Fix: initialize syscall table when kernel tracer is lazily initialized
[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
2656 ret = syscall_init_table();
2657 if (ret < 0) {
2658 ERR("Unable to populate syscall table. Syscall tracing won't "
2659 "work for this session daemon.");
2660 }
2661 return 0;
2662
2663 error_version:
2664 modprobe_remove_lttng_control();
2665 ret = close(kernel_tracer_fd);
2666 if (ret) {
2667 PERROR("close");
2668 }
2669 kernel_tracer_fd = -1;
2670 return LTTNG_ERR_KERN_VERSION;
2671
2672 error_modules:
2673 ret = close(kernel_tracer_fd);
2674 if (ret) {
2675 PERROR("close");
2676 }
2677
2678 error_open:
2679 modprobe_remove_lttng_control();
2680
2681 error:
2682 WARN("No kernel tracer available");
2683 kernel_tracer_fd = -1;
2684 if (!is_root) {
2685 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2686 } else {
2687 return LTTNG_ERR_KERN_NA;
2688 }
2689 }
2690
2691
2692 /*
2693 * Copy consumer output from the tracing session to the domain session. The
2694 * function also applies the right modification on a per domain basis for the
2695 * trace files destination directory.
2696 *
2697 * Should *NOT* be called with RCU read-side lock held.
2698 */
2699 static int copy_session_consumer(int domain, struct ltt_session *session)
2700 {
2701 int ret;
2702 const char *dir_name;
2703 struct consumer_output *consumer;
2704
2705 assert(session);
2706 assert(session->consumer);
2707
2708 switch (domain) {
2709 case LTTNG_DOMAIN_KERNEL:
2710 DBG3("Copying tracing session consumer output in kernel session");
2711 /*
2712 * XXX: We should audit the session creation and what this function
2713 * does "extra" in order to avoid a destroy since this function is used
2714 * in the domain session creation (kernel and ust) only. Same for UST
2715 * domain.
2716 */
2717 if (session->kernel_session->consumer) {
2718 consumer_output_put(session->kernel_session->consumer);
2719 }
2720 session->kernel_session->consumer =
2721 consumer_copy_output(session->consumer);
2722 /* Ease our life a bit for the next part */
2723 consumer = session->kernel_session->consumer;
2724 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2725 break;
2726 case LTTNG_DOMAIN_JUL:
2727 case LTTNG_DOMAIN_LOG4J:
2728 case LTTNG_DOMAIN_PYTHON:
2729 case LTTNG_DOMAIN_UST:
2730 DBG3("Copying tracing session consumer output in UST session");
2731 if (session->ust_session->consumer) {
2732 consumer_output_put(session->ust_session->consumer);
2733 }
2734 session->ust_session->consumer =
2735 consumer_copy_output(session->consumer);
2736 /* Ease our life a bit for the next part */
2737 consumer = session->ust_session->consumer;
2738 dir_name = DEFAULT_UST_TRACE_DIR;
2739 break;
2740 default:
2741 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2742 goto error;
2743 }
2744
2745 /* Append correct directory to subdir */
2746 strncat(consumer->subdir, dir_name,
2747 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2748 DBG3("Copy session consumer subdir %s", consumer->subdir);
2749
2750 ret = LTTNG_OK;
2751
2752 error:
2753 return ret;
2754 }
2755
2756 /*
2757 * Create an UST session and add it to the session ust list.
2758 *
2759 * Should *NOT* be called with RCU read-side lock held.
2760 */
2761 static int create_ust_session(struct ltt_session *session,
2762 struct lttng_domain *domain)
2763 {
2764 int ret;
2765 struct ltt_ust_session *lus = NULL;
2766
2767 assert(session);
2768 assert(domain);
2769 assert(session->consumer);
2770
2771 switch (domain->type) {
2772 case LTTNG_DOMAIN_JUL:
2773 case LTTNG_DOMAIN_LOG4J:
2774 case LTTNG_DOMAIN_PYTHON:
2775 case LTTNG_DOMAIN_UST:
2776 break;
2777 default:
2778 ERR("Unknown UST domain on create session %d", domain->type);
2779 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2780 goto error;
2781 }
2782
2783 DBG("Creating UST session");
2784
2785 lus = trace_ust_create_session(session->id);
2786 if (lus == NULL) {
2787 ret = LTTNG_ERR_UST_SESS_FAIL;
2788 goto error;
2789 }
2790
2791 lus->uid = session->uid;
2792 lus->gid = session->gid;
2793 lus->output_traces = session->output_traces;
2794 lus->snapshot_mode = session->snapshot_mode;
2795 lus->live_timer_interval = session->live_timer;
2796 session->ust_session = lus;
2797 if (session->shm_path[0]) {
2798 strncpy(lus->root_shm_path, session->shm_path,
2799 sizeof(lus->root_shm_path));
2800 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2801 strncpy(lus->shm_path, session->shm_path,
2802 sizeof(lus->shm_path));
2803 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2804 strncat(lus->shm_path, "/ust",
2805 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2806 }
2807 /* Copy session output to the newly created UST session */
2808 ret = copy_session_consumer(domain->type, session);
2809 if (ret != LTTNG_OK) {
2810 goto error;
2811 }
2812
2813 return LTTNG_OK;
2814
2815 error:
2816 free(lus);
2817 session->ust_session = NULL;
2818 return ret;
2819 }
2820
2821 /*
2822 * Create a kernel tracer session then create the default channel.
2823 */
2824 static int create_kernel_session(struct ltt_session *session)
2825 {
2826 int ret;
2827
2828 DBG("Creating kernel session");
2829
2830 ret = kernel_create_session(session, kernel_tracer_fd);
2831 if (ret < 0) {
2832 ret = LTTNG_ERR_KERN_SESS_FAIL;
2833 goto error;
2834 }
2835
2836 /* Code flow safety */
2837 assert(session->kernel_session);
2838
2839 /* Copy session output to the newly created Kernel session */
2840 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2841 if (ret != LTTNG_OK) {
2842 goto error;
2843 }
2844
2845 /* Create directory(ies) on local filesystem. */
2846 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2847 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2848 ret = run_as_mkdir_recursive(
2849 session->kernel_session->consumer->dst.trace_path,
2850 S_IRWXU | S_IRWXG, session->uid, session->gid);
2851 if (ret < 0) {
2852 if (errno != EEXIST) {
2853 ERR("Trace directory creation error");
2854 goto error;
2855 }
2856 }
2857 }
2858
2859 session->kernel_session->uid = session->uid;
2860 session->kernel_session->gid = session->gid;
2861 session->kernel_session->output_traces = session->output_traces;
2862 session->kernel_session->snapshot_mode = session->snapshot_mode;
2863
2864 return LTTNG_OK;
2865
2866 error:
2867 trace_kernel_destroy_session(session->kernel_session);
2868 session->kernel_session = NULL;
2869 return ret;
2870 }
2871
2872 /*
2873 * Count number of session permitted by uid/gid.
2874 */
2875 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2876 {
2877 unsigned int i = 0;
2878 struct ltt_session *session;
2879
2880 DBG("Counting number of available session for UID %d GID %d",
2881 uid, gid);
2882 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2883 /*
2884 * Only list the sessions the user can control.
2885 */
2886 if (!session_access_ok(session, uid, gid)) {
2887 continue;
2888 }
2889 i++;
2890 }
2891 return i;
2892 }
2893
2894 /*
2895 * Process the command requested by the lttng client within the command
2896 * context structure. This function make sure that the return structure (llm)
2897 * is set and ready for transmission before returning.
2898 *
2899 * Return any error encountered or 0 for success.
2900 *
2901 * "sock" is only used for special-case var. len data.
2902 *
2903 * Should *NOT* be called with RCU read-side lock held.
2904 */
2905 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2906 int *sock_error)
2907 {
2908 int ret = LTTNG_OK;
2909 int need_tracing_session = 1;
2910 int need_domain;
2911
2912 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2913
2914 assert(!rcu_read_ongoing());
2915
2916 *sock_error = 0;
2917
2918 switch (cmd_ctx->lsm->cmd_type) {
2919 case LTTNG_CREATE_SESSION:
2920 case LTTNG_CREATE_SESSION_SNAPSHOT:
2921 case LTTNG_CREATE_SESSION_LIVE:
2922 case LTTNG_DESTROY_SESSION:
2923 case LTTNG_LIST_SESSIONS:
2924 case LTTNG_LIST_DOMAINS:
2925 case LTTNG_START_TRACE:
2926 case LTTNG_STOP_TRACE:
2927 case LTTNG_DATA_PENDING:
2928 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2929 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2930 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2931 case LTTNG_SNAPSHOT_RECORD:
2932 case LTTNG_SAVE_SESSION:
2933 case LTTNG_SET_SESSION_SHM_PATH:
2934 case LTTNG_REGENERATE_METADATA:
2935 case LTTNG_REGENERATE_STATEDUMP:
2936 case LTTNG_REGISTER_TRIGGER:
2937 case LTTNG_UNREGISTER_TRIGGER:
2938 need_domain = 0;
2939 break;
2940 default:
2941 need_domain = 1;
2942 }
2943
2944 if (config.no_kernel && need_domain
2945 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2946 if (!is_root) {
2947 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2948 } else {
2949 ret = LTTNG_ERR_KERN_NA;
2950 }
2951 goto error;
2952 }
2953
2954 /* Deny register consumer if we already have a spawned consumer. */
2955 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2956 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2957 if (kconsumer_data.pid > 0) {
2958 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2959 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2960 goto error;
2961 }
2962 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2963 }
2964
2965 /*
2966 * Check for command that don't needs to allocate a returned payload. We do
2967 * this here so we don't have to make the call for no payload at each
2968 * command.
2969 */
2970 switch(cmd_ctx->lsm->cmd_type) {
2971 case LTTNG_LIST_SESSIONS:
2972 case LTTNG_LIST_TRACEPOINTS:
2973 case LTTNG_LIST_TRACEPOINT_FIELDS:
2974 case LTTNG_LIST_DOMAINS:
2975 case LTTNG_LIST_CHANNELS:
2976 case LTTNG_LIST_EVENTS:
2977 case LTTNG_LIST_SYSCALLS:
2978 case LTTNG_LIST_TRACKER_PIDS:
2979 case LTTNG_DATA_PENDING:
2980 break;
2981 default:
2982 /* Setup lttng message with no payload */
2983 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
2984 if (ret < 0) {
2985 /* This label does not try to unlock the session */
2986 goto init_setup_error;
2987 }
2988 }
2989
2990 /* Commands that DO NOT need a session. */
2991 switch (cmd_ctx->lsm->cmd_type) {
2992 case LTTNG_CREATE_SESSION:
2993 case LTTNG_CREATE_SESSION_SNAPSHOT:
2994 case LTTNG_CREATE_SESSION_LIVE:
2995 case LTTNG_LIST_SESSIONS:
2996 case LTTNG_LIST_TRACEPOINTS:
2997 case LTTNG_LIST_SYSCALLS:
2998 case LTTNG_LIST_TRACEPOINT_FIELDS:
2999 case LTTNG_SAVE_SESSION:
3000 case LTTNG_REGISTER_TRIGGER:
3001 case LTTNG_UNREGISTER_TRIGGER:
3002 need_tracing_session = 0;
3003 break;
3004 default:
3005 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3006 /*
3007 * We keep the session list lock across _all_ commands
3008 * for now, because the per-session lock does not
3009 * handle teardown properly.
3010 */
3011 session_lock_list();
3012 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3013 if (cmd_ctx->session == NULL) {
3014 ret = LTTNG_ERR_SESS_NOT_FOUND;
3015 goto error;
3016 } else {
3017 /* Acquire lock for the session */
3018 session_lock(cmd_ctx->session);
3019 }
3020 break;
3021 }
3022
3023 /*
3024 * Commands that need a valid session but should NOT create one if none
3025 * exists. Instead of creating one and destroying it when the command is
3026 * handled, process that right before so we save some round trip in useless
3027 * code path.
3028 */
3029 switch (cmd_ctx->lsm->cmd_type) {
3030 case LTTNG_DISABLE_CHANNEL:
3031 case LTTNG_DISABLE_EVENT:
3032 switch (cmd_ctx->lsm->domain.type) {
3033 case LTTNG_DOMAIN_KERNEL:
3034 if (!cmd_ctx->session->kernel_session) {
3035 ret = LTTNG_ERR_NO_CHANNEL;
3036 goto error;
3037 }
3038 break;
3039 case LTTNG_DOMAIN_JUL:
3040 case LTTNG_DOMAIN_LOG4J:
3041 case LTTNG_DOMAIN_PYTHON:
3042 case LTTNG_DOMAIN_UST:
3043 if (!cmd_ctx->session->ust_session) {
3044 ret = LTTNG_ERR_NO_CHANNEL;
3045 goto error;
3046 }
3047 break;
3048 default:
3049 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3050 goto error;
3051 }
3052 default:
3053 break;
3054 }
3055
3056 if (!need_domain) {
3057 goto skip_domain;
3058 }
3059
3060 /*
3061 * Check domain type for specific "pre-action".
