88877c700437d54713020f4dc06c66f543fe4d67
[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
53 #include "lttng-sessiond.h"
54 #include "buffer-registry.h"
55 #include "channel.h"
56 #include "cmd.h"
57 #include "consumer.h"
58 #include "context.h"
59 #include "event.h"
60 #include "kernel.h"
61 #include "kernel-consumer.h"
62 #include "modprobe.h"
63 #include "shm.h"
64 #include "ust-ctl.h"
65 #include "ust-consumer.h"
66 #include "utils.h"
67 #include "fd-limit.h"
68 #include "health-sessiond.h"
69 #include "testpoint.h"
70 #include "ust-thread.h"
71 #include "agent-thread.h"
72 #include "save.h"
73 #include "load-session-thread.h"
74 #include "notification-thread.h"
75 #include "notification-thread-commands.h"
76 #include "syscall.h"
77 #include "agent.h"
78 #include "ht-cleanup.h"
79 #include "sessiond-config.h"
80
81 static const char *help_msg =
82 #ifdef LTTNG_EMBED_HELP
83 #include <lttng-sessiond.8.h>
84 #else
85 NULL
86 #endif
87 ;
88
89 const char *progname;
90 static pid_t ppid; /* Parent PID for --sig-parent option */
91 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
92 static int lockfile_fd = -1;
93
94 /* Set to 1 when a SIGUSR1 signal is received. */
95 static int recv_child_signal;
96
97 /*
98 * Consumer daemon specific control data. Every value not initialized here is
99 * set to 0 by the static definition.
100 */
101 static struct consumer_data kconsumer_data = {
102 .type = LTTNG_CONSUMER_KERNEL,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .channel_monitor_pipe = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer64_data = {
112 .type = LTTNG_CONSUMER64_UST,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .channel_monitor_pipe = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121 static struct consumer_data ustconsumer32_data = {
122 .type = LTTNG_CONSUMER32_UST,
123 .err_sock = -1,
124 .cmd_sock = -1,
125 .channel_monitor_pipe = -1,
126 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
127 .lock = PTHREAD_MUTEX_INITIALIZER,
128 .cond = PTHREAD_COND_INITIALIZER,
129 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
130 };
131
132 /* Command line options */
133 static const struct option long_options[] = {
134 { "client-sock", required_argument, 0, 'c' },
135 { "apps-sock", required_argument, 0, 'a' },
136 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
137 { "kconsumerd-err-sock", required_argument, 0, '\0' },
138 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
139 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
140 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
141 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
142 { "consumerd32-path", required_argument, 0, '\0' },
143 { "consumerd32-libdir", required_argument, 0, '\0' },
144 { "consumerd64-path", required_argument, 0, '\0' },
145 { "consumerd64-libdir", required_argument, 0, '\0' },
146 { "daemonize", no_argument, 0, 'd' },
147 { "background", no_argument, 0, 'b' },
148 { "sig-parent", no_argument, 0, 'S' },
149 { "help", no_argument, 0, 'h' },
150 { "group", required_argument, 0, 'g' },
151 { "version", no_argument, 0, 'V' },
152 { "quiet", no_argument, 0, 'q' },
153 { "verbose", no_argument, 0, 'v' },
154 { "verbose-consumer", no_argument, 0, '\0' },
155 { "no-kernel", no_argument, 0, '\0' },
156 { "pidfile", required_argument, 0, 'p' },
157 { "agent-tcp-port", required_argument, 0, '\0' },
158 { "config", required_argument, 0, 'f' },
159 { "load", required_argument, 0, 'l' },
160 { "kmod-probes", required_argument, 0, '\0' },
161 { "extra-kmod-probes", required_argument, 0, '\0' },
162 { NULL, 0, 0, 0 }
163 };
164
165 struct sessiond_config config;
166
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options[] = { "help", "version", "config" };
169
170 /* Shared between threads */
171 static int dispatch_thread_exit;
172
173 /* Sockets and FDs */
174 static int client_sock = -1;
175 static int apps_sock = -1;
176 int kernel_tracer_fd = -1;
177 static int kernel_poll_pipe[2] = { -1, -1 };
178
179 /*
180 * Quit pipe for all threads. This permits a single cancellation point
181 * for all threads when receiving an event on the pipe.
182 */
183 static int thread_quit_pipe[2] = { -1, -1 };
184
185 /*
186 * This pipe is used to inform the thread managing application communication
187 * that a command is queued and ready to be processed.
188 */
189 static int apps_cmd_pipe[2] = { -1, -1 };
190
191 int apps_cmd_notify_pipe[2] = { -1, -1 };
192
193 /* Pthread, Mutexes and Semaphores */
194 static pthread_t apps_thread;
195 static pthread_t apps_notify_thread;
196 static pthread_t reg_apps_thread;
197 static pthread_t client_thread;
198 static pthread_t kernel_thread;
199 static pthread_t dispatch_thread;
200 static pthread_t health_thread;
201 static pthread_t ht_cleanup_thread;
202 static pthread_t agent_reg_thread;
203 static pthread_t load_session_thread;
204 static pthread_t notification_thread;
205
206 /*
207 * UST registration command queue. This queue is tied with a futex and uses a N
208 * wakers / 1 waiter implemented and detailed in futex.c/.h
209 *
210 * The thread_registration_apps and thread_dispatch_ust_registration uses this
211 * queue along with the wait/wake scheme. The thread_manage_apps receives down
212 * the line new application socket and monitors it for any I/O error or clean
213 * close that triggers an unregistration of the application.
214 */
215 static struct ust_cmd_queue ust_cmd_queue;
216
217 /*
218 * Pointer initialized before thread creation.
219 *
220 * This points to the tracing session list containing the session count and a
221 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
222 * MUST NOT be taken if you call a public function in session.c.
223 *
224 * The lock is nested inside the structure: session_list_ptr->lock. Please use
225 * session_lock_list and session_unlock_list for lock acquisition.
226 */
227 static struct ltt_session_list *session_list_ptr;
228
229 int ust_consumerd64_fd = -1;
230 int ust_consumerd32_fd = -1;
231
232 static const char *module_proc_lttng = "/proc/lttng";
233
234 /*
235 * Consumer daemon state which is changed when spawning it, killing it or in
236 * case of a fatal error.
237 */
238 enum consumerd_state {
239 CONSUMER_STARTED = 1,
240 CONSUMER_STOPPED = 2,
241 CONSUMER_ERROR = 3,
242 };
243
244 /*
245 * This consumer daemon state is used to validate if a client command will be
246 * able to reach the consumer. If not, the client is informed. For instance,
247 * doing a "lttng start" when the consumer state is set to ERROR will return an
248 * error to the client.
249 *
250 * The following example shows a possible race condition of this scheme:
251 *
252 * consumer thread error happens
253 * client cmd arrives
254 * client cmd checks state -> still OK
255 * consumer thread exit, sets error
256 * client cmd try to talk to consumer
257 * ...
258 *
259 * However, since the consumer is a different daemon, we have no way of making
260 * sure the command will reach it safely even with this state flag. This is why
261 * we consider that up to the state validation during command processing, the
262 * command is safe. After that, we can not guarantee the correctness of the
263 * client request vis-a-vis the consumer.
264 */
265 static enum consumerd_state ust_consumerd_state;
266 static enum consumerd_state kernel_consumerd_state;
267
268 /* Set in main() with the current page size. */
269 long page_size;
270
271 /* Application health monitoring */
272 struct health_app *health_sessiond;
273
274 /* Am I root or not. */
275 int is_root; /* Set to 1 if the daemon is running as root */
276
277 const char * const config_section_name = "sessiond";
278
279 /* Load session thread information to operate. */
280 struct load_session_thread_data *load_info;
281
282 /* Notification thread handle. */
283 struct notification_thread_handle *notification_thread_handle;
284
285 /* Global hash tables */
286 struct lttng_ht *agent_apps_ht_by_sock = NULL;
287
288 /*
289 * The initialization of the session daemon is done in multiple phases.
290 *
291 * While all threads are launched near-simultaneously, only some of them
292 * are needed to ensure the session daemon can start to respond to client
293 * requests.
294 *
295 * There are two important guarantees that we wish to offer with respect
296 * to the initialisation of the session daemon:
297 * - When the daemonize/background launcher process exits, the sessiond
298 * is fully able to respond to client requests,
299 * - Auto-loaded sessions are visible to clients.
300 *
301 * In order to achieve this, a number of support threads have to be launched
302 * to allow the "client" thread to function properly. Moreover, since the
303 * "load session" thread needs the client thread, we must provide a way
304 * for the "load session" thread to know that the "client" thread is up
305 * and running.
306 *
307 * Hence, the support threads decrement the lttng_sessiond_ready counter
308 * while the "client" threads waits for it to reach 0. Once the "client" thread
309 * unblocks, it posts the message_thread_ready semaphore which allows the
310 * "load session" thread to progress.
311 *
312 * This implies that the "load session" thread is the last to be initialized
313 * and will explicitly call sessiond_signal_parents(), which signals the parents
314 * that the session daemon is fully initialized.
315 *
316 * The three (3) support threads are:
317 * - agent_thread
318 * - notification_thread
319 * - health_thread
320 */
321 int lttng_sessiond_ready = 3;
322
323 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
324 {
325 return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0;
326 }
327
328 /* Notify parents that we are ready for cmd and health check */
329 LTTNG_HIDDEN
330 void sessiond_signal_parents(void)
331 {
332 /*
333 * Notify parent pid that we are ready to accept command
334 * for client side. This ppid is the one from the
335 * external process that spawned us.
336 */
337 if (config.sig_parent) {
338 kill(ppid, SIGUSR1);
339 }
340
341 /*
342 * Notify the parent of the fork() process that we are
343 * ready.
344 */
345 if (config.daemonize || config.background) {
346 kill(child_ppid, SIGUSR1);
347 }
348 }
349
350 LTTNG_HIDDEN
351 void sessiond_notify_ready(void)
352 {
353 /*
354 * The _return variant is used since the implied memory barriers are
355 * required.
356 */
357 (void) uatomic_sub_return(&lttng_sessiond_ready, 1);
358 }
359
360 static
361 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
362 int *a_pipe)
363 {
364 int ret;
365
366 assert(events);
367
368 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
369 if (ret < 0) {
370 goto error;
371 }
372
373 /* Add quit pipe */
374 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
375 if (ret < 0) {
376 goto error;
377 }
378
379 return 0;
380
381 error:
382 return ret;
383 }
384
385 /*
386 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
387 */
388 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
389 {
390 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
391 }
392
393 /*
394 * Init thread quit pipe.
395 *
396 * Return -1 on error or 0 if all pipes are created.
397 */
398 static int __init_thread_quit_pipe(int *a_pipe)
399 {
400 int ret, i;
401
402 ret = pipe(a_pipe);
403 if (ret < 0) {
404 PERROR("thread quit pipe");
405 goto error;
406 }
407
408 for (i = 0; i < 2; i++) {
409 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
410 if (ret < 0) {
411 PERROR("fcntl");
412 goto error;
413 }
414 }
415
416 error:
417 return ret;
418 }
419
420 static int init_thread_quit_pipe(void)
421 {
422 return __init_thread_quit_pipe(thread_quit_pipe);
423 }
424
425 /*
426 * Stop all threads by closing the thread quit pipe.
427 */
428 static void stop_threads(void)
429 {
430 int ret;
431
432 /* Stopping all threads */
433 DBG("Terminating all threads");
434 ret = notify_thread_pipe(thread_quit_pipe[1]);
435 if (ret < 0) {
436 ERR("write error on thread quit pipe");
437 }
438
439 /* Dispatch thread */
440 CMM_STORE_SHARED(dispatch_thread_exit, 1);
441 futex_nto1_wake(&ust_cmd_queue.futex);
442 }
443
444 /*
445 * Close every consumer sockets.
446 */
447 static void close_consumer_sockets(void)
448 {
449 int ret;
450
451 if (kconsumer_data.err_sock >= 0) {
452 ret = close(kconsumer_data.err_sock);
453 if (ret < 0) {
454 PERROR("kernel consumer err_sock close");
455 }
456 }
457 if (ustconsumer32_data.err_sock >= 0) {
458 ret = close(ustconsumer32_data.err_sock);
459 if (ret < 0) {
460 PERROR("UST consumerd32 err_sock close");
461 }
462 }
463 if (ustconsumer64_data.err_sock >= 0) {
464 ret = close(ustconsumer64_data.err_sock);
465 if (ret < 0) {
466 PERROR("UST consumerd64 err_sock close");
467 }
468 }
469 if (kconsumer_data.cmd_sock >= 0) {
470 ret = close(kconsumer_data.cmd_sock);
471 if (ret < 0) {
472 PERROR("kernel consumer cmd_sock close");
473 }
474 }
475 if (ustconsumer32_data.cmd_sock >= 0) {
476 ret = close(ustconsumer32_data.cmd_sock);
477 if (ret < 0) {
478 PERROR("UST consumerd32 cmd_sock close");
479 }
480 }
481 if (ustconsumer64_data.cmd_sock >= 0) {
482 ret = close(ustconsumer64_data.cmd_sock);
483 if (ret < 0) {
484 PERROR("UST consumerd64 cmd_sock close");
485 }
486 }
487 if (kconsumer_data.channel_monitor_pipe >= 0) {
488 ret = close(kconsumer_data.channel_monitor_pipe);
489 if (ret < 0) {
490 PERROR("kernel consumer channel monitor pipe close");
491 }
492 }
493 if (ustconsumer32_data.channel_monitor_pipe >= 0) {
494 ret = close(ustconsumer32_data.channel_monitor_pipe);
495 if (ret < 0) {
496 PERROR("UST consumerd32 channel monitor pipe close");
497 }
498 }
499 if (ustconsumer64_data.channel_monitor_pipe >= 0) {
500 ret = close(ustconsumer64_data.channel_monitor_pipe);
501 if (ret < 0) {
502 PERROR("UST consumerd64 channel monitor pipe close");
503 }
504 }
505 }
506
507 /*
508 * Wait on consumer process termination.
509 *
510 * Need to be called with the consumer data lock held or from a context
511 * ensuring no concurrent access to data (e.g: cleanup).
512 */
513 static void wait_consumer(struct consumer_data *consumer_data)
514 {
515 pid_t ret;
516 int status;
517
518 if (consumer_data->pid <= 0) {
519 return;
520 }
521
522 DBG("Waiting for complete teardown of consumerd (PID: %d)",
523 consumer_data->pid);
524 ret = waitpid(consumer_data->pid, &status, 0);
525 if (ret == -1) {
526 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
527 } else if (!WIFEXITED(status)) {
528 ERR("consumerd termination with error: %d",
529 WEXITSTATUS(ret));
530 }
531 consumer_data->pid = 0;
532 }
533
534 /*
535 * Cleanup the session daemon's data structures.
536 */
537 static void sessiond_cleanup(void)
538 {
539 int ret;
540 struct ltt_session *sess, *stmp;
541
542 DBG("Cleanup sessiond");
543
544 /*
545 * Close the thread quit pipe. It has already done its job,
546 * since we are now called.
547 */
548 utils_close_pipe(thread_quit_pipe);
549
550 /*
551 * If config.pid_file_path.value is undefined, the default file will be
552 * wiped when removing the rundir.
553 */
554 if (config.pid_file_path.value) {
555 ret = remove(config.pid_file_path.value);
556 if (ret < 0) {
557 PERROR("remove pidfile %s", config.pid_file_path.value);
558 }
559 }
560
561 DBG("Removing sessiond and consumerd content of directory %s",
562 config.rundir.value);
563
564 /* sessiond */
565 DBG("Removing %s", config.pid_file_path.value);
566 (void) unlink(config.pid_file_path.value);
567
568 DBG("Removing %s", config.agent_port_file_path.value);
569 (void) unlink(config.agent_port_file_path.value);
570
571 /* kconsumerd */
572 DBG("Removing %s", kconsumer_data.err_unix_sock_path);
573 (void) unlink(kconsumer_data.err_unix_sock_path);
574
575 DBG("Removing directory %s", config.kconsumerd_path.value);
576 (void) rmdir(config.kconsumerd_path.value);
577
578 /* ust consumerd 32 */
579 DBG("Removing %s", config.consumerd32_err_unix_sock_path.value);
580 (void) unlink(config.consumerd32_err_unix_sock_path.value);
581
582 DBG("Removing directory %s", config.consumerd32_path.value);
583 (void) rmdir(config.consumerd32_path.value);
584
585 /* ust consumerd 64 */
586 DBG("Removing %s", config.consumerd64_err_unix_sock_path.value);
587 (void) unlink(config.consumerd64_err_unix_sock_path.value);
588
589 DBG("Removing directory %s", config.consumerd64_path.value);
590 (void) rmdir(config.consumerd64_path.value);
591
592 DBG("Cleaning up all sessions");
593
594 /* Destroy session list mutex */
595 if (session_list_ptr != NULL) {
596 pthread_mutex_destroy(&session_list_ptr->lock);
597
598 /* Cleanup ALL session */
599 cds_list_for_each_entry_safe(sess, stmp,
600 &session_list_ptr->head, list) {
601 cmd_destroy_session(sess, kernel_poll_pipe[1]);
602 }
603 }
604
605 wait_consumer(&kconsumer_data);
606 wait_consumer(&ustconsumer64_data);
607 wait_consumer(&ustconsumer32_data);
608
609 DBG("Cleaning up all agent apps");
610 agent_app_ht_clean();
611
612 DBG("Closing all UST sockets");
613 ust_app_clean_list();
614 buffer_reg_destroy_registries();
615
616 if (is_root && !config.no_kernel) {
617 DBG2("Closing kernel fd");
618 if (kernel_tracer_fd >= 0) {
619 ret = close(kernel_tracer_fd);
620 if (ret) {
621 PERROR("close");
622 }
623 }
624 DBG("Unloading kernel modules");
625 modprobe_remove_lttng_all();
626 free(syscall_table);
627 }
628
629 close_consumer_sockets();
630
631 if (load_info) {
632 load_session_destroy_data(load_info);
633 free(load_info);
634 }
635
636 /*
637 * We do NOT rmdir rundir because there are other processes
638 * using it, for instance lttng-relayd, which can start in
639 * parallel with this teardown.
640 */
641 }
642
643 /*
644 * Cleanup the daemon's option data structures.
645 */
646 static void sessiond_cleanup_options(void)
647 {
648 DBG("Cleaning up options");
649
650 sessiond_config_fini(&config);
651
652 run_as_destroy_worker();
653 }
654
655 /*
656 * Send data on a unix socket using the liblttsessiondcomm API.
657 *
658 * Return lttcomm error code.
659 */
660 static int send_unix_sock(int sock, void *buf, size_t len)
661 {
662 /* Check valid length */
663 if (len == 0) {
664 return -1;
665 }
666
667 return lttcomm_send_unix_sock(sock, buf, len);
668 }
669
670 /*
671 * Free memory of a command context structure.
672 */
673 static void clean_command_ctx(struct command_ctx **cmd_ctx)
674 {
675 DBG("Clean command context structure");
676 if (*cmd_ctx) {
677 if ((*cmd_ctx)->llm) {
678 free((*cmd_ctx)->llm);
679 }
680 if ((*cmd_ctx)->lsm) {
681 free((*cmd_ctx)->lsm);
682 }
683 free(*cmd_ctx);
684 *cmd_ctx = NULL;
685 }
686 }
687
688 /*
689 * Notify UST applications using the shm mmap futex.
690 */
691 static int notify_ust_apps(int active)
692 {
693 char *wait_shm_mmap;
694
695 DBG("Notifying applications of session daemon state: %d", active);
696
697 /* See shm.c for this call implying mmap, shm and futex calls */
698 wait_shm_mmap = shm_ust_get_mmap(config.wait_shm_path.value, is_root);
699 if (wait_shm_mmap == NULL) {
700 goto error;
701 }
702
703 /* Wake waiting process */
704 futex_wait_update((int32_t *) wait_shm_mmap, active);
705
706 /* Apps notified successfully */
707 return 0;
708
709 error:
710 return -1;
711 }
712
713 /*
714 * Setup the outgoing data buffer for the response (llm) by allocating the
715 * right amount of memory and copying the original information from the lsm
716 * structure.
717 *
718 * Return 0 on success, negative value on error.
719 */
720 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
721 const void *payload_buf, size_t payload_len,
722 const void *cmd_header_buf, size_t cmd_header_len)
723 {
724 int ret = 0;
725 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
726 const size_t cmd_header_offset = header_len;
727 const size_t payload_offset = cmd_header_offset + cmd_header_len;
728 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
729
730 cmd_ctx->llm = zmalloc(total_msg_size);
731
732 if (cmd_ctx->llm == NULL) {
733 PERROR("zmalloc");
734 ret = -ENOMEM;
735 goto end;
736 }
737
738 /* Copy common data */
739 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
740 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
741 cmd_ctx->llm->cmd_header_size = cmd_header_len;
742 cmd_ctx->llm->data_size = payload_len;
743 cmd_ctx->lttng_msg_size = total_msg_size;
744
745 /* Copy command header */
746 if (cmd_header_len) {
747 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
748 cmd_header_len);
749 }
750
751 /* Copy payload */
752 if (payload_len) {
753 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
754 payload_len);
755 }
756
757 end:
758 return ret;
759 }
760
761 /*
762 * Version of setup_lttng_msg() without command header.
763 */
764 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
765 void *payload_buf, size_t payload_len)
766 {
767 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
768 }
769 /*
770 * Update the kernel poll set of all channel fd available over all tracing
771 * session. Add the wakeup pipe at the end of the set.
772 */
773 static int update_kernel_poll(struct lttng_poll_event *events)
774 {
775 int ret;
776 struct ltt_session *session;
777 struct ltt_kernel_channel *channel;
778
779 DBG("Updating kernel poll set");
780
781 session_lock_list();
782 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
783 session_lock(session);
784 if (session->kernel_session == NULL) {
785 session_unlock(session);
786 continue;
787 }
788
789 cds_list_for_each_entry(channel,
790 &session->kernel_session->channel_list.head, list) {
791 /* Add channel fd to the kernel poll set */
792 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
793 if (ret < 0) {
794 session_unlock(session);
795 goto error;
796 }
797 DBG("Channel fd %d added to kernel set", channel->fd);
798 }
799 session_unlock(session);
800 }
801 session_unlock_list();
802
803 return 0;
804
805 error:
806 session_unlock_list();
807 return -1;
808 }
809
810 /*
811 * Find the channel fd from 'fd' over all tracing session. When found, check
812 * for new channel stream and send those stream fds to the kernel consumer.
