Implement snapshot commands in lttng-sessiond
[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 * Sanitize the wait queue of the dispatch registration thread meaning removing
1339 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1340 * notify socket is never received.
1341 */
1342 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1343 {
1344 int ret, nb_fd = 0, i;
1345 unsigned int fd_added = 0;
1346 struct lttng_poll_event events;
1347 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1348
1349 assert(wait_queue);
1350
1351 lttng_poll_init(&events);
1352
1353 /* Just skip everything for an empty queue. */
1354 if (!wait_queue->count) {
1355 goto end;
1356 }
1357
1358 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1359 if (ret < 0) {
1360 goto error_create;
1361 }
1362
1363 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1364 &wait_queue->head, head) {
1365 assert(wait_node->app);
1366 ret = lttng_poll_add(&events, wait_node->app->sock,
1367 LPOLLHUP | LPOLLERR);
1368 if (ret < 0) {
1369 goto error;
1370 }
1371
1372 fd_added = 1;
1373 }
1374
1375 if (!fd_added) {
1376 goto end;
1377 }
1378
1379 /*
1380 * Poll but don't block so we can quickly identify the faulty events and
1381 * clean them afterwards from the wait queue.
1382 */
1383 ret = lttng_poll_wait(&events, 0);
1384 if (ret < 0) {
1385 goto error;
1386 }
1387 nb_fd = ret;
1388
1389 for (i = 0; i < nb_fd; i++) {
1390 /* Get faulty FD. */
1391 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1392 int pollfd = LTTNG_POLL_GETFD(&events, i);
1393
1394 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1395 &wait_queue->head, head) {
1396 if (pollfd == wait_node->app->sock &&
1397 (revents & (LPOLLHUP | LPOLLERR))) {
1398 cds_list_del(&wait_node->head);
1399 wait_queue->count--;
1400 ust_app_destroy(wait_node->app);
1401 free(wait_node);
1402 break;
1403 }
1404 }
1405 }
1406
1407 if (nb_fd > 0) {
1408 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1409 }
1410
1411 end:
1412 lttng_poll_clean(&events);
1413 return;
1414
1415 error:
1416 lttng_poll_clean(&events);
1417 error_create:
1418 ERR("Unable to sanitize wait queue");
1419 return;
1420 }
1421
1422 /*
1423 * Dispatch request from the registration threads to the application
1424 * communication thread.
1425 */
1426 static void *thread_dispatch_ust_registration(void *data)
1427 {
1428 int ret, err = -1;
1429 struct cds_wfq_node *node;
1430 struct ust_command *ust_cmd = NULL;
1431 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1432 struct ust_reg_wait_queue wait_queue = {
1433 .count = 0,
1434 };
1435
1436 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1437
1438 health_code_update();
1439
1440 CDS_INIT_LIST_HEAD(&wait_queue.head);
1441
1442 DBG("[thread] Dispatch UST command started");
1443
1444 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1445 health_code_update();
1446
1447 /* Atomically prepare the queue futex */
1448 futex_nto1_prepare(&ust_cmd_queue.futex);
1449
1450 do {
1451 struct ust_app *app = NULL;
1452 ust_cmd = NULL;
1453
1454 /*
1455 * Make sure we don't have node(s) that have hung up before receiving
1456 * the notify socket. This is to clean the list in order to avoid
1457 * memory leaks from notify socket that are never seen.
1458 */
1459 sanitize_wait_queue(&wait_queue);
1460
1461 health_code_update();
1462 /* Dequeue command for registration */
1463 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1464 if (node == NULL) {
1465 DBG("Woken up but nothing in the UST command queue");
1466 /* Continue thread execution */
1467 break;
1468 }
1469
1470 ust_cmd = caa_container_of(node, struct ust_command, node);
1471
1472 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1473 " gid:%d sock:%d name:%s (version %d.%d)",
1474 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1475 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1476 ust_cmd->sock, ust_cmd->reg_msg.name,
1477 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1478
1479 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1480 wait_node = zmalloc(sizeof(*wait_node));
1481 if (!wait_node) {
1482 PERROR("zmalloc wait_node dispatch");
1483 ret = close(ust_cmd->sock);
1484 if (ret < 0) {
1485 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1486 }
1487 lttng_fd_put(1, LTTNG_FD_APPS);
1488 free(ust_cmd);
1489 goto error;
1490 }
1491 CDS_INIT_LIST_HEAD(&wait_node->head);
1492
1493 /* Create application object if socket is CMD. */
1494 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1495 ust_cmd->sock);
1496 if (!wait_node->app) {
1497 ret = close(ust_cmd->sock);
1498 if (ret < 0) {
1499 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1500 }
1501 lttng_fd_put(1, LTTNG_FD_APPS);
1502 free(wait_node);
1503 free(ust_cmd);
1504 continue;
1505 }
1506 /*
1507 * Add application to the wait queue so we can set the notify
1508 * socket before putting this object in the global ht.
1509 */
1510 cds_list_add(&wait_node->head, &wait_queue.head);
1511 wait_queue.count++;
1512
1513 free(ust_cmd);
1514 /*
1515 * We have to continue here since we don't have the notify
1516 * socket and the application MUST be added to the hash table
1517 * only at that moment.
1518 */
1519 continue;
1520 } else {
1521 /*
1522 * Look for the application in the local wait queue and set the
1523 * notify socket if found.
1524 */
1525 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1526 &wait_queue.head, head) {
1527 health_code_update();
1528 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1529 wait_node->app->notify_sock = ust_cmd->sock;
1530 cds_list_del(&wait_node->head);
1531 wait_queue.count--;
1532 app = wait_node->app;
1533 free(wait_node);
1534 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1535 break;
1536 }
1537 }
1538
1539 /*
1540 * With no application at this stage the received socket is
1541 * basically useless so close it before we free the cmd data
1542 * structure for good.
1543 */
1544 if (!app) {
1545 ret = close(ust_cmd->sock);
1546 if (ret < 0) {
1547 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1548 }
1549 lttng_fd_put(1, LTTNG_FD_APPS);
1550 }
1551 free(ust_cmd);
1552 }
1553
1554 if (app) {
1555 /*
1556 * @session_lock_list
1557 *
1558 * Lock the global session list so from the register up to the
1559 * registration done message, no thread can see the application
1560 * and change its state.
1561 */
1562 session_lock_list();
1563 rcu_read_lock();
1564
1565 /*
1566 * Add application to the global hash table. This needs to be
1567 * done before the update to the UST registry can locate the
1568 * application.
1569 */
1570 ust_app_add(app);
1571
1572 /* Set app version. This call will print an error if needed. */
1573 (void) ust_app_version(app);
1574
1575 /* Send notify socket through the notify pipe. */
1576 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1577 app->notify_sock);
1578 if (ret < 0) {
1579 rcu_read_unlock();
1580 session_unlock_list();
1581 /* No notify thread, stop the UST tracing. */
1582 goto error;
1583 }
1584
1585 /*
1586 * Update newly registered application with the tracing
1587 * registry info already enabled information.
1588 */
1589 update_ust_app(app->sock);
1590
1591 /*
1592 * Don't care about return value. Let the manage apps threads
1593 * handle app unregistration upon socket close.
1594 */
1595 (void) ust_app_register_done(app->sock);
1596
1597 /*
1598 * Even if the application socket has been closed, send the app
1599 * to the thread and unregistration will take place at that
1600 * place.
1601 */
1602 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1603 if (ret < 0) {
1604 rcu_read_unlock();
1605 session_unlock_list();
1606 /* No apps. thread, stop the UST tracing. */
1607 goto error;
1608 }
1609
1610 rcu_read_unlock();
1611 session_unlock_list();
1612 }
1613 } while (node != NULL);
1614
1615 health_poll_entry();
1616 /* Futex wait on queue. Blocking call on futex() */
1617 futex_nto1_wait(&ust_cmd_queue.futex);
1618 health_poll_exit();
1619 }
1620 /* Normal exit, no error */
1621 err = 0;
1622
1623 error:
1624 /* Clean up wait queue. */
1625 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1626 &wait_queue.head, head) {
1627 cds_list_del(&wait_node->head);
1628 wait_queue.count--;
1629 free(wait_node);
1630 }
1631
1632 DBG("Dispatch thread dying");
1633 if (err) {
1634 health_error();
1635 ERR("Health error occurred in %s", __func__);
1636 }
1637 health_unregister();
1638 return NULL;
1639 }
1640
1641 /*
1642 * This thread manage application registration.
1643 */
1644 static void *thread_registration_apps(void *data)
1645 {
1646 int sock = -1, i, ret, pollfd, err = -1;
1647 uint32_t revents, nb_fd;
1648 struct lttng_poll_event events;
1649 /*
1650 * Get allocated in this thread, enqueued to a global queue, dequeued and
1651 * freed in the manage apps thread.
1652 */
1653 struct ust_command *ust_cmd = NULL;
1654
1655 DBG("[thread] Manage application registration started");
1656
1657 health_register(HEALTH_TYPE_APP_REG);
1658
1659 if (testpoint(thread_registration_apps)) {
1660 goto error_testpoint;
1661 }
1662
1663 ret = lttcomm_listen_unix_sock(apps_sock);
1664 if (ret < 0) {
1665 goto error_listen;
1666 }
1667
1668 /*
1669 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1670 * more will be added to this poll set.
1671 */
1672 ret = sessiond_set_thread_pollset(&events, 2);
1673 if (ret < 0) {
1674 goto error_create_poll;
1675 }
1676
1677 /* Add the application registration socket */
1678 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1679 if (ret < 0) {
1680 goto error_poll_add;
1681 }
1682
1683 /* Notify all applications to register */
1684 ret = notify_ust_apps(1);
1685 if (ret < 0) {
1686 ERR("Failed to notify applications or create the wait shared memory.\n"
1687 "Execution continues but there might be problem for already\n"
1688 "running applications that wishes to register.");
1689 }
1690
1691 while (1) {
1692 DBG("Accepting application registration");
1693
1694 /* Inifinite blocking call, waiting for transmission */
1695 restart:
1696 health_poll_entry();
1697 ret = lttng_poll_wait(&events, -1);
1698 health_poll_exit();
1699 if (ret < 0) {
1700 /*
1701 * Restart interrupted system call.
1702 */
1703 if (errno == EINTR) {
1704 goto restart;
1705 }
1706 goto error;
1707 }
1708
1709 nb_fd = ret;
1710
1711 for (i = 0; i < nb_fd; i++) {
1712 health_code_update();
1713
1714 /* Fetch once the poll data */
1715 revents = LTTNG_POLL_GETEV(&events, i);
1716 pollfd = LTTNG_POLL_GETFD(&events, i);
1717
1718 /* Thread quit pipe has been closed. Killing thread. */
1719 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1720 if (ret) {
1721 err = 0;
1722 goto exit;
1723 }
1724
1725 /* Event on the registration socket */
1726 if (pollfd == apps_sock) {
1727 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1728 ERR("Register apps socket poll error");
1729 goto error;
1730 } else if (revents & LPOLLIN) {
1731 sock = lttcomm_accept_unix_sock(apps_sock);
1732 if (sock < 0) {
1733 goto error;
1734 }
1735
1736 /*
1737 * Set the CLOEXEC flag. Return code is useless because
1738 * either way, the show must go on.
1739 */
1740 (void) utils_set_fd_cloexec(sock);
1741
1742 /* Create UST registration command for enqueuing */
1743 ust_cmd = zmalloc(sizeof(struct ust_command));
1744 if (ust_cmd == NULL) {
1745 PERROR("ust command zmalloc");
1746 goto error;
1747 }
1748
1749 /*
1750 * Using message-based transmissions to ensure we don't
1751 * have to deal with partially received messages.