3062 */
3063 switch (cmd_ctx->lsm->domain.type) {
3064 case LTTNG_DOMAIN_KERNEL:
3065 if (!is_root) {
3066 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3067 goto error;
3068 }
3069
3070 /* Kernel tracer check */
3071 if (kernel_tracer_fd == -1) {
3072 /* Basically, load kernel tracer modules */
3073 ret = init_kernel_tracer();
3074 if (ret != 0) {
3075 goto error;
3076 }
3077 }
3078
3079 /* Consumer is in an ERROR state. Report back to client */
3080 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3081 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3082 goto error;
3083 }
3084
3085 /* Need a session for kernel command */
3086 if (need_tracing_session) {
3087 if (cmd_ctx->session->kernel_session == NULL) {
3088 ret = create_kernel_session(cmd_ctx->session);
3089 if (ret < 0) {
3090 ret = LTTNG_ERR_KERN_SESS_FAIL;
3091 goto error;
3092 }
3093 }
3094
3095 /* Start the kernel consumer daemon */
3096 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3097 if (kconsumer_data.pid == 0 &&
3098 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3099 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3100 ret = start_consumerd(&kconsumer_data);
3101 if (ret < 0) {
3102 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3103 goto error;
3104 }
3105 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3106 } else {
3107 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3108 }
3109
3110 /*
3111 * The consumer was just spawned so we need to add the socket to
3112 * the consumer output of the session if exist.
3113 */
3114 ret = consumer_create_socket(&kconsumer_data,
3115 cmd_ctx->session->kernel_session->consumer);
3116 if (ret < 0) {
3117 goto error;
3118 }
3119 }
3120
3121 break;
3122 case LTTNG_DOMAIN_JUL:
3123 case LTTNG_DOMAIN_LOG4J:
3124 case LTTNG_DOMAIN_PYTHON:
3125 case LTTNG_DOMAIN_UST:
3126 {
3127 if (!ust_app_supported()) {
3128 ret = LTTNG_ERR_NO_UST;
3129 goto error;
3130 }
3131 /* Consumer is in an ERROR state. Report back to client */
3132 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3133 ret = LTTNG_ERR_NO_USTCONSUMERD;
3134 goto error;
3135 }
3136
3137 if (need_tracing_session) {
3138 /* Create UST session if none exist. */
3139 if (cmd_ctx->session->ust_session == NULL) {
3140 ret = create_ust_session(cmd_ctx->session,
3141 &cmd_ctx->lsm->domain);
3142 if (ret != LTTNG_OK) {
3143 goto error;
3144 }
3145 }
3146
3147 /* Start the UST consumer daemons */
3148 /* 64-bit */
3149 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3150 if (config.consumerd64_bin_path.value &&
3151 ustconsumer64_data.pid == 0 &&
3152 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3153 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3154 ret = start_consumerd(&ustconsumer64_data);
3155 if (ret < 0) {
3156 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3157 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3158 goto error;
3159 }
3160
3161 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3162 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3163 } else {
3164 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3165 }
3166
3167 /*
3168 * Setup socket for consumer 64 bit. No need for atomic access
3169 * since it was set above and can ONLY be set in this thread.
3170 */
3171 ret = consumer_create_socket(&ustconsumer64_data,
3172 cmd_ctx->session->ust_session->consumer);
3173 if (ret < 0) {
3174 goto error;
3175 }
3176
3177 /* 32-bit */
3178 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3179 if (config.consumerd32_bin_path.value &&
3180 ustconsumer32_data.pid == 0 &&
3181 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3182 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3183 ret = start_consumerd(&ustconsumer32_data);
3184 if (ret < 0) {
3185 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3186 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3187 goto error;
3188 }
3189
3190 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3191 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3192 } else {
3193 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3194 }
3195
3196 /*
3197 * Setup socket for consumer 64 bit. No need for atomic access
3198 * since it was set above and can ONLY be set in this thread.
3199 */
3200 ret = consumer_create_socket(&ustconsumer32_data,
3201 cmd_ctx->session->ust_session->consumer);
3202 if (ret < 0) {
3203 goto error;
3204 }
3205 }
3206 break;
3207 }
3208 default:
3209 break;
3210 }
3211 skip_domain:
3212
3213 /* Validate consumer daemon state when start/stop trace command */
3214 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3215 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3216 switch (cmd_ctx->lsm->domain.type) {
3217 case LTTNG_DOMAIN_NONE:
3218 break;
3219 case LTTNG_DOMAIN_JUL:
3220 case LTTNG_DOMAIN_LOG4J:
3221 case LTTNG_DOMAIN_PYTHON:
3222 case LTTNG_DOMAIN_UST:
3223 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3224 ret = LTTNG_ERR_NO_USTCONSUMERD;
3225 goto error;
3226 }
3227 break;
3228 case LTTNG_DOMAIN_KERNEL:
3229 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3230 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3231 goto error;
3232 }
3233 break;
3234 default:
3235 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3236 goto error;
3237 }
3238 }
3239
3240 /*
3241 * Check that the UID or GID match that of the tracing session.
3242 * The root user can interact with all sessions.
3243 */
3244 if (need_tracing_session) {
3245 if (!session_access_ok(cmd_ctx->session,
3246 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3247 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3248 ret = LTTNG_ERR_EPERM;
3249 goto error;
3250 }
3251 }
3252
3253 /*
3254 * Send relayd information to consumer as soon as we have a domain and a
3255 * session defined.
3256 */
3257 if (cmd_ctx->session && need_domain) {
3258 /*
3259 * Setup relayd if not done yet. If the relayd information was already
3260 * sent to the consumer, this call will gracefully return.
3261 */
3262 ret = cmd_setup_relayd(cmd_ctx->session);
3263 if (ret != LTTNG_OK) {
3264 goto error;
3265 }
3266 }
3267
3268 /* Process by command type */
3269 switch (cmd_ctx->lsm->cmd_type) {
3270 case LTTNG_ADD_CONTEXT:
3271 {
3272 /*
3273 * An LTTNG_ADD_CONTEXT command might have a supplementary
3274 * payload if the context being added is an application context.
3275 */
3276 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3277 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3278 char *provider_name = NULL, *context_name = NULL;
3279 size_t provider_name_len =
3280 cmd_ctx->lsm->u.context.provider_name_len;
3281 size_t context_name_len =
3282 cmd_ctx->lsm->u.context.context_name_len;
3283
3284 if (provider_name_len == 0 || context_name_len == 0) {
3285 /*
3286 * Application provider and context names MUST
3287 * be provided.
3288 */
3289 ret = -LTTNG_ERR_INVALID;
3290 goto error;
3291 }
3292
3293 provider_name = zmalloc(provider_name_len + 1);
3294 if (!provider_name) {
3295 ret = -LTTNG_ERR_NOMEM;
3296 goto error;
3297 }
3298 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3299 provider_name;
3300
3301 context_name = zmalloc(context_name_len + 1);
3302 if (!context_name) {
3303 ret = -LTTNG_ERR_NOMEM;
3304 goto error_add_context;
3305 }
3306 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3307 context_name;
3308
3309 ret = lttcomm_recv_unix_sock(sock, provider_name,
3310 provider_name_len);
3311 if (ret < 0) {
3312 goto error_add_context;
3313 }
3314
3315 ret = lttcomm_recv_unix_sock(sock, context_name,
3316 context_name_len);
3317 if (ret < 0) {
3318 goto error_add_context;
3319 }
3320 }
3321
3322 /*
3323 * cmd_add_context assumes ownership of the provider and context
3324 * names.
3325 */
3326 ret = cmd_add_context(cmd_ctx->session,
3327 cmd_ctx->lsm->domain.type,
3328 cmd_ctx->lsm->u.context.channel_name,
3329 &cmd_ctx->lsm->u.context.ctx,
3330 kernel_poll_pipe[1]);
3331
3332 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3333 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3334 error_add_context:
3335 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3336 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3337 if (ret < 0) {
3338 goto error;
3339 }
3340 break;
3341 }
3342 case LTTNG_DISABLE_CHANNEL:
3343 {
3344 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3345 cmd_ctx->lsm->u.disable.channel_name);
3346 break;
3347 }
3348 case LTTNG_DISABLE_EVENT:
3349 {
3350
3351 /*
3352 * FIXME: handle filter; for now we just receive the filter's
3353 * bytecode along with the filter expression which are sent by
3354 * liblttng-ctl and discard them.
3355 *
3356 * This fixes an issue where the client may block while sending
3357 * the filter payload and encounter an error because the session
3358 * daemon closes the socket without ever handling this data.
3359 */
3360 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3361 cmd_ctx->lsm->u.disable.bytecode_len;
3362
3363 if (count) {
3364 char data[LTTNG_FILTER_MAX_LEN];
3365
3366 DBG("Discarding disable event command payload of size %zu", count);
3367 while (count) {
3368 ret = lttcomm_recv_unix_sock(sock, data,
3369 count > sizeof(data) ? sizeof(data) : count);
3370 if (ret < 0) {
3371 goto error;
3372 }
3373
3374 count -= (size_t) ret;
3375 }
3376 }
3377 /* FIXME: passing packed structure to non-packed pointer */
3378 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3379 cmd_ctx->lsm->u.disable.channel_name,
3380 &cmd_ctx->lsm->u.disable.event);
3381 break;
3382 }
3383 case LTTNG_ENABLE_CHANNEL:
3384 {
3385 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3386 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3387 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3388 &cmd_ctx->lsm->u.channel.chan,
3389 kernel_poll_pipe[1]);
3390 break;
3391 }
3392 case LTTNG_TRACK_PID:
3393 {
3394 ret = cmd_track_pid(cmd_ctx->session,
3395 cmd_ctx->lsm->domain.type,
3396 cmd_ctx->lsm->u.pid_tracker.pid);
3397 break;
3398 }
3399 case LTTNG_UNTRACK_PID:
3400 {
3401 ret = cmd_untrack_pid(cmd_ctx->session,
3402 cmd_ctx->lsm->domain.type,
3403 cmd_ctx->lsm->u.pid_tracker.pid);
3404 break;
3405 }
3406 case LTTNG_ENABLE_EVENT:
3407 {
3408 struct lttng_event_exclusion *exclusion = NULL;
3409 struct lttng_filter_bytecode *bytecode = NULL;
3410 char *filter_expression = NULL;
3411
3412 /* Handle exclusion events and receive it from the client. */
3413 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3414 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3415
3416 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3417 (count * LTTNG_SYMBOL_NAME_LEN));
3418 if (!exclusion) {
3419 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3420 goto error;
3421 }
3422
3423 DBG("Receiving var len exclusion event list from client ...");
3424 exclusion->count = count;
3425 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3426 count * LTTNG_SYMBOL_NAME_LEN);
3427 if (ret <= 0) {
3428 DBG("Nothing recv() from client var len data... continuing");
3429 *sock_error = 1;
3430 free(exclusion);
3431 ret = LTTNG_ERR_EXCLUSION_INVAL;
3432 goto error;
3433 }
3434 }
3435
3436 /* Get filter expression from client. */
3437 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3438 size_t expression_len =
3439 cmd_ctx->lsm->u.enable.expression_len;
3440
3441 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3442 ret = LTTNG_ERR_FILTER_INVAL;
3443 free(exclusion);
3444 goto error;
3445 }
3446
3447 filter_expression = zmalloc(expression_len);
3448 if (!filter_expression) {
3449 free(exclusion);
3450 ret = LTTNG_ERR_FILTER_NOMEM;
3451 goto error;
3452 }
3453
3454 /* Receive var. len. data */
3455 DBG("Receiving var len filter's expression from client ...");
3456 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3457 expression_len);
3458 if (ret <= 0) {
3459 DBG("Nothing recv() from client car len data... continuing");
3460 *sock_error = 1;
3461 free(filter_expression);
3462 free(exclusion);
3463 ret = LTTNG_ERR_FILTER_INVAL;
3464 goto error;
3465 }
3466 }
3467
3468 /* Handle filter and get bytecode from client. */
3469 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3470 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3471
3472 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3473 ret = LTTNG_ERR_FILTER_INVAL;
3474 free(filter_expression);
3475 free(exclusion);
3476 goto error;
3477 }
3478
3479 bytecode = zmalloc(bytecode_len);
3480 if (!bytecode) {
3481 free(filter_expression);
3482 free(exclusion);
3483 ret = LTTNG_ERR_FILTER_NOMEM;
3484 goto error;
3485 }
3486
3487 /* Receive var. len. data */
3488 DBG("Receiving var len filter's bytecode from client ...");
3489 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3490 if (ret <= 0) {
3491 DBG("Nothing recv() from client car len data... continuing");
3492 *sock_error = 1;
3493 free(filter_expression);
3494 free(bytecode);
3495 free(exclusion);
3496 ret = LTTNG_ERR_FILTER_INVAL;
3497 goto error;
3498 }
3499
3500 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3501 free(filter_expression);
3502 free(bytecode);
3503 free(exclusion);
3504 ret = LTTNG_ERR_FILTER_INVAL;
3505 goto error;
3506 }
3507 }
3508
3509 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3510 cmd_ctx->lsm->u.enable.channel_name,
3511 &cmd_ctx->lsm->u.enable.event,
3512 filter_expression, bytecode, exclusion,
3513 kernel_poll_pipe[1]);
3514 break;
3515 }
3516 case LTTNG_LIST_TRACEPOINTS:
3517 {
3518 struct lttng_event *events;
3519 ssize_t nb_events;
3520
3521 session_lock_list();
3522 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3523 session_unlock_list();
3524 if (nb_events < 0) {
3525 /* Return value is a negative lttng_error_code. */
3526 ret = -nb_events;
3527 goto error;
3528 }
3529
3530 /*
3531 * Setup lttng message with payload size set to the event list size in
3532 * bytes and then copy list into the llm payload.
