ee50db21e38b2c29a400b1ccec0065f72f28900c
[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
55 #include "lttng-sessiond.h"
56 #include "buffer-registry.h"
57 #include "channel.h"
58 #include "cmd.h"
59 #include "consumer.h"
60 #include "context.h"
61 #include "event.h"
62 #include "kernel.h"
63 #include "kernel-consumer.h"
64 #include "modprobe.h"
65 #include "shm.h"
66 #include "ust-ctl.h"
67 #include "ust-consumer.h"
68 #include "utils.h"
69 #include "fd-limit.h"
70 #include "health-sessiond.h"
71 #include "testpoint.h"
72 #include "ust-thread.h"
73 #include "agent-thread.h"
74 #include "save.h"
75 #include "load-session-thread.h"
76 #include "notification-thread.h"
77 #include "notification-thread-commands.h"
78 #include "rotation-thread.h"
79 #include "lttng-syscall.h"
80 #include "agent.h"
81 #include "ht-cleanup.h"
82 #include "sessiond-config.h"
83 #include "sessiond-timer.h"
84
85 static const char *help_msg =
86 #ifdef LTTNG_EMBED_HELP
87 #include <lttng-sessiond.8.h>
88 #else
89 NULL
90 #endif
91 ;
92
93 const char *progname;
94 static pid_t ppid; /* Parent PID for --sig-parent option */
95 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
96 static int lockfile_fd = -1;
97
98 /* Set to 1 when a SIGUSR1 signal is received. */
99 static int recv_child_signal;
100
101 static struct lttng_kernel_tracer_version kernel_tracer_version;
102 static struct lttng_kernel_tracer_abi_version kernel_tracer_abi_version;
103
104 /*
105 * Consumer daemon specific control data. Every value not initialized here is
106 * set to 0 by the static definition.
107 */
108 static struct consumer_data kconsumer_data = {
109 .type = LTTNG_CONSUMER_KERNEL,
110 .err_sock = -1,
111 .cmd_sock = -1,
112 .channel_monitor_pipe = -1,
113 .channel_rotate_pipe = -1,
114 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
115 .lock = PTHREAD_MUTEX_INITIALIZER,
116 .cond = PTHREAD_COND_INITIALIZER,
117 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
118 };
119 static struct consumer_data ustconsumer64_data = {
120 .type = LTTNG_CONSUMER64_UST,
121 .err_sock = -1,
122 .cmd_sock = -1,
123 .channel_monitor_pipe = -1,
124 .channel_rotate_pipe = -1,
125 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
126 .lock = PTHREAD_MUTEX_INITIALIZER,
127 .cond = PTHREAD_COND_INITIALIZER,
128 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
129 };
130 static struct consumer_data ustconsumer32_data = {
131 .type = LTTNG_CONSUMER32_UST,
132 .err_sock = -1,
133 .cmd_sock = -1,
134 .channel_monitor_pipe = -1,
135 .channel_rotate_pipe = -1,
136 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
137 .lock = PTHREAD_MUTEX_INITIALIZER,
138 .cond = PTHREAD_COND_INITIALIZER,
139 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
140 };
141
142 /* Command line options */
143 static const struct option long_options[] = {
144 { "client-sock", required_argument, 0, 'c' },
145 { "apps-sock", required_argument, 0, 'a' },
146 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
147 { "kconsumerd-err-sock", required_argument, 0, '\0' },
148 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
149 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
150 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
151 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
152 { "consumerd32-path", required_argument, 0, '\0' },
153 { "consumerd32-libdir", required_argument, 0, '\0' },
154 { "consumerd64-path", required_argument, 0, '\0' },
155 { "consumerd64-libdir", required_argument, 0, '\0' },
156 { "daemonize", no_argument, 0, 'd' },
157 { "background", no_argument, 0, 'b' },
158 { "sig-parent", no_argument, 0, 'S' },
159 { "help", no_argument, 0, 'h' },
160 { "group", required_argument, 0, 'g' },
161 { "version", no_argument, 0, 'V' },
162 { "quiet", no_argument, 0, 'q' },
163 { "verbose", no_argument, 0, 'v' },
164 { "verbose-consumer", no_argument, 0, '\0' },
165 { "no-kernel", no_argument, 0, '\0' },
166 { "pidfile", required_argument, 0, 'p' },
167 { "agent-tcp-port", required_argument, 0, '\0' },
168 { "config", required_argument, 0, 'f' },
169 { "load", required_argument, 0, 'l' },
170 { "kmod-probes", required_argument, 0, '\0' },
171 { "extra-kmod-probes", required_argument, 0, '\0' },
172 { NULL, 0, 0, 0 }
173 };
174
175 struct sessiond_config config;
176
177 /* Command line options to ignore from configuration file */
178 static const char *config_ignore_options[] = { "help", "version", "config" };
179
180 /* Shared between threads */
181 static int dispatch_thread_exit;
182
183 /* Sockets and FDs */
184 static int client_sock = -1;
185 static int apps_sock = -1;
186 int kernel_tracer_fd = -1;
187 static int kernel_poll_pipe[2] = { -1, -1 };
188
189 /*
190 * Quit pipe for all threads. This permits a single cancellation point
191 * for all threads when receiving an event on the pipe.
192 */
193 static int thread_quit_pipe[2] = { -1, -1 };
194
195 /*
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
198 */
199 static int apps_cmd_pipe[2] = { -1, -1 };
200
201 int apps_cmd_notify_pipe[2] = { -1, -1 };
202
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread;
205 static pthread_t apps_notify_thread;
206 static pthread_t reg_apps_thread;
207 static pthread_t client_thread;
208 static pthread_t kernel_thread;
209 static pthread_t dispatch_thread;
210 static pthread_t health_thread;
211 static pthread_t ht_cleanup_thread;
212 static pthread_t agent_reg_thread;
213 static pthread_t load_session_thread;
214 static pthread_t notification_thread;
215 static pthread_t rotation_thread;
216 static pthread_t timer_thread;
217
218 /*
219 * UST registration command queue. This queue is tied with a futex and uses a N
220 * wakers / 1 waiter implemented and detailed in futex.c/.h
221 *
222 * The thread_registration_apps and thread_dispatch_ust_registration uses this
223 * queue along with the wait/wake scheme. The thread_manage_apps receives down
224 * the line new application socket and monitors it for any I/O error or clean
225 * close that triggers an unregistration of the application.
226 */
227 static struct ust_cmd_queue ust_cmd_queue;
228
229 /*
230 * Pointer initialized before thread creation.
231 *
232 * This points to the tracing session list containing the session count and a
233 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
234 * MUST NOT be taken if you call a public function in session.c.
235 *
236 * The lock is nested inside the structure: session_list_ptr->lock. Please use
237 * session_lock_list and session_unlock_list for lock acquisition.
238 */
239 static struct ltt_session_list *session_list_ptr;
240
241 int ust_consumerd64_fd = -1;
242 int ust_consumerd32_fd = -1;
243
244 static const char *module_proc_lttng = "/proc/lttng";
245
246 /*
247 * Consumer daemon state which is changed when spawning it, killing it or in
248 * case of a fatal error.
249 */
250 enum consumerd_state {
251 CONSUMER_STARTED = 1,
252 CONSUMER_STOPPED = 2,
253 CONSUMER_ERROR = 3,
254 };
255
256 /*
257 * This consumer daemon state is used to validate if a client command will be
258 * able to reach the consumer. If not, the client is informed. For instance,
259 * doing a "lttng start" when the consumer state is set to ERROR will return an
260 * error to the client.
261 *
262 * The following example shows a possible race condition of this scheme:
263 *
264 * consumer thread error happens
265 * client cmd arrives
266 * client cmd checks state -> still OK
267 * consumer thread exit, sets error
268 * client cmd try to talk to consumer
269 * ...
270 *
271 * However, since the consumer is a different daemon, we have no way of making
272 * sure the command will reach it safely even with this state flag. This is why
273 * we consider that up to the state validation during command processing, the
274 * command is safe. After that, we can not guarantee the correctness of the
275 * client request vis-a-vis the consumer.
276 */
277 static enum consumerd_state ust_consumerd_state;
278 static enum consumerd_state kernel_consumerd_state;
279
280 /* Set in main() with the current page size. */
281 long page_size;
282
283 /* Application health monitoring */
284 struct health_app *health_sessiond;
285
286 /* Am I root or not. */
287 int is_root; /* Set to 1 if the daemon is running as root */
288
289 const char * const config_section_name = "sessiond";
290
291 /* Load session thread information to operate. */
292 struct load_session_thread_data *load_info;
293
294 /* Notification thread handle. */
295 struct notification_thread_handle *notification_thread_handle;
296
297 /* Rotation thread handle. */
298 struct rotation_thread_handle *rotation_thread_handle;
299
300 /* Global hash tables */
301 struct lttng_ht *agent_apps_ht_by_sock = NULL;
302
303 /*
304 * The initialization of the session daemon is done in multiple phases.
305 *
306 * While all threads are launched near-simultaneously, only some of them
307 * are needed to ensure the session daemon can start to respond to client
308 * requests.
309 *
310 * There are two important guarantees that we wish to offer with respect
311 * to the initialisation of the session daemon:
312 * - When the daemonize/background launcher process exits, the sessiond
313 * is fully able to respond to client requests,
314 * - Auto-loaded sessions are visible to clients.
315 *
316 * In order to achieve this, a number of support threads have to be launched
317 * to allow the "client" thread to function properly. Moreover, since the
318 * "load session" thread needs the client thread, we must provide a way
319 * for the "load session" thread to know that the "client" thread is up
320 * and running.
321 *
322 * Hence, the support threads decrement the lttng_sessiond_ready counter
323 * while the "client" threads waits for it to reach 0. Once the "client" thread
324 * unblocks, it posts the message_thread_ready semaphore which allows the
325 * "load session" thread to progress.
326 *
327 * This implies that the "load session" thread is the last to be initialized
328 * and will explicitly call sessiond_signal_parents(), which signals the parents
329 * that the session daemon is fully initialized.
330 *
331 * The four (4) support threads are:
332 * - agent_thread
333 * - notification_thread
334 * - rotation_thread
335 * - health_thread
336 */
337 #define NR_LTTNG_SESSIOND_SUPPORT_THREADS 4
338 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_SUPPORT_THREADS;
339
340 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
341 {
342 return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0;
343 }
344
345 /* Notify parents that we are ready for cmd and health check */
346 LTTNG_HIDDEN
347 void sessiond_signal_parents(void)
348 {
349 /*
350 * Notify parent pid that we are ready to accept command
351 * for client side. This ppid is the one from the
352 * external process that spawned us.
353 */
354 if (config.sig_parent) {
355 kill(ppid, SIGUSR1);
356 }
357
358 /*
359 * Notify the parent of the fork() process that we are
360 * ready.
361 */
362 if (config.daemonize || config.background) {
363 kill(child_ppid, SIGUSR1);
364 }
365 }
366
367 LTTNG_HIDDEN
368 void sessiond_notify_ready(void)
369 {
370 /*
371 * This memory barrier is paired with the one performed by
372 * the client thread after it has seen that 'lttng_sessiond_ready' is 0.
373 *
374 * The purpose of these memory barriers is to ensure that all
375 * initialization operations of the various threads that call this
376 * function to signal that they are ready are commited/published
377 * before the client thread can see the 'lttng_sessiond_ready' counter
378 * reach 0.
379 *
380 * Note that this could be a 'write' memory barrier, but a full barrier
381 * is used in case the code using this utility changes. The performance
382 * implications of this choice are minimal since this is a slow path.
383 */
384 cmm_smp_mb();
385 uatomic_sub(&lttng_sessiond_ready, 1);
386 }
387
388 static
389 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
390 int *a_pipe)
391 {
392 int ret;
393
394 assert(events);
395
396 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
397 if (ret < 0) {
398 goto error;
399 }
400
401 /* Add quit pipe */
402 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
403 if (ret < 0) {
404 goto error;
405 }
406
407 return 0;
408
409 error:
410 return ret;
411 }
412
413 /*
414 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
415 */
416 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
417 {
418 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
419 }
420
421 /*
422 * Init thread quit pipe.
423 *
424 * Return -1 on error or 0 if all pipes are created.
425 */
426 static int __init_thread_quit_pipe(int *a_pipe)
427 {
428 int ret, i;
429
430 ret = pipe(a_pipe);
431 if (ret < 0) {
432 PERROR("thread quit pipe");
433 goto error;
434 }
435
436 for (i = 0; i < 2; i++) {
437 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
438 if (ret < 0) {
439 PERROR("fcntl");
440 goto error;
441 }
442 }
443
444 error:
445 return ret;
446 }
447
448 static int init_thread_quit_pipe(void)
449 {
450 return __init_thread_quit_pipe(thread_quit_pipe);
451 }
452
453 /*
454 * Stop all threads by closing the thread quit pipe.
455 */
456 static void stop_threads(void)
457 {
458 int ret;
459
460 /* Stopping all threads */
461 DBG("Terminating all threads");
462 ret = notify_thread_pipe(thread_quit_pipe[1]);
463 if (ret < 0) {
464 ERR("write error on thread quit pipe");
465 }
466
467 /* Dispatch thread */
468 CMM_STORE_SHARED(dispatch_thread_exit, 1);
469 futex_nto1_wake(&ust_cmd_queue.futex);
470 }
471
472 /*
473 * Close every consumer sockets.
474 */
475 static void close_consumer_sockets(void)
476 {
477 int ret;
478
479 if (kconsumer_data.err_sock >= 0) {
480 ret = close(kconsumer_data.err_sock);
481 if (ret < 0) {
482 PERROR("kernel consumer err_sock close");
483 }
484 }
485 if (ustconsumer32_data.err_sock >= 0) {
486 ret = close(ustconsumer32_data.err_sock);
487 if (ret < 0) {
488 PERROR("UST consumerd32 err_sock close");
489 }
490 }
491 if (ustconsumer64_data.err_sock >= 0) {
492 ret = close(ustconsumer64_data.err_sock);
493 if (ret < 0) {
494 PERROR("UST consumerd64 err_sock close");
495 }
496 }
497 if (kconsumer_data.cmd_sock >= 0) {
498 ret = close(kconsumer_data.cmd_sock);
499 if (ret < 0) {
500 PERROR("kernel consumer cmd_sock close");
501 }
502 }
503 if (ustconsumer32_data.cmd_sock >= 0) {
504 ret = close(ustconsumer32_data.cmd_sock);
505 if (ret < 0) {
506 PERROR("UST consumerd32 cmd_sock close");
507 }
508 }
509 if (ustconsumer64_data.cmd_sock >= 0) {
510 ret = close(ustconsumer64_data.cmd_sock);
511 if (ret < 0) {
512 PERROR("UST consumerd64 cmd_sock close");
513 }
514 }
515 if (kconsumer_data.channel_monitor_pipe >= 0) {
516 ret = close(kconsumer_data.channel_monitor_pipe);
517 if (ret < 0) {
518 PERROR("kernel consumer channel monitor pipe close");
519 }
520 }
521 if (ustconsumer32_data.channel_monitor_pipe >= 0) {
522 ret = close(ustconsumer32_data.channel_monitor_pipe);
523 if (ret < 0) {
524 PERROR("UST consumerd32 channel monitor pipe close");
525 }
526 }
527 if (ustconsumer64_data.channel_monitor_pipe >= 0) {
528 ret = close(ustconsumer64_data.channel_monitor_pipe);
529 if (ret < 0) {
530 PERROR("UST consumerd64 channel monitor pipe close");
531 }
532 }
533 if (kconsumer_data.channel_rotate_pipe >= 0) {
534 ret = close(kconsumer_data.channel_rotate_pipe);
535 if (ret < 0) {
536 PERROR("kernel consumer channel rotate pipe close");
537 }
538 }
539 if (ustconsumer32_data.channel_rotate_pipe >= 0) {
540 ret = close(ustconsumer32_data.channel_rotate_pipe);
541 if (ret < 0) {
542 PERROR("UST consumerd32 channel rotate pipe close");
543 }
544 }
545 if (ustconsumer64_data.channel_rotate_pipe >= 0) {
546 ret = close(ustconsumer64_data.channel_rotate_pipe);
547 if (ret < 0) {
548 PERROR("UST consumerd64 channel rotate pipe close");
549 }
550 }
551 }
552
553 /*
554 * Wait on consumer process termination.
555 *
556 * Need to be called with the consumer data lock held or from a context
557 * ensuring no concurrent access to data (e.g: cleanup).
558 */
559 static void wait_consumer(struct consumer_data *consumer_data)
560 {
561 pid_t ret;
562 int status;
563
564 if (consumer_data->pid <= 0) {
565 return;
566 }
567
568 DBG("Waiting for complete teardown of consumerd (PID: %d)",
569 consumer_data->pid);
570 ret = waitpid(consumer_data->pid, &status, 0);
571 if (ret == -1) {
572 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
573 } else if (!WIFEXITED(status)) {
574 ERR("consumerd termination with error: %d",
575 WEXITSTATUS(ret));
576 }
577 consumer_data->pid = 0;
578 }
579
580 /*
581 * Cleanup the session daemon's data structures.
582 */
583 static void sessiond_cleanup(void)
584 {
585 int ret;
586 struct ltt_session *sess, *stmp;
587
588 DBG("Cleanup sessiond");
589
590 /*
591 * Close the thread quit pipe. It has already done its job,
592 * since we are now called.
593 */
594 utils_close_pipe(thread_quit_pipe);
595
596 ret = remove(config.pid_file_path.value);
597 if (ret < 0) {
598 PERROR("remove pidfile %s", config.pid_file_path.value);
599 }
600
601 DBG("Removing sessiond and consumerd content of directory %s",
602 config.rundir.value);
603
604 /* sessiond */
605 DBG("Removing %s", config.pid_file_path.value);
606 (void) unlink(config.pid_file_path.value);
607
608 DBG("Removing %s", config.agent_port_file_path.value);
609 (void) unlink(config.agent_port_file_path.value);
610
611 /* kconsumerd */
612 DBG("Removing %s", kconsumer_data.err_unix_sock_path);
613 (void) unlink(kconsumer_data.err_unix_sock_path);
614
615 DBG("Removing directory %s", config.kconsumerd_path.value);
616 (void) rmdir(config.kconsumerd_path.value);
617
618 /* ust consumerd 32 */
619 DBG("Removing %s", config.consumerd32_err_unix_sock_path.value);
620 (void) unlink(config.consumerd32_err_unix_sock_path.value);
621
622 DBG("Removing directory %s", config.consumerd32_path.value);
623 (void) rmdir(config.consumerd32_path.value);
624
625 /* ust consumerd 64 */
626 DBG("Removing %s", config.consumerd64_err_unix_sock_path.value);
627 (void) unlink(config.consumerd64_err_unix_sock_path.value);
628
629 DBG("Removing directory %s", config.consumerd64_path.value);
630 (void) rmdir(config.consumerd64_path.value);
631
632 DBG("Cleaning up all sessions");
633
634 /* Destroy session list mutex */
635 if (session_list_ptr != NULL) {
636 pthread_mutex_destroy(&session_list_ptr->lock);
637
638 /* Cleanup ALL session */
639 cds_list_for_each_entry_safe(sess, stmp,
640 &session_list_ptr->head, list) {
641 cmd_destroy_session(sess, kernel_poll_pipe[1],
642 notification_thread_handle);
643 }
644 }
645
646 wait_consumer(&kconsumer_data);
647 wait_consumer(&ustconsumer64_data);
648 wait_consumer(&ustconsumer32_data);
649
650 DBG("Cleaning up all agent apps");
651 agent_app_ht_clean();
652
653 DBG("Closing all UST sockets");
654 ust_app_clean_list();
655 buffer_reg_destroy_registries();
656
657 if (is_root && !config.no_kernel) {
658 DBG2("Closing kernel fd");
659 if (kernel_tracer_fd >= 0) {
660 ret = close(kernel_tracer_fd);
661 if (ret) {
662 PERROR("close");
663 }
664 }
665 DBG("Unloading kernel modules");
666 modprobe_remove_lttng_all();
667 free(syscall_table);
668 }
669
670 close_consumer_sockets();
671
672 if (load_info) {
673 load_session_destroy_data(load_info);
674 free(load_info);
675 }
676
677 /*
678 * We do NOT rmdir rundir because there are other processes
679 * using it, for instance lttng-relayd, which can start in
680 * parallel with this teardown.
681 */
682 }
683
684 /*
685 * Cleanup the daemon's option data structures.
686 */
687 static void sessiond_cleanup_options(void)
688 {
689 DBG("Cleaning up options");
690
691 sessiond_config_fini(&config);
692
693 run_as_destroy_worker();
694 }
695
696 /*
697 * Send data on a unix socket using the liblttsessiondcomm API.
698 *
699 * Return lttcomm error code.
700 */
701 static int send_unix_sock(int sock, void *buf, size_t len)
702 {
703 /* Check valid length */
704 if (len == 0) {
705 return -1;
706 }
707
708 return lttcomm_send_unix_sock(sock, buf, len);
709 }
710
711 /*
712 * Free memory of a command context structure.
713 */
714 static void clean_command_ctx(struct command_ctx **cmd_ctx)
715 {
716 DBG("Clean command context structure");
717 if (*cmd_ctx) {
718 if ((*cmd_ctx)->llm) {
719 free((*cmd_ctx)->llm);
720 }
721 if ((*cmd_ctx)->lsm) {
722 free((*cmd_ctx)->lsm);
723 }
724 free(*cmd_ctx);
725 *cmd_ctx = NULL;
726 }
727 }
728
729 /*
730 * Notify UST applications using the shm mmap futex.
731 */
732 static int notify_ust_apps(int active)
733 {
734 char *wait_shm_mmap;
735
736 DBG("Notifying applications of session daemon state: %d", active);
737
738 /* See shm.c for this call implying mmap, shm and futex calls */
739 wait_shm_mmap = shm_ust_get_mmap(config.wait_shm_path.value, is_root);
740 if (wait_shm_mmap == NULL) {
741 goto error;
742 }
743
744 /* Wake waiting process */
745 futex_wait_update((int32_t *) wait_shm_mmap, active);
746
747 /* Apps notified successfully */
748 return 0;
749
750 error:
751 return -1;
752 }
753
754 /*
755 * Setup the outgoing data buffer for the response (llm) by allocating the
756 * right amount of memory and copying the original information from the lsm
757 * structure.
758 *
759 * Return 0 on success, negative value on error.
760 */
761 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
762 const void *payload_buf, size_t payload_len,
763 const void *cmd_header_buf, size_t cmd_header_len)
764 {
765 int ret = 0;
766 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
767 const size_t cmd_header_offset = header_len;
768 const size_t payload_offset = cmd_header_offset + cmd_header_len;
769 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
770
771 cmd_ctx->llm = zmalloc(total_msg_size);
772
773 if (cmd_ctx->llm == NULL) {
774 PERROR("zmalloc");
775 ret = -ENOMEM;
776 goto end;
777 }
778
779 /* Copy common data */
780 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
781 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
782 cmd_ctx->llm->cmd_header_size = cmd_header_len;
783 cmd_ctx->llm->data_size = payload_len;
784 cmd_ctx->lttng_msg_size = total_msg_size;
785
786 /* Copy command header */
787 if (cmd_header_len) {
788 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
789 cmd_header_len);
790 }
791
792 /* Copy payload */
793 if (payload_len) {
794 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
795 payload_len);
796 }
797
798 end:
799 return ret;
800 }
801
802 /*
803 * Version of setup_lttng_msg() without command header.
804 */
805 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
806 void *payload_buf, size_t payload_len)
807 {
808 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
809 }
810 /*
811 * Update the kernel poll set of all channel fd available over all tracing
812 * session. Add the wakeup pipe at the end of the set.
813 */
814 static int update_kernel_poll(struct lttng_poll_event *events)
815 {
816 int ret;
817 struct ltt_session *session;
818 struct ltt_kernel_channel *channel;
819
820 DBG("Updating kernel poll set");
821
822 session_lock_list();
823 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
824 session_lock(session);
825 if (session->kernel_session == NULL) {
826 session_unlock(session);
827 continue;
828 }
829
830 cds_list_for_each_entry(channel,
831 &session->kernel_session->channel_list.head, list) {
832 /* Add channel fd to the kernel poll set */
833 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
834 if (ret < 0) {
835 session_unlock(session);
836 goto error;
837 }
838 DBG("Channel fd %d added to kernel set", channel->fd);
839 }
840 session_unlock(session);
841 }
842 session_unlock_list();
843
844 return 0;
845
846 error:
847 session_unlock_list();
848 return -1;
849 }
850
851 /*
852 * Find the channel fd from 'fd' over all tracing session. When found, check
853 * for new channel stream and send those stream fds to the kernel consumer.
