Fix: hash table cleanup call_rcu deadlock
[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 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <inttypes.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <urcu/uatomic.h>
37 #include <unistd.h>
38 #include <config.h>
39
40 #include <common/common.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
47
48 #include "lttng-sessiond.h"
49 #include "buffer-registry.h"
50 #include "channel.h"
51 #include "cmd.h"
52 #include "consumer.h"
53 #include "context.h"
54 #include "event.h"
55 #include "kernel.h"
56 #include "kernel-consumer.h"
57 #include "modprobe.h"
58 #include "shm.h"
59 #include "ust-ctl.h"
60 #include "ust-consumer.h"
61 #include "utils.h"
62 #include "fd-limit.h"
63 #include "health.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
66
67 #define CONSUMERD_FILE "lttng-consumerd"
68
69 /* Const values */
70 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
71
72 const char *progname;
73 const char *opt_tracing_group;
74 static const char *opt_pidfile;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static char *rundir;
82
83 /*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
87 static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
91 .err_sock = -1,
92 .cmd_sock = -1,
93 .metadata_sock.fd = -1,
94 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
95 .lock = PTHREAD_MUTEX_INITIALIZER,
96 .cond = PTHREAD_COND_INITIALIZER,
97 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
98 };
99 static struct consumer_data ustconsumer64_data = {
100 .type = LTTNG_CONSUMER64_UST,
101 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .metadata_sock.fd = -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 ustconsumer32_data = {
112 .type = LTTNG_CONSUMER32_UST,
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .metadata_sock.fd = -1,
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
122 };
123
124 /* Shared between threads */
125 static int dispatch_thread_exit;
126
127 /* Global application Unix socket path */
128 static char apps_unix_sock_path[PATH_MAX];
129 /* Global client Unix socket path */
130 static char client_unix_sock_path[PATH_MAX];
131 /* global wait shm path for UST */
132 static char wait_shm_path[PATH_MAX];
133 /* Global health check unix path */
134 static char health_unix_sock_path[PATH_MAX];
135
136 /* Sockets and FDs */
137 static int client_sock = -1;
138 static int apps_sock = -1;
139 int kernel_tracer_fd = -1;
140 static int kernel_poll_pipe[2] = { -1, -1 };
141
142 /*
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
145 */
146 static int thread_quit_pipe[2] = { -1, -1 };
147
148 /*
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
151 */
152 static int apps_cmd_pipe[2] = { -1, -1 };
153
154 int apps_cmd_notify_pipe[2] = { -1, -1 };
155
156 /* Pthread, Mutexes and Semaphores */
157 static pthread_t apps_thread;
158 static pthread_t apps_notify_thread;
159 static pthread_t reg_apps_thread;
160 static pthread_t client_thread;
161 static pthread_t kernel_thread;
162 static pthread_t dispatch_thread;
163 static pthread_t health_thread;
164 static pthread_t ht_cleanup_thread;
165
166 /*
167 * UST registration command queue. This queue is tied with a futex and uses a N
168 * wakers / 1 waiter implemented and detailed in futex.c/.h
169 *
170 * The thread_manage_apps and thread_dispatch_ust_registration interact with
171 * this queue and the wait/wake scheme.
172 */
173 static struct ust_cmd_queue ust_cmd_queue;
174
175 /*
176 * Pointer initialized before thread creation.
177 *
178 * This points to the tracing session list containing the session count and a
179 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
180 * MUST NOT be taken if you call a public function in session.c.
181 *
182 * The lock is nested inside the structure: session_list_ptr->lock. Please use
183 * session_lock_list and session_unlock_list for lock acquisition.
184 */
185 static struct ltt_session_list *session_list_ptr;
186
187 int ust_consumerd64_fd = -1;
188 int ust_consumerd32_fd = -1;
189
190 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
191 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
192 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
193 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
194
195 static const char *module_proc_lttng = "/proc/lttng";
196
197 /*
198 * Consumer daemon state which is changed when spawning it, killing it or in
199 * case of a fatal error.
200 */
201 enum consumerd_state {
202 CONSUMER_STARTED = 1,
203 CONSUMER_STOPPED = 2,
204 CONSUMER_ERROR = 3,
205 };
206
207 /*
208 * This consumer daemon state is used to validate if a client command will be
209 * able to reach the consumer. If not, the client is informed. For instance,
210 * doing a "lttng start" when the consumer state is set to ERROR will return an
211 * error to the client.
212 *
213 * The following example shows a possible race condition of this scheme:
214 *
215 * consumer thread error happens
216 * client cmd arrives
217 * client cmd checks state -> still OK
218 * consumer thread exit, sets error
219 * client cmd try to talk to consumer
220 * ...
221 *
222 * However, since the consumer is a different daemon, we have no way of making
223 * sure the command will reach it safely even with this state flag. This is why
224 * we consider that up to the state validation during command processing, the
225 * command is safe. After that, we can not guarantee the correctness of the
226 * client request vis-a-vis the consumer.
227 */
228 static enum consumerd_state ust_consumerd_state;
229 static enum consumerd_state kernel_consumerd_state;
230
231 /*
232 * Socket timeout for receiving and sending in seconds.
233 */
234 static int app_socket_timeout;
235
236 /* Set in main() with the current page size. */
237 long page_size;
238
239 static
240 void setup_consumerd_path(void)
241 {
242 const char *bin, *libdir;
243
244 /*
245 * Allow INSTALL_BIN_PATH to be used as a target path for the
246 * native architecture size consumer if CONFIG_CONSUMER*_PATH
247 * has not been defined.
248 */
249 #if (CAA_BITS_PER_LONG == 32)
250 if (!consumerd32_bin[0]) {
251 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
252 }
253 if (!consumerd32_libdir[0]) {
254 consumerd32_libdir = INSTALL_LIB_PATH;
255 }
256 #elif (CAA_BITS_PER_LONG == 64)
257 if (!consumerd64_bin[0]) {
258 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
259 }
260 if (!consumerd64_libdir[0]) {
261 consumerd64_libdir = INSTALL_LIB_PATH;
262 }
263 #else
264 #error "Unknown bitness"
265 #endif
266
267 /*
268 * runtime env. var. overrides the build default.
269 */
270 bin = getenv("LTTNG_CONSUMERD32_BIN");
271 if (bin) {
272 consumerd32_bin = bin;
273 }
274 bin = getenv("LTTNG_CONSUMERD64_BIN");
275 if (bin) {
276 consumerd64_bin = bin;
277 }
278 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
279 if (libdir) {
280 consumerd32_libdir = libdir;
281 }
282 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
283 if (libdir) {
284 consumerd64_libdir = libdir;
285 }
286 }
287
288 /*
289 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
290 */
291 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
292 {
293 int ret;
294
295 assert(events);
296
297 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
298 if (ret < 0) {
299 goto error;
300 }
301
302 /* Add quit pipe */
303 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
304 if (ret < 0) {
305 goto error;
306 }
307
308 return 0;
309
310 error:
311 return ret;
312 }
313
314 /*
315 * Check if the thread quit pipe was triggered.
316 *
317 * Return 1 if it was triggered else 0;
318 */
319 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
320 {
321 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
322 return 1;
323 }
324
325 return 0;
326 }
327
328 /*
329 * Return group ID of the tracing group or -1 if not found.
330 */
331 static gid_t allowed_group(void)
332 {
333 struct group *grp;
334
335 if (opt_tracing_group) {
336 grp = getgrnam(opt_tracing_group);
337 } else {
338 grp = getgrnam(default_tracing_group);
339 }
340 if (!grp) {
341 return -1;
342 } else {
343 return grp->gr_gid;
344 }
345 }
346
347 /*
348 * Init thread quit pipe.
349 *
350 * Return -1 on error or 0 if all pipes are created.
351 */
352 static int init_thread_quit_pipe(void)
353 {
354 int ret, i;
355
356 ret = pipe(thread_quit_pipe);
357 if (ret < 0) {
358 PERROR("thread quit pipe");
359 goto error;
360 }
361
362 for (i = 0; i < 2; i++) {
363 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
364 if (ret < 0) {
365 PERROR("fcntl");
366 goto error;
367 }
368 }
369
370 error:
371 return ret;
372 }
373
374 /*
375 * Stop all threads by closing the thread quit pipe.
376 */
377 static void stop_threads(void)
378 {
379 int ret;
380
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret = notify_thread_pipe(thread_quit_pipe[1]);
384 if (ret < 0) {
385 ERR("write error on thread quit pipe");
386 }
387
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit, 1);
390 futex_nto1_wake(&ust_cmd_queue.futex);
391 }
392
393 /*
394 * Cleanup the daemon
395 */
396 static void cleanup(void)
397 {
398 int ret;
399 char *cmd = NULL;
400 struct ltt_session *sess, *stmp;
401
402 DBG("Cleaning up");
403
404 /* First thing first, stop all threads */
405 utils_close_pipe(thread_quit_pipe);
406
407 /*
408 * If opt_pidfile is undefined, the default file will be wiped when
409 * removing the rundir.
410 */
411 if (opt_pidfile) {
412 ret = remove(opt_pidfile);
413 if (ret < 0) {
414 PERROR("remove pidfile %s", opt_pidfile);
415 }
416 }
417
418 DBG("Removing %s directory", rundir);
419 ret = asprintf(&cmd, "rm -rf %s", rundir);
420 if (ret < 0) {
421 ERR("asprintf failed. Something is really wrong!");
422 }
423
424 /* Remove lttng run directory */
425 ret = system(cmd);
426 if (ret < 0) {
427 ERR("Unable to clean %s", rundir);
428 }
429 free(cmd);
430 free(rundir);
431
432 DBG("Cleaning up all sessions");
433
434 /* Destroy session list mutex */
435 if (session_list_ptr != NULL) {
436 pthread_mutex_destroy(&session_list_ptr->lock);
437
438 /* Cleanup ALL session */
439 cds_list_for_each_entry_safe(sess, stmp,
440 &session_list_ptr->head, list) {
441 cmd_destroy_session(sess, kernel_poll_pipe[1]);
442 }
443 }
444
445 DBG("Closing all UST sockets");
446 ust_app_clean_list();
447 buffer_reg_destroy_registries();
448
449 if (is_root && !opt_no_kernel) {
450 DBG2("Closing kernel fd");
451 if (kernel_tracer_fd >= 0) {
452 ret = close(kernel_tracer_fd);
453 if (ret) {
454 PERROR("close");
455 }
456 }
457 DBG("Unloading kernel modules");
458 modprobe_remove_lttng_all();
459 }
460
461 /* <fun> */
462 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
463 "Matthew, BEET driven development works!%c[%dm",
464 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
465 /* </fun> */
466 }
467
468 /*
469 * Send data on a unix socket using the liblttsessiondcomm API.
470 *
471 * Return lttcomm error code.
472 */
473 static int send_unix_sock(int sock, void *buf, size_t len)
474 {
475 /* Check valid length */
476 if (len == 0) {
477 return -1;
478 }
479
480 return lttcomm_send_unix_sock(sock, buf, len);
481 }
482
483 /*
484 * Free memory of a command context structure.
485 */
486 static void clean_command_ctx(struct command_ctx **cmd_ctx)
487 {
488 DBG("Clean command context structure");
489 if (*cmd_ctx) {
490 if ((*cmd_ctx)->llm) {
491 free((*cmd_ctx)->llm);
492 }
493 if ((*cmd_ctx)->lsm) {
494 free((*cmd_ctx)->lsm);
495 }
496 free(*cmd_ctx);
497 *cmd_ctx = NULL;
498 }
499 }
500
501 /*
502 * Notify UST applications using the shm mmap futex.
503 */
504 static int notify_ust_apps(int active)
505 {
506 char *wait_shm_mmap;
507
508 DBG("Notifying applications of session daemon state: %d", active);
509
510 /* See shm.c for this call implying mmap, shm and futex calls */
511 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
512 if (wait_shm_mmap == NULL) {
513 goto error;
514 }
515
516 /* Wake waiting process */
517 futex_wait_update((int32_t *) wait_shm_mmap, active);
518
519 /* Apps notified successfully */
520 return 0;
521
522 error:
523 return -1;
524 }
525
526 /*
527 * Setup the outgoing data buffer for the response (llm) by allocating the
528 * right amount of memory and copying the original information from the lsm
529 * structure.
530 *
531 * Return total size of the buffer pointed by buf.
532 */
533 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
534 {
535 int ret, buf_size;
536
537 buf_size = size;
538
539 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
540 if (cmd_ctx->llm == NULL) {
541 PERROR("zmalloc");
542 ret = -ENOMEM;
543 goto error;
544 }
545
546 /* Copy common data */
547 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
548 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
549
550 cmd_ctx->llm->data_size = size;
551 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
552
553 return buf_size;
554
555 error:
556 return ret;
557 }
558
559 /*
560 * Update the kernel poll set of all channel fd available over all tracing
561 * session. Add the wakeup pipe at the end of the set.
562 */
563 static int update_kernel_poll(struct lttng_poll_event *events)
564 {
565 int ret;
566 struct ltt_session *session;
567 struct ltt_kernel_channel *channel;
568
569 DBG("Updating kernel poll set");
570
571 session_lock_list();
572 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
573 session_lock(session);
574 if (session->kernel_session == NULL) {
575 session_unlock(session);
576 continue;
577 }
578
579 cds_list_for_each_entry(channel,
580 &session->kernel_session->channel_list.head, list) {
581 /* Add channel fd to the kernel poll set */
582 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
583 if (ret < 0) {
584 session_unlock(session);
585 goto error;
586 }
587 DBG("Channel fd %d added to kernel set", channel->fd);
588 }
589 session_unlock(session);
590 }
591 session_unlock_list();
592
593 return 0;
594
595 error:
596 session_unlock_list();
597 return -1;
598 }
599
600 /*
601 * Find the channel fd from 'fd' over all tracing session. When found, check
602 * for new channel stream and send those stream fds to the kernel consumer.
603 *
604 * Useful for CPU hotplug feature.
605 */
606 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
607 {
608 int ret = 0;
609 struct ltt_session *session;
610 struct ltt_kernel_session *ksess;
611 struct ltt_kernel_channel *channel;
612
613 DBG("Updating kernel streams for channel fd %d", fd);
614
615 session_lock_list();
616 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
617 session_lock(session);
618 if (session->kernel_session == NULL) {
619 session_unlock(session);
620 continue;
621 }
622 ksess = session->kernel_session;
623
624 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
625 if (channel->fd == fd) {
626 DBG("Channel found, updating kernel streams");
627 ret = kernel_open_channel_stream(channel);
628 if (ret < 0) {
629 goto error;
630 }
631
632 /*
633 * Have we already sent fds to the consumer? If yes, it means
634 * that tracing is started so it is safe to send our updated
635 * stream fds.
636 */
637 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
638 struct lttng_ht_iter iter;
639 struct consumer_socket *socket;
640
641 rcu_read_lock();
642 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
643 &iter.iter, socket, node.node) {
644 /* Code flow error */
645 assert(socket->fd >= 0);
646
647 pthread_mutex_lock(socket->lock);
648 ret = kernel_consumer_send_channel_stream(socket,
649 channel, ksess);
650 pthread_mutex_unlock(socket->lock);
651 if (ret < 0) {
652 rcu_read_unlock();
653 goto error;
654 }
655 }
656 rcu_read_unlock();
657 }
658 goto error;
659 }
660 }
661 session_unlock(session);
662 }
663 session_unlock_list();
664 return ret;
665
666 error:
667 session_unlock(session);
668 session_unlock_list();
669 return ret;
670 }
671
672 /*
673 * For each tracing session, update newly registered apps. The session list
674 * lock MUST be acquired before calling this.
