Fix: sessiond consumer thread should register as RCU thread
[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 rcu_register_thread();
1216 rcu_thread_online();
1217
1218 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1219
1220 health_code_update();
1221
1222 /*
1223 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1224 * metadata_sock. Nothing more will be added to this poll set.
1225 */
1226 ret = sessiond_set_thread_pollset(&events, 3);
1227 if (ret < 0) {
1228 goto error_poll;
1229 }
1230
1231 /*
1232 * The error socket here is already in a listening state which was done
1233 * just before spawning this thread to avoid a race between the consumer
1234 * daemon exec trying to connect and the listen() call.
1235 */
1236 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1237 if (ret < 0) {
1238 goto error;
1239 }
1240
1241 health_code_update();
1242
1243 /* Infinite blocking call, waiting for transmission */
1244 restart:
1245 health_poll_entry();
1246
1247 if (testpoint(sessiond_thread_manage_consumer)) {
1248 goto error;
1249 }
1250
1251 ret = lttng_poll_wait(&events, -1);
1252 health_poll_exit();
1253 if (ret < 0) {
1254 /*
1255 * Restart interrupted system call.
1256 */
1257 if (errno == EINTR) {
1258 goto restart;
1259 }
1260 goto error;
1261 }
1262
1263 nb_fd = ret;
1264
1265 for (i = 0; i < nb_fd; i++) {
1266 /* Fetch once the poll data */
1267 revents = LTTNG_POLL_GETEV(&events, i);
1268 pollfd = LTTNG_POLL_GETFD(&events, i);
1269
1270 health_code_update();
1271
1272 if (!revents) {
1273 /* No activity for this FD (poll implementation). */
1274 continue;
1275 }
1276
1277 /* Thread quit pipe has been closed. Killing thread. */
1278 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1279 if (ret) {
1280 err = 0;
1281 goto exit;
1282 }
1283
1284 /* Event on the registration socket */
1285 if (pollfd == consumer_data->err_sock) {
1286 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1287 ERR("consumer err socket poll error");
1288 goto error;
1289 }
1290 }
1291 }
1292
1293 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1294 if (sock < 0) {
1295 goto error;
1296 }
1297
1298 /*
1299 * Set the CLOEXEC flag. Return code is useless because either way, the
1300 * show must go on.
1301 */
1302 (void) utils_set_fd_cloexec(sock);
1303
1304 health_code_update();
1305
1306 DBG2("Receiving code from consumer err_sock");
1307
1308 /* Getting status code from kconsumerd */
1309 ret = lttcomm_recv_unix_sock(sock, &code,
1310 sizeof(enum lttcomm_return_code));
1311 if (ret <= 0) {
1312 goto error;
1313 }
1314
1315 health_code_update();
1316 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1317 /* Connect both socket, command and metadata. */
1318 consumer_data->cmd_sock =
1319 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1320 consumer_data->metadata_fd =
1321 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1322 if (consumer_data->cmd_sock < 0
1323 || consumer_data->metadata_fd < 0) {
1324 PERROR("consumer connect cmd socket");
1325 /* On error, signal condition and quit. */
1326 signal_consumer_condition(consumer_data, -1);
1327 goto error;
1328 }
1329 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1330 /* Create metadata socket lock. */
1331 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1332 if (consumer_data->metadata_sock.lock == NULL) {
1333 PERROR("zmalloc pthread mutex");
1334 ret = -1;
1335 goto error;
1336 }
1337 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1338
1339 signal_consumer_condition(consumer_data, 1);
1340 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1341 DBG("Consumer metadata socket ready (fd: %d)",
1342 consumer_data->metadata_fd);
1343 } else {
1344 ERR("consumer error when waiting for SOCK_READY : %s",
1345 lttcomm_get_readable_code(-code));
1346 goto error;
1347 }
1348
1349 /* Remove the consumerd error sock since we've established a connexion */
1350 ret = lttng_poll_del(&events, consumer_data->err_sock);
1351 if (ret < 0) {
1352 goto error;
1353 }
1354
1355 /* Add new accepted error socket. */
1356 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1357 if (ret < 0) {
1358 goto error;
1359 }
1360
1361 /* Add metadata socket that is successfully connected. */
1362 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1363 LPOLLIN | LPOLLRDHUP);
1364 if (ret < 0) {
1365 goto error;
1366 }
1367
1368 health_code_update();
1369
1370 /* Infinite blocking call, waiting for transmission */
1371 restart_poll:
1372 while (1) {
1373 health_code_update();
1374
1375 /* Exit the thread because the thread quit pipe has been triggered. */
1376 if (should_quit) {
1377 /* Not a health error. */
1378 err = 0;
1379 goto exit;
1380 }
1381
1382 health_poll_entry();
1383 ret = lttng_poll_wait(&events, -1);
1384 health_poll_exit();
1385 if (ret < 0) {
1386 /*
1387 * Restart interrupted system call.
1388 */
1389 if (errno == EINTR) {
1390 goto restart_poll;
1391 }
1392 goto error;
1393 }
1394
1395 nb_fd = ret;
1396
1397 for (i = 0; i < nb_fd; i++) {
1398 /* Fetch once the poll data */
1399 revents = LTTNG_POLL_GETEV(&events, i);
1400 pollfd = LTTNG_POLL_GETFD(&events, i);
1401
1402 health_code_update();
1403
1404 if (!revents) {
1405 /* No activity for this FD (poll implementation). */
1406 continue;
1407 }
1408
1409 /*
1410 * Thread quit pipe has been triggered, flag that we should stop
1411 * but continue the current loop to handle potential data from
1412 * consumer.
1413 */
1414 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1415
1416 if (pollfd == sock) {
1417 /* Event on the consumerd socket */
1418 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1419 ERR("consumer err socket second poll error");
1420 goto error;
1421 }
1422 health_code_update();
1423 /* Wait for any kconsumerd error */
1424 ret = lttcomm_recv_unix_sock(sock, &code,
1425 sizeof(enum lttcomm_return_code));
1426 if (ret <= 0) {
1427 ERR("consumer closed the command socket");
1428 goto error;
1429 }
1430
1431 ERR("consumer return code : %s",
1432 lttcomm_get_readable_code(-code));
1433
1434 goto exit;
1435 } else if (pollfd == consumer_data->metadata_fd) {
1436 /* UST metadata requests */
1437 ret = ust_consumer_metadata_request(
1438 &consumer_data->metadata_sock);
1439 if (ret < 0) {
1440 ERR("Handling metadata request");
1441 goto error;
1442 }
1443 }
1444 /* No need for an else branch all FDs are tested prior. */
1445 }
1446 health_code_update();
1447 }
1448
1449 exit:
1450 error:
1451 /*
1452 * We lock here because we are about to close the sockets and some other
1453 * thread might be using them so get exclusive access which will abort all
1454 * other consumer command by other threads.
1455 */
1456 pthread_mutex_lock(&consumer_data->lock);
1457
1458 /* Immediately set the consumerd state to stopped */
1459 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1460 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1461 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1462 consumer_data->type == LTTNG_CONSUMER32_UST) {
1463 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1464 } else {
1465 /* Code flow error... */
1466 assert(0);
1467 }
1468
1469 if (consumer_data->err_sock >= 0) {
1470 ret = close(consumer_data->err_sock);
1471 if (ret) {
1472 PERROR("close");
1473 }
1474 consumer_data->err_sock = -1;
1475 }
1476 if (consumer_data->cmd_sock >= 0) {
1477 ret = close(consumer_data->cmd_sock);
1478 if (ret) {
1479 PERROR("close");
1480 }
1481 consumer_data->cmd_sock = -1;
1482 }
1483 if (consumer_data->metadata_sock.fd_ptr &&
1484 *consumer_data->metadata_sock.fd_ptr >= 0) {
1485 ret = close(*consumer_data->metadata_sock.fd_ptr);
1486 if (ret) {
1487 PERROR("close");
1488 }
1489 }
1490 if (sock >= 0) {
1491 ret = close(sock);
1492 if (ret) {
1493 PERROR("close");
1494 }
1495 }
1496
1497 unlink(consumer_data->err_unix_sock_path);
1498 unlink(consumer_data->cmd_unix_sock_path);
1499 consumer_data->pid = 0;
1500 pthread_mutex_unlock(&consumer_data->lock);
1501
1502 /* Cleanup metadata socket mutex. */
1503 if (consumer_data->metadata_sock.lock) {
1504 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1505 free(consumer_data->metadata_sock.lock);
1506 }
1507 lttng_poll_clean(&events);
1508 error_poll:
1509 if (err) {
1510 health_error();
1511 ERR("Health error occurred in %s", __func__);
1512 }
1513 health_unregister(health_sessiond);
1514 DBG("consumer thread cleanup completed");
1515
1516 rcu_thread_offline();
1517 rcu_unregister_thread();
1518
1519 return NULL;
1520 }
1521
1522 /*
1523 * This thread manage application communication.
1524 */
1525 static void *thread_manage_apps(void *data)
1526 {
1527 int i, ret, pollfd, err = -1;
1528 ssize_t size_ret;
1529 uint32_t revents, nb_fd;
1530 struct lttng_poll_event events;
1531
1532 DBG("[thread] Manage application started");
1533
1534 rcu_register_thread();
1535 rcu_thread_online();
1536
1537 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1538
1539 if (testpoint(sessiond_thread_manage_apps)) {
1540 goto error_testpoint;
1541 }
1542
1543 health_code_update();
1544
1545 ret = sessiond_set_thread_pollset(&events, 2);
1546 if (ret < 0) {
1547 goto error_poll_create;
1548 }
1549
1550 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1551 if (ret < 0) {
1552 goto error;
1553 }
1554
1555 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1556 goto error;
1557 }
1558
1559 health_code_update();
1560
1561 while (1) {
1562 DBG("Apps thread polling");
1563
1564 /* Inifinite blocking call, waiting for transmission */
1565 restart:
1566 health_poll_entry();
1567 ret = lttng_poll_wait(&events, -1);
1568 DBG("Apps thread return from poll on %d fds",
1569 LTTNG_POLL_GETNB(&events));
1570 health_poll_exit();
1571 if (ret < 0) {
1572 /*
1573 * Restart interrupted system call.
1574 */
1575 if (errno == EINTR) {
1576 goto restart;
1577 }
1578 goto error;
1579 }
1580
1581 nb_fd = ret;
1582
1583 for (i = 0; i < nb_fd; i++) {
1584 /* Fetch once the poll data */
1585 revents = LTTNG_POLL_GETEV(&events, i);
1586 pollfd = LTTNG_POLL_GETFD(&events, i);
1587
1588 health_code_update();
1589
1590 if (!revents) {
1591 /* No activity for this FD (poll implementation). */
1592 continue;
1593 }
1594
1595 /* Thread quit pipe has been closed. Killing thread. */
1596 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1597 if (ret) {
1598 err = 0;
1599 goto exit;
1600 }
1601
1602 /* Inspect the apps cmd pipe */
1603 if (pollfd == apps_cmd_pipe[0]) {
1604 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1605 ERR("Apps command pipe error");
1606 goto error;
1607 } else if (revents & LPOLLIN) {
1608 int sock;
1609
1610 /* Empty pipe */
1611 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1612 if (size_ret < sizeof(sock)) {
1613 PERROR("read apps cmd pipe");
1614 goto error;
1615 }
1616
1617 health_code_update();
1618
1619 /*
1620 * We only monitor the error events of the socket. This
1621 * thread does not handle any incoming data from UST
1622 * (POLLIN).
1623 */
1624 ret = lttng_poll_add(&events, sock,
1625 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1626 if (ret < 0) {
1627 goto error;
1628 }
1629
1630 DBG("Apps with sock %d added to poll set", sock);
1631 }
1632 } else {
1633 /*
1634 * At this point, we know that a registered application made
1635 * the event at poll_wait.
1636 */
1637 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1638 /* Removing from the poll set */
1639 ret = lttng_poll_del(&events, pollfd);
1640 if (ret < 0) {
1641 goto error;
1642 }
1643
1644 /* Socket closed on remote end. */
1645 ust_app_unregister(pollfd);
1646 }
1647 }
1648
1649 health_code_update();
1650 }
1651 }
1652
1653 exit:
1654 error:
1655 lttng_poll_clean(&events);
1656 error_poll_create:
1657 error_testpoint:
1658 utils_close_pipe(apps_cmd_pipe);
1659 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1660
1661 /*
1662 * We don't clean the UST app hash table here since already registered
1663 * applications can still be controlled so let them be until the session
1664 * daemon dies or the applications stop.
1665 */
1666
1667 if (err) {
1668 health_error();
1669 ERR("Health error occurred in %s", __func__);
1670 }
1671 health_unregister(health_sessiond);
1672 DBG("Application communication apps thread cleanup complete");
1673 rcu_thread_offline();
1674 rcu_unregister_thread();
1675 return NULL;
1676 }
1677
1678 /*
1679 * Send a socket to a thread This is called from the dispatch UST registration
1680 * thread once all sockets are set for the application.
1681 *
1682 * The sock value can be invalid, we don't really care, the thread will handle
1683 * it and make the necessary cleanup if so.
1684 *
1685 * On success, return 0 else a negative value being the errno message of the
1686 * write().
1687 */
1688 static int send_socket_to_thread(int fd, int sock)
1689 {
1690 ssize_t ret;
1691
1692 /*
1693 * It's possible that the FD is set as invalid with -1 concurrently just
1694 * before calling this function being a shutdown state of the thread.
1695 */
1696 if (fd < 0) {
1697 ret = -EBADF;
1698 goto error;
1699 }
1700
1701 ret = lttng_write(fd, &sock, sizeof(sock));
1702 if (ret < sizeof(sock)) {
1703 PERROR("write apps pipe %d", fd);
1704 if (ret < 0) {
1705 ret = -errno;
1706 }
1707 goto error;
1708 }
1709
1710 /* All good. Don't send back the write positive ret value. */
1711 ret = 0;
1712 error:
1713 return (int) ret;
1714 }
1715
1716 /*
1717 * Sanitize the wait queue of the dispatch registration thread meaning removing
1718 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1719 * notify socket is never received.
1720 */
1721 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1722 {
1723 int ret, nb_fd = 0, i;
1724 unsigned int fd_added = 0;
1725 struct lttng_poll_event events;
1726 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1727
1728 assert(wait_queue);
1729
1730 lttng_poll_init(&events);
1731
1732 /* Just skip everything for an empty queue. */
1733 if (!wait_queue->count) {
1734 goto end;
1735 }
1736
1737 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1738 if (ret < 0) {
1739 goto error_create;
1740 }
1741
1742 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1743 &wait_queue->head, head) {
1744 assert(wait_node->app);
1745 ret = lttng_poll_add(&events, wait_node->app->sock,
1746 LPOLLHUP | LPOLLERR);
1747 if (ret < 0) {
1748 goto error;
1749 }
1750
1751 fd_added = 1;
1752 }
1753
1754 if (!fd_added) {
1755 goto end;
1756 }
1757
1758 /*
1759 * Poll but don't block so we can quickly identify the faulty events and
1760 * clean them afterwards from the wait queue.
1761 */
1762 ret = lttng_poll_wait(&events, 0);
1763 if (ret < 0) {
1764 goto error;
1765 }
1766 nb_fd = ret;
1767
1768 for (i = 0; i < nb_fd; i++) {
1769 /* Get faulty FD. */
1770 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1771 int pollfd = LTTNG_POLL_GETFD(&events, i);
1772
1773 if (!revents) {
1774 /* No activity for this FD (poll implementation). */
1775 continue;
1776 }
1777
1778 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1779 &wait_queue->head, head) {
1780 if (pollfd == wait_node->app->sock &&
1781 (revents & (LPOLLHUP | LPOLLERR))) {
1782 cds_list_del(&wait_node->head);
1783 wait_queue->count--;
1784 ust_app_destroy(wait_node->app);
1785 free(wait_node);
1786 break;
1787 }
1788 }
1789 }
1790
1791 if (nb_fd > 0) {
1792 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1793 }
1794
1795 end:
1796 lttng_poll_clean(&events);
1797 return;
1798
1799 error:
1800 lttng_poll_clean(&events);
1801 error_create:
1802 ERR("Unable to sanitize wait queue");
1803 return;
1804 }
1805
1806 /*
1807 * Dispatch request from the registration threads to the application
1808 * communication thread.
1809 */
1810 static void *thread_dispatch_ust_registration(void *data)
1811 {
1812 int ret, err = -1;
1813 struct cds_wfcq_node *node;
1814 struct ust_command *ust_cmd = NULL;
1815 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1816 struct ust_reg_wait_queue wait_queue = {
1817 .count = 0,
1818 };
1819
1820 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1821
1822 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1823 goto error_testpoint;
1824 }
1825
1826 health_code_update();
1827
1828 CDS_INIT_LIST_HEAD(&wait_queue.head);
1829
1830 DBG("[thread] Dispatch UST command started");
1831
1832 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1833 health_code_update();
1834
1835 /* Atomically prepare the queue futex */
1836 futex_nto1_prepare(&ust_cmd_queue.futex);
1837
1838 do {
1839 struct ust_app *app = NULL;
1840 ust_cmd = NULL;
1841
1842 /*
1843 * Make sure we don't have node(s) that have hung up before receiving
1844 * the notify socket. This is to clean the list in order to avoid
1845 * memory leaks from notify socket that are never seen.
