Fix: Discard disable event command filter payload
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _GNU_SOURCE
21 #define _LGPL_SOURCE
22 #include <getopt.h>
23 #include <grp.h>
24 #include <limits.h>
25 #include <paths.h>
26 #include <pthread.h>
27 #include <signal.h>
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <string.h>
31 #include <inttypes.h>
32 #include <sys/mman.h>
33 #include <sys/mount.h>
34 #include <sys/resource.h>
35 #include <sys/socket.h>
36 #include <sys/stat.h>
37 #include <sys/types.h>
38 #include <sys/wait.h>
39 #include <urcu/uatomic.h>
40 #include <unistd.h>
41 #include <config.h>
42
43 #include <common/common.h>
44 #include <common/compat/socket.h>
45 #include <common/compat/getenv.h>
46 #include <common/defaults.h>
47 #include <common/kernel-consumer/kernel-consumer.h>
48 #include <common/futex.h>
49 #include <common/relayd/relayd.h>
50 #include <common/utils.h>
51 #include <common/daemonize.h>
52 #include <common/config/config.h>
53
54 #include "lttng-sessiond.h"
55 #include "buffer-registry.h"
56 #include "channel.h"
57 #include "cmd.h"
58 #include "consumer.h"
59 #include "context.h"
60 #include "event.h"
61 #include "kernel.h"
62 #include "kernel-consumer.h"
63 #include "modprobe.h"
64 #include "shm.h"
65 #include "ust-ctl.h"
66 #include "ust-consumer.h"
67 #include "utils.h"
68 #include "fd-limit.h"
69 #include "health-sessiond.h"
70 #include "testpoint.h"
71 #include "ust-thread.h"
72 #include "agent-thread.h"
73 #include "save.h"
74 #include "load-session-thread.h"
75 #include "syscall.h"
76
77 #define CONSUMERD_FILE "lttng-consumerd"
78
79 const char *progname;
80 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
81 static int tracing_group_name_override;
82 static char *opt_pidfile;
83 static int opt_sig_parent;
84 static int opt_verbose_consumer;
85 static int opt_daemon, opt_background;
86 static int opt_no_kernel;
87 static char *opt_load_session_path;
88 static pid_t ppid; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
90 static char *rundir;
91 static int lockfile_fd = -1;
92
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal;
95
96 /*
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
99 */
100 static struct consumer_data kconsumer_data = {
101 .type = LTTNG_CONSUMER_KERNEL,
102 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
103 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
104 .err_sock = -1,
105 .cmd_sock = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer64_data = {
112 .type = LTTNG_CONSUMER64_UST,
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
118 .lock = PTHREAD_MUTEX_INITIALIZER,
119 .cond = PTHREAD_COND_INITIALIZER,
120 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
121 };
122 static struct consumer_data ustconsumer32_data = {
123 .type = LTTNG_CONSUMER32_UST,
124 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
125 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
126 .err_sock = -1,
127 .cmd_sock = -1,
128 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
129 .lock = PTHREAD_MUTEX_INITIALIZER,
130 .cond = PTHREAD_COND_INITIALIZER,
131 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
132 };
133
134 /* Command line options */
135 static const struct option long_options[] = {
136 { "client-sock", required_argument, 0, 'c' },
137 { "apps-sock", required_argument, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
144 { "consumerd32-path", required_argument, 0, '\0' },
145 { "consumerd32-libdir", required_argument, 0, '\0' },
146 { "consumerd64-path", required_argument, 0, '\0' },
147 { "consumerd64-libdir", required_argument, 0, '\0' },
148 { "daemonize", no_argument, 0, 'd' },
149 { "background", no_argument, 0, 'b' },
150 { "sig-parent", no_argument, 0, 'S' },
151 { "help", no_argument, 0, 'h' },
152 { "group", required_argument, 0, 'g' },
153 { "version", no_argument, 0, 'V' },
154 { "quiet", no_argument, 0, 'q' },
155 { "verbose", no_argument, 0, 'v' },
156 { "verbose-consumer", no_argument, 0, '\0' },
157 { "no-kernel", no_argument, 0, '\0' },
158 { "pidfile", required_argument, 0, 'p' },
159 { "agent-tcp-port", required_argument, 0, '\0' },
160 { "config", required_argument, 0, 'f' },
161 { "load", required_argument, 0, 'l' },
162 { "kmod-probes", required_argument, 0, '\0' },
163 { "extra-kmod-probes", required_argument, 0, '\0' },
164 { NULL, 0, 0, 0 }
165 };
166
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options[] = { "help", "version", "config" };
169
170 /* Shared between threads */
171 static int dispatch_thread_exit;
172
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path[PATH_MAX];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path[PATH_MAX];
177 /* global wait shm path for UST */
178 static char wait_shm_path[PATH_MAX];
179 /* Global health check unix path */
180 static char health_unix_sock_path[PATH_MAX];
181
182 /* Sockets and FDs */
183 static int client_sock = -1;
184 static int apps_sock = -1;
185 int kernel_tracer_fd = -1;
186 static int kernel_poll_pipe[2] = { -1, -1 };
187
188 /*
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
191 */
192 static int thread_quit_pipe[2] = { -1, -1 };
193 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
194
195 /*
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
198 */
199 static int apps_cmd_pipe[2] = { -1, -1 };
200
201 int apps_cmd_notify_pipe[2] = { -1, -1 };
202
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread;
205 static pthread_t apps_notify_thread;
206 static pthread_t reg_apps_thread;
207 static pthread_t client_thread;
208 static pthread_t kernel_thread;
209 static pthread_t dispatch_thread;
210 static pthread_t health_thread;
211 static pthread_t ht_cleanup_thread;
212 static pthread_t agent_reg_thread;
213 static pthread_t load_session_thread;
214
215 /*
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
218 *
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
223 */
224 static struct ust_cmd_queue ust_cmd_queue;
225
226 /*
227 * Pointer initialized before thread creation.
228 *
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
232 *
233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
234 * session_lock_list and session_unlock_list for lock acquisition.
235 */
236 static struct ltt_session_list *session_list_ptr;
237
238 int ust_consumerd64_fd = -1;
239 int ust_consumerd32_fd = -1;
240
241 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
242 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
243 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
244 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
245 static int consumerd32_bin_override;
246 static int consumerd64_bin_override;
247 static int consumerd32_libdir_override;
248 static int consumerd64_libdir_override;
249
250 static const char *module_proc_lttng = "/proc/lttng";
251
252 /*
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
255 */
256 enum consumerd_state {
257 CONSUMER_STARTED = 1,
258 CONSUMER_STOPPED = 2,
259 CONSUMER_ERROR = 3,
260 };
261
262 /*
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
267 *
268 * The following example shows a possible race condition of this scheme:
269 *
270 * consumer thread error happens
271 * client cmd arrives
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
275 * ...
276 *
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
282 */
283 static enum consumerd_state ust_consumerd_state;
284 static enum consumerd_state kernel_consumerd_state;
285
286 /*
287 * Socket timeout for receiving and sending in seconds.
288 */
289 static int app_socket_timeout;
290
291 /* Set in main() with the current page size. */
292 long page_size;
293
294 /* Application health monitoring */
295 struct health_app *health_sessiond;
296
297 /* Agent TCP port for registration. Used by the agent thread. */
298 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
299
300 /* Am I root or not. */
301 int is_root; /* Set to 1 if the daemon is running as root */
302
303 const char * const config_section_name = "sessiond";
304
305 /* Load session thread information to operate. */
306 struct load_session_thread_data *load_info;
307
308 /*
309 * Whether sessiond is ready for commands/health check requests.
310 * NR_LTTNG_SESSIOND_READY must match the number of calls to
311 * sessiond_notify_ready().
312 */
313 #define NR_LTTNG_SESSIOND_READY 3
314 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
315
316 /* Notify parents that we are ready for cmd and health check */
317 LTTNG_HIDDEN
318 void sessiond_notify_ready(void)
319 {
320 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
321 /*
322 * Notify parent pid that we are ready to accept command
323 * for client side. This ppid is the one from the
324 * external process that spawned us.
325 */
326 if (opt_sig_parent) {
327 kill(ppid, SIGUSR1);
328 }
329
330 /*
331 * Notify the parent of the fork() process that we are
332 * ready.
333 */
334 if (opt_daemon || opt_background) {
335 kill(child_ppid, SIGUSR1);
336 }
337 }
338 }
339
340 static
341 void setup_consumerd_path(void)
342 {
343 const char *bin, *libdir;
344
345 /*
346 * Allow INSTALL_BIN_PATH to be used as a target path for the
347 * native architecture size consumer if CONFIG_CONSUMER*_PATH
348 * has not been defined.
349 */
350 #if (CAA_BITS_PER_LONG == 32)
351 if (!consumerd32_bin[0]) {
352 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
353 }
354 if (!consumerd32_libdir[0]) {
355 consumerd32_libdir = INSTALL_LIB_PATH;
356 }
357 #elif (CAA_BITS_PER_LONG == 64)
358 if (!consumerd64_bin[0]) {
359 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
360 }
361 if (!consumerd64_libdir[0]) {
362 consumerd64_libdir = INSTALL_LIB_PATH;
363 }
364 #else
365 #error "Unknown bitness"
366 #endif
367
368 /*
369 * runtime env. var. overrides the build default.
370 */
371 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
372 if (bin) {
373 consumerd32_bin = bin;
374 }
375 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
376 if (bin) {
377 consumerd64_bin = bin;
378 }
379 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
380 if (libdir) {
381 consumerd32_libdir = libdir;
382 }
383 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
384 if (libdir) {
385 consumerd64_libdir = libdir;
386 }
387 }
388
389 static
390 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
391 int *a_pipe)
392 {
393 int ret;
394
395 assert(events);
396
397 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
398 if (ret < 0) {
399 goto error;
400 }
401
402 /* Add quit pipe */
403 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
404 if (ret < 0) {
405 goto error;
406 }
407
408 return 0;
409
410 error:
411 return ret;
412 }
413
414 /*
415 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
416 */
417 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
418 {
419 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
420 }
421
422 /*
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
424 */
425 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
426 size_t size)
427 {
428 return __sessiond_set_thread_pollset(events, size,
429 ht_cleanup_quit_pipe);
430 }
431
432 static
433 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
434 {
435 if (fd == a_pipe && (events & LPOLLIN)) {
436 return 1;
437 }
438 return 0;
439 }
440
441 /*
442 * Check if the thread quit pipe was triggered.
443 *
444 * Return 1 if it was triggered else 0;
445 */
446 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
447 {
448 return __sessiond_check_thread_quit_pipe(fd, events,
449 thread_quit_pipe[0]);
450 }
451
452 /*
453 * Check if the ht_cleanup thread quit pipe was triggered.
454 *
455 * Return 1 if it was triggered else 0;
456 */
457 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
458 {
459 return __sessiond_check_thread_quit_pipe(fd, events,
460 ht_cleanup_quit_pipe[0]);
461 }
462
463 /*
464 * Init thread quit pipe.
465 *
466 * Return -1 on error or 0 if all pipes are created.
467 */
468 static int __init_thread_quit_pipe(int *a_pipe)
469 {
470 int ret, i;
471
472 ret = pipe(a_pipe);
473 if (ret < 0) {
474 PERROR("thread quit pipe");
475 goto error;
476 }
477
478 for (i = 0; i < 2; i++) {
479 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
480 if (ret < 0) {
481 PERROR("fcntl");
482 goto error;
483 }
484 }
485
486 error:
487 return ret;
488 }
489
490 static int init_thread_quit_pipe(void)
491 {
492 return __init_thread_quit_pipe(thread_quit_pipe);
493 }
494
495 static int init_ht_cleanup_quit_pipe(void)
496 {
497 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
498 }
499
500 /*
501 * Stop all threads by closing the thread quit pipe.
502 */
503 static void stop_threads(void)
504 {
505 int ret;
506
507 /* Stopping all threads */
508 DBG("Terminating all threads");
509 ret = notify_thread_pipe(thread_quit_pipe[1]);
510 if (ret < 0) {
511 ERR("write error on thread quit pipe");
512 }
513
514 /* Dispatch thread */
515 CMM_STORE_SHARED(dispatch_thread_exit, 1);
516 futex_nto1_wake(&ust_cmd_queue.futex);
517 }
518
519 /*
520 * Close every consumer sockets.
521 */
522 static void close_consumer_sockets(void)
523 {
524 int ret;
525
526 if (kconsumer_data.err_sock >= 0) {
527 ret = close(kconsumer_data.err_sock);
528 if (ret < 0) {
529 PERROR("kernel consumer err_sock close");
530 }
531 }
532 if (ustconsumer32_data.err_sock >= 0) {
533 ret = close(ustconsumer32_data.err_sock);
534 if (ret < 0) {
535 PERROR("UST consumerd32 err_sock close");
536 }
537 }
538 if (ustconsumer64_data.err_sock >= 0) {
539 ret = close(ustconsumer64_data.err_sock);
540 if (ret < 0) {
541 PERROR("UST consumerd64 err_sock close");
542 }
543 }
544 if (kconsumer_data.cmd_sock >= 0) {
545 ret = close(kconsumer_data.cmd_sock);
546 if (ret < 0) {
547 PERROR("kernel consumer cmd_sock close");
548 }
549 }
550 if (ustconsumer32_data.cmd_sock >= 0) {
551 ret = close(ustconsumer32_data.cmd_sock);
552 if (ret < 0) {
553 PERROR("UST consumerd32 cmd_sock close");
554 }
555 }
556 if (ustconsumer64_data.cmd_sock >= 0) {
557 ret = close(ustconsumer64_data.cmd_sock);
558 if (ret < 0) {
559 PERROR("UST consumerd64 cmd_sock close");
560 }
561 }
562 }
563
564 /*
565 * Generate the full lock file path using the rundir.
566 *
567 * Return the snprintf() return value thus a negative value is an error.
568 */
569 static int generate_lock_file_path(char *path, size_t len)
570 {
571 int ret;
572
573 assert(path);
574 assert(rundir);
575
576 /* Build lockfile path from rundir. */
577 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
578 if (ret < 0) {
579 PERROR("snprintf lockfile path");
580 }
581
582 return ret;
583 }
584
585 /*
586 * Cleanup the session daemon's data structures.
587 */
588 static void sessiond_cleanup(void)
589 {
590 int ret;
591 struct ltt_session *sess, *stmp;
592 char path[PATH_MAX];
593
594 DBG("Cleanup sessiond");
595
596 /*
597 * Close the thread quit pipe. It has already done its job,
598 * since we are now called.
599 */
600 utils_close_pipe(thread_quit_pipe);
601
602 /*
603 * If opt_pidfile is undefined, the default file will be wiped when
604 * removing the rundir.
605 */
606 if (opt_pidfile) {
607 ret = remove(opt_pidfile);
608 if (ret < 0) {
609 PERROR("remove pidfile %s", opt_pidfile);
610 }
611 }
612
613 DBG("Removing sessiond and consumerd content of directory %s", rundir);
614
615 /* sessiond */
616 snprintf(path, PATH_MAX,
617 "%s/%s",
618 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
619 DBG("Removing %s", path);
620 (void) unlink(path);
621
622 snprintf(path, PATH_MAX, "%s/%s", rundir,
623 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
624 DBG("Removing %s", path);
625 (void) unlink(path);
626
627 /* kconsumerd */
628 snprintf(path, PATH_MAX,
629 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
630 rundir);
631 DBG("Removing %s", path);
632 (void) unlink(path);
633
634 snprintf(path, PATH_MAX,
635 DEFAULT_KCONSUMERD_PATH,
636 rundir);
637 DBG("Removing directory %s", path);
638 (void) rmdir(path);
639
640 /* ust consumerd 32 */
641 snprintf(path, PATH_MAX,
642 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
643 rundir);
644 DBG("Removing %s", path);
645 (void) unlink(path);
646
647 snprintf(path, PATH_MAX,
648 DEFAULT_USTCONSUMERD32_PATH,
649 rundir);
650 DBG("Removing directory %s", path);
651 (void) rmdir(path);
652
653 /* ust consumerd 64 */
654 snprintf(path, PATH_MAX,
655 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
656 rundir);
657 DBG("Removing %s", path);
658 (void) unlink(path);
659
660 snprintf(path, PATH_MAX,
661 DEFAULT_USTCONSUMERD64_PATH,
662 rundir);
663 DBG("Removing directory %s", path);
664 (void) rmdir(path);
665
666 DBG("Cleaning up all sessions");
667
668 /* Destroy session list mutex */
669 if (session_list_ptr != NULL) {
670 pthread_mutex_destroy(&session_list_ptr->lock);
671
672 /* Cleanup ALL session */
673 cds_list_for_each_entry_safe(sess, stmp,
674 &session_list_ptr->head, list) {
675 cmd_destroy_session(sess, kernel_poll_pipe[1]);
676 }
677 }
678
679 DBG("Cleaning up all agent apps");
680 agent_app_ht_clean();
681
682 DBG("Closing all UST sockets");
683 ust_app_clean_list();
684 buffer_reg_destroy_registries();
685
686 if (is_root && !opt_no_kernel) {
687 DBG2("Closing kernel fd");
688 if (kernel_tracer_fd >= 0) {
689 ret = close(kernel_tracer_fd);
690 if (ret) {
691 PERROR("close");
692 }
693 }
694 DBG("Unloading kernel modules");
695 modprobe_remove_lttng_all();
696 free(syscall_table);
697 }
698
699 close_consumer_sockets();
700
701 if (load_info) {
702 load_session_destroy_data(load_info);
703 free(load_info);
704 }
705
706 /*
707 * Cleanup lock file by deleting it and finaly closing it which will
708 * release the file system lock.
709 */
710 if (lockfile_fd >= 0) {
711 char lockfile_path[PATH_MAX];
712
713 ret = generate_lock_file_path(lockfile_path,
714 sizeof(lockfile_path));
715 if (ret > 0) {
716 ret = remove(lockfile_path);
717 if (ret < 0) {
718 PERROR("remove lock file");
719 }
720 ret = close(lockfile_fd);
721 if (ret < 0) {
722 PERROR("close lock file");
723 }
724 }
725 }
726
727 /*
728 * We do NOT rmdir rundir because there are other processes
729 * using it, for instance lttng-relayd, which can start in
730 * parallel with this teardown.
731 */
732
733 free(rundir);
734 }
735
736 /*
737 * Cleanup the daemon's option data structures.
738 */
739 static void sessiond_cleanup_options(void)
740 {
741 DBG("Cleaning up options");
742
743 /*
744 * If the override option is set, the pointer points to a *non* const
745 * thus freeing it even though the variable type is set to const.
746 */
747 if (tracing_group_name_override) {
748 free((void *) tracing_group_name);
749 }
750 if (consumerd32_bin_override) {
751 free((void *) consumerd32_bin);
752 }
753 if (consumerd64_bin_override) {
754 free((void *) consumerd64_bin);
755 }
756 if (consumerd32_libdir_override) {
757 free((void *) consumerd32_libdir);
758 }
759 if (consumerd64_libdir_override) {
760 free((void *) consumerd64_libdir);
761 }
762
763 free(opt_pidfile);
764 free(opt_load_session_path);
765 free(kmod_probes_list);
766 free(kmod_extra_probes_list);
767
768 /* <fun> */
769 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
770 "Matthew, BEET driven development works!%c[%dm",
771 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
772 /* </fun> */
773 }
774
775 /*
776 * Send data on a unix socket using the liblttsessiondcomm API.
777 *
778 * Return lttcomm error code.
779 */
780 static int send_unix_sock(int sock, void *buf, size_t len)
781 {
782 /* Check valid length */
783 if (len == 0) {
784 return -1;
785 }
786
787 return lttcomm_send_unix_sock(sock, buf, len);
788 }
789
790 /*
791 * Free memory of a command context structure.
792 */
793 static void clean_command_ctx(struct command_ctx **cmd_ctx)
794 {
795 DBG("Clean command context structure");
796 if (*cmd_ctx) {
797 if ((*cmd_ctx)->llm) {
798 free((*cmd_ctx)->llm);
799 }
800 if ((*cmd_ctx)->lsm) {
801 free((*cmd_ctx)->lsm);
802 }
803 free(*cmd_ctx);
804 *cmd_ctx = NULL;
805 }
806 }
807
808 /*
809 * Notify UST applications using the shm mmap futex.
810 */
811 static int notify_ust_apps(int active)
812 {
813 char *wait_shm_mmap;
814
815 DBG("Notifying applications of session daemon state: %d", active);
816
817 /* See shm.c for this call implying mmap, shm and futex calls */
818 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
819 if (wait_shm_mmap == NULL) {
820 goto error;
821 }
822
823 /* Wake waiting process */
824 futex_wait_update((int32_t *) wait_shm_mmap, active);
825
826 /* Apps notified successfully */
827 return 0;
828
829 error:
830 return -1;
831 }
832
833 /*
834 * Setup the outgoing data buffer for the response (llm) by allocating the
835 * right amount of memory and copying the original information from the lsm
836 * structure.
