Fix: app cmd leak on sessiond exit
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _GNU_SOURCE
21 #define _LGPL_SOURCE
22 #include <getopt.h>
23 #include <grp.h>
24 #include <limits.h>
25 #include <paths.h>
26 #include <pthread.h>
27 #include <signal.h>
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <string.h>
31 #include <inttypes.h>
32 #include <sys/mman.h>
33 #include <sys/mount.h>
34 #include <sys/resource.h>
35 #include <sys/socket.h>
36 #include <sys/stat.h>
37 #include <sys/types.h>
38 #include <sys/wait.h>
39 #include <urcu/uatomic.h>
40 #include <unistd.h>
41 #include <config.h>
42
43 #include <common/common.h>
44 #include <common/compat/socket.h>
45 #include <common/compat/getenv.h>
46 #include <common/defaults.h>
47 #include <common/kernel-consumer/kernel-consumer.h>
48 #include <common/futex.h>
49 #include <common/relayd/relayd.h>
50 #include <common/utils.h>
51 #include <common/daemonize.h>
52 #include <common/config/config.h>
53
54 #include "lttng-sessiond.h"
55 #include "buffer-registry.h"
56 #include "channel.h"
57 #include "cmd.h"
58 #include "consumer.h"
59 #include "context.h"
60 #include "event.h"
61 #include "kernel.h"
62 #include "kernel-consumer.h"
63 #include "modprobe.h"
64 #include "shm.h"
65 #include "ust-ctl.h"
66 #include "ust-consumer.h"
67 #include "utils.h"
68 #include "fd-limit.h"
69 #include "health-sessiond.h"
70 #include "testpoint.h"
71 #include "ust-thread.h"
72 #include "agent-thread.h"
73 #include "save.h"
74 #include "load-session-thread.h"
75 #include "syscall.h"
76
77 #define CONSUMERD_FILE "lttng-consumerd"
78
79 const char *progname;
80 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
81 static int tracing_group_name_override;
82 static char *opt_pidfile;
83 static int opt_sig_parent;
84 static int opt_verbose_consumer;
85 static int opt_daemon, opt_background;
86 static int opt_no_kernel;
87 static char *opt_load_session_path;
88 static pid_t ppid; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
90 static char *rundir;
91 static int lockfile_fd = -1;
92
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal;
95
96 /*
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
99 */
100 static struct consumer_data kconsumer_data = {
101 .type = LTTNG_CONSUMER_KERNEL,
102 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
103 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
104 .err_sock = -1,
105 .cmd_sock = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer64_data = {
112 .type = LTTNG_CONSUMER64_UST,
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
118 .lock = PTHREAD_MUTEX_INITIALIZER,
119 .cond = PTHREAD_COND_INITIALIZER,
120 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
121 };
122 static struct consumer_data ustconsumer32_data = {
123 .type = LTTNG_CONSUMER32_UST,
124 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
125 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
126 .err_sock = -1,
127 .cmd_sock = -1,
128 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
129 .lock = PTHREAD_MUTEX_INITIALIZER,
130 .cond = PTHREAD_COND_INITIALIZER,
131 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
132 };
133
134 /* Command line options */
135 static const struct option long_options[] = {
136 { "client-sock", required_argument, 0, 'c' },
137 { "apps-sock", required_argument, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
144 { "consumerd32-path", required_argument, 0, '\0' },
145 { "consumerd32-libdir", required_argument, 0, '\0' },
146 { "consumerd64-path", required_argument, 0, '\0' },
147 { "consumerd64-libdir", required_argument, 0, '\0' },
148 { "daemonize", no_argument, 0, 'd' },
149 { "background", no_argument, 0, 'b' },
150 { "sig-parent", no_argument, 0, 'S' },
151 { "help", no_argument, 0, 'h' },
152 { "group", required_argument, 0, 'g' },
153 { "version", no_argument, 0, 'V' },
154 { "quiet", no_argument, 0, 'q' },
155 { "verbose", no_argument, 0, 'v' },
156 { "verbose-consumer", no_argument, 0, '\0' },
157 { "no-kernel", no_argument, 0, '\0' },
158 { "pidfile", required_argument, 0, 'p' },
159 { "agent-tcp-port", required_argument, 0, '\0' },
160 { "config", required_argument, 0, 'f' },
161 { "load", required_argument, 0, 'l' },
162 { "kmod-probes", required_argument, 0, '\0' },
163 { "extra-kmod-probes", required_argument, 0, '\0' },
164 { NULL, 0, 0, 0 }
165 };
166
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options[] = { "help", "version", "config" };
169
170 /* Shared between threads */
171 static int dispatch_thread_exit;
172
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path[PATH_MAX];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path[PATH_MAX];
177 /* global wait shm path for UST */
178 static char wait_shm_path[PATH_MAX];
179 /* Global health check unix path */
180 static char health_unix_sock_path[PATH_MAX];
181
182 /* Sockets and FDs */
183 static int client_sock = -1;
184 static int apps_sock = -1;
185 int kernel_tracer_fd = -1;
186 static int kernel_poll_pipe[2] = { -1, -1 };
187
188 /*
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
191 */
192 static int thread_quit_pipe[2] = { -1, -1 };
193 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
194
195 /*
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
198 */
199 static int apps_cmd_pipe[2] = { -1, -1 };
200
201 int apps_cmd_notify_pipe[2] = { -1, -1 };
202
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread;
205 static pthread_t apps_notify_thread;
206 static pthread_t reg_apps_thread;
207 static pthread_t client_thread;
208 static pthread_t kernel_thread;
209 static pthread_t dispatch_thread;
210 static pthread_t health_thread;
211 static pthread_t ht_cleanup_thread;
212 static pthread_t agent_reg_thread;
213 static pthread_t load_session_thread;
214
215 /*
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
218 *
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
223 */
224 static struct ust_cmd_queue ust_cmd_queue;
225
226 /*
227 * Pointer initialized before thread creation.
228 *
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
232 *
233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
234 * session_lock_list and session_unlock_list for lock acquisition.
235 */
236 static struct ltt_session_list *session_list_ptr;
237
238 int ust_consumerd64_fd = -1;
239 int ust_consumerd32_fd = -1;
240
241 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
242 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
243 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
244 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
245 static int consumerd32_bin_override;
246 static int consumerd64_bin_override;
247 static int consumerd32_libdir_override;
248 static int consumerd64_libdir_override;
249
250 static const char *module_proc_lttng = "/proc/lttng";
251
252 /*
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
255 */
256 enum consumerd_state {
257 CONSUMER_STARTED = 1,
258 CONSUMER_STOPPED = 2,
259 CONSUMER_ERROR = 3,
260 };
261
262 /*
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
267 *
268 * The following example shows a possible race condition of this scheme:
269 *
270 * consumer thread error happens
271 * client cmd arrives
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
275 * ...
276 *
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
282 */
283 static enum consumerd_state ust_consumerd_state;
284 static enum consumerd_state kernel_consumerd_state;
285
286 /*
287 * Socket timeout for receiving and sending in seconds.
288 */
289 static int app_socket_timeout;
290
291 /* Set in main() with the current page size. */
292 long page_size;
293
294 /* Application health monitoring */
295 struct health_app *health_sessiond;
296
297 /* Agent TCP port for registration. Used by the agent thread. */
298 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
299
300 /* Am I root or not. */
301 int is_root; /* Set to 1 if the daemon is running as root */
302
303 const char * const config_section_name = "sessiond";
304
305 /* Load session thread information to operate. */
306 struct load_session_thread_data *load_info;
307
308 /*
309 * Whether sessiond is ready for commands/health check requests.
310 * NR_LTTNG_SESSIOND_READY must match the number of calls to
311 * sessiond_notify_ready().
312 */
313 #define NR_LTTNG_SESSIOND_READY 3
314 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
315
316 /* Notify parents that we are ready for cmd and health check */
317 LTTNG_HIDDEN
318 void sessiond_notify_ready(void)
319 {
320 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
321 /*
322 * Notify parent pid that we are ready to accept command
323 * for client side. This ppid is the one from the
324 * external process that spawned us.
325 */
326 if (opt_sig_parent) {
327 kill(ppid, SIGUSR1);
328 }
329
330 /*
331 * Notify the parent of the fork() process that we are
332 * ready.
333 */
334 if (opt_daemon || opt_background) {
335 kill(child_ppid, SIGUSR1);
336 }
337 }
338 }
339
340 static
341 void setup_consumerd_path(void)
342 {
343 const char *bin, *libdir;
344
345 /*
346 * Allow INSTALL_BIN_PATH to be used as a target path for the
347 * native architecture size consumer if CONFIG_CONSUMER*_PATH
348 * has not been defined.
349 */
350 #if (CAA_BITS_PER_LONG == 32)
351 if (!consumerd32_bin[0]) {
352 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
353 }
354 if (!consumerd32_libdir[0]) {
355 consumerd32_libdir = INSTALL_LIB_PATH;
356 }
357 #elif (CAA_BITS_PER_LONG == 64)
358 if (!consumerd64_bin[0]) {
359 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
360 }
361 if (!consumerd64_libdir[0]) {
362 consumerd64_libdir = INSTALL_LIB_PATH;
363 }
364 #else
365 #error "Unknown bitness"
366 #endif
367
368 /*
369 * runtime env. var. overrides the build default.
370 */
371 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
372 if (bin) {
373 consumerd32_bin = bin;
374 }
375 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
376 if (bin) {
377 consumerd64_bin = bin;
378 }
379 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
380 if (libdir) {
381 consumerd32_libdir = libdir;
382 }
383 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
384 if (libdir) {
385 consumerd64_libdir = libdir;
386 }
387 }
388
389 static
390 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
391 int *a_pipe)
392 {
393 int ret;
394
395 assert(events);
396
397 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
398 if (ret < 0) {
399 goto error;
400 }
401
402 /* Add quit pipe */
403 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
404 if (ret < 0) {
405 goto error;
406 }
407
408 return 0;
409
410 error:
411 return ret;
412 }
413
414 /*
415 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
416 */
417 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
418 {
419 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
420 }
421
422 /*
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
424 */
425 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
426 size_t size)
427 {
428 return __sessiond_set_thread_pollset(events, size,
429 ht_cleanup_quit_pipe);
430 }
431
432 static
433 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
434 {
435 if (fd == a_pipe && (events & LPOLLIN)) {
436 return 1;
437 }
438 return 0;
439 }
440
441 /*
442 * Check if the thread quit pipe was triggered.
443 *
444 * Return 1 if it was triggered else 0;
445 */
446 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
447 {
448 return __sessiond_check_thread_quit_pipe(fd, events,
449 thread_quit_pipe[0]);
450 }
451
452 /*
453 * Check if the ht_cleanup thread quit pipe was triggered.
454 *
455 * Return 1 if it was triggered else 0;
456 */
457 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
458 {
459 return __sessiond_check_thread_quit_pipe(fd, events,
460 ht_cleanup_quit_pipe[0]);
461 }
462
463 /*
464 * Init thread quit pipe.
465 *
466 * Return -1 on error or 0 if all pipes are created.
467 */
468 static int __init_thread_quit_pipe(int *a_pipe)
469 {
470 int ret, i;
471
472 ret = pipe(a_pipe);
473 if (ret < 0) {
474 PERROR("thread quit pipe");
475 goto error;
476 }
477
478 for (i = 0; i < 2; i++) {
479 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
480 if (ret < 0) {
481 PERROR("fcntl");
482 goto error;
483 }
484 }
485
486 error:
487 return ret;
488 }
489
490 static int init_thread_quit_pipe(void)
491 {
492 return __init_thread_quit_pipe(thread_quit_pipe);
493 }
494
495 static int init_ht_cleanup_quit_pipe(void)
496 {
497 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
498 }
499
500 /*
501 * Stop all threads by closing the thread quit pipe.
502 */
503 static void stop_threads(void)
504 {
505 int ret;
506
507 /* Stopping all threads */
508 DBG("Terminating all threads");
509 ret = notify_thread_pipe(thread_quit_pipe[1]);
510 if (ret < 0) {
511 ERR("write error on thread quit pipe");
512 }
513
514 /* Dispatch thread */
515 CMM_STORE_SHARED(dispatch_thread_exit, 1);
516 futex_nto1_wake(&ust_cmd_queue.futex);
517 }
518
519 /*
520 * Close every consumer sockets.
521 */
522 static void close_consumer_sockets(void)
523 {
524 int ret;
525
526 if (kconsumer_data.err_sock >= 0) {
527 ret = close(kconsumer_data.err_sock);
528 if (ret < 0) {
529 PERROR("kernel consumer err_sock close");
530 }
531 }
532 if (ustconsumer32_data.err_sock >= 0) {
533 ret = close(ustconsumer32_data.err_sock);
534 if (ret < 0) {
535 PERROR("UST consumerd32 err_sock close");
536 }
537 }
538 if (ustconsumer64_data.err_sock >= 0) {
539 ret = close(ustconsumer64_data.err_sock);
540 if (ret < 0) {
541 PERROR("UST consumerd64 err_sock close");
542 }
543 }
544 if (kconsumer_data.cmd_sock >= 0) {
545 ret = close(kconsumer_data.cmd_sock);
546 if (ret < 0) {
547 PERROR("kernel consumer cmd_sock close");
548 }
549 }
550 if (ustconsumer32_data.cmd_sock >= 0) {
551 ret = close(ustconsumer32_data.cmd_sock);
552 if (ret < 0) {
553 PERROR("UST consumerd32 cmd_sock close");
554 }
555 }
556 if (ustconsumer64_data.cmd_sock >= 0) {
557 ret = close(ustconsumer64_data.cmd_sock);
558 if (ret < 0) {
559 PERROR("UST consumerd64 cmd_sock close");
560 }
561 }
562 }
563
564 /*
565 * Generate the full lock file path using the rundir.
566 *
567 * Return the snprintf() return value thus a negative value is an error.
568 */
569 static int generate_lock_file_path(char *path, size_t len)
570 {
571 int ret;
572
573 assert(path);
574 assert(rundir);
575
576 /* Build lockfile path from rundir. */
577 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
578 if (ret < 0) {
579 PERROR("snprintf lockfile path");
580 }
581
582 return ret;
583 }
584
585 /*
586 * Cleanup the session daemon's data structures.
587 */
588 static void sessiond_cleanup(void)
589 {
590 int ret;
591 struct ltt_session *sess, *stmp;
592 char path[PATH_MAX];
593
594 DBG("Cleanup sessiond");
595
596 /*
597 * Close the thread quit pipe. It has already done its job,
598 * since we are now called.
599 */
600 utils_close_pipe(thread_quit_pipe);
601
602 /*
603 * If opt_pidfile is undefined, the default file will be wiped when
604 * removing the rundir.
605 */
606 if (opt_pidfile) {
607 ret = remove(opt_pidfile);
608 if (ret < 0) {
609 PERROR("remove pidfile %s", opt_pidfile);
610 }
611 }
612
613 DBG("Removing sessiond and consumerd content of directory %s", rundir);
614
615 /* sessiond */
616 snprintf(path, PATH_MAX,
617 "%s/%s",
618 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
619 DBG("Removing %s", path);
620 (void) unlink(path);
621
622 snprintf(path, PATH_MAX, "%s/%s", rundir,
623 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
624 DBG("Removing %s", path);
625 (void) unlink(path);
626
627 /* kconsumerd */
628 snprintf(path, PATH_MAX,
629 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
630 rundir);
631 DBG("Removing %s", path);
632 (void) unlink(path);
633
634 snprintf(path, PATH_MAX,
635 DEFAULT_KCONSUMERD_PATH,
636 rundir);
637 DBG("Removing directory %s", path);
638 (void) rmdir(path);
639
640 /* ust consumerd 32 */
641 snprintf(path, PATH_MAX,
642 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
643 rundir);
644 DBG("Removing %s", path);
645 (void) unlink(path);
646
647 snprintf(path, PATH_MAX,
648 DEFAULT_USTCONSUMERD32_PATH,
649 rundir);
650 DBG("Removing directory %s", path);
651 (void) rmdir(path);
652
653 /* ust consumerd 64 */
654 snprintf(path, PATH_MAX,
655 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
656 rundir);
657 DBG("Removing %s", path);
658 (void) unlink(path);
659
660 snprintf(path, PATH_MAX,
661 DEFAULT_USTCONSUMERD64_PATH,
662 rundir);
663 DBG("Removing directory %s", path);
664 (void) rmdir(path);
665
666 DBG("Cleaning up all sessions");
667
668 /* Destroy session list mutex */
669 if (session_list_ptr != NULL) {
670 pthread_mutex_destroy(&session_list_ptr->lock);
671
672 /* Cleanup ALL session */
673 cds_list_for_each_entry_safe(sess, stmp,
674 &session_list_ptr->head, list) {
675 cmd_destroy_session(sess, kernel_poll_pipe[1]);
676 }
677 }
678
679 DBG("Cleaning up all agent apps");
680 agent_app_ht_clean();
681
682 DBG("Closing all UST sockets");
683 ust_app_clean_list();
684 buffer_reg_destroy_registries();
685
686 if (is_root && !opt_no_kernel) {
687 DBG2("Closing kernel fd");
688 if (kernel_tracer_fd >= 0) {
689 ret = close(kernel_tracer_fd);
690 if (ret) {
691 PERROR("close");
692 }
693 }
694 DBG("Unloading kernel modules");
695 modprobe_remove_lttng_all();
696 free(syscall_table);
697 }
698
699 close_consumer_sockets();
700
701 if (load_info) {
702 load_session_destroy_data(load_info);
703 free(load_info);
704 }
705
706 /*
707 * Cleanup lock file by deleting it and finaly closing it which will
708 * release the file system lock.
709 */
710 if (lockfile_fd >= 0) {
711 char lockfile_path[PATH_MAX];
712
713 ret = generate_lock_file_path(lockfile_path,
714 sizeof(lockfile_path));
715 if (ret > 0) {
716 ret = remove(lockfile_path);
717 if (ret < 0) {
718 PERROR("remove lock file");
719 }
720 ret = close(lockfile_fd);
721 if (ret < 0) {
722 PERROR("close lock file");
723 }
724 }
725 }
726
727 /*
728 * We do NOT rmdir rundir because there are other processes
729 * using it, for instance lttng-relayd, which can start in
730 * parallel with this teardown.
731 */
732
733 free(rundir);
734 }
735
736 /*
737 * Cleanup the daemon's option data structures.
738 */
739 static void sessiond_cleanup_options(void)
740 {
741 DBG("Cleaning up options");
742
743 /*
744 * If the override option is set, the pointer points to a *non* const
745 * thus freeing it even though the variable type is set to const.
746 */
747 if (tracing_group_name_override) {
748 free((void *) tracing_group_name);
749 }
750 if (consumerd32_bin_override) {
751 free((void *) consumerd32_bin);
752 }
753 if (consumerd64_bin_override) {
754 free((void *) consumerd64_bin);
755 }
756 if (consumerd32_libdir_override) {
757 free((void *) consumerd32_libdir);
758 }
759 if (consumerd64_libdir_override) {
760 free((void *) consumerd64_libdir);
761 }
762
763 free(opt_pidfile);
764 free(opt_load_session_path);
765 free(kmod_probes_list);
766 free(kmod_extra_probes_list);
767
768 /* <fun> */
769 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
770 "Matthew, BEET driven development works!%c[%dm",
771 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
772 /* </fun> */
773 }
774
775 /*
776 * Send data on a unix socket using the liblttsessiondcomm API.
777 *
778 * Return lttcomm error code.
779 */
780 static int send_unix_sock(int sock, void *buf, size_t len)
781 {
782 /* Check valid length */
783 if (len == 0) {
784 return -1;
785 }
786
787 return lttcomm_send_unix_sock(sock, buf, len);
788 }
789
790 /*
791 * Free memory of a command context structure.
792 */
793 static void clean_command_ctx(struct command_ctx **cmd_ctx)
794 {
795 DBG("Clean command context structure");
796 if (*cmd_ctx) {
797 if ((*cmd_ctx)->llm) {
798 free((*cmd_ctx)->llm);
799 }
800 if ((*cmd_ctx)->lsm) {
801 free((*cmd_ctx)->lsm);
802 }
803 free(*cmd_ctx);
804 *cmd_ctx = NULL;
805 }
806 }
807
808 /*
809 * Notify UST applications using the shm mmap futex.
810 */
811 static int notify_ust_apps(int active)
812 {
813 char *wait_shm_mmap;
814
815 DBG("Notifying applications of session daemon state: %d", active);
816
817 /* See shm.c for this call implying mmap, shm and futex calls */
818 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
819 if (wait_shm_mmap == NULL) {
820 goto error;
821 }
822
823 /* Wake waiting process */
824 futex_wait_update((int32_t *) wait_shm_mmap, active);
825
826 /* Apps notified successfully */
827 return 0;
828
829 error:
830 return -1;
831 }
832
833 /*
834 * Setup the outgoing data buffer for the response (llm) by allocating the
835 * right amount of memory and copying the original information from the lsm
836 * structure.
