Fix: consumer err_sock cloexec
[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 * Close every consumer sockets.
395 */
396 static void close_consumer_sockets(void)
397 {
398 int ret;
399
400 if (kconsumer_data.err_sock >= 0) {
401 ret = close(kconsumer_data.err_sock);
402 if (ret < 0) {
403 PERROR("kernel consumer err_sock close");
404 }
405 }
406 if (ustconsumer32_data.err_sock >= 0) {
407 ret = close(ustconsumer32_data.err_sock);
408 if (ret < 0) {
409 PERROR("UST consumerd32 err_sock close");
410 }
411 }
412 if (ustconsumer64_data.err_sock >= 0) {
413 ret = close(ustconsumer64_data.err_sock);
414 if (ret < 0) {
415 PERROR("UST consumerd64 err_sock close");
416 }
417 }
418 if (kconsumer_data.cmd_sock >= 0) {
419 ret = close(kconsumer_data.cmd_sock);
420 if (ret < 0) {
421 PERROR("kernel consumer cmd_sock close");
422 }
423 }
424 if (ustconsumer32_data.cmd_sock >= 0) {
425 ret = close(ustconsumer32_data.cmd_sock);
426 if (ret < 0) {
427 PERROR("UST consumerd32 cmd_sock close");
428 }
429 }
430 if (ustconsumer64_data.cmd_sock >= 0) {
431 ret = close(ustconsumer64_data.cmd_sock);
432 if (ret < 0) {
433 PERROR("UST consumerd64 cmd_sock close");
434 }
435 }
436 }
437
438 /*
439 * Cleanup the daemon
440 */
441 static void cleanup(void)
442 {
443 int ret;
444 char *cmd = NULL;
445 struct ltt_session *sess, *stmp;
446
447 DBG("Cleaning up");
448
449 /* First thing first, stop all threads */
450 utils_close_pipe(thread_quit_pipe);
451
452 /*
453 * If opt_pidfile is undefined, the default file will be wiped when
454 * removing the rundir.
455 */
456 if (opt_pidfile) {
457 ret = remove(opt_pidfile);
458 if (ret < 0) {
459 PERROR("remove pidfile %s", opt_pidfile);
460 }
461 }
462
463 DBG("Removing %s directory", rundir);
464 ret = asprintf(&cmd, "rm -rf %s", rundir);
465 if (ret < 0) {
466 ERR("asprintf failed. Something is really wrong!");
467 }
468
469 /* Remove lttng run directory */
470 ret = system(cmd);
471 if (ret < 0) {
472 ERR("Unable to clean %s", rundir);
473 }
474 free(cmd);
475 free(rundir);
476
477 DBG("Cleaning up all sessions");
478
479 /* Destroy session list mutex */
480 if (session_list_ptr != NULL) {
481 pthread_mutex_destroy(&session_list_ptr->lock);
482
483 /* Cleanup ALL session */
484 cds_list_for_each_entry_safe(sess, stmp,
485 &session_list_ptr->head, list) {
486 cmd_destroy_session(sess, kernel_poll_pipe[1]);
487 }
488 }
489
490 DBG("Closing all UST sockets");
491 ust_app_clean_list();
492 buffer_reg_destroy_registries();
493
494 if (is_root && !opt_no_kernel) {
495 DBG2("Closing kernel fd");
496 if (kernel_tracer_fd >= 0) {
497 ret = close(kernel_tracer_fd);
498 if (ret) {
499 PERROR("close");
500 }
501 }
502 DBG("Unloading kernel modules");
503 modprobe_remove_lttng_all();
504 }
505
506 close_consumer_sockets();
507
508 /* <fun> */
509 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
510 "Matthew, BEET driven development works!%c[%dm",
511 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
512 /* </fun> */
513 }
514
515 /*
516 * Send data on a unix socket using the liblttsessiondcomm API.
517 *
518 * Return lttcomm error code.
519 */
520 static int send_unix_sock(int sock, void *buf, size_t len)
521 {
522 /* Check valid length */
523 if (len == 0) {
524 return -1;
525 }
526
527 return lttcomm_send_unix_sock(sock, buf, len);
528 }
529
530 /*
531 * Free memory of a command context structure.
532 */
533 static void clean_command_ctx(struct command_ctx **cmd_ctx)
534 {
535 DBG("Clean command context structure");
536 if (*cmd_ctx) {
537 if ((*cmd_ctx)->llm) {
538 free((*cmd_ctx)->llm);
539 }
540 if ((*cmd_ctx)->lsm) {
541 free((*cmd_ctx)->lsm);
542 }
543 free(*cmd_ctx);
544 *cmd_ctx = NULL;
545 }
546 }
547
548 /*
549 * Notify UST applications using the shm mmap futex.
550 */
551 static int notify_ust_apps(int active)
552 {
553 char *wait_shm_mmap;
554
555 DBG("Notifying applications of session daemon state: %d", active);
556
557 /* See shm.c for this call implying mmap, shm and futex calls */
558 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
559 if (wait_shm_mmap == NULL) {
560 goto error;
561 }
562
563 /* Wake waiting process */
564 futex_wait_update((int32_t *) wait_shm_mmap, active);
565
566 /* Apps notified successfully */
567 return 0;
568
569 error:
570 return -1;
571 }
572
573 /*
574 * Setup the outgoing data buffer for the response (llm) by allocating the
575 * right amount of memory and copying the original information from the lsm
576 * structure.
577 *
578 * Return total size of the buffer pointed by buf.
579 */
580 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
581 {
582 int ret, buf_size;
583
584 buf_size = size;
585
586 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
587 if (cmd_ctx->llm == NULL) {
588 PERROR("zmalloc");
589 ret = -ENOMEM;
590 goto error;
591 }
592
593 /* Copy common data */
594 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
595 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
596
597 cmd_ctx->llm->data_size = size;
598 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
599
600 return buf_size;
601
602 error:
603 return ret;
604 }
605
606 /*
607 * Update the kernel poll set of all channel fd available over all tracing
608 * session. Add the wakeup pipe at the end of the set.
609 */
610 static int update_kernel_poll(struct lttng_poll_event *events)
611 {
612 int ret;
613 struct ltt_session *session;
614 struct ltt_kernel_channel *channel;
615
616 DBG("Updating kernel poll set");
617
618 session_lock_list();
619 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
620 session_lock(session);
621 if (session->kernel_session == NULL) {
622 session_unlock(session);
623 continue;
624 }
625
626 cds_list_for_each_entry(channel,
627 &session->kernel_session->channel_list.head, list) {
628 /* Add channel fd to the kernel poll set */
629 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
630 if (ret < 0) {
631 session_unlock(session);
632 goto error;
633 }
634 DBG("Channel fd %d added to kernel set", channel->fd);
635 }
636 session_unlock(session);
637 }
638 session_unlock_list();
639
640 return 0;
641
642 error:
643 session_unlock_list();
644 return -1;
645 }
646
647 /*
648 * Find the channel fd from 'fd' over all tracing session. When found, check
649 * for new channel stream and send those stream fds to the kernel consumer.
650 *
651 * Useful for CPU hotplug feature.
652 */
653 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
654 {
655 int ret = 0;
656 struct ltt_session *session;
657 struct ltt_kernel_session *ksess;
658 struct ltt_kernel_channel *channel;
659
660 DBG("Updating kernel streams for channel fd %d", fd);
661
662 session_lock_list();
663 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
664 session_lock(session);
665 if (session->kernel_session == NULL) {
666 session_unlock(session);
667 continue;
668 }
669 ksess = session->kernel_session;
670
671 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
672 if (channel->fd == fd) {
673 DBG("Channel found, updating kernel streams");
674 ret = kernel_open_channel_stream(channel);
675 if (ret < 0) {
676 goto error;
677 }
678
679 /*
680 * Have we already sent fds to the consumer? If yes, it means
681 * that tracing is started so it is safe to send our updated
682 * stream fds.
683 */
684 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
685 struct lttng_ht_iter iter;
686 struct consumer_socket *socket;
687
688 rcu_read_lock();
689 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
690 &iter.iter, socket, node.node) {
691 /* Code flow error */
692 assert(socket->fd >= 0);
693
694 pthread_mutex_lock(socket->lock);
695 ret = kernel_consumer_send_channel_stream(socket,
696 channel, ksess);
697 pthread_mutex_unlock(socket->lock);
698 if (ret < 0) {
699 rcu_read_unlock();
700 goto error;
701 }
702 }
703 rcu_read_unlock();
704 }
705 goto error;
706 }
707 }
708 session_unlock(session);
709 }
710 session_unlock_list();
711 return ret;
712
713 error:
714 session_unlock(session);
715 session_unlock_list();
716 return ret;
717 }
718
719 /*
720 * For each tracing session, update newly registered apps. The session list
721 * lock MUST be acquired before calling this.
722 */
723 static void update_ust_app(int app_sock)
724 {
725 struct ltt_session *sess, *stmp;
726
727 /* For all tracing session(s) */
728 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
729 session_lock(sess);
730 if (sess->ust_session) {
731 ust_app_global_update(sess->ust_session, app_sock);
732 }
733 session_unlock(sess);
734 }
735 }
736
737 /*
738 * This thread manage event coming from the kernel.
739 *
740 * Features supported in this thread:
741 * -) CPU Hotplug
742 */
743 static void *thread_manage_kernel(void *data)
744 {
745 int ret, i, pollfd, update_poll_flag = 1, err = -1;
746 uint32_t revents, nb_fd;
747 char tmp;
748 struct lttng_poll_event events;
749
750 DBG("[thread] Thread manage kernel started");
751
752 health_register(HEALTH_TYPE_KERNEL);
753
754 /*
755 * This first step of the while is to clean this structure which could free
756 * non NULL pointers so initialize it before the loop.
757 */
758 lttng_poll_init(&events);
759
760 if (testpoint(thread_manage_kernel)) {
761 goto error_testpoint;
762 }
763
764 health_code_update();
765
766 if (testpoint(thread_manage_kernel_before_loop)) {
767 goto error_testpoint;
768 }
769
770 while (1) {
771 health_code_update();
772
773 if (update_poll_flag == 1) {
774 /* Clean events object. We are about to populate it again. */
775 lttng_poll_clean(&events);
776
777 ret = sessiond_set_thread_pollset(&events, 2);
778 if (ret < 0) {
779 goto error_poll_create;
780 }
781
782 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
783 if (ret < 0) {
784 goto error;
785 }
786
787 /* This will add the available kernel channel if any. */
788 ret = update_kernel_poll(&events);
789 if (ret < 0) {
790 goto error;
791 }
792 update_poll_flag = 0;
793 }
794
795 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
796
797 /* Poll infinite value of time */
798 restart:
799 health_poll_entry();
800 ret = lttng_poll_wait(&events, -1);
801 health_poll_exit();
802 if (ret < 0) {
803 /*
804 * Restart interrupted system call.
805 */
806 if (errno == EINTR) {
807 goto restart;
808 }
809 goto error;
810 } else if (ret == 0) {
811 /* Should not happen since timeout is infinite */
812 ERR("Return value of poll is 0 with an infinite timeout.\n"
813 "This should not have happened! Continuing...");
814 continue;
815 }
816
817 nb_fd = ret;
818
819 for (i = 0; i < nb_fd; i++) {
820 /* Fetch once the poll data */
821 revents = LTTNG_POLL_GETEV(&events, i);
822 pollfd = LTTNG_POLL_GETFD(&events, i);
823
824 health_code_update();
825
826 /* Thread quit pipe has been closed. Killing thread. */
827 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
828 if (ret) {
829 err = 0;
830 goto exit;
831 }
832
833 /* Check for data on kernel pipe */
834 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
835 do {
836 ret = read(kernel_poll_pipe[0], &tmp, 1);
837 } while (ret < 0 && errno == EINTR);
838 /*
839 * Ret value is useless here, if this pipe gets any actions an
840 * update is required anyway.
841 */
842 update_poll_flag = 1;
843 continue;
844 } else {
845 /*
846 * New CPU detected by the kernel. Adding kernel stream to
847 * kernel session and updating the kernel consumer
848 */
849 if (revents & LPOLLIN) {
850 ret = update_kernel_stream(&kconsumer_data, pollfd);
851 if (ret < 0) {
852 continue;
853 }
854 break;
855 /*
856 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
857 * and unregister kernel stream at this point.
858 */
859 }
860 }
861 }
862 }
863
864 exit:
865 error:
866 lttng_poll_clean(&events);
867 error_poll_create:
868 error_testpoint:
869 utils_close_pipe(kernel_poll_pipe);
870 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
871 if (err) {
872 health_error();
873 ERR("Health error occurred in %s", __func__);
874 WARN("Kernel thread died unexpectedly. "
875 "Kernel tracing can continue but CPU hotplug is disabled.");
876 }
877 health_unregister();
878 DBG("Kernel thread dying");
879 return NULL;
880 }
881
882 /*
883 * Signal pthread condition of the consumer data that the thread.
884 */
885 static void signal_consumer_condition(struct consumer_data *data, int state)
886 {
887 pthread_mutex_lock(&data->cond_mutex);
888
889 /*
890 * The state is set before signaling. It can be any value, it's the waiter
891 * job to correctly interpret this condition variable associated to the
892 * consumer pthread_cond.
893 *
894 * A value of 0 means that the corresponding thread of the consumer data
895 * was not started. 1 indicates that the thread has started and is ready
896 * for action. A negative value means that there was an error during the
897 * thread bootstrap.
898 */
899 data->consumer_thread_is_ready = state;
900 (void) pthread_cond_signal(&data->cond);
901
902 pthread_mutex_unlock(&data->cond_mutex);
903 }
904
905 /*
906 * This thread manage the consumer error sent back to the session daemon.
