Hide lttng_event_copy symbol
[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 _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <paths.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <inttypes.h>
31 #include <sys/mman.h>
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37 #include <sys/wait.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40 #include <ctype.h>
41
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/compat/getenv.h>
45 #include <common/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/session-config.h>
52 #include <common/dynamic-buffer.h>
53 #include <lttng/userspace-probe-internal.h>
54 #include <lttng/event-internal.h>
55
56 #include "lttng-sessiond.h"
57 #include "buffer-registry.h"
58 #include "channel.h"
59 #include "cmd.h"
60 #include "consumer.h"
61 #include "context.h"
62 #include "event.h"
63 #include "kernel.h"
64 #include "kernel-consumer.h"
65 #include "modprobe.h"
66 #include "shm.h"
67 #include "ust-ctl.h"
68 #include "ust-consumer.h"
69 #include "utils.h"
70 #include "fd-limit.h"
71 #include "health-sessiond.h"
72 #include "testpoint.h"
73 #include "ust-thread.h"
74 #include "agent-thread.h"
75 #include "save.h"
76 #include "load-session-thread.h"
77 #include "notification-thread.h"
78 #include "notification-thread-commands.h"
79 #include "rotation-thread.h"
80 #include "lttng-syscall.h"
81 #include "agent.h"
82 #include "ht-cleanup.h"
83 #include "sessiond-config.h"
84 #include "sessiond-timer.h"
85
86 static const char *help_msg =
87 #ifdef LTTNG_EMBED_HELP
88 #include <lttng-sessiond.8.h>
89 #else
90 NULL
91 #endif
92 ;
93
94 const char *progname;
95 static pid_t ppid; /* Parent PID for --sig-parent option */
96 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
97 static int lockfile_fd = -1;
98
99 /* Set to 1 when a SIGUSR1 signal is received. */
100 static int recv_child_signal;
101
102 static struct lttng_kernel_tracer_version kernel_tracer_version;
103 static struct lttng_kernel_tracer_abi_version kernel_tracer_abi_version;
104
105 /*
106 * Consumer daemon specific control data. Every value not initialized here is
107 * set to 0 by the static definition.
108 */
109 static struct consumer_data kconsumer_data = {
110 .type = LTTNG_CONSUMER_KERNEL,
111 .err_sock = -1,
112 .cmd_sock = -1,
113 .channel_monitor_pipe = -1,
114 .channel_rotate_pipe = -1,
115 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
116 .lock = PTHREAD_MUTEX_INITIALIZER,
117 .cond = PTHREAD_COND_INITIALIZER,
118 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
119 };
120 static struct consumer_data ustconsumer64_data = {
121 .type = LTTNG_CONSUMER64_UST,
122 .err_sock = -1,
123 .cmd_sock = -1,
124 .channel_monitor_pipe = -1,
125 .channel_rotate_pipe = -1,
126 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
127 .lock = PTHREAD_MUTEX_INITIALIZER,
128 .cond = PTHREAD_COND_INITIALIZER,
129 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
130 };
131 static struct consumer_data ustconsumer32_data = {
132 .type = LTTNG_CONSUMER32_UST,
133 .err_sock = -1,
134 .cmd_sock = -1,
135 .channel_monitor_pipe = -1,
136 .channel_rotate_pipe = -1,
137 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
138 .lock = PTHREAD_MUTEX_INITIALIZER,
139 .cond = PTHREAD_COND_INITIALIZER,
140 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
141 };
142
143 /* Command line options */
144 static const struct option long_options[] = {
145 { "client-sock", required_argument, 0, 'c' },
146 { "apps-sock", required_argument, 0, 'a' },
147 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
148 { "kconsumerd-err-sock", required_argument, 0, '\0' },
149 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
150 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
151 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
152 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
153 { "consumerd32-path", required_argument, 0, '\0' },
154 { "consumerd32-libdir", required_argument, 0, '\0' },
155 { "consumerd64-path", required_argument, 0, '\0' },
156 { "consumerd64-libdir", required_argument, 0, '\0' },
157 { "daemonize", no_argument, 0, 'd' },
158 { "background", no_argument, 0, 'b' },
159 { "sig-parent", no_argument, 0, 'S' },
160 { "help", no_argument, 0, 'h' },
161 { "group", required_argument, 0, 'g' },
162 { "version", no_argument, 0, 'V' },
163 { "quiet", no_argument, 0, 'q' },
164 { "verbose", no_argument, 0, 'v' },
165 { "verbose-consumer", no_argument, 0, '\0' },
166 { "no-kernel", no_argument, 0, '\0' },
167 { "pidfile", required_argument, 0, 'p' },
168 { "agent-tcp-port", required_argument, 0, '\0' },
169 { "config", required_argument, 0, 'f' },
170 { "load", required_argument, 0, 'l' },
171 { "kmod-probes", required_argument, 0, '\0' },
172 { "extra-kmod-probes", required_argument, 0, '\0' },
173 { NULL, 0, 0, 0 }
174 };
175
176 struct sessiond_config config;
177
178 /* Command line options to ignore from configuration file */
179 static const char *config_ignore_options[] = { "help", "version", "config" };
180
181 /* Shared between threads */
182 static int dispatch_thread_exit;
183
184 /* Sockets and FDs */
185 static int client_sock = -1;
186 static int apps_sock = -1;
187 int kernel_tracer_fd = -1;
188 static int kernel_poll_pipe[2] = { -1, -1 };
189
190 /*
191 * Quit pipe for all threads. This permits a single cancellation point
192 * for all threads when receiving an event on the pipe.
193 */
194 static int thread_quit_pipe[2] = { -1, -1 };
195
196 /*
197 * This pipe is used to inform the thread managing application communication
198 * that a command is queued and ready to be processed.
199 */
200 static int apps_cmd_pipe[2] = { -1, -1 };
201
202 int apps_cmd_notify_pipe[2] = { -1, -1 };
203
204 /* Pthread, Mutexes and Semaphores */
205 static pthread_t apps_thread;
206 static pthread_t apps_notify_thread;
207 static pthread_t reg_apps_thread;
208 static pthread_t client_thread;
209 static pthread_t kernel_thread;
210 static pthread_t dispatch_thread;
211 static pthread_t health_thread;
212 static pthread_t ht_cleanup_thread;
213 static pthread_t agent_reg_thread;
214 static pthread_t load_session_thread;
215 static pthread_t notification_thread;
216 static pthread_t rotation_thread;
217 static pthread_t timer_thread;
218
219 /*
220 * UST registration command queue. This queue is tied with a futex and uses a N
221 * wakers / 1 waiter implemented and detailed in futex.c/.h
222 *
223 * The thread_registration_apps and thread_dispatch_ust_registration uses this
224 * queue along with the wait/wake scheme. The thread_manage_apps receives down
225 * the line new application socket and monitors it for any I/O error or clean
226 * close that triggers an unregistration of the application.
227 */
228 static struct ust_cmd_queue ust_cmd_queue;
229
230 /*
231 * Pointer initialized before thread creation.
232 *
233 * This points to the tracing session list containing the session count and a
234 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
235 * MUST NOT be taken if you call a public function in session.c.
236 *
237 * The lock is nested inside the structure: session_list_ptr->lock. Please use
238 * session_lock_list and session_unlock_list for lock acquisition.
239 */
240 static struct ltt_session_list *session_list_ptr;
241
242 int ust_consumerd64_fd = -1;
243 int ust_consumerd32_fd = -1;
244
245 static const char *module_proc_lttng = "/proc/lttng";
246
247 /*
248 * Consumer daemon state which is changed when spawning it, killing it or in
249 * case of a fatal error.
250 */
251 enum consumerd_state {
252 CONSUMER_STARTED = 1,
253 CONSUMER_STOPPED = 2,
254 CONSUMER_ERROR = 3,
255 };
256
257 /*
258 * This consumer daemon state is used to validate if a client command will be
259 * able to reach the consumer. If not, the client is informed. For instance,
260 * doing a "lttng start" when the consumer state is set to ERROR will return an
261 * error to the client.
262 *
263 * The following example shows a possible race condition of this scheme:
264 *
265 * consumer thread error happens
266 * client cmd arrives
267 * client cmd checks state -> still OK
268 * consumer thread exit, sets error
269 * client cmd try to talk to consumer
270 * ...
271 *
272 * However, since the consumer is a different daemon, we have no way of making
273 * sure the command will reach it safely even with this state flag. This is why
274 * we consider that up to the state validation during command processing, the
275 * command is safe. After that, we can not guarantee the correctness of the
276 * client request vis-a-vis the consumer.
277 */
278 static enum consumerd_state ust_consumerd_state;
279 static enum consumerd_state kernel_consumerd_state;
280
281 /* Set in main() with the current page size. */
282 long page_size;
283
284 /* Application health monitoring */
285 struct health_app *health_sessiond;
286
287 /* Am I root or not. */
288 int is_root; /* Set to 1 if the daemon is running as root */
289
290 const char * const config_section_name = "sessiond";
291
292 /* Load session thread information to operate. */
293 struct load_session_thread_data *load_info;
294
295 /* Notification thread handle. */
296 struct notification_thread_handle *notification_thread_handle;
297
298 /* Rotation thread handle. */
299 struct rotation_thread_handle *rotation_thread_handle;
300
301 /* Global hash tables */
302 struct lttng_ht *agent_apps_ht_by_sock = NULL;
303
304 /*
305 * The initialization of the session daemon is done in multiple phases.
306 *
307 * While all threads are launched near-simultaneously, only some of them
308 * are needed to ensure the session daemon can start to respond to client
309 * requests.
310 *
311 * There are two important guarantees that we wish to offer with respect
312 * to the initialisation of the session daemon:
313 * - When the daemonize/background launcher process exits, the sessiond
314 * is fully able to respond to client requests,
315 * - Auto-loaded sessions are visible to clients.
316 *
317 * In order to achieve this, a number of support threads have to be launched
318 * to allow the "client" thread to function properly. Moreover, since the
319 * "load session" thread needs the client thread, we must provide a way
320 * for the "load session" thread to know that the "client" thread is up
321 * and running.
322 *
323 * Hence, the support threads decrement the lttng_sessiond_ready counter
324 * while the "client" threads waits for it to reach 0. Once the "client" thread
325 * unblocks, it posts the message_thread_ready semaphore which allows the
326 * "load session" thread to progress.
327 *
328 * This implies that the "load session" thread is the last to be initialized
329 * and will explicitly call sessiond_signal_parents(), which signals the parents
330 * that the session daemon is fully initialized.
331 *
332 * The four (4) support threads are:
333 * - agent_thread
334 * - notification_thread
335 * - rotation_thread
336 * - health_thread
337 */
338 #define NR_LTTNG_SESSIOND_SUPPORT_THREADS 4
339 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_SUPPORT_THREADS;
340
341 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
342 {
343 return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0;
344 }
345
346 /* Notify parents that we are ready for cmd and health check */
347 LTTNG_HIDDEN
348 void sessiond_signal_parents(void)
349 {
350 /*
351 * Notify parent pid that we are ready to accept command
352 * for client side. This ppid is the one from the
353 * external process that spawned us.
354 */
355 if (config.sig_parent) {
356 kill(ppid, SIGUSR1);
357 }
358
359 /*
360 * Notify the parent of the fork() process that we are
361 * ready.
362 */
363 if (config.daemonize || config.background) {
364 kill(child_ppid, SIGUSR1);
365 }
366 }
367
368 LTTNG_HIDDEN
369 void sessiond_notify_ready(void)
370 {
371 /*
372 * This memory barrier is paired with the one performed by
373 * the client thread after it has seen that 'lttng_sessiond_ready' is 0.
374 *
375 * The purpose of these memory barriers is to ensure that all
376 * initialization operations of the various threads that call this
377 * function to signal that they are ready are commited/published
378 * before the client thread can see the 'lttng_sessiond_ready' counter
379 * reach 0.
380 *
381 * Note that this could be a 'write' memory barrier, but a full barrier
382 * is used in case the code using this utility changes. The performance
383 * implications of this choice are minimal since this is a slow path.
384 */
385 cmm_smp_mb();
386 uatomic_sub(&lttng_sessiond_ready, 1);
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 * Init thread quit pipe.
424 *
425 * Return -1 on error or 0 if all pipes are created.
426 */
427 static int __init_thread_quit_pipe(int *a_pipe)
428 {
429 int ret, i;
430
431 ret = pipe(a_pipe);
432 if (ret < 0) {
433 PERROR("thread quit pipe");
434 goto error;
435 }
436
437 for (i = 0; i < 2; i++) {
438 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
439 if (ret < 0) {
440 PERROR("fcntl");
441 goto error;
442 }
443 }
444
445 error:
446 return ret;
447 }
448
449 static int init_thread_quit_pipe(void)
450 {
451 return __init_thread_quit_pipe(thread_quit_pipe);
452 }
453
454 /*
455 * Stop all threads by closing the thread quit pipe.
456 */
457 static void stop_threads(void)
458 {
459 int ret;
460
461 /* Stopping all threads */
462 DBG("Terminating all threads");
463 ret = notify_thread_pipe(thread_quit_pipe[1]);
464 if (ret < 0) {
465 ERR("write error on thread quit pipe");
466 }
467
468 /* Dispatch thread */
469 CMM_STORE_SHARED(dispatch_thread_exit, 1);
470 futex_nto1_wake(&ust_cmd_queue.futex);
471 }
472
473 /*
474 * Close every consumer sockets.
475 */
476 static void close_consumer_sockets(void)
477 {
478 int ret;
479
480 if (kconsumer_data.err_sock >= 0) {
481 ret = close(kconsumer_data.err_sock);
482 if (ret < 0) {
483 PERROR("kernel consumer err_sock close");
484 }
485 }
486 if (ustconsumer32_data.err_sock >= 0) {
487 ret = close(ustconsumer32_data.err_sock);
488 if (ret < 0) {
489 PERROR("UST consumerd32 err_sock close");
490 }
491 }
492 if (ustconsumer64_data.err_sock >= 0) {
493 ret = close(ustconsumer64_data.err_sock);
494 if (ret < 0) {
495 PERROR("UST consumerd64 err_sock close");
496 }
497 }
498 if (kconsumer_data.cmd_sock >= 0) {
499 ret = close(kconsumer_data.cmd_sock);
500 if (ret < 0) {
501 PERROR("kernel consumer cmd_sock close");
502 }
503 }
504 if (ustconsumer32_data.cmd_sock >= 0) {
505 ret = close(ustconsumer32_data.cmd_sock);
506 if (ret < 0) {
507 PERROR("UST consumerd32 cmd_sock close");
508 }
509 }
510 if (ustconsumer64_data.cmd_sock >= 0) {
511 ret = close(ustconsumer64_data.cmd_sock);
512 if (ret < 0) {
513 PERROR("UST consumerd64 cmd_sock close");
514 }
515 }
516 if (kconsumer_data.channel_monitor_pipe >= 0) {
517 ret = close(kconsumer_data.channel_monitor_pipe);
518 if (ret < 0) {
519 PERROR("kernel consumer channel monitor pipe close");
520 }
521 }
522 if (ustconsumer32_data.channel_monitor_pipe >= 0) {
523 ret = close(ustconsumer32_data.channel_monitor_pipe);
524 if (ret < 0) {
525 PERROR("UST consumerd32 channel monitor pipe close");
526 }
527 }
528 if (ustconsumer64_data.channel_monitor_pipe >= 0) {
529 ret = close(ustconsumer64_data.channel_monitor_pipe);
530 if (ret < 0) {
531 PERROR("UST consumerd64 channel monitor pipe close");
532 }
533 }
534 if (kconsumer_data.channel_rotate_pipe >= 0) {
535 ret = close(kconsumer_data.channel_rotate_pipe);
536 if (ret < 0) {
537 PERROR("kernel consumer channel rotate pipe close");
538 }
539 }
540 if (ustconsumer32_data.channel_rotate_pipe >= 0) {
541 ret = close(ustconsumer32_data.channel_rotate_pipe);
542 if (ret < 0) {
543 PERROR("UST consumerd32 channel rotate pipe close");
544 }
545 }
546 if (ustconsumer64_data.channel_rotate_pipe >= 0) {
547 ret = close(ustconsumer64_data.channel_rotate_pipe);
548 if (ret < 0) {
549 PERROR("UST consumerd64 channel rotate pipe close");
550 }
551 }
552 }
553
554 /*
555 * Wait on consumer process termination.
556 *
557 * Need to be called with the consumer data lock held or from a context
558 * ensuring no concurrent access to data (e.g: cleanup).
559 */
560 static void wait_consumer(struct consumer_data *consumer_data)
561 {
562 pid_t ret;
563 int status;
564
565 if (consumer_data->pid <= 0) {
566 return;
567 }
568
569 DBG("Waiting for complete teardown of consumerd (PID: %d)",
570 consumer_data->pid);
571 ret = waitpid(consumer_data->pid, &status, 0);
572 if (ret == -1) {
573 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
574 } else if (!WIFEXITED(status)) {
575 ERR("consumerd termination with error: %d",
576 WEXITSTATUS(ret));
577 }
578 consumer_data->pid = 0;
579 }
580
581 /*
582 * Cleanup the session daemon's data structures.
583 */
584 static void sessiond_cleanup(void)
585 {
586 int ret;
587 struct ltt_session *sess, *stmp;
588
589 DBG("Cleanup sessiond");
590
591 /*
592 * Close the thread quit pipe. It has already done its job,
593 * since we are now called.
594 */
595 utils_close_pipe(thread_quit_pipe);
596
597 ret = remove(config.pid_file_path.value);
598 if (ret < 0) {
599 PERROR("remove pidfile %s", config.pid_file_path.value);
600 }
601
602 DBG("Removing sessiond and consumerd content of directory %s",
603 config.rundir.value);
604
605 /* sessiond */
606 DBG("Removing %s", config.pid_file_path.value);
607 (void) unlink(config.pid_file_path.value);
608
609 DBG("Removing %s", config.agent_port_file_path.value);
610 (void) unlink(config.agent_port_file_path.value);
611
612 /* kconsumerd */
613 DBG("Removing %s", kconsumer_data.err_unix_sock_path);
614 (void) unlink(kconsumer_data.err_unix_sock_path);
615
616 DBG("Removing directory %s", config.kconsumerd_path.value);
617 (void) rmdir(config.kconsumerd_path.value);
618
619 /* ust consumerd 32 */
620 DBG("Removing %s", config.consumerd32_err_unix_sock_path.value);
621 (void) unlink(config.consumerd32_err_unix_sock_path.value);
622
623 DBG("Removing directory %s", config.consumerd32_path.value);
624 (void) rmdir(config.consumerd32_path.value);
625
626 /* ust consumerd 64 */
627 DBG("Removing %s", config.consumerd64_err_unix_sock_path.value);
628 (void) unlink(config.consumerd64_err_unix_sock_path.value);
629
630 DBG("Removing directory %s", config.consumerd64_path.value);
631 (void) rmdir(config.consumerd64_path.value);
632
633 DBG("Cleaning up all sessions");
634
635 /* Destroy session list mutex */
636 if (session_list_ptr != NULL) {
637 pthread_mutex_destroy(&session_list_ptr->lock);
638
639 /* Cleanup ALL session */
640 cds_list_for_each_entry_safe(sess, stmp,
641 &session_list_ptr->head, list) {
642 cmd_destroy_session(sess, kernel_poll_pipe[1],
643 notification_thread_handle);
644 }
645 }
646
647 wait_consumer(&kconsumer_data);
648 wait_consumer(&ustconsumer64_data);
649 wait_consumer(&ustconsumer32_data);
650
651 DBG("Cleaning up all agent apps");
652 agent_app_ht_clean();
653
654 DBG("Closing all UST sockets");
655 ust_app_clean_list();
656 buffer_reg_destroy_registries();
657
658 if (is_root && !config.no_kernel) {
659 DBG2("Closing kernel fd");
660 if (kernel_tracer_fd >= 0) {
661 ret = close(kernel_tracer_fd);
662 if (ret) {
663 PERROR("close");
664 }
665 }
666 DBG("Unloading kernel modules");
667 modprobe_remove_lttng_all();
668 free(syscall_table);
669 }
670
671 close_consumer_sockets();
672
673 if (load_info) {
674 load_session_destroy_data(load_info);
675 free(load_info);
676 }
677
678 /*
679 * We do NOT rmdir rundir because there are other processes
680 * using it, for instance lttng-relayd, which can start in
681 * parallel with this teardown.
682 */
683 }
684
685 /*
686 * Cleanup the daemon's option data structures.
687 */
688 static void sessiond_cleanup_options(void)
689 {
690 DBG("Cleaning up options");
691
692 sessiond_config_fini(&config);
693
694 run_as_destroy_worker();
695 }
696
697 /*
698 * Send data on a unix socket using the liblttsessiondcomm API.
699 *
700 * Return lttcomm error code.
701 */
702 static int send_unix_sock(int sock, void *buf, size_t len)
703 {
704 /* Check valid length */
705 if (len == 0) {
706 return -1;
707 }
708
709 return lttcomm_send_unix_sock(sock, buf, len);
710 }
711
712 /*
713 * Free memory of a command context structure.
714 */
715 static void clean_command_ctx(struct command_ctx **cmd_ctx)
716 {
717 DBG("Clean command context structure");
718 if (*cmd_ctx) {
719 if ((*cmd_ctx)->llm) {
720 free((*cmd_ctx)->llm);
721 }
722 if ((*cmd_ctx)->lsm) {
723 free((*cmd_ctx)->lsm);
724 }
725 free(*cmd_ctx);
726 *cmd_ctx = NULL;
727 }
728 }
729
730 /*
731 * Notify UST applications using the shm mmap futex.
732 */
733 static int notify_ust_apps(int active)
734 {
735 char *wait_shm_mmap;
736
737 DBG("Notifying applications of session daemon state: %d", active);
738
739 /* See shm.c for this call implying mmap, shm and futex calls */
740 wait_shm_mmap = shm_ust_get_mmap(config.wait_shm_path.value, is_root);
741 if (wait_shm_mmap == NULL) {
742 goto error;
743 }
744
745 /* Wake waiting process */
746 futex_wait_update((int32_t *) wait_shm_mmap, active);
747
748 /* Apps notified successfully */
749 return 0;
750
751 error:
752 return -1;
753 }
754
755 /*
756 * Setup the outgoing data buffer for the response (llm) by allocating the
757 * right amount of memory and copying the original information from the lsm
758 * structure.
759 *
760 * Return 0 on success, negative value on error.
761 */
762 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
763 const void *payload_buf, size_t payload_len,
764 const void *cmd_header_buf, size_t cmd_header_len)
765 {
766 int ret = 0;
767 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
768 const size_t cmd_header_offset = header_len;
769 const size_t payload_offset = cmd_header_offset + cmd_header_len;
770 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
771
772 cmd_ctx->llm = zmalloc(total_msg_size);
773
774 if (cmd_ctx->llm == NULL) {
775 PERROR("zmalloc");
776 ret = -ENOMEM;
777 goto end;
778 }
779
780 /* Copy common data */
781 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
782 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
783 cmd_ctx->llm->cmd_header_size = cmd_header_len;
784 cmd_ctx->llm->data_size = payload_len;
785 cmd_ctx->lttng_msg_size = total_msg_size;
786
787 /* Copy command header */
788 if (cmd_header_len) {
789 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
790 cmd_header_len);
791 }
792
793 /* Copy payload */
794 if (payload_len) {
795 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
796 payload_len);
797 }
798
799 end:
800 return ret;
801 }
802
803 /*
804 * Version of setup_lttng_msg() without command header.
805 */
806 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
807 void *payload_buf, size_t payload_len)
808 {
809 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
810 }
811 /*
812 * Update the kernel poll set of all channel fd available over all tracing
813 * session. Add the wakeup pipe at the end of the set.
814 */
815 static int update_kernel_poll(struct lttng_poll_event *events)
816 {
817 int ret;
818 struct ltt_session *session;
819 struct ltt_kernel_channel *channel;
820
821 DBG("Updating kernel poll set");
822
823 session_lock_list();
824 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
825 session_lock(session);
826 if (session->kernel_session == NULL) {
827 session_unlock(session);
828 continue;
829 }
830
831 cds_list_for_each_entry(channel,
832 &session->kernel_session->channel_list.head, list) {
833 /* Add channel fd to the kernel poll set */
834 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
835 if (ret < 0) {
836 session_unlock(session);
837 goto error;
838 }
839 DBG("Channel fd %d added to kernel set", channel->fd);
840 }
841 session_unlock(session);
842 }
843 session_unlock_list();
844
845 return 0;
846
847 error:
848 session_unlock_list();
849 return -1;
850 }
851
852 /*
853 * Find the channel fd from 'fd' over all tracing session. When found, check
854 * for new channel stream and send those stream fds to the kernel consumer.