675 */
676 static void update_ust_app(int app_sock)
677 {
678 struct ltt_session *sess, *stmp;
679
680 /* For all tracing session(s) */
681 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
682 session_lock(sess);
683 if (sess->ust_session) {
684 ust_app_global_update(sess->ust_session, app_sock);
685 }
686 session_unlock(sess);
687 }
688 }
689
690 /*
691 * This thread manage event coming from the kernel.
692 *
693 * Features supported in this thread:
694 * -) CPU Hotplug
695 */
696 static void *thread_manage_kernel(void *data)
697 {
698 int ret, i, pollfd, update_poll_flag = 1, err = -1;
699 uint32_t revents, nb_fd;
700 char tmp;
701 struct lttng_poll_event events;
702
703 DBG("[thread] Thread manage kernel started");
704
705 health_register(HEALTH_TYPE_KERNEL);
706
707 /*
708 * This first step of the while is to clean this structure which could free
709 * non NULL pointers so initialize it before the loop.
710 */
711 lttng_poll_init(&events);
712
713 if (testpoint(thread_manage_kernel)) {
714 goto error_testpoint;
715 }
716
717 health_code_update();
718
719 if (testpoint(thread_manage_kernel_before_loop)) {
720 goto error_testpoint;
721 }
722
723 while (1) {
724 health_code_update();
725
726 if (update_poll_flag == 1) {
727 /* Clean events object. We are about to populate it again. */
728 lttng_poll_clean(&events);
729
730 ret = sessiond_set_thread_pollset(&events, 2);
731 if (ret < 0) {
732 goto error_poll_create;
733 }
734
735 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
736 if (ret < 0) {
737 goto error;
738 }
739
740 /* This will add the available kernel channel if any. */
741 ret = update_kernel_poll(&events);
742 if (ret < 0) {
743 goto error;
744 }
745 update_poll_flag = 0;
746 }
747
748 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
749
750 /* Poll infinite value of time */
751 restart:
752 health_poll_entry();
753 ret = lttng_poll_wait(&events, -1);
754 health_poll_exit();
755 if (ret < 0) {
756 /*
757 * Restart interrupted system call.
758 */
759 if (errno == EINTR) {
760 goto restart;
761 }
762 goto error;
763 } else if (ret == 0) {
764 /* Should not happen since timeout is infinite */
765 ERR("Return value of poll is 0 with an infinite timeout.\n"
766 "This should not have happened! Continuing...");
767 continue;
768 }
769
770 nb_fd = ret;
771
772 for (i = 0; i < nb_fd; i++) {
773 /* Fetch once the poll data */
774 revents = LTTNG_POLL_GETEV(&events, i);
775 pollfd = LTTNG_POLL_GETFD(&events, i);
776
777 health_code_update();
778
779 /* Thread quit pipe has been closed. Killing thread. */
780 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
781 if (ret) {
782 err = 0;
783 goto exit;
784 }
785
786 /* Check for data on kernel pipe */
787 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
788 do {
789 ret = read(kernel_poll_pipe[0], &tmp, 1);
790 } while (ret < 0 && errno == EINTR);
791 /*
792 * Ret value is useless here, if this pipe gets any actions an
793 * update is required anyway.
794 */
795 update_poll_flag = 1;
796 continue;
797 } else {
798 /*
799 * New CPU detected by the kernel. Adding kernel stream to
800 * kernel session and updating the kernel consumer
801 */
802 if (revents & LPOLLIN) {
803 ret = update_kernel_stream(&kconsumer_data, pollfd);
804 if (ret < 0) {
805 continue;
806 }
807 break;
808 /*
809 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
810 * and unregister kernel stream at this point.
811 */
812 }
813 }
814 }
815 }
816
817 exit:
818 error:
819 lttng_poll_clean(&events);
820 error_poll_create:
821 error_testpoint:
822 utils_close_pipe(kernel_poll_pipe);
823 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
824 if (err) {
825 health_error();
826 ERR("Health error occurred in %s", __func__);
827 WARN("Kernel thread died unexpectedly. "
828 "Kernel tracing can continue but CPU hotplug is disabled.");
829 }
830 health_unregister();
831 DBG("Kernel thread dying");
832 return NULL;
833 }
834
835 /*
836 * Signal pthread condition of the consumer data that the thread.
837 */
838 static void signal_consumer_condition(struct consumer_data *data, int state)
839 {
840 pthread_mutex_lock(&data->cond_mutex);
841
842 /*
843 * The state is set before signaling. It can be any value, it's the waiter
844 * job to correctly interpret this condition variable associated to the
845 * consumer pthread_cond.
846 *
847 * A value of 0 means that the corresponding thread of the consumer data
848 * was not started. 1 indicates that the thread has started and is ready
849 * for action. A negative value means that there was an error during the
850 * thread bootstrap.
851 */
852 data->consumer_thread_is_ready = state;
853 (void) pthread_cond_signal(&data->cond);
854
855 pthread_mutex_unlock(&data->cond_mutex);
856 }
857
858 /*
859 * This thread manage the consumer error sent back to the session daemon.
860 */
861 static void *thread_manage_consumer(void *data)
862 {
863 int sock = -1, i, ret, pollfd, err = -1;
864 uint32_t revents, nb_fd;
865 enum lttcomm_return_code code;
866 struct lttng_poll_event events;
867 struct consumer_data *consumer_data = data;
868
869 DBG("[thread] Manage consumer started");
870
871 health_register(HEALTH_TYPE_CONSUMER);
872
873 health_code_update();
874
875 /*
876 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
877 * metadata_sock. Nothing more will be added to this poll set.
878 */
879 ret = sessiond_set_thread_pollset(&events, 3);
880 if (ret < 0) {
881 goto error_poll;
882 }
883
884 /*
885 * The error socket here is already in a listening state which was done
886 * just before spawning this thread to avoid a race between the consumer
887 * daemon exec trying to connect and the listen() call.
888 */
889 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
890 if (ret < 0) {
891 goto error;
892 }
893
894 health_code_update();
895
896 /* Infinite blocking call, waiting for transmission */
897 restart:
898 health_poll_entry();
899
900 if (testpoint(thread_manage_consumer)) {
901 goto error;
902 }
903
904 ret = lttng_poll_wait(&events, -1);
905 health_poll_exit();
906 if (ret < 0) {
907 /*
908 * Restart interrupted system call.
909 */
910 if (errno == EINTR) {
911 goto restart;
912 }
913 goto error;
914 }
915
916 nb_fd = ret;
917
918 for (i = 0; i < nb_fd; i++) {
919 /* Fetch once the poll data */
920 revents = LTTNG_POLL_GETEV(&events, i);
921 pollfd = LTTNG_POLL_GETFD(&events, i);
922
923 health_code_update();
924
925 /* Thread quit pipe has been closed. Killing thread. */
926 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
927 if (ret) {
928 err = 0;
929 goto exit;
930 }
931
932 /* Event on the registration socket */
933 if (pollfd == consumer_data->err_sock) {
934 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
935 ERR("consumer err socket poll error");
936 goto error;
937 }
938 }
939 }
940
941 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
942 if (sock < 0) {
943 goto error;
944 }
945
946 /*
947 * Set the CLOEXEC flag. Return code is useless because either way, the
948 * show must go on.
949 */
950 (void) utils_set_fd_cloexec(sock);
951
952 health_code_update();
953
954 DBG2("Receiving code from consumer err_sock");
955
956 /* Getting status code from kconsumerd */
957 ret = lttcomm_recv_unix_sock(sock, &code,
958 sizeof(enum lttcomm_return_code));
959 if (ret <= 0) {
960 goto error;
961 }
962
963 health_code_update();
964
965 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
966 /* Connect both socket, command and metadata. */
967 consumer_data->cmd_sock =
968 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
969 consumer_data->metadata_sock.fd =
970 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
971 if (consumer_data->cmd_sock < 0 ||
972 consumer_data->metadata_sock.fd < 0) {
973 PERROR("consumer connect cmd socket");
974 /* On error, signal condition and quit. */
975 signal_consumer_condition(consumer_data, -1);
976 goto error;
977 }
978 /* Create metadata socket lock. */
979 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
980 if (consumer_data->metadata_sock.lock == NULL) {
981 PERROR("zmalloc pthread mutex");
982 ret = -1;
983 goto error;
984 }
985 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
986
987 signal_consumer_condition(consumer_data, 1);
988 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
989 DBG("Consumer metadata socket ready (fd: %d)",
990 consumer_data->metadata_sock.fd);
991 } else {
992 ERR("consumer error when waiting for SOCK_READY : %s",
993 lttcomm_get_readable_code(-code));
994 goto error;
995 }
996
997 /* Remove the consumerd error sock since we've established a connexion */
998 ret = lttng_poll_del(&events, consumer_data->err_sock);
999 if (ret < 0) {
1000 goto error;
1001 }
1002
1003 /* Add new accepted error socket. */
1004 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1005 if (ret < 0) {
1006 goto error;
1007 }
1008
1009 /* Add metadata socket that is successfully connected. */
1010 ret = lttng_poll_add(&events, consumer_data->metadata_sock.fd,
1011 LPOLLIN | LPOLLRDHUP);
1012 if (ret < 0) {
1013 goto error;
1014 }
1015
1016 health_code_update();
1017
1018 /* Infinite blocking call, waiting for transmission */
1019 restart_poll:
1020 while (1) {
1021 health_poll_entry();
1022 ret = lttng_poll_wait(&events, -1);
1023 health_poll_exit();
1024 if (ret < 0) {
1025 /*
1026 * Restart interrupted system call.
1027 */
1028 if (errno == EINTR) {
1029 goto restart_poll;
1030 }
1031 goto error;
1032 }
1033
1034 nb_fd = ret;
1035
1036 for (i = 0; i < nb_fd; i++) {
1037 /* Fetch once the poll data */
1038 revents = LTTNG_POLL_GETEV(&events, i);
1039 pollfd = LTTNG_POLL_GETFD(&events, i);
1040
1041 health_code_update();
1042
1043 /* Thread quit pipe has been closed. Killing thread. */
1044 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1045 if (ret) {
1046 err = 0;
1047 goto exit;
1048 }
1049
1050 if (pollfd == sock) {
1051 /* Event on the consumerd socket */
1052 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1053 ERR("consumer err socket second poll error");
1054 goto error;
1055 }
1056 health_code_update();
1057 /* Wait for any kconsumerd error */
1058 ret = lttcomm_recv_unix_sock(sock, &code,
1059 sizeof(enum lttcomm_return_code));
1060 if (ret <= 0) {
1061 ERR("consumer closed the command socket");
1062 goto error;
1063 }
1064
1065 ERR("consumer return code : %s",
1066 lttcomm_get_readable_code(-code));
1067
1068 goto exit;
1069 } else if (pollfd == consumer_data->metadata_sock.fd) {
1070 /* UST metadata requests */
1071 ret = ust_consumer_metadata_request(
1072 &consumer_data->metadata_sock);
1073 if (ret < 0) {
1074 ERR("Handling metadata request");
1075 goto error;
1076 }
1077 break;
1078 } else {
1079 ERR("Unknown pollfd");
1080 goto error;
1081 }
1082 }
1083 health_code_update();
1084 }
1085
1086 exit:
1087 error:
1088 /* Immediately set the consumerd state to stopped */
1089 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1090 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1091 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1092 consumer_data->type == LTTNG_CONSUMER32_UST) {
1093 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1094 } else {
1095 /* Code flow error... */
1096 assert(0);
1097 }
1098
1099 if (consumer_data->err_sock >= 0) {
1100 ret = close(consumer_data->err_sock);
1101 if (ret) {
1102 PERROR("close");
1103 }
1104 }
1105 if (consumer_data->cmd_sock >= 0) {
1106 ret = close(consumer_data->cmd_sock);
1107 if (ret) {
1108 PERROR("close");
1109 }
1110 }
1111 if (consumer_data->metadata_sock.fd >= 0) {
1112 ret = close(consumer_data->metadata_sock.fd);
1113 if (ret) {
1114 PERROR("close");
1115 }
1116 }
1117 /* Cleanup metadata socket mutex. */
1118 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1119 free(consumer_data->metadata_sock.lock);
1120
1121 if (sock >= 0) {
1122 ret = close(sock);
1123 if (ret) {
1124 PERROR("close");
1125 }
1126 }
1127
1128 unlink(consumer_data->err_unix_sock_path);
1129 unlink(consumer_data->cmd_unix_sock_path);
1130 consumer_data->pid = 0;
1131
1132 lttng_poll_clean(&events);
1133 error_poll:
1134 if (err) {
1135 health_error();
1136 ERR("Health error occurred in %s", __func__);
1137 }
1138 health_unregister();
1139 DBG("consumer thread cleanup completed");
1140
1141 return NULL;
1142 }
1143
1144 /*
1145 * This thread manage application communication.
1146 */
1147 static void *thread_manage_apps(void *data)
1148 {
1149 int i, ret, pollfd, err = -1;
1150 uint32_t revents, nb_fd;
1151 struct lttng_poll_event events;
1152
1153 DBG("[thread] Manage application started");
1154
1155 rcu_register_thread();
1156 rcu_thread_online();
1157
1158 health_register(HEALTH_TYPE_APP_MANAGE);
1159
1160 if (testpoint(thread_manage_apps)) {
1161 goto error_testpoint;
1162 }
1163
1164 health_code_update();
1165
1166 ret = sessiond_set_thread_pollset(&events, 2);
1167 if (ret < 0) {
1168 goto error_poll_create;
1169 }
1170
1171 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1172 if (ret < 0) {
1173 goto error;
1174 }
1175
1176 if (testpoint(thread_manage_apps_before_loop)) {
1177 goto error;
1178 }
1179
1180 health_code_update();
1181
1182 while (1) {
1183 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1184
1185 /* Inifinite blocking call, waiting for transmission */
1186 restart:
1187 health_poll_entry();
1188 ret = lttng_poll_wait(&events, -1);
1189 health_poll_exit();
1190 if (ret < 0) {
1191 /*
1192 * Restart interrupted system call.
1193 */
1194 if (errno == EINTR) {
1195 goto restart;
1196 }
1197 goto error;
1198 }
1199
1200 nb_fd = ret;
1201
1202 for (i = 0; i < nb_fd; i++) {
1203 /* Fetch once the poll data */
1204 revents = LTTNG_POLL_GETEV(&events, i);
1205 pollfd = LTTNG_POLL_GETFD(&events, i);
1206
1207 health_code_update();
1208
1209 /* Thread quit pipe has been closed. Killing thread. */
1210 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1211 if (ret) {
1212 err = 0;
1213 goto exit;
1214 }
1215
1216 /* Inspect the apps cmd pipe */
1217 if (pollfd == apps_cmd_pipe[0]) {
1218 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1219 ERR("Apps command pipe error");
1220 goto error;
1221 } else if (revents & LPOLLIN) {
1222 int sock;
1223
1224 /* Empty pipe */
1225 do {
1226 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1227 } while (ret < 0 && errno == EINTR);
1228 if (ret < 0 || ret < sizeof(sock)) {
1229 PERROR("read apps cmd pipe");
1230 goto error;
1231 }
1232
1233 health_code_update();
1234
1235 /*
1236 * We only monitor the error events of the socket. This
1237 * thread does not handle any incoming data from UST
1238 * (POLLIN).