907 */
908 static void *thread_manage_consumer(void *data)
909 {
910 int sock = -1, i, ret, pollfd, err = -1;
911 uint32_t revents, nb_fd;
912 enum lttcomm_return_code code;
913 struct lttng_poll_event events;
914 struct consumer_data *consumer_data = data;
915
916 DBG("[thread] Manage consumer started");
917
918 health_register(HEALTH_TYPE_CONSUMER);
919
920 health_code_update();
921
922 /*
923 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
924 * metadata_sock. Nothing more will be added to this poll set.
925 */
926 ret = sessiond_set_thread_pollset(&events, 3);
927 if (ret < 0) {
928 goto error_poll;
929 }
930
931 /*
932 * The error socket here is already in a listening state which was done
933 * just before spawning this thread to avoid a race between the consumer
934 * daemon exec trying to connect and the listen() call.
935 */
936 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
937 if (ret < 0) {
938 goto error;
939 }
940
941 health_code_update();
942
943 /* Infinite blocking call, waiting for transmission */
944 restart:
945 health_poll_entry();
946
947 if (testpoint(thread_manage_consumer)) {
948 goto error;
949 }
950
951 ret = lttng_poll_wait(&events, -1);
952 health_poll_exit();
953 if (ret < 0) {
954 /*
955 * Restart interrupted system call.
956 */
957 if (errno == EINTR) {
958 goto restart;
959 }
960 goto error;
961 }
962
963 nb_fd = ret;
964
965 for (i = 0; i < nb_fd; i++) {
966 /* Fetch once the poll data */
967 revents = LTTNG_POLL_GETEV(&events, i);
968 pollfd = LTTNG_POLL_GETFD(&events, i);
969
970 health_code_update();
971
972 /* Thread quit pipe has been closed. Killing thread. */
973 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
974 if (ret) {
975 err = 0;
976 goto exit;
977 }
978
979 /* Event on the registration socket */
980 if (pollfd == consumer_data->err_sock) {
981 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
982 ERR("consumer err socket poll error");
983 goto error;
984 }
985 }
986 }
987
988 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
989 if (sock < 0) {
990 goto error;
991 }
992
993 /*
994 * Set the CLOEXEC flag. Return code is useless because either way, the
995 * show must go on.
996 */
997 (void) utils_set_fd_cloexec(sock);
998
999 health_code_update();
1000
1001 DBG2("Receiving code from consumer err_sock");
1002
1003 /* Getting status code from kconsumerd */
1004 ret = lttcomm_recv_unix_sock(sock, &code,
1005 sizeof(enum lttcomm_return_code));
1006 if (ret <= 0) {
1007 goto error;
1008 }
1009
1010 health_code_update();
1011
1012 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1013 /* Connect both socket, command and metadata. */
1014 consumer_data->cmd_sock =
1015 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1016 consumer_data->metadata_sock.fd =
1017 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1018 if (consumer_data->cmd_sock < 0 ||
1019 consumer_data->metadata_sock.fd < 0) {
1020 PERROR("consumer connect cmd socket");
1021 /* On error, signal condition and quit. */
1022 signal_consumer_condition(consumer_data, -1);
1023 goto error;
1024 }
1025 /* Create metadata socket lock. */
1026 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1027 if (consumer_data->metadata_sock.lock == NULL) {
1028 PERROR("zmalloc pthread mutex");
1029 ret = -1;
1030 goto error;
1031 }
1032 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1033
1034 signal_consumer_condition(consumer_data, 1);
1035 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1036 DBG("Consumer metadata socket ready (fd: %d)",
1037 consumer_data->metadata_sock.fd);
1038 } else {
1039 ERR("consumer error when waiting for SOCK_READY : %s",
1040 lttcomm_get_readable_code(-code));
1041 goto error;
1042 }
1043
1044 /* Remove the consumerd error sock since we've established a connexion */
1045 ret = lttng_poll_del(&events, consumer_data->err_sock);
1046 if (ret < 0) {
1047 goto error;
1048 }
1049
1050 /* Add new accepted error socket. */
1051 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1052 if (ret < 0) {
1053 goto error;
1054 }
1055
1056 /* Add metadata socket that is successfully connected. */
1057 ret = lttng_poll_add(&events, consumer_data->metadata_sock.fd,
1058 LPOLLIN | LPOLLRDHUP);
1059 if (ret < 0) {
1060 goto error;
1061 }
1062
1063 health_code_update();
1064
1065 /* Infinite blocking call, waiting for transmission */
1066 restart_poll:
1067 while (1) {
1068 health_poll_entry();
1069 ret = lttng_poll_wait(&events, -1);
1070 health_poll_exit();
1071 if (ret < 0) {
1072 /*
1073 * Restart interrupted system call.
1074 */
1075 if (errno == EINTR) {
1076 goto restart_poll;
1077 }
1078 goto error;
1079 }
1080
1081 nb_fd = ret;
1082
1083 for (i = 0; i < nb_fd; i++) {
1084 /* Fetch once the poll data */
1085 revents = LTTNG_POLL_GETEV(&events, i);
1086 pollfd = LTTNG_POLL_GETFD(&events, i);
1087
1088 health_code_update();
1089
1090 /* Thread quit pipe has been closed. Killing thread. */
1091 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1092 if (ret) {
1093 err = 0;
1094 goto exit;
1095 }
1096
1097 if (pollfd == sock) {
1098 /* Event on the consumerd socket */
1099 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1100 ERR("consumer err socket second poll error");
1101 goto error;
1102 }
1103 health_code_update();
1104 /* Wait for any kconsumerd error */
1105 ret = lttcomm_recv_unix_sock(sock, &code,
1106 sizeof(enum lttcomm_return_code));
1107 if (ret <= 0) {
1108 ERR("consumer closed the command socket");
1109 goto error;
1110 }
1111
1112 ERR("consumer return code : %s",
1113 lttcomm_get_readable_code(-code));
1114
1115 goto exit;
1116 } else if (pollfd == consumer_data->metadata_sock.fd) {
1117 /* UST metadata requests */
1118 ret = ust_consumer_metadata_request(
1119 &consumer_data->metadata_sock);
1120 if (ret < 0) {
1121 ERR("Handling metadata request");
1122 goto error;
1123 }
1124 break;
1125 } else {
1126 ERR("Unknown pollfd");
1127 goto error;
1128 }
1129 }
1130 health_code_update();
1131 }
1132
1133 exit:
1134 error:
1135 /* Immediately set the consumerd state to stopped */
1136 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1137 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1138 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1139 consumer_data->type == LTTNG_CONSUMER32_UST) {
1140 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1141 } else {
1142 /* Code flow error... */
1143 assert(0);
1144 }
1145
1146 if (consumer_data->err_sock >= 0) {
1147 ret = close(consumer_data->err_sock);
1148 if (ret) {
1149 PERROR("close");
1150 }
1151 consumer_data->err_sock = -1;
1152 }
1153 if (consumer_data->cmd_sock >= 0) {
1154 ret = close(consumer_data->cmd_sock);
1155 if (ret) {
1156 PERROR("close");
1157 }
1158 consumer_data->cmd_sock = -1;
1159 }
1160 if (consumer_data->metadata_sock.fd >= 0) {
1161 ret = close(consumer_data->metadata_sock.fd);
1162 if (ret) {
1163 PERROR("close");
1164 }
1165 }
1166 /* Cleanup metadata socket mutex. */
1167 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1168 free(consumer_data->metadata_sock.lock);
1169
1170 if (sock >= 0) {
1171 ret = close(sock);
1172 if (ret) {
1173 PERROR("close");
1174 }
1175 }
1176
1177 unlink(consumer_data->err_unix_sock_path);
1178 unlink(consumer_data->cmd_unix_sock_path);
1179 consumer_data->pid = 0;
1180
1181 lttng_poll_clean(&events);
1182 error_poll:
1183 if (err) {
1184 health_error();
1185 ERR("Health error occurred in %s", __func__);
1186 }
1187 health_unregister();
1188 DBG("consumer thread cleanup completed");
1189
1190 return NULL;
1191 }
1192
1193 /*
1194 * This thread manage application communication.
1195 */
1196 static void *thread_manage_apps(void *data)
1197 {
1198 int i, ret, pollfd, err = -1;
1199 uint32_t revents, nb_fd;
1200 struct lttng_poll_event events;
1201
1202 DBG("[thread] Manage application started");
1203
1204 rcu_register_thread();
1205 rcu_thread_online();
1206
1207 health_register(HEALTH_TYPE_APP_MANAGE);
1208
1209 if (testpoint(thread_manage_apps)) {
1210 goto error_testpoint;
1211 }
1212
1213 health_code_update();
1214
1215 ret = sessiond_set_thread_pollset(&events, 2);
1216 if (ret < 0) {
1217 goto error_poll_create;
1218 }
1219
1220 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1221 if (ret < 0) {
1222 goto error;
1223 }
1224
1225 if (testpoint(thread_manage_apps_before_loop)) {
1226 goto error;
1227 }
1228
1229 health_code_update();
1230
1231 while (1) {
1232 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1233
1234 /* Inifinite blocking call, waiting for transmission */
1235 restart:
1236 health_poll_entry();
1237 ret = lttng_poll_wait(&events, -1);
1238 health_poll_exit();
1239 if (ret < 0) {
1240 /*
1241 * Restart interrupted system call.
1242 */
1243 if (errno == EINTR) {
1244 goto restart;
1245 }
1246 goto error;
1247 }
1248
1249 nb_fd = ret;
1250
1251 for (i = 0; i < nb_fd; i++) {
1252 /* Fetch once the poll data */
1253 revents = LTTNG_POLL_GETEV(&events, i);
1254 pollfd = LTTNG_POLL_GETFD(&events, i);
1255
1256 health_code_update();
1257
1258 /* Thread quit pipe has been closed. Killing thread. */
1259 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1260 if (ret) {
1261 err = 0;
1262 goto exit;
1263 }
1264
1265 /* Inspect the apps cmd pipe */
1266 if (pollfd == apps_cmd_pipe[0]) {
1267 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1268 ERR("Apps command pipe error");
1269 goto error;
1270 } else if (revents & LPOLLIN) {
1271 int sock;
1272
1273 /* Empty pipe */
1274 do {
1275 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1276 } while (ret < 0 && errno == EINTR);
1277 if (ret < 0 || ret < sizeof(sock)) {
1278 PERROR("read apps cmd pipe");
1279 goto error;
1280 }
1281
1282 health_code_update();
1283
1284 /*
1285 * We only monitor the error events of the socket. This
1286 * thread does not handle any incoming data from UST
1287 * (POLLIN).
1288 */
1289 ret = lttng_poll_add(&events, sock,
1290 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1291 if (ret < 0) {
1292 goto error;
1293 }
1294
1295 /* Set socket timeout for both receiving and ending */
1296 (void) lttcomm_setsockopt_rcv_timeout(sock,
1297 app_socket_timeout);
1298 (void) lttcomm_setsockopt_snd_timeout(sock,
1299 app_socket_timeout);
1300
1301 DBG("Apps with sock %d added to poll set", sock);
1302
1303 health_code_update();
1304
1305 break;
1306 }
1307 } else {
1308 /*
1309 * At this point, we know that a registered application made
1310 * the event at poll_wait.
1311 */
1312 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1313 /* Removing from the poll set */
1314 ret = lttng_poll_del(&events, pollfd);
1315 if (ret < 0) {
1316 goto error;
1317 }
1318
1319 /* Socket closed on remote end. */
1320 ust_app_unregister(pollfd);
1321 break;
1322 }
1323 }
1324
1325 health_code_update();
1326 }
1327 }
1328
1329 exit:
1330 error:
1331 lttng_poll_clean(&events);
1332 error_poll_create:
1333 error_testpoint:
1334 utils_close_pipe(apps_cmd_pipe);
1335 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1336
1337 /*
1338 * We don't clean the UST app hash table here since already registered
1339 * applications can still be controlled so let them be until the session
1340 * daemon dies or the applications stop.
1341 */
1342
1343 if (err) {
1344 health_error();
1345 ERR("Health error occurred in %s", __func__);
1346 }
1347 health_unregister();
1348 DBG("Application communication apps thread cleanup complete");
1349 rcu_thread_offline();
1350 rcu_unregister_thread();
1351 return NULL;
1352 }
1353
1354 /*
1355 * Send a socket to a thread This is called from the dispatch UST registration
1356 * thread once all sockets are set for the application.
1357 *
1358 * On success, return 0 else a negative value being the errno message of the
1359 * write().
1360 */
1361 static int send_socket_to_thread(int fd, int sock)
1362 {
1363 int ret;
1364
1365 /* Sockets MUST be set or else this should not have been called. */
1366 assert(fd >= 0);
1367 assert(sock >= 0);
1368
1369 do {
1370 ret = write(fd, &sock, sizeof(sock));
1371 } while (ret < 0 && errno == EINTR);
1372 if (ret < 0 || ret != sizeof(sock)) {
1373 PERROR("write apps pipe %d", fd);
1374 if (ret < 0) {
1375 ret = -errno;
1376 }
1377 goto error;
1378 }
1379
1380 /* All good. Don't send back the write positive ret value. */
1381 ret = 0;
1382 error:
1383 return ret;
1384 }
1385
1386 /*
1387 * Sanitize the wait queue of the dispatch registration thread meaning removing
1388 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1389 * notify socket is never received.
1390 */
1391 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1392 {
1393 int ret, nb_fd = 0, i;
1394 unsigned int fd_added = 0;
1395 struct lttng_poll_event events;
1396 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1397
1398 assert(wait_queue);
1399
1400 lttng_poll_init(&events);
1401
1402 /* Just skip everything for an empty queue. */
1403 if (!wait_queue->count) {
1404 goto end;
1405 }
1406
1407 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1408 if (ret < 0) {
1409 goto error_create;
1410 }
1411
1412 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1413 &wait_queue->head, head) {
1414 assert(wait_node->app);
1415 ret = lttng_poll_add(&events, wait_node->app->sock,
1416 LPOLLHUP | LPOLLERR);
1417 if (ret < 0) {
1418 goto error;
1419 }
1420
1421 fd_added = 1;
1422 }
1423
1424 if (!fd_added) {
1425 goto end;
1426 }
1427
1428 /*
1429 * Poll but don't block so we can quickly identify the faulty events and
1430 * clean them afterwards from the wait queue.