813 *
814 * Useful for CPU hotplug feature.
815 */
816 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
817 {
818 int ret = 0;
819 struct ltt_session *session;
820 struct ltt_kernel_session *ksess;
821 struct ltt_kernel_channel *channel;
822
823 DBG("Updating kernel streams for channel fd %d", fd);
824
825 session_lock_list();
826 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
827 session_lock(session);
828 if (session->kernel_session == NULL) {
829 session_unlock(session);
830 continue;
831 }
832 ksess = session->kernel_session;
833
834 cds_list_for_each_entry(channel,
835 &ksess->channel_list.head, list) {
836 struct lttng_ht_iter iter;
837 struct consumer_socket *socket;
838
839 if (channel->fd != fd) {
840 continue;
841 }
842 DBG("Channel found, updating kernel streams");
843 ret = kernel_open_channel_stream(channel);
844 if (ret < 0) {
845 goto error;
846 }
847 /* Update the stream global counter */
848 ksess->stream_count_global += ret;
849
850 /*
851 * Have we already sent fds to the consumer? If yes, it
852 * means that tracing is started so it is safe to send
853 * our updated stream fds.
854 */
855 if (ksess->consumer_fds_sent != 1
856 || ksess->consumer == NULL) {
857 ret = -1;
858 goto error;
859 }
860
861 rcu_read_lock();
862 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
863 &iter.iter, socket, node.node) {
864 pthread_mutex_lock(socket->lock);
865 ret = kernel_consumer_send_channel_stream(socket,
866 channel, ksess,
867 session->output_traces ? 1 : 0);
868 pthread_mutex_unlock(socket->lock);
869 if (ret < 0) {
870 rcu_read_unlock();
871 goto error;
872 }
873 }
874 rcu_read_unlock();
875 }
876 session_unlock(session);
877 }
878 session_unlock_list();
879 return ret;
880
881 error:
882 session_unlock(session);
883 session_unlock_list();
884 return ret;
885 }
886
887 /*
888 * For each tracing session, update newly registered apps. The session list
889 * lock MUST be acquired before calling this.
890 */
891 static void update_ust_app(int app_sock)
892 {
893 struct ltt_session *sess, *stmp;
894
895 /* Consumer is in an ERROR state. Stop any application update. */
896 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
897 /* Stop the update process since the consumer is dead. */
898 return;
899 }
900
901 /* For all tracing session(s) */
902 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
903 struct ust_app *app;
904
905 session_lock(sess);
906 if (!sess->ust_session) {
907 goto unlock_session;
908 }
909
910 rcu_read_lock();
911 assert(app_sock >= 0);
912 app = ust_app_find_by_sock(app_sock);
913 if (app == NULL) {
914 /*
915 * Application can be unregistered before so
916 * this is possible hence simply stopping the
917 * update.
918 */
919 DBG3("UST app update failed to find app sock %d",
920 app_sock);
921 goto unlock_rcu;
922 }
923 ust_app_global_update(sess->ust_session, app);
924 unlock_rcu:
925 rcu_read_unlock();
926 unlock_session:
927 session_unlock(sess);
928 }
929 }
930
931 /*
932 * This thread manage event coming from the kernel.
933 *
934 * Features supported in this thread:
935 * -) CPU Hotplug
936 */
937 static void *thread_manage_kernel(void *data)
938 {
939 int ret, i, pollfd, update_poll_flag = 1, err = -1;
940 uint32_t revents, nb_fd;
941 char tmp;
942 struct lttng_poll_event events;
943
944 DBG("[thread] Thread manage kernel started");
945
946 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
947
948 /*
949 * This first step of the while is to clean this structure which could free
950 * non NULL pointers so initialize it before the loop.
951 */
952 lttng_poll_init(&events);
953
954 if (testpoint(sessiond_thread_manage_kernel)) {
955 goto error_testpoint;
956 }
957
958 health_code_update();
959
960 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
961 goto error_testpoint;
962 }
963
964 while (1) {
965 health_code_update();
966
967 if (update_poll_flag == 1) {
968 /* Clean events object. We are about to populate it again. */
969 lttng_poll_clean(&events);
970
971 ret = sessiond_set_thread_pollset(&events, 2);
972 if (ret < 0) {
973 goto error_poll_create;
974 }
975
976 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
977 if (ret < 0) {
978 goto error;
979 }
980
981 /* This will add the available kernel channel if any. */
982 ret = update_kernel_poll(&events);
983 if (ret < 0) {
984 goto error;
985 }
986 update_poll_flag = 0;
987 }
988
989 DBG("Thread kernel polling");
990
991 /* Poll infinite value of time */
992 restart:
993 health_poll_entry();
994 ret = lttng_poll_wait(&events, -1);
995 DBG("Thread kernel return from poll on %d fds",
996 LTTNG_POLL_GETNB(&events));
997 health_poll_exit();
998 if (ret < 0) {
999 /*
1000 * Restart interrupted system call.
1001 */
1002 if (errno == EINTR) {
1003 goto restart;
1004 }
1005 goto error;
1006 } else if (ret == 0) {
1007 /* Should not happen since timeout is infinite */
1008 ERR("Return value of poll is 0 with an infinite timeout.\n"
1009 "This should not have happened! Continuing...");
1010 continue;
1011 }
1012
1013 nb_fd = ret;
1014
1015 for (i = 0; i < nb_fd; i++) {
1016 /* Fetch once the poll data */
1017 revents = LTTNG_POLL_GETEV(&events, i);
1018 pollfd = LTTNG_POLL_GETFD(&events, i);
1019
1020 health_code_update();
1021
1022 if (!revents) {
1023 /* No activity for this FD (poll implementation). */
1024 continue;
1025 }
1026
1027 /* Thread quit pipe has been closed. Killing thread. */
1028 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1029 if (ret) {
1030 err = 0;
1031 goto exit;
1032 }
1033
1034 /* Check for data on kernel pipe */
1035 if (revents & LPOLLIN) {
1036 if (pollfd == kernel_poll_pipe[0]) {
1037 (void) lttng_read(kernel_poll_pipe[0],
1038 &tmp, 1);
1039 /*
1040 * Ret value is useless here, if this pipe gets any actions an
1041 * update is required anyway.
1042 */
1043 update_poll_flag = 1;
1044 continue;
1045 } else {
1046 /*
1047 * New CPU detected by the kernel. Adding kernel stream to
1048 * kernel session and updating the kernel consumer
1049 */
1050 ret = update_kernel_stream(&kconsumer_data, pollfd);
1051 if (ret < 0) {
1052 continue;
1053 }
1054 break;
1055 }
1056 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1057 update_poll_flag = 1;
1058 continue;
1059 } else {
1060 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1061 goto error;
1062 }
1063 }
1064 }
1065
1066 exit:
1067 error:
1068 lttng_poll_clean(&events);
1069 error_poll_create:
1070 error_testpoint:
1071 utils_close_pipe(kernel_poll_pipe);
1072 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1073 if (err) {
1074 health_error();
1075 ERR("Health error occurred in %s", __func__);
1076 WARN("Kernel thread died unexpectedly. "
1077 "Kernel tracing can continue but CPU hotplug is disabled.");
1078 }
1079 health_unregister(health_sessiond);
1080 DBG("Kernel thread dying");
1081 return NULL;
1082 }
1083
1084 /*
1085 * Signal pthread condition of the consumer data that the thread.
1086 */
1087 static void signal_consumer_condition(struct consumer_data *data, int state)
1088 {
1089 pthread_mutex_lock(&data->cond_mutex);
1090
1091 /*
1092 * The state is set before signaling. It can be any value, it's the waiter
1093 * job to correctly interpret this condition variable associated to the
1094 * consumer pthread_cond.
1095 *
1096 * A value of 0 means that the corresponding thread of the consumer data
1097 * was not started. 1 indicates that the thread has started and is ready
1098 * for action. A negative value means that there was an error during the
1099 * thread bootstrap.
1100 */
1101 data->consumer_thread_is_ready = state;
1102 (void) pthread_cond_signal(&data->cond);
1103
1104 pthread_mutex_unlock(&data->cond_mutex);
1105 }
1106
1107 /*
1108 * This thread manage the consumer error sent back to the session daemon.
1109 */
1110 static void *thread_manage_consumer(void *data)
1111 {
1112 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1113 uint32_t revents, nb_fd;
1114 enum lttcomm_return_code code;
1115 struct lttng_poll_event events;
1116 struct consumer_data *consumer_data = data;
1117 struct consumer_socket *cmd_socket_wrapper = NULL;
1118
1119 DBG("[thread] Manage consumer started");
1120
1121 rcu_register_thread();
1122 rcu_thread_online();
1123
1124 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1125
1126 health_code_update();
1127
1128 /*
1129 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1130 * metadata_sock. Nothing more will be added to this poll set.
1131 */
1132 ret = sessiond_set_thread_pollset(&events, 3);
1133 if (ret < 0) {
1134 goto error_poll;
1135 }
1136
1137 /*
1138 * The error socket here is already in a listening state which was done
1139 * just before spawning this thread to avoid a race between the consumer
1140 * daemon exec trying to connect and the listen() call.
1141 */
1142 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1143 if (ret < 0) {
1144 goto error;
1145 }
1146
1147 health_code_update();
1148
1149 /* Infinite blocking call, waiting for transmission */
1150 restart:
1151 health_poll_entry();
1152
1153 if (testpoint(sessiond_thread_manage_consumer)) {
1154 goto error;
1155 }
1156
1157 ret = lttng_poll_wait(&events, -1);
1158 health_poll_exit();
1159 if (ret < 0) {
1160 /*
1161 * Restart interrupted system call.
1162 */
1163 if (errno == EINTR) {
1164 goto restart;
1165 }
1166 goto error;
1167 }
1168
1169 nb_fd = ret;
1170
1171 for (i = 0; i < nb_fd; i++) {
1172 /* Fetch once the poll data */
1173 revents = LTTNG_POLL_GETEV(&events, i);
1174 pollfd = LTTNG_POLL_GETFD(&events, i);
1175
1176 health_code_update();
1177
1178 if (!revents) {
1179 /* No activity for this FD (poll implementation). */
1180 continue;
1181 }
1182
1183 /* Thread quit pipe has been closed. Killing thread. */
1184 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1185 if (ret) {
1186 err = 0;
1187 goto exit;
1188 }
1189
1190 /* Event on the registration socket */
1191 if (pollfd == consumer_data->err_sock) {
1192 if (revents & LPOLLIN) {
1193 continue;
1194 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1195 ERR("consumer err socket poll error");
1196 goto error;
1197 } else {
1198 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1199 goto error;
1200 }
1201 }
1202 }
1203
1204 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1205 if (sock < 0) {
1206 goto error;
1207 }
1208
1209 /*
1210 * Set the CLOEXEC flag. Return code is useless because either way, the
1211 * show must go on.
1212 */
1213 (void) utils_set_fd_cloexec(sock);
1214
1215 health_code_update();
1216
1217 DBG2("Receiving code from consumer err_sock");
1218
1219 /* Getting status code from kconsumerd */
1220 ret = lttcomm_recv_unix_sock(sock, &code,
1221 sizeof(enum lttcomm_return_code));
1222 if (ret <= 0) {
1223 goto error;
1224 }
1225
1226 health_code_update();
1227 if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1228 ERR("consumer error when waiting for SOCK_READY : %s",
1229 lttcomm_get_readable_code(-code));
1230 goto error;
1231 }
1232
1233 /* Connect both command and metadata sockets. */
1234 consumer_data->cmd_sock =
1235 lttcomm_connect_unix_sock(
1236 consumer_data->cmd_unix_sock_path);
1237 consumer_data->metadata_fd =
1238 lttcomm_connect_unix_sock(
1239 consumer_data->cmd_unix_sock_path);
1240 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1241 PERROR("consumer connect cmd socket");
1242 /* On error, signal condition and quit. */
1243 signal_consumer_condition(consumer_data, -1);
1244 goto error;
1245 }
1246
1247 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1248
1249 /* Create metadata socket lock. */
1250 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1251 if (consumer_data->metadata_sock.lock == NULL) {
1252 PERROR("zmalloc pthread mutex");
1253 goto error;
1254 }
1255 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1256
1257 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1258 DBG("Consumer metadata socket ready (fd: %d)",
1259 consumer_data->metadata_fd);
1260
1261 /*
1262 * Remove the consumerd error sock since we've established a connection.
1263 */
1264 ret = lttng_poll_del(&events, consumer_data->err_sock);
1265 if (ret < 0) {
1266 goto error;
1267 }
1268
1269 /* Add new accepted error socket. */
1270 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1271 if (ret < 0) {
1272 goto error;
1273 }
1274
1275 /* Add metadata socket that is successfully connected. */
1276 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1277 LPOLLIN | LPOLLRDHUP);
1278 if (ret < 0) {
1279 goto error;
1280 }
1281
1282 health_code_update();
1283
1284 /*
1285 * Transfer the write-end of the channel monitoring pipe to the
1286 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
1287 */
1288 cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock);
1289 if (!cmd_socket_wrapper) {
1290 goto error;
1291 }
1292 cmd_socket_wrapper->lock = &consumer_data->lock;
1293
1294 ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper,
1295 consumer_data->channel_monitor_pipe);
1296 if (ret) {
1297 goto error;
1298 }
1299 /* Discard the socket wrapper as it is no longer needed. */
1300 consumer_destroy_socket(cmd_socket_wrapper);
1301 cmd_socket_wrapper = NULL;
1302
1303 /* The thread is completely initialized, signal that it is ready. */
1304 signal_consumer_condition(consumer_data, 1);
1305
1306 /* Infinite blocking call, waiting for transmission */
1307 restart_poll:
1308 while (1) {
1309 health_code_update();
1310
1311 /* Exit the thread because the thread quit pipe has been triggered. */
1312 if (should_quit) {
1313 /* Not a health error. */
1314 err = 0;
1315 goto exit;
1316 }
1317
1318 health_poll_entry();
1319 ret = lttng_poll_wait(&events, -1);
1320 health_poll_exit();
1321 if (ret < 0) {
1322 /*
1323 * Restart interrupted system call.
1324 */
1325 if (errno == EINTR) {
1326 goto restart_poll;
1327 }
1328 goto error;
1329 }
1330
1331 nb_fd = ret;
1332
1333 for (i = 0; i < nb_fd; i++) {
1334 /* Fetch once the poll data */
1335 revents = LTTNG_POLL_GETEV(&events, i);
1336 pollfd = LTTNG_POLL_GETFD(&events, i);
1337
1338 health_code_update();
1339
1340 if (!revents) {
1341 /* No activity for this FD (poll implementation). */
1342 continue;
1343 }
1344
1345 /*
1346 * Thread quit pipe has been triggered, flag that we should stop
1347 * but continue the current loop to handle potential data from
1348 * consumer.
1349 */
1350 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1351
1352 if (pollfd == sock) {
1353 /* Event on the consumerd socket */
1354 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1355 && !(revents & LPOLLIN)) {
1356 ERR("consumer err socket second poll error");
1357 goto error;
1358 }
1359 health_code_update();
1360 /* Wait for any kconsumerd error */
1361 ret = lttcomm_recv_unix_sock(sock, &code,
1362 sizeof(enum lttcomm_return_code));
1363 if (ret <= 0) {
1364 ERR("consumer closed the command socket");
1365 goto error;
1366 }
1367
1368 ERR("consumer return code : %s",
1369 lttcomm_get_readable_code(-code));
1370
1371 goto exit;
1372 } else if (pollfd == consumer_data->metadata_fd) {
1373 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1374 && !(revents & LPOLLIN)) {
1375 ERR("consumer err metadata socket second poll error");
1376 goto error;
1377 }
1378 /* UST metadata requests */
1379 ret = ust_consumer_metadata_request(
1380 &consumer_data->metadata_sock);
1381 if (ret < 0) {
1382 ERR("Handling metadata request");
1383 goto error;
1384 }
1385 }
1386 /* No need for an else branch all FDs are tested prior. */
1387 }
1388 health_code_update();
1389 }
1390
1391 exit:
1392 error:
1393 /*
1394 * We lock here because we are about to close the sockets and some other
1395 * thread might be using them so get exclusive access which will abort all
1396 * other consumer command by other threads.
1397 */
1398 pthread_mutex_lock(&consumer_data->lock);
1399
1400 /* Immediately set the consumerd state to stopped */
1401 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1402 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1403 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1404 consumer_data->type == LTTNG_CONSUMER32_UST) {
1405 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1406 } else {
1407 /* Code flow error... */
1408 assert(0);
1409 }
1410
1411 if (consumer_data->err_sock >= 0) {
1412 ret = close(consumer_data->err_sock);
1413 if (ret) {
1414 PERROR("close");
1415 }
1416 consumer_data->err_sock = -1;
1417 }
1418 if (consumer_data->cmd_sock >= 0) {
1419 ret = close(consumer_data->cmd_sock);
1420 if (ret) {
1421 PERROR("close");
1422 }
1423 consumer_data->cmd_sock = -1;
1424 }
1425 if (consumer_data->metadata_sock.fd_ptr &&
1426 *consumer_data->metadata_sock.fd_ptr >= 0) {
1427 ret = close(*consumer_data->metadata_sock.fd_ptr);
1428 if (ret) {
1429 PERROR("close");
1430 }
1431 }
1432 if (sock >= 0) {
1433 ret = close(sock);
1434 if (ret) {
1435 PERROR("close");
1436 }
1437 }
1438
1439 unlink(consumer_data->err_unix_sock_path);
1440 unlink(consumer_data->cmd_unix_sock_path);
1441 pthread_mutex_unlock(&consumer_data->lock);
1442
1443 /* Cleanup metadata socket mutex. */
1444 if (consumer_data->metadata_sock.lock) {
1445 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1446 free(consumer_data->metadata_sock.lock);
1447 }
1448 lttng_poll_clean(&events);
1449
1450 if (cmd_socket_wrapper) {
1451 consumer_destroy_socket(cmd_socket_wrapper);
1452 }
1453 error_poll:
1454 if (err) {
1455 health_error();
1456 ERR("Health error occurred in %s", __func__);
1457 }
1458 health_unregister(health_sessiond);
1459 DBG("consumer thread cleanup completed");
1460
1461 rcu_thread_offline();
1462 rcu_unregister_thread();
1463
1464 return NULL;
1465 }
1466
1467 /*
1468 * This thread manage application communication.
1469 */
1470 static void *thread_manage_apps(void *data)
1471 {
1472 int i, ret, pollfd, err = -1;
1473 ssize_t size_ret;
1474 uint32_t revents, nb_fd;
1475 struct lttng_poll_event events;
1476
1477 DBG("[thread] Manage application started");
1478
1479 rcu_register_thread();
1480 rcu_thread_online();
1481
1482 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1483
1484 if (testpoint(sessiond_thread_manage_apps)) {
1485 goto error_testpoint;
1486 }
1487
1488 health_code_update();
1489
1490 ret = sessiond_set_thread_pollset(&events, 2);
1491 if (ret < 0) {
1492 goto error_poll_create;
1493 }
1494
1495 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1496 if (ret < 0) {
1497 goto error;
1498 }
1499
1500 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1501 goto error;
1502 }
1503
1504 health_code_update();
1505
1506 while (1) {
1507 DBG("Apps thread polling");
1508
1509 /* Inifinite blocking call, waiting for transmission */
1510 restart:
1511 health_poll_entry();
1512 ret = lttng_poll_wait(&events, -1);
1513 DBG("Apps thread return from poll on %d fds",
1514 LTTNG_POLL_GETNB(&events));
1515 health_poll_exit();
1516 if (ret < 0) {
1517 /*
1518 * Restart interrupted system call.
1519 */
1520 if (errno == EINTR) {
1521 goto restart;
1522 }
1523 goto error;
1524 }
1525
1526 nb_fd = ret;
1527
1528 for (i = 0; i < nb_fd; i++) {
1529 /* Fetch once the poll data */
1530 revents = LTTNG_POLL_GETEV(&events, i);
1531 pollfd = LTTNG_POLL_GETFD(&events, i);
1532
1533 health_code_update();
1534
1535 if (!revents) {
1536 /* No activity for this FD (poll implementation). */
1537 continue;
1538 }
1539
1540 /* Thread quit pipe has been closed. Killing thread. */
1541 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1542 if (ret) {
1543 err = 0;
1544 goto exit;
1545 }
1546
1547 /* Inspect the apps cmd pipe */
1548 if (pollfd == apps_cmd_pipe[0]) {
1549 if (revents & LPOLLIN) {
1550 int sock;
1551
1552 /* Empty pipe */
1553 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1554 if (size_ret < sizeof(sock)) {
1555 PERROR("read apps cmd pipe");
1556 goto error;
1557 }
1558
1559 health_code_update();
1560
1561 /*
1562 * Since this is a command socket (write then read),
1563 * we only monitor the error events of the socket.
1564 */
1565 ret = lttng_poll_add(&events, sock,
1566 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1567 if (ret < 0) {
1568 goto error;
1569 }
1570
1571 DBG("Apps with sock %d added to poll set", sock);
1572 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1573 ERR("Apps command pipe error");
1574 goto error;
1575 } else {
1576 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1577 goto error;
1578 }
1579 } else {
1580 /*
1581 * At this point, we know that a registered application made
1582 * the event at poll_wait.
