fa6fa4830e468efc0180a88e1888434f67955a25
[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 ret = remove(config.pid_file_path.value);
551 if (ret < 0) {
552 PERROR("remove pidfile %s", config.pid_file_path.value);
553 }
554
555 DBG("Removing sessiond and consumerd content of directory %s",
556 config.rundir.value);
557
558 /* sessiond */
559 DBG("Removing %s", config.pid_file_path.value);
560 (void) unlink(config.pid_file_path.value);
561
562 DBG("Removing %s", config.agent_port_file_path.value);
563 (void) unlink(config.agent_port_file_path.value);
564
565 /* kconsumerd */
566 DBG("Removing %s", kconsumer_data.err_unix_sock_path);
567 (void) unlink(kconsumer_data.err_unix_sock_path);
568
569 DBG("Removing directory %s", config.kconsumerd_path.value);
570 (void) rmdir(config.kconsumerd_path.value);
571
572 /* ust consumerd 32 */
573 DBG("Removing %s", config.consumerd32_err_unix_sock_path.value);
574 (void) unlink(config.consumerd32_err_unix_sock_path.value);
575
576 DBG("Removing directory %s", config.consumerd32_path.value);
577 (void) rmdir(config.consumerd32_path.value);
578
579 /* ust consumerd 64 */
580 DBG("Removing %s", config.consumerd64_err_unix_sock_path.value);
581 (void) unlink(config.consumerd64_err_unix_sock_path.value);
582
583 DBG("Removing directory %s", config.consumerd64_path.value);
584 (void) rmdir(config.consumerd64_path.value);
585
586 DBG("Cleaning up all sessions");
587
588 /* Destroy session list mutex */
589 if (session_list_ptr != NULL) {
590 pthread_mutex_destroy(&session_list_ptr->lock);
591
592 /* Cleanup ALL session */
593 cds_list_for_each_entry_safe(sess, stmp,
594 &session_list_ptr->head, list) {
595 cmd_destroy_session(sess, kernel_poll_pipe[1]);
596 }
597 }
598
599 wait_consumer(&kconsumer_data);
600 wait_consumer(&ustconsumer64_data);
601 wait_consumer(&ustconsumer32_data);
602
603 DBG("Cleaning up all agent apps");
604 agent_app_ht_clean();
605
606 DBG("Closing all UST sockets");
607 ust_app_clean_list();
608 buffer_reg_destroy_registries();
609
610 if (is_root && !config.no_kernel) {
611 DBG2("Closing kernel fd");
612 if (kernel_tracer_fd >= 0) {
613 ret = close(kernel_tracer_fd);
614 if (ret) {
615 PERROR("close");
616 }
617 }
618 DBG("Unloading kernel modules");
619 modprobe_remove_lttng_all();
620 free(syscall_table);
621 }
622
623 close_consumer_sockets();
624
625 if (load_info) {
626 load_session_destroy_data(load_info);
627 free(load_info);
628 }
629
630 /*
631 * We do NOT rmdir rundir because there are other processes
632 * using it, for instance lttng-relayd, which can start in
633 * parallel with this teardown.
634 */
635 }
636
637 /*
638 * Cleanup the daemon's option data structures.
639 */
640 static void sessiond_cleanup_options(void)
641 {
642 DBG("Cleaning up options");
643
644 sessiond_config_fini(&config);
645
646 run_as_destroy_worker();
647 }
648
649 /*
650 * Send data on a unix socket using the liblttsessiondcomm API.
651 *
652 * Return lttcomm error code.
653 */
654 static int send_unix_sock(int sock, void *buf, size_t len)
655 {
656 /* Check valid length */
657 if (len == 0) {
658 return -1;
659 }
660
661 return lttcomm_send_unix_sock(sock, buf, len);
662 }
663
664 /*
665 * Free memory of a command context structure.
666 */
667 static void clean_command_ctx(struct command_ctx **cmd_ctx)
668 {
669 DBG("Clean command context structure");
670 if (*cmd_ctx) {
671 if ((*cmd_ctx)->llm) {
672 free((*cmd_ctx)->llm);
673 }
674 if ((*cmd_ctx)->lsm) {
675 free((*cmd_ctx)->lsm);
676 }
677 free(*cmd_ctx);
678 *cmd_ctx = NULL;
679 }
680 }
681
682 /*
683 * Notify UST applications using the shm mmap futex.
684 */
685 static int notify_ust_apps(int active)
686 {
687 char *wait_shm_mmap;
688
689 DBG("Notifying applications of session daemon state: %d", active);
690
691 /* See shm.c for this call implying mmap, shm and futex calls */
692 wait_shm_mmap = shm_ust_get_mmap(config.wait_shm_path.value, is_root);
693 if (wait_shm_mmap == NULL) {
694 goto error;
695 }
696
697 /* Wake waiting process */
698 futex_wait_update((int32_t *) wait_shm_mmap, active);
699
700 /* Apps notified successfully */
701 return 0;
702
703 error:
704 return -1;
705 }
706
707 /*
708 * Setup the outgoing data buffer for the response (llm) by allocating the
709 * right amount of memory and copying the original information from the lsm
710 * structure.
711 *
712 * Return 0 on success, negative value on error.
713 */
714 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
715 const void *payload_buf, size_t payload_len,
716 const void *cmd_header_buf, size_t cmd_header_len)
717 {
718 int ret = 0;
719 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
720 const size_t cmd_header_offset = header_len;
721 const size_t payload_offset = cmd_header_offset + cmd_header_len;
722 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
723
724 cmd_ctx->llm = zmalloc(total_msg_size);
725
726 if (cmd_ctx->llm == NULL) {
727 PERROR("zmalloc");
728 ret = -ENOMEM;
729 goto end;
730 }
731
732 /* Copy common data */
733 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
734 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
735 cmd_ctx->llm->cmd_header_size = cmd_header_len;
736 cmd_ctx->llm->data_size = payload_len;
737 cmd_ctx->lttng_msg_size = total_msg_size;
738
739 /* Copy command header */
740 if (cmd_header_len) {
741 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
742 cmd_header_len);
743 }
744
745 /* Copy payload */
746 if (payload_len) {
747 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
748 payload_len);
749 }
750
751 end:
752 return ret;
753 }
754
755 /*
756 * Version of setup_lttng_msg() without command header.
757 */
758 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
759 void *payload_buf, size_t payload_len)
760 {
761 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
762 }
763 /*
764 * Update the kernel poll set of all channel fd available over all tracing
765 * session. Add the wakeup pipe at the end of the set.
766 */
767 static int update_kernel_poll(struct lttng_poll_event *events)
768 {
769 int ret;
770 struct ltt_session *session;
771 struct ltt_kernel_channel *channel;
772
773 DBG("Updating kernel poll set");
774
775 session_lock_list();
776 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
777 session_lock(session);
778 if (session->kernel_session == NULL) {
779 session_unlock(session);
780 continue;
781 }
782
783 cds_list_for_each_entry(channel,
784 &session->kernel_session->channel_list.head, list) {
785 /* Add channel fd to the kernel poll set */
786 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
787 if (ret < 0) {
788 session_unlock(session);
789 goto error;
790 }
791 DBG("Channel fd %d added to kernel set", channel->fd);
792 }
793 session_unlock(session);
794 }
795 session_unlock_list();
796
797 return 0;
798
799 error:
800 session_unlock_list();
801 return -1;
802 }
803
804 /*
805 * Find the channel fd from 'fd' over all tracing session. When found, check
806 * for new channel stream and send those stream fds to the kernel consumer.
807 *
808 * Useful for CPU hotplug feature.
809 */
810 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
811 {
812 int ret = 0;
813 struct ltt_session *session;
814 struct ltt_kernel_session *ksess;
815 struct ltt_kernel_channel *channel;
816
817 DBG("Updating kernel streams for channel fd %d", fd);
818
819 session_lock_list();
820 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
821 session_lock(session);
822 if (session->kernel_session == NULL) {
823 session_unlock(session);
824 continue;
825 }
826 ksess = session->kernel_session;
827
828 cds_list_for_each_entry(channel,
829 &ksess->channel_list.head, list) {
830 struct lttng_ht_iter iter;
831 struct consumer_socket *socket;
832
833 if (channel->fd != fd) {
834 continue;
835 }
836 DBG("Channel found, updating kernel streams");
837 ret = kernel_open_channel_stream(channel);
838 if (ret < 0) {
839 goto error;
840 }
841 /* Update the stream global counter */
842 ksess->stream_count_global += ret;
843
844 /*
845 * Have we already sent fds to the consumer? If yes, it
846 * means that tracing is started so it is safe to send
847 * our updated stream fds.
848 */
849 if (ksess->consumer_fds_sent != 1
850 || ksess->consumer == NULL) {
851 ret = -1;
852 goto error;
853 }
854
855 rcu_read_lock();
856 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
857 &iter.iter, socket, node.node) {
858 pthread_mutex_lock(socket->lock);
859 ret = kernel_consumer_send_channel_stream(socket,
860 channel, ksess,
861 session->output_traces ? 1 : 0);
862 pthread_mutex_unlock(socket->lock);
863 if (ret < 0) {
864 rcu_read_unlock();
865 goto error;
866 }
867 }
868 rcu_read_unlock();
869 }
870 session_unlock(session);
871 }
872 session_unlock_list();
873 return ret;
874
875 error:
876 session_unlock(session);
877 session_unlock_list();
878 return ret;
879 }
880
881 /*
882 * For each tracing session, update newly registered apps. The session list
883 * lock MUST be acquired before calling this.
884 */
885 static void update_ust_app(int app_sock)
886 {
887 struct ltt_session *sess, *stmp;
888
889 /* Consumer is in an ERROR state. Stop any application update. */
890 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
891 /* Stop the update process since the consumer is dead. */
892 return;
893 }
894
895 /* For all tracing session(s) */
896 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
897 struct ust_app *app;
898
899 session_lock(sess);
900 if (!sess->ust_session) {
901 goto unlock_session;
902 }
903
904 rcu_read_lock();
905 assert(app_sock >= 0);
906 app = ust_app_find_by_sock(app_sock);
907 if (app == NULL) {
908 /*
909 * Application can be unregistered before so
910 * this is possible hence simply stopping the
911 * update.
912 */
913 DBG3("UST app update failed to find app sock %d",
914 app_sock);
915 goto unlock_rcu;
916 }
917 ust_app_global_update(sess->ust_session, app);
918 unlock_rcu:
919 rcu_read_unlock();
920 unlock_session:
921 session_unlock(sess);
922 }
923 }
924
925 /*
926 * This thread manage event coming from the kernel.
927 *
928 * Features supported in this thread:
929 * -) CPU Hotplug
930 */
931 static void *thread_manage_kernel(void *data)
932 {
933 int ret, i, pollfd, update_poll_flag = 1, err = -1;
934 uint32_t revents, nb_fd;
935 char tmp;
936 struct lttng_poll_event events;
937
938 DBG("[thread] Thread manage kernel started");
939
940 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
941
942 /*
943 * This first step of the while is to clean this structure which could free
944 * non NULL pointers so initialize it before the loop.
945 */
946 lttng_poll_init(&events);
947
948 if (testpoint(sessiond_thread_manage_kernel)) {
949 goto error_testpoint;
950 }
951
952 health_code_update();
953
954 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
955 goto error_testpoint;
956 }
957
958 while (1) {
959 health_code_update();
960
961 if (update_poll_flag == 1) {
962 /* Clean events object. We are about to populate it again. */
963 lttng_poll_clean(&events);
964
965 ret = sessiond_set_thread_pollset(&events, 2);
966 if (ret < 0) {
967 goto error_poll_create;
968 }
969
970 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
971 if (ret < 0) {
972 goto error;
973 }
974
975 /* This will add the available kernel channel if any. */
976 ret = update_kernel_poll(&events);
977 if (ret < 0) {
978 goto error;
979 }
980 update_poll_flag = 0;
981 }
982
983 DBG("Thread kernel polling");
984
985 /* Poll infinite value of time */
986 restart:
987 health_poll_entry();
988 ret = lttng_poll_wait(&events, -1);
989 DBG("Thread kernel return from poll on %d fds",
990 LTTNG_POLL_GETNB(&events));
991 health_poll_exit();
992 if (ret < 0) {
993 /*
994 * Restart interrupted system call.
995 */
996 if (errno == EINTR) {
997 goto restart;
998 }
999 goto error;
1000 } else if (ret == 0) {
1001 /* Should not happen since timeout is infinite */
1002 ERR("Return value of poll is 0 with an infinite timeout.\n"
1003 "This should not have happened! Continuing...");
1004 continue;
1005 }
1006
1007 nb_fd = ret;
1008
1009 for (i = 0; i < nb_fd; i++) {
1010 /* Fetch once the poll data */
1011 revents = LTTNG_POLL_GETEV(&events, i);
1012 pollfd = LTTNG_POLL_GETFD(&events, i);
1013
1014 health_code_update();
1015
1016 if (!revents) {
1017 /* No activity for this FD (poll implementation). */
1018 continue;
1019 }
1020
1021 /* Thread quit pipe has been closed. Killing thread. */
1022 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1023 if (ret) {
1024 err = 0;
1025 goto exit;
1026 }
1027
1028 /* Check for data on kernel pipe */
1029 if (revents & LPOLLIN) {
1030 if (pollfd == kernel_poll_pipe[0]) {
1031 (void) lttng_read(kernel_poll_pipe[0],
1032 &tmp, 1);
1033 /*
1034 * Ret value is useless here, if this pipe gets any actions an
1035 * update is required anyway.
1036 */
1037 update_poll_flag = 1;
1038 continue;
1039 } else {
1040 /*
1041 * New CPU detected by the kernel. Adding kernel stream to
1042 * kernel session and updating the kernel consumer
1043 */
1044 ret = update_kernel_stream(&kconsumer_data, pollfd);
1045 if (ret < 0) {
1046 continue;
1047 }
1048 break;
1049 }
1050 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1051 update_poll_flag = 1;
1052 continue;
1053 } else {
1054 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1055 goto error;
1056 }
1057 }
1058 }
1059
1060 exit:
1061 error:
1062 lttng_poll_clean(&events);
1063 error_poll_create:
1064 error_testpoint:
1065 utils_close_pipe(kernel_poll_pipe);
1066 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1067 if (err) {
1068 health_error();
1069 ERR("Health error occurred in %s", __func__);
1070 WARN("Kernel thread died unexpectedly. "
1071 "Kernel tracing can continue but CPU hotplug is disabled.");
1072 }
1073 health_unregister(health_sessiond);
1074 DBG("Kernel thread dying");
1075 return NULL;
1076 }
1077
1078 /*
1079 * Signal pthread condition of the consumer data that the thread.
1080 */
1081 static void signal_consumer_condition(struct consumer_data *data, int state)
1082 {
1083 pthread_mutex_lock(&data->cond_mutex);
1084
1085 /*
1086 * The state is set before signaling. It can be any value, it's the waiter
1087 * job to correctly interpret this condition variable associated to the
1088 * consumer pthread_cond.
1089 *
1090 * A value of 0 means that the corresponding thread of the consumer data
1091 * was not started. 1 indicates that the thread has started and is ready
1092 * for action. A negative value means that there was an error during the
1093 * thread bootstrap.
1094 */
1095 data->consumer_thread_is_ready = state;
1096 (void) pthread_cond_signal(&data->cond);
1097
1098 pthread_mutex_unlock(&data->cond_mutex);
1099 }
1100
1101 /*
1102 * This thread manage the consumer error sent back to the session daemon.
1103 */
1104 static void *thread_manage_consumer(void *data)
1105 {
1106 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1107 uint32_t revents, nb_fd;
1108 enum lttcomm_return_code code;
1109 struct lttng_poll_event events;
1110 struct consumer_data *consumer_data = data;
1111 struct consumer_socket *cmd_socket_wrapper = NULL;
1112
1113 DBG("[thread] Manage consumer started");
1114
1115 rcu_register_thread();
1116 rcu_thread_online();
1117
1118 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1119
1120 health_code_update();
1121
1122 /*
1123 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1124 * metadata_sock. Nothing more will be added to this poll set.
1125 */
1126 ret = sessiond_set_thread_pollset(&events, 3);
1127 if (ret < 0) {
1128 goto error_poll;
1129 }
1130
1131 /*
1132 * The error socket here is already in a listening state which was done
1133 * just before spawning this thread to avoid a race between the consumer
1134 * daemon exec trying to connect and the listen() call.
1135 */
1136 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1137 if (ret < 0) {
1138 goto error;
1139 }
1140
1141 health_code_update();
1142
1143 /* Infinite blocking call, waiting for transmission */
1144 restart:
1145 health_poll_entry();
1146
1147 if (testpoint(sessiond_thread_manage_consumer)) {
1148 goto error;
1149 }
1150
1151 ret = lttng_poll_wait(&events, -1);
1152 health_poll_exit();
1153 if (ret < 0) {
1154 /*
1155 * Restart interrupted system call.
1156 */
1157 if (errno == EINTR) {
1158 goto restart;
1159 }
1160 goto error;
1161 }
1162
1163 nb_fd = ret;
1164
1165 for (i = 0; i < nb_fd; i++) {
1166 /* Fetch once the poll data */
1167 revents = LTTNG_POLL_GETEV(&events, i);
1168 pollfd = LTTNG_POLL_GETFD(&events, i);
1169
1170 health_code_update();
1171
1172 if (!revents) {
1173 /* No activity for this FD (poll implementation). */
1174 continue;
1175 }
1176
1177 /* Thread quit pipe has been closed. Killing thread. */
1178 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1179 if (ret) {
1180 err = 0;
1181 goto exit;
1182 }
1183
1184 /* Event on the registration socket */
1185 if (pollfd == consumer_data->err_sock) {
1186 if (revents & LPOLLIN) {
1187 continue;
1188 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1189 ERR("consumer err socket poll error");
1190 goto error;
1191 } else {
1192 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1193 goto error;
1194 }
1195 }
1196 }
1197
1198 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1199 if (sock < 0) {
1200 goto error;
1201 }
1202
1203 /*
1204 * Set the CLOEXEC flag. Return code is useless because either way, the
1205 * show must go on.
1206 */
1207 (void) utils_set_fd_cloexec(sock);
1208
1209 health_code_update();
1210
1211 DBG2("Receiving code from consumer err_sock");
1212
1213 /* Getting status code from kconsumerd */
1214 ret = lttcomm_recv_unix_sock(sock, &code,
1215 sizeof(enum lttcomm_return_code));
1216 if (ret <= 0) {
1217 goto error;
1218 }
1219
1220 health_code_update();
1221 if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1222 ERR("consumer error when waiting for SOCK_READY : %s",
1223 lttcomm_get_readable_code(-code));
1224 goto error;
1225 }
1226
1227 /* Connect both command and metadata sockets. */
1228 consumer_data->cmd_sock =
1229 lttcomm_connect_unix_sock(
1230 consumer_data->cmd_unix_sock_path);
1231 consumer_data->metadata_fd =
1232 lttcomm_connect_unix_sock(
1233 consumer_data->cmd_unix_sock_path);
1234 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1235 PERROR("consumer connect cmd socket");
1236 /* On error, signal condition and quit. */
1237 signal_consumer_condition(consumer_data, -1);
1238 goto error;
1239 }
1240
1241 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1242
1243 /* Create metadata socket lock. */
1244 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1245 if (consumer_data->metadata_sock.lock == NULL) {
1246 PERROR("zmalloc pthread mutex");
1247 goto error;
1248 }
1249 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1250
1251 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1252 DBG("Consumer metadata socket ready (fd: %d)",
1253 consumer_data->metadata_fd);
1254
1255 /*
1256 * Remove the consumerd error sock since we've established a connection.