1431 */
1432 ret = lttng_poll_wait(&events, 0);
1433 if (ret < 0) {
1434 goto error;
1435 }
1436 nb_fd = ret;
1437
1438 for (i = 0; i < nb_fd; i++) {
1439 /* Get faulty FD. */
1440 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1441 int pollfd = LTTNG_POLL_GETFD(&events, i);
1442
1443 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1444 &wait_queue->head, head) {
1445 if (pollfd == wait_node->app->sock &&
1446 (revents & (LPOLLHUP | LPOLLERR))) {
1447 cds_list_del(&wait_node->head);
1448 wait_queue->count--;
1449 ust_app_destroy(wait_node->app);
1450 free(wait_node);
1451 break;
1452 }
1453 }
1454 }
1455
1456 if (nb_fd > 0) {
1457 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1458 }
1459
1460 end:
1461 lttng_poll_clean(&events);
1462 return;
1463
1464 error:
1465 lttng_poll_clean(&events);
1466 error_create:
1467 ERR("Unable to sanitize wait queue");
1468 return;
1469 }
1470
1471 /*
1472 * Dispatch request from the registration threads to the application
1473 * communication thread.
1474 */
1475 static void *thread_dispatch_ust_registration(void *data)
1476 {
1477 int ret, err = -1;
1478 struct cds_wfq_node *node;
1479 struct ust_command *ust_cmd = NULL;
1480 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1481 struct ust_reg_wait_queue wait_queue = {
1482 .count = 0,
1483 };
1484
1485 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1486
1487 health_code_update();
1488
1489 CDS_INIT_LIST_HEAD(&wait_queue.head);
1490
1491 DBG("[thread] Dispatch UST command started");
1492
1493 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1494 health_code_update();
1495
1496 /* Atomically prepare the queue futex */
1497 futex_nto1_prepare(&ust_cmd_queue.futex);
1498
1499 do {
1500 struct ust_app *app = NULL;
1501 ust_cmd = NULL;
1502
1503 /*
1504 * Make sure we don't have node(s) that have hung up before receiving
1505 * the notify socket. This is to clean the list in order to avoid
1506 * memory leaks from notify socket that are never seen.
1507 */
1508 sanitize_wait_queue(&wait_queue);
1509
1510 health_code_update();
1511 /* Dequeue command for registration */
1512 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1513 if (node == NULL) {
1514 DBG("Woken up but nothing in the UST command queue");
1515 /* Continue thread execution */
1516 break;
1517 }
1518
1519 ust_cmd = caa_container_of(node, struct ust_command, node);
1520
1521 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1522 " gid:%d sock:%d name:%s (version %d.%d)",
1523 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1524 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1525 ust_cmd->sock, ust_cmd->reg_msg.name,
1526 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1527
1528 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1529 wait_node = zmalloc(sizeof(*wait_node));
1530 if (!wait_node) {
1531 PERROR("zmalloc wait_node dispatch");
1532 ret = close(ust_cmd->sock);
1533 if (ret < 0) {
1534 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1535 }
1536 lttng_fd_put(1, LTTNG_FD_APPS);
1537 free(ust_cmd);
1538 goto error;
1539 }
1540 CDS_INIT_LIST_HEAD(&wait_node->head);
1541
1542 /* Create application object if socket is CMD. */
1543 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1544 ust_cmd->sock);
1545 if (!wait_node->app) {
1546 ret = close(ust_cmd->sock);
1547 if (ret < 0) {
1548 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1549 }
1550 lttng_fd_put(1, LTTNG_FD_APPS);
1551 free(wait_node);
1552 free(ust_cmd);
1553 continue;
1554 }
1555 /*
1556 * Add application to the wait queue so we can set the notify
1557 * socket before putting this object in the global ht.
1558 */
1559 cds_list_add(&wait_node->head, &wait_queue.head);
1560 wait_queue.count++;
1561
1562 free(ust_cmd);
1563 /*
1564 * We have to continue here since we don't have the notify
1565 * socket and the application MUST be added to the hash table
1566 * only at that moment.
1567 */
1568 continue;
1569 } else {
1570 /*
1571 * Look for the application in the local wait queue and set the
1572 * notify socket if found.
1573 */
1574 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1575 &wait_queue.head, head) {
1576 health_code_update();
1577 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1578 wait_node->app->notify_sock = ust_cmd->sock;
1579 cds_list_del(&wait_node->head);
1580 wait_queue.count--;
1581 app = wait_node->app;
1582 free(wait_node);
1583 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1584 break;
1585 }
1586 }
1587
1588 /*
1589 * With no application at this stage the received socket is
1590 * basically useless so close it before we free the cmd data
1591 * structure for good.
1592 */
1593 if (!app) {
1594 ret = close(ust_cmd->sock);
1595 if (ret < 0) {
1596 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1597 }
1598 lttng_fd_put(1, LTTNG_FD_APPS);
1599 }
1600 free(ust_cmd);
1601 }
1602
1603 if (app) {
1604 /*
1605 * @session_lock_list
1606 *
1607 * Lock the global session list so from the register up to the
1608 * registration done message, no thread can see the application
1609 * and change its state.
1610 */
1611 session_lock_list();
1612 rcu_read_lock();
1613
1614 /*
1615 * Add application to the global hash table. This needs to be
1616 * done before the update to the UST registry can locate the
1617 * application.
1618 */
1619 ust_app_add(app);
1620
1621 /* Set app version. This call will print an error if needed. */
1622 (void) ust_app_version(app);
1623
1624 /* Send notify socket through the notify pipe. */
1625 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1626 app->notify_sock);
1627 if (ret < 0) {
1628 rcu_read_unlock();
1629 session_unlock_list();
1630 /* No notify thread, stop the UST tracing. */
1631 goto error;
1632 }
1633
1634 /*
1635 * Update newly registered application with the tracing
1636 * registry info already enabled information.
1637 */
1638 update_ust_app(app->sock);
1639
1640 /*
1641 * Don't care about return value. Let the manage apps threads
1642 * handle app unregistration upon socket close.
1643 */
1644 (void) ust_app_register_done(app->sock);
1645
1646 /*
1647 * Even if the application socket has been closed, send the app
1648 * to the thread and unregistration will take place at that
1649 * place.
1650 */
1651 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1652 if (ret < 0) {
1653 rcu_read_unlock();
1654 session_unlock_list();
1655 /* No apps. thread, stop the UST tracing. */
1656 goto error;
1657 }
1658
1659 rcu_read_unlock();
1660 session_unlock_list();
1661 }
1662 } while (node != NULL);
1663
1664 health_poll_entry();
1665 /* Futex wait on queue. Blocking call on futex() */
1666 futex_nto1_wait(&ust_cmd_queue.futex);
1667 health_poll_exit();
1668 }
1669 /* Normal exit, no error */
1670 err = 0;
1671
1672 error:
1673 /* Clean up wait queue. */
1674 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1675 &wait_queue.head, head) {
1676 cds_list_del(&wait_node->head);
1677 wait_queue.count--;
1678 free(wait_node);
1679 }
1680
1681 DBG("Dispatch thread dying");
1682 if (err) {
1683 health_error();
1684 ERR("Health error occurred in %s", __func__);
1685 }
1686 health_unregister();
1687 return NULL;
1688 }
1689
1690 /*
1691 * This thread manage application registration.
1692 */
1693 static void *thread_registration_apps(void *data)
1694 {
1695 int sock = -1, i, ret, pollfd, err = -1;
1696 uint32_t revents, nb_fd;
1697 struct lttng_poll_event events;
1698 /*
1699 * Get allocated in this thread, enqueued to a global queue, dequeued and
1700 * freed in the manage apps thread.
1701 */
1702 struct ust_command *ust_cmd = NULL;
1703
1704 DBG("[thread] Manage application registration started");
1705
1706 health_register(HEALTH_TYPE_APP_REG);
1707
1708 if (testpoint(thread_registration_apps)) {
1709 goto error_testpoint;
1710 }
1711
1712 ret = lttcomm_listen_unix_sock(apps_sock);
1713 if (ret < 0) {
1714 goto error_listen;
1715 }
1716
1717 /*
1718 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1719 * more will be added to this poll set.
1720 */
1721 ret = sessiond_set_thread_pollset(&events, 2);
1722 if (ret < 0) {
1723 goto error_create_poll;
1724 }
1725
1726 /* Add the application registration socket */
1727 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1728 if (ret < 0) {
1729 goto error_poll_add;
1730 }
1731
1732 /* Notify all applications to register */
1733 ret = notify_ust_apps(1);
1734 if (ret < 0) {
1735 ERR("Failed to notify applications or create the wait shared memory.\n"
1736 "Execution continues but there might be problem for already\n"
1737 "running applications that wishes to register.");
1738 }
1739
1740 while (1) {
1741 DBG("Accepting application registration");
1742
1743 /* Inifinite blocking call, waiting for transmission */
1744 restart:
1745 health_poll_entry();
1746 ret = lttng_poll_wait(&events, -1);
1747 health_poll_exit();
1748 if (ret < 0) {
1749 /*
1750 * Restart interrupted system call.
1751 */
1752 if (errno == EINTR) {
1753 goto restart;
1754 }
1755 goto error;
1756 }
1757
1758 nb_fd = ret;
1759
1760 for (i = 0; i < nb_fd; i++) {
1761 health_code_update();
1762
1763 /* Fetch once the poll data */
1764 revents = LTTNG_POLL_GETEV(&events, i);
1765 pollfd = LTTNG_POLL_GETFD(&events, i);
1766
1767 /* Thread quit pipe has been closed. Killing thread. */
1768 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1769 if (ret) {
1770 err = 0;
1771 goto exit;
1772 }
1773
1774 /* Event on the registration socket */
1775 if (pollfd == apps_sock) {
1776 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1777 ERR("Register apps socket poll error");
1778 goto error;
1779 } else if (revents & LPOLLIN) {
1780 sock = lttcomm_accept_unix_sock(apps_sock);
1781 if (sock < 0) {
1782 goto error;
1783 }
1784
1785 /*
1786 * Set the CLOEXEC flag. Return code is useless because
1787 * either way, the show must go on.
1788 */
1789 (void) utils_set_fd_cloexec(sock);
1790
1791 /* Create UST registration command for enqueuing */
1792 ust_cmd = zmalloc(sizeof(struct ust_command));
1793 if (ust_cmd == NULL) {
1794 PERROR("ust command zmalloc");
1795 goto error;
1796 }
1797
1798 /*
1799 * Using message-based transmissions to ensure we don't
1800 * have to deal with partially received messages.
1801 */
1802 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1803 if (ret < 0) {
1804 ERR("Exhausted file descriptors allowed for applications.");
1805 free(ust_cmd);
1806 ret = close(sock);
1807 if (ret) {
1808 PERROR("close");
1809 }
1810 sock = -1;
1811 continue;
1812 }
1813
1814 health_code_update();
1815 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1816 if (ret < 0) {
1817 free(ust_cmd);
1818 /* Close socket of the application. */
1819 ret = close(sock);
1820 if (ret) {
1821 PERROR("close");
1822 }
1823 lttng_fd_put(LTTNG_FD_APPS, 1);
1824 sock = -1;
1825 continue;
1826 }
1827 health_code_update();
1828
1829 ust_cmd->sock = sock;
1830 sock = -1;
1831
1832 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1833 " gid:%d sock:%d name:%s (version %d.%d)",
1834 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1835 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1836 ust_cmd->sock, ust_cmd->reg_msg.name,
1837 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1838
1839 /*
1840 * Lock free enqueue the registration request. The red pill
1841 * has been taken! This apps will be part of the *system*.
1842 */
1843 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1844
1845 /*
1846 * Wake the registration queue futex. Implicit memory
1847 * barrier with the exchange in cds_wfq_enqueue.
1848 */
1849 futex_nto1_wake(&ust_cmd_queue.futex);
1850 }
1851 }
1852 }
1853 }
1854
1855 exit:
1856 error:
1857 if (err) {
1858 health_error();
1859 ERR("Health error occurred in %s", __func__);
1860 }
1861
1862 /* Notify that the registration thread is gone */
1863 notify_ust_apps(0);
1864
1865 if (apps_sock >= 0) {
1866 ret = close(apps_sock);
1867 if (ret) {
1868 PERROR("close");
1869 }
1870 }
1871 if (sock >= 0) {
1872 ret = close(sock);
1873 if (ret) {
1874 PERROR("close");
1875 }
1876 lttng_fd_put(LTTNG_FD_APPS, 1);
1877 }
1878 unlink(apps_unix_sock_path);
1879
1880 error_poll_add:
1881 lttng_poll_clean(&events);
1882 error_listen:
1883 error_create_poll:
1884 error_testpoint:
1885 DBG("UST Registration thread cleanup complete");
1886 health_unregister();
1887
1888 return NULL;
1889 }
1890
1891 /*
1892 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1893 * exec or it will fails.
1894 */
1895 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1896 {
1897 int ret, clock_ret;
1898 struct timespec timeout;
1899
1900 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1901 consumer_data->consumer_thread_is_ready = 0;
1902
1903 /* Setup pthread condition */
1904 ret = pthread_condattr_init(&consumer_data->condattr);
1905 if (ret != 0) {
1906 errno = ret;
1907 PERROR("pthread_condattr_init consumer data");
1908 goto error;
1909 }
1910
1911 /*
1912 * Set the monotonic clock in order to make sure we DO NOT jump in time
1913 * between the clock_gettime() call and the timedwait call. See bug #324
1914 * for a more details and how we noticed it.