3533 */
3534 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3535 sizeof(struct lttng_event) * nb_events);
3536 free(events);
3537
3538 if (ret < 0) {
3539 goto setup_error;
3540 }
3541
3542 ret = LTTNG_OK;
3543 break;
3544 }
3545 case LTTNG_LIST_TRACEPOINT_FIELDS:
3546 {
3547 struct lttng_event_field *fields;
3548 ssize_t nb_fields;
3549
3550 session_lock_list();
3551 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3552 &fields);
3553 session_unlock_list();
3554 if (nb_fields < 0) {
3555 /* Return value is a negative lttng_error_code. */
3556 ret = -nb_fields;
3557 goto error;
3558 }
3559
3560 /*
3561 * Setup lttng message with payload size set to the event list size in
3562 * bytes and then copy list into the llm payload.
3563 */
3564 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3565 sizeof(struct lttng_event_field) * nb_fields);
3566 free(fields);
3567
3568 if (ret < 0) {
3569 goto setup_error;
3570 }
3571
3572 ret = LTTNG_OK;
3573 break;
3574 }
3575 case LTTNG_LIST_SYSCALLS:
3576 {
3577 struct lttng_event *events;
3578 ssize_t nb_events;
3579
3580 nb_events = cmd_list_syscalls(&events);
3581 if (nb_events < 0) {
3582 /* Return value is a negative lttng_error_code. */
3583 ret = -nb_events;
3584 goto error;
3585 }
3586
3587 /*
3588 * Setup lttng message with payload size set to the event list size in
3589 * bytes and then copy list into the llm payload.
3590 */
3591 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3592 sizeof(struct lttng_event) * nb_events);
3593 free(events);
3594
3595 if (ret < 0) {
3596 goto setup_error;
3597 }
3598
3599 ret = LTTNG_OK;
3600 break;
3601 }
3602 case LTTNG_LIST_TRACKER_PIDS:
3603 {
3604 int32_t *pids = NULL;
3605 ssize_t nr_pids;
3606
3607 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3608 cmd_ctx->lsm->domain.type, &pids);
3609 if (nr_pids < 0) {
3610 /* Return value is a negative lttng_error_code. */
3611 ret = -nr_pids;
3612 goto error;
3613 }
3614
3615 /*
3616 * Setup lttng message with payload size set to the event list size in
3617 * bytes and then copy list into the llm payload.
3618 */
3619 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3620 sizeof(int32_t) * nr_pids);
3621 free(pids);
3622
3623 if (ret < 0) {
3624 goto setup_error;
3625 }
3626
3627 ret = LTTNG_OK;
3628 break;
3629 }
3630 case LTTNG_SET_CONSUMER_URI:
3631 {
3632 size_t nb_uri, len;
3633 struct lttng_uri *uris;
3634
3635 nb_uri = cmd_ctx->lsm->u.uri.size;
3636 len = nb_uri * sizeof(struct lttng_uri);
3637
3638 if (nb_uri == 0) {
3639 ret = LTTNG_ERR_INVALID;
3640 goto error;
3641 }
3642
3643 uris = zmalloc(len);
3644 if (uris == NULL) {
3645 ret = LTTNG_ERR_FATAL;
3646 goto error;
3647 }
3648
3649 /* Receive variable len data */
3650 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3651 ret = lttcomm_recv_unix_sock(sock, uris, len);
3652 if (ret <= 0) {
3653 DBG("No URIs received from client... continuing");
3654 *sock_error = 1;
3655 ret = LTTNG_ERR_SESSION_FAIL;
3656 free(uris);
3657 goto error;
3658 }
3659
3660 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3661 free(uris);
3662 if (ret != LTTNG_OK) {
3663 goto error;
3664 }
3665
3666
3667 break;
3668 }
3669 case LTTNG_START_TRACE:
3670 {
3671 ret = cmd_start_trace(cmd_ctx->session);
3672 break;
3673 }
3674 case LTTNG_STOP_TRACE:
3675 {
3676 ret = cmd_stop_trace(cmd_ctx->session);
3677 break;
3678 }
3679 case LTTNG_CREATE_SESSION:
3680 {
3681 size_t nb_uri, len;
3682 struct lttng_uri *uris = NULL;
3683
3684 nb_uri = cmd_ctx->lsm->u.uri.size;
3685 len = nb_uri * sizeof(struct lttng_uri);
3686
3687 if (nb_uri > 0) {
3688 uris = zmalloc(len);
3689 if (uris == NULL) {
3690 ret = LTTNG_ERR_FATAL;
3691 goto error;
3692 }
3693
3694 /* Receive variable len data */
3695 DBG("Waiting for %zu URIs from client ...", nb_uri);
3696 ret = lttcomm_recv_unix_sock(sock, uris, len);
3697 if (ret <= 0) {
3698 DBG("No URIs received from client... continuing");
3699 *sock_error = 1;
3700 ret = LTTNG_ERR_SESSION_FAIL;
3701 free(uris);
3702 goto error;
3703 }
3704
3705 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3706 DBG("Creating session with ONE network URI is a bad call");
3707 ret = LTTNG_ERR_SESSION_FAIL;
3708 free(uris);
3709 goto error;
3710 }
3711 }
3712
3713 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3714 &cmd_ctx->creds, 0);
3715
3716 free(uris);
3717
3718 break;
3719 }
3720 case LTTNG_DESTROY_SESSION:
3721 {
3722 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3723
3724 /* Set session to NULL so we do not unlock it after free. */
3725 cmd_ctx->session = NULL;
3726 break;
3727 }
3728 case LTTNG_LIST_DOMAINS:
3729 {
3730 ssize_t nb_dom;
3731 struct lttng_domain *domains = NULL;
3732
3733 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3734 if (nb_dom < 0) {
3735 /* Return value is a negative lttng_error_code. */
3736 ret = -nb_dom;
3737 goto error;
3738 }
3739
3740 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3741 nb_dom * sizeof(struct lttng_domain));
3742 free(domains);
3743
3744 if (ret < 0) {
3745 goto setup_error;
3746 }
3747
3748 ret = LTTNG_OK;
3749 break;
3750 }
3751 case LTTNG_LIST_CHANNELS:
3752 {
3753 ssize_t payload_size;
3754 struct lttng_channel *channels = NULL;
3755
3756 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3757 cmd_ctx->session, &channels);
3758 if (payload_size < 0) {
3759 /* Return value is a negative lttng_error_code. */
3760 ret = -payload_size;
3761 goto error;
3762 }
3763
3764 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3765 payload_size);
3766 free(channels);
3767
3768 if (ret < 0) {
3769 goto setup_error;
3770 }
3771
3772 ret = LTTNG_OK;
3773 break;
3774 }
3775 case LTTNG_LIST_EVENTS:
3776 {
3777 ssize_t nb_event;
3778 struct lttng_event *events = NULL;
3779 struct lttcomm_event_command_header cmd_header;
3780 size_t total_size;
3781
3782 memset(&cmd_header, 0, sizeof(cmd_header));
3783 /* Extended infos are included at the end of events */
3784 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3785 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3786 &events, &total_size);
3787
3788 if (nb_event < 0) {
3789 /* Return value is a negative lttng_error_code. */
3790 ret = -nb_event;
3791 goto error;
3792 }
3793
3794 cmd_header.nb_events = nb_event;
3795 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3796 &cmd_header, sizeof(cmd_header));
3797 free(events);
3798
3799 if (ret < 0) {
3800 goto setup_error;
3801 }
3802
3803 ret = LTTNG_OK;
3804 break;
3805 }
3806 case LTTNG_LIST_SESSIONS:
3807 {
3808 unsigned int nr_sessions;
3809 void *sessions_payload;
3810 size_t payload_len;
3811
3812 session_lock_list();
3813 nr_sessions = lttng_sessions_count(
3814 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3815 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3816 payload_len = sizeof(struct lttng_session) * nr_sessions;
3817 sessions_payload = zmalloc(payload_len);
3818
3819 if (!sessions_payload) {
3820 session_unlock_list();
3821 ret = -ENOMEM;
3822 goto setup_error;
3823 }
3824
3825 cmd_list_lttng_sessions(sessions_payload,
3826 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3827 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3828 session_unlock_list();
3829
3830 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3831 payload_len);
3832 free(sessions_payload);
3833
3834 if (ret < 0) {
3835 goto setup_error;
3836 }
3837
3838 ret = LTTNG_OK;
3839 break;
3840 }
3841 case LTTNG_REGISTER_CONSUMER:
3842 {
3843 struct consumer_data *cdata;
3844
3845 switch (cmd_ctx->lsm->domain.type) {
3846 case LTTNG_DOMAIN_KERNEL:
3847 cdata = &kconsumer_data;
3848 break;
3849 default:
3850 ret = LTTNG_ERR_UND;
3851 goto error;
3852 }
3853
3854 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3855 cmd_ctx->lsm->u.reg.path, cdata);
3856 break;
3857 }
3858 case LTTNG_DATA_PENDING:
3859 {
3860 int pending_ret;
3861 uint8_t pending_ret_byte;
3862
3863 pending_ret = cmd_data_pending(cmd_ctx->session);
3864
3865 /*
3866 * FIXME
3867 *
3868 * This function may returns 0 or 1 to indicate whether or not
3869 * there is data pending. In case of error, it should return an
3870 * LTTNG_ERR code. However, some code paths may still return
3871 * a nondescript error code, which we handle by returning an
3872 * "unknown" error.
3873 */
3874 if (pending_ret == 0 || pending_ret == 1) {
3875 /*
3876 * ret will be set to LTTNG_OK at the end of
3877 * this function.