1257 */
1258 ret = lttng_poll_del(&events, consumer_data->err_sock);
1259 if (ret < 0) {
1260 goto error;
1261 }
1262
1263 /* Add new accepted error socket. */
1264 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1265 if (ret < 0) {
1266 goto error;
1267 }
1268
1269 /* Add metadata socket that is successfully connected. */
1270 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1271 LPOLLIN | LPOLLRDHUP);
1272 if (ret < 0) {
1273 goto error;
1274 }
1275
1276 health_code_update();
1277
1278 /*
1279 * Transfer the write-end of the channel monitoring pipe to the
1280 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
1281 */
1282 cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock);
1283 if (!cmd_socket_wrapper) {
1284 goto error;
1285 }
1286 cmd_socket_wrapper->lock = &consumer_data->lock;
1287
1288 ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper,
1289 consumer_data->channel_monitor_pipe);
1290 if (ret) {
1291 goto error;
1292 }
1293 /* Discard the socket wrapper as it is no longer needed. */
1294 consumer_destroy_socket(cmd_socket_wrapper);
1295 cmd_socket_wrapper = NULL;
1296
1297 /* The thread is completely initialized, signal that it is ready. */
1298 signal_consumer_condition(consumer_data, 1);
1299
1300 /* Infinite blocking call, waiting for transmission */
1301 restart_poll:
1302 while (1) {
1303 health_code_update();
1304
1305 /* Exit the thread because the thread quit pipe has been triggered. */
1306 if (should_quit) {
1307 /* Not a health error. */
1308 err = 0;
1309 goto exit;
1310 }
1311
1312 health_poll_entry();
1313 ret = lttng_poll_wait(&events, -1);
1314 health_poll_exit();
1315 if (ret < 0) {
1316 /*
1317 * Restart interrupted system call.
1318 */
1319 if (errno == EINTR) {
1320 goto restart_poll;
1321 }
1322 goto error;
1323 }
1324
1325 nb_fd = ret;
1326
1327 for (i = 0; i < nb_fd; i++) {
1328 /* Fetch once the poll data */
1329 revents = LTTNG_POLL_GETEV(&events, i);
1330 pollfd = LTTNG_POLL_GETFD(&events, i);
1331
1332 health_code_update();
1333
1334 if (!revents) {
1335 /* No activity for this FD (poll implementation). */
1336 continue;
1337 }
1338
1339 /*
1340 * Thread quit pipe has been triggered, flag that we should stop
1341 * but continue the current loop to handle potential data from
1342 * consumer.
1343 */
1344 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1345
1346 if (pollfd == sock) {
1347 /* Event on the consumerd socket */
1348 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1349 && !(revents & LPOLLIN)) {
1350 ERR("consumer err socket second poll error");
1351 goto error;
1352 }
1353 health_code_update();
1354 /* Wait for any kconsumerd error */
1355 ret = lttcomm_recv_unix_sock(sock, &code,
1356 sizeof(enum lttcomm_return_code));
1357 if (ret <= 0) {
1358 ERR("consumer closed the command socket");
1359 goto error;
1360 }
1361
1362 ERR("consumer return code : %s",
1363 lttcomm_get_readable_code(-code));
1364
1365 goto exit;
1366 } else if (pollfd == consumer_data->metadata_fd) {
1367 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1368 && !(revents & LPOLLIN)) {
1369 ERR("consumer err metadata socket second poll error");
1370 goto error;
1371 }
1372 /* UST metadata requests */
1373 ret = ust_consumer_metadata_request(
1374 &consumer_data->metadata_sock);
1375 if (ret < 0) {
1376 ERR("Handling metadata request");
1377 goto error;
1378 }
1379 }
1380 /* No need for an else branch all FDs are tested prior. */
1381 }
1382 health_code_update();
1383 }
1384
1385 exit:
1386 error:
1387 /*
1388 * We lock here because we are about to close the sockets and some other
1389 * thread might be using them so get exclusive access which will abort all
1390 * other consumer command by other threads.
1391 */
1392 pthread_mutex_lock(&consumer_data->lock);
1393
1394 /* Immediately set the consumerd state to stopped */
1395 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1396 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1397 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1398 consumer_data->type == LTTNG_CONSUMER32_UST) {
1399 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1400 } else {
1401 /* Code flow error... */
1402 assert(0);
1403 }
1404
1405 if (consumer_data->err_sock >= 0) {
1406 ret = close(consumer_data->err_sock);
1407 if (ret) {
1408 PERROR("close");
1409 }
1410 consumer_data->err_sock = -1;
1411 }
1412 if (consumer_data->cmd_sock >= 0) {
1413 ret = close(consumer_data->cmd_sock);
1414 if (ret) {
1415 PERROR("close");
1416 }
1417 consumer_data->cmd_sock = -1;
1418 }
1419 if (consumer_data->metadata_sock.fd_ptr &&
1420 *consumer_data->metadata_sock.fd_ptr >= 0) {
1421 ret = close(*consumer_data->metadata_sock.fd_ptr);
1422 if (ret) {
1423 PERROR("close");
1424 }
1425 }
1426 if (sock >= 0) {
1427 ret = close(sock);
1428 if (ret) {
1429 PERROR("close");
1430 }
1431 }
1432
1433 unlink(consumer_data->err_unix_sock_path);
1434 unlink(consumer_data->cmd_unix_sock_path);
1435 pthread_mutex_unlock(&consumer_data->lock);
1436
1437 /* Cleanup metadata socket mutex. */
1438 if (consumer_data->metadata_sock.lock) {
1439 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1440 free(consumer_data->metadata_sock.lock);
1441 }
1442 lttng_poll_clean(&events);
1443
1444 if (cmd_socket_wrapper) {
1445 consumer_destroy_socket(cmd_socket_wrapper);
1446 }
1447 error_poll:
1448 if (err) {
1449 health_error();
1450 ERR("Health error occurred in %s", __func__);
1451 }
1452 health_unregister(health_sessiond);
1453 DBG("consumer thread cleanup completed");
1454
1455 rcu_thread_offline();
1456 rcu_unregister_thread();
1457
1458 return NULL;
1459 }
1460
1461 /*
1462 * This thread manage application communication.
1463 */
1464 static void *thread_manage_apps(void *data)
1465 {
1466 int i, ret, pollfd, err = -1;
1467 ssize_t size_ret;
1468 uint32_t revents, nb_fd;
1469 struct lttng_poll_event events;
1470
1471 DBG("[thread] Manage application started");
1472
1473 rcu_register_thread();
1474 rcu_thread_online();
1475
1476 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1477
1478 if (testpoint(sessiond_thread_manage_apps)) {
1479 goto error_testpoint;
1480 }
1481
1482 health_code_update();
1483
1484 ret = sessiond_set_thread_pollset(&events, 2);
1485 if (ret < 0) {
1486 goto error_poll_create;
1487 }
1488
1489 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1490 if (ret < 0) {
1491 goto error;
1492 }
1493
1494 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1495 goto error;
1496 }
1497
1498 health_code_update();
1499
1500 while (1) {
1501 DBG("Apps thread polling");
1502
1503 /* Inifinite blocking call, waiting for transmission */
1504 restart:
1505 health_poll_entry();
1506 ret = lttng_poll_wait(&events, -1);
1507 DBG("Apps thread return from poll on %d fds",
1508 LTTNG_POLL_GETNB(&events));
1509 health_poll_exit();
1510 if (ret < 0) {
1511 /*
1512 * Restart interrupted system call.
1513 */
1514 if (errno == EINTR) {
1515 goto restart;
1516 }
1517 goto error;
1518 }
1519
1520 nb_fd = ret;
1521
1522 for (i = 0; i < nb_fd; i++) {
1523 /* Fetch once the poll data */
1524 revents = LTTNG_POLL_GETEV(&events, i);
1525 pollfd = LTTNG_POLL_GETFD(&events, i);
1526
1527 health_code_update();
1528
1529 if (!revents) {
1530 /* No activity for this FD (poll implementation). */
1531 continue;
1532 }
1533
1534 /* Thread quit pipe has been closed. Killing thread. */
1535 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1536 if (ret) {
1537 err = 0;
1538 goto exit;
1539 }
1540
1541 /* Inspect the apps cmd pipe */
1542 if (pollfd == apps_cmd_pipe[0]) {
1543 if (revents & LPOLLIN) {
1544 int sock;
1545
1546 /* Empty pipe */
1547 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1548 if (size_ret < sizeof(sock)) {
1549 PERROR("read apps cmd pipe");
1550 goto error;
1551 }
1552
1553 health_code_update();
1554
1555 /*
1556 * Since this is a command socket (write then read),
1557 * we only monitor the error events of the socket.
1558 */
1559 ret = lttng_poll_add(&events, sock,
1560 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1561 if (ret < 0) {
1562 goto error;
1563 }
1564
1565 DBG("Apps with sock %d added to poll set", sock);
1566 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1567 ERR("Apps command pipe error");
1568 goto error;
1569 } else {
1570 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1571 goto error;
1572 }
1573 } else {
1574 /*
1575 * At this point, we know that a registered application made
1576 * the event at poll_wait.
1577 */
1578 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1579 /* Removing from the poll set */
1580 ret = lttng_poll_del(&events, pollfd);
1581 if (ret < 0) {
1582 goto error;
1583 }
1584
1585 /* Socket closed on remote end. */
1586 ust_app_unregister(pollfd);
1587 } else {
1588 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1589 goto error;
1590 }
1591 }
1592
1593 health_code_update();
1594 }
1595 }
1596
1597 exit:
1598 error:
1599 lttng_poll_clean(&events);
1600 error_poll_create:
1601 error_testpoint:
1602 utils_close_pipe(apps_cmd_pipe);
1603 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1604
1605 /*
1606 * We don't clean the UST app hash table here since already registered
1607 * applications can still be controlled so let them be until the session
1608 * daemon dies or the applications stop.
1609 */
1610
1611 if (err) {
1612 health_error();
1613 ERR("Health error occurred in %s", __func__);
1614 }
1615 health_unregister(health_sessiond);
1616 DBG("Application communication apps thread cleanup complete");
1617 rcu_thread_offline();
1618 rcu_unregister_thread();
1619 return NULL;
1620 }
1621
1622 /*
1623 * Send a socket to a thread This is called from the dispatch UST registration
1624 * thread once all sockets are set for the application.
1625 *
1626 * The sock value can be invalid, we don't really care, the thread will handle
1627 * it and make the necessary cleanup if so.
1628 *
1629 * On success, return 0 else a negative value being the errno message of the
1630 * write().
1631 */
1632 static int send_socket_to_thread(int fd, int sock)
1633 {
1634 ssize_t ret;
1635
1636 /*
1637 * It's possible that the FD is set as invalid with -1 concurrently just
1638 * before calling this function being a shutdown state of the thread.
1639 */
1640 if (fd < 0) {
1641 ret = -EBADF;
1642 goto error;
1643 }
1644
1645 ret = lttng_write(fd, &sock, sizeof(sock));
1646 if (ret < sizeof(sock)) {
1647 PERROR("write apps pipe %d", fd);
1648 if (ret < 0) {
1649 ret = -errno;
1650 }
1651 goto error;
1652 }
1653
1654 /* All good. Don't send back the write positive ret value. */
1655 ret = 0;
1656 error:
1657 return (int) ret;
1658 }
1659
1660 /*
1661 * Sanitize the wait queue of the dispatch registration thread meaning removing
1662 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1663 * notify socket is never received.
1664 */
1665 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1666 {
1667 int ret, nb_fd = 0, i;
1668 unsigned int fd_added = 0;
1669 struct lttng_poll_event events;
1670 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1671
1672 assert(wait_queue);
1673
1674 lttng_poll_init(&events);
1675
1676 /* Just skip everything for an empty queue. */
1677 if (!wait_queue->count) {
1678 goto end;
1679 }
1680
1681 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1682 if (ret < 0) {
1683 goto error_create;
1684 }
1685
1686 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1687 &wait_queue->head, head) {
1688 assert(wait_node->app);
1689 ret = lttng_poll_add(&events, wait_node->app->sock,
1690 LPOLLHUP | LPOLLERR);
1691 if (ret < 0) {
1692 goto error;
1693 }
1694
1695 fd_added = 1;
1696 }
1697
1698 if (!fd_added) {
1699 goto end;
1700 }
1701
1702 /*
1703 * Poll but don't block so we can quickly identify the faulty events and
1704 * clean them afterwards from the wait queue.
1705 */
1706 ret = lttng_poll_wait(&events, 0);
1707 if (ret < 0) {
1708 goto error;
1709 }
1710 nb_fd = ret;
1711
1712 for (i = 0; i < nb_fd; i++) {
1713 /* Get faulty FD. */
1714 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1715 int pollfd = LTTNG_POLL_GETFD(&events, i);
1716
1717 if (!revents) {
1718 /* No activity for this FD (poll implementation). */
1719 continue;
1720 }
1721
1722 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1723 &wait_queue->head, head) {
1724 if (pollfd == wait_node->app->sock &&
1725 (revents & (LPOLLHUP | LPOLLERR))) {
1726 cds_list_del(&wait_node->head);
1727 wait_queue->count--;
1728 ust_app_destroy(wait_node->app);
1729 free(wait_node);
1730 /*
1731 * Silence warning of use-after-free in
1732 * cds_list_for_each_entry_safe which uses
1733 * __typeof__(*wait_node).
1734 */
1735 wait_node = NULL;
1736 break;
1737 } else {
1738 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1739 goto error;
1740 }
1741 }
1742 }
1743
1744 if (nb_fd > 0) {
1745 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1746 }
1747
1748 end:
1749 lttng_poll_clean(&events);
1750 return;
1751
1752 error:
1753 lttng_poll_clean(&events);
1754 error_create:
1755 ERR("Unable to sanitize wait queue");
1756 return;
1757 }
1758
1759 /*
1760 * Dispatch request from the registration threads to the application
1761 * communication thread.
1762 */
1763 static void *thread_dispatch_ust_registration(void *data)
1764 {
1765 int ret, err = -1;
1766 struct cds_wfcq_node *node;
1767 struct ust_command *ust_cmd = NULL;
1768 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1769 struct ust_reg_wait_queue wait_queue = {
1770 .count = 0,
1771 };
1772
1773 rcu_register_thread();
1774
1775 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1776
1777 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1778 goto error_testpoint;
1779 }
1780
1781 health_code_update();
1782
1783 CDS_INIT_LIST_HEAD(&wait_queue.head);
1784
1785 DBG("[thread] Dispatch UST command started");
1786
1787 for (;;) {
1788 health_code_update();
1789
1790 /* Atomically prepare the queue futex */
1791 futex_nto1_prepare(&ust_cmd_queue.futex);
1792
1793 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1794 break;
1795 }
1796
1797 do {
1798 struct ust_app *app = NULL;
1799 ust_cmd = NULL;
1800
1801 /*
1802 * Make sure we don't have node(s) that have hung up before receiving
1803 * the notify socket. This is to clean the list in order to avoid
1804 * memory leaks from notify socket that are never seen.
1805 */
1806 sanitize_wait_queue(&wait_queue);
1807
1808 health_code_update();
1809 /* Dequeue command for registration */
1810 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1811 if (node == NULL) {
1812 DBG("Woken up but nothing in the UST command queue");
1813 /* Continue thread execution */
1814 break;
1815 }
1816
1817 ust_cmd = caa_container_of(node, struct ust_command, node);
1818
1819 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1820 " gid:%d sock:%d name:%s (version %d.%d)",
1821 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1822 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1823 ust_cmd->sock, ust_cmd->reg_msg.name,
1824 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1825
1826 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1827 wait_node = zmalloc(sizeof(*wait_node));
1828 if (!wait_node) {
1829 PERROR("zmalloc wait_node dispatch");
1830 ret = close(ust_cmd->sock);
1831 if (ret < 0) {
1832 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1833 }
1834 lttng_fd_put(LTTNG_FD_APPS, 1);
1835 free(ust_cmd);
1836 goto error;
1837 }
1838 CDS_INIT_LIST_HEAD(&wait_node->head);
1839
1840 /* Create application object if socket is CMD. */
1841 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1842 ust_cmd->sock);
1843 if (!wait_node->app) {
1844 ret = close(ust_cmd->sock);
1845 if (ret < 0) {
1846 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1847 }
1848 lttng_fd_put(LTTNG_FD_APPS, 1);
1849 free(wait_node);
1850 free(ust_cmd);
1851 continue;
1852 }
1853 /*
1854 * Add application to the wait queue so we can set the notify
1855 * socket before putting this object in the global ht.
1856 */
1857 cds_list_add(&wait_node->head, &wait_queue.head);
1858 wait_queue.count++;
1859
1860 free(ust_cmd);
1861 /*
1862 * We have to continue here since we don't have the notify
1863 * socket and the application MUST be added to the hash table
1864 * only at that moment.
1865 */
1866 continue;
1867 } else {
1868 /*
1869 * Look for the application in the local wait queue and set the
1870 * notify socket if found.
1871 */
1872 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1873 &wait_queue.head, head) {
1874 health_code_update();
1875 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1876 wait_node->app->notify_sock = ust_cmd->sock;
1877 cds_list_del(&wait_node->head);
1878 wait_queue.count--;
1879 app = wait_node->app;
1880 free(wait_node);
1881 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1882 break;
1883 }
1884 }
1885
1886 /*
1887 * With no application at this stage the received socket is
1888 * basically useless so close it before we free the cmd data
1889 * structure for good.
1890 */
1891 if (!app) {
1892 ret = close(ust_cmd->sock);
1893 if (ret < 0) {
1894 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1895 }
1896 lttng_fd_put(LTTNG_FD_APPS, 1);
1897 }
1898 free(ust_cmd);
1899 }
1900
1901 if (app) {
1902 /*
1903 * @session_lock_list
1904 *
1905 * Lock the global session list so from the register up to the
1906 * registration done message, no thread can see the application
1907 * and change its state.
1908 */
1909 session_lock_list();
1910 rcu_read_lock();
1911
1912 /*
1913 * Add application to the global hash table. This needs to be
1914 * done before the update to the UST registry can locate the
1915 * application.
1916 */
1917 ust_app_add(app);
1918
1919 /* Set app version. This call will print an error if needed. */
1920 (void) ust_app_version(app);
1921
1922 /* Send notify socket through the notify pipe. */
1923 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1924 app->notify_sock);
1925 if (ret < 0) {
1926 rcu_read_unlock();
1927 session_unlock_list();
1928 /*
1929 * No notify thread, stop the UST tracing. However, this is
1930 * not an internal error of the this thread thus setting
1931 * the health error code to a normal exit.
1932 */
1933 err = 0;
1934 goto error;
1935 }
1936
1937 /*
1938 * Update newly registered application with the tracing
1939 * registry info already enabled information.
1940 */
1941 update_ust_app(app->sock);
1942
1943 /*
1944 * Don't care about return value. Let the manage apps threads
1945 * handle app unregistration upon socket close.
1946 */
1947 (void) ust_app_register_done(app);
1948
1949 /*
1950 * Even if the application socket has been closed, send the app
1951 * to the thread and unregistration will take place at that
1952 * place.
1953 */
1954 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1955 if (ret < 0) {
1956 rcu_read_unlock();
1957 session_unlock_list();
1958 /*
1959 * No apps. thread, stop the UST tracing. However, this is
1960 * not an internal error of the this thread thus setting
1961 * the health error code to a normal exit.
1962 */
1963 err = 0;
1964 goto error;
1965 }
1966
1967 rcu_read_unlock();
1968 session_unlock_list();
1969 }
1970 } while (node != NULL);
1971
1972 health_poll_entry();
1973 /* Futex wait on queue. Blocking call on futex() */
1974 futex_nto1_wait(&ust_cmd_queue.futex);
1975 health_poll_exit();
1976 }
1977 /* Normal exit, no error */
1978 err = 0;
1979
1980 error:
1981 /* Clean up wait queue. */
1982 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1983 &wait_queue.head, head) {
1984 cds_list_del(&wait_node->head);
1985 wait_queue.count--;
1986 free(wait_node);
1987 }
1988
1989 /* Empty command queue. */
1990 for (;;) {
1991 /* Dequeue command for registration */
1992 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1993 if (node == NULL) {
1994 break;
1995 }
1996 ust_cmd = caa_container_of(node, struct ust_command, node);
1997 ret = close(ust_cmd->sock);
1998 if (ret < 0) {
1999 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2000 }
2001 lttng_fd_put(LTTNG_FD_APPS, 1);
2002 free(ust_cmd);
2003 }
2004
2005 error_testpoint:
2006 DBG("Dispatch thread dying");
2007 if (err) {
2008 health_error();
2009 ERR("Health error occurred in %s", __func__);
2010 }
2011 health_unregister(health_sessiond);
2012 rcu_unregister_thread();
2013 return NULL;
2014 }
2015
2016 /*
2017 * This thread manage application registration.