1846 */
1847 sanitize_wait_queue(&wait_queue);
1848
1849 health_code_update();
1850 /* Dequeue command for registration */
1851 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1852 if (node == NULL) {
1853 DBG("Woken up but nothing in the UST command queue");
1854 /* Continue thread execution */
1855 break;
1856 }
1857
1858 ust_cmd = caa_container_of(node, struct ust_command, node);
1859
1860 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1861 " gid:%d sock:%d name:%s (version %d.%d)",
1862 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1863 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1864 ust_cmd->sock, ust_cmd->reg_msg.name,
1865 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1866
1867 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1868 wait_node = zmalloc(sizeof(*wait_node));
1869 if (!wait_node) {
1870 PERROR("zmalloc wait_node dispatch");
1871 ret = close(ust_cmd->sock);
1872 if (ret < 0) {
1873 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1874 }
1875 lttng_fd_put(LTTNG_FD_APPS, 1);
1876 free(ust_cmd);
1877 goto error;
1878 }
1879 CDS_INIT_LIST_HEAD(&wait_node->head);
1880
1881 /* Create application object if socket is CMD. */
1882 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1883 ust_cmd->sock);
1884 if (!wait_node->app) {
1885 ret = close(ust_cmd->sock);
1886 if (ret < 0) {
1887 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1888 }
1889 lttng_fd_put(LTTNG_FD_APPS, 1);
1890 free(wait_node);
1891 free(ust_cmd);
1892 continue;
1893 }
1894 /*
1895 * Add application to the wait queue so we can set the notify
1896 * socket before putting this object in the global ht.
1897 */
1898 cds_list_add(&wait_node->head, &wait_queue.head);
1899 wait_queue.count++;
1900
1901 free(ust_cmd);
1902 /*
1903 * We have to continue here since we don't have the notify
1904 * socket and the application MUST be added to the hash table
1905 * only at that moment.
1906 */
1907 continue;
1908 } else {
1909 /*
1910 * Look for the application in the local wait queue and set the
1911 * notify socket if found.
1912 */
1913 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1914 &wait_queue.head, head) {
1915 health_code_update();
1916 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1917 wait_node->app->notify_sock = ust_cmd->sock;
1918 cds_list_del(&wait_node->head);
1919 wait_queue.count--;
1920 app = wait_node->app;
1921 free(wait_node);
1922 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1923 break;
1924 }
1925 }
1926
1927 /*
1928 * With no application at this stage the received socket is
1929 * basically useless so close it before we free the cmd data
1930 * structure for good.
1931 */
1932 if (!app) {
1933 ret = close(ust_cmd->sock);
1934 if (ret < 0) {
1935 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1936 }
1937 lttng_fd_put(LTTNG_FD_APPS, 1);
1938 }
1939 free(ust_cmd);
1940 }
1941
1942 if (app) {
1943 /*
1944 * @session_lock_list
1945 *
1946 * Lock the global session list so from the register up to the
1947 * registration done message, no thread can see the application
1948 * and change its state.
1949 */
1950 session_lock_list();
1951 rcu_read_lock();
1952
1953 /*
1954 * Add application to the global hash table. This needs to be
1955 * done before the update to the UST registry can locate the
1956 * application.
1957 */
1958 ust_app_add(app);
1959
1960 /* Set app version. This call will print an error if needed. */
1961 (void) ust_app_version(app);
1962
1963 /* Send notify socket through the notify pipe. */
1964 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1965 app->notify_sock);
1966 if (ret < 0) {
1967 rcu_read_unlock();
1968 session_unlock_list();
1969 /*
1970 * No notify thread, stop the UST tracing. However, this is
1971 * not an internal error of the this thread thus setting
1972 * the health error code to a normal exit.
1973 */
1974 err = 0;
1975 goto error;
1976 }
1977
1978 /*
1979 * Update newly registered application with the tracing
1980 * registry info already enabled information.
1981 */
1982 update_ust_app(app->sock);
1983
1984 /*
1985 * Don't care about return value. Let the manage apps threads
1986 * handle app unregistration upon socket close.
1987 */
1988 (void) ust_app_register_done(app->sock);
1989
1990 /*
1991 * Even if the application socket has been closed, send the app
1992 * to the thread and unregistration will take place at that
1993 * place.
1994 */
1995 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1996 if (ret < 0) {
1997 rcu_read_unlock();
1998 session_unlock_list();
1999 /*
2000 * No apps. thread, stop the UST tracing. However, this is
2001 * not an internal error of the this thread thus setting
2002 * the health error code to a normal exit.
2003 */
2004 err = 0;
2005 goto error;
2006 }
2007
2008 rcu_read_unlock();
2009 session_unlock_list();
2010 }
2011 } while (node != NULL);
2012
2013 health_poll_entry();
2014 /* Futex wait on queue. Blocking call on futex() */
2015 futex_nto1_wait(&ust_cmd_queue.futex);
2016 health_poll_exit();
2017 }
2018 /* Normal exit, no error */
2019 err = 0;
2020
2021 error:
2022 /* Clean up wait queue. */
2023 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2024 &wait_queue.head, head) {
2025 cds_list_del(&wait_node->head);
2026 wait_queue.count--;
2027 free(wait_node);
2028 }
2029
2030 error_testpoint:
2031 DBG("Dispatch thread dying");
2032 if (err) {
2033 health_error();
2034 ERR("Health error occurred in %s", __func__);
2035 }
2036 health_unregister(health_sessiond);
2037 return NULL;
2038 }
2039
2040 /*
2041 * This thread manage application registration.
2042 */
2043 static void *thread_registration_apps(void *data)
2044 {
2045 int sock = -1, i, ret, pollfd, err = -1;
2046 uint32_t revents, nb_fd;
2047 struct lttng_poll_event events;
2048 /*
2049 * Get allocated in this thread, enqueued to a global queue, dequeued and
2050 * freed in the manage apps thread.
2051 */
2052 struct ust_command *ust_cmd = NULL;
2053
2054 DBG("[thread] Manage application registration started");
2055
2056 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2057
2058 if (testpoint(sessiond_thread_registration_apps)) {
2059 goto error_testpoint;
2060 }
2061
2062 ret = lttcomm_listen_unix_sock(apps_sock);
2063 if (ret < 0) {
2064 goto error_listen;
2065 }
2066
2067 /*
2068 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2069 * more will be added to this poll set.
2070 */
2071 ret = sessiond_set_thread_pollset(&events, 2);
2072 if (ret < 0) {
2073 goto error_create_poll;
2074 }
2075
2076 /* Add the application registration socket */
2077 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2078 if (ret < 0) {
2079 goto error_poll_add;
2080 }
2081
2082 /* Notify all applications to register */
2083 ret = notify_ust_apps(1);
2084 if (ret < 0) {
2085 ERR("Failed to notify applications or create the wait shared memory.\n"
2086 "Execution continues but there might be problem for already\n"
2087 "running applications that wishes to register.");
2088 }
2089
2090 while (1) {
2091 DBG("Accepting application registration");
2092
2093 /* Inifinite blocking call, waiting for transmission */
2094 restart:
2095 health_poll_entry();
2096 ret = lttng_poll_wait(&events, -1);
2097 health_poll_exit();
2098 if (ret < 0) {
2099 /*
2100 * Restart interrupted system call.
2101 */
2102 if (errno == EINTR) {
2103 goto restart;
2104 }
2105 goto error;
2106 }
2107
2108 nb_fd = ret;
2109
2110 for (i = 0; i < nb_fd; i++) {
2111 health_code_update();
2112
2113 /* Fetch once the poll data */
2114 revents = LTTNG_POLL_GETEV(&events, i);
2115 pollfd = LTTNG_POLL_GETFD(&events, i);
2116
2117 if (!revents) {
2118 /* No activity for this FD (poll implementation). */
2119 continue;
2120 }
2121
2122 /* Thread quit pipe has been closed. Killing thread. */
2123 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2124 if (ret) {
2125 err = 0;
2126 goto exit;
2127 }
2128
2129 /* Event on the registration socket */
2130 if (pollfd == apps_sock) {
2131 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2132 ERR("Register apps socket poll error");
2133 goto error;
2134 } else if (revents & LPOLLIN) {
2135 sock = lttcomm_accept_unix_sock(apps_sock);
2136 if (sock < 0) {
2137 goto error;
2138 }
2139
2140 /*
2141 * Set socket timeout for both receiving and ending.
2142 * app_socket_timeout is in seconds, whereas
2143 * lttcomm_setsockopt_rcv_timeout and
2144 * lttcomm_setsockopt_snd_timeout expect msec as
2145 * parameter.
2146 */
2147 (void) lttcomm_setsockopt_rcv_timeout(sock,
2148 app_socket_timeout * 1000);
2149 (void) lttcomm_setsockopt_snd_timeout(sock,
2150 app_socket_timeout * 1000);
2151
2152 /*
2153 * Set the CLOEXEC flag. Return code is useless because
2154 * either way, the show must go on.
2155 */
2156 (void) utils_set_fd_cloexec(sock);
2157
2158 /* Create UST registration command for enqueuing */
2159 ust_cmd = zmalloc(sizeof(struct ust_command));
2160 if (ust_cmd == NULL) {
2161 PERROR("ust command zmalloc");
2162 ret = close(sock);
2163 if (ret) {
2164 PERROR("close");
2165 }
2166 goto error;
2167 }
2168
2169 /*
2170 * Using message-based transmissions to ensure we don't
2171 * have to deal with partially received messages.
2172 */
2173 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2174 if (ret < 0) {
2175 ERR("Exhausted file descriptors allowed for applications.");
2176 free(ust_cmd);
2177 ret = close(sock);
2178 if (ret) {
2179 PERROR("close");
2180 }
2181 sock = -1;
2182 continue;
2183 }
2184
2185 health_code_update();
2186 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2187 if (ret < 0) {
2188 free(ust_cmd);
2189 /* Close socket of the application. */
2190 ret = close(sock);
2191 if (ret) {
2192 PERROR("close");
2193 }
2194 lttng_fd_put(LTTNG_FD_APPS, 1);
2195 sock = -1;
2196 continue;
2197 }
2198 health_code_update();
2199
2200 ust_cmd->sock = sock;
2201 sock = -1;
2202
2203 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2204 " gid:%d sock:%d name:%s (version %d.%d)",
2205 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2206 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2207 ust_cmd->sock, ust_cmd->reg_msg.name,
2208 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2209
2210 /*
2211 * Lock free enqueue the registration request. The red pill
2212 * has been taken! This apps will be part of the *system*.
2213 */
2214 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2215
2216 /*
2217 * Wake the registration queue futex. Implicit memory
2218 * barrier with the exchange in cds_wfcq_enqueue.
2219 */
2220 futex_nto1_wake(&ust_cmd_queue.futex);
2221 }
2222 }
2223 }
2224 }
2225
2226 exit:
2227 error:
2228 /* Notify that the registration thread is gone */
2229 notify_ust_apps(0);
2230
2231 if (apps_sock >= 0) {
2232 ret = close(apps_sock);
2233 if (ret) {
2234 PERROR("close");
2235 }
2236 }
2237 if (sock >= 0) {
2238 ret = close(sock);
2239 if (ret) {
2240 PERROR("close");
2241 }
2242 lttng_fd_put(LTTNG_FD_APPS, 1);
2243 }
2244 unlink(apps_unix_sock_path);
2245
2246 error_poll_add:
2247 lttng_poll_clean(&events);
2248 error_listen:
2249 error_create_poll:
2250 error_testpoint:
2251 DBG("UST Registration thread cleanup complete");
2252 if (err) {
2253 health_error();
2254 ERR("Health error occurred in %s", __func__);
2255 }
2256 health_unregister(health_sessiond);
2257
2258 return NULL;
2259 }
2260
2261 /*
2262 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2263 * exec or it will fails.
2264 */
2265 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2266 {
2267 int ret, clock_ret;
2268 struct timespec timeout;
2269
2270 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2271 consumer_data->consumer_thread_is_ready = 0;
2272
2273 /* Setup pthread condition */
2274 ret = pthread_condattr_init(&consumer_data->condattr);
2275 if (ret) {
2276 errno = ret;
2277 PERROR("pthread_condattr_init consumer data");
2278 goto error;
2279 }
2280
2281 /*
2282 * Set the monotonic clock in order to make sure we DO NOT jump in time
2283 * between the clock_gettime() call and the timedwait call. See bug #324
2284 * for a more details and how we noticed it.
2285 */
2286 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2287 if (ret) {
2288 errno = ret;
2289 PERROR("pthread_condattr_setclock consumer data");
2290 goto error;
2291 }
2292
2293 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2294 if (ret) {
2295 errno = ret;
2296 PERROR("pthread_cond_init consumer data");
2297 goto error;
2298 }
2299
2300 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2301 consumer_data);
2302 if (ret) {
2303 errno = ret;
2304 PERROR("pthread_create consumer");
2305 ret = -1;
2306 goto error;
2307 }
2308
2309 /* We are about to wait on a pthread condition */
2310 pthread_mutex_lock(&consumer_data->cond_mutex);
2311
2312 /* Get time for sem_timedwait absolute timeout */
2313 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2314 /*
2315 * Set the timeout for the condition timed wait even if the clock gettime
2316 * call fails since we might loop on that call and we want to avoid to
2317 * increment the timeout too many times.
2318 */
2319 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2320
2321 /*
2322 * The following loop COULD be skipped in some conditions so this is why we
2323 * set ret to 0 in order to make sure at least one round of the loop is
2324 * done.
2325 */
2326 ret = 0;
2327
2328 /*
2329 * Loop until the condition is reached or when a timeout is reached. Note
2330 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2331 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2332 * possible. This loop does not take any chances and works with both of
2333 * them.
2334 */
2335 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2336 if (clock_ret < 0) {
2337 PERROR("clock_gettime spawn consumer");
2338 /* Infinite wait for the consumerd thread to be ready */
2339 ret = pthread_cond_wait(&consumer_data->cond,
2340 &consumer_data->cond_mutex);
2341 } else {
2342 ret = pthread_cond_timedwait(&consumer_data->cond,
2343 &consumer_data->cond_mutex, &timeout);
2344 }
2345 }
2346
2347 /* Release the pthread condition */
2348 pthread_mutex_unlock(&consumer_data->cond_mutex);
2349
2350 if (ret != 0) {
2351 errno = ret;
2352 if (ret == ETIMEDOUT) {
2353 int pth_ret;
2354
2355 /*
2356 * Call has timed out so we kill the kconsumerd_thread and return
2357 * an error.
2358 */
2359 ERR("Condition timed out. The consumer thread was never ready."
2360 " Killing it");
2361 pth_ret = pthread_cancel(consumer_data->thread);
2362 if (pth_ret < 0) {
2363 PERROR("pthread_cancel consumer thread");
2364 }
2365 } else {
2366 PERROR("pthread_cond_wait failed consumer thread");
2367 }
2368 /* Caller is expecting a negative value on failure. */
2369 ret = -1;
2370 goto error;
2371 }
2372
2373 pthread_mutex_lock(&consumer_data->pid_mutex);
2374 if (consumer_data->pid == 0) {
2375 ERR("Consumerd did not start");
2376 pthread_mutex_unlock(&consumer_data->pid_mutex);
2377 goto error;
2378 }
2379 pthread_mutex_unlock(&consumer_data->pid_mutex);
2380
2381 return 0;
2382
2383 error:
2384 return ret;
2385 }
2386
2387 /*
2388 * Join consumer thread
2389 */
2390 static int join_consumer_thread(struct consumer_data *consumer_data)
2391 {
2392 void *status;
2393
2394 /* Consumer pid must be a real one. */
2395 if (consumer_data->pid > 0) {
2396 int ret;
2397 ret = kill(consumer_data->pid, SIGTERM);
2398 if (ret) {
2399 PERROR("Error killing consumer daemon");
2400 return ret;
2401 }
2402 return pthread_join(consumer_data->thread, &status);
2403 } else {
2404 return 0;
2405 }
2406 }
2407
2408 /*
2409 * Fork and exec a consumer daemon (consumerd).
2410 *
2411 * Return pid if successful else -1.
2412 */
2413 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2414 {
2415 int ret;
2416 pid_t pid;
2417 const char *consumer_to_use;
2418 const char *verbosity;
2419 struct stat st;
2420
2421 DBG("Spawning consumerd");
2422
2423 pid = fork();
2424 if (pid == 0) {
2425 /*
2426 * Exec consumerd.