3878 */
3879 } else if (pending_ret < 0) {
3880 ret = LTTNG_ERR_UNK;
3881 goto setup_error;
3882 } else {
3883 ret = pending_ret;
3884 goto setup_error;
3885 }
3886
3887 pending_ret_byte = (uint8_t) pending_ret;
3888
3889 /* 1 byte to return whether or not data is pending */
3890 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
3891 &pending_ret_byte, 1);
3892
3893 if (ret < 0) {
3894 goto setup_error;
3895 }
3896
3897 ret = LTTNG_OK;
3898 break;
3899 }
3900 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3901 {
3902 struct lttcomm_lttng_output_id reply;
3903
3904 ret = cmd_snapshot_add_output(cmd_ctx->session,
3905 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3906 if (ret != LTTNG_OK) {
3907 goto error;
3908 }
3909
3910 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
3911 sizeof(reply));
3912 if (ret < 0) {
3913 goto setup_error;
3914 }
3915
3916 /* Copy output list into message payload */
3917 ret = LTTNG_OK;
3918 break;
3919 }
3920 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3921 {
3922 ret = cmd_snapshot_del_output(cmd_ctx->session,
3923 &cmd_ctx->lsm->u.snapshot_output.output);
3924 break;
3925 }
3926 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3927 {
3928 ssize_t nb_output;
3929 struct lttng_snapshot_output *outputs = NULL;
3930
3931 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3932 if (nb_output < 0) {
3933 ret = -nb_output;
3934 goto error;
3935 }
3936
3937 assert((nb_output > 0 && outputs) || nb_output == 0);
3938 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
3939 nb_output * sizeof(struct lttng_snapshot_output));
3940 free(outputs);
3941
3942 if (ret < 0) {
3943 goto setup_error;
3944 }
3945
3946 ret = LTTNG_OK;
3947 break;
3948 }
3949 case LTTNG_SNAPSHOT_RECORD:
3950 {
3951 ret = cmd_snapshot_record(cmd_ctx->session,
3952 &cmd_ctx->lsm->u.snapshot_record.output,
3953 cmd_ctx->lsm->u.snapshot_record.wait);
3954 break;
3955 }
3956 case LTTNG_CREATE_SESSION_SNAPSHOT:
3957 {
3958 size_t nb_uri, len;
3959 struct lttng_uri *uris = NULL;
3960
3961 nb_uri = cmd_ctx->lsm->u.uri.size;
3962 len = nb_uri * sizeof(struct lttng_uri);
3963
3964 if (nb_uri > 0) {
3965 uris = zmalloc(len);
3966 if (uris == NULL) {
3967 ret = LTTNG_ERR_FATAL;
3968 goto error;
3969 }
3970
3971 /* Receive variable len data */
3972 DBG("Waiting for %zu URIs from client ...", nb_uri);
3973 ret = lttcomm_recv_unix_sock(sock, uris, len);
3974 if (ret <= 0) {
3975 DBG("No URIs received from client... continuing");
3976 *sock_error = 1;
3977 ret = LTTNG_ERR_SESSION_FAIL;
3978 free(uris);
3979 goto error;
3980 }
3981
3982 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3983 DBG("Creating session with ONE network URI is a bad call");
3984 ret = LTTNG_ERR_SESSION_FAIL;
3985 free(uris);
3986 goto error;
3987 }
3988 }
3989
3990 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3991 nb_uri, &cmd_ctx->creds);
3992 free(uris);
3993 break;
3994 }
3995 case LTTNG_CREATE_SESSION_LIVE:
3996 {
3997 size_t nb_uri, len;
3998 struct lttng_uri *uris = NULL;
3999
4000 nb_uri = cmd_ctx->lsm->u.uri.size;
4001 len = nb_uri * sizeof(struct lttng_uri);
4002
4003 if (nb_uri > 0) {
4004 uris = zmalloc(len);
4005 if (uris == NULL) {
4006 ret = LTTNG_ERR_FATAL;
4007 goto error;
4008 }
4009
4010 /* Receive variable len data */
4011 DBG("Waiting for %zu URIs from client ...", nb_uri);
4012 ret = lttcomm_recv_unix_sock(sock, uris, len);
4013 if (ret <= 0) {
4014 DBG("No URIs received from client... continuing");
4015 *sock_error = 1;
4016 ret = LTTNG_ERR_SESSION_FAIL;
4017 free(uris);
4018 goto error;
4019 }
4020
4021 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4022 DBG("Creating session with ONE network URI is a bad call");
4023 ret = LTTNG_ERR_SESSION_FAIL;
4024 free(uris);
4025 goto error;
4026 }
4027 }
4028
4029 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4030 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4031 free(uris);
4032 break;
4033 }
4034 case LTTNG_SAVE_SESSION:
4035 {
4036 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4037 &cmd_ctx->creds);
4038 break;
4039 }
4040 case LTTNG_SET_SESSION_SHM_PATH:
4041 {
4042 ret = cmd_set_session_shm_path(cmd_ctx->session,
4043 cmd_ctx->lsm->u.set_shm_path.shm_path);
4044 break;
4045 }
4046 case LTTNG_REGENERATE_METADATA:
4047 {
4048 ret = cmd_regenerate_metadata(cmd_ctx->session);
4049 break;
4050 }
4051 case LTTNG_REGENERATE_STATEDUMP:
4052 {
4053 ret = cmd_regenerate_statedump(cmd_ctx->session);
4054 break;
4055 }
4056 case LTTNG_REGISTER_TRIGGER:
4057 {
4058 ret = cmd_register_trigger(cmd_ctx, sock,
4059 notification_thread_handle);
4060 break;
4061 }
4062 case LTTNG_UNREGISTER_TRIGGER:
4063 {
4064 ret = cmd_unregister_trigger(cmd_ctx, sock,
4065 notification_thread_handle);
4066 break;
4067 }
4068 default:
4069 ret = LTTNG_ERR_UND;
4070 break;
4071 }
4072
4073 error:
4074 if (cmd_ctx->llm == NULL) {
4075 DBG("Missing llm structure. Allocating one.");
4076 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4077 goto setup_error;
4078 }
4079 }
4080 /* Set return code */
4081 cmd_ctx->llm->ret_code = ret;
4082 setup_error:
4083 if (cmd_ctx->session) {
4084 session_unlock(cmd_ctx->session);
4085 }
4086 if (need_tracing_session) {
4087 session_unlock_list();
4088 }
4089 init_setup_error:
4090 assert(!rcu_read_ongoing());
4091 return ret;
4092 }
4093
4094 /*
4095 * Thread managing health check socket.
4096 */
4097 static void *thread_manage_health(void *data)
4098 {
4099 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4100 uint32_t revents, nb_fd;
4101 struct lttng_poll_event events;
4102 struct health_comm_msg msg;
4103 struct health_comm_reply reply;
4104
4105 DBG("[thread] Manage health check started");
4106
4107 rcu_register_thread();
4108
4109 /* We might hit an error path before this is created. */
4110 lttng_poll_init(&events);
4111
4112 /* Create unix socket */
4113 sock = lttcomm_create_unix_sock(config.health_unix_sock_path.value);
4114 if (sock < 0) {
4115 ERR("Unable to create health check Unix socket");
4116 goto error;
4117 }
4118
4119 if (is_root) {
4120 gid_t gid;
4121
4122 /* lttng health client socket path permissions */
4123 ret = utils_get_group_id(config.tracing_group_name.value, true,
4124 &gid);
4125 if (ret) {
4126 /* Default to root group. */
4127 gid = 0;
4128 }
4129 ret = chown(config.health_unix_sock_path.value, 0,
4130 gid);
4131 if (ret < 0) {
4132 ERR("Unable to set group on %s", config.health_unix_sock_path.value);
4133 PERROR("chown");
4134 goto error;
4135 }
4136
4137 ret = chmod(config.health_unix_sock_path.value,
4138 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4139 if (ret < 0) {
4140 ERR("Unable to set permissions on %s", config.health_unix_sock_path.value);
4141 PERROR("chmod");
4142 goto error;
4143 }
4144 }
4145
4146 /*
4147 * Set the CLOEXEC flag. Return code is useless because either way, the
4148 * show must go on.
4149 */
4150 (void) utils_set_fd_cloexec(sock);
4151
4152 ret = lttcomm_listen_unix_sock(sock);
4153 if (ret < 0) {
4154 goto error;
4155 }
4156
4157 /*
4158 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4159 * more will be added to this poll set.
4160 */
4161 ret = sessiond_set_thread_pollset(&events, 2);
4162 if (ret < 0) {
4163 goto error;
4164 }
4165
4166 /* Add the application registration socket */
4167 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4168 if (ret < 0) {
4169 goto error;
4170 }
4171
4172 sessiond_notify_ready();
4173
4174 while (1) {
4175 DBG("Health check ready");
4176
4177 /* Inifinite blocking call, waiting for transmission */
4178 restart:
4179 ret = lttng_poll_wait(&events, -1);
4180 if (ret < 0) {
4181 /*
4182 * Restart interrupted system call.
4183 */
4184 if (errno == EINTR) {
4185 goto restart;
4186 }
4187 goto error;
4188 }
4189
4190 nb_fd = ret;
4191
4192 for (i = 0; i < nb_fd; i++) {
4193 /* Fetch once the poll data */
4194 revents = LTTNG_POLL_GETEV(&events, i);
4195 pollfd = LTTNG_POLL_GETFD(&events, i);
4196
4197 if (!revents) {
4198 /* No activity for this FD (poll implementation). */
4199 continue;
4200 }
4201
4202 /* Thread quit pipe has been closed. Killing thread. */
4203 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4204 if (ret) {
4205 err = 0;
4206 goto exit;
4207 }
4208
4209 /* Event on the registration socket */
4210 if (pollfd == sock) {
4211 if (revents & LPOLLIN) {
4212 continue;
4213 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4214 ERR("Health socket poll error");
4215 goto error;
4216 } else {
4217 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4218 goto error;
4219 }
4220 }
4221 }
4222
4223 new_sock = lttcomm_accept_unix_sock(sock);
4224 if (new_sock < 0) {
4225 goto error;
4226 }
4227
4228 /*
4229 * Set the CLOEXEC flag. Return code is useless because either way, the
4230 * show must go on.
4231 */
4232 (void) utils_set_fd_cloexec(new_sock);
4233
4234 DBG("Receiving data from client for health...");
4235 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4236 if (ret <= 0) {
4237 DBG("Nothing recv() from client... continuing");
4238 ret = close(new_sock);
4239 if (ret) {
4240 PERROR("close");
4241 }
4242 continue;
4243 }
4244
4245 rcu_thread_online();
4246
4247 memset(&reply, 0, sizeof(reply));
4248 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4249 /*
4250 * health_check_state returns 0 if health is
4251 * bad.
4252 */
4253 if (!health_check_state(health_sessiond, i)) {
4254 reply.ret_code |= 1ULL << i;
4255 }
4256 }
4257
4258 DBG2("Health check return value %" PRIx64, reply.ret_code);
4259
4260 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4261 if (ret < 0) {
4262 ERR("Failed to send health data back to client");
4263 }
4264
4265 /* End of transmission */
4266 ret = close(new_sock);
4267 if (ret) {
4268 PERROR("close");
4269 }
4270 }
4271
4272 exit:
4273 error:
4274 if (err) {
4275 ERR("Health error occurred in %s", __func__);
4276 }
4277 DBG("Health check thread dying");
4278 unlink(config.health_unix_sock_path.value);
4279 if (sock >= 0) {
4280 ret = close(sock);
4281 if (ret) {
4282 PERROR("close");
4283 }
4284 }
4285
4286 lttng_poll_clean(&events);
4287 stop_threads();
4288 rcu_unregister_thread();
4289 return NULL;
4290 }
4291
4292 /*
4293 * This thread manage all clients request using the unix client socket for
4294 * communication.
4295 */
4296 static void *thread_manage_clients(void *data)
4297 {
4298 int sock = -1, ret, i, pollfd, err = -1;
4299 int sock_error;
4300 uint32_t revents, nb_fd;
4301 struct command_ctx *cmd_ctx = NULL;
4302 struct lttng_poll_event events;
4303
4304 DBG("[thread] Manage client started");
4305
4306 rcu_register_thread();
4307
4308 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4309
4310 health_code_update();
4311
4312 ret = lttcomm_listen_unix_sock(client_sock);
4313 if (ret < 0) {
4314 goto error_listen;
4315 }
4316
4317 /*
4318 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4319 * more will be added to this poll set.
4320 */
4321 ret = sessiond_set_thread_pollset(&events, 2);
4322 if (ret < 0) {
4323 goto error_create_poll;
4324 }
4325
4326 /* Add the application registration socket */
4327 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4328 if (ret < 0) {
4329 goto error;
4330 }
4331
4332 ret = sem_post(&load_info->message_thread_ready);
4333 if (ret) {
4334 PERROR("sem_post message_thread_ready");
4335 goto error;
4336 }
4337
4338 /*
4339 * Wait until all support threads are initialized before accepting
4340 * commands.
4341 */
4342 while (uatomic_read(&lttng_sessiond_ready) != 0) {
4343 fd_set read_fds;
4344 struct timeval timeout;
4345
4346 FD_ZERO(&read_fds);
4347 FD_SET(thread_quit_pipe[0], &read_fds);
4348 memset(&timeout, 0, sizeof(timeout));
4349 timeout.tv_usec = 1000;
4350
4351 /*
4352 * If a support thread failed to launch, it may signal that
4353 * we must exit and the sessiond would never be marked as
4354 * "ready".
4355 *
4356 * The timeout is set to 1ms, which serves as a way to
4357 * pace down this check.
4358 */
4359 ret = select(thread_quit_pipe[0] + 1, &read_fds, NULL, NULL,
4360 &timeout);
4361 if (ret > 0 || (ret < 0 && errno != EINTR)) {
4362 goto exit;
4363 }
4364 }
4365 /*
4366 * This barrier is paired with the one in sessiond_notify_ready() to
4367 * ensure that loads accessing data initialized by the other threads,
4368 * on which this thread was waiting, are not performed before this point.
4369 *
4370 * Note that this could be a 'read' memory barrier, but a full barrier
4371 * is used in case the code changes. The performance implications of
4372 * this choice are minimal since this is a slow path.
4373 */
4374 cmm_smp_mb();
4375
4376 /* This testpoint is after we signal readiness to the parent. */
4377 if (testpoint(sessiond_thread_manage_clients)) {
4378 goto error;
4379 }
4380
4381 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4382 goto error;
4383 }
4384
4385 health_code_update();
4386
4387 while (1) {
4388 DBG("Accepting client command ...");
4389
4390 /* Inifinite blocking call, waiting for transmission */
4391 restart:
4392 health_poll_entry();
4393 ret = lttng_poll_wait(&events, -1);
4394 health_poll_exit();
4395 if (ret < 0) {
4396 /*
4397 * Restart interrupted system call.