854 *
855 * Useful for CPU hotplug feature.
856 */
857 static int update_kernel_stream(int fd)
858 {
859 int ret = 0;
860 struct ltt_session *session;
861 struct ltt_kernel_session *ksess;
862 struct ltt_kernel_channel *channel;
863
864 DBG("Updating kernel streams for channel fd %d", fd);
865
866 session_lock_list();
867 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
868 session_lock(session);
869 if (session->kernel_session == NULL) {
870 session_unlock(session);
871 continue;
872 }
873 ksess = session->kernel_session;
874
875 cds_list_for_each_entry(channel,
876 &ksess->channel_list.head, list) {
877 struct lttng_ht_iter iter;
878 struct consumer_socket *socket;
879
880 if (channel->fd != fd) {
881 continue;
882 }
883 DBG("Channel found, updating kernel streams");
884 ret = kernel_open_channel_stream(channel);
885 if (ret < 0) {
886 goto error;
887 }
888 /* Update the stream global counter */
889 ksess->stream_count_global += ret;
890
891 /*
892 * Have we already sent fds to the consumer? If yes, it
893 * means that tracing is started so it is safe to send
894 * our updated stream fds.
895 */
896 if (ksess->consumer_fds_sent != 1
897 || ksess->consumer == NULL) {
898 ret = -1;
899 goto error;
900 }
901
902 rcu_read_lock();
903 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
904 &iter.iter, socket, node.node) {
905 pthread_mutex_lock(socket->lock);
906 ret = kernel_consumer_send_channel_streams(socket,
907 channel, ksess,
908 session->output_traces ? 1 : 0);
909 pthread_mutex_unlock(socket->lock);
910 if (ret < 0) {
911 rcu_read_unlock();
912 goto error;
913 }
914 }
915 rcu_read_unlock();
916 }
917 session_unlock(session);
918 }
919 session_unlock_list();
920 return ret;
921
922 error:
923 session_unlock(session);
924 session_unlock_list();
925 return ret;
926 }
927
928 /*
929 * For each tracing session, update newly registered apps. The session list
930 * lock MUST be acquired before calling this.
931 */
932 static void update_ust_app(int app_sock)
933 {
934 struct ltt_session *sess, *stmp;
935
936 /* Consumer is in an ERROR state. Stop any application update. */
937 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
938 /* Stop the update process since the consumer is dead. */
939 return;
940 }
941
942 /* For all tracing session(s) */
943 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
944 struct ust_app *app;
945
946 session_lock(sess);
947 if (!sess->ust_session) {
948 goto unlock_session;
949 }
950
951 rcu_read_lock();
952 assert(app_sock >= 0);
953 app = ust_app_find_by_sock(app_sock);
954 if (app == NULL) {
955 /*
956 * Application can be unregistered before so
957 * this is possible hence simply stopping the
958 * update.
959 */
960 DBG3("UST app update failed to find app sock %d",
961 app_sock);
962 goto unlock_rcu;
963 }
964 ust_app_global_update(sess->ust_session, app);
965 unlock_rcu:
966 rcu_read_unlock();
967 unlock_session:
968 session_unlock(sess);
969 }
970 }
971
972 /*
973 * This thread manage event coming from the kernel.
974 *
975 * Features supported in this thread:
976 * -) CPU Hotplug
977 */
978 static void *thread_manage_kernel(void *data)
979 {
980 int ret, i, pollfd, update_poll_flag = 1, err = -1;
981 uint32_t revents, nb_fd;
982 char tmp;
983 struct lttng_poll_event events;
984
985 DBG("[thread] Thread manage kernel started");
986
987 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
988
989 /*
990 * This first step of the while is to clean this structure which could free
991 * non NULL pointers so initialize it before the loop.
992 */
993 lttng_poll_init(&events);
994
995 if (testpoint(sessiond_thread_manage_kernel)) {
996 goto error_testpoint;
997 }
998
999 health_code_update();
1000
1001 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1002 goto error_testpoint;
1003 }
1004
1005 while (1) {
1006 health_code_update();
1007
1008 if (update_poll_flag == 1) {
1009 /* Clean events object. We are about to populate it again. */
1010 lttng_poll_clean(&events);
1011
1012 ret = sessiond_set_thread_pollset(&events, 2);
1013 if (ret < 0) {
1014 goto error_poll_create;
1015 }
1016
1017 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1018 if (ret < 0) {
1019 goto error;
1020 }
1021
1022 /* This will add the available kernel channel if any. */
1023 ret = update_kernel_poll(&events);
1024 if (ret < 0) {
1025 goto error;
1026 }
1027 update_poll_flag = 0;
1028 }
1029
1030 DBG("Thread kernel polling");
1031
1032 /* Poll infinite value of time */
1033 restart:
1034 health_poll_entry();
1035 ret = lttng_poll_wait(&events, -1);
1036 DBG("Thread kernel return from poll on %d fds",
1037 LTTNG_POLL_GETNB(&events));
1038 health_poll_exit();
1039 if (ret < 0) {
1040 /*
1041 * Restart interrupted system call.
1042 */
1043 if (errno == EINTR) {
1044 goto restart;
1045 }
1046 goto error;
1047 } else if (ret == 0) {
1048 /* Should not happen since timeout is infinite */
1049 ERR("Return value of poll is 0 with an infinite timeout.\n"
1050 "This should not have happened! Continuing...");
1051 continue;
1052 }
1053
1054 nb_fd = ret;
1055
1056 for (i = 0; i < nb_fd; i++) {
1057 /* Fetch once the poll data */
1058 revents = LTTNG_POLL_GETEV(&events, i);
1059 pollfd = LTTNG_POLL_GETFD(&events, i);
1060
1061 health_code_update();
1062
1063 if (!revents) {
1064 /* No activity for this FD (poll implementation). */
1065 continue;
1066 }
1067
1068 /* Thread quit pipe has been closed. Killing thread. */
1069 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1070 if (ret) {
1071 err = 0;
1072 goto exit;
1073 }
1074
1075 /* Check for data on kernel pipe */
1076 if (revents & LPOLLIN) {
1077 if (pollfd == kernel_poll_pipe[0]) {
1078 (void) lttng_read(kernel_poll_pipe[0],
1079 &tmp, 1);
1080 /*
1081 * Ret value is useless here, if this pipe gets any actions an
1082 * update is required anyway.
1083 */
1084 update_poll_flag = 1;
1085 continue;
1086 } else {
1087 /*
1088 * New CPU detected by the kernel. Adding kernel stream to
1089 * kernel session and updating the kernel consumer
1090 */
1091 ret = update_kernel_stream(pollfd);
1092 if (ret < 0) {
1093 continue;
1094 }
1095 break;
1096 }
1097 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1098 update_poll_flag = 1;
1099 continue;
1100 } else {
1101 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1102 goto error;
1103 }
1104 }
1105 }
1106
1107 exit:
1108 error:
1109 lttng_poll_clean(&events);
1110 error_poll_create:
1111 error_testpoint:
1112 utils_close_pipe(kernel_poll_pipe);
1113 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1114 if (err) {
1115 health_error();
1116 ERR("Health error occurred in %s", __func__);
1117 WARN("Kernel thread died unexpectedly. "
1118 "Kernel tracing can continue but CPU hotplug is disabled.");
1119 }
1120 health_unregister(health_sessiond);
1121 DBG("Kernel thread dying");
1122 return NULL;
1123 }
1124
1125 /*
1126 * Signal pthread condition of the consumer data that the thread.
1127 */
1128 static void signal_consumer_condition(struct consumer_data *data, int state)
1129 {
1130 pthread_mutex_lock(&data->cond_mutex);
1131
1132 /*
1133 * The state is set before signaling. It can be any value, it's the waiter
1134 * job to correctly interpret this condition variable associated to the
1135 * consumer pthread_cond.
1136 *
1137 * A value of 0 means that the corresponding thread of the consumer data
1138 * was not started. 1 indicates that the thread has started and is ready
1139 * for action. A negative value means that there was an error during the
1140 * thread bootstrap.
1141 */
1142 data->consumer_thread_is_ready = state;
1143 (void) pthread_cond_signal(&data->cond);
1144
1145 pthread_mutex_unlock(&data->cond_mutex);
1146 }
1147
1148 /*
1149 * This thread manage the consumer error sent back to the session daemon.
1150 */
1151 static void *thread_manage_consumer(void *data)
1152 {
1153 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1154 uint32_t revents, nb_fd;
1155 enum lttcomm_return_code code;
1156 struct lttng_poll_event events;
1157 struct consumer_data *consumer_data = data;
1158 struct consumer_socket *cmd_socket_wrapper = NULL;
1159
1160 DBG("[thread] Manage consumer started");
1161
1162 rcu_register_thread();
1163 rcu_thread_online();
1164
1165 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1166
1167 health_code_update();
1168
1169 /*
1170 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1171 * metadata_sock. Nothing more will be added to this poll set.
1172 */
1173 ret = sessiond_set_thread_pollset(&events, 3);
1174 if (ret < 0) {
1175 goto error_poll;
1176 }
1177
1178 /*
1179 * The error socket here is already in a listening state which was done
1180 * just before spawning this thread to avoid a race between the consumer
1181 * daemon exec trying to connect and the listen() call.
1182 */
1183 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1184 if (ret < 0) {
1185 goto error;
1186 }
1187
1188 health_code_update();
1189
1190 /* Infinite blocking call, waiting for transmission */
1191 restart:
1192 health_poll_entry();
1193
1194 if (testpoint(sessiond_thread_manage_consumer)) {
1195 goto error;
1196 }
1197
1198 ret = lttng_poll_wait(&events, -1);
1199 health_poll_exit();
1200 if (ret < 0) {
1201 /*
1202 * Restart interrupted system call.
1203 */
1204 if (errno == EINTR) {
1205 goto restart;
1206 }
1207 goto error;
1208 }
1209
1210 nb_fd = ret;
1211
1212 for (i = 0; i < nb_fd; i++) {
1213 /* Fetch once the poll data */
1214 revents = LTTNG_POLL_GETEV(&events, i);
1215 pollfd = LTTNG_POLL_GETFD(&events, i);
1216
1217 health_code_update();
1218
1219 if (!revents) {
1220 /* No activity for this FD (poll implementation). */
1221 continue;
1222 }
1223
1224 /* Thread quit pipe has been closed. Killing thread. */
1225 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1226 if (ret) {
1227 err = 0;
1228 goto exit;
1229 }
1230
1231 /* Event on the registration socket */
1232 if (pollfd == consumer_data->err_sock) {
1233 if (revents & LPOLLIN) {
1234 continue;
1235 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1236 ERR("consumer err socket poll error");
1237 goto error;
1238 } else {
1239 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1240 goto error;
1241 }
1242 }
1243 }
1244
1245 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1246 if (sock < 0) {
1247 goto error;
1248 }
1249
1250 /*
1251 * Set the CLOEXEC flag. Return code is useless because either way, the
1252 * show must go on.
1253 */
1254 (void) utils_set_fd_cloexec(sock);
1255
1256 health_code_update();
1257
1258 DBG2("Receiving code from consumer err_sock");
1259
1260 /* Getting status code from kconsumerd */
1261 ret = lttcomm_recv_unix_sock(sock, &code,
1262 sizeof(enum lttcomm_return_code));
1263 if (ret <= 0) {
1264 goto error;
1265 }
1266
1267 health_code_update();
1268 if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1269 ERR("consumer error when waiting for SOCK_READY : %s",
1270 lttcomm_get_readable_code(-code));
1271 goto error;
1272 }
1273
1274 /* Connect both command and metadata sockets. */
1275 consumer_data->cmd_sock =
1276 lttcomm_connect_unix_sock(
1277 consumer_data->cmd_unix_sock_path);
1278 consumer_data->metadata_fd =
1279 lttcomm_connect_unix_sock(
1280 consumer_data->cmd_unix_sock_path);
1281 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1282 PERROR("consumer connect cmd socket");
1283 /* On error, signal condition and quit. */
1284 signal_consumer_condition(consumer_data, -1);
1285 goto error;
1286 }
1287
1288 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1289
1290 /* Create metadata socket lock. */
1291 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1292 if (consumer_data->metadata_sock.lock == NULL) {
1293 PERROR("zmalloc pthread mutex");
1294 goto error;
1295 }
1296 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1297
1298 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1299 DBG("Consumer metadata socket ready (fd: %d)",
1300 consumer_data->metadata_fd);
1301
1302 /*
1303 * Remove the consumerd error sock since we've established a connection.
1304 */
1305 ret = lttng_poll_del(&events, consumer_data->err_sock);
1306 if (ret < 0) {
1307 goto error;
1308 }
1309
1310 /* Add new accepted error socket. */
1311 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1312 if (ret < 0) {
1313 goto error;
1314 }
1315
1316 /* Add metadata socket that is successfully connected. */
1317 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1318 LPOLLIN | LPOLLRDHUP);
1319 if (ret < 0) {
1320 goto error;
1321 }
1322
1323 health_code_update();
1324
1325 /*
1326 * Transfer the write-end of the channel monitoring and rotate pipe
1327 * to the consumer by issuing a SET_CHANNEL_MONITOR_PIPE and
1328 * SET_CHANNEL_ROTATE_PIPE commands.
1329 */
1330 cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock);
1331 if (!cmd_socket_wrapper) {
1332 goto error;
1333 }
1334 cmd_socket_wrapper->lock = &consumer_data->lock;
1335
1336 ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper,
1337 consumer_data->channel_monitor_pipe);
1338 if (ret) {
1339 goto error;
1340 }
1341
1342 ret = consumer_send_channel_rotate_pipe(cmd_socket_wrapper,
1343 consumer_data->channel_rotate_pipe);
1344 if (ret) {
1345 goto error;
1346 }
1347
1348 /* Discard the socket wrapper as it is no longer needed. */
1349 consumer_destroy_socket(cmd_socket_wrapper);
1350 cmd_socket_wrapper = NULL;
1351
1352 /* The thread is completely initialized, signal that it is ready. */
1353 signal_consumer_condition(consumer_data, 1);
1354
1355 /* Infinite blocking call, waiting for transmission */
1356 restart_poll:
1357 while (1) {
1358 health_code_update();
1359
1360 /* Exit the thread because the thread quit pipe has been triggered. */
1361 if (should_quit) {
1362 /* Not a health error. */
1363 err = 0;
1364 goto exit;
1365 }
1366
1367 health_poll_entry();
1368 ret = lttng_poll_wait(&events, -1);
1369 health_poll_exit();
1370 if (ret < 0) {
1371 /*
1372 * Restart interrupted system call.
1373 */
1374 if (errno == EINTR) {
1375 goto restart_poll;
1376 }
1377 goto error;
1378 }
1379
1380 nb_fd = ret;
1381
1382 for (i = 0; i < nb_fd; i++) {
1383 /* Fetch once the poll data */
1384 revents = LTTNG_POLL_GETEV(&events, i);
1385 pollfd = LTTNG_POLL_GETFD(&events, i);
1386
1387 health_code_update();
1388
1389 if (!revents) {
1390 /* No activity for this FD (poll implementation). */
1391 continue;
1392 }
1393
1394 /*
1395 * Thread quit pipe has been triggered, flag that we should stop
1396 * but continue the current loop to handle potential data from
1397 * consumer.
1398 */
1399 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1400
1401 if (pollfd == sock) {
1402 /* Event on the consumerd socket */
1403 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1404 && !(revents & LPOLLIN)) {
1405 ERR("consumer err socket second poll error");
1406 goto error;
1407 }
1408 health_code_update();
1409 /* Wait for any kconsumerd error */
1410 ret = lttcomm_recv_unix_sock(sock, &code,
1411 sizeof(enum lttcomm_return_code));
1412 if (ret <= 0) {
1413 ERR("consumer closed the command socket");
1414 goto error;
1415 }
1416
1417 ERR("consumer return code : %s",
1418 lttcomm_get_readable_code(-code));
1419
1420 goto exit;
1421 } else if (pollfd == consumer_data->metadata_fd) {
1422 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1423 && !(revents & LPOLLIN)) {
1424 ERR("consumer err metadata socket second poll error");
1425 goto error;
1426 }
1427 /* UST metadata requests */
1428 ret = ust_consumer_metadata_request(
1429 &consumer_data->metadata_sock);
1430 if (ret < 0) {
1431 ERR("Handling metadata request");
1432 goto error;
1433 }
1434 }
1435 /* No need for an else branch all FDs are tested prior. */
1436 }
1437 health_code_update();
1438 }
1439
1440 exit:
1441 error:
1442 /*
1443 * We lock here because we are about to close the sockets and some other
1444 * thread might be using them so get exclusive access which will abort all
1445 * other consumer command by other threads.
1446 */
1447 pthread_mutex_lock(&consumer_data->lock);
1448
1449 /* Immediately set the consumerd state to stopped */
1450 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1451 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1452 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1453 consumer_data->type == LTTNG_CONSUMER32_UST) {
1454 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1455 } else {
1456 /* Code flow error... */
1457 assert(0);
1458 }
1459
1460 if (consumer_data->err_sock >= 0) {
1461 ret = close(consumer_data->err_sock);
1462 if (ret) {
1463 PERROR("close");
1464 }
1465 consumer_data->err_sock = -1;
1466 }
1467 if (consumer_data->cmd_sock >= 0) {
1468 ret = close(consumer_data->cmd_sock);
1469 if (ret) {
1470 PERROR("close");
1471 }
1472 consumer_data->cmd_sock = -1;
1473 }
1474 if (consumer_data->metadata_sock.fd_ptr &&
1475 *consumer_data->metadata_sock.fd_ptr >= 0) {
1476 ret = close(*consumer_data->metadata_sock.fd_ptr);
1477 if (ret) {
1478 PERROR("close");
1479 }
1480 }
1481 if (sock >= 0) {
1482 ret = close(sock);
1483 if (ret) {
1484 PERROR("close");
1485 }
1486 }
1487
1488 unlink(consumer_data->err_unix_sock_path);
1489 unlink(consumer_data->cmd_unix_sock_path);
1490 pthread_mutex_unlock(&consumer_data->lock);
1491
1492 /* Cleanup metadata socket mutex. */
1493 if (consumer_data->metadata_sock.lock) {
1494 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1495 free(consumer_data->metadata_sock.lock);
1496 }
1497 lttng_poll_clean(&events);
1498
1499 if (cmd_socket_wrapper) {
1500 consumer_destroy_socket(cmd_socket_wrapper);
1501 }
1502 error_poll:
1503 if (err) {
1504 health_error();
1505 ERR("Health error occurred in %s", __func__);
1506 }
1507 health_unregister(health_sessiond);
1508 DBG("consumer thread cleanup completed");
1509
1510 rcu_thread_offline();
1511 rcu_unregister_thread();
1512
1513 return NULL;
1514 }
1515
1516 /*
1517 * This thread receives application command sockets (FDs) on the
1518 * apps_cmd_pipe and waits (polls) on them until they are closed
1519 * or an error occurs.
1520 *
1521 * At that point, it flushes the data (tracing and metadata) associated
1522 * with this application and tears down ust app sessions and other
1523 * associated data structures through ust_app_unregister().
1524 *
1525 * Note that this thread never sends commands to the applications
1526 * through the command sockets; it merely listens for hang-ups
1527 * and errors on those sockets and cleans-up as they occur.
1528 */
1529 static void *thread_manage_apps(void *data)
1530 {
1531 int i, ret, pollfd, err = -1;
1532 ssize_t size_ret;
1533 uint32_t revents, nb_fd;
1534 struct lttng_poll_event events;
1535
1536 DBG("[thread] Manage application started");
1537
1538 rcu_register_thread();
1539 rcu_thread_online();
1540
1541 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1542
1543 if (testpoint(sessiond_thread_manage_apps)) {
1544 goto error_testpoint;
1545 }
1546
1547 health_code_update();
1548
1549 ret = sessiond_set_thread_pollset(&events, 2);
1550 if (ret < 0) {
1551 goto error_poll_create;
1552 }
1553
1554 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1555 if (ret < 0) {
1556 goto error;
1557 }
1558
1559 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1560 goto error;
1561 }
1562
1563 health_code_update();
1564
1565 while (1) {
1566 DBG("Apps thread polling");
1567
1568 /* Inifinite blocking call, waiting for transmission */
1569 restart:
1570 health_poll_entry();
1571 ret = lttng_poll_wait(&events, -1);
1572 DBG("Apps thread return from poll on %d fds",
1573 LTTNG_POLL_GETNB(&events));
1574 health_poll_exit();
1575 if (ret < 0) {
1576 /*
1577 * Restart interrupted system call.
1578 */
1579 if (errno == EINTR) {
1580 goto restart;
1581 }
1582 goto error;
1583 }
1584
1585 nb_fd = ret;
1586
1587 for (i = 0; i < nb_fd; i++) {
1588 /* Fetch once the poll data */
1589 revents = LTTNG_POLL_GETEV(&events, i);
1590 pollfd = LTTNG_POLL_GETFD(&events, i);
1591
1592 health_code_update();
1593
1594 if (!revents) {
1595 /* No activity for this FD (poll implementation). */
1596 continue;
1597 }
1598
1599 /* Thread quit pipe has been closed. Killing thread. */
1600 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1601 if (ret) {
1602 err = 0;
1603 goto exit;
1604 }
1605
1606 /* Inspect the apps cmd pipe */
1607 if (pollfd == apps_cmd_pipe[0]) {
1608 if (revents & LPOLLIN) {
1609 int sock;
1610
1611 /* Empty pipe */
1612 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1613 if (size_ret < sizeof(sock)) {
1614 PERROR("read apps cmd pipe");
1615 goto error;
1616 }
1617
1618 health_code_update();
1619
1620 /*
1621 * Since this is a command socket (write then read),
1622 * we only monitor the error events of the socket.
1623 */
1624 ret = lttng_poll_add(&events, sock,
1625 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1626 if (ret < 0) {
1627 goto error;
1628 }
1629
1630 DBG("Apps with sock %d added to poll set", sock);
1631 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1632 ERR("Apps command pipe error");
1633 goto error;
1634 } else {
1635 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1636 goto error;
1637 }
1638 } else {
1639 /*
1640 * At this point, we know that a registered application made
1641 * the event at poll_wait.
1642 */
1643 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1644 /* Removing from the poll set */
1645 ret = lttng_poll_del(&events, pollfd);
1646 if (ret < 0) {
1647 goto error;
1648 }
1649
1650 /* Socket closed on remote end. */
1651 ust_app_unregister(pollfd);
1652 } else {
1653 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1654 goto error;
1655 }
1656 }
1657
1658 health_code_update();
1659 }
1660 }
1661
1662 exit:
1663 error:
1664 lttng_poll_clean(&events);
1665 error_poll_create:
1666 error_testpoint:
1667 utils_close_pipe(apps_cmd_pipe);
1668 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1669
1670 /*
1671 * We don't clean the UST app hash table here since already registered
1672 * applications can still be controlled so let them be until the session
1673 * daemon dies or the applications stop.
1674 */
1675
1676 if (err) {
1677 health_error();
1678 ERR("Health error occurred in %s", __func__);
1679 }
1680 health_unregister(health_sessiond);
1681 DBG("Application communication apps thread cleanup complete");
1682 rcu_thread_offline();
1683 rcu_unregister_thread();
1684 return NULL;
1685 }
1686
1687 /*
1688 * Send a socket to a thread This is called from the dispatch UST registration
1689 * thread once all sockets are set for the application.
1690 *
1691 * The sock value can be invalid, we don't really care, the thread will handle
1692 * it and make the necessary cleanup if so.
1693 *
1694 * On success, return 0 else a negative value being the errno message of the
1695 * write().
1696 */
1697 static int send_socket_to_thread(int fd, int sock)
1698 {
1699 ssize_t ret;
1700
1701 /*
1702 * It's possible that the FD is set as invalid with -1 concurrently just
1703 * before calling this function being a shutdown state of the thread.
1704 */
1705 if (fd < 0) {
1706 ret = -EBADF;
1707 goto error;
1708 }
1709
1710 ret = lttng_write(fd, &sock, sizeof(sock));
1711 if (ret < sizeof(sock)) {
1712 PERROR("write apps pipe %d", fd);
1713 if (ret < 0) {
1714 ret = -errno;
1715 }
1716 goto error;
1717 }
1718
1719 /* All good. Don't send back the write positive ret value. */
1720 ret = 0;
1721 error:
1722 return (int) ret;
1723 }
1724
1725 /*
1726 * Sanitize the wait queue of the dispatch registration thread meaning removing
1727 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1728 * notify socket is never received.