1752 */
1753 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1754 if (ret < 0) {
1755 ERR("Exhausted file descriptors allowed for applications.");
1756 free(ust_cmd);
1757 ret = close(sock);
1758 if (ret) {
1759 PERROR("close");
1760 }
1761 sock = -1;
1762 continue;
1763 }
1764
1765 health_code_update();
1766 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1767 if (ret < 0) {
1768 free(ust_cmd);
1769 /* Close socket of the application. */
1770 ret = close(sock);
1771 if (ret) {
1772 PERROR("close");
1773 }
1774 lttng_fd_put(LTTNG_FD_APPS, 1);
1775 sock = -1;
1776 continue;
1777 }
1778 health_code_update();
1779
1780 ust_cmd->sock = sock;
1781 sock = -1;
1782
1783 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1784 " gid:%d sock:%d name:%s (version %d.%d)",
1785 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1786 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1787 ust_cmd->sock, ust_cmd->reg_msg.name,
1788 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1789
1790 /*
1791 * Lock free enqueue the registration request. The red pill
1792 * has been taken! This apps will be part of the *system*.
1793 */
1794 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1795
1796 /*
1797 * Wake the registration queue futex. Implicit memory
1798 * barrier with the exchange in cds_wfq_enqueue.
1799 */
1800 futex_nto1_wake(&ust_cmd_queue.futex);
1801 }
1802 }
1803 }
1804 }
1805
1806 exit:
1807 error:
1808 if (err) {
1809 health_error();
1810 ERR("Health error occurred in %s", __func__);
1811 }
1812
1813 /* Notify that the registration thread is gone */
1814 notify_ust_apps(0);
1815
1816 if (apps_sock >= 0) {
1817 ret = close(apps_sock);
1818 if (ret) {
1819 PERROR("close");
1820 }
1821 }
1822 if (sock >= 0) {
1823 ret = close(sock);
1824 if (ret) {
1825 PERROR("close");
1826 }
1827 lttng_fd_put(LTTNG_FD_APPS, 1);
1828 }
1829 unlink(apps_unix_sock_path);
1830
1831 error_poll_add:
1832 lttng_poll_clean(&events);
1833 error_listen:
1834 error_create_poll:
1835 error_testpoint:
1836 DBG("UST Registration thread cleanup complete");
1837 health_unregister();
1838
1839 return NULL;
1840 }
1841
1842 /*
1843 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1844 * exec or it will fails.
1845 */
1846 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1847 {
1848 int ret, clock_ret;
1849 struct timespec timeout;
1850
1851 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1852 consumer_data->consumer_thread_is_ready = 0;
1853
1854 /* Setup pthread condition */
1855 ret = pthread_condattr_init(&consumer_data->condattr);
1856 if (ret != 0) {
1857 errno = ret;
1858 PERROR("pthread_condattr_init consumer data");
1859 goto error;
1860 }
1861
1862 /*
1863 * Set the monotonic clock in order to make sure we DO NOT jump in time
1864 * between the clock_gettime() call and the timedwait call. See bug #324
1865 * for a more details and how we noticed it.
1866 */
1867 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1868 if (ret != 0) {
1869 errno = ret;
1870 PERROR("pthread_condattr_setclock consumer data");
1871 goto error;
1872 }
1873
1874 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1875 if (ret != 0) {
1876 errno = ret;
1877 PERROR("pthread_cond_init consumer data");
1878 goto error;
1879 }
1880
1881 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1882 consumer_data);
1883 if (ret != 0) {
1884 PERROR("pthread_create consumer");
1885 ret = -1;
1886 goto error;
1887 }
1888
1889 /* We are about to wait on a pthread condition */
1890 pthread_mutex_lock(&consumer_data->cond_mutex);
1891
1892 /* Get time for sem_timedwait absolute timeout */
1893 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1894 /*
1895 * Set the timeout for the condition timed wait even if the clock gettime
1896 * call fails since we might loop on that call and we want to avoid to
1897 * increment the timeout too many times.
1898 */
1899 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1900
1901 /*
1902 * The following loop COULD be skipped in some conditions so this is why we
1903 * set ret to 0 in order to make sure at least one round of the loop is
1904 * done.
1905 */
1906 ret = 0;
1907
1908 /*
1909 * Loop until the condition is reached or when a timeout is reached. Note
1910 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1911 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1912 * possible. This loop does not take any chances and works with both of
1913 * them.
1914 */
1915 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1916 if (clock_ret < 0) {
1917 PERROR("clock_gettime spawn consumer");
1918 /* Infinite wait for the consumerd thread to be ready */
1919 ret = pthread_cond_wait(&consumer_data->cond,
1920 &consumer_data->cond_mutex);
1921 } else {
1922 ret = pthread_cond_timedwait(&consumer_data->cond,
1923 &consumer_data->cond_mutex, &timeout);
1924 }
1925 }
1926
1927 /* Release the pthread condition */
1928 pthread_mutex_unlock(&consumer_data->cond_mutex);
1929
1930 if (ret != 0) {
1931 errno = ret;
1932 if (ret == ETIMEDOUT) {
1933 /*
1934 * Call has timed out so we kill the kconsumerd_thread and return
1935 * an error.
1936 */
1937 ERR("Condition timed out. The consumer thread was never ready."
1938 " Killing it");
1939 ret = pthread_cancel(consumer_data->thread);
1940 if (ret < 0) {
1941 PERROR("pthread_cancel consumer thread");
1942 }
1943 } else {
1944 PERROR("pthread_cond_wait failed consumer thread");
1945 }
1946 goto error;
1947 }
1948
1949 pthread_mutex_lock(&consumer_data->pid_mutex);
1950 if (consumer_data->pid == 0) {
1951 ERR("Consumerd did not start");
1952 pthread_mutex_unlock(&consumer_data->pid_mutex);
1953 goto error;
1954 }
1955 pthread_mutex_unlock(&consumer_data->pid_mutex);
1956
1957 return 0;
1958
1959 error:
1960 return ret;
1961 }
1962
1963 /*
1964 * Join consumer thread
1965 */
1966 static int join_consumer_thread(struct consumer_data *consumer_data)
1967 {
1968 void *status;
1969
1970 /* Consumer pid must be a real one. */
1971 if (consumer_data->pid > 0) {
1972 int ret;
1973 ret = kill(consumer_data->pid, SIGTERM);
1974 if (ret) {
1975 ERR("Error killing consumer daemon");
1976 return ret;
1977 }
1978 return pthread_join(consumer_data->thread, &status);
1979 } else {
1980 return 0;
1981 }
1982 }
1983
1984 /*
1985 * Fork and exec a consumer daemon (consumerd).
1986 *
1987 * Return pid if successful else -1.
1988 */
1989 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1990 {
1991 int ret;
1992 pid_t pid;
1993 const char *consumer_to_use;
1994 const char *verbosity;
1995 struct stat st;
1996
1997 DBG("Spawning consumerd");
1998
1999 pid = fork();
2000 if (pid == 0) {
2001 /*
2002 * Exec consumerd.
2003 */
2004 if (opt_verbose_consumer) {
2005 verbosity = "--verbose";
2006 } else {
2007 verbosity = "--quiet";
2008 }
2009 switch (consumer_data->type) {
2010 case LTTNG_CONSUMER_KERNEL:
2011 /*
2012 * Find out which consumerd to execute. We will first try the
2013 * 64-bit path, then the sessiond's installation directory, and
2014 * fallback on the 32-bit one,
2015 */
2016 DBG3("Looking for a kernel consumer at these locations:");
2017 DBG3(" 1) %s", consumerd64_bin);
2018 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2019 DBG3(" 3) %s", consumerd32_bin);
2020 if (stat(consumerd64_bin, &st) == 0) {
2021 DBG3("Found location #1");
2022 consumer_to_use = consumerd64_bin;
2023 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2024 DBG3("Found location #2");
2025 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2026 } else if (stat(consumerd32_bin, &st) == 0) {
2027 DBG3("Found location #3");
2028 consumer_to_use = consumerd32_bin;
2029 } else {
2030 DBG("Could not find any valid consumerd executable");
2031 break;
2032 }
2033 DBG("Using kernel consumer at: %s", consumer_to_use);
2034 execl(consumer_to_use,
2035 "lttng-consumerd", verbosity, "-k",
2036 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2037 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2038 NULL);
2039 break;
2040 case LTTNG_CONSUMER64_UST:
2041 {
2042 char *tmpnew = NULL;
2043
2044 if (consumerd64_libdir[0] != '\0') {
2045 char *tmp;
2046 size_t tmplen;
2047
2048 tmp = getenv("LD_LIBRARY_PATH");
2049 if (!tmp) {
2050 tmp = "";
2051 }
2052 tmplen = strlen("LD_LIBRARY_PATH=")
2053 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2054 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2055 if (!tmpnew) {
2056 ret = -ENOMEM;
2057 goto error;
2058 }
2059 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2060 strcat(tmpnew, consumerd64_libdir);
2061 if (tmp[0] != '\0') {
2062 strcat(tmpnew, ":");
2063 strcat(tmpnew, tmp);
2064 }
2065 ret = putenv(tmpnew);
2066 if (ret) {
2067 ret = -errno;
2068 free(tmpnew);
2069 goto error;
2070 }
2071 }
2072 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2073 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2074 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2075 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2076 NULL);
2077 if (consumerd64_libdir[0] != '\0') {
2078 free(tmpnew);
2079 }
2080 if (ret) {
2081 goto error;
2082 }
2083 break;
2084 }
2085 case LTTNG_CONSUMER32_UST:
2086 {
2087 char *tmpnew = NULL;
2088
2089 if (consumerd32_libdir[0] != '\0') {
2090 char *tmp;
2091 size_t tmplen;
2092
2093 tmp = getenv("LD_LIBRARY_PATH");
2094 if (!tmp) {
2095 tmp = "";
2096 }
2097 tmplen = strlen("LD_LIBRARY_PATH=")
2098 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2099 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2100 if (!tmpnew) {
2101 ret = -ENOMEM;
2102 goto error;
2103 }
2104 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2105 strcat(tmpnew, consumerd32_libdir);
2106 if (tmp[0] != '\0') {
2107 strcat(tmpnew, ":");
2108 strcat(tmpnew, tmp);
2109 }
2110 ret = putenv(tmpnew);
2111 if (ret) {
2112 ret = -errno;
2113 free(tmpnew);
2114 goto error;
2115 }
2116 }
2117 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2118 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2119 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2120 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2121 NULL);
2122 if (consumerd32_libdir[0] != '\0') {
2123 free(tmpnew);
2124 }
2125 if (ret) {
2126 goto error;
2127 }
2128 break;
2129 }
2130 default:
2131 PERROR("unknown consumer type");
2132 exit(EXIT_FAILURE);
2133 }
2134 if (errno != 0) {
2135 PERROR("kernel start consumer exec");
2136 }
2137 exit(EXIT_FAILURE);
2138 } else if (pid > 0) {
2139 ret = pid;
2140 } else {
2141 PERROR("start consumer fork");
2142 ret = -errno;
2143 }
2144 error:
2145 return ret;
2146 }
2147
2148 /*
2149 * Spawn the consumerd daemon and session daemon thread.
2150 */
2151 static int start_consumerd(struct consumer_data *consumer_data)
2152 {
2153 int ret;
2154
2155 /*
2156 * Set the listen() state on the socket since there is a possible race
2157 * between the exec() of the consumer daemon and this call if place in the
2158 * consumer thread. See bug #366 for more details.