837 *
838 * Return total size of the buffer pointed by buf.
839 */
840 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
841 {
842 int ret, buf_size;
843
844 buf_size = size;
845
846 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
847 if (cmd_ctx->llm == NULL) {
848 PERROR("zmalloc");
849 ret = -ENOMEM;
850 goto error;
851 }
852
853 /* Copy common data */
854 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
855 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
856
857 cmd_ctx->llm->data_size = size;
858 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
859
860 return buf_size;
861
862 error:
863 return ret;
864 }
865
866 /*
867 * Update the kernel poll set of all channel fd available over all tracing
868 * session. Add the wakeup pipe at the end of the set.
869 */
870 static int update_kernel_poll(struct lttng_poll_event *events)
871 {
872 int ret;
873 struct ltt_session *session;
874 struct ltt_kernel_channel *channel;
875
876 DBG("Updating kernel poll set");
877
878 session_lock_list();
879 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
880 session_lock(session);
881 if (session->kernel_session == NULL) {
882 session_unlock(session);
883 continue;
884 }
885
886 cds_list_for_each_entry(channel,
887 &session->kernel_session->channel_list.head, list) {
888 /* Add channel fd to the kernel poll set */
889 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
890 if (ret < 0) {
891 session_unlock(session);
892 goto error;
893 }
894 DBG("Channel fd %d added to kernel set", channel->fd);
895 }
896 session_unlock(session);
897 }
898 session_unlock_list();
899
900 return 0;
901
902 error:
903 session_unlock_list();
904 return -1;
905 }
906
907 /*
908 * Find the channel fd from 'fd' over all tracing session. When found, check
909 * for new channel stream and send those stream fds to the kernel consumer.
910 *
911 * Useful for CPU hotplug feature.
912 */
913 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
914 {
915 int ret = 0;
916 struct ltt_session *session;
917 struct ltt_kernel_session *ksess;
918 struct ltt_kernel_channel *channel;
919
920 DBG("Updating kernel streams for channel fd %d", fd);
921
922 session_lock_list();
923 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
924 session_lock(session);
925 if (session->kernel_session == NULL) {
926 session_unlock(session);
927 continue;
928 }
929 ksess = session->kernel_session;
930
931 cds_list_for_each_entry(channel,
932 &ksess->channel_list.head, list) {
933 struct lttng_ht_iter iter;
934 struct consumer_socket *socket;
935
936 if (channel->fd != fd) {
937 continue;
938 }
939 DBG("Channel found, updating kernel streams");
940 ret = kernel_open_channel_stream(channel);
941 if (ret < 0) {
942 goto error;
943 }
944 /* Update the stream global counter */
945 ksess->stream_count_global += ret;
946
947 /*
948 * Have we already sent fds to the consumer? If yes, it
949 * means that tracing is started so it is safe to send
950 * our updated stream fds.
951 */
952 if (ksess->consumer_fds_sent != 1
953 || ksess->consumer == NULL) {
954 ret = -1;
955 goto error;
956 }
957
958 rcu_read_lock();
959 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
960 &iter.iter, socket, node.node) {
961 pthread_mutex_lock(socket->lock);
962 ret = kernel_consumer_send_channel_stream(socket,
963 channel, ksess,
964 session->output_traces ? 1 : 0);
965 pthread_mutex_unlock(socket->lock);
966 if (ret < 0) {
967 rcu_read_unlock();
968 goto error;
969 }
970 }
971 rcu_read_unlock();
972 }
973 session_unlock(session);
974 }
975 session_unlock_list();
976 return ret;
977
978 error:
979 session_unlock(session);
980 session_unlock_list();
981 return ret;
982 }
983
984 /*
985 * For each tracing session, update newly registered apps. The session list
986 * lock MUST be acquired before calling this.
987 */
988 static void update_ust_app(int app_sock)
989 {
990 struct ltt_session *sess, *stmp;
991
992 /* Consumer is in an ERROR state. Stop any application update. */
993 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
994 /* Stop the update process since the consumer is dead. */
995 return;
996 }
997
998 /* For all tracing session(s) */
999 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1000 struct ust_app *app;
1001
1002 session_lock(sess);
1003 if (!sess->ust_session) {
1004 goto unlock_session;
1005 }
1006
1007 rcu_read_lock();
1008 assert(app_sock >= 0);
1009 app = ust_app_find_by_sock(app_sock);
1010 if (app == NULL) {
1011 /*
1012 * Application can be unregistered before so
1013 * this is possible hence simply stopping the
1014 * update.
1015 */
1016 DBG3("UST app update failed to find app sock %d",
1017 app_sock);
1018 goto unlock_rcu;
1019 }
1020 ust_app_global_update(sess->ust_session, app);
1021 unlock_rcu:
1022 rcu_read_unlock();
1023 unlock_session:
1024 session_unlock(sess);
1025 }
1026 }
1027
1028 /*
1029 * This thread manage event coming from the kernel.
1030 *
1031 * Features supported in this thread:
1032 * -) CPU Hotplug
1033 */
1034 static void *thread_manage_kernel(void *data)
1035 {
1036 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1037 uint32_t revents, nb_fd;
1038 char tmp;
1039 struct lttng_poll_event events;
1040
1041 DBG("[thread] Thread manage kernel started");
1042
1043 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1044
1045 /*
1046 * This first step of the while is to clean this structure which could free
1047 * non NULL pointers so initialize it before the loop.
1048 */
1049 lttng_poll_init(&events);
1050
1051 if (testpoint(sessiond_thread_manage_kernel)) {
1052 goto error_testpoint;
1053 }
1054
1055 health_code_update();
1056
1057 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1058 goto error_testpoint;
1059 }
1060
1061 while (1) {
1062 health_code_update();
1063
1064 if (update_poll_flag == 1) {
1065 /* Clean events object. We are about to populate it again. */
1066 lttng_poll_clean(&events);
1067
1068 ret = sessiond_set_thread_pollset(&events, 2);
1069 if (ret < 0) {
1070 goto error_poll_create;
1071 }
1072
1073 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1074 if (ret < 0) {
1075 goto error;
1076 }
1077
1078 /* This will add the available kernel channel if any. */
1079 ret = update_kernel_poll(&events);
1080 if (ret < 0) {
1081 goto error;
1082 }
1083 update_poll_flag = 0;
1084 }
1085
1086 DBG("Thread kernel polling");
1087
1088 /* Poll infinite value of time */
1089 restart:
1090 health_poll_entry();
1091 ret = lttng_poll_wait(&events, -1);
1092 DBG("Thread kernel return from poll on %d fds",
1093 LTTNG_POLL_GETNB(&events));
1094 health_poll_exit();
1095 if (ret < 0) {
1096 /*
1097 * Restart interrupted system call.
1098 */
1099 if (errno == EINTR) {
1100 goto restart;
1101 }
1102 goto error;
1103 } else if (ret == 0) {
1104 /* Should not happen since timeout is infinite */
1105 ERR("Return value of poll is 0 with an infinite timeout.\n"
1106 "This should not have happened! Continuing...");
1107 continue;
1108 }
1109
1110 nb_fd = ret;
1111
1112 for (i = 0; i < nb_fd; i++) {
1113 /* Fetch once the poll data */
1114 revents = LTTNG_POLL_GETEV(&events, i);
1115 pollfd = LTTNG_POLL_GETFD(&events, i);
1116
1117 health_code_update();
1118
1119 if (!revents) {
1120 /* No activity for this FD (poll implementation). */
1121 continue;
1122 }
1123
1124 /* Thread quit pipe has been closed. Killing thread. */
1125 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1126 if (ret) {
1127 err = 0;
1128 goto exit;
1129 }
1130
1131 /* Check for data on kernel pipe */
1132 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1133 (void) lttng_read(kernel_poll_pipe[0],
1134 &tmp, 1);
1135 /*
1136 * Ret value is useless here, if this pipe gets any actions an
1137 * update is required anyway.
1138 */
1139 update_poll_flag = 1;
1140 continue;
1141 } else {
1142 /*
1143 * New CPU detected by the kernel. Adding kernel stream to
1144 * kernel session and updating the kernel consumer
1145 */
1146 if (revents & LPOLLIN) {
1147 ret = update_kernel_stream(&kconsumer_data, pollfd);
1148 if (ret < 0) {
1149 continue;
1150 }
1151 break;
1152 /*
1153 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1154 * and unregister kernel stream at this point.
1155 */
1156 }
1157 }
1158 }
1159 }
1160
1161 exit:
1162 error:
1163 lttng_poll_clean(&events);
1164 error_poll_create:
1165 error_testpoint:
1166 utils_close_pipe(kernel_poll_pipe);
1167 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1168 if (err) {
1169 health_error();
1170 ERR("Health error occurred in %s", __func__);
1171 WARN("Kernel thread died unexpectedly. "
1172 "Kernel tracing can continue but CPU hotplug is disabled.");
1173 }
1174 health_unregister(health_sessiond);
1175 DBG("Kernel thread dying");
1176 return NULL;
1177 }
1178
1179 /*
1180 * Signal pthread condition of the consumer data that the thread.
1181 */
1182 static void signal_consumer_condition(struct consumer_data *data, int state)
1183 {
1184 pthread_mutex_lock(&data->cond_mutex);
1185
1186 /*
1187 * The state is set before signaling. It can be any value, it's the waiter
1188 * job to correctly interpret this condition variable associated to the
1189 * consumer pthread_cond.
1190 *
1191 * A value of 0 means that the corresponding thread of the consumer data
1192 * was not started. 1 indicates that the thread has started and is ready
1193 * for action. A negative value means that there was an error during the
1194 * thread bootstrap.
1195 */
1196 data->consumer_thread_is_ready = state;
1197 (void) pthread_cond_signal(&data->cond);
1198
1199 pthread_mutex_unlock(&data->cond_mutex);
1200 }
1201
1202 /*
1203 * This thread manage the consumer error sent back to the session daemon.
1204 */
1205 static void *thread_manage_consumer(void *data)
1206 {
1207 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1208 uint32_t revents, nb_fd;
1209 enum lttcomm_return_code code;
1210 struct lttng_poll_event events;
1211 struct consumer_data *consumer_data = data;
1212
1213 DBG("[thread] Manage consumer started");
1214
1215 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1216
1217 health_code_update();
1218
1219 /*
1220 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1221 * metadata_sock. Nothing more will be added to this poll set.
1222 */
1223 ret = sessiond_set_thread_pollset(&events, 3);
1224 if (ret < 0) {
1225 goto error_poll;
1226 }
1227
1228 /*
1229 * The error socket here is already in a listening state which was done
1230 * just before spawning this thread to avoid a race between the consumer
1231 * daemon exec trying to connect and the listen() call.
1232 */
1233 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1234 if (ret < 0) {
1235 goto error;
1236 }
1237
1238 health_code_update();
1239
1240 /* Infinite blocking call, waiting for transmission */
1241 restart:
1242 health_poll_entry();
1243
1244 if (testpoint(sessiond_thread_manage_consumer)) {
1245 goto error;
1246 }
1247
1248 ret = lttng_poll_wait(&events, -1);
1249 health_poll_exit();
1250 if (ret < 0) {
1251 /*
1252 * Restart interrupted system call.
1253 */
1254 if (errno == EINTR) {
1255 goto restart;
1256 }
1257 goto error;
1258 }
1259
1260 nb_fd = ret;
1261
1262 for (i = 0; i < nb_fd; i++) {
1263 /* Fetch once the poll data */
1264 revents = LTTNG_POLL_GETEV(&events, i);
1265 pollfd = LTTNG_POLL_GETFD(&events, i);
1266
1267 health_code_update();
1268
1269 if (!revents) {
1270 /* No activity for this FD (poll implementation). */
1271 continue;
1272 }
1273
1274 /* Thread quit pipe has been closed. Killing thread. */
1275 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1276 if (ret) {
1277 err = 0;
1278 goto exit;
1279 }
1280
1281 /* Event on the registration socket */
1282 if (pollfd == consumer_data->err_sock) {
1283 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1284 ERR("consumer err socket poll error");
1285 goto error;
1286 }
1287 }
1288 }
1289
1290 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1291 if (sock < 0) {
1292 goto error;
1293 }
1294
1295 /*
1296 * Set the CLOEXEC flag. Return code is useless because either way, the
1297 * show must go on.
1298 */
1299 (void) utils_set_fd_cloexec(sock);
1300
1301 health_code_update();
1302
1303 DBG2("Receiving code from consumer err_sock");
1304
1305 /* Getting status code from kconsumerd */
1306 ret = lttcomm_recv_unix_sock(sock, &code,
1307 sizeof(enum lttcomm_return_code));
1308 if (ret <= 0) {
1309 goto error;
1310 }
1311
1312 health_code_update();
1313 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1314 /* Connect both socket, command and metadata. */
1315 consumer_data->cmd_sock =
1316 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1317 consumer_data->metadata_fd =
1318 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1319 if (consumer_data->cmd_sock < 0
1320 || consumer_data->metadata_fd < 0) {
1321 PERROR("consumer connect cmd socket");
1322 /* On error, signal condition and quit. */
1323 signal_consumer_condition(consumer_data, -1);
1324 goto error;
1325 }
1326 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1327 /* Create metadata socket lock. */
1328 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1329 if (consumer_data->metadata_sock.lock == NULL) {
1330 PERROR("zmalloc pthread mutex");
1331 ret = -1;
1332 goto error;
1333 }
1334 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1335
1336 signal_consumer_condition(consumer_data, 1);
1337 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1338 DBG("Consumer metadata socket ready (fd: %d)",
1339 consumer_data->metadata_fd);
1340 } else {
1341 ERR("consumer error when waiting for SOCK_READY : %s",
1342 lttcomm_get_readable_code(-code));
1343 goto error;
1344 }
1345
1346 /* Remove the consumerd error sock since we've established a connexion */
1347 ret = lttng_poll_del(&events, consumer_data->err_sock);
1348 if (ret < 0) {
1349 goto error;
1350 }
1351
1352 /* Add new accepted error socket. */
1353 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1354 if (ret < 0) {
1355 goto error;
1356 }
1357
1358 /* Add metadata socket that is successfully connected. */
1359 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1360 LPOLLIN | LPOLLRDHUP);
1361 if (ret < 0) {
1362 goto error;
1363 }
1364
1365 health_code_update();
1366
1367 /* Infinite blocking call, waiting for transmission */
1368 restart_poll:
1369 while (1) {
1370 health_code_update();
1371
1372 /* Exit the thread because the thread quit pipe has been triggered. */
1373 if (should_quit) {
1374 /* Not a health error. */
1375 err = 0;
1376 goto exit;
1377 }
1378
1379 health_poll_entry();
1380 ret = lttng_poll_wait(&events, -1);
1381 health_poll_exit();
1382 if (ret < 0) {
1383 /*
1384 * Restart interrupted system call.
1385 */
1386 if (errno == EINTR) {
1387 goto restart_poll;
1388 }
1389 goto error;
1390 }
1391
1392 nb_fd = ret;
1393
1394 for (i = 0; i < nb_fd; i++) {
1395 /* Fetch once the poll data */
1396 revents = LTTNG_POLL_GETEV(&events, i);
1397 pollfd = LTTNG_POLL_GETFD(&events, i);
1398
1399 health_code_update();
1400
1401 if (!revents) {
1402 /* No activity for this FD (poll implementation). */
1403 continue;
1404 }
1405
1406 /*
1407 * Thread quit pipe has been triggered, flag that we should stop
1408 * but continue the current loop to handle potential data from
1409 * consumer.
1410 */
1411 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1412
1413 if (pollfd == sock) {
1414 /* Event on the consumerd socket */
1415 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1416 ERR("consumer err socket second poll error");
1417 goto error;
1418 }
1419 health_code_update();
1420 /* Wait for any kconsumerd error */
1421 ret = lttcomm_recv_unix_sock(sock, &code,
1422 sizeof(enum lttcomm_return_code));
1423 if (ret <= 0) {
1424 ERR("consumer closed the command socket");
1425 goto error;
1426 }
1427
1428 ERR("consumer return code : %s",
1429 lttcomm_get_readable_code(-code));
1430
1431 goto exit;
1432 } else if (pollfd == consumer_data->metadata_fd) {
1433 /* UST metadata requests */
1434 ret = ust_consumer_metadata_request(
1435 &consumer_data->metadata_sock);
1436 if (ret < 0) {
1437 ERR("Handling metadata request");
1438 goto error;
1439 }
1440 }
1441 /* No need for an else branch all FDs are tested prior. */
1442 }
1443 health_code_update();
1444 }
1445
1446 exit:
1447 error:
1448 /*
1449 * We lock here because we are about to close the sockets and some other
1450 * thread might be using them so get exclusive access which will abort all
1451 * other consumer command by other threads.
1452 */
1453 pthread_mutex_lock(&consumer_data->lock);
1454
1455 /* Immediately set the consumerd state to stopped */
1456 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1457 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1458 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1459 consumer_data->type == LTTNG_CONSUMER32_UST) {
1460 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1461 } else {
1462 /* Code flow error... */
1463 assert(0);
1464 }
1465
1466 if (consumer_data->err_sock >= 0) {
1467 ret = close(consumer_data->err_sock);
1468 if (ret) {
1469 PERROR("close");
1470 }
1471 consumer_data->err_sock = -1;
1472 }
1473 if (consumer_data->cmd_sock >= 0) {
1474 ret = close(consumer_data->cmd_sock);
1475 if (ret) {
1476 PERROR("close");
1477 }
1478 consumer_data->cmd_sock = -1;
1479 }
1480 if (consumer_data->metadata_sock.fd_ptr &&
1481 *consumer_data->metadata_sock.fd_ptr >= 0) {
1482 ret = close(*consumer_data->metadata_sock.fd_ptr);
1483 if (ret) {
1484 PERROR("close");
1485 }
1486 }
1487 if (sock >= 0) {
1488 ret = close(sock);
1489 if (ret) {
1490 PERROR("close");
1491 }
1492 }
1493
1494 unlink(consumer_data->err_unix_sock_path);
1495 unlink(consumer_data->cmd_unix_sock_path);
1496 consumer_data->pid = 0;
1497 pthread_mutex_unlock(&consumer_data->lock);
1498
1499 /* Cleanup metadata socket mutex. */
1500 if (consumer_data->metadata_sock.lock) {
1501 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1502 free(consumer_data->metadata_sock.lock);
1503 }
1504 lttng_poll_clean(&events);
1505 error_poll:
1506 if (err) {
1507 health_error();
1508 ERR("Health error occurred in %s", __func__);
1509 }
1510 health_unregister(health_sessiond);
1511 DBG("consumer thread cleanup completed");
1512
1513 return NULL;
1514 }
1515
1516 /*
1517 * This thread manage application communication.
1518 */
1519 static void *thread_manage_apps(void *data)
1520 {
1521 int i, ret, pollfd, err = -1;
1522 ssize_t size_ret;
1523 uint32_t revents, nb_fd;
1524 struct lttng_poll_event events;
1525
1526 DBG("[thread] Manage application started");
1527
1528 rcu_register_thread();
1529 rcu_thread_online();
1530
1531 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1532
1533 if (testpoint(sessiond_thread_manage_apps)) {
1534 goto error_testpoint;
1535 }
1536
1537 health_code_update();
1538
1539 ret = sessiond_set_thread_pollset(&events, 2);
1540 if (ret < 0) {
1541 goto error_poll_create;
1542 }
1543
1544 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1545 if (ret < 0) {
1546 goto error;
1547 }
1548
1549 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1550 goto error;
1551 }
1552
1553 health_code_update();
1554
1555 while (1) {
1556 DBG("Apps thread polling");
1557
1558 /* Inifinite blocking call, waiting for transmission */
1559 restart:
1560 health_poll_entry();
1561 ret = lttng_poll_wait(&events, -1);
1562 DBG("Apps thread return from poll on %d fds",
1563 LTTNG_POLL_GETNB(&events));
1564 health_poll_exit();
1565 if (ret < 0) {
1566 /*
1567 * Restart interrupted system call.