4398 */
4399 if (errno == EINTR) {
4400 goto restart;
4401 }
4402 goto error;
4403 }
4404
4405 nb_fd = ret;
4406
4407 for (i = 0; i < nb_fd; i++) {
4408 /* Fetch once the poll data */
4409 revents = LTTNG_POLL_GETEV(&events, i);
4410 pollfd = LTTNG_POLL_GETFD(&events, i);
4411
4412 health_code_update();
4413
4414 if (!revents) {
4415 /* No activity for this FD (poll implementation). */
4416 continue;
4417 }
4418
4419 /* Thread quit pipe has been closed. Killing thread. */
4420 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4421 if (ret) {
4422 err = 0;
4423 goto exit;
4424 }
4425
4426 /* Event on the registration socket */
4427 if (pollfd == client_sock) {
4428 if (revents & LPOLLIN) {
4429 continue;
4430 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4431 ERR("Client socket poll error");
4432 goto error;
4433 } else {
4434 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4435 goto error;
4436 }
4437 }
4438 }
4439
4440 DBG("Wait for client response");
4441
4442 health_code_update();
4443
4444 sock = lttcomm_accept_unix_sock(client_sock);
4445 if (sock < 0) {
4446 goto error;
4447 }
4448
4449 /*
4450 * Set the CLOEXEC flag. Return code is useless because either way, the
4451 * show must go on.
4452 */
4453 (void) utils_set_fd_cloexec(sock);
4454
4455 /* Set socket option for credentials retrieval */
4456 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4457 if (ret < 0) {
4458 goto error;
4459 }
4460
4461 /* Allocate context command to process the client request */
4462 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4463 if (cmd_ctx == NULL) {
4464 PERROR("zmalloc cmd_ctx");
4465 goto error;
4466 }
4467
4468 /* Allocate data buffer for reception */
4469 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4470 if (cmd_ctx->lsm == NULL) {
4471 PERROR("zmalloc cmd_ctx->lsm");
4472 goto error;
4473 }
4474
4475 cmd_ctx->llm = NULL;
4476 cmd_ctx->session = NULL;
4477
4478 health_code_update();
4479
4480 /*
4481 * Data is received from the lttng client. The struct
4482 * lttcomm_session_msg (lsm) contains the command and data request of
4483 * the client.
4484 */
4485 DBG("Receiving data from client ...");
4486 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4487 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4488 if (ret <= 0) {
4489 DBG("Nothing recv() from client... continuing");
4490 ret = close(sock);
4491 if (ret) {
4492 PERROR("close");
4493 }
4494 sock = -1;
4495 clean_command_ctx(&cmd_ctx);
4496 continue;
4497 }
4498
4499 health_code_update();
4500
4501 // TODO: Validate cmd_ctx including sanity check for
4502 // security purpose.
4503
4504 rcu_thread_online();
4505 /*
4506 * This function dispatch the work to the kernel or userspace tracer
4507 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4508 * informations for the client. The command context struct contains
4509 * everything this function may needs.
4510 */
4511 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4512 rcu_thread_offline();
4513 if (ret < 0) {
4514 ret = close(sock);
4515 if (ret) {
4516 PERROR("close");
4517 }
4518 sock = -1;
4519 /*
4520 * TODO: Inform client somehow of the fatal error. At
4521 * this point, ret < 0 means that a zmalloc failed
4522 * (ENOMEM). Error detected but still accept
4523 * command, unless a socket error has been
4524 * detected.
4525 */
4526 clean_command_ctx(&cmd_ctx);
4527 continue;
4528 }
4529
4530 health_code_update();
4531
4532 DBG("Sending response (size: %d, retcode: %s (%d))",
4533 cmd_ctx->lttng_msg_size,
4534 lttng_strerror(-cmd_ctx->llm->ret_code),
4535 cmd_ctx->llm->ret_code);
4536 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4537 if (ret < 0) {
4538 ERR("Failed to send data back to client");
4539 }
4540
4541 /* End of transmission */
4542 ret = close(sock);
4543 if (ret) {
4544 PERROR("close");
4545 }
4546 sock = -1;
4547
4548 clean_command_ctx(&cmd_ctx);
4549
4550 health_code_update();
4551 }
4552
4553 exit:
4554 error:
4555 if (sock >= 0) {
4556 ret = close(sock);
4557 if (ret) {
4558 PERROR("close");
4559 }
4560 }
4561
4562 lttng_poll_clean(&events);
4563 clean_command_ctx(&cmd_ctx);
4564
4565 error_listen:
4566 error_create_poll:
4567 unlink(config.client_unix_sock_path.value);
4568 if (client_sock >= 0) {
4569 ret = close(client_sock);
4570 if (ret) {
4571 PERROR("close");
4572 }
4573 }
4574
4575 if (err) {
4576 health_error();
4577 ERR("Health error occurred in %s", __func__);
4578 }
4579
4580 health_unregister(health_sessiond);
4581
4582 DBG("Client thread dying");
4583
4584 rcu_unregister_thread();
4585
4586 /*
4587 * Since we are creating the consumer threads, we own them, so we need
4588 * to join them before our thread exits.
4589 */
4590 ret = join_consumer_thread(&kconsumer_data);
4591 if (ret) {
4592 errno = ret;
4593 PERROR("join_consumer");
4594 }
4595
4596 ret = join_consumer_thread(&ustconsumer32_data);
4597 if (ret) {
4598 errno = ret;
4599 PERROR("join_consumer ust32");
4600 }
4601
4602 ret = join_consumer_thread(&ustconsumer64_data);
4603 if (ret) {
4604 errno = ret;
4605 PERROR("join_consumer ust64");
4606 }
4607 return NULL;
4608 }
4609
4610 static int string_match(const char *str1, const char *str2)
4611 {
4612 return (str1 && str2) && !strcmp(str1, str2);
4613 }
4614
4615 /*
4616 * Take an option from the getopt output and set it in the right variable to be
4617 * used later.
4618 *
4619 * Return 0 on success else a negative value.
4620 */
4621 static int set_option(int opt, const char *arg, const char *optname)
4622 {
4623 int ret = 0;
4624
4625 if (string_match(optname, "client-sock") || opt == 'c') {
4626 if (!arg || *arg == '\0') {
4627 ret = -EINVAL;
4628 goto end;
4629 }
4630 if (lttng_is_setuid_setgid()) {
4631 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4632 "-c, --client-sock");
4633 } else {
4634 config_string_set(&config.client_unix_sock_path,
4635 strdup(arg));
4636 if (!config.client_unix_sock_path.value) {
4637 ret = -ENOMEM;
4638 PERROR("strdup");
4639 }
4640 }
4641 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4642 if (!arg || *arg == '\0') {
4643 ret = -EINVAL;
4644 goto end;
4645 }
4646 if (lttng_is_setuid_setgid()) {
4647 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4648 "-a, --apps-sock");
4649 } else {
4650 config_string_set(&config.apps_unix_sock_path,
4651 strdup(arg));
4652 if (!config.apps_unix_sock_path.value) {
4653 ret = -ENOMEM;
4654 PERROR("strdup");
4655 }
4656 }
4657 } else if (string_match(optname, "daemonize") || opt == 'd') {
4658 config.daemonize = true;
4659 } else if (string_match(optname, "background") || opt == 'b') {
4660 config.background = true;
4661 } else if (string_match(optname, "group") || opt == 'g') {
4662 if (!arg || *arg == '\0') {
4663 ret = -EINVAL;
4664 goto end;
4665 }
4666 if (lttng_is_setuid_setgid()) {
4667 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4668 "-g, --group");
4669 } else {
4670 config_string_set(&config.tracing_group_name,
4671 strdup(arg));
4672 if (!config.tracing_group_name.value) {
4673 ret = -ENOMEM;
4674 PERROR("strdup");
4675 }
4676 }
4677 } else if (string_match(optname, "help") || opt == 'h') {
4678 ret = utils_show_help(8, "lttng-sessiond", help_msg);
4679 if (ret) {
4680 ERR("Cannot show --help for `lttng-sessiond`");
4681 perror("exec");
4682 }
4683 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4684 } else if (string_match(optname, "version") || opt == 'V') {
4685 fprintf(stdout, "%s\n", VERSION);
4686 exit(EXIT_SUCCESS);
4687 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4688 config.sig_parent = true;
4689 } else if (string_match(optname, "kconsumerd-err-sock")) {
4690 if (!arg || *arg == '\0') {
4691 ret = -EINVAL;
4692 goto end;
4693 }
4694 if (lttng_is_setuid_setgid()) {
4695 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4696 "--kconsumerd-err-sock");
4697 } else {
4698 config_string_set(&config.kconsumerd_err_unix_sock_path,
4699 strdup(arg));
4700 if (!config.kconsumerd_err_unix_sock_path.value) {
4701 ret = -ENOMEM;
4702 PERROR("strdup");
4703 }
4704 }
4705 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4706 if (!arg || *arg == '\0') {
4707 ret = -EINVAL;
4708 goto end;
4709 }
4710 if (lttng_is_setuid_setgid()) {
4711 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4712 "--kconsumerd-cmd-sock");
4713 } else {
4714 config_string_set(&config.kconsumerd_cmd_unix_sock_path,
4715 strdup(arg));
4716 if (!config.kconsumerd_cmd_unix_sock_path.value) {
4717 ret = -ENOMEM;
4718 PERROR("strdup");
4719 }
4720 }
4721 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4722 if (!arg || *arg == '\0') {
4723 ret = -EINVAL;
4724 goto end;
4725 }
4726 if (lttng_is_setuid_setgid()) {
4727 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4728 "--ustconsumerd64-err-sock");
4729 } else {
4730 config_string_set(&config.consumerd64_err_unix_sock_path,
4731 strdup(arg));
4732 if (!config.consumerd64_err_unix_sock_path.value) {
4733 ret = -ENOMEM;
4734 PERROR("strdup");
4735 }
4736 }
4737 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4738 if (!arg || *arg == '\0') {
4739 ret = -EINVAL;
4740 goto end;
4741 }
4742 if (lttng_is_setuid_setgid()) {
4743 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4744 "--ustconsumerd64-cmd-sock");
4745 } else {
4746 config_string_set(&config.consumerd64_cmd_unix_sock_path,
4747 strdup(arg));
4748 if (!config.consumerd64_cmd_unix_sock_path.value) {
4749 ret = -ENOMEM;
4750 PERROR("strdup");
4751 }
4752 }
4753 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4754 if (!arg || *arg == '\0') {
4755 ret = -EINVAL;
4756 goto end;
4757 }
4758 if (lttng_is_setuid_setgid()) {
4759 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4760 "--ustconsumerd32-err-sock");
4761 } else {
4762 config_string_set(&config.consumerd32_err_unix_sock_path,
4763 strdup(arg));
4764 if (!config.consumerd32_err_unix_sock_path.value) {
4765 ret = -ENOMEM;
4766 PERROR("strdup");
4767 }
4768 }
4769 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4770 if (!arg || *arg == '\0') {
4771 ret = -EINVAL;
4772 goto end;
4773 }
4774 if (lttng_is_setuid_setgid()) {
4775 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4776 "--ustconsumerd32-cmd-sock");
4777 } else {
4778 config_string_set(&config.consumerd32_cmd_unix_sock_path,
4779 strdup(arg));
4780 if (!config.consumerd32_cmd_unix_sock_path.value) {
4781 ret = -ENOMEM;
4782 PERROR("strdup");
4783 }
4784 }
4785 } else if (string_match(optname, "no-kernel")) {
4786 config.no_kernel = true;
4787 } else if (string_match(optname, "quiet") || opt == 'q') {
4788 config.quiet = true;
4789 } else if (string_match(optname, "verbose") || opt == 'v') {
4790 /* Verbose level can increase using multiple -v */
4791 if (arg) {
4792 /* Value obtained from config file */
4793 config.verbose = config_parse_value(arg);
4794 } else {
4795 /* -v used on command line */
4796 config.verbose++;
4797 }
4798 /* Clamp value to [0, 3] */
4799 config.verbose = config.verbose < 0 ? 0 :
4800 (config.verbose <= 3 ? config.verbose : 3);
4801 } else if (string_match(optname, "verbose-consumer")) {
4802 if (arg) {
4803 config.verbose_consumer = config_parse_value(arg);
4804 } else {
4805 config.verbose_consumer++;
4806 }
4807 } else if (string_match(optname, "consumerd32-path")) {
4808 if (!arg || *arg == '\0') {
4809 ret = -EINVAL;
4810 goto end;
4811 }
4812 if (lttng_is_setuid_setgid()) {
4813 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4814 "--consumerd32-path");
4815 } else {
4816 config_string_set(&config.consumerd32_bin_path,
4817 strdup(arg));
4818 if (!config.consumerd32_bin_path.value) {
4819 PERROR("strdup");