1729 */
1730 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1731 {
1732 int ret, nb_fd = 0, i;
1733 unsigned int fd_added = 0;
1734 struct lttng_poll_event events;
1735 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1736
1737 assert(wait_queue);
1738
1739 lttng_poll_init(&events);
1740
1741 /* Just skip everything for an empty queue. */
1742 if (!wait_queue->count) {
1743 goto end;
1744 }
1745
1746 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1747 if (ret < 0) {
1748 goto error_create;
1749 }
1750
1751 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1752 &wait_queue->head, head) {
1753 assert(wait_node->app);
1754 ret = lttng_poll_add(&events, wait_node->app->sock,
1755 LPOLLHUP | LPOLLERR);
1756 if (ret < 0) {
1757 goto error;
1758 }
1759
1760 fd_added = 1;
1761 }
1762
1763 if (!fd_added) {
1764 goto end;
1765 }
1766
1767 /*
1768 * Poll but don't block so we can quickly identify the faulty events and
1769 * clean them afterwards from the wait queue.
1770 */
1771 ret = lttng_poll_wait(&events, 0);
1772 if (ret < 0) {
1773 goto error;
1774 }
1775 nb_fd = ret;
1776
1777 for (i = 0; i < nb_fd; i++) {
1778 /* Get faulty FD. */
1779 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1780 int pollfd = LTTNG_POLL_GETFD(&events, i);
1781
1782 if (!revents) {
1783 /* No activity for this FD (poll implementation). */
1784 continue;
1785 }
1786
1787 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1788 &wait_queue->head, head) {
1789 if (pollfd == wait_node->app->sock &&
1790 (revents & (LPOLLHUP | LPOLLERR))) {
1791 cds_list_del(&wait_node->head);
1792 wait_queue->count--;
1793 ust_app_destroy(wait_node->app);
1794 free(wait_node);
1795 /*
1796 * Silence warning of use-after-free in
1797 * cds_list_for_each_entry_safe which uses
1798 * __typeof__(*wait_node).
1799 */
1800 wait_node = NULL;
1801 break;
1802 } else {
1803 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1804 goto error;
1805 }
1806 }
1807 }
1808
1809 if (nb_fd > 0) {
1810 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1811 }
1812
1813 end:
1814 lttng_poll_clean(&events);
1815 return;
1816
1817 error:
1818 lttng_poll_clean(&events);
1819 error_create:
1820 ERR("Unable to sanitize wait queue");
1821 return;
1822 }
1823
1824 /*
1825 * Dispatch request from the registration threads to the application
1826 * communication thread.
1827 */
1828 static void *thread_dispatch_ust_registration(void *data)
1829 {
1830 int ret, err = -1;
1831 struct cds_wfcq_node *node;
1832 struct ust_command *ust_cmd = NULL;
1833 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1834 struct ust_reg_wait_queue wait_queue = {
1835 .count = 0,
1836 };
1837
1838 rcu_register_thread();
1839
1840 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1841
1842 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1843 goto error_testpoint;
1844 }
1845
1846 health_code_update();
1847
1848 CDS_INIT_LIST_HEAD(&wait_queue.head);
1849
1850 DBG("[thread] Dispatch UST command started");
1851
1852 for (;;) {
1853 health_code_update();
1854
1855 /* Atomically prepare the queue futex */
1856 futex_nto1_prepare(&ust_cmd_queue.futex);
1857
1858 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1859 break;
1860 }
1861
1862 do {
1863 struct ust_app *app = NULL;
1864 ust_cmd = NULL;
1865
1866 /*
1867 * Make sure we don't have node(s) that have hung up before receiving
1868 * the notify socket. This is to clean the list in order to avoid
1869 * memory leaks from notify socket that are never seen.
1870 */
1871 sanitize_wait_queue(&wait_queue);
1872
1873 health_code_update();
1874 /* Dequeue command for registration */
1875 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1876 if (node == NULL) {
1877 DBG("Woken up but nothing in the UST command queue");
1878 /* Continue thread execution */
1879 break;
1880 }
1881
1882 ust_cmd = caa_container_of(node, struct ust_command, node);
1883
1884 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1885 " gid:%d sock:%d name:%s (version %d.%d)",
1886 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1887 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1888 ust_cmd->sock, ust_cmd->reg_msg.name,
1889 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1890
1891 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1892 wait_node = zmalloc(sizeof(*wait_node));
1893 if (!wait_node) {
1894 PERROR("zmalloc wait_node dispatch");
1895 ret = close(ust_cmd->sock);
1896 if (ret < 0) {
1897 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1898 }
1899 lttng_fd_put(LTTNG_FD_APPS, 1);
1900 free(ust_cmd);
1901 goto error;
1902 }
1903 CDS_INIT_LIST_HEAD(&wait_node->head);
1904
1905 /* Create application object if socket is CMD. */
1906 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1907 ust_cmd->sock);
1908 if (!wait_node->app) {
1909 ret = close(ust_cmd->sock);
1910 if (ret < 0) {
1911 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1912 }
1913 lttng_fd_put(LTTNG_FD_APPS, 1);
1914 free(wait_node);
1915 free(ust_cmd);
1916 continue;
1917 }
1918 /*
1919 * Add application to the wait queue so we can set the notify
1920 * socket before putting this object in the global ht.
1921 */
1922 cds_list_add(&wait_node->head, &wait_queue.head);
1923 wait_queue.count++;
1924
1925 free(ust_cmd);
1926 /*
1927 * We have to continue here since we don't have the notify
1928 * socket and the application MUST be added to the hash table
1929 * only at that moment.
1930 */
1931 continue;
1932 } else {
1933 /*
1934 * Look for the application in the local wait queue and set the
1935 * notify socket if found.
1936 */
1937 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1938 &wait_queue.head, head) {
1939 health_code_update();
1940 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1941 wait_node->app->notify_sock = ust_cmd->sock;
1942 cds_list_del(&wait_node->head);
1943 wait_queue.count--;
1944 app = wait_node->app;
1945 free(wait_node);
1946 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1947 break;
1948 }
1949 }
1950
1951 /*
1952 * With no application at this stage the received socket is
1953 * basically useless so close it before we free the cmd data
1954 * structure for good.
1955 */
1956 if (!app) {
1957 ret = close(ust_cmd->sock);
1958 if (ret < 0) {
1959 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1960 }
1961 lttng_fd_put(LTTNG_FD_APPS, 1);
1962 }
1963 free(ust_cmd);
1964 }
1965
1966 if (app) {
1967 /*
1968 * @session_lock_list
1969 *
1970 * Lock the global session list so from the register up to the
1971 * registration done message, no thread can see the application
1972 * and change its state.
1973 */
1974 session_lock_list();
1975 rcu_read_lock();
1976
1977 /*
1978 * Add application to the global hash table. This needs to be
1979 * done before the update to the UST registry can locate the
1980 * application.
1981 */
1982 ust_app_add(app);
1983
1984 /* Set app version. This call will print an error if needed. */
1985 (void) ust_app_version(app);
1986
1987 /* Send notify socket through the notify pipe. */
1988 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1989 app->notify_sock);
1990 if (ret < 0) {
1991 rcu_read_unlock();
1992 session_unlock_list();
1993 /*
1994 * No notify thread, stop the UST tracing. However, this is
1995 * not an internal error of the this thread thus setting
1996 * the health error code to a normal exit.
1997 */
1998 err = 0;
1999 goto error;
2000 }
2001
2002 /*
2003 * Update newly registered application with the tracing
2004 * registry info already enabled information.
2005 */
2006 update_ust_app(app->sock);
2007
2008 /*
2009 * Don't care about return value. Let the manage apps threads
2010 * handle app unregistration upon socket close.
2011 */
2012 (void) ust_app_register_done(app);
2013
2014 /*
2015 * Even if the application socket has been closed, send the app
2016 * to the thread and unregistration will take place at that
2017 * place.
2018 */
2019 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2020 if (ret < 0) {
2021 rcu_read_unlock();
2022 session_unlock_list();
2023 /*
2024 * No apps. thread, stop the UST tracing. However, this is
2025 * not an internal error of the this thread thus setting
2026 * the health error code to a normal exit.
2027 */
2028 err = 0;
2029 goto error;
2030 }
2031
2032 rcu_read_unlock();
2033 session_unlock_list();
2034 }
2035 } while (node != NULL);
2036
2037 health_poll_entry();
2038 /* Futex wait on queue. Blocking call on futex() */
2039 futex_nto1_wait(&ust_cmd_queue.futex);
2040 health_poll_exit();
2041 }
2042 /* Normal exit, no error */
2043 err = 0;
2044
2045 error:
2046 /* Clean up wait queue. */
2047 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2048 &wait_queue.head, head) {
2049 cds_list_del(&wait_node->head);
2050 wait_queue.count--;
2051 free(wait_node);
2052 }
2053
2054 /* Empty command queue. */
2055 for (;;) {
2056 /* Dequeue command for registration */
2057 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2058 if (node == NULL) {
2059 break;
2060 }
2061 ust_cmd = caa_container_of(node, struct ust_command, node);
2062 ret = close(ust_cmd->sock);
2063 if (ret < 0) {
2064 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2065 }
2066 lttng_fd_put(LTTNG_FD_APPS, 1);
2067 free(ust_cmd);
2068 }
2069
2070 error_testpoint:
2071 DBG("Dispatch thread dying");
2072 if (err) {
2073 health_error();
2074 ERR("Health error occurred in %s", __func__);
2075 }
2076 health_unregister(health_sessiond);
2077 rcu_unregister_thread();
2078 return NULL;
2079 }
2080
2081 /*
2082 * This thread manage application registration.
2083 */
2084 static void *thread_registration_apps(void *data)
2085 {
2086 int sock = -1, i, ret, pollfd, err = -1;
2087 uint32_t revents, nb_fd;
2088 struct lttng_poll_event events;
2089 /*
2090 * Get allocated in this thread, enqueued to a global queue, dequeued and
2091 * freed in the manage apps thread.
2092 */
2093 struct ust_command *ust_cmd = NULL;
2094
2095 DBG("[thread] Manage application registration started");
2096
2097 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2098
2099 if (testpoint(sessiond_thread_registration_apps)) {
2100 goto error_testpoint;
2101 }
2102
2103 ret = lttcomm_listen_unix_sock(apps_sock);
2104 if (ret < 0) {
2105 goto error_listen;
2106 }
2107
2108 /*
2109 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2110 * more will be added to this poll set.
2111 */
2112 ret = sessiond_set_thread_pollset(&events, 2);
2113 if (ret < 0) {
2114 goto error_create_poll;
2115 }
2116
2117 /* Add the application registration socket */
2118 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2119 if (ret < 0) {
2120 goto error_poll_add;
2121 }
2122
2123 /* Notify all applications to register */
2124 ret = notify_ust_apps(1);
2125 if (ret < 0) {
2126 ERR("Failed to notify applications or create the wait shared memory.\n"
2127 "Execution continues but there might be problem for already\n"
2128 "running applications that wishes to register.");
2129 }
2130
2131 while (1) {
2132 DBG("Accepting application registration");
2133
2134 /* Inifinite blocking call, waiting for transmission */
2135 restart:
2136 health_poll_entry();
2137 ret = lttng_poll_wait(&events, -1);
2138 health_poll_exit();
2139 if (ret < 0) {
2140 /*
2141 * Restart interrupted system call.
2142 */
2143 if (errno == EINTR) {
2144 goto restart;
2145 }
2146 goto error;
2147 }
2148
2149 nb_fd = ret;
2150
2151 for (i = 0; i < nb_fd; i++) {
2152 health_code_update();
2153
2154 /* Fetch once the poll data */
2155 revents = LTTNG_POLL_GETEV(&events, i);
2156 pollfd = LTTNG_POLL_GETFD(&events, i);
2157
2158 if (!revents) {
2159 /* No activity for this FD (poll implementation). */
2160 continue;
2161 }
2162
2163 /* Thread quit pipe has been closed. Killing thread. */
2164 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2165 if (ret) {
2166 err = 0;
2167 goto exit;
2168 }
2169
2170 /* Event on the registration socket */
2171 if (pollfd == apps_sock) {
2172 if (revents & LPOLLIN) {
2173 sock = lttcomm_accept_unix_sock(apps_sock);
2174 if (sock < 0) {
2175 goto error;
2176 }
2177
2178 /*
2179 * Set socket timeout for both receiving and ending.
2180 * app_socket_timeout is in seconds, whereas
2181 * lttcomm_setsockopt_rcv_timeout and
2182 * lttcomm_setsockopt_snd_timeout expect msec as
2183 * parameter.
2184 */
2185 if (config.app_socket_timeout >= 0) {
2186 (void) lttcomm_setsockopt_rcv_timeout(sock,
2187 config.app_socket_timeout * 1000);
2188 (void) lttcomm_setsockopt_snd_timeout(sock,
2189 config.app_socket_timeout * 1000);
2190 }
2191
2192 /*
2193 * Set the CLOEXEC flag. Return code is useless because
2194 * either way, the show must go on.
2195 */
2196 (void) utils_set_fd_cloexec(sock);
2197
2198 /* Create UST registration command for enqueuing */
2199 ust_cmd = zmalloc(sizeof(struct ust_command));
2200 if (ust_cmd == NULL) {
2201 PERROR("ust command zmalloc");
2202 ret = close(sock);
2203 if (ret) {
2204 PERROR("close");
2205 }
2206 goto error;
2207 }
2208
2209 /*
2210 * Using message-based transmissions to ensure we don't
2211 * have to deal with partially received messages.
2212 */
2213 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2214 if (ret < 0) {
2215 ERR("Exhausted file descriptors allowed for applications.");
2216 free(ust_cmd);
2217 ret = close(sock);
2218 if (ret) {
2219 PERROR("close");
2220 }
2221 sock = -1;
2222 continue;
2223 }
2224
2225 health_code_update();
2226 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2227 if (ret < 0) {
2228 free(ust_cmd);
2229 /* Close socket of the application. */
2230 ret = close(sock);
2231 if (ret) {
2232 PERROR("close");
2233 }
2234 lttng_fd_put(LTTNG_FD_APPS, 1);
2235 sock = -1;
2236 continue;
2237 }
2238 health_code_update();
2239
2240 ust_cmd->sock = sock;
2241 sock = -1;
2242
2243 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2244 " gid:%d sock:%d name:%s (version %d.%d)",
2245 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2246 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2247 ust_cmd->sock, ust_cmd->reg_msg.name,
2248 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2249
2250 /*
2251 * Lock free enqueue the registration request. The red pill
2252 * has been taken! This apps will be part of the *system*.
2253 */
2254 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2255
2256 /*
2257 * Wake the registration queue futex. Implicit memory
2258 * barrier with the exchange in cds_wfcq_enqueue.
2259 */
2260 futex_nto1_wake(&ust_cmd_queue.futex);
2261 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2262 ERR("Register apps socket poll error");
2263 goto error;
2264 } else {
2265 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2266 goto error;
2267 }
2268 }
2269 }
2270 }
2271
2272 exit:
2273 error:
2274 /* Notify that the registration thread is gone */
2275 notify_ust_apps(0);
2276
2277 if (apps_sock >= 0) {
2278 ret = close(apps_sock);
2279 if (ret) {
2280 PERROR("close");
2281 }
2282 }
2283 if (sock >= 0) {
2284 ret = close(sock);
2285 if (ret) {
2286 PERROR("close");
2287 }
2288 lttng_fd_put(LTTNG_FD_APPS, 1);
2289 }
2290 unlink(config.apps_unix_sock_path.value);
2291
2292 error_poll_add:
2293 lttng_poll_clean(&events);
2294 error_listen:
2295 error_create_poll:
2296 error_testpoint:
2297 DBG("UST Registration thread cleanup complete");
2298 if (err) {
2299 health_error();
2300 ERR("Health error occurred in %s", __func__);
2301 }
2302 health_unregister(health_sessiond);
2303
2304 return NULL;
2305 }
2306
2307 /*
2308 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2309 * exec or it will fails.
2310 */
2311 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2312 {
2313 int ret, clock_ret;
2314 struct timespec timeout;
2315
2316 /*
2317 * Make sure we set the readiness flag to 0 because we are NOT ready.
2318 * This access to consumer_thread_is_ready does not need to be
2319 * protected by consumer_data.cond_mutex (yet) since the consumer
2320 * management thread has not been started at this point.
2321 */
2322 consumer_data->consumer_thread_is_ready = 0;
2323
2324 /* Setup pthread condition */
2325 ret = pthread_condattr_init(&consumer_data->condattr);
2326 if (ret) {
2327 errno = ret;
2328 PERROR("pthread_condattr_init consumer data");
2329 goto error;
2330 }
2331
2332 /*
2333 * Set the monotonic clock in order to make sure we DO NOT jump in time
2334 * between the clock_gettime() call and the timedwait call. See bug #324
2335 * for a more details and how we noticed it.
2336 */
2337 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2338 if (ret) {
2339 errno = ret;
2340 PERROR("pthread_condattr_setclock consumer data");
2341 goto error;
2342 }
2343
2344 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2345 if (ret) {
2346 errno = ret;
2347 PERROR("pthread_cond_init consumer data");
2348 goto error;
2349 }
2350
2351 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2352 thread_manage_consumer, consumer_data);
2353 if (ret) {
2354 errno = ret;
2355 PERROR("pthread_create consumer");
2356 ret = -1;
2357 goto error;
2358 }
2359
2360 /* We are about to wait on a pthread condition */
2361 pthread_mutex_lock(&consumer_data->cond_mutex);
2362
2363 /* Get time for sem_timedwait absolute timeout */
2364 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2365 /*
2366 * Set the timeout for the condition timed wait even if the clock gettime
2367 * call fails since we might loop on that call and we want to avoid to
2368 * increment the timeout too many times.
2369 */
2370 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2371
2372 /*
2373 * The following loop COULD be skipped in some conditions so this is why we
2374 * set ret to 0 in order to make sure at least one round of the loop is
2375 * done.
2376 */
2377 ret = 0;
2378
2379 /*
2380 * Loop until the condition is reached or when a timeout is reached. Note
2381 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2382 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2383 * possible. This loop does not take any chances and works with both of
2384 * them.
2385 */
2386 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2387 if (clock_ret < 0) {
2388 PERROR("clock_gettime spawn consumer");
2389 /* Infinite wait for the consumerd thread to be ready */
2390 ret = pthread_cond_wait(&consumer_data->cond,
2391 &consumer_data->cond_mutex);
2392 } else {
2393 ret = pthread_cond_timedwait(&consumer_data->cond,
2394 &consumer_data->cond_mutex, &timeout);
2395 }
2396 }
2397
2398 /* Release the pthread condition */
2399 pthread_mutex_unlock(&consumer_data->cond_mutex);
2400
2401 if (ret != 0) {
2402 errno = ret;
2403 if (ret == ETIMEDOUT) {
2404 int pth_ret;
2405
2406 /*
2407 * Call has timed out so we kill the kconsumerd_thread and return
2408 * an error.
2409 */
2410 ERR("Condition timed out. The consumer thread was never ready."
2411 " Killing it");
2412 pth_ret = pthread_cancel(consumer_data->thread);
2413 if (pth_ret < 0) {
2414 PERROR("pthread_cancel consumer thread");
2415 }
2416 } else {
2417 PERROR("pthread_cond_wait failed consumer thread");
2418 }
2419 /* Caller is expecting a negative value on failure. */
2420 ret = -1;
2421 goto error;
2422 }
2423
2424 pthread_mutex_lock(&consumer_data->pid_mutex);
2425 if (consumer_data->pid == 0) {
2426 ERR("Consumerd did not start");
2427 pthread_mutex_unlock(&consumer_data->pid_mutex);
2428 goto error;
2429 }
2430 pthread_mutex_unlock(&consumer_data->pid_mutex);
2431
2432 return 0;
2433
2434 error:
2435 return ret;
2436 }
2437
2438 /*
2439 * Join consumer thread
2440 */
2441 static int join_consumer_thread(struct consumer_data *consumer_data)
2442 {
2443 void *status;
2444
2445 /* Consumer pid must be a real one. */
2446 if (consumer_data->pid > 0) {
2447 int ret;
2448 ret = kill(consumer_data->pid, SIGTERM);
2449 if (ret) {
2450 PERROR("Error killing consumer daemon");
2451 return ret;
2452 }
2453 return pthread_join(consumer_data->thread, &status);
2454 } else {
2455 return 0;
2456 }
2457 }
2458
2459 /*
2460 * Fork and exec a consumer daemon (consumerd).
2461 *
2462 * Return pid if successful else -1.
2463 */
2464 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2465 {
2466 int ret;
2467 pid_t pid;
2468 const char *consumer_to_use;
2469 const char *verbosity;
2470 struct stat st;
2471
2472 DBG("Spawning consumerd");
2473
2474 pid = fork();
2475 if (pid == 0) {
2476 /*
2477 * Exec consumerd.
2478 */
2479 if (config.verbose_consumer) {
2480 verbosity = "--verbose";
2481 } else if (lttng_opt_quiet) {
2482 verbosity = "--quiet";
2483 } else {
2484 verbosity = "";
2485 }
2486
2487 switch (consumer_data->type) {
2488 case LTTNG_CONSUMER_KERNEL:
2489 /*
2490 * Find out which consumerd to execute. We will first try the
2491 * 64-bit path, then the sessiond's installation directory, and
2492 * fallback on the 32-bit one,
2493 */
2494 DBG3("Looking for a kernel consumer at these locations:");
2495 DBG3(" 1) %s", config.consumerd64_bin_path.value ? : "NULL");
2496 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, DEFAULT_CONSUMERD_FILE);
2497 DBG3(" 3) %s", config.consumerd32_bin_path.value ? : "NULL");
2498 if (stat(config.consumerd64_bin_path.value, &st) == 0) {
2499 DBG3("Found location #1");
2500 consumer_to_use = config.consumerd64_bin_path.value;
2501 } else if (stat(INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE, &st) == 0) {
2502 DBG3("Found location #2");
2503 consumer_to_use = INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE;
2504 } else if (stat(config.consumerd32_bin_path.value, &st) == 0) {
2505 DBG3("Found location #3");
2506 consumer_to_use = config.consumerd32_bin_path.value;
2507 } else {
2508 DBG("Could not find any valid consumerd executable");
2509 ret = -EINVAL;
2510 goto error;
2511 }
2512 DBG("Using kernel consumer at: %s", consumer_to_use);
2513 (void) execl(consumer_to_use,
2514 "lttng-consumerd", verbosity, "-k",
2515 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2516 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2517 "--group", config.tracing_group_name.value,
2518 NULL);
2519 break;
2520 case LTTNG_CONSUMER64_UST:
2521 {
2522 if (config.consumerd64_lib_dir.value) {
2523 char *tmp;
2524 size_t tmplen;
2525 char *tmpnew;
2526
2527 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2528 if (!tmp) {
2529 tmp = "";
2530 }
2531 tmplen = strlen(config.consumerd64_lib_dir.value) + 1 /* : */ + strlen(tmp);
2532 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2533 if (!tmpnew) {
2534 ret = -ENOMEM;
2535 goto error;
2536 }
2537 strcat(tmpnew, config.consumerd64_lib_dir.value);
2538 if (tmp[0] != '\0') {
2539 strcat(tmpnew, ":");
2540 strcat(tmpnew, tmp);
2541 }
2542 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2543 free(tmpnew);
2544 if (ret) {
2545 ret = -errno;
2546 goto error;
2547 }
2548 }
2549 DBG("Using 64-bit UST consumer at: %s", config.consumerd64_bin_path.value);
2550 (void) execl(config.consumerd64_bin_path.value, "lttng-consumerd", verbosity, "-u",
2551 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2552 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2553 "--group", config.tracing_group_name.value,
2554 NULL);
2555 break;
2556 }
2557 case LTTNG_CONSUMER32_UST:
2558 {
2559 if (config.consumerd32_lib_dir.value) {
2560 char *tmp;
2561 size_t tmplen;
2562 char *tmpnew;
2563
2564 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2565 if (!tmp) {
2566 tmp = "";
2567 }
2568 tmplen = strlen(config.consumerd32_lib_dir.value) + 1 /* : */ + strlen(tmp);
2569 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2570 if (!tmpnew) {
2571 ret = -ENOMEM;
2572 goto error;
2573 }
2574 strcat(tmpnew, config.consumerd32_lib_dir.value);
2575 if (tmp[0] != '\0') {
2576 strcat(tmpnew, ":");
2577 strcat(tmpnew, tmp);
2578 }
2579 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2580 free(tmpnew);
2581 if (ret) {
2582 ret = -errno;
2583 goto error;
2584 }
2585 }
2586 DBG("Using 32-bit UST consumer at: %s", config.consumerd32_bin_path.value);
2587 (void) execl(config.consumerd32_bin_path.value, "lttng-consumerd", verbosity, "-u",
2588 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2589 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2590 "--group", config.tracing_group_name.value,
2591 NULL);
2592 break;
2593 }
2594 default:
2595 ERR("unknown consumer type");
2596 errno = 0;
2597 }
2598 if (errno != 0) {
2599 PERROR("Consumer execl()");
2600 }
2601 /* Reaching this point, we got a failure on our execl(). */
2602 exit(EXIT_FAILURE);
2603 } else if (pid > 0) {
2604 ret = pid;
2605 } else {
2606 PERROR("start consumer fork");
2607 ret = -errno;
2608 }
2609 error:
2610 return ret;
2611 }
2612
2613 /*
2614 * Spawn the consumerd daemon and session daemon thread.