855 *
856 * Useful for CPU hotplug feature.
857 */
858 static int update_kernel_stream(int fd)
859 {
860 int ret = 0;
861 struct ltt_session *session;
862 struct ltt_kernel_session *ksess;
863 struct ltt_kernel_channel *channel;
864
865 DBG("Updating kernel streams for channel fd %d", fd);
866
867 session_lock_list();
868 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
869 session_lock(session);
870 if (session->kernel_session == NULL) {
871 session_unlock(session);
872 continue;
873 }
874 ksess = session->kernel_session;
875
876 cds_list_for_each_entry(channel,
877 &ksess->channel_list.head, list) {
878 struct lttng_ht_iter iter;
879 struct consumer_socket *socket;
880
881 if (channel->fd != fd) {
882 continue;
883 }
884 DBG("Channel found, updating kernel streams");
885 ret = kernel_open_channel_stream(channel);
886 if (ret < 0) {
887 goto error;
888 }
889 /* Update the stream global counter */
890 ksess->stream_count_global += ret;
891
892 /*
893 * Have we already sent fds to the consumer? If yes, it
894 * means that tracing is started so it is safe to send
895 * our updated stream fds.
896 */
897 if (ksess->consumer_fds_sent != 1
898 || ksess->consumer == NULL) {
899 ret = -1;
900 goto error;
901 }
902
903 rcu_read_lock();
904 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
905 &iter.iter, socket, node.node) {
906 pthread_mutex_lock(socket->lock);
907 ret = kernel_consumer_send_channel_streams(socket,
908 channel, ksess,
909 session->output_traces ? 1 : 0);
910 pthread_mutex_unlock(socket->lock);
911 if (ret < 0) {
912 rcu_read_unlock();
913 goto error;
914 }
915 }
916 rcu_read_unlock();
917 }
918 session_unlock(session);
919 }
920 session_unlock_list();
921 return ret;
922
923 error:
924 session_unlock(session);
925 session_unlock_list();
926 return ret;
927 }
928
929 /*
930 * For each tracing session, update newly registered apps. The session list
931 * lock MUST be acquired before calling this.
932 */
933 static void update_ust_app(int app_sock)
934 {
935 struct ltt_session *sess, *stmp;
936
937 /* Consumer is in an ERROR state. Stop any application update. */
938 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
939 /* Stop the update process since the consumer is dead. */
940 return;
941 }
942
943 /* For all tracing session(s) */
944 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
945 struct ust_app *app;
946
947 session_lock(sess);
948 if (!sess->ust_session) {
949 goto unlock_session;
950 }
951
952 rcu_read_lock();
953 assert(app_sock >= 0);
954 app = ust_app_find_by_sock(app_sock);
955 if (app == NULL) {
956 /*
957 * Application can be unregistered before so
958 * this is possible hence simply stopping the
959 * update.
960 */
961 DBG3("UST app update failed to find app sock %d",
962 app_sock);
963 goto unlock_rcu;
964 }
965 ust_app_global_update(sess->ust_session, app);
966 unlock_rcu:
967 rcu_read_unlock();
968 unlock_session:
969 session_unlock(sess);
970 }
971 }
972
973 /*
974 * This thread manage event coming from the kernel.
975 *
976 * Features supported in this thread:
977 * -) CPU Hotplug
978 */
979 static void *thread_manage_kernel(void *data)
980 {
981 int ret, i, pollfd, update_poll_flag = 1, err = -1;
982 uint32_t revents, nb_fd;
983 char tmp;
984 struct lttng_poll_event events;
985
986 DBG("[thread] Thread manage kernel started");
987
988 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
989
990 /*
991 * This first step of the while is to clean this structure which could free
992 * non NULL pointers so initialize it before the loop.
993 */
994 lttng_poll_init(&events);
995
996 if (testpoint(sessiond_thread_manage_kernel)) {
997 goto error_testpoint;
998 }
999
1000 health_code_update();
1001
1002 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1003 goto error_testpoint;
1004 }
1005
1006 while (1) {
1007 health_code_update();
1008
1009 if (update_poll_flag == 1) {
1010 /* Clean events object. We are about to populate it again. */
1011 lttng_poll_clean(&events);
1012
1013 ret = sessiond_set_thread_pollset(&events, 2);
1014 if (ret < 0) {
1015 goto error_poll_create;
1016 }
1017
1018 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1019 if (ret < 0) {
1020 goto error;
1021 }
1022
1023 /* This will add the available kernel channel if any. */
1024 ret = update_kernel_poll(&events);
1025 if (ret < 0) {
1026 goto error;
1027 }
1028 update_poll_flag = 0;
1029 }
1030
1031 DBG("Thread kernel polling");
1032
1033 /* Poll infinite value of time */
1034 restart:
1035 health_poll_entry();
1036 ret = lttng_poll_wait(&events, -1);
1037 DBG("Thread kernel return from poll on %d fds",
1038 LTTNG_POLL_GETNB(&events));
1039 health_poll_exit();
1040 if (ret < 0) {
1041 /*
1042 * Restart interrupted system call.
1043 */
1044 if (errno == EINTR) {
1045 goto restart;
1046 }
1047 goto error;
1048 } else if (ret == 0) {
1049 /* Should not happen since timeout is infinite */
1050 ERR("Return value of poll is 0 with an infinite timeout.\n"
1051 "This should not have happened! Continuing...");
1052 continue;
1053 }
1054
1055 nb_fd = ret;
1056
1057 for (i = 0; i < nb_fd; i++) {
1058 /* Fetch once the poll data */
1059 revents = LTTNG_POLL_GETEV(&events, i);
1060 pollfd = LTTNG_POLL_GETFD(&events, i);
1061
1062 health_code_update();
1063
1064 if (!revents) {
1065 /* No activity for this FD (poll implementation). */
1066 continue;
1067 }
1068
1069 /* Thread quit pipe has been closed. Killing thread. */
1070 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1071 if (ret) {
1072 err = 0;
1073 goto exit;
1074 }
1075
1076 /* Check for data on kernel pipe */
1077 if (revents & LPOLLIN) {
1078 if (pollfd == kernel_poll_pipe[0]) {
1079 (void) lttng_read(kernel_poll_pipe[0],
1080 &tmp, 1);
1081 /*
1082 * Ret value is useless here, if this pipe gets any actions an
1083 * update is required anyway.
1084 */
1085 update_poll_flag = 1;
1086 continue;
1087 } else {
1088 /*
1089 * New CPU detected by the kernel. Adding kernel stream to
1090 * kernel session and updating the kernel consumer
1091 */
1092 ret = update_kernel_stream(pollfd);
1093 if (ret < 0) {
1094 continue;
1095 }
1096 break;
1097 }
1098 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1099 update_poll_flag = 1;
1100 continue;
1101 } else {
1102 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1103 goto error;
1104 }
1105 }
1106 }
1107
1108 exit:
1109 error:
1110 lttng_poll_clean(&events);
1111 error_poll_create:
1112 error_testpoint:
1113 utils_close_pipe(kernel_poll_pipe);
1114 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1115 if (err) {
1116 health_error();
1117 ERR("Health error occurred in %s", __func__);
1118 WARN("Kernel thread died unexpectedly. "
1119 "Kernel tracing can continue but CPU hotplug is disabled.");
1120 }
1121 health_unregister(health_sessiond);
1122 DBG("Kernel thread dying");
1123 return NULL;
1124 }
1125
1126 /*
1127 * Signal pthread condition of the consumer data that the thread.
1128 */
1129 static void signal_consumer_condition(struct consumer_data *data, int state)
1130 {
1131 pthread_mutex_lock(&data->cond_mutex);
1132
1133 /*
1134 * The state is set before signaling. It can be any value, it's the waiter
1135 * job to correctly interpret this condition variable associated to the
1136 * consumer pthread_cond.
1137 *
1138 * A value of 0 means that the corresponding thread of the consumer data
1139 * was not started. 1 indicates that the thread has started and is ready
1140 * for action. A negative value means that there was an error during the
1141 * thread bootstrap.
1142 */
1143 data->consumer_thread_is_ready = state;
1144 (void) pthread_cond_signal(&data->cond);
1145
1146 pthread_mutex_unlock(&data->cond_mutex);
1147 }
1148
1149 /*
1150 * This thread manage the consumer error sent back to the session daemon.
1151 */
1152 static void *thread_manage_consumer(void *data)
1153 {
1154 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1155 uint32_t revents, nb_fd;
1156 enum lttcomm_return_code code;
1157 struct lttng_poll_event events;
1158 struct consumer_data *consumer_data = data;
1159 struct consumer_socket *cmd_socket_wrapper = NULL;
1160
1161 DBG("[thread] Manage consumer started");
1162
1163 rcu_register_thread();
1164 rcu_thread_online();
1165
1166 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1167
1168 health_code_update();
1169
1170 /*
1171 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1172 * metadata_sock. Nothing more will be added to this poll set.
1173 */
1174 ret = sessiond_set_thread_pollset(&events, 3);
1175 if (ret < 0) {
1176 goto error_poll;
1177 }
1178
1179 /*
1180 * The error socket here is already in a listening state which was done
1181 * just before spawning this thread to avoid a race between the consumer
1182 * daemon exec trying to connect and the listen() call.
1183 */
1184 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1185 if (ret < 0) {
1186 goto error;
1187 }
1188
1189 health_code_update();
1190
1191 /* Infinite blocking call, waiting for transmission */
1192 restart:
1193 health_poll_entry();
1194
1195 if (testpoint(sessiond_thread_manage_consumer)) {
1196 goto error;
1197 }
1198
1199 ret = lttng_poll_wait(&events, -1);
1200 health_poll_exit();
1201 if (ret < 0) {
1202 /*
1203 * Restart interrupted system call.
1204 */
1205 if (errno == EINTR) {
1206 goto restart;
1207 }
1208 goto error;
1209 }
1210
1211 nb_fd = ret;
1212
1213 for (i = 0; i < nb_fd; i++) {
1214 /* Fetch once the poll data */
1215 revents = LTTNG_POLL_GETEV(&events, i);
1216 pollfd = LTTNG_POLL_GETFD(&events, i);
1217
1218 health_code_update();
1219
1220 if (!revents) {
1221 /* No activity for this FD (poll implementation). */
1222 continue;
1223 }
1224
1225 /* Thread quit pipe has been closed. Killing thread. */
1226 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1227 if (ret) {
1228 err = 0;
1229 goto exit;
1230 }
1231
1232 /* Event on the registration socket */
1233 if (pollfd == consumer_data->err_sock) {
1234 if (revents & LPOLLIN) {
1235 continue;
1236 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1237 ERR("consumer err socket poll error");
1238 goto error;
1239 } else {
1240 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1241 goto error;
1242 }
1243 }
1244 }
1245
1246 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1247 if (sock < 0) {
1248 goto error;
1249 }
1250
1251 /*
1252 * Set the CLOEXEC flag. Return code is useless because either way, the
1253 * show must go on.
1254 */
1255 (void) utils_set_fd_cloexec(sock);
1256
1257 health_code_update();
1258
1259 DBG2("Receiving code from consumer err_sock");
1260
1261 /* Getting status code from kconsumerd */
1262 ret = lttcomm_recv_unix_sock(sock, &code,
1263 sizeof(enum lttcomm_return_code));
1264 if (ret <= 0) {
1265 goto error;
1266 }
1267
1268 health_code_update();
1269 if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1270 ERR("consumer error when waiting for SOCK_READY : %s",
1271 lttcomm_get_readable_code(-code));
1272 goto error;
1273 }
1274
1275 /* Connect both command and metadata sockets. */
1276 consumer_data->cmd_sock =
1277 lttcomm_connect_unix_sock(
1278 consumer_data->cmd_unix_sock_path);
1279 consumer_data->metadata_fd =
1280 lttcomm_connect_unix_sock(
1281 consumer_data->cmd_unix_sock_path);
1282 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1283 PERROR("consumer connect cmd socket");
1284 /* On error, signal condition and quit. */
1285 signal_consumer_condition(consumer_data, -1);
1286 goto error;
1287 }
1288
1289 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1290
1291 /* Create metadata socket lock. */
1292 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1293 if (consumer_data->metadata_sock.lock == NULL) {
1294 PERROR("zmalloc pthread mutex");
1295 goto error;
1296 }
1297 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1298
1299 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1300 DBG("Consumer metadata socket ready (fd: %d)",
1301 consumer_data->metadata_fd);
1302
1303 /*
1304 * Remove the consumerd error sock since we've established a connection.
1305 */
1306 ret = lttng_poll_del(&events, consumer_data->err_sock);
1307 if (ret < 0) {
1308 goto error;
1309 }
1310
1311 /* Add new accepted error socket. */
1312 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1313 if (ret < 0) {
1314 goto error;
1315 }
1316
1317 /* Add metadata socket that is successfully connected. */
1318 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1319 LPOLLIN | LPOLLRDHUP);
1320 if (ret < 0) {
1321 goto error;
1322 }
1323
1324 health_code_update();
1325
1326 /*
1327 * Transfer the write-end of the channel monitoring and rotate pipe
1328 * to the consumer by issuing a SET_CHANNEL_MONITOR_PIPE and
1329 * SET_CHANNEL_ROTATE_PIPE commands.
1330 */
1331 cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock);
1332 if (!cmd_socket_wrapper) {
1333 goto error;
1334 }
1335 cmd_socket_wrapper->lock = &consumer_data->lock;
1336
1337 ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper,
1338 consumer_data->channel_monitor_pipe);
1339 if (ret) {
1340 goto error;
1341 }
1342
1343 ret = consumer_send_channel_rotate_pipe(cmd_socket_wrapper,
1344 consumer_data->channel_rotate_pipe);
1345 if (ret) {
1346 goto error;
1347 }
1348
1349 /* Discard the socket wrapper as it is no longer needed. */
1350 consumer_destroy_socket(cmd_socket_wrapper);
1351 cmd_socket_wrapper = NULL;
1352
1353 /* The thread is completely initialized, signal that it is ready. */
1354 signal_consumer_condition(consumer_data, 1);
1355
1356 /* Infinite blocking call, waiting for transmission */
1357 restart_poll:
1358 while (1) {
1359 health_code_update();
1360
1361 /* Exit the thread because the thread quit pipe has been triggered. */
1362 if (should_quit) {
1363 /* Not a health error. */
1364 err = 0;
1365 goto exit;
1366 }
1367
1368 health_poll_entry();
1369 ret = lttng_poll_wait(&events, -1);
1370 health_poll_exit();
1371 if (ret < 0) {
1372 /*
1373 * Restart interrupted system call.
1374 */
1375 if (errno == EINTR) {
1376 goto restart_poll;
1377 }
1378 goto error;
1379 }
1380
1381 nb_fd = ret;
1382
1383 for (i = 0; i < nb_fd; i++) {
1384 /* Fetch once the poll data */
1385 revents = LTTNG_POLL_GETEV(&events, i);
1386 pollfd = LTTNG_POLL_GETFD(&events, i);
1387
1388 health_code_update();
1389
1390 if (!revents) {
1391 /* No activity for this FD (poll implementation). */
1392 continue;
1393 }
1394
1395 /*
1396 * Thread quit pipe has been triggered, flag that we should stop
1397 * but continue the current loop to handle potential data from
1398 * consumer.
1399 */
1400 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1401
1402 if (pollfd == sock) {
1403 /* Event on the consumerd socket */
1404 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1405 && !(revents & LPOLLIN)) {
1406 ERR("consumer err socket second poll error");
1407 goto error;
1408 }
1409 health_code_update();
1410 /* Wait for any kconsumerd error */
1411 ret = lttcomm_recv_unix_sock(sock, &code,
1412 sizeof(enum lttcomm_return_code));
1413 if (ret <= 0) {
1414 ERR("consumer closed the command socket");
1415 goto error;
1416 }
1417
1418 ERR("consumer return code : %s",
1419 lttcomm_get_readable_code(-code));
1420
1421 goto exit;
1422 } else if (pollfd == consumer_data->metadata_fd) {
1423 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1424 && !(revents & LPOLLIN)) {
1425 ERR("consumer err metadata socket second poll error");
1426 goto error;
1427 }
1428 /* UST metadata requests */
1429 ret = ust_consumer_metadata_request(
1430 &consumer_data->metadata_sock);
1431 if (ret < 0) {
1432 ERR("Handling metadata request");
1433 goto error;
1434 }
1435 }
1436 /* No need for an else branch all FDs are tested prior. */
1437 }
1438 health_code_update();
1439 }
1440
1441 exit:
1442 error:
1443 /*
1444 * We lock here because we are about to close the sockets and some other
1445 * thread might be using them so get exclusive access which will abort all
1446 * other consumer command by other threads.
1447 */
1448 pthread_mutex_lock(&consumer_data->lock);
1449
1450 /* Immediately set the consumerd state to stopped */
1451 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1452 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1453 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1454 consumer_data->type == LTTNG_CONSUMER32_UST) {
1455 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1456 } else {
1457 /* Code flow error... */
1458 assert(0);
1459 }
1460
1461 if (consumer_data->err_sock >= 0) {
1462 ret = close(consumer_data->err_sock);
1463 if (ret) {
1464 PERROR("close");
1465 }
1466 consumer_data->err_sock = -1;
1467 }
1468 if (consumer_data->cmd_sock >= 0) {
1469 ret = close(consumer_data->cmd_sock);
1470 if (ret) {
1471 PERROR("close");
1472 }
1473 consumer_data->cmd_sock = -1;
1474 }
1475 if (consumer_data->metadata_sock.fd_ptr &&
1476 *consumer_data->metadata_sock.fd_ptr >= 0) {
1477 ret = close(*consumer_data->metadata_sock.fd_ptr);
1478 if (ret) {
1479 PERROR("close");
1480 }
1481 }
1482 if (sock >= 0) {
1483 ret = close(sock);
1484 if (ret) {
1485 PERROR("close");
1486 }
1487 }
1488
1489 unlink(consumer_data->err_unix_sock_path);
1490 unlink(consumer_data->cmd_unix_sock_path);
1491 pthread_mutex_unlock(&consumer_data->lock);
1492
1493 /* Cleanup metadata socket mutex. */
1494 if (consumer_data->metadata_sock.lock) {
1495 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1496 free(consumer_data->metadata_sock.lock);
1497 }
1498 lttng_poll_clean(&events);
1499
1500 if (cmd_socket_wrapper) {
1501 consumer_destroy_socket(cmd_socket_wrapper);
1502 }
1503 error_poll:
1504 if (err) {
1505 health_error();
1506 ERR("Health error occurred in %s", __func__);
1507 }
1508 health_unregister(health_sessiond);
1509 DBG("consumer thread cleanup completed");
1510
1511 rcu_thread_offline();
1512 rcu_unregister_thread();
1513
1514 return NULL;
1515 }
1516
1517 /*
1518 * This thread receives application command sockets (FDs) on the
1519 * apps_cmd_pipe and waits (polls) on them until they are closed
1520 * or an error occurs.
1521 *
1522 * At that point, it flushes the data (tracing and metadata) associated
1523 * with this application and tears down ust app sessions and other
1524 * associated data structures through ust_app_unregister().
1525 *
1526 * Note that this thread never sends commands to the applications
1527 * through the command sockets; it merely listens for hang-ups
1528 * and errors on those sockets and cleans-up as they occur.
1529 */
1530 static void *thread_manage_apps(void *data)
1531 {
1532 int i, ret, pollfd, err = -1;
1533 ssize_t size_ret;
1534 uint32_t revents, nb_fd;
1535 struct lttng_poll_event events;
1536
1537 DBG("[thread] Manage application started");
1538
1539 rcu_register_thread();
1540 rcu_thread_online();
1541
1542 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1543
1544 if (testpoint(sessiond_thread_manage_apps)) {
1545 goto error_testpoint;
1546 }
1547
1548 health_code_update();
1549
1550 ret = sessiond_set_thread_pollset(&events, 2);
1551 if (ret < 0) {
1552 goto error_poll_create;
1553 }
1554
1555 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1556 if (ret < 0) {
1557 goto error;
1558 }
1559
1560 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1561 goto error;
1562 }
1563
1564 health_code_update();
1565
1566 while (1) {
1567 DBG("Apps thread polling");
1568
1569 /* Inifinite blocking call, waiting for transmission */
1570 restart:
1571 health_poll_entry();
1572 ret = lttng_poll_wait(&events, -1);
1573 DBG("Apps thread return from poll on %d fds",
1574 LTTNG_POLL_GETNB(&events));
1575 health_poll_exit();
1576 if (ret < 0) {
1577 /*
1578 * Restart interrupted system call.
1579 */
1580 if (errno == EINTR) {
1581 goto restart;
1582 }
1583 goto error;
1584 }
1585
1586 nb_fd = ret;
1587
1588 for (i = 0; i < nb_fd; i++) {
1589 /* Fetch once the poll data */
1590 revents = LTTNG_POLL_GETEV(&events, i);
1591 pollfd = LTTNG_POLL_GETFD(&events, i);
1592
1593 health_code_update();
1594
1595 if (!revents) {
1596 /* No activity for this FD (poll implementation). */
1597 continue;
1598 }
1599
1600 /* Thread quit pipe has been closed. Killing thread. */
1601 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1602 if (ret) {
1603 err = 0;
1604 goto exit;
1605 }
1606
1607 /* Inspect the apps cmd pipe */
1608 if (pollfd == apps_cmd_pipe[0]) {
1609 if (revents & LPOLLIN) {
1610 int sock;
1611
1612 /* Empty pipe */
1613 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1614 if (size_ret < sizeof(sock)) {
1615 PERROR("read apps cmd pipe");
1616 goto error;
1617 }
1618
1619 health_code_update();
1620
1621 /*
1622 * Since this is a command socket (write then read),
1623 * we only monitor the error events of the socket.
1624 */
1625 ret = lttng_poll_add(&events, sock,
1626 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1627 if (ret < 0) {
1628 goto error;
1629 }
1630
1631 DBG("Apps with sock %d added to poll set", sock);
1632 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1633 ERR("Apps command pipe error");
1634 goto error;
1635 } else {
1636 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1637 goto error;
1638 }
1639 } else {
1640 /*
1641 * At this point, we know that a registered application made
1642 * the event at poll_wait.
1643 */
1644 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1645 /* Removing from the poll set */
1646 ret = lttng_poll_del(&events, pollfd);
1647 if (ret < 0) {
1648 goto error;
1649 }
1650
1651 /* Socket closed on remote end. */
1652 ust_app_unregister(pollfd);
1653 } else {
1654 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1655 goto error;
1656 }
1657 }
1658
1659 health_code_update();
1660 }
1661 }
1662
1663 exit:
1664 error:
1665 lttng_poll_clean(&events);
1666 error_poll_create:
1667 error_testpoint:
1668 utils_close_pipe(apps_cmd_pipe);
1669 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1670
1671 /*
1672 * We don't clean the UST app hash table here since already registered
1673 * applications can still be controlled so let them be until the session
1674 * daemon dies or the applications stop.
1675 */
1676
1677 if (err) {
1678 health_error();
1679 ERR("Health error occurred in %s", __func__);
1680 }
1681 health_unregister(health_sessiond);
1682 DBG("Application communication apps thread cleanup complete");
1683 rcu_thread_offline();
1684 rcu_unregister_thread();
1685 return NULL;
1686 }
1687
1688 /*
1689 * Send a socket to a thread This is called from the dispatch UST registration
1690 * thread once all sockets are set for the application.
1691 *
1692 * The sock value can be invalid, we don't really care, the thread will handle
1693 * it and make the necessary cleanup if so.
1694 *
1695 * On success, return 0 else a negative value being the errno message of the
1696 * write().
1697 */
1698 static int send_socket_to_thread(int fd, int sock)
1699 {
1700 ssize_t ret;
1701
1702 /*
1703 * It's possible that the FD is set as invalid with -1 concurrently just
1704 * before calling this function being a shutdown state of the thread.
1705 */
1706 if (fd < 0) {
1707 ret = -EBADF;
1708 goto error;
1709 }
1710
1711 ret = lttng_write(fd, &sock, sizeof(sock));
1712 if (ret < sizeof(sock)) {
1713 PERROR("write apps pipe %d", fd);
1714 if (ret < 0) {
1715 ret = -errno;
1716 }
1717 goto error;
1718 }
1719
1720 /* All good. Don't send back the write positive ret value. */
1721 ret = 0;
1722 error:
1723 return (int) ret;
1724 }
1725
1726 /*
1727 * Sanitize the wait queue of the dispatch registration thread meaning removing
1728 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1729 * notify socket is never received.