2018 */
2019 static void *thread_registration_apps(void *data)
2020 {
2021 int sock = -1, i, ret, pollfd, err = -1;
2022 uint32_t revents, nb_fd;
2023 struct lttng_poll_event events;
2024 /*
2025 * Get allocated in this thread, enqueued to a global queue, dequeued and
2026 * freed in the manage apps thread.
2027 */
2028 struct ust_command *ust_cmd = NULL;
2029
2030 DBG("[thread] Manage application registration started");
2031
2032 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2033
2034 if (testpoint(sessiond_thread_registration_apps)) {
2035 goto error_testpoint;
2036 }
2037
2038 ret = lttcomm_listen_unix_sock(apps_sock);
2039 if (ret < 0) {
2040 goto error_listen;
2041 }
2042
2043 /*
2044 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2045 * more will be added to this poll set.
2046 */
2047 ret = sessiond_set_thread_pollset(&events, 2);
2048 if (ret < 0) {
2049 goto error_create_poll;
2050 }
2051
2052 /* Add the application registration socket */
2053 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2054 if (ret < 0) {
2055 goto error_poll_add;
2056 }
2057
2058 /* Notify all applications to register */
2059 ret = notify_ust_apps(1);
2060 if (ret < 0) {
2061 ERR("Failed to notify applications or create the wait shared memory.\n"
2062 "Execution continues but there might be problem for already\n"
2063 "running applications that wishes to register.");
2064 }
2065
2066 while (1) {
2067 DBG("Accepting application registration");
2068
2069 /* Inifinite blocking call, waiting for transmission */
2070 restart:
2071 health_poll_entry();
2072 ret = lttng_poll_wait(&events, -1);
2073 health_poll_exit();
2074 if (ret < 0) {
2075 /*
2076 * Restart interrupted system call.
2077 */
2078 if (errno == EINTR) {
2079 goto restart;
2080 }
2081 goto error;
2082 }
2083
2084 nb_fd = ret;
2085
2086 for (i = 0; i < nb_fd; i++) {
2087 health_code_update();
2088
2089 /* Fetch once the poll data */
2090 revents = LTTNG_POLL_GETEV(&events, i);
2091 pollfd = LTTNG_POLL_GETFD(&events, i);
2092
2093 if (!revents) {
2094 /* No activity for this FD (poll implementation). */
2095 continue;
2096 }
2097
2098 /* Thread quit pipe has been closed. Killing thread. */
2099 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2100 if (ret) {
2101 err = 0;
2102 goto exit;
2103 }
2104
2105 /* Event on the registration socket */
2106 if (pollfd == apps_sock) {
2107 if (revents & LPOLLIN) {
2108 sock = lttcomm_accept_unix_sock(apps_sock);
2109 if (sock < 0) {
2110 goto error;
2111 }
2112
2113 /*
2114 * Set socket timeout for both receiving and ending.
2115 * app_socket_timeout is in seconds, whereas
2116 * lttcomm_setsockopt_rcv_timeout and
2117 * lttcomm_setsockopt_snd_timeout expect msec as
2118 * parameter.
2119 */
2120 if (config.app_socket_timeout >= 0) {
2121 (void) lttcomm_setsockopt_rcv_timeout(sock,
2122 config.app_socket_timeout * 1000);
2123 (void) lttcomm_setsockopt_snd_timeout(sock,
2124 config.app_socket_timeout * 1000);
2125 }
2126
2127 /*
2128 * Set the CLOEXEC flag. Return code is useless because
2129 * either way, the show must go on.
2130 */
2131 (void) utils_set_fd_cloexec(sock);
2132
2133 /* Create UST registration command for enqueuing */
2134 ust_cmd = zmalloc(sizeof(struct ust_command));
2135 if (ust_cmd == NULL) {
2136 PERROR("ust command zmalloc");
2137 ret = close(sock);
2138 if (ret) {
2139 PERROR("close");
2140 }
2141 goto error;
2142 }
2143
2144 /*
2145 * Using message-based transmissions to ensure we don't
2146 * have to deal with partially received messages.
2147 */
2148 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2149 if (ret < 0) {
2150 ERR("Exhausted file descriptors allowed for applications.");
2151 free(ust_cmd);
2152 ret = close(sock);
2153 if (ret) {
2154 PERROR("close");
2155 }
2156 sock = -1;
2157 continue;
2158 }
2159
2160 health_code_update();
2161 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2162 if (ret < 0) {
2163 free(ust_cmd);
2164 /* Close socket of the application. */
2165 ret = close(sock);
2166 if (ret) {
2167 PERROR("close");
2168 }
2169 lttng_fd_put(LTTNG_FD_APPS, 1);
2170 sock = -1;
2171 continue;
2172 }
2173 health_code_update();
2174
2175 ust_cmd->sock = sock;
2176 sock = -1;
2177
2178 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2179 " gid:%d sock:%d name:%s (version %d.%d)",
2180 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2181 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2182 ust_cmd->sock, ust_cmd->reg_msg.name,
2183 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2184
2185 /*
2186 * Lock free enqueue the registration request. The red pill
2187 * has been taken! This apps will be part of the *system*.
2188 */
2189 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2190
2191 /*
2192 * Wake the registration queue futex. Implicit memory
2193 * barrier with the exchange in cds_wfcq_enqueue.
2194 */
2195 futex_nto1_wake(&ust_cmd_queue.futex);
2196 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2197 ERR("Register apps socket poll error");
2198 goto error;
2199 } else {
2200 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2201 goto error;
2202 }
2203 }
2204 }
2205 }
2206
2207 exit:
2208 error:
2209 /* Notify that the registration thread is gone */
2210 notify_ust_apps(0);
2211
2212 if (apps_sock >= 0) {
2213 ret = close(apps_sock);
2214 if (ret) {
2215 PERROR("close");
2216 }
2217 }
2218 if (sock >= 0) {
2219 ret = close(sock);
2220 if (ret) {
2221 PERROR("close");
2222 }
2223 lttng_fd_put(LTTNG_FD_APPS, 1);
2224 }
2225 unlink(config.apps_unix_sock_path.value);
2226
2227 error_poll_add:
2228 lttng_poll_clean(&events);
2229 error_listen:
2230 error_create_poll:
2231 error_testpoint:
2232 DBG("UST Registration thread cleanup complete");
2233 if (err) {
2234 health_error();
2235 ERR("Health error occurred in %s", __func__);
2236 }
2237 health_unregister(health_sessiond);
2238
2239 return NULL;
2240 }
2241
2242 /*
2243 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2244 * exec or it will fails.
2245 */
2246 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2247 {
2248 int ret, clock_ret;
2249 struct timespec timeout;
2250
2251 /*
2252 * Make sure we set the readiness flag to 0 because we are NOT ready.
2253 * This access to consumer_thread_is_ready does not need to be
2254 * protected by consumer_data.cond_mutex (yet) since the consumer
2255 * management thread has not been started at this point.
2256 */
2257 consumer_data->consumer_thread_is_ready = 0;
2258
2259 /* Setup pthread condition */
2260 ret = pthread_condattr_init(&consumer_data->condattr);
2261 if (ret) {
2262 errno = ret;
2263 PERROR("pthread_condattr_init consumer data");
2264 goto error;
2265 }
2266
2267 /*
2268 * Set the monotonic clock in order to make sure we DO NOT jump in time
2269 * between the clock_gettime() call and the timedwait call. See bug #324
2270 * for a more details and how we noticed it.
2271 */
2272 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2273 if (ret) {
2274 errno = ret;
2275 PERROR("pthread_condattr_setclock consumer data");
2276 goto error;
2277 }
2278
2279 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2280 if (ret) {
2281 errno = ret;
2282 PERROR("pthread_cond_init consumer data");
2283 goto error;
2284 }
2285
2286 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2287 thread_manage_consumer, consumer_data);
2288 if (ret) {
2289 errno = ret;
2290 PERROR("pthread_create consumer");
2291 ret = -1;
2292 goto error;
2293 }
2294
2295 /* We are about to wait on a pthread condition */
2296 pthread_mutex_lock(&consumer_data->cond_mutex);
2297
2298 /* Get time for sem_timedwait absolute timeout */
2299 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2300 /*
2301 * Set the timeout for the condition timed wait even if the clock gettime
2302 * call fails since we might loop on that call and we want to avoid to
2303 * increment the timeout too many times.
2304 */
2305 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2306
2307 /*
2308 * The following loop COULD be skipped in some conditions so this is why we
2309 * set ret to 0 in order to make sure at least one round of the loop is
2310 * done.
2311 */
2312 ret = 0;
2313
2314 /*
2315 * Loop until the condition is reached or when a timeout is reached. Note
2316 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2317 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2318 * possible. This loop does not take any chances and works with both of
2319 * them.
2320 */
2321 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2322 if (clock_ret < 0) {
2323 PERROR("clock_gettime spawn consumer");
2324 /* Infinite wait for the consumerd thread to be ready */
2325 ret = pthread_cond_wait(&consumer_data->cond,
2326 &consumer_data->cond_mutex);
2327 } else {
2328 ret = pthread_cond_timedwait(&consumer_data->cond,
2329 &consumer_data->cond_mutex, &timeout);
2330 }
2331 }
2332
2333 /* Release the pthread condition */
2334 pthread_mutex_unlock(&consumer_data->cond_mutex);
2335
2336 if (ret != 0) {
2337 errno = ret;
2338 if (ret == ETIMEDOUT) {
2339 int pth_ret;
2340
2341 /*
2342 * Call has timed out so we kill the kconsumerd_thread and return
2343 * an error.
2344 */
2345 ERR("Condition timed out. The consumer thread was never ready."
2346 " Killing it");
2347 pth_ret = pthread_cancel(consumer_data->thread);
2348 if (pth_ret < 0) {
2349 PERROR("pthread_cancel consumer thread");
2350 }
2351 } else {
2352 PERROR("pthread_cond_wait failed consumer thread");
2353 }
2354 /* Caller is expecting a negative value on failure. */
2355 ret = -1;
2356 goto error;
2357 }
2358
2359 pthread_mutex_lock(&consumer_data->pid_mutex);
2360 if (consumer_data->pid == 0) {
2361 ERR("Consumerd did not start");
2362 pthread_mutex_unlock(&consumer_data->pid_mutex);
2363 goto error;
2364 }
2365 pthread_mutex_unlock(&consumer_data->pid_mutex);
2366
2367 return 0;
2368
2369 error:
2370 return ret;
2371 }
2372
2373 /*
2374 * Join consumer thread
2375 */
2376 static int join_consumer_thread(struct consumer_data *consumer_data)
2377 {
2378 void *status;
2379
2380 /* Consumer pid must be a real one. */
2381 if (consumer_data->pid > 0) {
2382 int ret;
2383 ret = kill(consumer_data->pid, SIGTERM);
2384 if (ret) {
2385 PERROR("Error killing consumer daemon");
2386 return ret;
2387 }
2388 return pthread_join(consumer_data->thread, &status);
2389 } else {
2390 return 0;
2391 }
2392 }
2393
2394 /*
2395 * Fork and exec a consumer daemon (consumerd).
2396 *
2397 * Return pid if successful else -1.
2398 */
2399 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2400 {
2401 int ret;
2402 pid_t pid;
2403 const char *consumer_to_use;
2404 const char *verbosity;
2405 struct stat st;
2406
2407 DBG("Spawning consumerd");
2408
2409 pid = fork();
2410 if (pid == 0) {
2411 /*
2412 * Exec consumerd.
2413 */
2414 if (config.verbose_consumer) {
2415 verbosity = "--verbose";
2416 } else if (lttng_opt_quiet) {
2417 verbosity = "--quiet";
2418 } else {
2419 verbosity = "";
2420 }
2421
2422 switch (consumer_data->type) {
2423 case LTTNG_CONSUMER_KERNEL:
2424 /*
2425 * Find out which consumerd to execute. We will first try the
2426 * 64-bit path, then the sessiond's installation directory, and
2427 * fallback on the 32-bit one,
2428 */
2429 DBG3("Looking for a kernel consumer at these locations:");
2430 DBG3(" 1) %s", config.consumerd64_bin_path.value ? : "NULL");
2431 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, DEFAULT_CONSUMERD_FILE);
2432 DBG3(" 3) %s", config.consumerd32_bin_path.value ? : "NULL");
2433 if (stat(config.consumerd64_bin_path.value, &st) == 0) {
2434 DBG3("Found location #1");
2435 consumer_to_use = config.consumerd64_bin_path.value;
2436 } else if (stat(INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE, &st) == 0) {
2437 DBG3("Found location #2");
2438 consumer_to_use = INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE;
2439 } else if (config.consumerd32_bin_path.value &&
2440 stat(config.consumerd32_bin_path.value, &st) == 0) {
2441 DBG3("Found location #3");
2442 consumer_to_use = config.consumerd32_bin_path.value;
2443 } else {
2444 DBG("Could not find any valid consumerd executable");
2445 ret = -EINVAL;
2446 goto error;
2447 }
2448 DBG("Using kernel consumer at: %s", consumer_to_use);
2449 (void) execl(consumer_to_use,
2450 "lttng-consumerd", verbosity, "-k",
2451 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2452 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2453 "--group", config.tracing_group_name.value,
2454 NULL);
2455 break;
2456 case LTTNG_CONSUMER64_UST:
2457 {
2458 char *tmpnew = NULL;
2459
2460 if (config.consumerd64_lib_dir.value) {
2461 char *tmp;
2462 size_t tmplen;
2463
2464 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2465 if (!tmp) {
2466 tmp = "";
2467 }
2468 tmplen = strlen(config.consumerd64_lib_dir.value) + 1 /* : */ + strlen(tmp);
2469 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2470 if (!tmpnew) {
2471 ret = -ENOMEM;
2472 goto error;
2473 }
2474 strcat(tmpnew, config.consumerd64_lib_dir.value);
2475 if (tmp[0] != '\0') {
2476 strcat(tmpnew, ":");
2477 strcat(tmpnew, tmp);
2478 }
2479 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2480 if (ret) {
2481 ret = -errno;
2482 free(tmpnew);
2483 goto error;
2484 }
2485 }
2486 DBG("Using 64-bit UST consumer at: %s", config.consumerd64_bin_path.value);
2487 (void) execl(config.consumerd64_bin_path.value, "lttng-consumerd", verbosity, "-u",
2488 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2489 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2490 "--group", config.tracing_group_name.value,
2491 NULL);
2492 free(tmpnew);
2493 break;
2494 }
2495 case LTTNG_CONSUMER32_UST:
2496 {
2497 char *tmpnew = NULL;
2498
2499 if (config.consumerd32_lib_dir.value) {
2500 char *tmp;
2501 size_t tmplen;
2502
2503 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2504 if (!tmp) {
2505 tmp = "";
2506 }
2507 tmplen = strlen(config.consumerd32_lib_dir.value) + 1 /* : */ + strlen(tmp);
2508 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2509 if (!tmpnew) {
2510 ret = -ENOMEM;
2511 goto error;
2512 }
2513 strcat(tmpnew, config.consumerd32_lib_dir.value);
2514 if (tmp[0] != '\0') {
2515 strcat(tmpnew, ":");
2516 strcat(tmpnew, tmp);
2517 }
2518 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2519 if (ret) {
2520 ret = -errno;
2521 free(tmpnew);
2522 goto error;
2523 }
2524 }
2525 DBG("Using 32-bit UST consumer at: %s", config.consumerd32_bin_path.value);
2526 (void) execl(config.consumerd32_bin_path.value, "lttng-consumerd", verbosity, "-u",
2527 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2528 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2529 "--group", config.tracing_group_name.value,
2530 NULL);
2531 free(tmpnew);
2532 break;
2533 }
2534 default:
2535 ERR("unknown consumer type");
2536 exit(EXIT_FAILURE);
2537 }
2538 if (errno != 0) {
2539 PERROR("Consumer execl()");
2540 }
2541 /* Reaching this point, we got a failure on our execl(). */
2542 exit(EXIT_FAILURE);
2543 } else if (pid > 0) {
2544 ret = pid;
2545 } else {
2546 PERROR("start consumer fork");
2547 ret = -errno;
2548 }
2549 error:
2550 return ret;
2551 }
2552
2553 /*
2554 * Spawn the consumerd daemon and session daemon thread.
2555 */
2556 static int start_consumerd(struct consumer_data *consumer_data)
2557 {
2558 int ret;
2559
2560 /*
2561 * Set the listen() state on the socket since there is a possible race
2562 * between the exec() of the consumer daemon and this call if place in the
2563 * consumer thread. See bug #366 for more details.
2564 */
2565 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2566 if (ret < 0) {
2567 goto error;
2568 }
2569
2570 pthread_mutex_lock(&consumer_data->pid_mutex);
2571 if (consumer_data->pid != 0) {
2572 pthread_mutex_unlock(&consumer_data->pid_mutex);
2573 goto end;
2574 }
2575
2576 ret = spawn_consumerd(consumer_data);
2577 if (ret < 0) {
2578 ERR("Spawning consumerd failed");
2579 pthread_mutex_unlock(&consumer_data->pid_mutex);
2580 goto error;
2581 }
2582
2583 /* Setting up the consumer_data pid */
2584 consumer_data->pid = ret;
2585 DBG2("Consumer pid %d", consumer_data->pid);
2586 pthread_mutex_unlock(&consumer_data->pid_mutex);
2587
2588 DBG2("Spawning consumer control thread");
2589 ret = spawn_consumer_thread(consumer_data);
2590 if (ret < 0) {
2591 ERR("Fatal error spawning consumer control thread");
2592 goto error;
2593 }
2594
2595 end:
2596 return 0;
2597
2598 error:
2599 /* Cleanup already created sockets on error. */
2600 if (consumer_data->err_sock >= 0) {
2601 int err;
2602
2603 err = close(consumer_data->err_sock);
2604 if (err < 0) {
2605 PERROR("close consumer data error socket");
2606 }
2607 }
2608 return ret;
2609 }
2610
2611 /*
2612 * Setup necessary data for kernel tracer action.
2613 */
2614 static int init_kernel_tracer(void)
2615 {
2616 int ret;
2617
2618 /* Modprobe lttng kernel modules */
2619 ret = modprobe_lttng_control();
2620 if (ret < 0) {
2621 goto error;
2622 }
2623
2624 /* Open debugfs lttng */
2625 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2626 if (kernel_tracer_fd < 0) {
2627 DBG("Failed to open %s", module_proc_lttng);
2628 goto error_open;
2629 }
2630
2631 /* Validate kernel version */
2632 ret = kernel_validate_version(kernel_tracer_fd);
2633 if (ret < 0) {
2634 goto error_version;
2635 }
2636
2637 ret = modprobe_lttng_data();
2638 if (ret < 0) {
2639 goto error_modules;
2640 }
2641
2642 ret = kernel_supports_ring_buffer_snapshot_sample_positions(
2643 kernel_tracer_fd);
2644 if (ret < 0) {
2645 goto error_modules;
2646 }
2647
2648 if (ret < 1) {
2649 WARN("Kernel tracer does not support buffer monitoring. "
2650 "The monitoring timer of channels in the kernel domain "
2651 "will be set to 0 (disabled).");
2652 }
2653
2654 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2655 return 0;
2656
2657 error_version:
2658 modprobe_remove_lttng_control();
2659 ret = close(kernel_tracer_fd);
2660 if (ret) {
2661 PERROR("close");
2662 }
2663 kernel_tracer_fd = -1;
2664 return LTTNG_ERR_KERN_VERSION;
2665
2666 error_modules:
2667 ret = close(kernel_tracer_fd);
2668 if (ret) {
2669 PERROR("close");
2670 }
2671
2672 error_open:
2673 modprobe_remove_lttng_control();
2674
2675 error:
2676 WARN("No kernel tracer available");
2677 kernel_tracer_fd = -1;
2678 if (!is_root) {
2679 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2680 } else {
2681 return LTTNG_ERR_KERN_NA;
2682 }
2683 }
2684
2685
2686 /*
2687 * Copy consumer output from the tracing session to the domain session. The
2688 * function also applies the right modification on a per domain basis for the
2689 * trace files destination directory.