2615 */
2616 static int start_consumerd(struct consumer_data *consumer_data)
2617 {
2618 int ret;
2619
2620 /*
2621 * Set the listen() state on the socket since there is a possible race
2622 * between the exec() of the consumer daemon and this call if place in the
2623 * consumer thread. See bug #366 for more details.
2624 */
2625 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2626 if (ret < 0) {
2627 goto error;
2628 }
2629
2630 pthread_mutex_lock(&consumer_data->pid_mutex);
2631 if (consumer_data->pid != 0) {
2632 pthread_mutex_unlock(&consumer_data->pid_mutex);
2633 goto end;
2634 }
2635
2636 ret = spawn_consumerd(consumer_data);
2637 if (ret < 0) {
2638 ERR("Spawning consumerd failed");
2639 pthread_mutex_unlock(&consumer_data->pid_mutex);
2640 goto error;
2641 }
2642
2643 /* Setting up the consumer_data pid */
2644 consumer_data->pid = ret;
2645 DBG2("Consumer pid %d", consumer_data->pid);
2646 pthread_mutex_unlock(&consumer_data->pid_mutex);
2647
2648 DBG2("Spawning consumer control thread");
2649 ret = spawn_consumer_thread(consumer_data);
2650 if (ret < 0) {
2651 ERR("Fatal error spawning consumer control thread");
2652 goto error;
2653 }
2654
2655 end:
2656 return 0;
2657
2658 error:
2659 /* Cleanup already created sockets on error. */
2660 if (consumer_data->err_sock >= 0) {
2661 int err;
2662
2663 err = close(consumer_data->err_sock);
2664 if (err < 0) {
2665 PERROR("close consumer data error socket");
2666 }
2667 }
2668 return ret;
2669 }
2670
2671 /*
2672 * Setup necessary data for kernel tracer action.
2673 */
2674 static int init_kernel_tracer(void)
2675 {
2676 int ret;
2677
2678 /* Modprobe lttng kernel modules */
2679 ret = modprobe_lttng_control();
2680 if (ret < 0) {
2681 goto error;
2682 }
2683
2684 /* Open debugfs lttng */
2685 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2686 if (kernel_tracer_fd < 0) {
2687 DBG("Failed to open %s", module_proc_lttng);
2688 goto error_open;
2689 }
2690
2691 /* Validate kernel version */
2692 ret = kernel_validate_version(kernel_tracer_fd, &kernel_tracer_version,
2693 &kernel_tracer_abi_version);
2694 if (ret < 0) {
2695 goto error_version;
2696 }
2697
2698 ret = modprobe_lttng_data();
2699 if (ret < 0) {
2700 goto error_modules;
2701 }
2702
2703 ret = kernel_supports_ring_buffer_snapshot_sample_positions(
2704 kernel_tracer_fd);
2705 if (ret < 0) {
2706 goto error_modules;
2707 }
2708
2709 if (ret < 1) {
2710 WARN("Kernel tracer does not support buffer monitoring. "
2711 "The monitoring timer of channels in the kernel domain "
2712 "will be set to 0 (disabled).");
2713 }
2714
2715 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2716 return 0;
2717
2718 error_version:
2719 modprobe_remove_lttng_control();
2720 ret = close(kernel_tracer_fd);
2721 if (ret) {
2722 PERROR("close");
2723 }
2724 kernel_tracer_fd = -1;
2725 return LTTNG_ERR_KERN_VERSION;
2726
2727 error_modules:
2728 ret = close(kernel_tracer_fd);
2729 if (ret) {
2730 PERROR("close");
2731 }
2732
2733 error_open:
2734 modprobe_remove_lttng_control();
2735
2736 error:
2737 WARN("No kernel tracer available");
2738 kernel_tracer_fd = -1;
2739 if (!is_root) {
2740 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2741 } else {
2742 return LTTNG_ERR_KERN_NA;
2743 }
2744 }
2745
2746
2747 /*
2748 * Copy consumer output from the tracing session to the domain session. The
2749 * function also applies the right modification on a per domain basis for the
2750 * trace files destination directory.
2751 *
2752 * Should *NOT* be called with RCU read-side lock held.
2753 */
2754 static int copy_session_consumer(int domain, struct ltt_session *session)
2755 {
2756 int ret;
2757 const char *dir_name;
2758 struct consumer_output *consumer;
2759
2760 assert(session);
2761 assert(session->consumer);
2762
2763 switch (domain) {
2764 case LTTNG_DOMAIN_KERNEL:
2765 DBG3("Copying tracing session consumer output in kernel session");
2766 /*
2767 * XXX: We should audit the session creation and what this function
2768 * does "extra" in order to avoid a destroy since this function is used
2769 * in the domain session creation (kernel and ust) only. Same for UST
2770 * domain.
2771 */
2772 if (session->kernel_session->consumer) {
2773 consumer_output_put(session->kernel_session->consumer);
2774 }
2775 session->kernel_session->consumer =
2776 consumer_copy_output(session->consumer);
2777 /* Ease our life a bit for the next part */
2778 consumer = session->kernel_session->consumer;
2779 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2780 break;
2781 case LTTNG_DOMAIN_JUL:
2782 case LTTNG_DOMAIN_LOG4J:
2783 case LTTNG_DOMAIN_PYTHON:
2784 case LTTNG_DOMAIN_UST:
2785 DBG3("Copying tracing session consumer output in UST session");
2786 if (session->ust_session->consumer) {
2787 consumer_output_put(session->ust_session->consumer);
2788 }
2789 session->ust_session->consumer =
2790 consumer_copy_output(session->consumer);
2791 /* Ease our life a bit for the next part */
2792 consumer = session->ust_session->consumer;
2793 dir_name = DEFAULT_UST_TRACE_DIR;
2794 break;
2795 default:
2796 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2797 goto error;
2798 }
2799
2800 /* Append correct directory to subdir */
2801 strncat(consumer->subdir, dir_name,
2802 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2803 DBG3("Copy session consumer subdir %s", consumer->subdir);
2804
2805 ret = LTTNG_OK;
2806
2807 error:
2808 return ret;
2809 }
2810
2811 /*
2812 * Create an UST session and add it to the session ust list.
2813 *
2814 * Should *NOT* be called with RCU read-side lock held.
2815 */
2816 static int create_ust_session(struct ltt_session *session,
2817 struct lttng_domain *domain)
2818 {
2819 int ret;
2820 struct ltt_ust_session *lus = NULL;
2821
2822 assert(session);
2823 assert(domain);
2824 assert(session->consumer);
2825
2826 switch (domain->type) {
2827 case LTTNG_DOMAIN_JUL:
2828 case LTTNG_DOMAIN_LOG4J:
2829 case LTTNG_DOMAIN_PYTHON:
2830 case LTTNG_DOMAIN_UST:
2831 break;
2832 default:
2833 ERR("Unknown UST domain on create session %d", domain->type);
2834 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2835 goto error;
2836 }
2837
2838 DBG("Creating UST session");
2839
2840 lus = trace_ust_create_session(session->id);
2841 if (lus == NULL) {
2842 ret = LTTNG_ERR_UST_SESS_FAIL;
2843 goto error;
2844 }
2845
2846 lus->uid = session->uid;
2847 lus->gid = session->gid;
2848 lus->output_traces = session->output_traces;
2849 lus->snapshot_mode = session->snapshot_mode;
2850 lus->live_timer_interval = session->live_timer;
2851 session->ust_session = lus;
2852 if (session->shm_path[0]) {
2853 strncpy(lus->root_shm_path, session->shm_path,
2854 sizeof(lus->root_shm_path));
2855 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2856 strncpy(lus->shm_path, session->shm_path,
2857 sizeof(lus->shm_path));
2858 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2859 strncat(lus->shm_path, "/ust",
2860 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2861 }
2862 /* Copy session output to the newly created UST session */
2863 ret = copy_session_consumer(domain->type, session);
2864 if (ret != LTTNG_OK) {
2865 goto error;
2866 }
2867
2868 return LTTNG_OK;
2869
2870 error:
2871 free(lus);
2872 session->ust_session = NULL;
2873 return ret;
2874 }
2875
2876 /*
2877 * Create a kernel tracer session then create the default channel.
2878 */
2879 static int create_kernel_session(struct ltt_session *session)
2880 {
2881 int ret;
2882
2883 DBG("Creating kernel session");
2884
2885 ret = kernel_create_session(session, kernel_tracer_fd);
2886 if (ret < 0) {
2887 ret = LTTNG_ERR_KERN_SESS_FAIL;
2888 goto error;
2889 }
2890
2891 /* Code flow safety */
2892 assert(session->kernel_session);
2893
2894 /* Copy session output to the newly created Kernel session */
2895 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2896 if (ret != LTTNG_OK) {
2897 goto error;
2898 }
2899
2900 session->kernel_session->uid = session->uid;
2901 session->kernel_session->gid = session->gid;
2902 session->kernel_session->output_traces = session->output_traces;
2903 session->kernel_session->snapshot_mode = session->snapshot_mode;
2904
2905 return LTTNG_OK;
2906
2907 error:
2908 trace_kernel_destroy_session(session->kernel_session);
2909 session->kernel_session = NULL;
2910 return ret;
2911 }
2912
2913 /*
2914 * Count number of session permitted by uid/gid.
2915 */
2916 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2917 {
2918 unsigned int i = 0;
2919 struct ltt_session *session;
2920
2921 DBG("Counting number of available session for UID %d GID %d",
2922 uid, gid);
2923 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2924 /*
2925 * Only list the sessions the user can control.
2926 */
2927 if (!session_access_ok(session, uid, gid)) {
2928 continue;
2929 }
2930 i++;
2931 }
2932 return i;
2933 }
2934
2935 static int receive_userspace_probe(struct command_ctx *cmd_ctx, int sock,
2936 int *sock_error, struct lttng_event *event)
2937 {
2938 int fd, ret;
2939 struct lttng_userspace_probe_location *probe_location;
2940 const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
2941 struct lttng_dynamic_buffer probe_location_buffer;
2942 struct lttng_buffer_view buffer_view;
2943
2944 /*
2945 * Create a buffer to store the serialized version of the probe
2946 * location.
2947 */
2948 lttng_dynamic_buffer_init(&probe_location_buffer);
2949 ret = lttng_dynamic_buffer_set_size(&probe_location_buffer,
2950 cmd_ctx->lsm->u.enable.userspace_probe_location_len);
2951 if (ret) {
2952 ret = LTTNG_ERR_NOMEM;
2953 goto error;
2954 }
2955
2956 /*
2957 * Receive the probe location.
2958 */
2959 ret = lttcomm_recv_unix_sock(sock, probe_location_buffer.data,
2960 probe_location_buffer.size);
2961 if (ret <= 0) {
2962 DBG("Nothing recv() from client var len data... continuing");
2963 *sock_error = 1;
2964 lttng_dynamic_buffer_reset(&probe_location_buffer);
2965 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2966 goto error;
2967 }
2968
2969 buffer_view = lttng_buffer_view_from_dynamic_buffer(
2970 &probe_location_buffer, 0, probe_location_buffer.size);
2971
2972 /*
2973 * Extract the probe location from the serialized version.
2974 */
2975 ret = lttng_userspace_probe_location_create_from_buffer(
2976 &buffer_view, &probe_location);
2977 if (ret < 0) {
2978 WARN("Failed to create a userspace probe location from the received buffer");
2979 lttng_dynamic_buffer_reset( &probe_location_buffer);
2980 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2981 goto error;
2982 }
2983
2984 /*
2985 * Receive the file descriptor to the target binary from the client.
2986 */
2987 DBG("Receiving userspace probe target FD from client ...");
2988 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
2989 if (ret <= 0) {
2990 DBG("Nothing recv() from client userspace probe fd... continuing");
2991 *sock_error = 1;
2992 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2993 goto error;
2994 }
2995
2996 /*
2997 * Set the file descriptor received from the client through the unix
2998 * socket in the probe location.
2999 */
3000 lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
3001 if (!lookup) {
3002 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
3003 goto error;
3004 }
3005
3006 /*
3007 * From the kernel tracer's perspective, all userspace probe event types
3008 * are all the same: a file and an offset.
3009 */
3010 switch (lttng_userspace_probe_location_lookup_method_get_type(lookup)) {
3011 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
3012 ret = lttng_userspace_probe_location_function_set_binary_fd(
3013 probe_location, fd);
3014 break;
3015 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
3016 ret = lttng_userspace_probe_location_tracepoint_set_binary_fd(
3017 probe_location, fd);
3018 break;
3019 default:
3020 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
3021 goto error;
3022 }
3023
3024 if (ret) {
3025 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
3026 goto error;
3027 }
3028
3029 /* Attach the probe location to the event. */
3030 ret = lttng_event_set_userspace_probe_location(event, probe_location);
3031 if (ret) {
3032 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
3033 goto error;
3034 }
3035
3036 lttng_dynamic_buffer_reset(&probe_location_buffer);
3037 error:
3038 return ret;
3039 }
3040
3041 /*
3042 * Check if the current kernel tracer supports the session rotation feature.
3043 * Return 1 if it does, 0 otherwise.
3044 */
3045 static int check_rotate_compatible(void)
3046 {
3047 int ret = 1;
3048
3049 if (kernel_tracer_version.major != 2 || kernel_tracer_version.minor < 11) {
3050 DBG("Kernel tracer version is not compatible with the rotation feature");
3051 ret = 0;
3052 }
3053
3054 return ret;
3055 }
3056
3057 /*
3058 * Process the command requested by the lttng client within the command
3059 * context structure. This function make sure that the return structure (llm)
3060 * is set and ready for transmission before returning.
3061 *
3062 * Return any error encountered or 0 for success.
3063 *
3064 * "sock" is only used for special-case var. len data.
3065 *
3066 * Should *NOT* be called with RCU read-side lock held.
3067 */
3068 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
3069 int *sock_error)
3070 {
3071 int ret = LTTNG_OK;
3072 int need_tracing_session = 1;
3073 int need_domain;
3074
3075 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
3076
3077 assert(!rcu_read_ongoing());
3078
3079 *sock_error = 0;
3080
3081 switch (cmd_ctx->lsm->cmd_type) {
3082 case LTTNG_CREATE_SESSION:
3083 case LTTNG_CREATE_SESSION_SNAPSHOT:
3084 case LTTNG_CREATE_SESSION_LIVE:
3085 case LTTNG_DESTROY_SESSION:
3086 case LTTNG_LIST_SESSIONS:
3087 case LTTNG_LIST_DOMAINS:
3088 case LTTNG_START_TRACE:
3089 case LTTNG_STOP_TRACE:
3090 case LTTNG_DATA_PENDING:
3091 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3092 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3093 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3094 case LTTNG_SNAPSHOT_RECORD:
3095 case LTTNG_SAVE_SESSION:
3096 case LTTNG_SET_SESSION_SHM_PATH:
3097 case LTTNG_REGENERATE_METADATA:
3098 case LTTNG_REGENERATE_STATEDUMP:
3099 case LTTNG_REGISTER_TRIGGER:
3100 case LTTNG_UNREGISTER_TRIGGER:
3101 case LTTNG_ROTATE_SESSION:
3102 case LTTNG_ROTATION_GET_INFO:
3103 case LTTNG_SESSION_GET_CURRENT_OUTPUT:
3104 case LTTNG_ROTATION_SET_SCHEDULE:
3105 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES:
3106 need_domain = 0;
3107 break;
3108 default:
3109 need_domain = 1;
3110 }
3111
3112 if (config.no_kernel && need_domain
3113 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3114 if (!is_root) {
3115 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3116 } else {
3117 ret = LTTNG_ERR_KERN_NA;
3118 }
3119 goto error;
3120 }
3121
3122 /* Deny register consumer if we already have a spawned consumer. */
3123 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3124 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3125 if (kconsumer_data.pid > 0) {
3126 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3127 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3128 goto error;
3129 }
3130 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3131 }
3132
3133 /*
3134 * Check for command that don't needs to allocate a returned payload. We do
3135 * this here so we don't have to make the call for no payload at each
3136 * command.
3137 */
3138 switch(cmd_ctx->lsm->cmd_type) {
3139 case LTTNG_LIST_SESSIONS:
3140 case LTTNG_LIST_TRACEPOINTS:
3141 case LTTNG_LIST_TRACEPOINT_FIELDS:
3142 case LTTNG_LIST_DOMAINS:
3143 case LTTNG_LIST_CHANNELS:
3144 case LTTNG_LIST_EVENTS:
3145 case LTTNG_LIST_SYSCALLS:
3146 case LTTNG_LIST_TRACKER_PIDS:
3147 case LTTNG_DATA_PENDING:
3148 case LTTNG_ROTATE_SESSION:
3149 case LTTNG_ROTATION_GET_INFO:
3150 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES:
3151 break;
3152 default:
3153 /* Setup lttng message with no payload */
3154 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
3155 if (ret < 0) {
3156 /* This label does not try to unlock the session */
3157 goto init_setup_error;
3158 }
3159 }
3160
3161 /* Commands that DO NOT need a session. */
3162 switch (cmd_ctx->lsm->cmd_type) {
3163 case LTTNG_CREATE_SESSION:
3164 case LTTNG_CREATE_SESSION_SNAPSHOT:
3165 case LTTNG_CREATE_SESSION_LIVE:
3166 case LTTNG_LIST_SESSIONS:
3167 case LTTNG_LIST_TRACEPOINTS:
3168 case LTTNG_LIST_SYSCALLS:
3169 case LTTNG_LIST_TRACEPOINT_FIELDS:
3170 case LTTNG_SAVE_SESSION:
3171 case LTTNG_REGISTER_TRIGGER:
3172 case LTTNG_UNREGISTER_TRIGGER:
3173 need_tracing_session = 0;
3174 break;
3175 default:
3176 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3177 /*
3178 * We keep the session list lock across _all_ commands
3179 * for now, because the per-session lock does not
3180 * handle teardown properly.
3181 */
3182 session_lock_list();
3183 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3184 if (cmd_ctx->session == NULL) {
3185 ret = LTTNG_ERR_SESS_NOT_FOUND;
3186 goto error;
3187 } else {
3188 /* Acquire lock for the session */
3189 session_lock(cmd_ctx->session);
3190 }
3191 break;
3192 }
3193
3194 /*
3195 * Commands that need a valid session but should NOT create one if none
3196 * exists. Instead of creating one and destroying it when the command is
3197 * handled, process that right before so we save some round trip in useless
3198 * code path.
3199 */
3200 switch (cmd_ctx->lsm->cmd_type) {
3201 case LTTNG_DISABLE_CHANNEL:
3202 case LTTNG_DISABLE_EVENT:
3203 switch (cmd_ctx->lsm->domain.type) {
3204 case LTTNG_DOMAIN_KERNEL:
3205 if (!cmd_ctx->session->kernel_session) {
3206 ret = LTTNG_ERR_NO_CHANNEL;
3207 goto error;
3208 }
3209 break;
3210 case LTTNG_DOMAIN_JUL:
3211 case LTTNG_DOMAIN_LOG4J:
3212 case LTTNG_DOMAIN_PYTHON:
3213 case LTTNG_DOMAIN_UST:
3214 if (!cmd_ctx->session->ust_session) {
3215 ret = LTTNG_ERR_NO_CHANNEL;
3216 goto error;
3217 }
3218 break;
3219 default:
3220 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3221 goto error;
3222 }
3223 default:
3224 break;
3225 }
3226
3227 if (!need_domain) {
3228 goto skip_domain;
3229 }
3230
3231 /*
3232 * Check domain type for specific "pre-action".
3233 */
3234 switch (cmd_ctx->lsm->domain.type) {
3235 case LTTNG_DOMAIN_KERNEL:
3236 if (!is_root) {
3237 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3238 goto error;
3239 }
3240
3241 /* Kernel tracer check */
3242 if (kernel_tracer_fd == -1) {
3243 /* Basically, load kernel tracer modules */
3244 ret = init_kernel_tracer();
3245 if (ret != 0) {
3246 goto error;
3247 }
3248 }
3249
3250 /* Consumer is in an ERROR state. Report back to client */
3251 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3252 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3253 goto error;
3254 }
3255
3256 /* Need a session for kernel command */
3257 if (need_tracing_session) {
3258 if (cmd_ctx->session->kernel_session == NULL) {
3259 ret = create_kernel_session(cmd_ctx->session);
3260 if (ret < 0) {
3261 ret = LTTNG_ERR_KERN_SESS_FAIL;
3262 goto error;
3263 }
3264 }
3265
3266 /* Start the kernel consumer daemon */
3267 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3268 if (kconsumer_data.pid == 0 &&
3269 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3270 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3271 ret = start_consumerd(&kconsumer_data);
3272 if (ret < 0) {
3273 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3274 goto error;
3275 }
3276 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3277 } else {
3278 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3279 }
3280
3281 /*
3282 * The consumer was just spawned so we need to add the socket to
3283 * the consumer output of the session if exist.
3284 */
3285 ret = consumer_create_socket(&kconsumer_data,
3286 cmd_ctx->session->kernel_session->consumer);
3287 if (ret < 0) {
3288 goto error;
3289 }
3290 }
3291
3292 break;
3293 case LTTNG_DOMAIN_JUL:
3294 case LTTNG_DOMAIN_LOG4J:
3295 case LTTNG_DOMAIN_PYTHON:
3296 case LTTNG_DOMAIN_UST:
3297 {
3298 if (!ust_app_supported()) {
3299 ret = LTTNG_ERR_NO_UST;
3300 goto error;
3301 }
3302 /* Consumer is in an ERROR state. Report back to client */
3303 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3304 ret = LTTNG_ERR_NO_USTCONSUMERD;
3305 goto error;
3306 }
3307
3308 if (need_tracing_session) {
3309 /* Create UST session if none exist. */
3310 if (cmd_ctx->session->ust_session == NULL) {
3311 ret = create_ust_session(cmd_ctx->session,
3312 &cmd_ctx->lsm->domain);
3313 if (ret != LTTNG_OK) {
3314 goto error;
3315 }
3316 }
3317
3318 /* Start the UST consumer daemons */
3319 /* 64-bit */
3320 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3321 if (config.consumerd64_bin_path.value &&
3322 ustconsumer64_data.pid == 0 &&
3323 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3324 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3325 ret = start_consumerd(&ustconsumer64_data);
3326 if (ret < 0) {
3327 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3328 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3329 goto error;
3330 }
3331
3332 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3333 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3334 } else {
3335 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3336 }
3337
3338 /*
3339 * Setup socket for consumer 64 bit. No need for atomic access
3340 * since it was set above and can ONLY be set in this thread.
3341 */
3342 ret = consumer_create_socket(&ustconsumer64_data,
3343 cmd_ctx->session->ust_session->consumer);
3344 if (ret < 0) {
3345 goto error;
3346 }
3347
3348 /* 32-bit */
3349 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3350 if (config.consumerd32_bin_path.value &&
3351 ustconsumer32_data.pid == 0 &&
3352 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3353 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3354 ret = start_consumerd(&ustconsumer32_data);
3355 if (ret < 0) {
3356 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3357 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3358 goto error;
3359 }
3360
3361 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3362 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3363 } else {
3364 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3365 }
3366
3367 /*
3368 * Setup socket for consumer 64 bit. No need for atomic access
3369 * since it was set above and can ONLY be set in this thread.