837 *
838 * Return total size of the buffer pointed by buf.
839 */
840 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
841 {
842 int ret, buf_size;
843
844 buf_size = size;
845
846 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
847 if (cmd_ctx->llm == NULL) {
848 PERROR("zmalloc");
849 ret = -ENOMEM;
850 goto error;
851 }
852
853 /* Copy common data */
854 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
855 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
856
857 cmd_ctx->llm->data_size = size;
858 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
859
860 return buf_size;
861
862 error:
863 return ret;
864 }
865
866 /*
867 * Update the kernel poll set of all channel fd available over all tracing
868 * session. Add the wakeup pipe at the end of the set.
869 */
870 static int update_kernel_poll(struct lttng_poll_event *events)
871 {
872 int ret;
873 struct ltt_session *session;
874 struct ltt_kernel_channel *channel;
875
876 DBG("Updating kernel poll set");
877
878 session_lock_list();
879 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
880 session_lock(session);
881 if (session->kernel_session == NULL) {
882 session_unlock(session);
883 continue;
884 }
885
886 cds_list_for_each_entry(channel,
887 &session->kernel_session->channel_list.head, list) {
888 /* Add channel fd to the kernel poll set */
889 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
890 if (ret < 0) {
891 session_unlock(session);
892 goto error;
893 }
894 DBG("Channel fd %d added to kernel set", channel->fd);
895 }
896 session_unlock(session);
897 }
898 session_unlock_list();
899
900 return 0;
901
902 error:
903 session_unlock_list();
904 return -1;
905 }
906
907 /*
908 * Find the channel fd from 'fd' over all tracing session. When found, check
909 * for new channel stream and send those stream fds to the kernel consumer.
910 *
911 * Useful for CPU hotplug feature.
912 */
913 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
914 {
915 int ret = 0;
916 struct ltt_session *session;
917 struct ltt_kernel_session *ksess;
918 struct ltt_kernel_channel *channel;
919
920 DBG("Updating kernel streams for channel fd %d", fd);
921
922 session_lock_list();
923 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
924 session_lock(session);
925 if (session->kernel_session == NULL) {
926 session_unlock(session);
927 continue;
928 }
929 ksess = session->kernel_session;
930
931 cds_list_for_each_entry(channel,
932 &ksess->channel_list.head, list) {
933 struct lttng_ht_iter iter;
934 struct consumer_socket *socket;
935
936 if (channel->fd != fd) {
937 continue;
938 }
939 DBG("Channel found, updating kernel streams");
940 ret = kernel_open_channel_stream(channel);
941 if (ret < 0) {
942 goto error;
943 }
944 /* Update the stream global counter */
945 ksess->stream_count_global += ret;
946
947 /*
948 * Have we already sent fds to the consumer? If yes, it
949 * means that tracing is started so it is safe to send
950 * our updated stream fds.
951 */
952 if (ksess->consumer_fds_sent != 1
953 || ksess->consumer == NULL) {
954 ret = -1;
955 goto error;
956 }
957
958 rcu_read_lock();
959 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
960 &iter.iter, socket, node.node) {
961 pthread_mutex_lock(socket->lock);
962 ret = kernel_consumer_send_channel_stream(socket,
963 channel, ksess,
964 session->output_traces ? 1 : 0);
965 pthread_mutex_unlock(socket->lock);
966 if (ret < 0) {
967 rcu_read_unlock();
968 goto error;
969 }
970 }
971 rcu_read_unlock();
972 }
973 session_unlock(session);
974 }
975 session_unlock_list();
976 return ret;
977
978 error:
979 session_unlock(session);
980 session_unlock_list();
981 return ret;
982 }
983
984 /*
985 * For each tracing session, update newly registered apps. The session list
986 * lock MUST be acquired before calling this.
987 */
988 static void update_ust_app(int app_sock)
989 {
990 struct ltt_session *sess, *stmp;
991
992 /* Consumer is in an ERROR state. Stop any application update. */
993 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
994 /* Stop the update process since the consumer is dead. */
995 return;
996 }
997
998 /* For all tracing session(s) */
999 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1000 struct ust_app *app;
1001
1002 session_lock(sess);
1003 if (!sess->ust_session) {
1004 goto unlock_session;
1005 }
1006
1007 rcu_read_lock();
1008 assert(app_sock >= 0);
1009 app = ust_app_find_by_sock(app_sock);
1010 if (app == NULL) {
1011 /*
1012 * Application can be unregistered before so
1013 * this is possible hence simply stopping the
1014 * update.
1015 */
1016 DBG3("UST app update failed to find app sock %d",
1017 app_sock);
1018 goto unlock_rcu;
1019 }
1020 ust_app_global_update(sess->ust_session, app);
1021 unlock_rcu:
1022 rcu_read_unlock();
1023 unlock_session:
1024 session_unlock(sess);
1025 }
1026 }
1027
1028 /*
1029 * This thread manage event coming from the kernel.
1030 *
1031 * Features supported in this thread:
1032 * -) CPU Hotplug
1033 */
1034 static void *thread_manage_kernel(void *data)
1035 {
1036 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1037 uint32_t revents, nb_fd;
1038 char tmp;
1039 struct lttng_poll_event events;
1040
1041 DBG("[thread] Thread manage kernel started");
1042
1043 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1044
1045 /*
1046 * This first step of the while is to clean this structure which could free
1047 * non NULL pointers so initialize it before the loop.
1048 */
1049 lttng_poll_init(&events);
1050
1051 if (testpoint(sessiond_thread_manage_kernel)) {
1052 goto error_testpoint;
1053 }
1054
1055 health_code_update();
1056
1057 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1058 goto error_testpoint;
1059 }
1060
1061 while (1) {
1062 health_code_update();
1063
1064 if (update_poll_flag == 1) {
1065 /* Clean events object. We are about to populate it again. */
1066 lttng_poll_clean(&events);
1067
1068 ret = sessiond_set_thread_pollset(&events, 2);
1069 if (ret < 0) {
1070 goto error_poll_create;
1071 }
1072
1073 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1074 if (ret < 0) {
1075 goto error;
1076 }
1077
1078 /* This will add the available kernel channel if any. */
1079 ret = update_kernel_poll(&events);
1080 if (ret < 0) {
1081 goto error;
1082 }
1083 update_poll_flag = 0;
1084 }
1085
1086 DBG("Thread kernel polling");
1087
1088 /* Poll infinite value of time */
1089 restart:
1090 health_poll_entry();
1091 ret = lttng_poll_wait(&events, -1);
1092 DBG("Thread kernel return from poll on %d fds",
1093 LTTNG_POLL_GETNB(&events));
1094 health_poll_exit();
1095 if (ret < 0) {
1096 /*
1097 * Restart interrupted system call.
1098 */
1099 if (errno == EINTR) {
1100 goto restart;
1101 }
1102 goto error;
1103 } else if (ret == 0) {
1104 /* Should not happen since timeout is infinite */
1105 ERR("Return value of poll is 0 with an infinite timeout.\n"
1106 "This should not have happened! Continuing...");
1107 continue;
1108 }
1109
1110 nb_fd = ret;
1111
1112 for (i = 0; i < nb_fd; i++) {
1113 /* Fetch once the poll data */
1114 revents = LTTNG_POLL_GETEV(&events, i);
1115 pollfd = LTTNG_POLL_GETFD(&events, i);
1116
1117 health_code_update();
1118
1119 if (!revents) {
1120 /* No activity for this FD (poll implementation). */
1121 continue;
1122 }
1123
1124 /* Thread quit pipe has been closed. Killing thread. */
1125 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1126 if (ret) {
1127 err = 0;
1128 goto exit;
1129 }
1130
1131 /* Check for data on kernel pipe */
1132 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1133 (void) lttng_read(kernel_poll_pipe[0],
1134 &tmp, 1);
1135 /*
1136 * Ret value is useless here, if this pipe gets any actions an
1137 * update is required anyway.
1138 */
1139 update_poll_flag = 1;
1140 continue;
1141 } else {
1142 /*
1143 * New CPU detected by the kernel. Adding kernel stream to
1144 * kernel session and updating the kernel consumer
1145 */
1146 if (revents & LPOLLIN) {
1147 ret = update_kernel_stream(&kconsumer_data, pollfd);
1148 if (ret < 0) {
1149 continue;
1150 }
1151 break;
1152 /*
1153 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1154 * and unregister kernel stream at this point.
1155 */
1156 }
1157 }
1158 }
1159 }
1160
1161 exit:
1162 error:
1163 lttng_poll_clean(&events);
1164 error_poll_create:
1165 error_testpoint:
1166 utils_close_pipe(kernel_poll_pipe);
1167 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1168 if (err) {
1169 health_error();
1170 ERR("Health error occurred in %s", __func__);
1171 WARN("Kernel thread died unexpectedly. "
1172 "Kernel tracing can continue but CPU hotplug is disabled.");
1173 }
1174 health_unregister(health_sessiond);
1175 DBG("Kernel thread dying");
1176 return NULL;
1177 }
1178
1179 /*
1180 * Signal pthread condition of the consumer data that the thread.
1181 */
1182 static void signal_consumer_condition(struct consumer_data *data, int state)
1183 {
1184 pthread_mutex_lock(&data->cond_mutex);
1185
1186 /*
1187 * The state is set before signaling. It can be any value, it's the waiter
1188 * job to correctly interpret this condition variable associated to the
1189 * consumer pthread_cond.
1190 *
1191 * A value of 0 means that the corresponding thread of the consumer data
1192 * was not started. 1 indicates that the thread has started and is ready
1193 * for action. A negative value means that there was an error during the
1194 * thread bootstrap.
1195 */
1196 data->consumer_thread_is_ready = state;
1197 (void) pthread_cond_signal(&data->cond);
1198
1199 pthread_mutex_unlock(&data->cond_mutex);
1200 }
1201
1202 /*
1203 * This thread manage the consumer error sent back to the session daemon.
1204 */
1205 static void *thread_manage_consumer(void *data)
1206 {
1207 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1208 uint32_t revents, nb_fd;
1209 enum lttcomm_return_code code;
1210 struct lttng_poll_event events;
1211 struct consumer_data *consumer_data = data;
1212
1213 DBG("[thread] Manage consumer started");
1214
1215 rcu_register_thread();
1216 rcu_thread_online();
1217
1218 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1219
1220 health_code_update();
1221
1222 /*
1223 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1224 * metadata_sock. Nothing more will be added to this poll set.
1225 */
1226 ret = sessiond_set_thread_pollset(&events, 3);
1227 if (ret < 0) {
1228 goto error_poll;
1229 }
1230
1231 /*
1232 * The error socket here is already in a listening state which was done
1233 * just before spawning this thread to avoid a race between the consumer
1234 * daemon exec trying to connect and the listen() call.
1235 */
1236 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1237 if (ret < 0) {
1238 goto error;
1239 }
1240
1241 health_code_update();
1242
1243 /* Infinite blocking call, waiting for transmission */
1244 restart:
1245 health_poll_entry();
1246
1247 if (testpoint(sessiond_thread_manage_consumer)) {
1248 goto error;
1249 }
1250
1251 ret = lttng_poll_wait(&events, -1);
1252 health_poll_exit();
1253 if (ret < 0) {
1254 /*
1255 * Restart interrupted system call.
1256 */
1257 if (errno == EINTR) {
1258 goto restart;
1259 }
1260 goto error;
1261 }
1262
1263 nb_fd = ret;
1264
1265 for (i = 0; i < nb_fd; i++) {
1266 /* Fetch once the poll data */
1267 revents = LTTNG_POLL_GETEV(&events, i);
1268 pollfd = LTTNG_POLL_GETFD(&events, i);
1269
1270 health_code_update();
1271
1272 if (!revents) {
1273 /* No activity for this FD (poll implementation). */
1274 continue;
1275 }
1276
1277 /* Thread quit pipe has been closed. Killing thread. */
1278 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1279 if (ret) {
1280 err = 0;
1281 goto exit;
1282 }
1283
1284 /* Event on the registration socket */
1285 if (pollfd == consumer_data->err_sock) {
1286 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1287 ERR("consumer err socket poll error");
1288 goto error;
1289 }
1290 }
1291 }
1292
1293 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1294 if (sock < 0) {
1295 goto error;
1296 }
1297
1298 /*
1299 * Set the CLOEXEC flag. Return code is useless because either way, the
1300 * show must go on.
1301 */
1302 (void) utils_set_fd_cloexec(sock);
1303
1304 health_code_update();
1305
1306 DBG2("Receiving code from consumer err_sock");
1307
1308 /* Getting status code from kconsumerd */
1309 ret = lttcomm_recv_unix_sock(sock, &code,
1310 sizeof(enum lttcomm_return_code));
1311 if (ret <= 0) {
1312 goto error;
1313 }
1314
1315 health_code_update();
1316 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1317 /* Connect both socket, command and metadata. */
1318 consumer_data->cmd_sock =
1319 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1320 consumer_data->metadata_fd =
1321 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1322 if (consumer_data->cmd_sock < 0
1323 || consumer_data->metadata_fd < 0) {
1324 PERROR("consumer connect cmd socket");
1325 /* On error, signal condition and quit. */
1326 signal_consumer_condition(consumer_data, -1);
1327 goto error;
1328 }
1329 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1330 /* Create metadata socket lock. */
1331 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1332 if (consumer_data->metadata_sock.lock == NULL) {
1333 PERROR("zmalloc pthread mutex");
1334 ret = -1;
1335 goto error;
1336 }
1337 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1338
1339 signal_consumer_condition(consumer_data, 1);
1340 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1341 DBG("Consumer metadata socket ready (fd: %d)",
1342 consumer_data->metadata_fd);
1343 } else {
1344 ERR("consumer error when waiting for SOCK_READY : %s",
1345 lttcomm_get_readable_code(-code));
1346 goto error;
1347 }
1348
1349 /* Remove the consumerd error sock since we've established a connexion */
1350 ret = lttng_poll_del(&events, consumer_data->err_sock);
1351 if (ret < 0) {
1352 goto error;
1353 }
1354
1355 /* Add new accepted error socket. */
1356 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1357 if (ret < 0) {
1358 goto error;
1359 }
1360
1361 /* Add metadata socket that is successfully connected. */
1362 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1363 LPOLLIN | LPOLLRDHUP);
1364 if (ret < 0) {
1365 goto error;
1366 }
1367
1368 health_code_update();
1369
1370 /* Infinite blocking call, waiting for transmission */
1371 restart_poll:
1372 while (1) {
1373 health_code_update();
1374
1375 /* Exit the thread because the thread quit pipe has been triggered. */
1376 if (should_quit) {
1377 /* Not a health error. */
1378 err = 0;
1379 goto exit;
1380 }
1381
1382 health_poll_entry();
1383 ret = lttng_poll_wait(&events, -1);
1384 health_poll_exit();
1385 if (ret < 0) {
1386 /*
1387 * Restart interrupted system call.
1388 */
1389 if (errno == EINTR) {
1390 goto restart_poll;
1391 }
1392 goto error;
1393 }
1394
1395 nb_fd = ret;
1396
1397 for (i = 0; i < nb_fd; i++) {
1398 /* Fetch once the poll data */
1399 revents = LTTNG_POLL_GETEV(&events, i);
1400 pollfd = LTTNG_POLL_GETFD(&events, i);
1401
1402 health_code_update();
1403
1404 if (!revents) {
1405 /* No activity for this FD (poll implementation). */
1406 continue;
1407 }
1408
1409 /*
1410 * Thread quit pipe has been triggered, flag that we should stop
1411 * but continue the current loop to handle potential data from
1412 * consumer.
1413 */
1414 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1415
1416 if (pollfd == sock) {
1417 /* Event on the consumerd socket */
1418 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1419 ERR("consumer err socket second poll error");
1420 goto error;
1421 }
1422 health_code_update();
1423 /* Wait for any kconsumerd error */
1424 ret = lttcomm_recv_unix_sock(sock, &code,
1425 sizeof(enum lttcomm_return_code));
1426 if (ret <= 0) {
1427 ERR("consumer closed the command socket");
1428 goto error;
1429 }
1430
1431 ERR("consumer return code : %s",
1432 lttcomm_get_readable_code(-code));
1433
1434 goto exit;
1435 } else if (pollfd == consumer_data->metadata_fd) {
1436 /* UST metadata requests */
1437 ret = ust_consumer_metadata_request(
1438 &consumer_data->metadata_sock);
1439 if (ret < 0) {
1440 ERR("Handling metadata request");
1441 goto error;
1442 }
1443 }
1444 /* No need for an else branch all FDs are tested prior. */
1445 }
1446 health_code_update();
1447 }
1448
1449 exit:
1450 error:
1451 /*
1452 * We lock here because we are about to close the sockets and some other
1453 * thread might be using them so get exclusive access which will abort all
1454 * other consumer command by other threads.
1455 */
1456 pthread_mutex_lock(&consumer_data->lock);
1457
1458 /* Immediately set the consumerd state to stopped */
1459 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1460 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1461 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1462 consumer_data->type == LTTNG_CONSUMER32_UST) {
1463 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1464 } else {
1465 /* Code flow error... */
1466 assert(0);
1467 }
1468
1469 if (consumer_data->err_sock >= 0) {
1470 ret = close(consumer_data->err_sock);
1471 if (ret) {
1472 PERROR("close");
1473 }
1474 consumer_data->err_sock = -1;
1475 }
1476 if (consumer_data->cmd_sock >= 0) {
1477 ret = close(consumer_data->cmd_sock);
1478 if (ret) {
1479 PERROR("close");
1480 }
1481 consumer_data->cmd_sock = -1;
1482 }
1483 if (consumer_data->metadata_sock.fd_ptr &&
1484 *consumer_data->metadata_sock.fd_ptr >= 0) {
1485 ret = close(*consumer_data->metadata_sock.fd_ptr);
1486 if (ret) {
1487 PERROR("close");
1488 }
1489 }
1490 if (sock >= 0) {
1491 ret = close(sock);
1492 if (ret) {
1493 PERROR("close");
1494 }
1495 }
1496
1497 unlink(consumer_data->err_unix_sock_path);
1498 unlink(consumer_data->cmd_unix_sock_path);
1499 consumer_data->pid = 0;
1500 pthread_mutex_unlock(&consumer_data->lock);
1501
1502 /* Cleanup metadata socket mutex. */
1503 if (consumer_data->metadata_sock.lock) {
1504 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1505 free(consumer_data->metadata_sock.lock);
1506 }
1507 lttng_poll_clean(&events);
1508 error_poll:
1509 if (err) {
1510 health_error();
1511 ERR("Health error occurred in %s", __func__);
1512 }
1513 health_unregister(health_sessiond);
1514 DBG("consumer thread cleanup completed");
1515
1516 rcu_thread_offline();
1517 rcu_unregister_thread();
1518
1519 return NULL;
1520 }
1521
1522 /*
1523 * This thread manage application communication.
1524 */
1525 static void *thread_manage_apps(void *data)
1526 {
1527 int i, ret, pollfd, err = -1;
1528 ssize_t size_ret;
1529 uint32_t revents, nb_fd;
1530 struct lttng_poll_event events;
1531
1532 DBG("[thread] Manage application started");
1533
1534 rcu_register_thread();
1535 rcu_thread_online();
1536
1537 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1538
1539 if (testpoint(sessiond_thread_manage_apps)) {
1540 goto error_testpoint;
1541 }
1542
1543 health_code_update();
1544
1545 ret = sessiond_set_thread_pollset(&events, 2);
1546 if (ret < 0) {
1547 goto error_poll_create;
1548 }
1549
1550 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1551 if (ret < 0) {
1552 goto error;
1553 }
1554
1555 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1556 goto error;
1557 }
1558
1559 health_code_update();
1560
1561 while (1) {
1562 DBG("Apps thread polling");
1563
1564 /* Inifinite blocking call, waiting for transmission */
1565 restart:
1566 health_poll_entry();
1567 ret = lttng_poll_wait(&events, -1);
1568 DBG("Apps thread return from poll on %d fds",
1569 LTTNG_POLL_GETNB(&events));
1570 health_poll_exit();
1571 if (ret < 0) {
1572 /*
1573 * Restart interrupted system call.