2159 */
2160 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2161 if (ret < 0) {
2162 goto error;
2163 }
2164
2165 pthread_mutex_lock(&consumer_data->pid_mutex);
2166 if (consumer_data->pid != 0) {
2167 pthread_mutex_unlock(&consumer_data->pid_mutex);
2168 goto end;
2169 }
2170
2171 ret = spawn_consumerd(consumer_data);
2172 if (ret < 0) {
2173 ERR("Spawning consumerd failed");
2174 pthread_mutex_unlock(&consumer_data->pid_mutex);
2175 goto error;
2176 }
2177
2178 /* Setting up the consumer_data pid */
2179 consumer_data->pid = ret;
2180 DBG2("Consumer pid %d", consumer_data->pid);
2181 pthread_mutex_unlock(&consumer_data->pid_mutex);
2182
2183 DBG2("Spawning consumer control thread");
2184 ret = spawn_consumer_thread(consumer_data);
2185 if (ret < 0) {
2186 ERR("Fatal error spawning consumer control thread");
2187 goto error;
2188 }
2189
2190 end:
2191 return 0;
2192
2193 error:
2194 /* Cleanup already created sockets on error. */
2195 if (consumer_data->err_sock >= 0) {
2196 int err;
2197
2198 err = close(consumer_data->err_sock);
2199 if (err < 0) {
2200 PERROR("close consumer data error socket");
2201 }
2202 }
2203 return ret;
2204 }
2205
2206 /*
2207 * Compute health status of each consumer. If one of them is zero (bad
2208 * state), we return 0.
2209 */
2210 static int check_consumer_health(void)
2211 {
2212 int ret;
2213
2214 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2215
2216 DBG3("Health consumer check %d", ret);
2217
2218 return ret;
2219 }
2220
2221 /*
2222 * Setup necessary data for kernel tracer action.
2223 */
2224 static int init_kernel_tracer(void)
2225 {
2226 int ret;
2227
2228 /* Modprobe lttng kernel modules */
2229 ret = modprobe_lttng_control();
2230 if (ret < 0) {
2231 goto error;
2232 }
2233
2234 /* Open debugfs lttng */
2235 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2236 if (kernel_tracer_fd < 0) {
2237 DBG("Failed to open %s", module_proc_lttng);
2238 ret = -1;
2239 goto error_open;
2240 }
2241
2242 /* Validate kernel version */
2243 ret = kernel_validate_version(kernel_tracer_fd);
2244 if (ret < 0) {
2245 goto error_version;
2246 }
2247
2248 ret = modprobe_lttng_data();
2249 if (ret < 0) {
2250 goto error_modules;
2251 }
2252
2253 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2254 return 0;
2255
2256 error_version:
2257 modprobe_remove_lttng_control();
2258 ret = close(kernel_tracer_fd);
2259 if (ret) {
2260 PERROR("close");
2261 }
2262 kernel_tracer_fd = -1;
2263 return LTTNG_ERR_KERN_VERSION;
2264
2265 error_modules:
2266 ret = close(kernel_tracer_fd);
2267 if (ret) {
2268 PERROR("close");
2269 }
2270
2271 error_open:
2272 modprobe_remove_lttng_control();
2273
2274 error:
2275 WARN("No kernel tracer available");
2276 kernel_tracer_fd = -1;
2277 if (!is_root) {
2278 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2279 } else {
2280 return LTTNG_ERR_KERN_NA;
2281 }
2282 }
2283
2284
2285 /*
2286 * Copy consumer output from the tracing session to the domain session. The
2287 * function also applies the right modification on a per domain basis for the
2288 * trace files destination directory.
2289 *
2290 * Should *NOT* be called with RCU read-side lock held.
2291 */
2292 static int copy_session_consumer(int domain, struct ltt_session *session)
2293 {
2294 int ret;
2295 const char *dir_name;
2296 struct consumer_output *consumer;
2297
2298 assert(session);
2299 assert(session->consumer);
2300
2301 switch (domain) {
2302 case LTTNG_DOMAIN_KERNEL:
2303 DBG3("Copying tracing session consumer output in kernel session");
2304 /*
2305 * XXX: We should audit the session creation and what this function
2306 * does "extra" in order to avoid a destroy since this function is used
2307 * in the domain session creation (kernel and ust) only. Same for UST
2308 * domain.
2309 */
2310 if (session->kernel_session->consumer) {
2311 consumer_destroy_output(session->kernel_session->consumer);
2312 }
2313 session->kernel_session->consumer =
2314 consumer_copy_output(session->consumer);
2315 /* Ease our life a bit for the next part */
2316 consumer = session->kernel_session->consumer;
2317 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2318 break;
2319 case LTTNG_DOMAIN_UST:
2320 DBG3("Copying tracing session consumer output in UST session");
2321 if (session->ust_session->consumer) {
2322 consumer_destroy_output(session->ust_session->consumer);
2323 }
2324 session->ust_session->consumer =
2325 consumer_copy_output(session->consumer);
2326 /* Ease our life a bit for the next part */
2327 consumer = session->ust_session->consumer;
2328 dir_name = DEFAULT_UST_TRACE_DIR;
2329 break;
2330 default:
2331 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2332 goto error;
2333 }
2334
2335 /* Append correct directory to subdir */
2336 strncat(consumer->subdir, dir_name,
2337 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2338 DBG3("Copy session consumer subdir %s", consumer->subdir);
2339
2340 ret = LTTNG_OK;
2341
2342 error:
2343 return ret;
2344 }
2345
2346 /*
2347 * Create an UST session and add it to the session ust list.
2348 *
2349 * Should *NOT* be called with RCU read-side lock held.
2350 */
2351 static int create_ust_session(struct ltt_session *session,
2352 struct lttng_domain *domain)
2353 {
2354 int ret;
2355 struct ltt_ust_session *lus = NULL;
2356
2357 assert(session);
2358 assert(domain);
2359 assert(session->consumer);
2360
2361 switch (domain->type) {
2362 case LTTNG_DOMAIN_UST:
2363 break;
2364 default:
2365 ERR("Unknown UST domain on create session %d", domain->type);
2366 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2367 goto error;
2368 }
2369
2370 DBG("Creating UST session");
2371
2372 lus = trace_ust_create_session(session->id);
2373 if (lus == NULL) {
2374 ret = LTTNG_ERR_UST_SESS_FAIL;
2375 goto error;
2376 }
2377
2378 lus->uid = session->uid;
2379 lus->gid = session->gid;
2380
2381 session->ust_session = lus;
2382
2383 /* Copy session output to the newly created UST session */
2384 ret = copy_session_consumer(domain->type, session);
2385 if (ret != LTTNG_OK) {
2386 goto error;
2387 }
2388
2389 return LTTNG_OK;
2390
2391 error:
2392 free(lus);
2393 session->ust_session = NULL;
2394 return ret;
2395 }
2396
2397 /*
2398 * Create a kernel tracer session then create the default channel.
2399 */
2400 static int create_kernel_session(struct ltt_session *session)
2401 {
2402 int ret;
2403
2404 DBG("Creating kernel session");
2405
2406 ret = kernel_create_session(session, kernel_tracer_fd);
2407 if (ret < 0) {
2408 ret = LTTNG_ERR_KERN_SESS_FAIL;
2409 goto error;
2410 }
2411
2412 /* Code flow safety */
2413 assert(session->kernel_session);
2414
2415 /* Copy session output to the newly created Kernel session */
2416 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2417 if (ret != LTTNG_OK) {
2418 goto error;
2419 }
2420
2421 /* Create directory(ies) on local filesystem. */
2422 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2423 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2424 ret = run_as_mkdir_recursive(
2425 session->kernel_session->consumer->dst.trace_path,
2426 S_IRWXU | S_IRWXG, session->uid, session->gid);
2427 if (ret < 0) {
2428 if (ret != -EEXIST) {
2429 ERR("Trace directory creation error");
2430 goto error;
2431 }
2432 }
2433 }
2434
2435 session->kernel_session->uid = session->uid;
2436 session->kernel_session->gid = session->gid;
2437
2438 return LTTNG_OK;
2439
2440 error:
2441 trace_kernel_destroy_session(session->kernel_session);
2442 session->kernel_session = NULL;
2443 return ret;
2444 }
2445
2446 /*
2447 * Count number of session permitted by uid/gid.
2448 */
2449 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2450 {
2451 unsigned int i = 0;
2452 struct ltt_session *session;
2453
2454 DBG("Counting number of available session for UID %d GID %d",
2455 uid, gid);
2456 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2457 /*
2458 * Only list the sessions the user can control.
2459 */
2460 if (!session_access_ok(session, uid, gid)) {
2461 continue;
2462 }
2463 i++;
2464 }
2465 return i;
2466 }
2467
2468 /*
2469 * Process the command requested by the lttng client within the command
2470 * context structure. This function make sure that the return structure (llm)
2471 * is set and ready for transmission before returning.
2472 *
2473 * Return any error encountered or 0 for success.
2474 *
2475 * "sock" is only used for special-case var. len data.
2476 *
2477 * Should *NOT* be called with RCU read-side lock held.
2478 */
2479 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2480 int *sock_error)
2481 {
2482 int ret = LTTNG_OK;
2483 int need_tracing_session = 1;
2484 int need_domain;
2485
2486 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2487
2488 *sock_error = 0;
2489
2490 switch (cmd_ctx->lsm->cmd_type) {
2491 case LTTNG_CREATE_SESSION:
2492 case LTTNG_DESTROY_SESSION:
2493 case LTTNG_LIST_SESSIONS:
2494 case LTTNG_LIST_DOMAINS:
2495 case LTTNG_START_TRACE:
2496 case LTTNG_STOP_TRACE:
2497 case LTTNG_DATA_PENDING:
2498 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2499 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2500 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2501 case LTTNG_SNAPSHOT_RECORD:
2502 need_domain = 0;
2503 break;
2504 default:
2505 need_domain = 1;
2506 }
2507
2508 if (opt_no_kernel && need_domain
2509 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2510 if (!is_root) {
2511 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2512 } else {
2513 ret = LTTNG_ERR_KERN_NA;
2514 }
2515 goto error;
2516 }
2517
2518 /* Deny register consumer if we already have a spawned consumer. */
2519 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2520 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2521 if (kconsumer_data.pid > 0) {
2522 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2523 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2524 goto error;
2525 }
2526 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2527 }
2528
2529 /*
2530 * Check for command that don't needs to allocate a returned payload. We do
2531 * this here so we don't have to make the call for no payload at each
2532 * command.
2533 */
2534 switch(cmd_ctx->lsm->cmd_type) {
2535 case LTTNG_LIST_SESSIONS:
2536 case LTTNG_LIST_TRACEPOINTS:
2537 case LTTNG_LIST_TRACEPOINT_FIELDS:
2538 case LTTNG_LIST_DOMAINS:
2539 case LTTNG_LIST_CHANNELS:
2540 case LTTNG_LIST_EVENTS:
2541 break;
2542 default:
2543 /* Setup lttng message with no payload */
2544 ret = setup_lttng_msg(cmd_ctx, 0);
2545 if (ret < 0) {
2546 /* This label does not try to unlock the session */
2547 goto init_setup_error;
2548 }
2549 }
2550
2551 /* Commands that DO NOT need a session. */
2552 switch (cmd_ctx->lsm->cmd_type) {
2553 case LTTNG_CREATE_SESSION:
2554 case LTTNG_CALIBRATE:
2555 case LTTNG_LIST_SESSIONS:
2556 case LTTNG_LIST_TRACEPOINTS:
2557 case LTTNG_LIST_TRACEPOINT_FIELDS:
2558 need_tracing_session = 0;
2559 break;
2560 default:
2561 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2562 /*
2563 * We keep the session list lock across _all_ commands
2564 * for now, because the per-session lock does not
2565 * handle teardown properly.