1583 */
1584 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1585 /* Removing from the poll set */
1586 ret = lttng_poll_del(&events, pollfd);
1587 if (ret < 0) {
1588 goto error;
1589 }
1590
1591 /* Socket closed on remote end. */
1592 ust_app_unregister(pollfd);
1593 } else {
1594 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1595 goto error;
1596 }
1597 }
1598
1599 health_code_update();
1600 }
1601 }
1602
1603 exit:
1604 error:
1605 lttng_poll_clean(&events);
1606 error_poll_create:
1607 error_testpoint:
1608 utils_close_pipe(apps_cmd_pipe);
1609 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1610
1611 /*
1612 * We don't clean the UST app hash table here since already registered
1613 * applications can still be controlled so let them be until the session
1614 * daemon dies or the applications stop.
1615 */
1616
1617 if (err) {
1618 health_error();
1619 ERR("Health error occurred in %s", __func__);
1620 }
1621 health_unregister(health_sessiond);
1622 DBG("Application communication apps thread cleanup complete");
1623 rcu_thread_offline();
1624 rcu_unregister_thread();
1625 return NULL;
1626 }
1627
1628 /*
1629 * Send a socket to a thread This is called from the dispatch UST registration
1630 * thread once all sockets are set for the application.
1631 *
1632 * The sock value can be invalid, we don't really care, the thread will handle
1633 * it and make the necessary cleanup if so.
1634 *
1635 * On success, return 0 else a negative value being the errno message of the
1636 * write().
1637 */
1638 static int send_socket_to_thread(int fd, int sock)
1639 {
1640 ssize_t ret;
1641
1642 /*
1643 * It's possible that the FD is set as invalid with -1 concurrently just
1644 * before calling this function being a shutdown state of the thread.
1645 */
1646 if (fd < 0) {
1647 ret = -EBADF;
1648 goto error;
1649 }
1650
1651 ret = lttng_write(fd, &sock, sizeof(sock));
1652 if (ret < sizeof(sock)) {
1653 PERROR("write apps pipe %d", fd);
1654 if (ret < 0) {
1655 ret = -errno;
1656 }
1657 goto error;
1658 }
1659
1660 /* All good. Don't send back the write positive ret value. */
1661 ret = 0;
1662 error:
1663 return (int) ret;
1664 }
1665
1666 /*
1667 * Sanitize the wait queue of the dispatch registration thread meaning removing
1668 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1669 * notify socket is never received.
1670 */
1671 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1672 {
1673 int ret, nb_fd = 0, i;
1674 unsigned int fd_added = 0;
1675 struct lttng_poll_event events;
1676 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1677
1678 assert(wait_queue);
1679
1680 lttng_poll_init(&events);
1681
1682 /* Just skip everything for an empty queue. */
1683 if (!wait_queue->count) {
1684 goto end;
1685 }
1686
1687 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1688 if (ret < 0) {
1689 goto error_create;
1690 }
1691
1692 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1693 &wait_queue->head, head) {
1694 assert(wait_node->app);
1695 ret = lttng_poll_add(&events, wait_node->app->sock,
1696 LPOLLHUP | LPOLLERR);
1697 if (ret < 0) {
1698 goto error;
1699 }
1700
1701 fd_added = 1;
1702 }
1703
1704 if (!fd_added) {
1705 goto end;
1706 }
1707
1708 /*
1709 * Poll but don't block so we can quickly identify the faulty events and
1710 * clean them afterwards from the wait queue.
1711 */
1712 ret = lttng_poll_wait(&events, 0);
1713 if (ret < 0) {
1714 goto error;
1715 }
1716 nb_fd = ret;
1717
1718 for (i = 0; i < nb_fd; i++) {
1719 /* Get faulty FD. */
1720 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1721 int pollfd = LTTNG_POLL_GETFD(&events, i);
1722
1723 if (!revents) {
1724 /* No activity for this FD (poll implementation). */
1725 continue;
1726 }
1727
1728 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1729 &wait_queue->head, head) {
1730 if (pollfd == wait_node->app->sock &&
1731 (revents & (LPOLLHUP | LPOLLERR))) {
1732 cds_list_del(&wait_node->head);
1733 wait_queue->count--;
1734 ust_app_destroy(wait_node->app);
1735 free(wait_node);
1736 /*
1737 * Silence warning of use-after-free in
1738 * cds_list_for_each_entry_safe which uses
1739 * __typeof__(*wait_node).
1740 */
1741 wait_node = NULL;
1742 break;
1743 } else {
1744 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1745 goto error;
1746 }
1747 }
1748 }
1749
1750 if (nb_fd > 0) {
1751 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1752 }
1753
1754 end:
1755 lttng_poll_clean(&events);
1756 return;
1757
1758 error:
1759 lttng_poll_clean(&events);
1760 error_create:
1761 ERR("Unable to sanitize wait queue");
1762 return;
1763 }
1764
1765 /*
1766 * Dispatch request from the registration threads to the application
1767 * communication thread.
1768 */
1769 static void *thread_dispatch_ust_registration(void *data)
1770 {
1771 int ret, err = -1;
1772 struct cds_wfcq_node *node;
1773 struct ust_command *ust_cmd = NULL;
1774 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1775 struct ust_reg_wait_queue wait_queue = {
1776 .count = 0,
1777 };
1778
1779 rcu_register_thread();
1780
1781 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1782
1783 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1784 goto error_testpoint;
1785 }
1786
1787 health_code_update();
1788
1789 CDS_INIT_LIST_HEAD(&wait_queue.head);
1790
1791 DBG("[thread] Dispatch UST command started");
1792
1793 for (;;) {
1794 health_code_update();
1795
1796 /* Atomically prepare the queue futex */
1797 futex_nto1_prepare(&ust_cmd_queue.futex);
1798
1799 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1800 break;
1801 }
1802
1803 do {
1804 struct ust_app *app = NULL;
1805 ust_cmd = NULL;
1806
1807 /*
1808 * Make sure we don't have node(s) that have hung up before receiving
1809 * the notify socket. This is to clean the list in order to avoid
1810 * memory leaks from notify socket that are never seen.
1811 */
1812 sanitize_wait_queue(&wait_queue);
1813
1814 health_code_update();
1815 /* Dequeue command for registration */
1816 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1817 if (node == NULL) {
1818 DBG("Woken up but nothing in the UST command queue");
1819 /* Continue thread execution */
1820 break;
1821 }
1822
1823 ust_cmd = caa_container_of(node, struct ust_command, node);
1824
1825 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1826 " gid:%d sock:%d name:%s (version %d.%d)",
1827 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1828 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1829 ust_cmd->sock, ust_cmd->reg_msg.name,
1830 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1831
1832 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1833 wait_node = zmalloc(sizeof(*wait_node));
1834 if (!wait_node) {
1835 PERROR("zmalloc wait_node dispatch");
1836 ret = close(ust_cmd->sock);
1837 if (ret < 0) {
1838 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1839 }
1840 lttng_fd_put(LTTNG_FD_APPS, 1);
1841 free(ust_cmd);
1842 goto error;
1843 }
1844 CDS_INIT_LIST_HEAD(&wait_node->head);
1845
1846 /* Create application object if socket is CMD. */
1847 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1848 ust_cmd->sock);
1849 if (!wait_node->app) {
1850 ret = close(ust_cmd->sock);
1851 if (ret < 0) {
1852 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1853 }
1854 lttng_fd_put(LTTNG_FD_APPS, 1);
1855 free(wait_node);
1856 free(ust_cmd);
1857 continue;
1858 }
1859 /*
1860 * Add application to the wait queue so we can set the notify
1861 * socket before putting this object in the global ht.
1862 */
1863 cds_list_add(&wait_node->head, &wait_queue.head);
1864 wait_queue.count++;
1865
1866 free(ust_cmd);
1867 /*
1868 * We have to continue here since we don't have the notify
1869 * socket and the application MUST be added to the hash table
1870 * only at that moment.
1871 */
1872 continue;
1873 } else {
1874 /*
1875 * Look for the application in the local wait queue and set the
1876 * notify socket if found.
1877 */
1878 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1879 &wait_queue.head, head) {
1880 health_code_update();
1881 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1882 wait_node->app->notify_sock = ust_cmd->sock;
1883 cds_list_del(&wait_node->head);
1884 wait_queue.count--;
1885 app = wait_node->app;
1886 free(wait_node);
1887 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1888 break;
1889 }
1890 }
1891
1892 /*
1893 * With no application at this stage the received socket is
1894 * basically useless so close it before we free the cmd data
1895 * structure for good.
1896 */
1897 if (!app) {
1898 ret = close(ust_cmd->sock);
1899 if (ret < 0) {
1900 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1901 }
1902 lttng_fd_put(LTTNG_FD_APPS, 1);
1903 }
1904 free(ust_cmd);
1905 }
1906
1907 if (app) {
1908 /*
1909 * @session_lock_list
1910 *
1911 * Lock the global session list so from the register up to the
1912 * registration done message, no thread can see the application
1913 * and change its state.
1914 */
1915 session_lock_list();
1916 rcu_read_lock();
1917
1918 /*
1919 * Add application to the global hash table. This needs to be
1920 * done before the update to the UST registry can locate the
1921 * application.
1922 */
1923 ust_app_add(app);
1924
1925 /* Set app version. This call will print an error if needed. */
1926 (void) ust_app_version(app);
1927
1928 /* Send notify socket through the notify pipe. */
1929 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1930 app->notify_sock);
1931 if (ret < 0) {
1932 rcu_read_unlock();
1933 session_unlock_list();
1934 /*
1935 * No notify thread, stop the UST tracing. However, this is
1936 * not an internal error of the this thread thus setting
1937 * the health error code to a normal exit.
1938 */
1939 err = 0;
1940 goto error;
1941 }
1942
1943 /*
1944 * Update newly registered application with the tracing
1945 * registry info already enabled information.
1946 */
1947 update_ust_app(app->sock);
1948
1949 /*
1950 * Don't care about return value. Let the manage apps threads
1951 * handle app unregistration upon socket close.
1952 */
1953 (void) ust_app_register_done(app);
1954
1955 /*
1956 * Even if the application socket has been closed, send the app
1957 * to the thread and unregistration will take place at that
1958 * place.
1959 */
1960 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1961 if (ret < 0) {
1962 rcu_read_unlock();
1963 session_unlock_list();
1964 /*
1965 * No apps. thread, stop the UST tracing. However, this is
1966 * not an internal error of the this thread thus setting
1967 * the health error code to a normal exit.
1968 */
1969 err = 0;
1970 goto error;
1971 }
1972
1973 rcu_read_unlock();
1974 session_unlock_list();
1975 }
1976 } while (node != NULL);
1977
1978 health_poll_entry();
1979 /* Futex wait on queue. Blocking call on futex() */
1980 futex_nto1_wait(&ust_cmd_queue.futex);
1981 health_poll_exit();
1982 }
1983 /* Normal exit, no error */
1984 err = 0;
1985
1986 error:
1987 /* Clean up wait queue. */
1988 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1989 &wait_queue.head, head) {
1990 cds_list_del(&wait_node->head);
1991 wait_queue.count--;
1992 free(wait_node);
1993 }
1994
1995 /* Empty command queue. */
1996 for (;;) {
1997 /* Dequeue command for registration */
1998 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1999 if (node == NULL) {
2000 break;
2001 }
2002 ust_cmd = caa_container_of(node, struct ust_command, node);
2003 ret = close(ust_cmd->sock);
2004 if (ret < 0) {
2005 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2006 }
2007 lttng_fd_put(LTTNG_FD_APPS, 1);
2008 free(ust_cmd);
2009 }
2010
2011 error_testpoint:
2012 DBG("Dispatch thread dying");
2013 if (err) {
2014 health_error();
2015 ERR("Health error occurred in %s", __func__);
2016 }
2017 health_unregister(health_sessiond);
2018 rcu_unregister_thread();
2019 return NULL;
2020 }
2021
2022 /*
2023 * This thread manage application registration.
2024 */
2025 static void *thread_registration_apps(void *data)
2026 {
2027 int sock = -1, i, ret, pollfd, err = -1;
2028 uint32_t revents, nb_fd;
2029 struct lttng_poll_event events;
2030 /*
2031 * Get allocated in this thread, enqueued to a global queue, dequeued and
2032 * freed in the manage apps thread.
2033 */
2034 struct ust_command *ust_cmd = NULL;
2035
2036 DBG("[thread] Manage application registration started");
2037
2038 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2039
2040 if (testpoint(sessiond_thread_registration_apps)) {
2041 goto error_testpoint;
2042 }
2043
2044 ret = lttcomm_listen_unix_sock(apps_sock);
2045 if (ret < 0) {
2046 goto error_listen;
2047 }
2048
2049 /*
2050 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2051 * more will be added to this poll set.
2052 */
2053 ret = sessiond_set_thread_pollset(&events, 2);
2054 if (ret < 0) {
2055 goto error_create_poll;
2056 }
2057
2058 /* Add the application registration socket */
2059 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2060 if (ret < 0) {
2061 goto error_poll_add;
2062 }
2063
2064 /* Notify all applications to register */
2065 ret = notify_ust_apps(1);
2066 if (ret < 0) {
2067 ERR("Failed to notify applications or create the wait shared memory.\n"
2068 "Execution continues but there might be problem for already\n"
2069 "running applications that wishes to register.");
2070 }
2071
2072 while (1) {
2073 DBG("Accepting application registration");
2074
2075 /* Inifinite blocking call, waiting for transmission */
2076 restart:
2077 health_poll_entry();
2078 ret = lttng_poll_wait(&events, -1);
2079 health_poll_exit();
2080 if (ret < 0) {
2081 /*
2082 * Restart interrupted system call.
2083 */
2084 if (errno == EINTR) {
2085 goto restart;
2086 }
2087 goto error;
2088 }
2089
2090 nb_fd = ret;
2091
2092 for (i = 0; i < nb_fd; i++) {
2093 health_code_update();
2094
2095 /* Fetch once the poll data */
2096 revents = LTTNG_POLL_GETEV(&events, i);
2097 pollfd = LTTNG_POLL_GETFD(&events, i);
2098
2099 if (!revents) {
2100 /* No activity for this FD (poll implementation). */
2101 continue;
2102 }
2103
2104 /* Thread quit pipe has been closed. Killing thread. */
2105 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2106 if (ret) {
2107 err = 0;
2108 goto exit;
2109 }
2110
2111 /* Event on the registration socket */
2112 if (pollfd == apps_sock) {
2113 if (revents & LPOLLIN) {
2114 sock = lttcomm_accept_unix_sock(apps_sock);
2115 if (sock < 0) {
2116 goto error;
2117 }
2118
2119 /*
2120 * Set socket timeout for both receiving and ending.
2121 * app_socket_timeout is in seconds, whereas
2122 * lttcomm_setsockopt_rcv_timeout and
2123 * lttcomm_setsockopt_snd_timeout expect msec as
2124 * parameter.
2125 */
2126 if (config.app_socket_timeout >= 0) {
2127 (void) lttcomm_setsockopt_rcv_timeout(sock,
2128 config.app_socket_timeout * 1000);
2129 (void) lttcomm_setsockopt_snd_timeout(sock,
2130 config.app_socket_timeout * 1000);
2131 }
2132
2133 /*
2134 * Set the CLOEXEC flag. Return code is useless because
2135 * either way, the show must go on.
2136 */
2137 (void) utils_set_fd_cloexec(sock);
2138
2139 /* Create UST registration command for enqueuing */
2140 ust_cmd = zmalloc(sizeof(struct ust_command));
2141 if (ust_cmd == NULL) {
2142 PERROR("ust command zmalloc");
2143 ret = close(sock);
2144 if (ret) {
2145 PERROR("close");
2146 }
2147 goto error;
2148 }
2149
2150 /*
2151 * Using message-based transmissions to ensure we don't
2152 * have to deal with partially received messages.
2153 */
2154 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2155 if (ret < 0) {
2156 ERR("Exhausted file descriptors allowed for applications.");
2157 free(ust_cmd);
2158 ret = close(sock);
2159 if (ret) {
2160 PERROR("close");
2161 }
2162 sock = -1;
2163 continue;
2164 }
2165
2166 health_code_update();
2167 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2168 if (ret < 0) {
2169 free(ust_cmd);
2170 /* Close socket of the application. */
2171 ret = close(sock);
2172 if (ret) {
2173 PERROR("close");
2174 }
2175 lttng_fd_put(LTTNG_FD_APPS, 1);
2176 sock = -1;
2177 continue;
2178 }
2179 health_code_update();
2180
2181 ust_cmd->sock = sock;
2182 sock = -1;
2183
2184 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2185 " gid:%d sock:%d name:%s (version %d.%d)",
2186 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2187 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2188 ust_cmd->sock, ust_cmd->reg_msg.name,
2189 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2190
2191 /*
2192 * Lock free enqueue the registration request. The red pill
2193 * has been taken! This apps will be part of the *system*.
2194 */
2195 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2196
2197 /*
2198 * Wake the registration queue futex. Implicit memory
2199 * barrier with the exchange in cds_wfcq_enqueue.
2200 */
2201 futex_nto1_wake(&ust_cmd_queue.futex);
2202 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2203 ERR("Register apps socket poll error");
2204 goto error;
2205 } else {
2206 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2207 goto error;
2208 }
2209 }
2210 }
2211 }
2212
2213 exit:
2214 error:
2215 /* Notify that the registration thread is gone */
2216 notify_ust_apps(0);
2217
2218 if (apps_sock >= 0) {
2219 ret = close(apps_sock);
2220 if (ret) {
2221 PERROR("close");
2222 }
2223 }
2224 if (sock >= 0) {
2225 ret = close(sock);
2226 if (ret) {
2227 PERROR("close");
2228 }
2229 lttng_fd_put(LTTNG_FD_APPS, 1);
2230 }
2231 unlink(config.apps_unix_sock_path.value);
2232
2233 error_poll_add:
2234 lttng_poll_clean(&events);
2235 error_listen:
2236 error_create_poll:
2237 error_testpoint:
2238 DBG("UST Registration thread cleanup complete");
2239 if (err) {
2240 health_error();
2241 ERR("Health error occurred in %s", __func__);
2242 }
2243 health_unregister(health_sessiond);
2244
2245 return NULL;
2246 }
2247
2248 /*
2249 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2250 * exec or it will fails.
2251 */
2252 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2253 {
2254 int ret, clock_ret;
2255 struct timespec timeout;
2256
2257 /*
2258 * Make sure we set the readiness flag to 0 because we are NOT ready.
2259 * This access to consumer_thread_is_ready does not need to be
2260 * protected by consumer_data.cond_mutex (yet) since the consumer
2261 * management thread has not been started at this point.
2262 */
2263 consumer_data->consumer_thread_is_ready = 0;
2264
2265 /* Setup pthread condition */
2266 ret = pthread_condattr_init(&consumer_data->condattr);
2267 if (ret) {
2268 errno = ret;
2269 PERROR("pthread_condattr_init consumer data");
2270 goto error;
2271 }
2272
2273 /*
2274 * Set the monotonic clock in order to make sure we DO NOT jump in time
2275 * between the clock_gettime() call and the timedwait call. See bug #324
2276 * for a more details and how we noticed it.
2277 */
2278 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2279 if (ret) {
2280 errno = ret;
2281 PERROR("pthread_condattr_setclock consumer data");
2282 goto error;
2283 }
2284
2285 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2286 if (ret) {
2287 errno = ret;
2288 PERROR("pthread_cond_init consumer data");
2289 goto error;
2290 }
2291
2292 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2293 thread_manage_consumer, consumer_data);
2294 if (ret) {
2295 errno = ret;
2296 PERROR("pthread_create consumer");
2297 ret = -1;
2298 goto error;
2299 }
2300
2301 /* We are about to wait on a pthread condition */
2302 pthread_mutex_lock(&consumer_data->cond_mutex);
2303
2304 /* Get time for sem_timedwait absolute timeout */
2305 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2306 /*
2307 * Set the timeout for the condition timed wait even if the clock gettime
2308 * call fails since we might loop on that call and we want to avoid to
2309 * increment the timeout too many times.
2310 */
2311 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2312
2313 /*
2314 * The following loop COULD be skipped in some conditions so this is why we
2315 * set ret to 0 in order to make sure at least one round of the loop is
2316 * done.
2317 */
2318 ret = 0;
2319
2320 /*
2321 * Loop until the condition is reached or when a timeout is reached. Note
2322 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2323 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2324 * possible. This loop does not take any chances and works with both of
2325 * them.
2326 */
2327 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2328 if (clock_ret < 0) {
2329 PERROR("clock_gettime spawn consumer");
2330 /* Infinite wait for the consumerd thread to be ready */
2331 ret = pthread_cond_wait(&consumer_data->cond,
2332 &consumer_data->cond_mutex);
2333 } else {
2334 ret = pthread_cond_timedwait(&consumer_data->cond,
2335 &consumer_data->cond_mutex, &timeout);
2336 }
2337 }
2338
2339 /* Release the pthread condition */
2340 pthread_mutex_unlock(&consumer_data->cond_mutex);
2341
2342 if (ret != 0) {
2343 errno = ret;
2344 if (ret == ETIMEDOUT) {
2345 int pth_ret;
2346
2347 /*
2348 * Call has timed out so we kill the kconsumerd_thread and return
2349 * an error.
2350 */
2351 ERR("Condition timed out. The consumer thread was never ready."
2352 " Killing it");
2353 pth_ret = pthread_cancel(consumer_data->thread);
2354 if (pth_ret < 0) {
2355 PERROR("pthread_cancel consumer thread");
2356 }
2357 } else {
2358 PERROR("pthread_cond_wait failed consumer thread");
2359 }
2360 /* Caller is expecting a negative value on failure. */
2361 ret = -1;
2362 goto error;
2363 }
2364
2365 pthread_mutex_lock(&consumer_data->pid_mutex);
2366 if (consumer_data->pid == 0) {
2367 ERR("Consumerd did not start");
2368 pthread_mutex_unlock(&consumer_data->pid_mutex);
2369 goto error;
2370 }
2371 pthread_mutex_unlock(&consumer_data->pid_mutex);
2372
2373 return 0;
2374
2375 error:
2376 return ret;
2377 }
2378
2379 /*
2380 * Join consumer thread
2381 */
2382 static int join_consumer_thread(struct consumer_data *consumer_data)
2383 {
2384 void *status;
2385
2386 /* Consumer pid must be a real one. */
2387 if (consumer_data->pid > 0) {
2388 int ret;
2389 ret = kill(consumer_data->pid, SIGTERM);
2390 if (ret) {
2391 PERROR("Error killing consumer daemon");
2392 return ret;
2393 }
2394 return pthread_join(consumer_data->thread, &status);
2395 } else {
2396 return 0;
2397 }
2398 }
2399
2400 /*
2401 * Fork and exec a consumer daemon (consumerd).