2427 */
2428 if (opt_verbose_consumer) {
2429 verbosity = "--verbose";
2430 } else if (lttng_opt_quiet) {
2431 verbosity = "--quiet";
2432 } else {
2433 verbosity = "";
2434 }
2435
2436 switch (consumer_data->type) {
2437 case LTTNG_CONSUMER_KERNEL:
2438 /*
2439 * Find out which consumerd to execute. We will first try the
2440 * 64-bit path, then the sessiond's installation directory, and
2441 * fallback on the 32-bit one,
2442 */
2443 DBG3("Looking for a kernel consumer at these locations:");
2444 DBG3(" 1) %s", consumerd64_bin);
2445 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2446 DBG3(" 3) %s", consumerd32_bin);
2447 if (stat(consumerd64_bin, &st) == 0) {
2448 DBG3("Found location #1");
2449 consumer_to_use = consumerd64_bin;
2450 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2451 DBG3("Found location #2");
2452 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2453 } else if (stat(consumerd32_bin, &st) == 0) {
2454 DBG3("Found location #3");
2455 consumer_to_use = consumerd32_bin;
2456 } else {
2457 DBG("Could not find any valid consumerd executable");
2458 ret = -EINVAL;
2459 break;
2460 }
2461 DBG("Using kernel consumer at: %s", consumer_to_use);
2462 ret = execl(consumer_to_use,
2463 "lttng-consumerd", verbosity, "-k",
2464 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2465 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2466 "--group", tracing_group_name,
2467 NULL);
2468 break;
2469 case LTTNG_CONSUMER64_UST:
2470 {
2471 char *tmpnew = NULL;
2472
2473 if (consumerd64_libdir[0] != '\0') {
2474 char *tmp;
2475 size_t tmplen;
2476
2477 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2478 if (!tmp) {
2479 tmp = "";
2480 }
2481 tmplen = strlen("LD_LIBRARY_PATH=")
2482 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2483 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2484 if (!tmpnew) {
2485 ret = -ENOMEM;
2486 goto error;
2487 }
2488 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2489 strcat(tmpnew, consumerd64_libdir);
2490 if (tmp[0] != '\0') {
2491 strcat(tmpnew, ":");
2492 strcat(tmpnew, tmp);
2493 }
2494 ret = putenv(tmpnew);
2495 if (ret) {
2496 ret = -errno;
2497 free(tmpnew);
2498 goto error;
2499 }
2500 }
2501 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2502 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2503 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2504 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2505 "--group", tracing_group_name,
2506 NULL);
2507 if (consumerd64_libdir[0] != '\0') {
2508 free(tmpnew);
2509 }
2510 break;
2511 }
2512 case LTTNG_CONSUMER32_UST:
2513 {
2514 char *tmpnew = NULL;
2515
2516 if (consumerd32_libdir[0] != '\0') {
2517 char *tmp;
2518 size_t tmplen;
2519
2520 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2521 if (!tmp) {
2522 tmp = "";
2523 }
2524 tmplen = strlen("LD_LIBRARY_PATH=")
2525 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2526 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2527 if (!tmpnew) {
2528 ret = -ENOMEM;
2529 goto error;
2530 }
2531 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2532 strcat(tmpnew, consumerd32_libdir);
2533 if (tmp[0] != '\0') {
2534 strcat(tmpnew, ":");
2535 strcat(tmpnew, tmp);
2536 }
2537 ret = putenv(tmpnew);
2538 if (ret) {
2539 ret = -errno;
2540 free(tmpnew);
2541 goto error;
2542 }
2543 }
2544 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2545 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2546 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2547 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2548 "--group", tracing_group_name,
2549 NULL);
2550 if (consumerd32_libdir[0] != '\0') {
2551 free(tmpnew);
2552 }
2553 break;
2554 }
2555 default:
2556 PERROR("unknown consumer type");
2557 exit(EXIT_FAILURE);
2558 }
2559 if (errno != 0) {
2560 PERROR("Consumer execl()");
2561 }
2562 /* Reaching this point, we got a failure on our execl(). */
2563 exit(EXIT_FAILURE);
2564 } else if (pid > 0) {
2565 ret = pid;
2566 } else {
2567 PERROR("start consumer fork");
2568 ret = -errno;
2569 }
2570 error:
2571 return ret;
2572 }
2573
2574 /*
2575 * Spawn the consumerd daemon and session daemon thread.
2576 */
2577 static int start_consumerd(struct consumer_data *consumer_data)
2578 {
2579 int ret;
2580
2581 /*
2582 * Set the listen() state on the socket since there is a possible race
2583 * between the exec() of the consumer daemon and this call if place in the
2584 * consumer thread. See bug #366 for more details.
2585 */
2586 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2587 if (ret < 0) {
2588 goto error;
2589 }
2590
2591 pthread_mutex_lock(&consumer_data->pid_mutex);
2592 if (consumer_data->pid != 0) {
2593 pthread_mutex_unlock(&consumer_data->pid_mutex);
2594 goto end;
2595 }
2596
2597 ret = spawn_consumerd(consumer_data);
2598 if (ret < 0) {
2599 ERR("Spawning consumerd failed");
2600 pthread_mutex_unlock(&consumer_data->pid_mutex);
2601 goto error;
2602 }
2603
2604 /* Setting up the consumer_data pid */
2605 consumer_data->pid = ret;
2606 DBG2("Consumer pid %d", consumer_data->pid);
2607 pthread_mutex_unlock(&consumer_data->pid_mutex);
2608
2609 DBG2("Spawning consumer control thread");
2610 ret = spawn_consumer_thread(consumer_data);
2611 if (ret < 0) {
2612 ERR("Fatal error spawning consumer control thread");
2613 goto error;
2614 }
2615
2616 end:
2617 return 0;
2618
2619 error:
2620 /* Cleanup already created sockets on error. */
2621 if (consumer_data->err_sock >= 0) {
2622 int err;
2623
2624 err = close(consumer_data->err_sock);
2625 if (err < 0) {
2626 PERROR("close consumer data error socket");
2627 }
2628 }
2629 return ret;
2630 }
2631
2632 /*
2633 * Setup necessary data for kernel tracer action.
2634 */
2635 static int init_kernel_tracer(void)
2636 {
2637 int ret;
2638
2639 /* Modprobe lttng kernel modules */
2640 ret = modprobe_lttng_control();
2641 if (ret < 0) {
2642 goto error;
2643 }
2644
2645 /* Open debugfs lttng */
2646 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2647 if (kernel_tracer_fd < 0) {
2648 DBG("Failed to open %s", module_proc_lttng);
2649 ret = -1;
2650 goto error_open;
2651 }
2652
2653 /* Validate kernel version */
2654 ret = kernel_validate_version(kernel_tracer_fd);
2655 if (ret < 0) {
2656 goto error_version;
2657 }
2658
2659 ret = modprobe_lttng_data();
2660 if (ret < 0) {
2661 goto error_modules;
2662 }
2663
2664 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2665 return 0;
2666
2667 error_version:
2668 modprobe_remove_lttng_control();
2669 ret = close(kernel_tracer_fd);
2670 if (ret) {
2671 PERROR("close");
2672 }
2673 kernel_tracer_fd = -1;
2674 return LTTNG_ERR_KERN_VERSION;
2675
2676 error_modules:
2677 ret = close(kernel_tracer_fd);
2678 if (ret) {
2679 PERROR("close");
2680 }
2681
2682 error_open:
2683 modprobe_remove_lttng_control();
2684
2685 error:
2686 WARN("No kernel tracer available");
2687 kernel_tracer_fd = -1;
2688 if (!is_root) {
2689 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2690 } else {
2691 return LTTNG_ERR_KERN_NA;
2692 }
2693 }
2694
2695
2696 /*
2697 * Copy consumer output from the tracing session to the domain session. The
2698 * function also applies the right modification on a per domain basis for the
2699 * trace files destination directory.
2700 *
2701 * Should *NOT* be called with RCU read-side lock held.
2702 */
2703 static int copy_session_consumer(int domain, struct ltt_session *session)
2704 {
2705 int ret;
2706 const char *dir_name;
2707 struct consumer_output *consumer;
2708
2709 assert(session);
2710 assert(session->consumer);
2711
2712 switch (domain) {
2713 case LTTNG_DOMAIN_KERNEL:
2714 DBG3("Copying tracing session consumer output in kernel session");
2715 /*
2716 * XXX: We should audit the session creation and what this function
2717 * does "extra" in order to avoid a destroy since this function is used
2718 * in the domain session creation (kernel and ust) only. Same for UST
2719 * domain.
2720 */
2721 if (session->kernel_session->consumer) {
2722 consumer_output_put(session->kernel_session->consumer);
2723 }
2724 session->kernel_session->consumer =
2725 consumer_copy_output(session->consumer);
2726 /* Ease our life a bit for the next part */
2727 consumer = session->kernel_session->consumer;
2728 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2729 break;
2730 case LTTNG_DOMAIN_JUL:
2731 case LTTNG_DOMAIN_LOG4J:
2732 case LTTNG_DOMAIN_PYTHON:
2733 case LTTNG_DOMAIN_UST:
2734 DBG3("Copying tracing session consumer output in UST session");
2735 if (session->ust_session->consumer) {
2736 consumer_output_put(session->ust_session->consumer);
2737 }
2738 session->ust_session->consumer =
2739 consumer_copy_output(session->consumer);
2740 /* Ease our life a bit for the next part */
2741 consumer = session->ust_session->consumer;
2742 dir_name = DEFAULT_UST_TRACE_DIR;
2743 break;
2744 default:
2745 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2746 goto error;
2747 }
2748
2749 /* Append correct directory to subdir */
2750 strncat(consumer->subdir, dir_name,
2751 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2752 DBG3("Copy session consumer subdir %s", consumer->subdir);
2753
2754 ret = LTTNG_OK;
2755
2756 error:
2757 return ret;
2758 }
2759
2760 /*
2761 * Create an UST session and add it to the session ust list.
2762 *
2763 * Should *NOT* be called with RCU read-side lock held.
2764 */
2765 static int create_ust_session(struct ltt_session *session,
2766 struct lttng_domain *domain)
2767 {
2768 int ret;
2769 struct ltt_ust_session *lus = NULL;
2770
2771 assert(session);
2772 assert(domain);
2773 assert(session->consumer);
2774
2775 switch (domain->type) {
2776 case LTTNG_DOMAIN_JUL:
2777 case LTTNG_DOMAIN_LOG4J:
2778 case LTTNG_DOMAIN_PYTHON:
2779 case LTTNG_DOMAIN_UST:
2780 break;
2781 default:
2782 ERR("Unknown UST domain on create session %d", domain->type);
2783 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2784 goto error;
2785 }
2786
2787 DBG("Creating UST session");
2788
2789 lus = trace_ust_create_session(session->id);
2790 if (lus == NULL) {
2791 ret = LTTNG_ERR_UST_SESS_FAIL;
2792 goto error;
2793 }
2794
2795 lus->uid = session->uid;
2796 lus->gid = session->gid;
2797 lus->output_traces = session->output_traces;
2798 lus->snapshot_mode = session->snapshot_mode;
2799 lus->live_timer_interval = session->live_timer;
2800 session->ust_session = lus;
2801 if (session->shm_path[0]) {
2802 strncpy(lus->root_shm_path, session->shm_path,
2803 sizeof(lus->root_shm_path));
2804 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2805 strncpy(lus->shm_path, session->shm_path,
2806 sizeof(lus->shm_path));
2807 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2808 strncat(lus->shm_path, "/ust",
2809 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2810 }
2811 /* Copy session output to the newly created UST session */
2812 ret = copy_session_consumer(domain->type, session);
2813 if (ret != LTTNG_OK) {
2814 goto error;
2815 }
2816
2817 return LTTNG_OK;
2818
2819 error:
2820 free(lus);
2821 session->ust_session = NULL;
2822 return ret;
2823 }
2824
2825 /*
2826 * Create a kernel tracer session then create the default channel.
2827 */
2828 static int create_kernel_session(struct ltt_session *session)
2829 {
2830 int ret;
2831
2832 DBG("Creating kernel session");
2833
2834 ret = kernel_create_session(session, kernel_tracer_fd);
2835 if (ret < 0) {
2836 ret = LTTNG_ERR_KERN_SESS_FAIL;
2837 goto error;
2838 }
2839
2840 /* Code flow safety */
2841 assert(session->kernel_session);
2842
2843 /* Copy session output to the newly created Kernel session */
2844 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2845 if (ret != LTTNG_OK) {
2846 goto error;
2847 }
2848
2849 /* Create directory(ies) on local filesystem. */
2850 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2851 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2852 ret = run_as_mkdir_recursive(
2853 session->kernel_session->consumer->dst.trace_path,
2854 S_IRWXU | S_IRWXG, session->uid, session->gid);
2855 if (ret < 0) {
2856 if (errno != EEXIST) {
2857 ERR("Trace directory creation error");
2858 goto error;
2859 }
2860 }
2861 }
2862
2863 session->kernel_session->uid = session->uid;
2864 session->kernel_session->gid = session->gid;
2865 session->kernel_session->output_traces = session->output_traces;
2866 session->kernel_session->snapshot_mode = session->snapshot_mode;
2867
2868 return LTTNG_OK;
2869
2870 error:
2871 trace_kernel_destroy_session(session->kernel_session);
2872 session->kernel_session = NULL;
2873 return ret;
2874 }
2875
2876 /*
2877 * Count number of session permitted by uid/gid.
2878 */
2879 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2880 {
2881 unsigned int i = 0;
2882 struct ltt_session *session;
2883
2884 DBG("Counting number of available session for UID %d GID %d",
2885 uid, gid);
2886 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2887 /*
2888 * Only list the sessions the user can control.
2889 */
2890 if (!session_access_ok(session, uid, gid)) {
2891 continue;
2892 }
2893 i++;
2894 }
2895 return i;
2896 }
2897
2898 /*
2899 * Process the command requested by the lttng client within the command
2900 * context structure. This function make sure that the return structure (llm)
2901 * is set and ready for transmission before returning.
2902 *
2903 * Return any error encountered or 0 for success.
2904 *
2905 * "sock" is only used for special-case var. len data.
2906 *
2907 * Should *NOT* be called with RCU read-side lock held.
2908 */
2909 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2910 int *sock_error)
2911 {
2912 int ret = LTTNG_OK;
2913 int need_tracing_session = 1;
2914 int need_domain;
2915
2916 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2917
2918 *sock_error = 0;
2919
2920 switch (cmd_ctx->lsm->cmd_type) {
2921 case LTTNG_CREATE_SESSION:
2922 case LTTNG_CREATE_SESSION_SNAPSHOT:
2923 case LTTNG_CREATE_SESSION_LIVE:
2924 case LTTNG_DESTROY_SESSION:
2925 case LTTNG_LIST_SESSIONS:
2926 case LTTNG_LIST_DOMAINS:
2927 case LTTNG_START_TRACE:
2928 case LTTNG_STOP_TRACE:
2929 case LTTNG_DATA_PENDING:
2930 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2931 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2932 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2933 case LTTNG_SNAPSHOT_RECORD:
2934 case LTTNG_SAVE_SESSION:
2935 case LTTNG_SET_SESSION_SHM_PATH:
2936 need_domain = 0;
2937 break;
2938 default:
2939 need_domain = 1;
2940 }
2941
2942 if (opt_no_kernel && need_domain
2943 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2944 if (!is_root) {
2945 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2946 } else {
2947 ret = LTTNG_ERR_KERN_NA;
2948 }
2949 goto error;
2950 }
2951
2952 /* Deny register consumer if we already have a spawned consumer. */
2953 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2954 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2955 if (kconsumer_data.pid > 0) {
2956 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2957 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2958 goto error;
2959 }
2960 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2961 }
2962
2963 /*
2964 * Check for command that don't needs to allocate a returned payload. We do
2965 * this here so we don't have to make the call for no payload at each
2966 * command.
2967 */
2968 switch(cmd_ctx->lsm->cmd_type) {
2969 case LTTNG_LIST_SESSIONS:
2970 case LTTNG_LIST_TRACEPOINTS:
2971 case LTTNG_LIST_TRACEPOINT_FIELDS:
2972 case LTTNG_LIST_DOMAINS:
2973 case LTTNG_LIST_CHANNELS:
2974 case LTTNG_LIST_EVENTS:
2975 case LTTNG_LIST_SYSCALLS:
2976 case LTTNG_LIST_TRACKER_PIDS:
2977 break;
2978 default:
2979 /* Setup lttng message with no payload */
2980 ret = setup_lttng_msg(cmd_ctx, 0);
2981 if (ret < 0) {
2982 /* This label does not try to unlock the session */
2983 goto init_setup_error;
2984 }
2985 }
2986
2987 /* Commands that DO NOT need a session. */
2988 switch (cmd_ctx->lsm->cmd_type) {
2989 case LTTNG_CREATE_SESSION:
2990 case LTTNG_CREATE_SESSION_SNAPSHOT:
2991 case LTTNG_CREATE_SESSION_LIVE:
2992 case LTTNG_CALIBRATE:
2993 case LTTNG_LIST_SESSIONS:
2994 case LTTNG_LIST_TRACEPOINTS:
2995 case LTTNG_LIST_SYSCALLS:
2996 case LTTNG_LIST_TRACEPOINT_FIELDS:
2997 case LTTNG_SAVE_SESSION:
2998 need_tracing_session = 0;
2999 break;
3000 default:
3001 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3002 /*
3003 * We keep the session list lock across _all_ commands
3004 * for now, because the per-session lock does not
3005 * handle teardown properly.