1568 */
1569 if (errno == EINTR) {
1570 goto restart;
1571 }
1572 goto error;
1573 }
1574
1575 nb_fd = ret;
1576
1577 for (i = 0; i < nb_fd; i++) {
1578 /* Fetch once the poll data */
1579 revents = LTTNG_POLL_GETEV(&events, i);
1580 pollfd = LTTNG_POLL_GETFD(&events, i);
1581
1582 health_code_update();
1583
1584 if (!revents) {
1585 /* No activity for this FD (poll implementation). */
1586 continue;
1587 }
1588
1589 /* Thread quit pipe has been closed. Killing thread. */
1590 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1591 if (ret) {
1592 err = 0;
1593 goto exit;
1594 }
1595
1596 /* Inspect the apps cmd pipe */
1597 if (pollfd == apps_cmd_pipe[0]) {
1598 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1599 ERR("Apps command pipe error");
1600 goto error;
1601 } else if (revents & LPOLLIN) {
1602 int sock;
1603
1604 /* Empty pipe */
1605 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1606 if (size_ret < sizeof(sock)) {
1607 PERROR("read apps cmd pipe");
1608 goto error;
1609 }
1610
1611 health_code_update();
1612
1613 /*
1614 * We only monitor the error events of the socket. This
1615 * thread does not handle any incoming data from UST
1616 * (POLLIN).
1617 */
1618 ret = lttng_poll_add(&events, sock,
1619 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1620 if (ret < 0) {
1621 goto error;
1622 }
1623
1624 DBG("Apps with sock %d added to poll set", sock);
1625 }
1626 } else {
1627 /*
1628 * At this point, we know that a registered application made
1629 * the event at poll_wait.
1630 */
1631 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1632 /* Removing from the poll set */
1633 ret = lttng_poll_del(&events, pollfd);
1634 if (ret < 0) {
1635 goto error;
1636 }
1637
1638 /* Socket closed on remote end. */
1639 ust_app_unregister(pollfd);
1640 }
1641 }
1642
1643 health_code_update();
1644 }
1645 }
1646
1647 exit:
1648 error:
1649 lttng_poll_clean(&events);
1650 error_poll_create:
1651 error_testpoint:
1652 utils_close_pipe(apps_cmd_pipe);
1653 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1654
1655 /*
1656 * We don't clean the UST app hash table here since already registered
1657 * applications can still be controlled so let them be until the session
1658 * daemon dies or the applications stop.
1659 */
1660
1661 if (err) {
1662 health_error();
1663 ERR("Health error occurred in %s", __func__);
1664 }
1665 health_unregister(health_sessiond);
1666 DBG("Application communication apps thread cleanup complete");
1667 rcu_thread_offline();
1668 rcu_unregister_thread();
1669 return NULL;
1670 }
1671
1672 /*
1673 * Send a socket to a thread This is called from the dispatch UST registration
1674 * thread once all sockets are set for the application.
1675 *
1676 * The sock value can be invalid, we don't really care, the thread will handle
1677 * it and make the necessary cleanup if so.
1678 *
1679 * On success, return 0 else a negative value being the errno message of the
1680 * write().
1681 */
1682 static int send_socket_to_thread(int fd, int sock)
1683 {
1684 ssize_t ret;
1685
1686 /*
1687 * It's possible that the FD is set as invalid with -1 concurrently just
1688 * before calling this function being a shutdown state of the thread.
1689 */
1690 if (fd < 0) {
1691 ret = -EBADF;
1692 goto error;
1693 }
1694
1695 ret = lttng_write(fd, &sock, sizeof(sock));
1696 if (ret < sizeof(sock)) {
1697 PERROR("write apps pipe %d", fd);
1698 if (ret < 0) {
1699 ret = -errno;
1700 }
1701 goto error;
1702 }
1703
1704 /* All good. Don't send back the write positive ret value. */
1705 ret = 0;
1706 error:
1707 return (int) ret;
1708 }
1709
1710 /*
1711 * Sanitize the wait queue of the dispatch registration thread meaning removing
1712 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1713 * notify socket is never received.
1714 */
1715 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1716 {
1717 int ret, nb_fd = 0, i;
1718 unsigned int fd_added = 0;
1719 struct lttng_poll_event events;
1720 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1721
1722 assert(wait_queue);
1723
1724 lttng_poll_init(&events);
1725
1726 /* Just skip everything for an empty queue. */
1727 if (!wait_queue->count) {
1728 goto end;
1729 }
1730
1731 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1732 if (ret < 0) {
1733 goto error_create;
1734 }
1735
1736 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1737 &wait_queue->head, head) {
1738 assert(wait_node->app);
1739 ret = lttng_poll_add(&events, wait_node->app->sock,
1740 LPOLLHUP | LPOLLERR);
1741 if (ret < 0) {
1742 goto error;
1743 }
1744
1745 fd_added = 1;
1746 }
1747
1748 if (!fd_added) {
1749 goto end;
1750 }
1751
1752 /*
1753 * Poll but don't block so we can quickly identify the faulty events and
1754 * clean them afterwards from the wait queue.
1755 */
1756 ret = lttng_poll_wait(&events, 0);
1757 if (ret < 0) {
1758 goto error;
1759 }
1760 nb_fd = ret;
1761
1762 for (i = 0; i < nb_fd; i++) {
1763 /* Get faulty FD. */
1764 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1765 int pollfd = LTTNG_POLL_GETFD(&events, i);
1766
1767 if (!revents) {
1768 /* No activity for this FD (poll implementation). */
1769 continue;
1770 }
1771
1772 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1773 &wait_queue->head, head) {
1774 if (pollfd == wait_node->app->sock &&
1775 (revents & (LPOLLHUP | LPOLLERR))) {
1776 cds_list_del(&wait_node->head);
1777 wait_queue->count--;
1778 ust_app_destroy(wait_node->app);
1779 free(wait_node);
1780 break;
1781 }
1782 }
1783 }
1784
1785 if (nb_fd > 0) {
1786 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1787 }
1788
1789 end:
1790 lttng_poll_clean(&events);
1791 return;
1792
1793 error:
1794 lttng_poll_clean(&events);
1795 error_create:
1796 ERR("Unable to sanitize wait queue");
1797 return;
1798 }
1799
1800 /*
1801 * Dispatch request from the registration threads to the application
1802 * communication thread.
1803 */
1804 static void *thread_dispatch_ust_registration(void *data)
1805 {
1806 int ret, err = -1;
1807 struct cds_wfcq_node *node;
1808 struct ust_command *ust_cmd = NULL;
1809 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1810 struct ust_reg_wait_queue wait_queue = {
1811 .count = 0,
1812 };
1813
1814 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1815
1816 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1817 goto error_testpoint;
1818 }
1819
1820 health_code_update();
1821
1822 CDS_INIT_LIST_HEAD(&wait_queue.head);
1823
1824 DBG("[thread] Dispatch UST command started");
1825
1826 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1827 health_code_update();
1828
1829 /* Atomically prepare the queue futex */
1830 futex_nto1_prepare(&ust_cmd_queue.futex);
1831
1832 do {
1833 struct ust_app *app = NULL;
1834 ust_cmd = NULL;
1835
1836 /*
1837 * Make sure we don't have node(s) that have hung up before receiving
1838 * the notify socket. This is to clean the list in order to avoid
1839 * memory leaks from notify socket that are never seen.
1840 */
1841 sanitize_wait_queue(&wait_queue);
1842
1843 health_code_update();
1844 /* Dequeue command for registration */
1845 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1846 if (node == NULL) {
1847 DBG("Woken up but nothing in the UST command queue");
1848 /* Continue thread execution */
1849 break;
1850 }
1851
1852 ust_cmd = caa_container_of(node, struct ust_command, node);
1853
1854 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1855 " gid:%d sock:%d name:%s (version %d.%d)",
1856 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1857 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1858 ust_cmd->sock, ust_cmd->reg_msg.name,
1859 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1860
1861 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1862 wait_node = zmalloc(sizeof(*wait_node));
1863 if (!wait_node) {
1864 PERROR("zmalloc wait_node dispatch");
1865 ret = close(ust_cmd->sock);
1866 if (ret < 0) {
1867 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1868 }
1869 lttng_fd_put(LTTNG_FD_APPS, 1);
1870 free(ust_cmd);
1871 goto error;
1872 }
1873 CDS_INIT_LIST_HEAD(&wait_node->head);
1874
1875 /* Create application object if socket is CMD. */
1876 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1877 ust_cmd->sock);
1878 if (!wait_node->app) {
1879 ret = close(ust_cmd->sock);
1880 if (ret < 0) {
1881 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1882 }
1883 lttng_fd_put(LTTNG_FD_APPS, 1);
1884 free(wait_node);
1885 free(ust_cmd);
1886 continue;
1887 }
1888 /*
1889 * Add application to the wait queue so we can set the notify
1890 * socket before putting this object in the global ht.
1891 */
1892 cds_list_add(&wait_node->head, &wait_queue.head);
1893 wait_queue.count++;
1894
1895 free(ust_cmd);
1896 /*
1897 * We have to continue here since we don't have the notify
1898 * socket and the application MUST be added to the hash table
1899 * only at that moment.
1900 */
1901 continue;
1902 } else {
1903 /*
1904 * Look for the application in the local wait queue and set the
1905 * notify socket if found.
1906 */
1907 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1908 &wait_queue.head, head) {
1909 health_code_update();
1910 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1911 wait_node->app->notify_sock = ust_cmd->sock;
1912 cds_list_del(&wait_node->head);
1913 wait_queue.count--;
1914 app = wait_node->app;
1915 free(wait_node);
1916 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1917 break;
1918 }
1919 }
1920
1921 /*
1922 * With no application at this stage the received socket is
1923 * basically useless so close it before we free the cmd data
1924 * structure for good.
1925 */
1926 if (!app) {
1927 ret = close(ust_cmd->sock);
1928 if (ret < 0) {
1929 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1930 }
1931 lttng_fd_put(LTTNG_FD_APPS, 1);
1932 }
1933 free(ust_cmd);
1934 }
1935
1936 if (app) {
1937 /*
1938 * @session_lock_list
1939 *
1940 * Lock the global session list so from the register up to the
1941 * registration done message, no thread can see the application
1942 * and change its state.
1943 */
1944 session_lock_list();
1945 rcu_read_lock();
1946
1947 /*
1948 * Add application to the global hash table. This needs to be
1949 * done before the update to the UST registry can locate the
1950 * application.
1951 */
1952 ust_app_add(app);
1953
1954 /* Set app version. This call will print an error if needed. */
1955 (void) ust_app_version(app);
1956
1957 /* Send notify socket through the notify pipe. */
1958 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1959 app->notify_sock);
1960 if (ret < 0) {
1961 rcu_read_unlock();
1962 session_unlock_list();
1963 /*
1964 * No notify thread, stop the UST tracing. However, this is
1965 * not an internal error of the this thread thus setting
1966 * the health error code to a normal exit.
1967 */
1968 err = 0;
1969 goto error;
1970 }
1971
1972 /*
1973 * Update newly registered application with the tracing
1974 * registry info already enabled information.
1975 */
1976 update_ust_app(app->sock);
1977
1978 /*
1979 * Don't care about return value. Let the manage apps threads
1980 * handle app unregistration upon socket close.
1981 */
1982 (void) ust_app_register_done(app->sock);
1983
1984 /*
1985 * Even if the application socket has been closed, send the app
1986 * to the thread and unregistration will take place at that
1987 * place.
1988 */
1989 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1990 if (ret < 0) {
1991 rcu_read_unlock();
1992 session_unlock_list();
1993 /*
1994 * No apps. thread, stop the UST tracing. However, this is
1995 * not an internal error of the this thread thus setting
1996 * the health error code to a normal exit.
1997 */
1998 err = 0;
1999 goto error;
2000 }
2001
2002 rcu_read_unlock();
2003 session_unlock_list();
2004 }
2005 } while (node != NULL);
2006
2007 health_poll_entry();
2008 /* Futex wait on queue. Blocking call on futex() */
2009 futex_nto1_wait(&ust_cmd_queue.futex);
2010 health_poll_exit();
2011 }
2012 /* Normal exit, no error */
2013 err = 0;
2014
2015 error:
2016 /* Clean up wait queue. */
2017 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2018 &wait_queue.head, head) {
2019 cds_list_del(&wait_node->head);
2020 wait_queue.count--;
2021 free(wait_node);
2022 }
2023
2024 error_testpoint:
2025 DBG("Dispatch thread dying");
2026 if (err) {
2027 health_error();
2028 ERR("Health error occurred in %s", __func__);
2029 }
2030 health_unregister(health_sessiond);
2031 return NULL;
2032 }
2033
2034 /*
2035 * This thread manage application registration.
2036 */
2037 static void *thread_registration_apps(void *data)
2038 {
2039 int sock = -1, i, ret, pollfd, err = -1;
2040 uint32_t revents, nb_fd;
2041 struct lttng_poll_event events;
2042 /*
2043 * Get allocated in this thread, enqueued to a global queue, dequeued and
2044 * freed in the manage apps thread.
2045 */
2046 struct ust_command *ust_cmd = NULL;
2047
2048 DBG("[thread] Manage application registration started");
2049
2050 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2051
2052 if (testpoint(sessiond_thread_registration_apps)) {
2053 goto error_testpoint;
2054 }
2055
2056 ret = lttcomm_listen_unix_sock(apps_sock);
2057 if (ret < 0) {
2058 goto error_listen;
2059 }
2060
2061 /*
2062 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2063 * more will be added to this poll set.
2064 */
2065 ret = sessiond_set_thread_pollset(&events, 2);
2066 if (ret < 0) {
2067 goto error_create_poll;
2068 }
2069
2070 /* Add the application registration socket */
2071 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2072 if (ret < 0) {
2073 goto error_poll_add;
2074 }
2075
2076 /* Notify all applications to register */
2077 ret = notify_ust_apps(1);
2078 if (ret < 0) {
2079 ERR("Failed to notify applications or create the wait shared memory.\n"
2080 "Execution continues but there might be problem for already\n"
2081 "running applications that wishes to register.");
2082 }
2083
2084 while (1) {
2085 DBG("Accepting application registration");
2086
2087 /* Inifinite blocking call, waiting for transmission */
2088 restart:
2089 health_poll_entry();
2090 ret = lttng_poll_wait(&events, -1);
2091 health_poll_exit();
2092 if (ret < 0) {
2093 /*
2094 * Restart interrupted system call.
2095 */
2096 if (errno == EINTR) {
2097 goto restart;
2098 }
2099 goto error;
2100 }
2101
2102 nb_fd = ret;
2103
2104 for (i = 0; i < nb_fd; i++) {
2105 health_code_update();
2106
2107 /* Fetch once the poll data */
2108 revents = LTTNG_POLL_GETEV(&events, i);
2109 pollfd = LTTNG_POLL_GETFD(&events, i);
2110
2111 if (!revents) {
2112 /* No activity for this FD (poll implementation). */
2113 continue;
2114 }
2115
2116 /* Thread quit pipe has been closed. Killing thread. */
2117 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2118 if (ret) {
2119 err = 0;
2120 goto exit;
2121 }
2122
2123 /* Event on the registration socket */
2124 if (pollfd == apps_sock) {
2125 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2126 ERR("Register apps socket poll error");
2127 goto error;
2128 } else if (revents & LPOLLIN) {
2129 sock = lttcomm_accept_unix_sock(apps_sock);
2130 if (sock < 0) {
2131 goto error;
2132 }
2133
2134 /*
2135 * Set socket timeout for both receiving and ending.
2136 * app_socket_timeout is in seconds, whereas
2137 * lttcomm_setsockopt_rcv_timeout and
2138 * lttcomm_setsockopt_snd_timeout expect msec as
2139 * parameter.
2140 */
2141 (void) lttcomm_setsockopt_rcv_timeout(sock,
2142 app_socket_timeout * 1000);
2143 (void) lttcomm_setsockopt_snd_timeout(sock,
2144 app_socket_timeout * 1000);
2145
2146 /*
2147 * Set the CLOEXEC flag. Return code is useless because
2148 * either way, the show must go on.
2149 */
2150 (void) utils_set_fd_cloexec(sock);
2151
2152 /* Create UST registration command for enqueuing */
2153 ust_cmd = zmalloc(sizeof(struct ust_command));
2154 if (ust_cmd == NULL) {
2155 PERROR("ust command zmalloc");
2156 goto error;
2157 }
2158
2159 /*
2160 * Using message-based transmissions to ensure we don't
2161 * have to deal with partially received messages.
2162 */
2163 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2164 if (ret < 0) {
2165 ERR("Exhausted file descriptors allowed for applications.");
2166 free(ust_cmd);
2167 ret = close(sock);
2168 if (ret) {
2169 PERROR("close");
2170 }
2171 sock = -1;
2172 continue;
2173 }
2174
2175 health_code_update();
2176 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2177 if (ret < 0) {
2178 free(ust_cmd);
2179 /* Close socket of the application. */
2180 ret = close(sock);
2181 if (ret) {
2182 PERROR("close");
2183 }
2184 lttng_fd_put(LTTNG_FD_APPS, 1);
2185 sock = -1;
2186 continue;
2187 }
2188 health_code_update();
2189
2190 ust_cmd->sock = sock;
2191 sock = -1;
2192
2193 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2194 " gid:%d sock:%d name:%s (version %d.%d)",
2195 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2196 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2197 ust_cmd->sock, ust_cmd->reg_msg.name,
2198 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2199
2200 /*
2201 * Lock free enqueue the registration request. The red pill
2202 * has been taken! This apps will be part of the *system*.
2203 */
2204 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2205
2206 /*
2207 * Wake the registration queue futex. Implicit memory
2208 * barrier with the exchange in cds_wfcq_enqueue.
2209 */
2210 futex_nto1_wake(&ust_cmd_queue.futex);
2211 }
2212 }
2213 }
2214 }
2215
2216 exit:
2217 error:
2218 /* Notify that the registration thread is gone */
2219 notify_ust_apps(0);
2220
2221 if (apps_sock >= 0) {
2222 ret = close(apps_sock);
2223 if (ret) {
2224 PERROR("close");
2225 }
2226 }
2227 if (sock >= 0) {
2228 ret = close(sock);
2229 if (ret) {
2230 PERROR("close");
2231 }
2232 lttng_fd_put(LTTNG_FD_APPS, 1);
2233 }
2234 unlink(apps_unix_sock_path);
2235
2236 error_poll_add:
2237 lttng_poll_clean(&events);
2238 error_listen:
2239 error_create_poll:
2240 error_testpoint:
2241 DBG("UST Registration thread cleanup complete");
2242 if (err) {
2243 health_error();
2244 ERR("Health error occurred in %s", __func__);
2245 }
2246 health_unregister(health_sessiond);
2247
2248 return NULL;
2249 }
2250
2251 /*
2252 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2253 * exec or it will fails.
2254 */
2255 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2256 {
2257 int ret, clock_ret;
2258 struct timespec timeout;
2259
2260 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2261 consumer_data->consumer_thread_is_ready = 0;
2262
2263 /* Setup pthread condition */
2264 ret = pthread_condattr_init(&consumer_data->condattr);
2265 if (ret) {
2266 errno = ret;
2267 PERROR("pthread_condattr_init consumer data");
2268 goto error;
2269 }
2270
2271 /*
2272 * Set the monotonic clock in order to make sure we DO NOT jump in time
2273 * between the clock_gettime() call and the timedwait call. See bug #324
2274 * for a more details and how we noticed it.
2275 */
2276 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2277 if (ret) {
2278 errno = ret;
2279 PERROR("pthread_condattr_setclock consumer data");
2280 goto error;
2281 }
2282
2283 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2284 if (ret) {
2285 errno = ret;
2286 PERROR("pthread_cond_init consumer data");
2287 goto error;
2288 }
2289
2290 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2291 consumer_data);
2292 if (ret) {
2293 errno = ret;
2294 PERROR("pthread_create consumer");
2295 ret = -1;
2296 goto error;
2297 }
2298
2299 /* We are about to wait on a pthread condition */
2300 pthread_mutex_lock(&consumer_data->cond_mutex);
2301
2302 /* Get time for sem_timedwait absolute timeout */
2303 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2304 /*
2305 * Set the timeout for the condition timed wait even if the clock gettime
2306 * call fails since we might loop on that call and we want to avoid to
2307 * increment the timeout too many times.
2308 */
2309 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2310
2311 /*
2312 * The following loop COULD be skipped in some conditions so this is why we
2313 * set ret to 0 in order to make sure at least one round of the loop is
2314 * done.
2315 */
2316 ret = 0;
2317
2318 /*
2319 * Loop until the condition is reached or when a timeout is reached. Note
2320 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2321 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2322 * possible. This loop does not take any chances and works with both of
2323 * them.
2324 */
2325 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2326 if (clock_ret < 0) {
2327 PERROR("clock_gettime spawn consumer");
2328 /* Infinite wait for the consumerd thread to be ready */
2329 ret = pthread_cond_wait(&consumer_data->cond,
2330 &consumer_data->cond_mutex);
2331 } else {
2332 ret = pthread_cond_timedwait(&consumer_data->cond,
2333 &consumer_data->cond_mutex, &timeout);
2334 }
2335 }
2336
2337 /* Release the pthread condition */
2338 pthread_mutex_unlock(&consumer_data->cond_mutex);
2339
2340 if (ret != 0) {
2341 errno = ret;
2342 if (ret == ETIMEDOUT) {
2343 int pth_ret;
2344
2345 /*
2346 * Call has timed out so we kill the kconsumerd_thread and return
2347 * an error.