2690 *
2691 * Should *NOT* be called with RCU read-side lock held.
2692 */
2693 static int copy_session_consumer(int domain, struct ltt_session *session)
2694 {
2695 int ret;
2696 const char *dir_name;
2697 struct consumer_output *consumer;
2698
2699 assert(session);
2700 assert(session->consumer);
2701
2702 switch (domain) {
2703 case LTTNG_DOMAIN_KERNEL:
2704 DBG3("Copying tracing session consumer output in kernel session");
2705 /*
2706 * XXX: We should audit the session creation and what this function
2707 * does "extra" in order to avoid a destroy since this function is used
2708 * in the domain session creation (kernel and ust) only. Same for UST
2709 * domain.
2710 */
2711 if (session->kernel_session->consumer) {
2712 consumer_output_put(session->kernel_session->consumer);
2713 }
2714 session->kernel_session->consumer =
2715 consumer_copy_output(session->consumer);
2716 /* Ease our life a bit for the next part */
2717 consumer = session->kernel_session->consumer;
2718 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2719 break;
2720 case LTTNG_DOMAIN_JUL:
2721 case LTTNG_DOMAIN_LOG4J:
2722 case LTTNG_DOMAIN_PYTHON:
2723 case LTTNG_DOMAIN_UST:
2724 DBG3("Copying tracing session consumer output in UST session");
2725 if (session->ust_session->consumer) {
2726 consumer_output_put(session->ust_session->consumer);
2727 }
2728 session->ust_session->consumer =
2729 consumer_copy_output(session->consumer);
2730 /* Ease our life a bit for the next part */
2731 consumer = session->ust_session->consumer;
2732 dir_name = DEFAULT_UST_TRACE_DIR;
2733 break;
2734 default:
2735 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2736 goto error;
2737 }
2738
2739 /* Append correct directory to subdir */
2740 strncat(consumer->subdir, dir_name,
2741 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2742 DBG3("Copy session consumer subdir %s", consumer->subdir);
2743
2744 ret = LTTNG_OK;
2745
2746 error:
2747 return ret;
2748 }
2749
2750 /*
2751 * Create an UST session and add it to the session ust list.
2752 *
2753 * Should *NOT* be called with RCU read-side lock held.
2754 */
2755 static int create_ust_session(struct ltt_session *session,
2756 struct lttng_domain *domain)
2757 {
2758 int ret;
2759 struct ltt_ust_session *lus = NULL;
2760
2761 assert(session);
2762 assert(domain);
2763 assert(session->consumer);
2764
2765 switch (domain->type) {
2766 case LTTNG_DOMAIN_JUL:
2767 case LTTNG_DOMAIN_LOG4J:
2768 case LTTNG_DOMAIN_PYTHON:
2769 case LTTNG_DOMAIN_UST:
2770 break;
2771 default:
2772 ERR("Unknown UST domain on create session %d", domain->type);
2773 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2774 goto error;
2775 }
2776
2777 DBG("Creating UST session");
2778
2779 lus = trace_ust_create_session(session->id);
2780 if (lus == NULL) {
2781 ret = LTTNG_ERR_UST_SESS_FAIL;
2782 goto error;
2783 }
2784
2785 lus->uid = session->uid;
2786 lus->gid = session->gid;
2787 lus->output_traces = session->output_traces;
2788 lus->snapshot_mode = session->snapshot_mode;
2789 lus->live_timer_interval = session->live_timer;
2790 session->ust_session = lus;
2791 if (session->shm_path[0]) {
2792 strncpy(lus->root_shm_path, session->shm_path,
2793 sizeof(lus->root_shm_path));
2794 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2795 strncpy(lus->shm_path, session->shm_path,
2796 sizeof(lus->shm_path));
2797 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2798 strncat(lus->shm_path, "/ust",
2799 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2800 }
2801 /* Copy session output to the newly created UST session */
2802 ret = copy_session_consumer(domain->type, session);
2803 if (ret != LTTNG_OK) {
2804 goto error;
2805 }
2806
2807 return LTTNG_OK;
2808
2809 error:
2810 free(lus);
2811 session->ust_session = NULL;
2812 return ret;
2813 }
2814
2815 /*
2816 * Create a kernel tracer session then create the default channel.
2817 */
2818 static int create_kernel_session(struct ltt_session *session)
2819 {
2820 int ret;
2821
2822 DBG("Creating kernel session");
2823
2824 ret = kernel_create_session(session, kernel_tracer_fd);
2825 if (ret < 0) {
2826 ret = LTTNG_ERR_KERN_SESS_FAIL;
2827 goto error;
2828 }
2829
2830 /* Code flow safety */
2831 assert(session->kernel_session);
2832
2833 /* Copy session output to the newly created Kernel session */
2834 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2835 if (ret != LTTNG_OK) {
2836 goto error;
2837 }
2838
2839 /* Create directory(ies) on local filesystem. */
2840 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2841 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2842 ret = run_as_mkdir_recursive(
2843 session->kernel_session->consumer->dst.trace_path,
2844 S_IRWXU | S_IRWXG, session->uid, session->gid);
2845 if (ret < 0) {
2846 if (errno != EEXIST) {
2847 ERR("Trace directory creation error");
2848 goto error;
2849 }
2850 }
2851 }
2852
2853 session->kernel_session->uid = session->uid;
2854 session->kernel_session->gid = session->gid;
2855 session->kernel_session->output_traces = session->output_traces;
2856 session->kernel_session->snapshot_mode = session->snapshot_mode;
2857
2858 return LTTNG_OK;
2859
2860 error:
2861 trace_kernel_destroy_session(session->kernel_session);
2862 session->kernel_session = NULL;
2863 return ret;
2864 }
2865
2866 /*
2867 * Count number of session permitted by uid/gid.
2868 */
2869 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2870 {
2871 unsigned int i = 0;
2872 struct ltt_session *session;
2873
2874 DBG("Counting number of available session for UID %d GID %d",
2875 uid, gid);
2876 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2877 /*
2878 * Only list the sessions the user can control.
2879 */
2880 if (!session_access_ok(session, uid, gid)) {
2881 continue;
2882 }
2883 i++;
2884 }
2885 return i;
2886 }
2887
2888 /*
2889 * Process the command requested by the lttng client within the command
2890 * context structure. This function make sure that the return structure (llm)
2891 * is set and ready for transmission before returning.
2892 *
2893 * Return any error encountered or 0 for success.
2894 *
2895 * "sock" is only used for special-case var. len data.
2896 *
2897 * Should *NOT* be called with RCU read-side lock held.
2898 */
2899 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2900 int *sock_error)
2901 {
2902 int ret = LTTNG_OK;
2903 int need_tracing_session = 1;
2904 int need_domain;
2905
2906 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2907
2908 assert(!rcu_read_ongoing());
2909
2910 *sock_error = 0;
2911
2912 switch (cmd_ctx->lsm->cmd_type) {
2913 case LTTNG_CREATE_SESSION:
2914 case LTTNG_CREATE_SESSION_SNAPSHOT:
2915 case LTTNG_CREATE_SESSION_LIVE:
2916 case LTTNG_DESTROY_SESSION:
2917 case LTTNG_LIST_SESSIONS:
2918 case LTTNG_LIST_DOMAINS:
2919 case LTTNG_START_TRACE:
2920 case LTTNG_STOP_TRACE:
2921 case LTTNG_DATA_PENDING:
2922 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2923 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2924 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2925 case LTTNG_SNAPSHOT_RECORD:
2926 case LTTNG_SAVE_SESSION:
2927 case LTTNG_SET_SESSION_SHM_PATH:
2928 case LTTNG_REGENERATE_METADATA:
2929 case LTTNG_REGENERATE_STATEDUMP:
2930 case LTTNG_REGISTER_TRIGGER:
2931 case LTTNG_UNREGISTER_TRIGGER:
2932 need_domain = 0;
2933 break;
2934 default:
2935 need_domain = 1;
2936 }
2937
2938 if (config.no_kernel && need_domain
2939 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2940 if (!is_root) {
2941 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2942 } else {
2943 ret = LTTNG_ERR_KERN_NA;
2944 }
2945 goto error;
2946 }
2947
2948 /* Deny register consumer if we already have a spawned consumer. */
2949 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2950 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2951 if (kconsumer_data.pid > 0) {
2952 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2953 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2954 goto error;
2955 }
2956 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2957 }
2958
2959 /*
2960 * Check for command that don't needs to allocate a returned payload. We do
2961 * this here so we don't have to make the call for no payload at each
2962 * command.
2963 */
2964 switch(cmd_ctx->lsm->cmd_type) {
2965 case LTTNG_LIST_SESSIONS:
2966 case LTTNG_LIST_TRACEPOINTS:
2967 case LTTNG_LIST_TRACEPOINT_FIELDS:
2968 case LTTNG_LIST_DOMAINS:
2969 case LTTNG_LIST_CHANNELS:
2970 case LTTNG_LIST_EVENTS:
2971 case LTTNG_LIST_SYSCALLS:
2972 case LTTNG_LIST_TRACKER_PIDS:
2973 case LTTNG_DATA_PENDING:
2974 break;
2975 default:
2976 /* Setup lttng message with no payload */
2977 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
2978 if (ret < 0) {
2979 /* This label does not try to unlock the session */
2980 goto init_setup_error;
2981 }
2982 }
2983
2984 /* Commands that DO NOT need a session. */
2985 switch (cmd_ctx->lsm->cmd_type) {
2986 case LTTNG_CREATE_SESSION:
2987 case LTTNG_CREATE_SESSION_SNAPSHOT:
2988 case LTTNG_CREATE_SESSION_LIVE:
2989 case LTTNG_LIST_SESSIONS:
2990 case LTTNG_LIST_TRACEPOINTS:
2991 case LTTNG_LIST_SYSCALLS:
2992 case LTTNG_LIST_TRACEPOINT_FIELDS:
2993 case LTTNG_SAVE_SESSION:
2994 case LTTNG_REGISTER_TRIGGER:
2995 case LTTNG_UNREGISTER_TRIGGER:
2996 need_tracing_session = 0;
2997 break;
2998 default:
2999 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3000 /*
3001 * We keep the session list lock across _all_ commands
3002 * for now, because the per-session lock does not
3003 * handle teardown properly.
3004 */
3005 session_lock_list();
3006 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3007 if (cmd_ctx->session == NULL) {
3008 ret = LTTNG_ERR_SESS_NOT_FOUND;
3009 goto error;
3010 } else {
3011 /* Acquire lock for the session */
3012 session_lock(cmd_ctx->session);
3013 }
3014 break;
3015 }
3016
3017 /*
3018 * Commands that need a valid session but should NOT create one if none
3019 * exists. Instead of creating one and destroying it when the command is
3020 * handled, process that right before so we save some round trip in useless
3021 * code path.
3022 */
3023 switch (cmd_ctx->lsm->cmd_type) {
3024 case LTTNG_DISABLE_CHANNEL:
3025 case LTTNG_DISABLE_EVENT:
3026 switch (cmd_ctx->lsm->domain.type) {
3027 case LTTNG_DOMAIN_KERNEL:
3028 if (!cmd_ctx->session->kernel_session) {
3029 ret = LTTNG_ERR_NO_CHANNEL;
3030 goto error;
3031 }
3032 break;
3033 case LTTNG_DOMAIN_JUL:
3034 case LTTNG_DOMAIN_LOG4J:
3035 case LTTNG_DOMAIN_PYTHON:
3036 case LTTNG_DOMAIN_UST:
3037 if (!cmd_ctx->session->ust_session) {
3038 ret = LTTNG_ERR_NO_CHANNEL;
3039 goto error;
3040 }
3041 break;
3042 default:
3043 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3044 goto error;
3045 }
3046 default:
3047 break;
3048 }
3049
3050 if (!need_domain) {
3051 goto skip_domain;
3052 }
3053
3054 /*
3055 * Check domain type for specific "pre-action".
3056 */
3057 switch (cmd_ctx->lsm->domain.type) {
3058 case LTTNG_DOMAIN_KERNEL:
3059 if (!is_root) {
3060 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3061 goto error;
3062 }
3063
3064 /* Kernel tracer check */
3065 if (kernel_tracer_fd == -1) {
3066 /* Basically, load kernel tracer modules */
3067 ret = init_kernel_tracer();
3068 if (ret != 0) {
3069 goto error;
3070 }
3071 }
3072
3073 /* Consumer is in an ERROR state. Report back to client */
3074 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3075 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3076 goto error;
3077 }
3078
3079 /* Need a session for kernel command */
3080 if (need_tracing_session) {
3081 if (cmd_ctx->session->kernel_session == NULL) {
3082 ret = create_kernel_session(cmd_ctx->session);
3083 if (ret < 0) {
3084 ret = LTTNG_ERR_KERN_SESS_FAIL;
3085 goto error;
3086 }
3087 }
3088
3089 /* Start the kernel consumer daemon */
3090 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3091 if (kconsumer_data.pid == 0 &&
3092 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3093 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3094 ret = start_consumerd(&kconsumer_data);
3095 if (ret < 0) {
3096 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3097 goto error;
3098 }
3099 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3100 } else {
3101 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3102 }
3103
3104 /*
3105 * The consumer was just spawned so we need to add the socket to
3106 * the consumer output of the session if exist.
3107 */
3108 ret = consumer_create_socket(&kconsumer_data,
3109 cmd_ctx->session->kernel_session->consumer);
3110 if (ret < 0) {
3111 goto error;
3112 }
3113 }
3114
3115 break;
3116 case LTTNG_DOMAIN_JUL:
3117 case LTTNG_DOMAIN_LOG4J:
3118 case LTTNG_DOMAIN_PYTHON:
3119 case LTTNG_DOMAIN_UST:
3120 {
3121 if (!ust_app_supported()) {
3122 ret = LTTNG_ERR_NO_UST;
3123 goto error;
3124 }
3125 /* Consumer is in an ERROR state. Report back to client */
3126 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3127 ret = LTTNG_ERR_NO_USTCONSUMERD;
3128 goto error;
3129 }
3130
3131 if (need_tracing_session) {
3132 /* Create UST session if none exist. */
3133 if (cmd_ctx->session->ust_session == NULL) {
3134 ret = create_ust_session(cmd_ctx->session,
3135 &cmd_ctx->lsm->domain);
3136 if (ret != LTTNG_OK) {
3137 goto error;
3138 }
3139 }
3140
3141 /* Start the UST consumer daemons */
3142 /* 64-bit */
3143 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3144 if (config.consumerd64_bin_path.value &&
3145 ustconsumer64_data.pid == 0 &&
3146 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3147 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3148 ret = start_consumerd(&ustconsumer64_data);
3149 if (ret < 0) {
3150 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3151 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3152 goto error;
3153 }
3154
3155 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3156 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3157 } else {
3158 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3159 }
3160
3161 /*
3162 * Setup socket for consumer 64 bit. No need for atomic access
3163 * since it was set above and can ONLY be set in this thread.
3164 */
3165 ret = consumer_create_socket(&ustconsumer64_data,
3166 cmd_ctx->session->ust_session->consumer);
3167 if (ret < 0) {
3168 goto error;
3169 }
3170
3171 /* 32-bit */
3172 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3173 if (config.consumerd32_bin_path.value &&
3174 ustconsumer32_data.pid == 0 &&
3175 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3176 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3177 ret = start_consumerd(&ustconsumer32_data);
3178 if (ret < 0) {
3179 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3180 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3181 goto error;
3182 }
3183
3184 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3185 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3186 } else {
3187 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3188 }
3189
3190 /*
3191 * Setup socket for consumer 64 bit. No need for atomic access
3192 * since it was set above and can ONLY be set in this thread.
3193 */
3194 ret = consumer_create_socket(&ustconsumer32_data,
3195 cmd_ctx->session->ust_session->consumer);
3196 if (ret < 0) {
3197 goto error;
3198 }
3199 }
3200 break;
3201 }
3202 default:
3203 break;
3204 }
3205 skip_domain:
3206
3207 /* Validate consumer daemon state when start/stop trace command */
3208 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3209 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3210 switch (cmd_ctx->lsm->domain.type) {
3211 case LTTNG_DOMAIN_NONE:
3212 break;
3213 case LTTNG_DOMAIN_JUL:
3214 case LTTNG_DOMAIN_LOG4J:
3215 case LTTNG_DOMAIN_PYTHON:
3216 case LTTNG_DOMAIN_UST:
3217 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3218 ret = LTTNG_ERR_NO_USTCONSUMERD;
3219 goto error;
3220 }
3221 break;
3222 case LTTNG_DOMAIN_KERNEL:
3223 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3224 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3225 goto error;
3226 }
3227 break;
3228 default:
3229 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3230 goto error;
3231 }
3232 }
3233
3234 /*
3235 * Check that the UID or GID match that of the tracing session.
3236 * The root user can interact with all sessions.
3237 */
3238 if (need_tracing_session) {
3239 if (!session_access_ok(cmd_ctx->session,
3240 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3241 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3242 ret = LTTNG_ERR_EPERM;
3243 goto error;
3244 }
3245 }
3246
3247 /*
3248 * Send relayd information to consumer as soon as we have a domain and a
3249 * session defined.
3250 */
3251 if (cmd_ctx->session && need_domain) {
3252 /*
3253 * Setup relayd if not done yet. If the relayd information was already
3254 * sent to the consumer, this call will gracefully return.
3255 */
3256 ret = cmd_setup_relayd(cmd_ctx->session);
3257 if (ret != LTTNG_OK) {
3258 goto error;
3259 }
3260 }
3261
3262 /* Process by command type */
3263 switch (cmd_ctx->lsm->cmd_type) {
3264 case LTTNG_ADD_CONTEXT:
3265 {
3266 /*
3267 * An LTTNG_ADD_CONTEXT command might have a supplementary
3268 * payload if the context being added is an application context.
3269 */
3270 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3271 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3272 char *provider_name = NULL, *context_name = NULL;
3273 size_t provider_name_len =
3274 cmd_ctx->lsm->u.context.provider_name_len;
3275 size_t context_name_len =
3276 cmd_ctx->lsm->u.context.context_name_len;
3277
3278 if (provider_name_len == 0 || context_name_len == 0) {
3279 /*
3280 * Application provider and context names MUST
3281 * be provided.
3282 */
3283 ret = -LTTNG_ERR_INVALID;
3284 goto error;
3285 }
3286
3287 provider_name = zmalloc(provider_name_len + 1);
3288 if (!provider_name) {
3289 ret = -LTTNG_ERR_NOMEM;
3290 goto error;
3291 }
3292 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3293 provider_name;
3294
3295 context_name = zmalloc(context_name_len + 1);
3296 if (!context_name) {
3297 ret = -LTTNG_ERR_NOMEM;
3298 goto error_add_context;
3299 }
3300 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3301 context_name;
3302
3303 ret = lttcomm_recv_unix_sock(sock, provider_name,
3304 provider_name_len);
3305 if (ret < 0) {
3306 goto error_add_context;
3307 }
3308
3309 ret = lttcomm_recv_unix_sock(sock, context_name,
3310 context_name_len);
3311 if (ret < 0) {
3312 goto error_add_context;
3313 }
3314 }
3315
3316 /*
3317 * cmd_add_context assumes ownership of the provider and context
3318 * names.
3319 */
3320 ret = cmd_add_context(cmd_ctx->session,
3321 cmd_ctx->lsm->domain.type,
3322 cmd_ctx->lsm->u.context.channel_name,
3323 &cmd_ctx->lsm->u.context.ctx,
3324 kernel_poll_pipe[1]);
3325
3326 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3327 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3328 error_add_context:
3329 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3330 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3331 if (ret < 0) {
3332 goto error;
3333 }
3334 break;
3335 }
3336 case LTTNG_DISABLE_CHANNEL:
3337 {
3338 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3339 cmd_ctx->lsm->u.disable.channel_name);
3340 break;
3341 }
3342 case LTTNG_DISABLE_EVENT:
3343 {
3344
3345 /*
3346 * FIXME: handle filter; for now we just receive the filter's
3347 * bytecode along with the filter expression which are sent by
3348 * liblttng-ctl and discard them.
3349 *
3350 * This fixes an issue where the client may block while sending
3351 * the filter payload and encounter an error because the session
3352 * daemon closes the socket without ever handling this data.