2402 *
2403 * Return pid if successful else -1.
2404 */
2405 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2406 {
2407 int ret;
2408 pid_t pid;
2409 const char *consumer_to_use;
2410 const char *verbosity;
2411 struct stat st;
2412
2413 DBG("Spawning consumerd");
2414
2415 pid = fork();
2416 if (pid == 0) {
2417 /*
2418 * Exec consumerd.
2419 */
2420 if (config.verbose_consumer) {
2421 verbosity = "--verbose";
2422 } else if (lttng_opt_quiet) {
2423 verbosity = "--quiet";
2424 } else {
2425 verbosity = "";
2426 }
2427
2428 switch (consumer_data->type) {
2429 case LTTNG_CONSUMER_KERNEL:
2430 /*
2431 * Find out which consumerd to execute. We will first try the
2432 * 64-bit path, then the sessiond's installation directory, and
2433 * fallback on the 32-bit one,
2434 */
2435 DBG3("Looking for a kernel consumer at these locations:");
2436 DBG3(" 1) %s", config.consumerd64_bin_path.value ? : "NULL");
2437 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, DEFAULT_CONSUMERD_FILE);
2438 DBG3(" 3) %s", config.consumerd32_bin_path.value ? : "NULL");
2439 if (stat(config.consumerd64_bin_path.value, &st) == 0) {
2440 DBG3("Found location #1");
2441 consumer_to_use = config.consumerd64_bin_path.value;
2442 } else if (stat(INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE, &st) == 0) {
2443 DBG3("Found location #2");
2444 consumer_to_use = INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE;
2445 } else if (stat(config.consumerd32_bin_path.value, &st) == 0) {
2446 DBG3("Found location #3");
2447 consumer_to_use = config.consumerd32_bin_path.value;
2448 } else {
2449 DBG("Could not find any valid consumerd executable");
2450 ret = -EINVAL;
2451 goto error;
2452 }
2453 DBG("Using kernel consumer at: %s", consumer_to_use);
2454 (void) execl(consumer_to_use,
2455 "lttng-consumerd", verbosity, "-k",
2456 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2457 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2458 "--group", config.tracing_group_name.value,
2459 NULL);
2460 break;
2461 case LTTNG_CONSUMER64_UST:
2462 {
2463 char *tmpnew = NULL;
2464
2465 if (config.consumerd64_lib_dir.value) {
2466 char *tmp;
2467 size_t tmplen;
2468
2469 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2470 if (!tmp) {
2471 tmp = "";
2472 }
2473 tmplen = strlen(config.consumerd64_lib_dir.value) + 1 /* : */ + strlen(tmp);
2474 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2475 if (!tmpnew) {
2476 ret = -ENOMEM;
2477 goto error;
2478 }
2479 strcat(tmpnew, config.consumerd64_lib_dir.value);
2480 if (tmp[0] != '\0') {
2481 strcat(tmpnew, ":");
2482 strcat(tmpnew, tmp);
2483 }
2484 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2485 if (ret) {
2486 ret = -errno;
2487 free(tmpnew);
2488 goto error;
2489 }
2490 }
2491 DBG("Using 64-bit UST consumer at: %s", config.consumerd64_bin_path.value);
2492 (void) execl(config.consumerd64_bin_path.value, "lttng-consumerd", verbosity, "-u",
2493 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2494 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2495 "--group", config.tracing_group_name.value,
2496 NULL);
2497 free(tmpnew);
2498 break;
2499 }
2500 case LTTNG_CONSUMER32_UST:
2501 {
2502 char *tmpnew = NULL;
2503
2504 if (config.consumerd32_lib_dir.value) {
2505 char *tmp;
2506 size_t tmplen;
2507
2508 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2509 if (!tmp) {
2510 tmp = "";
2511 }
2512 tmplen = strlen(config.consumerd32_lib_dir.value) + 1 /* : */ + strlen(tmp);
2513 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2514 if (!tmpnew) {
2515 ret = -ENOMEM;
2516 goto error;
2517 }
2518 strcat(tmpnew, config.consumerd32_lib_dir.value);
2519 if (tmp[0] != '\0') {
2520 strcat(tmpnew, ":");
2521 strcat(tmpnew, tmp);
2522 }
2523 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2524 if (ret) {
2525 ret = -errno;
2526 free(tmpnew);
2527 goto error;
2528 }
2529 }
2530 DBG("Using 32-bit UST consumer at: %s", config.consumerd32_bin_path.value);
2531 (void) execl(config.consumerd32_bin_path.value, "lttng-consumerd", verbosity, "-u",
2532 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2533 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2534 "--group", config.tracing_group_name.value,
2535 NULL);
2536 free(tmpnew);
2537 break;
2538 }
2539 default:
2540 ERR("unknown consumer type");
2541 exit(EXIT_FAILURE);
2542 }
2543 if (errno != 0) {
2544 PERROR("Consumer execl()");
2545 }
2546 /* Reaching this point, we got a failure on our execl(). */
2547 exit(EXIT_FAILURE);
2548 } else if (pid > 0) {
2549 ret = pid;
2550 } else {
2551 PERROR("start consumer fork");
2552 ret = -errno;
2553 }
2554 error:
2555 return ret;
2556 }
2557
2558 /*
2559 * Spawn the consumerd daemon and session daemon thread.
2560 */
2561 static int start_consumerd(struct consumer_data *consumer_data)
2562 {
2563 int ret;
2564
2565 /*
2566 * Set the listen() state on the socket since there is a possible race
2567 * between the exec() of the consumer daemon and this call if place in the
2568 * consumer thread. See bug #366 for more details.
2569 */
2570 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2571 if (ret < 0) {
2572 goto error;
2573 }
2574
2575 pthread_mutex_lock(&consumer_data->pid_mutex);
2576 if (consumer_data->pid != 0) {
2577 pthread_mutex_unlock(&consumer_data->pid_mutex);
2578 goto end;
2579 }
2580
2581 ret = spawn_consumerd(consumer_data);
2582 if (ret < 0) {
2583 ERR("Spawning consumerd failed");
2584 pthread_mutex_unlock(&consumer_data->pid_mutex);
2585 goto error;
2586 }
2587
2588 /* Setting up the consumer_data pid */
2589 consumer_data->pid = ret;
2590 DBG2("Consumer pid %d", consumer_data->pid);
2591 pthread_mutex_unlock(&consumer_data->pid_mutex);
2592
2593 DBG2("Spawning consumer control thread");
2594 ret = spawn_consumer_thread(consumer_data);
2595 if (ret < 0) {
2596 ERR("Fatal error spawning consumer control thread");
2597 goto error;
2598 }
2599
2600 end:
2601 return 0;
2602
2603 error:
2604 /* Cleanup already created sockets on error. */
2605 if (consumer_data->err_sock >= 0) {
2606 int err;
2607
2608 err = close(consumer_data->err_sock);
2609 if (err < 0) {
2610 PERROR("close consumer data error socket");
2611 }
2612 }
2613 return ret;
2614 }
2615
2616 /*
2617 * Setup necessary data for kernel tracer action.
2618 */
2619 static int init_kernel_tracer(void)
2620 {
2621 int ret;
2622
2623 /* Modprobe lttng kernel modules */
2624 ret = modprobe_lttng_control();
2625 if (ret < 0) {
2626 goto error;
2627 }
2628
2629 /* Open debugfs lttng */
2630 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2631 if (kernel_tracer_fd < 0) {
2632 DBG("Failed to open %s", module_proc_lttng);
2633 goto error_open;
2634 }
2635
2636 /* Validate kernel version */
2637 ret = kernel_validate_version(kernel_tracer_fd);
2638 if (ret < 0) {
2639 goto error_version;
2640 }
2641
2642 ret = modprobe_lttng_data();
2643 if (ret < 0) {
2644 goto error_modules;
2645 }
2646
2647 ret = kernel_supports_ring_buffer_snapshot_sample_positions(
2648 kernel_tracer_fd);
2649 if (ret < 0) {
2650 goto error_modules;
2651 }
2652
2653 if (ret < 1) {
2654 WARN("Kernel tracer does not support buffer monitoring. "
2655 "The monitoring timer of channels in the kernel domain "
2656 "will be set to 0 (disabled).");
2657 }
2658
2659 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2660 return 0;
2661
2662 error_version:
2663 modprobe_remove_lttng_control();
2664 ret = close(kernel_tracer_fd);
2665 if (ret) {
2666 PERROR("close");
2667 }
2668 kernel_tracer_fd = -1;
2669 return LTTNG_ERR_KERN_VERSION;
2670
2671 error_modules:
2672 ret = close(kernel_tracer_fd);
2673 if (ret) {
2674 PERROR("close");
2675 }
2676
2677 error_open:
2678 modprobe_remove_lttng_control();
2679
2680 error:
2681 WARN("No kernel tracer available");
2682 kernel_tracer_fd = -1;
2683 if (!is_root) {
2684 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2685 } else {
2686 return LTTNG_ERR_KERN_NA;
2687 }
2688 }
2689
2690
2691 /*
2692 * Copy consumer output from the tracing session to the domain session. The
2693 * function also applies the right modification on a per domain basis for the
2694 * trace files destination directory.
2695 *
2696 * Should *NOT* be called with RCU read-side lock held.
2697 */
2698 static int copy_session_consumer(int domain, struct ltt_session *session)
2699 {
2700 int ret;
2701 const char *dir_name;
2702 struct consumer_output *consumer;
2703
2704 assert(session);
2705 assert(session->consumer);
2706
2707 switch (domain) {
2708 case LTTNG_DOMAIN_KERNEL:
2709 DBG3("Copying tracing session consumer output in kernel session");
2710 /*
2711 * XXX: We should audit the session creation and what this function
2712 * does "extra" in order to avoid a destroy since this function is used
2713 * in the domain session creation (kernel and ust) only. Same for UST
2714 * domain.
2715 */
2716 if (session->kernel_session->consumer) {
2717 consumer_output_put(session->kernel_session->consumer);
2718 }
2719 session->kernel_session->consumer =
2720 consumer_copy_output(session->consumer);
2721 /* Ease our life a bit for the next part */
2722 consumer = session->kernel_session->consumer;
2723 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2724 break;
2725 case LTTNG_DOMAIN_JUL:
2726 case LTTNG_DOMAIN_LOG4J:
2727 case LTTNG_DOMAIN_PYTHON:
2728 case LTTNG_DOMAIN_UST:
2729 DBG3("Copying tracing session consumer output in UST session");
2730 if (session->ust_session->consumer) {
2731 consumer_output_put(session->ust_session->consumer);
2732 }
2733 session->ust_session->consumer =
2734 consumer_copy_output(session->consumer);
2735 /* Ease our life a bit for the next part */
2736 consumer = session->ust_session->consumer;
2737 dir_name = DEFAULT_UST_TRACE_DIR;
2738 break;
2739 default:
2740 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2741 goto error;
2742 }
2743
2744 /* Append correct directory to subdir */
2745 strncat(consumer->subdir, dir_name,
2746 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2747 DBG3("Copy session consumer subdir %s", consumer->subdir);
2748
2749 ret = LTTNG_OK;
2750
2751 error:
2752 return ret;
2753 }
2754
2755 /*
2756 * Create an UST session and add it to the session ust list.
2757 *
2758 * Should *NOT* be called with RCU read-side lock held.
2759 */
2760 static int create_ust_session(struct ltt_session *session,
2761 struct lttng_domain *domain)
2762 {
2763 int ret;
2764 struct ltt_ust_session *lus = NULL;
2765
2766 assert(session);
2767 assert(domain);
2768 assert(session->consumer);
2769
2770 switch (domain->type) {
2771 case LTTNG_DOMAIN_JUL:
2772 case LTTNG_DOMAIN_LOG4J:
2773 case LTTNG_DOMAIN_PYTHON:
2774 case LTTNG_DOMAIN_UST:
2775 break;
2776 default:
2777 ERR("Unknown UST domain on create session %d", domain->type);
2778 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2779 goto error;
2780 }
2781
2782 DBG("Creating UST session");
2783
2784 lus = trace_ust_create_session(session->id);
2785 if (lus == NULL) {
2786 ret = LTTNG_ERR_UST_SESS_FAIL;
2787 goto error;
2788 }
2789
2790 lus->uid = session->uid;
2791 lus->gid = session->gid;
2792 lus->output_traces = session->output_traces;
2793 lus->snapshot_mode = session->snapshot_mode;
2794 lus->live_timer_interval = session->live_timer;
2795 session->ust_session = lus;
2796 if (session->shm_path[0]) {
2797 strncpy(lus->root_shm_path, session->shm_path,
2798 sizeof(lus->root_shm_path));
2799 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2800 strncpy(lus->shm_path, session->shm_path,
2801 sizeof(lus->shm_path));
2802 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2803 strncat(lus->shm_path, "/ust",
2804 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2805 }
2806 /* Copy session output to the newly created UST session */
2807 ret = copy_session_consumer(domain->type, session);
2808 if (ret != LTTNG_OK) {
2809 goto error;
2810 }
2811
2812 return LTTNG_OK;
2813
2814 error:
2815 free(lus);
2816 session->ust_session = NULL;
2817 return ret;
2818 }
2819
2820 /*
2821 * Create a kernel tracer session then create the default channel.
2822 */
2823 static int create_kernel_session(struct ltt_session *session)
2824 {
2825 int ret;
2826
2827 DBG("Creating kernel session");
2828
2829 ret = kernel_create_session(session, kernel_tracer_fd);
2830 if (ret < 0) {
2831 ret = LTTNG_ERR_KERN_SESS_FAIL;
2832 goto error;
2833 }
2834
2835 /* Code flow safety */
2836 assert(session->kernel_session);
2837
2838 /* Copy session output to the newly created Kernel session */
2839 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2840 if (ret != LTTNG_OK) {
2841 goto error;
2842 }
2843
2844 /* Create directory(ies) on local filesystem. */
2845 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2846 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2847 ret = run_as_mkdir_recursive(
2848 session->kernel_session->consumer->dst.trace_path,
2849 S_IRWXU | S_IRWXG, session->uid, session->gid);
2850 if (ret < 0) {
2851 if (errno != EEXIST) {
2852 ERR("Trace directory creation error");
2853 goto error;
2854 }
2855 }
2856 }
2857
2858 session->kernel_session->uid = session->uid;
2859 session->kernel_session->gid = session->gid;
2860 session->kernel_session->output_traces = session->output_traces;
2861 session->kernel_session->snapshot_mode = session->snapshot_mode;
2862
2863 return LTTNG_OK;
2864
2865 error:
2866 trace_kernel_destroy_session(session->kernel_session);
2867 session->kernel_session = NULL;
2868 return ret;
2869 }
2870
2871 /*
2872 * Count number of session permitted by uid/gid.
2873 */
2874 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2875 {
2876 unsigned int i = 0;
2877 struct ltt_session *session;
2878
2879 DBG("Counting number of available session for UID %d GID %d",
2880 uid, gid);
2881 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2882 /*
2883 * Only list the sessions the user can control.
2884 */
2885 if (!session_access_ok(session, uid, gid)) {
2886 continue;
2887 }
2888 i++;
2889 }
2890 return i;
2891 }
2892
2893 /*
2894 * Process the command requested by the lttng client within the command
2895 * context structure. This function make sure that the return structure (llm)
2896 * is set and ready for transmission before returning.
2897 *
2898 * Return any error encountered or 0 for success.
2899 *
2900 * "sock" is only used for special-case var. len data.
2901 *
2902 * Should *NOT* be called with RCU read-side lock held.
2903 */
2904 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2905 int *sock_error)
2906 {
2907 int ret = LTTNG_OK;
2908 int need_tracing_session = 1;
2909 int need_domain;
2910
2911 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2912
2913 assert(!rcu_read_ongoing());
2914
2915 *sock_error = 0;
2916
2917 switch (cmd_ctx->lsm->cmd_type) {
2918 case LTTNG_CREATE_SESSION:
2919 case LTTNG_CREATE_SESSION_SNAPSHOT:
2920 case LTTNG_CREATE_SESSION_LIVE:
2921 case LTTNG_DESTROY_SESSION:
2922 case LTTNG_LIST_SESSIONS:
2923 case LTTNG_LIST_DOMAINS:
2924 case LTTNG_START_TRACE:
2925 case LTTNG_STOP_TRACE:
2926 case LTTNG_DATA_PENDING:
2927 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2928 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2929 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2930 case LTTNG_SNAPSHOT_RECORD:
2931 case LTTNG_SAVE_SESSION:
2932 case LTTNG_SET_SESSION_SHM_PATH:
2933 case LTTNG_REGENERATE_METADATA:
2934 case LTTNG_REGENERATE_STATEDUMP:
2935 case LTTNG_REGISTER_TRIGGER:
2936 case LTTNG_UNREGISTER_TRIGGER:
2937 need_domain = 0;
2938 break;
2939 default:
2940 need_domain = 1;
2941 }
2942
2943 if (config.no_kernel && need_domain
2944 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2945 if (!is_root) {
2946 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2947 } else {
2948 ret = LTTNG_ERR_KERN_NA;
2949 }
2950 goto error;
2951 }
2952
2953 /* Deny register consumer if we already have a spawned consumer. */
2954 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2955 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2956 if (kconsumer_data.pid > 0) {
2957 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2958 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2959 goto error;
2960 }
2961 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2962 }
2963
2964 /*
2965 * Check for command that don't needs to allocate a returned payload. We do
2966 * this here so we don't have to make the call for no payload at each
2967 * command.
2968 */
2969 switch(cmd_ctx->lsm->cmd_type) {
2970 case LTTNG_LIST_SESSIONS:
2971 case LTTNG_LIST_TRACEPOINTS:
2972 case LTTNG_LIST_TRACEPOINT_FIELDS:
2973 case LTTNG_LIST_DOMAINS:
2974 case LTTNG_LIST_CHANNELS:
2975 case LTTNG_LIST_EVENTS:
2976 case LTTNG_LIST_SYSCALLS:
2977 case LTTNG_LIST_TRACKER_PIDS:
2978 case LTTNG_DATA_PENDING:
2979 break;
2980 default:
2981 /* Setup lttng message with no payload */
2982 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
2983 if (ret < 0) {
2984 /* This label does not try to unlock the session */
2985 goto init_setup_error;
2986 }
2987 }
2988
2989 /* Commands that DO NOT need a session. */
2990 switch (cmd_ctx->lsm->cmd_type) {
2991 case LTTNG_CREATE_SESSION:
2992 case LTTNG_CREATE_SESSION_SNAPSHOT:
2993 case LTTNG_CREATE_SESSION_LIVE:
2994 case LTTNG_LIST_SESSIONS:
2995 case LTTNG_LIST_TRACEPOINTS:
2996 case LTTNG_LIST_SYSCALLS:
2997 case LTTNG_LIST_TRACEPOINT_FIELDS:
2998 case LTTNG_SAVE_SESSION:
2999 case LTTNG_REGISTER_TRIGGER:
3000 case LTTNG_UNREGISTER_TRIGGER:
3001 need_tracing_session = 0;
3002 break;
3003 default:
3004 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3005 /*
3006 * We keep the session list lock across _all_ commands
3007 * for now, because the per-session lock does not
3008 * handle teardown properly.
3009 */
3010 session_lock_list();
3011 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3012 if (cmd_ctx->session == NULL) {
3013 ret = LTTNG_ERR_SESS_NOT_FOUND;
3014 goto error;
3015 } else {
3016 /* Acquire lock for the session */
3017 session_lock(cmd_ctx->session);
3018 }
3019 break;
3020 }
3021
3022 /*
3023 * Commands that need a valid session but should NOT create one if none
3024 * exists. Instead of creating one and destroying it when the command is
3025 * handled, process that right before so we save some round trip in useless
3026 * code path.
3027 */
3028 switch (cmd_ctx->lsm->cmd_type) {
3029 case LTTNG_DISABLE_CHANNEL:
3030 case LTTNG_DISABLE_EVENT:
3031 switch (cmd_ctx->lsm->domain.type) {
3032 case LTTNG_DOMAIN_KERNEL:
3033 if (!cmd_ctx->session->kernel_session) {
3034 ret = LTTNG_ERR_NO_CHANNEL;
3035 goto error;
3036 }
3037 break;
3038 case LTTNG_DOMAIN_JUL:
3039 case LTTNG_DOMAIN_LOG4J:
3040 case LTTNG_DOMAIN_PYTHON:
3041 case LTTNG_DOMAIN_UST:
3042 if (!cmd_ctx->session->ust_session) {
3043 ret = LTTNG_ERR_NO_CHANNEL;
3044 goto error;
3045 }
3046 break;
3047 default:
3048 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3049 goto error;
3050 }
3051 default:
3052 break;
3053 }
3054
3055 if (!need_domain) {
3056 goto skip_domain;
3057 }
3058
3059 /*
3060 * Check domain type for specific "pre-action".
3061 */
3062 switch (cmd_ctx->lsm->domain.type) {
3063 case LTTNG_DOMAIN_KERNEL:
3064 if (!is_root) {
3065 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3066 goto error;
3067 }
3068
3069 /* Kernel tracer check */
3070 if (kernel_tracer_fd == -1) {
3071 /* Basically, load kernel tracer modules */
3072 ret = init_kernel_tracer();
3073 if (ret != 0) {
3074 goto error;
3075 }
3076 }
3077
3078 /* Consumer is in an ERROR state. Report back to client */
3079 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3080 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3081 goto error;
3082 }
3083
3084 /* Need a session for kernel command */
3085 if (need_tracing_session) {
3086 if (cmd_ctx->session->kernel_session == NULL) {
3087 ret = create_kernel_session(cmd_ctx->session);
3088 if (ret < 0) {
3089 ret = LTTNG_ERR_KERN_SESS_FAIL;
3090 goto error;
3091 }
3092 }
3093
3094 /* Start the kernel consumer daemon */
3095 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3096 if (kconsumer_data.pid == 0 &&
3097 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3098 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3099 ret = start_consumerd(&kconsumer_data);
3100 if (ret < 0) {
3101 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3102 goto error;
3103 }
3104 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3105 } else {
3106 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3107 }
3108
3109 /*
3110 * The consumer was just spawned so we need to add the socket to
3111 * the consumer output of the session if exist.