2348 */
2349 ERR("Condition timed out. The consumer thread was never ready."
2350 " Killing it");
2351 pth_ret = pthread_cancel(consumer_data->thread);
2352 if (pth_ret < 0) {
2353 PERROR("pthread_cancel consumer thread");
2354 }
2355 } else {
2356 PERROR("pthread_cond_wait failed consumer thread");
2357 }
2358 /* Caller is expecting a negative value on failure. */
2359 ret = -1;
2360 goto error;
2361 }
2362
2363 pthread_mutex_lock(&consumer_data->pid_mutex);
2364 if (consumer_data->pid == 0) {
2365 ERR("Consumerd did not start");
2366 pthread_mutex_unlock(&consumer_data->pid_mutex);
2367 goto error;
2368 }
2369 pthread_mutex_unlock(&consumer_data->pid_mutex);
2370
2371 return 0;
2372
2373 error:
2374 return ret;
2375 }
2376
2377 /*
2378 * Join consumer thread
2379 */
2380 static int join_consumer_thread(struct consumer_data *consumer_data)
2381 {
2382 void *status;
2383
2384 /* Consumer pid must be a real one. */
2385 if (consumer_data->pid > 0) {
2386 int ret;
2387 ret = kill(consumer_data->pid, SIGTERM);
2388 if (ret) {
2389 PERROR("Error killing consumer daemon");
2390 return ret;
2391 }
2392 return pthread_join(consumer_data->thread, &status);
2393 } else {
2394 return 0;
2395 }
2396 }
2397
2398 /*
2399 * Fork and exec a consumer daemon (consumerd).
2400 *
2401 * Return pid if successful else -1.
2402 */
2403 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2404 {
2405 int ret;
2406 pid_t pid;
2407 const char *consumer_to_use;
2408 const char *verbosity;
2409 struct stat st;
2410
2411 DBG("Spawning consumerd");
2412
2413 pid = fork();
2414 if (pid == 0) {
2415 /*
2416 * Exec consumerd.
2417 */
2418 if (opt_verbose_consumer) {
2419 verbosity = "--verbose";
2420 } else if (lttng_opt_quiet) {
2421 verbosity = "--quiet";
2422 } else {
2423 verbosity = "";
2424 }
2425
2426 switch (consumer_data->type) {
2427 case LTTNG_CONSUMER_KERNEL:
2428 /*
2429 * Find out which consumerd to execute. We will first try the
2430 * 64-bit path, then the sessiond's installation directory, and
2431 * fallback on the 32-bit one,
2432 */
2433 DBG3("Looking for a kernel consumer at these locations:");
2434 DBG3(" 1) %s", consumerd64_bin);
2435 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2436 DBG3(" 3) %s", consumerd32_bin);
2437 if (stat(consumerd64_bin, &st) == 0) {
2438 DBG3("Found location #1");
2439 consumer_to_use = consumerd64_bin;
2440 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2441 DBG3("Found location #2");
2442 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2443 } else if (stat(consumerd32_bin, &st) == 0) {
2444 DBG3("Found location #3");
2445 consumer_to_use = consumerd32_bin;
2446 } else {
2447 DBG("Could not find any valid consumerd executable");
2448 ret = -EINVAL;
2449 break;
2450 }
2451 DBG("Using kernel consumer at: %s", consumer_to_use);
2452 ret = execl(consumer_to_use,
2453 "lttng-consumerd", verbosity, "-k",
2454 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2455 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2456 "--group", tracing_group_name,
2457 NULL);
2458 break;
2459 case LTTNG_CONSUMER64_UST:
2460 {
2461 char *tmpnew = NULL;
2462
2463 if (consumerd64_libdir[0] != '\0') {
2464 char *tmp;
2465 size_t tmplen;
2466
2467 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2468 if (!tmp) {
2469 tmp = "";
2470 }
2471 tmplen = strlen("LD_LIBRARY_PATH=")
2472 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2473 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2474 if (!tmpnew) {
2475 ret = -ENOMEM;
2476 goto error;
2477 }
2478 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2479 strcat(tmpnew, consumerd64_libdir);
2480 if (tmp[0] != '\0') {
2481 strcat(tmpnew, ":");
2482 strcat(tmpnew, tmp);
2483 }
2484 ret = putenv(tmpnew);
2485 if (ret) {
2486 ret = -errno;
2487 free(tmpnew);
2488 goto error;
2489 }
2490 }
2491 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2492 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2493 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2494 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2495 "--group", tracing_group_name,
2496 NULL);
2497 if (consumerd64_libdir[0] != '\0') {
2498 free(tmpnew);
2499 }
2500 break;
2501 }
2502 case LTTNG_CONSUMER32_UST:
2503 {
2504 char *tmpnew = NULL;
2505
2506 if (consumerd32_libdir[0] != '\0') {
2507 char *tmp;
2508 size_t tmplen;
2509
2510 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2511 if (!tmp) {
2512 tmp = "";
2513 }
2514 tmplen = strlen("LD_LIBRARY_PATH=")
2515 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2516 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2517 if (!tmpnew) {
2518 ret = -ENOMEM;
2519 goto error;
2520 }
2521 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2522 strcat(tmpnew, consumerd32_libdir);
2523 if (tmp[0] != '\0') {
2524 strcat(tmpnew, ":");
2525 strcat(tmpnew, tmp);
2526 }
2527 ret = putenv(tmpnew);
2528 if (ret) {
2529 ret = -errno;
2530 free(tmpnew);
2531 goto error;
2532 }
2533 }
2534 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2535 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2536 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2537 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2538 "--group", tracing_group_name,
2539 NULL);
2540 if (consumerd32_libdir[0] != '\0') {
2541 free(tmpnew);
2542 }
2543 break;
2544 }
2545 default:
2546 PERROR("unknown consumer type");
2547 exit(EXIT_FAILURE);
2548 }
2549 if (errno != 0) {
2550 PERROR("Consumer execl()");
2551 }
2552 /* Reaching this point, we got a failure on our execl(). */
2553 exit(EXIT_FAILURE);
2554 } else if (pid > 0) {
2555 ret = pid;
2556 } else {
2557 PERROR("start consumer fork");
2558 ret = -errno;
2559 }
2560 error:
2561 return ret;
2562 }
2563
2564 /*
2565 * Spawn the consumerd daemon and session daemon thread.
2566 */
2567 static int start_consumerd(struct consumer_data *consumer_data)
2568 {
2569 int ret;
2570
2571 /*
2572 * Set the listen() state on the socket since there is a possible race
2573 * between the exec() of the consumer daemon and this call if place in the
2574 * consumer thread. See bug #366 for more details.
2575 */
2576 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2577 if (ret < 0) {
2578 goto error;
2579 }
2580
2581 pthread_mutex_lock(&consumer_data->pid_mutex);
2582 if (consumer_data->pid != 0) {
2583 pthread_mutex_unlock(&consumer_data->pid_mutex);
2584 goto end;
2585 }
2586
2587 ret = spawn_consumerd(consumer_data);
2588 if (ret < 0) {
2589 ERR("Spawning consumerd failed");
2590 pthread_mutex_unlock(&consumer_data->pid_mutex);
2591 goto error;
2592 }
2593
2594 /* Setting up the consumer_data pid */
2595 consumer_data->pid = ret;
2596 DBG2("Consumer pid %d", consumer_data->pid);
2597 pthread_mutex_unlock(&consumer_data->pid_mutex);
2598
2599 DBG2("Spawning consumer control thread");
2600 ret = spawn_consumer_thread(consumer_data);
2601 if (ret < 0) {
2602 ERR("Fatal error spawning consumer control thread");
2603 goto error;
2604 }
2605
2606 end:
2607 return 0;
2608
2609 error:
2610 /* Cleanup already created sockets on error. */
2611 if (consumer_data->err_sock >= 0) {
2612 int err;
2613
2614 err = close(consumer_data->err_sock);
2615 if (err < 0) {
2616 PERROR("close consumer data error socket");
2617 }
2618 }
2619 return ret;
2620 }
2621
2622 /*
2623 * Setup necessary data for kernel tracer action.
2624 */
2625 static int init_kernel_tracer(void)
2626 {
2627 int ret;
2628
2629 /* Modprobe lttng kernel modules */
2630 ret = modprobe_lttng_control();
2631 if (ret < 0) {
2632 goto error;
2633 }
2634
2635 /* Open debugfs lttng */
2636 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2637 if (kernel_tracer_fd < 0) {
2638 DBG("Failed to open %s", module_proc_lttng);
2639 ret = -1;
2640 goto error_open;
2641 }
2642
2643 /* Validate kernel version */
2644 ret = kernel_validate_version(kernel_tracer_fd);
2645 if (ret < 0) {
2646 goto error_version;
2647 }
2648
2649 ret = modprobe_lttng_data();
2650 if (ret < 0) {
2651 goto error_modules;
2652 }
2653
2654 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2655 return 0;
2656
2657 error_version:
2658 modprobe_remove_lttng_control();
2659 ret = close(kernel_tracer_fd);
2660 if (ret) {
2661 PERROR("close");
2662 }
2663 kernel_tracer_fd = -1;
2664 return LTTNG_ERR_KERN_VERSION;
2665
2666 error_modules:
2667 ret = close(kernel_tracer_fd);
2668 if (ret) {
2669 PERROR("close");
2670 }
2671
2672 error_open:
2673 modprobe_remove_lttng_control();
2674
2675 error:
2676 WARN("No kernel tracer available");
2677 kernel_tracer_fd = -1;
2678 if (!is_root) {
2679 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2680 } else {
2681 return LTTNG_ERR_KERN_NA;
2682 }
2683 }
2684
2685
2686 /*
2687 * Copy consumer output from the tracing session to the domain session. The
2688 * function also applies the right modification on a per domain basis for the
2689 * trace files destination directory.
2690 *
2691 * Should *NOT* be called with RCU read-side lock held.
2692 */
2693 static int copy_session_consumer(int domain, struct ltt_session *session)
2694 {
2695 int ret;
2696 const char *dir_name;
2697 struct consumer_output *consumer;
2698
2699 assert(session);
2700 assert(session->consumer);
2701
2702 switch (domain) {
2703 case LTTNG_DOMAIN_KERNEL:
2704 DBG3("Copying tracing session consumer output in kernel session");
2705 /*
2706 * XXX: We should audit the session creation and what this function
2707 * does "extra" in order to avoid a destroy since this function is used
2708 * in the domain session creation (kernel and ust) only. Same for UST
2709 * domain.
2710 */
2711 if (session->kernel_session->consumer) {
2712 consumer_destroy_output(session->kernel_session->consumer);
2713 }
2714 session->kernel_session->consumer =
2715 consumer_copy_output(session->consumer);
2716 /* Ease our life a bit for the next part */
2717 consumer = session->kernel_session->consumer;
2718 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2719 break;
2720 case LTTNG_DOMAIN_JUL:
2721 case LTTNG_DOMAIN_LOG4J:
2722 case LTTNG_DOMAIN_PYTHON:
2723 case LTTNG_DOMAIN_UST:
2724 DBG3("Copying tracing session consumer output in UST session");
2725 if (session->ust_session->consumer) {
2726 consumer_destroy_output(session->ust_session->consumer);
2727 }
2728 session->ust_session->consumer =
2729 consumer_copy_output(session->consumer);
2730 /* Ease our life a bit for the next part */
2731 consumer = session->ust_session->consumer;
2732 dir_name = DEFAULT_UST_TRACE_DIR;
2733 break;
2734 default:
2735 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2736 goto error;
2737 }
2738
2739 /* Append correct directory to subdir */
2740 strncat(consumer->subdir, dir_name,
2741 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2742 DBG3("Copy session consumer subdir %s", consumer->subdir);
2743
2744 ret = LTTNG_OK;
2745
2746 error:
2747 return ret;
2748 }
2749
2750 /*
2751 * Create an UST session and add it to the session ust list.
2752 *
2753 * Should *NOT* be called with RCU read-side lock held.
2754 */
2755 static int create_ust_session(struct ltt_session *session,
2756 struct lttng_domain *domain)
2757 {
2758 int ret;
2759 struct ltt_ust_session *lus = NULL;
2760
2761 assert(session);
2762 assert(domain);
2763 assert(session->consumer);
2764
2765 switch (domain->type) {
2766 case LTTNG_DOMAIN_JUL:
2767 case LTTNG_DOMAIN_LOG4J:
2768 case LTTNG_DOMAIN_PYTHON:
2769 case LTTNG_DOMAIN_UST:
2770 break;
2771 default:
2772 ERR("Unknown UST domain on create session %d", domain->type);
2773 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2774 goto error;
2775 }
2776
2777 DBG("Creating UST session");
2778
2779 lus = trace_ust_create_session(session->id);
2780 if (lus == NULL) {
2781 ret = LTTNG_ERR_UST_SESS_FAIL;
2782 goto error;
2783 }
2784
2785 lus->uid = session->uid;
2786 lus->gid = session->gid;
2787 lus->output_traces = session->output_traces;
2788 lus->snapshot_mode = session->snapshot_mode;
2789 lus->live_timer_interval = session->live_timer;
2790 session->ust_session = lus;
2791 if (session->shm_path[0]) {
2792 strncpy(lus->root_shm_path, session->shm_path,
2793 sizeof(lus->root_shm_path));
2794 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2795 strncpy(lus->shm_path, session->shm_path,
2796 sizeof(lus->shm_path));
2797 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2798 strncat(lus->shm_path, "/ust",
2799 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2800 }
2801 /* Copy session output to the newly created UST session */
2802 ret = copy_session_consumer(domain->type, session);
2803 if (ret != LTTNG_OK) {
2804 goto error;
2805 }
2806
2807 return LTTNG_OK;
2808
2809 error:
2810 free(lus);
2811 session->ust_session = NULL;
2812 return ret;
2813 }
2814
2815 /*
2816 * Create a kernel tracer session then create the default channel.
2817 */
2818 static int create_kernel_session(struct ltt_session *session)
2819 {
2820 int ret;
2821
2822 DBG("Creating kernel session");
2823
2824 ret = kernel_create_session(session, kernel_tracer_fd);
2825 if (ret < 0) {
2826 ret = LTTNG_ERR_KERN_SESS_FAIL;
2827 goto error;
2828 }
2829
2830 /* Code flow safety */
2831 assert(session->kernel_session);
2832
2833 /* Copy session output to the newly created Kernel session */
2834 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2835 if (ret != LTTNG_OK) {
2836 goto error;
2837 }
2838
2839 /* Create directory(ies) on local filesystem. */
2840 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2841 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2842 ret = run_as_mkdir_recursive(
2843 session->kernel_session->consumer->dst.trace_path,
2844 S_IRWXU | S_IRWXG, session->uid, session->gid);
2845 if (ret < 0) {
2846 if (ret != -EEXIST) {
2847 ERR("Trace directory creation error");
2848 goto error;
2849 }
2850 }
2851 }
2852
2853 session->kernel_session->uid = session->uid;
2854 session->kernel_session->gid = session->gid;
2855 session->kernel_session->output_traces = session->output_traces;
2856 session->kernel_session->snapshot_mode = session->snapshot_mode;
2857
2858 return LTTNG_OK;
2859
2860 error:
2861 trace_kernel_destroy_session(session->kernel_session);
2862 session->kernel_session = NULL;
2863 return ret;
2864 }
2865
2866 /*
2867 * Count number of session permitted by uid/gid.
2868 */
2869 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2870 {
2871 unsigned int i = 0;
2872 struct ltt_session *session;
2873
2874 DBG("Counting number of available session for UID %d GID %d",
2875 uid, gid);
2876 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2877 /*
2878 * Only list the sessions the user can control.
2879 */
2880 if (!session_access_ok(session, uid, gid)) {
2881 continue;
2882 }
2883 i++;
2884 }
2885 return i;
2886 }
2887
2888 /*
2889 * Process the command requested by the lttng client within the command
2890 * context structure. This function make sure that the return structure (llm)
2891 * is set and ready for transmission before returning.
2892 *
2893 * Return any error encountered or 0 for success.
2894 *
2895 * "sock" is only used for special-case var. len data.
2896 *
2897 * Should *NOT* be called with RCU read-side lock held.
2898 */
2899 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2900 int *sock_error)
2901 {
2902 int ret = LTTNG_OK;
2903 int need_tracing_session = 1;
2904 int need_domain;
2905
2906 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2907
2908 *sock_error = 0;
2909
2910 switch (cmd_ctx->lsm->cmd_type) {
2911 case LTTNG_CREATE_SESSION:
2912 case LTTNG_CREATE_SESSION_SNAPSHOT:
2913 case LTTNG_CREATE_SESSION_LIVE:
2914 case LTTNG_DESTROY_SESSION:
2915 case LTTNG_LIST_SESSIONS:
2916 case LTTNG_LIST_DOMAINS:
2917 case LTTNG_START_TRACE:
2918 case LTTNG_STOP_TRACE:
2919 case LTTNG_DATA_PENDING:
2920 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2921 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2922 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2923 case LTTNG_SNAPSHOT_RECORD:
2924 case LTTNG_SAVE_SESSION:
2925 case LTTNG_SET_SESSION_SHM_PATH:
2926 need_domain = 0;
2927 break;
2928 default:
2929 need_domain = 1;
2930 }
2931
2932 if (opt_no_kernel && need_domain
2933 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2934 if (!is_root) {
2935 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2936 } else {
2937 ret = LTTNG_ERR_KERN_NA;
2938 }
2939 goto error;
2940 }
2941
2942 /* Deny register consumer if we already have a spawned consumer. */
2943 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2944 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2945 if (kconsumer_data.pid > 0) {
2946 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2947 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2948 goto error;
2949 }
2950 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2951 }
2952
2953 /*
2954 * Check for command that don't needs to allocate a returned payload. We do
2955 * this here so we don't have to make the call for no payload at each
2956 * command.
2957 */
2958 switch(cmd_ctx->lsm->cmd_type) {
2959 case LTTNG_LIST_SESSIONS:
2960 case LTTNG_LIST_TRACEPOINTS:
2961 case LTTNG_LIST_TRACEPOINT_FIELDS:
2962 case LTTNG_LIST_DOMAINS:
2963 case LTTNG_LIST_CHANNELS:
2964 case LTTNG_LIST_EVENTS:
2965 case LTTNG_LIST_SYSCALLS:
2966 case LTTNG_LIST_TRACKER_PIDS:
2967 break;
2968 default:
2969 /* Setup lttng message with no payload */
2970 ret = setup_lttng_msg(cmd_ctx, 0);
2971 if (ret < 0) {
2972 /* This label does not try to unlock the session */
2973 goto init_setup_error;
2974 }
2975 }
2976
2977 /* Commands that DO NOT need a session. */
2978 switch (cmd_ctx->lsm->cmd_type) {
2979 case LTTNG_CREATE_SESSION:
2980 case LTTNG_CREATE_SESSION_SNAPSHOT:
2981 case LTTNG_CREATE_SESSION_LIVE:
2982 case LTTNG_CALIBRATE:
2983 case LTTNG_LIST_SESSIONS:
2984 case LTTNG_LIST_TRACEPOINTS:
2985 case LTTNG_LIST_SYSCALLS:
2986 case LTTNG_LIST_TRACEPOINT_FIELDS:
2987 case LTTNG_SAVE_SESSION:
2988 need_tracing_session = 0;
2989 break;
2990 default:
2991 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2992 /*
2993 * We keep the session list lock across _all_ commands
2994 * for now, because the per-session lock does not
2995 * handle teardown properly.
2996 */
2997 session_lock_list();
2998 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2999 if (cmd_ctx->session == NULL) {
3000 ret = LTTNG_ERR_SESS_NOT_FOUND;
3001 goto error;
3002 } else {
3003 /* Acquire lock for the session */
3004 session_lock(cmd_ctx->session);
3005 }
3006 break;
3007 }
3008
3009 /*
3010 * Commands that need a valid session but should NOT create one if none
3011 * exists. Instead of creating one and destroying it when the command is
3012 * handled, process that right before so we save some round trip in useless
3013 * code path.
3014 */
3015 switch (cmd_ctx->lsm->cmd_type) {
3016 case LTTNG_DISABLE_CHANNEL:
3017 case LTTNG_DISABLE_EVENT:
3018 switch (cmd_ctx->lsm->domain.type) {
3019 case LTTNG_DOMAIN_KERNEL:
3020 if (!cmd_ctx->session->kernel_session) {
3021 ret = LTTNG_ERR_NO_CHANNEL;
3022 goto error;
3023 }
3024 break;
3025 case LTTNG_DOMAIN_JUL:
3026 case LTTNG_DOMAIN_LOG4J:
3027 case LTTNG_DOMAIN_PYTHON:
3028 case LTTNG_DOMAIN_UST:
3029 if (!cmd_ctx->session->ust_session) {
3030 ret = LTTNG_ERR_NO_CHANNEL;
3031 goto error;
3032 }
3033 break;
3034 default:
3035 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3036 goto error;
3037 }
3038 default:
3039 break;
3040 }
3041
3042 if (!need_domain) {
3043 goto skip_domain;
3044 }
3045
3046 /*
3047 * Check domain type for specific "pre-action".