3353 */
3354 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3355 cmd_ctx->lsm->u.disable.bytecode_len;
3356
3357 if (count) {
3358 char data[LTTNG_FILTER_MAX_LEN];
3359
3360 DBG("Discarding disable event command payload of size %zu", count);
3361 while (count) {
3362 ret = lttcomm_recv_unix_sock(sock, data,
3363 count > sizeof(data) ? sizeof(data) : count);
3364 if (ret < 0) {
3365 goto error;
3366 }
3367
3368 count -= (size_t) ret;
3369 }
3370 }
3371 /* FIXME: passing packed structure to non-packed pointer */
3372 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3373 cmd_ctx->lsm->u.disable.channel_name,
3374 &cmd_ctx->lsm->u.disable.event);
3375 break;
3376 }
3377 case LTTNG_ENABLE_CHANNEL:
3378 {
3379 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3380 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3381 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3382 &cmd_ctx->lsm->u.channel.chan,
3383 kernel_poll_pipe[1]);
3384 break;
3385 }
3386 case LTTNG_TRACK_PID:
3387 {
3388 ret = cmd_track_pid(cmd_ctx->session,
3389 cmd_ctx->lsm->domain.type,
3390 cmd_ctx->lsm->u.pid_tracker.pid);
3391 break;
3392 }
3393 case LTTNG_UNTRACK_PID:
3394 {
3395 ret = cmd_untrack_pid(cmd_ctx->session,
3396 cmd_ctx->lsm->domain.type,
3397 cmd_ctx->lsm->u.pid_tracker.pid);
3398 break;
3399 }
3400 case LTTNG_ENABLE_EVENT:
3401 {
3402 struct lttng_event_exclusion *exclusion = NULL;
3403 struct lttng_filter_bytecode *bytecode = NULL;
3404 char *filter_expression = NULL;
3405
3406 /* Handle exclusion events and receive it from the client. */
3407 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3408 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3409
3410 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3411 (count * LTTNG_SYMBOL_NAME_LEN));
3412 if (!exclusion) {
3413 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3414 goto error;
3415 }
3416
3417 DBG("Receiving var len exclusion event list from client ...");
3418 exclusion->count = count;
3419 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3420 count * LTTNG_SYMBOL_NAME_LEN);
3421 if (ret <= 0) {
3422 DBG("Nothing recv() from client var len data... continuing");
3423 *sock_error = 1;
3424 free(exclusion);
3425 ret = LTTNG_ERR_EXCLUSION_INVAL;
3426 goto error;
3427 }
3428 }
3429
3430 /* Get filter expression from client. */
3431 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3432 size_t expression_len =
3433 cmd_ctx->lsm->u.enable.expression_len;
3434
3435 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3436 ret = LTTNG_ERR_FILTER_INVAL;
3437 free(exclusion);
3438 goto error;
3439 }
3440
3441 filter_expression = zmalloc(expression_len);
3442 if (!filter_expression) {
3443 free(exclusion);
3444 ret = LTTNG_ERR_FILTER_NOMEM;
3445 goto error;
3446 }
3447
3448 /* Receive var. len. data */
3449 DBG("Receiving var len filter's expression from client ...");
3450 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3451 expression_len);
3452 if (ret <= 0) {
3453 DBG("Nothing recv() from client car len data... continuing");
3454 *sock_error = 1;
3455 free(filter_expression);
3456 free(exclusion);
3457 ret = LTTNG_ERR_FILTER_INVAL;
3458 goto error;
3459 }
3460 }
3461
3462 /* Handle filter and get bytecode from client. */
3463 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3464 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3465
3466 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3467 ret = LTTNG_ERR_FILTER_INVAL;
3468 free(filter_expression);
3469 free(exclusion);
3470 goto error;
3471 }
3472
3473 bytecode = zmalloc(bytecode_len);
3474 if (!bytecode) {
3475 free(filter_expression);
3476 free(exclusion);
3477 ret = LTTNG_ERR_FILTER_NOMEM;
3478 goto error;
3479 }
3480
3481 /* Receive var. len. data */
3482 DBG("Receiving var len filter's bytecode from client ...");
3483 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3484 if (ret <= 0) {
3485 DBG("Nothing recv() from client car len data... continuing");
3486 *sock_error = 1;
3487 free(filter_expression);
3488 free(bytecode);
3489 free(exclusion);
3490 ret = LTTNG_ERR_FILTER_INVAL;
3491 goto error;
3492 }
3493
3494 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3495 free(filter_expression);
3496 free(bytecode);
3497 free(exclusion);
3498 ret = LTTNG_ERR_FILTER_INVAL;
3499 goto error;
3500 }
3501 }
3502
3503 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3504 cmd_ctx->lsm->u.enable.channel_name,
3505 &cmd_ctx->lsm->u.enable.event,
3506 filter_expression, bytecode, exclusion,
3507 kernel_poll_pipe[1]);
3508 break;
3509 }
3510 case LTTNG_LIST_TRACEPOINTS:
3511 {
3512 struct lttng_event *events;
3513 ssize_t nb_events;
3514
3515 session_lock_list();
3516 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3517 session_unlock_list();
3518 if (nb_events < 0) {
3519 /* Return value is a negative lttng_error_code. */
3520 ret = -nb_events;
3521 goto error;
3522 }
3523
3524 /*
3525 * Setup lttng message with payload size set to the event list size in
3526 * bytes and then copy list into the llm payload.
3527 */
3528 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3529 sizeof(struct lttng_event) * nb_events);
3530 free(events);
3531
3532 if (ret < 0) {
3533 goto setup_error;
3534 }
3535
3536 ret = LTTNG_OK;
3537 break;
3538 }
3539 case LTTNG_LIST_TRACEPOINT_FIELDS:
3540 {
3541 struct lttng_event_field *fields;
3542 ssize_t nb_fields;
3543
3544 session_lock_list();
3545 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3546 &fields);
3547 session_unlock_list();
3548 if (nb_fields < 0) {
3549 /* Return value is a negative lttng_error_code. */
3550 ret = -nb_fields;
3551 goto error;
3552 }
3553
3554 /*
3555 * Setup lttng message with payload size set to the event list size in
3556 * bytes and then copy list into the llm payload.
3557 */
3558 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3559 sizeof(struct lttng_event_field) * nb_fields);
3560 free(fields);
3561
3562 if (ret < 0) {
3563 goto setup_error;
3564 }
3565
3566 ret = LTTNG_OK;
3567 break;
3568 }
3569 case LTTNG_LIST_SYSCALLS:
3570 {
3571 struct lttng_event *events;
3572 ssize_t nb_events;
3573
3574 nb_events = cmd_list_syscalls(&events);
3575 if (nb_events < 0) {
3576 /* Return value is a negative lttng_error_code. */
3577 ret = -nb_events;
3578 goto error;
3579 }
3580
3581 /*
3582 * Setup lttng message with payload size set to the event list size in
3583 * bytes and then copy list into the llm payload.
3584 */
3585 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3586 sizeof(struct lttng_event) * nb_events);
3587 free(events);
3588
3589 if (ret < 0) {
3590 goto setup_error;
3591 }
3592
3593 ret = LTTNG_OK;
3594 break;
3595 }
3596 case LTTNG_LIST_TRACKER_PIDS:
3597 {
3598 int32_t *pids = NULL;
3599 ssize_t nr_pids;
3600
3601 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3602 cmd_ctx->lsm->domain.type, &pids);
3603 if (nr_pids < 0) {
3604 /* Return value is a negative lttng_error_code. */
3605 ret = -nr_pids;
3606 goto error;
3607 }
3608
3609 /*
3610 * Setup lttng message with payload size set to the event list size in
3611 * bytes and then copy list into the llm payload.
3612 */
3613 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3614 sizeof(int32_t) * nr_pids);
3615 free(pids);
3616
3617 if (ret < 0) {
3618 goto setup_error;
3619 }
3620
3621 ret = LTTNG_OK;
3622 break;
3623 }
3624 case LTTNG_SET_CONSUMER_URI:
3625 {
3626 size_t nb_uri, len;
3627 struct lttng_uri *uris;
3628
3629 nb_uri = cmd_ctx->lsm->u.uri.size;
3630 len = nb_uri * sizeof(struct lttng_uri);
3631
3632 if (nb_uri == 0) {
3633 ret = LTTNG_ERR_INVALID;
3634 goto error;
3635 }
3636
3637 uris = zmalloc(len);
3638 if (uris == NULL) {
3639 ret = LTTNG_ERR_FATAL;
3640 goto error;
3641 }
3642
3643 /* Receive variable len data */
3644 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3645 ret = lttcomm_recv_unix_sock(sock, uris, len);
3646 if (ret <= 0) {
3647 DBG("No URIs received from client... continuing");
3648 *sock_error = 1;
3649 ret = LTTNG_ERR_SESSION_FAIL;
3650 free(uris);
3651 goto error;
3652 }
3653
3654 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3655 free(uris);
3656 if (ret != LTTNG_OK) {
3657 goto error;
3658 }
3659
3660
3661 break;
3662 }
3663 case LTTNG_START_TRACE:
3664 {
3665 ret = cmd_start_trace(cmd_ctx->session);
3666 break;
3667 }
3668 case LTTNG_STOP_TRACE:
3669 {
3670 ret = cmd_stop_trace(cmd_ctx->session);
3671 break;
3672 }
3673 case LTTNG_CREATE_SESSION:
3674 {
3675 size_t nb_uri, len;
3676 struct lttng_uri *uris = NULL;
3677
3678 nb_uri = cmd_ctx->lsm->u.uri.size;
3679 len = nb_uri * sizeof(struct lttng_uri);
3680
3681 if (nb_uri > 0) {
3682 uris = zmalloc(len);
3683 if (uris == NULL) {
3684 ret = LTTNG_ERR_FATAL;
3685 goto error;
3686 }
3687
3688 /* Receive variable len data */
3689 DBG("Waiting for %zu URIs from client ...", nb_uri);
3690 ret = lttcomm_recv_unix_sock(sock, uris, len);
3691 if (ret <= 0) {
3692 DBG("No URIs received from client... continuing");
3693 *sock_error = 1;
3694 ret = LTTNG_ERR_SESSION_FAIL;
3695 free(uris);
3696 goto error;
3697 }
3698
3699 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3700 DBG("Creating session with ONE network URI is a bad call");
3701 ret = LTTNG_ERR_SESSION_FAIL;
3702 free(uris);
3703 goto error;
3704 }
3705 }
3706
3707 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3708 &cmd_ctx->creds, 0);
3709
3710 free(uris);
3711
3712 break;
3713 }
3714 case LTTNG_DESTROY_SESSION:
3715 {
3716 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3717
3718 /* Set session to NULL so we do not unlock it after free. */
3719 cmd_ctx->session = NULL;
3720 break;
3721 }
3722 case LTTNG_LIST_DOMAINS:
3723 {
3724 ssize_t nb_dom;
3725 struct lttng_domain *domains = NULL;
3726
3727 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3728 if (nb_dom < 0) {
3729 /* Return value is a negative lttng_error_code. */
3730 ret = -nb_dom;
3731 goto error;
3732 }
3733
3734 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3735 nb_dom * sizeof(struct lttng_domain));
3736 free(domains);
3737
3738 if (ret < 0) {
3739 goto setup_error;
3740 }
3741
3742 ret = LTTNG_OK;
3743 break;
3744 }
3745 case LTTNG_LIST_CHANNELS:
3746 {
3747 ssize_t payload_size;
3748 struct lttng_channel *channels = NULL;
3749
3750 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3751 cmd_ctx->session, &channels);
3752 if (payload_size < 0) {
3753 /* Return value is a negative lttng_error_code. */
3754 ret = -payload_size;
3755 goto error;
3756 }
3757
3758 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3759 payload_size);
3760 free(channels);
3761
3762 if (ret < 0) {
3763 goto setup_error;
3764 }
3765
3766 ret = LTTNG_OK;
3767 break;
3768 }
3769 case LTTNG_LIST_EVENTS:
3770 {
3771 ssize_t nb_event;
3772 struct lttng_event *events = NULL;
3773 struct lttcomm_event_command_header cmd_header;
3774 size_t total_size;
3775
3776 memset(&cmd_header, 0, sizeof(cmd_header));
3777 /* Extended infos are included at the end of events */
3778 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3779 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3780 &events, &total_size);
3781
3782 if (nb_event < 0) {
3783 /* Return value is a negative lttng_error_code. */
3784 ret = -nb_event;
3785 goto error;
3786 }
3787
3788 cmd_header.nb_events = nb_event;
3789 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3790 &cmd_header, sizeof(cmd_header));
3791 free(events);
3792
3793 if (ret < 0) {
3794 goto setup_error;
3795 }
3796
3797 ret = LTTNG_OK;
3798 break;
3799 }
3800 case LTTNG_LIST_SESSIONS:
3801 {
3802 unsigned int nr_sessions;
3803 void *sessions_payload;
3804 size_t payload_len;
3805
3806 session_lock_list();
3807 nr_sessions = lttng_sessions_count(
3808 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3809 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3810 payload_len = sizeof(struct lttng_session) * nr_sessions;
3811 sessions_payload = zmalloc(payload_len);
3812
3813 if (!sessions_payload) {
3814 session_unlock_list();
3815 ret = -ENOMEM;
3816 goto setup_error;
3817 }
3818
3819 cmd_list_lttng_sessions(sessions_payload,
3820 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3821 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3822 session_unlock_list();
3823
3824 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3825 payload_len);
3826 free(sessions_payload);
3827
3828 if (ret < 0) {
3829 goto setup_error;
3830 }
3831
3832 ret = LTTNG_OK;
3833 break;
3834 }
3835 case LTTNG_REGISTER_CONSUMER:
3836 {
3837 struct consumer_data *cdata;
3838
3839 switch (cmd_ctx->lsm->domain.type) {
3840 case LTTNG_DOMAIN_KERNEL:
3841 cdata = &kconsumer_data;
3842 break;
3843 default:
3844 ret = LTTNG_ERR_UND;
3845 goto error;
3846 }
3847
3848 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3849 cmd_ctx->lsm->u.reg.path, cdata);
3850 break;
3851 }
3852 case LTTNG_DATA_PENDING:
3853 {
3854 int pending_ret;
3855 uint8_t pending_ret_byte;
3856
3857 pending_ret = cmd_data_pending(cmd_ctx->session);
3858
3859 /*
3860 * FIXME
3861 *
3862 * This function may returns 0 or 1 to indicate whether or not
3863 * there is data pending. In case of error, it should return an
3864 * LTTNG_ERR code. However, some code paths may still return
3865 * a nondescript error code, which we handle by returning an
3866 * "unknown" error.
3867 */
3868 if (pending_ret == 0 || pending_ret == 1) {
3869 /*
3870 * ret will be set to LTTNG_OK at the end of
3871 * this function.
3872 */
3873 } else if (pending_ret < 0) {
3874 ret = LTTNG_ERR_UNK;
3875 goto setup_error;
3876 } else {
3877 ret = pending_ret;
3878 goto setup_error;
3879 }
3880
3881 pending_ret_byte = (uint8_t) pending_ret;
3882
3883 /* 1 byte to return whether or not data is pending */
3884 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
3885 &pending_ret_byte, 1);
3886
3887 if (ret < 0) {
3888 goto setup_error;
3889 }
3890
3891 ret = LTTNG_OK;
3892 break;
3893 }
3894 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3895 {
3896 struct lttcomm_lttng_output_id reply;
3897
3898 ret = cmd_snapshot_add_output(cmd_ctx->session,
3899 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3900 if (ret != LTTNG_OK) {
3901 goto error;
3902 }
3903
3904 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
3905 sizeof(reply));
3906 if (ret < 0) {
3907 goto setup_error;
3908 }
3909
3910 /* Copy output list into message payload */
3911 ret = LTTNG_OK;
3912 break;
3913 }
3914 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3915 {
3916 ret = cmd_snapshot_del_output(cmd_ctx->session,
3917 &cmd_ctx->lsm->u.snapshot_output.output);
3918 break;
3919 }
3920 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3921 {
3922 ssize_t nb_output;
3923 struct lttng_snapshot_output *outputs = NULL;
3924
3925 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3926 if (nb_output < 0) {
3927 ret = -nb_output;
3928 goto error;
3929 }
3930
3931 assert((nb_output > 0 && outputs) || nb_output == 0);
3932 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
3933 nb_output * sizeof(struct lttng_snapshot_output));
3934 free(outputs);
3935
3936 if (ret < 0) {
3937 goto setup_error;
3938 }
3939
3940 ret = LTTNG_OK;
3941 break;
3942 }
3943 case LTTNG_SNAPSHOT_RECORD:
3944 {
3945 ret = cmd_snapshot_record(cmd_ctx->session,
3946 &cmd_ctx->lsm->u.snapshot_record.output,
3947 cmd_ctx->lsm->u.snapshot_record.wait);
3948 break;
3949 }
3950 case LTTNG_CREATE_SESSION_SNAPSHOT:
3951 {
3952 size_t nb_uri, len;
3953 struct lttng_uri *uris = NULL;
3954
3955 nb_uri = cmd_ctx->lsm->u.uri.size;
3956 len = nb_uri * sizeof(struct lttng_uri);
3957
3958 if (nb_uri > 0) {
3959 uris = zmalloc(len);
3960 if (uris == NULL) {
3961 ret = LTTNG_ERR_FATAL;
3962 goto error;
3963 }
3964
3965 /* Receive variable len data */
3966 DBG("Waiting for %zu URIs from client ...", nb_uri);
3967 ret = lttcomm_recv_unix_sock(sock, uris, len);
3968 if (ret <= 0) {
3969 DBG("No URIs received from client... continuing");
3970 *sock_error = 1;
3971 ret = LTTNG_ERR_SESSION_FAIL;
3972 free(uris);
3973 goto error;
3974 }
3975
3976 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3977 DBG("Creating session with ONE network URI is a bad call");
3978 ret = LTTNG_ERR_SESSION_FAIL;
3979 free(uris);
3980 goto error;
3981 }
3982 }
3983
3984 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3985 nb_uri, &cmd_ctx->creds);
3986 free(uris);
3987 break;
3988 }
3989 case LTTNG_CREATE_SESSION_LIVE:
3990 {
3991 size_t nb_uri, len;
3992 struct lttng_uri *uris = NULL;
3993
3994 nb_uri = cmd_ctx->lsm->u.uri.size;
3995 len = nb_uri * sizeof(struct lttng_uri);
3996
3997 if (nb_uri > 0) {
3998 uris = zmalloc(len);
3999 if (uris == NULL) {
4000 ret = LTTNG_ERR_FATAL;
4001 goto error;
4002 }
4003
4004 /* Receive variable len data */
4005 DBG("Waiting for %zu URIs from client ...", nb_uri);
4006 ret = lttcomm_recv_unix_sock(sock, uris, len);
4007 if (ret <= 0) {
4008 DBG("No URIs received from client... continuing");
4009 *sock_error = 1;
4010 ret = LTTNG_ERR_SESSION_FAIL;
4011 free(uris);
4012 goto error;
4013 }
4014
4015 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4016 DBG("Creating session with ONE network URI is a bad call");
4017 ret = LTTNG_ERR_SESSION_FAIL;
4018 free(uris);
4019 goto error;
4020 }
4021 }
4022
4023 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4024 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4025 free(uris);
4026 break;
4027 }
4028 case LTTNG_SAVE_SESSION:
4029 {
4030 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4031 &cmd_ctx->creds);
4032 break;
4033 }
4034 case LTTNG_SET_SESSION_SHM_PATH:
4035 {
4036 ret = cmd_set_session_shm_path(cmd_ctx->session,
4037 cmd_ctx->lsm->u.set_shm_path.shm_path);
4038 break;
4039 }
4040 case LTTNG_REGENERATE_METADATA:
4041 {
4042 ret = cmd_regenerate_metadata(cmd_ctx->session);
4043 break;
4044 }
4045 case LTTNG_REGENERATE_STATEDUMP:
4046 {
4047 ret = cmd_regenerate_statedump(cmd_ctx->session);
4048 break;
4049 }
4050 case LTTNG_REGISTER_TRIGGER:
4051 {
4052 ret = cmd_register_trigger(cmd_ctx, sock,
4053 notification_thread_handle);
4054 break;
4055 }
4056 case LTTNG_UNREGISTER_TRIGGER:
4057 {
4058 ret = cmd_unregister_trigger(cmd_ctx, sock,
4059 notification_thread_handle);
4060 break;
4061 }
4062 default:
4063 ret = LTTNG_ERR_UND;
4064 break;
4065 }
4066
4067 error:
4068 if (cmd_ctx->llm == NULL) {
4069 DBG("Missing llm structure. Allocating one.");
4070 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4071 goto setup_error;
4072 }
4073 }
4074 /* Set return code */
4075 cmd_ctx->llm->ret_code = ret;
4076 setup_error:
4077 if (cmd_ctx->session) {
4078 session_unlock(cmd_ctx->session);
4079 }
4080 if (need_tracing_session) {
4081 session_unlock_list();
4082 }
4083 init_setup_error:
4084 assert(!rcu_read_ongoing());
4085 return ret;
4086 }
4087
4088 /*
4089 * Thread managing health check socket.