4820 ret = -ENOMEM;
4821 }
4822 }
4823 } else if (string_match(optname, "consumerd32-libdir")) {
4824 if (!arg || *arg == '\0') {
4825 ret = -EINVAL;
4826 goto end;
4827 }
4828 if (lttng_is_setuid_setgid()) {
4829 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4830 "--consumerd32-libdir");
4831 } else {
4832 config_string_set(&config.consumerd32_lib_dir,
4833 strdup(arg));
4834 if (!config.consumerd32_lib_dir.value) {
4835 PERROR("strdup");
4836 ret = -ENOMEM;
4837 }
4838 }
4839 } else if (string_match(optname, "consumerd64-path")) {
4840 if (!arg || *arg == '\0') {
4841 ret = -EINVAL;
4842 goto end;
4843 }
4844 if (lttng_is_setuid_setgid()) {
4845 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4846 "--consumerd64-path");
4847 } else {
4848 config_string_set(&config.consumerd64_bin_path,
4849 strdup(arg));
4850 if (!config.consumerd64_bin_path.value) {
4851 PERROR("strdup");
4852 ret = -ENOMEM;
4853 }
4854 }
4855 } else if (string_match(optname, "consumerd64-libdir")) {
4856 if (!arg || *arg == '\0') {
4857 ret = -EINVAL;
4858 goto end;
4859 }
4860 if (lttng_is_setuid_setgid()) {
4861 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4862 "--consumerd64-libdir");
4863 } else {
4864 config_string_set(&config.consumerd64_lib_dir,
4865 strdup(arg));
4866 if (!config.consumerd64_lib_dir.value) {
4867 PERROR("strdup");
4868 ret = -ENOMEM;
4869 }
4870 }
4871 } else if (string_match(optname, "pidfile") || opt == 'p') {
4872 if (!arg || *arg == '\0') {
4873 ret = -EINVAL;
4874 goto end;
4875 }
4876 if (lttng_is_setuid_setgid()) {
4877 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4878 "-p, --pidfile");
4879 } else {
4880 config_string_set(&config.pid_file_path, strdup(arg));
4881 if (!config.pid_file_path.value) {
4882 PERROR("strdup");
4883 ret = -ENOMEM;
4884 }
4885 }
4886 } else if (string_match(optname, "agent-tcp-port")) {
4887 if (!arg || *arg == '\0') {
4888 ret = -EINVAL;
4889 goto end;
4890 }
4891 if (lttng_is_setuid_setgid()) {
4892 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4893 "--agent-tcp-port");
4894 } else {
4895 unsigned long v;
4896
4897 errno = 0;
4898 v = strtoul(arg, NULL, 0);
4899 if (errno != 0 || !isdigit(arg[0])) {
4900 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4901 return -1;
4902 }
4903 if (v == 0 || v >= 65535) {
4904 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4905 return -1;
4906 }
4907 config.agent_tcp_port.begin = config.agent_tcp_port.end = (int) v;
4908 DBG3("Agent TCP port set to non default: %i", (int) v);
4909 }
4910 } else if (string_match(optname, "load") || opt == 'l') {
4911 if (!arg || *arg == '\0') {
4912 ret = -EINVAL;
4913 goto end;
4914 }
4915 if (lttng_is_setuid_setgid()) {
4916 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4917 "-l, --load");
4918 } else {
4919 config_string_set(&config.load_session_path, strdup(arg));
4920 if (!config.load_session_path.value) {
4921 PERROR("strdup");
4922 ret = -ENOMEM;
4923 }
4924 }
4925 } else if (string_match(optname, "kmod-probes")) {
4926 if (!arg || *arg == '\0') {
4927 ret = -EINVAL;
4928 goto end;
4929 }
4930 if (lttng_is_setuid_setgid()) {
4931 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4932 "--kmod-probes");
4933 } else {
4934 config_string_set(&config.kmod_probes_list, strdup(arg));
4935 if (!config.kmod_probes_list.value) {
4936 PERROR("strdup");
4937 ret = -ENOMEM;
4938 }
4939 }
4940 } else if (string_match(optname, "extra-kmod-probes")) {
4941 if (!arg || *arg == '\0') {
4942 ret = -EINVAL;
4943 goto end;
4944 }
4945 if (lttng_is_setuid_setgid()) {
4946 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4947 "--extra-kmod-probes");
4948 } else {
4949 config_string_set(&config.kmod_extra_probes_list,
4950 strdup(arg));
4951 if (!config.kmod_extra_probes_list.value) {
4952 PERROR("strdup");
4953 ret = -ENOMEM;
4954 }
4955 }
4956 } else if (string_match(optname, "config") || opt == 'f') {
4957 /* This is handled in set_options() thus silent skip. */
4958 goto end;
4959 } else {
4960 /* Unknown option or other error.
4961 * Error is printed by getopt, just return */
4962 ret = -1;
4963 }
4964
4965 end:
4966 if (ret == -EINVAL) {
4967 const char *opt_name = "unknown";
4968 int i;
4969
4970 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4971 i++) {
4972 if (opt == long_options[i].val) {
4973 opt_name = long_options[i].name;
4974 break;
4975 }
4976 }
4977
4978 WARN("Invalid argument provided for option \"%s\", using default value.",
4979 opt_name);
4980 }
4981
4982 return ret;
4983 }
4984
4985 /*
4986 * config_entry_handler_cb used to handle options read from a config file.
4987 * See config_entry_handler_cb comment in common/config/session-config.h for the
4988 * return value conventions.
4989 */
4990 static int config_entry_handler(const struct config_entry *entry, void *unused)
4991 {
4992 int ret = 0, i;
4993
4994 if (!entry || !entry->name || !entry->value) {
4995 ret = -EINVAL;
4996 goto end;
4997 }
4998
4999 /* Check if the option is to be ignored */
5000 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
5001 if (!strcmp(entry->name, config_ignore_options[i])) {
5002 goto end;
5003 }
5004 }
5005
5006 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
5007 i++) {
5008
5009 /* Ignore if not fully matched. */
5010 if (strcmp(entry->name, long_options[i].name)) {
5011 continue;
5012 }
5013
5014 /*
5015 * If the option takes no argument on the command line, we have to
5016 * check if the value is "true". We support non-zero numeric values,
5017 * true, on and yes.
5018 */
5019 if (!long_options[i].has_arg) {
5020 ret = config_parse_value(entry->value);
5021 if (ret <= 0) {
5022 if (ret) {
5023 WARN("Invalid configuration value \"%s\" for option %s",
5024 entry->value, entry->name);
5025 }
5026 /* False, skip boolean config option. */
5027 goto end;
5028 }
5029 }
5030
5031 ret = set_option(long_options[i].val, entry->value, entry->name);
5032 goto end;
5033 }
5034
5035 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5036
5037 end:
5038 return ret;
5039 }
5040
5041 /*
5042 * daemon configuration loading and argument parsing
5043 */
5044 static int set_options(int argc, char **argv)
5045 {
5046 int ret = 0, c = 0, option_index = 0;
5047 int orig_optopt = optopt, orig_optind = optind;
5048 char *optstring;
5049 const char *config_path = NULL;
5050
5051 optstring = utils_generate_optstring(long_options,
5052 sizeof(long_options) / sizeof(struct option));
5053 if (!optstring) {
5054 ret = -ENOMEM;
5055 goto end;
5056 }
5057
5058 /* Check for the --config option */
5059 while ((c = getopt_long(argc, argv, optstring, long_options,
5060 &option_index)) != -1) {
5061 if (c == '?') {
5062 ret = -EINVAL;
5063 goto end;
5064 } else if (c != 'f') {
5065 /* if not equal to --config option. */
5066 continue;
5067 }
5068
5069 if (lttng_is_setuid_setgid()) {
5070 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5071 "-f, --config");
5072 } else {
5073 config_path = utils_expand_path(optarg);
5074 if (!config_path) {
5075 ERR("Failed to resolve path: %s", optarg);
5076 }
5077 }
5078 }
5079
5080 ret = config_get_section_entries(config_path, config_section_name,
5081 config_entry_handler, NULL);
5082 if (ret) {
5083 if (ret > 0) {
5084 ERR("Invalid configuration option at line %i", ret);
5085 ret = -1;
5086 }
5087 goto end;
5088 }
5089
5090 /* Reset getopt's global state */
5091 optopt = orig_optopt;
5092 optind = orig_optind;
5093 while (1) {
5094 option_index = -1;
5095 /*
5096 * getopt_long() will not set option_index if it encounters a
5097 * short option.
5098 */
5099 c = getopt_long(argc, argv, optstring, long_options,
5100 &option_index);
5101 if (c == -1) {
5102 break;
5103 }
5104
5105 /*
5106 * Pass NULL as the long option name if popt left the index
5107 * unset.
5108 */
5109 ret = set_option(c, optarg,
5110 option_index < 0 ? NULL :
5111 long_options[option_index].name);
5112 if (ret < 0) {
5113 break;
5114 }
5115 }
5116
5117 end:
5118 free(optstring);
5119 return ret;
5120 }
5121
5122 /*
5123 * Creates the two needed socket by the daemon.
5124 * apps_sock - The communication socket for all UST apps.
5125 * client_sock - The communication of the cli tool (lttng).
5126 */
5127 static int init_daemon_socket(void)
5128 {
5129 int ret = 0;
5130 mode_t old_umask;
5131
5132 old_umask = umask(0);
5133
5134 /* Create client tool unix socket */
5135 client_sock = lttcomm_create_unix_sock(config.client_unix_sock_path.value);
5136 if (client_sock < 0) {
5137 ERR("Create unix sock failed: %s", config.client_unix_sock_path.value);
5138 ret = -1;
5139 goto end;
5140 }
5141
5142 /* Set the cloexec flag */
5143 ret = utils_set_fd_cloexec(client_sock);
5144 if (ret < 0) {
5145 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5146 "Continuing but note that the consumer daemon will have a "
5147 "reference to this socket on exec()", client_sock);
5148 }
5149
5150 /* File permission MUST be 660 */
5151 ret = chmod(config.client_unix_sock_path.value, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5152 if (ret < 0) {
5153 ERR("Set file permissions failed: %s", config.client_unix_sock_path.value);
5154 PERROR("chmod");
5155 goto end;
5156 }
5157
5158 /* Create the application unix socket */
5159 apps_sock = lttcomm_create_unix_sock(config.apps_unix_sock_path.value);
5160 if (apps_sock < 0) {
5161 ERR("Create unix sock failed: %s", config.apps_unix_sock_path.value);
5162 ret = -1;
5163 goto end;
5164 }
5165
5166 /* Set the cloexec flag */
5167 ret = utils_set_fd_cloexec(apps_sock);
5168 if (ret < 0) {
5169 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5170 "Continuing but note that the consumer daemon will have a "
5171 "reference to this socket on exec()", apps_sock);
5172 }
5173
5174 /* File permission MUST be 666 */
5175 ret = chmod(config.apps_unix_sock_path.value,
5176 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5177 if (ret < 0) {
5178 ERR("Set file permissions failed: %s", config.apps_unix_sock_path.value);
5179 PERROR("chmod");
5180 goto end;
5181 }
5182
5183 DBG3("Session daemon client socket %d and application socket %d created",
5184 client_sock, apps_sock);
5185
5186 end:
5187 umask(old_umask);
5188 return ret;
5189 }
5190
5191 /*
5192 * Create lockfile using the rundir and return its fd.
5193 */
5194 static int create_lockfile(void)
5195 {
5196 return utils_create_lock_file(config.lock_file_path.value);
5197 }
5198
5199 /*
5200 * Check if the global socket is available, and if a daemon is answering at the
5201 * other side. If yes, error is returned.
5202 *
5203 * Also attempts to create and hold the lock file.
5204 */
5205 static int check_existing_daemon(void)
5206 {
5207 int ret = 0;
5208
5209 /* Is there anybody out there ? */
5210 if (lttng_session_daemon_alive()) {
5211 ret = -EEXIST;
5212 goto end;
5213 }
5214
5215 lockfile_fd = create_lockfile();
5216 if (lockfile_fd < 0) {
5217 ret = -EEXIST;
5218 goto end;
5219 }
5220 end:
5221 return ret;
5222 }
5223
5224 static void sessiond_cleanup_lock_file(void)
5225 {
5226 int ret;
5227
5228 /*
5229 * Cleanup lock file by deleting it and finaly closing it which will
5230 * release the file system lock.
5231 */
5232 if (lockfile_fd >= 0) {
5233 ret = remove(config.lock_file_path.value);
5234 if (ret < 0) {
5235 PERROR("remove lock file");
5236 }
5237 ret = close(lockfile_fd);
5238 if (ret < 0) {
5239 PERROR("close lock file");
5240 }
5241 }
5242 }
5243
5244 /*
5245 * Set the tracing group gid onto the client socket.
5246 *
5247 * Race window between mkdir and chown is OK because we are going from more
5248 * permissive (root.root) to less permissive (root.tracing).