3370 */
3371 ret = consumer_create_socket(&ustconsumer32_data,
3372 cmd_ctx->session->ust_session->consumer);
3373 if (ret < 0) {
3374 goto error;
3375 }
3376 }
3377 break;
3378 }
3379 default:
3380 break;
3381 }
3382 skip_domain:
3383
3384 /* Validate consumer daemon state when start/stop trace command */
3385 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3386 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3387 switch (cmd_ctx->lsm->domain.type) {
3388 case LTTNG_DOMAIN_NONE:
3389 break;
3390 case LTTNG_DOMAIN_JUL:
3391 case LTTNG_DOMAIN_LOG4J:
3392 case LTTNG_DOMAIN_PYTHON:
3393 case LTTNG_DOMAIN_UST:
3394 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3395 ret = LTTNG_ERR_NO_USTCONSUMERD;
3396 goto error;
3397 }
3398 break;
3399 case LTTNG_DOMAIN_KERNEL:
3400 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3401 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3402 goto error;
3403 }
3404 break;
3405 default:
3406 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3407 goto error;
3408 }
3409 }
3410
3411 /*
3412 * Check that the UID or GID match that of the tracing session.
3413 * The root user can interact with all sessions.
3414 */
3415 if (need_tracing_session) {
3416 if (!session_access_ok(cmd_ctx->session,
3417 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3418 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3419 ret = LTTNG_ERR_EPERM;
3420 goto error;
3421 }
3422 }
3423
3424 /*
3425 * Send relayd information to consumer as soon as we have a domain and a
3426 * session defined.
3427 */
3428 if (cmd_ctx->session && need_domain) {
3429 /*
3430 * Setup relayd if not done yet. If the relayd information was already
3431 * sent to the consumer, this call will gracefully return.
3432 */
3433 ret = cmd_setup_relayd(cmd_ctx->session);
3434 if (ret != LTTNG_OK) {
3435 goto error;
3436 }
3437 }
3438
3439 /* Process by command type */
3440 switch (cmd_ctx->lsm->cmd_type) {
3441 case LTTNG_ADD_CONTEXT:
3442 {
3443 /*
3444 * An LTTNG_ADD_CONTEXT command might have a supplementary
3445 * payload if the context being added is an application context.
3446 */
3447 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3448 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3449 char *provider_name = NULL, *context_name = NULL;
3450 size_t provider_name_len =
3451 cmd_ctx->lsm->u.context.provider_name_len;
3452 size_t context_name_len =
3453 cmd_ctx->lsm->u.context.context_name_len;
3454
3455 if (provider_name_len == 0 || context_name_len == 0) {
3456 /*
3457 * Application provider and context names MUST
3458 * be provided.
3459 */
3460 ret = -LTTNG_ERR_INVALID;
3461 goto error;
3462 }
3463
3464 provider_name = zmalloc(provider_name_len + 1);
3465 if (!provider_name) {
3466 ret = -LTTNG_ERR_NOMEM;
3467 goto error;
3468 }
3469 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3470 provider_name;
3471
3472 context_name = zmalloc(context_name_len + 1);
3473 if (!context_name) {
3474 ret = -LTTNG_ERR_NOMEM;
3475 goto error_add_context;
3476 }
3477 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3478 context_name;
3479
3480 ret = lttcomm_recv_unix_sock(sock, provider_name,
3481 provider_name_len);
3482 if (ret < 0) {
3483 goto error_add_context;
3484 }
3485
3486 ret = lttcomm_recv_unix_sock(sock, context_name,
3487 context_name_len);
3488 if (ret < 0) {
3489 goto error_add_context;
3490 }
3491 }
3492
3493 /*
3494 * cmd_add_context assumes ownership of the provider and context
3495 * names.
3496 */
3497 ret = cmd_add_context(cmd_ctx->session,
3498 cmd_ctx->lsm->domain.type,
3499 cmd_ctx->lsm->u.context.channel_name,
3500 &cmd_ctx->lsm->u.context.ctx,
3501 kernel_poll_pipe[1]);
3502
3503 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3504 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3505 error_add_context:
3506 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3507 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3508 if (ret < 0) {
3509 goto error;
3510 }
3511 break;
3512 }
3513 case LTTNG_DISABLE_CHANNEL:
3514 {
3515 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3516 cmd_ctx->lsm->u.disable.channel_name);
3517 break;
3518 }
3519 case LTTNG_DISABLE_EVENT:
3520 {
3521
3522 /*
3523 * FIXME: handle filter; for now we just receive the filter's
3524 * bytecode along with the filter expression which are sent by
3525 * liblttng-ctl and discard them.
3526 *
3527 * This fixes an issue where the client may block while sending
3528 * the filter payload and encounter an error because the session
3529 * daemon closes the socket without ever handling this data.
3530 */
3531 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3532 cmd_ctx->lsm->u.disable.bytecode_len;
3533
3534 if (count) {
3535 char data[LTTNG_FILTER_MAX_LEN];
3536
3537 DBG("Discarding disable event command payload of size %zu", count);
3538 while (count) {
3539 ret = lttcomm_recv_unix_sock(sock, data,
3540 count > sizeof(data) ? sizeof(data) : count);
3541 if (ret < 0) {
3542 goto error;
3543 }
3544
3545 count -= (size_t) ret;
3546 }
3547 }
3548 /* FIXME: passing packed structure to non-packed pointer */
3549 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3550 cmd_ctx->lsm->u.disable.channel_name,
3551 &cmd_ctx->lsm->u.disable.event);
3552 break;
3553 }
3554 case LTTNG_ENABLE_CHANNEL:
3555 {
3556 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3557 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3558 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3559 &cmd_ctx->lsm->u.channel.chan,
3560 kernel_poll_pipe[1]);
3561 break;
3562 }
3563 case LTTNG_TRACK_PID:
3564 {
3565 ret = cmd_track_pid(cmd_ctx->session,
3566 cmd_ctx->lsm->domain.type,
3567 cmd_ctx->lsm->u.pid_tracker.pid);
3568 break;
3569 }
3570 case LTTNG_UNTRACK_PID:
3571 {
3572 ret = cmd_untrack_pid(cmd_ctx->session,
3573 cmd_ctx->lsm->domain.type,
3574 cmd_ctx->lsm->u.pid_tracker.pid);
3575 break;
3576 }
3577 case LTTNG_ENABLE_EVENT:
3578 {
3579 struct lttng_event *ev = NULL;
3580 struct lttng_event_exclusion *exclusion = NULL;
3581 struct lttng_filter_bytecode *bytecode = NULL;
3582 char *filter_expression = NULL;
3583
3584 /* Handle exclusion events and receive it from the client. */
3585 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3586 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3587
3588 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3589 (count * LTTNG_SYMBOL_NAME_LEN));
3590 if (!exclusion) {
3591 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3592 goto error;
3593 }
3594
3595 DBG("Receiving var len exclusion event list from client ...");
3596 exclusion->count = count;
3597 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3598 count * LTTNG_SYMBOL_NAME_LEN);
3599 if (ret <= 0) {
3600 DBG("Nothing recv() from client var len data... continuing");
3601 *sock_error = 1;
3602 free(exclusion);
3603 ret = LTTNG_ERR_EXCLUSION_INVAL;
3604 goto error;
3605 }
3606 }
3607
3608 /* Get filter expression from client. */
3609 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3610 size_t expression_len =
3611 cmd_ctx->lsm->u.enable.expression_len;
3612
3613 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3614 ret = LTTNG_ERR_FILTER_INVAL;
3615 free(exclusion);
3616 goto error;
3617 }
3618
3619 filter_expression = zmalloc(expression_len);
3620 if (!filter_expression) {
3621 free(exclusion);
3622 ret = LTTNG_ERR_FILTER_NOMEM;
3623 goto error;
3624 }
3625
3626 /* Receive var. len. data */
3627 DBG("Receiving var len filter's expression from client ...");
3628 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3629 expression_len);
3630 if (ret <= 0) {
3631 DBG("Nothing recv() from client var len data... continuing");
3632 *sock_error = 1;
3633 free(filter_expression);
3634 free(exclusion);
3635 ret = LTTNG_ERR_FILTER_INVAL;
3636 goto error;
3637 }
3638 }
3639
3640 /* Handle filter and get bytecode from client. */
3641 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3642 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3643
3644 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3645 ret = LTTNG_ERR_FILTER_INVAL;
3646 free(filter_expression);
3647 free(exclusion);
3648 goto error;
3649 }
3650
3651 bytecode = zmalloc(bytecode_len);
3652 if (!bytecode) {
3653 free(filter_expression);
3654 free(exclusion);
3655 ret = LTTNG_ERR_FILTER_NOMEM;
3656 goto error;
3657 }
3658
3659 /* Receive var. len. data */
3660 DBG("Receiving var len filter's bytecode from client ...");
3661 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3662 if (ret <= 0) {
3663 DBG("Nothing recv() from client var len data... continuing");
3664 *sock_error = 1;
3665 free(filter_expression);
3666 free(bytecode);
3667 free(exclusion);
3668 ret = LTTNG_ERR_FILTER_INVAL;
3669 goto error;
3670 }
3671
3672 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3673 free(filter_expression);
3674 free(bytecode);
3675 free(exclusion);
3676 ret = LTTNG_ERR_FILTER_INVAL;
3677 goto error;
3678 }
3679 }
3680
3681 ev = lttng_event_copy(&cmd_ctx->lsm->u.enable.event);
3682 if (!ev) {
3683 DBG("Failed to copy event: %s",
3684 cmd_ctx->lsm->u.enable.event.name);
3685 ret = LTTNG_ERR_NOMEM;
3686 goto error;
3687 }
3688
3689
3690 if (cmd_ctx->lsm->u.enable.userspace_probe_location_len > 0) {
3691 /* Expect a userspace probe description. */
3692 ret = receive_userspace_probe(cmd_ctx, sock, sock_error, ev);
3693 if (ret) {
3694 lttng_event_destroy(ev);
3695 goto error;
3696 }
3697 }
3698
3699 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3700 cmd_ctx->lsm->u.enable.channel_name,
3701 ev,
3702 filter_expression, bytecode, exclusion,
3703 kernel_poll_pipe[1]);
3704 lttng_event_destroy(ev);
3705 break;
3706 }
3707 case LTTNG_LIST_TRACEPOINTS:
3708 {
3709 struct lttng_event *events;
3710 ssize_t nb_events;
3711
3712 session_lock_list();
3713 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3714 session_unlock_list();
3715 if (nb_events < 0) {
3716 /* Return value is a negative lttng_error_code. */
3717 ret = -nb_events;
3718 goto error;
3719 }
3720
3721 /*
3722 * Setup lttng message with payload size set to the event list size in
3723 * bytes and then copy list into the llm payload.
3724 */
3725 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3726 sizeof(struct lttng_event) * nb_events);
3727 free(events);
3728
3729 if (ret < 0) {
3730 goto setup_error;
3731 }
3732
3733 ret = LTTNG_OK;
3734 break;
3735 }
3736 case LTTNG_LIST_TRACEPOINT_FIELDS:
3737 {
3738 struct lttng_event_field *fields;
3739 ssize_t nb_fields;
3740
3741 session_lock_list();
3742 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3743 &fields);
3744 session_unlock_list();
3745 if (nb_fields < 0) {
3746 /* Return value is a negative lttng_error_code. */
3747 ret = -nb_fields;
3748 goto error;
3749 }
3750
3751 /*
3752 * Setup lttng message with payload size set to the event list size in
3753 * bytes and then copy list into the llm payload.
3754 */
3755 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3756 sizeof(struct lttng_event_field) * nb_fields);
3757 free(fields);
3758
3759 if (ret < 0) {
3760 goto setup_error;
3761 }
3762
3763 ret = LTTNG_OK;
3764 break;
3765 }
3766 case LTTNG_LIST_SYSCALLS:
3767 {
3768 struct lttng_event *events;
3769 ssize_t nb_events;
3770
3771 nb_events = cmd_list_syscalls(&events);
3772 if (nb_events < 0) {
3773 /* Return value is a negative lttng_error_code. */
3774 ret = -nb_events;
3775 goto error;
3776 }
3777
3778 /*
3779 * Setup lttng message with payload size set to the event list size in
3780 * bytes and then copy list into the llm payload.
3781 */
3782 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3783 sizeof(struct lttng_event) * nb_events);
3784 free(events);
3785
3786 if (ret < 0) {
3787 goto setup_error;
3788 }
3789
3790 ret = LTTNG_OK;
3791 break;
3792 }
3793 case LTTNG_LIST_TRACKER_PIDS:
3794 {
3795 int32_t *pids = NULL;
3796 ssize_t nr_pids;
3797
3798 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3799 cmd_ctx->lsm->domain.type, &pids);
3800 if (nr_pids < 0) {
3801 /* Return value is a negative lttng_error_code. */
3802 ret = -nr_pids;
3803 goto error;
3804 }
3805
3806 /*
3807 * Setup lttng message with payload size set to the event list size in
3808 * bytes and then copy list into the llm payload.
3809 */
3810 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3811 sizeof(int32_t) * nr_pids);
3812 free(pids);
3813
3814 if (ret < 0) {
3815 goto setup_error;
3816 }
3817
3818 ret = LTTNG_OK;
3819 break;
3820 }
3821 case LTTNG_SET_CONSUMER_URI:
3822 {
3823 size_t nb_uri, len;
3824 struct lttng_uri *uris;
3825
3826 nb_uri = cmd_ctx->lsm->u.uri.size;
3827 len = nb_uri * sizeof(struct lttng_uri);
3828
3829 if (nb_uri == 0) {
3830 ret = LTTNG_ERR_INVALID;
3831 goto error;
3832 }
3833
3834 uris = zmalloc(len);
3835 if (uris == NULL) {
3836 ret = LTTNG_ERR_FATAL;
3837 goto error;
3838 }
3839
3840 /* Receive variable len data */
3841 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3842 ret = lttcomm_recv_unix_sock(sock, uris, len);
3843 if (ret <= 0) {
3844 DBG("No URIs received from client... continuing");
3845 *sock_error = 1;
3846 ret = LTTNG_ERR_SESSION_FAIL;
3847 free(uris);
3848 goto error;
3849 }
3850
3851 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3852 free(uris);
3853 if (ret != LTTNG_OK) {
3854 goto error;
3855 }
3856
3857
3858 break;
3859 }
3860 case LTTNG_START_TRACE:
3861 {
3862 /*
3863 * On the first start, if we have a kernel session and we have
3864 * enabled time or size-based rotations, we have to make sure
3865 * the kernel tracer supports it.
3866 */
3867 if (!cmd_ctx->session->has_been_started && \
3868 cmd_ctx->session->kernel_session && \
3869 (cmd_ctx->session->rotate_timer_period || \
3870 cmd_ctx->session->rotate_size) && \
3871 !check_rotate_compatible()) {
3872 DBG("Kernel tracer version is not compatible with the rotation feature");
3873 ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
3874 goto error;
3875 }
3876 ret = cmd_start_trace(cmd_ctx->session);
3877 break;
3878 }
3879 case LTTNG_STOP_TRACE:
3880 {
3881 ret = cmd_stop_trace(cmd_ctx->session);
3882 break;
3883 }
3884 case LTTNG_CREATE_SESSION:
3885 {
3886 size_t nb_uri, len;
3887 struct lttng_uri *uris = NULL;
3888
3889 nb_uri = cmd_ctx->lsm->u.uri.size;
3890 len = nb_uri * sizeof(struct lttng_uri);
3891
3892 if (nb_uri > 0) {
3893 uris = zmalloc(len);
3894 if (uris == NULL) {
3895 ret = LTTNG_ERR_FATAL;
3896 goto error;
3897 }
3898
3899 /* Receive variable len data */
3900 DBG("Waiting for %zu URIs from client ...", nb_uri);
3901 ret = lttcomm_recv_unix_sock(sock, uris, len);
3902 if (ret <= 0) {
3903 DBG("No URIs received from client... continuing");
3904 *sock_error = 1;
3905 ret = LTTNG_ERR_SESSION_FAIL;
3906 free(uris);
3907 goto error;
3908 }
3909
3910 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3911 DBG("Creating session with ONE network URI is a bad call");
3912 ret = LTTNG_ERR_SESSION_FAIL;
3913 free(uris);
3914 goto error;
3915 }
3916 }
3917
3918 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3919 &cmd_ctx->creds, 0);
3920
3921 free(uris);
3922
3923 break;
3924 }
3925 case LTTNG_DESTROY_SESSION:
3926 {
3927 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1],
3928 notification_thread_handle);
3929
3930 /* Set session to NULL so we do not unlock it after free. */
3931 cmd_ctx->session = NULL;
3932 break;
3933 }
3934 case LTTNG_LIST_DOMAINS:
3935 {
3936 ssize_t nb_dom;
3937 struct lttng_domain *domains = NULL;
3938
3939 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3940 if (nb_dom < 0) {
3941 /* Return value is a negative lttng_error_code. */
3942 ret = -nb_dom;
3943 goto error;
3944 }
3945
3946 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3947 nb_dom * sizeof(struct lttng_domain));
3948 free(domains);
3949
3950 if (ret < 0) {
3951 goto setup_error;
3952 }
3953
3954 ret = LTTNG_OK;
3955 break;
3956 }
3957 case LTTNG_LIST_CHANNELS:
3958 {
3959 ssize_t payload_size;
3960 struct lttng_channel *channels = NULL;
3961
3962 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3963 cmd_ctx->session, &channels);
3964 if (payload_size < 0) {
3965 /* Return value is a negative lttng_error_code. */
3966 ret = -payload_size;
3967 goto error;
3968 }
3969
3970 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3971 payload_size);
3972 free(channels);
3973
3974 if (ret < 0) {
3975 goto setup_error;
3976 }
3977
3978 ret = LTTNG_OK;
3979 break;
3980 }
3981 case LTTNG_LIST_EVENTS:
3982 {
3983 ssize_t nb_event;
3984 struct lttng_event *events = NULL;
3985 struct lttcomm_event_command_header cmd_header;
3986 size_t total_size;
3987
3988 memset(&cmd_header, 0, sizeof(cmd_header));
3989 /* Extended infos are included at the end of events */
3990 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3991 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3992 &events, &total_size);
3993
3994 if (nb_event < 0) {
3995 /* Return value is a negative lttng_error_code. */
3996 ret = -nb_event;
3997 goto error;
3998 }
3999
4000 cmd_header.nb_events = nb_event;
4001 ret = setup_lttng_msg(cmd_ctx, events, total_size,
4002 &cmd_header, sizeof(cmd_header));
4003 free(events);
4004
4005 if (ret < 0) {
4006 goto setup_error;
4007 }
4008
4009 ret = LTTNG_OK;
4010 break;
4011 }
4012 case LTTNG_LIST_SESSIONS:
4013 {
4014 unsigned int nr_sessions;
4015 void *sessions_payload;
4016 size_t payload_len;
4017
4018 session_lock_list();
4019 nr_sessions = lttng_sessions_count(
4020 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
4021 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
4022 payload_len = sizeof(struct lttng_session) * nr_sessions;
4023 sessions_payload = zmalloc(payload_len);
4024
4025 if (!sessions_payload) {
4026 session_unlock_list();
4027 ret = -ENOMEM;
4028 goto setup_error;
4029 }
4030
4031 cmd_list_lttng_sessions(sessions_payload,
4032 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
4033 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
4034 session_unlock_list();
4035
4036 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
4037 payload_len);
4038 free(sessions_payload);
4039
4040 if (ret < 0) {
4041 goto setup_error;
4042 }
4043
4044 ret = LTTNG_OK;
4045 break;
4046 }
4047 case LTTNG_REGISTER_CONSUMER:
4048 {
4049 struct consumer_data *cdata;
4050
4051 switch (cmd_ctx->lsm->domain.type) {
4052 case LTTNG_DOMAIN_KERNEL:
4053 cdata = &kconsumer_data;
4054 break;
4055 default:
4056 ret = LTTNG_ERR_UND;
4057 goto error;
4058 }
4059
4060 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
4061 cmd_ctx->lsm->u.reg.path, cdata);
4062 break;
4063 }
4064 case LTTNG_DATA_PENDING:
4065 {
4066 int pending_ret;
4067 uint8_t pending_ret_byte;
4068
4069 pending_ret = cmd_data_pending(cmd_ctx->session);
4070
4071 /*
4072 * FIXME
4073 *
4074 * This function may returns 0 or 1 to indicate whether or not
4075 * there is data pending. In case of error, it should return an
4076 * LTTNG_ERR code. However, some code paths may still return
4077 * a nondescript error code, which we handle by returning an
4078 * "unknown" error.
4079 */
4080 if (pending_ret == 0 || pending_ret == 1) {
4081 /*
4082 * ret will be set to LTTNG_OK at the end of
4083 * this function.