3048 */
3049 switch (cmd_ctx->lsm->domain.type) {
3050 case LTTNG_DOMAIN_KERNEL:
3051 if (!is_root) {
3052 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3053 goto error;
3054 }
3055
3056 /* Kernel tracer check */
3057 if (kernel_tracer_fd == -1) {
3058 /* Basically, load kernel tracer modules */
3059 ret = init_kernel_tracer();
3060 if (ret != 0) {
3061 goto error;
3062 }
3063 }
3064
3065 /* Consumer is in an ERROR state. Report back to client */
3066 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3067 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3068 goto error;
3069 }
3070
3071 /* Need a session for kernel command */
3072 if (need_tracing_session) {
3073 if (cmd_ctx->session->kernel_session == NULL) {
3074 ret = create_kernel_session(cmd_ctx->session);
3075 if (ret < 0) {
3076 ret = LTTNG_ERR_KERN_SESS_FAIL;
3077 goto error;
3078 }
3079 }
3080
3081 /* Start the kernel consumer daemon */
3082 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3083 if (kconsumer_data.pid == 0 &&
3084 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3085 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3086 ret = start_consumerd(&kconsumer_data);
3087 if (ret < 0) {
3088 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3089 goto error;
3090 }
3091 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3092 } else {
3093 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3094 }
3095
3096 /*
3097 * The consumer was just spawned so we need to add the socket to
3098 * the consumer output of the session if exist.
3099 */
3100 ret = consumer_create_socket(&kconsumer_data,
3101 cmd_ctx->session->kernel_session->consumer);
3102 if (ret < 0) {
3103 goto error;
3104 }
3105 }
3106
3107 break;
3108 case LTTNG_DOMAIN_JUL:
3109 case LTTNG_DOMAIN_LOG4J:
3110 case LTTNG_DOMAIN_PYTHON:
3111 case LTTNG_DOMAIN_UST:
3112 {
3113 if (!ust_app_supported()) {
3114 ret = LTTNG_ERR_NO_UST;
3115 goto error;
3116 }
3117 /* Consumer is in an ERROR state. Report back to client */
3118 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3119 ret = LTTNG_ERR_NO_USTCONSUMERD;
3120 goto error;
3121 }
3122
3123 if (need_tracing_session) {
3124 /* Create UST session if none exist. */
3125 if (cmd_ctx->session->ust_session == NULL) {
3126 ret = create_ust_session(cmd_ctx->session,
3127 &cmd_ctx->lsm->domain);
3128 if (ret != LTTNG_OK) {
3129 goto error;
3130 }
3131 }
3132
3133 /* Start the UST consumer daemons */
3134 /* 64-bit */
3135 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3136 if (consumerd64_bin[0] != '\0' &&
3137 ustconsumer64_data.pid == 0 &&
3138 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3139 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3140 ret = start_consumerd(&ustconsumer64_data);
3141 if (ret < 0) {
3142 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3143 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3144 goto error;
3145 }
3146
3147 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3148 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3149 } else {
3150 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3151 }
3152
3153 /*
3154 * Setup socket for consumer 64 bit. No need for atomic access
3155 * since it was set above and can ONLY be set in this thread.
3156 */
3157 ret = consumer_create_socket(&ustconsumer64_data,
3158 cmd_ctx->session->ust_session->consumer);
3159 if (ret < 0) {
3160 goto error;
3161 }
3162
3163 /* 32-bit */
3164 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3165 if (consumerd32_bin[0] != '\0' &&
3166 ustconsumer32_data.pid == 0 &&
3167 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3168 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3169 ret = start_consumerd(&ustconsumer32_data);
3170 if (ret < 0) {
3171 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3172 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3173 goto error;
3174 }
3175
3176 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3177 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3178 } else {
3179 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3180 }
3181
3182 /*
3183 * Setup socket for consumer 64 bit. No need for atomic access
3184 * since it was set above and can ONLY be set in this thread.
3185 */
3186 ret = consumer_create_socket(&ustconsumer32_data,
3187 cmd_ctx->session->ust_session->consumer);
3188 if (ret < 0) {
3189 goto error;
3190 }
3191 }
3192 break;
3193 }
3194 default:
3195 break;
3196 }
3197 skip_domain:
3198
3199 /* Validate consumer daemon state when start/stop trace command */
3200 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3201 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3202 switch (cmd_ctx->lsm->domain.type) {
3203 case LTTNG_DOMAIN_JUL:
3204 case LTTNG_DOMAIN_LOG4J:
3205 case LTTNG_DOMAIN_PYTHON:
3206 case LTTNG_DOMAIN_UST:
3207 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3208 ret = LTTNG_ERR_NO_USTCONSUMERD;
3209 goto error;
3210 }
3211 break;
3212 case LTTNG_DOMAIN_KERNEL:
3213 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3214 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3215 goto error;
3216 }
3217 break;
3218 }
3219 }
3220
3221 /*
3222 * Check that the UID or GID match that of the tracing session.
3223 * The root user can interact with all sessions.
3224 */
3225 if (need_tracing_session) {
3226 if (!session_access_ok(cmd_ctx->session,
3227 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3228 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3229 ret = LTTNG_ERR_EPERM;
3230 goto error;
3231 }
3232 }
3233
3234 /*
3235 * Send relayd information to consumer as soon as we have a domain and a
3236 * session defined.
3237 */
3238 if (cmd_ctx->session && need_domain) {
3239 /*
3240 * Setup relayd if not done yet. If the relayd information was already
3241 * sent to the consumer, this call will gracefully return.
3242 */
3243 ret = cmd_setup_relayd(cmd_ctx->session);
3244 if (ret != LTTNG_OK) {
3245 goto error;
3246 }
3247 }
3248
3249 /* Process by command type */
3250 switch (cmd_ctx->lsm->cmd_type) {
3251 case LTTNG_ADD_CONTEXT:
3252 {
3253 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3254 cmd_ctx->lsm->u.context.channel_name,
3255 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3256 break;
3257 }
3258 case LTTNG_DISABLE_CHANNEL:
3259 {
3260 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3261 cmd_ctx->lsm->u.disable.channel_name);
3262 break;
3263 }
3264 case LTTNG_DISABLE_EVENT:
3265 {
3266
3267 /*
3268 * FIXME: handle filter; for now we just receive the filter's
3269 * bytecode along with the filter expression which are sent by
3270 * liblttng-ctl and discard them.
3271 *
3272 * This fixes an issue where the client may block while sending
3273 * the filter payload and encounter an error because the session
3274 * daemon closes the socket without ever handling this data.
3275 */
3276 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3277 cmd_ctx->lsm->u.disable.bytecode_len;
3278
3279 if (count) {
3280 char data[LTTNG_FILTER_MAX_LEN];
3281
3282 DBG("Discarding disable event command payload of size %zu", count);
3283 while (count) {
3284 ret = lttcomm_recv_unix_sock(sock, data,
3285 count > sizeof(data) ? sizeof(data) : count);
3286 if (ret < 0) {
3287 goto error;
3288 }
3289
3290 count -= (size_t) ret;
3291 }
3292 }
3293 /* FIXME: passing packed structure to non-packed pointer */
3294 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3295 cmd_ctx->lsm->u.disable.channel_name,
3296 &cmd_ctx->lsm->u.disable.event);
3297 break;
3298 }
3299 case LTTNG_ENABLE_CHANNEL:
3300 {
3301 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3302 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3303 break;
3304 }
3305 case LTTNG_TRACK_PID:
3306 {
3307 ret = cmd_track_pid(cmd_ctx->session,
3308 cmd_ctx->lsm->domain.type,
3309 cmd_ctx->lsm->u.pid_tracker.pid);
3310 break;
3311 }
3312 case LTTNG_UNTRACK_PID:
3313 {
3314 ret = cmd_untrack_pid(cmd_ctx->session,
3315 cmd_ctx->lsm->domain.type,
3316 cmd_ctx->lsm->u.pid_tracker.pid);
3317 break;
3318 }
3319 case LTTNG_ENABLE_EVENT:
3320 {
3321 struct lttng_event_exclusion *exclusion = NULL;
3322 struct lttng_filter_bytecode *bytecode = NULL;
3323 char *filter_expression = NULL;
3324
3325 /* Handle exclusion events and receive it from the client. */
3326 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3327 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3328
3329 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3330 (count * LTTNG_SYMBOL_NAME_LEN));
3331 if (!exclusion) {
3332 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3333 goto error;
3334 }
3335
3336 DBG("Receiving var len exclusion event list from client ...");
3337 exclusion->count = count;
3338 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3339 count * LTTNG_SYMBOL_NAME_LEN);
3340 if (ret <= 0) {
3341 DBG("Nothing recv() from client var len data... continuing");
3342 *sock_error = 1;
3343 free(exclusion);
3344 ret = LTTNG_ERR_EXCLUSION_INVAL;
3345 goto error;
3346 }
3347 }
3348
3349 /* Get filter expression from client. */
3350 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3351 size_t expression_len =
3352 cmd_ctx->lsm->u.enable.expression_len;
3353
3354 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3355 ret = LTTNG_ERR_FILTER_INVAL;
3356 free(exclusion);
3357 goto error;
3358 }
3359
3360 filter_expression = zmalloc(expression_len);
3361 if (!filter_expression) {
3362 free(exclusion);
3363 ret = LTTNG_ERR_FILTER_NOMEM;
3364 goto error;
3365 }
3366
3367 /* Receive var. len. data */
3368 DBG("Receiving var len filter's expression from client ...");
3369 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3370 expression_len);
3371 if (ret <= 0) {
3372 DBG("Nothing recv() from client car len data... continuing");
3373 *sock_error = 1;
3374 free(filter_expression);
3375 free(exclusion);
3376 ret = LTTNG_ERR_FILTER_INVAL;
3377 goto error;
3378 }
3379 }
3380
3381 /* Handle filter and get bytecode from client. */
3382 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3383 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3384
3385 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3386 ret = LTTNG_ERR_FILTER_INVAL;
3387 free(filter_expression);
3388 free(exclusion);
3389 goto error;
3390 }
3391
3392 bytecode = zmalloc(bytecode_len);
3393 if (!bytecode) {
3394 free(filter_expression);
3395 free(exclusion);
3396 ret = LTTNG_ERR_FILTER_NOMEM;
3397 goto error;
3398 }
3399
3400 /* Receive var. len. data */
3401 DBG("Receiving var len filter's bytecode from client ...");
3402 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3403 if (ret <= 0) {
3404 DBG("Nothing recv() from client car len data... continuing");
3405 *sock_error = 1;
3406 free(filter_expression);
3407 free(bytecode);
3408 free(exclusion);
3409 ret = LTTNG_ERR_FILTER_INVAL;
3410 goto error;
3411 }
3412
3413 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3414 free(filter_expression);
3415 free(bytecode);
3416 free(exclusion);
3417 ret = LTTNG_ERR_FILTER_INVAL;
3418 goto error;
3419 }
3420 }
3421
3422 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3423 cmd_ctx->lsm->u.enable.channel_name,
3424 &cmd_ctx->lsm->u.enable.event,
3425 filter_expression, bytecode, exclusion,
3426 kernel_poll_pipe[1]);
3427 break;
3428 }
3429 case LTTNG_LIST_TRACEPOINTS:
3430 {
3431 struct lttng_event *events;
3432 ssize_t nb_events;
3433
3434 session_lock_list();
3435 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3436 session_unlock_list();
3437 if (nb_events < 0) {
3438 /* Return value is a negative lttng_error_code. */
3439 ret = -nb_events;
3440 goto error;
3441 }
3442
3443 /*
3444 * Setup lttng message with payload size set to the event list size in
3445 * bytes and then copy list into the llm payload.
3446 */
3447 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3448 if (ret < 0) {
3449 free(events);
3450 goto setup_error;
3451 }
3452
3453 /* Copy event list into message payload */
3454 memcpy(cmd_ctx->llm->payload, events,
3455 sizeof(struct lttng_event) * nb_events);
3456
3457 free(events);
3458
3459 ret = LTTNG_OK;
3460 break;
3461 }
3462 case LTTNG_LIST_TRACEPOINT_FIELDS:
3463 {
3464 struct lttng_event_field *fields;
3465 ssize_t nb_fields;
3466
3467 session_lock_list();
3468 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3469 &fields);
3470 session_unlock_list();
3471 if (nb_fields < 0) {
3472 /* Return value is a negative lttng_error_code. */
3473 ret = -nb_fields;
3474 goto error;
3475 }
3476
3477 /*
3478 * Setup lttng message with payload size set to the event list size in
3479 * bytes and then copy list into the llm payload.
3480 */
3481 ret = setup_lttng_msg(cmd_ctx,
3482 sizeof(struct lttng_event_field) * nb_fields);
3483 if (ret < 0) {
3484 free(fields);
3485 goto setup_error;
3486 }
3487
3488 /* Copy event list into message payload */
3489 memcpy(cmd_ctx->llm->payload, fields,
3490 sizeof(struct lttng_event_field) * nb_fields);
3491
3492 free(fields);
3493
3494 ret = LTTNG_OK;
3495 break;
3496 }
3497 case LTTNG_LIST_SYSCALLS:
3498 {
3499 struct lttng_event *events;
3500 ssize_t nb_events;
3501
3502 nb_events = cmd_list_syscalls(&events);
3503 if (nb_events < 0) {
3504 /* Return value is a negative lttng_error_code. */
3505 ret = -nb_events;
3506 goto error;
3507 }
3508
3509 /*
3510 * Setup lttng message with payload size set to the event list size in
3511 * bytes and then copy list into the llm payload.
3512 */
3513 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3514 if (ret < 0) {
3515 free(events);
3516 goto setup_error;
3517 }
3518
3519 /* Copy event list into message payload */
3520 memcpy(cmd_ctx->llm->payload, events,
3521 sizeof(struct lttng_event) * nb_events);
3522
3523 free(events);
3524
3525 ret = LTTNG_OK;
3526 break;
3527 }
3528 case LTTNG_LIST_TRACKER_PIDS:
3529 {
3530 int32_t *pids = NULL;
3531 ssize_t nr_pids;
3532
3533 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3534 cmd_ctx->lsm->domain.type, &pids);
3535 if (nr_pids < 0) {
3536 /* Return value is a negative lttng_error_code. */
3537 ret = -nr_pids;
3538 goto error;
3539 }
3540
3541 /*
3542 * Setup lttng message with payload size set to the event list size in
3543 * bytes and then copy list into the llm payload.