2566 */
2567 session_lock_list();
2568 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2569 if (cmd_ctx->session == NULL) {
2570 ret = LTTNG_ERR_SESS_NOT_FOUND;
2571 goto error;
2572 } else {
2573 /* Acquire lock for the session */
2574 session_lock(cmd_ctx->session);
2575 }
2576 break;
2577 }
2578
2579 if (!need_domain) {
2580 goto skip_domain;
2581 }
2582
2583 /*
2584 * Check domain type for specific "pre-action".
2585 */
2586 switch (cmd_ctx->lsm->domain.type) {
2587 case LTTNG_DOMAIN_KERNEL:
2588 if (!is_root) {
2589 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2590 goto error;
2591 }
2592
2593 /* Kernel tracer check */
2594 if (kernel_tracer_fd == -1) {
2595 /* Basically, load kernel tracer modules */
2596 ret = init_kernel_tracer();
2597 if (ret != 0) {
2598 goto error;
2599 }
2600 }
2601
2602 /* Consumer is in an ERROR state. Report back to client */
2603 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2604 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2605 goto error;
2606 }
2607
2608 /* Need a session for kernel command */
2609 if (need_tracing_session) {
2610 if (cmd_ctx->session->kernel_session == NULL) {
2611 ret = create_kernel_session(cmd_ctx->session);
2612 if (ret < 0) {
2613 ret = LTTNG_ERR_KERN_SESS_FAIL;
2614 goto error;
2615 }
2616 }
2617
2618 /* Start the kernel consumer daemon */
2619 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2620 if (kconsumer_data.pid == 0 &&
2621 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2622 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2623 ret = start_consumerd(&kconsumer_data);
2624 if (ret < 0) {
2625 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2626 goto error;
2627 }
2628 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2629 } else {
2630 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2631 }
2632
2633 /*
2634 * The consumer was just spawned so we need to add the socket to
2635 * the consumer output of the session if exist.
2636 */
2637 ret = consumer_create_socket(&kconsumer_data,
2638 cmd_ctx->session->kernel_session->consumer);
2639 if (ret < 0) {
2640 goto error;
2641 }
2642 }
2643
2644 break;
2645 case LTTNG_DOMAIN_UST:
2646 {
2647 /* Consumer is in an ERROR state. Report back to client */
2648 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2649 ret = LTTNG_ERR_NO_USTCONSUMERD;
2650 goto error;
2651 }
2652
2653 if (need_tracing_session) {
2654 /* Create UST session if none exist. */
2655 if (cmd_ctx->session->ust_session == NULL) {
2656 ret = create_ust_session(cmd_ctx->session,
2657 &cmd_ctx->lsm->domain);
2658 if (ret != LTTNG_OK) {
2659 goto error;
2660 }
2661 }
2662
2663 /* Start the UST consumer daemons */
2664 /* 64-bit */
2665 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2666 if (consumerd64_bin[0] != '\0' &&
2667 ustconsumer64_data.pid == 0 &&
2668 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2669 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2670 ret = start_consumerd(&ustconsumer64_data);
2671 if (ret < 0) {
2672 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2673 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2674 goto error;
2675 }
2676
2677 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2678 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2679 } else {
2680 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2681 }
2682
2683 /*
2684 * Setup socket for consumer 64 bit. No need for atomic access
2685 * since it was set above and can ONLY be set in this thread.
2686 */
2687 ret = consumer_create_socket(&ustconsumer64_data,
2688 cmd_ctx->session->ust_session->consumer);
2689 if (ret < 0) {
2690 goto error;
2691 }
2692
2693 /* 32-bit */
2694 if (consumerd32_bin[0] != '\0' &&
2695 ustconsumer32_data.pid == 0 &&
2696 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2697 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2698 ret = start_consumerd(&ustconsumer32_data);
2699 if (ret < 0) {
2700 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2701 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2702 goto error;
2703 }
2704
2705 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2706 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2707 } else {
2708 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2709 }
2710
2711 /*
2712 * Setup socket for consumer 64 bit. No need for atomic access
2713 * since it was set above and can ONLY be set in this thread.
2714 */
2715 ret = consumer_create_socket(&ustconsumer32_data,
2716 cmd_ctx->session->ust_session->consumer);
2717 if (ret < 0) {
2718 goto error;
2719 }
2720 }
2721 break;
2722 }
2723 default:
2724 break;
2725 }
2726 skip_domain:
2727
2728 /* Validate consumer daemon state when start/stop trace command */
2729 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2730 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2731 switch (cmd_ctx->lsm->domain.type) {
2732 case LTTNG_DOMAIN_UST:
2733 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2734 ret = LTTNG_ERR_NO_USTCONSUMERD;
2735 goto error;
2736 }
2737 break;
2738 case LTTNG_DOMAIN_KERNEL:
2739 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2740 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2741 goto error;
2742 }
2743 break;
2744 }
2745 }
2746
2747 /*
2748 * Check that the UID or GID match that of the tracing session.
2749 * The root user can interact with all sessions.
2750 */
2751 if (need_tracing_session) {
2752 if (!session_access_ok(cmd_ctx->session,
2753 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2754 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2755 ret = LTTNG_ERR_EPERM;
2756 goto error;
2757 }
2758 }
2759
2760 /*
2761 * Send relayd information to consumer as soon as we have a domain and a
2762 * session defined.
2763 */
2764 if (cmd_ctx->session && need_domain) {
2765 /*
2766 * Setup relayd if not done yet. If the relayd information was already
2767 * sent to the consumer, this call will gracefully return.
2768 */
2769 ret = cmd_setup_relayd(cmd_ctx->session);
2770 if (ret != LTTNG_OK) {
2771 goto error;
2772 }
2773 }
2774
2775 /* Process by command type */
2776 switch (cmd_ctx->lsm->cmd_type) {
2777 case LTTNG_ADD_CONTEXT:
2778 {
2779 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2780 cmd_ctx->lsm->u.context.channel_name,
2781 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2782 break;
2783 }
2784 case LTTNG_DISABLE_CHANNEL:
2785 {
2786 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2787 cmd_ctx->lsm->u.disable.channel_name);
2788 break;
2789 }
2790 case LTTNG_DISABLE_EVENT:
2791 {
2792 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2793 cmd_ctx->lsm->u.disable.channel_name,
2794 cmd_ctx->lsm->u.disable.name);
2795 break;
2796 }
2797 case LTTNG_DISABLE_ALL_EVENT:
2798 {
2799 DBG("Disabling all events");
2800
2801 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2802 cmd_ctx->lsm->u.disable.channel_name);
2803 break;
2804 }
2805 case LTTNG_ENABLE_CHANNEL:
2806 {
2807 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2808 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2809 break;
2810 }
2811 case LTTNG_ENABLE_EVENT:
2812 {
2813 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2814 cmd_ctx->lsm->u.enable.channel_name,
2815 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2816 break;
2817 }
2818 case LTTNG_ENABLE_ALL_EVENT:
2819 {
2820 DBG("Enabling all events");
2821
2822 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2823 cmd_ctx->lsm->u.enable.channel_name,
2824 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2825 break;
2826 }
2827 case LTTNG_LIST_TRACEPOINTS:
2828 {
2829 struct lttng_event *events;
2830 ssize_t nb_events;
2831
2832 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2833 if (nb_events < 0) {
2834 /* Return value is a negative lttng_error_code. */
2835 ret = -nb_events;
2836 goto error;
2837 }
2838
2839 /*
2840 * Setup lttng message with payload size set to the event list size in
2841 * bytes and then copy list into the llm payload.
2842 */
2843 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2844 if (ret < 0) {
2845 free(events);
2846 goto setup_error;
2847 }
2848
2849 /* Copy event list into message payload */
2850 memcpy(cmd_ctx->llm->payload, events,
2851 sizeof(struct lttng_event) * nb_events);
2852
2853 free(events);
2854
2855 ret = LTTNG_OK;
2856 break;
2857 }
2858 case LTTNG_LIST_TRACEPOINT_FIELDS:
2859 {
2860 struct lttng_event_field *fields;
2861 ssize_t nb_fields;
2862
2863 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2864 &fields);
2865 if (nb_fields < 0) {
2866 /* Return value is a negative lttng_error_code. */
2867 ret = -nb_fields;
2868 goto error;
2869 }
2870
2871 /*
2872 * Setup lttng message with payload size set to the event list size in
2873 * bytes and then copy list into the llm payload.
2874 */
2875 ret = setup_lttng_msg(cmd_ctx,
2876 sizeof(struct lttng_event_field) * nb_fields);
2877 if (ret < 0) {
2878 free(fields);
2879 goto setup_error;
2880 }
2881
2882 /* Copy event list into message payload */
2883 memcpy(cmd_ctx->llm->payload, fields,
2884 sizeof(struct lttng_event_field) * nb_fields);
2885
2886 free(fields);
2887
2888 ret = LTTNG_OK;
2889 break;
2890 }
2891 case LTTNG_SET_CONSUMER_URI:
2892 {
2893 size_t nb_uri, len;
2894 struct lttng_uri *uris;
2895
2896 nb_uri = cmd_ctx->lsm->u.uri.size;
2897 len = nb_uri * sizeof(struct lttng_uri);
2898
2899 if (nb_uri == 0) {
2900 ret = LTTNG_ERR_INVALID;
2901 goto error;
2902 }
2903
2904 uris = zmalloc(len);
2905 if (uris == NULL) {
2906 ret = LTTNG_ERR_FATAL;
2907 goto error;
2908 }
2909
2910 /* Receive variable len data */
2911 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2912 ret = lttcomm_recv_unix_sock(sock, uris, len);
2913 if (ret <= 0) {
2914 DBG("No URIs received from client... continuing");
2915 *sock_error = 1;
2916 ret = LTTNG_ERR_SESSION_FAIL;
2917 free(uris);
2918 goto error;
2919 }
2920
2921 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2922 nb_uri, uris);
2923 if (ret != LTTNG_OK) {
2924 free(uris);
2925 goto error;
2926 }
2927
2928 /*
2929 * XXX: 0 means that this URI should be applied on the session. Should
2930 * be a DOMAIN enuam.