4090 */
4091 static void *thread_manage_health(void *data)
4092 {
4093 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4094 uint32_t revents, nb_fd;
4095 struct lttng_poll_event events;
4096 struct health_comm_msg msg;
4097 struct health_comm_reply reply;
4098
4099 DBG("[thread] Manage health check started");
4100
4101 rcu_register_thread();
4102
4103 /* We might hit an error path before this is created. */
4104 lttng_poll_init(&events);
4105
4106 /* Create unix socket */
4107 sock = lttcomm_create_unix_sock(config.health_unix_sock_path.value);
4108 if (sock < 0) {
4109 ERR("Unable to create health check Unix socket");
4110 goto error;
4111 }
4112
4113 if (is_root) {
4114 /* lttng health client socket path permissions */
4115 ret = chown(config.health_unix_sock_path.value, 0,
4116 utils_get_group_id(config.tracing_group_name.value));
4117 if (ret < 0) {
4118 ERR("Unable to set group on %s", config.health_unix_sock_path.value);
4119 PERROR("chown");
4120 goto error;
4121 }
4122
4123 ret = chmod(config.health_unix_sock_path.value,
4124 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4125 if (ret < 0) {
4126 ERR("Unable to set permissions on %s", config.health_unix_sock_path.value);
4127 PERROR("chmod");
4128 goto error;
4129 }
4130 }
4131
4132 /*
4133 * Set the CLOEXEC flag. Return code is useless because either way, the
4134 * show must go on.
4135 */
4136 (void) utils_set_fd_cloexec(sock);
4137
4138 ret = lttcomm_listen_unix_sock(sock);
4139 if (ret < 0) {
4140 goto error;
4141 }
4142
4143 /*
4144 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4145 * more will be added to this poll set.
4146 */
4147 ret = sessiond_set_thread_pollset(&events, 2);
4148 if (ret < 0) {
4149 goto error;
4150 }
4151
4152 /* Add the application registration socket */
4153 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4154 if (ret < 0) {
4155 goto error;
4156 }
4157
4158 sessiond_notify_ready();
4159
4160 while (1) {
4161 DBG("Health check ready");
4162
4163 /* Inifinite blocking call, waiting for transmission */
4164 restart:
4165 ret = lttng_poll_wait(&events, -1);
4166 if (ret < 0) {
4167 /*
4168 * Restart interrupted system call.
4169 */
4170 if (errno == EINTR) {
4171 goto restart;
4172 }
4173 goto error;
4174 }
4175
4176 nb_fd = ret;
4177
4178 for (i = 0; i < nb_fd; i++) {
4179 /* Fetch once the poll data */
4180 revents = LTTNG_POLL_GETEV(&events, i);
4181 pollfd = LTTNG_POLL_GETFD(&events, i);
4182
4183 if (!revents) {
4184 /* No activity for this FD (poll implementation). */
4185 continue;
4186 }
4187
4188 /* Thread quit pipe has been closed. Killing thread. */
4189 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4190 if (ret) {
4191 err = 0;
4192 goto exit;
4193 }
4194
4195 /* Event on the registration socket */
4196 if (pollfd == sock) {
4197 if (revents & LPOLLIN) {
4198 continue;
4199 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4200 ERR("Health socket poll error");
4201 goto error;
4202 } else {
4203 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4204 goto error;
4205 }
4206 }
4207 }
4208
4209 new_sock = lttcomm_accept_unix_sock(sock);
4210 if (new_sock < 0) {
4211 goto error;
4212 }
4213
4214 /*
4215 * Set the CLOEXEC flag. Return code is useless because either way, the
4216 * show must go on.
4217 */
4218 (void) utils_set_fd_cloexec(new_sock);
4219
4220 DBG("Receiving data from client for health...");
4221 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4222 if (ret <= 0) {
4223 DBG("Nothing recv() from client... continuing");
4224 ret = close(new_sock);
4225 if (ret) {
4226 PERROR("close");
4227 }
4228 continue;
4229 }
4230
4231 rcu_thread_online();
4232
4233 memset(&reply, 0, sizeof(reply));
4234 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4235 /*
4236 * health_check_state returns 0 if health is
4237 * bad.
4238 */
4239 if (!health_check_state(health_sessiond, i)) {
4240 reply.ret_code |= 1ULL << i;
4241 }
4242 }
4243
4244 DBG2("Health check return value %" PRIx64, reply.ret_code);
4245
4246 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4247 if (ret < 0) {
4248 ERR("Failed to send health data back to client");
4249 }
4250
4251 /* End of transmission */
4252 ret = close(new_sock);
4253 if (ret) {
4254 PERROR("close");
4255 }
4256 }
4257
4258 exit:
4259 error:
4260 if (err) {
4261 ERR("Health error occurred in %s", __func__);
4262 }
4263 DBG("Health check thread dying");
4264 unlink(config.health_unix_sock_path.value);
4265 if (sock >= 0) {
4266 ret = close(sock);
4267 if (ret) {
4268 PERROR("close");
4269 }
4270 }
4271
4272 lttng_poll_clean(&events);
4273 stop_threads();
4274 rcu_unregister_thread();
4275 return NULL;
4276 }
4277
4278 /*
4279 * This thread manage all clients request using the unix client socket for
4280 * communication.
4281 */
4282 static void *thread_manage_clients(void *data)
4283 {
4284 int sock = -1, ret, i, pollfd, err = -1;
4285 int sock_error;
4286 uint32_t revents, nb_fd;
4287 struct command_ctx *cmd_ctx = NULL;
4288 struct lttng_poll_event events;
4289
4290 DBG("[thread] Manage client started");
4291
4292 rcu_register_thread();
4293
4294 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4295
4296 health_code_update();
4297
4298 ret = lttcomm_listen_unix_sock(client_sock);
4299 if (ret < 0) {
4300 goto error_listen;
4301 }
4302
4303 /*
4304 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4305 * more will be added to this poll set.
4306 */
4307 ret = sessiond_set_thread_pollset(&events, 2);
4308 if (ret < 0) {
4309 goto error_create_poll;
4310 }
4311
4312 /* Add the application registration socket */
4313 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4314 if (ret < 0) {
4315 goto error;
4316 }
4317
4318 ret = sem_post(&load_info->message_thread_ready);
4319 if (ret) {
4320 PERROR("sem_post message_thread_ready");
4321 goto error;
4322 }
4323
4324 /*
4325 * Wait until all support threads are initialized before accepting
4326 * commands.
4327 */
4328 while (uatomic_read(&lttng_sessiond_ready) != 0) {
4329 fd_set read_fds;
4330 struct timeval timeout;
4331
4332 FD_ZERO(&read_fds);
4333 FD_SET(thread_quit_pipe[0], &read_fds);
4334 memset(&timeout, 0, sizeof(timeout));
4335 timeout.tv_usec = 1000;
4336
4337 /*
4338 * If a support thread failed to launch, it may signal that
4339 * we must exit and the sessiond would never be marked as
4340 * "ready".
4341 *
4342 * The timeout is set to 1ms, which serves as a way to
4343 * pace down this check.
4344 */
4345 ret = select(thread_quit_pipe[0] + 1, &read_fds, NULL, NULL,
4346 &timeout);
4347 if (ret > 0 || (ret < 0 && errno != EINTR)) {
4348 goto exit;
4349 }
4350 }
4351 /*
4352 * This barrier is paired with the one in sessiond_notify_ready() to
4353 * ensure that loads accessing data initialized by the other threads,
4354 * on which this thread was waiting, are not performed before this point.
4355 *
4356 * Note that this could be a 'read' memory barrier, but a full barrier
4357 * is used in case the code changes. The performance implications of
4358 * this choice are minimal since this is a slow path.
4359 */
4360 cmm_smp_mb();
4361
4362 /* This testpoint is after we signal readiness to the parent. */
4363 if (testpoint(sessiond_thread_manage_clients)) {
4364 goto error;
4365 }
4366
4367 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4368 goto error;
4369 }
4370
4371 health_code_update();
4372
4373 while (1) {
4374 DBG("Accepting client command ...");
4375
4376 /* Inifinite blocking call, waiting for transmission */
4377 restart:
4378 health_poll_entry();
4379 ret = lttng_poll_wait(&events, -1);
4380 health_poll_exit();
4381 if (ret < 0) {
4382 /*
4383 * Restart interrupted system call.
4384 */
4385 if (errno == EINTR) {
4386 goto restart;
4387 }
4388 goto error;
4389 }
4390
4391 nb_fd = ret;
4392
4393 for (i = 0; i < nb_fd; i++) {
4394 /* Fetch once the poll data */
4395 revents = LTTNG_POLL_GETEV(&events, i);
4396 pollfd = LTTNG_POLL_GETFD(&events, i);
4397
4398 health_code_update();
4399
4400 if (!revents) {
4401 /* No activity for this FD (poll implementation). */
4402 continue;
4403 }
4404
4405 /* Thread quit pipe has been closed. Killing thread. */
4406 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4407 if (ret) {
4408 err = 0;
4409 goto exit;
4410 }
4411
4412 /* Event on the registration socket */
4413 if (pollfd == client_sock) {
4414 if (revents & LPOLLIN) {
4415 continue;
4416 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4417 ERR("Client socket poll error");
4418 goto error;
4419 } else {
4420 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4421 goto error;
4422 }
4423 }
4424 }
4425
4426 DBG("Wait for client response");
4427
4428 health_code_update();
4429
4430 sock = lttcomm_accept_unix_sock(client_sock);
4431 if (sock < 0) {
4432 goto error;
4433 }
4434
4435 /*
4436 * Set the CLOEXEC flag. Return code is useless because either way, the
4437 * show must go on.
4438 */
4439 (void) utils_set_fd_cloexec(sock);
4440
4441 /* Set socket option for credentials retrieval */
4442 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4443 if (ret < 0) {
4444 goto error;
4445 }
4446
4447 /* Allocate context command to process the client request */
4448 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4449 if (cmd_ctx == NULL) {
4450 PERROR("zmalloc cmd_ctx");
4451 goto error;
4452 }
4453
4454 /* Allocate data buffer for reception */
4455 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4456 if (cmd_ctx->lsm == NULL) {
4457 PERROR("zmalloc cmd_ctx->lsm");
4458 goto error;
4459 }
4460
4461 cmd_ctx->llm = NULL;
4462 cmd_ctx->session = NULL;
4463
4464 health_code_update();
4465
4466 /*
4467 * Data is received from the lttng client. The struct
4468 * lttcomm_session_msg (lsm) contains the command and data request of
4469 * the client.
4470 */
4471 DBG("Receiving data from client ...");
4472 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4473 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4474 if (ret <= 0) {
4475 DBG("Nothing recv() from client... continuing");
4476 ret = close(sock);
4477 if (ret) {
4478 PERROR("close");
4479 }
4480 sock = -1;
4481 clean_command_ctx(&cmd_ctx);
4482 continue;
4483 }
4484
4485 health_code_update();
4486
4487 // TODO: Validate cmd_ctx including sanity check for
4488 // security purpose.
4489
4490 rcu_thread_online();
4491 /*
4492 * This function dispatch the work to the kernel or userspace tracer
4493 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4494 * informations for the client. The command context struct contains
4495 * everything this function may needs.
4496 */
4497 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4498 rcu_thread_offline();
4499 if (ret < 0) {
4500 ret = close(sock);
4501 if (ret) {
4502 PERROR("close");
4503 }
4504 sock = -1;
4505 /*
4506 * TODO: Inform client somehow of the fatal error. At
4507 * this point, ret < 0 means that a zmalloc failed
4508 * (ENOMEM). Error detected but still accept
4509 * command, unless a socket error has been
4510 * detected.
4511 */
4512 clean_command_ctx(&cmd_ctx);
4513 continue;
4514 }
4515
4516 health_code_update();
4517
4518 DBG("Sending response (size: %d, retcode: %s (%d))",
4519 cmd_ctx->lttng_msg_size,
4520 lttng_strerror(-cmd_ctx->llm->ret_code),
4521 cmd_ctx->llm->ret_code);
4522 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4523 if (ret < 0) {
4524 ERR("Failed to send data back to client");
4525 }
4526
4527 /* End of transmission */
4528 ret = close(sock);
4529 if (ret) {
4530 PERROR("close");
4531 }
4532 sock = -1;
4533
4534 clean_command_ctx(&cmd_ctx);
4535
4536 health_code_update();
4537 }
4538
4539 exit:
4540 error:
4541 if (sock >= 0) {
4542 ret = close(sock);
4543 if (ret) {
4544 PERROR("close");
4545 }
4546 }
4547
4548 lttng_poll_clean(&events);
4549 clean_command_ctx(&cmd_ctx);
4550
4551 error_listen:
4552 error_create_poll:
4553 unlink(config.client_unix_sock_path.value);
4554 if (client_sock >= 0) {
4555 ret = close(client_sock);
4556 if (ret) {
4557 PERROR("close");
4558 }
4559 }
4560
4561 if (err) {
4562 health_error();
4563 ERR("Health error occurred in %s", __func__);
4564 }
4565
4566 health_unregister(health_sessiond);
4567
4568 DBG("Client thread dying");
4569
4570 rcu_unregister_thread();
4571
4572 /*
4573 * Since we are creating the consumer threads, we own them, so we need
4574 * to join them before our thread exits.
4575 */
4576 ret = join_consumer_thread(&kconsumer_data);
4577 if (ret) {
4578 errno = ret;
4579 PERROR("join_consumer");
4580 }
4581
4582 ret = join_consumer_thread(&ustconsumer32_data);
4583 if (ret) {
4584 errno = ret;
4585 PERROR("join_consumer ust32");
4586 }
4587
4588 ret = join_consumer_thread(&ustconsumer64_data);
4589 if (ret) {
4590 errno = ret;
4591 PERROR("join_consumer ust64");
4592 }
4593 return NULL;
4594 }
4595
4596 static int string_match(const char *str1, const char *str2)
4597 {
4598 return (str1 && str2) && !strcmp(str1, str2);
4599 }
4600
4601 /*
4602 * Take an option from the getopt output and set it in the right variable to be
4603 * used later.
4604 *
4605 * Return 0 on success else a negative value.