3544 */
3545 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3546 if (ret < 0) {
3547 free(pids);
3548 goto setup_error;
3549 }
3550
3551 /* Copy event list into message payload */
3552 memcpy(cmd_ctx->llm->payload, pids,
3553 sizeof(int) * nr_pids);
3554
3555 free(pids);
3556
3557 ret = LTTNG_OK;
3558 break;
3559 }
3560 case LTTNG_SET_CONSUMER_URI:
3561 {
3562 size_t nb_uri, len;
3563 struct lttng_uri *uris;
3564
3565 nb_uri = cmd_ctx->lsm->u.uri.size;
3566 len = nb_uri * sizeof(struct lttng_uri);
3567
3568 if (nb_uri == 0) {
3569 ret = LTTNG_ERR_INVALID;
3570 goto error;
3571 }
3572
3573 uris = zmalloc(len);
3574 if (uris == NULL) {
3575 ret = LTTNG_ERR_FATAL;
3576 goto error;
3577 }
3578
3579 /* Receive variable len data */
3580 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3581 ret = lttcomm_recv_unix_sock(sock, uris, len);
3582 if (ret <= 0) {
3583 DBG("No URIs received from client... continuing");
3584 *sock_error = 1;
3585 ret = LTTNG_ERR_SESSION_FAIL;
3586 free(uris);
3587 goto error;
3588 }
3589
3590 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3591 free(uris);
3592 if (ret != LTTNG_OK) {
3593 goto error;
3594 }
3595
3596
3597 break;
3598 }
3599 case LTTNG_START_TRACE:
3600 {
3601 ret = cmd_start_trace(cmd_ctx->session);
3602 break;
3603 }
3604 case LTTNG_STOP_TRACE:
3605 {
3606 ret = cmd_stop_trace(cmd_ctx->session);
3607 break;
3608 }
3609 case LTTNG_CREATE_SESSION:
3610 {
3611 size_t nb_uri, len;
3612 struct lttng_uri *uris = NULL;
3613
3614 nb_uri = cmd_ctx->lsm->u.uri.size;
3615 len = nb_uri * sizeof(struct lttng_uri);
3616
3617 if (nb_uri > 0) {
3618 uris = zmalloc(len);
3619 if (uris == NULL) {
3620 ret = LTTNG_ERR_FATAL;
3621 goto error;
3622 }
3623
3624 /* Receive variable len data */
3625 DBG("Waiting for %zu URIs from client ...", nb_uri);
3626 ret = lttcomm_recv_unix_sock(sock, uris, len);
3627 if (ret <= 0) {
3628 DBG("No URIs received from client... continuing");
3629 *sock_error = 1;
3630 ret = LTTNG_ERR_SESSION_FAIL;
3631 free(uris);
3632 goto error;
3633 }
3634
3635 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3636 DBG("Creating session with ONE network URI is a bad call");
3637 ret = LTTNG_ERR_SESSION_FAIL;
3638 free(uris);
3639 goto error;
3640 }
3641 }
3642
3643 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3644 &cmd_ctx->creds, 0);
3645
3646 free(uris);
3647
3648 break;
3649 }
3650 case LTTNG_DESTROY_SESSION:
3651 {
3652 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3653
3654 /* Set session to NULL so we do not unlock it after free. */
3655 cmd_ctx->session = NULL;
3656 break;
3657 }
3658 case LTTNG_LIST_DOMAINS:
3659 {
3660 ssize_t nb_dom;
3661 struct lttng_domain *domains = NULL;
3662
3663 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3664 if (nb_dom < 0) {
3665 /* Return value is a negative lttng_error_code. */
3666 ret = -nb_dom;
3667 goto error;
3668 }
3669
3670 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3671 if (ret < 0) {
3672 free(domains);
3673 goto setup_error;
3674 }
3675
3676 /* Copy event list into message payload */
3677 memcpy(cmd_ctx->llm->payload, domains,
3678 nb_dom * sizeof(struct lttng_domain));
3679
3680 free(domains);
3681
3682 ret = LTTNG_OK;
3683 break;
3684 }
3685 case LTTNG_LIST_CHANNELS:
3686 {
3687 int nb_chan;
3688 struct lttng_channel *channels = NULL;
3689
3690 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3691 cmd_ctx->session, &channels);
3692 if (nb_chan < 0) {
3693 /* Return value is a negative lttng_error_code. */
3694 ret = -nb_chan;
3695 goto error;
3696 }
3697
3698 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3699 if (ret < 0) {
3700 free(channels);
3701 goto setup_error;
3702 }
3703
3704 /* Copy event list into message payload */
3705 memcpy(cmd_ctx->llm->payload, channels,
3706 nb_chan * sizeof(struct lttng_channel));
3707
3708 free(channels);
3709
3710 ret = LTTNG_OK;
3711 break;
3712 }
3713 case LTTNG_LIST_EVENTS:
3714 {
3715 ssize_t nb_event;
3716 struct lttng_event *events = NULL;
3717
3718 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3719 cmd_ctx->lsm->u.list.channel_name, &events);
3720 if (nb_event < 0) {
3721 /* Return value is a negative lttng_error_code. */
3722 ret = -nb_event;
3723 goto error;
3724 }
3725
3726 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3727 if (ret < 0) {
3728 free(events);
3729 goto setup_error;
3730 }
3731
3732 /* Copy event list into message payload */
3733 memcpy(cmd_ctx->llm->payload, events,
3734 nb_event * sizeof(struct lttng_event));
3735
3736 free(events);
3737
3738 ret = LTTNG_OK;
3739 break;
3740 }
3741 case LTTNG_LIST_SESSIONS:
3742 {
3743 unsigned int nr_sessions;
3744
3745 session_lock_list();
3746 nr_sessions = lttng_sessions_count(
3747 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3748 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3749
3750 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3751 if (ret < 0) {
3752 session_unlock_list();
3753 goto setup_error;
3754 }
3755
3756 /* Filled the session array */
3757 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3758 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3759 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3760
3761 session_unlock_list();
3762
3763 ret = LTTNG_OK;
3764 break;
3765 }
3766 case LTTNG_CALIBRATE:
3767 {
3768 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3769 &cmd_ctx->lsm->u.calibrate);
3770 break;
3771 }
3772 case LTTNG_REGISTER_CONSUMER:
3773 {
3774 struct consumer_data *cdata;
3775
3776 switch (cmd_ctx->lsm->domain.type) {
3777 case LTTNG_DOMAIN_KERNEL:
3778 cdata = &kconsumer_data;
3779 break;
3780 default:
3781 ret = LTTNG_ERR_UND;
3782 goto error;
3783 }
3784
3785 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3786 cmd_ctx->lsm->u.reg.path, cdata);
3787 break;
3788 }
3789 case LTTNG_DATA_PENDING:
3790 {
3791 ret = cmd_data_pending(cmd_ctx->session);
3792 break;
3793 }
3794 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3795 {
3796 struct lttcomm_lttng_output_id reply;
3797
3798 ret = cmd_snapshot_add_output(cmd_ctx->session,
3799 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3800 if (ret != LTTNG_OK) {
3801 goto error;
3802 }
3803
3804 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3805 if (ret < 0) {
3806 goto setup_error;
3807 }
3808
3809 /* Copy output list into message payload */
3810 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3811 ret = LTTNG_OK;
3812 break;
3813 }
3814 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3815 {
3816 ret = cmd_snapshot_del_output(cmd_ctx->session,
3817 &cmd_ctx->lsm->u.snapshot_output.output);
3818 break;
3819 }
3820 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3821 {
3822 ssize_t nb_output;
3823 struct lttng_snapshot_output *outputs = NULL;
3824
3825 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3826 if (nb_output < 0) {
3827 ret = -nb_output;
3828 goto error;
3829 }
3830
3831 ret = setup_lttng_msg(cmd_ctx,
3832 nb_output * sizeof(struct lttng_snapshot_output));
3833 if (ret < 0) {
3834 free(outputs);
3835 goto setup_error;
3836 }
3837
3838 if (outputs) {
3839 /* Copy output list into message payload */
3840 memcpy(cmd_ctx->llm->payload, outputs,
3841 nb_output * sizeof(struct lttng_snapshot_output));
3842 free(outputs);
3843 }
3844
3845 ret = LTTNG_OK;
3846 break;
3847 }
3848 case LTTNG_SNAPSHOT_RECORD:
3849 {
3850 ret = cmd_snapshot_record(cmd_ctx->session,
3851 &cmd_ctx->lsm->u.snapshot_record.output,
3852 cmd_ctx->lsm->u.snapshot_record.wait);
3853 break;
3854 }
3855 case LTTNG_CREATE_SESSION_SNAPSHOT:
3856 {
3857 size_t nb_uri, len;
3858 struct lttng_uri *uris = NULL;
3859
3860 nb_uri = cmd_ctx->lsm->u.uri.size;
3861 len = nb_uri * sizeof(struct lttng_uri);
3862
3863 if (nb_uri > 0) {
3864 uris = zmalloc(len);
3865 if (uris == NULL) {
3866 ret = LTTNG_ERR_FATAL;
3867 goto error;
3868 }
3869
3870 /* Receive variable len data */
3871 DBG("Waiting for %zu URIs from client ...", nb_uri);
3872 ret = lttcomm_recv_unix_sock(sock, uris, len);
3873 if (ret <= 0) {
3874 DBG("No URIs received from client... continuing");
3875 *sock_error = 1;
3876 ret = LTTNG_ERR_SESSION_FAIL;
3877 free(uris);
3878 goto error;
3879 }
3880
3881 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3882 DBG("Creating session with ONE network URI is a bad call");
3883 ret = LTTNG_ERR_SESSION_FAIL;
3884 free(uris);
3885 goto error;
3886 }
3887 }
3888
3889 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3890 nb_uri, &cmd_ctx->creds);
3891 free(uris);
3892 break;
3893 }
3894 case LTTNG_CREATE_SESSION_LIVE:
3895 {
3896 size_t nb_uri, len;
3897 struct lttng_uri *uris = NULL;
3898
3899 nb_uri = cmd_ctx->lsm->u.uri.size;
3900 len = nb_uri * sizeof(struct lttng_uri);
3901
3902 if (nb_uri > 0) {
3903 uris = zmalloc(len);
3904 if (uris == NULL) {
3905 ret = LTTNG_ERR_FATAL;
3906 goto error;
3907 }
3908
3909 /* Receive variable len data */
3910 DBG("Waiting for %zu URIs from client ...", nb_uri);
3911 ret = lttcomm_recv_unix_sock(sock, uris, len);
3912 if (ret <= 0) {
3913 DBG("No URIs received from client... continuing");
3914 *sock_error = 1;
3915 ret = LTTNG_ERR_SESSION_FAIL;
3916 free(uris);
3917 goto error;
3918 }
3919
3920 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3921 DBG("Creating session with ONE network URI is a bad call");
3922 ret = LTTNG_ERR_SESSION_FAIL;
3923 free(uris);
3924 goto error;
3925 }
3926 }
3927
3928 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3929 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3930 free(uris);
3931 break;
3932 }
3933 case LTTNG_SAVE_SESSION:
3934 {
3935 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3936 &cmd_ctx->creds);
3937 break;
3938 }
3939 case LTTNG_SET_SESSION_SHM_PATH:
3940 {
3941 ret = cmd_set_session_shm_path(cmd_ctx->session,
3942 cmd_ctx->lsm->u.set_shm_path.shm_path);
3943 break;
3944 }
3945 default:
3946 ret = LTTNG_ERR_UND;
3947 break;
3948 }
3949
3950 error:
3951 if (cmd_ctx->llm == NULL) {
3952 DBG("Missing llm structure. Allocating one.");
3953 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3954 goto setup_error;
3955 }
3956 }
3957 /* Set return code */
3958 cmd_ctx->llm->ret_code = ret;
3959 setup_error:
3960 if (cmd_ctx->session) {
3961 session_unlock(cmd_ctx->session);
3962 }
3963 if (need_tracing_session) {
3964 session_unlock_list();
3965 }
3966 init_setup_error:
3967 return ret;
3968 }
3969
3970 /*
3971 * Thread managing health check socket.
3972 */
3973 static void *thread_manage_health(void *data)
3974 {
3975 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3976 uint32_t revents, nb_fd;
3977 struct lttng_poll_event events;
3978 struct health_comm_msg msg;
3979 struct health_comm_reply reply;
3980
3981 DBG("[thread] Manage health check started");
3982
3983 rcu_register_thread();
3984
3985 /* We might hit an error path before this is created. */
3986 lttng_poll_init(&events);
3987
3988 /* Create unix socket */
3989 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3990 if (sock < 0) {
3991 ERR("Unable to create health check Unix socket");
3992 ret = -1;
3993 goto error;
3994 }
3995
3996 if (is_root) {
3997 /* lttng health client socket path permissions */
3998 ret = chown(health_unix_sock_path, 0,
3999 utils_get_group_id(tracing_group_name));
4000 if (ret < 0) {
4001 ERR("Unable to set group on %s", health_unix_sock_path);
4002 PERROR("chown");
4003 ret = -1;
4004 goto error;
4005 }
4006
4007 ret = chmod(health_unix_sock_path,
4008 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4009 if (ret < 0) {
4010 ERR("Unable to set permissions on %s", health_unix_sock_path);
4011 PERROR("chmod");
4012 ret = -1;
4013 goto error;
4014 }
4015 }
4016
4017 /*
4018 * Set the CLOEXEC flag. Return code is useless because either way, the
4019 * show must go on.
4020 */
4021 (void) utils_set_fd_cloexec(sock);
4022
4023 ret = lttcomm_listen_unix_sock(sock);
4024 if (ret < 0) {
4025 goto error;
4026 }
4027
4028 /*
4029 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4030 * more will be added to this poll set.
4031 */
4032 ret = sessiond_set_thread_pollset(&events, 2);
4033 if (ret < 0) {
4034 goto error;
4035 }
4036
4037 /* Add the application registration socket */
4038 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4039 if (ret < 0) {
4040 goto error;
4041 }
4042
4043 sessiond_notify_ready();
4044
4045 while (1) {
4046 DBG("Health check ready");
4047
4048 /* Inifinite blocking call, waiting for transmission */
4049 restart:
4050 ret = lttng_poll_wait(&events, -1);
4051 if (ret < 0) {
4052 /*
4053 * Restart interrupted system call.
4054 */
4055 if (errno == EINTR) {
4056 goto restart;
4057 }
4058 goto error;
4059 }
4060
4061 nb_fd = ret;
4062
4063 for (i = 0; i < nb_fd; i++) {
4064 /* Fetch once the poll data */
4065 revents = LTTNG_POLL_GETEV(&events, i);
4066 pollfd = LTTNG_POLL_GETFD(&events, i);
4067
4068 if (!revents) {
4069 /* No activity for this FD (poll implementation). */
4070 continue;
4071 }
4072
4073 /* Thread quit pipe has been closed. Killing thread. */
4074 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4075 if (ret) {
4076 err = 0;
4077 goto exit;
4078 }
4079
4080 /* Event on the registration socket */
4081 if (pollfd == sock) {
4082 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4083 ERR("Health socket poll error");
4084 goto error;
4085 }
4086 }
4087 }
4088
4089 new_sock = lttcomm_accept_unix_sock(sock);
4090 if (new_sock < 0) {
4091 goto error;
4092 }
4093
4094 /*
4095 * Set the CLOEXEC flag. Return code is useless because either way, the
4096 * show must go on.
4097 */
4098 (void) utils_set_fd_cloexec(new_sock);
4099
4100 DBG("Receiving data from client for health...");
4101 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4102 if (ret <= 0) {
4103 DBG("Nothing recv() from client... continuing");
4104 ret = close(new_sock);
4105 if (ret) {
4106 PERROR("close");
4107 }
4108 new_sock = -1;
4109 continue;
4110 }
4111
4112 rcu_thread_online();
4113
4114 memset(&reply, 0, sizeof(reply));
4115 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4116 /*
4117 * health_check_state returns 0 if health is
4118 * bad.
4119 */
4120 if (!health_check_state(health_sessiond, i)) {
4121 reply.ret_code |= 1ULL << i;
4122 }
4123 }
4124
4125 DBG2("Health check return value %" PRIx64, reply.ret_code);
4126
4127 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4128 if (ret < 0) {
4129 ERR("Failed to send health data back to client");
4130 }
4131
4132 /* End of transmission */
4133 ret = close(new_sock);
4134 if (ret) {
4135 PERROR("close");
4136 }
4137 new_sock = -1;
4138 }
4139
4140 exit:
4141 error:
4142 if (err) {
4143 ERR("Health error occurred in %s", __func__);
4144 }
4145 DBG("Health check thread dying");
4146 unlink(health_unix_sock_path);
4147 if (sock >= 0) {
4148 ret = close(sock);
4149 if (ret) {
4150 PERROR("close");
4151 }
4152 }
4153
4154 lttng_poll_clean(&events);
4155
4156 rcu_unregister_thread();
4157 return NULL;
4158 }
4159
4160 /*
4161 * This thread manage all clients request using the unix client socket for
4162 * communication.
4163 */
4164 static void *thread_manage_clients(void *data)
4165 {
4166 int sock = -1, ret, i, pollfd, err = -1;
4167 int sock_error;
4168 uint32_t revents, nb_fd;
4169 struct command_ctx *cmd_ctx = NULL;
4170 struct lttng_poll_event events;
4171
4172 DBG("[thread] Manage client started");
4173
4174 rcu_register_thread();
4175
4176 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4177
4178 health_code_update();
4179
4180 ret = lttcomm_listen_unix_sock(client_sock);
4181 if (ret < 0) {
4182 goto error_listen;
4183 }
4184
4185 /*
4186 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4187 * more will be added to this poll set.
4188 */
4189 ret = sessiond_set_thread_pollset(&events, 2);
4190 if (ret < 0) {
4191 goto error_create_poll;
4192 }
4193
4194 /* Add the application registration socket */
4195 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4196 if (ret < 0) {
4197 goto error;
4198 }
4199
4200 sessiond_notify_ready();
4201 ret = sem_post(&load_info->message_thread_ready);
4202 if (ret) {
4203 PERROR("sem_post message_thread_ready");
4204 goto error;
4205 }
4206
4207 /* This testpoint is after we signal readiness to the parent. */
4208 if (testpoint(sessiond_thread_manage_clients)) {
4209 goto error;
4210 }
4211
4212 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4213 goto error;
4214 }
4215
4216 health_code_update();
4217
4218 while (1) {
4219 DBG("Accepting client command ...");
4220
4221 /* Inifinite blocking call, waiting for transmission */
4222 restart:
4223 health_poll_entry();
4224 ret = lttng_poll_wait(&events, -1);
4225 health_poll_exit();
4226 if (ret < 0) {
4227 /*
4228 * Restart interrupted system call.
4229 */
4230 if (errno == EINTR) {
4231 goto restart;
4232 }
4233 goto error;
4234 }
4235
4236 nb_fd = ret;
4237
4238 for (i = 0; i < nb_fd; i++) {
4239 /* Fetch once the poll data */
4240 revents = LTTNG_POLL_GETEV(&events, i);
4241 pollfd = LTTNG_POLL_GETFD(&events, i);
4242
4243 health_code_update();
4244
4245 if (!revents) {
4246 /* No activity for this FD (poll implementation). */
4247 continue;
4248 }
4249
4250 /* Thread quit pipe has been closed. Killing thread. */
4251 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4252 if (ret) {
4253 err = 0;
4254 goto exit;
4255 }
4256
4257 /* Event on the registration socket */
4258 if (pollfd == client_sock) {
4259 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4260 ERR("Client socket poll error");
4261 goto error;
4262 }
4263 }
4264 }
4265
4266 DBG("Wait for client response");
4267
4268 health_code_update();
4269
4270 sock = lttcomm_accept_unix_sock(client_sock);
4271 if (sock < 0) {
4272 goto error;
4273 }
4274
4275 /*
4276 * Set the CLOEXEC flag. Return code is useless because either way, the
4277 * show must go on.
4278 */
4279 (void) utils_set_fd_cloexec(sock);
4280
4281 /* Set socket option for credentials retrieval */
4282 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4283 if (ret < 0) {
4284 goto error;
4285 }
4286
4287 /* Allocate context command to process the client request */
4288 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4289 if (cmd_ctx == NULL) {
4290 PERROR("zmalloc cmd_ctx");
4291 goto error;
4292 }
4293
4294 /* Allocate data buffer for reception */
4295 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4296 if (cmd_ctx->lsm == NULL) {
4297 PERROR("zmalloc cmd_ctx->lsm");
4298 goto error;
4299 }
4300
4301 cmd_ctx->llm = NULL;
4302 cmd_ctx->session = NULL;
4303
4304 health_code_update();
4305
4306 /*
4307 * Data is received from the lttng client. The struct
4308 * lttcomm_session_msg (lsm) contains the command and data request of
4309 * the client.
4310 */
4311 DBG("Receiving data from client ...");
4312 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4313 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4314 if (ret <= 0) {
4315 DBG("Nothing recv() from client... continuing");
4316 ret = close(sock);
4317 if (ret) {
4318 PERROR("close");
4319 }
4320 sock = -1;
4321 clean_command_ctx(&cmd_ctx);
4322 continue;
4323 }
4324
4325 health_code_update();
4326
4327 // TODO: Validate cmd_ctx including sanity check for
4328 // security purpose.
4329
4330 rcu_thread_online();
4331 /*
4332 * This function dispatch the work to the kernel or userspace tracer
4333 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4334 * informations for the client. The command context struct contains
4335 * everything this function may needs.
4336 */
4337 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4338 rcu_thread_offline();
4339 if (ret < 0) {
4340 ret = close(sock);
4341 if (ret) {
4342 PERROR("close");
4343 }
4344 sock = -1;
4345 /*
4346 * TODO: Inform client somehow of the fatal error. At
4347 * this point, ret < 0 means that a zmalloc failed
4348 * (ENOMEM). Error detected but still accept
4349 * command, unless a socket error has been
4350 * detected.
4351 */
4352 clean_command_ctx(&cmd_ctx);
4353 continue;
4354 }
4355
4356 health_code_update();
4357
4358 DBG("Sending response (size: %d, retcode: %s)",
4359 cmd_ctx->lttng_msg_size,
4360 lttng_strerror(-cmd_ctx->llm->ret_code));
4361 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4362 if (ret < 0) {
4363 ERR("Failed to send data back to client");
4364 }
4365
4366 /* End of transmission */
4367 ret = close(sock);
4368 if (ret) {
4369 PERROR("close");
4370 }
4371 sock = -1;
4372
4373 clean_command_ctx(&cmd_ctx);
4374
4375 health_code_update();
4376 }
4377
4378 exit:
4379 error:
4380 if (sock >= 0) {
4381 ret = close(sock);
4382 if (ret) {
4383 PERROR("close");
4384 }
4385 }
4386
4387 lttng_poll_clean(&events);
4388 clean_command_ctx(&cmd_ctx);
4389
4390 error_listen:
4391 error_create_poll:
4392 unlink(client_unix_sock_path);
4393 if (client_sock >= 0) {
4394 ret = close(client_sock);
4395 if (ret) {
4396 PERROR("close");
4397 }
4398 }
4399
4400 if (err) {
4401 health_error();
4402 ERR("Health error occurred in %s", __func__);
4403 }
4404
4405 health_unregister(health_sessiond);
4406
4407 DBG("Client thread dying");
4408
4409 rcu_unregister_thread();
4410
4411 /*
4412 * Since we are creating the consumer threads, we own them, so we need
4413 * to join them before our thread exits.
4414 */
4415 ret = join_consumer_thread(&kconsumer_data);
4416 if (ret) {
4417 errno = ret;
4418 PERROR("join_consumer");
4419 }
4420
4421 ret = join_consumer_thread(&ustconsumer32_data);
4422 if (ret) {
4423 errno = ret;
4424 PERROR("join_consumer ust32");
4425 }
4426
4427 ret = join_consumer_thread(&ustconsumer64_data);
4428 if (ret) {
4429 errno = ret;
4430 PERROR("join_consumer ust64");
4431 }
4432 return NULL;
4433 }
4434
4435
4436 /*
4437 * usage function on stderr
4438 */
4439 static void usage(void)
4440 {
4441 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4442 fprintf(stderr, " -h, --help Display this usage.\n");
4443 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4444 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4445 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4446 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4447 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4448 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4449 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4450 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4451 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4452 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4453 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4454 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4455 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4456 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4457 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4458 fprintf(stderr, " -V, --version Show version number.\n");
4459 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4460 fprintf(stderr, " -q, --quiet No output at all.\n");
4461 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4462 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4463 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4464 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4465 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4466 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4467 fprintf(stderr, " -l --load PATH Load session configuration\n");
4468 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4469 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4470 }
4471
4472 static int string_match(const char *str1, const char *str2)
4473 {
4474 return (str1 && str2) && !strcmp(str1, str2);
4475 }
4476
4477 /*
4478 * Take an option from the getopt output and set it in the right variable to be
4479 * used later.
4480 *
4481 * Return 0 on success else a negative value.
4482 */
4483 static int set_option(int opt, const char *arg, const char *optname)
4484 {
4485 int ret = 0;
4486
4487 if (arg && arg[0] == '\0') {
4488 /*
4489 * This only happens if the value is read from daemon config
4490 * file. This means the option requires an argument and the
4491 * configuration file contains a line such as:
4492 * my_option =
4493 */
4494 ret = -EINVAL;
4495 goto end;
4496 }
4497
4498 if (string_match(optname, "client-sock") || opt == 'c') {
4499 if (lttng_is_setuid_setgid()) {
4500 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4501 "-c, --client-sock");
4502 } else {
4503 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4504 }
4505 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4506 if (lttng_is_setuid_setgid()) {
4507 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4508 "-a, --apps-sock");
4509 } else {
4510 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4511 }
4512 } else if (string_match(optname, "daemonize") || opt == 'd') {
4513 opt_daemon = 1;
4514 } else if (string_match(optname, "background") || opt == 'b') {
4515 opt_background = 1;
4516 } else if (string_match(optname, "group") || opt == 'g') {
4517 if (lttng_is_setuid_setgid()) {
4518 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4519 "-g, --group");
4520 } else {
4521 /*
4522 * If the override option is set, the pointer points to a
4523 * *non* const thus freeing it even though the variable type is
4524 * set to const.