1915 */
1916 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1917 if (ret != 0) {
1918 errno = ret;
1919 PERROR("pthread_condattr_setclock consumer data");
1920 goto error;
1921 }
1922
1923 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1924 if (ret != 0) {
1925 errno = ret;
1926 PERROR("pthread_cond_init consumer data");
1927 goto error;
1928 }
1929
1930 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1931 consumer_data);
1932 if (ret != 0) {
1933 PERROR("pthread_create consumer");
1934 ret = -1;
1935 goto error;
1936 }
1937
1938 /* We are about to wait on a pthread condition */
1939 pthread_mutex_lock(&consumer_data->cond_mutex);
1940
1941 /* Get time for sem_timedwait absolute timeout */
1942 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1943 /*
1944 * Set the timeout for the condition timed wait even if the clock gettime
1945 * call fails since we might loop on that call and we want to avoid to
1946 * increment the timeout too many times.
1947 */
1948 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1949
1950 /*
1951 * The following loop COULD be skipped in some conditions so this is why we
1952 * set ret to 0 in order to make sure at least one round of the loop is
1953 * done.
1954 */
1955 ret = 0;
1956
1957 /*
1958 * Loop until the condition is reached or when a timeout is reached. Note
1959 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1960 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1961 * possible. This loop does not take any chances and works with both of
1962 * them.
1963 */
1964 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1965 if (clock_ret < 0) {
1966 PERROR("clock_gettime spawn consumer");
1967 /* Infinite wait for the consumerd thread to be ready */
1968 ret = pthread_cond_wait(&consumer_data->cond,
1969 &consumer_data->cond_mutex);
1970 } else {
1971 ret = pthread_cond_timedwait(&consumer_data->cond,
1972 &consumer_data->cond_mutex, &timeout);
1973 }
1974 }
1975
1976 /* Release the pthread condition */
1977 pthread_mutex_unlock(&consumer_data->cond_mutex);
1978
1979 if (ret != 0) {
1980 errno = ret;
1981 if (ret == ETIMEDOUT) {
1982 /*
1983 * Call has timed out so we kill the kconsumerd_thread and return
1984 * an error.
1985 */
1986 ERR("Condition timed out. The consumer thread was never ready."
1987 " Killing it");
1988 ret = pthread_cancel(consumer_data->thread);
1989 if (ret < 0) {
1990 PERROR("pthread_cancel consumer thread");
1991 }
1992 } else {
1993 PERROR("pthread_cond_wait failed consumer thread");
1994 }
1995 goto error;
1996 }
1997
1998 pthread_mutex_lock(&consumer_data->pid_mutex);
1999 if (consumer_data->pid == 0) {
2000 ERR("Consumerd did not start");
2001 pthread_mutex_unlock(&consumer_data->pid_mutex);
2002 goto error;
2003 }
2004 pthread_mutex_unlock(&consumer_data->pid_mutex);
2005
2006 return 0;
2007
2008 error:
2009 return ret;
2010 }
2011
2012 /*
2013 * Join consumer thread
2014 */
2015 static int join_consumer_thread(struct consumer_data *consumer_data)
2016 {
2017 void *status;
2018
2019 /* Consumer pid must be a real one. */
2020 if (consumer_data->pid > 0) {
2021 int ret;
2022 ret = kill(consumer_data->pid, SIGTERM);
2023 if (ret) {
2024 ERR("Error killing consumer daemon");
2025 return ret;
2026 }
2027 return pthread_join(consumer_data->thread, &status);
2028 } else {
2029 return 0;
2030 }
2031 }
2032
2033 /*
2034 * Fork and exec a consumer daemon (consumerd).
2035 *
2036 * Return pid if successful else -1.
2037 */
2038 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2039 {
2040 int ret;
2041 pid_t pid;
2042 const char *consumer_to_use;
2043 const char *verbosity;
2044 struct stat st;
2045
2046 DBG("Spawning consumerd");
2047
2048 pid = fork();
2049 if (pid == 0) {
2050 /*
2051 * Exec consumerd.
2052 */
2053 if (opt_verbose_consumer) {
2054 verbosity = "--verbose";
2055 } else {
2056 verbosity = "--quiet";
2057 }
2058 switch (consumer_data->type) {
2059 case LTTNG_CONSUMER_KERNEL:
2060 /*
2061 * Find out which consumerd to execute. We will first try the
2062 * 64-bit path, then the sessiond's installation directory, and
2063 * fallback on the 32-bit one,
2064 */
2065 DBG3("Looking for a kernel consumer at these locations:");
2066 DBG3(" 1) %s", consumerd64_bin);
2067 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2068 DBG3(" 3) %s", consumerd32_bin);
2069 if (stat(consumerd64_bin, &st) == 0) {
2070 DBG3("Found location #1");
2071 consumer_to_use = consumerd64_bin;
2072 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2073 DBG3("Found location #2");
2074 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2075 } else if (stat(consumerd32_bin, &st) == 0) {
2076 DBG3("Found location #3");
2077 consumer_to_use = consumerd32_bin;
2078 } else {
2079 DBG("Could not find any valid consumerd executable");
2080 break;
2081 }
2082 DBG("Using kernel consumer at: %s", consumer_to_use);
2083 execl(consumer_to_use,
2084 "lttng-consumerd", verbosity, "-k",
2085 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2086 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2087 NULL);
2088 break;
2089 case LTTNG_CONSUMER64_UST:
2090 {
2091 char *tmpnew = NULL;
2092
2093 if (consumerd64_libdir[0] != '\0') {
2094 char *tmp;
2095 size_t tmplen;
2096
2097 tmp = getenv("LD_LIBRARY_PATH");
2098 if (!tmp) {
2099 tmp = "";
2100 }
2101 tmplen = strlen("LD_LIBRARY_PATH=")
2102 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2103 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2104 if (!tmpnew) {
2105 ret = -ENOMEM;
2106 goto error;
2107 }
2108 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2109 strcat(tmpnew, consumerd64_libdir);
2110 if (tmp[0] != '\0') {
2111 strcat(tmpnew, ":");
2112 strcat(tmpnew, tmp);
2113 }
2114 ret = putenv(tmpnew);
2115 if (ret) {
2116 ret = -errno;
2117 free(tmpnew);
2118 goto error;
2119 }
2120 }
2121 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2122 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2123 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2124 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2125 NULL);
2126 if (consumerd64_libdir[0] != '\0') {
2127 free(tmpnew);
2128 }
2129 if (ret) {
2130 goto error;
2131 }
2132 break;
2133 }
2134 case LTTNG_CONSUMER32_UST:
2135 {
2136 char *tmpnew = NULL;
2137
2138 if (consumerd32_libdir[0] != '\0') {
2139 char *tmp;
2140 size_t tmplen;
2141
2142 tmp = getenv("LD_LIBRARY_PATH");
2143 if (!tmp) {
2144 tmp = "";
2145 }
2146 tmplen = strlen("LD_LIBRARY_PATH=")
2147 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2148 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2149 if (!tmpnew) {
2150 ret = -ENOMEM;
2151 goto error;
2152 }
2153 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2154 strcat(tmpnew, consumerd32_libdir);
2155 if (tmp[0] != '\0') {
2156 strcat(tmpnew, ":");
2157 strcat(tmpnew, tmp);
2158 }
2159 ret = putenv(tmpnew);
2160 if (ret) {
2161 ret = -errno;
2162 free(tmpnew);
2163 goto error;
2164 }
2165 }
2166 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2167 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2168 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2169 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2170 NULL);
2171 if (consumerd32_libdir[0] != '\0') {
2172 free(tmpnew);
2173 }
2174 if (ret) {
2175 goto error;
2176 }
2177 break;
2178 }
2179 default:
2180 PERROR("unknown consumer type");
2181 exit(EXIT_FAILURE);
2182 }
2183 if (errno != 0) {
2184 PERROR("kernel start consumer exec");
2185 }
2186 exit(EXIT_FAILURE);
2187 } else if (pid > 0) {
2188 ret = pid;
2189 } else {
2190 PERROR("start consumer fork");
2191 ret = -errno;
2192 }
2193 error:
2194 return ret;
2195 }
2196
2197 /*
2198 * Spawn the consumerd daemon and session daemon thread.
2199 */
2200 static int start_consumerd(struct consumer_data *consumer_data)
2201 {
2202 int ret;
2203
2204 /*
2205 * Set the listen() state on the socket since there is a possible race
2206 * between the exec() of the consumer daemon and this call if place in the
2207 * consumer thread. See bug #366 for more details.
2208 */
2209 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2210 if (ret < 0) {
2211 goto error;
2212 }
2213
2214 pthread_mutex_lock(&consumer_data->pid_mutex);
2215 if (consumer_data->pid != 0) {
2216 pthread_mutex_unlock(&consumer_data->pid_mutex);
2217 goto end;
2218 }
2219
2220 ret = spawn_consumerd(consumer_data);
2221 if (ret < 0) {
2222 ERR("Spawning consumerd failed");
2223 pthread_mutex_unlock(&consumer_data->pid_mutex);
2224 goto error;
2225 }
2226
2227 /* Setting up the consumer_data pid */
2228 consumer_data->pid = ret;
2229 DBG2("Consumer pid %d", consumer_data->pid);
2230 pthread_mutex_unlock(&consumer_data->pid_mutex);
2231
2232 DBG2("Spawning consumer control thread");
2233 ret = spawn_consumer_thread(consumer_data);
2234 if (ret < 0) {
2235 ERR("Fatal error spawning consumer control thread");
2236 goto error;
2237 }
2238
2239 end:
2240 return 0;
2241
2242 error:
2243 /* Cleanup already created sockets on error. */
2244 if (consumer_data->err_sock >= 0) {
2245 int err;
2246
2247 err = close(consumer_data->err_sock);
2248 if (err < 0) {
2249 PERROR("close consumer data error socket");
2250 }
2251 }
2252 return ret;
2253 }
2254
2255 /*
2256 * Compute health status of each consumer. If one of them is zero (bad
2257 * state), we return 0.
2258 */
2259 static int check_consumer_health(void)
2260 {
2261 int ret;
2262
2263 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2264
2265 DBG3("Health consumer check %d", ret);
2266
2267 return ret;
2268 }
2269
2270 /*
2271 * Setup necessary data for kernel tracer action.
2272 */
2273 static int init_kernel_tracer(void)
2274 {
2275 int ret;
2276
2277 /* Modprobe lttng kernel modules */
2278 ret = modprobe_lttng_control();
2279 if (ret < 0) {
2280 goto error;
2281 }
2282
2283 /* Open debugfs lttng */
2284 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2285 if (kernel_tracer_fd < 0) {
2286 DBG("Failed to open %s", module_proc_lttng);
2287 ret = -1;
2288 goto error_open;
2289 }
2290
2291 /* Validate kernel version */
2292 ret = kernel_validate_version(kernel_tracer_fd);
2293 if (ret < 0) {
2294 goto error_version;
2295 }
2296
2297 ret = modprobe_lttng_data();
2298 if (ret < 0) {
2299 goto error_modules;
2300 }
2301
2302 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2303 return 0;
2304
2305 error_version:
2306 modprobe_remove_lttng_control();
2307 ret = close(kernel_tracer_fd);
2308 if (ret) {
2309 PERROR("close");
2310 }
2311 kernel_tracer_fd = -1;
2312 return LTTNG_ERR_KERN_VERSION;
2313
2314 error_modules:
2315 ret = close(kernel_tracer_fd);
2316 if (ret) {
2317 PERROR("close");
2318 }
2319
2320 error_open:
2321 modprobe_remove_lttng_control();
2322
2323 error:
2324 WARN("No kernel tracer available");
2325 kernel_tracer_fd = -1;
2326 if (!is_root) {
2327 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2328 } else {
2329 return LTTNG_ERR_KERN_NA;
2330 }
2331 }
2332
2333
2334 /*
2335 * Copy consumer output from the tracing session to the domain session. The
2336 * function also applies the right modification on a per domain basis for the
2337 * trace files destination directory.
2338 *
2339 * Should *NOT* be called with RCU read-side lock held.
2340 */
2341 static int copy_session_consumer(int domain, struct ltt_session *session)
2342 {
2343 int ret;
2344 const char *dir_name;
2345 struct consumer_output *consumer;
2346
2347 assert(session);
2348 assert(session->consumer);
2349
2350 switch (domain) {
2351 case LTTNG_DOMAIN_KERNEL:
2352 DBG3("Copying tracing session consumer output in kernel session");
2353 /*
2354 * XXX: We should audit the session creation and what this function
2355 * does "extra" in order to avoid a destroy since this function is used
2356 * in the domain session creation (kernel and ust) only. Same for UST
2357 * domain.
2358 */
2359 if (session->kernel_session->consumer) {
2360 consumer_destroy_output(session->kernel_session->consumer);
2361 }
2362 session->kernel_session->consumer =
2363 consumer_copy_output(session->consumer);
2364 /* Ease our life a bit for the next part */
2365 consumer = session->kernel_session->consumer;
2366 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2367 break;
2368 case LTTNG_DOMAIN_UST:
2369 DBG3("Copying tracing session consumer output in UST session");
2370 if (session->ust_session->consumer) {
2371 consumer_destroy_output(session->ust_session->consumer);
2372 }
2373 session->ust_session->consumer =
2374 consumer_copy_output(session->consumer);
2375 /* Ease our life a bit for the next part */
2376 consumer = session->ust_session->consumer;
2377 dir_name = DEFAULT_UST_TRACE_DIR;
2378 break;
2379 default:
2380 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2381 goto error;
2382 }
2383
2384 /* Append correct directory to subdir */
2385 strncat(consumer->subdir, dir_name,
2386 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2387 DBG3("Copy session consumer subdir %s", consumer->subdir);
2388
2389 ret = LTTNG_OK;
2390
2391 error:
2392 return ret;
2393 }
2394
2395 /*
2396 * Create an UST session and add it to the session ust list.
2397 *
2398 * Should *NOT* be called with RCU read-side lock held.