1574 */
1575 if (errno == EINTR) {
1576 goto restart;
1577 }
1578 goto error;
1579 }
1580
1581 nb_fd = ret;
1582
1583 for (i = 0; i < nb_fd; i++) {
1584 /* Fetch once the poll data */
1585 revents = LTTNG_POLL_GETEV(&events, i);
1586 pollfd = LTTNG_POLL_GETFD(&events, i);
1587
1588 health_code_update();
1589
1590 if (!revents) {
1591 /* No activity for this FD (poll implementation). */
1592 continue;
1593 }
1594
1595 /* Thread quit pipe has been closed. Killing thread. */
1596 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1597 if (ret) {
1598 err = 0;
1599 goto exit;
1600 }
1601
1602 /* Inspect the apps cmd pipe */
1603 if (pollfd == apps_cmd_pipe[0]) {
1604 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1605 ERR("Apps command pipe error");
1606 goto error;
1607 } else if (revents & LPOLLIN) {
1608 int sock;
1609
1610 /* Empty pipe */
1611 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1612 if (size_ret < sizeof(sock)) {
1613 PERROR("read apps cmd pipe");
1614 goto error;
1615 }
1616
1617 health_code_update();
1618
1619 /*
1620 * We only monitor the error events of the socket. This
1621 * thread does not handle any incoming data from UST
1622 * (POLLIN).
1623 */
1624 ret = lttng_poll_add(&events, sock,
1625 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1626 if (ret < 0) {
1627 goto error;
1628 }
1629
1630 DBG("Apps with sock %d added to poll set", sock);
1631 }
1632 } else {
1633 /*
1634 * At this point, we know that a registered application made
1635 * the event at poll_wait.
1636 */
1637 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1638 /* Removing from the poll set */
1639 ret = lttng_poll_del(&events, pollfd);
1640 if (ret < 0) {
1641 goto error;
1642 }
1643
1644 /* Socket closed on remote end. */
1645 ust_app_unregister(pollfd);
1646 }
1647 }
1648
1649 health_code_update();
1650 }
1651 }
1652
1653 exit:
1654 error:
1655 lttng_poll_clean(&events);
1656 error_poll_create:
1657 error_testpoint:
1658 utils_close_pipe(apps_cmd_pipe);
1659 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1660
1661 /*
1662 * We don't clean the UST app hash table here since already registered
1663 * applications can still be controlled so let them be until the session
1664 * daemon dies or the applications stop.
1665 */
1666
1667 if (err) {
1668 health_error();
1669 ERR("Health error occurred in %s", __func__);
1670 }
1671 health_unregister(health_sessiond);
1672 DBG("Application communication apps thread cleanup complete");
1673 rcu_thread_offline();
1674 rcu_unregister_thread();
1675 return NULL;
1676 }
1677
1678 /*
1679 * Send a socket to a thread This is called from the dispatch UST registration
1680 * thread once all sockets are set for the application.
1681 *
1682 * The sock value can be invalid, we don't really care, the thread will handle
1683 * it and make the necessary cleanup if so.
1684 *
1685 * On success, return 0 else a negative value being the errno message of the
1686 * write().
1687 */
1688 static int send_socket_to_thread(int fd, int sock)
1689 {
1690 ssize_t ret;
1691
1692 /*
1693 * It's possible that the FD is set as invalid with -1 concurrently just
1694 * before calling this function being a shutdown state of the thread.
1695 */
1696 if (fd < 0) {
1697 ret = -EBADF;
1698 goto error;
1699 }
1700
1701 ret = lttng_write(fd, &sock, sizeof(sock));
1702 if (ret < sizeof(sock)) {
1703 PERROR("write apps pipe %d", fd);
1704 if (ret < 0) {
1705 ret = -errno;
1706 }
1707 goto error;
1708 }
1709
1710 /* All good. Don't send back the write positive ret value. */
1711 ret = 0;
1712 error:
1713 return (int) ret;
1714 }
1715
1716 /*
1717 * Sanitize the wait queue of the dispatch registration thread meaning removing
1718 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1719 * notify socket is never received.
1720 */
1721 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1722 {
1723 int ret, nb_fd = 0, i;
1724 unsigned int fd_added = 0;
1725 struct lttng_poll_event events;
1726 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1727
1728 assert(wait_queue);
1729
1730 lttng_poll_init(&events);
1731
1732 /* Just skip everything for an empty queue. */
1733 if (!wait_queue->count) {
1734 goto end;
1735 }
1736
1737 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1738 if (ret < 0) {
1739 goto error_create;
1740 }
1741
1742 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1743 &wait_queue->head, head) {
1744 assert(wait_node->app);
1745 ret = lttng_poll_add(&events, wait_node->app->sock,
1746 LPOLLHUP | LPOLLERR);
1747 if (ret < 0) {
1748 goto error;
1749 }
1750
1751 fd_added = 1;
1752 }
1753
1754 if (!fd_added) {
1755 goto end;
1756 }
1757
1758 /*
1759 * Poll but don't block so we can quickly identify the faulty events and
1760 * clean them afterwards from the wait queue.
1761 */
1762 ret = lttng_poll_wait(&events, 0);
1763 if (ret < 0) {
1764 goto error;
1765 }
1766 nb_fd = ret;
1767
1768 for (i = 0; i < nb_fd; i++) {
1769 /* Get faulty FD. */
1770 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1771 int pollfd = LTTNG_POLL_GETFD(&events, i);
1772
1773 if (!revents) {
1774 /* No activity for this FD (poll implementation). */
1775 continue;
1776 }
1777
1778 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1779 &wait_queue->head, head) {
1780 if (pollfd == wait_node->app->sock &&
1781 (revents & (LPOLLHUP | LPOLLERR))) {
1782 cds_list_del(&wait_node->head);
1783 wait_queue->count--;
1784 ust_app_destroy(wait_node->app);
1785 free(wait_node);
1786 break;
1787 }
1788 }
1789 }
1790
1791 if (nb_fd > 0) {
1792 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1793 }
1794
1795 end:
1796 lttng_poll_clean(&events);
1797 return;
1798
1799 error:
1800 lttng_poll_clean(&events);
1801 error_create:
1802 ERR("Unable to sanitize wait queue");
1803 return;
1804 }
1805
1806 /*
1807 * Dispatch request from the registration threads to the application
1808 * communication thread.
1809 */
1810 static void *thread_dispatch_ust_registration(void *data)
1811 {
1812 int ret, err = -1;
1813 struct cds_wfcq_node *node;
1814 struct ust_command *ust_cmd = NULL;
1815 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1816 struct ust_reg_wait_queue wait_queue = {
1817 .count = 0,
1818 };
1819
1820 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1821
1822 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1823 goto error_testpoint;
1824 }
1825
1826 health_code_update();
1827
1828 CDS_INIT_LIST_HEAD(&wait_queue.head);
1829
1830 DBG("[thread] Dispatch UST command started");
1831
1832 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1833 health_code_update();
1834
1835 /* Atomically prepare the queue futex */
1836 futex_nto1_prepare(&ust_cmd_queue.futex);
1837
1838 do {
1839 struct ust_app *app = NULL;
1840 ust_cmd = NULL;
1841
1842 /*
1843 * Make sure we don't have node(s) that have hung up before receiving
1844 * the notify socket. This is to clean the list in order to avoid
1845 * memory leaks from notify socket that are never seen.
1846 */
1847 sanitize_wait_queue(&wait_queue);
1848
1849 health_code_update();
1850 /* Dequeue command for registration */
1851 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1852 if (node == NULL) {
1853 DBG("Woken up but nothing in the UST command queue");
1854 /* Continue thread execution */
1855 break;
1856 }
1857
1858 ust_cmd = caa_container_of(node, struct ust_command, node);
1859
1860 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1861 " gid:%d sock:%d name:%s (version %d.%d)",
1862 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1863 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1864 ust_cmd->sock, ust_cmd->reg_msg.name,
1865 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1866
1867 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1868 wait_node = zmalloc(sizeof(*wait_node));
1869 if (!wait_node) {
1870 PERROR("zmalloc wait_node dispatch");
1871 ret = close(ust_cmd->sock);
1872 if (ret < 0) {
1873 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1874 }
1875 lttng_fd_put(LTTNG_FD_APPS, 1);
1876 free(ust_cmd);
1877 goto error;
1878 }
1879 CDS_INIT_LIST_HEAD(&wait_node->head);
1880
1881 /* Create application object if socket is CMD. */
1882 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1883 ust_cmd->sock);
1884 if (!wait_node->app) {
1885 ret = close(ust_cmd->sock);
1886 if (ret < 0) {
1887 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1888 }
1889 lttng_fd_put(LTTNG_FD_APPS, 1);
1890 free(wait_node);
1891 free(ust_cmd);
1892 continue;
1893 }
1894 /*
1895 * Add application to the wait queue so we can set the notify
1896 * socket before putting this object in the global ht.
1897 */
1898 cds_list_add(&wait_node->head, &wait_queue.head);
1899 wait_queue.count++;
1900
1901 free(ust_cmd);
1902 /*
1903 * We have to continue here since we don't have the notify
1904 * socket and the application MUST be added to the hash table
1905 * only at that moment.
1906 */
1907 continue;
1908 } else {
1909 /*
1910 * Look for the application in the local wait queue and set the
1911 * notify socket if found.
1912 */
1913 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1914 &wait_queue.head, head) {
1915 health_code_update();
1916 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1917 wait_node->app->notify_sock = ust_cmd->sock;
1918 cds_list_del(&wait_node->head);
1919 wait_queue.count--;
1920 app = wait_node->app;
1921 free(wait_node);
1922 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1923 break;
1924 }
1925 }
1926
1927 /*
1928 * With no application at this stage the received socket is
1929 * basically useless so close it before we free the cmd data
1930 * structure for good.
1931 */
1932 if (!app) {
1933 ret = close(ust_cmd->sock);
1934 if (ret < 0) {
1935 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1936 }
1937 lttng_fd_put(LTTNG_FD_APPS, 1);
1938 }
1939 free(ust_cmd);
1940 }
1941
1942 if (app) {
1943 /*
1944 * @session_lock_list
1945 *
1946 * Lock the global session list so from the register up to the
1947 * registration done message, no thread can see the application
1948 * and change its state.
1949 */
1950 session_lock_list();
1951 rcu_read_lock();
1952
1953 /*
1954 * Add application to the global hash table. This needs to be
1955 * done before the update to the UST registry can locate the
1956 * application.
1957 */
1958 ust_app_add(app);
1959
1960 /* Set app version. This call will print an error if needed. */
1961 (void) ust_app_version(app);
1962
1963 /* Send notify socket through the notify pipe. */
1964 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1965 app->notify_sock);
1966 if (ret < 0) {
1967 rcu_read_unlock();
1968 session_unlock_list();
1969 /*
1970 * No notify thread, stop the UST tracing. However, this is
1971 * not an internal error of the this thread thus setting
1972 * the health error code to a normal exit.
1973 */
1974 err = 0;
1975 goto error;
1976 }
1977
1978 /*
1979 * Update newly registered application with the tracing
1980 * registry info already enabled information.
1981 */
1982 update_ust_app(app->sock);
1983
1984 /*
1985 * Don't care about return value. Let the manage apps threads
1986 * handle app unregistration upon socket close.
1987 */
1988 (void) ust_app_register_done(app->sock);
1989
1990 /*
1991 * Even if the application socket has been closed, send the app
1992 * to the thread and unregistration will take place at that
1993 * place.
1994 */
1995 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1996 if (ret < 0) {
1997 rcu_read_unlock();
1998 session_unlock_list();
1999 /*
2000 * No apps. thread, stop the UST tracing. However, this is
2001 * not an internal error of the this thread thus setting
2002 * the health error code to a normal exit.
2003 */
2004 err = 0;
2005 goto error;
2006 }
2007
2008 rcu_read_unlock();
2009 session_unlock_list();
2010 }
2011 } while (node != NULL);
2012
2013 health_poll_entry();
2014 /* Futex wait on queue. Blocking call on futex() */
2015 futex_nto1_wait(&ust_cmd_queue.futex);
2016 health_poll_exit();
2017 }
2018 /* Normal exit, no error */
2019 err = 0;
2020
2021 error:
2022 /* Clean up wait queue. */
2023 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2024 &wait_queue.head, head) {
2025 cds_list_del(&wait_node->head);
2026 wait_queue.count--;
2027 free(wait_node);
2028 }
2029
2030 /* Empty command queue. */
2031 for (;;) {
2032 /* Dequeue command for registration */
2033 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2034 if (node == NULL) {
2035 break;
2036 }
2037 ust_cmd = caa_container_of(node, struct ust_command, node);
2038 ret = close(ust_cmd->sock);
2039 if (ret < 0) {
2040 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2041 }
2042 lttng_fd_put(LTTNG_FD_APPS, 1);
2043 free(ust_cmd);
2044 }
2045
2046 error_testpoint:
2047 DBG("Dispatch thread dying");
2048 if (err) {
2049 health_error();
2050 ERR("Health error occurred in %s", __func__);
2051 }
2052 health_unregister(health_sessiond);
2053 return NULL;
2054 }
2055
2056 /*
2057 * This thread manage application registration.
2058 */
2059 static void *thread_registration_apps(void *data)
2060 {
2061 int sock = -1, i, ret, pollfd, err = -1;
2062 uint32_t revents, nb_fd;
2063 struct lttng_poll_event events;
2064 /*
2065 * Get allocated in this thread, enqueued to a global queue, dequeued and
2066 * freed in the manage apps thread.
2067 */
2068 struct ust_command *ust_cmd = NULL;
2069
2070 DBG("[thread] Manage application registration started");
2071
2072 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2073
2074 if (testpoint(sessiond_thread_registration_apps)) {
2075 goto error_testpoint;
2076 }
2077
2078 ret = lttcomm_listen_unix_sock(apps_sock);
2079 if (ret < 0) {
2080 goto error_listen;
2081 }
2082
2083 /*
2084 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2085 * more will be added to this poll set.
2086 */
2087 ret = sessiond_set_thread_pollset(&events, 2);
2088 if (ret < 0) {
2089 goto error_create_poll;
2090 }
2091
2092 /* Add the application registration socket */
2093 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2094 if (ret < 0) {
2095 goto error_poll_add;
2096 }
2097
2098 /* Notify all applications to register */
2099 ret = notify_ust_apps(1);
2100 if (ret < 0) {
2101 ERR("Failed to notify applications or create the wait shared memory.\n"
2102 "Execution continues but there might be problem for already\n"
2103 "running applications that wishes to register.");
2104 }
2105
2106 while (1) {
2107 DBG("Accepting application registration");
2108
2109 /* Inifinite blocking call, waiting for transmission */
2110 restart:
2111 health_poll_entry();
2112 ret = lttng_poll_wait(&events, -1);
2113 health_poll_exit();
2114 if (ret < 0) {
2115 /*
2116 * Restart interrupted system call.
2117 */
2118 if (errno == EINTR) {
2119 goto restart;
2120 }
2121 goto error;
2122 }
2123
2124 nb_fd = ret;
2125
2126 for (i = 0; i < nb_fd; i++) {
2127 health_code_update();
2128
2129 /* Fetch once the poll data */
2130 revents = LTTNG_POLL_GETEV(&events, i);
2131 pollfd = LTTNG_POLL_GETFD(&events, i);
2132
2133 if (!revents) {
2134 /* No activity for this FD (poll implementation). */
2135 continue;
2136 }
2137
2138 /* Thread quit pipe has been closed. Killing thread. */
2139 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2140 if (ret) {
2141 err = 0;
2142 goto exit;
2143 }
2144
2145 /* Event on the registration socket */
2146 if (pollfd == apps_sock) {
2147 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2148 ERR("Register apps socket poll error");
2149 goto error;
2150 } else if (revents & LPOLLIN) {
2151 sock = lttcomm_accept_unix_sock(apps_sock);
2152 if (sock < 0) {
2153 goto error;
2154 }
2155
2156 /*
2157 * Set socket timeout for both receiving and ending.
2158 * app_socket_timeout is in seconds, whereas
2159 * lttcomm_setsockopt_rcv_timeout and
2160 * lttcomm_setsockopt_snd_timeout expect msec as
2161 * parameter.
2162 */
2163 (void) lttcomm_setsockopt_rcv_timeout(sock,
2164 app_socket_timeout * 1000);
2165 (void) lttcomm_setsockopt_snd_timeout(sock,
2166 app_socket_timeout * 1000);
2167
2168 /*
2169 * Set the CLOEXEC flag. Return code is useless because
2170 * either way, the show must go on.
2171 */
2172 (void) utils_set_fd_cloexec(sock);
2173
2174 /* Create UST registration command for enqueuing */
2175 ust_cmd = zmalloc(sizeof(struct ust_command));
2176 if (ust_cmd == NULL) {
2177 PERROR("ust command zmalloc");
2178 ret = close(sock);
2179 if (ret) {
2180 PERROR("close");
2181 }
2182 goto error;
2183 }
2184
2185 /*
2186 * Using message-based transmissions to ensure we don't
2187 * have to deal with partially received messages.
2188 */
2189 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2190 if (ret < 0) {
2191 ERR("Exhausted file descriptors allowed for applications.");
2192 free(ust_cmd);
2193 ret = close(sock);
2194 if (ret) {
2195 PERROR("close");
2196 }
2197 sock = -1;
2198 continue;
2199 }
2200
2201 health_code_update();
2202 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2203 if (ret < 0) {
2204 free(ust_cmd);
2205 /* Close socket of the application. */
2206 ret = close(sock);
2207 if (ret) {
2208 PERROR("close");
2209 }
2210 lttng_fd_put(LTTNG_FD_APPS, 1);
2211 sock = -1;
2212 continue;
2213 }
2214 health_code_update();
2215
2216 ust_cmd->sock = sock;
2217 sock = -1;
2218
2219 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2220 " gid:%d sock:%d name:%s (version %d.%d)",
2221 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2222 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2223 ust_cmd->sock, ust_cmd->reg_msg.name,
2224 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2225
2226 /*
2227 * Lock free enqueue the registration request. The red pill
2228 * has been taken! This apps will be part of the *system*.
2229 */
2230 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2231
2232 /*
2233 * Wake the registration queue futex. Implicit memory
2234 * barrier with the exchange in cds_wfcq_enqueue.
2235 */
2236 futex_nto1_wake(&ust_cmd_queue.futex);
2237 }
2238 }
2239 }
2240 }
2241
2242 exit:
2243 error:
2244 /* Notify that the registration thread is gone */
2245 notify_ust_apps(0);
2246
2247 if (apps_sock >= 0) {
2248 ret = close(apps_sock);
2249 if (ret) {
2250 PERROR("close");
2251 }
2252 }
2253 if (sock >= 0) {
2254 ret = close(sock);
2255 if (ret) {
2256 PERROR("close");
2257 }
2258 lttng_fd_put(LTTNG_FD_APPS, 1);
2259 }
2260 unlink(apps_unix_sock_path);
2261
2262 error_poll_add:
2263 lttng_poll_clean(&events);
2264 error_listen:
2265 error_create_poll:
2266 error_testpoint:
2267 DBG("UST Registration thread cleanup complete");
2268 if (err) {
2269 health_error();
2270 ERR("Health error occurred in %s", __func__);
2271 }
2272 health_unregister(health_sessiond);
2273
2274 return NULL;
2275 }
2276
2277 /*
2278 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2279 * exec or it will fails.
2280 */
2281 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2282 {
2283 int ret, clock_ret;
2284 struct timespec timeout;
2285
2286 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2287 consumer_data->consumer_thread_is_ready = 0;
2288
2289 /* Setup pthread condition */
2290 ret = pthread_condattr_init(&consumer_data->condattr);
2291 if (ret) {
2292 errno = ret;
2293 PERROR("pthread_condattr_init consumer data");
2294 goto error;
2295 }
2296
2297 /*
2298 * Set the monotonic clock in order to make sure we DO NOT jump in time
2299 * between the clock_gettime() call and the timedwait call. See bug #324
2300 * for a more details and how we noticed it.
2301 */
2302 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2303 if (ret) {
2304 errno = ret;
2305 PERROR("pthread_condattr_setclock consumer data");
2306 goto error;
2307 }
2308
2309 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2310 if (ret) {
2311 errno = ret;
2312 PERROR("pthread_cond_init consumer data");
2313 goto error;
2314 }
2315
2316 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2317 consumer_data);
2318 if (ret) {
2319 errno = ret;
2320 PERROR("pthread_create consumer");
2321 ret = -1;
2322 goto error;
2323 }
2324
2325 /* We are about to wait on a pthread condition */
2326 pthread_mutex_lock(&consumer_data->cond_mutex);
2327
2328 /* Get time for sem_timedwait absolute timeout */
2329 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2330 /*
2331 * Set the timeout for the condition timed wait even if the clock gettime
2332 * call fails since we might loop on that call and we want to avoid to
2333 * increment the timeout too many times.
2334 */
2335 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2336
2337 /*
2338 * The following loop COULD be skipped in some conditions so this is why we
2339 * set ret to 0 in order to make sure at least one round of the loop is
2340 * done.
2341 */
2342 ret = 0;
2343
2344 /*
2345 * Loop until the condition is reached or when a timeout is reached. Note
2346 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2347 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2348 * possible. This loop does not take any chances and works with both of
2349 * them.