1730 */
1731 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1732 {
1733 int ret, nb_fd = 0, i;
1734 unsigned int fd_added = 0;
1735 struct lttng_poll_event events;
1736 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1737
1738 assert(wait_queue);
1739
1740 lttng_poll_init(&events);
1741
1742 /* Just skip everything for an empty queue. */
1743 if (!wait_queue->count) {
1744 goto end;
1745 }
1746
1747 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1748 if (ret < 0) {
1749 goto error_create;
1750 }
1751
1752 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1753 &wait_queue->head, head) {
1754 assert(wait_node->app);
1755 ret = lttng_poll_add(&events, wait_node->app->sock,
1756 LPOLLHUP | LPOLLERR);
1757 if (ret < 0) {
1758 goto error;
1759 }
1760
1761 fd_added = 1;
1762 }
1763
1764 if (!fd_added) {
1765 goto end;
1766 }
1767
1768 /*
1769 * Poll but don't block so we can quickly identify the faulty events and
1770 * clean them afterwards from the wait queue.
1771 */
1772 ret = lttng_poll_wait(&events, 0);
1773 if (ret < 0) {
1774 goto error;
1775 }
1776 nb_fd = ret;
1777
1778 for (i = 0; i < nb_fd; i++) {
1779 /* Get faulty FD. */
1780 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1781 int pollfd = LTTNG_POLL_GETFD(&events, i);
1782
1783 if (!revents) {
1784 /* No activity for this FD (poll implementation). */
1785 continue;
1786 }
1787
1788 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1789 &wait_queue->head, head) {
1790 if (pollfd == wait_node->app->sock &&
1791 (revents & (LPOLLHUP | LPOLLERR))) {
1792 cds_list_del(&wait_node->head);
1793 wait_queue->count--;
1794 ust_app_destroy(wait_node->app);
1795 free(wait_node);
1796 /*
1797 * Silence warning of use-after-free in
1798 * cds_list_for_each_entry_safe which uses
1799 * __typeof__(*wait_node).
1800 */
1801 wait_node = NULL;
1802 break;
1803 } else {
1804 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1805 goto error;
1806 }
1807 }
1808 }
1809
1810 if (nb_fd > 0) {
1811 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1812 }
1813
1814 end:
1815 lttng_poll_clean(&events);
1816 return;
1817
1818 error:
1819 lttng_poll_clean(&events);
1820 error_create:
1821 ERR("Unable to sanitize wait queue");
1822 return;
1823 }
1824
1825 /*
1826 * Dispatch request from the registration threads to the application
1827 * communication thread.
1828 */
1829 static void *thread_dispatch_ust_registration(void *data)
1830 {
1831 int ret, err = -1;
1832 struct cds_wfcq_node *node;
1833 struct ust_command *ust_cmd = NULL;
1834 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1835 struct ust_reg_wait_queue wait_queue = {
1836 .count = 0,
1837 };
1838
1839 rcu_register_thread();
1840
1841 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1842
1843 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1844 goto error_testpoint;
1845 }
1846
1847 health_code_update();
1848
1849 CDS_INIT_LIST_HEAD(&wait_queue.head);
1850
1851 DBG("[thread] Dispatch UST command started");
1852
1853 for (;;) {
1854 health_code_update();
1855
1856 /* Atomically prepare the queue futex */
1857 futex_nto1_prepare(&ust_cmd_queue.futex);
1858
1859 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1860 break;
1861 }
1862
1863 do {
1864 struct ust_app *app = NULL;
1865 ust_cmd = NULL;
1866
1867 /*
1868 * Make sure we don't have node(s) that have hung up before receiving
1869 * the notify socket. This is to clean the list in order to avoid
1870 * memory leaks from notify socket that are never seen.
1871 */
1872 sanitize_wait_queue(&wait_queue);
1873
1874 health_code_update();
1875 /* Dequeue command for registration */
1876 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1877 if (node == NULL) {
1878 DBG("Woken up but nothing in the UST command queue");
1879 /* Continue thread execution */
1880 break;
1881 }
1882
1883 ust_cmd = caa_container_of(node, struct ust_command, node);
1884
1885 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1886 " gid:%d sock:%d name:%s (version %d.%d)",
1887 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1888 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1889 ust_cmd->sock, ust_cmd->reg_msg.name,
1890 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1891
1892 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1893 wait_node = zmalloc(sizeof(*wait_node));
1894 if (!wait_node) {
1895 PERROR("zmalloc wait_node dispatch");
1896 ret = close(ust_cmd->sock);
1897 if (ret < 0) {
1898 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1899 }
1900 lttng_fd_put(LTTNG_FD_APPS, 1);
1901 free(ust_cmd);
1902 goto error;
1903 }
1904 CDS_INIT_LIST_HEAD(&wait_node->head);
1905
1906 /* Create application object if socket is CMD. */
1907 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1908 ust_cmd->sock);
1909 if (!wait_node->app) {
1910 ret = close(ust_cmd->sock);
1911 if (ret < 0) {
1912 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1913 }
1914 lttng_fd_put(LTTNG_FD_APPS, 1);
1915 free(wait_node);
1916 free(ust_cmd);
1917 continue;
1918 }
1919 /*
1920 * Add application to the wait queue so we can set the notify
1921 * socket before putting this object in the global ht.
1922 */
1923 cds_list_add(&wait_node->head, &wait_queue.head);
1924 wait_queue.count++;
1925
1926 free(ust_cmd);
1927 /*
1928 * We have to continue here since we don't have the notify
1929 * socket and the application MUST be added to the hash table
1930 * only at that moment.
1931 */
1932 continue;
1933 } else {
1934 /*
1935 * Look for the application in the local wait queue and set the
1936 * notify socket if found.
1937 */
1938 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1939 &wait_queue.head, head) {
1940 health_code_update();
1941 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1942 wait_node->app->notify_sock = ust_cmd->sock;
1943 cds_list_del(&wait_node->head);
1944 wait_queue.count--;
1945 app = wait_node->app;
1946 free(wait_node);
1947 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1948 break;
1949 }
1950 }
1951
1952 /*
1953 * With no application at this stage the received socket is
1954 * basically useless so close it before we free the cmd data
1955 * structure for good.
1956 */
1957 if (!app) {
1958 ret = close(ust_cmd->sock);
1959 if (ret < 0) {
1960 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1961 }
1962 lttng_fd_put(LTTNG_FD_APPS, 1);
1963 }
1964 free(ust_cmd);
1965 }
1966
1967 if (app) {
1968 /*
1969 * @session_lock_list
1970 *
1971 * Lock the global session list so from the register up to the
1972 * registration done message, no thread can see the application
1973 * and change its state.
1974 */
1975 session_lock_list();
1976 rcu_read_lock();
1977
1978 /*
1979 * Add application to the global hash table. This needs to be
1980 * done before the update to the UST registry can locate the
1981 * application.
1982 */
1983 ust_app_add(app);
1984
1985 /* Set app version. This call will print an error if needed. */
1986 (void) ust_app_version(app);
1987
1988 /* Send notify socket through the notify pipe. */
1989 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1990 app->notify_sock);
1991 if (ret < 0) {
1992 rcu_read_unlock();
1993 session_unlock_list();
1994 /*
1995 * No notify thread, stop the UST tracing. However, this is
1996 * not an internal error of the this thread thus setting
1997 * the health error code to a normal exit.
1998 */
1999 err = 0;
2000 goto error;
2001 }
2002
2003 /*
2004 * Update newly registered application with the tracing
2005 * registry info already enabled information.
2006 */
2007 update_ust_app(app->sock);
2008
2009 /*
2010 * Don't care about return value. Let the manage apps threads
2011 * handle app unregistration upon socket close.
2012 */
2013 (void) ust_app_register_done(app);
2014
2015 /*
2016 * Even if the application socket has been closed, send the app
2017 * to the thread and unregistration will take place at that
2018 * place.
2019 */
2020 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2021 if (ret < 0) {
2022 rcu_read_unlock();
2023 session_unlock_list();
2024 /*
2025 * No apps. thread, stop the UST tracing. However, this is
2026 * not an internal error of the this thread thus setting
2027 * the health error code to a normal exit.
2028 */
2029 err = 0;
2030 goto error;
2031 }
2032
2033 rcu_read_unlock();
2034 session_unlock_list();
2035 }
2036 } while (node != NULL);
2037
2038 health_poll_entry();
2039 /* Futex wait on queue. Blocking call on futex() */
2040 futex_nto1_wait(&ust_cmd_queue.futex);
2041 health_poll_exit();
2042 }
2043 /* Normal exit, no error */
2044 err = 0;
2045
2046 error:
2047 /* Clean up wait queue. */
2048 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2049 &wait_queue.head, head) {
2050 cds_list_del(&wait_node->head);
2051 wait_queue.count--;
2052 free(wait_node);
2053 }
2054
2055 /* Empty command queue. */
2056 for (;;) {
2057 /* Dequeue command for registration */
2058 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2059 if (node == NULL) {
2060 break;
2061 }
2062 ust_cmd = caa_container_of(node, struct ust_command, node);
2063 ret = close(ust_cmd->sock);
2064 if (ret < 0) {
2065 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2066 }
2067 lttng_fd_put(LTTNG_FD_APPS, 1);
2068 free(ust_cmd);
2069 }
2070
2071 error_testpoint:
2072 DBG("Dispatch thread dying");
2073 if (err) {
2074 health_error();
2075 ERR("Health error occurred in %s", __func__);
2076 }
2077 health_unregister(health_sessiond);
2078 rcu_unregister_thread();
2079 return NULL;
2080 }
2081
2082 /*
2083 * This thread manage application registration.
2084 */
2085 static void *thread_registration_apps(void *data)
2086 {
2087 int sock = -1, i, ret, pollfd, err = -1;
2088 uint32_t revents, nb_fd;
2089 struct lttng_poll_event events;
2090 /*
2091 * Get allocated in this thread, enqueued to a global queue, dequeued and
2092 * freed in the manage apps thread.
2093 */
2094 struct ust_command *ust_cmd = NULL;
2095
2096 DBG("[thread] Manage application registration started");
2097
2098 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2099
2100 if (testpoint(sessiond_thread_registration_apps)) {
2101 goto error_testpoint;
2102 }
2103
2104 ret = lttcomm_listen_unix_sock(apps_sock);
2105 if (ret < 0) {
2106 goto error_listen;
2107 }
2108
2109 /*
2110 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2111 * more will be added to this poll set.
2112 */
2113 ret = sessiond_set_thread_pollset(&events, 2);
2114 if (ret < 0) {
2115 goto error_create_poll;
2116 }
2117
2118 /* Add the application registration socket */
2119 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2120 if (ret < 0) {
2121 goto error_poll_add;
2122 }
2123
2124 /* Notify all applications to register */
2125 ret = notify_ust_apps(1);
2126 if (ret < 0) {
2127 ERR("Failed to notify applications or create the wait shared memory.\n"
2128 "Execution continues but there might be problem for already\n"
2129 "running applications that wishes to register.");
2130 }
2131
2132 while (1) {
2133 DBG("Accepting application registration");
2134
2135 /* Inifinite blocking call, waiting for transmission */
2136 restart:
2137 health_poll_entry();
2138 ret = lttng_poll_wait(&events, -1);
2139 health_poll_exit();
2140 if (ret < 0) {
2141 /*
2142 * Restart interrupted system call.
2143 */
2144 if (errno == EINTR) {
2145 goto restart;
2146 }
2147 goto error;
2148 }
2149
2150 nb_fd = ret;
2151
2152 for (i = 0; i < nb_fd; i++) {
2153 health_code_update();
2154
2155 /* Fetch once the poll data */
2156 revents = LTTNG_POLL_GETEV(&events, i);
2157 pollfd = LTTNG_POLL_GETFD(&events, i);
2158
2159 if (!revents) {
2160 /* No activity for this FD (poll implementation). */
2161 continue;
2162 }
2163
2164 /* Thread quit pipe has been closed. Killing thread. */
2165 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2166 if (ret) {
2167 err = 0;
2168 goto exit;
2169 }
2170
2171 /* Event on the registration socket */
2172 if (pollfd == apps_sock) {
2173 if (revents & LPOLLIN) {
2174 sock = lttcomm_accept_unix_sock(apps_sock);
2175 if (sock < 0) {
2176 goto error;
2177 }
2178
2179 /*
2180 * Set socket timeout for both receiving and ending.
2181 * app_socket_timeout is in seconds, whereas
2182 * lttcomm_setsockopt_rcv_timeout and
2183 * lttcomm_setsockopt_snd_timeout expect msec as
2184 * parameter.
2185 */
2186 if (config.app_socket_timeout >= 0) {
2187 (void) lttcomm_setsockopt_rcv_timeout(sock,
2188 config.app_socket_timeout * 1000);
2189 (void) lttcomm_setsockopt_snd_timeout(sock,
2190 config.app_socket_timeout * 1000);
2191 }
2192
2193 /*
2194 * Set the CLOEXEC flag. Return code is useless because
2195 * either way, the show must go on.
2196 */
2197 (void) utils_set_fd_cloexec(sock);
2198
2199 /* Create UST registration command for enqueuing */
2200 ust_cmd = zmalloc(sizeof(struct ust_command));
2201 if (ust_cmd == NULL) {
2202 PERROR("ust command zmalloc");
2203 ret = close(sock);
2204 if (ret) {
2205 PERROR("close");
2206 }
2207 goto error;
2208 }
2209
2210 /*
2211 * Using message-based transmissions to ensure we don't
2212 * have to deal with partially received messages.
2213 */
2214 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2215 if (ret < 0) {
2216 ERR("Exhausted file descriptors allowed for applications.");
2217 free(ust_cmd);
2218 ret = close(sock);
2219 if (ret) {
2220 PERROR("close");
2221 }
2222 sock = -1;
2223 continue;
2224 }
2225
2226 health_code_update();
2227 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2228 if (ret < 0) {
2229 free(ust_cmd);
2230 /* Close socket of the application. */
2231 ret = close(sock);
2232 if (ret) {
2233 PERROR("close");
2234 }
2235 lttng_fd_put(LTTNG_FD_APPS, 1);
2236 sock = -1;
2237 continue;
2238 }
2239 health_code_update();
2240
2241 ust_cmd->sock = sock;
2242 sock = -1;
2243
2244 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2245 " gid:%d sock:%d name:%s (version %d.%d)",
2246 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2247 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2248 ust_cmd->sock, ust_cmd->reg_msg.name,
2249 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2250
2251 /*
2252 * Lock free enqueue the registration request. The red pill
2253 * has been taken! This apps will be part of the *system*.
2254 */
2255 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2256
2257 /*
2258 * Wake the registration queue futex. Implicit memory
2259 * barrier with the exchange in cds_wfcq_enqueue.
2260 */
2261 futex_nto1_wake(&ust_cmd_queue.futex);
2262 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2263 ERR("Register apps socket poll error");
2264 goto error;
2265 } else {
2266 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2267 goto error;
2268 }
2269 }
2270 }
2271 }
2272
2273 exit:
2274 error:
2275 /* Notify that the registration thread is gone */
2276 notify_ust_apps(0);
2277
2278 if (apps_sock >= 0) {
2279 ret = close(apps_sock);
2280 if (ret) {
2281 PERROR("close");
2282 }
2283 }
2284 if (sock >= 0) {
2285 ret = close(sock);
2286 if (ret) {
2287 PERROR("close");
2288 }
2289 lttng_fd_put(LTTNG_FD_APPS, 1);
2290 }
2291 unlink(config.apps_unix_sock_path.value);
2292
2293 error_poll_add:
2294 lttng_poll_clean(&events);
2295 error_listen:
2296 error_create_poll:
2297 error_testpoint:
2298 DBG("UST Registration thread cleanup complete");
2299 if (err) {
2300 health_error();
2301 ERR("Health error occurred in %s", __func__);
2302 }
2303 health_unregister(health_sessiond);
2304
2305 return NULL;
2306 }
2307
2308 /*
2309 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2310 * exec or it will fails.
2311 */
2312 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2313 {
2314 int ret, clock_ret;
2315 struct timespec timeout;
2316
2317 /*
2318 * Make sure we set the readiness flag to 0 because we are NOT ready.
2319 * This access to consumer_thread_is_ready does not need to be
2320 * protected by consumer_data.cond_mutex (yet) since the consumer
2321 * management thread has not been started at this point.
2322 */
2323 consumer_data->consumer_thread_is_ready = 0;
2324
2325 /* Setup pthread condition */
2326 ret = pthread_condattr_init(&consumer_data->condattr);
2327 if (ret) {
2328 errno = ret;
2329 PERROR("pthread_condattr_init consumer data");
2330 goto error;
2331 }
2332
2333 /*
2334 * Set the monotonic clock in order to make sure we DO NOT jump in time
2335 * between the clock_gettime() call and the timedwait call. See bug #324
2336 * for a more details and how we noticed it.
2337 */
2338 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2339 if (ret) {
2340 errno = ret;
2341 PERROR("pthread_condattr_setclock consumer data");
2342 goto error;
2343 }
2344
2345 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2346 if (ret) {
2347 errno = ret;
2348 PERROR("pthread_cond_init consumer data");
2349 goto error;
2350 }
2351
2352 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2353 thread_manage_consumer, consumer_data);
2354 if (ret) {
2355 errno = ret;
2356 PERROR("pthread_create consumer");
2357 ret = -1;
2358 goto error;
2359 }
2360
2361 /* We are about to wait on a pthread condition */
2362 pthread_mutex_lock(&consumer_data->cond_mutex);
2363
2364 /* Get time for sem_timedwait absolute timeout */
2365 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2366 /*
2367 * Set the timeout for the condition timed wait even if the clock gettime
2368 * call fails since we might loop on that call and we want to avoid to
2369 * increment the timeout too many times.
2370 */
2371 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2372
2373 /*
2374 * The following loop COULD be skipped in some conditions so this is why we
2375 * set ret to 0 in order to make sure at least one round of the loop is
2376 * done.
2377 */
2378 ret = 0;
2379
2380 /*
2381 * Loop until the condition is reached or when a timeout is reached. Note
2382 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2383 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2384 * possible. This loop does not take any chances and works with both of
2385 * them.
2386 */
2387 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2388 if (clock_ret < 0) {
2389 PERROR("clock_gettime spawn consumer");
2390 /* Infinite wait for the consumerd thread to be ready */
2391 ret = pthread_cond_wait(&consumer_data->cond,
2392 &consumer_data->cond_mutex);
2393 } else {
2394 ret = pthread_cond_timedwait(&consumer_data->cond,
2395 &consumer_data->cond_mutex, &timeout);
2396 }
2397 }
2398
2399 /* Release the pthread condition */
2400 pthread_mutex_unlock(&consumer_data->cond_mutex);
2401
2402 if (ret != 0) {
2403 errno = ret;
2404 if (ret == ETIMEDOUT) {
2405 int pth_ret;
2406
2407 /*
2408 * Call has timed out so we kill the kconsumerd_thread and return
2409 * an error.
2410 */
2411 ERR("Condition timed out. The consumer thread was never ready."
2412 " Killing it");
2413 pth_ret = pthread_cancel(consumer_data->thread);
2414 if (pth_ret < 0) {
2415 PERROR("pthread_cancel consumer thread");
2416 }
2417 } else {
2418 PERROR("pthread_cond_wait failed consumer thread");
2419 }
2420 /* Caller is expecting a negative value on failure. */
2421 ret = -1;
2422 goto error;
2423 }
2424
2425 pthread_mutex_lock(&consumer_data->pid_mutex);
2426 if (consumer_data->pid == 0) {
2427 ERR("Consumerd did not start");
2428 pthread_mutex_unlock(&consumer_data->pid_mutex);
2429 goto error;
2430 }
2431 pthread_mutex_unlock(&consumer_data->pid_mutex);
2432
2433 return 0;
2434
2435 error:
2436 return ret;
2437 }
2438
2439 /*
2440 * Join consumer thread
2441 */
2442 static int join_consumer_thread(struct consumer_data *consumer_data)
2443 {
2444 void *status;
2445
2446 /* Consumer pid must be a real one. */
2447 if (consumer_data->pid > 0) {
2448 int ret;
2449 ret = kill(consumer_data->pid, SIGTERM);
2450 if (ret) {
2451 PERROR("Error killing consumer daemon");
2452 return ret;
2453 }
2454 return pthread_join(consumer_data->thread, &status);
2455 } else {
2456 return 0;
2457 }
2458 }
2459
2460 /*
2461 * Fork and exec a consumer daemon (consumerd).
2462 *
2463 * Return pid if successful else -1.
2464 */
2465 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2466 {
2467 int ret;
2468 pid_t pid;
2469 const char *consumer_to_use;
2470 const char *verbosity;
2471 struct stat st;
2472
2473 DBG("Spawning consumerd");
2474
2475 pid = fork();
2476 if (pid == 0) {
2477 /*
2478 * Exec consumerd.
2479 */
2480 if (config.verbose_consumer) {
2481 verbosity = "--verbose";
2482 } else if (lttng_opt_quiet) {
2483 verbosity = "--quiet";
2484 } else {
2485 verbosity = "";
2486 }
2487
2488 switch (consumer_data->type) {
2489 case LTTNG_CONSUMER_KERNEL:
2490 /*
2491 * Find out which consumerd to execute. We will first try the
2492 * 64-bit path, then the sessiond's installation directory, and
2493 * fallback on the 32-bit one,
2494 */
2495 DBG3("Looking for a kernel consumer at these locations:");
2496 DBG3(" 1) %s", config.consumerd64_bin_path.value ? : "NULL");
2497 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, DEFAULT_CONSUMERD_FILE);
2498 DBG3(" 3) %s", config.consumerd32_bin_path.value ? : "NULL");
2499 if (stat(config.consumerd64_bin_path.value, &st) == 0) {
2500 DBG3("Found location #1");
2501 consumer_to_use = config.consumerd64_bin_path.value;
2502 } else if (stat(INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE, &st) == 0) {
2503 DBG3("Found location #2");
2504 consumer_to_use = INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE;
2505 } else if (stat(config.consumerd32_bin_path.value, &st) == 0) {
2506 DBG3("Found location #3");
2507 consumer_to_use = config.consumerd32_bin_path.value;
2508 } else {
2509 DBG("Could not find any valid consumerd executable");
2510 ret = -EINVAL;
2511 goto error;
2512 }
2513 DBG("Using kernel consumer at: %s", consumer_to_use);
2514 (void) execl(consumer_to_use,
2515 "lttng-consumerd", verbosity, "-k",
2516 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2517 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2518 "--group", config.tracing_group_name.value,
2519 NULL);
2520 break;
2521 case LTTNG_CONSUMER64_UST:
2522 {
2523 if (config.consumerd64_lib_dir.value) {
2524 char *tmp;
2525 size_t tmplen;
2526 char *tmpnew;
2527
2528 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2529 if (!tmp) {
2530 tmp = "";
2531 }
2532 tmplen = strlen(config.consumerd64_lib_dir.value) + 1 /* : */ + strlen(tmp);
2533 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2534 if (!tmpnew) {
2535 ret = -ENOMEM;
2536 goto error;
2537 }
2538 strcat(tmpnew, config.consumerd64_lib_dir.value);
2539 if (tmp[0] != '\0') {
2540 strcat(tmpnew, ":");
2541 strcat(tmpnew, tmp);
2542 }
2543 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2544 free(tmpnew);
2545 if (ret) {
2546 ret = -errno;
2547 goto error;
2548 }
2549 }
2550 DBG("Using 64-bit UST consumer at: %s", config.consumerd64_bin_path.value);
2551 (void) execl(config.consumerd64_bin_path.value, "lttng-consumerd", verbosity, "-u",
2552 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2553 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2554 "--group", config.tracing_group_name.value,
2555 NULL);
2556 break;
2557 }
2558 case LTTNG_CONSUMER32_UST:
2559 {
2560 if (config.consumerd32_lib_dir.value) {
2561 char *tmp;
2562 size_t tmplen;
2563 char *tmpnew;
2564
2565 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2566 if (!tmp) {
2567 tmp = "";
2568 }
2569 tmplen = strlen(config.consumerd32_lib_dir.value) + 1 /* : */ + strlen(tmp);
2570 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2571 if (!tmpnew) {
2572 ret = -ENOMEM;
2573 goto error;
2574 }
2575 strcat(tmpnew, config.consumerd32_lib_dir.value);
2576 if (tmp[0] != '\0') {
2577 strcat(tmpnew, ":");
2578 strcat(tmpnew, tmp);
2579 }
2580 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2581 free(tmpnew);
2582 if (ret) {
2583 ret = -errno;
2584 goto error;
2585 }
2586 }
2587 DBG("Using 32-bit UST consumer at: %s", config.consumerd32_bin_path.value);
2588 (void) execl(config.consumerd32_bin_path.value, "lttng-consumerd", verbosity, "-u",
2589 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2590 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2591 "--group", config.tracing_group_name.value,
2592 NULL);
2593 break;
2594 }
2595 default:
2596 ERR("unknown consumer type");
2597 errno = 0;
2598 }
2599 if (errno != 0) {
2600 PERROR("Consumer execl()");
2601 }
2602 /* Reaching this point, we got a failure on our execl(). */
2603 exit(EXIT_FAILURE);
2604 } else if (pid > 0) {
2605 ret = pid;
2606 } else {
2607 PERROR("start consumer fork");
2608 ret = -errno;
2609 }
2610 error:
2611 return ret;
2612 }
2613
2614 /*
2615 * Spawn the consumerd daemon and session daemon thread.