2399 */
2400 static int create_ust_session(struct ltt_session *session,
2401 struct lttng_domain *domain)
2402 {
2403 int ret;
2404 struct ltt_ust_session *lus = NULL;
2405
2406 assert(session);
2407 assert(domain);
2408 assert(session->consumer);
2409
2410 switch (domain->type) {
2411 case LTTNG_DOMAIN_UST:
2412 break;
2413 default:
2414 ERR("Unknown UST domain on create session %d", domain->type);
2415 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2416 goto error;
2417 }
2418
2419 DBG("Creating UST session");
2420
2421 lus = trace_ust_create_session(session->id);
2422 if (lus == NULL) {
2423 ret = LTTNG_ERR_UST_SESS_FAIL;
2424 goto error;
2425 }
2426
2427 lus->uid = session->uid;
2428 lus->gid = session->gid;
2429 session->ust_session = lus;
2430
2431 /* Copy session output to the newly created UST session */
2432 ret = copy_session_consumer(domain->type, session);
2433 if (ret != LTTNG_OK) {
2434 goto error;
2435 }
2436
2437 return LTTNG_OK;
2438
2439 error:
2440 free(lus);
2441 session->ust_session = NULL;
2442 return ret;
2443 }
2444
2445 /*
2446 * Create a kernel tracer session then create the default channel.
2447 */
2448 static int create_kernel_session(struct ltt_session *session)
2449 {
2450 int ret;
2451
2452 DBG("Creating kernel session");
2453
2454 ret = kernel_create_session(session, kernel_tracer_fd);
2455 if (ret < 0) {
2456 ret = LTTNG_ERR_KERN_SESS_FAIL;
2457 goto error;
2458 }
2459
2460 /* Code flow safety */
2461 assert(session->kernel_session);
2462
2463 /* Copy session output to the newly created Kernel session */
2464 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2465 if (ret != LTTNG_OK) {
2466 goto error;
2467 }
2468
2469 /* Create directory(ies) on local filesystem. */
2470 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2471 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2472 ret = run_as_mkdir_recursive(
2473 session->kernel_session->consumer->dst.trace_path,
2474 S_IRWXU | S_IRWXG, session->uid, session->gid);
2475 if (ret < 0) {
2476 if (ret != -EEXIST) {
2477 ERR("Trace directory creation error");
2478 goto error;
2479 }
2480 }
2481 }
2482
2483 session->kernel_session->uid = session->uid;
2484 session->kernel_session->gid = session->gid;
2485
2486 return LTTNG_OK;
2487
2488 error:
2489 trace_kernel_destroy_session(session->kernel_session);
2490 session->kernel_session = NULL;
2491 return ret;
2492 }
2493
2494 /*
2495 * Count number of session permitted by uid/gid.
2496 */
2497 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2498 {
2499 unsigned int i = 0;
2500 struct ltt_session *session;
2501
2502 DBG("Counting number of available session for UID %d GID %d",
2503 uid, gid);
2504 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2505 /*
2506 * Only list the sessions the user can control.
2507 */
2508 if (!session_access_ok(session, uid, gid)) {
2509 continue;
2510 }
2511 i++;
2512 }
2513 return i;
2514 }
2515
2516 /*
2517 * Process the command requested by the lttng client within the command
2518 * context structure. This function make sure that the return structure (llm)
2519 * is set and ready for transmission before returning.
2520 *
2521 * Return any error encountered or 0 for success.
2522 *
2523 * "sock" is only used for special-case var. len data.
2524 *
2525 * Should *NOT* be called with RCU read-side lock held.
2526 */
2527 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2528 int *sock_error)
2529 {
2530 int ret = LTTNG_OK;
2531 int need_tracing_session = 1;
2532 int need_domain;
2533
2534 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2535
2536 *sock_error = 0;
2537
2538 switch (cmd_ctx->lsm->cmd_type) {
2539 case LTTNG_CREATE_SESSION:
2540 case LTTNG_DESTROY_SESSION:
2541 case LTTNG_LIST_SESSIONS:
2542 case LTTNG_LIST_DOMAINS:
2543 case LTTNG_START_TRACE:
2544 case LTTNG_STOP_TRACE:
2545 case LTTNG_DATA_PENDING:
2546 need_domain = 0;
2547 break;
2548 default:
2549 need_domain = 1;
2550 }
2551
2552 if (opt_no_kernel && need_domain
2553 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2554 if (!is_root) {
2555 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2556 } else {
2557 ret = LTTNG_ERR_KERN_NA;
2558 }
2559 goto error;
2560 }
2561
2562 /* Deny register consumer if we already have a spawned consumer. */
2563 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2564 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2565 if (kconsumer_data.pid > 0) {
2566 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2567 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2568 goto error;
2569 }
2570 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2571 }
2572
2573 /*
2574 * Check for command that don't needs to allocate a returned payload. We do
2575 * this here so we don't have to make the call for no payload at each
2576 * command.
2577 */
2578 switch(cmd_ctx->lsm->cmd_type) {
2579 case LTTNG_LIST_SESSIONS:
2580 case LTTNG_LIST_TRACEPOINTS:
2581 case LTTNG_LIST_TRACEPOINT_FIELDS:
2582 case LTTNG_LIST_DOMAINS:
2583 case LTTNG_LIST_CHANNELS:
2584 case LTTNG_LIST_EVENTS:
2585 break;
2586 default:
2587 /* Setup lttng message with no payload */
2588 ret = setup_lttng_msg(cmd_ctx, 0);
2589 if (ret < 0) {
2590 /* This label does not try to unlock the session */
2591 goto init_setup_error;
2592 }
2593 }
2594
2595 /* Commands that DO NOT need a session. */
2596 switch (cmd_ctx->lsm->cmd_type) {
2597 case LTTNG_CREATE_SESSION:
2598 case LTTNG_CALIBRATE:
2599 case LTTNG_LIST_SESSIONS:
2600 case LTTNG_LIST_TRACEPOINTS:
2601 case LTTNG_LIST_TRACEPOINT_FIELDS:
2602 need_tracing_session = 0;
2603 break;
2604 default:
2605 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2606 /*
2607 * We keep the session list lock across _all_ commands
2608 * for now, because the per-session lock does not
2609 * handle teardown properly.
2610 */
2611 session_lock_list();
2612 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2613 if (cmd_ctx->session == NULL) {
2614 ret = LTTNG_ERR_SESS_NOT_FOUND;
2615 goto error;
2616 } else {
2617 /* Acquire lock for the session */
2618 session_lock(cmd_ctx->session);
2619 }
2620 break;
2621 }
2622
2623 if (!need_domain) {
2624 goto skip_domain;
2625 }
2626
2627 /*
2628 * Check domain type for specific "pre-action".
2629 */
2630 switch (cmd_ctx->lsm->domain.type) {
2631 case LTTNG_DOMAIN_KERNEL:
2632 if (!is_root) {
2633 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2634 goto error;
2635 }
2636
2637 /* Kernel tracer check */
2638 if (kernel_tracer_fd == -1) {
2639 /* Basically, load kernel tracer modules */
2640 ret = init_kernel_tracer();
2641 if (ret != 0) {
2642 goto error;
2643 }
2644 }
2645
2646 /* Consumer is in an ERROR state. Report back to client */
2647 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2648 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2649 goto error;
2650 }
2651
2652 /* Need a session for kernel command */
2653 if (need_tracing_session) {
2654 if (cmd_ctx->session->kernel_session == NULL) {
2655 ret = create_kernel_session(cmd_ctx->session);
2656 if (ret < 0) {
2657 ret = LTTNG_ERR_KERN_SESS_FAIL;
2658 goto error;
2659 }
2660 }
2661
2662 /* Start the kernel consumer daemon */
2663 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2664 if (kconsumer_data.pid == 0 &&
2665 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2666 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2667 ret = start_consumerd(&kconsumer_data);
2668 if (ret < 0) {
2669 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2670 goto error;
2671 }
2672 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2673 } else {
2674 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2675 }
2676
2677 /*
2678 * The consumer was just spawned so we need to add the socket to
2679 * the consumer output of the session if exist.
2680 */
2681 ret = consumer_create_socket(&kconsumer_data,
2682 cmd_ctx->session->kernel_session->consumer);
2683 if (ret < 0) {
2684 goto error;
2685 }
2686 }
2687
2688 break;
2689 case LTTNG_DOMAIN_UST:
2690 {
2691 /* Consumer is in an ERROR state. Report back to client */
2692 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2693 ret = LTTNG_ERR_NO_USTCONSUMERD;
2694 goto error;
2695 }
2696
2697 if (need_tracing_session) {
2698 /* Create UST session if none exist. */
2699 if (cmd_ctx->session->ust_session == NULL) {
2700 ret = create_ust_session(cmd_ctx->session,
2701 &cmd_ctx->lsm->domain);
2702 if (ret != LTTNG_OK) {
2703 goto error;
2704 }
2705 }
2706
2707 /* Start the UST consumer daemons */
2708 /* 64-bit */
2709 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2710 if (consumerd64_bin[0] != '\0' &&
2711 ustconsumer64_data.pid == 0 &&
2712 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2713 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2714 ret = start_consumerd(&ustconsumer64_data);
2715 if (ret < 0) {
2716 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2717 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2718 goto error;
2719 }
2720
2721 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2722 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2723 } else {
2724 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2725 }
2726
2727 /*
2728 * Setup socket for consumer 64 bit. No need for atomic access
2729 * since it was set above and can ONLY be set in this thread.
2730 */
2731 ret = consumer_create_socket(&ustconsumer64_data,
2732 cmd_ctx->session->ust_session->consumer);
2733 if (ret < 0) {
2734 goto error;
2735 }
2736
2737 /* 32-bit */
2738 if (consumerd32_bin[0] != '\0' &&
2739 ustconsumer32_data.pid == 0 &&
2740 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2741 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2742 ret = start_consumerd(&ustconsumer32_data);
2743 if (ret < 0) {
2744 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2745 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2746 goto error;
2747 }
2748
2749 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2750 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2751 } else {
2752 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2753 }
2754
2755 /*
2756 * Setup socket for consumer 64 bit. No need for atomic access
2757 * since it was set above and can ONLY be set in this thread.
2758 */
2759 ret = consumer_create_socket(&ustconsumer32_data,
2760 cmd_ctx->session->ust_session->consumer);
2761 if (ret < 0) {
2762 goto error;
2763 }
2764 }
2765 break;
2766 }
2767 default:
2768 break;
2769 }
2770 skip_domain:
2771
2772 /* Validate consumer daemon state when start/stop trace command */
2773 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2774 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2775 switch (cmd_ctx->lsm->domain.type) {
2776 case LTTNG_DOMAIN_UST:
2777 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2778 ret = LTTNG_ERR_NO_USTCONSUMERD;
2779 goto error;
2780 }
2781 break;
2782 case LTTNG_DOMAIN_KERNEL:
2783 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2784 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2785 goto error;
2786 }
2787 break;
2788 }
2789 }
2790
2791 /*
2792 * Check that the UID or GID match that of the tracing session.
2793 * The root user can interact with all sessions.
2794 */
2795 if (need_tracing_session) {
2796 if (!session_access_ok(cmd_ctx->session,
2797 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2798 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2799 ret = LTTNG_ERR_EPERM;
2800 goto error;
2801 }
2802 }
2803
2804 /*
2805 * Send relayd information to consumer as soon as we have a domain and a
2806 * session defined.
2807 */
2808 if (cmd_ctx->session && need_domain) {
2809 /*
2810 * Setup relayd if not done yet. If the relayd information was already
2811 * sent to the consumer, this call will gracefully return.
2812 */
2813 ret = cmd_setup_relayd(cmd_ctx->session);
2814 if (ret != LTTNG_OK) {
2815 goto error;
2816 }
2817 }
2818
2819 /* Process by command type */
2820 switch (cmd_ctx->lsm->cmd_type) {
2821 case LTTNG_ADD_CONTEXT:
2822 {
2823 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2824 cmd_ctx->lsm->u.context.channel_name,
2825 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2826 break;
2827 }
2828 case LTTNG_DISABLE_CHANNEL:
2829 {
2830 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2831 cmd_ctx->lsm->u.disable.channel_name);
2832 break;
2833 }
2834 case LTTNG_DISABLE_EVENT:
2835 {
2836 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2837 cmd_ctx->lsm->u.disable.channel_name,
2838 cmd_ctx->lsm->u.disable.name);
2839 break;
2840 }
2841 case LTTNG_DISABLE_ALL_EVENT:
2842 {
2843 DBG("Disabling all events");
2844
2845 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2846 cmd_ctx->lsm->u.disable.channel_name);
2847 break;
2848 }
2849 case LTTNG_ENABLE_CHANNEL:
2850 {
2851 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2852 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2853 break;
2854 }
2855 case LTTNG_ENABLE_EVENT:
2856 {
2857 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2858 cmd_ctx->lsm->u.enable.channel_name,
2859 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2860 break;
2861 }
2862 case LTTNG_ENABLE_ALL_EVENT:
2863 {
2864 DBG("Enabling all events");
2865
2866 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2867 cmd_ctx->lsm->u.enable.channel_name,
2868 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2869 break;
2870 }
2871 case LTTNG_LIST_TRACEPOINTS:
2872 {
2873 struct lttng_event *events;
2874 ssize_t nb_events;
2875
2876 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2877 if (nb_events < 0) {
2878 /* Return value is a negative lttng_error_code. */
2879 ret = -nb_events;
2880 goto error;
2881 }
2882
2883 /*
2884 * Setup lttng message with payload size set to the event list size in
2885 * bytes and then copy list into the llm payload.
2886 */
2887 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2888 if (ret < 0) {
2889 free(events);
2890 goto setup_error;
2891 }
2892
2893 /* Copy event list into message payload */
2894 memcpy(cmd_ctx->llm->payload, events,
2895 sizeof(struct lttng_event) * nb_events);
2896
2897 free(events);
2898
2899 ret = LTTNG_OK;
2900 break;
2901 }
2902 case LTTNG_LIST_TRACEPOINT_FIELDS:
2903 {
2904 struct lttng_event_field *fields;
2905 ssize_t nb_fields;
2906
2907 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2908 &fields);
2909 if (nb_fields < 0) {
2910 /* Return value is a negative lttng_error_code. */
2911 ret = -nb_fields;
2912 goto error;
2913 }
2914
2915 /*
2916 * Setup lttng message with payload size set to the event list size in
2917 * bytes and then copy list into the llm payload.