4084 */
4085 } else if (pending_ret < 0) {
4086 ret = LTTNG_ERR_UNK;
4087 goto setup_error;
4088 } else {
4089 ret = pending_ret;
4090 goto setup_error;
4091 }
4092
4093 pending_ret_byte = (uint8_t) pending_ret;
4094
4095 /* 1 byte to return whether or not data is pending */
4096 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
4097 &pending_ret_byte, 1);
4098
4099 if (ret < 0) {
4100 goto setup_error;
4101 }
4102
4103 ret = LTTNG_OK;
4104 break;
4105 }
4106 case LTTNG_SNAPSHOT_ADD_OUTPUT:
4107 {
4108 struct lttcomm_lttng_output_id reply;
4109
4110 ret = cmd_snapshot_add_output(cmd_ctx->session,
4111 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
4112 if (ret != LTTNG_OK) {
4113 goto error;
4114 }
4115
4116 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
4117 sizeof(reply));
4118 if (ret < 0) {
4119 goto setup_error;
4120 }
4121
4122 /* Copy output list into message payload */
4123 ret = LTTNG_OK;
4124 break;
4125 }
4126 case LTTNG_SNAPSHOT_DEL_OUTPUT:
4127 {
4128 ret = cmd_snapshot_del_output(cmd_ctx->session,
4129 &cmd_ctx->lsm->u.snapshot_output.output);
4130 break;
4131 }
4132 case LTTNG_SNAPSHOT_LIST_OUTPUT:
4133 {
4134 ssize_t nb_output;
4135 struct lttng_snapshot_output *outputs = NULL;
4136
4137 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
4138 if (nb_output < 0) {
4139 ret = -nb_output;
4140 goto error;
4141 }
4142
4143 assert((nb_output > 0 && outputs) || nb_output == 0);
4144 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
4145 nb_output * sizeof(struct lttng_snapshot_output));
4146 free(outputs);
4147
4148 if (ret < 0) {
4149 goto setup_error;
4150 }
4151
4152 ret = LTTNG_OK;
4153 break;
4154 }
4155 case LTTNG_SNAPSHOT_RECORD:
4156 {
4157 ret = cmd_snapshot_record(cmd_ctx->session,
4158 &cmd_ctx->lsm->u.snapshot_record.output,
4159 cmd_ctx->lsm->u.snapshot_record.wait);
4160 break;
4161 }
4162 case LTTNG_CREATE_SESSION_SNAPSHOT:
4163 {
4164 size_t nb_uri, len;
4165 struct lttng_uri *uris = NULL;
4166
4167 nb_uri = cmd_ctx->lsm->u.uri.size;
4168 len = nb_uri * sizeof(struct lttng_uri);
4169
4170 if (nb_uri > 0) {
4171 uris = zmalloc(len);
4172 if (uris == NULL) {
4173 ret = LTTNG_ERR_FATAL;
4174 goto error;
4175 }
4176
4177 /* Receive variable len data */
4178 DBG("Waiting for %zu URIs from client ...", nb_uri);
4179 ret = lttcomm_recv_unix_sock(sock, uris, len);
4180 if (ret <= 0) {
4181 DBG("No URIs received from client... continuing");
4182 *sock_error = 1;
4183 ret = LTTNG_ERR_SESSION_FAIL;
4184 free(uris);
4185 goto error;
4186 }
4187
4188 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4189 DBG("Creating session with ONE network URI is a bad call");
4190 ret = LTTNG_ERR_SESSION_FAIL;
4191 free(uris);
4192 goto error;
4193 }
4194 }
4195
4196 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
4197 nb_uri, &cmd_ctx->creds);
4198 free(uris);
4199 break;
4200 }
4201 case LTTNG_CREATE_SESSION_LIVE:
4202 {
4203 size_t nb_uri, len;
4204 struct lttng_uri *uris = NULL;
4205
4206 nb_uri = cmd_ctx->lsm->u.uri.size;
4207 len = nb_uri * sizeof(struct lttng_uri);
4208
4209 if (nb_uri > 0) {
4210 uris = zmalloc(len);
4211 if (uris == NULL) {
4212 ret = LTTNG_ERR_FATAL;
4213 goto error;
4214 }
4215
4216 /* Receive variable len data */
4217 DBG("Waiting for %zu URIs from client ...", nb_uri);
4218 ret = lttcomm_recv_unix_sock(sock, uris, len);
4219 if (ret <= 0) {
4220 DBG("No URIs received from client... continuing");
4221 *sock_error = 1;
4222 ret = LTTNG_ERR_SESSION_FAIL;
4223 free(uris);
4224 goto error;
4225 }
4226
4227 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4228 DBG("Creating session with ONE network URI is a bad call");
4229 ret = LTTNG_ERR_SESSION_FAIL;
4230 free(uris);
4231 goto error;
4232 }
4233 }
4234
4235 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4236 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4237 free(uris);
4238 break;
4239 }
4240 case LTTNG_SAVE_SESSION:
4241 {
4242 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4243 &cmd_ctx->creds);
4244 break;
4245 }
4246 case LTTNG_SET_SESSION_SHM_PATH:
4247 {
4248 ret = cmd_set_session_shm_path(cmd_ctx->session,
4249 cmd_ctx->lsm->u.set_shm_path.shm_path);
4250 break;
4251 }
4252 case LTTNG_REGENERATE_METADATA:
4253 {
4254 ret = cmd_regenerate_metadata(cmd_ctx->session);
4255 break;
4256 }
4257 case LTTNG_REGENERATE_STATEDUMP:
4258 {
4259 ret = cmd_regenerate_statedump(cmd_ctx->session);
4260 break;
4261 }
4262 case LTTNG_REGISTER_TRIGGER:
4263 {
4264 ret = cmd_register_trigger(cmd_ctx, sock,
4265 notification_thread_handle);
4266 break;
4267 }
4268 case LTTNG_UNREGISTER_TRIGGER:
4269 {
4270 ret = cmd_unregister_trigger(cmd_ctx, sock,
4271 notification_thread_handle);
4272 break;
4273 }
4274 case LTTNG_ROTATE_SESSION:
4275 {
4276 struct lttng_rotate_session_return rotate_return;
4277
4278 DBG("Client rotate session \"%s\"", cmd_ctx->session->name);
4279
4280 memset(&rotate_return, 0, sizeof(rotate_return));
4281 if (cmd_ctx->session->kernel_session && !check_rotate_compatible()) {
4282 DBG("Kernel tracer version is not compatible with the rotation feature");
4283 ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
4284 goto error;
4285 }
4286
4287 ret = cmd_rotate_session(cmd_ctx->session, &rotate_return);
4288 if (ret < 0) {
4289 ret = -ret;
4290 goto error;
4291 }
4292
4293 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &rotate_return,
4294 sizeof(rotate_return));
4295 if (ret < 0) {
4296 ret = -ret;
4297 goto error;
4298 }
4299
4300 ret = LTTNG_OK;
4301 break;
4302 }
4303 case LTTNG_ROTATION_GET_INFO:
4304 {
4305 struct lttng_rotation_get_info_return get_info_return;
4306
4307 memset(&get_info_return, 0, sizeof(get_info_return));
4308 ret = cmd_rotate_get_info(cmd_ctx->session, &get_info_return,
4309 cmd_ctx->lsm->u.get_rotation_info.rotation_id);
4310 if (ret < 0) {
4311 ret = -ret;
4312 goto error;
4313 }
4314
4315 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &get_info_return,
4316 sizeof(get_info_return));
4317 if (ret < 0) {
4318 ret = -ret;
4319 goto error;
4320 }
4321
4322 ret = LTTNG_OK;
4323 break;
4324 }
4325 case LTTNG_SESSION_GET_CURRENT_OUTPUT:
4326 {
4327 struct lttng_session_get_current_output_return output_return;
4328
4329 memset(&output_return, 0, sizeof(output_return));
4330 ret = cmd_session_get_current_output(cmd_ctx->session,
4331 &output_return);
4332 if (ret < 0) {
4333 ret = -ret;
4334 goto error;
4335 }
4336
4337 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &output_return,
4338 sizeof(output_return));
4339 if (ret < 0) {
4340 ret = -ret;
4341 goto error;
4342 }
4343
4344 ret = LTTNG_OK;
4345 break;
4346 }
4347 case LTTNG_ROTATION_SET_SCHEDULE:
4348 {
4349 bool set_schedule;
4350 enum lttng_rotation_schedule_type schedule_type;
4351 uint64_t value;
4352
4353 if (cmd_ctx->session->kernel_session && !check_rotate_compatible()) {
4354 DBG("Kernel tracer version does not support session rotations");
4355 ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
4356 goto error;
4357 }
4358
4359 set_schedule = cmd_ctx->lsm->u.rotation_set_schedule.set == 1;
4360 schedule_type = (enum lttng_rotation_schedule_type) cmd_ctx->lsm->u.rotation_set_schedule.type;
4361 value = cmd_ctx->lsm->u.rotation_set_schedule.value;
4362
4363 ret = cmd_rotation_set_schedule(cmd_ctx->session,
4364 set_schedule,
4365 schedule_type,
4366 value,
4367 notification_thread_handle);
4368 if (ret != LTTNG_OK) {
4369 goto error;
4370 }
4371
4372 break;
4373 }
4374 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES:
4375 {
4376 struct lttng_session_list_schedules_return schedules = {
4377 .periodic.set = !!cmd_ctx->session->rotate_timer_period,
4378 .periodic.value = cmd_ctx->session->rotate_timer_period,
4379 .size.set = !!cmd_ctx->session->rotate_size,
4380 .size.value = cmd_ctx->session->rotate_size,
4381 };
4382
4383 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &schedules,
4384 sizeof(schedules));
4385 if (ret < 0) {
4386 ret = -ret;
4387 goto error;
4388 }
4389
4390 ret = LTTNG_OK;
4391 break;
4392 }
4393 default:
4394 ret = LTTNG_ERR_UND;
4395 break;
4396 }
4397
4398 error:
4399 if (cmd_ctx->llm == NULL) {
4400 DBG("Missing llm structure. Allocating one.");
4401 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4402 goto setup_error;
4403 }
4404 }
4405 /* Set return code */
4406 cmd_ctx->llm->ret_code = ret;
4407 setup_error:
4408 if (cmd_ctx->session) {
4409 session_unlock(cmd_ctx->session);
4410 }
4411 if (need_tracing_session) {
4412 session_unlock_list();
4413 }
4414 init_setup_error:
4415 assert(!rcu_read_ongoing());
4416 return ret;
4417 }
4418
4419 /*
4420 * Thread managing health check socket.
4421 */
4422 static void *thread_manage_health(void *data)
4423 {
4424 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4425 uint32_t revents, nb_fd;
4426 struct lttng_poll_event events;
4427 struct health_comm_msg msg;
4428 struct health_comm_reply reply;
4429
4430 DBG("[thread] Manage health check started");
4431
4432 rcu_register_thread();
4433
4434 /* We might hit an error path before this is created. */
4435 lttng_poll_init(&events);
4436
4437 /* Create unix socket */
4438 sock = lttcomm_create_unix_sock(config.health_unix_sock_path.value);
4439 if (sock < 0) {
4440 ERR("Unable to create health check Unix socket");
4441 goto error;
4442 }
4443
4444 if (is_root) {
4445 /* lttng health client socket path permissions */
4446 ret = chown(config.health_unix_sock_path.value, 0,
4447 utils_get_group_id(config.tracing_group_name.value));
4448 if (ret < 0) {
4449 ERR("Unable to set group on %s", config.health_unix_sock_path.value);
4450 PERROR("chown");
4451 goto error;
4452 }
4453
4454 ret = chmod(config.health_unix_sock_path.value,
4455 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4456 if (ret < 0) {
4457 ERR("Unable to set permissions on %s", config.health_unix_sock_path.value);
4458 PERROR("chmod");
4459 goto error;
4460 }
4461 }
4462
4463 /*
4464 * Set the CLOEXEC flag. Return code is useless because either way, the
4465 * show must go on.
4466 */
4467 (void) utils_set_fd_cloexec(sock);
4468
4469 ret = lttcomm_listen_unix_sock(sock);
4470 if (ret < 0) {
4471 goto error;
4472 }
4473
4474 /*
4475 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4476 * more will be added to this poll set.
4477 */
4478 ret = sessiond_set_thread_pollset(&events, 2);
4479 if (ret < 0) {
4480 goto error;
4481 }
4482
4483 /* Add the application registration socket */
4484 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4485 if (ret < 0) {
4486 goto error;
4487 }
4488
4489 sessiond_notify_ready();
4490
4491 while (1) {
4492 DBG("Health check ready");
4493
4494 /* Inifinite blocking call, waiting for transmission */
4495 restart:
4496 ret = lttng_poll_wait(&events, -1);
4497 if (ret < 0) {
4498 /*
4499 * Restart interrupted system call.
4500 */
4501 if (errno == EINTR) {
4502 goto restart;
4503 }
4504 goto error;
4505 }
4506
4507 nb_fd = ret;
4508
4509 for (i = 0; i < nb_fd; i++) {
4510 /* Fetch once the poll data */
4511 revents = LTTNG_POLL_GETEV(&events, i);
4512 pollfd = LTTNG_POLL_GETFD(&events, i);
4513
4514 if (!revents) {
4515 /* No activity for this FD (poll implementation). */
4516 continue;
4517 }
4518
4519 /* Thread quit pipe has been closed. Killing thread. */
4520 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4521 if (ret) {
4522 err = 0;
4523 goto exit;
4524 }
4525
4526 /* Event on the registration socket */
4527 if (pollfd == sock) {
4528 if (revents & LPOLLIN) {
4529 continue;
4530 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4531 ERR("Health socket poll error");
4532 goto error;
4533 } else {
4534 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4535 goto error;
4536 }
4537 }
4538 }
4539
4540 new_sock = lttcomm_accept_unix_sock(sock);
4541 if (new_sock < 0) {
4542 goto error;
4543 }
4544
4545 /*
4546 * Set the CLOEXEC flag. Return code is useless because either way, the
4547 * show must go on.
4548 */
4549 (void) utils_set_fd_cloexec(new_sock);
4550
4551 DBG("Receiving data from client for health...");
4552 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4553 if (ret <= 0) {
4554 DBG("Nothing recv() from client... continuing");
4555 ret = close(new_sock);
4556 if (ret) {
4557 PERROR("close");
4558 }
4559 continue;
4560 }
4561
4562 rcu_thread_online();
4563
4564 memset(&reply, 0, sizeof(reply));
4565 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4566 /*
4567 * health_check_state returns 0 if health is
4568 * bad.
4569 */
4570 if (!health_check_state(health_sessiond, i)) {
4571 reply.ret_code |= 1ULL << i;
4572 }
4573 }
4574
4575 DBG2("Health check return value %" PRIx64, reply.ret_code);
4576
4577 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4578 if (ret < 0) {
4579 ERR("Failed to send health data back to client");
4580 }
4581
4582 /* End of transmission */
4583 ret = close(new_sock);
4584 if (ret) {
4585 PERROR("close");
4586 }
4587 }
4588
4589 exit:
4590 error:
4591 if (err) {
4592 ERR("Health error occurred in %s", __func__);
4593 }
4594 DBG("Health check thread dying");
4595 unlink(config.health_unix_sock_path.value);
4596 if (sock >= 0) {
4597 ret = close(sock);
4598 if (ret) {
4599 PERROR("close");
4600 }
4601 }
4602
4603 lttng_poll_clean(&events);
4604 stop_threads();
4605 rcu_unregister_thread();
4606 return NULL;
4607 }
4608
4609 /*
4610 * This thread manage all clients request using the unix client socket for
4611 * communication.
4612 */
4613 static void *thread_manage_clients(void *data)
4614 {
4615 int sock = -1, ret, i, pollfd, err = -1;
4616 int sock_error;
4617 uint32_t revents, nb_fd;
4618 struct command_ctx *cmd_ctx = NULL;
4619 struct lttng_poll_event events;
4620
4621 DBG("[thread] Manage client started");
4622
4623 rcu_register_thread();
4624
4625 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4626
4627 health_code_update();
4628
4629 ret = lttcomm_listen_unix_sock(client_sock);
4630 if (ret < 0) {
4631 goto error_listen;
4632 }
4633
4634 /*
4635 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4636 * more will be added to this poll set.
4637 */
4638 ret = sessiond_set_thread_pollset(&events, 2);
4639 if (ret < 0) {
4640 goto error_create_poll;
4641 }
4642
4643 /* Add the application registration socket */
4644 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4645 if (ret < 0) {
4646 goto error;
4647 }
4648
4649 ret = sem_post(&load_info->message_thread_ready);
4650 if (ret) {
4651 PERROR("sem_post message_thread_ready");
4652 goto error;
4653 }
4654
4655 /*
4656 * Wait until all support threads are initialized before accepting
4657 * commands.
4658 */
4659 while (uatomic_read(&lttng_sessiond_ready) != 0) {
4660 fd_set read_fds;
4661 struct timeval timeout;
4662
4663 FD_ZERO(&read_fds);
4664 FD_SET(thread_quit_pipe[0], &read_fds);
4665 memset(&timeout, 0, sizeof(timeout));
4666 timeout.tv_usec = 1000;
4667
4668 /*
4669 * If a support thread failed to launch, it may signal that
4670 * we must exit and the sessiond would never be marked as
4671 * "ready".
4672 *
4673 * The timeout is set to 1ms, which serves as a way to
4674 * pace down this check.
4675 */
4676 ret = select(thread_quit_pipe[0] + 1, &read_fds, NULL, NULL,
4677 &timeout);
4678 if (ret > 0 || (ret < 0 && errno != EINTR)) {
4679 goto exit;
4680 }
4681 }
4682 /*
4683 * This barrier is paired with the one in sessiond_notify_ready() to
4684 * ensure that loads accessing data initialized by the other threads,
4685 * on which this thread was waiting, are not performed before this point.
4686 *
4687 * Note that this could be a 'read' memory barrier, but a full barrier
4688 * is used in case the code changes. The performance implications of
4689 * this choice are minimal since this is a slow path.
4690 */
4691 cmm_smp_mb();
4692
4693 /* This testpoint is after we signal readiness to the parent. */
4694 if (testpoint(sessiond_thread_manage_clients)) {
4695 goto error;
4696 }
4697
4698 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4699 goto error;
4700 }
4701
4702 health_code_update();
4703
4704 while (1) {
4705 const struct cmd_completion_handler *cmd_completion_handler;
4706
4707 DBG("Accepting client command ...");
4708
4709 /* Inifinite blocking call, waiting for transmission */
4710 restart:
4711 health_poll_entry();
4712 ret = lttng_poll_wait(&events, -1);
4713 health_poll_exit();
4714 if (ret < 0) {
4715 /*
4716 * Restart interrupted system call.
4717 */
4718 if (errno == EINTR) {
4719 goto restart;
4720 }
4721 goto error;
4722 }
4723
4724 nb_fd = ret;
4725
4726 for (i = 0; i < nb_fd; i++) {
4727 /* Fetch once the poll data */
4728 revents = LTTNG_POLL_GETEV(&events, i);
4729 pollfd = LTTNG_POLL_GETFD(&events, i);
4730
4731 health_code_update();
4732
4733 if (!revents) {
4734 /* No activity for this FD (poll implementation). */
4735 continue;
4736 }
4737
4738 /* Thread quit pipe has been closed. Killing thread. */
4739 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4740 if (ret) {
4741 err = 0;
4742 goto exit;
4743 }
4744
4745 /* Event on the registration socket */
4746 if (pollfd == client_sock) {
4747 if (revents & LPOLLIN) {
4748 continue;
4749 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4750 ERR("Client socket poll error");
4751 goto error;
4752 } else {
4753 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4754 goto error;
4755 }
4756 }
4757 }
4758
4759 DBG("Wait for client response");
4760
4761 health_code_update();
4762
4763 sock = lttcomm_accept_unix_sock(client_sock);
4764 if (sock < 0) {
4765 goto error;
4766 }
4767
4768 /*
4769 * Set the CLOEXEC flag. Return code is useless because either way, the
4770 * show must go on.
4771 */
4772 (void) utils_set_fd_cloexec(sock);
4773
4774 /* Set socket option for credentials retrieval */
4775 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4776 if (ret < 0) {
4777 goto error;
4778 }
4779
4780 /* Allocate context command to process the client request */
4781 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4782 if (cmd_ctx == NULL) {
4783 PERROR("zmalloc cmd_ctx");
4784 goto error;
4785 }
4786
4787 /* Allocate data buffer for reception */
4788 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4789 if (cmd_ctx->lsm == NULL) {
4790 PERROR("zmalloc cmd_ctx->lsm");
4791 goto error;
4792 }
4793
4794 cmd_ctx->llm = NULL;
4795 cmd_ctx->session = NULL;
4796
4797 health_code_update();
4798
4799 /*
4800 * Data is received from the lttng client. The struct
4801 * lttcomm_session_msg (lsm) contains the command and data request of
4802 * the client.
4803 */
4804 DBG("Receiving data from client ...");
4805 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4806 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4807 if (ret <= 0) {
4808 DBG("Nothing recv() from client... continuing");
4809 ret = close(sock);
4810 if (ret) {
4811 PERROR("close");
4812 }
4813 sock = -1;
4814 clean_command_ctx(&cmd_ctx);
4815 continue;
4816 }
4817
4818 health_code_update();
4819
4820 // TODO: Validate cmd_ctx including sanity check for
4821 // security purpose.
4822
4823 rcu_thread_online();
4824 /*
4825 * This function dispatch the work to the kernel or userspace tracer
4826 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4827 * informations for the client. The command context struct contains
4828 * everything this function may needs.
4829 */
4830 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4831 rcu_thread_offline();
4832 if (ret < 0) {
4833 ret = close(sock);
4834 if (ret) {
4835 PERROR("close");
4836 }
4837 sock = -1;
4838 /*
4839 * TODO: Inform client somehow of the fatal error. At
4840 * this point, ret < 0 means that a zmalloc failed
4841 * (ENOMEM). Error detected but still accept
4842 * command, unless a socket error has been
4843 * detected.
4844 */
4845 clean_command_ctx(&cmd_ctx);
4846 continue;
4847 }
4848
4849 cmd_completion_handler = cmd_pop_completion_handler();
4850 if (cmd_completion_handler) {
4851 enum lttng_error_code completion_code;
4852
4853 completion_code = cmd_completion_handler->run(
4854 cmd_completion_handler->data);
4855 if (completion_code != LTTNG_OK) {
4856 clean_command_ctx(&cmd_ctx);
4857 continue;
4858 }
4859 }
4860
4861 health_code_update();
4862
4863 DBG("Sending response (size: %d, retcode: %s (%d))",
4864 cmd_ctx->lttng_msg_size,
4865 lttng_strerror(-cmd_ctx->llm->ret_code),
4866 cmd_ctx->llm->ret_code);
4867 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4868 if (ret < 0) {
4869 ERR("Failed to send data back to client");
4870 }
4871
4872 /* End of transmission */
4873 ret = close(sock);
4874 if (ret) {
4875 PERROR("close");
4876 }
4877 sock = -1;
4878
4879 clean_command_ctx(&cmd_ctx);
4880
4881 health_code_update();
4882 }
4883
4884 exit:
4885 error:
4886 if (sock >= 0) {
4887 ret = close(sock);
4888 if (ret) {
4889 PERROR("close");
4890 }
4891 }
4892
4893 lttng_poll_clean(&events);
4894 clean_command_ctx(&cmd_ctx);
4895
4896 error_listen:
4897 error_create_poll:
4898 unlink(config.client_unix_sock_path.value);
4899 if (client_sock >= 0) {
4900 ret = close(client_sock);
4901 if (ret) {
4902 PERROR("close");
4903 }
4904 }
4905
4906 if (err) {
4907 health_error();
4908 ERR("Health error occurred in %s", __func__);
4909 }
4910
4911 health_unregister(health_sessiond);
4912
4913 DBG("Client thread dying");
4914
4915 rcu_unregister_thread();
4916
4917 /*
4918 * Since we are creating the consumer threads, we own them, so we need
4919 * to join them before our thread exits.
4920 */
4921 ret = join_consumer_thread(&kconsumer_data);
4922 if (ret) {
4923 errno = ret;
4924 PERROR("join_consumer");
4925 }
4926
4927 ret = join_consumer_thread(&ustconsumer32_data);
4928 if (ret) {
4929 errno = ret;
4930 PERROR("join_consumer ust32");
4931 }
4932
4933 ret = join_consumer_thread(&ustconsumer64_data);
4934 if (ret) {
4935 errno = ret;
4936 PERROR("join_consumer ust64");
4937 }
4938 return NULL;
4939 }
4940
4941 static int string_match(const char *str1, const char *str2)
4942 {
4943 return (str1 && str2) && !strcmp(str1, str2);
4944 }
4945
4946 /*
4947 * Take an option from the getopt output and set it in the right variable to be
4948 * used later.
4949 *
4950 * Return 0 on success else a negative value.