5249 */
5250 static int set_permissions(char *rundir)
5251 {
5252 int ret;
5253 gid_t gid;
5254
5255 ret = utils_get_group_id(config.tracing_group_name.value, true, &gid);
5256 if (ret) {
5257 /* Default to root group. */
5258 gid = 0;
5259 }
5260
5261 /* Set lttng run dir */
5262 ret = chown(rundir, 0, gid);
5263 if (ret < 0) {
5264 ERR("Unable to set group on %s", rundir);
5265 PERROR("chown");
5266 }
5267
5268 /*
5269 * Ensure all applications and tracing group can search the run
5270 * dir. Allow everyone to read the directory, since it does not
5271 * buy us anything to hide its content.
5272 */
5273 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5274 if (ret < 0) {
5275 ERR("Unable to set permissions on %s", rundir);
5276 PERROR("chmod");
5277 }
5278
5279 /* lttng client socket path */
5280 ret = chown(config.client_unix_sock_path.value, 0, gid);
5281 if (ret < 0) {
5282 ERR("Unable to set group on %s", config.client_unix_sock_path.value);
5283 PERROR("chown");
5284 }
5285
5286 /* kconsumer error socket path */
5287 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5288 if (ret < 0) {
5289 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5290 PERROR("chown");
5291 }
5292
5293 /* 64-bit ustconsumer error socket path */
5294 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5295 if (ret < 0) {
5296 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5297 PERROR("chown");
5298 }
5299
5300 /* 32-bit ustconsumer compat32 error socket path */
5301 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5302 if (ret < 0) {
5303 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5304 PERROR("chown");
5305 }
5306
5307 DBG("All permissions are set");
5308
5309 return ret;
5310 }
5311
5312 /*
5313 * Create the lttng run directory needed for all global sockets and pipe.
5314 */
5315 static int create_lttng_rundir(void)
5316 {
5317 int ret;
5318
5319 DBG3("Creating LTTng run directory: %s", config.rundir.value);
5320
5321 ret = mkdir(config.rundir.value, S_IRWXU);
5322 if (ret < 0) {
5323 if (errno != EEXIST) {
5324 ERR("Unable to create %s", config.rundir.value);
5325 goto error;
5326 } else {
5327 ret = 0;
5328 }
5329 }
5330
5331 error:
5332 return ret;
5333 }
5334
5335 /*
5336 * Setup sockets and directory needed by the consumerds' communication with the
5337 * session daemon.
5338 */
5339 static int set_consumer_sockets(struct consumer_data *consumer_data)
5340 {
5341 int ret;
5342 char *path = NULL;
5343
5344 switch (consumer_data->type) {
5345 case LTTNG_CONSUMER_KERNEL:
5346 path = config.kconsumerd_path.value;
5347 break;
5348 case LTTNG_CONSUMER64_UST:
5349 path = config.consumerd64_path.value;
5350 break;
5351 case LTTNG_CONSUMER32_UST:
5352 path = config.consumerd32_path.value;
5353 break;
5354 default:
5355 ERR("Consumer type unknown");
5356 ret = -EINVAL;
5357 goto error;
5358 }
5359 assert(path);
5360
5361 DBG2("Creating consumer directory: %s", path);
5362
5363 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5364 if (ret < 0 && errno != EEXIST) {
5365 PERROR("mkdir");
5366 ERR("Failed to create %s", path);
5367 goto error;
5368 }
5369 if (is_root) {
5370 gid_t gid;
5371
5372 ret = utils_get_group_id(config.tracing_group_name.value, true,
5373 &gid);
5374 if (ret) {
5375 /* Default to root group. */
5376 gid = 0;
5377 }
5378
5379 ret = chown(path, 0, gid);
5380 if (ret < 0) {
5381 ERR("Unable to set group on %s", path);
5382 PERROR("chown");
5383 goto error;
5384 }
5385 }
5386
5387 /* Create the consumerd error unix socket */
5388 consumer_data->err_sock =
5389 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5390 if (consumer_data->err_sock < 0) {
5391 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5392 ret = -1;
5393 goto error;
5394 }
5395
5396 /*
5397 * Set the CLOEXEC flag. Return code is useless because either way, the
5398 * show must go on.
5399 */
5400 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5401 if (ret < 0) {
5402 PERROR("utils_set_fd_cloexec");
5403 /* continue anyway */
5404 }
5405
5406 /* File permission MUST be 660 */
5407 ret = chmod(consumer_data->err_unix_sock_path,
5408 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5409 if (ret < 0) {
5410 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5411 PERROR("chmod");
5412 goto error;
5413 }
5414
5415 error:
5416 return ret;
5417 }
5418
5419 /*
5420 * Signal handler for the daemon
5421 *
5422 * Simply stop all worker threads, leaving main() return gracefully after
5423 * joining all threads and calling cleanup().
5424 */
5425 static void sighandler(int sig)
5426 {
5427 switch (sig) {
5428 case SIGINT:
5429 DBG("SIGINT caught");
5430 stop_threads();
5431 break;
5432 case SIGTERM:
5433 DBG("SIGTERM caught");
5434 stop_threads();
5435 break;
5436 case SIGUSR1:
5437 CMM_STORE_SHARED(recv_child_signal, 1);
5438 break;
5439 default:
5440 break;
5441 }
5442 }
5443
5444 /*
5445 * Setup signal handler for :
5446 * SIGINT, SIGTERM, SIGPIPE
5447 */
5448 static int set_signal_handler(void)
5449 {
5450 int ret = 0;
5451 struct sigaction sa;
5452 sigset_t sigset;
5453
5454 if ((ret = sigemptyset(&sigset)) < 0) {
5455 PERROR("sigemptyset");
5456 return ret;
5457 }
5458
5459 sa.sa_mask = sigset;
5460 sa.sa_flags = 0;
5461
5462 sa.sa_handler = sighandler;
5463 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5464 PERROR("sigaction");
5465 return ret;
5466 }
5467
5468 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5469 PERROR("sigaction");
5470 return ret;
5471 }
5472
5473 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5474 PERROR("sigaction");
5475 return ret;
5476 }
5477
5478 sa.sa_handler = SIG_IGN;
5479 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5480 PERROR("sigaction");
5481 return ret;
5482 }
5483
5484 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5485
5486 return ret;
5487 }
5488
5489 /*
5490 * Set open files limit to unlimited. This daemon can open a large number of
5491 * file descriptors in order to consume multiple kernel traces.
5492 */
5493 static void set_ulimit(void)
5494 {
5495 int ret;
5496 struct rlimit lim;
5497
5498 /* The kernel does not allow an infinite limit for open files */
5499 lim.rlim_cur = 65535;
5500 lim.rlim_max = 65535;
5501
5502 ret = setrlimit(RLIMIT_NOFILE, &lim);
5503 if (ret < 0) {
5504 PERROR("failed to set open files limit");
5505 }
5506 }
5507
5508 static int write_pidfile(void)
5509 {
5510 return utils_create_pid_file(getpid(), config.pid_file_path.value);
5511 }
5512
5513 static int set_clock_plugin_env(void)
5514 {
5515 int ret = 0;
5516 char *env_value = NULL;
5517
5518 if (!config.lttng_ust_clock_plugin.value) {
5519 goto end;
5520 }
5521
5522 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
5523 config.lttng_ust_clock_plugin.value);
5524 if (ret < 0) {
5525 PERROR("asprintf");
5526 goto end;
5527 }
5528
5529 ret = putenv(env_value);
5530 if (ret) {
5531 free(env_value);
5532 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5533 goto end;
5534 }
5535
5536 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5537 config.lttng_ust_clock_plugin.value);
5538 end:
5539 return ret;
5540 }
5541
5542 /*
5543 * main
5544 */
5545 int main(int argc, char **argv)
5546 {
5547 int ret = 0, retval = 0;
5548 void *status;
5549 const char *env_app_timeout;
5550 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5551 *ust64_channel_monitor_pipe = NULL,
5552 *kernel_channel_monitor_pipe = NULL;
5553 bool notification_thread_running = false;
5554
5555 init_kernel_workarounds();
5556
5557 rcu_register_thread();
5558
5559 if (set_signal_handler()) {
5560 retval = -1;
5561 goto exit_set_signal_handler;
5562 }
5563
5564 page_size = sysconf(_SC_PAGESIZE);
5565 if (page_size < 0) {
5566 PERROR("sysconf _SC_PAGESIZE");
5567 page_size = LONG_MAX;
5568 WARN("Fallback page size to %ld", page_size);
5569 }
5570
5571 ret = sessiond_config_init(&config);
5572 if (ret) {
5573 retval = -1;
5574 goto exit_set_signal_handler;
5575 }
5576
5577 /*
5578 * Init config from environment variables.
5579 * Command line option override env configuration per-doc. Do env first.
5580 */
5581 sessiond_config_apply_env_config(&config);
5582
5583 /*
5584 * Parse arguments and load the daemon configuration file.
5585 *
5586 * We have an exit_options exit path to free memory reserved by
5587 * set_options. This is needed because the rest of sessiond_cleanup()
5588 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5589 * depends on set_options.
5590 */
5591 progname = argv[0];
5592 if (set_options(argc, argv)) {
5593 retval = -1;
5594 goto exit_options;
5595 }
5596
5597 /*
5598 * Resolve all paths received as arguments, configuration option, or
5599 * through environment variable as absolute paths. This is necessary
5600 * since daemonizing causes the sessiond's current working directory
5601 * to '/'.
5602 */
5603 ret = sessiond_config_resolve_paths(&config);
5604 if (ret) {
5605 goto exit_options;
5606 }
5607
5608 /* Apply config. */
5609 lttng_opt_verbose = config.verbose;
5610 lttng_opt_quiet = config.quiet;
5611 kconsumer_data.err_unix_sock_path =
5612 config.kconsumerd_err_unix_sock_path.value;
5613 kconsumer_data.cmd_unix_sock_path =
5614 config.kconsumerd_cmd_unix_sock_path.value;
5615 ustconsumer32_data.err_unix_sock_path =
5616 config.consumerd32_err_unix_sock_path.value;
5617 ustconsumer32_data.cmd_unix_sock_path =
5618 config.consumerd32_cmd_unix_sock_path.value;
5619 ustconsumer64_data.err_unix_sock_path =
5620 config.consumerd64_err_unix_sock_path.value;
5621 ustconsumer64_data.cmd_unix_sock_path =
5622 config.consumerd64_cmd_unix_sock_path.value;
5623 set_clock_plugin_env();
5624
5625 sessiond_config_log(&config);
5626
5627 if (create_lttng_rundir()) {
5628 retval = -1;
5629 goto exit_options;
5630 }
5631
5632 /* Abort launch if a session daemon is already running. */
5633 if (check_existing_daemon()) {
5634 ERR("A session daemon is already running.");
5635 retval = -1;
5636 goto exit_options;
5637 }
5638
5639 /* Daemonize */
5640 if (config.daemonize || config.background) {
5641 int i;
5642
5643 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5644 !config.background);
5645 if (ret < 0) {
5646 retval = -1;
5647 goto exit_options;
5648 }
5649
5650 /*
5651 * We are in the child. Make sure all other file descriptors are
5652 * closed, in case we are called with more opened file
5653 * descriptors than the standard ones and the lock file.
5654 */
5655 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5656 if (i == lockfile_fd) {
5657 continue;
5658 }
5659 (void) close(i);
5660 }
5661 }
5662
5663 if (run_as_create_worker(argv[0]) < 0) {
5664 goto exit_create_run_as_worker_cleanup;
5665 }
5666
5667 /*
5668 * Starting from here, we can create threads. This needs to be after
5669 * lttng_daemonize due to RCU.
5670 */
5671
5672 /*
5673 * Initialize the health check subsystem. This call should set the
5674 * appropriate time values.
5675 */
5676 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5677 if (!health_sessiond) {
5678 PERROR("health_app_create error");
5679 retval = -1;
5680 goto exit_health_sessiond_cleanup;
5681 }
5682
5683 /* Create thread to clean up RCU hash tables */
5684 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
5685 retval = -1;
5686 goto exit_ht_cleanup;
5687 }
5688
5689 /* Create thread quit pipe */
5690 if (init_thread_quit_pipe()) {
5691 retval = -1;
5692 goto exit_init_data;
5693 }
5694
5695 /* Check if daemon is UID = 0 */
5696 is_root = !getuid();
5697 if (is_root) {
5698 /* Create global run dir with root access */
5699
5700 kernel_channel_monitor_pipe = lttng_pipe_open(0);
5701 if (!kernel_channel_monitor_pipe) {
5702 ERR("Failed to create kernel consumer channel monitor pipe");
5703 retval = -1;
5704 goto exit_init_data;
5705 }
5706 kconsumer_data.channel_monitor_pipe =
5707 lttng_pipe_release_writefd(
5708 kernel_channel_monitor_pipe);
5709 if (kconsumer_data.channel_monitor_pipe < 0) {
5710 retval = -1;
5711 goto exit_init_data;
5712 }
5713 }
5714
5715 /* Set consumer initial state */
5716 kernel_consumerd_state = CONSUMER_STOPPED;
5717 ust_consumerd_state = CONSUMER_STOPPED;
5718
5719 ust32_channel_monitor_pipe = lttng_pipe_open(0);
5720 if (!ust32_channel_monitor_pipe) {
5721 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5722 retval = -1;
5723 goto exit_init_data;
5724 }
5725 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5726 ust32_channel_monitor_pipe);
5727 if (ustconsumer32_data.channel_monitor_pipe < 0) {
5728 retval = -1;
5729 goto exit_init_data;
5730 }
5731
5732 ust64_channel_monitor_pipe = lttng_pipe_open(0);
5733 if (!ust64_channel_monitor_pipe) {
5734 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5735 retval = -1;
5736 goto exit_init_data;
5737 }
5738 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5739 ust64_channel_monitor_pipe);
5740 if (ustconsumer64_data.channel_monitor_pipe < 0) {
5741 retval = -1;
5742 goto exit_init_data;
5743 }
5744
5745 /*
5746 * Init UST app hash table. Alloc hash table before this point since
5747 * cleanup() can get called after that point.