2918 */
2919 ret = setup_lttng_msg(cmd_ctx,
2920 sizeof(struct lttng_event_field) * nb_fields);
2921 if (ret < 0) {
2922 free(fields);
2923 goto setup_error;
2924 }
2925
2926 /* Copy event list into message payload */
2927 memcpy(cmd_ctx->llm->payload, fields,
2928 sizeof(struct lttng_event_field) * nb_fields);
2929
2930 free(fields);
2931
2932 ret = LTTNG_OK;
2933 break;
2934 }
2935 case LTTNG_SET_CONSUMER_URI:
2936 {
2937 size_t nb_uri, len;
2938 struct lttng_uri *uris;
2939
2940 nb_uri = cmd_ctx->lsm->u.uri.size;
2941 len = nb_uri * sizeof(struct lttng_uri);
2942
2943 if (nb_uri == 0) {
2944 ret = LTTNG_ERR_INVALID;
2945 goto error;
2946 }
2947
2948 uris = zmalloc(len);
2949 if (uris == NULL) {
2950 ret = LTTNG_ERR_FATAL;
2951 goto error;
2952 }
2953
2954 /* Receive variable len data */
2955 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2956 ret = lttcomm_recv_unix_sock(sock, uris, len);
2957 if (ret <= 0) {
2958 DBG("No URIs received from client... continuing");
2959 *sock_error = 1;
2960 ret = LTTNG_ERR_SESSION_FAIL;
2961 free(uris);
2962 goto error;
2963 }
2964
2965 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2966 nb_uri, uris);
2967 if (ret != LTTNG_OK) {
2968 free(uris);
2969 goto error;
2970 }
2971
2972 /*
2973 * XXX: 0 means that this URI should be applied on the session. Should
2974 * be a DOMAIN enuam.
2975 */
2976 if (cmd_ctx->lsm->domain.type == 0) {
2977 /* Add the URI for the UST session if a consumer is present. */
2978 if (cmd_ctx->session->ust_session &&
2979 cmd_ctx->session->ust_session->consumer) {
2980 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2981 nb_uri, uris);
2982 } else if (cmd_ctx->session->kernel_session &&
2983 cmd_ctx->session->kernel_session->consumer) {
2984 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2985 cmd_ctx->session, nb_uri, uris);
2986 }
2987 }
2988
2989 free(uris);
2990
2991 break;
2992 }
2993 case LTTNG_START_TRACE:
2994 {
2995 ret = cmd_start_trace(cmd_ctx->session);
2996 break;
2997 }
2998 case LTTNG_STOP_TRACE:
2999 {
3000 ret = cmd_stop_trace(cmd_ctx->session);
3001 break;
3002 }
3003 case LTTNG_CREATE_SESSION:
3004 {
3005 size_t nb_uri, len;
3006 struct lttng_uri *uris = NULL;
3007
3008 nb_uri = cmd_ctx->lsm->u.uri.size;
3009 len = nb_uri * sizeof(struct lttng_uri);
3010
3011 if (nb_uri > 0) {
3012 uris = zmalloc(len);
3013 if (uris == NULL) {
3014 ret = LTTNG_ERR_FATAL;
3015 goto error;
3016 }
3017
3018 /* Receive variable len data */
3019 DBG("Waiting for %zu URIs from client ...", nb_uri);
3020 ret = lttcomm_recv_unix_sock(sock, uris, len);
3021 if (ret <= 0) {
3022 DBG("No URIs received from client... continuing");
3023 *sock_error = 1;
3024 ret = LTTNG_ERR_SESSION_FAIL;
3025 free(uris);
3026 goto error;
3027 }
3028
3029 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3030 DBG("Creating session with ONE network URI is a bad call");
3031 ret = LTTNG_ERR_SESSION_FAIL;
3032 free(uris);
3033 goto error;
3034 }
3035 }
3036
3037 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3038 &cmd_ctx->creds);
3039
3040 free(uris);
3041
3042 break;
3043 }
3044 case LTTNG_DESTROY_SESSION:
3045 {
3046 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3047
3048 /* Set session to NULL so we do not unlock it after free. */
3049 cmd_ctx->session = NULL;
3050 break;
3051 }
3052 case LTTNG_LIST_DOMAINS:
3053 {
3054 ssize_t nb_dom;
3055 struct lttng_domain *domains;
3056
3057 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3058 if (nb_dom < 0) {
3059 /* Return value is a negative lttng_error_code. */
3060 ret = -nb_dom;
3061 goto error;
3062 }
3063
3064 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3065 if (ret < 0) {
3066 free(domains);
3067 goto setup_error;
3068 }
3069
3070 /* Copy event list into message payload */
3071 memcpy(cmd_ctx->llm->payload, domains,
3072 nb_dom * sizeof(struct lttng_domain));
3073
3074 free(domains);
3075
3076 ret = LTTNG_OK;
3077 break;
3078 }
3079 case LTTNG_LIST_CHANNELS:
3080 {
3081 int nb_chan;
3082 struct lttng_channel *channels;
3083
3084 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3085 cmd_ctx->session, &channels);
3086 if (nb_chan < 0) {
3087 /* Return value is a negative lttng_error_code. */
3088 ret = -nb_chan;
3089 goto error;
3090 }
3091
3092 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3093 if (ret < 0) {
3094 free(channels);
3095 goto setup_error;
3096 }
3097
3098 /* Copy event list into message payload */
3099 memcpy(cmd_ctx->llm->payload, channels,
3100 nb_chan * sizeof(struct lttng_channel));
3101
3102 free(channels);
3103
3104 ret = LTTNG_OK;
3105 break;
3106 }
3107 case LTTNG_LIST_EVENTS:
3108 {
3109 ssize_t nb_event;
3110 struct lttng_event *events = NULL;
3111
3112 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3113 cmd_ctx->lsm->u.list.channel_name, &events);
3114 if (nb_event < 0) {
3115 /* Return value is a negative lttng_error_code. */
3116 ret = -nb_event;
3117 goto error;
3118 }
3119
3120 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3121 if (ret < 0) {
3122 free(events);
3123 goto setup_error;
3124 }
3125
3126 /* Copy event list into message payload */
3127 memcpy(cmd_ctx->llm->payload, events,
3128 nb_event * sizeof(struct lttng_event));
3129
3130 free(events);
3131
3132 ret = LTTNG_OK;
3133 break;
3134 }
3135 case LTTNG_LIST_SESSIONS:
3136 {
3137 unsigned int nr_sessions;
3138
3139 session_lock_list();
3140 nr_sessions = lttng_sessions_count(
3141 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3142 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3143
3144 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3145 if (ret < 0) {
3146 session_unlock_list();
3147 goto setup_error;
3148 }
3149
3150 /* Filled the session array */
3151 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3152 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3153 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3154
3155 session_unlock_list();
3156
3157 ret = LTTNG_OK;
3158 break;
3159 }
3160 case LTTNG_CALIBRATE:
3161 {
3162 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3163 &cmd_ctx->lsm->u.calibrate);
3164 break;
3165 }
3166 case LTTNG_REGISTER_CONSUMER:
3167 {
3168 struct consumer_data *cdata;
3169
3170 switch (cmd_ctx->lsm->domain.type) {
3171 case LTTNG_DOMAIN_KERNEL:
3172 cdata = &kconsumer_data;
3173 break;
3174 default:
3175 ret = LTTNG_ERR_UND;
3176 goto error;
3177 }
3178
3179 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3180 cmd_ctx->lsm->u.reg.path, cdata);
3181 break;
3182 }
3183 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3184 {
3185 struct lttng_filter_bytecode *bytecode;
3186
3187 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3188 ret = LTTNG_ERR_FILTER_INVAL;
3189 goto error;
3190 }
3191 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3192 ret = LTTNG_ERR_FILTER_INVAL;
3193 goto error;
3194 }
3195 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3196 if (!bytecode) {
3197 ret = LTTNG_ERR_FILTER_NOMEM;
3198 goto error;
3199 }
3200 /* Receive var. len. data */
3201 DBG("Receiving var len data from client ...");
3202 ret = lttcomm_recv_unix_sock(sock, bytecode,
3203 cmd_ctx->lsm->u.enable.bytecode_len);
3204 if (ret <= 0) {
3205 DBG("Nothing recv() from client var len data... continuing");
3206 *sock_error = 1;
3207 ret = LTTNG_ERR_FILTER_INVAL;
3208 goto error;
3209 }
3210
3211 if (bytecode->len + sizeof(*bytecode)
3212 != cmd_ctx->lsm->u.enable.bytecode_len) {
3213 free(bytecode);
3214 ret = LTTNG_ERR_FILTER_INVAL;
3215 goto error;
3216 }
3217
3218 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3219 cmd_ctx->lsm->u.enable.channel_name,
3220 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3221 break;
3222 }
3223 case LTTNG_DATA_PENDING:
3224 {
3225 ret = cmd_data_pending(cmd_ctx->session);
3226 break;
3227 }
3228 default:
3229 ret = LTTNG_ERR_UND;
3230 break;
3231 }
3232
3233 error:
3234 if (cmd_ctx->llm == NULL) {
3235 DBG("Missing llm structure. Allocating one.");
3236 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3237 goto setup_error;
3238 }
3239 }
3240 /* Set return code */
3241 cmd_ctx->llm->ret_code = ret;
3242 setup_error:
3243 if (cmd_ctx->session) {
3244 session_unlock(cmd_ctx->session);
3245 }
3246 if (need_tracing_session) {
3247 session_unlock_list();
3248 }
3249 init_setup_error:
3250 return ret;
3251 }
3252
3253 /*
3254 * Thread managing health check socket.
3255 */
3256 static void *thread_manage_health(void *data)
3257 {
3258 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3259 uint32_t revents, nb_fd;
3260 struct lttng_poll_event events;
3261 struct lttcomm_health_msg msg;
3262 struct lttcomm_health_data reply;
3263
3264 DBG("[thread] Manage health check started");
3265
3266 rcu_register_thread();
3267
3268 /* We might hit an error path before this is created. */
3269 lttng_poll_init(&events);
3270
3271 /* Create unix socket */
3272 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3273 if (sock < 0) {
3274 ERR("Unable to create health check Unix socket");
3275 ret = -1;
3276 goto error;
3277 }
3278
3279 /*
3280 * Set the CLOEXEC flag. Return code is useless because either way, the
3281 * show must go on.
3282 */
3283 (void) utils_set_fd_cloexec(sock);
3284
3285 ret = lttcomm_listen_unix_sock(sock);
3286 if (ret < 0) {
3287 goto error;
3288 }
3289
3290 /*
3291 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3292 * more will be added to this poll set.
3293 */
3294 ret = sessiond_set_thread_pollset(&events, 2);
3295 if (ret < 0) {
3296 goto error;
3297 }
3298
3299 /* Add the application registration socket */
3300 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3301 if (ret < 0) {
3302 goto error;
3303 }
3304
3305 while (1) {
3306 DBG("Health check ready");
3307
3308 /* Inifinite blocking call, waiting for transmission */
3309 restart:
3310 ret = lttng_poll_wait(&events, -1);
3311 if (ret < 0) {
3312 /*
3313 * Restart interrupted system call.
3314 */
3315 if (errno == EINTR) {
3316 goto restart;
3317 }
3318 goto error;
3319 }
3320
3321 nb_fd = ret;
3322
3323 for (i = 0; i < nb_fd; i++) {
3324 /* Fetch once the poll data */
3325 revents = LTTNG_POLL_GETEV(&events, i);
3326 pollfd = LTTNG_POLL_GETFD(&events, i);
3327
3328 /* Thread quit pipe has been closed. Killing thread. */
3329 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3330 if (ret) {
3331 err = 0;
3332 goto exit;
3333 }
3334
3335 /* Event on the registration socket */
3336 if (pollfd == sock) {
3337 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3338 ERR("Health socket poll error");
3339 goto error;
3340 }
3341 }
3342 }
3343
3344 new_sock = lttcomm_accept_unix_sock(sock);
3345 if (new_sock < 0) {
3346 goto error;
3347 }
3348
3349 /*
3350 * Set the CLOEXEC flag. Return code is useless because either way, the
3351 * show must go on.
3352 */
3353 (void) utils_set_fd_cloexec(new_sock);
3354
3355 DBG("Receiving data from client for health...");
3356 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3357 if (ret <= 0) {
3358 DBG("Nothing recv() from client... continuing");
3359 ret = close(new_sock);
3360 if (ret) {
3361 PERROR("close");
3362 }
3363 new_sock = -1;
3364 continue;
3365 }
3366
3367 rcu_thread_online();
3368
3369 switch (msg.component) {
3370 case LTTNG_HEALTH_CMD:
3371 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3372 break;
3373 case LTTNG_HEALTH_APP_MANAGE:
3374 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3375 break;
3376 case LTTNG_HEALTH_APP_REG:
3377 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3378 break;
3379 case LTTNG_HEALTH_KERNEL:
3380 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3381 break;
3382 case LTTNG_HEALTH_CONSUMER:
3383 reply.ret_code = check_consumer_health();
3384 break;
3385 case LTTNG_HEALTH_HT_CLEANUP:
3386 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3387 break;
3388 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3389 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3390 break;
3391 case LTTNG_HEALTH_APP_REG_DISPATCH:
3392 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3393 break;
3394 case LTTNG_HEALTH_ALL:
3395 reply.ret_code =
3396 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3397 health_check_state(HEALTH_TYPE_APP_REG) &&
3398 health_check_state(HEALTH_TYPE_CMD) &&
3399 health_check_state(HEALTH_TYPE_KERNEL) &&
3400 check_consumer_health() &&
3401 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
3402 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3403 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3404 break;
3405 default:
3406 reply.ret_code = LTTNG_ERR_UND;
3407 break;
3408 }
3409
3410 /*
3411 * Flip ret value since 0 is a success and 1 indicates a bad health for
3412 * the client where in the sessiond it is the opposite. Again, this is
3413 * just to make things easier for us poor developer which enjoy a lot
3414 * lazyness.