4951 */
4952 static int set_option(int opt, const char *arg, const char *optname)
4953 {
4954 int ret = 0;
4955
4956 if (string_match(optname, "client-sock") || opt == 'c') {
4957 if (!arg || *arg == '\0') {
4958 ret = -EINVAL;
4959 goto end;
4960 }
4961 if (lttng_is_setuid_setgid()) {
4962 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4963 "-c, --client-sock");
4964 } else {
4965 config_string_set(&config.client_unix_sock_path,
4966 strdup(arg));
4967 if (!config.client_unix_sock_path.value) {
4968 ret = -ENOMEM;
4969 PERROR("strdup");
4970 }
4971 }
4972 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4973 if (!arg || *arg == '\0') {
4974 ret = -EINVAL;
4975 goto end;
4976 }
4977 if (lttng_is_setuid_setgid()) {
4978 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4979 "-a, --apps-sock");
4980 } else {
4981 config_string_set(&config.apps_unix_sock_path,
4982 strdup(arg));
4983 if (!config.apps_unix_sock_path.value) {
4984 ret = -ENOMEM;
4985 PERROR("strdup");
4986 }
4987 }
4988 } else if (string_match(optname, "daemonize") || opt == 'd') {
4989 config.daemonize = true;
4990 } else if (string_match(optname, "background") || opt == 'b') {
4991 config.background = true;
4992 } else if (string_match(optname, "group") || opt == 'g') {
4993 if (!arg || *arg == '\0') {
4994 ret = -EINVAL;
4995 goto end;
4996 }
4997 if (lttng_is_setuid_setgid()) {
4998 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4999 "-g, --group");
5000 } else {
5001 config_string_set(&config.tracing_group_name,
5002 strdup(arg));
5003 if (!config.tracing_group_name.value) {
5004 ret = -ENOMEM;
5005 PERROR("strdup");
5006 }
5007 }
5008 } else if (string_match(optname, "help") || opt == 'h') {
5009 ret = utils_show_help(8, "lttng-sessiond", help_msg);
5010 if (ret) {
5011 ERR("Cannot show --help for `lttng-sessiond`");
5012 perror("exec");
5013 }
5014 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
5015 } else if (string_match(optname, "version") || opt == 'V') {
5016 fprintf(stdout, "%s\n", VERSION);
5017 exit(EXIT_SUCCESS);
5018 } else if (string_match(optname, "sig-parent") || opt == 'S') {
5019 config.sig_parent = true;
5020 } else if (string_match(optname, "kconsumerd-err-sock")) {
5021 if (!arg || *arg == '\0') {
5022 ret = -EINVAL;
5023 goto end;
5024 }
5025 if (lttng_is_setuid_setgid()) {
5026 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5027 "--kconsumerd-err-sock");
5028 } else {
5029 config_string_set(&config.kconsumerd_err_unix_sock_path,
5030 strdup(arg));
5031 if (!config.kconsumerd_err_unix_sock_path.value) {
5032 ret = -ENOMEM;
5033 PERROR("strdup");
5034 }
5035 }
5036 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
5037 if (!arg || *arg == '\0') {
5038 ret = -EINVAL;
5039 goto end;
5040 }
5041 if (lttng_is_setuid_setgid()) {
5042 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5043 "--kconsumerd-cmd-sock");
5044 } else {
5045 config_string_set(&config.kconsumerd_cmd_unix_sock_path,
5046 strdup(arg));
5047 if (!config.kconsumerd_cmd_unix_sock_path.value) {
5048 ret = -ENOMEM;
5049 PERROR("strdup");
5050 }
5051 }
5052 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
5053 if (!arg || *arg == '\0') {
5054 ret = -EINVAL;
5055 goto end;
5056 }
5057 if (lttng_is_setuid_setgid()) {
5058 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5059 "--ustconsumerd64-err-sock");
5060 } else {
5061 config_string_set(&config.consumerd64_err_unix_sock_path,
5062 strdup(arg));
5063 if (!config.consumerd64_err_unix_sock_path.value) {
5064 ret = -ENOMEM;
5065 PERROR("strdup");
5066 }
5067 }
5068 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
5069 if (!arg || *arg == '\0') {
5070 ret = -EINVAL;
5071 goto end;
5072 }
5073 if (lttng_is_setuid_setgid()) {
5074 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5075 "--ustconsumerd64-cmd-sock");
5076 } else {
5077 config_string_set(&config.consumerd64_cmd_unix_sock_path,
5078 strdup(arg));
5079 if (!config.consumerd64_cmd_unix_sock_path.value) {
5080 ret = -ENOMEM;
5081 PERROR("strdup");
5082 }
5083 }
5084 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
5085 if (!arg || *arg == '\0') {
5086 ret = -EINVAL;
5087 goto end;
5088 }
5089 if (lttng_is_setuid_setgid()) {
5090 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5091 "--ustconsumerd32-err-sock");
5092 } else {
5093 config_string_set(&config.consumerd32_err_unix_sock_path,
5094 strdup(arg));
5095 if (!config.consumerd32_err_unix_sock_path.value) {
5096 ret = -ENOMEM;
5097 PERROR("strdup");
5098 }
5099 }
5100 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
5101 if (!arg || *arg == '\0') {
5102 ret = -EINVAL;
5103 goto end;
5104 }
5105 if (lttng_is_setuid_setgid()) {
5106 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5107 "--ustconsumerd32-cmd-sock");
5108 } else {
5109 config_string_set(&config.consumerd32_cmd_unix_sock_path,
5110 strdup(arg));
5111 if (!config.consumerd32_cmd_unix_sock_path.value) {
5112 ret = -ENOMEM;
5113 PERROR("strdup");
5114 }
5115 }
5116 } else if (string_match(optname, "no-kernel")) {
5117 config.no_kernel = true;
5118 } else if (string_match(optname, "quiet") || opt == 'q') {
5119 config.quiet = true;
5120 } else if (string_match(optname, "verbose") || opt == 'v') {
5121 /* Verbose level can increase using multiple -v */
5122 if (arg) {
5123 /* Value obtained from config file */
5124 config.verbose = config_parse_value(arg);
5125 } else {
5126 /* -v used on command line */
5127 config.verbose++;
5128 }
5129 /* Clamp value to [0, 3] */
5130 config.verbose = config.verbose < 0 ? 0 :
5131 (config.verbose <= 3 ? config.verbose : 3);
5132 } else if (string_match(optname, "verbose-consumer")) {
5133 if (arg) {
5134 config.verbose_consumer = config_parse_value(arg);
5135 } else {
5136 config.verbose_consumer++;
5137 }
5138 } else if (string_match(optname, "consumerd32-path")) {
5139 if (!arg || *arg == '\0') {
5140 ret = -EINVAL;
5141 goto end;
5142 }
5143 if (lttng_is_setuid_setgid()) {
5144 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5145 "--consumerd32-path");
5146 } else {
5147 config_string_set(&config.consumerd32_bin_path,
5148 strdup(arg));
5149 if (!config.consumerd32_bin_path.value) {
5150 PERROR("strdup");
5151 ret = -ENOMEM;
5152 }
5153 }
5154 } else if (string_match(optname, "consumerd32-libdir")) {
5155 if (!arg || *arg == '\0') {
5156 ret = -EINVAL;
5157 goto end;
5158 }
5159 if (lttng_is_setuid_setgid()) {
5160 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5161 "--consumerd32-libdir");
5162 } else {
5163 config_string_set(&config.consumerd32_lib_dir,
5164 strdup(arg));
5165 if (!config.consumerd32_lib_dir.value) {
5166 PERROR("strdup");
5167 ret = -ENOMEM;
5168 }
5169 }
5170 } else if (string_match(optname, "consumerd64-path")) {
5171 if (!arg || *arg == '\0') {
5172 ret = -EINVAL;
5173 goto end;
5174 }
5175 if (lttng_is_setuid_setgid()) {
5176 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5177 "--consumerd64-path");
5178 } else {
5179 config_string_set(&config.consumerd64_bin_path,
5180 strdup(arg));
5181 if (!config.consumerd64_bin_path.value) {
5182 PERROR("strdup");
5183 ret = -ENOMEM;
5184 }
5185 }
5186 } else if (string_match(optname, "consumerd64-libdir")) {
5187 if (!arg || *arg == '\0') {
5188 ret = -EINVAL;
5189 goto end;
5190 }
5191 if (lttng_is_setuid_setgid()) {
5192 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5193 "--consumerd64-libdir");
5194 } else {
5195 config_string_set(&config.consumerd64_lib_dir,
5196 strdup(arg));
5197 if (!config.consumerd64_lib_dir.value) {
5198 PERROR("strdup");
5199 ret = -ENOMEM;
5200 }
5201 }
5202 } else if (string_match(optname, "pidfile") || opt == 'p') {
5203 if (!arg || *arg == '\0') {
5204 ret = -EINVAL;
5205 goto end;
5206 }
5207 if (lttng_is_setuid_setgid()) {
5208 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5209 "-p, --pidfile");
5210 } else {
5211 config_string_set(&config.pid_file_path, strdup(arg));
5212 if (!config.pid_file_path.value) {
5213 PERROR("strdup");
5214 ret = -ENOMEM;
5215 }
5216 }
5217 } else if (string_match(optname, "agent-tcp-port")) {
5218 if (!arg || *arg == '\0') {
5219 ret = -EINVAL;
5220 goto end;
5221 }
5222 if (lttng_is_setuid_setgid()) {
5223 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5224 "--agent-tcp-port");
5225 } else {
5226 unsigned long v;
5227
5228 errno = 0;
5229 v = strtoul(arg, NULL, 0);
5230 if (errno != 0 || !isdigit(arg[0])) {
5231 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
5232 return -1;
5233 }
5234 if (v == 0 || v >= 65535) {
5235 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
5236 return -1;
5237 }
5238 config.agent_tcp_port.begin = config.agent_tcp_port.end = (int) v;
5239 DBG3("Agent TCP port set to non default: %i", (int) v);
5240 }
5241 } else if (string_match(optname, "load") || opt == 'l') {
5242 if (!arg || *arg == '\0') {
5243 ret = -EINVAL;
5244 goto end;
5245 }
5246 if (lttng_is_setuid_setgid()) {
5247 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5248 "-l, --load");
5249 } else {
5250 config_string_set(&config.load_session_path, strdup(arg));
5251 if (!config.load_session_path.value) {
5252 PERROR("strdup");
5253 ret = -ENOMEM;
5254 }
5255 }
5256 } else if (string_match(optname, "kmod-probes")) {
5257 if (!arg || *arg == '\0') {
5258 ret = -EINVAL;
5259 goto end;
5260 }
5261 if (lttng_is_setuid_setgid()) {
5262 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5263 "--kmod-probes");
5264 } else {
5265 config_string_set(&config.kmod_probes_list, strdup(arg));
5266 if (!config.kmod_probes_list.value) {
5267 PERROR("strdup");
5268 ret = -ENOMEM;
5269 }
5270 }
5271 } else if (string_match(optname, "extra-kmod-probes")) {
5272 if (!arg || *arg == '\0') {
5273 ret = -EINVAL;
5274 goto end;
5275 }
5276 if (lttng_is_setuid_setgid()) {
5277 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5278 "--extra-kmod-probes");
5279 } else {
5280 config_string_set(&config.kmod_extra_probes_list,
5281 strdup(arg));
5282 if (!config.kmod_extra_probes_list.value) {
5283 PERROR("strdup");
5284 ret = -ENOMEM;
5285 }
5286 }
5287 } else if (string_match(optname, "config") || opt == 'f') {
5288 /* This is handled in set_options() thus silent skip. */
5289 goto end;
5290 } else {
5291 /* Unknown option or other error.
5292 * Error is printed by getopt, just return */
5293 ret = -1;
5294 }
5295
5296 end:
5297 if (ret == -EINVAL) {
5298 const char *opt_name = "unknown";
5299 int i;
5300
5301 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
5302 i++) {
5303 if (opt == long_options[i].val) {
5304 opt_name = long_options[i].name;
5305 break;
5306 }
5307 }
5308
5309 WARN("Invalid argument provided for option \"%s\", using default value.",
5310 opt_name);
5311 }
5312
5313 return ret;
5314 }
5315
5316 /*
5317 * config_entry_handler_cb used to handle options read from a config file.
5318 * See config_entry_handler_cb comment in common/config/session-config.h for the
5319 * return value conventions.
5320 */
5321 static int config_entry_handler(const struct config_entry *entry, void *unused)
5322 {
5323 int ret = 0, i;
5324
5325 if (!entry || !entry->name || !entry->value) {
5326 ret = -EINVAL;
5327 goto end;
5328 }
5329
5330 /* Check if the option is to be ignored */
5331 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
5332 if (!strcmp(entry->name, config_ignore_options[i])) {
5333 goto end;
5334 }
5335 }
5336
5337 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
5338 i++) {
5339
5340 /* Ignore if not fully matched. */
5341 if (strcmp(entry->name, long_options[i].name)) {
5342 continue;
5343 }
5344
5345 /*
5346 * If the option takes no argument on the command line, we have to
5347 * check if the value is "true". We support non-zero numeric values,
5348 * true, on and yes.
5349 */
5350 if (!long_options[i].has_arg) {
5351 ret = config_parse_value(entry->value);
5352 if (ret <= 0) {
5353 if (ret) {
5354 WARN("Invalid configuration value \"%s\" for option %s",
5355 entry->value, entry->name);
5356 }
5357 /* False, skip boolean config option. */
5358 goto end;
5359 }
5360 }
5361
5362 ret = set_option(long_options[i].val, entry->value, entry->name);
5363 goto end;
5364 }
5365
5366 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5367
5368 end:
5369 return ret;
5370 }
5371
5372 /*
5373 * daemon configuration loading and argument parsing
5374 */
5375 static int set_options(int argc, char **argv)
5376 {
5377 int ret = 0, c = 0, option_index = 0;
5378 int orig_optopt = optopt, orig_optind = optind;
5379 char *optstring;
5380 const char *config_path = NULL;
5381
5382 optstring = utils_generate_optstring(long_options,
5383 sizeof(long_options) / sizeof(struct option));
5384 if (!optstring) {
5385 ret = -ENOMEM;
5386 goto end;
5387 }
5388
5389 /* Check for the --config option */
5390 while ((c = getopt_long(argc, argv, optstring, long_options,
5391 &option_index)) != -1) {
5392 if (c == '?') {
5393 ret = -EINVAL;
5394 goto end;
5395 } else if (c != 'f') {
5396 /* if not equal to --config option. */
5397 continue;
5398 }
5399
5400 if (lttng_is_setuid_setgid()) {
5401 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5402 "-f, --config");
5403 } else {
5404 config_path = utils_expand_path(optarg);
5405 if (!config_path) {
5406 ERR("Failed to resolve path: %s", optarg);
5407 }
5408 }
5409 }
5410
5411 ret = config_get_section_entries(config_path, config_section_name,
5412 config_entry_handler, NULL);
5413 if (ret) {
5414 if (ret > 0) {
5415 ERR("Invalid configuration option at line %i", ret);
5416 ret = -1;
5417 }
5418 goto end;
5419 }
5420
5421 /* Reset getopt's global state */
5422 optopt = orig_optopt;
5423 optind = orig_optind;
5424 while (1) {
5425 option_index = -1;
5426 /*
5427 * getopt_long() will not set option_index if it encounters a
5428 * short option.
5429 */
5430 c = getopt_long(argc, argv, optstring, long_options,
5431 &option_index);
5432 if (c == -1) {
5433 break;
5434 }
5435
5436 /*
5437 * Pass NULL as the long option name if popt left the index
5438 * unset.
5439 */
5440 ret = set_option(c, optarg,
5441 option_index < 0 ? NULL :
5442 long_options[option_index].name);
5443 if (ret < 0) {
5444 break;
5445 }
5446 }
5447
5448 end:
5449 free(optstring);
5450 return ret;
5451 }
5452
5453 /*
5454 * Creates the two needed socket by the daemon.
5455 * apps_sock - The communication socket for all UST apps.
5456 * client_sock - The communication of the cli tool (lttng).
5457 */
5458 static int init_daemon_socket(void)
5459 {
5460 int ret = 0;
5461 mode_t old_umask;
5462
5463 old_umask = umask(0);
5464
5465 /* Create client tool unix socket */
5466 client_sock = lttcomm_create_unix_sock(config.client_unix_sock_path.value);
5467 if (client_sock < 0) {
5468 ERR("Create unix sock failed: %s", config.client_unix_sock_path.value);
5469 ret = -1;
5470 goto end;
5471 }
5472
5473 /* Set the cloexec flag */
5474 ret = utils_set_fd_cloexec(client_sock);
5475 if (ret < 0) {
5476 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5477 "Continuing but note that the consumer daemon will have a "
5478 "reference to this socket on exec()", client_sock);
5479 }
5480
5481 /* File permission MUST be 660 */
5482 ret = chmod(config.client_unix_sock_path.value, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5483 if (ret < 0) {
5484 ERR("Set file permissions failed: %s", config.client_unix_sock_path.value);
5485 PERROR("chmod");
5486 goto end;
5487 }
5488
5489 /* Create the application unix socket */
5490 apps_sock = lttcomm_create_unix_sock(config.apps_unix_sock_path.value);
5491 if (apps_sock < 0) {
5492 ERR("Create unix sock failed: %s", config.apps_unix_sock_path.value);
5493 ret = -1;
5494 goto end;
5495 }
5496
5497 /* Set the cloexec flag */
5498 ret = utils_set_fd_cloexec(apps_sock);
5499 if (ret < 0) {
5500 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5501 "Continuing but note that the consumer daemon will have a "
5502 "reference to this socket on exec()", apps_sock);
5503 }
5504
5505 /* File permission MUST be 666 */
5506 ret = chmod(config.apps_unix_sock_path.value,
5507 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5508 if (ret < 0) {
5509 ERR("Set file permissions failed: %s", config.apps_unix_sock_path.value);
5510 PERROR("chmod");
5511 goto end;
5512 }
5513
5514 DBG3("Session daemon client socket %d and application socket %d created",
5515 client_sock, apps_sock);
5516
5517 end:
5518 umask(old_umask);
5519 return ret;
5520 }
5521
5522 /*
5523 * Create lockfile using the rundir and return its fd.
5524 */
5525 static int create_lockfile(void)
5526 {
5527 return utils_create_lock_file(config.lock_file_path.value);
5528 }
5529
5530 /*
5531 * Check if the global socket is available, and if a daemon is answering at the
5532 * other side. If yes, error is returned.
5533 *
5534 * Also attempts to create and hold the lock file.
5535 */
5536 static int check_existing_daemon(void)
5537 {
5538 int ret = 0;
5539
5540 /* Is there anybody out there ? */
5541 if (lttng_session_daemon_alive()) {
5542 ret = -EEXIST;
5543 goto end;
5544 }
5545
5546 lockfile_fd = create_lockfile();
5547 if (lockfile_fd < 0) {
5548 ret = -EEXIST;
5549 goto end;
5550 }
5551 end:
5552 return ret;
5553 }
5554
5555 static void sessiond_cleanup_lock_file(void)
5556 {
5557 int ret;
5558
5559 /*
5560 * Cleanup lock file by deleting it and finaly closing it which will
5561 * release the file system lock.
5562 */
5563 if (lockfile_fd >= 0) {
5564 ret = remove(config.lock_file_path.value);
5565 if (ret < 0) {
5566 PERROR("remove lock file");
5567 }
5568 ret = close(lockfile_fd);
5569 if (ret < 0) {
5570 PERROR("close lock file");
5571 }
5572 }
5573 }
5574
5575 /*
5576 * Set the tracing group gid onto the client socket.
5577 *
5578 * Race window between mkdir and chown is OK because we are going from more
5579 * permissive (root.root) to less permissive (root.tracing).
5580 */
5581 static int set_permissions(char *rundir)
5582 {
5583 int ret;
5584 gid_t gid;
5585
5586 gid = utils_get_group_id(config.tracing_group_name.value);
5587
5588 /* Set lttng run dir */
5589 ret = chown(rundir, 0, gid);
5590 if (ret < 0) {
5591 ERR("Unable to set group on %s", rundir);
5592 PERROR("chown");
5593 }
5594
5595 /*
5596 * Ensure all applications and tracing group can search the run
5597 * dir. Allow everyone to read the directory, since it does not
5598 * buy us anything to hide its content.
5599 */
5600 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5601 if (ret < 0) {
5602 ERR("Unable to set permissions on %s", rundir);
5603 PERROR("chmod");
5604 }
5605
5606 /* lttng client socket path */
5607 ret = chown(config.client_unix_sock_path.value, 0, gid);
5608 if (ret < 0) {
5609 ERR("Unable to set group on %s", config.client_unix_sock_path.value);
5610 PERROR("chown");
5611 }
5612
5613 /* kconsumer error socket path */
5614 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5615 if (ret < 0) {
5616 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5617 PERROR("chown");
5618 }
5619
5620 /* 64-bit ustconsumer error socket path */
5621 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5622 if (ret < 0) {
5623 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5624 PERROR("chown");
5625 }
5626
5627 /* 32-bit ustconsumer compat32 error socket path */
5628 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5629 if (ret < 0) {
5630 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5631 PERROR("chown");
5632 }
5633
5634 DBG("All permissions are set");
5635
5636 return ret;
5637 }
5638
5639 /*
5640 * Create the lttng run directory needed for all global sockets and pipe.
5641 */
5642 static int create_lttng_rundir(void)
5643 {
5644 int ret;
5645
5646 DBG3("Creating LTTng run directory: %s", config.rundir.value);
5647
5648 ret = mkdir(config.rundir.value, S_IRWXU);
5649 if (ret < 0) {
5650 if (errno != EEXIST) {
5651 ERR("Unable to create %s", config.rundir.value);
5652 goto error;
5653 } else {
5654 ret = 0;
5655 }
5656 }
5657
5658 error:
5659 return ret;
5660 }
5661
5662 /*
5663 * Setup sockets and directory needed by the consumerds' communication with the
5664 * session daemon.
5665 */
5666 static int set_consumer_sockets(struct consumer_data *consumer_data)
5667 {
5668 int ret;
5669 char *path = NULL;
5670
5671 switch (consumer_data->type) {
5672 case LTTNG_CONSUMER_KERNEL:
5673 path = config.kconsumerd_path.value;
5674 break;
5675 case LTTNG_CONSUMER64_UST:
5676 path = config.consumerd64_path.value;
5677 break;
5678 case LTTNG_CONSUMER32_UST:
5679 path = config.consumerd32_path.value;
5680 break;
5681 default:
5682 ERR("Consumer type unknown");
5683 ret = -EINVAL;
5684 goto error;
5685 }
5686 assert(path);
5687
5688 DBG2("Creating consumer directory: %s", path);
5689
5690 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5691 if (ret < 0 && errno != EEXIST) {
5692 PERROR("mkdir");
5693 ERR("Failed to create %s", path);
5694 goto error;
5695 }
5696 if (is_root) {
5697 ret = chown(path, 0, utils_get_group_id(config.tracing_group_name.value));
5698 if (ret < 0) {
5699 ERR("Unable to set group on %s", path);
5700 PERROR("chown");
5701 goto error;
5702 }
5703 }
5704
5705 /* Create the consumerd error unix socket */
5706 consumer_data->err_sock =
5707 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5708 if (consumer_data->err_sock < 0) {
5709 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5710 ret = -1;
5711 goto error;
5712 }
5713
5714 /*
5715 * Set the CLOEXEC flag. Return code is useless because either way, the
5716 * show must go on.
5717 */
5718 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5719 if (ret < 0) {
5720 PERROR("utils_set_fd_cloexec");
5721 /* continue anyway */
5722 }
5723
5724 /* File permission MUST be 660 */
5725 ret = chmod(consumer_data->err_unix_sock_path,
5726 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5727 if (ret < 0) {
5728 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5729 PERROR("chmod");
5730 goto error;
5731 }
5732
5733 error:
5734 return ret;
5735 }
5736
5737 /*
5738 * Signal handler for the daemon
5739 *
5740 * Simply stop all worker threads, leaving main() return gracefully after
5741 * joining all threads and calling cleanup().
5742 */
5743 static void sighandler(int sig)
5744 {
5745 switch (sig) {
5746 case SIGINT:
5747 DBG("SIGINT caught");
5748 stop_threads();
5749 break;
5750 case SIGTERM:
5751 DBG("SIGTERM caught");
5752 stop_threads();
5753 break;
5754 case SIGUSR1:
5755 CMM_STORE_SHARED(recv_child_signal, 1);
5756 break;
5757 default:
5758 break;
5759 }
5760 }
5761
5762 /*
5763 * Setup signal handler for :
5764 * SIGINT, SIGTERM, SIGPIPE
5765 */
5766 static int set_signal_handler(void)
5767 {
5768 int ret = 0;
5769 struct sigaction sa;
5770 sigset_t sigset;
5771
5772 if ((ret = sigemptyset(&sigset)) < 0) {
5773 PERROR("sigemptyset");
5774 return ret;
5775 }
5776
5777 sa.sa_mask = sigset;
5778 sa.sa_flags = 0;
5779
5780 sa.sa_handler = sighandler;
5781 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5782 PERROR("sigaction");
5783 return ret;
5784 }
5785
5786 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5787 PERROR("sigaction");
5788 return ret;
5789 }
5790
5791 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5792 PERROR("sigaction");
5793 return ret;
5794 }
5795
5796 sa.sa_handler = SIG_IGN;
5797 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5798 PERROR("sigaction");
5799 return ret;
5800 }
5801
5802 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5803
5804 return ret;
5805 }
5806
5807 /*
5808 * Set open files limit to unlimited. This daemon can open a large number of
5809 * file descriptors in order to consume multiple kernel traces.
5810 */
5811 static void set_ulimit(void)
5812 {
5813 int ret;
5814 struct rlimit lim;
5815
5816 /* The kernel does not allow an infinite limit for open files */
5817 lim.rlim_cur = 65535;
5818 lim.rlim_max = 65535;
5819
5820 ret = setrlimit(RLIMIT_NOFILE, &lim);
5821 if (ret < 0) {
5822 PERROR("failed to set open files limit");
5823 }
5824 }
5825
5826 static int write_pidfile(void)
5827 {
5828 return utils_create_pid_file(getpid(), config.pid_file_path.value);
5829 }
5830
5831 static int set_clock_plugin_env(void)
5832 {
5833 int ret = 0;
5834 char *env_value = NULL;
5835
5836 if (!config.lttng_ust_clock_plugin.value) {
5837 goto end;
5838 }
5839
5840 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
5841 config.lttng_ust_clock_plugin.value);
5842 if (ret < 0) {
5843 PERROR("asprintf");
5844 goto end;
5845 }
5846
5847 ret = putenv(env_value);
5848 if (ret) {
5849 free(env_value);
5850 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5851 goto end;
5852 }
5853
5854 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5855 config.lttng_ust_clock_plugin.value);
5856 end:
5857 return ret;
5858 }
5859
5860 static
5861 struct rotation_thread_timer_queue *create_rotate_timer_queue(void)
5862 {
5863 struct rotation_thread_timer_queue *queue = NULL;
5864
5865 queue = zmalloc(sizeof(struct rotation_thread_timer_queue));
5866 if (!queue) {
5867 PERROR("Failed to allocate timer rotate queue");
5868 goto end;
5869 }
5870
5871 queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK);
5872 CDS_INIT_LIST_HEAD(&queue->list);
5873 pthread_mutex_init(&queue->lock, NULL);
5874
5875 end:
5876 return queue;
5877 }
5878
5879 static
5880 void destroy_rotate_timer_queue(struct rotation_thread_timer_queue *queue)
5881 {
5882 struct sessiond_rotation_timer *node, *tmp_node;
5883
5884 if (!queue) {
5885 return;
5886 }
5887
5888 lttng_pipe_destroy(queue->event_pipe);
5889
5890 pthread_mutex_lock(&queue->lock);
5891 /* Empty wait queue. */
5892 cds_list_for_each_entry_safe(node, tmp_node, &queue->list, head) {
5893 cds_list_del(&node->head);
5894 free(node);
5895 }
5896 pthread_mutex_unlock(&queue->lock);
5897
5898 pthread_mutex_destroy(&queue->lock);
5899 free(queue);
5900 }
5901
5902 /*
5903 * main
5904 */
5905 int main(int argc, char **argv)
5906 {
5907 int ret = 0, retval = 0;
5908 void *status;
5909 const char *env_app_timeout;
5910 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5911 *ust64_channel_monitor_pipe = NULL,
5912 *kernel_channel_monitor_pipe = NULL;
5913 bool notification_thread_launched = false;
5914 bool rotation_thread_launched = false;
5915 bool timer_thread_launched = false;
5916 struct lttng_pipe *ust32_channel_rotate_pipe = NULL,
5917 *ust64_channel_rotate_pipe = NULL,
5918 *kernel_channel_rotate_pipe = NULL;
5919 struct timer_thread_parameters timer_thread_ctx;
5920 /* Queue of rotation jobs populated by the sessiond-timer. */
5921 struct rotation_thread_timer_queue *rotation_timer_queue = NULL;
5922 sem_t notification_thread_ready;
5923
5924 init_kernel_workarounds();
5925
5926 rcu_register_thread();
5927
5928 if (set_signal_handler()) {
5929 retval = -1;
5930 goto exit_set_signal_handler;
5931 }
5932
5933 if (sessiond_timer_signal_init()) {
5934 retval = -1;
5935 goto exit_set_signal_handler;
5936 }
5937
5938 page_size = sysconf(_SC_PAGESIZE);
5939 if (page_size < 0) {
5940 PERROR("sysconf _SC_PAGESIZE");
5941 page_size = LONG_MAX;
5942 WARN("Fallback page size to %ld", page_size);
5943 }
5944
5945 ret = sessiond_config_init(&config);
5946 if (ret) {
5947 retval = -1;
5948 goto exit_set_signal_handler;
5949 }
5950
5951 /*
5952 * Init config from environment variables.