5748 */
5749 if (ust_app_ht_alloc()) {
5750 ERR("Failed to allocate UST app hash table");
5751 retval = -1;
5752 goto exit_init_data;
5753 }
5754
5755 /*
5756 * Initialize agent app hash table. We allocate the hash table here
5757 * since cleanup() can get called after this point.
5758 */
5759 if (agent_app_ht_alloc()) {
5760 ERR("Failed to allocate Agent app hash table");
5761 retval = -1;
5762 goto exit_init_data;
5763 }
5764
5765 /*
5766 * These actions must be executed as root. We do that *after* setting up
5767 * the sockets path because we MUST make the check for another daemon using
5768 * those paths *before* trying to set the kernel consumer sockets and init
5769 * kernel tracer.
5770 */
5771 if (is_root) {
5772 if (set_consumer_sockets(&kconsumer_data)) {
5773 retval = -1;
5774 goto exit_init_data;
5775 }
5776
5777 /* Setup kernel tracer */
5778 if (!config.no_kernel) {
5779 init_kernel_tracer();
5780 }
5781
5782 /* Set ulimit for open files */
5783 set_ulimit();
5784 }
5785 /* init lttng_fd tracking must be done after set_ulimit. */
5786 lttng_fd_init();
5787
5788 if (set_consumer_sockets(&ustconsumer64_data)) {
5789 retval = -1;
5790 goto exit_init_data;
5791 }
5792
5793 if (set_consumer_sockets(&ustconsumer32_data)) {
5794 retval = -1;
5795 goto exit_init_data;
5796 }
5797
5798 /* Setup the needed unix socket */
5799 if (init_daemon_socket()) {
5800 retval = -1;
5801 goto exit_init_data;
5802 }
5803
5804 /* Set credentials to socket */
5805 if (is_root && set_permissions(config.rundir.value)) {
5806 retval = -1;
5807 goto exit_init_data;
5808 }
5809
5810 /* Get parent pid if -S, --sig-parent is specified. */
5811 if (config.sig_parent) {
5812 ppid = getppid();
5813 }
5814
5815 /* Setup the kernel pipe for waking up the kernel thread */
5816 if (is_root && !config.no_kernel) {
5817 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5818 retval = -1;
5819 goto exit_init_data;
5820 }
5821 }
5822
5823 /* Setup the thread apps communication pipe. */
5824 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5825 retval = -1;
5826 goto exit_init_data;
5827 }
5828
5829 /* Setup the thread apps notify communication pipe. */
5830 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5831 retval = -1;
5832 goto exit_init_data;
5833 }
5834
5835 /* Initialize global buffer per UID and PID registry. */
5836 buffer_reg_init_uid_registry();
5837 buffer_reg_init_pid_registry();
5838
5839 /* Init UST command queue. */
5840 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5841
5842 /*
5843 * Get session list pointer. This pointer MUST NOT be free'd. This list
5844 * is statically declared in session.c
5845 */
5846 session_list_ptr = session_get_list();
5847
5848 cmd_init();
5849
5850 /* Check for the application socket timeout env variable. */
5851 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5852 if (env_app_timeout) {
5853 config.app_socket_timeout = atoi(env_app_timeout);
5854 } else {
5855 config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5856 }
5857
5858 ret = write_pidfile();
5859 if (ret) {
5860 ERR("Error in write_pidfile");
5861 retval = -1;
5862 goto exit_init_data;
5863 }
5864
5865 /* Initialize communication library */
5866 lttcomm_init();
5867 /* Initialize TCP timeout values */
5868 lttcomm_inet_init();
5869
5870 if (load_session_init_data(&load_info) < 0) {
5871 retval = -1;
5872 goto exit_init_data;
5873 }
5874 load_info->path = config.load_session_path.value;
5875
5876 /* Create health-check thread. */
5877 ret = pthread_create(&health_thread, default_pthread_attr(),
5878 thread_manage_health, (void *) NULL);
5879 if (ret) {
5880 errno = ret;
5881 PERROR("pthread_create health");
5882 retval = -1;
5883 goto exit_health;
5884 }
5885
5886 /* notification_thread_data acquires the pipes' read side. */
5887 notification_thread_handle = notification_thread_handle_create(
5888 ust32_channel_monitor_pipe,
5889 ust64_channel_monitor_pipe,
5890 kernel_channel_monitor_pipe);
5891 if (!notification_thread_handle) {
5892 retval = -1;
5893 ERR("Failed to create notification thread shared data");
5894 stop_threads();
5895 goto exit_notification;
5896 }
5897
5898 /* Create notification thread. */
5899 ret = pthread_create(&notification_thread, default_pthread_attr(),
5900 thread_notification, notification_thread_handle);
5901 if (ret) {
5902 errno = ret;
5903 PERROR("pthread_create notification");
5904 retval = -1;
5905 stop_threads();
5906 goto exit_notification;
5907 }
5908 notification_thread_running = true;
5909
5910 /* Create thread to manage the client socket */
5911 ret = pthread_create(&client_thread, default_pthread_attr(),
5912 thread_manage_clients, (void *) NULL);
5913 if (ret) {
5914 errno = ret;
5915 PERROR("pthread_create clients");
5916 retval = -1;
5917 stop_threads();
5918 goto exit_client;
5919 }
5920
5921 /* Create thread to dispatch registration */
5922 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
5923 thread_dispatch_ust_registration, (void *) NULL);
5924 if (ret) {
5925 errno = ret;
5926 PERROR("pthread_create dispatch");
5927 retval = -1;
5928 stop_threads();
5929 goto exit_dispatch;
5930 }
5931
5932 /* Create thread to manage application registration. */
5933 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
5934 thread_registration_apps, (void *) NULL);
5935 if (ret) {
5936 errno = ret;
5937 PERROR("pthread_create registration");
5938 retval = -1;
5939 stop_threads();
5940 goto exit_reg_apps;
5941 }
5942
5943 /* Create thread to manage application socket */
5944 ret = pthread_create(&apps_thread, default_pthread_attr(),
5945 thread_manage_apps, (void *) NULL);
5946 if (ret) {
5947 errno = ret;
5948 PERROR("pthread_create apps");
5949 retval = -1;
5950 stop_threads();
5951 goto exit_apps;
5952 }
5953
5954 /* Create thread to manage application notify socket */
5955 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
5956 ust_thread_manage_notify, (void *) NULL);
5957 if (ret) {
5958 errno = ret;
5959 PERROR("pthread_create notify");
5960 retval = -1;
5961 stop_threads();
5962 goto exit_apps_notify;
5963 }
5964
5965 /* Create agent registration thread. */
5966 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
5967 agent_thread_manage_registration, (void *) NULL);
5968 if (ret) {
5969 errno = ret;
5970 PERROR("pthread_create agent");
5971 retval = -1;
5972 stop_threads();
5973 goto exit_agent_reg;
5974 }
5975
5976 /* Don't start this thread if kernel tracing is not requested nor root */
5977 if (is_root && !config.no_kernel) {
5978 /* Create kernel thread to manage kernel event */
5979 ret = pthread_create(&kernel_thread, default_pthread_attr(),
5980 thread_manage_kernel, (void *) NULL);
5981 if (ret) {
5982 errno = ret;
5983 PERROR("pthread_create kernel");
5984 retval = -1;
5985 stop_threads();
5986 goto exit_kernel;
5987 }
5988 }
5989
5990 /* Create session loading thread. */
5991 ret = pthread_create(&load_session_thread, default_pthread_attr(),
5992 thread_load_session, load_info);
5993 if (ret) {
5994 errno = ret;
5995 PERROR("pthread_create load_session_thread");
5996 retval = -1;
5997 stop_threads();
5998 goto exit_load_session;
5999 }
6000
6001 /*
6002 * This is where we start awaiting program completion (e.g. through
6003 * signal that asks threads to teardown).
6004 */
6005
6006 ret = pthread_join(load_session_thread, &status);
6007 if (ret) {
6008 errno = ret;
6009 PERROR("pthread_join load_session_thread");
6010 retval = -1;
6011 }
6012 exit_load_session:
6013
6014 if (is_root && !config.no_kernel) {
6015 ret = pthread_join(kernel_thread, &status);
6016 if (ret) {
6017 errno = ret;
6018 PERROR("pthread_join");
6019 retval = -1;
6020 }
6021 }
6022 exit_kernel:
6023
6024 ret = pthread_join(agent_reg_thread, &status);
6025 if (ret) {
6026 errno = ret;
6027 PERROR("pthread_join agent");
6028 retval = -1;
6029 }
6030 exit_agent_reg:
6031
6032 ret = pthread_join(apps_notify_thread, &status);
6033 if (ret) {
6034 errno = ret;
6035 PERROR("pthread_join apps notify");
6036 retval = -1;
6037 }
6038 exit_apps_notify:
6039
6040 ret = pthread_join(apps_thread, &status);
6041 if (ret) {
6042 errno = ret;
6043 PERROR("pthread_join apps");
6044 retval = -1;
6045 }
6046 exit_apps:
6047
6048 ret = pthread_join(reg_apps_thread, &status);
6049 if (ret) {
6050 errno = ret;
6051 PERROR("pthread_join");
6052 retval = -1;
6053 }
6054 exit_reg_apps:
6055
6056 /*
6057 * Join dispatch thread after joining reg_apps_thread to ensure
6058 * we don't leak applications in the queue.
6059 */
6060 ret = pthread_join(dispatch_thread, &status);
6061 if (ret) {
6062 errno = ret;
6063 PERROR("pthread_join");
6064 retval = -1;
6065 }
6066 exit_dispatch:
6067
6068 ret = pthread_join(client_thread, &status);
6069 if (ret) {
6070 errno = ret;
6071 PERROR("pthread_join");
6072 retval = -1;
6073 }
6074
6075 exit_client:
6076 exit_notification:
6077 ret = pthread_join(health_thread, &status);
6078 if (ret) {
6079 errno = ret;
6080 PERROR("pthread_join health thread");
6081 retval = -1;
6082 }
6083
6084 exit_health:
6085 exit_init_data:
6086 /*
6087 * Wait for all pending call_rcu work to complete before tearing
6088 * down data structures. call_rcu worker may be trying to
6089 * perform lookups in those structures.
6090 */
6091 rcu_barrier();
6092 /*
6093 * sessiond_cleanup() is called when no other thread is running, except
6094 * the ht_cleanup thread, which is needed to destroy the hash tables.
6095 */
6096 rcu_thread_online();
6097 sessiond_cleanup();
6098
6099 /*
6100 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6101 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6102 * the queue is empty before shutting down the clean-up thread.
6103 */
6104 rcu_barrier();
6105
6106 /*
6107 * The teardown of the notification system is performed after the
6108 * session daemon's teardown in order to allow it to be notified
6109 * of the active session and channels at the moment of the teardown.
6110 */
6111 if (notification_thread_handle) {
6112 if (notification_thread_running) {
6113 notification_thread_command_quit(
6114 notification_thread_handle);
6115 ret = pthread_join(notification_thread, &status);
6116 if (ret) {
6117 errno = ret;
6118 PERROR("pthread_join notification thread");
6119 retval = -1;
6120 }
6121 }
6122 notification_thread_handle_destroy(notification_thread_handle);
6123 }
6124
6125 rcu_thread_offline();
6126 rcu_unregister_thread();
6127
6128 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6129 if (ret) {
6130 retval = -1;
6131 }
6132 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6133 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6134 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6135 exit_ht_cleanup:
6136
6137 health_app_destroy(health_sessiond);
6138 exit_health_sessiond_cleanup:
6139 exit_create_run_as_worker_cleanup:
6140
6141 exit_options:
6142 sessiond_cleanup_lock_file();
6143 sessiond_cleanup_options();
6144
6145 exit_set_signal_handler:
6146 if (!retval) {
6147 exit(EXIT_SUCCESS);
6148 } else {
6149 exit(EXIT_FAILURE);
6150 }
6151 }
This page took 0.260454 seconds and 4 git commands to generate.