
5746 ERR("Error in write_agent_port");
5747 retval = -1;
5748 goto exit_init_data;
5749 }
5750
5751 /* Initialize communication library */
5752 lttcomm_init();
5753 /* Initialize TCP timeout values */
5754 lttcomm_inet_init();
5755
5756 if (load_session_init_data(&load_info) < 0) {
5757 retval = -1;
5758 goto exit_init_data;
5759 }
5760 load_info->path = opt_load_session_path;
5761
5762 /* Create health-check thread */
5763 ret = pthread_create(&health_thread, NULL,
5764 thread_manage_health, (void *) NULL);
5765 if (ret) {
5766 errno = ret;
5767 PERROR("pthread_create health");
5768 retval = -1;
5769 goto exit_health;
5770 }
5771
5772 /* Create thread to manage the client socket */
5773 ret = pthread_create(&client_thread, NULL,
5774 thread_manage_clients, (void *) NULL);
5775 if (ret) {
5776 errno = ret;
5777 PERROR("pthread_create clients");
5778 retval = -1;
5779 goto exit_client;
5780 }
5781
5782 /* Create thread to dispatch registration */
5783 ret = pthread_create(&dispatch_thread, NULL,
5784 thread_dispatch_ust_registration, (void *) NULL);
5785 if (ret) {
5786 errno = ret;
5787 PERROR("pthread_create dispatch");
5788 retval = -1;
5789 goto exit_dispatch;
5790 }
5791
5792 /* Create thread to manage application registration. */
5793 ret = pthread_create(&reg_apps_thread, NULL,
5794 thread_registration_apps, (void *) NULL);
5795 if (ret) {
5796 errno = ret;
5797 PERROR("pthread_create registration");
5798 retval = -1;
5799 goto exit_reg_apps;
5800 }
5801
5802 /* Create thread to manage application socket */
5803 ret = pthread_create(&apps_thread, NULL,
5804 thread_manage_apps, (void *) NULL);
5805 if (ret) {
5806 errno = ret;
5807 PERROR("pthread_create apps");
5808 retval = -1;
5809 goto exit_apps;
5810 }
5811
5812 /* Create thread to manage application notify socket */
5813 ret = pthread_create(&apps_notify_thread, NULL,
5814 ust_thread_manage_notify, (void *) NULL);
5815 if (ret) {
5816 errno = ret;
5817 PERROR("pthread_create notify");
5818 retval = -1;
5819 goto exit_apps_notify;
5820 }
5821
5822 /* Create agent registration thread. */
5823 ret = pthread_create(&agent_reg_thread, NULL,
5824 agent_thread_manage_registration, (void *) NULL);
5825 if (ret) {
5826 errno = ret;
5827 PERROR("pthread_create agent");
5828 retval = -1;
5829 goto exit_agent_reg;
5830 }
5831
5832 /* Don't start this thread if kernel tracing is not requested nor root */
5833 if (is_root && !opt_no_kernel) {
5834 /* Create kernel thread to manage kernel event */
5835 ret = pthread_create(&kernel_thread, NULL,
5836 thread_manage_kernel, (void *) NULL);
5837 if (ret) {
5838 errno = ret;
5839 PERROR("pthread_create kernel");
5840 retval = -1;
5841 goto exit_kernel;
5842 }
5843 }
5844
5845 /* Create session loading thread. */
5846 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5847 load_info);
5848 if (ret) {
5849 errno = ret;
5850 PERROR("pthread_create load_session_thread");
5851 retval = -1;
5852 goto exit_load_session;
5853 }
5854
5855 /*
5856 * This is where we start awaiting program completion (e.g. through
5857 * signal that asks threads to teardown).
5858 */
5859
5860 ret = pthread_join(load_session_thread, &status);
5861 if (ret) {
5862 errno = ret;
5863 PERROR("pthread_join load_session_thread");
5864 retval = -1;
5865 }
5866 exit_load_session:
5867
5868 if (is_root && !opt_no_kernel) {
5869 ret = pthread_join(kernel_thread, &status);
5870 if (ret) {
5871 errno = ret;
5872 PERROR("pthread_join");
5873 retval = -1;
5874 }
5875 }
5876 exit_kernel:
5877
5878 ret = pthread_join(agent_reg_thread, &status);
5879 if (ret) {
5880 errno = ret;
5881 PERROR("pthread_join agent");
5882 retval = -1;
5883 }
5884 exit_agent_reg:
5885
5886 ret = pthread_join(apps_notify_thread, &status);
5887 if (ret) {
5888 errno = ret;
5889 PERROR("pthread_join apps notify");
5890 retval = -1;
5891 }
5892 exit_apps_notify:
5893
5894 ret = pthread_join(apps_thread, &status);
5895 if (ret) {
5896 errno = ret;
5897 PERROR("pthread_join apps");
5898 retval = -1;
5899 }
5900 exit_apps:
5901
5902 ret = pthread_join(reg_apps_thread, &status);
5903 if (ret) {
5904 errno = ret;
5905 PERROR("pthread_join");
5906 retval = -1;
5907 }
5908 exit_reg_apps:
5909
5910 ret = pthread_join(dispatch_thread, &status);
5911 if (ret) {
5912 errno = ret;
5913 PERROR("pthread_join");
5914 retval = -1;
5915 }
5916 exit_dispatch:
5917
5918 ret = pthread_join(client_thread, &status);
5919 if (ret) {
5920 errno = ret;
5921 PERROR("pthread_join");
5922 retval = -1;
5923 }
5924 exit_client:
5925
5926 ret = pthread_join(health_thread, &status);
5927 if (ret) {
5928 errno = ret;
5929 PERROR("pthread_join health thread");
5930 retval = -1;
5931 }
5932 exit_health:
5933
5934 exit_init_data:
5935 /*
5936 * sessiond_cleanup() is called when no other thread is running, except
5937 * the ht_cleanup thread, which is needed to destroy the hash tables.
5938 */
5939 rcu_thread_online();
5940 sessiond_cleanup();
5941 rcu_thread_offline();
5942 rcu_unregister_thread();
5943
5944 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
5945 if (ret < 0) {
5946 ERR("write error on ht_cleanup quit pipe");
5947 retval = -1;
5948 }
5949
5950 ret = pthread_join(ht_cleanup_thread, &status);
5951 if (ret) {
5952 errno = ret;
5953 PERROR("pthread_join ht cleanup thread");
5954 retval = -1;
5955 }
5956 exit_ht_cleanup:
5957 exit_set_max_size:
5958
5959 utils_close_pipe(ht_cleanup_pipe);
5960 exit_ht_cleanup_pipe:
5961
5962 /*
5963 * Close the ht_cleanup quit pipe.
5964 */
5965 utils_close_pipe(ht_cleanup_quit_pipe);
5966 exit_ht_cleanup_quit_pipe:
5967
5968 health_app_destroy(health_sessiond);
5969 exit_health_sessiond_cleanup:
5970
5971 exit_options:
5972 sessiond_cleanup_options();
5973
5974 exit_set_signal_handler:
5975 if (!retval) {
5976 exit(EXIT_SUCCESS);
5977 } else {
5978 exit(EXIT_FAILURE);
5979 }
5980 }
This page took 0.19177 seconds and 5 git commands to generate.