2931 */
2932 if (cmd_ctx->lsm->domain.type == 0) {
2933 /* Add the URI for the UST session if a consumer is present. */
2934 if (cmd_ctx->session->ust_session &&
2935 cmd_ctx->session->ust_session->consumer) {
2936 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2937 nb_uri, uris);
2938 } else if (cmd_ctx->session->kernel_session &&
2939 cmd_ctx->session->kernel_session->consumer) {
2940 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2941 cmd_ctx->session, nb_uri, uris);
2942 }
2943 }
2944
2945 free(uris);
2946
2947 break;
2948 }
2949 case LTTNG_START_TRACE:
2950 {
2951 ret = cmd_start_trace(cmd_ctx->session);
2952 break;
2953 }
2954 case LTTNG_STOP_TRACE:
2955 {
2956 ret = cmd_stop_trace(cmd_ctx->session);
2957 break;
2958 }
2959 case LTTNG_CREATE_SESSION:
2960 {
2961 size_t nb_uri, len;
2962 struct lttng_uri *uris = NULL;
2963
2964 nb_uri = cmd_ctx->lsm->u.uri.size;
2965 len = nb_uri * sizeof(struct lttng_uri);
2966
2967 if (nb_uri > 0) {
2968 uris = zmalloc(len);
2969 if (uris == NULL) {
2970 ret = LTTNG_ERR_FATAL;
2971 goto error;
2972 }
2973
2974 /* Receive variable len data */
2975 DBG("Waiting for %zu URIs from client ...", nb_uri);
2976 ret = lttcomm_recv_unix_sock(sock, uris, len);
2977 if (ret <= 0) {
2978 DBG("No URIs received from client... continuing");
2979 *sock_error = 1;
2980 ret = LTTNG_ERR_SESSION_FAIL;
2981 free(uris);
2982 goto error;
2983 }
2984
2985 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2986 DBG("Creating session with ONE network URI is a bad call");
2987 ret = LTTNG_ERR_SESSION_FAIL;
2988 free(uris);
2989 goto error;
2990 }
2991 }
2992
2993 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2994 &cmd_ctx->creds);
2995
2996 free(uris);
2997
2998 break;
2999 }
3000 case LTTNG_DESTROY_SESSION:
3001 {
3002 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3003
3004 /* Set session to NULL so we do not unlock it after free. */
3005 cmd_ctx->session = NULL;
3006 break;
3007 }
3008 case LTTNG_LIST_DOMAINS:
3009 {
3010 ssize_t nb_dom;
3011 struct lttng_domain *domains;
3012
3013 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3014 if (nb_dom < 0) {
3015 /* Return value is a negative lttng_error_code. */
3016 ret = -nb_dom;
3017 goto error;
3018 }
3019
3020 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3021 if (ret < 0) {
3022 free(domains);
3023 goto setup_error;
3024 }
3025
3026 /* Copy event list into message payload */
3027 memcpy(cmd_ctx->llm->payload, domains,
3028 nb_dom * sizeof(struct lttng_domain));
3029
3030 free(domains);
3031
3032 ret = LTTNG_OK;
3033 break;
3034 }
3035 case LTTNG_LIST_CHANNELS:
3036 {
3037 int nb_chan;
3038 struct lttng_channel *channels;
3039
3040 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3041 cmd_ctx->session, &channels);
3042 if (nb_chan < 0) {
3043 /* Return value is a negative lttng_error_code. */
3044 ret = -nb_chan;
3045 goto error;
3046 }
3047
3048 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3049 if (ret < 0) {
3050 free(channels);
3051 goto setup_error;
3052 }
3053
3054 /* Copy event list into message payload */
3055 memcpy(cmd_ctx->llm->payload, channels,
3056 nb_chan * sizeof(struct lttng_channel));
3057
3058 free(channels);
3059
3060 ret = LTTNG_OK;
3061 break;
3062 }
3063 case LTTNG_LIST_EVENTS:
3064 {
3065 ssize_t nb_event;
3066 struct lttng_event *events = NULL;
3067
3068 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3069 cmd_ctx->lsm->u.list.channel_name, &events);
3070 if (nb_event < 0) {
3071 /* Return value is a negative lttng_error_code. */
3072 ret = -nb_event;
3073 goto error;
3074 }
3075
3076 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3077 if (ret < 0) {
3078 free(events);
3079 goto setup_error;
3080 }
3081
3082 /* Copy event list into message payload */
3083 memcpy(cmd_ctx->llm->payload, events,
3084 nb_event * sizeof(struct lttng_event));
3085
3086 free(events);
3087
3088 ret = LTTNG_OK;
3089 break;
3090 }
3091 case LTTNG_LIST_SESSIONS:
3092 {
3093 unsigned int nr_sessions;
3094
3095 session_lock_list();
3096 nr_sessions = lttng_sessions_count(
3097 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3098 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3099
3100 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3101 if (ret < 0) {
3102 session_unlock_list();
3103 goto setup_error;
3104 }
3105
3106 /* Filled the session array */
3107 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3108 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3109 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3110
3111 session_unlock_list();
3112
3113 ret = LTTNG_OK;
3114 break;
3115 }
3116 case LTTNG_CALIBRATE:
3117 {
3118 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3119 &cmd_ctx->lsm->u.calibrate);
3120 break;
3121 }
3122 case LTTNG_REGISTER_CONSUMER:
3123 {
3124 struct consumer_data *cdata;
3125
3126 switch (cmd_ctx->lsm->domain.type) {
3127 case LTTNG_DOMAIN_KERNEL:
3128 cdata = &kconsumer_data;
3129 break;
3130 default:
3131 ret = LTTNG_ERR_UND;
3132 goto error;
3133 }
3134
3135 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3136 cmd_ctx->lsm->u.reg.path, cdata);
3137 break;
3138 }
3139 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3140 {
3141 struct lttng_filter_bytecode *bytecode;
3142
3143 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3144 ret = LTTNG_ERR_FILTER_INVAL;
3145 goto error;
3146 }
3147 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3148 ret = LTTNG_ERR_FILTER_INVAL;
3149 goto error;
3150 }
3151 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3152 if (!bytecode) {
3153 ret = LTTNG_ERR_FILTER_NOMEM;
3154 goto error;
3155 }
3156 /* Receive var. len. data */
3157 DBG("Receiving var len data from client ...");
3158 ret = lttcomm_recv_unix_sock(sock, bytecode,
3159 cmd_ctx->lsm->u.enable.bytecode_len);
3160 if (ret <= 0) {
3161 DBG("Nothing recv() from client var len data... continuing");
3162 *sock_error = 1;
3163 ret = LTTNG_ERR_FILTER_INVAL;
3164 goto error;
3165 }
3166
3167 if (bytecode->len + sizeof(*bytecode)
3168 != cmd_ctx->lsm->u.enable.bytecode_len) {
3169 free(bytecode);
3170 ret = LTTNG_ERR_FILTER_INVAL;
3171 goto error;
3172 }
3173
3174 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3175 cmd_ctx->lsm->u.enable.channel_name,
3176 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3177 break;
3178 }
3179 case LTTNG_DATA_PENDING:
3180 {
3181 ret = cmd_data_pending(cmd_ctx->session);
3182 break;
3183 }
3184 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3185 {
3186 struct lttcomm_lttng_output_id reply;
3187
3188 ret = cmd_snapshot_add_output(cmd_ctx->session,
3189 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3190 if (ret != LTTNG_OK) {
3191 goto error;
3192 }
3193
3194 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3195 if (ret < 0) {
3196 goto setup_error;
3197 }
3198
3199 /* Copy output list into message payload */
3200 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3201 ret = LTTNG_OK;
3202 break;
3203 }
3204 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3205 {
3206 ret = cmd_snapshot_del_output(cmd_ctx->session,
3207 &cmd_ctx->lsm->u.snapshot_output.output);
3208 break;
3209 }
3210 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3211 {
3212 ssize_t nb_output;
3213 struct lttng_snapshot_output *outputs = NULL;
3214
3215 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3216 if (nb_output < 0) {
3217 ret = -nb_output;
3218 goto error;
3219 }
3220
3221 ret = setup_lttng_msg(cmd_ctx,
3222 nb_output * sizeof(struct lttng_snapshot_output));
3223 if (ret < 0) {
3224 free(outputs);
3225 goto setup_error;
3226 }
3227
3228 if (outputs) {
3229 /* Copy output list into message payload */
3230 memcpy(cmd_ctx->llm->payload, outputs,
3231 nb_output * sizeof(struct lttng_snapshot_output));
3232 free(outputs);
3233 }
3234
3235 ret = LTTNG_OK;
3236 break;
3237 }
3238 case LTTNG_SNAPSHOT_RECORD:
3239 {
3240 ret = cmd_snapshot_record(cmd_ctx->session,
3241 &cmd_ctx->lsm->u.snapshot_record.output,
3242 cmd_ctx->lsm->u.snapshot_record.wait);
3243 break;
3244 }
3245 default:
3246 ret = LTTNG_ERR_UND;
3247 break;
3248 }
3249
3250 error:
3251 if (cmd_ctx->llm == NULL) {
3252 DBG("Missing llm structure. Allocating one.");
3253 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3254 goto setup_error;
3255 }
3256 }
3257 /* Set return code */
3258 cmd_ctx->llm->ret_code = ret;
3259 setup_error:
3260 if (cmd_ctx->session) {
3261 session_unlock(cmd_ctx->session);
3262 }
3263 if (need_tracing_session) {
3264 session_unlock_list();
3265 }
3266 init_setup_error:
3267 return ret;
3268 }
3269
3270 /*
3271 * Thread managing health check socket.
3272 */
3273 static void *thread_manage_health(void *data)
3274 {
3275 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3276 uint32_t revents, nb_fd;
3277 struct lttng_poll_event events;
3278 struct lttcomm_health_msg msg;
3279 struct lttcomm_health_data reply;
3280
3281 DBG("[thread] Manage health check started");
3282
3283 rcu_register_thread();
3284
3285 /* We might hit an error path before this is created. */
3286 lttng_poll_init(&events);
3287
3288 /* Create unix socket */
3289 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3290 if (sock < 0) {
3291 ERR("Unable to create health check Unix socket");
3292 ret = -1;
3293 goto error;
3294 }
3295
3296 /*
3297 * Set the CLOEXEC flag. Return code is useless because either way, the
3298 * show must go on.
3299 */
3300 (void) utils_set_fd_cloexec(sock);
3301
3302 ret = lttcomm_listen_unix_sock(sock);
3303 if (ret < 0) {
3304 goto error;
3305 }
3306
3307 /*
3308 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3309 * more will be added to this poll set.
3310 */
3311 ret = sessiond_set_thread_pollset(&events, 2);
3312 if (ret < 0) {
3313 goto error;
3314 }
3315
3316 /* Add the application registration socket */
3317 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3318 if (ret < 0) {
3319 goto error;
3320 }
3321
3322 while (1) {
3323 DBG("Health check ready");
3324
3325 /* Inifinite blocking call, waiting for transmission */
3326 restart:
3327 ret = lttng_poll_wait(&events, -1);
3328 if (ret < 0) {
3329 /*
3330 * Restart interrupted system call.
3331 */
3332 if (errno == EINTR) {
3333 goto restart;
3334 }
3335 goto error;
3336 }
3337
3338 nb_fd = ret;
3339
3340 for (i = 0; i < nb_fd; i++) {
3341 /* Fetch once the poll data */
3342 revents = LTTNG_POLL_GETEV(&events, i);
3343 pollfd = LTTNG_POLL_GETFD(&events, i);
3344
3345 /* Thread quit pipe has been closed. Killing thread. */
3346 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3347 if (ret) {
3348 err = 0;
3349 goto exit;
3350 }
3351
3352 /* Event on the registration socket */
3353 if (pollfd == sock) {
3354 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3355 ERR("Health socket poll error");
3356 goto error;
3357 }
3358 }
3359 }
3360
3361 new_sock = lttcomm_accept_unix_sock(sock);
3362 if (new_sock < 0) {
3363 goto error;
3364 }
3365
3366 /*
3367 * Set the CLOEXEC flag. Return code is useless because either way, the
3368 * show must go on.
3369 */
3370 (void) utils_set_fd_cloexec(new_sock);
3371
3372 DBG("Receiving data from client for health...");
3373 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3374 if (ret <= 0) {
3375 DBG("Nothing recv() from client... continuing");
3376 ret = close(new_sock);
3377 if (ret) {
3378 PERROR("close");
3379 }
3380 new_sock = -1;
3381 continue;
3382 }
3383
3384 rcu_thread_online();
3385
3386 switch (msg.component) {
3387 case LTTNG_HEALTH_CMD:
3388 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3389 break;
3390 case LTTNG_HEALTH_APP_MANAGE:
3391 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3392 break;
3393 case LTTNG_HEALTH_APP_REG:
3394 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3395 break;
3396 case LTTNG_HEALTH_KERNEL:
3397 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3398 break;
3399 case LTTNG_HEALTH_CONSUMER:
3400 reply.ret_code = check_consumer_health();
3401 break;
3402 case LTTNG_HEALTH_HT_CLEANUP:
3403 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3404 break;
3405 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3406 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3407 break;
3408 case LTTNG_HEALTH_APP_REG_DISPATCH:
3409 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3410 break;
3411 case LTTNG_HEALTH_ALL:
3412 reply.ret_code =
3413 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3414 health_check_state(HEALTH_TYPE_APP_REG) &&
3415 health_check_state(HEALTH_TYPE_CMD) &&
3416 health_check_state(HEALTH_TYPE_KERNEL) &&
3417 check_consumer_health() &&
3418 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
3419 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3420 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3421 break;
3422 default:
3423 reply.ret_code = LTTNG_ERR_UND;
3424 break;
3425 }
3426
3427 /*
3428 * Flip ret value since 0 is a success and 1 indicates a bad health for
3429 * the client where in the sessiond it is the opposite. Again, this is
3430 * just to make things easier for us poor developer which enjoy a lot
3431 * lazyness.