2616 */
2617 static int start_consumerd(struct consumer_data *consumer_data)
2618 {
2619 int ret;
2620
2621 /*
2622 * Set the listen() state on the socket since there is a possible race
2623 * between the exec() of the consumer daemon and this call if place in the
2624 * consumer thread. See bug #366 for more details.
2625 */
2626 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2627 if (ret < 0) {
2628 goto error;
2629 }
2630
2631 pthread_mutex_lock(&consumer_data->pid_mutex);
2632 if (consumer_data->pid != 0) {
2633 pthread_mutex_unlock(&consumer_data->pid_mutex);
2634 goto end;
2635 }
2636
2637 ret = spawn_consumerd(consumer_data);
2638 if (ret < 0) {
2639 ERR("Spawning consumerd failed");
2640 pthread_mutex_unlock(&consumer_data->pid_mutex);
2641 goto error;
2642 }
2643
2644 /* Setting up the consumer_data pid */
2645 consumer_data->pid = ret;
2646 DBG2("Consumer pid %d", consumer_data->pid);
2647 pthread_mutex_unlock(&consumer_data->pid_mutex);
2648
2649 DBG2("Spawning consumer control thread");
2650 ret = spawn_consumer_thread(consumer_data);
2651 if (ret < 0) {
2652 ERR("Fatal error spawning consumer control thread");
2653 goto error;
2654 }
2655
2656 end:
2657 return 0;
2658
2659 error:
2660 /* Cleanup already created sockets on error. */
2661 if (consumer_data->err_sock >= 0) {
2662 int err;
2663
2664 err = close(consumer_data->err_sock);
2665 if (err < 0) {
2666 PERROR("close consumer data error socket");
2667 }
2668 }
2669 return ret;
2670 }
2671
2672 /*
2673 * Setup necessary data for kernel tracer action.
2674 */
2675 static int init_kernel_tracer(void)
2676 {
2677 int ret;
2678
2679 /* Modprobe lttng kernel modules */
2680 ret = modprobe_lttng_control();
2681 if (ret < 0) {
2682 goto error;
2683 }
2684
2685 /* Open debugfs lttng */
2686 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2687 if (kernel_tracer_fd < 0) {
2688 DBG("Failed to open %s", module_proc_lttng);
2689 goto error_open;
2690 }
2691
2692 /* Validate kernel version */
2693 ret = kernel_validate_version(kernel_tracer_fd, &kernel_tracer_version,
2694 &kernel_tracer_abi_version);
2695 if (ret < 0) {
2696 goto error_version;
2697 }
2698
2699 ret = modprobe_lttng_data();
2700 if (ret < 0) {
2701 goto error_modules;
2702 }
2703
2704 ret = kernel_supports_ring_buffer_snapshot_sample_positions(
2705 kernel_tracer_fd);
2706 if (ret < 0) {
2707 goto error_modules;
2708 }
2709
2710 if (ret < 1) {
2711 WARN("Kernel tracer does not support buffer monitoring. "
2712 "The monitoring timer of channels in the kernel domain "
2713 "will be set to 0 (disabled).");
2714 }
2715
2716 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2717 return 0;
2718
2719 error_version:
2720 modprobe_remove_lttng_control();
2721 ret = close(kernel_tracer_fd);
2722 if (ret) {
2723 PERROR("close");
2724 }
2725 kernel_tracer_fd = -1;
2726 return LTTNG_ERR_KERN_VERSION;
2727
2728 error_modules:
2729 ret = close(kernel_tracer_fd);
2730 if (ret) {
2731 PERROR("close");
2732 }
2733
2734 error_open:
2735 modprobe_remove_lttng_control();
2736
2737 error:
2738 WARN("No kernel tracer available");
2739 kernel_tracer_fd = -1;
2740 if (!is_root) {
2741 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2742 } else {
2743 return LTTNG_ERR_KERN_NA;
2744 }
2745 }
2746
2747
2748 /*
2749 * Copy consumer output from the tracing session to the domain session. The
2750 * function also applies the right modification on a per domain basis for the
2751 * trace files destination directory.
2752 *
2753 * Should *NOT* be called with RCU read-side lock held.
2754 */
2755 static int copy_session_consumer(int domain, struct ltt_session *session)
2756 {
2757 int ret;
2758 const char *dir_name;
2759 struct consumer_output *consumer;
2760
2761 assert(session);
2762 assert(session->consumer);
2763
2764 switch (domain) {
2765 case LTTNG_DOMAIN_KERNEL:
2766 DBG3("Copying tracing session consumer output in kernel session");
2767 /*
2768 * XXX: We should audit the session creation and what this function
2769 * does "extra" in order to avoid a destroy since this function is used
2770 * in the domain session creation (kernel and ust) only. Same for UST
2771 * domain.
2772 */
2773 if (session->kernel_session->consumer) {
2774 consumer_output_put(session->kernel_session->consumer);
2775 }
2776 session->kernel_session->consumer =
2777 consumer_copy_output(session->consumer);
2778 /* Ease our life a bit for the next part */
2779 consumer = session->kernel_session->consumer;
2780 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2781 break;
2782 case LTTNG_DOMAIN_JUL:
2783 case LTTNG_DOMAIN_LOG4J:
2784 case LTTNG_DOMAIN_PYTHON:
2785 case LTTNG_DOMAIN_UST:
2786 DBG3("Copying tracing session consumer output in UST session");
2787 if (session->ust_session->consumer) {
2788 consumer_output_put(session->ust_session->consumer);
2789 }
2790 session->ust_session->consumer =
2791 consumer_copy_output(session->consumer);
2792 /* Ease our life a bit for the next part */
2793 consumer = session->ust_session->consumer;
2794 dir_name = DEFAULT_UST_TRACE_DIR;
2795 break;
2796 default:
2797 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2798 goto error;
2799 }
2800
2801 /* Append correct directory to subdir */
2802 strncat(consumer->subdir, dir_name,
2803 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2804 DBG3("Copy session consumer subdir %s", consumer->subdir);
2805
2806 ret = LTTNG_OK;
2807
2808 error:
2809 return ret;
2810 }
2811
2812 /*
2813 * Create an UST session and add it to the session ust list.
2814 *
2815 * Should *NOT* be called with RCU read-side lock held.
2816 */
2817 static int create_ust_session(struct ltt_session *session,
2818 struct lttng_domain *domain)
2819 {
2820 int ret;
2821 struct ltt_ust_session *lus = NULL;
2822
2823 assert(session);
2824 assert(domain);
2825 assert(session->consumer);
2826
2827 switch (domain->type) {
2828 case LTTNG_DOMAIN_JUL:
2829 case LTTNG_DOMAIN_LOG4J:
2830 case LTTNG_DOMAIN_PYTHON:
2831 case LTTNG_DOMAIN_UST:
2832 break;
2833 default:
2834 ERR("Unknown UST domain on create session %d", domain->type);
2835 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2836 goto error;
2837 }
2838
2839 DBG("Creating UST session");
2840
2841 lus = trace_ust_create_session(session->id);
2842 if (lus == NULL) {
2843 ret = LTTNG_ERR_UST_SESS_FAIL;
2844 goto error;
2845 }
2846
2847 lus->uid = session->uid;
2848 lus->gid = session->gid;
2849 lus->output_traces = session->output_traces;
2850 lus->snapshot_mode = session->snapshot_mode;
2851 lus->live_timer_interval = session->live_timer;
2852 session->ust_session = lus;
2853 if (session->shm_path[0]) {
2854 strncpy(lus->root_shm_path, session->shm_path,
2855 sizeof(lus->root_shm_path));
2856 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2857 strncpy(lus->shm_path, session->shm_path,
2858 sizeof(lus->shm_path));
2859 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2860 strncat(lus->shm_path, "/ust",
2861 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2862 }
2863 /* Copy session output to the newly created UST session */
2864 ret = copy_session_consumer(domain->type, session);
2865 if (ret != LTTNG_OK) {
2866 goto error;
2867 }
2868
2869 return LTTNG_OK;
2870
2871 error:
2872 free(lus);
2873 session->ust_session = NULL;
2874 return ret;
2875 }
2876
2877 /*
2878 * Create a kernel tracer session then create the default channel.
2879 */
2880 static int create_kernel_session(struct ltt_session *session)
2881 {
2882 int ret;
2883
2884 DBG("Creating kernel session");
2885
2886 ret = kernel_create_session(session, kernel_tracer_fd);
2887 if (ret < 0) {
2888 ret = LTTNG_ERR_KERN_SESS_FAIL;
2889 goto error;
2890 }
2891
2892 /* Code flow safety */
2893 assert(session->kernel_session);
2894
2895 /* Copy session output to the newly created Kernel session */
2896 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2897 if (ret != LTTNG_OK) {
2898 goto error;
2899 }
2900
2901 session->kernel_session->uid = session->uid;
2902 session->kernel_session->gid = session->gid;
2903 session->kernel_session->output_traces = session->output_traces;
2904 session->kernel_session->snapshot_mode = session->snapshot_mode;
2905
2906 return LTTNG_OK;
2907
2908 error:
2909 trace_kernel_destroy_session(session->kernel_session);
2910 session->kernel_session = NULL;
2911 return ret;
2912 }
2913
2914 /*
2915 * Count number of session permitted by uid/gid.
2916 */
2917 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2918 {
2919 unsigned int i = 0;
2920 struct ltt_session *session;
2921
2922 DBG("Counting number of available session for UID %d GID %d",
2923 uid, gid);
2924 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2925 /*
2926 * Only list the sessions the user can control.
2927 */
2928 if (!session_access_ok(session, uid, gid)) {
2929 continue;
2930 }
2931 i++;
2932 }
2933 return i;
2934 }
2935
2936 static int receive_userspace_probe(struct command_ctx *cmd_ctx, int sock,
2937 int *sock_error, struct lttng_event *event)
2938 {
2939 int fd, ret;
2940 struct lttng_userspace_probe_location *probe_location;
2941 const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
2942 struct lttng_dynamic_buffer probe_location_buffer;
2943 struct lttng_buffer_view buffer_view;
2944
2945 /*
2946 * Create a buffer to store the serialized version of the probe
2947 * location.
2948 */
2949 lttng_dynamic_buffer_init(&probe_location_buffer);
2950 ret = lttng_dynamic_buffer_set_size(&probe_location_buffer,
2951 cmd_ctx->lsm->u.enable.userspace_probe_location_len);
2952 if (ret) {
2953 ret = LTTNG_ERR_NOMEM;
2954 goto error;
2955 }
2956
2957 /*
2958 * Receive the probe location.
2959 */
2960 ret = lttcomm_recv_unix_sock(sock, probe_location_buffer.data,
2961 probe_location_buffer.size);
2962 if (ret <= 0) {
2963 DBG("Nothing recv() from client var len data... continuing");
2964 *sock_error = 1;
2965 lttng_dynamic_buffer_reset(&probe_location_buffer);
2966 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2967 goto error;
2968 }
2969
2970 buffer_view = lttng_buffer_view_from_dynamic_buffer(
2971 &probe_location_buffer, 0, probe_location_buffer.size);
2972
2973 /*
2974 * Extract the probe location from the serialized version.
2975 */
2976 ret = lttng_userspace_probe_location_create_from_buffer(
2977 &buffer_view, &probe_location);
2978 if (ret < 0) {
2979 WARN("Failed to create a userspace probe location from the received buffer");
2980 lttng_dynamic_buffer_reset( &probe_location_buffer);
2981 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2982 goto error;
2983 }
2984
2985 /*
2986 * Receive the file descriptor to the target binary from the client.
2987 */
2988 DBG("Receiving userspace probe target FD from client ...");
2989 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
2990 if (ret <= 0) {
2991 DBG("Nothing recv() from client userspace probe fd... continuing");
2992 *sock_error = 1;
2993 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2994 goto error;
2995 }
2996
2997 /*
2998 * Set the file descriptor received from the client through the unix
2999 * socket in the probe location.
3000 */
3001 lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
3002 if (!lookup) {
3003 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
3004 goto error;
3005 }
3006
3007 /*
3008 * From the kernel tracer's perspective, all userspace probe event types
3009 * are all the same: a file and an offset.
3010 */
3011 switch (lttng_userspace_probe_location_lookup_method_get_type(lookup)) {
3012 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
3013 ret = lttng_userspace_probe_location_function_set_binary_fd(
3014 probe_location, fd);
3015 break;
3016 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
3017 ret = lttng_userspace_probe_location_tracepoint_set_binary_fd(
3018 probe_location, fd);
3019 break;
3020 default:
3021 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
3022 goto error;
3023 }
3024
3025 if (ret) {
3026 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
3027 goto error;
3028 }
3029
3030 /* Attach the probe location to the event. */
3031 ret = lttng_event_set_userspace_probe_location(event, probe_location);
3032 if (ret) {
3033 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
3034 goto error;
3035 }
3036
3037 lttng_dynamic_buffer_reset(&probe_location_buffer);
3038 error:
3039 return ret;
3040 }
3041
3042 /*
3043 * Check if the current kernel tracer supports the session rotation feature.
3044 * Return 1 if it does, 0 otherwise.
3045 */
3046 static int check_rotate_compatible(void)
3047 {
3048 int ret = 1;
3049
3050 if (kernel_tracer_version.major != 2 || kernel_tracer_version.minor < 11) {
3051 DBG("Kernel tracer version is not compatible with the rotation feature");
3052 ret = 0;
3053 }
3054
3055 return ret;
3056 }
3057
3058 /*
3059 * Process the command requested by the lttng client within the command
3060 * context structure. This function make sure that the return structure (llm)
3061 * is set and ready for transmission before returning.
3062 *
3063 * Return any error encountered or 0 for success.
3064 *
3065 * "sock" is only used for special-case var. len data.
3066 *
3067 * Should *NOT* be called with RCU read-side lock held.
3068 */
3069 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
3070 int *sock_error)
3071 {
3072 int ret = LTTNG_OK;
3073 int need_tracing_session = 1;
3074 int need_domain;
3075
3076 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
3077
3078 assert(!rcu_read_ongoing());
3079
3080 *sock_error = 0;
3081
3082 switch (cmd_ctx->lsm->cmd_type) {
3083 case LTTNG_CREATE_SESSION:
3084 case LTTNG_CREATE_SESSION_SNAPSHOT:
3085 case LTTNG_CREATE_SESSION_LIVE:
3086 case LTTNG_DESTROY_SESSION:
3087 case LTTNG_LIST_SESSIONS:
3088 case LTTNG_LIST_DOMAINS:
3089 case LTTNG_START_TRACE:
3090 case LTTNG_STOP_TRACE:
3091 case LTTNG_DATA_PENDING:
3092 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3093 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3094 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3095 case LTTNG_SNAPSHOT_RECORD:
3096 case LTTNG_SAVE_SESSION:
3097 case LTTNG_SET_SESSION_SHM_PATH:
3098 case LTTNG_REGENERATE_METADATA:
3099 case LTTNG_REGENERATE_STATEDUMP:
3100 case LTTNG_REGISTER_TRIGGER:
3101 case LTTNG_UNREGISTER_TRIGGER:
3102 case LTTNG_ROTATE_SESSION:
3103 case LTTNG_ROTATION_GET_INFO:
3104 case LTTNG_SESSION_GET_CURRENT_OUTPUT:
3105 case LTTNG_ROTATION_SET_SCHEDULE:
3106 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES:
3107 need_domain = 0;
3108 break;
3109 default:
3110 need_domain = 1;
3111 }
3112
3113 if (config.no_kernel && need_domain
3114 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3115 if (!is_root) {
3116 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3117 } else {
3118 ret = LTTNG_ERR_KERN_NA;
3119 }
3120 goto error;
3121 }
3122
3123 /* Deny register consumer if we already have a spawned consumer. */
3124 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3125 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3126 if (kconsumer_data.pid > 0) {
3127 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3128 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3129 goto error;
3130 }
3131 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3132 }
3133
3134 /*
3135 * Check for command that don't needs to allocate a returned payload. We do
3136 * this here so we don't have to make the call for no payload at each
3137 * command.
3138 */
3139 switch(cmd_ctx->lsm->cmd_type) {
3140 case LTTNG_LIST_SESSIONS:
3141 case LTTNG_LIST_TRACEPOINTS:
3142 case LTTNG_LIST_TRACEPOINT_FIELDS:
3143 case LTTNG_LIST_DOMAINS:
3144 case LTTNG_LIST_CHANNELS:
3145 case LTTNG_LIST_EVENTS:
3146 case LTTNG_LIST_SYSCALLS:
3147 case LTTNG_LIST_TRACKER_PIDS:
3148 case LTTNG_DATA_PENDING:
3149 case LTTNG_ROTATE_SESSION:
3150 case LTTNG_ROTATION_GET_INFO:
3151 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES:
3152 break;
3153 default:
3154 /* Setup lttng message with no payload */
3155 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
3156 if (ret < 0) {
3157 /* This label does not try to unlock the session */
3158 goto init_setup_error;
3159 }
3160 }
3161
3162 /* Commands that DO NOT need a session. */
3163 switch (cmd_ctx->lsm->cmd_type) {
3164 case LTTNG_CREATE_SESSION:
3165 case LTTNG_CREATE_SESSION_SNAPSHOT:
3166 case LTTNG_CREATE_SESSION_LIVE:
3167 case LTTNG_LIST_SESSIONS:
3168 case LTTNG_LIST_TRACEPOINTS:
3169 case LTTNG_LIST_SYSCALLS:
3170 case LTTNG_LIST_TRACEPOINT_FIELDS:
3171 case LTTNG_SAVE_SESSION:
3172 case LTTNG_REGISTER_TRIGGER:
3173 case LTTNG_UNREGISTER_TRIGGER:
3174 need_tracing_session = 0;
3175 break;
3176 default:
3177 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3178 /*
3179 * We keep the session list lock across _all_ commands
3180 * for now, because the per-session lock does not
3181 * handle teardown properly.
3182 */
3183 session_lock_list();
3184 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3185 if (cmd_ctx->session == NULL) {
3186 ret = LTTNG_ERR_SESS_NOT_FOUND;
3187 goto error;
3188 } else {
3189 /* Acquire lock for the session */
3190 session_lock(cmd_ctx->session);
3191 }
3192 break;
3193 }
3194
3195 /*
3196 * Commands that need a valid session but should NOT create one if none
3197 * exists. Instead of creating one and destroying it when the command is
3198 * handled, process that right before so we save some round trip in useless
3199 * code path.
3200 */
3201 switch (cmd_ctx->lsm->cmd_type) {
3202 case LTTNG_DISABLE_CHANNEL:
3203 case LTTNG_DISABLE_EVENT:
3204 switch (cmd_ctx->lsm->domain.type) {
3205 case LTTNG_DOMAIN_KERNEL:
3206 if (!cmd_ctx->session->kernel_session) {
3207 ret = LTTNG_ERR_NO_CHANNEL;
3208 goto error;
3209 }
3210 break;
3211 case LTTNG_DOMAIN_JUL:
3212 case LTTNG_DOMAIN_LOG4J:
3213 case LTTNG_DOMAIN_PYTHON:
3214 case LTTNG_DOMAIN_UST:
3215 if (!cmd_ctx->session->ust_session) {
3216 ret = LTTNG_ERR_NO_CHANNEL;
3217 goto error;
3218 }
3219 break;
3220 default:
3221 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3222 goto error;
3223 }
3224 default:
3225 break;
3226 }
3227
3228 if (!need_domain) {
3229 goto skip_domain;
3230 }
3231
3232 /*
3233 * Check domain type for specific "pre-action".
3234 */
3235 switch (cmd_ctx->lsm->domain.type) {
3236 case LTTNG_DOMAIN_KERNEL:
3237 if (!is_root) {
3238 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3239 goto error;
3240 }
3241
3242 /* Kernel tracer check */
3243 if (kernel_tracer_fd == -1) {
3244 /* Basically, load kernel tracer modules */
3245 ret = init_kernel_tracer();
3246 if (ret != 0) {
3247 goto error;
3248 }
3249 }
3250
3251 /* Consumer is in an ERROR state. Report back to client */
3252 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3253 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3254 goto error;
3255 }
3256
3257 /* Need a session for kernel command */
3258 if (need_tracing_session) {
3259 if (cmd_ctx->session->kernel_session == NULL) {
3260 ret = create_kernel_session(cmd_ctx->session);
3261 if (ret < 0) {
3262 ret = LTTNG_ERR_KERN_SESS_FAIL;
3263 goto error;
3264 }
3265 }
3266
3267 /* Start the kernel consumer daemon */
3268 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3269 if (kconsumer_data.pid == 0 &&
3270 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3271 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3272 ret = start_consumerd(&kconsumer_data);
3273 if (ret < 0) {
3274 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3275 goto error;
3276 }
3277 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3278 } else {
3279 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3280 }
3281
3282 /*
3283 * The consumer was just spawned so we need to add the socket to
3284 * the consumer output of the session if exist.
3285 */
3286 ret = consumer_create_socket(&kconsumer_data,
3287 cmd_ctx->session->kernel_session->consumer);
3288 if (ret < 0) {
3289 goto error;
3290 }
3291 }
3292
3293 break;
3294 case LTTNG_DOMAIN_JUL:
3295 case LTTNG_DOMAIN_LOG4J:
3296 case LTTNG_DOMAIN_PYTHON:
3297 case LTTNG_DOMAIN_UST:
3298 {
3299 if (!ust_app_supported()) {
3300 ret = LTTNG_ERR_NO_UST;
3301 goto error;
3302 }
3303 /* Consumer is in an ERROR state. Report back to client */
3304 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3305 ret = LTTNG_ERR_NO_USTCONSUMERD;
3306 goto error;
3307 }
3308
3309 if (need_tracing_session) {
3310 /* Create UST session if none exist. */
3311 if (cmd_ctx->session->ust_session == NULL) {
3312 ret = create_ust_session(cmd_ctx->session,
3313 &cmd_ctx->lsm->domain);
3314 if (ret != LTTNG_OK) {
3315 goto error;
3316 }
3317 }
3318
3319 /* Start the UST consumer daemons */
3320 /* 64-bit */
3321 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3322 if (config.consumerd64_bin_path.value &&
3323 ustconsumer64_data.pid == 0 &&
3324 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3325 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3326 ret = start_consumerd(&ustconsumer64_data);
3327 if (ret < 0) {
3328 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3329 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3330 goto error;
3331 }
3332
3333 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3334 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3335 } else {
3336 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3337 }
3338
3339 /*
3340 * Setup socket for consumer 64 bit. No need for atomic access
3341 * since it was set above and can ONLY be set in this thread.
3342 */
3343 ret = consumer_create_socket(&ustconsumer64_data,
3344 cmd_ctx->session->ust_session->consumer);
3345 if (ret < 0) {
3346 goto error;
3347 }
3348
3349 /* 32-bit */
3350 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3351 if (config.consumerd32_bin_path.value &&
3352 ustconsumer32_data.pid == 0 &&
3353 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3354 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3355 ret = start_consumerd(&ustconsumer32_data);
3356 if (ret < 0) {
3357 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3358 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3359 goto error;
3360 }
3361
3362 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3363 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3364 } else {
3365 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3366 }
3367
3368 /*
3369 * Setup socket for consumer 64 bit. No need for atomic access
3370 * since it was set above and can ONLY be set in this thread.