1239 */
1240 ret = lttng_poll_add(&events, sock,
1241 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1242 if (ret < 0) {
1243 goto error;
1244 }
1245
1246 /* Set socket timeout for both receiving and ending */
1247 (void) lttcomm_setsockopt_rcv_timeout(sock,
1248 app_socket_timeout);
1249 (void) lttcomm_setsockopt_snd_timeout(sock,
1250 app_socket_timeout);
1251
1252 DBG("Apps with sock %d added to poll set", sock);
1253
1254 health_code_update();
1255
1256 break;
1257 }
1258 } else {
1259 /*
1260 * At this point, we know that a registered application made
1261 * the event at poll_wait.
1262 */
1263 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1264 /* Removing from the poll set */
1265 ret = lttng_poll_del(&events, pollfd);
1266 if (ret < 0) {
1267 goto error;
1268 }
1269
1270 /* Socket closed on remote end. */
1271 ust_app_unregister(pollfd);
1272 break;
1273 }
1274 }
1275
1276 health_code_update();
1277 }
1278 }
1279
1280 exit:
1281 error:
1282 lttng_poll_clean(&events);
1283 error_poll_create:
1284 error_testpoint:
1285 utils_close_pipe(apps_cmd_pipe);
1286 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1287
1288 /*
1289 * We don't clean the UST app hash table here since already registered
1290 * applications can still be controlled so let them be until the session
1291 * daemon dies or the applications stop.
1292 */
1293
1294 if (err) {
1295 health_error();
1296 ERR("Health error occurred in %s", __func__);
1297 }
1298 health_unregister();
1299 DBG("Application communication apps thread cleanup complete");
1300 rcu_thread_offline();
1301 rcu_unregister_thread();
1302 return NULL;
1303 }
1304
1305 /*
1306 * Send a socket to a thread This is called from the dispatch UST registration
1307 * thread once all sockets are set for the application.
1308 *
1309 * On success, return 0 else a negative value being the errno message of the
1310 * write().
1311 */
1312 static int send_socket_to_thread(int fd, int sock)
1313 {
1314 int ret;
1315
1316 /* Sockets MUST be set or else this should not have been called. */
1317 assert(fd >= 0);
1318 assert(sock >= 0);
1319
1320 do {
1321 ret = write(fd, &sock, sizeof(sock));
1322 } while (ret < 0 && errno == EINTR);
1323 if (ret < 0 || ret != sizeof(sock)) {
1324 PERROR("write apps pipe %d", fd);
1325 if (ret < 0) {
1326 ret = -errno;
1327 }
1328 goto error;
1329 }
1330
1331 /* All good. Don't send back the write positive ret value. */
1332 ret = 0;
1333 error:
1334 return ret;
1335 }
1336
1337 /*
1338 * Dispatch request from the registration threads to the application
1339 * communication thread.
1340 */
1341 static void *thread_dispatch_ust_registration(void *data)
1342 {
1343 int ret;
1344 struct cds_wfq_node *node;
1345 struct ust_command *ust_cmd = NULL;
1346 struct {
1347 struct ust_app *app;
1348 struct cds_list_head head;
1349 } *wait_node = NULL, *tmp_wait_node;
1350
1351 CDS_LIST_HEAD(wait_queue);
1352
1353 DBG("[thread] Dispatch UST command started");
1354
1355 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1356 /* Atomically prepare the queue futex */
1357 futex_nto1_prepare(&ust_cmd_queue.futex);
1358
1359 do {
1360 struct ust_app *app = NULL;
1361 ust_cmd = NULL;
1362
1363 /* Dequeue command for registration */
1364 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1365 if (node == NULL) {
1366 DBG("Woken up but nothing in the UST command queue");
1367 /* Continue thread execution */
1368 break;
1369 }
1370
1371 ust_cmd = caa_container_of(node, struct ust_command, node);
1372
1373 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1374 " gid:%d sock:%d name:%s (version %d.%d)",
1375 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1376 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1377 ust_cmd->sock, ust_cmd->reg_msg.name,
1378 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1379
1380 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1381 wait_node = zmalloc(sizeof(*wait_node));
1382 if (!wait_node) {
1383 PERROR("zmalloc wait_node dispatch");
1384 ret = close(ust_cmd->sock);
1385 if (ret < 0) {
1386 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1387 }
1388 lttng_fd_put(1, LTTNG_FD_APPS);
1389 free(ust_cmd);
1390 goto error;
1391 }
1392 CDS_INIT_LIST_HEAD(&wait_node->head);
1393
1394 /* Create application object if socket is CMD. */
1395 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1396 ust_cmd->sock);
1397 if (!wait_node->app) {
1398 ret = close(ust_cmd->sock);
1399 if (ret < 0) {
1400 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1401 }
1402 lttng_fd_put(1, LTTNG_FD_APPS);
1403 free(wait_node);
1404 free(ust_cmd);
1405 continue;
1406 }
1407 /*
1408 * Add application to the wait queue so we can set the notify
1409 * socket before putting this object in the global ht.
1410 */
1411 cds_list_add(&wait_node->head, &wait_queue);
1412
1413 free(ust_cmd);
1414 /*
1415 * We have to continue here since we don't have the notify
1416 * socket and the application MUST be added to the hash table
1417 * only at that moment.
1418 */
1419 continue;
1420 } else {
1421 /*
1422 * Look for the application in the local wait queue and set the
1423 * notify socket if found.
1424 */
1425 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1426 &wait_queue, head) {
1427 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1428 wait_node->app->notify_sock = ust_cmd->sock;
1429 cds_list_del(&wait_node->head);
1430 app = wait_node->app;
1431 free(wait_node);
1432 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1433 break;
1434 }
1435 }
1436
1437 /*
1438 * With no application at this stage the received socket is
1439 * basically useless so close it before we free the cmd data
1440 * structure for good.
1441 */
1442 if (!app) {
1443 ret = close(ust_cmd->sock);
1444 if (ret < 0) {
1445 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1446 }
1447 lttng_fd_put(1, LTTNG_FD_APPS);
1448 }
1449 free(ust_cmd);
1450 }
1451
1452 if (app) {
1453 /*
1454 * @session_lock_list
1455 *
1456 * Lock the global session list so from the register up to the
1457 * registration done message, no thread can see the application
1458 * and change its state.
1459 */
1460 session_lock_list();
1461 rcu_read_lock();
1462
1463 /*
1464 * Add application to the global hash table. This needs to be
1465 * done before the update to the UST registry can locate the
1466 * application.
1467 */
1468 ust_app_add(app);
1469
1470 /* Set app version. This call will print an error if needed. */
1471 (void) ust_app_version(app);
1472
1473 /* Send notify socket through the notify pipe. */
1474 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1475 app->notify_sock);
1476 if (ret < 0) {
1477 rcu_read_unlock();
1478 session_unlock_list();
1479 /* No notify thread, stop the UST tracing. */
1480 goto error;
1481 }
1482
1483 /*
1484 * Update newly registered application with the tracing
1485 * registry info already enabled information.
1486 */
1487 update_ust_app(app->sock);
1488
1489 /*
1490 * Don't care about return value. Let the manage apps threads
1491 * handle app unregistration upon socket close.
1492 */
1493 (void) ust_app_register_done(app->sock);
1494
1495 /*
1496 * Even if the application socket has been closed, send the app
1497 * to the thread and unregistration will take place at that
1498 * place.
1499 */
1500 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1501 if (ret < 0) {
1502 rcu_read_unlock();
1503 session_unlock_list();
1504 /* No apps. thread, stop the UST tracing. */
1505 goto error;
1506 }
1507
1508 rcu_read_unlock();
1509 session_unlock_list();
1510 }
1511 } while (node != NULL);
1512
1513 /* Futex wait on queue. Blocking call on futex() */
1514 futex_nto1_wait(&ust_cmd_queue.futex);
1515 }
1516
1517 error:
1518 /* Clean up wait queue. */
1519 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1520 &wait_queue, head) {
1521 cds_list_del(&wait_node->head);
1522 free(wait_node);
1523 }
1524
1525 DBG("Dispatch thread dying");
1526 return NULL;
1527 }
1528
1529 /*
1530 * This thread manage application registration.
1531 */
1532 static void *thread_registration_apps(void *data)
1533 {
1534 int sock = -1, i, ret, pollfd, err = -1;
1535 uint32_t revents, nb_fd;
1536 struct lttng_poll_event events;
1537 /*
1538 * Get allocated in this thread, enqueued to a global queue, dequeued and
1539 * freed in the manage apps thread.
1540 */
1541 struct ust_command *ust_cmd = NULL;
1542
1543 DBG("[thread] Manage application registration started");
1544
1545 health_register(HEALTH_TYPE_APP_REG);
1546
1547 if (testpoint(thread_registration_apps)) {
1548 goto error_testpoint;
1549 }
1550
1551 ret = lttcomm_listen_unix_sock(apps_sock);
1552 if (ret < 0) {
1553 goto error_listen;
1554 }
1555
1556 /*
1557 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1558 * more will be added to this poll set.
1559 */
1560 ret = sessiond_set_thread_pollset(&events, 2);
1561 if (ret < 0) {
1562 goto error_create_poll;
1563 }
1564
1565 /* Add the application registration socket */
1566 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1567 if (ret < 0) {
1568 goto error_poll_add;
1569 }
1570
1571 /* Notify all applications to register */
1572 ret = notify_ust_apps(1);
1573 if (ret < 0) {
1574 ERR("Failed to notify applications or create the wait shared memory.\n"
1575 "Execution continues but there might be problem for already\n"
1576 "running applications that wishes to register.");
1577 }
1578
1579 while (1) {
1580 DBG("Accepting application registration");
1581
1582 /* Inifinite blocking call, waiting for transmission */
1583 restart:
1584 health_poll_entry();
1585 ret = lttng_poll_wait(&events, -1);
1586 health_poll_exit();
1587 if (ret < 0) {
1588 /*
1589 * Restart interrupted system call.
1590 */
1591 if (errno == EINTR) {
1592 goto restart;
1593 }
1594 goto error;
1595 }
1596
1597 nb_fd = ret;
1598
1599 for (i = 0; i < nb_fd; i++) {
1600 health_code_update();
1601
1602 /* Fetch once the poll data */
1603 revents = LTTNG_POLL_GETEV(&events, i);
1604 pollfd = LTTNG_POLL_GETFD(&events, i);
1605
1606 /* Thread quit pipe has been closed. Killing thread. */
1607 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1608 if (ret) {
1609 err = 0;
1610 goto exit;
1611 }
1612
1613 /* Event on the registration socket */
1614 if (pollfd == apps_sock) {
1615 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1616 ERR("Register apps socket poll error");
1617 goto error;
1618 } else if (revents & LPOLLIN) {
1619 sock = lttcomm_accept_unix_sock(apps_sock);
1620 if (sock < 0) {
1621 goto error;
1622 }
1623
1624 /*
1625 * Set the CLOEXEC flag. Return code is useless because
1626 * either way, the show must go on.
1627 */
1628 (void) utils_set_fd_cloexec(sock);
1629
1630 /* Create UST registration command for enqueuing */
1631 ust_cmd = zmalloc(sizeof(struct ust_command));
1632 if (ust_cmd == NULL) {
1633 PERROR("ust command zmalloc");
1634 goto error;
1635 }
1636
1637 /*
1638 * Using message-based transmissions to ensure we don't
1639 * have to deal with partially received messages.
1640 */
1641 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1642 if (ret < 0) {
1643 ERR("Exhausted file descriptors allowed for applications.");
1644 free(ust_cmd);
1645 ret = close(sock);
1646 if (ret) {
1647 PERROR("close");
1648 }
1649 sock = -1;
1650 continue;
1651 }
1652
1653 health_code_update();
1654 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1655 if (ret < 0) {
1656 free(ust_cmd);
1657 /* Close socket of the application. */
1658 ret = close(sock);
1659 if (ret) {
1660 PERROR("close");
1661 }
1662 lttng_fd_put(LTTNG_FD_APPS, 1);
1663 sock = -1;
1664 continue;
1665 }
1666 health_code_update();
1667
1668 ust_cmd->sock = sock;
1669 sock = -1;
1670
1671 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1672 " gid:%d sock:%d name:%s (version %d.%d)",
1673 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1674 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1675 ust_cmd->sock, ust_cmd->reg_msg.name,
1676 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1677
1678 /*
1679 * Lock free enqueue the registration request. The red pill
1680 * has been taken! This apps will be part of the *system*.
1681 */
1682 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1683
1684 /*
1685 * Wake the registration queue futex. Implicit memory
1686 * barrier with the exchange in cds_wfq_enqueue.
1687 */
1688 futex_nto1_wake(&ust_cmd_queue.futex);
1689 }
1690 }
1691 }
1692 }
1693
1694 exit:
1695 error:
1696 if (err) {
1697 health_error();
1698 ERR("Health error occurred in %s", __func__);
1699 }
1700
1701 /* Notify that the registration thread is gone */
1702 notify_ust_apps(0);
1703
1704 if (apps_sock >= 0) {
1705 ret = close(apps_sock);
1706 if (ret) {
1707 PERROR("close");
1708 }
1709 }
1710 if (sock >= 0) {
1711 ret = close(sock);
1712 if (ret) {
1713 PERROR("close");
1714 }
1715 lttng_fd_put(LTTNG_FD_APPS, 1);
1716 }
1717 unlink(apps_unix_sock_path);
1718
1719 error_poll_add:
1720 lttng_poll_clean(&events);
1721 error_listen:
1722 error_create_poll:
1723 error_testpoint:
1724 DBG("UST Registration thread cleanup complete");
1725 health_unregister();
1726
1727 return NULL;
1728 }
1729
1730 /*
1731 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1732 * exec or it will fails.
1733 */
1734 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1735 {
1736 int ret, clock_ret;
1737 struct timespec timeout;
1738
1739 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1740 consumer_data->consumer_thread_is_ready = 0;
1741
1742 /* Setup pthread condition */
1743 ret = pthread_condattr_init(&consumer_data->condattr);
1744 if (ret != 0) {
1745 errno = ret;
1746 PERROR("pthread_condattr_init consumer data");
1747 goto error;
1748 }
1749
1750 /*
1751 * Set the monotonic clock in order to make sure we DO NOT jump in time
1752 * between the clock_gettime() call and the timedwait call. See bug #324
1753 * for a more details and how we noticed it.