3112 */
3113 ret = consumer_create_socket(&kconsumer_data,
3114 cmd_ctx->session->kernel_session->consumer);
3115 if (ret < 0) {
3116 goto error;
3117 }
3118 }
3119
3120 break;
3121 case LTTNG_DOMAIN_JUL:
3122 case LTTNG_DOMAIN_LOG4J:
3123 case LTTNG_DOMAIN_PYTHON:
3124 case LTTNG_DOMAIN_UST:
3125 {
3126 if (!ust_app_supported()) {
3127 ret = LTTNG_ERR_NO_UST;
3128 goto error;
3129 }
3130 /* Consumer is in an ERROR state. Report back to client */
3131 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3132 ret = LTTNG_ERR_NO_USTCONSUMERD;
3133 goto error;
3134 }
3135
3136 if (need_tracing_session) {
3137 /* Create UST session if none exist. */
3138 if (cmd_ctx->session->ust_session == NULL) {
3139 ret = create_ust_session(cmd_ctx->session,
3140 &cmd_ctx->lsm->domain);
3141 if (ret != LTTNG_OK) {
3142 goto error;
3143 }
3144 }
3145
3146 /* Start the UST consumer daemons */
3147 /* 64-bit */
3148 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3149 if (config.consumerd64_bin_path.value &&
3150 ustconsumer64_data.pid == 0 &&
3151 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3152 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3153 ret = start_consumerd(&ustconsumer64_data);
3154 if (ret < 0) {
3155 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3156 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3157 goto error;
3158 }
3159
3160 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3161 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3162 } else {
3163 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3164 }
3165
3166 /*
3167 * Setup socket for consumer 64 bit. No need for atomic access
3168 * since it was set above and can ONLY be set in this thread.
3169 */
3170 ret = consumer_create_socket(&ustconsumer64_data,
3171 cmd_ctx->session->ust_session->consumer);
3172 if (ret < 0) {
3173 goto error;
3174 }
3175
3176 /* 32-bit */
3177 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3178 if (config.consumerd32_bin_path.value &&
3179 ustconsumer32_data.pid == 0 &&
3180 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3181 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3182 ret = start_consumerd(&ustconsumer32_data);
3183 if (ret < 0) {
3184 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3185 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3186 goto error;
3187 }
3188
3189 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3190 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3191 } else {
3192 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3193 }
3194
3195 /*
3196 * Setup socket for consumer 64 bit. No need for atomic access
3197 * since it was set above and can ONLY be set in this thread.
3198 */
3199 ret = consumer_create_socket(&ustconsumer32_data,
3200 cmd_ctx->session->ust_session->consumer);
3201 if (ret < 0) {
3202 goto error;
3203 }
3204 }
3205 break;
3206 }
3207 default:
3208 break;
3209 }
3210 skip_domain:
3211
3212 /* Validate consumer daemon state when start/stop trace command */
3213 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3214 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3215 switch (cmd_ctx->lsm->domain.type) {
3216 case LTTNG_DOMAIN_NONE:
3217 break;
3218 case LTTNG_DOMAIN_JUL:
3219 case LTTNG_DOMAIN_LOG4J:
3220 case LTTNG_DOMAIN_PYTHON:
3221 case LTTNG_DOMAIN_UST:
3222 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3223 ret = LTTNG_ERR_NO_USTCONSUMERD;
3224 goto error;
3225 }
3226 break;
3227 case LTTNG_DOMAIN_KERNEL:
3228 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3229 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3230 goto error;
3231 }
3232 break;
3233 default:
3234 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3235 goto error;
3236 }
3237 }
3238
3239 /*
3240 * Check that the UID or GID match that of the tracing session.
3241 * The root user can interact with all sessions.
3242 */
3243 if (need_tracing_session) {
3244 if (!session_access_ok(cmd_ctx->session,
3245 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3246 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3247 ret = LTTNG_ERR_EPERM;
3248 goto error;
3249 }
3250 }
3251
3252 /*
3253 * Send relayd information to consumer as soon as we have a domain and a
3254 * session defined.
3255 */
3256 if (cmd_ctx->session && need_domain) {
3257 /*
3258 * Setup relayd if not done yet. If the relayd information was already
3259 * sent to the consumer, this call will gracefully return.
3260 */
3261 ret = cmd_setup_relayd(cmd_ctx->session);
3262 if (ret != LTTNG_OK) {
3263 goto error;
3264 }
3265 }
3266
3267 /* Process by command type */
3268 switch (cmd_ctx->lsm->cmd_type) {
3269 case LTTNG_ADD_CONTEXT:
3270 {
3271 /*
3272 * An LTTNG_ADD_CONTEXT command might have a supplementary
3273 * payload if the context being added is an application context.
3274 */
3275 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3276 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3277 char *provider_name = NULL, *context_name = NULL;
3278 size_t provider_name_len =
3279 cmd_ctx->lsm->u.context.provider_name_len;
3280 size_t context_name_len =
3281 cmd_ctx->lsm->u.context.context_name_len;
3282
3283 if (provider_name_len == 0 || context_name_len == 0) {
3284 /*
3285 * Application provider and context names MUST
3286 * be provided.
3287 */
3288 ret = -LTTNG_ERR_INVALID;
3289 goto error;
3290 }
3291
3292 provider_name = zmalloc(provider_name_len + 1);
3293 if (!provider_name) {
3294 ret = -LTTNG_ERR_NOMEM;
3295 goto error;
3296 }
3297 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3298 provider_name;
3299
3300 context_name = zmalloc(context_name_len + 1);
3301 if (!context_name) {
3302 ret = -LTTNG_ERR_NOMEM;
3303 goto error_add_context;
3304 }
3305 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3306 context_name;
3307
3308 ret = lttcomm_recv_unix_sock(sock, provider_name,
3309 provider_name_len);
3310 if (ret < 0) {
3311 goto error_add_context;
3312 }
3313
3314 ret = lttcomm_recv_unix_sock(sock, context_name,
3315 context_name_len);
3316 if (ret < 0) {
3317 goto error_add_context;
3318 }
3319 }
3320
3321 /*
3322 * cmd_add_context assumes ownership of the provider and context
3323 * names.
3324 */
3325 ret = cmd_add_context(cmd_ctx->session,
3326 cmd_ctx->lsm->domain.type,
3327 cmd_ctx->lsm->u.context.channel_name,
3328 &cmd_ctx->lsm->u.context.ctx,
3329 kernel_poll_pipe[1]);
3330
3331 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3332 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3333 error_add_context:
3334 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3335 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3336 if (ret < 0) {
3337 goto error;
3338 }
3339 break;
3340 }
3341 case LTTNG_DISABLE_CHANNEL:
3342 {
3343 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3344 cmd_ctx->lsm->u.disable.channel_name);
3345 break;
3346 }
3347 case LTTNG_DISABLE_EVENT:
3348 {
3349
3350 /*
3351 * FIXME: handle filter; for now we just receive the filter's
3352 * bytecode along with the filter expression which are sent by
3353 * liblttng-ctl and discard them.
3354 *
3355 * This fixes an issue where the client may block while sending
3356 * the filter payload and encounter an error because the session
3357 * daemon closes the socket without ever handling this data.
3358 */
3359 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3360 cmd_ctx->lsm->u.disable.bytecode_len;
3361
3362 if (count) {
3363 char data[LTTNG_FILTER_MAX_LEN];
3364
3365 DBG("Discarding disable event command payload of size %zu", count);
3366 while (count) {
3367 ret = lttcomm_recv_unix_sock(sock, data,
3368 count > sizeof(data) ? sizeof(data) : count);
3369 if (ret < 0) {
3370 goto error;
3371 }
3372
3373 count -= (size_t) ret;
3374 }
3375 }
3376 /* FIXME: passing packed structure to non-packed pointer */
3377 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3378 cmd_ctx->lsm->u.disable.channel_name,
3379 &cmd_ctx->lsm->u.disable.event);
3380 break;
3381 }
3382 case LTTNG_ENABLE_CHANNEL:
3383 {
3384 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3385 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3386 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3387 &cmd_ctx->lsm->u.channel.chan,
3388 kernel_poll_pipe[1]);
3389 break;
3390 }
3391 case LTTNG_TRACK_PID:
3392 {
3393 ret = cmd_track_pid(cmd_ctx->session,
3394 cmd_ctx->lsm->domain.type,
3395 cmd_ctx->lsm->u.pid_tracker.pid);
3396 break;
3397 }
3398 case LTTNG_UNTRACK_PID:
3399 {
3400 ret = cmd_untrack_pid(cmd_ctx->session,
3401 cmd_ctx->lsm->domain.type,
3402 cmd_ctx->lsm->u.pid_tracker.pid);
3403 break;
3404 }
3405 case LTTNG_ENABLE_EVENT:
3406 {
3407 struct lttng_event_exclusion *exclusion = NULL;
3408 struct lttng_filter_bytecode *bytecode = NULL;
3409 char *filter_expression = NULL;
3410
3411 /* Handle exclusion events and receive it from the client. */
3412 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3413 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3414
3415 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3416 (count * LTTNG_SYMBOL_NAME_LEN));
3417 if (!exclusion) {
3418 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3419 goto error;
3420 }
3421
3422 DBG("Receiving var len exclusion event list from client ...");
3423 exclusion->count = count;
3424 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3425 count * LTTNG_SYMBOL_NAME_LEN);
3426 if (ret <= 0) {
3427 DBG("Nothing recv() from client var len data... continuing");
3428 *sock_error = 1;
3429 free(exclusion);
3430 ret = LTTNG_ERR_EXCLUSION_INVAL;
3431 goto error;
3432 }
3433 }
3434
3435 /* Get filter expression from client. */
3436 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3437 size_t expression_len =
3438 cmd_ctx->lsm->u.enable.expression_len;
3439
3440 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3441 ret = LTTNG_ERR_FILTER_INVAL;
3442 free(exclusion);
3443 goto error;
3444 }
3445
3446 filter_expression = zmalloc(expression_len);
3447 if (!filter_expression) {
3448 free(exclusion);
3449 ret = LTTNG_ERR_FILTER_NOMEM;
3450 goto error;
3451 }
3452
3453 /* Receive var. len. data */
3454 DBG("Receiving var len filter's expression from client ...");
3455 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3456 expression_len);
3457 if (ret <= 0) {
3458 DBG("Nothing recv() from client car len data... continuing");
3459 *sock_error = 1;
3460 free(filter_expression);
3461 free(exclusion);
3462 ret = LTTNG_ERR_FILTER_INVAL;
3463 goto error;
3464 }
3465 }
3466
3467 /* Handle filter and get bytecode from client. */
3468 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3469 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3470
3471 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3472 ret = LTTNG_ERR_FILTER_INVAL;
3473 free(filter_expression);
3474 free(exclusion);
3475 goto error;
3476 }
3477
3478 bytecode = zmalloc(bytecode_len);
3479 if (!bytecode) {
3480 free(filter_expression);
3481 free(exclusion);
3482 ret = LTTNG_ERR_FILTER_NOMEM;
3483 goto error;
3484 }
3485
3486 /* Receive var. len. data */
3487 DBG("Receiving var len filter's bytecode from client ...");
3488 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3489 if (ret <= 0) {
3490 DBG("Nothing recv() from client car len data... continuing");
3491 *sock_error = 1;
3492 free(filter_expression);
3493 free(bytecode);
3494 free(exclusion);
3495 ret = LTTNG_ERR_FILTER_INVAL;
3496 goto error;
3497 }
3498
3499 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3500 free(filter_expression);
3501 free(bytecode);
3502 free(exclusion);
3503 ret = LTTNG_ERR_FILTER_INVAL;
3504 goto error;
3505 }
3506 }
3507
3508 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3509 cmd_ctx->lsm->u.enable.channel_name,
3510 &cmd_ctx->lsm->u.enable.event,
3511 filter_expression, bytecode, exclusion,
3512 kernel_poll_pipe[1]);
3513 break;
3514 }
3515 case LTTNG_LIST_TRACEPOINTS:
3516 {
3517 struct lttng_event *events;
3518 ssize_t nb_events;
3519
3520 session_lock_list();
3521 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3522 session_unlock_list();
3523 if (nb_events < 0) {
3524 /* Return value is a negative lttng_error_code. */
3525 ret = -nb_events;
3526 goto error;
3527 }
3528
3529 /*
3530 * Setup lttng message with payload size set to the event list size in
3531 * bytes and then copy list into the llm payload.
3532 */
3533 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3534 sizeof(struct lttng_event) * nb_events);
3535 free(events);
3536
3537 if (ret < 0) {
3538 goto setup_error;
3539 }
3540
3541 ret = LTTNG_OK;
3542 break;
3543 }
3544 case LTTNG_LIST_TRACEPOINT_FIELDS:
3545 {
3546 struct lttng_event_field *fields;
3547 ssize_t nb_fields;
3548
3549 session_lock_list();
3550 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3551 &fields);
3552 session_unlock_list();
3553 if (nb_fields < 0) {
3554 /* Return value is a negative lttng_error_code. */
3555 ret = -nb_fields;
3556 goto error;
3557 }
3558
3559 /*
3560 * Setup lttng message with payload size set to the event list size in
3561 * bytes and then copy list into the llm payload.
3562 */
3563 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3564 sizeof(struct lttng_event_field) * nb_fields);
3565 free(fields);
3566
3567 if (ret < 0) {
3568 goto setup_error;
3569 }
3570
3571 ret = LTTNG_OK;
3572 break;
3573 }
3574 case LTTNG_LIST_SYSCALLS:
3575 {
3576 struct lttng_event *events;
3577 ssize_t nb_events;
3578
3579 nb_events = cmd_list_syscalls(&events);
3580 if (nb_events < 0) {
3581 /* Return value is a negative lttng_error_code. */
3582 ret = -nb_events;
3583 goto error;
3584 }
3585
3586 /*
3587 * Setup lttng message with payload size set to the event list size in
3588 * bytes and then copy list into the llm payload.
3589 */
3590 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3591 sizeof(struct lttng_event) * nb_events);
3592 free(events);
3593
3594 if (ret < 0) {
3595 goto setup_error;
3596 }
3597
3598 ret = LTTNG_OK;
3599 break;
3600 }
3601 case LTTNG_LIST_TRACKER_PIDS:
3602 {
3603 int32_t *pids = NULL;
3604 ssize_t nr_pids;
3605
3606 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3607 cmd_ctx->lsm->domain.type, &pids);
3608 if (nr_pids < 0) {
3609 /* Return value is a negative lttng_error_code. */
3610 ret = -nr_pids;
3611 goto error;
3612 }
3613
3614 /*
3615 * Setup lttng message with payload size set to the event list size in
3616 * bytes and then copy list into the llm payload.
3617 */
3618 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3619 sizeof(int32_t) * nr_pids);
3620 free(pids);
3621
3622 if (ret < 0) {
3623 goto setup_error;
3624 }
3625
3626 ret = LTTNG_OK;
3627 break;
3628 }
3629 case LTTNG_SET_CONSUMER_URI:
3630 {
3631 size_t nb_uri, len;
3632 struct lttng_uri *uris;
3633
3634 nb_uri = cmd_ctx->lsm->u.uri.size;
3635 len = nb_uri * sizeof(struct lttng_uri);
3636
3637 if (nb_uri == 0) {
3638 ret = LTTNG_ERR_INVALID;
3639 goto error;
3640 }
3641
3642 uris = zmalloc(len);
3643 if (uris == NULL) {
3644 ret = LTTNG_ERR_FATAL;
3645 goto error;
3646 }
3647
3648 /* Receive variable len data */
3649 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3650 ret = lttcomm_recv_unix_sock(sock, uris, len);
3651 if (ret <= 0) {
3652 DBG("No URIs received from client... continuing");
3653 *sock_error = 1;
3654 ret = LTTNG_ERR_SESSION_FAIL;
3655 free(uris);
3656 goto error;
3657 }
3658
3659 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3660 free(uris);
3661 if (ret != LTTNG_OK) {
3662 goto error;
3663 }
3664
3665
3666 break;
3667 }
3668 case LTTNG_START_TRACE:
3669 {
3670 ret = cmd_start_trace(cmd_ctx->session);
3671 break;
3672 }
3673 case LTTNG_STOP_TRACE:
3674 {
3675 ret = cmd_stop_trace(cmd_ctx->session);
3676 break;
3677 }
3678 case LTTNG_CREATE_SESSION:
3679 {
3680 size_t nb_uri, len;
3681 struct lttng_uri *uris = NULL;
3682
3683 nb_uri = cmd_ctx->lsm->u.uri.size;
3684 len = nb_uri * sizeof(struct lttng_uri);
3685
3686 if (nb_uri > 0) {
3687 uris = zmalloc(len);
3688 if (uris == NULL) {
3689 ret = LTTNG_ERR_FATAL;
3690 goto error;
3691 }
3692
3693 /* Receive variable len data */
3694 DBG("Waiting for %zu URIs from client ...", nb_uri);
3695 ret = lttcomm_recv_unix_sock(sock, uris, len);
3696 if (ret <= 0) {
3697 DBG("No URIs received from client... continuing");
3698 *sock_error = 1;
3699 ret = LTTNG_ERR_SESSION_FAIL;
3700 free(uris);
3701 goto error;
3702 }
3703
3704 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3705 DBG("Creating session with ONE network URI is a bad call");
3706 ret = LTTNG_ERR_SESSION_FAIL;
3707 free(uris);
3708 goto error;
3709 }
3710 }
3711
3712 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3713 &cmd_ctx->creds, 0);
3714
3715 free(uris);
3716
3717 break;
3718 }
3719 case LTTNG_DESTROY_SESSION:
3720 {
3721 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3722
3723 /* Set session to NULL so we do not unlock it after free. */
3724 cmd_ctx->session = NULL;
3725 break;
3726 }
3727 case LTTNG_LIST_DOMAINS:
3728 {
3729 ssize_t nb_dom;
3730 struct lttng_domain *domains = NULL;
3731
3732 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3733 if (nb_dom < 0) {
3734 /* Return value is a negative lttng_error_code. */
3735 ret = -nb_dom;
3736 goto error;
3737 }
3738
3739 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3740 nb_dom * sizeof(struct lttng_domain));
3741 free(domains);
3742
3743 if (ret < 0) {
3744 goto setup_error;
3745 }
3746
3747 ret = LTTNG_OK;
3748 break;
3749 }
3750 case LTTNG_LIST_CHANNELS:
3751 {
3752 ssize_t payload_size;
3753 struct lttng_channel *channels = NULL;
3754
3755 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3756 cmd_ctx->session, &channels);
3757 if (payload_size < 0) {
3758 /* Return value is a negative lttng_error_code. */
3759 ret = -payload_size;
3760 goto error;
3761 }
3762
3763 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3764 payload_size);
3765 free(channels);
3766
3767 if (ret < 0) {
3768 goto setup_error;
3769 }
3770
3771 ret = LTTNG_OK;
3772 break;
3773 }
3774 case LTTNG_LIST_EVENTS:
3775 {
3776 ssize_t nb_event;
3777 struct lttng_event *events = NULL;
3778 struct lttcomm_event_command_header cmd_header;
3779 size_t total_size;
3780
3781 memset(&cmd_header, 0, sizeof(cmd_header));
3782 /* Extended infos are included at the end of events */
3783 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3784 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3785 &events, &total_size);
3786
3787 if (nb_event < 0) {
3788 /* Return value is a negative lttng_error_code. */
3789 ret = -nb_event;
3790 goto error;
3791 }
3792
3793 cmd_header.nb_events = nb_event;
3794 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3795 &cmd_header, sizeof(cmd_header));
3796 free(events);
3797
3798 if (ret < 0) {
3799 goto setup_error;
3800 }
3801
3802 ret = LTTNG_OK;
3803 break;
3804 }
3805 case LTTNG_LIST_SESSIONS:
3806 {
3807 unsigned int nr_sessions;
3808 void *sessions_payload;
3809 size_t payload_len;
3810
3811 session_lock_list();
3812 nr_sessions = lttng_sessions_count(
3813 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3814 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3815 payload_len = sizeof(struct lttng_session) * nr_sessions;
3816 sessions_payload = zmalloc(payload_len);
3817
3818 if (!sessions_payload) {
3819 session_unlock_list();
3820 ret = -ENOMEM;
3821 goto setup_error;
3822 }
3823
3824 cmd_list_lttng_sessions(sessions_payload,
3825 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3826 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3827 session_unlock_list();
3828
3829 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3830 payload_len);
3831 free(sessions_payload);
3832
3833 if (ret < 0) {
3834 goto setup_error;
3835 }
3836
3837 ret = LTTNG_OK;
3838 break;
3839 }
3840 case LTTNG_REGISTER_CONSUMER:
3841 {
3842 struct consumer_data *cdata;
3843
3844 switch (cmd_ctx->lsm->domain.type) {
3845 case LTTNG_DOMAIN_KERNEL:
3846 cdata = &kconsumer_data;
3847 break;
3848 default:
3849 ret = LTTNG_ERR_UND;
3850 goto error;
3851 }
3852
3853 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3854 cmd_ctx->lsm->u.reg.path, cdata);
3855 break;
3856 }
3857 case LTTNG_DATA_PENDING:
3858 {
3859 int pending_ret;
3860 uint8_t pending_ret_byte;
3861
3862 pending_ret = cmd_data_pending(cmd_ctx->session);
3863
3864 /*
3865 * FIXME
3866 *
3867 * This function may returns 0 or 1 to indicate whether or not
3868 * there is data pending. In case of error, it should return an
3869 * LTTNG_ERR code. However, some code paths may still return
3870 * a nondescript error code, which we handle by returning an
3871 * "unknown" error.