3371 */
3372 ret = consumer_create_socket(&ustconsumer32_data,
3373 cmd_ctx->session->ust_session->consumer);
3374 if (ret < 0) {
3375 goto error;
3376 }
3377 }
3378 break;
3379 }
3380 default:
3381 break;
3382 }
3383 skip_domain:
3384
3385 /* Validate consumer daemon state when start/stop trace command */
3386 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3387 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3388 switch (cmd_ctx->lsm->domain.type) {
3389 case LTTNG_DOMAIN_NONE:
3390 break;
3391 case LTTNG_DOMAIN_JUL:
3392 case LTTNG_DOMAIN_LOG4J:
3393 case LTTNG_DOMAIN_PYTHON:
3394 case LTTNG_DOMAIN_UST:
3395 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3396 ret = LTTNG_ERR_NO_USTCONSUMERD;
3397 goto error;
3398 }
3399 break;
3400 case LTTNG_DOMAIN_KERNEL:
3401 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3402 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3403 goto error;
3404 }
3405 break;
3406 default:
3407 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3408 goto error;
3409 }
3410 }
3411
3412 /*
3413 * Check that the UID or GID match that of the tracing session.
3414 * The root user can interact with all sessions.
3415 */
3416 if (need_tracing_session) {
3417 if (!session_access_ok(cmd_ctx->session,
3418 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3419 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3420 ret = LTTNG_ERR_EPERM;
3421 goto error;
3422 }
3423 }
3424
3425 /*
3426 * Send relayd information to consumer as soon as we have a domain and a
3427 * session defined.
3428 */
3429 if (cmd_ctx->session && need_domain) {
3430 /*
3431 * Setup relayd if not done yet. If the relayd information was already
3432 * sent to the consumer, this call will gracefully return.
3433 */
3434 ret = cmd_setup_relayd(cmd_ctx->session);
3435 if (ret != LTTNG_OK) {
3436 goto error;
3437 }
3438 }
3439
3440 /* Process by command type */
3441 switch (cmd_ctx->lsm->cmd_type) {
3442 case LTTNG_ADD_CONTEXT:
3443 {
3444 /*
3445 * An LTTNG_ADD_CONTEXT command might have a supplementary
3446 * payload if the context being added is an application context.
3447 */
3448 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3449 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3450 char *provider_name = NULL, *context_name = NULL;
3451 size_t provider_name_len =
3452 cmd_ctx->lsm->u.context.provider_name_len;
3453 size_t context_name_len =
3454 cmd_ctx->lsm->u.context.context_name_len;
3455
3456 if (provider_name_len == 0 || context_name_len == 0) {
3457 /*
3458 * Application provider and context names MUST
3459 * be provided.
3460 */
3461 ret = -LTTNG_ERR_INVALID;
3462 goto error;
3463 }
3464
3465 provider_name = zmalloc(provider_name_len + 1);
3466 if (!provider_name) {
3467 ret = -LTTNG_ERR_NOMEM;
3468 goto error;
3469 }
3470 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3471 provider_name;
3472
3473 context_name = zmalloc(context_name_len + 1);
3474 if (!context_name) {
3475 ret = -LTTNG_ERR_NOMEM;
3476 goto error_add_context;
3477 }
3478 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3479 context_name;
3480
3481 ret = lttcomm_recv_unix_sock(sock, provider_name,
3482 provider_name_len);
3483 if (ret < 0) {
3484 goto error_add_context;
3485 }
3486
3487 ret = lttcomm_recv_unix_sock(sock, context_name,
3488 context_name_len);
3489 if (ret < 0) {
3490 goto error_add_context;
3491 }
3492 }
3493
3494 /*
3495 * cmd_add_context assumes ownership of the provider and context
3496 * names.
3497 */
3498 ret = cmd_add_context(cmd_ctx->session,
3499 cmd_ctx->lsm->domain.type,
3500 cmd_ctx->lsm->u.context.channel_name,
3501 &cmd_ctx->lsm->u.context.ctx,
3502 kernel_poll_pipe[1]);
3503
3504 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3505 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3506 error_add_context:
3507 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3508 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3509 if (ret < 0) {
3510 goto error;
3511 }
3512 break;
3513 }
3514 case LTTNG_DISABLE_CHANNEL:
3515 {
3516 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3517 cmd_ctx->lsm->u.disable.channel_name);
3518 break;
3519 }
3520 case LTTNG_DISABLE_EVENT:
3521 {
3522
3523 /*
3524 * FIXME: handle filter; for now we just receive the filter's
3525 * bytecode along with the filter expression which are sent by
3526 * liblttng-ctl and discard them.
3527 *
3528 * This fixes an issue where the client may block while sending
3529 * the filter payload and encounter an error because the session
3530 * daemon closes the socket without ever handling this data.
3531 */
3532 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3533 cmd_ctx->lsm->u.disable.bytecode_len;
3534
3535 if (count) {
3536 char data[LTTNG_FILTER_MAX_LEN];
3537
3538 DBG("Discarding disable event command payload of size %zu", count);
3539 while (count) {
3540 ret = lttcomm_recv_unix_sock(sock, data,
3541 count > sizeof(data) ? sizeof(data) : count);
3542 if (ret < 0) {
3543 goto error;
3544 }
3545
3546 count -= (size_t) ret;
3547 }
3548 }
3549 /* FIXME: passing packed structure to non-packed pointer */
3550 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3551 cmd_ctx->lsm->u.disable.channel_name,
3552 &cmd_ctx->lsm->u.disable.event);
3553 break;
3554 }
3555 case LTTNG_ENABLE_CHANNEL:
3556 {
3557 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3558 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3559 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3560 &cmd_ctx->lsm->u.channel.chan,
3561 kernel_poll_pipe[1]);
3562 break;
3563 }
3564 case LTTNG_TRACK_PID:
3565 {
3566 ret = cmd_track_pid(cmd_ctx->session,
3567 cmd_ctx->lsm->domain.type,
3568 cmd_ctx->lsm->u.pid_tracker.pid);
3569 break;
3570 }
3571 case LTTNG_UNTRACK_PID:
3572 {
3573 ret = cmd_untrack_pid(cmd_ctx->session,
3574 cmd_ctx->lsm->domain.type,
3575 cmd_ctx->lsm->u.pid_tracker.pid);
3576 break;
3577 }
3578 case LTTNG_ENABLE_EVENT:
3579 {
3580 struct lttng_event *ev = NULL;
3581 struct lttng_event_exclusion *exclusion = NULL;
3582 struct lttng_filter_bytecode *bytecode = NULL;
3583 char *filter_expression = NULL;
3584
3585 /* Handle exclusion events and receive it from the client. */
3586 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3587 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3588
3589 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3590 (count * LTTNG_SYMBOL_NAME_LEN));
3591 if (!exclusion) {
3592 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3593 goto error;
3594 }
3595
3596 DBG("Receiving var len exclusion event list from client ...");
3597 exclusion->count = count;
3598 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3599 count * LTTNG_SYMBOL_NAME_LEN);
3600 if (ret <= 0) {
3601 DBG("Nothing recv() from client var len data... continuing");
3602 *sock_error = 1;
3603 free(exclusion);
3604 ret = LTTNG_ERR_EXCLUSION_INVAL;
3605 goto error;
3606 }
3607 }
3608
3609 /* Get filter expression from client. */
3610 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3611 size_t expression_len =
3612 cmd_ctx->lsm->u.enable.expression_len;
3613
3614 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3615 ret = LTTNG_ERR_FILTER_INVAL;
3616 free(exclusion);
3617 goto error;
3618 }
3619
3620 filter_expression = zmalloc(expression_len);
3621 if (!filter_expression) {
3622 free(exclusion);
3623 ret = LTTNG_ERR_FILTER_NOMEM;
3624 goto error;
3625 }
3626
3627 /* Receive var. len. data */
3628 DBG("Receiving var len filter's expression from client ...");
3629 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3630 expression_len);
3631 if (ret <= 0) {
3632 DBG("Nothing recv() from client var len data... continuing");
3633 *sock_error = 1;
3634 free(filter_expression);
3635 free(exclusion);
3636 ret = LTTNG_ERR_FILTER_INVAL;
3637 goto error;
3638 }
3639 }
3640
3641 /* Handle filter and get bytecode from client. */
3642 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3643 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3644
3645 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3646 ret = LTTNG_ERR_FILTER_INVAL;
3647 free(filter_expression);
3648 free(exclusion);
3649 goto error;
3650 }
3651
3652 bytecode = zmalloc(bytecode_len);
3653 if (!bytecode) {
3654 free(filter_expression);
3655 free(exclusion);
3656 ret = LTTNG_ERR_FILTER_NOMEM;
3657 goto error;
3658 }
3659
3660 /* Receive var. len. data */
3661 DBG("Receiving var len filter's bytecode from client ...");
3662 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3663 if (ret <= 0) {
3664 DBG("Nothing recv() from client var len data... continuing");
3665 *sock_error = 1;
3666 free(filter_expression);
3667 free(bytecode);
3668 free(exclusion);
3669 ret = LTTNG_ERR_FILTER_INVAL;
3670 goto error;
3671 }
3672
3673 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3674 free(filter_expression);
3675 free(bytecode);
3676 free(exclusion);
3677 ret = LTTNG_ERR_FILTER_INVAL;
3678 goto error;
3679 }
3680 }
3681
3682 ev = lttng_event_copy(&cmd_ctx->lsm->u.enable.event);
3683 if (!ev) {
3684 DBG("Failed to copy event: %s",
3685 cmd_ctx->lsm->u.enable.event.name);
3686 ret = LTTNG_ERR_NOMEM;
3687 goto error;
3688 }
3689
3690
3691 if (cmd_ctx->lsm->u.enable.userspace_probe_location_len > 0) {
3692 /* Expect a userspace probe description. */
3693 ret = receive_userspace_probe(cmd_ctx, sock, sock_error, ev);
3694 if (ret) {
3695 lttng_event_destroy(ev);
3696 goto error;
3697 }
3698 }
3699
3700 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3701 cmd_ctx->lsm->u.enable.channel_name,
3702 ev,
3703 filter_expression, bytecode, exclusion,
3704 kernel_poll_pipe[1]);
3705 lttng_event_destroy(ev);
3706 break;
3707 }
3708 case LTTNG_LIST_TRACEPOINTS:
3709 {
3710 struct lttng_event *events;
3711 ssize_t nb_events;
3712
3713 session_lock_list();
3714 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3715 session_unlock_list();
3716 if (nb_events < 0) {
3717 /* Return value is a negative lttng_error_code. */
3718 ret = -nb_events;
3719 goto error;
3720 }
3721
3722 /*
3723 * Setup lttng message with payload size set to the event list size in
3724 * bytes and then copy list into the llm payload.
3725 */
3726 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3727 sizeof(struct lttng_event) * nb_events);
3728 free(events);
3729
3730 if (ret < 0) {
3731 goto setup_error;
3732 }
3733
3734 ret = LTTNG_OK;
3735 break;
3736 }
3737 case LTTNG_LIST_TRACEPOINT_FIELDS:
3738 {
3739 struct lttng_event_field *fields;
3740 ssize_t nb_fields;
3741
3742 session_lock_list();
3743 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3744 &fields);
3745 session_unlock_list();
3746 if (nb_fields < 0) {
3747 /* Return value is a negative lttng_error_code. */
3748 ret = -nb_fields;
3749 goto error;
3750 }
3751
3752 /*
3753 * Setup lttng message with payload size set to the event list size in
3754 * bytes and then copy list into the llm payload.
3755 */
3756 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3757 sizeof(struct lttng_event_field) * nb_fields);
3758 free(fields);
3759
3760 if (ret < 0) {
3761 goto setup_error;
3762 }
3763
3764 ret = LTTNG_OK;
3765 break;
3766 }
3767 case LTTNG_LIST_SYSCALLS:
3768 {
3769 struct lttng_event *events;
3770 ssize_t nb_events;
3771
3772 nb_events = cmd_list_syscalls(&events);
3773 if (nb_events < 0) {
3774 /* Return value is a negative lttng_error_code. */
3775 ret = -nb_events;
3776 goto error;
3777 }
3778
3779 /*
3780 * Setup lttng message with payload size set to the event list size in
3781 * bytes and then copy list into the llm payload.
3782 */
3783 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3784 sizeof(struct lttng_event) * nb_events);
3785 free(events);
3786
3787 if (ret < 0) {
3788 goto setup_error;
3789 }
3790
3791 ret = LTTNG_OK;
3792 break;
3793 }
3794 case LTTNG_LIST_TRACKER_PIDS:
3795 {
3796 int32_t *pids = NULL;
3797 ssize_t nr_pids;
3798
3799 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3800 cmd_ctx->lsm->domain.type, &pids);
3801 if (nr_pids < 0) {
3802 /* Return value is a negative lttng_error_code. */
3803 ret = -nr_pids;
3804 goto error;
3805 }
3806
3807 /*
3808 * Setup lttng message with payload size set to the event list size in
3809 * bytes and then copy list into the llm payload.
3810 */
3811 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3812 sizeof(int32_t) * nr_pids);
3813 free(pids);
3814
3815 if (ret < 0) {
3816 goto setup_error;
3817 }
3818
3819 ret = LTTNG_OK;
3820 break;
3821 }
3822 case LTTNG_SET_CONSUMER_URI:
3823 {
3824 size_t nb_uri, len;
3825 struct lttng_uri *uris;
3826
3827 nb_uri = cmd_ctx->lsm->u.uri.size;
3828 len = nb_uri * sizeof(struct lttng_uri);
3829
3830 if (nb_uri == 0) {
3831 ret = LTTNG_ERR_INVALID;
3832 goto error;
3833 }
3834
3835 uris = zmalloc(len);
3836 if (uris == NULL) {
3837 ret = LTTNG_ERR_FATAL;
3838 goto error;
3839 }
3840
3841 /* Receive variable len data */
3842 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3843 ret = lttcomm_recv_unix_sock(sock, uris, len);
3844 if (ret <= 0) {
3845 DBG("No URIs received from client... continuing");
3846 *sock_error = 1;
3847 ret = LTTNG_ERR_SESSION_FAIL;
3848 free(uris);
3849 goto error;
3850 }
3851
3852 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3853 free(uris);
3854 if (ret != LTTNG_OK) {
3855 goto error;
3856 }
3857
3858
3859 break;
3860 }
3861 case LTTNG_START_TRACE:
3862 {
3863 /*
3864 * On the first start, if we have a kernel session and we have
3865 * enabled time or size-based rotations, we have to make sure
3866 * the kernel tracer supports it.
3867 */
3868 if (!cmd_ctx->session->has_been_started && \
3869 cmd_ctx->session->kernel_session && \
3870 (cmd_ctx->session->rotate_timer_period || \
3871 cmd_ctx->session->rotate_size) && \
3872 !check_rotate_compatible()) {
3873 DBG("Kernel tracer version is not compatible with the rotation feature");
3874 ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
3875 goto error;
3876 }
3877 ret = cmd_start_trace(cmd_ctx->session);
3878 break;
3879 }
3880 case LTTNG_STOP_TRACE:
3881 {
3882 ret = cmd_stop_trace(cmd_ctx->session);
3883 break;
3884 }
3885 case LTTNG_CREATE_SESSION:
3886 {
3887 size_t nb_uri, len;
3888 struct lttng_uri *uris = NULL;
3889
3890 nb_uri = cmd_ctx->lsm->u.uri.size;
3891 len = nb_uri * sizeof(struct lttng_uri);
3892
3893 if (nb_uri > 0) {
3894 uris = zmalloc(len);
3895 if (uris == NULL) {
3896 ret = LTTNG_ERR_FATAL;
3897 goto error;
3898 }
3899
3900 /* Receive variable len data */
3901 DBG("Waiting for %zu URIs from client ...", nb_uri);
3902 ret = lttcomm_recv_unix_sock(sock, uris, len);
3903 if (ret <= 0) {
3904 DBG("No URIs received from client... continuing");
3905 *sock_error = 1;
3906 ret = LTTNG_ERR_SESSION_FAIL;
3907 free(uris);
3908 goto error;
3909 }
3910
3911 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3912 DBG("Creating session with ONE network URI is a bad call");
3913 ret = LTTNG_ERR_SESSION_FAIL;
3914 free(uris);
3915 goto error;
3916 }
3917 }
3918
3919 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3920 &cmd_ctx->creds, 0);
3921
3922 free(uris);
3923
3924 break;
3925 }
3926 case LTTNG_DESTROY_SESSION:
3927 {
3928 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1],
3929 notification_thread_handle);
3930
3931 /* Set session to NULL so we do not unlock it after free. */
3932 cmd_ctx->session = NULL;
3933 break;
3934 }
3935 case LTTNG_LIST_DOMAINS:
3936 {
3937 ssize_t nb_dom;
3938 struct lttng_domain *domains = NULL;
3939
3940 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3941 if (nb_dom < 0) {
3942 /* Return value is a negative lttng_error_code. */
3943 ret = -nb_dom;
3944 goto error;
3945 }
3946
3947 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3948 nb_dom * sizeof(struct lttng_domain));
3949 free(domains);
3950
3951 if (ret < 0) {
3952 goto setup_error;
3953 }
3954
3955 ret = LTTNG_OK;
3956 break;
3957 }
3958 case LTTNG_LIST_CHANNELS:
3959 {
3960 ssize_t payload_size;
3961 struct lttng_channel *channels = NULL;
3962
3963 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3964 cmd_ctx->session, &channels);
3965 if (payload_size < 0) {
3966 /* Return value is a negative lttng_error_code. */
3967 ret = -payload_size;
3968 goto error;
3969 }
3970
3971 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3972 payload_size);
3973 free(channels);
3974
3975 if (ret < 0) {
3976 goto setup_error;
3977 }
3978
3979 ret = LTTNG_OK;
3980 break;
3981 }
3982 case LTTNG_LIST_EVENTS:
3983 {
3984 ssize_t nb_event;
3985 struct lttng_event *events = NULL;
3986 struct lttcomm_event_command_header cmd_header;
3987 size_t total_size;
3988
3989 memset(&cmd_header, 0, sizeof(cmd_header));
3990 /* Extended infos are included at the end of events */
3991 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3992 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3993 &events, &total_size);
3994
3995 if (nb_event < 0) {
3996 /* Return value is a negative lttng_error_code. */
3997 ret = -nb_event;
3998 goto error;
3999 }
4000
4001 cmd_header.nb_events = nb_event;
4002 ret = setup_lttng_msg(cmd_ctx, events, total_size,
4003 &cmd_header, sizeof(cmd_header));
4004 free(events);
4005
4006 if (ret < 0) {
4007 goto setup_error;
4008 }
4009
4010 ret = LTTNG_OK;
4011 break;
4012 }
4013 case LTTNG_LIST_SESSIONS:
4014 {
4015 unsigned int nr_sessions;
4016 void *sessions_payload;
4017 size_t payload_len;
4018
4019 session_lock_list();
4020 nr_sessions = lttng_sessions_count(
4021 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
4022 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
4023 payload_len = sizeof(struct lttng_session) * nr_sessions;
4024 sessions_payload = zmalloc(payload_len);
4025
4026 if (!sessions_payload) {
4027 session_unlock_list();
4028 ret = -ENOMEM;
4029 goto setup_error;
4030 }
4031
4032 cmd_list_lttng_sessions(sessions_payload,
4033 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
4034 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
4035 session_unlock_list();
4036
4037 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
4038 payload_len);
4039 free(sessions_payload);
4040
4041 if (ret < 0) {
4042 goto setup_error;
4043 }
4044
4045 ret = LTTNG_OK;
4046 break;
4047 }
4048 case LTTNG_REGISTER_CONSUMER:
4049 {
4050 struct consumer_data *cdata;
4051
4052 switch (cmd_ctx->lsm->domain.type) {
4053 case LTTNG_DOMAIN_KERNEL:
4054 cdata = &kconsumer_data;
4055 break;
4056 default:
4057 ret = LTTNG_ERR_UND;
4058 goto error;
4059 }
4060
4061 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
4062 cmd_ctx->lsm->u.reg.path, cdata);
4063 break;
4064 }
4065 case LTTNG_DATA_PENDING:
4066 {
4067 int pending_ret;
4068 uint8_t pending_ret_byte;
4069
4070 pending_ret = cmd_data_pending(cmd_ctx->session);
4071
4072 /*
4073 * FIXME
4074 *
4075 * This function may returns 0 or 1 to indicate whether or not
4076 * there is data pending. In case of error, it should return an
4077 * LTTNG_ERR code. However, some code paths may still return
4078 * a nondescript error code, which we handle by returning an
4079 * "unknown" error.
4080 */
4081 if (pending_ret == 0 || pending_ret == 1) {
4082 /*
4083 * ret will be set to LTTNG_OK at the end of
4084 * this function.