1754 */
1755 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1756 if (ret != 0) {
1757 errno = ret;
1758 PERROR("pthread_condattr_setclock consumer data");
1759 goto error;
1760 }
1761
1762 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1763 if (ret != 0) {
1764 errno = ret;
1765 PERROR("pthread_cond_init consumer data");
1766 goto error;
1767 }
1768
1769 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1770 consumer_data);
1771 if (ret != 0) {
1772 PERROR("pthread_create consumer");
1773 ret = -1;
1774 goto error;
1775 }
1776
1777 /* We are about to wait on a pthread condition */
1778 pthread_mutex_lock(&consumer_data->cond_mutex);
1779
1780 /* Get time for sem_timedwait absolute timeout */
1781 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1782 /*
1783 * Set the timeout for the condition timed wait even if the clock gettime
1784 * call fails since we might loop on that call and we want to avoid to
1785 * increment the timeout too many times.
1786 */
1787 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1788
1789 /*
1790 * The following loop COULD be skipped in some conditions so this is why we
1791 * set ret to 0 in order to make sure at least one round of the loop is
1792 * done.
1793 */
1794 ret = 0;
1795
1796 /*
1797 * Loop until the condition is reached or when a timeout is reached. Note
1798 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1799 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1800 * possible. This loop does not take any chances and works with both of
1801 * them.
1802 */
1803 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1804 if (clock_ret < 0) {
1805 PERROR("clock_gettime spawn consumer");
1806 /* Infinite wait for the consumerd thread to be ready */
1807 ret = pthread_cond_wait(&consumer_data->cond,
1808 &consumer_data->cond_mutex);
1809 } else {
1810 ret = pthread_cond_timedwait(&consumer_data->cond,
1811 &consumer_data->cond_mutex, &timeout);
1812 }
1813 }
1814
1815 /* Release the pthread condition */
1816 pthread_mutex_unlock(&consumer_data->cond_mutex);
1817
1818 if (ret != 0) {
1819 errno = ret;
1820 if (ret == ETIMEDOUT) {
1821 /*
1822 * Call has timed out so we kill the kconsumerd_thread and return
1823 * an error.
1824 */
1825 ERR("Condition timed out. The consumer thread was never ready."
1826 " Killing it");
1827 ret = pthread_cancel(consumer_data->thread);
1828 if (ret < 0) {
1829 PERROR("pthread_cancel consumer thread");
1830 }
1831 } else {
1832 PERROR("pthread_cond_wait failed consumer thread");
1833 }
1834 goto error;
1835 }
1836
1837 pthread_mutex_lock(&consumer_data->pid_mutex);
1838 if (consumer_data->pid == 0) {
1839 ERR("Consumerd did not start");
1840 pthread_mutex_unlock(&consumer_data->pid_mutex);
1841 goto error;
1842 }
1843 pthread_mutex_unlock(&consumer_data->pid_mutex);
1844
1845 return 0;
1846
1847 error:
1848 return ret;
1849 }
1850
1851 /*
1852 * Join consumer thread
1853 */
1854 static int join_consumer_thread(struct consumer_data *consumer_data)
1855 {
1856 void *status;
1857
1858 /* Consumer pid must be a real one. */
1859 if (consumer_data->pid > 0) {
1860 int ret;
1861 ret = kill(consumer_data->pid, SIGTERM);
1862 if (ret) {
1863 ERR("Error killing consumer daemon");
1864 return ret;
1865 }
1866 return pthread_join(consumer_data->thread, &status);
1867 } else {
1868 return 0;
1869 }
1870 }
1871
1872 /*
1873 * Fork and exec a consumer daemon (consumerd).
1874 *
1875 * Return pid if successful else -1.
1876 */
1877 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1878 {
1879 int ret;
1880 pid_t pid;
1881 const char *consumer_to_use;
1882 const char *verbosity;
1883 struct stat st;
1884
1885 DBG("Spawning consumerd");
1886
1887 pid = fork();
1888 if (pid == 0) {
1889 /*
1890 * Exec consumerd.
1891 */
1892 if (opt_verbose_consumer) {
1893 verbosity = "--verbose";
1894 } else {
1895 verbosity = "--quiet";
1896 }
1897 switch (consumer_data->type) {
1898 case LTTNG_CONSUMER_KERNEL:
1899 /*
1900 * Find out which consumerd to execute. We will first try the
1901 * 64-bit path, then the sessiond's installation directory, and
1902 * fallback on the 32-bit one,
1903 */
1904 DBG3("Looking for a kernel consumer at these locations:");
1905 DBG3(" 1) %s", consumerd64_bin);
1906 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1907 DBG3(" 3) %s", consumerd32_bin);
1908 if (stat(consumerd64_bin, &st) == 0) {
1909 DBG3("Found location #1");
1910 consumer_to_use = consumerd64_bin;
1911 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1912 DBG3("Found location #2");
1913 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1914 } else if (stat(consumerd32_bin, &st) == 0) {
1915 DBG3("Found location #3");
1916 consumer_to_use = consumerd32_bin;
1917 } else {
1918 DBG("Could not find any valid consumerd executable");
1919 break;
1920 }
1921 DBG("Using kernel consumer at: %s", consumer_to_use);
1922 execl(consumer_to_use,
1923 "lttng-consumerd", verbosity, "-k",
1924 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1925 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1926 NULL);
1927 break;
1928 case LTTNG_CONSUMER64_UST:
1929 {
1930 char *tmpnew = NULL;
1931
1932 if (consumerd64_libdir[0] != '\0') {
1933 char *tmp;
1934 size_t tmplen;
1935
1936 tmp = getenv("LD_LIBRARY_PATH");
1937 if (!tmp) {
1938 tmp = "";
1939 }
1940 tmplen = strlen("LD_LIBRARY_PATH=")
1941 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1942 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1943 if (!tmpnew) {
1944 ret = -ENOMEM;
1945 goto error;
1946 }
1947 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1948 strcat(tmpnew, consumerd64_libdir);
1949 if (tmp[0] != '\0') {
1950 strcat(tmpnew, ":");
1951 strcat(tmpnew, tmp);
1952 }
1953 ret = putenv(tmpnew);
1954 if (ret) {
1955 ret = -errno;
1956 free(tmpnew);
1957 goto error;
1958 }
1959 }
1960 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1961 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1962 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1963 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1964 NULL);
1965 if (consumerd64_libdir[0] != '\0') {
1966 free(tmpnew);
1967 }
1968 if (ret) {
1969 goto error;
1970 }
1971 break;
1972 }
1973 case LTTNG_CONSUMER32_UST:
1974 {
1975 char *tmpnew = NULL;
1976
1977 if (consumerd32_libdir[0] != '\0') {
1978 char *tmp;
1979 size_t tmplen;
1980
1981 tmp = getenv("LD_LIBRARY_PATH");
1982 if (!tmp) {
1983 tmp = "";
1984 }
1985 tmplen = strlen("LD_LIBRARY_PATH=")
1986 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1987 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1988 if (!tmpnew) {
1989 ret = -ENOMEM;
1990 goto error;
1991 }
1992 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1993 strcat(tmpnew, consumerd32_libdir);
1994 if (tmp[0] != '\0') {
1995 strcat(tmpnew, ":");
1996 strcat(tmpnew, tmp);
1997 }
1998 ret = putenv(tmpnew);
1999 if (ret) {
2000 ret = -errno;
2001 free(tmpnew);
2002 goto error;
2003 }
2004 }
2005 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2006 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2007 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2008 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2009 NULL);
2010 if (consumerd32_libdir[0] != '\0') {
2011 free(tmpnew);
2012 }
2013 if (ret) {
2014 goto error;
2015 }
2016 break;
2017 }
2018 default:
2019 PERROR("unknown consumer type");
2020 exit(EXIT_FAILURE);
2021 }
2022 if (errno != 0) {
2023 PERROR("kernel start consumer exec");
2024 }
2025 exit(EXIT_FAILURE);
2026 } else if (pid > 0) {
2027 ret = pid;
2028 } else {
2029 PERROR("start consumer fork");
2030 ret = -errno;
2031 }
2032 error:
2033 return ret;
2034 }
2035
2036 /*
2037 * Spawn the consumerd daemon and session daemon thread.
2038 */
2039 static int start_consumerd(struct consumer_data *consumer_data)
2040 {
2041 int ret;
2042
2043 /*
2044 * Set the listen() state on the socket since there is a possible race
2045 * between the exec() of the consumer daemon and this call if place in the
2046 * consumer thread. See bug #366 for more details.
2047 */
2048 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2049 if (ret < 0) {
2050 goto error;
2051 }
2052
2053 pthread_mutex_lock(&consumer_data->pid_mutex);
2054 if (consumer_data->pid != 0) {
2055 pthread_mutex_unlock(&consumer_data->pid_mutex);
2056 goto end;
2057 }
2058
2059 ret = spawn_consumerd(consumer_data);
2060 if (ret < 0) {
2061 ERR("Spawning consumerd failed");
2062 pthread_mutex_unlock(&consumer_data->pid_mutex);
2063 goto error;
2064 }
2065
2066 /* Setting up the consumer_data pid */
2067 consumer_data->pid = ret;
2068 DBG2("Consumer pid %d", consumer_data->pid);
2069 pthread_mutex_unlock(&consumer_data->pid_mutex);
2070
2071 DBG2("Spawning consumer control thread");
2072 ret = spawn_consumer_thread(consumer_data);
2073 if (ret < 0) {
2074 ERR("Fatal error spawning consumer control thread");
2075 goto error;
2076 }
2077
2078 end:
2079 return 0;
2080
2081 error:
2082 /* Cleanup already created sockets on error. */
2083 if (consumer_data->err_sock >= 0) {
2084 int err;
2085
2086 err = close(consumer_data->err_sock);
2087 if (err < 0) {
2088 PERROR("close consumer data error socket");
2089 }
2090 }
2091 return ret;
2092 }
2093
2094 /*
2095 * Compute health status of each consumer. If one of them is zero (bad
2096 * state), we return 0.
2097 */
2098 static int check_consumer_health(void)
2099 {
2100 int ret;
2101
2102 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2103
2104 DBG3("Health consumer check %d", ret);
2105
2106 return ret;
2107 }
2108
2109 /*
2110 * Setup necessary data for kernel tracer action.
2111 */
2112 static int init_kernel_tracer(void)
2113 {
2114 int ret;
2115
2116 /* Modprobe lttng kernel modules */
2117 ret = modprobe_lttng_control();
2118 if (ret < 0) {
2119 goto error;
2120 }
2121
2122 /* Open debugfs lttng */
2123 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2124 if (kernel_tracer_fd < 0) {
2125 DBG("Failed to open %s", module_proc_lttng);
2126 ret = -1;
2127 goto error_open;
2128 }
2129
2130 /* Validate kernel version */
2131 ret = kernel_validate_version(kernel_tracer_fd);
2132 if (ret < 0) {
2133 goto error_version;
2134 }
2135
2136 ret = modprobe_lttng_data();
2137 if (ret < 0) {
2138 goto error_modules;
2139 }
2140
2141 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2142 return 0;
2143
2144 error_version:
2145 modprobe_remove_lttng_control();
2146 ret = close(kernel_tracer_fd);
2147 if (ret) {
2148 PERROR("close");
2149 }
2150 kernel_tracer_fd = -1;
2151 return LTTNG_ERR_KERN_VERSION;
2152
2153 error_modules:
2154 ret = close(kernel_tracer_fd);
2155 if (ret) {
2156 PERROR("close");
2157 }
2158
2159 error_open:
2160 modprobe_remove_lttng_control();
2161
2162 error:
2163 WARN("No kernel tracer available");
2164 kernel_tracer_fd = -1;
2165 if (!is_root) {
2166 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2167 } else {
2168 return LTTNG_ERR_KERN_NA;
2169 }
2170 }
2171
2172
2173 /*
2174 * Copy consumer output from the tracing session to the domain session. The
2175 * function also applies the right modification on a per domain basis for the
2176 * trace files destination directory.
2177 *
2178 * Should *NOT* be called with RCU read-side lock held.
2179 */
2180 static int copy_session_consumer(int domain, struct ltt_session *session)
2181 {
2182 int ret;
2183 const char *dir_name;
2184 struct consumer_output *consumer;
2185
2186 assert(session);
2187 assert(session->consumer);
2188
2189 switch (domain) {
2190 case LTTNG_DOMAIN_KERNEL:
2191 DBG3("Copying tracing session consumer output in kernel session");
2192 /*
2193 * XXX: We should audit the session creation and what this function
2194 * does "extra" in order to avoid a destroy since this function is used
2195 * in the domain session creation (kernel and ust) only. Same for UST
2196 * domain.
2197 */
2198 if (session->kernel_session->consumer) {
2199 consumer_destroy_output(session->kernel_session->consumer);
2200 }
2201 session->kernel_session->consumer =
2202 consumer_copy_output(session->consumer);
2203 /* Ease our life a bit for the next part */
2204 consumer = session->kernel_session->consumer;
2205 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2206 break;
2207 case LTTNG_DOMAIN_UST:
2208 DBG3("Copying tracing session consumer output in UST session");
2209 if (session->ust_session->consumer) {
2210 consumer_destroy_output(session->ust_session->consumer);
2211 }
2212 session->ust_session->consumer =
2213 consumer_copy_output(session->consumer);
2214 /* Ease our life a bit for the next part */
2215 consumer = session->ust_session->consumer;
2216 dir_name = DEFAULT_UST_TRACE_DIR;
2217 break;
2218 default:
2219 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2220 goto error;
2221 }
2222
2223 /* Append correct directory to subdir */
2224 strncat(consumer->subdir, dir_name,
2225 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2226 DBG3("Copy session consumer subdir %s", consumer->subdir);
2227
2228 ret = LTTNG_OK;
2229
2230 error:
2231 return ret;
2232 }
2233
2234 /*
2235 * Create an UST session and add it to the session ust list.
2236 *
2237 * Should *NOT* be called with RCU read-side lock held.
2238 */
2239 static int create_ust_session(struct ltt_session *session,
2240 struct lttng_domain *domain)
2241 {
2242 int ret;
2243 struct ltt_ust_session *lus = NULL;
2244
2245 assert(session);
2246 assert(domain);
2247 assert(session->consumer);
2248
2249 switch (domain->type) {
2250 case LTTNG_DOMAIN_UST:
2251 break;
2252 default:
2253 ERR("Unknown UST domain on create session %d", domain->type);
2254 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2255 goto error;
2256 }
2257
2258 DBG("Creating UST session");
2259
2260 lus = trace_ust_create_session(session->id);
2261 if (lus == NULL) {
2262 ret = LTTNG_ERR_UST_SESS_FAIL;
2263 goto error;
2264 }
2265
2266 lus->uid = session->uid;
2267 lus->gid = session->gid;
2268 session->ust_session = lus;
2269
2270 /* Copy session output to the newly created UST session */
2271 ret = copy_session_consumer(domain->type, session);
2272 if (ret != LTTNG_OK) {
2273 goto error;
2274 }
2275
2276 return LTTNG_OK;
2277
2278 error:
2279 free(lus);
2280 session->ust_session = NULL;
2281 return ret;
2282 }
2283
2284 /*
2285 * Create a kernel tracer session then create the default channel.