5953 * Command line option override env configuration per-doc. Do env first.
5954 */
5955 sessiond_config_apply_env_config(&config);
5956
5957 /*
5958 * Parse arguments and load the daemon configuration file.
5959 *
5960 * We have an exit_options exit path to free memory reserved by
5961 * set_options. This is needed because the rest of sessiond_cleanup()
5962 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5963 * depends on set_options.
5964 */
5965 progname = argv[0];
5966 if (set_options(argc, argv)) {
5967 retval = -1;
5968 goto exit_options;
5969 }
5970
5971 /*
5972 * Resolve all paths received as arguments, configuration option, or
5973 * through environment variable as absolute paths. This is necessary
5974 * since daemonizing causes the sessiond's current working directory
5975 * to '/'.
5976 */
5977 ret = sessiond_config_resolve_paths(&config);
5978 if (ret) {
5979 goto exit_options;
5980 }
5981
5982 /* Apply config. */
5983 lttng_opt_verbose = config.verbose;
5984 lttng_opt_quiet = config.quiet;
5985 kconsumer_data.err_unix_sock_path =
5986 config.kconsumerd_err_unix_sock_path.value;
5987 kconsumer_data.cmd_unix_sock_path =
5988 config.kconsumerd_cmd_unix_sock_path.value;
5989 ustconsumer32_data.err_unix_sock_path =
5990 config.consumerd32_err_unix_sock_path.value;
5991 ustconsumer32_data.cmd_unix_sock_path =
5992 config.consumerd32_cmd_unix_sock_path.value;
5993 ustconsumer64_data.err_unix_sock_path =
5994 config.consumerd64_err_unix_sock_path.value;
5995 ustconsumer64_data.cmd_unix_sock_path =
5996 config.consumerd64_cmd_unix_sock_path.value;
5997 set_clock_plugin_env();
5998
5999 sessiond_config_log(&config);
6000
6001 if (create_lttng_rundir()) {
6002 retval = -1;
6003 goto exit_options;
6004 }
6005
6006 /* Abort launch if a session daemon is already running. */
6007 if (check_existing_daemon()) {
6008 ERR("A session daemon is already running.");
6009 retval = -1;
6010 goto exit_options;
6011 }
6012
6013 /* Daemonize */
6014 if (config.daemonize || config.background) {
6015 int i;
6016
6017 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
6018 !config.background);
6019 if (ret < 0) {
6020 retval = -1;
6021 goto exit_options;
6022 }
6023
6024 /*
6025 * We are in the child. Make sure all other file descriptors are
6026 * closed, in case we are called with more opened file
6027 * descriptors than the standard ones and the lock file.
6028 */
6029 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
6030 if (i == lockfile_fd) {
6031 continue;
6032 }
6033 (void) close(i);
6034 }
6035 }
6036
6037 if (run_as_create_worker(argv[0]) < 0) {
6038 goto exit_create_run_as_worker_cleanup;
6039 }
6040
6041 /*
6042 * Starting from here, we can create threads. This needs to be after
6043 * lttng_daemonize due to RCU.
6044 */
6045
6046 /*
6047 * Initialize the health check subsystem. This call should set the
6048 * appropriate time values.
6049 */
6050 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
6051 if (!health_sessiond) {
6052 PERROR("health_app_create error");
6053 retval = -1;
6054 goto exit_health_sessiond_cleanup;
6055 }
6056
6057 /* Create thread to clean up RCU hash tables */
6058 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
6059 retval = -1;
6060 goto exit_ht_cleanup;
6061 }
6062
6063 /* Create thread quit pipe */
6064 if (init_thread_quit_pipe()) {
6065 retval = -1;
6066 goto exit_init_data;
6067 }
6068
6069 /* Check if daemon is UID = 0 */
6070 is_root = !getuid();
6071 if (is_root) {
6072 /* Create global run dir with root access */
6073
6074 kernel_channel_monitor_pipe = lttng_pipe_open(0);
6075 if (!kernel_channel_monitor_pipe) {
6076 ERR("Failed to create kernel consumer channel monitor pipe");
6077 retval = -1;
6078 goto exit_init_data;
6079 }
6080 kconsumer_data.channel_monitor_pipe =
6081 lttng_pipe_release_writefd(
6082 kernel_channel_monitor_pipe);
6083 if (kconsumer_data.channel_monitor_pipe < 0) {
6084 retval = -1;
6085 goto exit_init_data;
6086 }
6087 kernel_channel_rotate_pipe = lttng_pipe_open(0);
6088 if (!kernel_channel_rotate_pipe) {
6089 ERR("Failed to create kernel consumer channel rotate pipe");
6090 retval = -1;
6091 goto exit_init_data;
6092 }
6093 kconsumer_data.channel_rotate_pipe =
6094 lttng_pipe_release_writefd(
6095 kernel_channel_rotate_pipe);
6096 if (kconsumer_data.channel_rotate_pipe < 0) {
6097 retval = -1;
6098 goto exit_init_data;
6099 }
6100 }
6101
6102 /* Set consumer initial state */
6103 kernel_consumerd_state = CONSUMER_STOPPED;
6104 ust_consumerd_state = CONSUMER_STOPPED;
6105
6106 ust32_channel_monitor_pipe = lttng_pipe_open(0);
6107 if (!ust32_channel_monitor_pipe) {
6108 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
6109 retval = -1;
6110 goto exit_init_data;
6111 }
6112 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
6113 ust32_channel_monitor_pipe);
6114 if (ustconsumer32_data.channel_monitor_pipe < 0) {
6115 retval = -1;
6116 goto exit_init_data;
6117 }
6118 ust32_channel_rotate_pipe = lttng_pipe_open(0);
6119 if (!ust32_channel_rotate_pipe) {
6120 ERR("Failed to create 32-bit user space consumer channel rotate pipe");
6121 retval = -1;
6122 goto exit_init_data;
6123 }
6124 ustconsumer32_data.channel_rotate_pipe = lttng_pipe_release_writefd(
6125 ust32_channel_rotate_pipe);
6126 if (ustconsumer32_data.channel_rotate_pipe < 0) {
6127 retval = -1;
6128 goto exit_init_data;
6129 }
6130
6131 /*
6132 * The rotation_timer_queue structure is shared between the sessiond timer
6133 * thread and the rotation thread. The main() keeps the ownership and
6134 * destroys it when both threads have quit.
6135 */
6136 rotation_timer_queue = create_rotate_timer_queue();
6137 if (!rotation_timer_queue) {
6138 retval = -1;
6139 goto exit_init_data;
6140 }
6141 timer_thread_ctx.rotation_timer_queue = rotation_timer_queue;
6142
6143 ust64_channel_monitor_pipe = lttng_pipe_open(0);
6144 if (!ust64_channel_monitor_pipe) {
6145 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
6146 retval = -1;
6147 goto exit_init_data;
6148 }
6149 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
6150 ust64_channel_monitor_pipe);
6151 if (ustconsumer64_data.channel_monitor_pipe < 0) {
6152 retval = -1;
6153 goto exit_init_data;
6154 }
6155 ust64_channel_rotate_pipe = lttng_pipe_open(0);
6156 if (!ust64_channel_rotate_pipe) {
6157 ERR("Failed to create 64-bit user space consumer channel rotate pipe");
6158 retval = -1;
6159 goto exit_init_data;
6160 }
6161 ustconsumer64_data.channel_rotate_pipe = lttng_pipe_release_writefd(
6162 ust64_channel_rotate_pipe);
6163 if (ustconsumer64_data.channel_rotate_pipe < 0) {
6164 retval = -1;
6165 goto exit_init_data;
6166 }
6167
6168 /*
6169 * Init UST app hash table. Alloc hash table before this point since
6170 * cleanup() can get called after that point.
6171 */
6172 if (ust_app_ht_alloc()) {
6173 ERR("Failed to allocate UST app hash table");
6174 retval = -1;
6175 goto exit_init_data;
6176 }
6177
6178 /*
6179 * Initialize agent app hash table. We allocate the hash table here
6180 * since cleanup() can get called after this point.
6181 */
6182 if (agent_app_ht_alloc()) {
6183 ERR("Failed to allocate Agent app hash table");
6184 retval = -1;
6185 goto exit_init_data;
6186 }
6187
6188 /*
6189 * These actions must be executed as root. We do that *after* setting up
6190 * the sockets path because we MUST make the check for another daemon using
6191 * those paths *before* trying to set the kernel consumer sockets and init
6192 * kernel tracer.
6193 */
6194 if (is_root) {
6195 if (set_consumer_sockets(&kconsumer_data)) {
6196 retval = -1;
6197 goto exit_init_data;
6198 }
6199
6200 /* Setup kernel tracer */
6201 if (!config.no_kernel) {
6202 init_kernel_tracer();
6203 if (kernel_tracer_fd >= 0) {
6204 ret = syscall_init_table();
6205 if (ret < 0) {
6206 ERR("Unable to populate syscall table. "
6207 "Syscall tracing won't work "
6208 "for this session daemon.");
6209 }
6210 }
6211 }
6212
6213 /* Set ulimit for open files */
6214 set_ulimit();
6215 }
6216 /* init lttng_fd tracking must be done after set_ulimit. */
6217 lttng_fd_init();
6218
6219 if (set_consumer_sockets(&ustconsumer64_data)) {
6220 retval = -1;
6221 goto exit_init_data;
6222 }
6223
6224 if (set_consumer_sockets(&ustconsumer32_data)) {
6225 retval = -1;
6226 goto exit_init_data;
6227 }
6228
6229 /* Setup the needed unix socket */
6230 if (init_daemon_socket()) {
6231 retval = -1;
6232 goto exit_init_data;
6233 }
6234
6235 /* Set credentials to socket */
6236 if (is_root && set_permissions(config.rundir.value)) {
6237 retval = -1;
6238 goto exit_init_data;
6239 }
6240
6241 /* Get parent pid if -S, --sig-parent is specified. */
6242 if (config.sig_parent) {
6243 ppid = getppid();
6244 }
6245
6246 /* Setup the kernel pipe for waking up the kernel thread */
6247 if (is_root && !config.no_kernel) {
6248 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
6249 retval = -1;
6250 goto exit_init_data;
6251 }
6252 }
6253
6254 /* Setup the thread apps communication pipe. */
6255 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
6256 retval = -1;
6257 goto exit_init_data;
6258 }
6259
6260 /* Setup the thread apps notify communication pipe. */
6261 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
6262 retval = -1;
6263 goto exit_init_data;
6264 }
6265
6266 /* Initialize global buffer per UID and PID registry. */
6267 buffer_reg_init_uid_registry();
6268 buffer_reg_init_pid_registry();
6269
6270 /* Init UST command queue. */
6271 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
6272
6273 /*
6274 * Get session list pointer. This pointer MUST NOT be free'd. This list
6275 * is statically declared in session.c
6276 */
6277 session_list_ptr = session_get_list();
6278
6279 cmd_init();
6280
6281 /* Check for the application socket timeout env variable. */
6282 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
6283 if (env_app_timeout) {
6284 config.app_socket_timeout = atoi(env_app_timeout);
6285 } else {
6286 config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
6287 }
6288
6289 ret = write_pidfile();
6290 if (ret) {
6291 ERR("Error in write_pidfile");
6292 retval = -1;
6293 goto exit_init_data;
6294 }
6295
6296 /* Initialize communication library */
6297 lttcomm_init();
6298 /* Initialize TCP timeout values */
6299 lttcomm_inet_init();
6300
6301 if (load_session_init_data(&load_info) < 0) {
6302 retval = -1;
6303 goto exit_init_data;
6304 }
6305 load_info->path = config.load_session_path.value;
6306
6307 /* Create health-check thread. */
6308 ret = pthread_create(&health_thread, default_pthread_attr(),
6309 thread_manage_health, (void *) NULL);
6310 if (ret) {
6311 errno = ret;
6312 PERROR("pthread_create health");
6313 retval = -1;
6314 goto exit_health;
6315 }
6316
6317 /*
6318 * The rotation thread needs the notification thread to be ready before
6319 * creating the rotate_notification_channel, so we use this semaphore as
6320 * a rendez-vous point.
6321 */
6322 sem_init(&notification_thread_ready, 0, 0);
6323
6324 /* notification_thread_data acquires the pipes' read side. */
6325 notification_thread_handle = notification_thread_handle_create(
6326 ust32_channel_monitor_pipe,
6327 ust64_channel_monitor_pipe,
6328 kernel_channel_monitor_pipe,
6329 &notification_thread_ready);
6330 if (!notification_thread_handle) {
6331 retval = -1;
6332 ERR("Failed to create notification thread shared data");
6333 stop_threads();
6334 goto exit_notification;
6335 }
6336
6337 /* Create notification thread. */
6338 ret = pthread_create(&notification_thread, default_pthread_attr(),
6339 thread_notification, notification_thread_handle);
6340 if (ret) {
6341 errno = ret;
6342 PERROR("pthread_create notification");
6343 retval = -1;
6344 stop_threads();
6345 goto exit_notification;
6346 }
6347 notification_thread_launched = true;
6348
6349 /* Create timer thread. */
6350 ret = pthread_create(&timer_thread, default_pthread_attr(),
6351 sessiond_timer_thread, &timer_thread_ctx);
6352 if (ret) {
6353 errno = ret;
6354 PERROR("pthread_create timer");
6355 retval = -1;
6356 stop_threads();
6357 goto exit_notification;
6358 }
6359 timer_thread_launched = true;
6360
6361 /* rotation_thread_data acquires the pipes' read side. */
6362 rotation_thread_handle = rotation_thread_handle_create(
6363 ust32_channel_rotate_pipe,
6364 ust64_channel_rotate_pipe,
6365 kernel_channel_rotate_pipe,
6366 thread_quit_pipe[0],
6367 rotation_timer_queue,
6368 notification_thread_handle,
6369 &notification_thread_ready);
6370 if (!rotation_thread_handle) {
6371 retval = -1;
6372 ERR("Failed to create rotation thread shared data");
6373 stop_threads();
6374 goto exit_rotation;
6375 }
6376
6377 /* Create rotation thread. */
6378 ret = pthread_create(&rotation_thread, default_pthread_attr(),
6379 thread_rotation, rotation_thread_handle);
6380 if (ret) {
6381 errno = ret;
6382 PERROR("pthread_create rotation");
6383 retval = -1;
6384 stop_threads();
6385 goto exit_rotation;
6386 }
6387 rotation_thread_launched = true;
6388
6389 /* Create thread to manage the client socket */
6390 ret = pthread_create(&client_thread, default_pthread_attr(),
6391 thread_manage_clients, (void *) NULL);
6392 if (ret) {
6393 errno = ret;
6394 PERROR("pthread_create clients");
6395 retval = -1;
6396 stop_threads();
6397 goto exit_client;
6398 }
6399
6400 /* Create thread to dispatch registration */
6401 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
6402 thread_dispatch_ust_registration, (void *) NULL);
6403 if (ret) {
6404 errno = ret;
6405 PERROR("pthread_create dispatch");
6406 retval = -1;
6407 stop_threads();
6408 goto exit_dispatch;
6409 }
6410
6411 /* Create thread to manage application registration. */
6412 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
6413 thread_registration_apps, (void *) NULL);
6414 if (ret) {
6415 errno = ret;
6416 PERROR("pthread_create registration");
6417 retval = -1;
6418 stop_threads();
6419 goto exit_reg_apps;
6420 }
6421
6422 /* Create thread to manage application socket */
6423 ret = pthread_create(&apps_thread, default_pthread_attr(),
6424 thread_manage_apps, (void *) NULL);
6425 if (ret) {
6426 errno = ret;
6427 PERROR("pthread_create apps");
6428 retval = -1;
6429 stop_threads();
6430 goto exit_apps;
6431 }
6432
6433 /* Create thread to manage application notify socket */
6434 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
6435 ust_thread_manage_notify, (void *) NULL);
6436 if (ret) {
6437 errno = ret;
6438 PERROR("pthread_create notify");
6439 retval = -1;
6440 stop_threads();
6441 goto exit_apps_notify;
6442 }
6443
6444 /* Create agent registration thread. */
6445 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
6446 agent_thread_manage_registration, (void *) NULL);
6447 if (ret) {
6448 errno = ret;
6449 PERROR("pthread_create agent");
6450 retval = -1;
6451 stop_threads();
6452 goto exit_agent_reg;
6453 }
6454
6455 /* Don't start this thread if kernel tracing is not requested nor root */
6456 if (is_root && !config.no_kernel) {
6457 /* Create kernel thread to manage kernel event */
6458 ret = pthread_create(&kernel_thread, default_pthread_attr(),
6459 thread_manage_kernel, (void *) NULL);
6460 if (ret) {
6461 errno = ret;
6462 PERROR("pthread_create kernel");
6463 retval = -1;
6464 stop_threads();
6465 goto exit_kernel;
6466 }
6467 }
6468
6469 /* Create session loading thread. */
6470 ret = pthread_create(&load_session_thread, default_pthread_attr(),
6471 thread_load_session, load_info);
6472 if (ret) {
6473 errno = ret;
6474 PERROR("pthread_create load_session_thread");
6475 retval = -1;
6476 stop_threads();
6477 goto exit_load_session;
6478 }
6479
6480 /*
6481 * This is where we start awaiting program completion (e.g. through
6482 * signal that asks threads to teardown).
6483 */
6484
6485 ret = pthread_join(load_session_thread, &status);
6486 if (ret) {
6487 errno = ret;
6488 PERROR("pthread_join load_session_thread");
6489 retval = -1;
6490 }
6491 exit_load_session:
6492
6493 if (is_root && !config.no_kernel) {
6494 ret = pthread_join(kernel_thread, &status);
6495 if (ret) {
6496 errno = ret;
6497 PERROR("pthread_join");
6498 retval = -1;
6499 }
6500 }
6501 exit_kernel:
6502
6503 ret = pthread_join(agent_reg_thread, &status);
6504 if (ret) {
6505 errno = ret;
6506 PERROR("pthread_join agent");
6507 retval = -1;
6508 }
6509 exit_agent_reg:
6510
6511 ret = pthread_join(apps_notify_thread, &status);
6512 if (ret) {
6513 errno = ret;
6514 PERROR("pthread_join apps notify");
6515 retval = -1;
6516 }
6517 exit_apps_notify:
6518
6519 ret = pthread_join(apps_thread, &status);
6520 if (ret) {
6521 errno = ret;
6522 PERROR("pthread_join apps");
6523 retval = -1;
6524 }
6525 exit_apps:
6526
6527 ret = pthread_join(reg_apps_thread, &status);
6528 if (ret) {
6529 errno = ret;
6530 PERROR("pthread_join");
6531 retval = -1;
6532 }
6533 exit_reg_apps:
6534
6535 /*
6536 * Join dispatch thread after joining reg_apps_thread to ensure
6537 * we don't leak applications in the queue.
6538 */
6539 ret = pthread_join(dispatch_thread, &status);
6540 if (ret) {
6541 errno = ret;
6542 PERROR("pthread_join");
6543 retval = -1;
6544 }
6545 exit_dispatch:
6546
6547 ret = pthread_join(client_thread, &status);
6548 if (ret) {
6549 errno = ret;
6550 PERROR("pthread_join");
6551 retval = -1;
6552 }
6553
6554 exit_client:
6555 exit_rotation:
6556 exit_notification:
6557 sem_destroy(&notification_thread_ready);
6558 ret = pthread_join(health_thread, &status);
6559 if (ret) {
6560 errno = ret;
6561 PERROR("pthread_join health thread");
6562 retval = -1;
6563 }
6564
6565 exit_health:
6566 exit_init_data:
6567 /*
6568 * Wait for all pending call_rcu work to complete before tearing
6569 * down data structures. call_rcu worker may be trying to
6570 * perform lookups in those structures.
6571 */
6572 rcu_barrier();
6573 /*
6574 * sessiond_cleanup() is called when no other thread is running, except
6575 * the ht_cleanup thread, which is needed to destroy the hash tables.
6576 */
6577 rcu_thread_online();
6578 sessiond_cleanup();
6579
6580 /*
6581 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6582 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6583 * the queue is empty before shutting down the clean-up thread.
6584 */
6585 rcu_barrier();
6586
6587 /*
6588 * The teardown of the notification system is performed after the
6589 * session daemon's teardown in order to allow it to be notified
6590 * of the active session and channels at the moment of the teardown.
6591 */
6592 if (notification_thread_handle) {
6593 if (notification_thread_launched) {
6594 notification_thread_command_quit(
6595 notification_thread_handle);
6596 ret = pthread_join(notification_thread, &status);
6597 if (ret) {
6598 errno = ret;
6599 PERROR("pthread_join notification thread");
6600 retval = -1;
6601 }
6602 }
6603 notification_thread_handle_destroy(notification_thread_handle);
6604 }
6605
6606 if (rotation_thread_handle) {
6607 if (rotation_thread_launched) {
6608 ret = pthread_join(rotation_thread, &status);
6609 if (ret) {
6610 errno = ret;
6611 PERROR("pthread_join rotation thread");
6612 retval = -1;
6613 }
6614 }
6615 rotation_thread_handle_destroy(rotation_thread_handle);
6616 }
6617
6618 if (timer_thread_launched) {
6619 kill(getpid(), LTTNG_SESSIOND_SIG_EXIT);
6620 ret = pthread_join(timer_thread, &status);
6621 if (ret) {
6622 errno = ret;
6623 PERROR("pthread_join timer thread");
6624 retval = -1;
6625 }
6626 }
6627
6628 /*
6629 * After the rotation and timer thread have quit, we can safely destroy
6630 * the rotation_timer_queue.
6631 */
6632 destroy_rotate_timer_queue(rotation_timer_queue);
6633
6634 rcu_thread_offline();
6635 rcu_unregister_thread();
6636
6637 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6638 if (ret) {
6639 retval = -1;
6640 }
6641 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6642 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6643 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6644 lttng_pipe_destroy(ust32_channel_rotate_pipe);
6645 lttng_pipe_destroy(ust64_channel_rotate_pipe);
6646 lttng_pipe_destroy(kernel_channel_rotate_pipe);
6647 exit_ht_cleanup:
6648
6649 health_app_destroy(health_sessiond);
6650 exit_health_sessiond_cleanup:
6651 exit_create_run_as_worker_cleanup:
6652
6653 exit_options:
6654 sessiond_cleanup_lock_file();
6655 sessiond_cleanup_options();
6656
6657 exit_set_signal_handler:
6658 if (!retval) {
6659 exit(EXIT_SUCCESS);
6660 } else {
6661 exit(EXIT_FAILURE);
6662 }
6663 }
This page took 0.209223 seconds and 4 git commands to generate.