3006 */
3007 session_lock_list();
3008 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3009 if (cmd_ctx->session == NULL) {
3010 ret = LTTNG_ERR_SESS_NOT_FOUND;
3011 goto error;
3012 } else {
3013 /* Acquire lock for the session */
3014 session_lock(cmd_ctx->session);
3015 }
3016 break;
3017 }
3018
3019 /*
3020 * Commands that need a valid session but should NOT create one if none
3021 * exists. Instead of creating one and destroying it when the command is
3022 * handled, process that right before so we save some round trip in useless
3023 * code path.
3024 */
3025 switch (cmd_ctx->lsm->cmd_type) {
3026 case LTTNG_DISABLE_CHANNEL:
3027 case LTTNG_DISABLE_EVENT:
3028 switch (cmd_ctx->lsm->domain.type) {
3029 case LTTNG_DOMAIN_KERNEL:
3030 if (!cmd_ctx->session->kernel_session) {
3031 ret = LTTNG_ERR_NO_CHANNEL;
3032 goto error;
3033 }
3034 break;
3035 case LTTNG_DOMAIN_JUL:
3036 case LTTNG_DOMAIN_LOG4J:
3037 case LTTNG_DOMAIN_PYTHON:
3038 case LTTNG_DOMAIN_UST:
3039 if (!cmd_ctx->session->ust_session) {
3040 ret = LTTNG_ERR_NO_CHANNEL;
3041 goto error;
3042 }
3043 break;
3044 default:
3045 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3046 goto error;
3047 }
3048 default:
3049 break;
3050 }
3051
3052 if (!need_domain) {
3053 goto skip_domain;
3054 }
3055
3056 /*
3057 * Check domain type for specific "pre-action".
3058 */
3059 switch (cmd_ctx->lsm->domain.type) {
3060 case LTTNG_DOMAIN_KERNEL:
3061 if (!is_root) {
3062 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3063 goto error;
3064 }
3065
3066 /* Kernel tracer check */
3067 if (kernel_tracer_fd == -1) {
3068 /* Basically, load kernel tracer modules */
3069 ret = init_kernel_tracer();
3070 if (ret != 0) {
3071 goto error;
3072 }
3073 }
3074
3075 /* Consumer is in an ERROR state. Report back to client */
3076 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3077 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3078 goto error;
3079 }
3080
3081 /* Need a session for kernel command */
3082 if (need_tracing_session) {
3083 if (cmd_ctx->session->kernel_session == NULL) {
3084 ret = create_kernel_session(cmd_ctx->session);
3085 if (ret < 0) {
3086 ret = LTTNG_ERR_KERN_SESS_FAIL;
3087 goto error;
3088 }
3089 }
3090
3091 /* Start the kernel consumer daemon */
3092 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3093 if (kconsumer_data.pid == 0 &&
3094 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3095 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3096 ret = start_consumerd(&kconsumer_data);
3097 if (ret < 0) {
3098 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3099 goto error;
3100 }
3101 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3102 } else {
3103 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3104 }
3105
3106 /*
3107 * The consumer was just spawned so we need to add the socket to
3108 * the consumer output of the session if exist.
3109 */
3110 ret = consumer_create_socket(&kconsumer_data,
3111 cmd_ctx->session->kernel_session->consumer);
3112 if (ret < 0) {
3113 goto error;
3114 }
3115 }
3116
3117 break;
3118 case LTTNG_DOMAIN_JUL:
3119 case LTTNG_DOMAIN_LOG4J:
3120 case LTTNG_DOMAIN_PYTHON:
3121 case LTTNG_DOMAIN_UST:
3122 {
3123 if (!ust_app_supported()) {
3124 ret = LTTNG_ERR_NO_UST;
3125 goto error;
3126 }
3127 /* Consumer is in an ERROR state. Report back to client */
3128 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3129 ret = LTTNG_ERR_NO_USTCONSUMERD;
3130 goto error;
3131 }
3132
3133 if (need_tracing_session) {
3134 /* Create UST session if none exist. */
3135 if (cmd_ctx->session->ust_session == NULL) {
3136 ret = create_ust_session(cmd_ctx->session,
3137 &cmd_ctx->lsm->domain);
3138 if (ret != LTTNG_OK) {
3139 goto error;
3140 }
3141 }
3142
3143 /* Start the UST consumer daemons */
3144 /* 64-bit */
3145 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3146 if (consumerd64_bin[0] != '\0' &&
3147 ustconsumer64_data.pid == 0 &&
3148 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3149 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3150 ret = start_consumerd(&ustconsumer64_data);
3151 if (ret < 0) {
3152 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3153 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3154 goto error;
3155 }
3156
3157 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3158 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3159 } else {
3160 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3161 }
3162
3163 /*
3164 * Setup socket for consumer 64 bit. No need for atomic access
3165 * since it was set above and can ONLY be set in this thread.
3166 */
3167 ret = consumer_create_socket(&ustconsumer64_data,
3168 cmd_ctx->session->ust_session->consumer);
3169 if (ret < 0) {
3170 goto error;
3171 }
3172
3173 /* 32-bit */
3174 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3175 if (consumerd32_bin[0] != '\0' &&
3176 ustconsumer32_data.pid == 0 &&
3177 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3178 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3179 ret = start_consumerd(&ustconsumer32_data);
3180 if (ret < 0) {
3181 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3182 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3183 goto error;
3184 }
3185
3186 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3187 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3188 } else {
3189 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3190 }
3191
3192 /*
3193 * Setup socket for consumer 64 bit. No need for atomic access
3194 * since it was set above and can ONLY be set in this thread.
3195 */
3196 ret = consumer_create_socket(&ustconsumer32_data,
3197 cmd_ctx->session->ust_session->consumer);
3198 if (ret < 0) {
3199 goto error;
3200 }
3201 }
3202 break;
3203 }
3204 default:
3205 break;
3206 }
3207 skip_domain:
3208
3209 /* Validate consumer daemon state when start/stop trace command */
3210 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3211 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3212 switch (cmd_ctx->lsm->domain.type) {
3213 case LTTNG_DOMAIN_JUL:
3214 case LTTNG_DOMAIN_LOG4J:
3215 case LTTNG_DOMAIN_PYTHON:
3216 case LTTNG_DOMAIN_UST:
3217 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3218 ret = LTTNG_ERR_NO_USTCONSUMERD;
3219 goto error;
3220 }
3221 break;
3222 case LTTNG_DOMAIN_KERNEL:
3223 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3224 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3225 goto error;
3226 }
3227 break;
3228 }
3229 }
3230
3231 /*
3232 * Check that the UID or GID match that of the tracing session.
3233 * The root user can interact with all sessions.
3234 */
3235 if (need_tracing_session) {
3236 if (!session_access_ok(cmd_ctx->session,
3237 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3238 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3239 ret = LTTNG_ERR_EPERM;
3240 goto error;
3241 }
3242 }
3243
3244 /*
3245 * Send relayd information to consumer as soon as we have a domain and a
3246 * session defined.
3247 */
3248 if (cmd_ctx->session && need_domain) {
3249 /*
3250 * Setup relayd if not done yet. If the relayd information was already
3251 * sent to the consumer, this call will gracefully return.
3252 */
3253 ret = cmd_setup_relayd(cmd_ctx->session);
3254 if (ret != LTTNG_OK) {
3255 goto error;
3256 }
3257 }
3258
3259 /* Process by command type */
3260 switch (cmd_ctx->lsm->cmd_type) {
3261 case LTTNG_ADD_CONTEXT:
3262 {
3263 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3264 cmd_ctx->lsm->u.context.channel_name,
3265 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3266 break;
3267 }
3268 case LTTNG_DISABLE_CHANNEL:
3269 {
3270 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3271 cmd_ctx->lsm->u.disable.channel_name);
3272 break;
3273 }
3274 case LTTNG_DISABLE_EVENT:
3275 {
3276
3277 /*
3278 * FIXME: handle filter; for now we just receive the filter's
3279 * bytecode along with the filter expression which are sent by
3280 * liblttng-ctl and discard them.
3281 *
3282 * This fixes an issue where the client may block while sending
3283 * the filter payload and encounter an error because the session
3284 * daemon closes the socket without ever handling this data.
3285 */
3286 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3287 cmd_ctx->lsm->u.disable.bytecode_len;
3288
3289 if (count) {
3290 char data[LTTNG_FILTER_MAX_LEN];
3291
3292 DBG("Discarding disable event command payload of size %zu", count);
3293 while (count) {
3294 ret = lttcomm_recv_unix_sock(sock, data,
3295 count > sizeof(data) ? sizeof(data) : count);
3296 if (ret < 0) {
3297 goto error;
3298 }
3299
3300 count -= (size_t) ret;
3301 }
3302 }
3303 /* FIXME: passing packed structure to non-packed pointer */
3304 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3305 cmd_ctx->lsm->u.disable.channel_name,
3306 &cmd_ctx->lsm->u.disable.event);
3307 break;
3308 }
3309 case LTTNG_ENABLE_CHANNEL:
3310 {
3311 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3312 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3313 break;
3314 }
3315 case LTTNG_TRACK_PID:
3316 {
3317 ret = cmd_track_pid(cmd_ctx->session,
3318 cmd_ctx->lsm->domain.type,
3319 cmd_ctx->lsm->u.pid_tracker.pid);
3320 break;
3321 }
3322 case LTTNG_UNTRACK_PID:
3323 {
3324 ret = cmd_untrack_pid(cmd_ctx->session,
3325 cmd_ctx->lsm->domain.type,
3326 cmd_ctx->lsm->u.pid_tracker.pid);
3327 break;
3328 }
3329 case LTTNG_ENABLE_EVENT:
3330 {
3331 struct lttng_event_exclusion *exclusion = NULL;
3332 struct lttng_filter_bytecode *bytecode = NULL;
3333 char *filter_expression = NULL;
3334
3335 /* Handle exclusion events and receive it from the client. */
3336 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3337 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3338
3339 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3340 (count * LTTNG_SYMBOL_NAME_LEN));
3341 if (!exclusion) {
3342 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3343 goto error;
3344 }
3345
3346 DBG("Receiving var len exclusion event list from client ...");
3347 exclusion->count = count;
3348 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3349 count * LTTNG_SYMBOL_NAME_LEN);
3350 if (ret <= 0) {
3351 DBG("Nothing recv() from client var len data... continuing");
3352 *sock_error = 1;
3353 free(exclusion);
3354 ret = LTTNG_ERR_EXCLUSION_INVAL;
3355 goto error;
3356 }
3357 }
3358
3359 /* Get filter expression from client. */
3360 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3361 size_t expression_len =
3362 cmd_ctx->lsm->u.enable.expression_len;
3363
3364 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3365 ret = LTTNG_ERR_FILTER_INVAL;
3366 free(exclusion);
3367 goto error;
3368 }
3369
3370 filter_expression = zmalloc(expression_len);
3371 if (!filter_expression) {
3372 free(exclusion);
3373 ret = LTTNG_ERR_FILTER_NOMEM;
3374 goto error;
3375 }
3376
3377 /* Receive var. len. data */
3378 DBG("Receiving var len filter's expression from client ...");
3379 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3380 expression_len);
3381 if (ret <= 0) {
3382 DBG("Nothing recv() from client car len data... continuing");
3383 *sock_error = 1;
3384 free(filter_expression);
3385 free(exclusion);
3386 ret = LTTNG_ERR_FILTER_INVAL;
3387 goto error;
3388 }
3389 }
3390
3391 /* Handle filter and get bytecode from client. */
3392 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3393 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3394
3395 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3396 ret = LTTNG_ERR_FILTER_INVAL;
3397 free(filter_expression);
3398 free(exclusion);
3399 goto error;
3400 }
3401
3402 bytecode = zmalloc(bytecode_len);
3403 if (!bytecode) {
3404 free(filter_expression);
3405 free(exclusion);
3406 ret = LTTNG_ERR_FILTER_NOMEM;
3407 goto error;
3408 }
3409
3410 /* Receive var. len. data */
3411 DBG("Receiving var len filter's bytecode from client ...");
3412 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3413 if (ret <= 0) {
3414 DBG("Nothing recv() from client car len data... continuing");
3415 *sock_error = 1;
3416 free(filter_expression);
3417 free(bytecode);
3418 free(exclusion);
3419 ret = LTTNG_ERR_FILTER_INVAL;
3420 goto error;
3421 }
3422
3423 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3424 free(filter_expression);
3425 free(bytecode);
3426 free(exclusion);
3427 ret = LTTNG_ERR_FILTER_INVAL;
3428 goto error;
3429 }
3430 }
3431
3432 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3433 cmd_ctx->lsm->u.enable.channel_name,
3434 &cmd_ctx->lsm->u.enable.event,
3435 filter_expression, bytecode, exclusion,
3436 kernel_poll_pipe[1]);
3437 break;
3438 }
3439 case LTTNG_LIST_TRACEPOINTS:
3440 {
3441 struct lttng_event *events;
3442 ssize_t nb_events;
3443
3444 session_lock_list();
3445 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3446 session_unlock_list();
3447 if (nb_events < 0) {
3448 /* Return value is a negative lttng_error_code. */
3449 ret = -nb_events;
3450 goto error;
3451 }
3452
3453 /*
3454 * Setup lttng message with payload size set to the event list size in
3455 * bytes and then copy list into the llm payload.
3456 */
3457 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3458 if (ret < 0) {
3459 free(events);
3460 goto setup_error;
3461 }
3462
3463 /* Copy event list into message payload */
3464 memcpy(cmd_ctx->llm->payload, events,
3465 sizeof(struct lttng_event) * nb_events);
3466
3467 free(events);
3468
3469 ret = LTTNG_OK;
3470 break;
3471 }
3472 case LTTNG_LIST_TRACEPOINT_FIELDS:
3473 {
3474 struct lttng_event_field *fields;
3475 ssize_t nb_fields;
3476
3477 session_lock_list();
3478 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3479 &fields);
3480 session_unlock_list();
3481 if (nb_fields < 0) {
3482 /* Return value is a negative lttng_error_code. */
3483 ret = -nb_fields;
3484 goto error;
3485 }
3486
3487 /*
3488 * Setup lttng message with payload size set to the event list size in
3489 * bytes and then copy list into the llm payload.
3490 */
3491 ret = setup_lttng_msg(cmd_ctx,
3492 sizeof(struct lttng_event_field) * nb_fields);
3493 if (ret < 0) {
3494 free(fields);
3495 goto setup_error;
3496 }
3497
3498 /* Copy event list into message payload */
3499 memcpy(cmd_ctx->llm->payload, fields,
3500 sizeof(struct lttng_event_field) * nb_fields);
3501
3502 free(fields);
3503
3504 ret = LTTNG_OK;
3505 break;
3506 }
3507 case LTTNG_LIST_SYSCALLS:
3508 {
3509 struct lttng_event *events;
3510 ssize_t nb_events;
3511
3512 nb_events = cmd_list_syscalls(&events);
3513 if (nb_events < 0) {
3514 /* Return value is a negative lttng_error_code. */
3515 ret = -nb_events;
3516 goto error;
3517 }
3518
3519 /*
3520 * Setup lttng message with payload size set to the event list size in
3521 * bytes and then copy list into the llm payload.
3522 */
3523 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3524 if (ret < 0) {
3525 free(events);
3526 goto setup_error;
3527 }
3528
3529 /* Copy event list into message payload */
3530 memcpy(cmd_ctx->llm->payload, events,
3531 sizeof(struct lttng_event) * nb_events);
3532
3533 free(events);
3534
3535 ret = LTTNG_OK;
3536 break;
3537 }
3538 case LTTNG_LIST_TRACKER_PIDS:
3539 {
3540 int32_t *pids = NULL;
3541 ssize_t nr_pids;
3542
3543 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3544 cmd_ctx->lsm->domain.type, &pids);
3545 if (nr_pids < 0) {
3546 /* Return value is a negative lttng_error_code. */
3547 ret = -nr_pids;
3548 goto error;
3549 }
3550
3551 /*
3552 * Setup lttng message with payload size set to the event list size in
3553 * bytes and then copy list into the llm payload.