4606 */
4607 static int set_option(int opt, const char *arg, const char *optname)
4608 {
4609 int ret = 0;
4610
4611 if (string_match(optname, "client-sock") || opt == 'c') {
4612 if (!arg || *arg == '\0') {
4613 ret = -EINVAL;
4614 goto end;
4615 }
4616 if (lttng_is_setuid_setgid()) {
4617 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4618 "-c, --client-sock");
4619 } else {
4620 config_string_set(&config.client_unix_sock_path,
4621 strdup(arg));
4622 if (!config.client_unix_sock_path.value) {
4623 ret = -ENOMEM;
4624 PERROR("strdup");
4625 }
4626 }
4627 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4628 if (!arg || *arg == '\0') {
4629 ret = -EINVAL;
4630 goto end;
4631 }
4632 if (lttng_is_setuid_setgid()) {
4633 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4634 "-a, --apps-sock");
4635 } else {
4636 config_string_set(&config.apps_unix_sock_path,
4637 strdup(arg));
4638 if (!config.apps_unix_sock_path.value) {
4639 ret = -ENOMEM;
4640 PERROR("strdup");
4641 }
4642 }
4643 } else if (string_match(optname, "daemonize") || opt == 'd') {
4644 config.daemonize = true;
4645 } else if (string_match(optname, "background") || opt == 'b') {
4646 config.background = true;
4647 } else if (string_match(optname, "group") || opt == 'g') {
4648 if (!arg || *arg == '\0') {
4649 ret = -EINVAL;
4650 goto end;
4651 }
4652 if (lttng_is_setuid_setgid()) {
4653 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4654 "-g, --group");
4655 } else {
4656 config_string_set(&config.tracing_group_name,
4657 strdup(arg));
4658 if (!config.tracing_group_name.value) {
4659 ret = -ENOMEM;
4660 PERROR("strdup");
4661 }
4662 }
4663 } else if (string_match(optname, "help") || opt == 'h') {
4664 ret = utils_show_help(8, "lttng-sessiond", help_msg);
4665 if (ret) {
4666 ERR("Cannot show --help for `lttng-sessiond`");
4667 perror("exec");
4668 }
4669 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4670 } else if (string_match(optname, "version") || opt == 'V') {
4671 fprintf(stdout, "%s\n", VERSION);
4672 exit(EXIT_SUCCESS);
4673 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4674 config.sig_parent = true;
4675 } else if (string_match(optname, "kconsumerd-err-sock")) {
4676 if (!arg || *arg == '\0') {
4677 ret = -EINVAL;
4678 goto end;
4679 }
4680 if (lttng_is_setuid_setgid()) {
4681 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4682 "--kconsumerd-err-sock");
4683 } else {
4684 config_string_set(&config.kconsumerd_err_unix_sock_path,
4685 strdup(arg));
4686 if (!config.kconsumerd_err_unix_sock_path.value) {
4687 ret = -ENOMEM;
4688 PERROR("strdup");
4689 }
4690 }
4691 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4692 if (!arg || *arg == '\0') {
4693 ret = -EINVAL;
4694 goto end;
4695 }
4696 if (lttng_is_setuid_setgid()) {
4697 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4698 "--kconsumerd-cmd-sock");
4699 } else {
4700 config_string_set(&config.kconsumerd_cmd_unix_sock_path,
4701 strdup(arg));
4702 if (!config.kconsumerd_cmd_unix_sock_path.value) {
4703 ret = -ENOMEM;
4704 PERROR("strdup");
4705 }
4706 }
4707 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4708 if (!arg || *arg == '\0') {
4709 ret = -EINVAL;
4710 goto end;
4711 }
4712 if (lttng_is_setuid_setgid()) {
4713 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4714 "--ustconsumerd64-err-sock");
4715 } else {
4716 config_string_set(&config.consumerd64_err_unix_sock_path,
4717 strdup(arg));
4718 if (!config.consumerd64_err_unix_sock_path.value) {
4719 ret = -ENOMEM;
4720 PERROR("strdup");
4721 }
4722 }
4723 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4724 if (!arg || *arg == '\0') {
4725 ret = -EINVAL;
4726 goto end;
4727 }
4728 if (lttng_is_setuid_setgid()) {
4729 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4730 "--ustconsumerd64-cmd-sock");
4731 } else {
4732 config_string_set(&config.consumerd64_cmd_unix_sock_path,
4733 strdup(arg));
4734 if (!config.consumerd64_cmd_unix_sock_path.value) {
4735 ret = -ENOMEM;
4736 PERROR("strdup");
4737 }
4738 }
4739 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4740 if (!arg || *arg == '\0') {
4741 ret = -EINVAL;
4742 goto end;
4743 }
4744 if (lttng_is_setuid_setgid()) {
4745 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4746 "--ustconsumerd32-err-sock");
4747 } else {
4748 config_string_set(&config.consumerd32_err_unix_sock_path,
4749 strdup(arg));
4750 if (!config.consumerd32_err_unix_sock_path.value) {
4751 ret = -ENOMEM;
4752 PERROR("strdup");
4753 }
4754 }
4755 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4756 if (!arg || *arg == '\0') {
4757 ret = -EINVAL;
4758 goto end;
4759 }
4760 if (lttng_is_setuid_setgid()) {
4761 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4762 "--ustconsumerd32-cmd-sock");
4763 } else {
4764 config_string_set(&config.consumerd32_cmd_unix_sock_path,
4765 strdup(arg));
4766 if (!config.consumerd32_cmd_unix_sock_path.value) {
4767 ret = -ENOMEM;
4768 PERROR("strdup");
4769 }
4770 }
4771 } else if (string_match(optname, "no-kernel")) {
4772 config.no_kernel = true;
4773 } else if (string_match(optname, "quiet") || opt == 'q') {
4774 config.quiet = true;
4775 } else if (string_match(optname, "verbose") || opt == 'v') {
4776 /* Verbose level can increase using multiple -v */
4777 if (arg) {
4778 /* Value obtained from config file */
4779 config.verbose = config_parse_value(arg);
4780 } else {
4781 /* -v used on command line */
4782 config.verbose++;
4783 }
4784 /* Clamp value to [0, 3] */
4785 config.verbose = config.verbose < 0 ? 0 :
4786 (config.verbose <= 3 ? config.verbose : 3);
4787 } else if (string_match(optname, "verbose-consumer")) {
4788 if (arg) {
4789 config.verbose_consumer = config_parse_value(arg);
4790 } else {
4791 config.verbose_consumer++;
4792 }
4793 } else if (string_match(optname, "consumerd32-path")) {
4794 if (!arg || *arg == '\0') {
4795 ret = -EINVAL;
4796 goto end;
4797 }
4798 if (lttng_is_setuid_setgid()) {
4799 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4800 "--consumerd32-path");
4801 } else {
4802 config_string_set(&config.consumerd32_bin_path,
4803 strdup(arg));
4804 if (!config.consumerd32_bin_path.value) {
4805 PERROR("strdup");
4806 ret = -ENOMEM;
4807 }
4808 }
4809 } else if (string_match(optname, "consumerd32-libdir")) {
4810 if (!arg || *arg == '\0') {
4811 ret = -EINVAL;
4812 goto end;
4813 }
4814 if (lttng_is_setuid_setgid()) {
4815 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4816 "--consumerd32-libdir");
4817 } else {
4818 config_string_set(&config.consumerd32_lib_dir,
4819 strdup(arg));
4820 if (!config.consumerd32_lib_dir.value) {
4821 PERROR("strdup");
4822 ret = -ENOMEM;
4823 }
4824 }
4825 } else if (string_match(optname, "consumerd64-path")) {
4826 if (!arg || *arg == '\0') {
4827 ret = -EINVAL;
4828 goto end;
4829 }
4830 if (lttng_is_setuid_setgid()) {
4831 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4832 "--consumerd64-path");
4833 } else {
4834 config_string_set(&config.consumerd64_bin_path,
4835 strdup(arg));
4836 if (!config.consumerd64_bin_path.value) {
4837 PERROR("strdup");
4838 ret = -ENOMEM;
4839 }
4840 }
4841 } else if (string_match(optname, "consumerd64-libdir")) {
4842 if (!arg || *arg == '\0') {
4843 ret = -EINVAL;
4844 goto end;
4845 }
4846 if (lttng_is_setuid_setgid()) {
4847 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4848 "--consumerd64-libdir");
4849 } else {
4850 config_string_set(&config.consumerd64_lib_dir,
4851 strdup(arg));
4852 if (!config.consumerd64_lib_dir.value) {
4853 PERROR("strdup");
4854 ret = -ENOMEM;
4855 }
4856 }
4857 } else if (string_match(optname, "pidfile") || opt == 'p') {
4858 if (!arg || *arg == '\0') {
4859 ret = -EINVAL;
4860 goto end;
4861 }
4862 if (lttng_is_setuid_setgid()) {
4863 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4864 "-p, --pidfile");
4865 } else {
4866 config_string_set(&config.pid_file_path, strdup(arg));
4867 if (!config.pid_file_path.value) {
4868 PERROR("strdup");
4869 ret = -ENOMEM;
4870 }
4871 }
4872 } else if (string_match(optname, "agent-tcp-port")) {
4873 if (!arg || *arg == '\0') {
4874 ret = -EINVAL;
4875 goto end;
4876 }
4877 if (lttng_is_setuid_setgid()) {
4878 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4879 "--agent-tcp-port");
4880 } else {
4881 unsigned long v;
4882
4883 errno = 0;
4884 v = strtoul(arg, NULL, 0);
4885 if (errno != 0 || !isdigit(arg[0])) {
4886 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4887 return -1;
4888 }
4889 if (v == 0 || v >= 65535) {
4890 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4891 return -1;
4892 }
4893 config.agent_tcp_port.begin = config.agent_tcp_port.end = (int) v;
4894 DBG3("Agent TCP port set to non default: %i", (int) v);
4895 }
4896 } else if (string_match(optname, "load") || opt == 'l') {
4897 if (!arg || *arg == '\0') {
4898 ret = -EINVAL;
4899 goto end;
4900 }
4901 if (lttng_is_setuid_setgid()) {
4902 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4903 "-l, --load");
4904 } else {
4905 config_string_set(&config.load_session_path, strdup(arg));
4906 if (!config.load_session_path.value) {
4907 PERROR("strdup");
4908 ret = -ENOMEM;
4909 }
4910 }
4911 } else if (string_match(optname, "kmod-probes")) {
4912 if (!arg || *arg == '\0') {
4913 ret = -EINVAL;
4914 goto end;
4915 }
4916 if (lttng_is_setuid_setgid()) {
4917 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4918 "--kmod-probes");
4919 } else {
4920 config_string_set(&config.kmod_probes_list, strdup(arg));
4921 if (!config.kmod_probes_list.value) {
4922 PERROR("strdup");
4923 ret = -ENOMEM;
4924 }
4925 }
4926 } else if (string_match(optname, "extra-kmod-probes")) {
4927 if (!arg || *arg == '\0') {
4928 ret = -EINVAL;
4929 goto end;
4930 }
4931 if (lttng_is_setuid_setgid()) {
4932 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4933 "--extra-kmod-probes");
4934 } else {
4935 config_string_set(&config.kmod_extra_probes_list,
4936 strdup(arg));
4937 if (!config.kmod_extra_probes_list.value) {
4938 PERROR("strdup");
4939 ret = -ENOMEM;
4940 }
4941 }
4942 } else if (string_match(optname, "config") || opt == 'f') {
4943 /* This is handled in set_options() thus silent skip. */
4944 goto end;
4945 } else {
4946 /* Unknown option or other error.
4947 * Error is printed by getopt, just return */
4948 ret = -1;
4949 }
4950
4951 end:
4952 if (ret == -EINVAL) {
4953 const char *opt_name = "unknown";
4954 int i;
4955
4956 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4957 i++) {
4958 if (opt == long_options[i].val) {
4959 opt_name = long_options[i].name;
4960 break;
4961 }
4962 }
4963
4964 WARN("Invalid argument provided for option \"%s\", using default value.",
4965 opt_name);
4966 }
4967
4968 return ret;
4969 }
4970
4971 /*
4972 * config_entry_handler_cb used to handle options read from a config file.
4973 * See config_entry_handler_cb comment in common/config/session-config.h for the
4974 * return value conventions.
4975 */
4976 static int config_entry_handler(const struct config_entry *entry, void *unused)
4977 {
4978 int ret = 0, i;
4979
4980 if (!entry || !entry->name || !entry->value) {
4981 ret = -EINVAL;
4982 goto end;
4983 }
4984
4985 /* Check if the option is to be ignored */
4986 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4987 if (!strcmp(entry->name, config_ignore_options[i])) {
4988 goto end;
4989 }
4990 }
4991
4992 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4993 i++) {
4994
4995 /* Ignore if not fully matched. */
4996 if (strcmp(entry->name, long_options[i].name)) {
4997 continue;
4998 }
4999
5000 /*
5001 * If the option takes no argument on the command line, we have to
5002 * check if the value is "true". We support non-zero numeric values,
5003 * true, on and yes.
5004 */
5005 if (!long_options[i].has_arg) {
5006 ret = config_parse_value(entry->value);
5007 if (ret <= 0) {
5008 if (ret) {
5009 WARN("Invalid configuration value \"%s\" for option %s",
5010 entry->value, entry->name);
5011 }
5012 /* False, skip boolean config option. */
5013 goto end;
5014 }
5015 }
5016
5017 ret = set_option(long_options[i].val, entry->value, entry->name);
5018 goto end;
5019 }
5020
5021 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5022
5023 end:
5024 return ret;
5025 }
5026
5027 /*
5028 * daemon configuration loading and argument parsing
5029 */
5030 static int set_options(int argc, char **argv)
5031 {
5032 int ret = 0, c = 0, option_index = 0;
5033 int orig_optopt = optopt, orig_optind = optind;
5034 char *optstring;
5035 const char *config_path = NULL;
5036
5037 optstring = utils_generate_optstring(long_options,
5038 sizeof(long_options) / sizeof(struct option));
5039 if (!optstring) {
5040 ret = -ENOMEM;
5041 goto end;
5042 }
5043
5044 /* Check for the --config option */
5045 while ((c = getopt_long(argc, argv, optstring, long_options,
5046 &option_index)) != -1) {
5047 if (c == '?') {
5048 ret = -EINVAL;
5049 goto end;
5050 } else if (c != 'f') {
5051 /* if not equal to --config option. */
5052 continue;
5053 }
5054
5055 if (lttng_is_setuid_setgid()) {
5056 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5057 "-f, --config");
5058 } else {
5059 config_path = utils_expand_path(optarg);
5060 if (!config_path) {
5061 ERR("Failed to resolve path: %s", optarg);
5062 }
5063 }
5064 }
5065
5066 ret = config_get_section_entries(config_path, config_section_name,
5067 config_entry_handler, NULL);
5068 if (ret) {
5069 if (ret > 0) {
5070 ERR("Invalid configuration option at line %i", ret);
5071 ret = -1;
5072 }
5073 goto end;
5074 }
5075
5076 /* Reset getopt's global state */
5077 optopt = orig_optopt;
5078 optind = orig_optind;
5079 while (1) {
5080 option_index = -1;
5081 /*
5082 * getopt_long() will not set option_index if it encounters a
5083 * short option.
5084 */
5085 c = getopt_long(argc, argv, optstring, long_options,
5086 &option_index);
5087 if (c == -1) {
5088 break;
5089 }
5090
5091 /*
5092 * Pass NULL as the long option name if popt left the index
5093 * unset.
5094 */
5095 ret = set_option(c, optarg,
5096 option_index < 0 ? NULL :
5097 long_options[option_index].name);
5098 if (ret < 0) {
5099 break;
5100 }
5101 }
5102
5103 end:
5104 free(optstring);
5105 return ret;
5106 }
5107
5108 /*
5109 * Creates the two needed socket by the daemon.
5110 * apps_sock - The communication socket for all UST apps.
5111 * client_sock - The communication of the cli tool (lttng).
5112 */
5113 static int init_daemon_socket(void)
5114 {
5115 int ret = 0;
5116 mode_t old_umask;
5117
5118 old_umask = umask(0);
5119
5120 /* Create client tool unix socket */
5121 client_sock = lttcomm_create_unix_sock(config.client_unix_sock_path.value);
5122 if (client_sock < 0) {
5123 ERR("Create unix sock failed: %s", config.client_unix_sock_path.value);
5124 ret = -1;
5125 goto end;
5126 }
5127
5128 /* Set the cloexec flag */
5129 ret = utils_set_fd_cloexec(client_sock);
5130 if (ret < 0) {
5131 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5132 "Continuing but note that the consumer daemon will have a "
5133 "reference to this socket on exec()", client_sock);
5134 }
5135
5136 /* File permission MUST be 660 */
5137 ret = chmod(config.client_unix_sock_path.value, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5138 if (ret < 0) {
5139 ERR("Set file permissions failed: %s", config.client_unix_sock_path.value);
5140 PERROR("chmod");
5141 goto end;
5142 }
5143
5144 /* Create the application unix socket */
5145 apps_sock = lttcomm_create_unix_sock(config.apps_unix_sock_path.value);
5146 if (apps_sock < 0) {
5147 ERR("Create unix sock failed: %s", config.apps_unix_sock_path.value);
5148 ret = -1;
5149 goto end;
5150 }
5151
5152 /* Set the cloexec flag */
5153 ret = utils_set_fd_cloexec(apps_sock);
5154 if (ret < 0) {
5155 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5156 "Continuing but note that the consumer daemon will have a "
5157 "reference to this socket on exec()", apps_sock);
5158 }
5159
5160 /* File permission MUST be 666 */
5161 ret = chmod(config.apps_unix_sock_path.value,
5162 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5163 if (ret < 0) {
5164 ERR("Set file permissions failed: %s", config.apps_unix_sock_path.value);
5165 PERROR("chmod");
5166 goto end;
5167 }
5168
5169 DBG3("Session daemon client socket %d and application socket %d created",
5170 client_sock, apps_sock);
5171
5172 end:
5173 umask(old_umask);
5174 return ret;
5175 }
5176
5177 /*
5178 * Create lockfile using the rundir and return its fd.
5179 */
5180 static int create_lockfile(void)
5181 {
5182 return utils_create_lock_file(config.lock_file_path.value);
5183 }
5184
5185 /*
5186 * Check if the global socket is available, and if a daemon is answering at the
5187 * other side. If yes, error is returned.
5188 *
5189 * Also attempts to create and hold the lock file.
5190 */
5191 static int check_existing_daemon(void)
5192 {
5193 int ret = 0;
5194
5195 /* Is there anybody out there ? */
5196 if (lttng_session_daemon_alive()) {
5197 ret = -EEXIST;
5198 goto end;
5199 }
5200
5201 lockfile_fd = create_lockfile();
5202 if (lockfile_fd < 0) {
5203 ret = -EEXIST;
5204 goto end;
5205 }
5206 end:
5207 return ret;
5208 }
5209
5210 static void sessiond_cleanup_lock_file(void)
5211 {
5212 int ret;
5213
5214 /*
5215 * Cleanup lock file by deleting it and finaly closing it which will
5216 * release the file system lock.
5217 */
5218 if (lockfile_fd >= 0) {
5219 ret = remove(config.lock_file_path.value);
5220 if (ret < 0) {
5221 PERROR("remove lock file");
5222 }
5223 ret = close(lockfile_fd);
5224 if (ret < 0) {
5225 PERROR("close lock file");
5226 }
5227 }
5228 }
5229
5230 /*
5231 * Set the tracing group gid onto the client socket.
5232 *
5233 * Race window between mkdir and chown is OK because we are going from more
5234 * permissive (root.root) to less permissive (root.tracing).
5235 */
5236 static int set_permissions(char *rundir)
5237 {
5238 int ret;
5239 gid_t gid;
5240
5241 gid = utils_get_group_id(config.tracing_group_name.value);
5242
5243 /* Set lttng run dir */
5244 ret = chown(rundir, 0, gid);
5245 if (ret < 0) {
5246 ERR("Unable to set group on %s", rundir);
5247 PERROR("chown");
5248 }
5249
5250 /*
5251 * Ensure all applications and tracing group can search the run
5252 * dir. Allow everyone to read the directory, since it does not
5253 * buy us anything to hide its content.
5254 */
5255 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5256 if (ret < 0) {
5257 ERR("Unable to set permissions on %s", rundir);
5258 PERROR("chmod");
5259 }
5260
5261 /* lttng client socket path */
5262 ret = chown(config.client_unix_sock_path.value, 0, gid);
5263 if (ret < 0) {
5264 ERR("Unable to set group on %s", config.client_unix_sock_path.value);
5265 PERROR("chown");
5266 }
5267
5268 /* kconsumer error socket path */
5269 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5270 if (ret < 0) {
5271 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5272 PERROR("chown");
5273 }
5274
5275 /* 64-bit ustconsumer error socket path */
5276 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5277 if (ret < 0) {
5278 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5279 PERROR("chown");
5280 }
5281
5282 /* 32-bit ustconsumer compat32 error socket path */
5283 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5284 if (ret < 0) {
5285 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5286 PERROR("chown");
5287 }
5288
5289 DBG("All permissions are set");
5290
5291 return ret;
5292 }
5293
5294 /*
5295 * Create the lttng run directory needed for all global sockets and pipe.
5296 */
5297 static int create_lttng_rundir(void)
5298 {
5299 int ret;
5300
5301 DBG3("Creating LTTng run directory: %s", config.rundir.value);
5302
5303 ret = mkdir(config.rundir.value, S_IRWXU);
5304 if (ret < 0) {
5305 if (errno != EEXIST) {
5306 ERR("Unable to create %s", config.rundir.value);
5307 goto error;
5308 } else {
5309 ret = 0;
5310 }
5311 }
5312
5313 error:
5314 return ret;
5315 }
5316
5317 /*
5318 * Setup sockets and directory needed by the consumerds' communication with the
5319 * session daemon.
5320 */
5321 static int set_consumer_sockets(struct consumer_data *consumer_data)
5322 {
5323 int ret;
5324 char *path = NULL;
5325
5326 switch (consumer_data->type) {
5327 case LTTNG_CONSUMER_KERNEL:
5328 path = config.kconsumerd_path.value;
5329 break;
5330 case LTTNG_CONSUMER64_UST:
5331 path = config.consumerd64_path.value;
5332 break;
5333 case LTTNG_CONSUMER32_UST:
5334 path = config.consumerd32_path.value;
5335 break;
5336 default:
5337 ERR("Consumer type unknown");
5338 ret = -EINVAL;
5339 goto error;
5340 }
5341 assert(path);
5342
5343 DBG2("Creating consumer directory: %s", path);
5344
5345 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5346 if (ret < 0 && errno != EEXIST) {
5347 PERROR("mkdir");
5348 ERR("Failed to create %s", path);
5349 goto error;
5350 }
5351 if (is_root) {
5352 ret = chown(path, 0, utils_get_group_id(config.tracing_group_name.value));
5353 if (ret < 0) {
5354 ERR("Unable to set group on %s", path);
5355 PERROR("chown");
5356 goto error;
5357 }
5358 }
5359
5360 /* Create the consumerd error unix socket */
5361 consumer_data->err_sock =
5362 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5363 if (consumer_data->err_sock < 0) {
5364 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5365 ret = -1;
5366 goto error;
5367 }
5368
5369 /*
5370 * Set the CLOEXEC flag. Return code is useless because either way, the
5371 * show must go on.
5372 */
5373 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5374 if (ret < 0) {
5375 PERROR("utils_set_fd_cloexec");
5376 /* continue anyway */
5377 }
5378
5379 /* File permission MUST be 660 */
5380 ret = chmod(consumer_data->err_unix_sock_path,
5381 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5382 if (ret < 0) {
5383 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5384 PERROR("chmod");
5385 goto error;
5386 }
5387
5388 error:
5389 return ret;
5390 }
5391
5392 /*
5393 * Signal handler for the daemon
5394 *
5395 * Simply stop all worker threads, leaving main() return gracefully after
5396 * joining all threads and calling cleanup().
5397 */
5398 static void sighandler(int sig)
5399 {
5400 switch (sig) {
5401 case SIGINT:
5402 DBG("SIGINT caught");
5403 stop_threads();
5404 break;
5405 case SIGTERM:
5406 DBG("SIGTERM caught");
5407 stop_threads();
5408 break;
5409 case SIGUSR1:
5410 CMM_STORE_SHARED(recv_child_signal, 1);
5411 break;
5412 default:
5413 break;
5414 }
5415 }
5416
5417 /*
5418 * Setup signal handler for :
5419 * SIGINT, SIGTERM, SIGPIPE
5420 */
5421 static int set_signal_handler(void)
5422 {
5423 int ret = 0;
5424 struct sigaction sa;
5425 sigset_t sigset;
5426
5427 if ((ret = sigemptyset(&sigset)) < 0) {
5428 PERROR("sigemptyset");
5429 return ret;
5430 }
5431
5432 sa.sa_mask = sigset;
5433 sa.sa_flags = 0;
5434
5435 sa.sa_handler = sighandler;
5436 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5437 PERROR("sigaction");
5438 return ret;
5439 }
5440
5441 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5442 PERROR("sigaction");
5443 return ret;
5444 }
5445
5446 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5447 PERROR("sigaction");
5448 return ret;
5449 }
5450
5451 sa.sa_handler = SIG_IGN;
5452 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5453 PERROR("sigaction");
5454 return ret;
5455 }
5456
5457 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5458
5459 return ret;
5460 }
5461
5462 /*
5463 * Set open files limit to unlimited. This daemon can open a large number of
5464 * file descriptors in order to consume multiple kernel traces.