2286 */
2287 static int create_kernel_session(struct ltt_session *session)
2288 {
2289 int ret;
2290
2291 DBG("Creating kernel session");
2292
2293 ret = kernel_create_session(session, kernel_tracer_fd);
2294 if (ret < 0) {
2295 ret = LTTNG_ERR_KERN_SESS_FAIL;
2296 goto error;
2297 }
2298
2299 /* Code flow safety */
2300 assert(session->kernel_session);
2301
2302 /* Copy session output to the newly created Kernel session */
2303 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2304 if (ret != LTTNG_OK) {
2305 goto error;
2306 }
2307
2308 /* Create directory(ies) on local filesystem. */
2309 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2310 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2311 ret = run_as_mkdir_recursive(
2312 session->kernel_session->consumer->dst.trace_path,
2313 S_IRWXU | S_IRWXG, session->uid, session->gid);
2314 if (ret < 0) {
2315 if (ret != -EEXIST) {
2316 ERR("Trace directory creation error");
2317 goto error;
2318 }
2319 }
2320 }
2321
2322 session->kernel_session->uid = session->uid;
2323 session->kernel_session->gid = session->gid;
2324
2325 return LTTNG_OK;
2326
2327 error:
2328 trace_kernel_destroy_session(session->kernel_session);
2329 session->kernel_session = NULL;
2330 return ret;
2331 }
2332
2333 /*
2334 * Count number of session permitted by uid/gid.
2335 */
2336 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2337 {
2338 unsigned int i = 0;
2339 struct ltt_session *session;
2340
2341 DBG("Counting number of available session for UID %d GID %d",
2342 uid, gid);
2343 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2344 /*
2345 * Only list the sessions the user can control.
2346 */
2347 if (!session_access_ok(session, uid, gid)) {
2348 continue;
2349 }
2350 i++;
2351 }
2352 return i;
2353 }
2354
2355 /*
2356 * Process the command requested by the lttng client within the command
2357 * context structure. This function make sure that the return structure (llm)
2358 * is set and ready for transmission before returning.
2359 *
2360 * Return any error encountered or 0 for success.
2361 *
2362 * "sock" is only used for special-case var. len data.
2363 *
2364 * Should *NOT* be called with RCU read-side lock held.
2365 */
2366 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2367 int *sock_error)
2368 {
2369 int ret = LTTNG_OK;
2370 int need_tracing_session = 1;
2371 int need_domain;
2372
2373 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2374
2375 *sock_error = 0;
2376
2377 switch (cmd_ctx->lsm->cmd_type) {
2378 case LTTNG_CREATE_SESSION:
2379 case LTTNG_DESTROY_SESSION:
2380 case LTTNG_LIST_SESSIONS:
2381 case LTTNG_LIST_DOMAINS:
2382 case LTTNG_START_TRACE:
2383 case LTTNG_STOP_TRACE:
2384 case LTTNG_DATA_PENDING:
2385 need_domain = 0;
2386 break;
2387 default:
2388 need_domain = 1;
2389 }
2390
2391 if (opt_no_kernel && need_domain
2392 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2393 if (!is_root) {
2394 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2395 } else {
2396 ret = LTTNG_ERR_KERN_NA;
2397 }
2398 goto error;
2399 }
2400
2401 /* Deny register consumer if we already have a spawned consumer. */
2402 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2403 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2404 if (kconsumer_data.pid > 0) {
2405 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2406 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2407 goto error;
2408 }
2409 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2410 }
2411
2412 /*
2413 * Check for command that don't needs to allocate a returned payload. We do
2414 * this here so we don't have to make the call for no payload at each
2415 * command.
2416 */
2417 switch(cmd_ctx->lsm->cmd_type) {
2418 case LTTNG_LIST_SESSIONS:
2419 case LTTNG_LIST_TRACEPOINTS:
2420 case LTTNG_LIST_TRACEPOINT_FIELDS:
2421 case LTTNG_LIST_DOMAINS:
2422 case LTTNG_LIST_CHANNELS:
2423 case LTTNG_LIST_EVENTS:
2424 break;
2425 default:
2426 /* Setup lttng message with no payload */
2427 ret = setup_lttng_msg(cmd_ctx, 0);
2428 if (ret < 0) {
2429 /* This label does not try to unlock the session */
2430 goto init_setup_error;
2431 }
2432 }
2433
2434 /* Commands that DO NOT need a session. */
2435 switch (cmd_ctx->lsm->cmd_type) {
2436 case LTTNG_CREATE_SESSION:
2437 case LTTNG_CALIBRATE:
2438 case LTTNG_LIST_SESSIONS:
2439 case LTTNG_LIST_TRACEPOINTS:
2440 case LTTNG_LIST_TRACEPOINT_FIELDS:
2441 need_tracing_session = 0;
2442 break;
2443 default:
2444 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2445 /*
2446 * We keep the session list lock across _all_ commands
2447 * for now, because the per-session lock does not
2448 * handle teardown properly.
2449 */
2450 session_lock_list();
2451 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2452 if (cmd_ctx->session == NULL) {
2453 if (cmd_ctx->lsm->session.name != NULL) {
2454 ret = LTTNG_ERR_SESS_NOT_FOUND;
2455 } else {
2456 /* If no session name specified */
2457 ret = LTTNG_ERR_SELECT_SESS;
2458 }
2459 goto error;
2460 } else {
2461 /* Acquire lock for the session */
2462 session_lock(cmd_ctx->session);
2463 }
2464 break;
2465 }
2466
2467 if (!need_domain) {
2468 goto skip_domain;
2469 }
2470
2471 /*
2472 * Check domain type for specific "pre-action".
2473 */
2474 switch (cmd_ctx->lsm->domain.type) {
2475 case LTTNG_DOMAIN_KERNEL:
2476 if (!is_root) {
2477 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2478 goto error;
2479 }
2480
2481 /* Kernel tracer check */
2482 if (kernel_tracer_fd == -1) {
2483 /* Basically, load kernel tracer modules */
2484 ret = init_kernel_tracer();
2485 if (ret != 0) {
2486 goto error;
2487 }
2488 }
2489
2490 /* Consumer is in an ERROR state. Report back to client */
2491 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2492 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2493 goto error;
2494 }
2495
2496 /* Need a session for kernel command */
2497 if (need_tracing_session) {
2498 if (cmd_ctx->session->kernel_session == NULL) {
2499 ret = create_kernel_session(cmd_ctx->session);
2500 if (ret < 0) {
2501 ret = LTTNG_ERR_KERN_SESS_FAIL;
2502 goto error;
2503 }
2504 }
2505
2506 /* Start the kernel consumer daemon */
2507 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2508 if (kconsumer_data.pid == 0 &&
2509 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2510 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2511 ret = start_consumerd(&kconsumer_data);
2512 if (ret < 0) {
2513 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2514 goto error;
2515 }
2516 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2517 } else {
2518 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2519 }
2520
2521 /*
2522 * The consumer was just spawned so we need to add the socket to
2523 * the consumer output of the session if exist.
2524 */
2525 ret = consumer_create_socket(&kconsumer_data,
2526 cmd_ctx->session->kernel_session->consumer);
2527 if (ret < 0) {
2528 goto error;
2529 }
2530 }
2531
2532 break;
2533 case LTTNG_DOMAIN_UST:
2534 {
2535 /* Consumer is in an ERROR state. Report back to client */
2536 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2537 ret = LTTNG_ERR_NO_USTCONSUMERD;
2538 goto error;
2539 }
2540
2541 if (need_tracing_session) {
2542 /* Create UST session if none exist. */
2543 if (cmd_ctx->session->ust_session == NULL) {
2544 ret = create_ust_session(cmd_ctx->session,
2545 &cmd_ctx->lsm->domain);
2546 if (ret != LTTNG_OK) {
2547 goto error;
2548 }
2549 }
2550
2551 /* Start the UST consumer daemons */
2552 /* 64-bit */
2553 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2554 if (consumerd64_bin[0] != '\0' &&
2555 ustconsumer64_data.pid == 0 &&
2556 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2557 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2558 ret = start_consumerd(&ustconsumer64_data);
2559 if (ret < 0) {
2560 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2561 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2562 goto error;
2563 }
2564
2565 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2566 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2567 } else {
2568 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2569 }
2570
2571 /*
2572 * Setup socket for consumer 64 bit. No need for atomic access
2573 * since it was set above and can ONLY be set in this thread.
2574 */
2575 ret = consumer_create_socket(&ustconsumer64_data,
2576 cmd_ctx->session->ust_session->consumer);
2577 if (ret < 0) {
2578 goto error;
2579 }
2580
2581 /* 32-bit */
2582 if (consumerd32_bin[0] != '\0' &&
2583 ustconsumer32_data.pid == 0 &&
2584 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2585 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2586 ret = start_consumerd(&ustconsumer32_data);
2587 if (ret < 0) {
2588 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2589 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2590 goto error;
2591 }
2592
2593 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2594 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2595 } else {
2596 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2597 }
2598
2599 /*
2600 * Setup socket for consumer 64 bit. No need for atomic access
2601 * since it was set above and can ONLY be set in this thread.
2602 */
2603 ret = consumer_create_socket(&ustconsumer32_data,
2604 cmd_ctx->session->ust_session->consumer);
2605 if (ret < 0) {
2606 goto error;
2607 }
2608 }
2609 break;
2610 }
2611 default:
2612 break;
2613 }
2614 skip_domain:
2615
2616 /* Validate consumer daemon state when start/stop trace command */
2617 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2618 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2619 switch (cmd_ctx->lsm->domain.type) {
2620 case LTTNG_DOMAIN_UST:
2621 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2622 ret = LTTNG_ERR_NO_USTCONSUMERD;
2623 goto error;
2624 }
2625 break;
2626 case LTTNG_DOMAIN_KERNEL:
2627 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2628 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2629 goto error;
2630 }
2631 break;
2632 }
2633 }
2634
2635 /*
2636 * Check that the UID or GID match that of the tracing session.
2637 * The root user can interact with all sessions.
2638 */
2639 if (need_tracing_session) {
2640 if (!session_access_ok(cmd_ctx->session,
2641 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2642 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2643 ret = LTTNG_ERR_EPERM;
2644 goto error;
2645 }
2646 }
2647
2648 /*
2649 * Send relayd information to consumer as soon as we have a domain and a
2650 * session defined.
2651 */
2652 if (cmd_ctx->session && need_domain) {
2653 /*
2654 * Setup relayd if not done yet. If the relayd information was already
2655 * sent to the consumer, this call will gracefully return.
2656 */
2657 ret = cmd_setup_relayd(cmd_ctx->session);
2658 if (ret != LTTNG_OK) {
2659 goto error;
2660 }
2661 }
2662
2663 /* Process by command type */
2664 switch (cmd_ctx->lsm->cmd_type) {
2665 case LTTNG_ADD_CONTEXT:
2666 {
2667 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2668 cmd_ctx->lsm->u.context.channel_name,
2669 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2670 break;
2671 }
2672 case LTTNG_DISABLE_CHANNEL:
2673 {
2674 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2675 cmd_ctx->lsm->u.disable.channel_name);
2676 break;
2677 }
2678 case LTTNG_DISABLE_EVENT:
2679 {
2680 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2681 cmd_ctx->lsm->u.disable.channel_name,
2682 cmd_ctx->lsm->u.disable.name);
2683 break;
2684 }
2685 case LTTNG_DISABLE_ALL_EVENT:
2686 {
2687 DBG("Disabling all events");
2688
2689 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2690 cmd_ctx->lsm->u.disable.channel_name);
2691 break;
2692 }
2693 case LTTNG_ENABLE_CHANNEL:
2694 {
2695 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2696 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2697 break;
2698 }
2699 case LTTNG_ENABLE_EVENT:
2700 {
2701 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2702 cmd_ctx->lsm->u.enable.channel_name,
2703 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2704 break;
2705 }
2706 case LTTNG_ENABLE_ALL_EVENT:
2707 {
2708 DBG("Enabling all events");
2709
2710 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2711 cmd_ctx->lsm->u.enable.channel_name,
2712 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2713 break;
2714 }
2715 case LTTNG_LIST_TRACEPOINTS:
2716 {
2717 struct lttng_event *events;
2718 ssize_t nb_events;
2719
2720 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2721 if (nb_events < 0) {
2722 /* Return value is a negative lttng_error_code. */
2723 ret = -nb_events;
2724 goto error;
2725 }
2726
2727 /*
2728 * Setup lttng message with payload size set to the event list size in
2729 * bytes and then copy list into the llm payload.
2730 */
2731 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2732 if (ret < 0) {
2733 free(events);
2734 goto setup_error;
2735 }
2736
2737 /* Copy event list into message payload */
2738 memcpy(cmd_ctx->llm->payload, events,
2739 sizeof(struct lttng_event) * nb_events);
2740
2741 free(events);
2742
2743 ret = LTTNG_OK;
2744 break;
2745 }
2746 case LTTNG_LIST_TRACEPOINT_FIELDS:
2747 {
2748 struct lttng_event_field *fields;
2749 ssize_t nb_fields;
2750
2751 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2752 &fields);
2753 if (nb_fields < 0) {
2754 /* Return value is a negative lttng_error_code. */
2755 ret = -nb_fields;
2756 goto error;
2757 }
2758
2759 /*
2760 * Setup lttng message with payload size set to the event list size in
2761 * bytes and then copy list into the llm payload.
2762 */
2763 ret = setup_lttng_msg(cmd_ctx,
2764 sizeof(struct lttng_event_field) * nb_fields);
2765 if (ret < 0) {
2766 free(fields);
2767 goto setup_error;
2768 }
2769
2770 /* Copy event list into message payload */
2771 memcpy(cmd_ctx->llm->payload, fields,
2772 sizeof(struct lttng_event_field) * nb_fields);
2773
2774 free(fields);
2775
2776 ret = LTTNG_OK;
2777 break;
2778 }
2779 case LTTNG_SET_CONSUMER_URI:
2780 {
2781 size_t nb_uri, len;
2782 struct lttng_uri *uris;
2783
2784 nb_uri = cmd_ctx->lsm->u.uri.size;
2785 len = nb_uri * sizeof(struct lttng_uri);
2786
2787 if (nb_uri == 0) {
2788 ret = LTTNG_ERR_INVALID;
2789 goto error;
2790 }
2791
2792 uris = zmalloc(len);
2793 if (uris == NULL) {
2794 ret = LTTNG_ERR_FATAL;
2795 goto error;
2796 }
2797
2798 /* Receive variable len data */
2799 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2800 ret = lttcomm_recv_unix_sock(sock, uris, len);
2801 if (ret <= 0) {
2802 DBG("No URIs received from client... continuing");
2803 *sock_error = 1;
2804 ret = LTTNG_ERR_SESSION_FAIL;
2805 free(uris);
2806 goto error;
2807 }
2808
2809 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2810 nb_uri, uris);
2811 if (ret != LTTNG_OK) {
2812 free(uris);
2813 goto error;
2814 }
2815
2816 /*
2817 * XXX: 0 means that this URI should be applied on the session. Should
2818 * be a DOMAIN enuam.