3415 */
3416 if (reply.ret_code == 0 || reply.ret_code == 1) {
3417 reply.ret_code = !reply.ret_code;
3418 }
3419
3420 DBG2("Health check return value %d", reply.ret_code);
3421
3422 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3423 if (ret < 0) {
3424 ERR("Failed to send health data back to client");
3425 }
3426
3427 /* End of transmission */
3428 ret = close(new_sock);
3429 if (ret) {
3430 PERROR("close");
3431 }
3432 new_sock = -1;
3433 }
3434
3435 exit:
3436 error:
3437 if (err) {
3438 ERR("Health error occurred in %s", __func__);
3439 }
3440 DBG("Health check thread dying");
3441 unlink(health_unix_sock_path);
3442 if (sock >= 0) {
3443 ret = close(sock);
3444 if (ret) {
3445 PERROR("close");
3446 }
3447 }
3448
3449 lttng_poll_clean(&events);
3450
3451 rcu_unregister_thread();
3452 return NULL;
3453 }
3454
3455 /*
3456 * This thread manage all clients request using the unix client socket for
3457 * communication.
3458 */
3459 static void *thread_manage_clients(void *data)
3460 {
3461 int sock = -1, ret, i, pollfd, err = -1;
3462 int sock_error;
3463 uint32_t revents, nb_fd;
3464 struct command_ctx *cmd_ctx = NULL;
3465 struct lttng_poll_event events;
3466
3467 DBG("[thread] Manage client started");
3468
3469 rcu_register_thread();
3470
3471 health_register(HEALTH_TYPE_CMD);
3472
3473 if (testpoint(thread_manage_clients)) {
3474 goto error_testpoint;
3475 }
3476
3477 health_code_update();
3478
3479 ret = lttcomm_listen_unix_sock(client_sock);
3480 if (ret < 0) {
3481 goto error_listen;
3482 }
3483
3484 /*
3485 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3486 * more will be added to this poll set.
3487 */
3488 ret = sessiond_set_thread_pollset(&events, 2);
3489 if (ret < 0) {
3490 goto error_create_poll;
3491 }
3492
3493 /* Add the application registration socket */
3494 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3495 if (ret < 0) {
3496 goto error;
3497 }
3498
3499 /*
3500 * Notify parent pid that we are ready to accept command for client side.
3501 */
3502 if (opt_sig_parent) {
3503 kill(ppid, SIGUSR1);
3504 }
3505
3506 if (testpoint(thread_manage_clients_before_loop)) {
3507 goto error;
3508 }
3509
3510 health_code_update();
3511
3512 while (1) {
3513 DBG("Accepting client command ...");
3514
3515 /* Inifinite blocking call, waiting for transmission */
3516 restart:
3517 health_poll_entry();
3518 ret = lttng_poll_wait(&events, -1);
3519 health_poll_exit();
3520 if (ret < 0) {
3521 /*
3522 * Restart interrupted system call.
3523 */
3524 if (errno == EINTR) {
3525 goto restart;
3526 }
3527 goto error;
3528 }
3529
3530 nb_fd = ret;
3531
3532 for (i = 0; i < nb_fd; i++) {
3533 /* Fetch once the poll data */
3534 revents = LTTNG_POLL_GETEV(&events, i);
3535 pollfd = LTTNG_POLL_GETFD(&events, i);
3536
3537 health_code_update();
3538
3539 /* Thread quit pipe has been closed. Killing thread. */
3540 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3541 if (ret) {
3542 err = 0;
3543 goto exit;
3544 }
3545
3546 /* Event on the registration socket */
3547 if (pollfd == client_sock) {
3548 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3549 ERR("Client socket poll error");
3550 goto error;
3551 }
3552 }
3553 }
3554
3555 DBG("Wait for client response");
3556
3557 health_code_update();
3558
3559 sock = lttcomm_accept_unix_sock(client_sock);
3560 if (sock < 0) {
3561 goto error;
3562 }
3563
3564 /*
3565 * Set the CLOEXEC flag. Return code is useless because either way, the
3566 * show must go on.
3567 */
3568 (void) utils_set_fd_cloexec(sock);
3569
3570 /* Set socket option for credentials retrieval */
3571 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3572 if (ret < 0) {
3573 goto error;
3574 }
3575
3576 /* Allocate context command to process the client request */
3577 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3578 if (cmd_ctx == NULL) {
3579 PERROR("zmalloc cmd_ctx");
3580 goto error;
3581 }
3582
3583 /* Allocate data buffer for reception */
3584 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3585 if (cmd_ctx->lsm == NULL) {
3586 PERROR("zmalloc cmd_ctx->lsm");
3587 goto error;
3588 }
3589
3590 cmd_ctx->llm = NULL;
3591 cmd_ctx->session = NULL;
3592
3593 health_code_update();
3594
3595 /*
3596 * Data is received from the lttng client. The struct
3597 * lttcomm_session_msg (lsm) contains the command and data request of
3598 * the client.
3599 */
3600 DBG("Receiving data from client ...");
3601 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3602 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3603 if (ret <= 0) {
3604 DBG("Nothing recv() from client... continuing");
3605 ret = close(sock);
3606 if (ret) {
3607 PERROR("close");
3608 }
3609 sock = -1;
3610 clean_command_ctx(&cmd_ctx);
3611 continue;
3612 }
3613
3614 health_code_update();
3615
3616 // TODO: Validate cmd_ctx including sanity check for
3617 // security purpose.
3618
3619 rcu_thread_online();
3620 /*
3621 * This function dispatch the work to the kernel or userspace tracer
3622 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3623 * informations for the client. The command context struct contains
3624 * everything this function may needs.
3625 */
3626 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3627 rcu_thread_offline();
3628 if (ret < 0) {
3629 ret = close(sock);
3630 if (ret) {
3631 PERROR("close");
3632 }
3633 sock = -1;
3634 /*
3635 * TODO: Inform client somehow of the fatal error. At
3636 * this point, ret < 0 means that a zmalloc failed
3637 * (ENOMEM). Error detected but still accept
3638 * command, unless a socket error has been
3639 * detected.
3640 */
3641 clean_command_ctx(&cmd_ctx);
3642 continue;
3643 }
3644
3645 health_code_update();
3646
3647 DBG("Sending response (size: %d, retcode: %s)",
3648 cmd_ctx->lttng_msg_size,
3649 lttng_strerror(-cmd_ctx->llm->ret_code));
3650 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3651 if (ret < 0) {
3652 ERR("Failed to send data back to client");
3653 }
3654
3655 /* End of transmission */
3656 ret = close(sock);
3657 if (ret) {
3658 PERROR("close");
3659 }
3660 sock = -1;
3661
3662 clean_command_ctx(&cmd_ctx);
3663
3664 health_code_update();
3665 }
3666
3667 exit:
3668 error:
3669 if (sock >= 0) {
3670 ret = close(sock);
3671 if (ret) {
3672 PERROR("close");
3673 }
3674 }
3675
3676 lttng_poll_clean(&events);
3677 clean_command_ctx(&cmd_ctx);
3678
3679 error_listen:
3680 error_create_poll:
3681 error_testpoint:
3682 unlink(client_unix_sock_path);
3683 if (client_sock >= 0) {
3684 ret = close(client_sock);
3685 if (ret) {
3686 PERROR("close");
3687 }
3688 }
3689
3690 if (err) {
3691 health_error();
3692 ERR("Health error occurred in %s", __func__);
3693 }
3694
3695 health_unregister();
3696
3697 DBG("Client thread dying");
3698
3699 rcu_unregister_thread();
3700 return NULL;
3701 }
3702
3703
3704 /*
3705 * usage function on stderr
3706 */
3707 static void usage(void)
3708 {
3709 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3710 fprintf(stderr, " -h, --help Display this usage.\n");
3711 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3712 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3713 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3714 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3715 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3716 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3717 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3718 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3719 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3720 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3721 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3722 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3723 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3724 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3725 fprintf(stderr, " -V, --version Show version number.\n");
3726 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3727 fprintf(stderr, " -q, --quiet No output at all.\n");
3728 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3729 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3730 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3731 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3732 }
3733
3734 /*
3735 * daemon argument parsing
3736 */
3737 static int parse_args(int argc, char **argv)
3738 {
3739 int c;
3740
3741 static struct option long_options[] = {
3742 { "client-sock", 1, 0, 'c' },
3743 { "apps-sock", 1, 0, 'a' },
3744 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3745 { "kconsumerd-err-sock", 1, 0, 'E' },
3746 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3747 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3748 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3749 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3750 { "consumerd32-path", 1, 0, 'u' },
3751 { "consumerd32-libdir", 1, 0, 'U' },
3752 { "consumerd64-path", 1, 0, 't' },
3753 { "consumerd64-libdir", 1, 0, 'T' },
3754 { "daemonize", 0, 0, 'd' },
3755 { "sig-parent", 0, 0, 'S' },
3756 { "help", 0, 0, 'h' },
3757 { "group", 1, 0, 'g' },
3758 { "version", 0, 0, 'V' },
3759 { "quiet", 0, 0, 'q' },
3760 { "verbose", 0, 0, 'v' },
3761 { "verbose-consumer", 0, 0, 'Z' },
3762 { "no-kernel", 0, 0, 'N' },
3763 { "pidfile", 1, 0, 'p' },
3764 { NULL, 0, 0, 0 }
3765 };
3766
3767 while (1) {
3768 int option_index = 0;
3769 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3770 long_options, &option_index);
3771 if (c == -1) {
3772 break;
3773 }
3774
3775 switch (c) {
3776 case 0:
3777 fprintf(stderr, "option %s", long_options[option_index].name);
3778 if (optarg) {
3779 fprintf(stderr, " with arg %s\n", optarg);
3780 }
3781 break;
3782 case 'c':
3783 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3784 break;
3785 case 'a':
3786 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3787 break;
3788 case 'd':
3789 opt_daemon = 1;
3790 break;
3791 case 'g':
3792 opt_tracing_group = optarg;
3793 break;
3794 case 'h':
3795 usage();
3796 exit(EXIT_FAILURE);
3797 case 'V':
3798 fprintf(stdout, "%s\n", VERSION);
3799 exit(EXIT_SUCCESS);
3800 case 'S':
3801 opt_sig_parent = 1;
3802 break;
3803 case 'E':
3804 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3805 break;
3806 case 'C':
3807 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3808 break;
3809 case 'F':
3810 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3811 break;
3812 case 'D':
3813 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3814 break;
3815 case 'H':
3816 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3817 break;
3818 case 'G':
3819 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3820 break;
3821 case 'N':
3822 opt_no_kernel = 1;
3823 break;
3824 case 'q':
3825 lttng_opt_quiet = 1;
3826 break;
3827 case 'v':
3828 /* Verbose level can increase using multiple -v */
3829 lttng_opt_verbose += 1;
3830 break;
3831 case 'Z':
3832 opt_verbose_consumer += 1;
3833 break;
3834 case 'u':
3835 consumerd32_bin= optarg;
3836 break;
3837 case 'U':
3838 consumerd32_libdir = optarg;
3839 break;
3840 case 't':
3841 consumerd64_bin = optarg;
3842 break;
3843 case 'T':
3844 consumerd64_libdir = optarg;
3845 break;
3846 case 'p':
3847 opt_pidfile = optarg;
3848 break;
3849 default:
3850 /* Unknown option or other error.
3851 * Error is printed by getopt, just return */
3852 return -1;
3853 }
3854 }
3855
3856 return 0;
3857 }
3858
3859 /*
3860 * Creates the two needed socket by the daemon.
3861 * apps_sock - The communication socket for all UST apps.
3862 * client_sock - The communication of the cli tool (lttng).
3863 */
3864 static int init_daemon_socket(void)
3865 {
3866 int ret = 0;
3867 mode_t old_umask;
3868
3869 old_umask = umask(0);
3870
3871 /* Create client tool unix socket */
3872 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3873 if (client_sock < 0) {
3874 ERR("Create unix sock failed: %s", client_unix_sock_path);
3875 ret = -1;
3876 goto end;
3877 }
3878
3879 /* Set the cloexec flag */
3880 ret = utils_set_fd_cloexec(client_sock);
3881 if (ret < 0) {
3882 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3883 "Continuing but note that the consumer daemon will have a "
3884 "reference to this socket on exec()", client_sock);
3885 }
3886
3887 /* File permission MUST be 660 */
3888 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3889 if (ret < 0) {
3890 ERR("Set file permissions failed: %s", client_unix_sock_path);
3891 PERROR("chmod");
3892 goto end;
3893 }
3894
3895 /* Create the application unix socket */
3896 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3897 if (apps_sock < 0) {
3898 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3899 ret = -1;
3900 goto end;
3901 }
3902
3903 /* Set the cloexec flag */
3904 ret = utils_set_fd_cloexec(apps_sock);
3905 if (ret < 0) {
3906 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3907 "Continuing but note that the consumer daemon will have a "
3908 "reference to this socket on exec()", apps_sock);
3909 }
3910
3911 /* File permission MUST be 666 */
3912 ret = chmod(apps_unix_sock_path,
3913 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3914 if (ret < 0) {
3915 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3916 PERROR("chmod");
3917 goto end;
3918 }
3919
3920 DBG3("Session daemon client socket %d and application socket %d created",
3921 client_sock, apps_sock);
3922
3923 end:
3924 umask(old_umask);
3925 return ret;
3926 }
3927
3928 /*
3929 * Check if the global socket is available, and if a daemon is answering at the
3930 * other side. If yes, error is returned.
3931 */
3932 static int check_existing_daemon(void)
3933 {
3934 /* Is there anybody out there ? */
3935 if (lttng_session_daemon_alive()) {
3936 return -EEXIST;
3937 }
3938
3939 return 0;
3940 }
3941
3942 /*
3943 * Set the tracing group gid onto the client socket.