5465 */
5466 static void set_ulimit(void)
5467 {
5468 int ret;
5469 struct rlimit lim;
5470
5471 /* The kernel does not allow an infinite limit for open files */
5472 lim.rlim_cur = 65535;
5473 lim.rlim_max = 65535;
5474
5475 ret = setrlimit(RLIMIT_NOFILE, &lim);
5476 if (ret < 0) {
5477 PERROR("failed to set open files limit");
5478 }
5479 }
5480
5481 static int write_pidfile(void)
5482 {
5483 return utils_create_pid_file(getpid(), config.pid_file_path.value);
5484 }
5485
5486 static int set_clock_plugin_env(void)
5487 {
5488 int ret = 0;
5489 char *env_value = NULL;
5490
5491 if (!config.lttng_ust_clock_plugin.value) {
5492 goto end;
5493 }
5494
5495 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
5496 config.lttng_ust_clock_plugin.value);
5497 if (ret < 0) {
5498 PERROR("asprintf");
5499 goto end;
5500 }
5501
5502 ret = putenv(env_value);
5503 if (ret) {
5504 free(env_value);
5505 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5506 goto end;
5507 }
5508
5509 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5510 config.lttng_ust_clock_plugin.value);
5511 end:
5512 return ret;
5513 }
5514
5515 /*
5516 * main
5517 */
5518 int main(int argc, char **argv)
5519 {
5520 int ret = 0, retval = 0;
5521 void *status;
5522 const char *env_app_timeout;
5523 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5524 *ust64_channel_monitor_pipe = NULL,
5525 *kernel_channel_monitor_pipe = NULL;
5526 bool notification_thread_running = false;
5527
5528 init_kernel_workarounds();
5529
5530 rcu_register_thread();
5531
5532 if (set_signal_handler()) {
5533 retval = -1;
5534 goto exit_set_signal_handler;
5535 }
5536
5537 page_size = sysconf(_SC_PAGESIZE);
5538 if (page_size < 0) {
5539 PERROR("sysconf _SC_PAGESIZE");
5540 page_size = LONG_MAX;
5541 WARN("Fallback page size to %ld", page_size);
5542 }
5543
5544 ret = sessiond_config_init(&config);
5545 if (ret) {
5546 retval = -1;
5547 goto exit_set_signal_handler;
5548 }
5549
5550 /*
5551 * Init config from environment variables.
5552 * Command line option override env configuration per-doc. Do env first.
5553 */
5554 sessiond_config_apply_env_config(&config);
5555
5556 /*
5557 * Parse arguments and load the daemon configuration file.
5558 *
5559 * We have an exit_options exit path to free memory reserved by
5560 * set_options. This is needed because the rest of sessiond_cleanup()
5561 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5562 * depends on set_options.
5563 */
5564 progname = argv[0];
5565 if (set_options(argc, argv)) {
5566 retval = -1;
5567 goto exit_options;
5568 }
5569
5570 /*
5571 * Resolve all paths received as arguments, configuration option, or
5572 * through environment variable as absolute paths. This is necessary
5573 * since daemonizing causes the sessiond's current working directory
5574 * to '/'.
5575 */
5576 ret = sessiond_config_resolve_paths(&config);
5577 if (ret) {
5578 goto exit_options;
5579 }
5580
5581 /* Apply config. */
5582 lttng_opt_verbose = config.verbose;
5583 lttng_opt_quiet = config.quiet;
5584 kconsumer_data.err_unix_sock_path =
5585 config.kconsumerd_err_unix_sock_path.value;
5586 kconsumer_data.cmd_unix_sock_path =
5587 config.kconsumerd_cmd_unix_sock_path.value;
5588 ustconsumer32_data.err_unix_sock_path =
5589 config.consumerd32_err_unix_sock_path.value;
5590 ustconsumer32_data.cmd_unix_sock_path =
5591 config.consumerd32_cmd_unix_sock_path.value;
5592 ustconsumer64_data.err_unix_sock_path =
5593 config.consumerd64_err_unix_sock_path.value;
5594 ustconsumer64_data.cmd_unix_sock_path =
5595 config.consumerd64_cmd_unix_sock_path.value;
5596 set_clock_plugin_env();
5597
5598 sessiond_config_log(&config);
5599
5600 if (create_lttng_rundir()) {
5601 retval = -1;
5602 goto exit_options;
5603 }
5604
5605 /* Abort launch if a session daemon is already running. */
5606 if (check_existing_daemon()) {
5607 ERR("A session daemon is already running.");
5608 retval = -1;
5609 goto exit_options;
5610 }
5611
5612 /* Daemonize */
5613 if (config.daemonize || config.background) {
5614 int i;
5615
5616 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5617 !config.background);
5618 if (ret < 0) {
5619 retval = -1;
5620 goto exit_options;
5621 }
5622
5623 /*
5624 * We are in the child. Make sure all other file descriptors are
5625 * closed, in case we are called with more opened file
5626 * descriptors than the standard ones and the lock file.
5627 */
5628 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5629 if (i == lockfile_fd) {
5630 continue;
5631 }
5632 (void) close(i);
5633 }
5634 }
5635
5636 if (run_as_create_worker(argv[0]) < 0) {
5637 goto exit_create_run_as_worker_cleanup;
5638 }
5639
5640 /*
5641 * Starting from here, we can create threads. This needs to be after
5642 * lttng_daemonize due to RCU.
5643 */
5644
5645 /*
5646 * Initialize the health check subsystem. This call should set the
5647 * appropriate time values.
5648 */
5649 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5650 if (!health_sessiond) {
5651 PERROR("health_app_create error");
5652 retval = -1;
5653 goto exit_health_sessiond_cleanup;
5654 }
5655
5656 /* Create thread to clean up RCU hash tables */
5657 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
5658 retval = -1;
5659 goto exit_ht_cleanup;
5660 }
5661
5662 /* Create thread quit pipe */
5663 if (init_thread_quit_pipe()) {
5664 retval = -1;
5665 goto exit_init_data;
5666 }
5667
5668 /* Check if daemon is UID = 0 */
5669 is_root = !getuid();
5670 if (is_root) {
5671 /* Create global run dir with root access */
5672
5673 kernel_channel_monitor_pipe = lttng_pipe_open(0);
5674 if (!kernel_channel_monitor_pipe) {
5675 ERR("Failed to create kernel consumer channel monitor pipe");
5676 retval = -1;
5677 goto exit_init_data;
5678 }
5679 kconsumer_data.channel_monitor_pipe =
5680 lttng_pipe_release_writefd(
5681 kernel_channel_monitor_pipe);
5682 if (kconsumer_data.channel_monitor_pipe < 0) {
5683 retval = -1;
5684 goto exit_init_data;
5685 }
5686 }
5687
5688 /* Set consumer initial state */
5689 kernel_consumerd_state = CONSUMER_STOPPED;
5690 ust_consumerd_state = CONSUMER_STOPPED;
5691
5692 ust32_channel_monitor_pipe = lttng_pipe_open(0);
5693 if (!ust32_channel_monitor_pipe) {
5694 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5695 retval = -1;
5696 goto exit_init_data;
5697 }
5698 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5699 ust32_channel_monitor_pipe);
5700 if (ustconsumer32_data.channel_monitor_pipe < 0) {
5701 retval = -1;
5702 goto exit_init_data;
5703 }
5704
5705 ust64_channel_monitor_pipe = lttng_pipe_open(0);
5706 if (!ust64_channel_monitor_pipe) {
5707 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5708 retval = -1;
5709 goto exit_init_data;
5710 }
5711 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5712 ust64_channel_monitor_pipe);
5713 if (ustconsumer64_data.channel_monitor_pipe < 0) {
5714 retval = -1;
5715 goto exit_init_data;
5716 }
5717
5718 /*
5719 * Init UST app hash table. Alloc hash table before this point since
5720 * cleanup() can get called after that point.
5721 */
5722 if (ust_app_ht_alloc()) {
5723 ERR("Failed to allocate UST app hash table");
5724 retval = -1;
5725 goto exit_init_data;
5726 }
5727
5728 /*
5729 * Initialize agent app hash table. We allocate the hash table here
5730 * since cleanup() can get called after this point.
5731 */
5732 if (agent_app_ht_alloc()) {
5733 ERR("Failed to allocate Agent app hash table");
5734 retval = -1;
5735 goto exit_init_data;
5736 }
5737
5738 /*
5739 * These actions must be executed as root. We do that *after* setting up
5740 * the sockets path because we MUST make the check for another daemon using
5741 * those paths *before* trying to set the kernel consumer sockets and init
5742 * kernel tracer.
5743 */
5744 if (is_root) {
5745 if (set_consumer_sockets(&kconsumer_data)) {
5746 retval = -1;
5747 goto exit_init_data;
5748 }
5749
5750 /* Setup kernel tracer */
5751 if (!config.no_kernel) {
5752 init_kernel_tracer();
5753 if (kernel_tracer_fd >= 0) {
5754 ret = syscall_init_table();
5755 if (ret < 0) {
5756 ERR("Unable to populate syscall table. "
5757 "Syscall tracing won't work "
5758 "for this session daemon.");
5759 }
5760 }
5761 }
5762
5763 /* Set ulimit for open files */
5764 set_ulimit();
5765 }
5766 /* init lttng_fd tracking must be done after set_ulimit. */
5767 lttng_fd_init();
5768
5769 if (set_consumer_sockets(&ustconsumer64_data)) {
5770 retval = -1;
5771 goto exit_init_data;
5772 }
5773
5774 if (set_consumer_sockets(&ustconsumer32_data)) {
5775 retval = -1;
5776 goto exit_init_data;
5777 }
5778
5779 /* Setup the needed unix socket */
5780 if (init_daemon_socket()) {
5781 retval = -1;
5782 goto exit_init_data;
5783 }
5784
5785 /* Set credentials to socket */
5786 if (is_root && set_permissions(config.rundir.value)) {
5787 retval = -1;
5788 goto exit_init_data;
5789 }
5790
5791 /* Get parent pid if -S, --sig-parent is specified. */
5792 if (config.sig_parent) {
5793 ppid = getppid();
5794 }
5795
5796 /* Setup the kernel pipe for waking up the kernel thread */
5797 if (is_root && !config.no_kernel) {
5798 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5799 retval = -1;
5800 goto exit_init_data;
5801 }
5802 }
5803
5804 /* Setup the thread apps communication pipe. */
5805 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5806 retval = -1;
5807 goto exit_init_data;
5808 }
5809
5810 /* Setup the thread apps notify communication pipe. */
5811 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5812 retval = -1;
5813 goto exit_init_data;
5814 }
5815
5816 /* Initialize global buffer per UID and PID registry. */
5817 buffer_reg_init_uid_registry();
5818 buffer_reg_init_pid_registry();
5819
5820 /* Init UST command queue. */
5821 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5822
5823 /*
5824 * Get session list pointer. This pointer MUST NOT be free'd. This list
5825 * is statically declared in session.c
5826 */
5827 session_list_ptr = session_get_list();
5828
5829 cmd_init();
5830
5831 /* Check for the application socket timeout env variable. */
5832 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5833 if (env_app_timeout) {
5834 config.app_socket_timeout = atoi(env_app_timeout);
5835 } else {
5836 config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5837 }
5838
5839 ret = write_pidfile();
5840 if (ret) {
5841 ERR("Error in write_pidfile");
5842 retval = -1;
5843 goto exit_init_data;
5844 }
5845
5846 /* Initialize communication library */
5847 lttcomm_init();
5848 /* Initialize TCP timeout values */
5849 lttcomm_inet_init();
5850
5851 if (load_session_init_data(&load_info) < 0) {
5852 retval = -1;
5853 goto exit_init_data;
5854 }
5855 load_info->path = config.load_session_path.value;
5856
5857 /* Create health-check thread. */
5858 ret = pthread_create(&health_thread, default_pthread_attr(),
5859 thread_manage_health, (void *) NULL);
5860 if (ret) {
5861 errno = ret;
5862 PERROR("pthread_create health");
5863 retval = -1;
5864 goto exit_health;
5865 }
5866
5867 /* notification_thread_data acquires the pipes' read side. */
5868 notification_thread_handle = notification_thread_handle_create(
5869 ust32_channel_monitor_pipe,
5870 ust64_channel_monitor_pipe,
5871 kernel_channel_monitor_pipe);
5872 if (!notification_thread_handle) {
5873 retval = -1;
5874 ERR("Failed to create notification thread shared data");
5875 stop_threads();
5876 goto exit_notification;
5877 }
5878
5879 /* Create notification thread. */
5880 ret = pthread_create(&notification_thread, default_pthread_attr(),
5881 thread_notification, notification_thread_handle);
5882 if (ret) {
5883 errno = ret;
5884 PERROR("pthread_create notification");
5885 retval = -1;
5886 stop_threads();
5887 goto exit_notification;
5888 }
5889 notification_thread_running = true;
5890
5891 /* Create thread to manage the client socket */
5892 ret = pthread_create(&client_thread, default_pthread_attr(),
5893 thread_manage_clients, (void *) NULL);
5894 if (ret) {
5895 errno = ret;
5896 PERROR("pthread_create clients");
5897 retval = -1;
5898 stop_threads();
5899 goto exit_client;
5900 }
5901
5902 /* Create thread to dispatch registration */
5903 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
5904 thread_dispatch_ust_registration, (void *) NULL);
5905 if (ret) {
5906 errno = ret;
5907 PERROR("pthread_create dispatch");
5908 retval = -1;
5909 stop_threads();
5910 goto exit_dispatch;
5911 }
5912
5913 /* Create thread to manage application registration. */
5914 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
5915 thread_registration_apps, (void *) NULL);
5916 if (ret) {
5917 errno = ret;
5918 PERROR("pthread_create registration");
5919 retval = -1;
5920 stop_threads();
5921 goto exit_reg_apps;
5922 }
5923
5924 /* Create thread to manage application socket */
5925 ret = pthread_create(&apps_thread, default_pthread_attr(),
5926 thread_manage_apps, (void *) NULL);
5927 if (ret) {
5928 errno = ret;
5929 PERROR("pthread_create apps");
5930 retval = -1;
5931 stop_threads();
5932 goto exit_apps;
5933 }
5934
5935 /* Create thread to manage application notify socket */
5936 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
5937 ust_thread_manage_notify, (void *) NULL);
5938 if (ret) {
5939 errno = ret;
5940 PERROR("pthread_create notify");
5941 retval = -1;
5942 stop_threads();
5943 goto exit_apps_notify;
5944 }
5945
5946 /* Create agent registration thread. */
5947 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
5948 agent_thread_manage_registration, (void *) NULL);
5949 if (ret) {
5950 errno = ret;
5951 PERROR("pthread_create agent");
5952 retval = -1;
5953 stop_threads();
5954 goto exit_agent_reg;
5955 }
5956
5957 /* Don't start this thread if kernel tracing is not requested nor root */
5958 if (is_root && !config.no_kernel) {
5959 /* Create kernel thread to manage kernel event */
5960 ret = pthread_create(&kernel_thread, default_pthread_attr(),
5961 thread_manage_kernel, (void *) NULL);
5962 if (ret) {
5963 errno = ret;
5964 PERROR("pthread_create kernel");
5965 retval = -1;
5966 stop_threads();
5967 goto exit_kernel;
5968 }
5969 }
5970
5971 /* Create session loading thread. */
5972 ret = pthread_create(&load_session_thread, default_pthread_attr(),
5973 thread_load_session, load_info);
5974 if (ret) {
5975 errno = ret;
5976 PERROR("pthread_create load_session_thread");
5977 retval = -1;
5978 stop_threads();
5979 goto exit_load_session;
5980 }
5981
5982 /*
5983 * This is where we start awaiting program completion (e.g. through
5984 * signal that asks threads to teardown).
5985 */
5986
5987 ret = pthread_join(load_session_thread, &status);
5988 if (ret) {
5989 errno = ret;
5990 PERROR("pthread_join load_session_thread");
5991 retval = -1;
5992 }
5993 exit_load_session:
5994
5995 if (is_root && !config.no_kernel) {
5996 ret = pthread_join(kernel_thread, &status);
5997 if (ret) {
5998 errno = ret;
5999 PERROR("pthread_join");
6000 retval = -1;
6001 }
6002 }
6003 exit_kernel:
6004
6005 ret = pthread_join(agent_reg_thread, &status);
6006 if (ret) {
6007 errno = ret;
6008 PERROR("pthread_join agent");
6009 retval = -1;
6010 }
6011 exit_agent_reg:
6012
6013 ret = pthread_join(apps_notify_thread, &status);
6014 if (ret) {
6015 errno = ret;
6016 PERROR("pthread_join apps notify");
6017 retval = -1;
6018 }
6019 exit_apps_notify:
6020
6021 ret = pthread_join(apps_thread, &status);
6022 if (ret) {
6023 errno = ret;
6024 PERROR("pthread_join apps");
6025 retval = -1;
6026 }
6027 exit_apps:
6028
6029 ret = pthread_join(reg_apps_thread, &status);
6030 if (ret) {
6031 errno = ret;
6032 PERROR("pthread_join");
6033 retval = -1;
6034 }
6035 exit_reg_apps:
6036
6037 /*
6038 * Join dispatch thread after joining reg_apps_thread to ensure
6039 * we don't leak applications in the queue.
6040 */
6041 ret = pthread_join(dispatch_thread, &status);
6042 if (ret) {
6043 errno = ret;
6044 PERROR("pthread_join");
6045 retval = -1;
6046 }
6047 exit_dispatch:
6048
6049 ret = pthread_join(client_thread, &status);
6050 if (ret) {
6051 errno = ret;
6052 PERROR("pthread_join");
6053 retval = -1;
6054 }
6055
6056 exit_client:
6057 exit_notification:
6058 ret = pthread_join(health_thread, &status);
6059 if (ret) {
6060 errno = ret;
6061 PERROR("pthread_join health thread");
6062 retval = -1;
6063 }
6064
6065 exit_health:
6066 exit_init_data:
6067 /*
6068 * Wait for all pending call_rcu work to complete before tearing
6069 * down data structures. call_rcu worker may be trying to
6070 * perform lookups in those structures.
6071 */
6072 rcu_barrier();
6073 /*
6074 * sessiond_cleanup() is called when no other thread is running, except
6075 * the ht_cleanup thread, which is needed to destroy the hash tables.
6076 */
6077 rcu_thread_online();
6078 sessiond_cleanup();
6079
6080 /*
6081 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6082 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6083 * the queue is empty before shutting down the clean-up thread.
6084 */
6085 rcu_barrier();
6086
6087 /*
6088 * The teardown of the notification system is performed after the
6089 * session daemon's teardown in order to allow it to be notified
6090 * of the active session and channels at the moment of the teardown.
6091 */
6092 if (notification_thread_handle) {
6093 if (notification_thread_running) {
6094 notification_thread_command_quit(
6095 notification_thread_handle);
6096 ret = pthread_join(notification_thread, &status);
6097 if (ret) {
6098 errno = ret;
6099 PERROR("pthread_join notification thread");
6100 retval = -1;
6101 }
6102 }
6103 notification_thread_handle_destroy(notification_thread_handle);
6104 }
6105
6106 rcu_thread_offline();
6107 rcu_unregister_thread();
6108
6109 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6110 if (ret) {
6111 retval = -1;
6112 }
6113 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6114 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6115 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6116 exit_ht_cleanup:
6117
6118 health_app_destroy(health_sessiond);
6119 exit_health_sessiond_cleanup:
6120 exit_create_run_as_worker_cleanup:
6121
6122 exit_options:
6123 sessiond_cleanup_lock_file();
6124 sessiond_cleanup_options();
6125
6126 exit_set_signal_handler:
6127 if (!retval) {
6128 exit(EXIT_SUCCESS);
6129 } else {
6130 exit(EXIT_FAILURE);
6131 }
6132 }
This page took 0.204909 seconds and 4 git commands to generate.