3554 */
3555 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3556 if (ret < 0) {
3557 free(pids);
3558 goto setup_error;
3559 }
3560
3561 /* Copy event list into message payload */
3562 memcpy(cmd_ctx->llm->payload, pids,
3563 sizeof(int) * nr_pids);
3564
3565 free(pids);
3566
3567 ret = LTTNG_OK;
3568 break;
3569 }
3570 case LTTNG_SET_CONSUMER_URI:
3571 {
3572 size_t nb_uri, len;
3573 struct lttng_uri *uris;
3574
3575 nb_uri = cmd_ctx->lsm->u.uri.size;
3576 len = nb_uri * sizeof(struct lttng_uri);
3577
3578 if (nb_uri == 0) {
3579 ret = LTTNG_ERR_INVALID;
3580 goto error;
3581 }
3582
3583 uris = zmalloc(len);
3584 if (uris == NULL) {
3585 ret = LTTNG_ERR_FATAL;
3586 goto error;
3587 }
3588
3589 /* Receive variable len data */
3590 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3591 ret = lttcomm_recv_unix_sock(sock, uris, len);
3592 if (ret <= 0) {
3593 DBG("No URIs received from client... continuing");
3594 *sock_error = 1;
3595 ret = LTTNG_ERR_SESSION_FAIL;
3596 free(uris);
3597 goto error;
3598 }
3599
3600 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3601 free(uris);
3602 if (ret != LTTNG_OK) {
3603 goto error;
3604 }
3605
3606
3607 break;
3608 }
3609 case LTTNG_START_TRACE:
3610 {
3611 ret = cmd_start_trace(cmd_ctx->session);
3612 break;
3613 }
3614 case LTTNG_STOP_TRACE:
3615 {
3616 ret = cmd_stop_trace(cmd_ctx->session);
3617 break;
3618 }
3619 case LTTNG_CREATE_SESSION:
3620 {
3621 size_t nb_uri, len;
3622 struct lttng_uri *uris = NULL;
3623
3624 nb_uri = cmd_ctx->lsm->u.uri.size;
3625 len = nb_uri * sizeof(struct lttng_uri);
3626
3627 if (nb_uri > 0) {
3628 uris = zmalloc(len);
3629 if (uris == NULL) {
3630 ret = LTTNG_ERR_FATAL;
3631 goto error;
3632 }
3633
3634 /* Receive variable len data */
3635 DBG("Waiting for %zu URIs from client ...", nb_uri);
3636 ret = lttcomm_recv_unix_sock(sock, uris, len);
3637 if (ret <= 0) {
3638 DBG("No URIs received from client... continuing");
3639 *sock_error = 1;
3640 ret = LTTNG_ERR_SESSION_FAIL;
3641 free(uris);
3642 goto error;
3643 }
3644
3645 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3646 DBG("Creating session with ONE network URI is a bad call");
3647 ret = LTTNG_ERR_SESSION_FAIL;
3648 free(uris);
3649 goto error;
3650 }
3651 }
3652
3653 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3654 &cmd_ctx->creds, 0);
3655
3656 free(uris);
3657
3658 break;
3659 }
3660 case LTTNG_DESTROY_SESSION:
3661 {
3662 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3663
3664 /* Set session to NULL so we do not unlock it after free. */
3665 cmd_ctx->session = NULL;
3666 break;
3667 }
3668 case LTTNG_LIST_DOMAINS:
3669 {
3670 ssize_t nb_dom;
3671 struct lttng_domain *domains = NULL;
3672
3673 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3674 if (nb_dom < 0) {
3675 /* Return value is a negative lttng_error_code. */
3676 ret = -nb_dom;
3677 goto error;
3678 }
3679
3680 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3681 if (ret < 0) {
3682 free(domains);
3683 goto setup_error;
3684 }
3685
3686 /* Copy event list into message payload */
3687 memcpy(cmd_ctx->llm->payload, domains,
3688 nb_dom * sizeof(struct lttng_domain));
3689
3690 free(domains);
3691
3692 ret = LTTNG_OK;
3693 break;
3694 }
3695 case LTTNG_LIST_CHANNELS:
3696 {
3697 int nb_chan;
3698 struct lttng_channel *channels = NULL;
3699
3700 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3701 cmd_ctx->session, &channels);
3702 if (nb_chan < 0) {
3703 /* Return value is a negative lttng_error_code. */
3704 ret = -nb_chan;
3705 goto error;
3706 }
3707
3708 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3709 if (ret < 0) {
3710 free(channels);
3711 goto setup_error;
3712 }
3713
3714 /* Copy event list into message payload */
3715 memcpy(cmd_ctx->llm->payload, channels,
3716 nb_chan * sizeof(struct lttng_channel));
3717
3718 free(channels);
3719
3720 ret = LTTNG_OK;
3721 break;
3722 }
3723 case LTTNG_LIST_EVENTS:
3724 {
3725 ssize_t nb_event;
3726 struct lttng_event *events = NULL;
3727
3728 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3729 cmd_ctx->lsm->u.list.channel_name, &events);
3730 if (nb_event < 0) {
3731 /* Return value is a negative lttng_error_code. */
3732 ret = -nb_event;
3733 goto error;
3734 }
3735
3736 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3737 if (ret < 0) {
3738 free(events);
3739 goto setup_error;
3740 }
3741
3742 /* Copy event list into message payload */
3743 memcpy(cmd_ctx->llm->payload, events,
3744 nb_event * sizeof(struct lttng_event));
3745
3746 free(events);
3747
3748 ret = LTTNG_OK;
3749 break;
3750 }
3751 case LTTNG_LIST_SESSIONS:
3752 {
3753 unsigned int nr_sessions;
3754
3755 session_lock_list();
3756 nr_sessions = lttng_sessions_count(
3757 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3758 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3759
3760 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3761 if (ret < 0) {
3762 session_unlock_list();
3763 goto setup_error;
3764 }
3765
3766 /* Filled the session array */
3767 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3768 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3769 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3770
3771 session_unlock_list();
3772
3773 ret = LTTNG_OK;
3774 break;
3775 }
3776 case LTTNG_CALIBRATE:
3777 {
3778 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3779 &cmd_ctx->lsm->u.calibrate);
3780 break;
3781 }
3782 case LTTNG_REGISTER_CONSUMER:
3783 {
3784 struct consumer_data *cdata;
3785
3786 switch (cmd_ctx->lsm->domain.type) {
3787 case LTTNG_DOMAIN_KERNEL:
3788 cdata = &kconsumer_data;
3789 break;
3790 default:
3791 ret = LTTNG_ERR_UND;
3792 goto error;
3793 }
3794
3795 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3796 cmd_ctx->lsm->u.reg.path, cdata);
3797 break;
3798 }
3799 case LTTNG_DATA_PENDING:
3800 {
3801 ret = cmd_data_pending(cmd_ctx->session);
3802 break;
3803 }
3804 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3805 {
3806 struct lttcomm_lttng_output_id reply;
3807
3808 ret = cmd_snapshot_add_output(cmd_ctx->session,
3809 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3810 if (ret != LTTNG_OK) {
3811 goto error;
3812 }
3813
3814 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3815 if (ret < 0) {
3816 goto setup_error;
3817 }
3818
3819 /* Copy output list into message payload */
3820 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3821 ret = LTTNG_OK;
3822 break;
3823 }
3824 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3825 {
3826 ret = cmd_snapshot_del_output(cmd_ctx->session,
3827 &cmd_ctx->lsm->u.snapshot_output.output);
3828 break;
3829 }
3830 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3831 {
3832 ssize_t nb_output;
3833 struct lttng_snapshot_output *outputs = NULL;
3834
3835 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3836 if (nb_output < 0) {
3837 ret = -nb_output;
3838 goto error;
3839 }
3840
3841 ret = setup_lttng_msg(cmd_ctx,
3842 nb_output * sizeof(struct lttng_snapshot_output));
3843 if (ret < 0) {
3844 free(outputs);
3845 goto setup_error;
3846 }
3847
3848 if (outputs) {
3849 /* Copy output list into message payload */
3850 memcpy(cmd_ctx->llm->payload, outputs,
3851 nb_output * sizeof(struct lttng_snapshot_output));
3852 free(outputs);
3853 }
3854
3855 ret = LTTNG_OK;
3856 break;
3857 }
3858 case LTTNG_SNAPSHOT_RECORD:
3859 {
3860 ret = cmd_snapshot_record(cmd_ctx->session,
3861 &cmd_ctx->lsm->u.snapshot_record.output,
3862 cmd_ctx->lsm->u.snapshot_record.wait);
3863 break;
3864 }
3865 case LTTNG_CREATE_SESSION_SNAPSHOT:
3866 {
3867 size_t nb_uri, len;
3868 struct lttng_uri *uris = NULL;
3869
3870 nb_uri = cmd_ctx->lsm->u.uri.size;
3871 len = nb_uri * sizeof(struct lttng_uri);
3872
3873 if (nb_uri > 0) {
3874 uris = zmalloc(len);
3875 if (uris == NULL) {
3876 ret = LTTNG_ERR_FATAL;
3877 goto error;
3878 }
3879
3880 /* Receive variable len data */
3881 DBG("Waiting for %zu URIs from client ...", nb_uri);
3882 ret = lttcomm_recv_unix_sock(sock, uris, len);
3883 if (ret <= 0) {
3884 DBG("No URIs received from client... continuing");
3885 *sock_error = 1;
3886 ret = LTTNG_ERR_SESSION_FAIL;
3887 free(uris);
3888 goto error;
3889 }
3890
3891 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3892 DBG("Creating session with ONE network URI is a bad call");
3893 ret = LTTNG_ERR_SESSION_FAIL;
3894 free(uris);
3895 goto error;
3896 }
3897 }
3898
3899 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3900 nb_uri, &cmd_ctx->creds);
3901 free(uris);
3902 break;
3903 }
3904 case LTTNG_CREATE_SESSION_LIVE:
3905 {
3906 size_t nb_uri, len;
3907 struct lttng_uri *uris = NULL;
3908
3909 nb_uri = cmd_ctx->lsm->u.uri.size;
3910 len = nb_uri * sizeof(struct lttng_uri);
3911
3912 if (nb_uri > 0) {
3913 uris = zmalloc(len);
3914 if (uris == NULL) {
3915 ret = LTTNG_ERR_FATAL;
3916 goto error;
3917 }
3918
3919 /* Receive variable len data */
3920 DBG("Waiting for %zu URIs from client ...", nb_uri);
3921 ret = lttcomm_recv_unix_sock(sock, uris, len);
3922 if (ret <= 0) {
3923 DBG("No URIs received from client... continuing");
3924 *sock_error = 1;
3925 ret = LTTNG_ERR_SESSION_FAIL;
3926 free(uris);
3927 goto error;
3928 }
3929
3930 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3931 DBG("Creating session with ONE network URI is a bad call");
3932 ret = LTTNG_ERR_SESSION_FAIL;
3933 free(uris);
3934 goto error;
3935 }
3936 }
3937
3938 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3939 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3940 free(uris);
3941 break;
3942 }
3943 case LTTNG_SAVE_SESSION:
3944 {
3945 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3946 &cmd_ctx->creds);
3947 break;
3948 }
3949 case LTTNG_SET_SESSION_SHM_PATH:
3950 {
3951 ret = cmd_set_session_shm_path(cmd_ctx->session,
3952 cmd_ctx->lsm->u.set_shm_path.shm_path);
3953 break;
3954 }
3955 default:
3956 ret = LTTNG_ERR_UND;
3957 break;
3958 }
3959
3960 error:
3961 if (cmd_ctx->llm == NULL) {
3962 DBG("Missing llm structure. Allocating one.");
3963 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3964 goto setup_error;
3965 }
3966 }
3967 /* Set return code */
3968 cmd_ctx->llm->ret_code = ret;
3969 setup_error:
3970 if (cmd_ctx->session) {
3971 session_unlock(cmd_ctx->session);
3972 }
3973 if (need_tracing_session) {
3974 session_unlock_list();
3975 }
3976 init_setup_error:
3977 return ret;
3978 }
3979
3980 /*
3981 * Thread managing health check socket.
3982 */
3983 static void *thread_manage_health(void *data)
3984 {
3985 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3986 uint32_t revents, nb_fd;
3987 struct lttng_poll_event events;
3988 struct health_comm_msg msg;
3989 struct health_comm_reply reply;
3990
3991 DBG("[thread] Manage health check started");
3992
3993 rcu_register_thread();
3994
3995 /* We might hit an error path before this is created. */
3996 lttng_poll_init(&events);
3997
3998 /* Create unix socket */
3999 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4000 if (sock < 0) {
4001 ERR("Unable to create health check Unix socket");
4002 ret = -1;
4003 goto error;
4004 }
4005
4006 if (is_root) {
4007 /* lttng health client socket path permissions */
4008 ret = chown(health_unix_sock_path, 0,
4009 utils_get_group_id(tracing_group_name));
4010 if (ret < 0) {
4011 ERR("Unable to set group on %s", health_unix_sock_path);
4012 PERROR("chown");
4013 ret = -1;
4014 goto error;
4015 }
4016
4017 ret = chmod(health_unix_sock_path,
4018 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4019 if (ret < 0) {
4020 ERR("Unable to set permissions on %s", health_unix_sock_path);
4021 PERROR("chmod");
4022 ret = -1;
4023 goto error;
4024 }
4025 }
4026
4027 /*
4028 * Set the CLOEXEC flag. Return code is useless because either way, the
4029 * show must go on.
4030 */
4031 (void) utils_set_fd_cloexec(sock);
4032
4033 ret = lttcomm_listen_unix_sock(sock);
4034 if (ret < 0) {
4035 goto error;
4036 }
4037
4038 /*
4039 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4040 * more will be added to this poll set.
4041 */
4042 ret = sessiond_set_thread_pollset(&events, 2);
4043 if (ret < 0) {
4044 goto error;
4045 }
4046
4047 /* Add the application registration socket */
4048 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4049 if (ret < 0) {
4050 goto error;
4051 }
4052
4053 sessiond_notify_ready();
4054
4055 while (1) {
4056 DBG("Health check ready");
4057
4058 /* Inifinite blocking call, waiting for transmission */
4059 restart:
4060 ret = lttng_poll_wait(&events, -1);
4061 if (ret < 0) {
4062 /*
4063 * Restart interrupted system call.
4064 */
4065 if (errno == EINTR) {
4066 goto restart;
4067 }
4068 goto error;
4069 }
4070
4071 nb_fd = ret;
4072
4073 for (i = 0; i < nb_fd; i++) {
4074 /* Fetch once the poll data */
4075 revents = LTTNG_POLL_GETEV(&events, i);
4076 pollfd = LTTNG_POLL_GETFD(&events, i);
4077
4078 if (!revents) {
4079 /* No activity for this FD (poll implementation). */
4080 continue;
4081 }
4082
4083 /* Thread quit pipe has been closed. Killing thread. */
4084 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4085 if (ret) {
4086 err = 0;
4087 goto exit;
4088 }
4089
4090 /* Event on the registration socket */
4091 if (pollfd == sock) {
4092 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4093 ERR("Health socket poll error");
4094 goto error;
4095 }
4096 }
4097 }
4098
4099 new_sock = lttcomm_accept_unix_sock(sock);
4100 if (new_sock < 0) {
4101 goto error;
4102 }
4103
4104 /*
4105 * Set the CLOEXEC flag. Return code is useless because either way, the
4106 * show must go on.
4107 */
4108 (void) utils_set_fd_cloexec(new_sock);
4109
4110 DBG("Receiving data from client for health...");
4111 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4112 if (ret <= 0) {
4113 DBG("Nothing recv() from client... continuing");
4114 ret = close(new_sock);
4115 if (ret) {
4116 PERROR("close");
4117 }
4118 new_sock = -1;
4119 continue;
4120 }
4121
4122 rcu_thread_online();
4123
4124 memset(&reply, 0, sizeof(reply));
4125 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4126 /*
4127 * health_check_state returns 0 if health is
4128 * bad.
4129 */
4130 if (!health_check_state(health_sessiond, i)) {
4131 reply.ret_code |= 1ULL << i;
4132 }
4133 }
4134
4135 DBG2("Health check return value %" PRIx64, reply.ret_code);
4136
4137 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4138 if (ret < 0) {
4139 ERR("Failed to send health data back to client");
4140 }
4141
4142 /* End of transmission */
4143 ret = close(new_sock);
4144 if (ret) {
4145 PERROR("close");
4146 }
4147 new_sock = -1;
4148 }
4149
4150 exit:
4151 error:
4152 if (err) {
4153 ERR("Health error occurred in %s", __func__);
4154 }
4155 DBG("Health check thread dying");
4156 unlink(health_unix_sock_path);
4157 if (sock >= 0) {
4158 ret = close(sock);
4159 if (ret) {
4160 PERROR("close");
4161 }
4162 }
4163
4164 lttng_poll_clean(&events);
4165
4166 rcu_unregister_thread();
4167 return NULL;
4168 }
4169
4170 /*
4171 * This thread manage all clients request using the unix client socket for
4172 * communication.
4173 */
4174 static void *thread_manage_clients(void *data)
4175 {
4176 int sock = -1, ret, i, pollfd, err = -1;
4177 int sock_error;
4178 uint32_t revents, nb_fd;
4179 struct command_ctx *cmd_ctx = NULL;
4180 struct lttng_poll_event events;
4181
4182 DBG("[thread] Manage client started");
4183
4184 rcu_register_thread();
4185
4186 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4187
4188 health_code_update();
4189
4190 ret = lttcomm_listen_unix_sock(client_sock);
4191 if (ret < 0) {
4192 goto error_listen;
4193 }
4194
4195 /*
4196 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4197 * more will be added to this poll set.