3872 */
3873 if (pending_ret == 0 || pending_ret == 1) {
3874 /*
3875 * ret will be set to LTTNG_OK at the end of
3876 * this function.
3877 */
3878 } else if (pending_ret < 0) {
3879 ret = LTTNG_ERR_UNK;
3880 goto setup_error;
3881 } else {
3882 ret = pending_ret;
3883 goto setup_error;
3884 }
3885
3886 pending_ret_byte = (uint8_t) pending_ret;
3887
3888 /* 1 byte to return whether or not data is pending */
3889 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
3890 &pending_ret_byte, 1);
3891
3892 if (ret < 0) {
3893 goto setup_error;
3894 }
3895
3896 ret = LTTNG_OK;
3897 break;
3898 }
3899 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3900 {
3901 struct lttcomm_lttng_output_id reply;
3902
3903 ret = cmd_snapshot_add_output(cmd_ctx->session,
3904 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3905 if (ret != LTTNG_OK) {
3906 goto error;
3907 }
3908
3909 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
3910 sizeof(reply));
3911 if (ret < 0) {
3912 goto setup_error;
3913 }
3914
3915 /* Copy output list into message payload */
3916 ret = LTTNG_OK;
3917 break;
3918 }
3919 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3920 {
3921 ret = cmd_snapshot_del_output(cmd_ctx->session,
3922 &cmd_ctx->lsm->u.snapshot_output.output);
3923 break;
3924 }
3925 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3926 {
3927 ssize_t nb_output;
3928 struct lttng_snapshot_output *outputs = NULL;
3929
3930 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3931 if (nb_output < 0) {
3932 ret = -nb_output;
3933 goto error;
3934 }
3935
3936 assert((nb_output > 0 && outputs) || nb_output == 0);
3937 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
3938 nb_output * sizeof(struct lttng_snapshot_output));
3939 free(outputs);
3940
3941 if (ret < 0) {
3942 goto setup_error;
3943 }
3944
3945 ret = LTTNG_OK;
3946 break;
3947 }
3948 case LTTNG_SNAPSHOT_RECORD:
3949 {
3950 ret = cmd_snapshot_record(cmd_ctx->session,
3951 &cmd_ctx->lsm->u.snapshot_record.output,
3952 cmd_ctx->lsm->u.snapshot_record.wait);
3953 break;
3954 }
3955 case LTTNG_CREATE_SESSION_SNAPSHOT:
3956 {
3957 size_t nb_uri, len;
3958 struct lttng_uri *uris = NULL;
3959
3960 nb_uri = cmd_ctx->lsm->u.uri.size;
3961 len = nb_uri * sizeof(struct lttng_uri);
3962
3963 if (nb_uri > 0) {
3964 uris = zmalloc(len);
3965 if (uris == NULL) {
3966 ret = LTTNG_ERR_FATAL;
3967 goto error;
3968 }
3969
3970 /* Receive variable len data */
3971 DBG("Waiting for %zu URIs from client ...", nb_uri);
3972 ret = lttcomm_recv_unix_sock(sock, uris, len);
3973 if (ret <= 0) {
3974 DBG("No URIs received from client... continuing");
3975 *sock_error = 1;
3976 ret = LTTNG_ERR_SESSION_FAIL;
3977 free(uris);
3978 goto error;
3979 }
3980
3981 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3982 DBG("Creating session with ONE network URI is a bad call");
3983 ret = LTTNG_ERR_SESSION_FAIL;
3984 free(uris);
3985 goto error;
3986 }
3987 }
3988
3989 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3990 nb_uri, &cmd_ctx->creds);
3991 free(uris);
3992 break;
3993 }
3994 case LTTNG_CREATE_SESSION_LIVE:
3995 {
3996 size_t nb_uri, len;
3997 struct lttng_uri *uris = NULL;
3998
3999 nb_uri = cmd_ctx->lsm->u.uri.size;
4000 len = nb_uri * sizeof(struct lttng_uri);
4001
4002 if (nb_uri > 0) {
4003 uris = zmalloc(len);
4004 if (uris == NULL) {
4005 ret = LTTNG_ERR_FATAL;
4006 goto error;
4007 }
4008
4009 /* Receive variable len data */
4010 DBG("Waiting for %zu URIs from client ...", nb_uri);
4011 ret = lttcomm_recv_unix_sock(sock, uris, len);
4012 if (ret <= 0) {
4013 DBG("No URIs received from client... continuing");
4014 *sock_error = 1;
4015 ret = LTTNG_ERR_SESSION_FAIL;
4016 free(uris);
4017 goto error;
4018 }
4019
4020 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4021 DBG("Creating session with ONE network URI is a bad call");
4022 ret = LTTNG_ERR_SESSION_FAIL;
4023 free(uris);
4024 goto error;
4025 }
4026 }
4027
4028 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4029 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4030 free(uris);
4031 break;
4032 }
4033 case LTTNG_SAVE_SESSION:
4034 {
4035 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4036 &cmd_ctx->creds);
4037 break;
4038 }
4039 case LTTNG_SET_SESSION_SHM_PATH:
4040 {
4041 ret = cmd_set_session_shm_path(cmd_ctx->session,
4042 cmd_ctx->lsm->u.set_shm_path.shm_path);
4043 break;
4044 }
4045 case LTTNG_REGENERATE_METADATA:
4046 {
4047 ret = cmd_regenerate_metadata(cmd_ctx->session);
4048 break;
4049 }
4050 case LTTNG_REGENERATE_STATEDUMP:
4051 {
4052 ret = cmd_regenerate_statedump(cmd_ctx->session);
4053 break;
4054 }
4055 case LTTNG_REGISTER_TRIGGER:
4056 {
4057 ret = cmd_register_trigger(cmd_ctx, sock,
4058 notification_thread_handle);
4059 break;
4060 }
4061 case LTTNG_UNREGISTER_TRIGGER:
4062 {
4063 ret = cmd_unregister_trigger(cmd_ctx, sock,
4064 notification_thread_handle);
4065 break;
4066 }
4067 default:
4068 ret = LTTNG_ERR_UND;
4069 break;
4070 }
4071
4072 error:
4073 if (cmd_ctx->llm == NULL) {
4074 DBG("Missing llm structure. Allocating one.");
4075 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4076 goto setup_error;
4077 }
4078 }
4079 /* Set return code */
4080 cmd_ctx->llm->ret_code = ret;
4081 setup_error:
4082 if (cmd_ctx->session) {
4083 session_unlock(cmd_ctx->session);
4084 }
4085 if (need_tracing_session) {
4086 session_unlock_list();
4087 }
4088 init_setup_error:
4089 assert(!rcu_read_ongoing());
4090 return ret;
4091 }
4092
4093 /*
4094 * Thread managing health check socket.
4095 */
4096 static void *thread_manage_health(void *data)
4097 {
4098 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4099 uint32_t revents, nb_fd;
4100 struct lttng_poll_event events;
4101 struct health_comm_msg msg;
4102 struct health_comm_reply reply;
4103
4104 DBG("[thread] Manage health check started");
4105
4106 rcu_register_thread();
4107
4108 /* We might hit an error path before this is created. */
4109 lttng_poll_init(&events);
4110
4111 /* Create unix socket */
4112 sock = lttcomm_create_unix_sock(config.health_unix_sock_path.value);
4113 if (sock < 0) {
4114 ERR("Unable to create health check Unix socket");
4115 goto error;
4116 }
4117
4118 if (is_root) {
4119 /* lttng health client socket path permissions */
4120 ret = chown(config.health_unix_sock_path.value, 0,
4121 utils_get_group_id(config.tracing_group_name.value));
4122 if (ret < 0) {
4123 ERR("Unable to set group on %s", config.health_unix_sock_path.value);
4124 PERROR("chown");
4125 goto error;
4126 }
4127
4128 ret = chmod(config.health_unix_sock_path.value,
4129 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4130 if (ret < 0) {
4131 ERR("Unable to set permissions on %s", config.health_unix_sock_path.value);
4132 PERROR("chmod");
4133 goto error;
4134 }
4135 }
4136
4137 /*
4138 * Set the CLOEXEC flag. Return code is useless because either way, the
4139 * show must go on.
4140 */
4141 (void) utils_set_fd_cloexec(sock);
4142
4143 ret = lttcomm_listen_unix_sock(sock);
4144 if (ret < 0) {
4145 goto error;
4146 }
4147
4148 /*
4149 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4150 * more will be added to this poll set.
4151 */
4152 ret = sessiond_set_thread_pollset(&events, 2);
4153 if (ret < 0) {
4154 goto error;
4155 }
4156
4157 /* Add the application registration socket */
4158 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4159 if (ret < 0) {
4160 goto error;
4161 }
4162
4163 sessiond_notify_ready();
4164
4165 while (1) {
4166 DBG("Health check ready");
4167
4168 /* Inifinite blocking call, waiting for transmission */
4169 restart:
4170 ret = lttng_poll_wait(&events, -1);
4171 if (ret < 0) {
4172 /*
4173 * Restart interrupted system call.
4174 */
4175 if (errno == EINTR) {
4176 goto restart;
4177 }
4178 goto error;
4179 }
4180
4181 nb_fd = ret;
4182
4183 for (i = 0; i < nb_fd; i++) {
4184 /* Fetch once the poll data */
4185 revents = LTTNG_POLL_GETEV(&events, i);
4186 pollfd = LTTNG_POLL_GETFD(&events, i);
4187
4188 if (!revents) {
4189 /* No activity for this FD (poll implementation). */
4190 continue;
4191 }
4192
4193 /* Thread quit pipe has been closed. Killing thread. */
4194 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4195 if (ret) {
4196 err = 0;
4197 goto exit;
4198 }
4199
4200 /* Event on the registration socket */
4201 if (pollfd == sock) {
4202 if (revents & LPOLLIN) {
4203 continue;
4204 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4205 ERR("Health socket poll error");
4206 goto error;
4207 } else {
4208 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4209 goto error;
4210 }
4211 }
4212 }
4213
4214 new_sock = lttcomm_accept_unix_sock(sock);
4215 if (new_sock < 0) {
4216 goto error;
4217 }
4218
4219 /*
4220 * Set the CLOEXEC flag. Return code is useless because either way, the
4221 * show must go on.
4222 */
4223 (void) utils_set_fd_cloexec(new_sock);
4224
4225 DBG("Receiving data from client for health...");
4226 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4227 if (ret <= 0) {
4228 DBG("Nothing recv() from client... continuing");
4229 ret = close(new_sock);
4230 if (ret) {
4231 PERROR("close");
4232 }
4233 continue;
4234 }
4235
4236 rcu_thread_online();
4237
4238 memset(&reply, 0, sizeof(reply));
4239 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4240 /*
4241 * health_check_state returns 0 if health is
4242 * bad.
4243 */
4244 if (!health_check_state(health_sessiond, i)) {
4245 reply.ret_code |= 1ULL << i;
4246 }
4247 }
4248
4249 DBG2("Health check return value %" PRIx64, reply.ret_code);
4250
4251 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4252 if (ret < 0) {
4253 ERR("Failed to send health data back to client");
4254 }
4255
4256 /* End of transmission */
4257 ret = close(new_sock);
4258 if (ret) {
4259 PERROR("close");
4260 }
4261 }
4262
4263 exit:
4264 error:
4265 if (err) {
4266 ERR("Health error occurred in %s", __func__);
4267 }
4268 DBG("Health check thread dying");
4269 unlink(config.health_unix_sock_path.value);
4270 if (sock >= 0) {
4271 ret = close(sock);
4272 if (ret) {
4273 PERROR("close");
4274 }
4275 }
4276
4277 lttng_poll_clean(&events);
4278 stop_threads();
4279 rcu_unregister_thread();
4280 return NULL;
4281 }
4282
4283 /*
4284 * This thread manage all clients request using the unix client socket for
4285 * communication.
4286 */
4287 static void *thread_manage_clients(void *data)
4288 {
4289 int sock = -1, ret, i, pollfd, err = -1;
4290 int sock_error;
4291 uint32_t revents, nb_fd;
4292 struct command_ctx *cmd_ctx = NULL;
4293 struct lttng_poll_event events;
4294
4295 DBG("[thread] Manage client started");
4296
4297 rcu_register_thread();
4298
4299 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4300
4301 health_code_update();
4302
4303 ret = lttcomm_listen_unix_sock(client_sock);
4304 if (ret < 0) {
4305 goto error_listen;
4306 }
4307
4308 /*
4309 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4310 * more will be added to this poll set.
4311 */
4312 ret = sessiond_set_thread_pollset(&events, 2);
4313 if (ret < 0) {
4314 goto error_create_poll;
4315 }
4316
4317 /* Add the application registration socket */
4318 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4319 if (ret < 0) {
4320 goto error;
4321 }
4322
4323 ret = sem_post(&load_info->message_thread_ready);
4324 if (ret) {
4325 PERROR("sem_post message_thread_ready");
4326 goto error;
4327 }
4328
4329 /*
4330 * Wait until all support threads are initialized before accepting
4331 * commands.
4332 */
4333 while (uatomic_read(&lttng_sessiond_ready) != 0) {
4334 fd_set read_fds;
4335 struct timeval timeout;
4336
4337 FD_ZERO(&read_fds);
4338 FD_SET(thread_quit_pipe[0], &read_fds);
4339 memset(&timeout, 0, sizeof(timeout));
4340 timeout.tv_usec = 1000;
4341
4342 /*
4343 * If a support thread failed to launch, it may signal that
4344 * we must exit and the sessiond would never be marked as
4345 * "ready".
4346 *
4347 * The timeout is set to 1ms, which serves as a way to
4348 * pace down this check.
4349 */
4350 ret = select(thread_quit_pipe[0] + 1, &read_fds, NULL, NULL,
4351 &timeout);
4352 if (ret > 0 || (ret < 0 && errno != EINTR)) {
4353 goto exit;
4354 }
4355 }
4356 /*
4357 * This barrier is paired with the one in sessiond_notify_ready() to
4358 * ensure that loads accessing data initialized by the other threads,
4359 * on which this thread was waiting, are not performed before this point.
4360 *
4361 * Note that this could be a 'read' memory barrier, but a full barrier
4362 * is used in case the code changes. The performance implications of
4363 * this choice are minimal since this is a slow path.
4364 */
4365 cmm_smp_mb();
4366
4367 /* This testpoint is after we signal readiness to the parent. */
4368 if (testpoint(sessiond_thread_manage_clients)) {
4369 goto error;
4370 }
4371
4372 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4373 goto error;
4374 }
4375
4376 health_code_update();
4377
4378 while (1) {
4379 DBG("Accepting client command ...");
4380
4381 /* Inifinite blocking call, waiting for transmission */
4382 restart:
4383 health_poll_entry();
4384 ret = lttng_poll_wait(&events, -1);
4385 health_poll_exit();
4386 if (ret < 0) {
4387 /*
4388 * Restart interrupted system call.
4389 */
4390 if (errno == EINTR) {
4391 goto restart;
4392 }
4393 goto error;
4394 }
4395
4396 nb_fd = ret;
4397
4398 for (i = 0; i < nb_fd; i++) {
4399 /* Fetch once the poll data */
4400 revents = LTTNG_POLL_GETEV(&events, i);
4401 pollfd = LTTNG_POLL_GETFD(&events, i);
4402
4403 health_code_update();
4404
4405 if (!revents) {
4406 /* No activity for this FD (poll implementation). */
4407 continue;
4408 }
4409
4410 /* Thread quit pipe has been closed. Killing thread. */
4411 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4412 if (ret) {
4413 err = 0;
4414 goto exit;
4415 }
4416
4417 /* Event on the registration socket */
4418 if (pollfd == client_sock) {
4419 if (revents & LPOLLIN) {
4420 continue;
4421 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4422 ERR("Client socket poll error");
4423 goto error;
4424 } else {
4425 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4426 goto error;
4427 }
4428 }
4429 }
4430
4431 DBG("Wait for client response");
4432
4433 health_code_update();
4434
4435 sock = lttcomm_accept_unix_sock(client_sock);
4436 if (sock < 0) {
4437 goto error;
4438 }
4439
4440 /*
4441 * Set the CLOEXEC flag. Return code is useless because either way, the
4442 * show must go on.
4443 */
4444 (void) utils_set_fd_cloexec(sock);
4445
4446 /* Set socket option for credentials retrieval */
4447 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4448 if (ret < 0) {
4449 goto error;
4450 }
4451
4452 /* Allocate context command to process the client request */
4453 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4454 if (cmd_ctx == NULL) {
4455 PERROR("zmalloc cmd_ctx");
4456 goto error;
4457 }
4458
4459 /* Allocate data buffer for reception */
4460 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4461 if (cmd_ctx->lsm == NULL) {
4462 PERROR("zmalloc cmd_ctx->lsm");
4463 goto error;
4464 }
4465
4466 cmd_ctx->llm = NULL;
4467 cmd_ctx->session = NULL;
4468
4469 health_code_update();
4470
4471 /*
4472 * Data is received from the lttng client. The struct
4473 * lttcomm_session_msg (lsm) contains the command and data request of
4474 * the client.
4475 */
4476 DBG("Receiving data from client ...");
4477 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4478 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4479 if (ret <= 0) {
4480 DBG("Nothing recv() from client... continuing");
4481 ret = close(sock);
4482 if (ret) {
4483 PERROR("close");
4484 }
4485 sock = -1;
4486 clean_command_ctx(&cmd_ctx);
4487 continue;
4488 }
4489
4490 health_code_update();
4491
4492 // TODO: Validate cmd_ctx including sanity check for
4493 // security purpose.
4494
4495 rcu_thread_online();
4496 /*
4497 * This function dispatch the work to the kernel or userspace tracer
4498 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4499 * informations for the client. The command context struct contains
4500 * everything this function may needs.
4501 */
4502 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4503 rcu_thread_offline();
4504 if (ret < 0) {
4505 ret = close(sock);
4506 if (ret) {
4507 PERROR("close");
4508 }
4509 sock = -1;
4510 /*
4511 * TODO: Inform client somehow of the fatal error. At
4512 * this point, ret < 0 means that a zmalloc failed
4513 * (ENOMEM). Error detected but still accept
4514 * command, unless a socket error has been
4515 * detected.
4516 */
4517 clean_command_ctx(&cmd_ctx);
4518 continue;
4519 }
4520
4521 health_code_update();
4522
4523 DBG("Sending response (size: %d, retcode: %s (%d))",
4524 cmd_ctx->lttng_msg_size,
4525 lttng_strerror(-cmd_ctx->llm->ret_code),
4526 cmd_ctx->llm->ret_code);
4527 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4528 if (ret < 0) {
4529 ERR("Failed to send data back to client");
4530 }
4531
4532 /* End of transmission */
4533 ret = close(sock);
4534 if (ret) {
4535 PERROR("close");
4536 }
4537 sock = -1;
4538
4539 clean_command_ctx(&cmd_ctx);
4540
4541 health_code_update();
4542 }
4543
4544 exit:
4545 error:
4546 if (sock >= 0) {
4547 ret = close(sock);
4548 if (ret) {
4549 PERROR("close");
4550 }
4551 }
4552
4553 lttng_poll_clean(&events);
4554 clean_command_ctx(&cmd_ctx);
4555
4556 error_listen:
4557 error_create_poll:
4558 unlink(config.client_unix_sock_path.value);
4559 if (client_sock >= 0) {
4560 ret = close(client_sock);
4561 if (ret) {
4562 PERROR("close");
4563 }
4564 }
4565
4566 if (err) {
4567 health_error();
4568 ERR("Health error occurred in %s", __func__);
4569 }
4570
4571 health_unregister(health_sessiond);
4572
4573 DBG("Client thread dying");
4574
4575 rcu_unregister_thread();
4576
4577 /*
4578 * Since we are creating the consumer threads, we own them, so we need
4579 * to join them before our thread exits.
4580 */
4581 ret = join_consumer_thread(&kconsumer_data);
4582 if (ret) {
4583 errno = ret;
4584 PERROR("join_consumer");
4585 }
4586
4587 ret = join_consumer_thread(&ustconsumer32_data);
4588 if (ret) {
4589 errno = ret;
4590 PERROR("join_consumer ust32");
4591 }
4592
4593 ret = join_consumer_thread(&ustconsumer64_data);
4594 if (ret) {
4595 errno = ret;
4596 PERROR("join_consumer ust64");
4597 }
4598 return NULL;
4599 }
4600
4601 static int string_match(const char *str1, const char *str2)
4602 {
4603 return (str1 && str2) && !strcmp(str1, str2);
4604 }
4605
4606 /*
4607 * Take an option from the getopt output and set it in the right variable to be
4608 * used later.
4609 *
4610 * Return 0 on success else a negative value.