4085 */
4086 } else if (pending_ret < 0) {
4087 ret = LTTNG_ERR_UNK;
4088 goto setup_error;
4089 } else {
4090 ret = pending_ret;
4091 goto setup_error;
4092 }
4093
4094 pending_ret_byte = (uint8_t) pending_ret;
4095
4096 /* 1 byte to return whether or not data is pending */
4097 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
4098 &pending_ret_byte, 1);
4099
4100 if (ret < 0) {
4101 goto setup_error;
4102 }
4103
4104 ret = LTTNG_OK;
4105 break;
4106 }
4107 case LTTNG_SNAPSHOT_ADD_OUTPUT:
4108 {
4109 struct lttcomm_lttng_output_id reply;
4110
4111 ret = cmd_snapshot_add_output(cmd_ctx->session,
4112 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
4113 if (ret != LTTNG_OK) {
4114 goto error;
4115 }
4116
4117 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
4118 sizeof(reply));
4119 if (ret < 0) {
4120 goto setup_error;
4121 }
4122
4123 /* Copy output list into message payload */
4124 ret = LTTNG_OK;
4125 break;
4126 }
4127 case LTTNG_SNAPSHOT_DEL_OUTPUT:
4128 {
4129 ret = cmd_snapshot_del_output(cmd_ctx->session,
4130 &cmd_ctx->lsm->u.snapshot_output.output);
4131 break;
4132 }
4133 case LTTNG_SNAPSHOT_LIST_OUTPUT:
4134 {
4135 ssize_t nb_output;
4136 struct lttng_snapshot_output *outputs = NULL;
4137
4138 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
4139 if (nb_output < 0) {
4140 ret = -nb_output;
4141 goto error;
4142 }
4143
4144 assert((nb_output > 0 && outputs) || nb_output == 0);
4145 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
4146 nb_output * sizeof(struct lttng_snapshot_output));
4147 free(outputs);
4148
4149 if (ret < 0) {
4150 goto setup_error;
4151 }
4152
4153 ret = LTTNG_OK;
4154 break;
4155 }
4156 case LTTNG_SNAPSHOT_RECORD:
4157 {
4158 ret = cmd_snapshot_record(cmd_ctx->session,
4159 &cmd_ctx->lsm->u.snapshot_record.output,
4160 cmd_ctx->lsm->u.snapshot_record.wait);
4161 break;
4162 }
4163 case LTTNG_CREATE_SESSION_SNAPSHOT:
4164 {
4165 size_t nb_uri, len;
4166 struct lttng_uri *uris = NULL;
4167
4168 nb_uri = cmd_ctx->lsm->u.uri.size;
4169 len = nb_uri * sizeof(struct lttng_uri);
4170
4171 if (nb_uri > 0) {
4172 uris = zmalloc(len);
4173 if (uris == NULL) {
4174 ret = LTTNG_ERR_FATAL;
4175 goto error;
4176 }
4177
4178 /* Receive variable len data */
4179 DBG("Waiting for %zu URIs from client ...", nb_uri);
4180 ret = lttcomm_recv_unix_sock(sock, uris, len);
4181 if (ret <= 0) {
4182 DBG("No URIs received from client... continuing");
4183 *sock_error = 1;
4184 ret = LTTNG_ERR_SESSION_FAIL;
4185 free(uris);
4186 goto error;
4187 }
4188
4189 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4190 DBG("Creating session with ONE network URI is a bad call");
4191 ret = LTTNG_ERR_SESSION_FAIL;
4192 free(uris);
4193 goto error;
4194 }
4195 }
4196
4197 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
4198 nb_uri, &cmd_ctx->creds);
4199 free(uris);
4200 break;
4201 }
4202 case LTTNG_CREATE_SESSION_LIVE:
4203 {
4204 size_t nb_uri, len;
4205 struct lttng_uri *uris = NULL;
4206
4207 nb_uri = cmd_ctx->lsm->u.uri.size;
4208 len = nb_uri * sizeof(struct lttng_uri);
4209
4210 if (nb_uri > 0) {
4211 uris = zmalloc(len);
4212 if (uris == NULL) {
4213 ret = LTTNG_ERR_FATAL;
4214 goto error;
4215 }
4216
4217 /* Receive variable len data */
4218 DBG("Waiting for %zu URIs from client ...", nb_uri);
4219 ret = lttcomm_recv_unix_sock(sock, uris, len);
4220 if (ret <= 0) {
4221 DBG("No URIs received from client... continuing");
4222 *sock_error = 1;
4223 ret = LTTNG_ERR_SESSION_FAIL;
4224 free(uris);
4225 goto error;
4226 }
4227
4228 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4229 DBG("Creating session with ONE network URI is a bad call");
4230 ret = LTTNG_ERR_SESSION_FAIL;
4231 free(uris);
4232 goto error;
4233 }
4234 }
4235
4236 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4237 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4238 free(uris);
4239 break;
4240 }
4241 case LTTNG_SAVE_SESSION:
4242 {
4243 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4244 &cmd_ctx->creds);
4245 break;
4246 }
4247 case LTTNG_SET_SESSION_SHM_PATH:
4248 {
4249 ret = cmd_set_session_shm_path(cmd_ctx->session,
4250 cmd_ctx->lsm->u.set_shm_path.shm_path);
4251 break;
4252 }
4253 case LTTNG_REGENERATE_METADATA:
4254 {
4255 ret = cmd_regenerate_metadata(cmd_ctx->session);
4256 break;
4257 }
4258 case LTTNG_REGENERATE_STATEDUMP:
4259 {
4260 ret = cmd_regenerate_statedump(cmd_ctx->session);
4261 break;
4262 }
4263 case LTTNG_REGISTER_TRIGGER:
4264 {
4265 ret = cmd_register_trigger(cmd_ctx, sock,
4266 notification_thread_handle);
4267 break;
4268 }
4269 case LTTNG_UNREGISTER_TRIGGER:
4270 {
4271 ret = cmd_unregister_trigger(cmd_ctx, sock,
4272 notification_thread_handle);
4273 break;
4274 }
4275 case LTTNG_ROTATE_SESSION:
4276 {
4277 struct lttng_rotate_session_return rotate_return;
4278
4279 DBG("Client rotate session \"%s\"", cmd_ctx->session->name);
4280
4281 memset(&rotate_return, 0, sizeof(rotate_return));
4282 if (cmd_ctx->session->kernel_session && !check_rotate_compatible()) {
4283 DBG("Kernel tracer version is not compatible with the rotation feature");
4284 ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
4285 goto error;
4286 }
4287
4288 ret = cmd_rotate_session(cmd_ctx->session, &rotate_return);
4289 if (ret < 0) {
4290 ret = -ret;
4291 goto error;
4292 }
4293
4294 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &rotate_return,
4295 sizeof(rotate_return));
4296 if (ret < 0) {
4297 ret = -ret;
4298 goto error;
4299 }
4300
4301 ret = LTTNG_OK;
4302 break;
4303 }
4304 case LTTNG_ROTATION_GET_INFO:
4305 {
4306 struct lttng_rotation_get_info_return get_info_return;
4307
4308 memset(&get_info_return, 0, sizeof(get_info_return));
4309 ret = cmd_rotate_get_info(cmd_ctx->session, &get_info_return,
4310 cmd_ctx->lsm->u.get_rotation_info.rotation_id);
4311 if (ret < 0) {
4312 ret = -ret;
4313 goto error;
4314 }
4315
4316 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &get_info_return,
4317 sizeof(get_info_return));
4318 if (ret < 0) {
4319 ret = -ret;
4320 goto error;
4321 }
4322
4323 ret = LTTNG_OK;
4324 break;
4325 }
4326 case LTTNG_SESSION_GET_CURRENT_OUTPUT:
4327 {
4328 struct lttng_session_get_current_output_return output_return;
4329
4330 memset(&output_return, 0, sizeof(output_return));
4331 ret = cmd_session_get_current_output(cmd_ctx->session,
4332 &output_return);
4333 if (ret < 0) {
4334 ret = -ret;
4335 goto error;
4336 }
4337
4338 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &output_return,
4339 sizeof(output_return));
4340 if (ret < 0) {
4341 ret = -ret;
4342 goto error;
4343 }
4344
4345 ret = LTTNG_OK;
4346 break;
4347 }
4348 case LTTNG_ROTATION_SET_SCHEDULE:
4349 {
4350 bool set_schedule;
4351 enum lttng_rotation_schedule_type schedule_type;
4352 uint64_t value;
4353
4354 if (cmd_ctx->session->kernel_session && !check_rotate_compatible()) {
4355 DBG("Kernel tracer version does not support session rotations");
4356 ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
4357 goto error;
4358 }
4359
4360 set_schedule = cmd_ctx->lsm->u.rotation_set_schedule.set == 1;
4361 schedule_type = (enum lttng_rotation_schedule_type) cmd_ctx->lsm->u.rotation_set_schedule.type;
4362 value = cmd_ctx->lsm->u.rotation_set_schedule.value;
4363
4364 ret = cmd_rotation_set_schedule(cmd_ctx->session,
4365 set_schedule,
4366 schedule_type,
4367 value,
4368 notification_thread_handle);
4369 if (ret != LTTNG_OK) {
4370 goto error;
4371 }
4372
4373 break;
4374 }
4375 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES:
4376 {
4377 struct lttng_session_list_schedules_return schedules = {
4378 .periodic.set = !!cmd_ctx->session->rotate_timer_period,
4379 .periodic.value = cmd_ctx->session->rotate_timer_period,
4380 .size.set = !!cmd_ctx->session->rotate_size,
4381 .size.value = cmd_ctx->session->rotate_size,
4382 };
4383
4384 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &schedules,
4385 sizeof(schedules));
4386 if (ret < 0) {
4387 ret = -ret;
4388 goto error;
4389 }
4390
4391 ret = LTTNG_OK;
4392 break;
4393 }
4394 default:
4395 ret = LTTNG_ERR_UND;
4396 break;
4397 }
4398
4399 error:
4400 if (cmd_ctx->llm == NULL) {
4401 DBG("Missing llm structure. Allocating one.");
4402 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4403 goto setup_error;
4404 }
4405 }
4406 /* Set return code */
4407 cmd_ctx->llm->ret_code = ret;
4408 setup_error:
4409 if (cmd_ctx->session) {
4410 session_unlock(cmd_ctx->session);
4411 }
4412 if (need_tracing_session) {
4413 session_unlock_list();
4414 }
4415 init_setup_error:
4416 assert(!rcu_read_ongoing());
4417 return ret;
4418 }
4419
4420 /*
4421 * Thread managing health check socket.
4422 */
4423 static void *thread_manage_health(void *data)
4424 {
4425 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4426 uint32_t revents, nb_fd;
4427 struct lttng_poll_event events;
4428 struct health_comm_msg msg;
4429 struct health_comm_reply reply;
4430
4431 DBG("[thread] Manage health check started");
4432
4433 rcu_register_thread();
4434
4435 /* We might hit an error path before this is created. */
4436 lttng_poll_init(&events);
4437
4438 /* Create unix socket */
4439 sock = lttcomm_create_unix_sock(config.health_unix_sock_path.value);
4440 if (sock < 0) {
4441 ERR("Unable to create health check Unix socket");
4442 goto error;
4443 }
4444
4445 if (is_root) {
4446 /* lttng health client socket path permissions */
4447 ret = chown(config.health_unix_sock_path.value, 0,
4448 utils_get_group_id(config.tracing_group_name.value));
4449 if (ret < 0) {
4450 ERR("Unable to set group on %s", config.health_unix_sock_path.value);
4451 PERROR("chown");
4452 goto error;
4453 }
4454
4455 ret = chmod(config.health_unix_sock_path.value,
4456 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4457 if (ret < 0) {
4458 ERR("Unable to set permissions on %s", config.health_unix_sock_path.value);
4459 PERROR("chmod");
4460 goto error;
4461 }
4462 }
4463
4464 /*
4465 * Set the CLOEXEC flag. Return code is useless because either way, the
4466 * show must go on.
4467 */
4468 (void) utils_set_fd_cloexec(sock);
4469
4470 ret = lttcomm_listen_unix_sock(sock);
4471 if (ret < 0) {
4472 goto error;
4473 }
4474
4475 /*
4476 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4477 * more will be added to this poll set.
4478 */
4479 ret = sessiond_set_thread_pollset(&events, 2);
4480 if (ret < 0) {
4481 goto error;
4482 }
4483
4484 /* Add the application registration socket */
4485 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4486 if (ret < 0) {
4487 goto error;
4488 }
4489
4490 sessiond_notify_ready();
4491
4492 while (1) {
4493 DBG("Health check ready");
4494
4495 /* Inifinite blocking call, waiting for transmission */
4496 restart:
4497 ret = lttng_poll_wait(&events, -1);
4498 if (ret < 0) {
4499 /*
4500 * Restart interrupted system call.
4501 */
4502 if (errno == EINTR) {
4503 goto restart;
4504 }
4505 goto error;
4506 }
4507
4508 nb_fd = ret;
4509
4510 for (i = 0; i < nb_fd; i++) {
4511 /* Fetch once the poll data */
4512 revents = LTTNG_POLL_GETEV(&events, i);
4513 pollfd = LTTNG_POLL_GETFD(&events, i);
4514
4515 if (!revents) {
4516 /* No activity for this FD (poll implementation). */
4517 continue;
4518 }
4519
4520 /* Thread quit pipe has been closed. Killing thread. */
4521 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4522 if (ret) {
4523 err = 0;
4524 goto exit;
4525 }
4526
4527 /* Event on the registration socket */
4528 if (pollfd == sock) {
4529 if (revents & LPOLLIN) {
4530 continue;
4531 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4532 ERR("Health socket poll error");
4533 goto error;
4534 } else {
4535 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4536 goto error;
4537 }
4538 }
4539 }
4540
4541 new_sock = lttcomm_accept_unix_sock(sock);
4542 if (new_sock < 0) {
4543 goto error;
4544 }
4545
4546 /*
4547 * Set the CLOEXEC flag. Return code is useless because either way, the
4548 * show must go on.
4549 */
4550 (void) utils_set_fd_cloexec(new_sock);
4551
4552 DBG("Receiving data from client for health...");
4553 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4554 if (ret <= 0) {
4555 DBG("Nothing recv() from client... continuing");
4556 ret = close(new_sock);
4557 if (ret) {
4558 PERROR("close");
4559 }
4560 continue;
4561 }
4562
4563 rcu_thread_online();
4564
4565 memset(&reply, 0, sizeof(reply));
4566 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4567 /*
4568 * health_check_state returns 0 if health is
4569 * bad.
4570 */
4571 if (!health_check_state(health_sessiond, i)) {
4572 reply.ret_code |= 1ULL << i;
4573 }
4574 }
4575
4576 DBG2("Health check return value %" PRIx64, reply.ret_code);
4577
4578 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4579 if (ret < 0) {
4580 ERR("Failed to send health data back to client");
4581 }
4582
4583 /* End of transmission */
4584 ret = close(new_sock);
4585 if (ret) {
4586 PERROR("close");
4587 }
4588 }
4589
4590 exit:
4591 error:
4592 if (err) {
4593 ERR("Health error occurred in %s", __func__);
4594 }
4595 DBG("Health check thread dying");
4596 unlink(config.health_unix_sock_path.value);
4597 if (sock >= 0) {
4598 ret = close(sock);
4599 if (ret) {
4600 PERROR("close");
4601 }
4602 }
4603
4604 lttng_poll_clean(&events);
4605 stop_threads();
4606 rcu_unregister_thread();
4607 return NULL;
4608 }
4609
4610 /*
4611 * This thread manage all clients request using the unix client socket for
4612 * communication.
4613 */
4614 static void *thread_manage_clients(void *data)
4615 {
4616 int sock = -1, ret, i, pollfd, err = -1;
4617 int sock_error;
4618 uint32_t revents, nb_fd;
4619 struct command_ctx *cmd_ctx = NULL;
4620 struct lttng_poll_event events;
4621
4622 DBG("[thread] Manage client started");
4623
4624 rcu_register_thread();
4625
4626 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4627
4628 health_code_update();
4629
4630 ret = lttcomm_listen_unix_sock(client_sock);
4631 if (ret < 0) {
4632 goto error_listen;
4633 }
4634
4635 /*
4636 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4637 * more will be added to this poll set.
4638 */
4639 ret = sessiond_set_thread_pollset(&events, 2);
4640 if (ret < 0) {
4641 goto error_create_poll;
4642 }
4643
4644 /* Add the application registration socket */
4645 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4646 if (ret < 0) {
4647 goto error;
4648 }
4649
4650 ret = sem_post(&load_info->message_thread_ready);
4651 if (ret) {
4652 PERROR("sem_post message_thread_ready");
4653 goto error;
4654 }
4655
4656 /*
4657 * Wait until all support threads are initialized before accepting
4658 * commands.
4659 */
4660 while (uatomic_read(&lttng_sessiond_ready) != 0) {
4661 fd_set read_fds;
4662 struct timeval timeout;
4663
4664 FD_ZERO(&read_fds);
4665 FD_SET(thread_quit_pipe[0], &read_fds);
4666 memset(&timeout, 0, sizeof(timeout));
4667 timeout.tv_usec = 1000;
4668
4669 /*
4670 * If a support thread failed to launch, it may signal that
4671 * we must exit and the sessiond would never be marked as
4672 * "ready".
4673 *
4674 * The timeout is set to 1ms, which serves as a way to
4675 * pace down this check.
4676 */
4677 ret = select(thread_quit_pipe[0] + 1, &read_fds, NULL, NULL,
4678 &timeout);
4679 if (ret > 0 || (ret < 0 && errno != EINTR)) {
4680 goto exit;
4681 }
4682 }
4683 /*
4684 * This barrier is paired with the one in sessiond_notify_ready() to
4685 * ensure that loads accessing data initialized by the other threads,
4686 * on which this thread was waiting, are not performed before this point.
4687 *
4688 * Note that this could be a 'read' memory barrier, but a full barrier
4689 * is used in case the code changes. The performance implications of
4690 * this choice are minimal since this is a slow path.
4691 */
4692 cmm_smp_mb();
4693
4694 /* This testpoint is after we signal readiness to the parent. */
4695 if (testpoint(sessiond_thread_manage_clients)) {
4696 goto error;
4697 }
4698
4699 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4700 goto error;
4701 }
4702
4703 health_code_update();
4704
4705 while (1) {
4706 const struct cmd_completion_handler *cmd_completion_handler;
4707
4708 DBG("Accepting client command ...");
4709
4710 /* Inifinite blocking call, waiting for transmission */
4711 restart:
4712 health_poll_entry();
4713 ret = lttng_poll_wait(&events, -1);
4714 health_poll_exit();
4715 if (ret < 0) {
4716 /*
4717 * Restart interrupted system call.
4718 */
4719 if (errno == EINTR) {
4720 goto restart;
4721 }
4722 goto error;
4723 }
4724
4725 nb_fd = ret;
4726
4727 for (i = 0; i < nb_fd; i++) {
4728 /* Fetch once the poll data */
4729 revents = LTTNG_POLL_GETEV(&events, i);
4730 pollfd = LTTNG_POLL_GETFD(&events, i);
4731
4732 health_code_update();
4733
4734 if (!revents) {
4735 /* No activity for this FD (poll implementation). */
4736 continue;
4737 }
4738
4739 /* Thread quit pipe has been closed. Killing thread. */
4740 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4741 if (ret) {
4742 err = 0;
4743 goto exit;
4744 }
4745
4746 /* Event on the registration socket */
4747 if (pollfd == client_sock) {
4748 if (revents & LPOLLIN) {
4749 continue;
4750 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4751 ERR("Client socket poll error");
4752 goto error;
4753 } else {
4754 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4755 goto error;
4756 }
4757 }
4758 }
4759
4760 DBG("Wait for client response");
4761
4762 health_code_update();
4763
4764 sock = lttcomm_accept_unix_sock(client_sock);
4765 if (sock < 0) {
4766 goto error;
4767 }
4768
4769 /*
4770 * Set the CLOEXEC flag. Return code is useless because either way, the
4771 * show must go on.
4772 */
4773 (void) utils_set_fd_cloexec(sock);
4774
4775 /* Set socket option for credentials retrieval */
4776 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4777 if (ret < 0) {
4778 goto error;
4779 }
4780
4781 /* Allocate context command to process the client request */
4782 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4783 if (cmd_ctx == NULL) {
4784 PERROR("zmalloc cmd_ctx");
4785 goto error;
4786 }
4787
4788 /* Allocate data buffer for reception */
4789 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4790 if (cmd_ctx->lsm == NULL) {
4791 PERROR("zmalloc cmd_ctx->lsm");
4792 goto error;
4793 }
4794
4795 cmd_ctx->llm = NULL;
4796 cmd_ctx->session = NULL;
4797
4798 health_code_update();
4799
4800 /*
4801 * Data is received from the lttng client. The struct
4802 * lttcomm_session_msg (lsm) contains the command and data request of
4803 * the client.
4804 */
4805 DBG("Receiving data from client ...");
4806 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4807 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4808 if (ret <= 0) {
4809 DBG("Nothing recv() from client... continuing");
4810 ret = close(sock);
4811 if (ret) {
4812 PERROR("close");
4813 }
4814 sock = -1;
4815 clean_command_ctx(&cmd_ctx);
4816 continue;
4817 }
4818
4819 health_code_update();
4820
4821 // TODO: Validate cmd_ctx including sanity check for
4822 // security purpose.
4823
4824 rcu_thread_online();
4825 /*
4826 * This function dispatch the work to the kernel or userspace tracer
4827 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4828 * informations for the client. The command context struct contains
4829 * everything this function may needs.
4830 */
4831 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4832 rcu_thread_offline();
4833 if (ret < 0) {
4834 ret = close(sock);
4835 if (ret) {
4836 PERROR("close");
4837 }
4838 sock = -1;
4839 /*
4840 * TODO: Inform client somehow of the fatal error. At
4841 * this point, ret < 0 means that a zmalloc failed
4842 * (ENOMEM). Error detected but still accept
4843 * command, unless a socket error has been
4844 * detected.
4845 */
4846 clean_command_ctx(&cmd_ctx);
4847 continue;
4848 }
4849
4850 cmd_completion_handler = cmd_pop_completion_handler();
4851 if (cmd_completion_handler) {
4852 enum lttng_error_code completion_code;
4853
4854 completion_code = cmd_completion_handler->run(
4855 cmd_completion_handler->data);
4856 if (completion_code != LTTNG_OK) {
4857 clean_command_ctx(&cmd_ctx);
4858 continue;
4859 }
4860 }
4861
4862 health_code_update();
4863
4864 DBG("Sending response (size: %d, retcode: %s (%d))",
4865 cmd_ctx->lttng_msg_size,
4866 lttng_strerror(-cmd_ctx->llm->ret_code),
4867 cmd_ctx->llm->ret_code);
4868 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4869 if (ret < 0) {
4870 ERR("Failed to send data back to client");
4871 }
4872
4873 /* End of transmission */
4874 ret = close(sock);
4875 if (ret) {
4876 PERROR("close");
4877 }
4878 sock = -1;
4879
4880 clean_command_ctx(&cmd_ctx);
4881
4882 health_code_update();
4883 }
4884
4885 exit:
4886 error:
4887 if (sock >= 0) {
4888 ret = close(sock);
4889 if (ret) {
4890 PERROR("close");
4891 }
4892 }
4893
4894 lttng_poll_clean(&events);
4895 clean_command_ctx(&cmd_ctx);
4896
4897 error_listen:
4898 error_create_poll:
4899 unlink(config.client_unix_sock_path.value);
4900 if (client_sock >= 0) {
4901 ret = close(client_sock);
4902 if (ret) {
4903 PERROR("close");
4904 }
4905 }
4906
4907 if (err) {
4908 health_error();
4909 ERR("Health error occurred in %s", __func__);
4910 }
4911
4912 health_unregister(health_sessiond);
4913
4914 DBG("Client thread dying");
4915
4916 rcu_unregister_thread();
4917
4918 /*
4919 * Since we are creating the consumer threads, we own them, so we need
4920 * to join them before our thread exits.
4921 */
4922 ret = join_consumer_thread(&kconsumer_data);
4923 if (ret) {
4924 errno = ret;
4925 PERROR("join_consumer");
4926 }
4927
4928 ret = join_consumer_thread(&ustconsumer32_data);
4929 if (ret) {
4930 errno = ret;
4931 PERROR("join_consumer ust32");
4932 }
4933
4934 ret = join_consumer_thread(&ustconsumer64_data);
4935 if (ret) {
4936 errno = ret;
4937 PERROR("join_consumer ust64");
4938 }
4939 return NULL;
4940 }
4941
4942 static int string_match(const char *str1, const char *str2)
4943 {
4944 return (str1 && str2) && !strcmp(str1, str2);
4945 }
4946
4947 /*
4948 * Take an option from the getopt output and set it in the right variable to be
4949 * used later.
4950 *
4951 * Return 0 on success else a negative value.