4198 */
4199 ret = sessiond_set_thread_pollset(&events, 2);
4200 if (ret < 0) {
4201 goto error_create_poll;
4202 }
4203
4204 /* Add the application registration socket */
4205 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4206 if (ret < 0) {
4207 goto error;
4208 }
4209
4210 sessiond_notify_ready();
4211 ret = sem_post(&load_info->message_thread_ready);
4212 if (ret) {
4213 PERROR("sem_post message_thread_ready");
4214 goto error;
4215 }
4216
4217 /* This testpoint is after we signal readiness to the parent. */
4218 if (testpoint(sessiond_thread_manage_clients)) {
4219 goto error;
4220 }
4221
4222 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4223 goto error;
4224 }
4225
4226 health_code_update();
4227
4228 while (1) {
4229 DBG("Accepting client command ...");
4230
4231 /* Inifinite blocking call, waiting for transmission */
4232 restart:
4233 health_poll_entry();
4234 ret = lttng_poll_wait(&events, -1);
4235 health_poll_exit();
4236 if (ret < 0) {
4237 /*
4238 * Restart interrupted system call.
4239 */
4240 if (errno == EINTR) {
4241 goto restart;
4242 }
4243 goto error;
4244 }
4245
4246 nb_fd = ret;
4247
4248 for (i = 0; i < nb_fd; i++) {
4249 /* Fetch once the poll data */
4250 revents = LTTNG_POLL_GETEV(&events, i);
4251 pollfd = LTTNG_POLL_GETFD(&events, i);
4252
4253 health_code_update();
4254
4255 if (!revents) {
4256 /* No activity for this FD (poll implementation). */
4257 continue;
4258 }
4259
4260 /* Thread quit pipe has been closed. Killing thread. */
4261 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4262 if (ret) {
4263 err = 0;
4264 goto exit;
4265 }
4266
4267 /* Event on the registration socket */
4268 if (pollfd == client_sock) {
4269 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4270 ERR("Client socket poll error");
4271 goto error;
4272 }
4273 }
4274 }
4275
4276 DBG("Wait for client response");
4277
4278 health_code_update();
4279
4280 sock = lttcomm_accept_unix_sock(client_sock);
4281 if (sock < 0) {
4282 goto error;
4283 }
4284
4285 /*
4286 * Set the CLOEXEC flag. Return code is useless because either way, the
4287 * show must go on.
4288 */
4289 (void) utils_set_fd_cloexec(sock);
4290
4291 /* Set socket option for credentials retrieval */
4292 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4293 if (ret < 0) {
4294 goto error;
4295 }
4296
4297 /* Allocate context command to process the client request */
4298 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4299 if (cmd_ctx == NULL) {
4300 PERROR("zmalloc cmd_ctx");
4301 goto error;
4302 }
4303
4304 /* Allocate data buffer for reception */
4305 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4306 if (cmd_ctx->lsm == NULL) {
4307 PERROR("zmalloc cmd_ctx->lsm");
4308 goto error;
4309 }
4310
4311 cmd_ctx->llm = NULL;
4312 cmd_ctx->session = NULL;
4313
4314 health_code_update();
4315
4316 /*
4317 * Data is received from the lttng client. The struct
4318 * lttcomm_session_msg (lsm) contains the command and data request of
4319 * the client.
4320 */
4321 DBG("Receiving data from client ...");
4322 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4323 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4324 if (ret <= 0) {
4325 DBG("Nothing recv() from client... continuing");
4326 ret = close(sock);
4327 if (ret) {
4328 PERROR("close");
4329 }
4330 sock = -1;
4331 clean_command_ctx(&cmd_ctx);
4332 continue;
4333 }
4334
4335 health_code_update();
4336
4337 // TODO: Validate cmd_ctx including sanity check for
4338 // security purpose.
4339
4340 rcu_thread_online();
4341 /*
4342 * This function dispatch the work to the kernel or userspace tracer
4343 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4344 * informations for the client. The command context struct contains
4345 * everything this function may needs.
4346 */
4347 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4348 rcu_thread_offline();
4349 if (ret < 0) {
4350 ret = close(sock);
4351 if (ret) {
4352 PERROR("close");
4353 }
4354 sock = -1;
4355 /*
4356 * TODO: Inform client somehow of the fatal error. At
4357 * this point, ret < 0 means that a zmalloc failed
4358 * (ENOMEM). Error detected but still accept
4359 * command, unless a socket error has been
4360 * detected.
4361 */
4362 clean_command_ctx(&cmd_ctx);
4363 continue;
4364 }
4365
4366 health_code_update();
4367
4368 DBG("Sending response (size: %d, retcode: %s)",
4369 cmd_ctx->lttng_msg_size,
4370 lttng_strerror(-cmd_ctx->llm->ret_code));
4371 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4372 if (ret < 0) {
4373 ERR("Failed to send data back to client");
4374 }
4375
4376 /* End of transmission */
4377 ret = close(sock);
4378 if (ret) {
4379 PERROR("close");
4380 }
4381 sock = -1;
4382
4383 clean_command_ctx(&cmd_ctx);
4384
4385 health_code_update();
4386 }
4387
4388 exit:
4389 error:
4390 if (sock >= 0) {
4391 ret = close(sock);
4392 if (ret) {
4393 PERROR("close");
4394 }
4395 }
4396
4397 lttng_poll_clean(&events);
4398 clean_command_ctx(&cmd_ctx);
4399
4400 error_listen:
4401 error_create_poll:
4402 unlink(client_unix_sock_path);
4403 if (client_sock >= 0) {
4404 ret = close(client_sock);
4405 if (ret) {
4406 PERROR("close");
4407 }
4408 }
4409
4410 if (err) {
4411 health_error();
4412 ERR("Health error occurred in %s", __func__);
4413 }
4414
4415 health_unregister(health_sessiond);
4416
4417 DBG("Client thread dying");
4418
4419 rcu_unregister_thread();
4420
4421 /*
4422 * Since we are creating the consumer threads, we own them, so we need
4423 * to join them before our thread exits.
4424 */
4425 ret = join_consumer_thread(&kconsumer_data);
4426 if (ret) {
4427 errno = ret;
4428 PERROR("join_consumer");
4429 }
4430
4431 ret = join_consumer_thread(&ustconsumer32_data);
4432 if (ret) {
4433 errno = ret;
4434 PERROR("join_consumer ust32");
4435 }
4436
4437 ret = join_consumer_thread(&ustconsumer64_data);
4438 if (ret) {
4439 errno = ret;
4440 PERROR("join_consumer ust64");
4441 }
4442 return NULL;
4443 }
4444
4445
4446 /*
4447 * usage function on stderr
4448 */
4449 static void usage(void)
4450 {
4451 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4452 fprintf(stderr, " -h, --help Display this usage.\n");
4453 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4454 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4455 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4456 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4457 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4458 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4459 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4460 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4461 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4462 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4463 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4464 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4465 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4466 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4467 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4468 fprintf(stderr, " -V, --version Show version number.\n");
4469 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4470 fprintf(stderr, " -q, --quiet No output at all.\n");
4471 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4472 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4473 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4474 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4475 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4476 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4477 fprintf(stderr, " -l --load PATH Load session configuration\n");
4478 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4479 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4480 }
4481
4482 static int string_match(const char *str1, const char *str2)
4483 {
4484 return (str1 && str2) && !strcmp(str1, str2);
4485 }
4486
4487 /*
4488 * Take an option from the getopt output and set it in the right variable to be
4489 * used later.
4490 *
4491 * Return 0 on success else a negative value.
4492 */
4493 static int set_option(int opt, const char *arg, const char *optname)
4494 {
4495 int ret = 0;
4496
4497 if (arg && arg[0] == '\0') {
4498 /*
4499 * This only happens if the value is read from daemon config
4500 * file. This means the option requires an argument and the
4501 * configuration file contains a line such as:
4502 * my_option =
4503 */
4504 ret = -EINVAL;
4505 goto end;
4506 }
4507
4508 if (string_match(optname, "client-sock") || opt == 'c') {
4509 if (lttng_is_setuid_setgid()) {
4510 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4511 "-c, --client-sock");
4512 } else {
4513 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4514 }
4515 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4516 if (lttng_is_setuid_setgid()) {
4517 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4518 "-a, --apps-sock");
4519 } else {
4520 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4521 }
4522 } else if (string_match(optname, "daemonize") || opt == 'd') {
4523 opt_daemon = 1;
4524 } else if (string_match(optname, "background") || opt == 'b') {
4525 opt_background = 1;
4526 } else if (string_match(optname, "group") || opt == 'g') {
4527 if (lttng_is_setuid_setgid()) {
4528 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4529 "-g, --group");
4530 } else {
4531 /*
4532 * If the override option is set, the pointer points to a
4533 * *non* const thus freeing it even though the variable type is
4534 * set to const.
4535 */
4536 if (tracing_group_name_override) {
4537 free((void *) tracing_group_name);
4538 }
4539 tracing_group_name = strdup(arg);
4540 if (!tracing_group_name) {
4541 PERROR("strdup");
4542 ret = -ENOMEM;
4543 }
4544 tracing_group_name_override = 1;
4545 }
4546 } else if (string_match(optname, "help") || opt == 'h') {
4547 usage();
4548 exit(EXIT_SUCCESS);
4549 } else if (string_match(optname, "version") || opt == 'V') {
4550 fprintf(stdout, "%s\n", VERSION);
4551 exit(EXIT_SUCCESS);
4552 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4553 opt_sig_parent = 1;
4554 } else if (string_match(optname, "kconsumerd-err-sock")) {
4555 if (lttng_is_setuid_setgid()) {
4556 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4557 "--kconsumerd-err-sock");
4558 } else {
4559 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4560 }
4561 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4562 if (lttng_is_setuid_setgid()) {
4563 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4564 "--kconsumerd-cmd-sock");
4565 } else {
4566 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4567 }
4568 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4569 if (lttng_is_setuid_setgid()) {
4570 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4571 "--ustconsumerd64-err-sock");
4572 } else {
4573 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4574 }
4575 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4576 if (lttng_is_setuid_setgid()) {
4577 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4578 "--ustconsumerd64-cmd-sock");
4579 } else {
4580 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4581 }
4582 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4583 if (lttng_is_setuid_setgid()) {
4584 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4585 "--ustconsumerd32-err-sock");
4586 } else {
4587 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4588 }
4589 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4590 if (lttng_is_setuid_setgid()) {
4591 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4592 "--ustconsumerd32-cmd-sock");
4593 } else {
4594 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4595 }
4596 } else if (string_match(optname, "no-kernel")) {
4597 opt_no_kernel = 1;
4598 } else if (string_match(optname, "quiet") || opt == 'q') {
4599 lttng_opt_quiet = 1;
4600 } else if (string_match(optname, "verbose") || opt == 'v') {
4601 /* Verbose level can increase using multiple -v */
4602 if (arg) {
4603 /* Value obtained from config file */
4604 lttng_opt_verbose = config_parse_value(arg);
4605 } else {
4606 /* -v used on command line */
4607 lttng_opt_verbose++;
4608 }
4609 /* Clamp value to [0, 3] */
4610 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4611 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4612 } else if (string_match(optname, "verbose-consumer")) {
4613 if (arg) {
4614 opt_verbose_consumer = config_parse_value(arg);
4615 } else {
4616 opt_verbose_consumer += 1;
4617 }
4618 } else if (string_match(optname, "consumerd32-path")) {
4619 if (lttng_is_setuid_setgid()) {
4620 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4621 "--consumerd32-path");
4622 } else {
4623 if (consumerd32_bin_override) {
4624 free((void *) consumerd32_bin);
4625 }
4626 consumerd32_bin = strdup(arg);
4627 if (!consumerd32_bin) {
4628 PERROR("strdup");
4629 ret = -ENOMEM;
4630 }
4631 consumerd32_bin_override = 1;
4632 }
4633 } else if (string_match(optname, "consumerd32-libdir")) {
4634 if (lttng_is_setuid_setgid()) {
4635 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4636 "--consumerd32-libdir");
4637 } else {
4638 if (consumerd32_libdir_override) {
4639 free((void *) consumerd32_libdir);
4640 }
4641 consumerd32_libdir = strdup(arg);
4642 if (!consumerd32_libdir) {
4643 PERROR("strdup");
4644 ret = -ENOMEM;
4645 }
4646 consumerd32_libdir_override = 1;
4647 }
4648 } else if (string_match(optname, "consumerd64-path")) {
4649 if (lttng_is_setuid_setgid()) {
4650 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4651 "--consumerd64-path");
4652 } else {
4653 if (consumerd64_bin_override) {
4654 free((void *) consumerd64_bin);
4655 }
4656 consumerd64_bin = strdup(arg);
4657 if (!consumerd64_bin) {
4658 PERROR("strdup");
4659 ret = -ENOMEM;
4660 }
4661 consumerd64_bin_override = 1;
4662 }
4663 } else if (string_match(optname, "consumerd64-libdir")) {
4664 if (lttng_is_setuid_setgid()) {
4665 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4666 "--consumerd64-libdir");
4667 } else {
4668 if (consumerd64_libdir_override) {
4669 free((void *) consumerd64_libdir);
4670 }
4671 consumerd64_libdir = strdup(arg);
4672 if (!consumerd64_libdir) {
4673 PERROR("strdup");
4674 ret = -ENOMEM;
4675 }
4676 consumerd64_libdir_override = 1;
4677 }
4678 } else if (string_match(optname, "pidfile") || opt == 'p') {
4679 if (lttng_is_setuid_setgid()) {
4680 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4681 "-p, --pidfile");
4682 } else {
4683 free(opt_pidfile);
4684 opt_pidfile = strdup(arg);
4685 if (!opt_pidfile) {
4686 PERROR("strdup");
4687 ret = -ENOMEM;
4688 }
4689 }
4690 } else if (string_match(optname, "agent-tcp-port")) {
4691 if (lttng_is_setuid_setgid()) {
4692 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4693 "--agent-tcp-port");
4694 } else {
4695 unsigned long v;
4696
4697 if (!arg) {
4698 ret = -EINVAL;
4699 goto end;
4700 }
4701 errno = 0;
4702 v = strtoul(arg, NULL, 0);
4703 if (errno != 0 || !isdigit(arg[0])) {
4704 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4705 return -1;
4706 }
4707 if (v == 0 || v >= 65535) {
4708 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4709 return -1;
4710 }
4711 agent_tcp_port = (uint32_t) v;
4712 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4713 }
4714 } else if (string_match(optname, "load") || opt == 'l') {
4715 if (lttng_is_setuid_setgid()) {
4716 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4717 "-l, --load");
4718 } else {
4719 free(opt_load_session_path);
4720 opt_load_session_path = strdup(arg);
4721 if (!opt_load_session_path) {
4722 PERROR("strdup");
4723 ret = -ENOMEM;
4724 }
4725 }
4726 } else if (string_match(optname, "kmod-probes")) {
4727 if (lttng_is_setuid_setgid()) {
4728 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4729 "--kmod-probes");
4730 } else {
4731 free(kmod_probes_list);
4732 kmod_probes_list = strdup(arg);
4733 if (!kmod_probes_list) {
4734 PERROR("strdup");
4735 ret = -ENOMEM;
4736 }
4737 }
4738 } else if (string_match(optname, "extra-kmod-probes")) {
4739 if (lttng_is_setuid_setgid()) {
4740 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4741 "--extra-kmod-probes");
4742 } else {
4743 free(kmod_extra_probes_list);
4744 kmod_extra_probes_list = strdup(arg);
4745 if (!kmod_extra_probes_list) {
4746 PERROR("strdup");
4747 ret = -ENOMEM;
4748 }
4749 }
4750 } else if (string_match(optname, "config") || opt == 'f') {
4751 /* This is handled in set_options() thus silent skip. */
4752 goto end;
4753 } else {
4754 /* Unknown option or other error.
4755 * Error is printed by getopt, just return */
4756 ret = -1;
4757 }
4758
4759 end:
4760 if (ret == -EINVAL) {
4761 const char *opt_name = "unknown";
4762 int i;
4763
4764 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4765 i++) {
4766 if (opt == long_options[i].val) {
4767 opt_name = long_options[i].name;
4768 break;
4769 }
4770 }
4771
4772 WARN("Invalid argument provided for option \"%s\", using default value.",
4773 opt_name);
4774 }
4775
4776 return ret;
4777 }
4778
4779 /*
4780 * config_entry_handler_cb used to handle options read from a config file.
4781 * See config_entry_handler_cb comment in common/config/config.h for the
4782 * return value conventions.