4611 */
4612 static int set_option(int opt, const char *arg, const char *optname)
4613 {
4614 int ret = 0;
4615
4616 if (string_match(optname, "client-sock") || opt == 'c') {
4617 if (!arg || *arg == '\0') {
4618 ret = -EINVAL;
4619 goto end;
4620 }
4621 if (lttng_is_setuid_setgid()) {
4622 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4623 "-c, --client-sock");
4624 } else {
4625 config_string_set(&config.client_unix_sock_path,
4626 strdup(arg));
4627 if (!config.client_unix_sock_path.value) {
4628 ret = -ENOMEM;
4629 PERROR("strdup");
4630 }
4631 }
4632 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4633 if (!arg || *arg == '\0') {
4634 ret = -EINVAL;
4635 goto end;
4636 }
4637 if (lttng_is_setuid_setgid()) {
4638 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4639 "-a, --apps-sock");
4640 } else {
4641 config_string_set(&config.apps_unix_sock_path,
4642 strdup(arg));
4643 if (!config.apps_unix_sock_path.value) {
4644 ret = -ENOMEM;
4645 PERROR("strdup");
4646 }
4647 }
4648 } else if (string_match(optname, "daemonize") || opt == 'd') {
4649 config.daemonize = true;
4650 } else if (string_match(optname, "background") || opt == 'b') {
4651 config.background = true;
4652 } else if (string_match(optname, "group") || opt == 'g') {
4653 if (!arg || *arg == '\0') {
4654 ret = -EINVAL;
4655 goto end;
4656 }
4657 if (lttng_is_setuid_setgid()) {
4658 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4659 "-g, --group");
4660 } else {
4661 config_string_set(&config.tracing_group_name,
4662 strdup(arg));
4663 if (!config.tracing_group_name.value) {
4664 ret = -ENOMEM;
4665 PERROR("strdup");
4666 }
4667 }
4668 } else if (string_match(optname, "help") || opt == 'h') {
4669 ret = utils_show_help(8, "lttng-sessiond", help_msg);
4670 if (ret) {
4671 ERR("Cannot show --help for `lttng-sessiond`");
4672 perror("exec");
4673 }
4674 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4675 } else if (string_match(optname, "version") || opt == 'V') {
4676 fprintf(stdout, "%s\n", VERSION);
4677 exit(EXIT_SUCCESS);
4678 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4679 config.sig_parent = true;
4680 } else if (string_match(optname, "kconsumerd-err-sock")) {
4681 if (!arg || *arg == '\0') {
4682 ret = -EINVAL;
4683 goto end;
4684 }
4685 if (lttng_is_setuid_setgid()) {
4686 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4687 "--kconsumerd-err-sock");
4688 } else {
4689 config_string_set(&config.kconsumerd_err_unix_sock_path,
4690 strdup(arg));
4691 if (!config.kconsumerd_err_unix_sock_path.value) {
4692 ret = -ENOMEM;
4693 PERROR("strdup");
4694 }
4695 }
4696 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4697 if (!arg || *arg == '\0') {
4698 ret = -EINVAL;
4699 goto end;
4700 }
4701 if (lttng_is_setuid_setgid()) {
4702 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4703 "--kconsumerd-cmd-sock");
4704 } else {
4705 config_string_set(&config.kconsumerd_cmd_unix_sock_path,
4706 strdup(arg));
4707 if (!config.kconsumerd_cmd_unix_sock_path.value) {
4708 ret = -ENOMEM;
4709 PERROR("strdup");
4710 }
4711 }
4712 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4713 if (!arg || *arg == '\0') {
4714 ret = -EINVAL;
4715 goto end;
4716 }
4717 if (lttng_is_setuid_setgid()) {
4718 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4719 "--ustconsumerd64-err-sock");
4720 } else {
4721 config_string_set(&config.consumerd64_err_unix_sock_path,
4722 strdup(arg));
4723 if (!config.consumerd64_err_unix_sock_path.value) {
4724 ret = -ENOMEM;
4725 PERROR("strdup");
4726 }
4727 }
4728 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4729 if (!arg || *arg == '\0') {
4730 ret = -EINVAL;
4731 goto end;
4732 }
4733 if (lttng_is_setuid_setgid()) {
4734 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4735 "--ustconsumerd64-cmd-sock");
4736 } else {
4737 config_string_set(&config.consumerd64_cmd_unix_sock_path,
4738 strdup(arg));
4739 if (!config.consumerd64_cmd_unix_sock_path.value) {
4740 ret = -ENOMEM;
4741 PERROR("strdup");
4742 }
4743 }
4744 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4745 if (!arg || *arg == '\0') {
4746 ret = -EINVAL;
4747 goto end;
4748 }
4749 if (lttng_is_setuid_setgid()) {
4750 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4751 "--ustconsumerd32-err-sock");
4752 } else {
4753 config_string_set(&config.consumerd32_err_unix_sock_path,
4754 strdup(arg));
4755 if (!config.consumerd32_err_unix_sock_path.value) {
4756 ret = -ENOMEM;
4757 PERROR("strdup");
4758 }
4759 }
4760 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4761 if (!arg || *arg == '\0') {
4762 ret = -EINVAL;
4763 goto end;
4764 }
4765 if (lttng_is_setuid_setgid()) {
4766 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4767 "--ustconsumerd32-cmd-sock");
4768 } else {
4769 config_string_set(&config.consumerd32_cmd_unix_sock_path,
4770 strdup(arg));
4771 if (!config.consumerd32_cmd_unix_sock_path.value) {
4772 ret = -ENOMEM;
4773 PERROR("strdup");
4774 }
4775 }
4776 } else if (string_match(optname, "no-kernel")) {
4777 config.no_kernel = true;
4778 } else if (string_match(optname, "quiet") || opt == 'q') {
4779 config.quiet = true;
4780 } else if (string_match(optname, "verbose") || opt == 'v') {
4781 /* Verbose level can increase using multiple -v */
4782 if (arg) {
4783 /* Value obtained from config file */
4784 config.verbose = config_parse_value(arg);
4785 } else {
4786 /* -v used on command line */
4787 config.verbose++;
4788 }
4789 /* Clamp value to [0, 3] */
4790 config.verbose = config.verbose < 0 ? 0 :
4791 (config.verbose <= 3 ? config.verbose : 3);
4792 } else if (string_match(optname, "verbose-consumer")) {
4793 if (arg) {
4794 config.verbose_consumer = config_parse_value(arg);
4795 } else {
4796 config.verbose_consumer++;
4797 }
4798 } else if (string_match(optname, "consumerd32-path")) {
4799 if (!arg || *arg == '\0') {
4800 ret = -EINVAL;
4801 goto end;
4802 }
4803 if (lttng_is_setuid_setgid()) {
4804 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4805 "--consumerd32-path");
4806 } else {
4807 config_string_set(&config.consumerd32_bin_path,
4808 strdup(arg));
4809 if (!config.consumerd32_bin_path.value) {
4810 PERROR("strdup");
4811 ret = -ENOMEM;
4812 }
4813 }
4814 } else if (string_match(optname, "consumerd32-libdir")) {
4815 if (!arg || *arg == '\0') {
4816 ret = -EINVAL;
4817 goto end;
4818 }
4819 if (lttng_is_setuid_setgid()) {
4820 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4821 "--consumerd32-libdir");
4822 } else {
4823 config_string_set(&config.consumerd32_lib_dir,
4824 strdup(arg));
4825 if (!config.consumerd32_lib_dir.value) {
4826 PERROR("strdup");
4827 ret = -ENOMEM;
4828 }
4829 }
4830 } else if (string_match(optname, "consumerd64-path")) {
4831 if (!arg || *arg == '\0') {
4832 ret = -EINVAL;
4833 goto end;
4834 }
4835 if (lttng_is_setuid_setgid()) {
4836 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4837 "--consumerd64-path");
4838 } else {
4839 config_string_set(&config.consumerd64_bin_path,
4840 strdup(arg));
4841 if (!config.consumerd64_bin_path.value) {
4842 PERROR("strdup");
4843 ret = -ENOMEM;
4844 }
4845 }
4846 } else if (string_match(optname, "consumerd64-libdir")) {
4847 if (!arg || *arg == '\0') {
4848 ret = -EINVAL;
4849 goto end;
4850 }
4851 if (lttng_is_setuid_setgid()) {
4852 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4853 "--consumerd64-libdir");
4854 } else {
4855 config_string_set(&config.consumerd64_lib_dir,
4856 strdup(arg));
4857 if (!config.consumerd64_lib_dir.value) {
4858 PERROR("strdup");
4859 ret = -ENOMEM;
4860 }
4861 }
4862 } else if (string_match(optname, "pidfile") || opt == 'p') {
4863 if (!arg || *arg == '\0') {
4864 ret = -EINVAL;
4865 goto end;
4866 }
4867 if (lttng_is_setuid_setgid()) {
4868 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4869 "-p, --pidfile");
4870 } else {
4871 config_string_set(&config.pid_file_path, strdup(arg));
4872 if (!config.pid_file_path.value) {
4873 PERROR("strdup");
4874 ret = -ENOMEM;
4875 }
4876 }
4877 } else if (string_match(optname, "agent-tcp-port")) {
4878 if (!arg || *arg == '\0') {
4879 ret = -EINVAL;
4880 goto end;
4881 }
4882 if (lttng_is_setuid_setgid()) {
4883 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4884 "--agent-tcp-port");
4885 } else {
4886 unsigned long v;
4887
4888 errno = 0;
4889 v = strtoul(arg, NULL, 0);
4890 if (errno != 0 || !isdigit(arg[0])) {
4891 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4892 return -1;
4893 }
4894 if (v == 0 || v >= 65535) {
4895 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4896 return -1;
4897 }
4898 config.agent_tcp_port.begin = config.agent_tcp_port.end = (int) v;
4899 DBG3("Agent TCP port set to non default: %i", (int) v);
4900 }
4901 } else if (string_match(optname, "load") || opt == 'l') {
4902 if (!arg || *arg == '\0') {
4903 ret = -EINVAL;
4904 goto end;
4905 }
4906 if (lttng_is_setuid_setgid()) {
4907 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4908 "-l, --load");
4909 } else {
4910 config_string_set(&config.load_session_path, strdup(arg));
4911 if (!config.load_session_path.value) {
4912 PERROR("strdup");
4913 ret = -ENOMEM;
4914 }
4915 }
4916 } else if (string_match(optname, "kmod-probes")) {
4917 if (!arg || *arg == '\0') {
4918 ret = -EINVAL;
4919 goto end;
4920 }
4921 if (lttng_is_setuid_setgid()) {
4922 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4923 "--kmod-probes");
4924 } else {
4925 config_string_set(&config.kmod_probes_list, strdup(arg));
4926 if (!config.kmod_probes_list.value) {
4927 PERROR("strdup");
4928 ret = -ENOMEM;
4929 }
4930 }
4931 } else if (string_match(optname, "extra-kmod-probes")) {
4932 if (!arg || *arg == '\0') {
4933 ret = -EINVAL;
4934 goto end;
4935 }
4936 if (lttng_is_setuid_setgid()) {
4937 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4938 "--extra-kmod-probes");
4939 } else {
4940 config_string_set(&config.kmod_extra_probes_list,
4941 strdup(arg));
4942 if (!config.kmod_extra_probes_list.value) {
4943 PERROR("strdup");
4944 ret = -ENOMEM;
4945 }
4946 }
4947 } else if (string_match(optname, "config") || opt == 'f') {
4948 /* This is handled in set_options() thus silent skip. */
4949 goto end;
4950 } else {
4951 /* Unknown option or other error.
4952 * Error is printed by getopt, just return */
4953 ret = -1;
4954 }
4955
4956 end:
4957 if (ret == -EINVAL) {
4958 const char *opt_name = "unknown";
4959 int i;
4960
4961 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4962 i++) {
4963 if (opt == long_options[i].val) {
4964 opt_name = long_options[i].name;
4965 break;
4966 }
4967 }
4968
4969 WARN("Invalid argument provided for option \"%s\", using default value.",
4970 opt_name);
4971 }
4972
4973 return ret;
4974 }
4975
4976 /*
4977 * config_entry_handler_cb used to handle options read from a config file.
4978 * See config_entry_handler_cb comment in common/config/session-config.h for the
4979 * return value conventions.
4980 */
4981 static int config_entry_handler(const struct config_entry *entry, void *unused)
4982 {
4983 int ret = 0, i;
4984
4985 if (!entry || !entry->name || !entry->value) {
4986 ret = -EINVAL;
4987 goto end;
4988 }
4989
4990 /* Check if the option is to be ignored */
4991 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4992 if (!strcmp(entry->name, config_ignore_options[i])) {
4993 goto end;
4994 }
4995 }
4996
4997 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4998 i++) {
4999
5000 /* Ignore if not fully matched. */
5001 if (strcmp(entry->name, long_options[i].name)) {
5002 continue;
5003 }
5004
5005 /*
5006 * If the option takes no argument on the command line, we have to
5007 * check if the value is "true". We support non-zero numeric values,
5008 * true, on and yes.
5009 */
5010 if (!long_options[i].has_arg) {
5011 ret = config_parse_value(entry->value);
5012 if (ret <= 0) {
5013 if (ret) {
5014 WARN("Invalid configuration value \"%s\" for option %s",
5015 entry->value, entry->name);
5016 }
5017 /* False, skip boolean config option. */
5018 goto end;
5019 }
5020 }
5021
5022 ret = set_option(long_options[i].val, entry->value, entry->name);
5023 goto end;
5024 }
5025
5026 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5027
5028 end:
5029 return ret;
5030 }
5031
5032 /*
5033 * daemon configuration loading and argument parsing
5034 */
5035 static int set_options(int argc, char **argv)
5036 {
5037 int ret = 0, c = 0, option_index = 0;
5038 int orig_optopt = optopt, orig_optind = optind;
5039 char *optstring;
5040 const char *config_path = NULL;
5041
5042 optstring = utils_generate_optstring(long_options,
5043 sizeof(long_options) / sizeof(struct option));
5044 if (!optstring) {
5045 ret = -ENOMEM;
5046 goto end;
5047 }
5048
5049 /* Check for the --config option */
5050 while ((c = getopt_long(argc, argv, optstring, long_options,
5051 &option_index)) != -1) {
5052 if (c == '?') {
5053 ret = -EINVAL;
5054 goto end;
5055 } else if (c != 'f') {
5056 /* if not equal to --config option. */
5057 continue;
5058 }
5059
5060 if (lttng_is_setuid_setgid()) {
5061 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5062 "-f, --config");
5063 } else {
5064 config_path = utils_expand_path(optarg);
5065 if (!config_path) {
5066 ERR("Failed to resolve path: %s", optarg);
5067 }
5068 }
5069 }
5070
5071 ret = config_get_section_entries(config_path, config_section_name,
5072 config_entry_handler, NULL);
5073 if (ret) {
5074 if (ret > 0) {
5075 ERR("Invalid configuration option at line %i", ret);
5076 ret = -1;
5077 }
5078 goto end;
5079 }
5080
5081 /* Reset getopt's global state */
5082 optopt = orig_optopt;
5083 optind = orig_optind;
5084 while (1) {
5085 option_index = -1;
5086 /*
5087 * getopt_long() will not set option_index if it encounters a
5088 * short option.
5089 */
5090 c = getopt_long(argc, argv, optstring, long_options,
5091 &option_index);
5092 if (c == -1) {
5093 break;
5094 }
5095
5096 /*
5097 * Pass NULL as the long option name if popt left the index
5098 * unset.
5099 */
5100 ret = set_option(c, optarg,
5101 option_index < 0 ? NULL :
5102 long_options[option_index].name);
5103 if (ret < 0) {
5104 break;
5105 }
5106 }
5107
5108 end:
5109 free(optstring);
5110 return ret;
5111 }
5112
5113 /*
5114 * Creates the two needed socket by the daemon.
5115 * apps_sock - The communication socket for all UST apps.
5116 * client_sock - The communication of the cli tool (lttng).
5117 */
5118 static int init_daemon_socket(void)
5119 {
5120 int ret = 0;
5121 mode_t old_umask;
5122
5123 old_umask = umask(0);
5124
5125 /* Create client tool unix socket */
5126 client_sock = lttcomm_create_unix_sock(config.client_unix_sock_path.value);
5127 if (client_sock < 0) {
5128 ERR("Create unix sock failed: %s", config.client_unix_sock_path.value);
5129 ret = -1;
5130 goto end;
5131 }
5132
5133 /* Set the cloexec flag */
5134 ret = utils_set_fd_cloexec(client_sock);
5135 if (ret < 0) {
5136 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5137 "Continuing but note that the consumer daemon will have a "
5138 "reference to this socket on exec()", client_sock);
5139 }
5140
5141 /* File permission MUST be 660 */
5142 ret = chmod(config.client_unix_sock_path.value, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5143 if (ret < 0) {
5144 ERR("Set file permissions failed: %s", config.client_unix_sock_path.value);
5145 PERROR("chmod");
5146 goto end;
5147 }
5148
5149 /* Create the application unix socket */
5150 apps_sock = lttcomm_create_unix_sock(config.apps_unix_sock_path.value);
5151 if (apps_sock < 0) {
5152 ERR("Create unix sock failed: %s", config.apps_unix_sock_path.value);
5153 ret = -1;
5154 goto end;
5155 }
5156
5157 /* Set the cloexec flag */
5158 ret = utils_set_fd_cloexec(apps_sock);
5159 if (ret < 0) {
5160 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5161 "Continuing but note that the consumer daemon will have a "
5162 "reference to this socket on exec()", apps_sock);
5163 }
5164
5165 /* File permission MUST be 666 */
5166 ret = chmod(config.apps_unix_sock_path.value,
5167 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5168 if (ret < 0) {
5169 ERR("Set file permissions failed: %s", config.apps_unix_sock_path.value);
5170 PERROR("chmod");
5171 goto end;
5172 }
5173
5174 DBG3("Session daemon client socket %d and application socket %d created",
5175 client_sock, apps_sock);
5176
5177 end:
5178 umask(old_umask);
5179 return ret;
5180 }
5181
5182 /*
5183 * Create lockfile using the rundir and return its fd.
5184 */
5185 static int create_lockfile(void)
5186 {
5187 return utils_create_lock_file(config.lock_file_path.value);
5188 }
5189
5190 /*
5191 * Check if the global socket is available, and if a daemon is answering at the
5192 * other side. If yes, error is returned.
5193 *
5194 * Also attempts to create and hold the lock file.
5195 */
5196 static int check_existing_daemon(void)
5197 {
5198 int ret = 0;
5199
5200 /* Is there anybody out there ? */
5201 if (lttng_session_daemon_alive()) {
5202 ret = -EEXIST;
5203 goto end;
5204 }
5205
5206 lockfile_fd = create_lockfile();
5207 if (lockfile_fd < 0) {
5208 ret = -EEXIST;
5209 goto end;
5210 }
5211 end:
5212 return ret;
5213 }
5214
5215 static void sessiond_cleanup_lock_file(void)
5216 {
5217 int ret;
5218
5219 /*
5220 * Cleanup lock file by deleting it and finaly closing it which will
5221 * release the file system lock.
5222 */
5223 if (lockfile_fd >= 0) {
5224 ret = remove(config.lock_file_path.value);
5225 if (ret < 0) {
5226 PERROR("remove lock file");
5227 }
5228 ret = close(lockfile_fd);
5229 if (ret < 0) {
5230 PERROR("close lock file");
5231 }
5232 }
5233 }
5234
5235 /*
5236 * Set the tracing group gid onto the client socket.
5237 *
5238 * Race window between mkdir and chown is OK because we are going from more
5239 * permissive (root.root) to less permissive (root.tracing).
5240 */
5241 static int set_permissions(char *rundir)
5242 {
5243 int ret;
5244 gid_t gid;
5245
5246 gid = utils_get_group_id(config.tracing_group_name.value);
5247
5248 /* Set lttng run dir */
5249 ret = chown(rundir, 0, gid);
5250 if (ret < 0) {
5251 ERR("Unable to set group on %s", rundir);
5252 PERROR("chown");
5253 }
5254
5255 /*
5256 * Ensure all applications and tracing group can search the run
5257 * dir. Allow everyone to read the directory, since it does not
5258 * buy us anything to hide its content.
5259 */
5260 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5261 if (ret < 0) {
5262 ERR("Unable to set permissions on %s", rundir);
5263 PERROR("chmod");
5264 }
5265
5266 /* lttng client socket path */
5267 ret = chown(config.client_unix_sock_path.value, 0, gid);
5268 if (ret < 0) {
5269 ERR("Unable to set group on %s", config.client_unix_sock_path.value);
5270 PERROR("chown");
5271 }
5272
5273 /* kconsumer error socket path */
5274 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5275 if (ret < 0) {
5276 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5277 PERROR("chown");
5278 }
5279
5280 /* 64-bit ustconsumer error socket path */
5281 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5282 if (ret < 0) {
5283 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5284 PERROR("chown");
5285 }
5286
5287 /* 32-bit ustconsumer compat32 error socket path */
5288 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5289 if (ret < 0) {
5290 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5291 PERROR("chown");
5292 }
5293
5294 DBG("All permissions are set");
5295
5296 return ret;
5297 }
5298
5299 /*
5300 * Create the lttng run directory needed for all global sockets and pipe.
5301 */
5302 static int create_lttng_rundir(void)
5303 {
5304 int ret;
5305
5306 DBG3("Creating LTTng run directory: %s", config.rundir.value);
5307
5308 ret = mkdir(config.rundir.value, S_IRWXU);
5309 if (ret < 0) {
5310 if (errno != EEXIST) {
5311 ERR("Unable to create %s", config.rundir.value);
5312 goto error;
5313 } else {
5314 ret = 0;
5315 }
5316 }
5317
5318 error:
5319 return ret;
5320 }
5321
5322 /*
5323 * Setup sockets and directory needed by the consumerds' communication with the
5324 * session daemon.
5325 */
5326 static int set_consumer_sockets(struct consumer_data *consumer_data)
5327 {
5328 int ret;
5329 char *path = NULL;
5330
5331 switch (consumer_data->type) {
5332 case LTTNG_CONSUMER_KERNEL:
5333 path = config.kconsumerd_path.value;
5334 break;
5335 case LTTNG_CONSUMER64_UST:
5336 path = config.consumerd64_path.value;
5337 break;
5338 case LTTNG_CONSUMER32_UST:
5339 path = config.consumerd32_path.value;
5340 break;
5341 default:
5342 ERR("Consumer type unknown");
5343 ret = -EINVAL;
5344 goto error;
5345 }
5346 assert(path);
5347
5348 DBG2("Creating consumer directory: %s", path);
5349
5350 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5351 if (ret < 0 && errno != EEXIST) {
5352 PERROR("mkdir");
5353 ERR("Failed to create %s", path);
5354 goto error;
5355 }
5356 if (is_root) {
5357 ret = chown(path, 0, utils_get_group_id(config.tracing_group_name.value));
5358 if (ret < 0) {
5359 ERR("Unable to set group on %s", path);
5360 PERROR("chown");
5361 goto error;
5362 }
5363 }
5364
5365 /* Create the consumerd error unix socket */
5366 consumer_data->err_sock =
5367 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5368 if (consumer_data->err_sock < 0) {
5369 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5370 ret = -1;
5371 goto error;
5372 }
5373
5374 /*
5375 * Set the CLOEXEC flag. Return code is useless because either way, the
5376 * show must go on.