3432 */
3433 if (reply.ret_code == 0 || reply.ret_code == 1) {
3434 reply.ret_code = !reply.ret_code;
3435 }
3436
3437 DBG2("Health check return value %d", reply.ret_code);
3438
3439 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3440 if (ret < 0) {
3441 ERR("Failed to send health data back to client");
3442 }
3443
3444 /* End of transmission */
3445 ret = close(new_sock);
3446 if (ret) {
3447 PERROR("close");
3448 }
3449 new_sock = -1;
3450 }
3451
3452 exit:
3453 error:
3454 if (err) {
3455 ERR("Health error occurred in %s", __func__);
3456 }
3457 DBG("Health check thread dying");
3458 unlink(health_unix_sock_path);
3459 if (sock >= 0) {
3460 ret = close(sock);
3461 if (ret) {
3462 PERROR("close");
3463 }
3464 }
3465
3466 lttng_poll_clean(&events);
3467
3468 rcu_unregister_thread();
3469 return NULL;
3470 }
3471
3472 /*
3473 * This thread manage all clients request using the unix client socket for
3474 * communication.
3475 */
3476 static void *thread_manage_clients(void *data)
3477 {
3478 int sock = -1, ret, i, pollfd, err = -1;
3479 int sock_error;
3480 uint32_t revents, nb_fd;
3481 struct command_ctx *cmd_ctx = NULL;
3482 struct lttng_poll_event events;
3483
3484 DBG("[thread] Manage client started");
3485
3486 rcu_register_thread();
3487
3488 health_register(HEALTH_TYPE_CMD);
3489
3490 if (testpoint(thread_manage_clients)) {
3491 goto error_testpoint;
3492 }
3493
3494 health_code_update();
3495
3496 ret = lttcomm_listen_unix_sock(client_sock);
3497 if (ret < 0) {
3498 goto error_listen;
3499 }
3500
3501 /*
3502 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3503 * more will be added to this poll set.
3504 */
3505 ret = sessiond_set_thread_pollset(&events, 2);
3506 if (ret < 0) {
3507 goto error_create_poll;
3508 }
3509
3510 /* Add the application registration socket */
3511 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3512 if (ret < 0) {
3513 goto error;
3514 }
3515
3516 /*
3517 * Notify parent pid that we are ready to accept command for client side.
3518 */
3519 if (opt_sig_parent) {
3520 kill(ppid, SIGUSR1);
3521 }
3522
3523 if (testpoint(thread_manage_clients_before_loop)) {
3524 goto error;
3525 }
3526
3527 health_code_update();
3528
3529 while (1) {
3530 DBG("Accepting client command ...");
3531
3532 /* Inifinite blocking call, waiting for transmission */
3533 restart:
3534 health_poll_entry();
3535 ret = lttng_poll_wait(&events, -1);
3536 health_poll_exit();
3537 if (ret < 0) {
3538 /*
3539 * Restart interrupted system call.
3540 */
3541 if (errno == EINTR) {
3542 goto restart;
3543 }
3544 goto error;
3545 }
3546
3547 nb_fd = ret;
3548
3549 for (i = 0; i < nb_fd; i++) {
3550 /* Fetch once the poll data */
3551 revents = LTTNG_POLL_GETEV(&events, i);
3552 pollfd = LTTNG_POLL_GETFD(&events, i);
3553
3554 health_code_update();
3555
3556 /* Thread quit pipe has been closed. Killing thread. */
3557 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3558 if (ret) {
3559 err = 0;
3560 goto exit;
3561 }
3562
3563 /* Event on the registration socket */
3564 if (pollfd == client_sock) {
3565 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3566 ERR("Client socket poll error");
3567 goto error;
3568 }
3569 }
3570 }
3571
3572 DBG("Wait for client response");
3573
3574 health_code_update();
3575
3576 sock = lttcomm_accept_unix_sock(client_sock);
3577 if (sock < 0) {
3578 goto error;
3579 }
3580
3581 /*
3582 * Set the CLOEXEC flag. Return code is useless because either way, the
3583 * show must go on.
3584 */
3585 (void) utils_set_fd_cloexec(sock);
3586
3587 /* Set socket option for credentials retrieval */
3588 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3589 if (ret < 0) {
3590 goto error;
3591 }
3592
3593 /* Allocate context command to process the client request */
3594 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3595 if (cmd_ctx == NULL) {
3596 PERROR("zmalloc cmd_ctx");
3597 goto error;
3598 }
3599
3600 /* Allocate data buffer for reception */
3601 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3602 if (cmd_ctx->lsm == NULL) {
3603 PERROR("zmalloc cmd_ctx->lsm");
3604 goto error;
3605 }
3606
3607 cmd_ctx->llm = NULL;
3608 cmd_ctx->session = NULL;
3609
3610 health_code_update();
3611
3612 /*
3613 * Data is received from the lttng client. The struct
3614 * lttcomm_session_msg (lsm) contains the command and data request of
3615 * the client.
3616 */
3617 DBG("Receiving data from client ...");
3618 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3619 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3620 if (ret <= 0) {
3621 DBG("Nothing recv() from client... continuing");
3622 ret = close(sock);
3623 if (ret) {
3624 PERROR("close");
3625 }
3626 sock = -1;
3627 clean_command_ctx(&cmd_ctx);
3628 continue;
3629 }
3630
3631 health_code_update();
3632
3633 // TODO: Validate cmd_ctx including sanity check for
3634 // security purpose.
3635
3636 rcu_thread_online();
3637 /*
3638 * This function dispatch the work to the kernel or userspace tracer
3639 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3640 * informations for the client. The command context struct contains
3641 * everything this function may needs.
3642 */
3643 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3644 rcu_thread_offline();
3645 if (ret < 0) {
3646 ret = close(sock);
3647 if (ret) {
3648 PERROR("close");
3649 }
3650 sock = -1;
3651 /*
3652 * TODO: Inform client somehow of the fatal error. At
3653 * this point, ret < 0 means that a zmalloc failed
3654 * (ENOMEM). Error detected but still accept
3655 * command, unless a socket error has been
3656 * detected.
3657 */
3658 clean_command_ctx(&cmd_ctx);
3659 continue;
3660 }
3661
3662 health_code_update();
3663
3664 DBG("Sending response (size: %d, retcode: %s)",
3665 cmd_ctx->lttng_msg_size,
3666 lttng_strerror(-cmd_ctx->llm->ret_code));
3667 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3668 if (ret < 0) {
3669 ERR("Failed to send data back to client");
3670 }
3671
3672 /* End of transmission */
3673 ret = close(sock);
3674 if (ret) {
3675 PERROR("close");
3676 }
3677 sock = -1;
3678
3679 clean_command_ctx(&cmd_ctx);
3680
3681 health_code_update();
3682 }
3683
3684 exit:
3685 error:
3686 if (sock >= 0) {
3687 ret = close(sock);
3688 if (ret) {
3689 PERROR("close");
3690 }
3691 }
3692
3693 lttng_poll_clean(&events);
3694 clean_command_ctx(&cmd_ctx);
3695
3696 error_listen:
3697 error_create_poll:
3698 error_testpoint:
3699 unlink(client_unix_sock_path);
3700 if (client_sock >= 0) {
3701 ret = close(client_sock);
3702 if (ret) {
3703 PERROR("close");
3704 }
3705 }
3706
3707 if (err) {
3708 health_error();
3709 ERR("Health error occurred in %s", __func__);
3710 }
3711
3712 health_unregister();
3713
3714 DBG("Client thread dying");
3715
3716 rcu_unregister_thread();
3717 return NULL;
3718 }
3719
3720
3721 /*
3722 * usage function on stderr
3723 */
3724 static void usage(void)
3725 {
3726 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3727 fprintf(stderr, " -h, --help Display this usage.\n");
3728 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3729 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3730 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3731 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3732 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3733 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3734 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3735 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3736 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3737 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3738 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3739 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3740 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3741 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3742 fprintf(stderr, " -V, --version Show version number.\n");
3743 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3744 fprintf(stderr, " -q, --quiet No output at all.\n");
3745 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3746 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3747 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3748 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3749 }
3750
3751 /*
3752 * daemon argument parsing
3753 */
3754 static int parse_args(int argc, char **argv)
3755 {
3756 int c;
3757
3758 static struct option long_options[] = {
3759 { "client-sock", 1, 0, 'c' },
3760 { "apps-sock", 1, 0, 'a' },
3761 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3762 { "kconsumerd-err-sock", 1, 0, 'E' },
3763 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3764 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3765 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3766 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3767 { "consumerd32-path", 1, 0, 'u' },
3768 { "consumerd32-libdir", 1, 0, 'U' },
3769 { "consumerd64-path", 1, 0, 't' },
3770 { "consumerd64-libdir", 1, 0, 'T' },
3771 { "daemonize", 0, 0, 'd' },
3772 { "sig-parent", 0, 0, 'S' },
3773 { "help", 0, 0, 'h' },
3774 { "group", 1, 0, 'g' },
3775 { "version", 0, 0, 'V' },
3776 { "quiet", 0, 0, 'q' },
3777 { "verbose", 0, 0, 'v' },
3778 { "verbose-consumer", 0, 0, 'Z' },
3779 { "no-kernel", 0, 0, 'N' },
3780 { "pidfile", 1, 0, 'p' },
3781 { NULL, 0, 0, 0 }
3782 };
3783
3784 while (1) {
3785 int option_index = 0;
3786 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3787 long_options, &option_index);
3788 if (c == -1) {
3789 break;
3790 }
3791
3792 switch (c) {
3793 case 0:
3794 fprintf(stderr, "option %s", long_options[option_index].name);
3795 if (optarg) {
3796 fprintf(stderr, " with arg %s\n", optarg);
3797 }
3798 break;
3799 case 'c':
3800 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3801 break;
3802 case 'a':
3803 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3804 break;
3805 case 'd':
3806 opt_daemon = 1;
3807 break;
3808 case 'g':
3809 opt_tracing_group = optarg;
3810 break;
3811 case 'h':
3812 usage();
3813 exit(EXIT_FAILURE);
3814 case 'V':
3815 fprintf(stdout, "%s\n", VERSION);
3816 exit(EXIT_SUCCESS);
3817 case 'S':
3818 opt_sig_parent = 1;
3819 break;
3820 case 'E':
3821 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3822 break;
3823 case 'C':
3824 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3825 break;
3826 case 'F':
3827 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3828 break;
3829 case 'D':
3830 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3831 break;
3832 case 'H':
3833 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3834 break;
3835 case 'G':
3836 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3837 break;
3838 case 'N':
3839 opt_no_kernel = 1;
3840 break;
3841 case 'q':
3842 lttng_opt_quiet = 1;
3843 break;
3844 case 'v':
3845 /* Verbose level can increase using multiple -v */
3846 lttng_opt_verbose += 1;
3847 break;
3848 case 'Z':
3849 opt_verbose_consumer += 1;
3850 break;
3851 case 'u':
3852 consumerd32_bin= optarg;
3853 break;
3854 case 'U':
3855 consumerd32_libdir = optarg;
3856 break;
3857 case 't':
3858 consumerd64_bin = optarg;
3859 break;
3860 case 'T':
3861 consumerd64_libdir = optarg;
3862 break;
3863 case 'p':
3864 opt_pidfile = optarg;
3865 break;
3866 default:
3867 /* Unknown option or other error.