4783 */
4784 static int config_entry_handler(const struct config_entry *entry, void *unused)
4785 {
4786 int ret = 0, i;
4787
4788 if (!entry || !entry->name || !entry->value) {
4789 ret = -EINVAL;
4790 goto end;
4791 }
4792
4793 /* Check if the option is to be ignored */
4794 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4795 if (!strcmp(entry->name, config_ignore_options[i])) {
4796 goto end;
4797 }
4798 }
4799
4800 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4801 i++) {
4802
4803 /* Ignore if not fully matched. */
4804 if (strcmp(entry->name, long_options[i].name)) {
4805 continue;
4806 }
4807
4808 /*
4809 * If the option takes no argument on the command line, we have to
4810 * check if the value is "true". We support non-zero numeric values,
4811 * true, on and yes.
4812 */
4813 if (!long_options[i].has_arg) {
4814 ret = config_parse_value(entry->value);
4815 if (ret <= 0) {
4816 if (ret) {
4817 WARN("Invalid configuration value \"%s\" for option %s",
4818 entry->value, entry->name);
4819 }
4820 /* False, skip boolean config option. */
4821 goto end;
4822 }
4823 }
4824
4825 ret = set_option(long_options[i].val, entry->value, entry->name);
4826 goto end;
4827 }
4828
4829 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4830
4831 end:
4832 return ret;
4833 }
4834
4835 /*
4836 * daemon configuration loading and argument parsing
4837 */
4838 static int set_options(int argc, char **argv)
4839 {
4840 int ret = 0, c = 0, option_index = 0;
4841 int orig_optopt = optopt, orig_optind = optind;
4842 char *optstring;
4843 const char *config_path = NULL;
4844
4845 optstring = utils_generate_optstring(long_options,
4846 sizeof(long_options) / sizeof(struct option));
4847 if (!optstring) {
4848 ret = -ENOMEM;
4849 goto end;
4850 }
4851
4852 /* Check for the --config option */
4853 while ((c = getopt_long(argc, argv, optstring, long_options,
4854 &option_index)) != -1) {
4855 if (c == '?') {
4856 ret = -EINVAL;
4857 goto end;
4858 } else if (c != 'f') {
4859 /* if not equal to --config option. */
4860 continue;
4861 }
4862
4863 if (lttng_is_setuid_setgid()) {
4864 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4865 "-f, --config");
4866 } else {
4867 config_path = utils_expand_path(optarg);
4868 if (!config_path) {
4869 ERR("Failed to resolve path: %s", optarg);
4870 }
4871 }
4872 }
4873
4874 ret = config_get_section_entries(config_path, config_section_name,
4875 config_entry_handler, NULL);
4876 if (ret) {
4877 if (ret > 0) {
4878 ERR("Invalid configuration option at line %i", ret);
4879 ret = -1;
4880 }
4881 goto end;
4882 }
4883
4884 /* Reset getopt's global state */
4885 optopt = orig_optopt;
4886 optind = orig_optind;
4887 while (1) {
4888 option_index = -1;
4889 /*
4890 * getopt_long() will not set option_index if it encounters a
4891 * short option.
4892 */
4893 c = getopt_long(argc, argv, optstring, long_options,
4894 &option_index);
4895 if (c == -1) {
4896 break;
4897 }
4898
4899 /*
4900 * Pass NULL as the long option name if popt left the index
4901 * unset.
4902 */
4903 ret = set_option(c, optarg,
4904 option_index < 0 ? NULL :
4905 long_options[option_index].name);
4906 if (ret < 0) {
4907 break;
4908 }
4909 }
4910
4911 end:
4912 free(optstring);
4913 return ret;
4914 }
4915
4916 /*
4917 * Creates the two needed socket by the daemon.
4918 * apps_sock - The communication socket for all UST apps.
4919 * client_sock - The communication of the cli tool (lttng).
4920 */
4921 static int init_daemon_socket(void)
4922 {
4923 int ret = 0;
4924 mode_t old_umask;
4925
4926 old_umask = umask(0);
4927
4928 /* Create client tool unix socket */
4929 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4930 if (client_sock < 0) {
4931 ERR("Create unix sock failed: %s", client_unix_sock_path);
4932 ret = -1;
4933 goto end;
4934 }
4935
4936 /* Set the cloexec flag */
4937 ret = utils_set_fd_cloexec(client_sock);
4938 if (ret < 0) {
4939 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4940 "Continuing but note that the consumer daemon will have a "
4941 "reference to this socket on exec()", client_sock);
4942 }
4943
4944 /* File permission MUST be 660 */
4945 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4946 if (ret < 0) {
4947 ERR("Set file permissions failed: %s", client_unix_sock_path);
4948 PERROR("chmod");
4949 goto end;
4950 }
4951
4952 /* Create the application unix socket */
4953 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4954 if (apps_sock < 0) {
4955 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4956 ret = -1;
4957 goto end;
4958 }
4959
4960 /* Set the cloexec flag */
4961 ret = utils_set_fd_cloexec(apps_sock);
4962 if (ret < 0) {
4963 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4964 "Continuing but note that the consumer daemon will have a "
4965 "reference to this socket on exec()", apps_sock);
4966 }
4967
4968 /* File permission MUST be 666 */
4969 ret = chmod(apps_unix_sock_path,
4970 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4971 if (ret < 0) {
4972 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4973 PERROR("chmod");
4974 goto end;
4975 }
4976
4977 DBG3("Session daemon client socket %d and application socket %d created",
4978 client_sock, apps_sock);
4979
4980 end:
4981 umask(old_umask);
4982 return ret;
4983 }
4984
4985 /*
4986 * Check if the global socket is available, and if a daemon is answering at the
4987 * other side. If yes, error is returned.
4988 */
4989 static int check_existing_daemon(void)
4990 {
4991 /* Is there anybody out there ? */
4992 if (lttng_session_daemon_alive()) {
4993 return -EEXIST;
4994 }
4995
4996 return 0;
4997 }
4998
4999 /*
5000 * Set the tracing group gid onto the client socket.
5001 *
5002 * Race window between mkdir and chown is OK because we are going from more
5003 * permissive (root.root) to less permissive (root.tracing).
5004 */
5005 static int set_permissions(char *rundir)
5006 {
5007 int ret;
5008 gid_t gid;
5009
5010 gid = utils_get_group_id(tracing_group_name);
5011
5012 /* Set lttng run dir */
5013 ret = chown(rundir, 0, gid);
5014 if (ret < 0) {
5015 ERR("Unable to set group on %s", rundir);
5016 PERROR("chown");
5017 }
5018
5019 /*
5020 * Ensure all applications and tracing group can search the run
5021 * dir. Allow everyone to read the directory, since it does not
5022 * buy us anything to hide its content.
5023 */
5024 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5025 if (ret < 0) {
5026 ERR("Unable to set permissions on %s", rundir);
5027 PERROR("chmod");
5028 }
5029
5030 /* lttng client socket path */
5031 ret = chown(client_unix_sock_path, 0, gid);
5032 if (ret < 0) {
5033 ERR("Unable to set group on %s", client_unix_sock_path);
5034 PERROR("chown");
5035 }
5036
5037 /* kconsumer error socket path */
5038 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5039 if (ret < 0) {
5040 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5041 PERROR("chown");
5042 }
5043
5044 /* 64-bit ustconsumer error socket path */
5045 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5046 if (ret < 0) {
5047 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5048 PERROR("chown");
5049 }
5050
5051 /* 32-bit ustconsumer compat32 error socket path */
5052 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5053 if (ret < 0) {
5054 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5055 PERROR("chown");
5056 }
5057
5058 DBG("All permissions are set");
5059
5060 return ret;
5061 }
5062
5063 /*
5064 * Create the lttng run directory needed for all global sockets and pipe.
5065 */
5066 static int create_lttng_rundir(const char *rundir)
5067 {
5068 int ret;
5069
5070 DBG3("Creating LTTng run directory: %s", rundir);
5071
5072 ret = mkdir(rundir, S_IRWXU);
5073 if (ret < 0) {
5074 if (errno != EEXIST) {
5075 ERR("Unable to create %s", rundir);
5076 goto error;
5077 } else {
5078 ret = 0;
5079 }
5080 }
5081
5082 error:
5083 return ret;
5084 }
5085
5086 /*
5087 * Setup sockets and directory needed by the kconsumerd communication with the
5088 * session daemon.
5089 */
5090 static int set_consumer_sockets(struct consumer_data *consumer_data,
5091 const char *rundir)
5092 {
5093 int ret;
5094 char path[PATH_MAX];
5095
5096 switch (consumer_data->type) {
5097 case LTTNG_CONSUMER_KERNEL:
5098 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5099 break;
5100 case LTTNG_CONSUMER64_UST:
5101 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5102 break;
5103 case LTTNG_CONSUMER32_UST:
5104 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5105 break;
5106 default:
5107 ERR("Consumer type unknown");
5108 ret = -EINVAL;
5109 goto error;
5110 }
5111
5112 DBG2("Creating consumer directory: %s", path);
5113
5114 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5115 if (ret < 0) {
5116 if (errno != EEXIST) {
5117 PERROR("mkdir");
5118 ERR("Failed to create %s", path);
5119 goto error;
5120 }
5121 ret = -1;
5122 }
5123 if (is_root) {
5124 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5125 if (ret < 0) {
5126 ERR("Unable to set group on %s", path);
5127 PERROR("chown");
5128 goto error;
5129 }
5130 }
5131
5132 /* Create the kconsumerd error unix socket */
5133 consumer_data->err_sock =
5134 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5135 if (consumer_data->err_sock < 0) {
5136 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5137 ret = -1;
5138 goto error;
5139 }
5140
5141 /*
5142 * Set the CLOEXEC flag. Return code is useless because either way, the
5143 * show must go on.
5144 */
5145 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5146 if (ret < 0) {
5147 PERROR("utils_set_fd_cloexec");
5148 /* continue anyway */
5149 }
5150
5151 /* File permission MUST be 660 */
5152 ret = chmod(consumer_data->err_unix_sock_path,
5153 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5154 if (ret < 0) {
5155 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5156 PERROR("chmod");
5157 goto error;
5158 }
5159
5160 error:
5161 return ret;
5162 }
5163
5164 /*
5165 * Signal handler for the daemon
5166 *
5167 * Simply stop all worker threads, leaving main() return gracefully after
5168 * joining all threads and calling cleanup().
5169 */
5170 static void sighandler(int sig)
5171 {
5172 switch (sig) {
5173 case SIGPIPE:
5174 DBG("SIGPIPE caught");
5175 return;
5176 case SIGINT:
5177 DBG("SIGINT caught");
5178 stop_threads();
5179 break;
5180 case SIGTERM:
5181 DBG("SIGTERM caught");
5182 stop_threads();
5183 break;
5184 case SIGUSR1:
5185 CMM_STORE_SHARED(recv_child_signal, 1);
5186 break;
5187 default:
5188 break;
5189 }
5190 }
5191
5192 /*
5193 * Setup signal handler for :
5194 * SIGINT, SIGTERM, SIGPIPE
5195 */
5196 static int set_signal_handler(void)
5197 {
5198 int ret = 0;
5199 struct sigaction sa;
5200 sigset_t sigset;
5201
5202 if ((ret = sigemptyset(&sigset)) < 0) {
5203 PERROR("sigemptyset");
5204 return ret;
5205 }
5206
5207 sa.sa_handler = sighandler;
5208 sa.sa_mask = sigset;
5209 sa.sa_flags = 0;
5210 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5211 PERROR("sigaction");
5212 return ret;
5213 }
5214
5215 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5216 PERROR("sigaction");
5217 return ret;
5218 }
5219
5220 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5221 PERROR("sigaction");
5222 return ret;
5223 }
5224
5225 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5226 PERROR("sigaction");
5227 return ret;
5228 }
5229
5230 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5231
5232 return ret;
5233 }
5234
5235 /*
5236 * Set open files limit to unlimited. This daemon can open a large number of
5237 * file descriptors in order to consumer multiple kernel traces.
5238 */
5239 static void set_ulimit(void)
5240 {
5241 int ret;
5242 struct rlimit lim;
5243
5244 /* The kernel does not allowed an infinite limit for open files */
5245 lim.rlim_cur = 65535;
5246 lim.rlim_max = 65535;
5247
5248 ret = setrlimit(RLIMIT_NOFILE, &lim);
5249 if (ret < 0) {
5250 PERROR("failed to set open files limit");
5251 }
5252 }
5253
5254 /*
5255 * Write pidfile using the rundir and opt_pidfile.
5256 */
5257 static int write_pidfile(void)
5258 {
5259 int ret;
5260 char pidfile_path[PATH_MAX];
5261
5262 assert(rundir);
5263
5264 if (opt_pidfile) {
5265 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5266 } else {
5267 /* Build pidfile path from rundir and opt_pidfile. */
5268 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5269 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5270 if (ret < 0) {
5271 PERROR("snprintf pidfile path");
5272 goto error;
5273 }
5274 }
5275
5276 /*
5277 * Create pid file in rundir.
5278 */
5279 ret = utils_create_pid_file(getpid(), pidfile_path);
5280 error:
5281 return ret;
5282 }
5283
5284 /*
5285 * Create lockfile using the rundir and return its fd.
5286 */
5287 static int create_lockfile(void)
5288 {
5289 int ret;
5290 char lockfile_path[PATH_MAX];
5291
5292 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5293 if (ret < 0) {
5294 goto error;
5295 }
5296
5297 ret = utils_create_lock_file(lockfile_path);
5298 error:
5299 return ret;
5300 }
5301
5302 /*
5303 * Write agent TCP port using the rundir.
5304 */
5305 static int write_agent_port(void)
5306 {
5307 int ret;
5308 char path[PATH_MAX];
5309
5310 assert(rundir);
5311
5312 ret = snprintf(path, sizeof(path), "%s/"
5313 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5314 if (ret < 0) {
5315 PERROR("snprintf agent port path");
5316 goto error;
5317 }
5318
5319 /*
5320 * Create TCP agent port file in rundir.
5321 */
5322 ret = utils_create_pid_file(agent_tcp_port, path);
5323
5324 error:
5325 return ret;
5326 }
5327
5328 /*
5329 * main
5330 */
5331 int main(int argc, char **argv)
5332 {
5333 int ret = 0, retval = 0;
5334 void *status;
5335 const char *home_path, *env_app_timeout;
5336
5337 /* Initialize agent apps ht global variable */
5338 agent_apps_ht_by_sock = NULL;
5339
5340 init_kernel_workarounds();
5341
5342 rcu_register_thread();
5343
5344 if (set_signal_handler()) {
5345 retval = -1;
5346 goto exit_set_signal_handler;
5347 }
5348
5349 setup_consumerd_path();
5350
5351 page_size = sysconf(_SC_PAGESIZE);
5352 if (page_size < 0) {
5353 PERROR("sysconf _SC_PAGESIZE");
5354 page_size = LONG_MAX;
5355 WARN("Fallback page size to %ld", page_size);
5356 }
5357
5358 /*
5359 * Parse arguments and load the daemon configuration file.
5360 *
5361 * We have an exit_options exit path to free memory reserved by
5362 * set_options. This is needed because the rest of sessiond_cleanup()
5363 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5364 * depends on set_options.
5365 */
5366 progname = argv[0];
5367 if (set_options(argc, argv)) {
5368 retval = -1;
5369 goto exit_options;
5370 }
5371
5372 /* Daemonize */
5373 if (opt_daemon || opt_background) {
5374 int i;
5375
5376 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5377 !opt_background);
5378 if (ret < 0) {
5379 retval = -1;
5380 goto exit_options;
5381 }
5382
5383 /*
5384 * We are in the child. Make sure all other file descriptors are
5385 * closed, in case we are called with more opened file
5386 * descriptors than the standard ones.
5387 */
5388 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5389 (void) close(i);
5390 }
5391 }
5392
5393 /*
5394 * Starting from here, we can create threads. This needs to be after
5395 * lttng_daemonize due to RCU.
5396 */
5397
5398 /*
5399 * Initialize the health check subsystem. This call should set the
5400 * appropriate time values.