4952 */
4953 static int set_option(int opt, const char *arg, const char *optname)
4954 {
4955 int ret = 0;
4956
4957 if (string_match(optname, "client-sock") || opt == 'c') {
4958 if (!arg || *arg == '\0') {
4959 ret = -EINVAL;
4960 goto end;
4961 }
4962 if (lttng_is_setuid_setgid()) {
4963 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4964 "-c, --client-sock");
4965 } else {
4966 config_string_set(&config.client_unix_sock_path,
4967 strdup(arg));
4968 if (!config.client_unix_sock_path.value) {
4969 ret = -ENOMEM;
4970 PERROR("strdup");
4971 }
4972 }
4973 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4974 if (!arg || *arg == '\0') {
4975 ret = -EINVAL;
4976 goto end;
4977 }
4978 if (lttng_is_setuid_setgid()) {
4979 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4980 "-a, --apps-sock");
4981 } else {
4982 config_string_set(&config.apps_unix_sock_path,
4983 strdup(arg));
4984 if (!config.apps_unix_sock_path.value) {
4985 ret = -ENOMEM;
4986 PERROR("strdup");
4987 }
4988 }
4989 } else if (string_match(optname, "daemonize") || opt == 'd') {
4990 config.daemonize = true;
4991 } else if (string_match(optname, "background") || opt == 'b') {
4992 config.background = true;
4993 } else if (string_match(optname, "group") || opt == 'g') {
4994 if (!arg || *arg == '\0') {
4995 ret = -EINVAL;
4996 goto end;
4997 }
4998 if (lttng_is_setuid_setgid()) {
4999 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5000 "-g, --group");
5001 } else {
5002 config_string_set(&config.tracing_group_name,
5003 strdup(arg));
5004 if (!config.tracing_group_name.value) {
5005 ret = -ENOMEM;
5006 PERROR("strdup");
5007 }
5008 }
5009 } else if (string_match(optname, "help") || opt == 'h') {
5010 ret = utils_show_help(8, "lttng-sessiond", help_msg);
5011 if (ret) {
5012 ERR("Cannot show --help for `lttng-sessiond`");
5013 perror("exec");
5014 }
5015 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
5016 } else if (string_match(optname, "version") || opt == 'V') {
5017 fprintf(stdout, "%s\n", VERSION);
5018 exit(EXIT_SUCCESS);
5019 } else if (string_match(optname, "sig-parent") || opt == 'S') {
5020 config.sig_parent = true;
5021 } else if (string_match(optname, "kconsumerd-err-sock")) {
5022 if (!arg || *arg == '\0') {
5023 ret = -EINVAL;
5024 goto end;
5025 }
5026 if (lttng_is_setuid_setgid()) {
5027 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5028 "--kconsumerd-err-sock");
5029 } else {
5030 config_string_set(&config.kconsumerd_err_unix_sock_path,
5031 strdup(arg));
5032 if (!config.kconsumerd_err_unix_sock_path.value) {
5033 ret = -ENOMEM;
5034 PERROR("strdup");
5035 }
5036 }
5037 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
5038 if (!arg || *arg == '\0') {
5039 ret = -EINVAL;
5040 goto end;
5041 }
5042 if (lttng_is_setuid_setgid()) {
5043 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5044 "--kconsumerd-cmd-sock");
5045 } else {
5046 config_string_set(&config.kconsumerd_cmd_unix_sock_path,
5047 strdup(arg));
5048 if (!config.kconsumerd_cmd_unix_sock_path.value) {
5049 ret = -ENOMEM;
5050 PERROR("strdup");
5051 }
5052 }
5053 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
5054 if (!arg || *arg == '\0') {
5055 ret = -EINVAL;
5056 goto end;
5057 }
5058 if (lttng_is_setuid_setgid()) {
5059 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5060 "--ustconsumerd64-err-sock");
5061 } else {
5062 config_string_set(&config.consumerd64_err_unix_sock_path,
5063 strdup(arg));
5064 if (!config.consumerd64_err_unix_sock_path.value) {
5065 ret = -ENOMEM;
5066 PERROR("strdup");
5067 }
5068 }
5069 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
5070 if (!arg || *arg == '\0') {
5071 ret = -EINVAL;
5072 goto end;
5073 }
5074 if (lttng_is_setuid_setgid()) {
5075 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5076 "--ustconsumerd64-cmd-sock");
5077 } else {
5078 config_string_set(&config.consumerd64_cmd_unix_sock_path,
5079 strdup(arg));
5080 if (!config.consumerd64_cmd_unix_sock_path.value) {
5081 ret = -ENOMEM;
5082 PERROR("strdup");
5083 }
5084 }
5085 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
5086 if (!arg || *arg == '\0') {
5087 ret = -EINVAL;
5088 goto end;
5089 }
5090 if (lttng_is_setuid_setgid()) {
5091 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5092 "--ustconsumerd32-err-sock");
5093 } else {
5094 config_string_set(&config.consumerd32_err_unix_sock_path,
5095 strdup(arg));
5096 if (!config.consumerd32_err_unix_sock_path.value) {
5097 ret = -ENOMEM;
5098 PERROR("strdup");
5099 }
5100 }
5101 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
5102 if (!arg || *arg == '\0') {
5103 ret = -EINVAL;
5104 goto end;
5105 }
5106 if (lttng_is_setuid_setgid()) {
5107 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5108 "--ustconsumerd32-cmd-sock");
5109 } else {
5110 config_string_set(&config.consumerd32_cmd_unix_sock_path,
5111 strdup(arg));
5112 if (!config.consumerd32_cmd_unix_sock_path.value) {
5113 ret = -ENOMEM;
5114 PERROR("strdup");
5115 }
5116 }
5117 } else if (string_match(optname, "no-kernel")) {
5118 config.no_kernel = true;
5119 } else if (string_match(optname, "quiet") || opt == 'q') {
5120 config.quiet = true;
5121 } else if (string_match(optname, "verbose") || opt == 'v') {
5122 /* Verbose level can increase using multiple -v */
5123 if (arg) {
5124 /* Value obtained from config file */
5125 config.verbose = config_parse_value(arg);
5126 } else {
5127 /* -v used on command line */
5128 config.verbose++;
5129 }
5130 /* Clamp value to [0, 3] */
5131 config.verbose = config.verbose < 0 ? 0 :
5132 (config.verbose <= 3 ? config.verbose : 3);
5133 } else if (string_match(optname, "verbose-consumer")) {
5134 if (arg) {
5135 config.verbose_consumer = config_parse_value(arg);
5136 } else {
5137 config.verbose_consumer++;
5138 }
5139 } else if (string_match(optname, "consumerd32-path")) {
5140 if (!arg || *arg == '\0') {
5141 ret = -EINVAL;
5142 goto end;
5143 }
5144 if (lttng_is_setuid_setgid()) {
5145 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5146 "--consumerd32-path");
5147 } else {
5148 config_string_set(&config.consumerd32_bin_path,
5149 strdup(arg));
5150 if (!config.consumerd32_bin_path.value) {
5151 PERROR("strdup");
5152 ret = -ENOMEM;
5153 }
5154 }
5155 } else if (string_match(optname, "consumerd32-libdir")) {
5156 if (!arg || *arg == '\0') {
5157 ret = -EINVAL;
5158 goto end;
5159 }
5160 if (lttng_is_setuid_setgid()) {
5161 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5162 "--consumerd32-libdir");
5163 } else {
5164 config_string_set(&config.consumerd32_lib_dir,
5165 strdup(arg));
5166 if (!config.consumerd32_lib_dir.value) {
5167 PERROR("strdup");
5168 ret = -ENOMEM;
5169 }
5170 }
5171 } else if (string_match(optname, "consumerd64-path")) {
5172 if (!arg || *arg == '\0') {
5173 ret = -EINVAL;
5174 goto end;
5175 }
5176 if (lttng_is_setuid_setgid()) {
5177 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5178 "--consumerd64-path");
5179 } else {
5180 config_string_set(&config.consumerd64_bin_path,
5181 strdup(arg));
5182 if (!config.consumerd64_bin_path.value) {
5183 PERROR("strdup");
5184 ret = -ENOMEM;
5185 }
5186 }
5187 } else if (string_match(optname, "consumerd64-libdir")) {
5188 if (!arg || *arg == '\0') {
5189 ret = -EINVAL;
5190 goto end;
5191 }
5192 if (lttng_is_setuid_setgid()) {
5193 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5194 "--consumerd64-libdir");
5195 } else {
5196 config_string_set(&config.consumerd64_lib_dir,
5197 strdup(arg));
5198 if (!config.consumerd64_lib_dir.value) {
5199 PERROR("strdup");
5200 ret = -ENOMEM;
5201 }
5202 }
5203 } else if (string_match(optname, "pidfile") || opt == 'p') {
5204 if (!arg || *arg == '\0') {
5205 ret = -EINVAL;
5206 goto end;
5207 }
5208 if (lttng_is_setuid_setgid()) {
5209 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5210 "-p, --pidfile");
5211 } else {
5212 config_string_set(&config.pid_file_path, strdup(arg));
5213 if (!config.pid_file_path.value) {
5214 PERROR("strdup");
5215 ret = -ENOMEM;
5216 }
5217 }
5218 } else if (string_match(optname, "agent-tcp-port")) {
5219 if (!arg || *arg == '\0') {
5220 ret = -EINVAL;
5221 goto end;
5222 }
5223 if (lttng_is_setuid_setgid()) {
5224 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5225 "--agent-tcp-port");
5226 } else {
5227 unsigned long v;
5228
5229 errno = 0;
5230 v = strtoul(arg, NULL, 0);
5231 if (errno != 0 || !isdigit(arg[0])) {
5232 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
5233 return -1;
5234 }
5235 if (v == 0 || v >= 65535) {
5236 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
5237 return -1;
5238 }
5239 config.agent_tcp_port.begin = config.agent_tcp_port.end = (int) v;
5240 DBG3("Agent TCP port set to non default: %i", (int) v);
5241 }
5242 } else if (string_match(optname, "load") || opt == 'l') {
5243 if (!arg || *arg == '\0') {
5244 ret = -EINVAL;
5245 goto end;
5246 }
5247 if (lttng_is_setuid_setgid()) {
5248 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5249 "-l, --load");
5250 } else {
5251 config_string_set(&config.load_session_path, strdup(arg));
5252 if (!config.load_session_path.value) {
5253 PERROR("strdup");
5254 ret = -ENOMEM;
5255 }
5256 }
5257 } else if (string_match(optname, "kmod-probes")) {
5258 if (!arg || *arg == '\0') {
5259 ret = -EINVAL;
5260 goto end;
5261 }
5262 if (lttng_is_setuid_setgid()) {
5263 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5264 "--kmod-probes");
5265 } else {
5266 config_string_set(&config.kmod_probes_list, strdup(arg));
5267 if (!config.kmod_probes_list.value) {
5268 PERROR("strdup");
5269 ret = -ENOMEM;
5270 }
5271 }
5272 } else if (string_match(optname, "extra-kmod-probes")) {
5273 if (!arg || *arg == '\0') {
5274 ret = -EINVAL;
5275 goto end;
5276 }
5277 if (lttng_is_setuid_setgid()) {
5278 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5279 "--extra-kmod-probes");
5280 } else {
5281 config_string_set(&config.kmod_extra_probes_list,
5282 strdup(arg));
5283 if (!config.kmod_extra_probes_list.value) {
5284 PERROR("strdup");
5285 ret = -ENOMEM;
5286 }
5287 }
5288 } else if (string_match(optname, "config") || opt == 'f') {
5289 /* This is handled in set_options() thus silent skip. */
5290 goto end;
5291 } else {
5292 /* Unknown option or other error.
5293 * Error is printed by getopt, just return */
5294 ret = -1;
5295 }
5296
5297 end:
5298 if (ret == -EINVAL) {
5299 const char *opt_name = "unknown";
5300 int i;
5301
5302 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
5303 i++) {
5304 if (opt == long_options[i].val) {
5305 opt_name = long_options[i].name;
5306 break;
5307 }
5308 }
5309
5310 WARN("Invalid argument provided for option \"%s\", using default value.",
5311 opt_name);
5312 }
5313
5314 return ret;
5315 }
5316
5317 /*
5318 * config_entry_handler_cb used to handle options read from a config file.
5319 * See config_entry_handler_cb comment in common/config/session-config.h for the
5320 * return value conventions.
5321 */
5322 static int config_entry_handler(const struct config_entry *entry, void *unused)
5323 {
5324 int ret = 0, i;
5325
5326 if (!entry || !entry->name || !entry->value) {
5327 ret = -EINVAL;
5328 goto end;
5329 }
5330
5331 /* Check if the option is to be ignored */
5332 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
5333 if (!strcmp(entry->name, config_ignore_options[i])) {
5334 goto end;
5335 }
5336 }
5337
5338 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
5339 i++) {
5340
5341 /* Ignore if not fully matched. */
5342 if (strcmp(entry->name, long_options[i].name)) {
5343 continue;
5344 }
5345
5346 /*
5347 * If the option takes no argument on the command line, we have to
5348 * check if the value is "true". We support non-zero numeric values,
5349 * true, on and yes.
5350 */
5351 if (!long_options[i].has_arg) {
5352 ret = config_parse_value(entry->value);
5353 if (ret <= 0) {
5354 if (ret) {
5355 WARN("Invalid configuration value \"%s\" for option %s",
5356 entry->value, entry->name);
5357 }
5358 /* False, skip boolean config option. */
5359 goto end;
5360 }
5361 }
5362
5363 ret = set_option(long_options[i].val, entry->value, entry->name);
5364 goto end;
5365 }
5366
5367 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5368
5369 end:
5370 return ret;
5371 }
5372
5373 /*
5374 * daemon configuration loading and argument parsing
5375 */
5376 static int set_options(int argc, char **argv)
5377 {
5378 int ret = 0, c = 0, option_index = 0;
5379 int orig_optopt = optopt, orig_optind = optind;
5380 char *optstring;
5381 const char *config_path = NULL;
5382
5383 optstring = utils_generate_optstring(long_options,
5384 sizeof(long_options) / sizeof(struct option));
5385 if (!optstring) {
5386 ret = -ENOMEM;
5387 goto end;
5388 }
5389
5390 /* Check for the --config option */
5391 while ((c = getopt_long(argc, argv, optstring, long_options,
5392 &option_index)) != -1) {
5393 if (c == '?') {
5394 ret = -EINVAL;
5395 goto end;
5396 } else if (c != 'f') {
5397 /* if not equal to --config option. */
5398 continue;
5399 }
5400
5401 if (lttng_is_setuid_setgid()) {
5402 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5403 "-f, --config");
5404 } else {
5405 config_path = utils_expand_path(optarg);
5406 if (!config_path) {
5407 ERR("Failed to resolve path: %s", optarg);
5408 }
5409 }
5410 }
5411
5412 ret = config_get_section_entries(config_path, config_section_name,
5413 config_entry_handler, NULL);
5414 if (ret) {
5415 if (ret > 0) {
5416 ERR("Invalid configuration option at line %i", ret);
5417 ret = -1;
5418 }
5419 goto end;
5420 }
5421
5422 /* Reset getopt's global state */
5423 optopt = orig_optopt;
5424 optind = orig_optind;
5425 while (1) {
5426 option_index = -1;
5427 /*
5428 * getopt_long() will not set option_index if it encounters a
5429 * short option.
5430 */
5431 c = getopt_long(argc, argv, optstring, long_options,
5432 &option_index);
5433 if (c == -1) {
5434 break;
5435 }
5436
5437 /*
5438 * Pass NULL as the long option name if popt left the index
5439 * unset.
5440 */
5441 ret = set_option(c, optarg,
5442 option_index < 0 ? NULL :
5443 long_options[option_index].name);
5444 if (ret < 0) {
5445 break;
5446 }
5447 }
5448
5449 end:
5450 free(optstring);
5451 return ret;
5452 }
5453
5454 /*
5455 * Creates the two needed socket by the daemon.
5456 * apps_sock - The communication socket for all UST apps.
5457 * client_sock - The communication of the cli tool (lttng).
5458 */
5459 static int init_daemon_socket(void)
5460 {
5461 int ret = 0;
5462 mode_t old_umask;
5463
5464 old_umask = umask(0);
5465
5466 /* Create client tool unix socket */
5467 client_sock = lttcomm_create_unix_sock(config.client_unix_sock_path.value);
5468 if (client_sock < 0) {
5469 ERR("Create unix sock failed: %s", config.client_unix_sock_path.value);
5470 ret = -1;
5471 goto end;
5472 }
5473
5474 /* Set the cloexec flag */
5475 ret = utils_set_fd_cloexec(client_sock);
5476 if (ret < 0) {
5477 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5478 "Continuing but note that the consumer daemon will have a "
5479 "reference to this socket on exec()", client_sock);
5480 }
5481
5482 /* File permission MUST be 660 */
5483 ret = chmod(config.client_unix_sock_path.value, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5484 if (ret < 0) {
5485 ERR("Set file permissions failed: %s", config.client_unix_sock_path.value);
5486 PERROR("chmod");
5487 goto end;
5488 }
5489
5490 /* Create the application unix socket */
5491 apps_sock = lttcomm_create_unix_sock(config.apps_unix_sock_path.value);
5492 if (apps_sock < 0) {
5493 ERR("Create unix sock failed: %s", config.apps_unix_sock_path.value);
5494 ret = -1;
5495 goto end;
5496 }
5497
5498 /* Set the cloexec flag */
5499 ret = utils_set_fd_cloexec(apps_sock);
5500 if (ret < 0) {
5501 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5502 "Continuing but note that the consumer daemon will have a "
5503 "reference to this socket on exec()", apps_sock);
5504 }
5505
5506 /* File permission MUST be 666 */
5507 ret = chmod(config.apps_unix_sock_path.value,
5508 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5509 if (ret < 0) {
5510 ERR("Set file permissions failed: %s", config.apps_unix_sock_path.value);
5511 PERROR("chmod");
5512 goto end;
5513 }
5514
5515 DBG3("Session daemon client socket %d and application socket %d created",
5516 client_sock, apps_sock);
5517
5518 end:
5519 umask(old_umask);
5520 return ret;
5521 }
5522
5523 /*
5524 * Create lockfile using the rundir and return its fd.
5525 */
5526 static int create_lockfile(void)
5527 {
5528 return utils_create_lock_file(config.lock_file_path.value);
5529 }
5530
5531 /*
5532 * Check if the global socket is available, and if a daemon is answering at the
5533 * other side. If yes, error is returned.
5534 *
5535 * Also attempts to create and hold the lock file.
5536 */
5537 static int check_existing_daemon(void)
5538 {
5539 int ret = 0;
5540
5541 /* Is there anybody out there ? */
5542 if (lttng_session_daemon_alive()) {
5543 ret = -EEXIST;
5544 goto end;
5545 }
5546
5547 lockfile_fd = create_lockfile();
5548 if (lockfile_fd < 0) {
5549 ret = -EEXIST;
5550 goto end;
5551 }
5552 end:
5553 return ret;
5554 }
5555
5556 static void sessiond_cleanup_lock_file(void)
5557 {
5558 int ret;
5559
5560 /*
5561 * Cleanup lock file by deleting it and finaly closing it which will
5562 * release the file system lock.
5563 */
5564 if (lockfile_fd >= 0) {
5565 ret = remove(config.lock_file_path.value);
5566 if (ret < 0) {
5567 PERROR("remove lock file");
5568 }
5569 ret = close(lockfile_fd);
5570 if (ret < 0) {
5571 PERROR("close lock file");
5572 }
5573 }
5574 }
5575
5576 /*
5577 * Set the tracing group gid onto the client socket.
5578 *
5579 * Race window between mkdir and chown is OK because we are going from more
5580 * permissive (root.root) to less permissive (root.tracing).
5581 */
5582 static int set_permissions(char *rundir)
5583 {
5584 int ret;
5585 gid_t gid;
5586
5587 gid = utils_get_group_id(config.tracing_group_name.value);
5588
5589 /* Set lttng run dir */
5590 ret = chown(rundir, 0, gid);
5591 if (ret < 0) {
5592 ERR("Unable to set group on %s", rundir);
5593 PERROR("chown");
5594 }
5595
5596 /*
5597 * Ensure all applications and tracing group can search the run
5598 * dir. Allow everyone to read the directory, since it does not
5599 * buy us anything to hide its content.
5600 */
5601 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5602 if (ret < 0) {
5603 ERR("Unable to set permissions on %s", rundir);
5604 PERROR("chmod");
5605 }
5606
5607 /* lttng client socket path */
5608 ret = chown(config.client_unix_sock_path.value, 0, gid);
5609 if (ret < 0) {
5610 ERR("Unable to set group on %s", config.client_unix_sock_path.value);
5611 PERROR("chown");
5612 }
5613
5614 /* kconsumer error socket path */
5615 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5616 if (ret < 0) {
5617 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5618 PERROR("chown");
5619 }
5620
5621 /* 64-bit ustconsumer error socket path */
5622 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5623 if (ret < 0) {
5624 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5625 PERROR("chown");
5626 }
5627
5628 /* 32-bit ustconsumer compat32 error socket path */
5629 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5630 if (ret < 0) {
5631 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5632 PERROR("chown");
5633 }
5634
5635 DBG("All permissions are set");
5636
5637 return ret;
5638 }
5639
5640 /*
5641 * Create the lttng run directory needed for all global sockets and pipe.
5642 */
5643 static int create_lttng_rundir(void)
5644 {
5645 int ret;
5646
5647 DBG3("Creating LTTng run directory: %s", config.rundir.value);
5648
5649 ret = mkdir(config.rundir.value, S_IRWXU);
5650 if (ret < 0) {
5651 if (errno != EEXIST) {
5652 ERR("Unable to create %s", config.rundir.value);
5653 goto error;
5654 } else {
5655 ret = 0;
5656 }
5657 }
5658
5659 error:
5660 return ret;
5661 }
5662
5663 /*
5664 * Setup sockets and directory needed by the consumerds' communication with the
5665 * session daemon.
5666 */
5667 static int set_consumer_sockets(struct consumer_data *consumer_data)
5668 {
5669 int ret;
5670 char *path = NULL;
5671
5672 switch (consumer_data->type) {
5673 case LTTNG_CONSUMER_KERNEL:
5674 path = config.kconsumerd_path.value;
5675 break;
5676 case LTTNG_CONSUMER64_UST:
5677 path = config.consumerd64_path.value;
5678 break;
5679 case LTTNG_CONSUMER32_UST:
5680 path = config.consumerd32_path.value;
5681 break;
5682 default:
5683 ERR("Consumer type unknown");
5684 ret = -EINVAL;
5685 goto error;
5686 }
5687 assert(path);
5688
5689 DBG2("Creating consumer directory: %s", path);
5690
5691 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5692 if (ret < 0 && errno != EEXIST) {
5693 PERROR("mkdir");
5694 ERR("Failed to create %s", path);
5695 goto error;
5696 }
5697 if (is_root) {
5698 ret = chown(path, 0, utils_get_group_id(config.tracing_group_name.value));
5699 if (ret < 0) {
5700 ERR("Unable to set group on %s", path);
5701 PERROR("chown");
5702 goto error;
5703 }
5704 }
5705
5706 /* Create the consumerd error unix socket */
5707 consumer_data->err_sock =
5708 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5709 if (consumer_data->err_sock < 0) {
5710 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5711 ret = -1;
5712 goto error;
5713 }
5714
5715 /*
5716 * Set the CLOEXEC flag. Return code is useless because either way, the
5717 * show must go on.
5718 */
5719 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5720 if (ret < 0) {
5721 PERROR("utils_set_fd_cloexec");
5722 /* continue anyway */
5723 }
5724
5725 /* File permission MUST be 660 */
5726 ret = chmod(consumer_data->err_unix_sock_path,
5727 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5728 if (ret < 0) {
5729 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5730 PERROR("chmod");
5731 goto error;
5732 }
5733
5734 error:
5735 return ret;
5736 }
5737
5738 /*
5739 * Signal handler for the daemon
5740 *
5741 * Simply stop all worker threads, leaving main() return gracefully after
5742 * joining all threads and calling cleanup().
5743 */
5744 static void sighandler(int sig)
5745 {
5746 switch (sig) {
5747 case SIGINT:
5748 DBG("SIGINT caught");
5749 stop_threads();
5750 break;
5751 case SIGTERM:
5752 DBG("SIGTERM caught");
5753 stop_threads();
5754 break;
5755 case SIGUSR1:
5756 CMM_STORE_SHARED(recv_child_signal, 1);
5757 break;
5758 default:
5759 break;
5760 }
5761 }
5762
5763 /*
5764 * Setup signal handler for :
5765 * SIGINT, SIGTERM, SIGPIPE
5766 */
5767 static int set_signal_handler(void)
5768 {
5769 int ret = 0;
5770 struct sigaction sa;
5771 sigset_t sigset;
5772
5773 if ((ret = sigemptyset(&sigset)) < 0) {
5774 PERROR("sigemptyset");
5775 return ret;
5776 }
5777
5778 sa.sa_mask = sigset;
5779 sa.sa_flags = 0;
5780
5781 sa.sa_handler = sighandler;
5782 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5783 PERROR("sigaction");
5784 return ret;
5785 }
5786
5787 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5788 PERROR("sigaction");
5789 return ret;
5790 }
5791
5792 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5793 PERROR("sigaction");
5794 return ret;
5795 }
5796
5797 sa.sa_handler = SIG_IGN;
5798 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5799 PERROR("sigaction");
5800 return ret;
5801 }
5802
5803 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5804
5805 return ret;
5806 }
5807
5808 /*
5809 * Set open files limit to unlimited. This daemon can open a large number of
5810 * file descriptors in order to consume multiple kernel traces.
5811 */
5812 static void set_ulimit(void)
5813 {
5814 int ret;
5815 struct rlimit lim;
5816
5817 /* The kernel does not allow an infinite limit for open files */
5818 lim.rlim_cur = 65535;
5819 lim.rlim_max = 65535;
5820
5821 ret = setrlimit(RLIMIT_NOFILE, &lim);
5822 if (ret < 0) {
5823 PERROR("failed to set open files limit");
5824 }
5825 }
5826
5827 static int write_pidfile(void)
5828 {
5829 return utils_create_pid_file(getpid(), config.pid_file_path.value);
5830 }
5831
5832 static int set_clock_plugin_env(void)
5833 {
5834 int ret = 0;
5835 char *env_value = NULL;
5836
5837 if (!config.lttng_ust_clock_plugin.value) {
5838 goto end;
5839 }
5840
5841 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
5842 config.lttng_ust_clock_plugin.value);
5843 if (ret < 0) {
5844 PERROR("asprintf");
5845 goto end;
5846 }
5847
5848 ret = putenv(env_value);
5849 if (ret) {
5850 free(env_value);
5851 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5852 goto end;
5853 }
5854
5855 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5856 config.lttng_ust_clock_plugin.value);
5857 end:
5858 return ret;
5859 }
5860
5861 static
5862 struct rotation_thread_timer_queue *create_rotate_timer_queue(void)
5863 {
5864 struct rotation_thread_timer_queue *queue = NULL;
5865
5866 queue = zmalloc(sizeof(struct rotation_thread_timer_queue));
5867 if (!queue) {
5868 PERROR("Failed to allocate timer rotate queue");
5869 goto end;
5870 }
5871
5872 queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK);
5873 CDS_INIT_LIST_HEAD(&queue->list);
5874 pthread_mutex_init(&queue->lock, NULL);
5875
5876 end:
5877 return queue;
5878 }
5879
5880 static
5881 void destroy_rotate_timer_queue(struct rotation_thread_timer_queue *queue)
5882 {
5883 struct sessiond_rotation_timer *node, *tmp_node;
5884
5885 if (!queue) {
5886 return;
5887 }
5888
5889 lttng_pipe_destroy(queue->event_pipe);
5890
5891 pthread_mutex_lock(&queue->lock);
5892 /* Empty wait queue. */
5893 cds_list_for_each_entry_safe(node, tmp_node, &queue->list, head) {
5894 cds_list_del(&node->head);
5895 free(node);
5896 }
5897 pthread_mutex_unlock(&queue->lock);
5898
5899 pthread_mutex_destroy(&queue->lock);
5900 free(queue);
5901 }
5902
5903 /*
5904 * main
5905 */
5906 int main(int argc, char **argv)
5907 {
5908 int ret = 0, retval = 0;
5909 void *status;
5910 const char *env_app_timeout;
5911 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5912 *ust64_channel_monitor_pipe = NULL,
5913 *kernel_channel_monitor_pipe = NULL;
5914 bool notification_thread_launched = false;
5915 bool rotation_thread_launched = false;
5916 bool timer_thread_launched = false;
5917 struct lttng_pipe *ust32_channel_rotate_pipe = NULL,
5918 *ust64_channel_rotate_pipe = NULL,
5919 *kernel_channel_rotate_pipe = NULL;
5920 struct timer_thread_parameters timer_thread_ctx;
5921 /* Queue of rotation jobs populated by the sessiond-timer. */
5922 struct rotation_thread_timer_queue *rotation_timer_queue = NULL;
5923 sem_t notification_thread_ready;
5924
5925 init_kernel_workarounds();
5926
5927 rcu_register_thread();
5928
5929 if (set_signal_handler()) {
5930 retval = -1;
5931 goto exit_set_signal_handler;
5932 }
5933
5934 if (sessiond_timer_signal_init()) {
5935 retval = -1;
5936 goto exit_set_signal_handler;
5937 }
5938
5939 page_size = sysconf(_SC_PAGESIZE);
5940 if (page_size < 0) {
5941 PERROR("sysconf _SC_PAGESIZE");
5942 page_size = LONG_MAX;
5943 WARN("Fallback page size to %ld", page_size);
5944 }
5945
5946 ret = sessiond_config_init(&config);
5947 if (ret) {
5948 retval = -1;
5949 goto exit_set_signal_handler;
5950 }
5951
5952 /*
5953 * Init config from environment variables.