2819 */
2820 if (cmd_ctx->lsm->domain.type == 0) {
2821 /* Add the URI for the UST session if a consumer is present. */
2822 if (cmd_ctx->session->ust_session &&
2823 cmd_ctx->session->ust_session->consumer) {
2824 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2825 nb_uri, uris);
2826 } else if (cmd_ctx->session->kernel_session &&
2827 cmd_ctx->session->kernel_session->consumer) {
2828 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2829 cmd_ctx->session, nb_uri, uris);
2830 }
2831 }
2832
2833 free(uris);
2834
2835 break;
2836 }
2837 case LTTNG_START_TRACE:
2838 {
2839 ret = cmd_start_trace(cmd_ctx->session);
2840 break;
2841 }
2842 case LTTNG_STOP_TRACE:
2843 {
2844 ret = cmd_stop_trace(cmd_ctx->session);
2845 break;
2846 }
2847 case LTTNG_CREATE_SESSION:
2848 {
2849 size_t nb_uri, len;
2850 struct lttng_uri *uris = NULL;
2851
2852 nb_uri = cmd_ctx->lsm->u.uri.size;
2853 len = nb_uri * sizeof(struct lttng_uri);
2854
2855 if (nb_uri > 0) {
2856 uris = zmalloc(len);
2857 if (uris == NULL) {
2858 ret = LTTNG_ERR_FATAL;
2859 goto error;
2860 }
2861
2862 /* Receive variable len data */
2863 DBG("Waiting for %zu URIs from client ...", nb_uri);
2864 ret = lttcomm_recv_unix_sock(sock, uris, len);
2865 if (ret <= 0) {
2866 DBG("No URIs received from client... continuing");
2867 *sock_error = 1;
2868 ret = LTTNG_ERR_SESSION_FAIL;
2869 free(uris);
2870 goto error;
2871 }
2872
2873 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2874 DBG("Creating session with ONE network URI is a bad call");
2875 ret = LTTNG_ERR_SESSION_FAIL;
2876 free(uris);
2877 goto error;
2878 }
2879 }
2880
2881 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2882 &cmd_ctx->creds);
2883
2884 free(uris);
2885
2886 break;
2887 }
2888 case LTTNG_DESTROY_SESSION:
2889 {
2890 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2891
2892 /* Set session to NULL so we do not unlock it after free. */
2893 cmd_ctx->session = NULL;
2894 break;
2895 }
2896 case LTTNG_LIST_DOMAINS:
2897 {
2898 ssize_t nb_dom;
2899 struct lttng_domain *domains;
2900
2901 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2902 if (nb_dom < 0) {
2903 /* Return value is a negative lttng_error_code. */
2904 ret = -nb_dom;
2905 goto error;
2906 }
2907
2908 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2909 if (ret < 0) {
2910 free(domains);
2911 goto setup_error;
2912 }
2913
2914 /* Copy event list into message payload */
2915 memcpy(cmd_ctx->llm->payload, domains,
2916 nb_dom * sizeof(struct lttng_domain));
2917
2918 free(domains);
2919
2920 ret = LTTNG_OK;
2921 break;
2922 }
2923 case LTTNG_LIST_CHANNELS:
2924 {
2925 int nb_chan;
2926 struct lttng_channel *channels;
2927
2928 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2929 cmd_ctx->session, &channels);
2930 if (nb_chan < 0) {
2931 /* Return value is a negative lttng_error_code. */
2932 ret = -nb_chan;
2933 goto error;
2934 }
2935
2936 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2937 if (ret < 0) {
2938 free(channels);
2939 goto setup_error;
2940 }
2941
2942 /* Copy event list into message payload */
2943 memcpy(cmd_ctx->llm->payload, channels,
2944 nb_chan * sizeof(struct lttng_channel));
2945
2946 free(channels);
2947
2948 ret = LTTNG_OK;
2949 break;
2950 }
2951 case LTTNG_LIST_EVENTS:
2952 {
2953 ssize_t nb_event;
2954 struct lttng_event *events = NULL;
2955
2956 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2957 cmd_ctx->lsm->u.list.channel_name, &events);
2958 if (nb_event < 0) {
2959 /* Return value is a negative lttng_error_code. */
2960 ret = -nb_event;
2961 goto error;
2962 }
2963
2964 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2965 if (ret < 0) {
2966 free(events);
2967 goto setup_error;
2968 }
2969
2970 /* Copy event list into message payload */
2971 memcpy(cmd_ctx->llm->payload, events,
2972 nb_event * sizeof(struct lttng_event));
2973
2974 free(events);
2975
2976 ret = LTTNG_OK;
2977 break;
2978 }
2979 case LTTNG_LIST_SESSIONS:
2980 {
2981 unsigned int nr_sessions;
2982
2983 session_lock_list();
2984 nr_sessions = lttng_sessions_count(
2985 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2986 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2987
2988 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2989 if (ret < 0) {
2990 session_unlock_list();
2991 goto setup_error;
2992 }
2993
2994 /* Filled the session array */
2995 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2996 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2997 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2998
2999 session_unlock_list();
3000
3001 ret = LTTNG_OK;
3002 break;
3003 }
3004 case LTTNG_CALIBRATE:
3005 {
3006 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3007 &cmd_ctx->lsm->u.calibrate);
3008 break;
3009 }
3010 case LTTNG_REGISTER_CONSUMER:
3011 {
3012 struct consumer_data *cdata;
3013
3014 switch (cmd_ctx->lsm->domain.type) {
3015 case LTTNG_DOMAIN_KERNEL:
3016 cdata = &kconsumer_data;
3017 break;
3018 default:
3019 ret = LTTNG_ERR_UND;
3020 goto error;
3021 }
3022
3023 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3024 cmd_ctx->lsm->u.reg.path, cdata);
3025 break;
3026 }
3027 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3028 {
3029 struct lttng_filter_bytecode *bytecode;
3030
3031 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3032 ret = LTTNG_ERR_FILTER_INVAL;
3033 goto error;
3034 }
3035 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3036 ret = LTTNG_ERR_FILTER_INVAL;
3037 goto error;
3038 }
3039 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3040 if (!bytecode) {
3041 ret = LTTNG_ERR_FILTER_NOMEM;
3042 goto error;
3043 }
3044 /* Receive var. len. data */
3045 DBG("Receiving var len data from client ...");
3046 ret = lttcomm_recv_unix_sock(sock, bytecode,
3047 cmd_ctx->lsm->u.enable.bytecode_len);
3048 if (ret <= 0) {
3049 DBG("Nothing recv() from client var len data... continuing");
3050 *sock_error = 1;
3051 ret = LTTNG_ERR_FILTER_INVAL;
3052 goto error;
3053 }
3054
3055 if (bytecode->len + sizeof(*bytecode)
3056 != cmd_ctx->lsm->u.enable.bytecode_len) {
3057 free(bytecode);
3058 ret = LTTNG_ERR_FILTER_INVAL;
3059 goto error;
3060 }
3061
3062 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3063 cmd_ctx->lsm->u.enable.channel_name,
3064 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3065 break;
3066 }
3067 case LTTNG_DATA_PENDING:
3068 {
3069 ret = cmd_data_pending(cmd_ctx->session);
3070 break;
3071 }
3072 default:
3073 ret = LTTNG_ERR_UND;
3074 break;
3075 }
3076
3077 error:
3078 if (cmd_ctx->llm == NULL) {
3079 DBG("Missing llm structure. Allocating one.");
3080 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3081 goto setup_error;
3082 }
3083 }
3084 /* Set return code */
3085 cmd_ctx->llm->ret_code = ret;
3086 setup_error:
3087 if (cmd_ctx->session) {
3088 session_unlock(cmd_ctx->session);
3089 }
3090 if (need_tracing_session) {
3091 session_unlock_list();
3092 }
3093 init_setup_error:
3094 return ret;
3095 }
3096
3097 /*
3098 * Thread managing health check socket.
3099 */
3100 static void *thread_manage_health(void *data)
3101 {
3102 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3103 uint32_t revents, nb_fd;
3104 struct lttng_poll_event events;
3105 struct lttcomm_health_msg msg;
3106 struct lttcomm_health_data reply;
3107
3108 DBG("[thread] Manage health check started");
3109
3110 rcu_register_thread();
3111
3112 /* We might hit an error path before this is created. */
3113 lttng_poll_init(&events);
3114
3115 /* Create unix socket */
3116 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3117 if (sock < 0) {
3118 ERR("Unable to create health check Unix socket");
3119 ret = -1;
3120 goto error;
3121 }
3122
3123 /*
3124 * Set the CLOEXEC flag. Return code is useless because either way, the
3125 * show must go on.
3126 */
3127 (void) utils_set_fd_cloexec(sock);
3128
3129 ret = lttcomm_listen_unix_sock(sock);
3130 if (ret < 0) {
3131 goto error;
3132 }
3133
3134 /*
3135 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3136 * more will be added to this poll set.
3137 */
3138 ret = sessiond_set_thread_pollset(&events, 2);
3139 if (ret < 0) {
3140 goto error;
3141 }
3142
3143 /* Add the application registration socket */
3144 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3145 if (ret < 0) {
3146 goto error;
3147 }
3148
3149 while (1) {
3150 DBG("Health check ready");
3151
3152 /* Inifinite blocking call, waiting for transmission */
3153 restart:
3154 ret = lttng_poll_wait(&events, -1);
3155 if (ret < 0) {
3156 /*
3157 * Restart interrupted system call.
3158 */
3159 if (errno == EINTR) {
3160 goto restart;
3161 }
3162 goto error;
3163 }
3164
3165 nb_fd = ret;
3166
3167 for (i = 0; i < nb_fd; i++) {
3168 /* Fetch once the poll data */
3169 revents = LTTNG_POLL_GETEV(&events, i);
3170 pollfd = LTTNG_POLL_GETFD(&events, i);
3171
3172 /* Thread quit pipe has been closed. Killing thread. */
3173 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3174 if (ret) {
3175 err = 0;
3176 goto exit;
3177 }
3178
3179 /* Event on the registration socket */
3180 if (pollfd == sock) {
3181 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3182 ERR("Health socket poll error");
3183 goto error;
3184 }
3185 }
3186 }
3187
3188 new_sock = lttcomm_accept_unix_sock(sock);
3189 if (new_sock < 0) {
3190 goto error;
3191 }
3192
3193 /*
3194 * Set the CLOEXEC flag. Return code is useless because either way, the
3195 * show must go on.
3196 */
3197 (void) utils_set_fd_cloexec(new_sock);
3198
3199 DBG("Receiving data from client for health...");
3200 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3201 if (ret <= 0) {
3202 DBG("Nothing recv() from client... continuing");
3203 ret = close(new_sock);
3204 if (ret) {
3205 PERROR("close");
3206 }
3207 new_sock = -1;
3208 continue;
3209 }
3210
3211 rcu_thread_online();
3212
3213 switch (msg.component) {
3214 case LTTNG_HEALTH_CMD:
3215 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3216 break;
3217 case LTTNG_HEALTH_APP_MANAGE:
3218 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3219 break;
3220 case LTTNG_HEALTH_APP_REG:
3221 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3222 break;
3223 case LTTNG_HEALTH_KERNEL:
3224 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3225 break;
3226 case LTTNG_HEALTH_CONSUMER:
3227 reply.ret_code = check_consumer_health();
3228 break;
3229 case LTTNG_HEALTH_HT_CLEANUP:
3230 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3231 break;
3232 case LTTNG_HEALTH_ALL:
3233 reply.ret_code =
3234 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3235 health_check_state(HEALTH_TYPE_APP_REG) &&
3236 health_check_state(HEALTH_TYPE_CMD) &&
3237 health_check_state(HEALTH_TYPE_KERNEL) &&
3238 check_consumer_health() &&
3239 health_check_state(HEALTH_TYPE_HT_CLEANUP);
3240 break;
3241 default:
3242 reply.ret_code = LTTNG_ERR_UND;
3243 break;
3244 }
3245
3246 /*
3247 * Flip ret value since 0 is a success and 1 indicates a bad health for
3248 * the client where in the sessiond it is the opposite. Again, this is
3249 * just to make things easier for us poor developer which enjoy a lot
3250 * lazyness.
3251 */
3252 if (reply.ret_code == 0 || reply.ret_code == 1) {
3253 reply.ret_code = !reply.ret_code;
3254 }
3255
3256 DBG2("Health check return value %d", reply.ret_code);
3257
3258 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3259 if (ret < 0) {
3260 ERR("Failed to send health data back to client");
3261 }
3262
3263 /* End of transmission */
3264 ret = close(new_sock);
3265 if (ret) {
3266 PERROR("close");
3267 }
3268 new_sock = -1;
3269 }
3270
3271 exit:
3272 error:
3273 if (err) {
3274 ERR("Health error occurred in %s", __func__);
3275 }
3276 DBG("Health check thread dying");
3277 unlink(health_unix_sock_path);
3278 if (sock >= 0) {
3279 ret = close(sock);
3280 if (ret) {
3281 PERROR("close");
3282 }
3283 }
3284
3285 lttng_poll_clean(&events);
3286
3287 rcu_unregister_thread();
3288 return NULL;
3289 }
3290
3291 /*
3292 * This thread manage all clients request using the unix client socket for
3293 * communication.
3294 */
3295 static void *thread_manage_clients(void *data)
3296 {
3297 int sock = -1, ret, i, pollfd, err = -1;
3298 int sock_error;
3299 uint32_t revents, nb_fd;
3300 struct command_ctx *cmd_ctx = NULL;
3301 struct lttng_poll_event events;
3302
3303 DBG("[thread] Manage client started");
3304
3305 rcu_register_thread();
3306
3307 health_register(HEALTH_TYPE_CMD);
3308
3309 if (testpoint(thread_manage_clients)) {
3310 goto error_testpoint;
3311 }
3312
3313 health_code_update();
3314
3315 ret = lttcomm_listen_unix_sock(client_sock);
3316 if (ret < 0) {
3317 goto error_listen;
3318 }
3319
3320 /*
3321 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3322 * more will be added to this poll set.
3323 */
3324 ret = sessiond_set_thread_pollset(&events, 2);
3325 if (ret < 0) {
3326 goto error_create_poll;
3327 }
3328
3329 /* Add the application registration socket */
3330 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3331 if (ret < 0) {
3332 goto error;
3333 }
3334
3335 /*
3336 * Notify parent pid that we are ready to accept command for client side.
3337 */
3338 if (opt_sig_parent) {
3339 kill(ppid, SIGUSR1);
3340 }
3341
3342 if (testpoint(thread_manage_clients_before_loop)) {
3343 goto error;
3344 }
3345
3346 health_code_update();
3347
3348 while (1) {
3349 DBG("Accepting client command ...");
3350
3351 /* Inifinite blocking call, waiting for transmission */
3352 restart:
3353 health_poll_entry();
3354 ret = lttng_poll_wait(&events, -1);
3355 health_poll_exit();
3356 if (ret < 0) {
3357 /*
3358 * Restart interrupted system call.
3359 */
3360 if (errno == EINTR) {
3361 goto restart;
3362 }
3363 goto error;
3364 }
3365
3366 nb_fd = ret;
3367
3368 for (i = 0; i < nb_fd; i++) {
3369 /* Fetch once the poll data */
3370 revents = LTTNG_POLL_GETEV(&events, i);
3371 pollfd = LTTNG_POLL_GETFD(&events, i);
3372
3373 health_code_update();
3374
3375 /* Thread quit pipe has been closed. Killing thread. */
3376 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3377 if (ret) {
3378 err = 0;
3379 goto exit;
3380 }
3381
3382 /* Event on the registration socket */
3383 if (pollfd == client_sock) {
3384 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3385 ERR("Client socket poll error");
3386 goto error;
3387 }
3388 }
3389 }
3390
3391 DBG("Wait for client response");
3392
3393 health_code_update();
3394
3395 sock = lttcomm_accept_unix_sock(client_sock);
3396 if (sock < 0) {
3397 goto error;
3398 }
3399
3400 /*
3401 * Set the CLOEXEC flag. Return code is useless because either way, the
3402 * show must go on.