5377 */
5378 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5379 if (ret < 0) {
5380 PERROR("utils_set_fd_cloexec");
5381 /* continue anyway */
5382 }
5383
5384 /* File permission MUST be 660 */
5385 ret = chmod(consumer_data->err_unix_sock_path,
5386 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5387 if (ret < 0) {
5388 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5389 PERROR("chmod");
5390 goto error;
5391 }
5392
5393 error:
5394 return ret;
5395 }
5396
5397 /*
5398 * Signal handler for the daemon
5399 *
5400 * Simply stop all worker threads, leaving main() return gracefully after
5401 * joining all threads and calling cleanup().
5402 */
5403 static void sighandler(int sig)
5404 {
5405 switch (sig) {
5406 case SIGINT:
5407 DBG("SIGINT caught");
5408 stop_threads();
5409 break;
5410 case SIGTERM:
5411 DBG("SIGTERM caught");
5412 stop_threads();
5413 break;
5414 case SIGUSR1:
5415 CMM_STORE_SHARED(recv_child_signal, 1);
5416 break;
5417 default:
5418 break;
5419 }
5420 }
5421
5422 /*
5423 * Setup signal handler for :
5424 * SIGINT, SIGTERM, SIGPIPE
5425 */
5426 static int set_signal_handler(void)
5427 {
5428 int ret = 0;
5429 struct sigaction sa;
5430 sigset_t sigset;
5431
5432 if ((ret = sigemptyset(&sigset)) < 0) {
5433 PERROR("sigemptyset");
5434 return ret;
5435 }
5436
5437 sa.sa_mask = sigset;
5438 sa.sa_flags = 0;
5439
5440 sa.sa_handler = sighandler;
5441 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5442 PERROR("sigaction");
5443 return ret;
5444 }
5445
5446 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5447 PERROR("sigaction");
5448 return ret;
5449 }
5450
5451 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5452 PERROR("sigaction");
5453 return ret;
5454 }
5455
5456 sa.sa_handler = SIG_IGN;
5457 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5458 PERROR("sigaction");
5459 return ret;
5460 }
5461
5462 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5463
5464 return ret;
5465 }
5466
5467 /*
5468 * Set open files limit to unlimited. This daemon can open a large number of
5469 * file descriptors in order to consume multiple kernel traces.
5470 */
5471 static void set_ulimit(void)
5472 {
5473 int ret;
5474 struct rlimit lim;
5475
5476 /* The kernel does not allow an infinite limit for open files */
5477 lim.rlim_cur = 65535;
5478 lim.rlim_max = 65535;
5479
5480 ret = setrlimit(RLIMIT_NOFILE, &lim);
5481 if (ret < 0) {
5482 PERROR("failed to set open files limit");
5483 }
5484 }
5485
5486 static int write_pidfile(void)
5487 {
5488 return utils_create_pid_file(getpid(), config.pid_file_path.value);
5489 }
5490
5491 static int set_clock_plugin_env(void)
5492 {
5493 int ret = 0;
5494 char *env_value = NULL;
5495
5496 if (!config.lttng_ust_clock_plugin.value) {
5497 goto end;
5498 }
5499
5500 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
5501 config.lttng_ust_clock_plugin.value);
5502 if (ret < 0) {
5503 PERROR("asprintf");
5504 goto end;
5505 }
5506
5507 ret = putenv(env_value);
5508 if (ret) {
5509 free(env_value);
5510 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5511 goto end;
5512 }
5513
5514 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5515 config.lttng_ust_clock_plugin.value);
5516 end:
5517 return ret;
5518 }
5519
5520 /*
5521 * main
5522 */
5523 int main(int argc, char **argv)
5524 {
5525 int ret = 0, retval = 0;
5526 void *status;
5527 const char *env_app_timeout;
5528 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5529 *ust64_channel_monitor_pipe = NULL,
5530 *kernel_channel_monitor_pipe = NULL;
5531 bool notification_thread_running = false;
5532
5533 init_kernel_workarounds();
5534
5535 rcu_register_thread();
5536
5537 if (set_signal_handler()) {
5538 retval = -1;
5539 goto exit_set_signal_handler;
5540 }
5541
5542 page_size = sysconf(_SC_PAGESIZE);
5543 if (page_size < 0) {
5544 PERROR("sysconf _SC_PAGESIZE");
5545 page_size = LONG_MAX;
5546 WARN("Fallback page size to %ld", page_size);
5547 }
5548
5549 ret = sessiond_config_init(&config);
5550 if (ret) {
5551 retval = -1;
5552 goto exit_set_signal_handler;
5553 }
5554
5555 /*
5556 * Init config from environment variables.
5557 * Command line option override env configuration per-doc. Do env first.
5558 */
5559 sessiond_config_apply_env_config(&config);
5560
5561 /*
5562 * Parse arguments and load the daemon configuration file.
5563 *
5564 * We have an exit_options exit path to free memory reserved by
5565 * set_options. This is needed because the rest of sessiond_cleanup()
5566 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5567 * depends on set_options.
5568 */
5569 progname = argv[0];
5570 if (set_options(argc, argv)) {
5571 retval = -1;
5572 goto exit_options;
5573 }
5574
5575 /*
5576 * Resolve all paths received as arguments, configuration option, or
5577 * through environment variable as absolute paths. This is necessary
5578 * since daemonizing causes the sessiond's current working directory
5579 * to '/'.
5580 */
5581 ret = sessiond_config_resolve_paths(&config);
5582 if (ret) {
5583 goto exit_options;
5584 }
5585
5586 /* Apply config. */
5587 lttng_opt_verbose = config.verbose;
5588 lttng_opt_quiet = config.quiet;
5589 kconsumer_data.err_unix_sock_path =
5590 config.kconsumerd_err_unix_sock_path.value;
5591 kconsumer_data.cmd_unix_sock_path =
5592 config.kconsumerd_cmd_unix_sock_path.value;
5593 ustconsumer32_data.err_unix_sock_path =
5594 config.consumerd32_err_unix_sock_path.value;
5595 ustconsumer32_data.cmd_unix_sock_path =
5596 config.consumerd32_cmd_unix_sock_path.value;
5597 ustconsumer64_data.err_unix_sock_path =
5598 config.consumerd64_err_unix_sock_path.value;
5599 ustconsumer64_data.cmd_unix_sock_path =
5600 config.consumerd64_cmd_unix_sock_path.value;
5601 set_clock_plugin_env();
5602
5603 sessiond_config_log(&config);
5604
5605 if (create_lttng_rundir()) {
5606 retval = -1;
5607 goto exit_options;
5608 }
5609
5610 /* Abort launch if a session daemon is already running. */
5611 if (check_existing_daemon()) {
5612 ERR("A session daemon is already running.");
5613 retval = -1;
5614 goto exit_options;
5615 }
5616
5617 /* Daemonize */
5618 if (config.daemonize || config.background) {
5619 int i;
5620
5621 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5622 !config.background);
5623 if (ret < 0) {
5624 retval = -1;
5625 goto exit_options;
5626 }
5627
5628 /*
5629 * We are in the child. Make sure all other file descriptors are
5630 * closed, in case we are called with more opened file
5631 * descriptors than the standard ones and the lock file.
5632 */
5633 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5634 if (i == lockfile_fd) {
5635 continue;
5636 }
5637 (void) close(i);
5638 }
5639 }
5640
5641 if (run_as_create_worker(argv[0]) < 0) {
5642 goto exit_create_run_as_worker_cleanup;
5643 }
5644
5645 /*
5646 * Starting from here, we can create threads. This needs to be after
5647 * lttng_daemonize due to RCU.
5648 */
5649
5650 /*
5651 * Initialize the health check subsystem. This call should set the
5652 * appropriate time values.
5653 */
5654 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5655 if (!health_sessiond) {
5656 PERROR("health_app_create error");
5657 retval = -1;
5658 goto exit_health_sessiond_cleanup;
5659 }
5660
5661 /* Create thread to clean up RCU hash tables */
5662 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
5663 retval = -1;
5664 goto exit_ht_cleanup;
5665 }
5666
5667 /* Create thread quit pipe */
5668 if (init_thread_quit_pipe()) {
5669 retval = -1;
5670 goto exit_init_data;
5671 }
5672
5673 /* Check if daemon is UID = 0 */
5674 is_root = !getuid();
5675 if (is_root) {
5676 /* Create global run dir with root access */
5677
5678 kernel_channel_monitor_pipe = lttng_pipe_open(0);
5679 if (!kernel_channel_monitor_pipe) {
5680 ERR("Failed to create kernel consumer channel monitor pipe");
5681 retval = -1;
5682 goto exit_init_data;
5683 }
5684 kconsumer_data.channel_monitor_pipe =
5685 lttng_pipe_release_writefd(
5686 kernel_channel_monitor_pipe);
5687 if (kconsumer_data.channel_monitor_pipe < 0) {
5688 retval = -1;
5689 goto exit_init_data;
5690 }
5691 }
5692
5693 /* Set consumer initial state */
5694 kernel_consumerd_state = CONSUMER_STOPPED;
5695 ust_consumerd_state = CONSUMER_STOPPED;
5696
5697 ust32_channel_monitor_pipe = lttng_pipe_open(0);
5698 if (!ust32_channel_monitor_pipe) {
5699 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5700 retval = -1;
5701 goto exit_init_data;
5702 }
5703 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5704 ust32_channel_monitor_pipe);
5705 if (ustconsumer32_data.channel_monitor_pipe < 0) {
5706 retval = -1;
5707 goto exit_init_data;
5708 }
5709
5710 ust64_channel_monitor_pipe = lttng_pipe_open(0);
5711 if (!ust64_channel_monitor_pipe) {
5712 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5713 retval = -1;
5714 goto exit_init_data;
5715 }
5716 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5717 ust64_channel_monitor_pipe);
5718 if (ustconsumer64_data.channel_monitor_pipe < 0) {
5719 retval = -1;
5720 goto exit_init_data;
5721 }
5722
5723 /*
5724 * Init UST app hash table. Alloc hash table before this point since
5725 * cleanup() can get called after that point.
5726 */
5727 if (ust_app_ht_alloc()) {
5728 ERR("Failed to allocate UST app hash table");
5729 retval = -1;
5730 goto exit_init_data;
5731 }
5732
5733 /*
5734 * Initialize agent app hash table. We allocate the hash table here
5735 * since cleanup() can get called after this point.
5736 */
5737 if (agent_app_ht_alloc()) {
5738 ERR("Failed to allocate Agent app hash table");
5739 retval = -1;
5740 goto exit_init_data;
5741 }
5742
5743 /*
5744 * These actions must be executed as root. We do that *after* setting up
5745 * the sockets path because we MUST make the check for another daemon using
5746 * those paths *before* trying to set the kernel consumer sockets and init
5747 * kernel tracer.
5748 */
5749 if (is_root) {
5750 if (set_consumer_sockets(&kconsumer_data)) {
5751 retval = -1;
5752 goto exit_init_data;
5753 }
5754
5755 /* Setup kernel tracer */
5756 if (!config.no_kernel) {
5757 init_kernel_tracer();
5758 if (kernel_tracer_fd >= 0) {
5759 ret = syscall_init_table();
5760 if (ret < 0) {
5761 ERR("Unable to populate syscall table. "
5762 "Syscall tracing won't work "
5763 "for this session daemon.");
5764 }
5765 }
5766 }
5767
5768 /* Set ulimit for open files */
5769 set_ulimit();
5770 }
5771 /* init lttng_fd tracking must be done after set_ulimit. */
5772 lttng_fd_init();
5773
5774 if (set_consumer_sockets(&ustconsumer64_data)) {
5775 retval = -1;
5776 goto exit_init_data;
5777 }
5778
5779 if (set_consumer_sockets(&ustconsumer32_data)) {
5780 retval = -1;
5781 goto exit_init_data;
5782 }
5783
5784 /* Setup the needed unix socket */
5785 if (init_daemon_socket()) {
5786 retval = -1;
5787 goto exit_init_data;
5788 }
5789
5790 /* Set credentials to socket */
5791 if (is_root && set_permissions(config.rundir.value)) {
5792 retval = -1;
5793 goto exit_init_data;
5794 }
5795
5796 /* Get parent pid if -S, --sig-parent is specified. */
5797 if (config.sig_parent) {
5798 ppid = getppid();
5799 }
5800
5801 /* Setup the kernel pipe for waking up the kernel thread */
5802 if (is_root && !config.no_kernel) {
5803 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5804 retval = -1;
5805 goto exit_init_data;
5806 }
5807 }
5808
5809 /* Setup the thread apps communication pipe. */
5810 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5811 retval = -1;
5812 goto exit_init_data;
5813 }
5814
5815 /* Setup the thread apps notify communication pipe. */
5816 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5817 retval = -1;
5818 goto exit_init_data;
5819 }
5820
5821 /* Initialize global buffer per UID and PID registry. */
5822 buffer_reg_init_uid_registry();
5823 buffer_reg_init_pid_registry();
5824
5825 /* Init UST command queue. */
5826 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5827
5828 /*
5829 * Get session list pointer. This pointer MUST NOT be free'd. This list
5830 * is statically declared in session.c
5831 */
5832 session_list_ptr = session_get_list();
5833
5834 cmd_init();
5835
5836 /* Check for the application socket timeout env variable. */
5837 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5838 if (env_app_timeout) {
5839 config.app_socket_timeout = atoi(env_app_timeout);
5840 } else {
5841 config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5842 }
5843
5844 ret = write_pidfile();
5845 if (ret) {
5846 ERR("Error in write_pidfile");
5847 retval = -1;
5848 goto exit_init_data;
5849 }
5850
5851 /* Initialize communication library */
5852 lttcomm_init();
5853 /* Initialize TCP timeout values */
5854 lttcomm_inet_init();
5855
5856 if (load_session_init_data(&load_info) < 0) {
5857 retval = -1;
5858 goto exit_init_data;
5859 }
5860 load_info->path = config.load_session_path.value;
5861
5862 /* Create health-check thread. */
5863 ret = pthread_create(&health_thread, default_pthread_attr(),
5864 thread_manage_health, (void *) NULL);
5865 if (ret) {
5866 errno = ret;
5867 PERROR("pthread_create health");
5868 retval = -1;
5869 goto exit_health;
5870 }
5871
5872 /* notification_thread_data acquires the pipes' read side. */
5873 notification_thread_handle = notification_thread_handle_create(
5874 ust32_channel_monitor_pipe,
5875 ust64_channel_monitor_pipe,
5876 kernel_channel_monitor_pipe);
5877 if (!notification_thread_handle) {
5878 retval = -1;
5879 ERR("Failed to create notification thread shared data");
5880 stop_threads();
5881 goto exit_notification;
5882 }
5883
5884 /* Create notification thread. */
5885 ret = pthread_create(&notification_thread, default_pthread_attr(),
5886 thread_notification, notification_thread_handle);
5887 if (ret) {
5888 errno = ret;
5889 PERROR("pthread_create notification");
5890 retval = -1;
5891 stop_threads();
5892 goto exit_notification;
5893 }
5894 notification_thread_running = true;
5895
5896 /* Create thread to manage the client socket */
5897 ret = pthread_create(&client_thread, default_pthread_attr(),
5898 thread_manage_clients, (void *) NULL);
5899 if (ret) {
5900 errno = ret;
5901 PERROR("pthread_create clients");
5902 retval = -1;
5903 stop_threads();
5904 goto exit_client;
5905 }
5906
5907 /* Create thread to dispatch registration */
5908 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
5909 thread_dispatch_ust_registration, (void *) NULL);
5910 if (ret) {
5911 errno = ret;
5912 PERROR("pthread_create dispatch");
5913 retval = -1;
5914 stop_threads();
5915 goto exit_dispatch;
5916 }
5917
5918 /* Create thread to manage application registration. */
5919 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
5920 thread_registration_apps, (void *) NULL);
5921 if (ret) {
5922 errno = ret;
5923 PERROR("pthread_create registration");
5924 retval = -1;
5925 stop_threads();
5926 goto exit_reg_apps;
5927 }
5928
5929 /* Create thread to manage application socket */
5930 ret = pthread_create(&apps_thread, default_pthread_attr(),
5931 thread_manage_apps, (void *) NULL);
5932 if (ret) {
5933 errno = ret;
5934 PERROR("pthread_create apps");
5935 retval = -1;
5936 stop_threads();
5937 goto exit_apps;
5938 }
5939
5940 /* Create thread to manage application notify socket */
5941 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
5942 ust_thread_manage_notify, (void *) NULL);
5943 if (ret) {
5944 errno = ret;
5945 PERROR("pthread_create notify");
5946 retval = -1;
5947 stop_threads();
5948 goto exit_apps_notify;
5949 }
5950
5951 /* Create agent registration thread. */
5952 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
5953 agent_thread_manage_registration, (void *) NULL);
5954 if (ret) {
5955 errno = ret;
5956 PERROR("pthread_create agent");
5957 retval = -1;
5958 stop_threads();
5959 goto exit_agent_reg;
5960 }
5961
5962 /* Don't start this thread if kernel tracing is not requested nor root */
5963 if (is_root && !config.no_kernel) {
5964 /* Create kernel thread to manage kernel event */
5965 ret = pthread_create(&kernel_thread, default_pthread_attr(),
5966 thread_manage_kernel, (void *) NULL);
5967 if (ret) {
5968 errno = ret;
5969 PERROR("pthread_create kernel");
5970 retval = -1;
5971 stop_threads();
5972 goto exit_kernel;
5973 }
5974 }
5975
5976 /* Create session loading thread. */
5977 ret = pthread_create(&load_session_thread, default_pthread_attr(),
5978 thread_load_session, load_info);
5979 if (ret) {
5980 errno = ret;
5981 PERROR("pthread_create load_session_thread");
5982 retval = -1;
5983 stop_threads();
5984 goto exit_load_session;
5985 }
5986
5987 /*
5988 * This is where we start awaiting program completion (e.g. through
5989 * signal that asks threads to teardown).
5990 */
5991
5992 ret = pthread_join(load_session_thread, &status);
5993 if (ret) {
5994 errno = ret;
5995 PERROR("pthread_join load_session_thread");
5996 retval = -1;
5997 }
5998 exit_load_session:
5999
6000 if (is_root && !config.no_kernel) {
6001 ret = pthread_join(kernel_thread, &status);
6002 if (ret) {
6003 errno = ret;
6004 PERROR("pthread_join");
6005 retval = -1;
6006 }
6007 }
6008 exit_kernel:
6009
6010 ret = pthread_join(agent_reg_thread, &status);
6011 if (ret) {
6012 errno = ret;
6013 PERROR("pthread_join agent");
6014 retval = -1;
6015 }
6016 exit_agent_reg:
6017
6018 ret = pthread_join(apps_notify_thread, &status);
6019 if (ret) {
6020 errno = ret;
6021 PERROR("pthread_join apps notify");
6022 retval = -1;
6023 }
6024 exit_apps_notify:
6025
6026 ret = pthread_join(apps_thread, &status);
6027 if (ret) {
6028 errno = ret;
6029 PERROR("pthread_join apps");
6030 retval = -1;
6031 }
6032 exit_apps:
6033
6034 ret = pthread_join(reg_apps_thread, &status);
6035 if (ret) {
6036 errno = ret;
6037 PERROR("pthread_join");
6038 retval = -1;
6039 }
6040 exit_reg_apps:
6041
6042 /*
6043 * Join dispatch thread after joining reg_apps_thread to ensure
6044 * we don't leak applications in the queue.
6045 */
6046 ret = pthread_join(dispatch_thread, &status);
6047 if (ret) {
6048 errno = ret;
6049 PERROR("pthread_join");
6050 retval = -1;
6051 }
6052 exit_dispatch:
6053
6054 ret = pthread_join(client_thread, &status);
6055 if (ret) {
6056 errno = ret;
6057 PERROR("pthread_join");
6058 retval = -1;
6059 }
6060
6061 exit_client:
6062 exit_notification:
6063 ret = pthread_join(health_thread, &status);
6064 if (ret) {
6065 errno = ret;
6066 PERROR("pthread_join health thread");
6067 retval = -1;
6068 }
6069
6070 exit_health:
6071 exit_init_data:
6072 /*
6073 * Wait for all pending call_rcu work to complete before tearing
6074 * down data structures. call_rcu worker may be trying to
6075 * perform lookups in those structures.
6076 */
6077 rcu_barrier();
6078 /*
6079 * sessiond_cleanup() is called when no other thread is running, except
6080 * the ht_cleanup thread, which is needed to destroy the hash tables.
6081 */
6082 rcu_thread_online();
6083 sessiond_cleanup();
6084
6085 /*
6086 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6087 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6088 * the queue is empty before shutting down the clean-up thread.
6089 */
6090 rcu_barrier();
6091
6092 /*
6093 * The teardown of the notification system is performed after the
6094 * session daemon's teardown in order to allow it to be notified
6095 * of the active session and channels at the moment of the teardown.
6096 */
6097 if (notification_thread_handle) {
6098 if (notification_thread_running) {
6099 notification_thread_command_quit(
6100 notification_thread_handle);
6101 ret = pthread_join(notification_thread, &status);
6102 if (ret) {
6103 errno = ret;
6104 PERROR("pthread_join notification thread");
6105 retval = -1;
6106 }
6107 }
6108 notification_thread_handle_destroy(notification_thread_handle);
6109 }
6110
6111 rcu_thread_offline();
6112 rcu_unregister_thread();
6113
6114 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6115 if (ret) {
6116 retval = -1;
6117 }
6118 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6119 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6120 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6121 exit_ht_cleanup:
6122
6123 health_app_destroy(health_sessiond);
6124 exit_health_sessiond_cleanup:
6125 exit_create_run_as_worker_cleanup:
6126
6127 exit_options:
6128 sessiond_cleanup_lock_file();
6129 sessiond_cleanup_options();
6130
6131 exit_set_signal_handler:
6132 if (!retval) {
6133 exit(EXIT_SUCCESS);
6134 } else {
6135 exit(EXIT_FAILURE);
6136 }
6137 }
This page took 0.182207 seconds and 4 git commands to generate.