5954 * Command line option override env configuration per-doc. Do env first.
5955 */
5956 sessiond_config_apply_env_config(&config);
5957
5958 /*
5959 * Parse arguments and load the daemon configuration file.
5960 *
5961 * We have an exit_options exit path to free memory reserved by
5962 * set_options. This is needed because the rest of sessiond_cleanup()
5963 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5964 * depends on set_options.
5965 */
5966 progname = argv[0];
5967 if (set_options(argc, argv)) {
5968 retval = -1;
5969 goto exit_options;
5970 }
5971
5972 /*
5973 * Resolve all paths received as arguments, configuration option, or
5974 * through environment variable as absolute paths. This is necessary
5975 * since daemonizing causes the sessiond's current working directory
5976 * to '/'.
5977 */
5978 ret = sessiond_config_resolve_paths(&config);
5979 if (ret) {
5980 goto exit_options;
5981 }
5982
5983 /* Apply config. */
5984 lttng_opt_verbose = config.verbose;
5985 lttng_opt_quiet = config.quiet;
5986 kconsumer_data.err_unix_sock_path =
5987 config.kconsumerd_err_unix_sock_path.value;
5988 kconsumer_data.cmd_unix_sock_path =
5989 config.kconsumerd_cmd_unix_sock_path.value;
5990 ustconsumer32_data.err_unix_sock_path =
5991 config.consumerd32_err_unix_sock_path.value;
5992 ustconsumer32_data.cmd_unix_sock_path =
5993 config.consumerd32_cmd_unix_sock_path.value;
5994 ustconsumer64_data.err_unix_sock_path =
5995 config.consumerd64_err_unix_sock_path.value;
5996 ustconsumer64_data.cmd_unix_sock_path =
5997 config.consumerd64_cmd_unix_sock_path.value;
5998 set_clock_plugin_env();
5999
6000 sessiond_config_log(&config);
6001
6002 if (create_lttng_rundir()) {
6003 retval = -1;
6004 goto exit_options;
6005 }
6006
6007 /* Abort launch if a session daemon is already running. */
6008 if (check_existing_daemon()) {
6009 ERR("A session daemon is already running.");
6010 retval = -1;
6011 goto exit_options;
6012 }
6013
6014 /* Daemonize */
6015 if (config.daemonize || config.background) {
6016 int i;
6017
6018 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
6019 !config.background);
6020 if (ret < 0) {
6021 retval = -1;
6022 goto exit_options;
6023 }
6024
6025 /*
6026 * We are in the child. Make sure all other file descriptors are
6027 * closed, in case we are called with more opened file
6028 * descriptors than the standard ones and the lock file.
6029 */
6030 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
6031 if (i == lockfile_fd) {
6032 continue;
6033 }
6034 (void) close(i);
6035 }
6036 }
6037
6038 if (run_as_create_worker(argv[0]) < 0) {
6039 goto exit_create_run_as_worker_cleanup;
6040 }
6041
6042 /*
6043 * Starting from here, we can create threads. This needs to be after
6044 * lttng_daemonize due to RCU.
6045 */
6046
6047 /*
6048 * Initialize the health check subsystem. This call should set the
6049 * appropriate time values.
6050 */
6051 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
6052 if (!health_sessiond) {
6053 PERROR("health_app_create error");
6054 retval = -1;
6055 goto exit_health_sessiond_cleanup;
6056 }
6057
6058 /* Create thread to clean up RCU hash tables */
6059 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
6060 retval = -1;
6061 goto exit_ht_cleanup;
6062 }
6063
6064 /* Create thread quit pipe */
6065 if (init_thread_quit_pipe()) {
6066 retval = -1;
6067 goto exit_init_data;
6068 }
6069
6070 /* Check if daemon is UID = 0 */
6071 is_root = !getuid();
6072 if (is_root) {
6073 /* Create global run dir with root access */
6074
6075 kernel_channel_monitor_pipe = lttng_pipe_open(0);
6076 if (!kernel_channel_monitor_pipe) {
6077 ERR("Failed to create kernel consumer channel monitor pipe");
6078 retval = -1;
6079 goto exit_init_data;
6080 }
6081 kconsumer_data.channel_monitor_pipe =
6082 lttng_pipe_release_writefd(
6083 kernel_channel_monitor_pipe);
6084 if (kconsumer_data.channel_monitor_pipe < 0) {
6085 retval = -1;
6086 goto exit_init_data;
6087 }
6088 kernel_channel_rotate_pipe = lttng_pipe_open(0);
6089 if (!kernel_channel_rotate_pipe) {
6090 ERR("Failed to create kernel consumer channel rotate pipe");
6091 retval = -1;
6092 goto exit_init_data;
6093 }
6094 kconsumer_data.channel_rotate_pipe =
6095 lttng_pipe_release_writefd(
6096 kernel_channel_rotate_pipe);
6097 if (kconsumer_data.channel_rotate_pipe < 0) {
6098 retval = -1;
6099 goto exit_init_data;
6100 }
6101 }
6102
6103 /* Set consumer initial state */
6104 kernel_consumerd_state = CONSUMER_STOPPED;
6105 ust_consumerd_state = CONSUMER_STOPPED;
6106
6107 ust32_channel_monitor_pipe = lttng_pipe_open(0);
6108 if (!ust32_channel_monitor_pipe) {
6109 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
6110 retval = -1;
6111 goto exit_init_data;
6112 }
6113 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
6114 ust32_channel_monitor_pipe);
6115 if (ustconsumer32_data.channel_monitor_pipe < 0) {
6116 retval = -1;
6117 goto exit_init_data;
6118 }
6119 ust32_channel_rotate_pipe = lttng_pipe_open(0);
6120 if (!ust32_channel_rotate_pipe) {
6121 ERR("Failed to create 32-bit user space consumer channel rotate pipe");
6122 retval = -1;
6123 goto exit_init_data;
6124 }
6125 ustconsumer32_data.channel_rotate_pipe = lttng_pipe_release_writefd(
6126 ust32_channel_rotate_pipe);
6127 if (ustconsumer32_data.channel_rotate_pipe < 0) {
6128 retval = -1;
6129 goto exit_init_data;
6130 }
6131
6132 /*
6133 * The rotation_timer_queue structure is shared between the sessiond timer
6134 * thread and the rotation thread. The main() keeps the ownership and
6135 * destroys it when both threads have quit.
6136 */
6137 rotation_timer_queue = create_rotate_timer_queue();
6138 if (!rotation_timer_queue) {
6139 retval = -1;
6140 goto exit_init_data;
6141 }
6142 timer_thread_ctx.rotation_timer_queue = rotation_timer_queue;
6143
6144 ust64_channel_monitor_pipe = lttng_pipe_open(0);
6145 if (!ust64_channel_monitor_pipe) {
6146 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
6147 retval = -1;
6148 goto exit_init_data;
6149 }
6150 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
6151 ust64_channel_monitor_pipe);
6152 if (ustconsumer64_data.channel_monitor_pipe < 0) {
6153 retval = -1;
6154 goto exit_init_data;
6155 }
6156 ust64_channel_rotate_pipe = lttng_pipe_open(0);
6157 if (!ust64_channel_rotate_pipe) {
6158 ERR("Failed to create 64-bit user space consumer channel rotate pipe");
6159 retval = -1;
6160 goto exit_init_data;
6161 }
6162 ustconsumer64_data.channel_rotate_pipe = lttng_pipe_release_writefd(
6163 ust64_channel_rotate_pipe);
6164 if (ustconsumer64_data.channel_rotate_pipe < 0) {
6165 retval = -1;
6166 goto exit_init_data;
6167 }
6168
6169 /*
6170 * Init UST app hash table. Alloc hash table before this point since
6171 * cleanup() can get called after that point.
6172 */
6173 if (ust_app_ht_alloc()) {
6174 ERR("Failed to allocate UST app hash table");
6175 retval = -1;
6176 goto exit_init_data;
6177 }
6178
6179 /*
6180 * Initialize agent app hash table. We allocate the hash table here
6181 * since cleanup() can get called after this point.
6182 */
6183 if (agent_app_ht_alloc()) {
6184 ERR("Failed to allocate Agent app hash table");
6185 retval = -1;
6186 goto exit_init_data;
6187 }
6188
6189 /*
6190 * These actions must be executed as root. We do that *after* setting up
6191 * the sockets path because we MUST make the check for another daemon using
6192 * those paths *before* trying to set the kernel consumer sockets and init
6193 * kernel tracer.
6194 */
6195 if (is_root) {
6196 if (set_consumer_sockets(&kconsumer_data)) {
6197 retval = -1;
6198 goto exit_init_data;
6199 }
6200
6201 /* Setup kernel tracer */
6202 if (!config.no_kernel) {
6203 init_kernel_tracer();
6204 if (kernel_tracer_fd >= 0) {
6205 ret = syscall_init_table();
6206 if (ret < 0) {
6207 ERR("Unable to populate syscall table. "
6208 "Syscall tracing won't work "
6209 "for this session daemon.");
6210 }
6211 }
6212 }
6213
6214 /* Set ulimit for open files */
6215 set_ulimit();
6216 }
6217 /* init lttng_fd tracking must be done after set_ulimit. */
6218 lttng_fd_init();
6219
6220 if (set_consumer_sockets(&ustconsumer64_data)) {
6221 retval = -1;
6222 goto exit_init_data;
6223 }
6224
6225 if (set_consumer_sockets(&ustconsumer32_data)) {
6226 retval = -1;
6227 goto exit_init_data;
6228 }
6229
6230 /* Setup the needed unix socket */
6231 if (init_daemon_socket()) {
6232 retval = -1;
6233 goto exit_init_data;
6234 }
6235
6236 /* Set credentials to socket */
6237 if (is_root && set_permissions(config.rundir.value)) {
6238 retval = -1;
6239 goto exit_init_data;
6240 }
6241
6242 /* Get parent pid if -S, --sig-parent is specified. */
6243 if (config.sig_parent) {
6244 ppid = getppid();
6245 }
6246
6247 /* Setup the kernel pipe for waking up the kernel thread */
6248 if (is_root && !config.no_kernel) {
6249 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
6250 retval = -1;
6251 goto exit_init_data;
6252 }
6253 }
6254
6255 /* Setup the thread apps communication pipe. */
6256 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
6257 retval = -1;
6258 goto exit_init_data;
6259 }
6260
6261 /* Setup the thread apps notify communication pipe. */
6262 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
6263 retval = -1;
6264 goto exit_init_data;
6265 }
6266
6267 /* Initialize global buffer per UID and PID registry. */
6268 buffer_reg_init_uid_registry();
6269 buffer_reg_init_pid_registry();
6270
6271 /* Init UST command queue. */
6272 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
6273
6274 /*
6275 * Get session list pointer. This pointer MUST NOT be free'd. This list
6276 * is statically declared in session.c
6277 */
6278 session_list_ptr = session_get_list();
6279
6280 cmd_init();
6281
6282 /* Check for the application socket timeout env variable. */
6283 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
6284 if (env_app_timeout) {
6285 config.app_socket_timeout = atoi(env_app_timeout);
6286 } else {
6287 config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
6288 }
6289
6290 ret = write_pidfile();
6291 if (ret) {
6292 ERR("Error in write_pidfile");
6293 retval = -1;
6294 goto exit_init_data;
6295 }
6296
6297 /* Initialize communication library */
6298 lttcomm_init();
6299 /* Initialize TCP timeout values */
6300 lttcomm_inet_init();
6301
6302 if (load_session_init_data(&load_info) < 0) {
6303 retval = -1;
6304 goto exit_init_data;
6305 }
6306 load_info->path = config.load_session_path.value;
6307
6308 /* Create health-check thread. */
6309 ret = pthread_create(&health_thread, default_pthread_attr(),
6310 thread_manage_health, (void *) NULL);
6311 if (ret) {
6312 errno = ret;
6313 PERROR("pthread_create health");
6314 retval = -1;
6315 goto exit_health;
6316 }
6317
6318 /*
6319 * The rotation thread needs the notification thread to be ready before
6320 * creating the rotate_notification_channel, so we use this semaphore as
6321 * a rendez-vous point.
6322 */
6323 sem_init(&notification_thread_ready, 0, 0);
6324
6325 /* notification_thread_data acquires the pipes' read side. */
6326 notification_thread_handle = notification_thread_handle_create(
6327 ust32_channel_monitor_pipe,
6328 ust64_channel_monitor_pipe,
6329 kernel_channel_monitor_pipe,
6330 &notification_thread_ready);
6331 if (!notification_thread_handle) {
6332 retval = -1;
6333 ERR("Failed to create notification thread shared data");
6334 stop_threads();
6335 goto exit_notification;
6336 }
6337
6338 /* Create notification thread. */
6339 ret = pthread_create(&notification_thread, default_pthread_attr(),
6340 thread_notification, notification_thread_handle);
6341 if (ret) {
6342 errno = ret;
6343 PERROR("pthread_create notification");
6344 retval = -1;
6345 stop_threads();
6346 goto exit_notification;
6347 }
6348 notification_thread_launched = true;
6349
6350 /* Create timer thread. */
6351 ret = pthread_create(&timer_thread, default_pthread_attr(),
6352 sessiond_timer_thread, &timer_thread_ctx);
6353 if (ret) {
6354 errno = ret;
6355 PERROR("pthread_create timer");
6356 retval = -1;
6357 stop_threads();
6358 goto exit_notification;
6359 }
6360 timer_thread_launched = true;
6361
6362 /* rotation_thread_data acquires the pipes' read side. */
6363 rotation_thread_handle = rotation_thread_handle_create(
6364 ust32_channel_rotate_pipe,
6365 ust64_channel_rotate_pipe,
6366 kernel_channel_rotate_pipe,
6367 thread_quit_pipe[0],
6368 rotation_timer_queue,
6369 notification_thread_handle,
6370 &notification_thread_ready);
6371 if (!rotation_thread_handle) {
6372 retval = -1;
6373 ERR("Failed to create rotation thread shared data");
6374 stop_threads();
6375 goto exit_rotation;
6376 }
6377
6378 /* Create rotation thread. */
6379 ret = pthread_create(&rotation_thread, default_pthread_attr(),
6380 thread_rotation, rotation_thread_handle);
6381 if (ret) {
6382 errno = ret;
6383 PERROR("pthread_create rotation");
6384 retval = -1;
6385 stop_threads();
6386 goto exit_rotation;
6387 }
6388 rotation_thread_launched = true;
6389
6390 /* Create thread to manage the client socket */
6391 ret = pthread_create(&client_thread, default_pthread_attr(),
6392 thread_manage_clients, (void *) NULL);
6393 if (ret) {
6394 errno = ret;
6395 PERROR("pthread_create clients");
6396 retval = -1;
6397 stop_threads();
6398 goto exit_client;
6399 }
6400
6401 /* Create thread to dispatch registration */
6402 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
6403 thread_dispatch_ust_registration, (void *) NULL);
6404 if (ret) {
6405 errno = ret;
6406 PERROR("pthread_create dispatch");
6407 retval = -1;
6408 stop_threads();
6409 goto exit_dispatch;
6410 }
6411
6412 /* Create thread to manage application registration. */
6413 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
6414 thread_registration_apps, (void *) NULL);
6415 if (ret) {
6416 errno = ret;
6417 PERROR("pthread_create registration");
6418 retval = -1;
6419 stop_threads();
6420 goto exit_reg_apps;
6421 }
6422
6423 /* Create thread to manage application socket */
6424 ret = pthread_create(&apps_thread, default_pthread_attr(),
6425 thread_manage_apps, (void *) NULL);
6426 if (ret) {
6427 errno = ret;
6428 PERROR("pthread_create apps");
6429 retval = -1;
6430 stop_threads();
6431 goto exit_apps;
6432 }
6433
6434 /* Create thread to manage application notify socket */
6435 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
6436 ust_thread_manage_notify, (void *) NULL);
6437 if (ret) {
6438 errno = ret;
6439 PERROR("pthread_create notify");
6440 retval = -1;
6441 stop_threads();
6442 goto exit_apps_notify;
6443 }
6444
6445 /* Create agent registration thread. */
6446 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
6447 agent_thread_manage_registration, (void *) NULL);
6448 if (ret) {
6449 errno = ret;
6450 PERROR("pthread_create agent");
6451 retval = -1;
6452 stop_threads();
6453 goto exit_agent_reg;
6454 }
6455
6456 /* Don't start this thread if kernel tracing is not requested nor root */
6457 if (is_root && !config.no_kernel) {
6458 /* Create kernel thread to manage kernel event */
6459 ret = pthread_create(&kernel_thread, default_pthread_attr(),
6460 thread_manage_kernel, (void *) NULL);
6461 if (ret) {
6462 errno = ret;
6463 PERROR("pthread_create kernel");
6464 retval = -1;
6465 stop_threads();
6466 goto exit_kernel;
6467 }
6468 }
6469
6470 /* Create session loading thread. */
6471 ret = pthread_create(&load_session_thread, default_pthread_attr(),
6472 thread_load_session, load_info);
6473 if (ret) {
6474 errno = ret;
6475 PERROR("pthread_create load_session_thread");
6476 retval = -1;
6477 stop_threads();
6478 goto exit_load_session;
6479 }
6480
6481 /*
6482 * This is where we start awaiting program completion (e.g. through
6483 * signal that asks threads to teardown).
6484 */
6485
6486 ret = pthread_join(load_session_thread, &status);
6487 if (ret) {
6488 errno = ret;
6489 PERROR("pthread_join load_session_thread");
6490 retval = -1;
6491 }
6492 exit_load_session:
6493
6494 if (is_root && !config.no_kernel) {
6495 ret = pthread_join(kernel_thread, &status);
6496 if (ret) {
6497 errno = ret;
6498 PERROR("pthread_join");
6499 retval = -1;
6500 }
6501 }
6502 exit_kernel:
6503
6504 ret = pthread_join(agent_reg_thread, &status);
6505 if (ret) {
6506 errno = ret;
6507 PERROR("pthread_join agent");
6508 retval = -1;
6509 }
6510 exit_agent_reg:
6511
6512 ret = pthread_join(apps_notify_thread, &status);
6513 if (ret) {
6514 errno = ret;
6515 PERROR("pthread_join apps notify");
6516 retval = -1;
6517 }
6518 exit_apps_notify:
6519
6520 ret = pthread_join(apps_thread, &status);
6521 if (ret) {
6522 errno = ret;
6523 PERROR("pthread_join apps");
6524 retval = -1;
6525 }
6526 exit_apps:
6527
6528 ret = pthread_join(reg_apps_thread, &status);
6529 if (ret) {
6530 errno = ret;
6531 PERROR("pthread_join");
6532 retval = -1;
6533 }
6534 exit_reg_apps:
6535
6536 /*
6537 * Join dispatch thread after joining reg_apps_thread to ensure
6538 * we don't leak applications in the queue.
6539 */
6540 ret = pthread_join(dispatch_thread, &status);
6541 if (ret) {
6542 errno = ret;
6543 PERROR("pthread_join");
6544 retval = -1;
6545 }
6546 exit_dispatch:
6547
6548 ret = pthread_join(client_thread, &status);
6549 if (ret) {
6550 errno = ret;
6551 PERROR("pthread_join");
6552 retval = -1;
6553 }
6554
6555 exit_client:
6556 exit_rotation:
6557 exit_notification:
6558 sem_destroy(&notification_thread_ready);
6559 ret = pthread_join(health_thread, &status);
6560 if (ret) {
6561 errno = ret;
6562 PERROR("pthread_join health thread");
6563 retval = -1;
6564 }
6565
6566 exit_health:
6567 exit_init_data:
6568 /*
6569 * Wait for all pending call_rcu work to complete before tearing
6570 * down data structures. call_rcu worker may be trying to
6571 * perform lookups in those structures.
6572 */
6573 rcu_barrier();
6574 /*
6575 * sessiond_cleanup() is called when no other thread is running, except
6576 * the ht_cleanup thread, which is needed to destroy the hash tables.
6577 */
6578 rcu_thread_online();
6579 sessiond_cleanup();
6580
6581 /*
6582 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6583 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6584 * the queue is empty before shutting down the clean-up thread.
6585 */
6586 rcu_barrier();
6587
6588 /*
6589 * The teardown of the notification system is performed after the
6590 * session daemon's teardown in order to allow it to be notified
6591 * of the active session and channels at the moment of the teardown.
6592 */
6593 if (notification_thread_handle) {
6594 if (notification_thread_launched) {
6595 notification_thread_command_quit(
6596 notification_thread_handle);
6597 ret = pthread_join(notification_thread, &status);
6598 if (ret) {
6599 errno = ret;
6600 PERROR("pthread_join notification thread");
6601 retval = -1;
6602 }
6603 }
6604 notification_thread_handle_destroy(notification_thread_handle);
6605 }
6606
6607 if (rotation_thread_handle) {
6608 if (rotation_thread_launched) {
6609 ret = pthread_join(rotation_thread, &status);
6610 if (ret) {
6611 errno = ret;
6612 PERROR("pthread_join rotation thread");
6613 retval = -1;
6614 }
6615 }
6616 rotation_thread_handle_destroy(rotation_thread_handle);
6617 }
6618
6619 if (timer_thread_launched) {
6620 kill(getpid(), LTTNG_SESSIOND_SIG_EXIT);
6621 ret = pthread_join(timer_thread, &status);
6622 if (ret) {
6623 errno = ret;
6624 PERROR("pthread_join timer thread");
6625 retval = -1;
6626 }
6627 }
6628
6629 /*
6630 * After the rotation and timer thread have quit, we can safely destroy
6631 * the rotation_timer_queue.
6632 */
6633 destroy_rotate_timer_queue(rotation_timer_queue);
6634
6635 rcu_thread_offline();
6636 rcu_unregister_thread();
6637
6638 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6639 if (ret) {
6640 retval = -1;
6641 }
6642 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6643 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6644 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6645 lttng_pipe_destroy(ust32_channel_rotate_pipe);
6646 lttng_pipe_destroy(ust64_channel_rotate_pipe);
6647 lttng_pipe_destroy(kernel_channel_rotate_pipe);
6648 exit_ht_cleanup:
6649
6650 health_app_destroy(health_sessiond);
6651 exit_health_sessiond_cleanup:
6652 exit_create_run_as_worker_cleanup:
6653
6654 exit_options:
6655 sessiond_cleanup_lock_file();
6656 sessiond_cleanup_options();
6657
6658 exit_set_signal_handler:
6659 if (!retval) {
6660 exit(EXIT_SUCCESS);
6661 } else {
6662 exit(EXIT_FAILURE);
6663 }
6664 }
This page took 0.239884 seconds and 6 git commands to generate.