2350 */
2351 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2352 if (clock_ret < 0) {
2353 PERROR("clock_gettime spawn consumer");
2354 /* Infinite wait for the consumerd thread to be ready */
2355 ret = pthread_cond_wait(&consumer_data->cond,
2356 &consumer_data->cond_mutex);
2357 } else {
2358 ret = pthread_cond_timedwait(&consumer_data->cond,
2359 &consumer_data->cond_mutex, &timeout);
2360 }
2361 }
2362
2363 /* Release the pthread condition */
2364 pthread_mutex_unlock(&consumer_data->cond_mutex);
2365
2366 if (ret != 0) {
2367 errno = ret;
2368 if (ret == ETIMEDOUT) {
2369 int pth_ret;
2370
2371 /*
2372 * Call has timed out so we kill the kconsumerd_thread and return
2373 * an error.
2374 */
2375 ERR("Condition timed out. The consumer thread was never ready."
2376 " Killing it");
2377 pth_ret = pthread_cancel(consumer_data->thread);
2378 if (pth_ret < 0) {
2379 PERROR("pthread_cancel consumer thread");
2380 }
2381 } else {
2382 PERROR("pthread_cond_wait failed consumer thread");
2383 }
2384 /* Caller is expecting a negative value on failure. */
2385 ret = -1;
2386 goto error;
2387 }
2388
2389 pthread_mutex_lock(&consumer_data->pid_mutex);
2390 if (consumer_data->pid == 0) {
2391 ERR("Consumerd did not start");
2392 pthread_mutex_unlock(&consumer_data->pid_mutex);
2393 goto error;
2394 }
2395 pthread_mutex_unlock(&consumer_data->pid_mutex);
2396
2397 return 0;
2398
2399 error:
2400 return ret;
2401 }
2402
2403 /*
2404 * Join consumer thread
2405 */
2406 static int join_consumer_thread(struct consumer_data *consumer_data)
2407 {
2408 void *status;
2409
2410 /* Consumer pid must be a real one. */
2411 if (consumer_data->pid > 0) {
2412 int ret;
2413 ret = kill(consumer_data->pid, SIGTERM);
2414 if (ret) {
2415 PERROR("Error killing consumer daemon");
2416 return ret;
2417 }
2418 return pthread_join(consumer_data->thread, &status);
2419 } else {
2420 return 0;
2421 }
2422 }
2423
2424 /*
2425 * Fork and exec a consumer daemon (consumerd).
2426 *
2427 * Return pid if successful else -1.
2428 */
2429 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2430 {
2431 int ret;
2432 pid_t pid;
2433 const char *consumer_to_use;
2434 const char *verbosity;
2435 struct stat st;
2436
2437 DBG("Spawning consumerd");
2438
2439 pid = fork();
2440 if (pid == 0) {
2441 /*
2442 * Exec consumerd.
2443 */
2444 if (opt_verbose_consumer) {
2445 verbosity = "--verbose";
2446 } else if (lttng_opt_quiet) {
2447 verbosity = "--quiet";
2448 } else {
2449 verbosity = "";
2450 }
2451
2452 switch (consumer_data->type) {
2453 case LTTNG_CONSUMER_KERNEL:
2454 /*
2455 * Find out which consumerd to execute. We will first try the
2456 * 64-bit path, then the sessiond's installation directory, and
2457 * fallback on the 32-bit one,
2458 */
2459 DBG3("Looking for a kernel consumer at these locations:");
2460 DBG3(" 1) %s", consumerd64_bin);
2461 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2462 DBG3(" 3) %s", consumerd32_bin);
2463 if (stat(consumerd64_bin, &st) == 0) {
2464 DBG3("Found location #1");
2465 consumer_to_use = consumerd64_bin;
2466 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2467 DBG3("Found location #2");
2468 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2469 } else if (stat(consumerd32_bin, &st) == 0) {
2470 DBG3("Found location #3");
2471 consumer_to_use = consumerd32_bin;
2472 } else {
2473 DBG("Could not find any valid consumerd executable");
2474 ret = -EINVAL;
2475 break;
2476 }
2477 DBG("Using kernel consumer at: %s", consumer_to_use);
2478 ret = execl(consumer_to_use,
2479 "lttng-consumerd", verbosity, "-k",
2480 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2481 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2482 "--group", tracing_group_name,
2483 NULL);
2484 break;
2485 case LTTNG_CONSUMER64_UST:
2486 {
2487 char *tmpnew = NULL;
2488
2489 if (consumerd64_libdir[0] != '\0') {
2490 char *tmp;
2491 size_t tmplen;
2492
2493 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2494 if (!tmp) {
2495 tmp = "";
2496 }
2497 tmplen = strlen("LD_LIBRARY_PATH=")
2498 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2499 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2500 if (!tmpnew) {
2501 ret = -ENOMEM;
2502 goto error;
2503 }
2504 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2505 strcat(tmpnew, consumerd64_libdir);
2506 if (tmp[0] != '\0') {
2507 strcat(tmpnew, ":");
2508 strcat(tmpnew, tmp);
2509 }
2510 ret = putenv(tmpnew);
2511 if (ret) {
2512 ret = -errno;
2513 free(tmpnew);
2514 goto error;
2515 }
2516 }
2517 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2518 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2519 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2520 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2521 "--group", tracing_group_name,
2522 NULL);
2523 if (consumerd64_libdir[0] != '\0') {
2524 free(tmpnew);
2525 }
2526 break;
2527 }
2528 case LTTNG_CONSUMER32_UST:
2529 {
2530 char *tmpnew = NULL;
2531
2532 if (consumerd32_libdir[0] != '\0') {
2533 char *tmp;
2534 size_t tmplen;
2535
2536 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2537 if (!tmp) {
2538 tmp = "";
2539 }
2540 tmplen = strlen("LD_LIBRARY_PATH=")
2541 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2542 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2543 if (!tmpnew) {
2544 ret = -ENOMEM;
2545 goto error;
2546 }
2547 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2548 strcat(tmpnew, consumerd32_libdir);
2549 if (tmp[0] != '\0') {
2550 strcat(tmpnew, ":");
2551 strcat(tmpnew, tmp);
2552 }
2553 ret = putenv(tmpnew);
2554 if (ret) {
2555 ret = -errno;
2556 free(tmpnew);
2557 goto error;
2558 }
2559 }
2560 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2561 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2562 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2563 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2564 "--group", tracing_group_name,
2565 NULL);
2566 if (consumerd32_libdir[0] != '\0') {
2567 free(tmpnew);
2568 }
2569 break;
2570 }
2571 default:
2572 PERROR("unknown consumer type");
2573 exit(EXIT_FAILURE);
2574 }
2575 if (errno != 0) {
2576 PERROR("Consumer execl()");
2577 }
2578 /* Reaching this point, we got a failure on our execl(). */
2579 exit(EXIT_FAILURE);
2580 } else if (pid > 0) {
2581 ret = pid;
2582 } else {
2583 PERROR("start consumer fork");
2584 ret = -errno;
2585 }
2586 error:
2587 return ret;
2588 }
2589
2590 /*
2591 * Spawn the consumerd daemon and session daemon thread.
2592 */
2593 static int start_consumerd(struct consumer_data *consumer_data)
2594 {
2595 int ret;
2596
2597 /*
2598 * Set the listen() state on the socket since there is a possible race
2599 * between the exec() of the consumer daemon and this call if place in the
2600 * consumer thread. See bug #366 for more details.
2601 */
2602 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2603 if (ret < 0) {
2604 goto error;
2605 }
2606
2607 pthread_mutex_lock(&consumer_data->pid_mutex);
2608 if (consumer_data->pid != 0) {
2609 pthread_mutex_unlock(&consumer_data->pid_mutex);
2610 goto end;
2611 }
2612
2613 ret = spawn_consumerd(consumer_data);
2614 if (ret < 0) {
2615 ERR("Spawning consumerd failed");
2616 pthread_mutex_unlock(&consumer_data->pid_mutex);
2617 goto error;
2618 }
2619
2620 /* Setting up the consumer_data pid */
2621 consumer_data->pid = ret;
2622 DBG2("Consumer pid %d", consumer_data->pid);
2623 pthread_mutex_unlock(&consumer_data->pid_mutex);
2624
2625 DBG2("Spawning consumer control thread");
2626 ret = spawn_consumer_thread(consumer_data);
2627 if (ret < 0) {
2628 ERR("Fatal error spawning consumer control thread");
2629 goto error;
2630 }
2631
2632 end:
2633 return 0;
2634
2635 error:
2636 /* Cleanup already created sockets on error. */
2637 if (consumer_data->err_sock >= 0) {
2638 int err;
2639
2640 err = close(consumer_data->err_sock);
2641 if (err < 0) {
2642 PERROR("close consumer data error socket");
2643 }
2644 }
2645 return ret;
2646 }
2647
2648 /*
2649 * Setup necessary data for kernel tracer action.
2650 */
2651 static int init_kernel_tracer(void)
2652 {
2653 int ret;
2654
2655 /* Modprobe lttng kernel modules */
2656 ret = modprobe_lttng_control();
2657 if (ret < 0) {
2658 goto error;
2659 }
2660
2661 /* Open debugfs lttng */
2662 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2663 if (kernel_tracer_fd < 0) {
2664 DBG("Failed to open %s", module_proc_lttng);
2665 ret = -1;
2666 goto error_open;
2667 }
2668
2669 /* Validate kernel version */
2670 ret = kernel_validate_version(kernel_tracer_fd);
2671 if (ret < 0) {
2672 goto error_version;
2673 }
2674
2675 ret = modprobe_lttng_data();
2676 if (ret < 0) {
2677 goto error_modules;
2678 }
2679
2680 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2681 return 0;
2682
2683 error_version:
2684 modprobe_remove_lttng_control();
2685 ret = close(kernel_tracer_fd);
2686 if (ret) {
2687 PERROR("close");
2688 }
2689 kernel_tracer_fd = -1;
2690 return LTTNG_ERR_KERN_VERSION;
2691
2692 error_modules:
2693 ret = close(kernel_tracer_fd);
2694 if (ret) {
2695 PERROR("close");
2696 }
2697
2698 error_open:
2699 modprobe_remove_lttng_control();
2700
2701 error:
2702 WARN("No kernel tracer available");
2703 kernel_tracer_fd = -1;
2704 if (!is_root) {
2705 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2706 } else {
2707 return LTTNG_ERR_KERN_NA;
2708 }
2709 }
2710
2711
2712 /*
2713 * Copy consumer output from the tracing session to the domain session. The
2714 * function also applies the right modification on a per domain basis for the
2715 * trace files destination directory.
2716 *
2717 * Should *NOT* be called with RCU read-side lock held.
2718 */
2719 static int copy_session_consumer(int domain, struct ltt_session *session)
2720 {
2721 int ret;
2722 const char *dir_name;
2723 struct consumer_output *consumer;
2724
2725 assert(session);
2726 assert(session->consumer);
2727
2728 switch (domain) {
2729 case LTTNG_DOMAIN_KERNEL:
2730 DBG3("Copying tracing session consumer output in kernel session");
2731 /*
2732 * XXX: We should audit the session creation and what this function
2733 * does "extra" in order to avoid a destroy since this function is used
2734 * in the domain session creation (kernel and ust) only. Same for UST
2735 * domain.
2736 */
2737 if (session->kernel_session->consumer) {
2738 consumer_output_put(session->kernel_session->consumer);
2739 }
2740 session->kernel_session->consumer =
2741 consumer_copy_output(session->consumer);
2742 /* Ease our life a bit for the next part */
2743 consumer = session->kernel_session->consumer;
2744 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2745 break;
2746 case LTTNG_DOMAIN_JUL:
2747 case LTTNG_DOMAIN_LOG4J:
2748 case LTTNG_DOMAIN_PYTHON:
2749 case LTTNG_DOMAIN_UST:
2750 DBG3("Copying tracing session consumer output in UST session");
2751 if (session->ust_session->consumer) {
2752 consumer_output_put(session->ust_session->consumer);
2753 }
2754 session->ust_session->consumer =
2755 consumer_copy_output(session->consumer);
2756 /* Ease our life a bit for the next part */
2757 consumer = session->ust_session->consumer;
2758 dir_name = DEFAULT_UST_TRACE_DIR;
2759 break;
2760 default:
2761 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2762 goto error;
2763 }
2764
2765 /* Append correct directory to subdir */
2766 strncat(consumer->subdir, dir_name,
2767 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2768 DBG3("Copy session consumer subdir %s", consumer->subdir);
2769
2770 ret = LTTNG_OK;
2771
2772 error:
2773 return ret;
2774 }
2775
2776 /*
2777 * Create an UST session and add it to the session ust list.
2778 *
2779 * Should *NOT* be called with RCU read-side lock held.
2780 */
2781 static int create_ust_session(struct ltt_session *session,
2782 struct lttng_domain *domain)
2783 {
2784 int ret;
2785 struct ltt_ust_session *lus = NULL;
2786
2787 assert(session);
2788 assert(domain);
2789 assert(session->consumer);
2790
2791 switch (domain->type) {
2792 case LTTNG_DOMAIN_JUL:
2793 case LTTNG_DOMAIN_LOG4J:
2794 case LTTNG_DOMAIN_PYTHON:
2795 case LTTNG_DOMAIN_UST:
2796 break;
2797 default:
2798 ERR("Unknown UST domain on create session %d", domain->type);
2799 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2800 goto error;
2801 }
2802
2803 DBG("Creating UST session");
2804
2805 lus = trace_ust_create_session(session->id);
2806 if (lus == NULL) {
2807 ret = LTTNG_ERR_UST_SESS_FAIL;
2808 goto error;
2809 }
2810
2811 lus->uid = session->uid;
2812 lus->gid = session->gid;
2813 lus->output_traces = session->output_traces;
2814 lus->snapshot_mode = session->snapshot_mode;
2815 lus->live_timer_interval = session->live_timer;
2816 session->ust_session = lus;
2817 if (session->shm_path[0]) {
2818 strncpy(lus->root_shm_path, session->shm_path,
2819 sizeof(lus->root_shm_path));
2820 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2821 strncpy(lus->shm_path, session->shm_path,
2822 sizeof(lus->shm_path));
2823 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2824 strncat(lus->shm_path, "/ust",
2825 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2826 }
2827 /* Copy session output to the newly created UST session */
2828 ret = copy_session_consumer(domain->type, session);
2829 if (ret != LTTNG_OK) {
2830 goto error;
2831 }
2832
2833 return LTTNG_OK;
2834
2835 error:
2836 free(lus);
2837 session->ust_session = NULL;
2838 return ret;
2839 }
2840
2841 /*
2842 * Create a kernel tracer session then create the default channel.
2843 */
2844 static int create_kernel_session(struct ltt_session *session)
2845 {
2846 int ret;
2847
2848 DBG("Creating kernel session");
2849
2850 ret = kernel_create_session(session, kernel_tracer_fd);
2851 if (ret < 0) {
2852 ret = LTTNG_ERR_KERN_SESS_FAIL;
2853 goto error;
2854 }
2855
2856 /* Code flow safety */
2857 assert(session->kernel_session);
2858
2859 /* Copy session output to the newly created Kernel session */
2860 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2861 if (ret != LTTNG_OK) {
2862 goto error;
2863 }
2864
2865 /* Create directory(ies) on local filesystem. */
2866 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2867 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2868 ret = run_as_mkdir_recursive(
2869 session->kernel_session->consumer->dst.trace_path,
2870 S_IRWXU | S_IRWXG, session->uid, session->gid);
2871 if (ret < 0) {
2872 if (errno != EEXIST) {
2873 ERR("Trace directory creation error");
2874 goto error;
2875 }
2876 }
2877 }
2878
2879 session->kernel_session->uid = session->uid;
2880 session->kernel_session->gid = session->gid;
2881 session->kernel_session->output_traces = session->output_traces;
2882 session->kernel_session->snapshot_mode = session->snapshot_mode;
2883
2884 return LTTNG_OK;
2885
2886 error:
2887 trace_kernel_destroy_session(session->kernel_session);
2888 session->kernel_session = NULL;
2889 return ret;
2890 }
2891
2892 /*
2893 * Count number of session permitted by uid/gid.
2894 */
2895 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2896 {
2897 unsigned int i = 0;
2898 struct ltt_session *session;
2899
2900 DBG("Counting number of available session for UID %d GID %d",
2901 uid, gid);
2902 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2903 /*
2904 * Only list the sessions the user can control.
2905 */
2906 if (!session_access_ok(session, uid, gid)) {
2907 continue;
2908 }
2909 i++;
2910 }
2911 return i;
2912 }
2913
2914 /*
2915 * Process the command requested by the lttng client within the command
2916 * context structure. This function make sure that the return structure (llm)
2917 * is set and ready for transmission before returning.
2918 *
2919 * Return any error encountered or 0 for success.
2920 *
2921 * "sock" is only used for special-case var. len data.
2922 *
2923 * Should *NOT* be called with RCU read-side lock held.
2924 */
2925 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2926 int *sock_error)
2927 {
2928 int ret = LTTNG_OK;
2929 int need_tracing_session = 1;
2930 int need_domain;
2931
2932 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2933
2934 *sock_error = 0;
2935
2936 switch (cmd_ctx->lsm->cmd_type) {
2937 case LTTNG_CREATE_SESSION:
2938 case LTTNG_CREATE_SESSION_SNAPSHOT:
2939 case LTTNG_CREATE_SESSION_LIVE:
2940 case LTTNG_DESTROY_SESSION:
2941 case LTTNG_LIST_SESSIONS:
2942 case LTTNG_LIST_DOMAINS:
2943 case LTTNG_START_TRACE:
2944 case LTTNG_STOP_TRACE:
2945 case LTTNG_DATA_PENDING:
2946 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2947 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2948 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2949 case LTTNG_SNAPSHOT_RECORD:
2950 case LTTNG_SAVE_SESSION:
2951 case LTTNG_SET_SESSION_SHM_PATH:
2952 need_domain = 0;
2953 break;
2954 default:
2955 need_domain = 1;
2956 }
2957
2958 if (opt_no_kernel && need_domain
2959 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2960 if (!is_root) {
2961 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2962 } else {
2963 ret = LTTNG_ERR_KERN_NA;
2964 }
2965 goto error;
2966 }
2967
2968 /* Deny register consumer if we already have a spawned consumer. */
2969 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2970 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2971 if (kconsumer_data.pid > 0) {
2972 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2973 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2974 goto error;
2975 }
2976 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2977 }
2978
2979 /*
2980 * Check for command that don't needs to allocate a returned payload. We do
2981 * this here so we don't have to make the call for no payload at each
2982 * command.
2983 */
2984 switch(cmd_ctx->lsm->cmd_type) {
2985 case LTTNG_LIST_SESSIONS:
2986 case LTTNG_LIST_TRACEPOINTS:
2987 case LTTNG_LIST_TRACEPOINT_FIELDS:
2988 case LTTNG_LIST_DOMAINS:
2989 case LTTNG_LIST_CHANNELS:
2990 case LTTNG_LIST_EVENTS:
2991 case LTTNG_LIST_SYSCALLS:
2992 case LTTNG_LIST_TRACKER_PIDS:
2993 break;
2994 default:
2995 /* Setup lttng message with no payload */
2996 ret = setup_lttng_msg(cmd_ctx, 0);
2997 if (ret < 0) {
2998 /* This label does not try to unlock the session */
2999 goto init_setup_error;
3000 }
3001 }
3002
3003 /* Commands that DO NOT need a session. */
3004 switch (cmd_ctx->lsm->cmd_type) {
3005 case LTTNG_CREATE_SESSION:
3006 case LTTNG_CREATE_SESSION_SNAPSHOT:
3007 case LTTNG_CREATE_SESSION_LIVE:
3008 case LTTNG_CALIBRATE:
3009 case LTTNG_LIST_SESSIONS:
3010 case LTTNG_LIST_TRACEPOINTS:
3011 case LTTNG_LIST_SYSCALLS:
3012 case LTTNG_LIST_TRACEPOINT_FIELDS:
3013 case LTTNG_SAVE_SESSION:
3014 need_tracing_session = 0;
3015 break;
3016 default:
3017 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3018 /*
3019 * We keep the session list lock across _all_ commands
3020 * for now, because the per-session lock does not
3021 * handle teardown properly.
3022 */
3023 session_lock_list();
3024 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3025 if (cmd_ctx->session == NULL) {
3026 ret = LTTNG_ERR_SESS_NOT_FOUND;
3027 goto error;
3028 } else {
3029 /* Acquire lock for the session */
3030 session_lock(cmd_ctx->session);
3031 }
3032 break;
3033 }
3034
3035 /*
3036 * Commands that need a valid session but should NOT create one if none
3037 * exists. Instead of creating one and destroying it when the command is
3038 * handled, process that right before so we save some round trip in useless
3039 * code path.