3403 */
3404 (void) utils_set_fd_cloexec(sock);
3405
3406 /* Set socket option for credentials retrieval */
3407 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3408 if (ret < 0) {
3409 goto error;
3410 }
3411
3412 /* Allocate context command to process the client request */
3413 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3414 if (cmd_ctx == NULL) {
3415 PERROR("zmalloc cmd_ctx");
3416 goto error;
3417 }
3418
3419 /* Allocate data buffer for reception */
3420 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3421 if (cmd_ctx->lsm == NULL) {
3422 PERROR("zmalloc cmd_ctx->lsm");
3423 goto error;
3424 }
3425
3426 cmd_ctx->llm = NULL;
3427 cmd_ctx->session = NULL;
3428
3429 health_code_update();
3430
3431 /*
3432 * Data is received from the lttng client. The struct
3433 * lttcomm_session_msg (lsm) contains the command and data request of
3434 * the client.
3435 */
3436 DBG("Receiving data from client ...");
3437 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3438 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3439 if (ret <= 0) {
3440 DBG("Nothing recv() from client... continuing");
3441 ret = close(sock);
3442 if (ret) {
3443 PERROR("close");
3444 }
3445 sock = -1;
3446 clean_command_ctx(&cmd_ctx);
3447 continue;
3448 }
3449
3450 health_code_update();
3451
3452 // TODO: Validate cmd_ctx including sanity check for
3453 // security purpose.
3454
3455 rcu_thread_online();
3456 /*
3457 * This function dispatch the work to the kernel or userspace tracer
3458 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3459 * informations for the client. The command context struct contains
3460 * everything this function may needs.
3461 */
3462 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3463 rcu_thread_offline();
3464 if (ret < 0) {
3465 ret = close(sock);
3466 if (ret) {
3467 PERROR("close");
3468 }
3469 sock = -1;
3470 /*
3471 * TODO: Inform client somehow of the fatal error. At
3472 * this point, ret < 0 means that a zmalloc failed
3473 * (ENOMEM). Error detected but still accept
3474 * command, unless a socket error has been
3475 * detected.
3476 */
3477 clean_command_ctx(&cmd_ctx);
3478 continue;
3479 }
3480
3481 health_code_update();
3482
3483 DBG("Sending response (size: %d, retcode: %s)",
3484 cmd_ctx->lttng_msg_size,
3485 lttng_strerror(-cmd_ctx->llm->ret_code));
3486 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3487 if (ret < 0) {
3488 ERR("Failed to send data back to client");
3489 }
3490
3491 /* End of transmission */
3492 ret = close(sock);
3493 if (ret) {
3494 PERROR("close");
3495 }
3496 sock = -1;
3497
3498 clean_command_ctx(&cmd_ctx);
3499
3500 health_code_update();
3501 }
3502
3503 exit:
3504 error:
3505 if (sock >= 0) {
3506 ret = close(sock);
3507 if (ret) {
3508 PERROR("close");
3509 }
3510 }
3511
3512 lttng_poll_clean(&events);
3513 clean_command_ctx(&cmd_ctx);
3514
3515 error_listen:
3516 error_create_poll:
3517 error_testpoint:
3518 unlink(client_unix_sock_path);
3519 if (client_sock >= 0) {
3520 ret = close(client_sock);
3521 if (ret) {
3522 PERROR("close");
3523 }
3524 }
3525
3526 if (err) {
3527 health_error();
3528 ERR("Health error occurred in %s", __func__);
3529 }
3530
3531 health_unregister();
3532
3533 DBG("Client thread dying");
3534
3535 rcu_unregister_thread();
3536 return NULL;
3537 }
3538
3539
3540 /*
3541 * usage function on stderr
3542 */
3543 static void usage(void)
3544 {
3545 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3546 fprintf(stderr, " -h, --help Display this usage.\n");
3547 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3548 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3549 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3550 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3551 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3552 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3553 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3554 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3555 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3556 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3557 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3558 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3559 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3560 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3561 fprintf(stderr, " -V, --version Show version number.\n");
3562 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3563 fprintf(stderr, " -q, --quiet No output at all.\n");
3564 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3565 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3566 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3567 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3568 }
3569
3570 /*
3571 * daemon argument parsing
3572 */
3573 static int parse_args(int argc, char **argv)
3574 {
3575 int c;
3576
3577 static struct option long_options[] = {
3578 { "client-sock", 1, 0, 'c' },
3579 { "apps-sock", 1, 0, 'a' },
3580 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3581 { "kconsumerd-err-sock", 1, 0, 'E' },
3582 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3583 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3584 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3585 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3586 { "consumerd32-path", 1, 0, 'u' },
3587 { "consumerd32-libdir", 1, 0, 'U' },
3588 { "consumerd64-path", 1, 0, 't' },
3589 { "consumerd64-libdir", 1, 0, 'T' },
3590 { "daemonize", 0, 0, 'd' },
3591 { "sig-parent", 0, 0, 'S' },
3592 { "help", 0, 0, 'h' },
3593 { "group", 1, 0, 'g' },
3594 { "version", 0, 0, 'V' },
3595 { "quiet", 0, 0, 'q' },
3596 { "verbose", 0, 0, 'v' },
3597 { "verbose-consumer", 0, 0, 'Z' },
3598 { "no-kernel", 0, 0, 'N' },
3599 { "pidfile", 1, 0, 'p' },
3600 { NULL, 0, 0, 0 }
3601 };
3602
3603 while (1) {
3604 int option_index = 0;
3605 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3606 long_options, &option_index);
3607 if (c == -1) {
3608 break;
3609 }
3610
3611 switch (c) {
3612 case 0:
3613 fprintf(stderr, "option %s", long_options[option_index].name);
3614 if (optarg) {
3615 fprintf(stderr, " with arg %s\n", optarg);
3616 }
3617 break;
3618 case 'c':
3619 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3620 break;
3621 case 'a':
3622 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3623 break;
3624 case 'd':
3625 opt_daemon = 1;
3626 break;
3627 case 'g':
3628 opt_tracing_group = optarg;
3629 break;
3630 case 'h':
3631 usage();
3632 exit(EXIT_FAILURE);
3633 case 'V':
3634 fprintf(stdout, "%s\n", VERSION);
3635 exit(EXIT_SUCCESS);
3636 case 'S':
3637 opt_sig_parent = 1;
3638 break;
3639 case 'E':
3640 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3641 break;
3642 case 'C':
3643 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3644 break;
3645 case 'F':
3646 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3647 break;
3648 case 'D':
3649 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3650 break;
3651 case 'H':
3652 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3653 break;
3654 case 'G':
3655 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3656 break;
3657 case 'N':
3658 opt_no_kernel = 1;
3659 break;
3660 case 'q':
3661 lttng_opt_quiet = 1;
3662 break;
3663 case 'v':
3664 /* Verbose level can increase using multiple -v */
3665 lttng_opt_verbose += 1;
3666 break;
3667 case 'Z':
3668 opt_verbose_consumer += 1;
3669 break;
3670 case 'u':
3671 consumerd32_bin= optarg;
3672 break;
3673 case 'U':
3674 consumerd32_libdir = optarg;
3675 break;
3676 case 't':
3677 consumerd64_bin = optarg;
3678 break;
3679 case 'T':
3680 consumerd64_libdir = optarg;
3681 break;
3682 case 'p':
3683 opt_pidfile = optarg;
3684 break;
3685 default:
3686 /* Unknown option or other error.
3687 * Error is printed by getopt, just return */
3688 return -1;
3689 }
3690 }
3691
3692 return 0;
3693 }
3694
3695 /*
3696 * Creates the two needed socket by the daemon.
3697 * apps_sock - The communication socket for all UST apps.
3698 * client_sock - The communication of the cli tool (lttng).
3699 */
3700 static int init_daemon_socket(void)
3701 {
3702 int ret = 0;
3703 mode_t old_umask;
3704
3705 old_umask = umask(0);
3706
3707 /* Create client tool unix socket */
3708 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3709 if (client_sock < 0) {
3710 ERR("Create unix sock failed: %s", client_unix_sock_path);
3711 ret = -1;
3712 goto end;
3713 }
3714
3715 /* Set the cloexec flag */
3716 ret = utils_set_fd_cloexec(client_sock);
3717 if (ret < 0) {
3718 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3719 "Continuing but note that the consumer daemon will have a "
3720 "reference to this socket on exec()", client_sock);
3721 }
3722
3723 /* File permission MUST be 660 */
3724 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3725 if (ret < 0) {
3726 ERR("Set file permissions failed: %s", client_unix_sock_path);
3727 PERROR("chmod");
3728 goto end;
3729 }
3730
3731 /* Create the application unix socket */
3732 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3733 if (apps_sock < 0) {
3734 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3735 ret = -1;
3736 goto end;
3737 }
3738
3739 /* Set the cloexec flag */
3740 ret = utils_set_fd_cloexec(apps_sock);
3741 if (ret < 0) {
3742 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3743 "Continuing but note that the consumer daemon will have a "
3744 "reference to this socket on exec()", apps_sock);
3745 }
3746
3747 /* File permission MUST be 666 */
3748 ret = chmod(apps_unix_sock_path,
3749 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3750 if (ret < 0) {
3751 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3752 PERROR("chmod");
3753 goto end;
3754 }
3755
3756 DBG3("Session daemon client socket %d and application socket %d created",
3757 client_sock, apps_sock);
3758
3759 end:
3760 umask(old_umask);
3761 return ret;
3762 }
3763
3764 /*
3765 * Check if the global socket is available, and if a daemon is answering at the
3766 * other side. If yes, error is returned.
3767 */
3768 static int check_existing_daemon(void)
3769 {
3770 /* Is there anybody out there ? */
3771 if (lttng_session_daemon_alive()) {
3772 return -EEXIST;
3773 }
3774
3775 return 0;
3776 }
3777
3778 /*
3779 * Set the tracing group gid onto the client socket.
3780 *
3781 * Race window between mkdir and chown is OK because we are going from more
3782 * permissive (root.root) to less permissive (root.tracing).
3783 */
3784 static int set_permissions(char *rundir)
3785 {
3786 int ret;
3787 gid_t gid;
3788
3789 ret = allowed_group();
3790 if (ret < 0) {
3791 WARN("No tracing group detected");
3792 ret = 0;
3793 goto end;
3794 }
3795
3796 gid = ret;
3797
3798 /* Set lttng run dir */
3799 ret = chown(rundir, 0, gid);
3800 if (ret < 0) {
3801 ERR("Unable to set group on %s", rundir);
3802 PERROR("chown");
3803 }
3804
3805 /* Ensure tracing group can search the run dir */
3806 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3807 if (ret < 0) {
3808 ERR("Unable to set permissions on %s", rundir);
3809 PERROR("chmod");
3810 }
3811
3812 /* lttng client socket path */
3813 ret = chown(client_unix_sock_path, 0, gid);
3814 if (ret < 0) {
3815 ERR("Unable to set group on %s", client_unix_sock_path);
3816 PERROR("chown");
3817 }
3818
3819 /* kconsumer error socket path */
3820 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3821 if (ret < 0) {
3822 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3823 PERROR("chown");
3824 }
3825
3826 /* 64-bit ustconsumer error socket path */
3827 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3828 if (ret < 0) {
3829 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3830 PERROR("chown");
3831 }
3832
3833 /* 32-bit ustconsumer compat32 error socket path */
3834 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3835 if (ret < 0) {
3836 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3837 PERROR("chown");
3838 }
3839
3840 DBG("All permissions are set");
3841
3842 end:
3843 return ret;
3844 }
3845
3846 /*
3847 * Create the lttng run directory needed for all global sockets and pipe.
3848 */
3849 static int create_lttng_rundir(const char *rundir)
3850 {
3851 int ret;
3852
3853 DBG3("Creating LTTng run directory: %s", rundir);
3854
3855 ret = mkdir(rundir, S_IRWXU);
3856 if (ret < 0) {
3857 if (errno != EEXIST) {
3858 ERR("Unable to create %s", rundir);
3859 goto error;
3860 } else {
3861 ret = 0;
3862 }
3863 }
3864
3865 error:
3866 return ret;
3867 }
3868
3869 /*
3870 * Setup sockets and directory needed by the kconsumerd communication with the
3871 * session daemon.
3872 */
3873 static int set_consumer_sockets(struct consumer_data *consumer_data,
3874 const char *rundir)
3875 {
3876 int ret;
3877 char path[PATH_MAX];
3878
3879 switch (consumer_data->type) {
3880 case LTTNG_CONSUMER_KERNEL:
3881 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3882 break;
3883 case LTTNG_CONSUMER64_UST:
3884 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3885 break;
3886 case LTTNG_CONSUMER32_UST:
3887 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3888 break;
3889 default:
3890 ERR("Consumer type unknown");
3891 ret = -EINVAL;
3892 goto error;
3893 }
3894
3895 DBG2("Creating consumer directory: %s", path);
3896
3897 ret = mkdir(path, S_IRWXU);
3898 if (ret < 0) {
3899 if (errno != EEXIST) {
3900 PERROR("mkdir");
3901 ERR("Failed to create %s", path);
3902 goto error;
3903 }
3904 ret = -1;
3905 }
3906
3907 /* Create the kconsumerd error unix socket */
3908 consumer_data->err_sock =
3909 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3910 if (consumer_data->err_sock < 0) {
3911 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3912 ret = -1;
3913 goto error;
3914 }
3915
3916 /* File permission MUST be 660 */
3917 ret = chmod(consumer_data->err_unix_sock_path,
3918 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3919 if (ret < 0) {
3920 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3921 PERROR("chmod");
3922 goto error;
3923 }
3924
3925 error:
3926 return ret;
3927 }
3928
3929 /*
3930 * Signal handler for the daemon
3931 *
3932 * Simply stop all worker threads, leaving main() return gracefully after
3933 * joining all threads and calling cleanup().
3934 */
3935 static void sighandler(int sig)
3936 {
3937 switch (sig) {
3938 case SIGPIPE:
3939 DBG("SIGPIPE caught");
3940 return;
3941 case SIGINT:
3942 DBG("SIGINT caught");
3943 stop_threads();
3944 break;
3945 case SIGTERM:
3946 DBG("SIGTERM caught");
3947 stop_threads();
3948 break;
3949 default:
3950 break;
3951 }
3952 }
3953
3954 /*
3955 * Setup signal handler for :
3956 * SIGINT, SIGTERM, SIGPIPE
3957 */
3958 static int set_signal_handler(void)
3959 {
3960 int ret = 0;
3961 struct sigaction sa;
3962 sigset_t sigset;
3963
3964 if ((ret = sigemptyset(&sigset)) < 0) {
3965 PERROR("sigemptyset");
3966 return ret;
3967 }
3968
3969 sa.sa_handler = sighandler;
3970 sa.sa_mask = sigset;
3971 sa.sa_flags = 0;
3972 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3973 PERROR("sigaction");
3974 return ret;
3975 }
3976
3977 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3978 PERROR("sigaction");
3979 return ret;
3980 }
3981
3982 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3983 PERROR("sigaction");
3984 return ret;
3985 }
3986
3987 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3988
3989 return ret;
3990 }
3991
3992 /*
3993 * Set open files limit to unlimited. This daemon can open a large number of
3994 * file descriptors in order to consumer multiple kernel traces.