3868 * Error is printed by getopt, just return */
3869 return -1;
3870 }
3871 }
3872
3873 return 0;
3874 }
3875
3876 /*
3877 * Creates the two needed socket by the daemon.
3878 * apps_sock - The communication socket for all UST apps.
3879 * client_sock - The communication of the cli tool (lttng).
3880 */
3881 static int init_daemon_socket(void)
3882 {
3883 int ret = 0;
3884 mode_t old_umask;
3885
3886 old_umask = umask(0);
3887
3888 /* Create client tool unix socket */
3889 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3890 if (client_sock < 0) {
3891 ERR("Create unix sock failed: %s", client_unix_sock_path);
3892 ret = -1;
3893 goto end;
3894 }
3895
3896 /* Set the cloexec flag */
3897 ret = utils_set_fd_cloexec(client_sock);
3898 if (ret < 0) {
3899 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3900 "Continuing but note that the consumer daemon will have a "
3901 "reference to this socket on exec()", client_sock);
3902 }
3903
3904 /* File permission MUST be 660 */
3905 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3906 if (ret < 0) {
3907 ERR("Set file permissions failed: %s", client_unix_sock_path);
3908 PERROR("chmod");
3909 goto end;
3910 }
3911
3912 /* Create the application unix socket */
3913 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3914 if (apps_sock < 0) {
3915 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3916 ret = -1;
3917 goto end;
3918 }
3919
3920 /* Set the cloexec flag */
3921 ret = utils_set_fd_cloexec(apps_sock);
3922 if (ret < 0) {
3923 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3924 "Continuing but note that the consumer daemon will have a "
3925 "reference to this socket on exec()", apps_sock);
3926 }
3927
3928 /* File permission MUST be 666 */
3929 ret = chmod(apps_unix_sock_path,
3930 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3931 if (ret < 0) {
3932 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3933 PERROR("chmod");
3934 goto end;
3935 }
3936
3937 DBG3("Session daemon client socket %d and application socket %d created",
3938 client_sock, apps_sock);
3939
3940 end:
3941 umask(old_umask);
3942 return ret;
3943 }
3944
3945 /*
3946 * Check if the global socket is available, and if a daemon is answering at the
3947 * other side. If yes, error is returned.
3948 */
3949 static int check_existing_daemon(void)
3950 {
3951 /* Is there anybody out there ? */
3952 if (lttng_session_daemon_alive()) {
3953 return -EEXIST;
3954 }
3955
3956 return 0;
3957 }
3958
3959 /*
3960 * Set the tracing group gid onto the client socket.
3961 *
3962 * Race window between mkdir and chown is OK because we are going from more
3963 * permissive (root.root) to less permissive (root.tracing).
3964 */
3965 static int set_permissions(char *rundir)
3966 {
3967 int ret;
3968 gid_t gid;
3969
3970 ret = allowed_group();
3971 if (ret < 0) {
3972 WARN("No tracing group detected");
3973 ret = 0;
3974 goto end;
3975 }
3976
3977 gid = ret;
3978
3979 /* Set lttng run dir */
3980 ret = chown(rundir, 0, gid);
3981 if (ret < 0) {
3982 ERR("Unable to set group on %s", rundir);
3983 PERROR("chown");
3984 }
3985
3986 /* Ensure tracing group can search the run dir */
3987 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3988 if (ret < 0) {
3989 ERR("Unable to set permissions on %s", rundir);
3990 PERROR("chmod");
3991 }
3992
3993 /* lttng client socket path */
3994 ret = chown(client_unix_sock_path, 0, gid);
3995 if (ret < 0) {
3996 ERR("Unable to set group on %s", client_unix_sock_path);
3997 PERROR("chown");
3998 }
3999
4000 /* kconsumer error socket path */
4001 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
4002 if (ret < 0) {
4003 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4004 PERROR("chown");
4005 }
4006
4007 /* 64-bit ustconsumer error socket path */
4008 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
4009 if (ret < 0) {
4010 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4011 PERROR("chown");
4012 }
4013
4014 /* 32-bit ustconsumer compat32 error socket path */
4015 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
4016 if (ret < 0) {
4017 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4018 PERROR("chown");
4019 }
4020
4021 DBG("All permissions are set");
4022
4023 end:
4024 return ret;
4025 }
4026
4027 /*
4028 * Create the lttng run directory needed for all global sockets and pipe.
4029 */
4030 static int create_lttng_rundir(const char *rundir)
4031 {
4032 int ret;
4033
4034 DBG3("Creating LTTng run directory: %s", rundir);
4035
4036 ret = mkdir(rundir, S_IRWXU);
4037 if (ret < 0) {
4038 if (errno != EEXIST) {
4039 ERR("Unable to create %s", rundir);
4040 goto error;
4041 } else {
4042 ret = 0;
4043 }
4044 }
4045
4046 error:
4047 return ret;
4048 }
4049
4050 /*
4051 * Setup sockets and directory needed by the kconsumerd communication with the
4052 * session daemon.
4053 */
4054 static int set_consumer_sockets(struct consumer_data *consumer_data,
4055 const char *rundir)
4056 {
4057 int ret;
4058 char path[PATH_MAX];
4059
4060 switch (consumer_data->type) {
4061 case LTTNG_CONSUMER_KERNEL:
4062 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4063 break;
4064 case LTTNG_CONSUMER64_UST:
4065 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4066 break;
4067 case LTTNG_CONSUMER32_UST:
4068 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4069 break;
4070 default:
4071 ERR("Consumer type unknown");
4072 ret = -EINVAL;
4073 goto error;
4074 }
4075
4076 DBG2("Creating consumer directory: %s", path);
4077
4078 ret = mkdir(path, S_IRWXU);
4079 if (ret < 0) {
4080 if (errno != EEXIST) {
4081 PERROR("mkdir");
4082 ERR("Failed to create %s", path);
4083 goto error;
4084 }
4085 ret = -1;
4086 }
4087
4088 /* Create the kconsumerd error unix socket */
4089 consumer_data->err_sock =
4090 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4091 if (consumer_data->err_sock < 0) {
4092 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4093 ret = -1;
4094 goto error;
4095 }
4096
4097 /* File permission MUST be 660 */
4098 ret = chmod(consumer_data->err_unix_sock_path,
4099 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4100 if (ret < 0) {
4101 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4102 PERROR("chmod");
4103 goto error;
4104 }
4105
4106 error:
4107 return ret;
4108 }
4109
4110 /*
4111 * Signal handler for the daemon
4112 *
4113 * Simply stop all worker threads, leaving main() return gracefully after
4114 * joining all threads and calling cleanup().
4115 */
4116 static void sighandler(int sig)
4117 {
4118 switch (sig) {
4119 case SIGPIPE:
4120 DBG("SIGPIPE caught");
4121 return;
4122 case SIGINT:
4123 DBG("SIGINT caught");
4124 stop_threads();
4125 break;
4126 case SIGTERM:
4127 DBG("SIGTERM caught");
4128 stop_threads();
4129 break;
4130 default:
4131 break;
4132 }
4133 }
4134
4135 /*
4136 * Setup signal handler for :
4137 * SIGINT, SIGTERM, SIGPIPE
4138 */
4139 static int set_signal_handler(void)
4140 {
4141 int ret = 0;
4142 struct sigaction sa;
4143 sigset_t sigset;
4144
4145 if ((ret = sigemptyset(&sigset)) < 0) {
4146 PERROR("sigemptyset");
4147 return ret;
4148 }
4149
4150 sa.sa_handler = sighandler;
4151 sa.sa_mask = sigset;
4152 sa.sa_flags = 0;
4153 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4154 PERROR("sigaction");
4155 return ret;
4156 }
4157
4158 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4159 PERROR("sigaction");
4160 return ret;
4161 }
4162
4163 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4164 PERROR("sigaction");
4165 return ret;
4166 }
4167
4168 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4169
4170 return ret;
4171 }
4172
4173 /*
4174 * Set open files limit to unlimited. This daemon can open a large number of
4175 * file descriptors in order to consumer multiple kernel traces.
4176 */
4177 static void set_ulimit(void)
4178 {
4179 int ret;
4180 struct rlimit lim;
4181
4182 /* The kernel does not allowed an infinite limit for open files */
4183 lim.rlim_cur = 65535;
4184 lim.rlim_max = 65535;
4185
4186 ret = setrlimit(RLIMIT_NOFILE, &lim);
4187 if (ret < 0) {
4188 PERROR("failed to set open files limit");
4189 }
4190 }
4191
4192 /*
4193 * Write pidfile using the rundir and opt_pidfile.
4194 */
4195 static void write_pidfile(void)
4196 {
4197 int ret;
4198 char pidfile_path[PATH_MAX];
4199
4200 assert(rundir);
4201
4202 if (opt_pidfile) {
4203 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4204 } else {
4205 /* Build pidfile path from rundir and opt_pidfile. */
4206 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4207 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4208 if (ret < 0) {
4209 PERROR("snprintf pidfile path");
4210 goto error;
4211 }
4212 }
4213
4214 /*
4215 * Create pid file in rundir. Return value is of no importance. The
4216 * execution will continue even though we are not able to write the file.
4217 */
4218 (void) utils_create_pid_file(getpid(), pidfile_path);
4219
4220 error:
4221 return;
4222 }
4223
4224 /*
4225 * main
4226 */
4227 int main(int argc, char **argv)
4228 {
4229 int ret = 0;
4230 void *status;
4231 const char *home_path, *env_app_timeout;
4232
4233 init_kernel_workarounds();
4234
4235 rcu_register_thread();
4236
4237 setup_consumerd_path();
4238
4239 page_size = sysconf(_SC_PAGESIZE);
4240 if (page_size < 0) {
4241 PERROR("sysconf _SC_PAGESIZE");
4242 page_size = LONG_MAX;
4243 WARN("Fallback page size to %ld", page_size);
4244 }
4245
4246 /* Parse arguments */
4247 progname = argv[0];
4248 if ((ret = parse_args(argc, argv)) < 0) {
4249 goto error;
4250 }
4251
4252 /* Daemonize */
4253 if (opt_daemon) {
4254 int i;
4255
4256 /*
4257 * fork
4258 * child: setsid, close FD 0, 1, 2, chdir /
4259 * parent: exit (if fork is successful)
4260 */
4261 ret = daemon(0, 0);
4262 if (ret < 0) {
4263 PERROR("daemon");
4264 goto error;
4265 }
4266 /*
4267 * We are in the child. Make sure all other file
4268 * descriptors are closed, in case we are called with
4269 * more opened file descriptors than the standard ones.