3040 */
3041 switch (cmd_ctx->lsm->cmd_type) {
3042 case LTTNG_DISABLE_CHANNEL:
3043 case LTTNG_DISABLE_EVENT:
3044 switch (cmd_ctx->lsm->domain.type) {
3045 case LTTNG_DOMAIN_KERNEL:
3046 if (!cmd_ctx->session->kernel_session) {
3047 ret = LTTNG_ERR_NO_CHANNEL;
3048 goto error;
3049 }
3050 break;
3051 case LTTNG_DOMAIN_JUL:
3052 case LTTNG_DOMAIN_LOG4J:
3053 case LTTNG_DOMAIN_PYTHON:
3054 case LTTNG_DOMAIN_UST:
3055 if (!cmd_ctx->session->ust_session) {
3056 ret = LTTNG_ERR_NO_CHANNEL;
3057 goto error;
3058 }
3059 break;
3060 default:
3061 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3062 goto error;
3063 }
3064 default:
3065 break;
3066 }
3067
3068 if (!need_domain) {
3069 goto skip_domain;
3070 }
3071
3072 /*
3073 * Check domain type for specific "pre-action".
3074 */
3075 switch (cmd_ctx->lsm->domain.type) {
3076 case LTTNG_DOMAIN_KERNEL:
3077 if (!is_root) {
3078 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3079 goto error;
3080 }
3081
3082 /* Kernel tracer check */
3083 if (kernel_tracer_fd == -1) {
3084 /* Basically, load kernel tracer modules */
3085 ret = init_kernel_tracer();
3086 if (ret != 0) {
3087 goto error;
3088 }
3089 }
3090
3091 /* Consumer is in an ERROR state. Report back to client */
3092 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3093 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3094 goto error;
3095 }
3096
3097 /* Need a session for kernel command */
3098 if (need_tracing_session) {
3099 if (cmd_ctx->session->kernel_session == NULL) {
3100 ret = create_kernel_session(cmd_ctx->session);
3101 if (ret < 0) {
3102 ret = LTTNG_ERR_KERN_SESS_FAIL;
3103 goto error;
3104 }
3105 }
3106
3107 /* Start the kernel consumer daemon */
3108 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3109 if (kconsumer_data.pid == 0 &&
3110 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3111 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3112 ret = start_consumerd(&kconsumer_data);
3113 if (ret < 0) {
3114 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3115 goto error;
3116 }
3117 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3118 } else {
3119 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3120 }
3121
3122 /*
3123 * The consumer was just spawned so we need to add the socket to
3124 * the consumer output of the session if exist.
3125 */
3126 ret = consumer_create_socket(&kconsumer_data,
3127 cmd_ctx->session->kernel_session->consumer);
3128 if (ret < 0) {
3129 goto error;
3130 }
3131 }
3132
3133 break;
3134 case LTTNG_DOMAIN_JUL:
3135 case LTTNG_DOMAIN_LOG4J:
3136 case LTTNG_DOMAIN_PYTHON:
3137 case LTTNG_DOMAIN_UST:
3138 {
3139 if (!ust_app_supported()) {
3140 ret = LTTNG_ERR_NO_UST;
3141 goto error;
3142 }
3143 /* Consumer is in an ERROR state. Report back to client */
3144 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3145 ret = LTTNG_ERR_NO_USTCONSUMERD;
3146 goto error;
3147 }
3148
3149 if (need_tracing_session) {
3150 /* Create UST session if none exist. */
3151 if (cmd_ctx->session->ust_session == NULL) {
3152 ret = create_ust_session(cmd_ctx->session,
3153 &cmd_ctx->lsm->domain);
3154 if (ret != LTTNG_OK) {
3155 goto error;
3156 }
3157 }
3158
3159 /* Start the UST consumer daemons */
3160 /* 64-bit */
3161 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3162 if (consumerd64_bin[0] != '\0' &&
3163 ustconsumer64_data.pid == 0 &&
3164 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3165 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3166 ret = start_consumerd(&ustconsumer64_data);
3167 if (ret < 0) {
3168 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3169 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3170 goto error;
3171 }
3172
3173 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3174 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3175 } else {
3176 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3177 }
3178
3179 /*
3180 * Setup socket for consumer 64 bit. No need for atomic access
3181 * since it was set above and can ONLY be set in this thread.
3182 */
3183 ret = consumer_create_socket(&ustconsumer64_data,
3184 cmd_ctx->session->ust_session->consumer);
3185 if (ret < 0) {
3186 goto error;
3187 }
3188
3189 /* 32-bit */
3190 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3191 if (consumerd32_bin[0] != '\0' &&
3192 ustconsumer32_data.pid == 0 &&
3193 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3194 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3195 ret = start_consumerd(&ustconsumer32_data);
3196 if (ret < 0) {
3197 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3198 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3199 goto error;
3200 }
3201
3202 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3203 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3204 } else {
3205 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3206 }
3207
3208 /*
3209 * Setup socket for consumer 64 bit. No need for atomic access
3210 * since it was set above and can ONLY be set in this thread.
3211 */
3212 ret = consumer_create_socket(&ustconsumer32_data,
3213 cmd_ctx->session->ust_session->consumer);
3214 if (ret < 0) {
3215 goto error;
3216 }
3217 }
3218 break;
3219 }
3220 default:
3221 break;
3222 }
3223 skip_domain:
3224
3225 /* Validate consumer daemon state when start/stop trace command */
3226 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3227 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3228 switch (cmd_ctx->lsm->domain.type) {
3229 case LTTNG_DOMAIN_JUL:
3230 case LTTNG_DOMAIN_LOG4J:
3231 case LTTNG_DOMAIN_PYTHON:
3232 case LTTNG_DOMAIN_UST:
3233 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3234 ret = LTTNG_ERR_NO_USTCONSUMERD;
3235 goto error;
3236 }
3237 break;
3238 case LTTNG_DOMAIN_KERNEL:
3239 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3240 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3241 goto error;
3242 }
3243 break;
3244 }
3245 }
3246
3247 /*
3248 * Check that the UID or GID match that of the tracing session.
3249 * The root user can interact with all sessions.
3250 */
3251 if (need_tracing_session) {
3252 if (!session_access_ok(cmd_ctx->session,
3253 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3254 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3255 ret = LTTNG_ERR_EPERM;
3256 goto error;
3257 }
3258 }
3259
3260 /*
3261 * Send relayd information to consumer as soon as we have a domain and a
3262 * session defined.
3263 */
3264 if (cmd_ctx->session && need_domain) {
3265 /*
3266 * Setup relayd if not done yet. If the relayd information was already
3267 * sent to the consumer, this call will gracefully return.
3268 */
3269 ret = cmd_setup_relayd(cmd_ctx->session);
3270 if (ret != LTTNG_OK) {
3271 goto error;
3272 }
3273 }
3274
3275 /* Process by command type */
3276 switch (cmd_ctx->lsm->cmd_type) {
3277 case LTTNG_ADD_CONTEXT:
3278 {
3279 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3280 cmd_ctx->lsm->u.context.channel_name,
3281 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3282 break;
3283 }
3284 case LTTNG_DISABLE_CHANNEL:
3285 {
3286 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3287 cmd_ctx->lsm->u.disable.channel_name);
3288 break;
3289 }
3290 case LTTNG_DISABLE_EVENT:
3291 {
3292
3293 /*
3294 * FIXME: handle filter; for now we just receive the filter's
3295 * bytecode along with the filter expression which are sent by
3296 * liblttng-ctl and discard them.
3297 *
3298 * This fixes an issue where the client may block while sending
3299 * the filter payload and encounter an error because the session
3300 * daemon closes the socket without ever handling this data.
3301 */
3302 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3303 cmd_ctx->lsm->u.disable.bytecode_len;
3304
3305 if (count) {
3306 char data[LTTNG_FILTER_MAX_LEN];
3307
3308 DBG("Discarding disable event command payload of size %zu", count);
3309 while (count) {
3310 ret = lttcomm_recv_unix_sock(sock, data,
3311 count > sizeof(data) ? sizeof(data) : count);
3312 if (ret < 0) {
3313 goto error;
3314 }
3315
3316 count -= (size_t) ret;
3317 }
3318 }
3319 /* FIXME: passing packed structure to non-packed pointer */
3320 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3321 cmd_ctx->lsm->u.disable.channel_name,
3322 &cmd_ctx->lsm->u.disable.event);
3323 break;
3324 }
3325 case LTTNG_ENABLE_CHANNEL:
3326 {
3327 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3328 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3329 break;
3330 }
3331 case LTTNG_TRACK_PID:
3332 {
3333 ret = cmd_track_pid(cmd_ctx->session,
3334 cmd_ctx->lsm->domain.type,
3335 cmd_ctx->lsm->u.pid_tracker.pid);
3336 break;
3337 }
3338 case LTTNG_UNTRACK_PID:
3339 {
3340 ret = cmd_untrack_pid(cmd_ctx->session,
3341 cmd_ctx->lsm->domain.type,
3342 cmd_ctx->lsm->u.pid_tracker.pid);
3343 break;
3344 }
3345 case LTTNG_ENABLE_EVENT:
3346 {
3347 struct lttng_event_exclusion *exclusion = NULL;
3348 struct lttng_filter_bytecode *bytecode = NULL;
3349 char *filter_expression = NULL;
3350
3351 /* Handle exclusion events and receive it from the client. */
3352 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3353 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3354
3355 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3356 (count * LTTNG_SYMBOL_NAME_LEN));
3357 if (!exclusion) {
3358 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3359 goto error;
3360 }
3361
3362 DBG("Receiving var len exclusion event list from client ...");
3363 exclusion->count = count;
3364 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3365 count * LTTNG_SYMBOL_NAME_LEN);
3366 if (ret <= 0) {
3367 DBG("Nothing recv() from client var len data... continuing");
3368 *sock_error = 1;
3369 free(exclusion);
3370 ret = LTTNG_ERR_EXCLUSION_INVAL;
3371 goto error;
3372 }
3373 }
3374
3375 /* Get filter expression from client. */
3376 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3377 size_t expression_len =
3378 cmd_ctx->lsm->u.enable.expression_len;
3379
3380 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3381 ret = LTTNG_ERR_FILTER_INVAL;
3382 free(exclusion);
3383 goto error;
3384 }
3385
3386 filter_expression = zmalloc(expression_len);
3387 if (!filter_expression) {
3388 free(exclusion);
3389 ret = LTTNG_ERR_FILTER_NOMEM;
3390 goto error;
3391 }
3392
3393 /* Receive var. len. data */
3394 DBG("Receiving var len filter's expression from client ...");
3395 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3396 expression_len);
3397 if (ret <= 0) {
3398 DBG("Nothing recv() from client car len data... continuing");
3399 *sock_error = 1;
3400 free(filter_expression);
3401 free(exclusion);
3402 ret = LTTNG_ERR_FILTER_INVAL;
3403 goto error;
3404 }
3405 }
3406
3407 /* Handle filter and get bytecode from client. */
3408 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3409 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3410
3411 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3412 ret = LTTNG_ERR_FILTER_INVAL;
3413 free(filter_expression);
3414 free(exclusion);
3415 goto error;
3416 }
3417
3418 bytecode = zmalloc(bytecode_len);
3419 if (!bytecode) {
3420 free(filter_expression);
3421 free(exclusion);
3422 ret = LTTNG_ERR_FILTER_NOMEM;
3423 goto error;
3424 }
3425
3426 /* Receive var. len. data */
3427 DBG("Receiving var len filter's bytecode from client ...");
3428 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3429 if (ret <= 0) {
3430 DBG("Nothing recv() from client car len data... continuing");
3431 *sock_error = 1;
3432 free(filter_expression);
3433 free(bytecode);
3434 free(exclusion);
3435 ret = LTTNG_ERR_FILTER_INVAL;
3436 goto error;
3437 }
3438
3439 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3440 free(filter_expression);
3441 free(bytecode);
3442 free(exclusion);
3443 ret = LTTNG_ERR_FILTER_INVAL;
3444 goto error;
3445 }
3446 }
3447
3448 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3449 cmd_ctx->lsm->u.enable.channel_name,
3450 &cmd_ctx->lsm->u.enable.event,
3451 filter_expression, bytecode, exclusion,
3452 kernel_poll_pipe[1]);
3453 break;
3454 }
3455 case LTTNG_LIST_TRACEPOINTS:
3456 {
3457 struct lttng_event *events;
3458 ssize_t nb_events;
3459
3460 session_lock_list();
3461 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3462 session_unlock_list();
3463 if (nb_events < 0) {
3464 /* Return value is a negative lttng_error_code. */
3465 ret = -nb_events;
3466 goto error;
3467 }
3468
3469 /*
3470 * Setup lttng message with payload size set to the event list size in
3471 * bytes and then copy list into the llm payload.
3472 */
3473 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3474 if (ret < 0) {
3475 free(events);
3476 goto setup_error;
3477 }
3478
3479 /* Copy event list into message payload */
3480 memcpy(cmd_ctx->llm->payload, events,
3481 sizeof(struct lttng_event) * nb_events);
3482
3483 free(events);
3484
3485 ret = LTTNG_OK;
3486 break;
3487 }
3488 case LTTNG_LIST_TRACEPOINT_FIELDS:
3489 {
3490 struct lttng_event_field *fields;
3491 ssize_t nb_fields;
3492
3493 session_lock_list();
3494 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3495 &fields);
3496 session_unlock_list();
3497 if (nb_fields < 0) {
3498 /* Return value is a negative lttng_error_code. */
3499 ret = -nb_fields;
3500 goto error;
3501 }
3502
3503 /*
3504 * Setup lttng message with payload size set to the event list size in
3505 * bytes and then copy list into the llm payload.
3506 */
3507 ret = setup_lttng_msg(cmd_ctx,
3508 sizeof(struct lttng_event_field) * nb_fields);
3509 if (ret < 0) {
3510 free(fields);
3511 goto setup_error;
3512 }
3513
3514 /* Copy event list into message payload */
3515 memcpy(cmd_ctx->llm->payload, fields,
3516 sizeof(struct lttng_event_field) * nb_fields);
3517
3518 free(fields);
3519
3520 ret = LTTNG_OK;
3521 break;
3522 }
3523 case LTTNG_LIST_SYSCALLS:
3524 {
3525 struct lttng_event *events;
3526 ssize_t nb_events;
3527
3528 nb_events = cmd_list_syscalls(&events);
3529 if (nb_events < 0) {
3530 /* Return value is a negative lttng_error_code. */
3531 ret = -nb_events;
3532 goto error;
3533 }
3534
3535 /*
3536 * Setup lttng message with payload size set to the event list size in
3537 * bytes and then copy list into the llm payload.
3538 */
3539 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3540 if (ret < 0) {
3541 free(events);
3542 goto setup_error;
3543 }
3544
3545 /* Copy event list into message payload */
3546 memcpy(cmd_ctx->llm->payload, events,
3547 sizeof(struct lttng_event) * nb_events);
3548
3549 free(events);
3550
3551 ret = LTTNG_OK;
3552 break;
3553 }
3554 case LTTNG_LIST_TRACKER_PIDS:
3555 {
3556 int32_t *pids = NULL;
3557 ssize_t nr_pids;
3558
3559 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3560 cmd_ctx->lsm->domain.type, &pids);
3561 if (nr_pids < 0) {
3562 /* Return value is a negative lttng_error_code. */
3563 ret = -nr_pids;
3564 goto error;
3565 }
3566
3567 /*
3568 * Setup lttng message with payload size set to the event list size in
3569 * bytes and then copy list into the llm payload.