3995 */
3996 static void set_ulimit(void)
3997 {
3998 int ret;
3999 struct rlimit lim;
4000
4001 /* The kernel does not allowed an infinite limit for open files */
4002 lim.rlim_cur = 65535;
4003 lim.rlim_max = 65535;
4004
4005 ret = setrlimit(RLIMIT_NOFILE, &lim);
4006 if (ret < 0) {
4007 PERROR("failed to set open files limit");
4008 }
4009 }
4010
4011 /*
4012 * Write pidfile using the rundir and opt_pidfile.
4013 */
4014 static void write_pidfile(void)
4015 {
4016 int ret;
4017 char pidfile_path[PATH_MAX];
4018
4019 assert(rundir);
4020
4021 if (opt_pidfile) {
4022 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4023 } else {
4024 /* Build pidfile path from rundir and opt_pidfile. */
4025 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4026 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4027 if (ret < 0) {
4028 PERROR("snprintf pidfile path");
4029 goto error;
4030 }
4031 }
4032
4033 /*
4034 * Create pid file in rundir. Return value is of no importance. The
4035 * execution will continue even though we are not able to write the file.
4036 */
4037 (void) utils_create_pid_file(getpid(), pidfile_path);
4038
4039 error:
4040 return;
4041 }
4042
4043 /*
4044 * main
4045 */
4046 int main(int argc, char **argv)
4047 {
4048 int ret = 0;
4049 void *status;
4050 const char *home_path, *env_app_timeout;
4051
4052 init_kernel_workarounds();
4053
4054 rcu_register_thread();
4055
4056 setup_consumerd_path();
4057
4058 page_size = sysconf(_SC_PAGESIZE);
4059 if (page_size < 0) {
4060 PERROR("sysconf _SC_PAGESIZE");
4061 page_size = LONG_MAX;
4062 WARN("Fallback page size to %ld", page_size);
4063 }
4064
4065 /* Parse arguments */
4066 progname = argv[0];
4067 if ((ret = parse_args(argc, argv)) < 0) {
4068 goto error;
4069 }
4070
4071 /* Daemonize */
4072 if (opt_daemon) {
4073 int i;
4074
4075 /*
4076 * fork
4077 * child: setsid, close FD 0, 1, 2, chdir /
4078 * parent: exit (if fork is successful)
4079 */
4080 ret = daemon(0, 0);
4081 if (ret < 0) {
4082 PERROR("daemon");
4083 goto error;
4084 }
4085 /*
4086 * We are in the child. Make sure all other file
4087 * descriptors are closed, in case we are called with
4088 * more opened file descriptors than the standard ones.
4089 */
4090 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4091 (void) close(i);
4092 }
4093 }
4094
4095 /* Create thread quit pipe */
4096 if ((ret = init_thread_quit_pipe()) < 0) {
4097 goto error;
4098 }
4099
4100 /* Check if daemon is UID = 0 */
4101 is_root = !getuid();
4102
4103 if (is_root) {
4104 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4105
4106 /* Create global run dir with root access */
4107 ret = create_lttng_rundir(rundir);
4108 if (ret < 0) {
4109 goto error;
4110 }
4111
4112 if (strlen(apps_unix_sock_path) == 0) {
4113 snprintf(apps_unix_sock_path, PATH_MAX,
4114 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4115 }
4116
4117 if (strlen(client_unix_sock_path) == 0) {
4118 snprintf(client_unix_sock_path, PATH_MAX,
4119 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4120 }
4121
4122 /* Set global SHM for ust */
4123 if (strlen(wait_shm_path) == 0) {
4124 snprintf(wait_shm_path, PATH_MAX,
4125 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4126 }
4127
4128 if (strlen(health_unix_sock_path) == 0) {
4129 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4130 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4131 }
4132
4133 /* Setup kernel consumerd path */
4134 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4135 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4136 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4137 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4138
4139 DBG2("Kernel consumer err path: %s",
4140 kconsumer_data.err_unix_sock_path);
4141 DBG2("Kernel consumer cmd path: %s",
4142 kconsumer_data.cmd_unix_sock_path);
4143 } else {
4144 home_path = get_home_dir();
4145 if (home_path == NULL) {
4146 /* TODO: Add --socket PATH option */
4147 ERR("Can't get HOME directory for sockets creation.");
4148 ret = -EPERM;
4149 goto error;
4150 }
4151
4152 /*
4153 * Create rundir from home path. This will create something like
4154 * $HOME/.lttng
4155 */
4156 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4157 if (ret < 0) {
4158 ret = -ENOMEM;
4159 goto error;
4160 }
4161
4162 ret = create_lttng_rundir(rundir);
4163 if (ret < 0) {
4164 goto error;
4165 }
4166
4167 if (strlen(apps_unix_sock_path) == 0) {
4168 snprintf(apps_unix_sock_path, PATH_MAX,
4169 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4170 }
4171
4172 /* Set the cli tool unix socket path */
4173 if (strlen(client_unix_sock_path) == 0) {
4174 snprintf(client_unix_sock_path, PATH_MAX,
4175 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4176 }
4177
4178 /* Set global SHM for ust */
4179 if (strlen(wait_shm_path) == 0) {
4180 snprintf(wait_shm_path, PATH_MAX,
4181 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4182 }
4183
4184 /* Set health check Unix path */
4185 if (strlen(health_unix_sock_path) == 0) {
4186 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4187 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4188 }
4189 }
4190
4191 /* Set consumer initial state */
4192 kernel_consumerd_state = CONSUMER_STOPPED;
4193 ust_consumerd_state = CONSUMER_STOPPED;
4194
4195 DBG("Client socket path %s", client_unix_sock_path);
4196 DBG("Application socket path %s", apps_unix_sock_path);
4197 DBG("Application wait path %s", wait_shm_path);
4198 DBG("LTTng run directory path: %s", rundir);
4199
4200 /* 32 bits consumerd path setup */
4201 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4202 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4203 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4204 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4205
4206 DBG2("UST consumer 32 bits err path: %s",
4207 ustconsumer32_data.err_unix_sock_path);
4208 DBG2("UST consumer 32 bits cmd path: %s",
4209 ustconsumer32_data.cmd_unix_sock_path);
4210
4211 /* 64 bits consumerd path setup */
4212 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4213 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4214 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4215 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4216
4217 DBG2("UST consumer 64 bits err path: %s",
4218 ustconsumer64_data.err_unix_sock_path);
4219 DBG2("UST consumer 64 bits cmd path: %s",
4220 ustconsumer64_data.cmd_unix_sock_path);
4221
4222 /*
4223 * See if daemon already exist.
4224 */
4225 if ((ret = check_existing_daemon()) < 0) {
4226 ERR("Already running daemon.\n");
4227 /*
4228 * We do not goto exit because we must not cleanup()
4229 * because a daemon is already running.
4230 */
4231 goto error;
4232 }
4233
4234 /*
4235 * Init UST app hash table. Alloc hash table before this point since
4236 * cleanup() can get called after that point.
4237 */
4238 ust_app_ht_alloc();
4239
4240 /* After this point, we can safely call cleanup() with "goto exit" */
4241
4242 /*
4243 * These actions must be executed as root. We do that *after* setting up
4244 * the sockets path because we MUST make the check for another daemon using
4245 * those paths *before* trying to set the kernel consumer sockets and init
4246 * kernel tracer.
4247 */
4248 if (is_root) {
4249 ret = set_consumer_sockets(&kconsumer_data, rundir);
4250 if (ret < 0) {
4251 goto exit;
4252 }
4253
4254 /* Setup kernel tracer */
4255 if (!opt_no_kernel) {
4256 init_kernel_tracer();
4257 }
4258
4259 /* Set ulimit for open files */
4260 set_ulimit();
4261 }
4262 /* init lttng_fd tracking must be done after set_ulimit. */
4263 lttng_fd_init();
4264
4265 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4266 if (ret < 0) {
4267 goto exit;
4268 }
4269
4270 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4271 if (ret < 0) {
4272 goto exit;
4273 }
4274
4275 if ((ret = set_signal_handler()) < 0) {
4276 goto exit;
4277 }
4278
4279 /* Setup the needed unix socket */
4280 if ((ret = init_daemon_socket()) < 0) {
4281 goto exit;
4282 }
4283
4284 /* Set credentials to socket */
4285 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4286 goto exit;
4287 }
4288
4289 /* Get parent pid if -S, --sig-parent is specified. */
4290 if (opt_sig_parent) {
4291 ppid = getppid();
4292 }
4293
4294 /* Setup the kernel pipe for waking up the kernel thread */
4295 if (is_root && !opt_no_kernel) {
4296 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4297 goto exit;
4298 }
4299 }
4300
4301 /* Setup the thread ht_cleanup communication pipe. */
4302 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4303 goto exit;
4304 }
4305
4306 /* Setup the thread apps communication pipe. */
4307 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4308 goto exit;
4309 }
4310
4311 /* Setup the thread apps notify communication pipe. */
4312 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4313 goto exit;
4314 }
4315
4316 /* Initialize global buffer per UID and PID registry. */
4317 buffer_reg_init_uid_registry();
4318 buffer_reg_init_pid_registry();
4319
4320 /* Init UST command queue. */
4321 cds_wfq_init(&ust_cmd_queue.queue);
4322
4323 /*
4324 * Get session list pointer. This pointer MUST NOT be free(). This list is
4325 * statically declared in session.c
4326 */
4327 session_list_ptr = session_get_list();
4328
4329 /* Set up max poll set size */
4330 lttng_poll_set_max_size();
4331
4332 cmd_init();
4333
4334 /* Check for the application socket timeout env variable. */
4335 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4336 if (env_app_timeout) {
4337 app_socket_timeout = atoi(env_app_timeout);
4338 } else {
4339 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4340 }
4341
4342 write_pidfile();
4343
4344 /* Create thread to manage the client socket */
4345 ret = pthread_create(&ht_cleanup_thread, NULL,
4346 thread_ht_cleanup, (void *) NULL);
4347 if (ret != 0) {
4348 PERROR("pthread_create ht_cleanup");
4349 goto exit_ht_cleanup;
4350 }
4351
4352 /* Create thread to manage the client socket */
4353 ret = pthread_create(&health_thread, NULL,
4354 thread_manage_health, (void *) NULL);
4355 if (ret != 0) {
4356 PERROR("pthread_create health");
4357 goto exit_health;
4358 }
4359
4360 /* Create thread to manage the client socket */
4361 ret = pthread_create(&client_thread, NULL,
4362 thread_manage_clients, (void *) NULL);
4363 if (ret != 0) {
4364 PERROR("pthread_create clients");
4365 goto exit_client;
4366 }
4367
4368 /* Create thread to dispatch registration */
4369 ret = pthread_create(&dispatch_thread, NULL,
4370 thread_dispatch_ust_registration, (void *) NULL);
4371 if (ret != 0) {
4372 PERROR("pthread_create dispatch");
4373 goto exit_dispatch;
4374 }
4375
4376 /* Create thread to manage application registration. */
4377 ret = pthread_create(&reg_apps_thread, NULL,
4378 thread_registration_apps, (void *) NULL);
4379 if (ret != 0) {
4380 PERROR("pthread_create registration");
4381 goto exit_reg_apps;
4382 }
4383
4384 /* Create thread to manage application socket */
4385 ret = pthread_create(&apps_thread, NULL,
4386 thread_manage_apps, (void *) NULL);
4387 if (ret != 0) {
4388 PERROR("pthread_create apps");
4389 goto exit_apps;
4390 }
4391
4392 /* Create thread to manage application notify socket */
4393 ret = pthread_create(&apps_notify_thread, NULL,
4394 ust_thread_manage_notify, (void *) NULL);
4395 if (ret != 0) {
4396 PERROR("pthread_create apps");
4397 goto exit_apps;
4398 }
4399
4400 /* Don't start this thread if kernel tracing is not requested nor root */
4401 if (is_root && !opt_no_kernel) {
4402 /* Create kernel thread to manage kernel event */
4403 ret = pthread_create(&kernel_thread, NULL,
4404 thread_manage_kernel, (void *) NULL);
4405 if (ret != 0) {
4406 PERROR("pthread_create kernel");
4407 goto exit_kernel;
4408 }
4409
4410 ret = pthread_join(kernel_thread, &status);
4411 if (ret != 0) {
4412 PERROR("pthread_join");
4413 goto error; /* join error, exit without cleanup */
4414 }
4415 }
4416
4417 exit_kernel:
4418 ret = pthread_join(apps_thread, &status);
4419 if (ret != 0) {
4420 PERROR("pthread_join");
4421 goto error; /* join error, exit without cleanup */
4422 }
4423
4424 exit_apps:
4425 ret = pthread_join(reg_apps_thread, &status);
4426 if (ret != 0) {
4427 PERROR("pthread_join");
4428 goto error; /* join error, exit without cleanup */
4429 }
4430
4431 exit_reg_apps:
4432 ret = pthread_join(dispatch_thread, &status);
4433 if (ret != 0) {
4434 PERROR("pthread_join");
4435 goto error; /* join error, exit without cleanup */
4436 }
4437
4438 exit_dispatch:
4439 ret = pthread_join(client_thread, &status);
4440 if (ret != 0) {
4441 PERROR("pthread_join");
4442 goto error; /* join error, exit without cleanup */
4443 }
4444
4445 ret = join_consumer_thread(&kconsumer_data);
4446 if (ret != 0) {
4447 PERROR("join_consumer");
4448 goto error; /* join error, exit without cleanup */
4449 }
4450
4451 ret = join_consumer_thread(&ustconsumer32_data);
4452 if (ret != 0) {
4453 PERROR("join_consumer ust32");
4454 goto error; /* join error, exit without cleanup */
4455 }
4456
4457 ret = join_consumer_thread(&ustconsumer64_data);
4458 if (ret != 0) {
4459 PERROR("join_consumer ust64");
4460 goto error; /* join error, exit without cleanup */
4461 }
4462
4463 exit_client:
4464 ret = pthread_join(health_thread, &status);
4465 if (ret != 0) {
4466 PERROR("pthread_join health thread");
4467 goto error; /* join error, exit without cleanup */
4468 }
4469
4470 exit_health:
4471 ret = pthread_join(ht_cleanup_thread, &status);
4472 if (ret != 0) {
4473 PERROR("pthread_join ht cleanup thread");
4474 goto error; /* join error, exit without cleanup */
4475 }
4476 exit_ht_cleanup:
4477 exit:
4478 /*
4479 * cleanup() is called when no other thread is running.
4480 */
4481 rcu_thread_online();
4482 cleanup();
4483 rcu_thread_offline();
4484 rcu_unregister_thread();
4485 if (!ret) {
4486 exit(EXIT_SUCCESS);
4487 }
4488 error:
4489 exit(EXIT_FAILURE);
4490 }
This page took 0.207768 seconds and 4 git commands to generate.