4525 */
4526 if (tracing_group_name_override) {
4527 free((void *) tracing_group_name);
4528 }
4529 tracing_group_name = strdup(arg);
4530 if (!tracing_group_name) {
4531 PERROR("strdup");
4532 ret = -ENOMEM;
4533 }
4534 tracing_group_name_override = 1;
4535 }
4536 } else if (string_match(optname, "help") || opt == 'h') {
4537 usage();
4538 exit(EXIT_SUCCESS);
4539 } else if (string_match(optname, "version") || opt == 'V') {
4540 fprintf(stdout, "%s\n", VERSION);
4541 exit(EXIT_SUCCESS);
4542 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4543 opt_sig_parent = 1;
4544 } else if (string_match(optname, "kconsumerd-err-sock")) {
4545 if (lttng_is_setuid_setgid()) {
4546 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4547 "--kconsumerd-err-sock");
4548 } else {
4549 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4550 }
4551 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4552 if (lttng_is_setuid_setgid()) {
4553 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4554 "--kconsumerd-cmd-sock");
4555 } else {
4556 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4557 }
4558 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4559 if (lttng_is_setuid_setgid()) {
4560 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4561 "--ustconsumerd64-err-sock");
4562 } else {
4563 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4564 }
4565 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4566 if (lttng_is_setuid_setgid()) {
4567 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4568 "--ustconsumerd64-cmd-sock");
4569 } else {
4570 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4571 }
4572 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4573 if (lttng_is_setuid_setgid()) {
4574 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4575 "--ustconsumerd32-err-sock");
4576 } else {
4577 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4578 }
4579 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4580 if (lttng_is_setuid_setgid()) {
4581 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4582 "--ustconsumerd32-cmd-sock");
4583 } else {
4584 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4585 }
4586 } else if (string_match(optname, "no-kernel")) {
4587 opt_no_kernel = 1;
4588 } else if (string_match(optname, "quiet") || opt == 'q') {
4589 lttng_opt_quiet = 1;
4590 } else if (string_match(optname, "verbose") || opt == 'v') {
4591 /* Verbose level can increase using multiple -v */
4592 if (arg) {
4593 /* Value obtained from config file */
4594 lttng_opt_verbose = config_parse_value(arg);
4595 } else {
4596 /* -v used on command line */
4597 lttng_opt_verbose++;
4598 }
4599 /* Clamp value to [0, 3] */
4600 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4601 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4602 } else if (string_match(optname, "verbose-consumer")) {
4603 if (arg) {
4604 opt_verbose_consumer = config_parse_value(arg);
4605 } else {
4606 opt_verbose_consumer += 1;
4607 }
4608 } else if (string_match(optname, "consumerd32-path")) {
4609 if (lttng_is_setuid_setgid()) {
4610 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4611 "--consumerd32-path");
4612 } else {
4613 if (consumerd32_bin_override) {
4614 free((void *) consumerd32_bin);
4615 }
4616 consumerd32_bin = strdup(arg);
4617 if (!consumerd32_bin) {
4618 PERROR("strdup");
4619 ret = -ENOMEM;
4620 }
4621 consumerd32_bin_override = 1;
4622 }
4623 } else if (string_match(optname, "consumerd32-libdir")) {
4624 if (lttng_is_setuid_setgid()) {
4625 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4626 "--consumerd32-libdir");
4627 } else {
4628 if (consumerd32_libdir_override) {
4629 free((void *) consumerd32_libdir);
4630 }
4631 consumerd32_libdir = strdup(arg);
4632 if (!consumerd32_libdir) {
4633 PERROR("strdup");
4634 ret = -ENOMEM;
4635 }
4636 consumerd32_libdir_override = 1;
4637 }
4638 } else if (string_match(optname, "consumerd64-path")) {
4639 if (lttng_is_setuid_setgid()) {
4640 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4641 "--consumerd64-path");
4642 } else {
4643 if (consumerd64_bin_override) {
4644 free((void *) consumerd64_bin);
4645 }
4646 consumerd64_bin = strdup(arg);
4647 if (!consumerd64_bin) {
4648 PERROR("strdup");
4649 ret = -ENOMEM;
4650 }
4651 consumerd64_bin_override = 1;
4652 }
4653 } else if (string_match(optname, "consumerd64-libdir")) {
4654 if (lttng_is_setuid_setgid()) {
4655 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4656 "--consumerd64-libdir");
4657 } else {
4658 if (consumerd64_libdir_override) {
4659 free((void *) consumerd64_libdir);
4660 }
4661 consumerd64_libdir = strdup(arg);
4662 if (!consumerd64_libdir) {
4663 PERROR("strdup");
4664 ret = -ENOMEM;
4665 }
4666 consumerd64_libdir_override = 1;
4667 }
4668 } else if (string_match(optname, "pidfile") || opt == 'p') {
4669 if (lttng_is_setuid_setgid()) {
4670 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4671 "-p, --pidfile");
4672 } else {
4673 free(opt_pidfile);
4674 opt_pidfile = strdup(arg);
4675 if (!opt_pidfile) {
4676 PERROR("strdup");
4677 ret = -ENOMEM;
4678 }
4679 }
4680 } else if (string_match(optname, "agent-tcp-port")) {
4681 if (lttng_is_setuid_setgid()) {
4682 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4683 "--agent-tcp-port");
4684 } else {
4685 unsigned long v;
4686
4687 if (!arg) {
4688 ret = -EINVAL;
4689 goto end;
4690 }
4691 errno = 0;
4692 v = strtoul(arg, NULL, 0);
4693 if (errno != 0 || !isdigit(arg[0])) {
4694 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4695 return -1;
4696 }
4697 if (v == 0 || v >= 65535) {
4698 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4699 return -1;
4700 }
4701 agent_tcp_port = (uint32_t) v;
4702 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4703 }
4704 } else if (string_match(optname, "load") || opt == 'l') {
4705 if (lttng_is_setuid_setgid()) {
4706 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4707 "-l, --load");
4708 } else {
4709 free(opt_load_session_path);
4710 opt_load_session_path = strdup(arg);
4711 if (!opt_load_session_path) {
4712 PERROR("strdup");
4713 ret = -ENOMEM;
4714 }
4715 }
4716 } else if (string_match(optname, "kmod-probes")) {
4717 if (lttng_is_setuid_setgid()) {
4718 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4719 "--kmod-probes");
4720 } else {
4721 free(kmod_probes_list);
4722 kmod_probes_list = strdup(arg);
4723 if (!kmod_probes_list) {
4724 PERROR("strdup");
4725 ret = -ENOMEM;
4726 }
4727 }
4728 } else if (string_match(optname, "extra-kmod-probes")) {
4729 if (lttng_is_setuid_setgid()) {
4730 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4731 "--extra-kmod-probes");
4732 } else {
4733 free(kmod_extra_probes_list);
4734 kmod_extra_probes_list = strdup(arg);
4735 if (!kmod_extra_probes_list) {
4736 PERROR("strdup");
4737 ret = -ENOMEM;
4738 }
4739 }
4740 } else if (string_match(optname, "config") || opt == 'f') {
4741 /* This is handled in set_options() thus silent skip. */
4742 goto end;
4743 } else {
4744 /* Unknown option or other error.
4745 * Error is printed by getopt, just return */
4746 ret = -1;
4747 }
4748
4749 end:
4750 if (ret == -EINVAL) {
4751 const char *opt_name = "unknown";
4752 int i;
4753
4754 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4755 i++) {
4756 if (opt == long_options[i].val) {
4757 opt_name = long_options[i].name;
4758 break;
4759 }
4760 }
4761
4762 WARN("Invalid argument provided for option \"%s\", using default value.",
4763 opt_name);
4764 }
4765
4766 return ret;
4767 }
4768
4769 /*
4770 * config_entry_handler_cb used to handle options read from a config file.
4771 * See config_entry_handler_cb comment in common/config/config.h for the
4772 * return value conventions.
4773 */
4774 static int config_entry_handler(const struct config_entry *entry, void *unused)
4775 {
4776 int ret = 0, i;
4777
4778 if (!entry || !entry->name || !entry->value) {
4779 ret = -EINVAL;
4780 goto end;
4781 }
4782
4783 /* Check if the option is to be ignored */
4784 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4785 if (!strcmp(entry->name, config_ignore_options[i])) {
4786 goto end;
4787 }
4788 }
4789
4790 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4791 i++) {
4792
4793 /* Ignore if not fully matched. */
4794 if (strcmp(entry->name, long_options[i].name)) {
4795 continue;
4796 }
4797
4798 /*
4799 * If the option takes no argument on the command line, we have to
4800 * check if the value is "true". We support non-zero numeric values,
4801 * true, on and yes.
4802 */
4803 if (!long_options[i].has_arg) {
4804 ret = config_parse_value(entry->value);
4805 if (ret <= 0) {
4806 if (ret) {
4807 WARN("Invalid configuration value \"%s\" for option %s",
4808 entry->value, entry->name);
4809 }
4810 /* False, skip boolean config option. */
4811 goto end;
4812 }
4813 }
4814
4815 ret = set_option(long_options[i].val, entry->value, entry->name);
4816 goto end;
4817 }
4818
4819 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4820
4821 end:
4822 return ret;
4823 }
4824
4825 /*
4826 * daemon configuration loading and argument parsing
4827 */
4828 static int set_options(int argc, char **argv)
4829 {
4830 int ret = 0, c = 0, option_index = 0;
4831 int orig_optopt = optopt, orig_optind = optind;
4832 char *optstring;
4833 const char *config_path = NULL;
4834
4835 optstring = utils_generate_optstring(long_options,
4836 sizeof(long_options) / sizeof(struct option));
4837 if (!optstring) {
4838 ret = -ENOMEM;
4839 goto end;
4840 }
4841
4842 /* Check for the --config option */
4843 while ((c = getopt_long(argc, argv, optstring, long_options,
4844 &option_index)) != -1) {
4845 if (c == '?') {
4846 ret = -EINVAL;
4847 goto end;
4848 } else if (c != 'f') {
4849 /* if not equal to --config option. */
4850 continue;
4851 }
4852
4853 if (lttng_is_setuid_setgid()) {
4854 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4855 "-f, --config");
4856 } else {
4857 config_path = utils_expand_path(optarg);
4858 if (!config_path) {
4859 ERR("Failed to resolve path: %s", optarg);
4860 }
4861 }
4862 }
4863
4864 ret = config_get_section_entries(config_path, config_section_name,
4865 config_entry_handler, NULL);
4866 if (ret) {
4867 if (ret > 0) {
4868 ERR("Invalid configuration option at line %i", ret);
4869 ret = -1;
4870 }
4871 goto end;
4872 }
4873
4874 /* Reset getopt's global state */
4875 optopt = orig_optopt;
4876 optind = orig_optind;
4877 while (1) {
4878 option_index = -1;
4879 /*
4880 * getopt_long() will not set option_index if it encounters a
4881 * short option.
4882 */
4883 c = getopt_long(argc, argv, optstring, long_options,
4884 &option_index);
4885 if (c == -1) {
4886 break;
4887 }
4888
4889 /*
4890 * Pass NULL as the long option name if popt left the index
4891 * unset.
4892 */
4893 ret = set_option(c, optarg,
4894 option_index < 0 ? NULL :
4895 long_options[option_index].name);
4896 if (ret < 0) {
4897 break;
4898 }
4899 }
4900
4901 end:
4902 free(optstring);
4903 return ret;
4904 }
4905
4906 /*
4907 * Creates the two needed socket by the daemon.
4908 * apps_sock - The communication socket for all UST apps.
4909 * client_sock - The communication of the cli tool (lttng).
4910 */
4911 static int init_daemon_socket(void)
4912 {
4913 int ret = 0;
4914 mode_t old_umask;
4915
4916 old_umask = umask(0);
4917
4918 /* Create client tool unix socket */
4919 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4920 if (client_sock < 0) {
4921 ERR("Create unix sock failed: %s", client_unix_sock_path);
4922 ret = -1;
4923 goto end;
4924 }
4925
4926 /* Set the cloexec flag */
4927 ret = utils_set_fd_cloexec(client_sock);
4928 if (ret < 0) {
4929 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4930 "Continuing but note that the consumer daemon will have a "
4931 "reference to this socket on exec()", client_sock);
4932 }
4933
4934 /* File permission MUST be 660 */
4935 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4936 if (ret < 0) {
4937 ERR("Set file permissions failed: %s", client_unix_sock_path);
4938 PERROR("chmod");
4939 goto end;
4940 }
4941
4942 /* Create the application unix socket */
4943 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4944 if (apps_sock < 0) {
4945 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4946 ret = -1;
4947 goto end;
4948 }
4949
4950 /* Set the cloexec flag */
4951 ret = utils_set_fd_cloexec(apps_sock);
4952 if (ret < 0) {
4953 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4954 "Continuing but note that the consumer daemon will have a "
4955 "reference to this socket on exec()", apps_sock);
4956 }
4957
4958 /* File permission MUST be 666 */
4959 ret = chmod(apps_unix_sock_path,
4960 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4961 if (ret < 0) {
4962 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4963 PERROR("chmod");
4964 goto end;
4965 }
4966
4967 DBG3("Session daemon client socket %d and application socket %d created",
4968 client_sock, apps_sock);
4969
4970 end:
4971 umask(old_umask);
4972 return ret;
4973 }
4974
4975 /*
4976 * Check if the global socket is available, and if a daemon is answering at the
4977 * other side. If yes, error is returned.
4978 */
4979 static int check_existing_daemon(void)
4980 {
4981 /* Is there anybody out there ? */
4982 if (lttng_session_daemon_alive()) {
4983 return -EEXIST;
4984 }
4985
4986 return 0;
4987 }
4988
4989 /*
4990 * Set the tracing group gid onto the client socket.
4991 *
4992 * Race window between mkdir and chown is OK because we are going from more
4993 * permissive (root.root) to less permissive (root.tracing).
4994 */
4995 static int set_permissions(char *rundir)
4996 {
4997 int ret;
4998 gid_t gid;
4999
5000 gid = utils_get_group_id(tracing_group_name);
5001
5002 /* Set lttng run dir */
5003 ret = chown(rundir, 0, gid);
5004 if (ret < 0) {
5005 ERR("Unable to set group on %s", rundir);
5006 PERROR("chown");
5007 }
5008
5009 /*
5010 * Ensure all applications and tracing group can search the run
5011 * dir. Allow everyone to read the directory, since it does not
5012 * buy us anything to hide its content.
5013 */
5014 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5015 if (ret < 0) {
5016 ERR("Unable to set permissions on %s", rundir);
5017 PERROR("chmod");
5018 }
5019
5020 /* lttng client socket path */
5021 ret = chown(client_unix_sock_path, 0, gid);
5022 if (ret < 0) {
5023 ERR("Unable to set group on %s", client_unix_sock_path);
5024 PERROR("chown");
5025 }
5026
5027 /* kconsumer error socket path */
5028 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5029 if (ret < 0) {
5030 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5031 PERROR("chown");
5032 }
5033
5034 /* 64-bit ustconsumer error socket path */
5035 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5036 if (ret < 0) {
5037 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5038 PERROR("chown");
5039 }
5040
5041 /* 32-bit ustconsumer compat32 error socket path */
5042 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5043 if (ret < 0) {
5044 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5045 PERROR("chown");
5046 }
5047
5048 DBG("All permissions are set");
5049
5050 return ret;
5051 }
5052
5053 /*
5054 * Create the lttng run directory needed for all global sockets and pipe.
5055 */
5056 static int create_lttng_rundir(const char *rundir)
5057 {
5058 int ret;
5059
5060 DBG3("Creating LTTng run directory: %s", rundir);
5061
5062 ret = mkdir(rundir, S_IRWXU);
5063 if (ret < 0) {
5064 if (errno != EEXIST) {
5065 ERR("Unable to create %s", rundir);
5066 goto error;
5067 } else {
5068 ret = 0;
5069 }
5070 }
5071
5072 error:
5073 return ret;
5074 }
5075
5076 /*
5077 * Setup sockets and directory needed by the kconsumerd communication with the
5078 * session daemon.
5079 */
5080 static int set_consumer_sockets(struct consumer_data *consumer_data,
5081 const char *rundir)
5082 {
5083 int ret;
5084 char path[PATH_MAX];
5085
5086 switch (consumer_data->type) {
5087 case LTTNG_CONSUMER_KERNEL:
5088 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5089 break;
5090 case LTTNG_CONSUMER64_UST:
5091 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5092 break;
5093 case LTTNG_CONSUMER32_UST:
5094 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5095 break;
5096 default:
5097 ERR("Consumer type unknown");
5098 ret = -EINVAL;
5099 goto error;
5100 }
5101
5102 DBG2("Creating consumer directory: %s", path);
5103
5104 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5105 if (ret < 0) {
5106 if (errno != EEXIST) {
5107 PERROR("mkdir");
5108 ERR("Failed to create %s", path);
5109 goto error;
5110 }
5111 ret = -1;
5112 }
5113 if (is_root) {
5114 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5115 if (ret < 0) {
5116 ERR("Unable to set group on %s", path);
5117 PERROR("chown");
5118 goto error;
5119 }
5120 }
5121
5122 /* Create the kconsumerd error unix socket */
5123 consumer_data->err_sock =
5124 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5125 if (consumer_data->err_sock < 0) {
5126 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5127 ret = -1;
5128 goto error;
5129 }
5130
5131 /*
5132 * Set the CLOEXEC flag. Return code is useless because either way, the
5133 * show must go on.
5134 */
5135 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5136 if (ret < 0) {
5137 PERROR("utils_set_fd_cloexec");
5138 /* continue anyway */
5139 }
5140
5141 /* File permission MUST be 660 */
5142 ret = chmod(consumer_data->err_unix_sock_path,
5143 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5144 if (ret < 0) {
5145 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5146 PERROR("chmod");
5147 goto error;
5148 }
5149
5150 error:
5151 return ret;
5152 }
5153
5154 /*
5155 * Signal handler for the daemon
5156 *
5157 * Simply stop all worker threads, leaving main() return gracefully after
5158 * joining all threads and calling cleanup().
5159 */
5160 static void sighandler(int sig)
5161 {
5162 switch (sig) {
5163 case SIGPIPE:
5164 DBG("SIGPIPE caught");
5165 return;
5166 case SIGINT:
5167 DBG("SIGINT caught");
5168 stop_threads();
5169 break;
5170 case SIGTERM:
5171 DBG("SIGTERM caught");
5172 stop_threads();
5173 break;
5174 case SIGUSR1:
5175 CMM_STORE_SHARED(recv_child_signal, 1);
5176 break;
5177 default:
5178 break;
5179 }
5180 }
5181
5182 /*
5183 * Setup signal handler for :
5184 * SIGINT, SIGTERM, SIGPIPE
5185 */
5186 static int set_signal_handler(void)
5187 {
5188 int ret = 0;
5189 struct sigaction sa;
5190 sigset_t sigset;
5191
5192 if ((ret = sigemptyset(&sigset)) < 0) {
5193 PERROR("sigemptyset");
5194 return ret;
5195 }
5196
5197 sa.sa_handler = sighandler;
5198 sa.sa_mask = sigset;
5199 sa.sa_flags = 0;
5200 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5201 PERROR("sigaction");
5202 return ret;
5203 }
5204
5205 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5206 PERROR("sigaction");
5207 return ret;
5208 }
5209
5210 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5211 PERROR("sigaction");
5212 return ret;
5213 }
5214
5215 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5216 PERROR("sigaction");
5217 return ret;
5218 }
5219
5220 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5221
5222 return ret;
5223 }
5224
5225 /*
5226 * Set open files limit to unlimited. This daemon can open a large number of
5227 * file descriptors in order to consumer multiple kernel traces.
5228 */
5229 static void set_ulimit(void)
5230 {
5231 int ret;
5232 struct rlimit lim;
5233
5234 /* The kernel does not allowed an infinite limit for open files */
5235 lim.rlim_cur = 65535;
5236 lim.rlim_max = 65535;
5237
5238 ret = setrlimit(RLIMIT_NOFILE, &lim);
5239 if (ret < 0) {
5240 PERROR("failed to set open files limit");
5241 }
5242 }
5243
5244 /*
5245 * Write pidfile using the rundir and opt_pidfile.
5246 */
5247 static int write_pidfile(void)
5248 {
5249 int ret;
5250 char pidfile_path[PATH_MAX];
5251
5252 assert(rundir);
5253
5254 if (opt_pidfile) {
5255 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5256 } else {
5257 /* Build pidfile path from rundir and opt_pidfile. */
5258 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5259 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5260 if (ret < 0) {
5261 PERROR("snprintf pidfile path");
5262 goto error;
5263 }
5264 }
5265
5266 /*
5267 * Create pid file in rundir.
5268 */
5269 ret = utils_create_pid_file(getpid(), pidfile_path);
5270 error:
5271 return ret;
5272 }
5273
5274 /*
5275 * Create lockfile using the rundir and return its fd.