3570 */
3571 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3572 if (ret < 0) {
3573 free(pids);
3574 goto setup_error;
3575 }
3576
3577 /* Copy event list into message payload */
3578 memcpy(cmd_ctx->llm->payload, pids,
3579 sizeof(int) * nr_pids);
3580
3581 free(pids);
3582
3583 ret = LTTNG_OK;
3584 break;
3585 }
3586 case LTTNG_SET_CONSUMER_URI:
3587 {
3588 size_t nb_uri, len;
3589 struct lttng_uri *uris;
3590
3591 nb_uri = cmd_ctx->lsm->u.uri.size;
3592 len = nb_uri * sizeof(struct lttng_uri);
3593
3594 if (nb_uri == 0) {
3595 ret = LTTNG_ERR_INVALID;
3596 goto error;
3597 }
3598
3599 uris = zmalloc(len);
3600 if (uris == NULL) {
3601 ret = LTTNG_ERR_FATAL;
3602 goto error;
3603 }
3604
3605 /* Receive variable len data */
3606 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3607 ret = lttcomm_recv_unix_sock(sock, uris, len);
3608 if (ret <= 0) {
3609 DBG("No URIs received from client... continuing");
3610 *sock_error = 1;
3611 ret = LTTNG_ERR_SESSION_FAIL;
3612 free(uris);
3613 goto error;
3614 }
3615
3616 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3617 free(uris);
3618 if (ret != LTTNG_OK) {
3619 goto error;
3620 }
3621
3622
3623 break;
3624 }
3625 case LTTNG_START_TRACE:
3626 {
3627 ret = cmd_start_trace(cmd_ctx->session);
3628 break;
3629 }
3630 case LTTNG_STOP_TRACE:
3631 {
3632 ret = cmd_stop_trace(cmd_ctx->session);
3633 break;
3634 }
3635 case LTTNG_CREATE_SESSION:
3636 {
3637 size_t nb_uri, len;
3638 struct lttng_uri *uris = NULL;
3639
3640 nb_uri = cmd_ctx->lsm->u.uri.size;
3641 len = nb_uri * sizeof(struct lttng_uri);
3642
3643 if (nb_uri > 0) {
3644 uris = zmalloc(len);
3645 if (uris == NULL) {
3646 ret = LTTNG_ERR_FATAL;
3647 goto error;
3648 }
3649
3650 /* Receive variable len data */
3651 DBG("Waiting for %zu URIs from client ...", nb_uri);
3652 ret = lttcomm_recv_unix_sock(sock, uris, len);
3653 if (ret <= 0) {
3654 DBG("No URIs received from client... continuing");
3655 *sock_error = 1;
3656 ret = LTTNG_ERR_SESSION_FAIL;
3657 free(uris);
3658 goto error;
3659 }
3660
3661 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3662 DBG("Creating session with ONE network URI is a bad call");
3663 ret = LTTNG_ERR_SESSION_FAIL;
3664 free(uris);
3665 goto error;
3666 }
3667 }
3668
3669 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3670 &cmd_ctx->creds, 0);
3671
3672 free(uris);
3673
3674 break;
3675 }
3676 case LTTNG_DESTROY_SESSION:
3677 {
3678 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3679
3680 /* Set session to NULL so we do not unlock it after free. */
3681 cmd_ctx->session = NULL;
3682 break;
3683 }
3684 case LTTNG_LIST_DOMAINS:
3685 {
3686 ssize_t nb_dom;
3687 struct lttng_domain *domains = NULL;
3688
3689 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3690 if (nb_dom < 0) {
3691 /* Return value is a negative lttng_error_code. */
3692 ret = -nb_dom;
3693 goto error;
3694 }
3695
3696 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3697 if (ret < 0) {
3698 free(domains);
3699 goto setup_error;
3700 }
3701
3702 /* Copy event list into message payload */
3703 memcpy(cmd_ctx->llm->payload, domains,
3704 nb_dom * sizeof(struct lttng_domain));
3705
3706 free(domains);
3707
3708 ret = LTTNG_OK;
3709 break;
3710 }
3711 case LTTNG_LIST_CHANNELS:
3712 {
3713 int nb_chan;
3714 struct lttng_channel *channels = NULL;
3715
3716 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3717 cmd_ctx->session, &channels);
3718 if (nb_chan < 0) {
3719 /* Return value is a negative lttng_error_code. */
3720 ret = -nb_chan;
3721 goto error;
3722 }
3723
3724 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3725 if (ret < 0) {
3726 free(channels);
3727 goto setup_error;
3728 }
3729
3730 /* Copy event list into message payload */
3731 memcpy(cmd_ctx->llm->payload, channels,
3732 nb_chan * sizeof(struct lttng_channel));
3733
3734 free(channels);
3735
3736 ret = LTTNG_OK;
3737 break;
3738 }
3739 case LTTNG_LIST_EVENTS:
3740 {
3741 ssize_t nb_event;
3742 struct lttng_event *events = NULL;
3743
3744 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3745 cmd_ctx->lsm->u.list.channel_name, &events);
3746 if (nb_event < 0) {
3747 /* Return value is a negative lttng_error_code. */
3748 ret = -nb_event;
3749 goto error;
3750 }
3751
3752 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3753 if (ret < 0) {
3754 free(events);
3755 goto setup_error;
3756 }
3757
3758 /* Copy event list into message payload */
3759 memcpy(cmd_ctx->llm->payload, events,
3760 nb_event * sizeof(struct lttng_event));
3761
3762 free(events);
3763
3764 ret = LTTNG_OK;
3765 break;
3766 }
3767 case LTTNG_LIST_SESSIONS:
3768 {
3769 unsigned int nr_sessions;
3770
3771 session_lock_list();
3772 nr_sessions = lttng_sessions_count(
3773 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3774 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3775
3776 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3777 if (ret < 0) {
3778 session_unlock_list();
3779 goto setup_error;
3780 }
3781
3782 /* Filled the session array */
3783 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3784 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3785 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3786
3787 session_unlock_list();
3788
3789 ret = LTTNG_OK;
3790 break;
3791 }
3792 case LTTNG_CALIBRATE:
3793 {
3794 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3795 &cmd_ctx->lsm->u.calibrate);
3796 break;
3797 }
3798 case LTTNG_REGISTER_CONSUMER:
3799 {
3800 struct consumer_data *cdata;
3801
3802 switch (cmd_ctx->lsm->domain.type) {
3803 case LTTNG_DOMAIN_KERNEL:
3804 cdata = &kconsumer_data;
3805 break;
3806 default:
3807 ret = LTTNG_ERR_UND;
3808 goto error;
3809 }
3810
3811 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3812 cmd_ctx->lsm->u.reg.path, cdata);
3813 break;
3814 }
3815 case LTTNG_DATA_PENDING:
3816 {
3817 ret = cmd_data_pending(cmd_ctx->session);
3818 break;
3819 }
3820 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3821 {
3822 struct lttcomm_lttng_output_id reply;
3823
3824 ret = cmd_snapshot_add_output(cmd_ctx->session,
3825 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3826 if (ret != LTTNG_OK) {
3827 goto error;
3828 }
3829
3830 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3831 if (ret < 0) {
3832 goto setup_error;
3833 }
3834
3835 /* Copy output list into message payload */
3836 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3837 ret = LTTNG_OK;
3838 break;
3839 }
3840 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3841 {
3842 ret = cmd_snapshot_del_output(cmd_ctx->session,
3843 &cmd_ctx->lsm->u.snapshot_output.output);
3844 break;
3845 }
3846 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3847 {
3848 ssize_t nb_output;
3849 struct lttng_snapshot_output *outputs = NULL;
3850
3851 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3852 if (nb_output < 0) {
3853 ret = -nb_output;
3854 goto error;
3855 }
3856
3857 ret = setup_lttng_msg(cmd_ctx,
3858 nb_output * sizeof(struct lttng_snapshot_output));
3859 if (ret < 0) {
3860 free(outputs);
3861 goto setup_error;
3862 }
3863
3864 if (outputs) {
3865 /* Copy output list into message payload */
3866 memcpy(cmd_ctx->llm->payload, outputs,
3867 nb_output * sizeof(struct lttng_snapshot_output));
3868 free(outputs);
3869 }
3870
3871 ret = LTTNG_OK;
3872 break;
3873 }
3874 case LTTNG_SNAPSHOT_RECORD:
3875 {
3876 ret = cmd_snapshot_record(cmd_ctx->session,
3877 &cmd_ctx->lsm->u.snapshot_record.output,
3878 cmd_ctx->lsm->u.snapshot_record.wait);
3879 break;
3880 }
3881 case LTTNG_CREATE_SESSION_SNAPSHOT:
3882 {
3883 size_t nb_uri, len;
3884 struct lttng_uri *uris = NULL;
3885
3886 nb_uri = cmd_ctx->lsm->u.uri.size;
3887 len = nb_uri * sizeof(struct lttng_uri);
3888
3889 if (nb_uri > 0) {
3890 uris = zmalloc(len);
3891 if (uris == NULL) {
3892 ret = LTTNG_ERR_FATAL;
3893 goto error;
3894 }
3895
3896 /* Receive variable len data */
3897 DBG("Waiting for %zu URIs from client ...", nb_uri);
3898 ret = lttcomm_recv_unix_sock(sock, uris, len);
3899 if (ret <= 0) {
3900 DBG("No URIs received from client... continuing");
3901 *sock_error = 1;
3902 ret = LTTNG_ERR_SESSION_FAIL;
3903 free(uris);
3904 goto error;
3905 }
3906
3907 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3908 DBG("Creating session with ONE network URI is a bad call");
3909 ret = LTTNG_ERR_SESSION_FAIL;
3910 free(uris);
3911 goto error;
3912 }
3913 }
3914
3915 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3916 nb_uri, &cmd_ctx->creds);
3917 free(uris);
3918 break;
3919 }
3920 case LTTNG_CREATE_SESSION_LIVE:
3921 {
3922 size_t nb_uri, len;
3923 struct lttng_uri *uris = NULL;
3924
3925 nb_uri = cmd_ctx->lsm->u.uri.size;
3926 len = nb_uri * sizeof(struct lttng_uri);
3927
3928 if (nb_uri > 0) {
3929 uris = zmalloc(len);
3930 if (uris == NULL) {
3931 ret = LTTNG_ERR_FATAL;
3932 goto error;
3933 }
3934
3935 /* Receive variable len data */
3936 DBG("Waiting for %zu URIs from client ...", nb_uri);
3937 ret = lttcomm_recv_unix_sock(sock, uris, len);
3938 if (ret <= 0) {
3939 DBG("No URIs received from client... continuing");
3940 *sock_error = 1;
3941 ret = LTTNG_ERR_SESSION_FAIL;
3942 free(uris);
3943 goto error;
3944 }
3945
3946 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3947 DBG("Creating session with ONE network URI is a bad call");
3948 ret = LTTNG_ERR_SESSION_FAIL;
3949 free(uris);
3950 goto error;
3951 }
3952 }
3953
3954 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3955 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3956 free(uris);
3957 break;
3958 }
3959 case LTTNG_SAVE_SESSION:
3960 {
3961 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3962 &cmd_ctx->creds);
3963 break;
3964 }
3965 case LTTNG_SET_SESSION_SHM_PATH:
3966 {
3967 ret = cmd_set_session_shm_path(cmd_ctx->session,
3968 cmd_ctx->lsm->u.set_shm_path.shm_path);
3969 break;
3970 }
3971 default:
3972 ret = LTTNG_ERR_UND;
3973 break;
3974 }
3975
3976 error:
3977 if (cmd_ctx->llm == NULL) {
3978 DBG("Missing llm structure. Allocating one.");
3979 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3980 goto setup_error;
3981 }
3982 }
3983 /* Set return code */
3984 cmd_ctx->llm->ret_code = ret;
3985 setup_error:
3986 if (cmd_ctx->session) {
3987 session_unlock(cmd_ctx->session);
3988 }
3989 if (need_tracing_session) {
3990 session_unlock_list();
3991 }
3992 init_setup_error:
3993 return ret;
3994 }
3995
3996 /*
3997 * Thread managing health check socket.
3998 */
3999 static void *thread_manage_health(void *data)
4000 {
4001 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4002 uint32_t revents, nb_fd;
4003 struct lttng_poll_event events;
4004 struct health_comm_msg msg;
4005 struct health_comm_reply reply;
4006
4007 DBG("[thread] Manage health check started");
4008
4009 rcu_register_thread();
4010
4011 /* We might hit an error path before this is created. */
4012 lttng_poll_init(&events);
4013
4014 /* Create unix socket */
4015 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4016 if (sock < 0) {
4017 ERR("Unable to create health check Unix socket");
4018 ret = -1;
4019 goto error;
4020 }
4021
4022 if (is_root) {
4023 /* lttng health client socket path permissions */
4024 ret = chown(health_unix_sock_path, 0,
4025 utils_get_group_id(tracing_group_name));
4026 if (ret < 0) {
4027 ERR("Unable to set group on %s", health_unix_sock_path);
4028 PERROR("chown");
4029 ret = -1;
4030 goto error;
4031 }
4032
4033 ret = chmod(health_unix_sock_path,
4034 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4035 if (ret < 0) {
4036 ERR("Unable to set permissions on %s", health_unix_sock_path);
4037 PERROR("chmod");
4038 ret = -1;
4039 goto error;
4040 }
4041 }
4042
4043 /*
4044 * Set the CLOEXEC flag. Return code is useless because either way, the
4045 * show must go on.
4046 */
4047 (void) utils_set_fd_cloexec(sock);
4048
4049 ret = lttcomm_listen_unix_sock(sock);
4050 if (ret < 0) {
4051 goto error;
4052 }
4053
4054 /*
4055 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4056 * more will be added to this poll set.
4057 */
4058 ret = sessiond_set_thread_pollset(&events, 2);
4059 if (ret < 0) {
4060 goto error;
4061 }
4062
4063 /* Add the application registration socket */
4064 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4065 if (ret < 0) {
4066 goto error;
4067 }
4068
4069 sessiond_notify_ready();
4070
4071 while (1) {
4072 DBG("Health check ready");
4073
4074 /* Inifinite blocking call, waiting for transmission */
4075 restart:
4076 ret = lttng_poll_wait(&events, -1);
4077 if (ret < 0) {
4078 /*
4079 * Restart interrupted system call.
4080 */
4081 if (errno == EINTR) {
4082 goto restart;
4083 }
4084 goto error;
4085 }
4086
4087 nb_fd = ret;
4088
4089 for (i = 0; i < nb_fd; i++) {
4090 /* Fetch once the poll data */
4091 revents = LTTNG_POLL_GETEV(&events, i);
4092 pollfd = LTTNG_POLL_GETFD(&events, i);
4093
4094 if (!revents) {
4095 /* No activity for this FD (poll implementation). */
4096 continue;
4097 }
4098
4099 /* Thread quit pipe has been closed. Killing thread. */
4100 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4101 if (ret) {
4102 err = 0;
4103 goto exit;
4104 }
4105
4106 /* Event on the registration socket */
4107 if (pollfd == sock) {
4108 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4109 ERR("Health socket poll error");
4110 goto error;
4111 }
4112 }
4113 }
4114
4115 new_sock = lttcomm_accept_unix_sock(sock);
4116 if (new_sock < 0) {
4117 goto error;
4118 }
4119
4120 /*
4121 * Set the CLOEXEC flag. Return code is useless because either way, the
4122 * show must go on.
4123 */
4124 (void) utils_set_fd_cloexec(new_sock);
4125
4126 DBG("Receiving data from client for health...");
4127 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4128 if (ret <= 0) {
4129 DBG("Nothing recv() from client... continuing");
4130 ret = close(new_sock);
4131 if (ret) {
4132 PERROR("close");
4133 }
4134 new_sock = -1;
4135 continue;
4136 }
4137
4138 rcu_thread_online();
4139
4140 memset(&reply, 0, sizeof(reply));
4141 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4142 /*
4143 * health_check_state returns 0 if health is
4144 * bad.
4145 */
4146 if (!health_check_state(health_sessiond, i)) {
4147 reply.ret_code |= 1ULL << i;
4148 }
4149 }
4150
4151 DBG2("Health check return value %" PRIx64, reply.ret_code);
4152
4153 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4154 if (ret < 0) {
4155 ERR("Failed to send health data back to client");
4156 }
4157
4158 /* End of transmission */
4159 ret = close(new_sock);
4160 if (ret) {
4161 PERROR("close");
4162 }
4163 new_sock = -1;
4164 }
4165
4166 exit:
4167 error:
4168 if (err) {
4169 ERR("Health error occurred in %s", __func__);
4170 }
4171 DBG("Health check thread dying");
4172 unlink(health_unix_sock_path);
4173 if (sock >= 0) {
4174 ret = close(sock);
4175 if (ret) {
4176 PERROR("close");
4177 }
4178 }
4179
4180 lttng_poll_clean(&events);
4181
4182 rcu_unregister_thread();
4183 return NULL;
4184 }
4185
4186 /*
4187 * This thread manage all clients request using the unix client socket for
4188 * communication.
4189 */
4190 static void *thread_manage_clients(void *data)
4191 {
4192 int sock = -1, ret, i, pollfd, err = -1;
4193 int sock_error;
4194 uint32_t revents, nb_fd;
4195 struct command_ctx *cmd_ctx = NULL;
4196 struct lttng_poll_event events;
4197
4198 DBG("[thread] Manage client started");
4199
4200 rcu_register_thread();
4201
4202 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4203
4204 health_code_update();
4205
4206 ret = lttcomm_listen_unix_sock(client_sock);
4207 if (ret < 0) {
4208 goto error_listen;
4209 }
4210
4211 /*
4212 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4213 * more will be added to this poll set.
4214 */
4215 ret = sessiond_set_thread_pollset(&events, 2);
4216 if (ret < 0) {
4217 goto error_create_poll;
4218 }
4219
4220 /* Add the application registration socket */
4221 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4222 if (ret < 0) {
4223 goto error;
4224 }
4225
4226 sessiond_notify_ready();
4227 ret = sem_post(&load_info->message_thread_ready);
4228 if (ret) {
4229 PERROR("sem_post message_thread_ready");
4230 goto error;
4231 }
4232
4233 /* This testpoint is after we signal readiness to the parent. */
4234 if (testpoint(sessiond_thread_manage_clients)) {
4235 goto error;
4236 }
4237
4238 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4239 goto error;
4240 }
4241
4242 health_code_update();
4243
4244 while (1) {
4245 DBG("Accepting client command ...");
4246
4247 /* Inifinite blocking call, waiting for transmission */
4248 restart:
4249 health_poll_entry();
4250 ret = lttng_poll_wait(&events, -1);
4251 health_poll_exit();
4252 if (ret < 0) {
4253 /*
4254 * Restart interrupted system call.
4255 */
4256 if (errno == EINTR) {
4257 goto restart;
4258 }
4259 goto error;
4260 }
4261
4262 nb_fd = ret;
4263
4264 for (i = 0; i < nb_fd; i++) {
4265 /* Fetch once the poll data */
4266 revents = LTTNG_POLL_GETEV(&events, i);
4267 pollfd = LTTNG_POLL_GETFD(&events, i);
4268
4269 health_code_update();
4270
4271 if (!revents) {
4272 /* No activity for this FD (poll implementation). */
4273 continue;
4274 }
4275
4276 /* Thread quit pipe has been closed. Killing thread. */
4277 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4278 if (ret) {
4279 err = 0;
4280 goto exit;
4281 }
4282
4283 /* Event on the registration socket */
4284 if (pollfd == client_sock) {
4285 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4286 ERR("Client socket poll error");
4287 goto error;
4288 }
4289 }
4290 }
4291
4292 DBG("Wait for client response");
4293
4294 health_code_update();
4295
4296 sock = lttcomm_accept_unix_sock(client_sock);
4297 if (sock < 0) {
4298 goto error;
4299 }
4300
4301 /*
4302 * Set the CLOEXEC flag. Return code is useless because either way, the
4303 * show must go on.
4304 */
4305 (void) utils_set_fd_cloexec(sock);
4306
4307 /* Set socket option for credentials retrieval */
4308 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4309 if (ret < 0) {
4310 goto error;
4311 }
4312
4313 /* Allocate context command to process the client request */
4314 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4315 if (cmd_ctx == NULL) {
4316 PERROR("zmalloc cmd_ctx");
4317 goto error;
4318 }
4319
4320 /* Allocate data buffer for reception */
4321 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4322 if (cmd_ctx->lsm == NULL) {
4323 PERROR("zmalloc cmd_ctx->lsm");
4324 goto error;
4325 }
4326
4327 cmd_ctx->llm = NULL;
4328 cmd_ctx->session = NULL;
4329
4330 health_code_update();
4331
4332 /*
4333 * Data is received from the lttng client. The struct
4334 * lttcomm_session_msg (lsm) contains the command and data request of
4335 * the client.
4336 */
4337 DBG("Receiving data from client ...");
4338 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4339 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4340 if (ret <= 0) {
4341 DBG("Nothing recv() from client... continuing");
4342 ret = close(sock);
4343 if (ret) {
4344 PERROR("close");
4345 }
4346 sock = -1;
4347 clean_command_ctx(&cmd_ctx);
4348 continue;
4349 }
4350
4351 health_code_update();
4352
4353 // TODO: Validate cmd_ctx including sanity check for
4354 // security purpose.
4355
4356 rcu_thread_online();
4357 /*
4358 * This function dispatch the work to the kernel or userspace tracer
4359 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4360 * informations for the client. The command context struct contains
4361 * everything this function may needs.
4362 */
4363 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4364 rcu_thread_offline();
4365 if (ret < 0) {
4366 ret = close(sock);
4367 if (ret) {
4368 PERROR("close");
4369 }
4370 sock = -1;
4371 /*
4372 * TODO: Inform client somehow of the fatal error. At
4373 * this point, ret < 0 means that a zmalloc failed
4374 * (ENOMEM). Error detected but still accept
4375 * command, unless a socket error has been
4376 * detected.
4377 */
4378 clean_command_ctx(&cmd_ctx);
4379 continue;
4380 }
4381
4382 health_code_update();
4383
4384 DBG("Sending response (size: %d, retcode: %s)",
4385 cmd_ctx->lttng_msg_size,
4386 lttng_strerror(-cmd_ctx->llm->ret_code));
4387 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4388 if (ret < 0) {
4389 ERR("Failed to send data back to client");
4390 }
4391
4392 /* End of transmission */
4393 ret = close(sock);
4394 if (ret) {
4395 PERROR("close");
4396 }
4397 sock = -1;
4398
4399 clean_command_ctx(&cmd_ctx);
4400
4401 health_code_update();
4402 }
4403
4404 exit:
4405 error:
4406 if (sock >= 0) {
4407 ret = close(sock);
4408 if (ret) {
4409 PERROR("close");
4410 }
4411 }
4412
4413 lttng_poll_clean(&events);
4414 clean_command_ctx(&cmd_ctx);
4415
4416 error_listen:
4417 error_create_poll:
4418 unlink(client_unix_sock_path);
4419 if (client_sock >= 0) {
4420 ret = close(client_sock);
4421 if (ret) {
4422 PERROR("close");
4423 }
4424 }
4425
4426 if (err) {
4427 health_error();
4428 ERR("Health error occurred in %s", __func__);
4429 }
4430
4431 health_unregister(health_sessiond);
4432
4433 DBG("Client thread dying");
4434
4435 rcu_unregister_thread();
4436
4437 /*
4438 * Since we are creating the consumer threads, we own them, so we need
4439 * to join them before our thread exits.
4440 */
4441 ret = join_consumer_thread(&kconsumer_data);
4442 if (ret) {
4443 errno = ret;
4444 PERROR("join_consumer");
4445 }
4446
4447 ret = join_consumer_thread(&ustconsumer32_data);
4448 if (ret) {
4449 errno = ret;
4450 PERROR("join_consumer ust32");
4451 }
4452
4453 ret = join_consumer_thread(&ustconsumer64_data);
4454 if (ret) {
4455 errno = ret;
4456 PERROR("join_consumer ust64");
4457 }
4458 return NULL;
4459 }
4460
4461
4462 /*
4463 * usage function on stderr
4464 */
4465 static void usage(void)
4466 {
4467 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4468 fprintf(stderr, " -h, --help Display this usage.\n");
4469 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4470 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4471 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4472 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4473 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4474 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4475 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4476 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4477 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4478 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4479 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4480 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4481 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4482 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4483 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4484 fprintf(stderr, " -V, --version Show version number.\n");
4485 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4486 fprintf(stderr, " -q, --quiet No output at all.\n");
4487 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4488 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4489 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4490 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4491 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4492 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4493 fprintf(stderr, " -l --load PATH Load session configuration\n");
4494 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4495 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4496 }
4497
4498 static int string_match(const char *str1, const char *str2)
4499 {
4500 return (str1 && str2) && !strcmp(str1, str2);
4501 }
4502
4503 /*
4504 * Take an option from the getopt output and set it in the right variable to be
4505 * used later.
4506 *
4507 * Return 0 on success else a negative value.
4508 */
4509 static int set_option(int opt, const char *arg, const char *optname)
4510 {
4511 int ret = 0;
4512
4513 if (arg && arg[0] == '\0') {
4514 /*
4515 * This only happens if the value is read from daemon config
4516 * file. This means the option requires an argument and the
4517 * configuration file contains a line such as:
4518 * my_option =
4519 */
4520 ret = -EINVAL;
4521 goto end;
4522 }
4523
4524 if (string_match(optname, "client-sock") || opt == 'c') {
4525 if (lttng_is_setuid_setgid()) {
4526 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4527 "-c, --client-sock");
4528 } else {
4529 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4530 }
4531 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4532 if (lttng_is_setuid_setgid()) {
4533 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4534 "-a, --apps-sock");
4535 } else {
4536 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4537 }
4538 } else if (string_match(optname, "daemonize") || opt == 'd') {
4539 opt_daemon = 1;
4540 } else if (string_match(optname, "background") || opt == 'b') {
4541 opt_background = 1;
4542 } else if (string_match(optname, "group") || opt == 'g') {
4543 if (lttng_is_setuid_setgid()) {
4544 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4545 "-g, --group");
4546 } else {
4547 /*
4548 * If the override option is set, the pointer points to a
4549 * *non* const thus freeing it even though the variable type is
4550 * set to const.