4270 */
4271 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4272 (void) close(i);
4273 }
4274 }
4275
4276 /* Create thread quit pipe */
4277 if ((ret = init_thread_quit_pipe()) < 0) {
4278 goto error;
4279 }
4280
4281 /* Check if daemon is UID = 0 */
4282 is_root = !getuid();
4283
4284 if (is_root) {
4285 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4286
4287 /* Create global run dir with root access */
4288 ret = create_lttng_rundir(rundir);
4289 if (ret < 0) {
4290 goto error;
4291 }
4292
4293 if (strlen(apps_unix_sock_path) == 0) {
4294 snprintf(apps_unix_sock_path, PATH_MAX,
4295 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4296 }
4297
4298 if (strlen(client_unix_sock_path) == 0) {
4299 snprintf(client_unix_sock_path, PATH_MAX,
4300 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4301 }
4302
4303 /* Set global SHM for ust */
4304 if (strlen(wait_shm_path) == 0) {
4305 snprintf(wait_shm_path, PATH_MAX,
4306 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4307 }
4308
4309 if (strlen(health_unix_sock_path) == 0) {
4310 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4311 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4312 }
4313
4314 /* Setup kernel consumerd path */
4315 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4316 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4317 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4318 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4319
4320 DBG2("Kernel consumer err path: %s",
4321 kconsumer_data.err_unix_sock_path);
4322 DBG2("Kernel consumer cmd path: %s",
4323 kconsumer_data.cmd_unix_sock_path);
4324 } else {
4325 home_path = utils_get_home_dir();
4326 if (home_path == NULL) {
4327 /* TODO: Add --socket PATH option */
4328 ERR("Can't get HOME directory for sockets creation.");
4329 ret = -EPERM;
4330 goto error;
4331 }
4332
4333 /*
4334 * Create rundir from home path. This will create something like
4335 * $HOME/.lttng
4336 */
4337 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4338 if (ret < 0) {
4339 ret = -ENOMEM;
4340 goto error;
4341 }
4342
4343 ret = create_lttng_rundir(rundir);
4344 if (ret < 0) {
4345 goto error;
4346 }
4347
4348 if (strlen(apps_unix_sock_path) == 0) {
4349 snprintf(apps_unix_sock_path, PATH_MAX,
4350 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4351 }
4352
4353 /* Set the cli tool unix socket path */
4354 if (strlen(client_unix_sock_path) == 0) {
4355 snprintf(client_unix_sock_path, PATH_MAX,
4356 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4357 }
4358
4359 /* Set global SHM for ust */
4360 if (strlen(wait_shm_path) == 0) {
4361 snprintf(wait_shm_path, PATH_MAX,
4362 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4363 }
4364
4365 /* Set health check Unix path */
4366 if (strlen(health_unix_sock_path) == 0) {
4367 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4368 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4369 }
4370 }
4371
4372 /* Set consumer initial state */
4373 kernel_consumerd_state = CONSUMER_STOPPED;
4374 ust_consumerd_state = CONSUMER_STOPPED;
4375
4376 DBG("Client socket path %s", client_unix_sock_path);
4377 DBG("Application socket path %s", apps_unix_sock_path);
4378 DBG("Application wait path %s", wait_shm_path);
4379 DBG("LTTng run directory path: %s", rundir);
4380
4381 /* 32 bits consumerd path setup */
4382 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4383 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4384 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4385 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4386
4387 DBG2("UST consumer 32 bits err path: %s",
4388 ustconsumer32_data.err_unix_sock_path);
4389 DBG2("UST consumer 32 bits cmd path: %s",
4390 ustconsumer32_data.cmd_unix_sock_path);
4391
4392 /* 64 bits consumerd path setup */
4393 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4394 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4395 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4396 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4397
4398 DBG2("UST consumer 64 bits err path: %s",
4399 ustconsumer64_data.err_unix_sock_path);
4400 DBG2("UST consumer 64 bits cmd path: %s",
4401 ustconsumer64_data.cmd_unix_sock_path);
4402
4403 /*
4404 * See if daemon already exist.
4405 */
4406 if ((ret = check_existing_daemon()) < 0) {
4407 ERR("Already running daemon.\n");
4408 /*
4409 * We do not goto exit because we must not cleanup()
4410 * because a daemon is already running.
4411 */
4412 goto error;
4413 }
4414
4415 /*
4416 * Init UST app hash table. Alloc hash table before this point since
4417 * cleanup() can get called after that point.
4418 */
4419 ust_app_ht_alloc();
4420
4421 /* After this point, we can safely call cleanup() with "goto exit" */
4422
4423 /*
4424 * These actions must be executed as root. We do that *after* setting up
4425 * the sockets path because we MUST make the check for another daemon using
4426 * those paths *before* trying to set the kernel consumer sockets and init
4427 * kernel tracer.
4428 */
4429 if (is_root) {
4430 ret = set_consumer_sockets(&kconsumer_data, rundir);
4431 if (ret < 0) {
4432 goto exit;
4433 }
4434
4435 /* Setup kernel tracer */
4436 if (!opt_no_kernel) {
4437 init_kernel_tracer();
4438 }
4439
4440 /* Set ulimit for open files */
4441 set_ulimit();
4442 }
4443 /* init lttng_fd tracking must be done after set_ulimit. */
4444 lttng_fd_init();
4445
4446 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4447 if (ret < 0) {
4448 goto exit;
4449 }
4450
4451 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4452 if (ret < 0) {
4453 goto exit;
4454 }
4455
4456 if ((ret = set_signal_handler()) < 0) {
4457 goto exit;
4458 }
4459
4460 /* Setup the needed unix socket */
4461 if ((ret = init_daemon_socket()) < 0) {
4462 goto exit;
4463 }
4464
4465 /* Set credentials to socket */
4466 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4467 goto exit;
4468 }
4469
4470 /* Get parent pid if -S, --sig-parent is specified. */
4471 if (opt_sig_parent) {
4472 ppid = getppid();
4473 }
4474
4475 /* Setup the kernel pipe for waking up the kernel thread */
4476 if (is_root && !opt_no_kernel) {
4477 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4478 goto exit;
4479 }
4480 }
4481
4482 /* Setup the thread ht_cleanup communication pipe. */
4483 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4484 goto exit;
4485 }
4486
4487 /* Setup the thread apps communication pipe. */
4488 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4489 goto exit;
4490 }
4491
4492 /* Setup the thread apps notify communication pipe. */
4493 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4494 goto exit;
4495 }
4496
4497 /* Initialize global buffer per UID and PID registry. */
4498 buffer_reg_init_uid_registry();
4499 buffer_reg_init_pid_registry();
4500
4501 /* Init UST command queue. */
4502 cds_wfq_init(&ust_cmd_queue.queue);
4503
4504 /*
4505 * Get session list pointer. This pointer MUST NOT be free(). This list is
4506 * statically declared in session.c
4507 */
4508 session_list_ptr = session_get_list();
4509
4510 /* Set up max poll set size */
4511 lttng_poll_set_max_size();
4512
4513 cmd_init();
4514
4515 /* Check for the application socket timeout env variable. */
4516 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4517 if (env_app_timeout) {
4518 app_socket_timeout = atoi(env_app_timeout);
4519 } else {
4520 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4521 }
4522
4523 write_pidfile();
4524
4525 /* Create thread to manage the client socket */
4526 ret = pthread_create(&ht_cleanup_thread, NULL,
4527 thread_ht_cleanup, (void *) NULL);
4528 if (ret != 0) {
4529 PERROR("pthread_create ht_cleanup");
4530 goto exit_ht_cleanup;
4531 }
4532
4533 /* Create thread to manage the client socket */
4534 ret = pthread_create(&health_thread, NULL,
4535 thread_manage_health, (void *) NULL);
4536 if (ret != 0) {
4537 PERROR("pthread_create health");
4538 goto exit_health;
4539 }
4540
4541 /* Create thread to manage the client socket */
4542 ret = pthread_create(&client_thread, NULL,
4543 thread_manage_clients, (void *) NULL);
4544 if (ret != 0) {
4545 PERROR("pthread_create clients");
4546 goto exit_client;
4547 }
4548
4549 /* Create thread to dispatch registration */
4550 ret = pthread_create(&dispatch_thread, NULL,
4551 thread_dispatch_ust_registration, (void *) NULL);
4552 if (ret != 0) {
4553 PERROR("pthread_create dispatch");
4554 goto exit_dispatch;
4555 }
4556
4557 /* Create thread to manage application registration. */
4558 ret = pthread_create(&reg_apps_thread, NULL,
4559 thread_registration_apps, (void *) NULL);
4560 if (ret != 0) {
4561 PERROR("pthread_create registration");
4562 goto exit_reg_apps;
4563 }
4564
4565 /* Create thread to manage application socket */
4566 ret = pthread_create(&apps_thread, NULL,
4567 thread_manage_apps, (void *) NULL);
4568 if (ret != 0) {
4569 PERROR("pthread_create apps");
4570 goto exit_apps;
4571 }
4572
4573 /* Create thread to manage application notify socket */
4574 ret = pthread_create(&apps_notify_thread, NULL,
4575 ust_thread_manage_notify, (void *) NULL);
4576 if (ret != 0) {
4577 PERROR("pthread_create apps");
4578 goto exit_apps;
4579 }
4580
4581 /* Don't start this thread if kernel tracing is not requested nor root */
4582 if (is_root && !opt_no_kernel) {
4583 /* Create kernel thread to manage kernel event */
4584 ret = pthread_create(&kernel_thread, NULL,
4585 thread_manage_kernel, (void *) NULL);
4586 if (ret != 0) {
4587 PERROR("pthread_create kernel");
4588 goto exit_kernel;
4589 }
4590
4591 ret = pthread_join(kernel_thread, &status);
4592 if (ret != 0) {
4593 PERROR("pthread_join");
4594 goto error; /* join error, exit without cleanup */
4595 }
4596 }
4597
4598 exit_kernel:
4599 ret = pthread_join(apps_thread, &status);
4600 if (ret != 0) {
4601 PERROR("pthread_join");
4602 goto error; /* join error, exit without cleanup */
4603 }
4604
4605 exit_apps:
4606 ret = pthread_join(reg_apps_thread, &status);
4607 if (ret != 0) {
4608 PERROR("pthread_join");
4609 goto error; /* join error, exit without cleanup */
4610 }
4611
4612 exit_reg_apps:
4613 ret = pthread_join(dispatch_thread, &status);
4614 if (ret != 0) {
4615 PERROR("pthread_join");
4616 goto error; /* join error, exit without cleanup */
4617 }
4618
4619 exit_dispatch:
4620 ret = pthread_join(client_thread, &status);
4621 if (ret != 0) {
4622 PERROR("pthread_join");
4623 goto error; /* join error, exit without cleanup */
4624 }
4625
4626 ret = join_consumer_thread(&kconsumer_data);
4627 if (ret != 0) {
4628 PERROR("join_consumer");
4629 goto error; /* join error, exit without cleanup */
4630 }
4631
4632 ret = join_consumer_thread(&ustconsumer32_data);
4633 if (ret != 0) {
4634 PERROR("join_consumer ust32");
4635 goto error; /* join error, exit without cleanup */
4636 }
4637
4638 ret = join_consumer_thread(&ustconsumer64_data);
4639 if (ret != 0) {
4640 PERROR("join_consumer ust64");
4641 goto error; /* join error, exit without cleanup */
4642 }
4643
4644 exit_client:
4645 ret = pthread_join(health_thread, &status);
4646 if (ret != 0) {
4647 PERROR("pthread_join health thread");
4648 goto error; /* join error, exit without cleanup */
4649 }
4650
4651 exit_health:
4652 ret = pthread_join(ht_cleanup_thread, &status);
4653 if (ret != 0) {
4654 PERROR("pthread_join ht cleanup thread");
4655 goto error; /* join error, exit without cleanup */
4656 }
4657 exit_ht_cleanup:
4658 exit:
4659 /*
4660 * cleanup() is called when no other thread is running.
4661 */
4662 rcu_thread_online();
4663 cleanup();
4664 rcu_thread_offline();
4665 rcu_unregister_thread();
4666 if (!ret) {
4667 exit(EXIT_SUCCESS);
4668 }
4669 error:
4670 exit(EXIT_FAILURE);
4671 }
This page took 0.197051 seconds and 5 git commands to generate.