5401 */
5402 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5403 if (!health_sessiond) {
5404 PERROR("health_app_create error");
5405 retval = -1;
5406 goto exit_health_sessiond_cleanup;
5407 }
5408
5409 if (init_ht_cleanup_quit_pipe()) {
5410 retval = -1;
5411 goto exit_ht_cleanup_quit_pipe;
5412 }
5413
5414 /* Setup the thread ht_cleanup communication pipe. */
5415 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5416 retval = -1;
5417 goto exit_ht_cleanup_pipe;
5418 }
5419
5420 /* Set up max poll set size */
5421 if (lttng_poll_set_max_size()) {
5422 retval = -1;
5423 goto exit_set_max_size;
5424 }
5425
5426 /* Create thread to clean up RCU hash tables */
5427 ret = pthread_create(&ht_cleanup_thread, NULL,
5428 thread_ht_cleanup, (void *) NULL);
5429 if (ret) {
5430 errno = ret;
5431 PERROR("pthread_create ht_cleanup");
5432 retval = -1;
5433 goto exit_ht_cleanup;
5434 }
5435
5436 /* Create thread quit pipe */
5437 if (init_thread_quit_pipe()) {
5438 retval = -1;
5439 goto exit_init_data;
5440 }
5441
5442 /* Check if daemon is UID = 0 */
5443 is_root = !getuid();
5444
5445 if (is_root) {
5446 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5447 if (!rundir) {
5448 retval = -1;
5449 goto exit_init_data;
5450 }
5451
5452 /* Create global run dir with root access */
5453 if (create_lttng_rundir(rundir)) {
5454 retval = -1;
5455 goto exit_init_data;
5456 }
5457
5458 if (strlen(apps_unix_sock_path) == 0) {
5459 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5460 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5461 if (ret < 0) {
5462 retval = -1;
5463 goto exit_init_data;
5464 }
5465 }
5466
5467 if (strlen(client_unix_sock_path) == 0) {
5468 ret = snprintf(client_unix_sock_path, PATH_MAX,
5469 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5470 if (ret < 0) {
5471 retval = -1;
5472 goto exit_init_data;
5473 }
5474 }
5475
5476 /* Set global SHM for ust */
5477 if (strlen(wait_shm_path) == 0) {
5478 ret = snprintf(wait_shm_path, PATH_MAX,
5479 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5480 if (ret < 0) {
5481 retval = -1;
5482 goto exit_init_data;
5483 }
5484 }
5485
5486 if (strlen(health_unix_sock_path) == 0) {
5487 ret = snprintf(health_unix_sock_path,
5488 sizeof(health_unix_sock_path),
5489 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5490 if (ret < 0) {
5491 retval = -1;
5492 goto exit_init_data;
5493 }
5494 }
5495
5496 /* Setup kernel consumerd path */
5497 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5498 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5499 if (ret < 0) {
5500 retval = -1;
5501 goto exit_init_data;
5502 }
5503 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5504 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5505 if (ret < 0) {
5506 retval = -1;
5507 goto exit_init_data;
5508 }
5509
5510 DBG2("Kernel consumer err path: %s",
5511 kconsumer_data.err_unix_sock_path);
5512 DBG2("Kernel consumer cmd path: %s",
5513 kconsumer_data.cmd_unix_sock_path);
5514 } else {
5515 home_path = utils_get_home_dir();
5516 if (home_path == NULL) {
5517 /* TODO: Add --socket PATH option */
5518 ERR("Can't get HOME directory for sockets creation.");
5519 retval = -1;
5520 goto exit_init_data;
5521 }
5522
5523 /*
5524 * Create rundir from home path. This will create something like
5525 * $HOME/.lttng
5526 */
5527 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5528 if (ret < 0) {
5529 retval = -1;
5530 goto exit_init_data;
5531 }
5532
5533 if (create_lttng_rundir(rundir)) {
5534 retval = -1;
5535 goto exit_init_data;
5536 }
5537
5538 if (strlen(apps_unix_sock_path) == 0) {
5539 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5540 DEFAULT_HOME_APPS_UNIX_SOCK,
5541 home_path);
5542 if (ret < 0) {
5543 retval = -1;
5544 goto exit_init_data;
5545 }
5546 }
5547
5548 /* Set the cli tool unix socket path */
5549 if (strlen(client_unix_sock_path) == 0) {
5550 ret = snprintf(client_unix_sock_path, PATH_MAX,
5551 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5552 home_path);
5553 if (ret < 0) {
5554 retval = -1;
5555 goto exit_init_data;
5556 }
5557 }
5558
5559 /* Set global SHM for ust */
5560 if (strlen(wait_shm_path) == 0) {
5561 ret = snprintf(wait_shm_path, PATH_MAX,
5562 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5563 getuid());
5564 if (ret < 0) {
5565 retval = -1;
5566 goto exit_init_data;
5567 }
5568 }
5569
5570 /* Set health check Unix path */
5571 if (strlen(health_unix_sock_path) == 0) {
5572 ret = snprintf(health_unix_sock_path,
5573 sizeof(health_unix_sock_path),
5574 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5575 home_path);
5576 if (ret < 0) {
5577 retval = -1;
5578 goto exit_init_data;
5579 }
5580 }
5581 }
5582
5583 lockfile_fd = create_lockfile();
5584 if (lockfile_fd < 0) {
5585 retval = -1;
5586 goto exit_init_data;
5587 }
5588
5589 /* Set consumer initial state */
5590 kernel_consumerd_state = CONSUMER_STOPPED;
5591 ust_consumerd_state = CONSUMER_STOPPED;
5592
5593 DBG("Client socket path %s", client_unix_sock_path);
5594 DBG("Application socket path %s", apps_unix_sock_path);
5595 DBG("Application wait path %s", wait_shm_path);
5596 DBG("LTTng run directory path: %s", rundir);
5597
5598 /* 32 bits consumerd path setup */
5599 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5600 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5601 if (ret < 0) {
5602 PERROR("snprintf 32-bit consumer error socket path");
5603 retval = -1;
5604 goto exit_init_data;
5605 }
5606 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5607 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5608 if (ret < 0) {
5609 PERROR("snprintf 32-bit consumer command socket path");
5610 retval = -1;
5611 goto exit_init_data;
5612 }
5613
5614 DBG2("UST consumer 32 bits err path: %s",
5615 ustconsumer32_data.err_unix_sock_path);
5616 DBG2("UST consumer 32 bits cmd path: %s",
5617 ustconsumer32_data.cmd_unix_sock_path);
5618
5619 /* 64 bits consumerd path setup */
5620 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5621 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5622 if (ret < 0) {
5623 PERROR("snprintf 64-bit consumer error socket path");
5624 retval = -1;
5625 goto exit_init_data;
5626 }
5627 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5628 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5629 if (ret < 0) {
5630 PERROR("snprintf 64-bit consumer command socket path");
5631 retval = -1;
5632 goto exit_init_data;
5633 }
5634
5635 DBG2("UST consumer 64 bits err path: %s",
5636 ustconsumer64_data.err_unix_sock_path);
5637 DBG2("UST consumer 64 bits cmd path: %s",
5638 ustconsumer64_data.cmd_unix_sock_path);
5639
5640 /*
5641 * See if daemon already exist.
5642 */
5643 if (check_existing_daemon()) {
5644 ERR("Already running daemon.\n");
5645 /*
5646 * We do not goto exit because we must not cleanup()
5647 * because a daemon is already running.
5648 */
5649 retval = -1;
5650 goto exit_init_data;
5651 }
5652
5653 /*
5654 * Init UST app hash table. Alloc hash table before this point since
5655 * cleanup() can get called after that point.
5656 */
5657 if (ust_app_ht_alloc()) {
5658 ERR("Failed to allocate UST app hash table");
5659 retval = -1;
5660 goto exit_init_data;
5661 }
5662
5663 /*
5664 * Initialize agent app hash table. We allocate the hash table here
5665 * since cleanup() can get called after this point.
5666 */
5667 if (agent_app_ht_alloc()) {
5668 ERR("Failed to allocate Agent app hash table");
5669 retval = -1;
5670 goto exit_init_data;
5671 }
5672
5673 /*
5674 * These actions must be executed as root. We do that *after* setting up
5675 * the sockets path because we MUST make the check for another daemon using
5676 * those paths *before* trying to set the kernel consumer sockets and init
5677 * kernel tracer.
5678 */
5679 if (is_root) {
5680 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5681 retval = -1;
5682 goto exit_init_data;
5683 }
5684
5685 /* Setup kernel tracer */
5686 if (!opt_no_kernel) {
5687 init_kernel_tracer();
5688 if (kernel_tracer_fd >= 0) {
5689 ret = syscall_init_table();
5690 if (ret < 0) {
5691 ERR("Unable to populate syscall table. "
5692 "Syscall tracing won't work "
5693 "for this session daemon.");
5694 }
5695 }
5696 }
5697
5698 /* Set ulimit for open files */
5699 set_ulimit();
5700 }
5701 /* init lttng_fd tracking must be done after set_ulimit. */
5702 lttng_fd_init();
5703
5704 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5705 retval = -1;
5706 goto exit_init_data;
5707 }
5708
5709 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5710 retval = -1;
5711 goto exit_init_data;
5712 }
5713
5714 /* Setup the needed unix socket */
5715 if (init_daemon_socket()) {
5716 retval = -1;
5717 goto exit_init_data;
5718 }
5719
5720 /* Set credentials to socket */
5721 if (is_root && set_permissions(rundir)) {
5722 retval = -1;
5723 goto exit_init_data;
5724 }
5725
5726 /* Get parent pid if -S, --sig-parent is specified. */
5727 if (opt_sig_parent) {
5728 ppid = getppid();
5729 }
5730
5731 /* Setup the kernel pipe for waking up the kernel thread */
5732 if (is_root && !opt_no_kernel) {
5733 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5734 retval = -1;
5735 goto exit_init_data;
5736 }
5737 }
5738
5739 /* Setup the thread apps communication pipe. */
5740 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5741 retval = -1;
5742 goto exit_init_data;
5743 }
5744
5745 /* Setup the thread apps notify communication pipe. */
5746 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5747 retval = -1;
5748 goto exit_init_data;
5749 }
5750
5751 /* Initialize global buffer per UID and PID registry. */
5752 buffer_reg_init_uid_registry();
5753 buffer_reg_init_pid_registry();
5754
5755 /* Init UST command queue. */
5756 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5757
5758 /*
5759 * Get session list pointer. This pointer MUST NOT be free'd. This list
5760 * is statically declared in session.c
5761 */
5762 session_list_ptr = session_get_list();
5763
5764 cmd_init();
5765
5766 /* Check for the application socket timeout env variable. */
5767 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5768 if (env_app_timeout) {
5769 app_socket_timeout = atoi(env_app_timeout);
5770 } else {
5771 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5772 }
5773
5774 ret = write_pidfile();
5775 if (ret) {
5776 ERR("Error in write_pidfile");
5777 retval = -1;
5778 goto exit_init_data;
5779 }
5780 ret = write_agent_port();
5781 if (ret) {
5782 ERR("Error in write_agent_port");
5783 retval = -1;
5784 goto exit_init_data;
5785 }
5786
5787 /* Initialize communication library */
5788 lttcomm_init();
5789 /* Initialize TCP timeout values */
5790 lttcomm_inet_init();
5791
5792 if (load_session_init_data(&load_info) < 0) {
5793 retval = -1;
5794 goto exit_init_data;
5795 }
5796 load_info->path = opt_load_session_path;
5797
5798 /* Create health-check thread */
5799 ret = pthread_create(&health_thread, NULL,
5800 thread_manage_health, (void *) NULL);
5801 if (ret) {
5802 errno = ret;
5803 PERROR("pthread_create health");
5804 retval = -1;
5805 goto exit_health;
5806 }
5807
5808 /* Create thread to manage the client socket */
5809 ret = pthread_create(&client_thread, NULL,
5810 thread_manage_clients, (void *) NULL);
5811 if (ret) {
5812 errno = ret;
5813 PERROR("pthread_create clients");
5814 retval = -1;
5815 goto exit_client;
5816 }
5817
5818 /* Create thread to dispatch registration */
5819 ret = pthread_create(&dispatch_thread, NULL,
5820 thread_dispatch_ust_registration, (void *) NULL);
5821 if (ret) {
5822 errno = ret;
5823 PERROR("pthread_create dispatch");
5824 retval = -1;
5825 goto exit_dispatch;
5826 }
5827
5828 /* Create thread to manage application registration. */
5829 ret = pthread_create(&reg_apps_thread, NULL,
5830 thread_registration_apps, (void *) NULL);
5831 if (ret) {
5832 errno = ret;
5833 PERROR("pthread_create registration");
5834 retval = -1;
5835 goto exit_reg_apps;
5836 }
5837
5838 /* Create thread to manage application socket */
5839 ret = pthread_create(&apps_thread, NULL,
5840 thread_manage_apps, (void *) NULL);
5841 if (ret) {
5842 errno = ret;
5843 PERROR("pthread_create apps");
5844 retval = -1;
5845 goto exit_apps;
5846 }
5847
5848 /* Create thread to manage application notify socket */
5849 ret = pthread_create(&apps_notify_thread, NULL,
5850 ust_thread_manage_notify, (void *) NULL);
5851 if (ret) {
5852 errno = ret;
5853 PERROR("pthread_create notify");
5854 retval = -1;
5855 goto exit_apps_notify;
5856 }
5857
5858 /* Create agent registration thread. */
5859 ret = pthread_create(&agent_reg_thread, NULL,
5860 agent_thread_manage_registration, (void *) NULL);
5861 if (ret) {
5862 errno = ret;
5863 PERROR("pthread_create agent");
5864 retval = -1;
5865 goto exit_agent_reg;
5866 }
5867
5868 /* Don't start this thread if kernel tracing is not requested nor root */
5869 if (is_root && !opt_no_kernel) {
5870 /* Create kernel thread to manage kernel event */
5871 ret = pthread_create(&kernel_thread, NULL,
5872 thread_manage_kernel, (void *) NULL);
5873 if (ret) {
5874 errno = ret;
5875 PERROR("pthread_create kernel");
5876 retval = -1;
5877 goto exit_kernel;
5878 }
5879 }
5880
5881 /* Create session loading thread. */
5882 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5883 load_info);
5884 if (ret) {
5885 errno = ret;
5886 PERROR("pthread_create load_session_thread");
5887 retval = -1;
5888 goto exit_load_session;
5889 }
5890
5891 /*
5892 * This is where we start awaiting program completion (e.g. through
5893 * signal that asks threads to teardown).
5894 */
5895
5896 ret = pthread_join(load_session_thread, &status);
5897 if (ret) {
5898 errno = ret;
5899 PERROR("pthread_join load_session_thread");
5900 retval = -1;
5901 }
5902 exit_load_session:
5903
5904 if (is_root && !opt_no_kernel) {
5905 ret = pthread_join(kernel_thread, &status);
5906 if (ret) {
5907 errno = ret;
5908 PERROR("pthread_join");
5909 retval = -1;
5910 }
5911 }
5912 exit_kernel:
5913
5914 ret = pthread_join(agent_reg_thread, &status);
5915 if (ret) {
5916 errno = ret;
5917 PERROR("pthread_join agent");
5918 retval = -1;
5919 }
5920 exit_agent_reg:
5921
5922 ret = pthread_join(apps_notify_thread, &status);
5923 if (ret) {
5924 errno = ret;
5925 PERROR("pthread_join apps notify");
5926 retval = -1;
5927 }
5928 exit_apps_notify:
5929
5930 ret = pthread_join(apps_thread, &status);
5931 if (ret) {
5932 errno = ret;
5933 PERROR("pthread_join apps");
5934 retval = -1;
5935 }
5936 exit_apps:
5937
5938 ret = pthread_join(reg_apps_thread, &status);
5939 if (ret) {
5940 errno = ret;
5941 PERROR("pthread_join");
5942 retval = -1;
5943 }
5944 exit_reg_apps:
5945
5946 ret = pthread_join(dispatch_thread, &status);
5947 if (ret) {
5948 errno = ret;
5949 PERROR("pthread_join");
5950 retval = -1;
5951 }
5952 exit_dispatch:
5953
5954 ret = pthread_join(client_thread, &status);
5955 if (ret) {
5956 errno = ret;
5957 PERROR("pthread_join");
5958 retval = -1;
5959 }
5960 exit_client:
5961
5962 ret = pthread_join(health_thread, &status);
5963 if (ret) {
5964 errno = ret;
5965 PERROR("pthread_join health thread");
5966 retval = -1;
5967 }
5968 exit_health:
5969
5970 exit_init_data:
5971 /*
5972 * sessiond_cleanup() is called when no other thread is running, except
5973 * the ht_cleanup thread, which is needed to destroy the hash tables.
5974 */
5975 rcu_thread_online();
5976 sessiond_cleanup();
5977 rcu_thread_offline();
5978 rcu_unregister_thread();
5979
5980 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
5981 if (ret < 0) {
5982 ERR("write error on ht_cleanup quit pipe");
5983 retval = -1;
5984 }
5985
5986 ret = pthread_join(ht_cleanup_thread, &status);
5987 if (ret) {
5988 errno = ret;
5989 PERROR("pthread_join ht cleanup thread");
5990 retval = -1;
5991 }
5992 exit_ht_cleanup:
5993 exit_set_max_size:
5994
5995 utils_close_pipe(ht_cleanup_pipe);
5996 exit_ht_cleanup_pipe:
5997
5998 /*
5999 * Close the ht_cleanup quit pipe.
6000 */
6001 utils_close_pipe(ht_cleanup_quit_pipe);
6002 exit_ht_cleanup_quit_pipe:
6003
6004 health_app_destroy(health_sessiond);
6005 exit_health_sessiond_cleanup:
6006
6007 exit_options:
6008 sessiond_cleanup_options();
6009
6010 exit_set_signal_handler:
6011 if (!retval) {
6012 exit(EXIT_SUCCESS);
6013 } else {
6014 exit(EXIT_FAILURE);
6015 }
6016 }
This page took 0.142216 seconds and 5 git commands to generate.