4551 */
4552 if (tracing_group_name_override) {
4553 free((void *) tracing_group_name);
4554 }
4555 tracing_group_name = strdup(arg);
4556 if (!tracing_group_name) {
4557 PERROR("strdup");
4558 ret = -ENOMEM;
4559 }
4560 tracing_group_name_override = 1;
4561 }
4562 } else if (string_match(optname, "help") || opt == 'h') {
4563 usage();
4564 exit(EXIT_SUCCESS);
4565 } else if (string_match(optname, "version") || opt == 'V') {
4566 fprintf(stdout, "%s\n", VERSION);
4567 exit(EXIT_SUCCESS);
4568 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4569 opt_sig_parent = 1;
4570 } else if (string_match(optname, "kconsumerd-err-sock")) {
4571 if (lttng_is_setuid_setgid()) {
4572 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4573 "--kconsumerd-err-sock");
4574 } else {
4575 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4576 }
4577 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4578 if (lttng_is_setuid_setgid()) {
4579 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4580 "--kconsumerd-cmd-sock");
4581 } else {
4582 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4583 }
4584 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4585 if (lttng_is_setuid_setgid()) {
4586 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4587 "--ustconsumerd64-err-sock");
4588 } else {
4589 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4590 }
4591 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4592 if (lttng_is_setuid_setgid()) {
4593 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4594 "--ustconsumerd64-cmd-sock");
4595 } else {
4596 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4597 }
4598 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4599 if (lttng_is_setuid_setgid()) {
4600 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4601 "--ustconsumerd32-err-sock");
4602 } else {
4603 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4604 }
4605 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4606 if (lttng_is_setuid_setgid()) {
4607 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4608 "--ustconsumerd32-cmd-sock");
4609 } else {
4610 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4611 }
4612 } else if (string_match(optname, "no-kernel")) {
4613 opt_no_kernel = 1;
4614 } else if (string_match(optname, "quiet") || opt == 'q') {
4615 lttng_opt_quiet = 1;
4616 } else if (string_match(optname, "verbose") || opt == 'v') {
4617 /* Verbose level can increase using multiple -v */
4618 if (arg) {
4619 /* Value obtained from config file */
4620 lttng_opt_verbose = config_parse_value(arg);
4621 } else {
4622 /* -v used on command line */
4623 lttng_opt_verbose++;
4624 }
4625 /* Clamp value to [0, 3] */
4626 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4627 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4628 } else if (string_match(optname, "verbose-consumer")) {
4629 if (arg) {
4630 opt_verbose_consumer = config_parse_value(arg);
4631 } else {
4632 opt_verbose_consumer += 1;
4633 }
4634 } else if (string_match(optname, "consumerd32-path")) {
4635 if (lttng_is_setuid_setgid()) {
4636 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4637 "--consumerd32-path");
4638 } else {
4639 if (consumerd32_bin_override) {
4640 free((void *) consumerd32_bin);
4641 }
4642 consumerd32_bin = strdup(arg);
4643 if (!consumerd32_bin) {
4644 PERROR("strdup");
4645 ret = -ENOMEM;
4646 }
4647 consumerd32_bin_override = 1;
4648 }
4649 } else if (string_match(optname, "consumerd32-libdir")) {
4650 if (lttng_is_setuid_setgid()) {
4651 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4652 "--consumerd32-libdir");
4653 } else {
4654 if (consumerd32_libdir_override) {
4655 free((void *) consumerd32_libdir);
4656 }
4657 consumerd32_libdir = strdup(arg);
4658 if (!consumerd32_libdir) {
4659 PERROR("strdup");
4660 ret = -ENOMEM;
4661 }
4662 consumerd32_libdir_override = 1;
4663 }
4664 } else if (string_match(optname, "consumerd64-path")) {
4665 if (lttng_is_setuid_setgid()) {
4666 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4667 "--consumerd64-path");
4668 } else {
4669 if (consumerd64_bin_override) {
4670 free((void *) consumerd64_bin);
4671 }
4672 consumerd64_bin = strdup(arg);
4673 if (!consumerd64_bin) {
4674 PERROR("strdup");
4675 ret = -ENOMEM;
4676 }
4677 consumerd64_bin_override = 1;
4678 }
4679 } else if (string_match(optname, "consumerd64-libdir")) {
4680 if (lttng_is_setuid_setgid()) {
4681 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4682 "--consumerd64-libdir");
4683 } else {
4684 if (consumerd64_libdir_override) {
4685 free((void *) consumerd64_libdir);
4686 }
4687 consumerd64_libdir = strdup(arg);
4688 if (!consumerd64_libdir) {
4689 PERROR("strdup");
4690 ret = -ENOMEM;
4691 }
4692 consumerd64_libdir_override = 1;
4693 }
4694 } else if (string_match(optname, "pidfile") || opt == 'p') {
4695 if (lttng_is_setuid_setgid()) {
4696 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4697 "-p, --pidfile");
4698 } else {
4699 free(opt_pidfile);
4700 opt_pidfile = strdup(arg);
4701 if (!opt_pidfile) {
4702 PERROR("strdup");
4703 ret = -ENOMEM;
4704 }
4705 }
4706 } else if (string_match(optname, "agent-tcp-port")) {
4707 if (lttng_is_setuid_setgid()) {
4708 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4709 "--agent-tcp-port");
4710 } else {
4711 unsigned long v;
4712
4713 if (!arg) {
4714 ret = -EINVAL;
4715 goto end;
4716 }
4717 errno = 0;
4718 v = strtoul(arg, NULL, 0);
4719 if (errno != 0 || !isdigit(arg[0])) {
4720 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4721 return -1;
4722 }
4723 if (v == 0 || v >= 65535) {
4724 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4725 return -1;
4726 }
4727 agent_tcp_port = (uint32_t) v;
4728 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4729 }
4730 } else if (string_match(optname, "load") || opt == 'l') {
4731 if (lttng_is_setuid_setgid()) {
4732 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4733 "-l, --load");
4734 } else {
4735 free(opt_load_session_path);
4736 opt_load_session_path = strdup(arg);
4737 if (!opt_load_session_path) {
4738 PERROR("strdup");
4739 ret = -ENOMEM;
4740 }
4741 }
4742 } else if (string_match(optname, "kmod-probes")) {
4743 if (lttng_is_setuid_setgid()) {
4744 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4745 "--kmod-probes");
4746 } else {
4747 free(kmod_probes_list);
4748 kmod_probes_list = strdup(arg);
4749 if (!kmod_probes_list) {
4750 PERROR("strdup");
4751 ret = -ENOMEM;
4752 }
4753 }
4754 } else if (string_match(optname, "extra-kmod-probes")) {
4755 if (lttng_is_setuid_setgid()) {
4756 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4757 "--extra-kmod-probes");
4758 } else {
4759 free(kmod_extra_probes_list);
4760 kmod_extra_probes_list = strdup(arg);
4761 if (!kmod_extra_probes_list) {
4762 PERROR("strdup");
4763 ret = -ENOMEM;
4764 }
4765 }
4766 } else if (string_match(optname, "config") || opt == 'f') {
4767 /* This is handled in set_options() thus silent skip. */
4768 goto end;
4769 } else {
4770 /* Unknown option or other error.
4771 * Error is printed by getopt, just return */
4772 ret = -1;
4773 }
4774
4775 end:
4776 if (ret == -EINVAL) {
4777 const char *opt_name = "unknown";
4778 int i;
4779
4780 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4781 i++) {
4782 if (opt == long_options[i].val) {
4783 opt_name = long_options[i].name;
4784 break;
4785 }
4786 }
4787
4788 WARN("Invalid argument provided for option \"%s\", using default value.",
4789 opt_name);
4790 }
4791
4792 return ret;
4793 }
4794
4795 /*
4796 * config_entry_handler_cb used to handle options read from a config file.
4797 * See config_entry_handler_cb comment in common/config/config.h for the
4798 * return value conventions.
4799 */
4800 static int config_entry_handler(const struct config_entry *entry, void *unused)
4801 {
4802 int ret = 0, i;
4803
4804 if (!entry || !entry->name || !entry->value) {
4805 ret = -EINVAL;
4806 goto end;
4807 }
4808
4809 /* Check if the option is to be ignored */
4810 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4811 if (!strcmp(entry->name, config_ignore_options[i])) {
4812 goto end;
4813 }
4814 }
4815
4816 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4817 i++) {
4818
4819 /* Ignore if not fully matched. */
4820 if (strcmp(entry->name, long_options[i].name)) {
4821 continue;
4822 }
4823
4824 /*
4825 * If the option takes no argument on the command line, we have to
4826 * check if the value is "true". We support non-zero numeric values,
4827 * true, on and yes.
4828 */
4829 if (!long_options[i].has_arg) {
4830 ret = config_parse_value(entry->value);
4831 if (ret <= 0) {
4832 if (ret) {
4833 WARN("Invalid configuration value \"%s\" for option %s",
4834 entry->value, entry->name);
4835 }
4836 /* False, skip boolean config option. */
4837 goto end;
4838 }
4839 }
4840
4841 ret = set_option(long_options[i].val, entry->value, entry->name);
4842 goto end;
4843 }
4844
4845 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4846
4847 end:
4848 return ret;
4849 }
4850
4851 /*
4852 * daemon configuration loading and argument parsing
4853 */
4854 static int set_options(int argc, char **argv)
4855 {
4856 int ret = 0, c = 0, option_index = 0;
4857 int orig_optopt = optopt, orig_optind = optind;
4858 char *optstring;
4859 const char *config_path = NULL;
4860
4861 optstring = utils_generate_optstring(long_options,
4862 sizeof(long_options) / sizeof(struct option));
4863 if (!optstring) {
4864 ret = -ENOMEM;
4865 goto end;
4866 }
4867
4868 /* Check for the --config option */
4869 while ((c = getopt_long(argc, argv, optstring, long_options,
4870 &option_index)) != -1) {
4871 if (c == '?') {
4872 ret = -EINVAL;
4873 goto end;
4874 } else if (c != 'f') {
4875 /* if not equal to --config option. */
4876 continue;
4877 }
4878
4879 if (lttng_is_setuid_setgid()) {
4880 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4881 "-f, --config");
4882 } else {
4883 config_path = utils_expand_path(optarg);
4884 if (!config_path) {
4885 ERR("Failed to resolve path: %s", optarg);
4886 }
4887 }
4888 }
4889
4890 ret = config_get_section_entries(config_path, config_section_name,
4891 config_entry_handler, NULL);
4892 if (ret) {
4893 if (ret > 0) {
4894 ERR("Invalid configuration option at line %i", ret);
4895 ret = -1;
4896 }
4897 goto end;
4898 }
4899
4900 /* Reset getopt's global state */
4901 optopt = orig_optopt;
4902 optind = orig_optind;
4903 while (1) {
4904 option_index = -1;
4905 /*
4906 * getopt_long() will not set option_index if it encounters a
4907 * short option.
4908 */
4909 c = getopt_long(argc, argv, optstring, long_options,
4910 &option_index);
4911 if (c == -1) {
4912 break;
4913 }
4914
4915 /*
4916 * Pass NULL as the long option name if popt left the index
4917 * unset.
4918 */
4919 ret = set_option(c, optarg,
4920 option_index < 0 ? NULL :
4921 long_options[option_index].name);
4922 if (ret < 0) {
4923 break;
4924 }
4925 }
4926
4927 end:
4928 free(optstring);
4929 return ret;
4930 }
4931
4932 /*
4933 * Creates the two needed socket by the daemon.
4934 * apps_sock - The communication socket for all UST apps.
4935 * client_sock - The communication of the cli tool (lttng).
4936 */
4937 static int init_daemon_socket(void)
4938 {
4939 int ret = 0;
4940 mode_t old_umask;
4941
4942 old_umask = umask(0);
4943
4944 /* Create client tool unix socket */
4945 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4946 if (client_sock < 0) {
4947 ERR("Create unix sock failed: %s", client_unix_sock_path);
4948 ret = -1;
4949 goto end;
4950 }
4951
4952 /* Set the cloexec flag */
4953 ret = utils_set_fd_cloexec(client_sock);
4954 if (ret < 0) {
4955 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4956 "Continuing but note that the consumer daemon will have a "
4957 "reference to this socket on exec()", client_sock);
4958 }
4959
4960 /* File permission MUST be 660 */
4961 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4962 if (ret < 0) {
4963 ERR("Set file permissions failed: %s", client_unix_sock_path);
4964 PERROR("chmod");
4965 goto end;
4966 }
4967
4968 /* Create the application unix socket */
4969 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4970 if (apps_sock < 0) {
4971 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4972 ret = -1;
4973 goto end;
4974 }
4975
4976 /* Set the cloexec flag */
4977 ret = utils_set_fd_cloexec(apps_sock);
4978 if (ret < 0) {
4979 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4980 "Continuing but note that the consumer daemon will have a "
4981 "reference to this socket on exec()", apps_sock);
4982 }
4983
4984 /* File permission MUST be 666 */
4985 ret = chmod(apps_unix_sock_path,
4986 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4987 if (ret < 0) {
4988 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4989 PERROR("chmod");
4990 goto end;
4991 }
4992
4993 DBG3("Session daemon client socket %d and application socket %d created",
4994 client_sock, apps_sock);
4995
4996 end:
4997 umask(old_umask);
4998 return ret;
4999 }
5000
5001 /*
5002 * Check if the global socket is available, and if a daemon is answering at the
5003 * other side. If yes, error is returned.
5004 */
5005 static int check_existing_daemon(void)
5006 {
5007 /* Is there anybody out there ? */
5008 if (lttng_session_daemon_alive()) {
5009 return -EEXIST;
5010 }
5011
5012 return 0;
5013 }
5014
5015 /*
5016 * Set the tracing group gid onto the client socket.
5017 *
5018 * Race window between mkdir and chown is OK because we are going from more
5019 * permissive (root.root) to less permissive (root.tracing).
5020 */
5021 static int set_permissions(char *rundir)
5022 {
5023 int ret;
5024 gid_t gid;
5025
5026 gid = utils_get_group_id(tracing_group_name);
5027
5028 /* Set lttng run dir */
5029 ret = chown(rundir, 0, gid);
5030 if (ret < 0) {
5031 ERR("Unable to set group on %s", rundir);
5032 PERROR("chown");
5033 }
5034
5035 /*
5036 * Ensure all applications and tracing group can search the run
5037 * dir. Allow everyone to read the directory, since it does not
5038 * buy us anything to hide its content.
5039 */
5040 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5041 if (ret < 0) {
5042 ERR("Unable to set permissions on %s", rundir);
5043 PERROR("chmod");
5044 }
5045
5046 /* lttng client socket path */
5047 ret = chown(client_unix_sock_path, 0, gid);
5048 if (ret < 0) {
5049 ERR("Unable to set group on %s", client_unix_sock_path);
5050 PERROR("chown");
5051 }
5052
5053 /* kconsumer error socket path */
5054 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5055 if (ret < 0) {
5056 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5057 PERROR("chown");
5058 }
5059
5060 /* 64-bit ustconsumer error socket path */
5061 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5062 if (ret < 0) {
5063 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5064 PERROR("chown");
5065 }
5066
5067 /* 32-bit ustconsumer compat32 error socket path */
5068 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5069 if (ret < 0) {
5070 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5071 PERROR("chown");
5072 }
5073
5074 DBG("All permissions are set");
5075
5076 return ret;
5077 }
5078
5079 /*
5080 * Create the lttng run directory needed for all global sockets and pipe.
5081 */
5082 static int create_lttng_rundir(const char *rundir)
5083 {
5084 int ret;
5085
5086 DBG3("Creating LTTng run directory: %s", rundir);
5087
5088 ret = mkdir(rundir, S_IRWXU);
5089 if (ret < 0) {
5090 if (errno != EEXIST) {
5091 ERR("Unable to create %s", rundir);
5092 goto error;
5093 } else {
5094 ret = 0;
5095 }
5096 }
5097
5098 error:
5099 return ret;
5100 }
5101
5102 /*
5103 * Setup sockets and directory needed by the kconsumerd communication with the
5104 * session daemon.
5105 */
5106 static int set_consumer_sockets(struct consumer_data *consumer_data,
5107 const char *rundir)
5108 {
5109 int ret;
5110 char path[PATH_MAX];
5111
5112 switch (consumer_data->type) {
5113 case LTTNG_CONSUMER_KERNEL:
5114 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5115 break;
5116 case LTTNG_CONSUMER64_UST:
5117 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5118 break;
5119 case LTTNG_CONSUMER32_UST:
5120 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5121 break;
5122 default:
5123 ERR("Consumer type unknown");
5124 ret = -EINVAL;
5125 goto error;
5126 }
5127
5128 DBG2("Creating consumer directory: %s", path);
5129
5130 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5131 if (ret < 0) {
5132 if (errno != EEXIST) {
5133 PERROR("mkdir");
5134 ERR("Failed to create %s", path);
5135 goto error;
5136 }
5137 ret = -1;
5138 }
5139 if (is_root) {
5140 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5141 if (ret < 0) {
5142 ERR("Unable to set group on %s", path);
5143 PERROR("chown");
5144 goto error;
5145 }
5146 }
5147
5148 /* Create the kconsumerd error unix socket */
5149 consumer_data->err_sock =
5150 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5151 if (consumer_data->err_sock < 0) {
5152 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5153 ret = -1;
5154 goto error;
5155 }
5156
5157 /*
5158 * Set the CLOEXEC flag. Return code is useless because either way, the
5159 * show must go on.
5160 */
5161 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5162 if (ret < 0) {
5163 PERROR("utils_set_fd_cloexec");
5164 /* continue anyway */
5165 }
5166
5167 /* File permission MUST be 660 */
5168 ret = chmod(consumer_data->err_unix_sock_path,
5169 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5170 if (ret < 0) {
5171 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5172 PERROR("chmod");
5173 goto error;
5174 }
5175
5176 error:
5177 return ret;
5178 }
5179
5180 /*
5181 * Signal handler for the daemon
5182 *
5183 * Simply stop all worker threads, leaving main() return gracefully after
5184 * joining all threads and calling cleanup().
5185 */
5186 static void sighandler(int sig)
5187 {
5188 switch (sig) {
5189 case SIGPIPE:
5190 DBG("SIGPIPE caught");
5191 return;
5192 case SIGINT:
5193 DBG("SIGINT caught");
5194 stop_threads();
5195 break;
5196 case SIGTERM:
5197 DBG("SIGTERM caught");
5198 stop_threads();
5199 break;
5200 case SIGUSR1:
5201 CMM_STORE_SHARED(recv_child_signal, 1);
5202 break;
5203 default:
5204 break;
5205 }
5206 }
5207
5208 /*
5209 * Setup signal handler for :
5210 * SIGINT, SIGTERM, SIGPIPE
5211 */
5212 static int set_signal_handler(void)
5213 {
5214 int ret = 0;
5215 struct sigaction sa;
5216 sigset_t sigset;
5217
5218 if ((ret = sigemptyset(&sigset)) < 0) {
5219 PERROR("sigemptyset");
5220 return ret;
5221 }
5222
5223 sa.sa_handler = sighandler;
5224 sa.sa_mask = sigset;
5225 sa.sa_flags = 0;
5226 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5227 PERROR("sigaction");
5228 return ret;
5229 }
5230
5231 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5232 PERROR("sigaction");
5233 return ret;
5234 }
5235
5236 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5237 PERROR("sigaction");
5238 return ret;
5239 }
5240
5241 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5242 PERROR("sigaction");
5243 return ret;
5244 }
5245
5246 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5247
5248 return ret;
5249 }
5250
5251 /*
5252 * Set open files limit to unlimited. This daemon can open a large number of
5253 * file descriptors in order to consumer multiple kernel traces.
5254 */
5255 static void set_ulimit(void)
5256 {
5257 int ret;
5258 struct rlimit lim;
5259
5260 /* The kernel does not allowed an infinite limit for open files */
5261 lim.rlim_cur = 65535;
5262 lim.rlim_max = 65535;
5263
5264 ret = setrlimit(RLIMIT_NOFILE, &lim);
5265 if (ret < 0) {
5266 PERROR("failed to set open files limit");
5267 }
5268 }
5269
5270 /*
5271 * Write pidfile using the rundir and opt_pidfile.
5272 */
5273 static int write_pidfile(void)
5274 {
5275 int ret;
5276 char pidfile_path[PATH_MAX];
5277
5278 assert(rundir);
5279
5280 if (opt_pidfile) {
5281 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5282 } else {
5283 /* Build pidfile path from rundir and opt_pidfile. */
5284 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5285 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5286 if (ret < 0) {
5287 PERROR("snprintf pidfile path");
5288 goto error;
5289 }
5290 }
5291
5292 /*
5293 * Create pid file in rundir.
5294 */
5295 ret = utils_create_pid_file(getpid(), pidfile_path);
5296 error:
5297 return ret;
5298 }
5299
5300 /*
5301 * Create lockfile using the rundir and return its fd.