5276 */
5277 static int create_lockfile(void)
5278 {
5279 int ret;
5280 char lockfile_path[PATH_MAX];
5281
5282 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5283 if (ret < 0) {
5284 goto error;
5285 }
5286
5287 ret = utils_create_lock_file(lockfile_path);
5288 error:
5289 return ret;
5290 }
5291
5292 /*
5293 * Write agent TCP port using the rundir.
5294 */
5295 static int write_agent_port(void)
5296 {
5297 int ret;
5298 char path[PATH_MAX];
5299
5300 assert(rundir);
5301
5302 ret = snprintf(path, sizeof(path), "%s/"
5303 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5304 if (ret < 0) {
5305 PERROR("snprintf agent port path");
5306 goto error;
5307 }
5308
5309 /*
5310 * Create TCP agent port file in rundir.
5311 */
5312 ret = utils_create_pid_file(agent_tcp_port, path);
5313
5314 error:
5315 return ret;
5316 }
5317
5318 /*
5319 * main
5320 */
5321 int main(int argc, char **argv)
5322 {
5323 int ret = 0, retval = 0;
5324 void *status;
5325 const char *home_path, *env_app_timeout;
5326
5327 /* Initialize agent apps ht global variable */
5328 agent_apps_ht_by_sock = NULL;
5329
5330 init_kernel_workarounds();
5331
5332 rcu_register_thread();
5333
5334 if (set_signal_handler()) {
5335 retval = -1;
5336 goto exit_set_signal_handler;
5337 }
5338
5339 setup_consumerd_path();
5340
5341 page_size = sysconf(_SC_PAGESIZE);
5342 if (page_size < 0) {
5343 PERROR("sysconf _SC_PAGESIZE");
5344 page_size = LONG_MAX;
5345 WARN("Fallback page size to %ld", page_size);
5346 }
5347
5348 /*
5349 * Parse arguments and load the daemon configuration file.
5350 *
5351 * We have an exit_options exit path to free memory reserved by
5352 * set_options. This is needed because the rest of sessiond_cleanup()
5353 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5354 * depends on set_options.
5355 */
5356 progname = argv[0];
5357 if (set_options(argc, argv)) {
5358 retval = -1;
5359 goto exit_options;
5360 }
5361
5362 /* Daemonize */
5363 if (opt_daemon || opt_background) {
5364 int i;
5365
5366 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5367 !opt_background);
5368 if (ret < 0) {
5369 retval = -1;
5370 goto exit_options;
5371 }
5372
5373 /*
5374 * We are in the child. Make sure all other file descriptors are
5375 * closed, in case we are called with more opened file
5376 * descriptors than the standard ones.
5377 */
5378 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5379 (void) close(i);
5380 }
5381 }
5382
5383 /*
5384 * Starting from here, we can create threads. This needs to be after
5385 * lttng_daemonize due to RCU.
5386 */
5387
5388 /*
5389 * Initialize the health check subsystem. This call should set the
5390 * appropriate time values.
5391 */
5392 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5393 if (!health_sessiond) {
5394 PERROR("health_app_create error");
5395 retval = -1;
5396 goto exit_health_sessiond_cleanup;
5397 }
5398
5399 if (init_ht_cleanup_quit_pipe()) {
5400 retval = -1;
5401 goto exit_ht_cleanup_quit_pipe;
5402 }
5403
5404 /* Setup the thread ht_cleanup communication pipe. */
5405 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5406 retval = -1;
5407 goto exit_ht_cleanup_pipe;
5408 }
5409
5410 /* Set up max poll set size */
5411 if (lttng_poll_set_max_size()) {
5412 retval = -1;
5413 goto exit_set_max_size;
5414 }
5415
5416 /* Create thread to clean up RCU hash tables */
5417 ret = pthread_create(&ht_cleanup_thread, NULL,
5418 thread_ht_cleanup, (void *) NULL);
5419 if (ret) {
5420 errno = ret;
5421 PERROR("pthread_create ht_cleanup");
5422 retval = -1;
5423 goto exit_ht_cleanup;
5424 }
5425
5426 /* Create thread quit pipe */
5427 if (init_thread_quit_pipe()) {
5428 retval = -1;
5429 goto exit_init_data;
5430 }
5431
5432 /* Check if daemon is UID = 0 */
5433 is_root = !getuid();
5434
5435 if (is_root) {
5436 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5437 if (!rundir) {
5438 retval = -1;
5439 goto exit_init_data;
5440 }
5441
5442 /* Create global run dir with root access */
5443 if (create_lttng_rundir(rundir)) {
5444 retval = -1;
5445 goto exit_init_data;
5446 }
5447
5448 if (strlen(apps_unix_sock_path) == 0) {
5449 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5450 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5451 if (ret < 0) {
5452 retval = -1;
5453 goto exit_init_data;
5454 }
5455 }
5456
5457 if (strlen(client_unix_sock_path) == 0) {
5458 ret = snprintf(client_unix_sock_path, PATH_MAX,
5459 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5460 if (ret < 0) {
5461 retval = -1;
5462 goto exit_init_data;
5463 }
5464 }
5465
5466 /* Set global SHM for ust */
5467 if (strlen(wait_shm_path) == 0) {
5468 ret = snprintf(wait_shm_path, PATH_MAX,
5469 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5470 if (ret < 0) {
5471 retval = -1;
5472 goto exit_init_data;
5473 }
5474 }
5475
5476 if (strlen(health_unix_sock_path) == 0) {
5477 ret = snprintf(health_unix_sock_path,
5478 sizeof(health_unix_sock_path),
5479 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5480 if (ret < 0) {
5481 retval = -1;
5482 goto exit_init_data;
5483 }
5484 }
5485
5486 /* Setup kernel consumerd path */
5487 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5488 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5489 if (ret < 0) {
5490 retval = -1;
5491 goto exit_init_data;
5492 }
5493 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5494 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5495 if (ret < 0) {
5496 retval = -1;
5497 goto exit_init_data;
5498 }
5499
5500 DBG2("Kernel consumer err path: %s",
5501 kconsumer_data.err_unix_sock_path);
5502 DBG2("Kernel consumer cmd path: %s",
5503 kconsumer_data.cmd_unix_sock_path);
5504 } else {
5505 home_path = utils_get_home_dir();
5506 if (home_path == NULL) {
5507 /* TODO: Add --socket PATH option */
5508 ERR("Can't get HOME directory for sockets creation.");
5509 retval = -1;
5510 goto exit_init_data;
5511 }
5512
5513 /*
5514 * Create rundir from home path. This will create something like
5515 * $HOME/.lttng
5516 */
5517 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5518 if (ret < 0) {
5519 retval = -1;
5520 goto exit_init_data;
5521 }
5522
5523 if (create_lttng_rundir(rundir)) {
5524 retval = -1;
5525 goto exit_init_data;
5526 }
5527
5528 if (strlen(apps_unix_sock_path) == 0) {
5529 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5530 DEFAULT_HOME_APPS_UNIX_SOCK,
5531 home_path);
5532 if (ret < 0) {
5533 retval = -1;
5534 goto exit_init_data;
5535 }
5536 }
5537
5538 /* Set the cli tool unix socket path */
5539 if (strlen(client_unix_sock_path) == 0) {
5540 ret = snprintf(client_unix_sock_path, PATH_MAX,
5541 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5542 home_path);
5543 if (ret < 0) {
5544 retval = -1;
5545 goto exit_init_data;
5546 }
5547 }
5548
5549 /* Set global SHM for ust */
5550 if (strlen(wait_shm_path) == 0) {
5551 ret = snprintf(wait_shm_path, PATH_MAX,
5552 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5553 getuid());
5554 if (ret < 0) {
5555 retval = -1;
5556 goto exit_init_data;
5557 }
5558 }
5559
5560 /* Set health check Unix path */
5561 if (strlen(health_unix_sock_path) == 0) {
5562 ret = snprintf(health_unix_sock_path,
5563 sizeof(health_unix_sock_path),
5564 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5565 home_path);
5566 if (ret < 0) {
5567 retval = -1;
5568 goto exit_init_data;
5569 }
5570 }
5571 }
5572
5573 lockfile_fd = create_lockfile();
5574 if (lockfile_fd < 0) {
5575 retval = -1;
5576 goto exit_init_data;
5577 }
5578
5579 /* Set consumer initial state */
5580 kernel_consumerd_state = CONSUMER_STOPPED;
5581 ust_consumerd_state = CONSUMER_STOPPED;
5582
5583 DBG("Client socket path %s", client_unix_sock_path);
5584 DBG("Application socket path %s", apps_unix_sock_path);
5585 DBG("Application wait path %s", wait_shm_path);
5586 DBG("LTTng run directory path: %s", rundir);
5587
5588 /* 32 bits consumerd path setup */
5589 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5590 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5591 if (ret < 0) {
5592 PERROR("snprintf 32-bit consumer error socket path");
5593 retval = -1;
5594 goto exit_init_data;
5595 }
5596 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5597 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5598 if (ret < 0) {
5599 PERROR("snprintf 32-bit consumer command socket path");
5600 retval = -1;
5601 goto exit_init_data;
5602 }
5603
5604 DBG2("UST consumer 32 bits err path: %s",
5605 ustconsumer32_data.err_unix_sock_path);
5606 DBG2("UST consumer 32 bits cmd path: %s",
5607 ustconsumer32_data.cmd_unix_sock_path);
5608
5609 /* 64 bits consumerd path setup */
5610 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5611 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5612 if (ret < 0) {
5613 PERROR("snprintf 64-bit consumer error socket path");
5614 retval = -1;
5615 goto exit_init_data;
5616 }
5617 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5618 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5619 if (ret < 0) {
5620 PERROR("snprintf 64-bit consumer command socket path");
5621 retval = -1;
5622 goto exit_init_data;
5623 }
5624
5625 DBG2("UST consumer 64 bits err path: %s",
5626 ustconsumer64_data.err_unix_sock_path);
5627 DBG2("UST consumer 64 bits cmd path: %s",
5628 ustconsumer64_data.cmd_unix_sock_path);
5629
5630 /*
5631 * See if daemon already exist.
5632 */
5633 if (check_existing_daemon()) {
5634 ERR("Already running daemon.\n");
5635 /*
5636 * We do not goto exit because we must not cleanup()
5637 * because a daemon is already running.
5638 */
5639 retval = -1;
5640 goto exit_init_data;
5641 }
5642
5643 /*
5644 * Init UST app hash table. Alloc hash table before this point since
5645 * cleanup() can get called after that point.
5646 */
5647 if (ust_app_ht_alloc()) {
5648 ERR("Failed to allocate UST app hash table");
5649 retval = -1;
5650 goto exit_init_data;
5651 }
5652
5653 /*
5654 * Initialize agent app hash table. We allocate the hash table here
5655 * since cleanup() can get called after this point.
5656 */
5657 if (agent_app_ht_alloc()) {
5658 ERR("Failed to allocate Agent app hash table");
5659 retval = -1;
5660 goto exit_init_data;
5661 }
5662
5663 /*
5664 * These actions must be executed as root. We do that *after* setting up
5665 * the sockets path because we MUST make the check for another daemon using
5666 * those paths *before* trying to set the kernel consumer sockets and init
5667 * kernel tracer.
5668 */
5669 if (is_root) {
5670 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5671 retval = -1;
5672 goto exit_init_data;
5673 }
5674
5675 /* Setup kernel tracer */
5676 if (!opt_no_kernel) {
5677 init_kernel_tracer();
5678 if (kernel_tracer_fd >= 0) {
5679 ret = syscall_init_table();
5680 if (ret < 0) {
5681 ERR("Unable to populate syscall table. "
5682 "Syscall tracing won't work "
5683 "for this session daemon.");
5684 }
5685 }
5686 }
5687
5688 /* Set ulimit for open files */
5689 set_ulimit();
5690 }
5691 /* init lttng_fd tracking must be done after set_ulimit. */
5692 lttng_fd_init();
5693
5694 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5695 retval = -1;
5696 goto exit_init_data;
5697 }
5698
5699 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5700 retval = -1;
5701 goto exit_init_data;
5702 }
5703
5704 /* Setup the needed unix socket */
5705 if (init_daemon_socket()) {
5706 retval = -1;
5707 goto exit_init_data;
5708 }
5709
5710 /* Set credentials to socket */
5711 if (is_root && set_permissions(rundir)) {
5712 retval = -1;
5713 goto exit_init_data;
5714 }
5715
5716 /* Get parent pid if -S, --sig-parent is specified. */
5717 if (opt_sig_parent) {
5718 ppid = getppid();
5719 }
5720
5721 /* Setup the kernel pipe for waking up the kernel thread */
5722 if (is_root && !opt_no_kernel) {
5723 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5724 retval = -1;
5725 goto exit_init_data;
5726 }
5727 }
5728
5729 /* Setup the thread apps communication pipe. */
5730 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5731 retval = -1;
5732 goto exit_init_data;
5733 }
5734
5735 /* Setup the thread apps notify communication pipe. */
5736 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5737 retval = -1;
5738 goto exit_init_data;
5739 }
5740
5741 /* Initialize global buffer per UID and PID registry. */
5742 buffer_reg_init_uid_registry();
5743 buffer_reg_init_pid_registry();
5744
5745 /* Init UST command queue. */
5746 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5747
5748 /*
5749 * Get session list pointer. This pointer MUST NOT be free'd. This list
5750 * is statically declared in session.c
5751 */
5752 session_list_ptr = session_get_list();
5753
5754 cmd_init();
5755
5756 /* Check for the application socket timeout env variable. */
5757 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5758 if (env_app_timeout) {
5759 app_socket_timeout = atoi(env_app_timeout);
5760 } else {
5761 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5762 }
5763
5764 ret = write_pidfile();
5765 if (ret) {
5766 ERR("Error in write_pidfile");
5767 retval = -1;
5768 goto exit_init_data;
5769 }
5770 ret = write_agent_port();
5771 if (ret) {
5772 ERR("Error in write_agent_port");
5773 retval = -1;
5774 goto exit_init_data;
5775 }
5776
5777 /* Initialize communication library */
5778 lttcomm_init();
5779 /* Initialize TCP timeout values */
5780 lttcomm_inet_init();
5781
5782 if (load_session_init_data(&load_info) < 0) {
5783 retval = -1;
5784 goto exit_init_data;
5785 }
5786 load_info->path = opt_load_session_path;
5787
5788 /* Create health-check thread */
5789 ret = pthread_create(&health_thread, NULL,
5790 thread_manage_health, (void *) NULL);
5791 if (ret) {
5792 errno = ret;
5793 PERROR("pthread_create health");
5794 retval = -1;
5795 goto exit_health;
5796 }
5797
5798 /* Create thread to manage the client socket */
5799 ret = pthread_create(&client_thread, NULL,
5800 thread_manage_clients, (void *) NULL);
5801 if (ret) {
5802 errno = ret;
5803 PERROR("pthread_create clients");
5804 retval = -1;
5805 goto exit_client;
5806 }
5807
5808 /* Create thread to dispatch registration */
5809 ret = pthread_create(&dispatch_thread, NULL,
5810 thread_dispatch_ust_registration, (void *) NULL);
5811 if (ret) {
5812 errno = ret;
5813 PERROR("pthread_create dispatch");
5814 retval = -1;
5815 goto exit_dispatch;
5816 }
5817
5818 /* Create thread to manage application registration. */
5819 ret = pthread_create(&reg_apps_thread, NULL,
5820 thread_registration_apps, (void *) NULL);
5821 if (ret) {
5822 errno = ret;
5823 PERROR("pthread_create registration");
5824 retval = -1;
5825 goto exit_reg_apps;
5826 }
5827
5828 /* Create thread to manage application socket */
5829 ret = pthread_create(&apps_thread, NULL,
5830 thread_manage_apps, (void *) NULL);
5831 if (ret) {
5832 errno = ret;
5833 PERROR("pthread_create apps");
5834 retval = -1;
5835 goto exit_apps;
5836 }
5837
5838 /* Create thread to manage application notify socket */
5839 ret = pthread_create(&apps_notify_thread, NULL,
5840 ust_thread_manage_notify, (void *) NULL);
5841 if (ret) {
5842 errno = ret;
5843 PERROR("pthread_create notify");
5844 retval = -1;
5845 goto exit_apps_notify;
5846 }
5847
5848 /* Create agent registration thread. */
5849 ret = pthread_create(&agent_reg_thread, NULL,
5850 agent_thread_manage_registration, (void *) NULL);
5851 if (ret) {
5852 errno = ret;
5853 PERROR("pthread_create agent");
5854 retval = -1;
5855 goto exit_agent_reg;
5856 }
5857
5858 /* Don't start this thread if kernel tracing is not requested nor root */
5859 if (is_root && !opt_no_kernel) {
5860 /* Create kernel thread to manage kernel event */
5861 ret = pthread_create(&kernel_thread, NULL,
5862 thread_manage_kernel, (void *) NULL);
5863 if (ret) {
5864 errno = ret;
5865 PERROR("pthread_create kernel");
5866 retval = -1;
5867 goto exit_kernel;
5868 }
5869 }
5870
5871 /* Create session loading thread. */
5872 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5873 load_info);
5874 if (ret) {
5875 errno = ret;
5876 PERROR("pthread_create load_session_thread");
5877 retval = -1;
5878 goto exit_load_session;
5879 }
5880
5881 /*
5882 * This is where we start awaiting program completion (e.g. through
5883 * signal that asks threads to teardown).
5884 */
5885
5886 ret = pthread_join(load_session_thread, &status);
5887 if (ret) {
5888 errno = ret;
5889 PERROR("pthread_join load_session_thread");
5890 retval = -1;
5891 }
5892 exit_load_session:
5893
5894 if (is_root && !opt_no_kernel) {
5895 ret = pthread_join(kernel_thread, &status);
5896 if (ret) {
5897 errno = ret;
5898 PERROR("pthread_join");
5899 retval = -1;
5900 }
5901 }
5902 exit_kernel:
5903
5904 ret = pthread_join(agent_reg_thread, &status);
5905 if (ret) {
5906 errno = ret;
5907 PERROR("pthread_join agent");
5908 retval = -1;
5909 }
5910 exit_agent_reg:
5911
5912 ret = pthread_join(apps_notify_thread, &status);
5913 if (ret) {
5914 errno = ret;
5915 PERROR("pthread_join apps notify");
5916 retval = -1;
5917 }
5918 exit_apps_notify:
5919
5920 ret = pthread_join(apps_thread, &status);
5921 if (ret) {
5922 errno = ret;
5923 PERROR("pthread_join apps");
5924 retval = -1;
5925 }
5926 exit_apps:
5927
5928 ret = pthread_join(reg_apps_thread, &status);
5929 if (ret) {
5930 errno = ret;
5931 PERROR("pthread_join");
5932 retval = -1;
5933 }
5934 exit_reg_apps:
5935
5936 ret = pthread_join(dispatch_thread, &status);
5937 if (ret) {
5938 errno = ret;
5939 PERROR("pthread_join");
5940 retval = -1;
5941 }
5942 exit_dispatch:
5943
5944 ret = pthread_join(client_thread, &status);
5945 if (ret) {
5946 errno = ret;
5947 PERROR("pthread_join");
5948 retval = -1;
5949 }
5950 exit_client:
5951
5952 ret = pthread_join(health_thread, &status);
5953 if (ret) {
5954 errno = ret;
5955 PERROR("pthread_join health thread");
5956 retval = -1;
5957 }
5958 exit_health:
5959
5960 exit_init_data:
5961 /*
5962 * sessiond_cleanup() is called when no other thread is running, except
5963 * the ht_cleanup thread, which is needed to destroy the hash tables.
5964 */
5965 rcu_thread_online();
5966 sessiond_cleanup();
5967 rcu_thread_offline();
5968 rcu_unregister_thread();
5969
5970 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
5971 if (ret < 0) {
5972 ERR("write error on ht_cleanup quit pipe");
5973 retval = -1;
5974 }
5975
5976 ret = pthread_join(ht_cleanup_thread, &status);
5977 if (ret) {
5978 errno = ret;
5979 PERROR("pthread_join ht cleanup thread");
5980 retval = -1;
5981 }
5982 exit_ht_cleanup:
5983 exit_set_max_size:
5984
5985 utils_close_pipe(ht_cleanup_pipe);
5986 exit_ht_cleanup_pipe:
5987
5988 /*
5989 * Close the ht_cleanup quit pipe.
5990 */
5991 utils_close_pipe(ht_cleanup_quit_pipe);
5992 exit_ht_cleanup_quit_pipe:
5993
5994 health_app_destroy(health_sessiond);
5995 exit_health_sessiond_cleanup:
5996
5997 exit_options:
5998 sessiond_cleanup_options();
5999
6000 exit_set_signal_handler:
6001 if (!retval) {
6002 exit(EXIT_SUCCESS);
6003 } else {
6004 exit(EXIT_FAILURE);
6005 }
6006 }
This page took 0.184521 seconds and 5 git commands to generate.