5302 */
5303 static int create_lockfile(void)
5304 {
5305 int ret;
5306 char lockfile_path[PATH_MAX];
5307
5308 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5309 if (ret < 0) {
5310 goto error;
5311 }
5312
5313 ret = utils_create_lock_file(lockfile_path);
5314 error:
5315 return ret;
5316 }
5317
5318 /*
5319 * Write agent TCP port using the rundir.
5320 */
5321 static int write_agent_port(void)
5322 {
5323 int ret;
5324 char path[PATH_MAX];
5325
5326 assert(rundir);
5327
5328 ret = snprintf(path, sizeof(path), "%s/"
5329 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5330 if (ret < 0) {
5331 PERROR("snprintf agent port path");
5332 goto error;
5333 }
5334
5335 /*
5336 * Create TCP agent port file in rundir.
5337 */
5338 ret = utils_create_pid_file(agent_tcp_port, path);
5339
5340 error:
5341 return ret;
5342 }
5343
5344 /*
5345 * main
5346 */
5347 int main(int argc, char **argv)
5348 {
5349 int ret = 0, retval = 0;
5350 void *status;
5351 const char *home_path, *env_app_timeout;
5352
5353 /* Initialize agent apps ht global variable */
5354 agent_apps_ht_by_sock = NULL;
5355
5356 init_kernel_workarounds();
5357
5358 rcu_register_thread();
5359
5360 if (set_signal_handler()) {
5361 retval = -1;
5362 goto exit_set_signal_handler;
5363 }
5364
5365 setup_consumerd_path();
5366
5367 page_size = sysconf(_SC_PAGESIZE);
5368 if (page_size < 0) {
5369 PERROR("sysconf _SC_PAGESIZE");
5370 page_size = LONG_MAX;
5371 WARN("Fallback page size to %ld", page_size);
5372 }
5373
5374 /*
5375 * Parse arguments and load the daemon configuration file.
5376 *
5377 * We have an exit_options exit path to free memory reserved by
5378 * set_options. This is needed because the rest of sessiond_cleanup()
5379 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5380 * depends on set_options.
5381 */
5382 progname = argv[0];
5383 if (set_options(argc, argv)) {
5384 retval = -1;
5385 goto exit_options;
5386 }
5387
5388 /* Daemonize */
5389 if (opt_daemon || opt_background) {
5390 int i;
5391
5392 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5393 !opt_background);
5394 if (ret < 0) {
5395 retval = -1;
5396 goto exit_options;
5397 }
5398
5399 /*
5400 * We are in the child. Make sure all other file descriptors are
5401 * closed, in case we are called with more opened file
5402 * descriptors than the standard ones.
5403 */
5404 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5405 (void) close(i);
5406 }
5407 }
5408
5409 /*
5410 * Starting from here, we can create threads. This needs to be after
5411 * lttng_daemonize due to RCU.
5412 */
5413
5414 /*
5415 * Initialize the health check subsystem. This call should set the
5416 * appropriate time values.
5417 */
5418 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5419 if (!health_sessiond) {
5420 PERROR("health_app_create error");
5421 retval = -1;
5422 goto exit_health_sessiond_cleanup;
5423 }
5424
5425 if (init_ht_cleanup_quit_pipe()) {
5426 retval = -1;
5427 goto exit_ht_cleanup_quit_pipe;
5428 }
5429
5430 /* Setup the thread ht_cleanup communication pipe. */
5431 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5432 retval = -1;
5433 goto exit_ht_cleanup_pipe;
5434 }
5435
5436 /* Set up max poll set size */
5437 if (lttng_poll_set_max_size()) {
5438 retval = -1;
5439 goto exit_set_max_size;
5440 }
5441
5442 /* Create thread to clean up RCU hash tables */
5443 ret = pthread_create(&ht_cleanup_thread, NULL,
5444 thread_ht_cleanup, (void *) NULL);
5445 if (ret) {
5446 errno = ret;
5447 PERROR("pthread_create ht_cleanup");
5448 retval = -1;
5449 goto exit_ht_cleanup;
5450 }
5451
5452 /* Create thread quit pipe */
5453 if (init_thread_quit_pipe()) {
5454 retval = -1;
5455 goto exit_init_data;
5456 }
5457
5458 /* Check if daemon is UID = 0 */
5459 is_root = !getuid();
5460
5461 if (is_root) {
5462 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5463 if (!rundir) {
5464 retval = -1;
5465 goto exit_init_data;
5466 }
5467
5468 /* Create global run dir with root access */
5469 if (create_lttng_rundir(rundir)) {
5470 retval = -1;
5471 goto exit_init_data;
5472 }
5473
5474 if (strlen(apps_unix_sock_path) == 0) {
5475 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5476 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5477 if (ret < 0) {
5478 retval = -1;
5479 goto exit_init_data;
5480 }
5481 }
5482
5483 if (strlen(client_unix_sock_path) == 0) {
5484 ret = snprintf(client_unix_sock_path, PATH_MAX,
5485 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5486 if (ret < 0) {
5487 retval = -1;
5488 goto exit_init_data;
5489 }
5490 }
5491
5492 /* Set global SHM for ust */
5493 if (strlen(wait_shm_path) == 0) {
5494 ret = snprintf(wait_shm_path, PATH_MAX,
5495 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5496 if (ret < 0) {
5497 retval = -1;
5498 goto exit_init_data;
5499 }
5500 }
5501
5502 if (strlen(health_unix_sock_path) == 0) {
5503 ret = snprintf(health_unix_sock_path,
5504 sizeof(health_unix_sock_path),
5505 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5506 if (ret < 0) {
5507 retval = -1;
5508 goto exit_init_data;
5509 }
5510 }
5511
5512 /* Setup kernel consumerd path */
5513 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5514 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5515 if (ret < 0) {
5516 retval = -1;
5517 goto exit_init_data;
5518 }
5519 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5520 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5521 if (ret < 0) {
5522 retval = -1;
5523 goto exit_init_data;
5524 }
5525
5526 DBG2("Kernel consumer err path: %s",
5527 kconsumer_data.err_unix_sock_path);
5528 DBG2("Kernel consumer cmd path: %s",
5529 kconsumer_data.cmd_unix_sock_path);
5530 } else {
5531 home_path = utils_get_home_dir();
5532 if (home_path == NULL) {
5533 /* TODO: Add --socket PATH option */
5534 ERR("Can't get HOME directory for sockets creation.");
5535 retval = -1;
5536 goto exit_init_data;
5537 }
5538
5539 /*
5540 * Create rundir from home path. This will create something like
5541 * $HOME/.lttng
5542 */
5543 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5544 if (ret < 0) {
5545 retval = -1;
5546 goto exit_init_data;
5547 }
5548
5549 if (create_lttng_rundir(rundir)) {
5550 retval = -1;
5551 goto exit_init_data;
5552 }
5553
5554 if (strlen(apps_unix_sock_path) == 0) {
5555 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5556 DEFAULT_HOME_APPS_UNIX_SOCK,
5557 home_path);
5558 if (ret < 0) {
5559 retval = -1;
5560 goto exit_init_data;
5561 }
5562 }
5563
5564 /* Set the cli tool unix socket path */
5565 if (strlen(client_unix_sock_path) == 0) {
5566 ret = snprintf(client_unix_sock_path, PATH_MAX,
5567 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5568 home_path);
5569 if (ret < 0) {
5570 retval = -1;
5571 goto exit_init_data;
5572 }
5573 }
5574
5575 /* Set global SHM for ust */
5576 if (strlen(wait_shm_path) == 0) {
5577 ret = snprintf(wait_shm_path, PATH_MAX,
5578 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5579 getuid());
5580 if (ret < 0) {
5581 retval = -1;
5582 goto exit_init_data;
5583 }
5584 }
5585
5586 /* Set health check Unix path */
5587 if (strlen(health_unix_sock_path) == 0) {
5588 ret = snprintf(health_unix_sock_path,
5589 sizeof(health_unix_sock_path),
5590 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5591 home_path);
5592 if (ret < 0) {
5593 retval = -1;
5594 goto exit_init_data;
5595 }
5596 }
5597 }
5598
5599 lockfile_fd = create_lockfile();
5600 if (lockfile_fd < 0) {
5601 retval = -1;
5602 goto exit_init_data;
5603 }
5604
5605 /* Set consumer initial state */
5606 kernel_consumerd_state = CONSUMER_STOPPED;
5607 ust_consumerd_state = CONSUMER_STOPPED;
5608
5609 DBG("Client socket path %s", client_unix_sock_path);
5610 DBG("Application socket path %s", apps_unix_sock_path);
5611 DBG("Application wait path %s", wait_shm_path);
5612 DBG("LTTng run directory path: %s", rundir);
5613
5614 /* 32 bits consumerd path setup */
5615 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5616 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5617 if (ret < 0) {
5618 PERROR("snprintf 32-bit consumer error socket path");
5619 retval = -1;
5620 goto exit_init_data;
5621 }
5622 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5623 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5624 if (ret < 0) {
5625 PERROR("snprintf 32-bit consumer command socket path");
5626 retval = -1;
5627 goto exit_init_data;
5628 }
5629
5630 DBG2("UST consumer 32 bits err path: %s",
5631 ustconsumer32_data.err_unix_sock_path);
5632 DBG2("UST consumer 32 bits cmd path: %s",
5633 ustconsumer32_data.cmd_unix_sock_path);
5634
5635 /* 64 bits consumerd path setup */
5636 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5637 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5638 if (ret < 0) {
5639 PERROR("snprintf 64-bit consumer error socket path");
5640 retval = -1;
5641 goto exit_init_data;
5642 }
5643 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5644 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5645 if (ret < 0) {
5646 PERROR("snprintf 64-bit consumer command socket path");
5647 retval = -1;
5648 goto exit_init_data;
5649 }
5650
5651 DBG2("UST consumer 64 bits err path: %s",
5652 ustconsumer64_data.err_unix_sock_path);
5653 DBG2("UST consumer 64 bits cmd path: %s",
5654 ustconsumer64_data.cmd_unix_sock_path);
5655
5656 /*
5657 * See if daemon already exist.
5658 */
5659 if (check_existing_daemon()) {
5660 ERR("Already running daemon.\n");
5661 /*
5662 * We do not goto exit because we must not cleanup()
5663 * because a daemon is already running.
5664 */
5665 retval = -1;
5666 goto exit_init_data;
5667 }
5668
5669 /*
5670 * Init UST app hash table. Alloc hash table before this point since
5671 * cleanup() can get called after that point.
5672 */
5673 if (ust_app_ht_alloc()) {
5674 ERR("Failed to allocate UST app hash table");
5675 retval = -1;
5676 goto exit_init_data;
5677 }
5678
5679 /*
5680 * Initialize agent app hash table. We allocate the hash table here
5681 * since cleanup() can get called after this point.
5682 */
5683 if (agent_app_ht_alloc()) {
5684 ERR("Failed to allocate Agent app hash table");
5685 retval = -1;
5686 goto exit_init_data;
5687 }
5688
5689 /*
5690 * These actions must be executed as root. We do that *after* setting up
5691 * the sockets path because we MUST make the check for another daemon using
5692 * those paths *before* trying to set the kernel consumer sockets and init
5693 * kernel tracer.
5694 */
5695 if (is_root) {
5696 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5697 retval = -1;
5698 goto exit_init_data;
5699 }
5700
5701 /* Setup kernel tracer */
5702 if (!opt_no_kernel) {
5703 init_kernel_tracer();
5704 if (kernel_tracer_fd >= 0) {
5705 ret = syscall_init_table();
5706 if (ret < 0) {
5707 ERR("Unable to populate syscall table. "
5708 "Syscall tracing won't work "
5709 "for this session daemon.");
5710 }
5711 }
5712 }
5713
5714 /* Set ulimit for open files */
5715 set_ulimit();
5716 }
5717 /* init lttng_fd tracking must be done after set_ulimit. */
5718 lttng_fd_init();
5719
5720 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5721 retval = -1;
5722 goto exit_init_data;
5723 }
5724
5725 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5726 retval = -1;
5727 goto exit_init_data;
5728 }
5729
5730 /* Setup the needed unix socket */
5731 if (init_daemon_socket()) {
5732 retval = -1;
5733 goto exit_init_data;
5734 }
5735
5736 /* Set credentials to socket */
5737 if (is_root && set_permissions(rundir)) {
5738 retval = -1;
5739 goto exit_init_data;
5740 }
5741
5742 /* Get parent pid if -S, --sig-parent is specified. */
5743 if (opt_sig_parent) {
5744 ppid = getppid();
5745 }
5746
5747 /* Setup the kernel pipe for waking up the kernel thread */
5748 if (is_root && !opt_no_kernel) {
5749 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5750 retval = -1;
5751 goto exit_init_data;
5752 }
5753 }
5754
5755 /* Setup the thread apps communication pipe. */
5756 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5757 retval = -1;
5758 goto exit_init_data;
5759 }
5760
5761 /* Setup the thread apps notify communication pipe. */
5762 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5763 retval = -1;
5764 goto exit_init_data;
5765 }
5766
5767 /* Initialize global buffer per UID and PID registry. */
5768 buffer_reg_init_uid_registry();
5769 buffer_reg_init_pid_registry();
5770
5771 /* Init UST command queue. */
5772 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5773
5774 /*
5775 * Get session list pointer. This pointer MUST NOT be free'd. This list
5776 * is statically declared in session.c
5777 */
5778 session_list_ptr = session_get_list();
5779
5780 cmd_init();
5781
5782 /* Check for the application socket timeout env variable. */
5783 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5784 if (env_app_timeout) {
5785 app_socket_timeout = atoi(env_app_timeout);
5786 } else {
5787 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5788 }
5789
5790 ret = write_pidfile();
5791 if (ret) {
5792 ERR("Error in write_pidfile");
5793 retval = -1;
5794 goto exit_init_data;
5795 }
5796 ret = write_agent_port();
5797 if (ret) {
5798 ERR("Error in write_agent_port");
5799 retval = -1;
5800 goto exit_init_data;
5801 }
5802
5803 /* Initialize communication library */
5804 lttcomm_init();
5805 /* Initialize TCP timeout values */
5806 lttcomm_inet_init();
5807
5808 if (load_session_init_data(&load_info) < 0) {
5809 retval = -1;
5810 goto exit_init_data;
5811 }
5812 load_info->path = opt_load_session_path;
5813
5814 /* Create health-check thread */
5815 ret = pthread_create(&health_thread, NULL,
5816 thread_manage_health, (void *) NULL);
5817 if (ret) {
5818 errno = ret;
5819 PERROR("pthread_create health");
5820 retval = -1;
5821 goto exit_health;
5822 }
5823
5824 /* Create thread to manage the client socket */
5825 ret = pthread_create(&client_thread, NULL,
5826 thread_manage_clients, (void *) NULL);
5827 if (ret) {
5828 errno = ret;
5829 PERROR("pthread_create clients");
5830 retval = -1;
5831 goto exit_client;
5832 }
5833
5834 /* Create thread to dispatch registration */
5835 ret = pthread_create(&dispatch_thread, NULL,
5836 thread_dispatch_ust_registration, (void *) NULL);
5837 if (ret) {
5838 errno = ret;
5839 PERROR("pthread_create dispatch");
5840 retval = -1;
5841 goto exit_dispatch;
5842 }
5843
5844 /* Create thread to manage application registration. */
5845 ret = pthread_create(&reg_apps_thread, NULL,
5846 thread_registration_apps, (void *) NULL);
5847 if (ret) {
5848 errno = ret;
5849 PERROR("pthread_create registration");
5850 retval = -1;
5851 goto exit_reg_apps;
5852 }
5853
5854 /* Create thread to manage application socket */
5855 ret = pthread_create(&apps_thread, NULL,
5856 thread_manage_apps, (void *) NULL);
5857 if (ret) {
5858 errno = ret;
5859 PERROR("pthread_create apps");
5860 retval = -1;
5861 goto exit_apps;
5862 }
5863
5864 /* Create thread to manage application notify socket */
5865 ret = pthread_create(&apps_notify_thread, NULL,
5866 ust_thread_manage_notify, (void *) NULL);
5867 if (ret) {
5868 errno = ret;
5869 PERROR("pthread_create notify");
5870 retval = -1;
5871 goto exit_apps_notify;
5872 }
5873
5874 /* Create agent registration thread. */
5875 ret = pthread_create(&agent_reg_thread, NULL,
5876 agent_thread_manage_registration, (void *) NULL);
5877 if (ret) {
5878 errno = ret;
5879 PERROR("pthread_create agent");
5880 retval = -1;
5881 goto exit_agent_reg;
5882 }
5883
5884 /* Don't start this thread if kernel tracing is not requested nor root */
5885 if (is_root && !opt_no_kernel) {
5886 /* Create kernel thread to manage kernel event */
5887 ret = pthread_create(&kernel_thread, NULL,
5888 thread_manage_kernel, (void *) NULL);
5889 if (ret) {
5890 errno = ret;
5891 PERROR("pthread_create kernel");
5892 retval = -1;
5893 goto exit_kernel;
5894 }
5895 }
5896
5897 /* Create session loading thread. */
5898 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5899 load_info);
5900 if (ret) {
5901 errno = ret;
5902 PERROR("pthread_create load_session_thread");
5903 retval = -1;
5904 goto exit_load_session;
5905 }
5906
5907 /*
5908 * This is where we start awaiting program completion (e.g. through
5909 * signal that asks threads to teardown).
5910 */
5911
5912 ret = pthread_join(load_session_thread, &status);
5913 if (ret) {
5914 errno = ret;
5915 PERROR("pthread_join load_session_thread");
5916 retval = -1;
5917 }
5918 exit_load_session:
5919
5920 if (is_root && !opt_no_kernel) {
5921 ret = pthread_join(kernel_thread, &status);
5922 if (ret) {
5923 errno = ret;
5924 PERROR("pthread_join");
5925 retval = -1;
5926 }
5927 }
5928 exit_kernel:
5929
5930 ret = pthread_join(agent_reg_thread, &status);
5931 if (ret) {
5932 errno = ret;
5933 PERROR("pthread_join agent");
5934 retval = -1;
5935 }
5936 exit_agent_reg:
5937
5938 ret = pthread_join(apps_notify_thread, &status);
5939 if (ret) {
5940 errno = ret;
5941 PERROR("pthread_join apps notify");
5942 retval = -1;
5943 }
5944 exit_apps_notify:
5945
5946 ret = pthread_join(apps_thread, &status);
5947 if (ret) {
5948 errno = ret;
5949 PERROR("pthread_join apps");
5950 retval = -1;
5951 }
5952 exit_apps:
5953
5954 ret = pthread_join(reg_apps_thread, &status);
5955 if (ret) {
5956 errno = ret;
5957 PERROR("pthread_join");
5958 retval = -1;
5959 }
5960 exit_reg_apps:
5961
5962 /*
5963 * Join dispatch thread after joining reg_apps_thread to ensure
5964 * we don't leak applications in the queue.
5965 */
5966 ret = pthread_join(dispatch_thread, &status);
5967 if (ret) {
5968 errno = ret;
5969 PERROR("pthread_join");
5970 retval = -1;
5971 }
5972 exit_dispatch:
5973
5974 ret = pthread_join(client_thread, &status);
5975 if (ret) {
5976 errno = ret;
5977 PERROR("pthread_join");
5978 retval = -1;
5979 }
5980 exit_client:
5981
5982 ret = pthread_join(health_thread, &status);
5983 if (ret) {
5984 errno = ret;
5985 PERROR("pthread_join health thread");
5986 retval = -1;
5987 }
5988 exit_health:
5989
5990 exit_init_data:
5991 /*
5992 * sessiond_cleanup() is called when no other thread is running, except
5993 * the ht_cleanup thread, which is needed to destroy the hash tables.
5994 */
5995 rcu_thread_online();
5996 sessiond_cleanup();
5997 rcu_thread_offline();
5998 rcu_unregister_thread();
5999
6000 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6001 if (ret < 0) {
6002 ERR("write error on ht_cleanup quit pipe");
6003 retval = -1;
6004 }
6005
6006 ret = pthread_join(ht_cleanup_thread, &status);
6007 if (ret) {
6008 errno = ret;
6009 PERROR("pthread_join ht cleanup thread");
6010 retval = -1;
6011 }
6012 exit_ht_cleanup:
6013 exit_set_max_size:
6014
6015 utils_close_pipe(ht_cleanup_pipe);
6016 exit_ht_cleanup_pipe:
6017
6018 /*
6019 * Close the ht_cleanup quit pipe.
6020 */
6021 utils_close_pipe(ht_cleanup_quit_pipe);
6022 exit_ht_cleanup_quit_pipe:
6023
6024 health_app_destroy(health_sessiond);
6025 exit_health_sessiond_cleanup:
6026
6027 exit_options:
6028 sessiond_cleanup_options();
6029
6030 exit_set_signal_handler:
6031 if (!retval) {
6032 exit(EXIT_SUCCESS);
6033 } else {
6034 exit(EXIT_FAILURE);
6035 }
6036 }
This page took 0.15235 seconds and 5 git commands to generate.