3944 *
3945 * Race window between mkdir and chown is OK because we are going from more
3946 * permissive (root.root) to less permissive (root.tracing).
3947 */
3948 static int set_permissions(char *rundir)
3949 {
3950 int ret;
3951 gid_t gid;
3952
3953 ret = allowed_group();
3954 if (ret < 0) {
3955 WARN("No tracing group detected");
3956 ret = 0;
3957 goto end;
3958 }
3959
3960 gid = ret;
3961
3962 /* Set lttng run dir */
3963 ret = chown(rundir, 0, gid);
3964 if (ret < 0) {
3965 ERR("Unable to set group on %s", rundir);
3966 PERROR("chown");
3967 }
3968
3969 /* Ensure tracing group can search the run dir */
3970 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3971 if (ret < 0) {
3972 ERR("Unable to set permissions on %s", rundir);
3973 PERROR("chmod");
3974 }
3975
3976 /* lttng client socket path */
3977 ret = chown(client_unix_sock_path, 0, gid);
3978 if (ret < 0) {
3979 ERR("Unable to set group on %s", client_unix_sock_path);
3980 PERROR("chown");
3981 }
3982
3983 /* kconsumer error socket path */
3984 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3985 if (ret < 0) {
3986 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3987 PERROR("chown");
3988 }
3989
3990 /* 64-bit ustconsumer error socket path */
3991 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3992 if (ret < 0) {
3993 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3994 PERROR("chown");
3995 }
3996
3997 /* 32-bit ustconsumer compat32 error socket path */
3998 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3999 if (ret < 0) {
4000 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4001 PERROR("chown");
4002 }
4003
4004 DBG("All permissions are set");
4005
4006 end:
4007 return ret;
4008 }
4009
4010 /*
4011 * Create the lttng run directory needed for all global sockets and pipe.
4012 */
4013 static int create_lttng_rundir(const char *rundir)
4014 {
4015 int ret;
4016
4017 DBG3("Creating LTTng run directory: %s", rundir);
4018
4019 ret = mkdir(rundir, S_IRWXU);
4020 if (ret < 0) {
4021 if (errno != EEXIST) {
4022 ERR("Unable to create %s", rundir);
4023 goto error;
4024 } else {
4025 ret = 0;
4026 }
4027 }
4028
4029 error:
4030 return ret;
4031 }
4032
4033 /*
4034 * Setup sockets and directory needed by the kconsumerd communication with the
4035 * session daemon.
4036 */
4037 static int set_consumer_sockets(struct consumer_data *consumer_data,
4038 const char *rundir)
4039 {
4040 int ret;
4041 char path[PATH_MAX];
4042
4043 switch (consumer_data->type) {
4044 case LTTNG_CONSUMER_KERNEL:
4045 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4046 break;
4047 case LTTNG_CONSUMER64_UST:
4048 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4049 break;
4050 case LTTNG_CONSUMER32_UST:
4051 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4052 break;
4053 default:
4054 ERR("Consumer type unknown");
4055 ret = -EINVAL;
4056 goto error;
4057 }
4058
4059 DBG2("Creating consumer directory: %s", path);
4060
4061 ret = mkdir(path, S_IRWXU);
4062 if (ret < 0) {
4063 if (errno != EEXIST) {
4064 PERROR("mkdir");
4065 ERR("Failed to create %s", path);
4066 goto error;
4067 }
4068 ret = -1;
4069 }
4070
4071 /* Create the kconsumerd error unix socket */
4072 consumer_data->err_sock =
4073 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4074 if (consumer_data->err_sock < 0) {
4075 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4076 ret = -1;
4077 goto error;
4078 }
4079
4080 /*
4081 * Set the CLOEXEC flag. Return code is useless because either way, the
4082 * show must go on.
4083 */
4084 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4085 if (ret < 0) {
4086 PERROR("utils_set_fd_cloexec");
4087 /* continue anyway */
4088 }
4089
4090 /* File permission MUST be 660 */
4091 ret = chmod(consumer_data->err_unix_sock_path,
4092 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4093 if (ret < 0) {
4094 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4095 PERROR("chmod");
4096 goto error;
4097 }
4098
4099 error:
4100 return ret;
4101 }
4102
4103 /*
4104 * Signal handler for the daemon
4105 *
4106 * Simply stop all worker threads, leaving main() return gracefully after
4107 * joining all threads and calling cleanup().
4108 */
4109 static void sighandler(int sig)
4110 {
4111 switch (sig) {
4112 case SIGPIPE:
4113 DBG("SIGPIPE caught");
4114 return;
4115 case SIGINT:
4116 DBG("SIGINT caught");
4117 stop_threads();
4118 break;
4119 case SIGTERM:
4120 DBG("SIGTERM caught");
4121 stop_threads();
4122 break;
4123 default:
4124 break;
4125 }
4126 }
4127
4128 /*
4129 * Setup signal handler for :
4130 * SIGINT, SIGTERM, SIGPIPE
4131 */
4132 static int set_signal_handler(void)
4133 {
4134 int ret = 0;
4135 struct sigaction sa;
4136 sigset_t sigset;
4137
4138 if ((ret = sigemptyset(&sigset)) < 0) {
4139 PERROR("sigemptyset");
4140 return ret;
4141 }
4142
4143 sa.sa_handler = sighandler;
4144 sa.sa_mask = sigset;
4145 sa.sa_flags = 0;
4146 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4147 PERROR("sigaction");
4148 return ret;
4149 }
4150
4151 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4152 PERROR("sigaction");
4153 return ret;
4154 }
4155
4156 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4157 PERROR("sigaction");
4158 return ret;
4159 }
4160
4161 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4162
4163 return ret;
4164 }
4165
4166 /*
4167 * Set open files limit to unlimited. This daemon can open a large number of
4168 * file descriptors in order to consumer multiple kernel traces.
4169 */
4170 static void set_ulimit(void)
4171 {
4172 int ret;
4173 struct rlimit lim;
4174
4175 /* The kernel does not allowed an infinite limit for open files */
4176 lim.rlim_cur = 65535;
4177 lim.rlim_max = 65535;
4178
4179 ret = setrlimit(RLIMIT_NOFILE, &lim);
4180 if (ret < 0) {
4181 PERROR("failed to set open files limit");
4182 }
4183 }
4184
4185 /*
4186 * Write pidfile using the rundir and opt_pidfile.
4187 */
4188 static void write_pidfile(void)
4189 {
4190 int ret;
4191 char pidfile_path[PATH_MAX];
4192
4193 assert(rundir);
4194
4195 if (opt_pidfile) {
4196 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4197 } else {
4198 /* Build pidfile path from rundir and opt_pidfile. */
4199 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4200 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4201 if (ret < 0) {
4202 PERROR("snprintf pidfile path");
4203 goto error;
4204 }
4205 }
4206
4207 /*
4208 * Create pid file in rundir. Return value is of no importance. The
4209 * execution will continue even though we are not able to write the file.
4210 */
4211 (void) utils_create_pid_file(getpid(), pidfile_path);
4212
4213 error:
4214 return;
4215 }
4216
4217 /*
4218 * main
4219 */
4220 int main(int argc, char **argv)
4221 {
4222 int ret = 0;
4223 void *status;
4224 const char *home_path, *env_app_timeout;
4225
4226 init_kernel_workarounds();
4227
4228 rcu_register_thread();
4229
4230 setup_consumerd_path();
4231
4232 page_size = sysconf(_SC_PAGESIZE);
4233 if (page_size < 0) {
4234 PERROR("sysconf _SC_PAGESIZE");
4235 page_size = LONG_MAX;
4236 WARN("Fallback page size to %ld", page_size);
4237 }
4238
4239 /* Parse arguments */
4240 progname = argv[0];
4241 if ((ret = parse_args(argc, argv)) < 0) {
4242 goto error;
4243 }
4244
4245 /* Daemonize */
4246 if (opt_daemon) {
4247 int i;
4248
4249 /*
4250 * fork
4251 * child: setsid, close FD 0, 1, 2, chdir /
4252 * parent: exit (if fork is successful)
4253 */
4254 ret = daemon(0, 0);
4255 if (ret < 0) {
4256 PERROR("daemon");
4257 goto error;
4258 }
4259 /*
4260 * We are in the child. Make sure all other file
4261 * descriptors are closed, in case we are called with
4262 * more opened file descriptors than the standard ones.
4263 */
4264 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4265 (void) close(i);
4266 }
4267 }
4268
4269 /* Create thread quit pipe */
4270 if ((ret = init_thread_quit_pipe()) < 0) {
4271 goto error;
4272 }
4273
4274 /* Check if daemon is UID = 0 */
4275 is_root = !getuid();
4276
4277 if (is_root) {
4278 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4279
4280 /* Create global run dir with root access */
4281 ret = create_lttng_rundir(rundir);
4282 if (ret < 0) {
4283 goto error;
4284 }
4285
4286 if (strlen(apps_unix_sock_path) == 0) {
4287 snprintf(apps_unix_sock_path, PATH_MAX,
4288 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4289 }
4290
4291 if (strlen(client_unix_sock_path) == 0) {
4292 snprintf(client_unix_sock_path, PATH_MAX,
4293 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4294 }
4295
4296 /* Set global SHM for ust */
4297 if (strlen(wait_shm_path) == 0) {
4298 snprintf(wait_shm_path, PATH_MAX,
4299 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4300 }
4301
4302 if (strlen(health_unix_sock_path) == 0) {
4303 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4304 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4305 }
4306
4307 /* Setup kernel consumerd path */
4308 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4309 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4310 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4311 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4312
4313 DBG2("Kernel consumer err path: %s",
4314 kconsumer_data.err_unix_sock_path);
4315 DBG2("Kernel consumer cmd path: %s",
4316 kconsumer_data.cmd_unix_sock_path);
4317 } else {
4318 home_path = utils_get_home_dir();
4319 if (home_path == NULL) {
4320 /* TODO: Add --socket PATH option */
4321 ERR("Can't get HOME directory for sockets creation.");
4322 ret = -EPERM;
4323 goto error;
4324 }
4325
4326 /*
4327 * Create rundir from home path. This will create something like
4328 * $HOME/.lttng
4329 */
4330 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4331 if (ret < 0) {
4332 ret = -ENOMEM;
4333 goto error;
4334 }
4335
4336 ret = create_lttng_rundir(rundir);
4337 if (ret < 0) {
4338 goto error;
4339 }
4340
4341 if (strlen(apps_unix_sock_path) == 0) {
4342 snprintf(apps_unix_sock_path, PATH_MAX,
4343 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4344 }
4345
4346 /* Set the cli tool unix socket path */
4347 if (strlen(client_unix_sock_path) == 0) {
4348 snprintf(client_unix_sock_path, PATH_MAX,
4349 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4350 }
4351
4352 /* Set global SHM for ust */
4353 if (strlen(wait_shm_path) == 0) {
4354 snprintf(wait_shm_path, PATH_MAX,
4355 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4356 }
4357
4358 /* Set health check Unix path */
4359 if (strlen(health_unix_sock_path) == 0) {
4360 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4361 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4362 }
4363 }
4364
4365 /* Set consumer initial state */
4366 kernel_consumerd_state = CONSUMER_STOPPED;
4367 ust_consumerd_state = CONSUMER_STOPPED;
4368
4369 DBG("Client socket path %s", client_unix_sock_path);
4370 DBG("Application socket path %s", apps_unix_sock_path);
4371 DBG("Application wait path %s", wait_shm_path);
4372 DBG("LTTng run directory path: %s", rundir);
4373
4374 /* 32 bits consumerd path setup */
4375 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4376 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4377 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4378 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4379
4380 DBG2("UST consumer 32 bits err path: %s",
4381 ustconsumer32_data.err_unix_sock_path);
4382 DBG2("UST consumer 32 bits cmd path: %s",
4383 ustconsumer32_data.cmd_unix_sock_path);
4384
4385 /* 64 bits consumerd path setup */
4386 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4387 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4388 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4389 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4390
4391 DBG2("UST consumer 64 bits err path: %s",
4392 ustconsumer64_data.err_unix_sock_path);
4393 DBG2("UST consumer 64 bits cmd path: %s",
4394 ustconsumer64_data.cmd_unix_sock_path);
4395
4396 /*
4397 * See if daemon already exist.
4398 */
4399 if ((ret = check_existing_daemon()) < 0) {
4400 ERR("Already running daemon.\n");
4401 /*
4402 * We do not goto exit because we must not cleanup()
4403 * because a daemon is already running.
4404 */
4405 goto error;
4406 }
4407
4408 /*
4409 * Init UST app hash table. Alloc hash table before this point since
4410 * cleanup() can get called after that point.
4411 */
4412 ust_app_ht_alloc();
4413
4414 /* After this point, we can safely call cleanup() with "goto exit" */
4415
4416 /*
4417 * These actions must be executed as root. We do that *after* setting up
4418 * the sockets path because we MUST make the check for another daemon using
4419 * those paths *before* trying to set the kernel consumer sockets and init
4420 * kernel tracer.
4421 */
4422 if (is_root) {
4423 ret = set_consumer_sockets(&kconsumer_data, rundir);
4424 if (ret < 0) {
4425 goto exit;
4426 }
4427
4428 /* Setup kernel tracer */
4429 if (!opt_no_kernel) {
4430 init_kernel_tracer();
4431 }
4432
4433 /* Set ulimit for open files */
4434 set_ulimit();
4435 }
4436 /* init lttng_fd tracking must be done after set_ulimit. */
4437 lttng_fd_init();
4438
4439 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4440 if (ret < 0) {
4441 goto exit;
4442 }
4443
4444 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4445 if (ret < 0) {
4446 goto exit;
4447 }
4448
4449 if ((ret = set_signal_handler()) < 0) {
4450 goto exit;
4451 }
4452
4453 /* Setup the needed unix socket */
4454 if ((ret = init_daemon_socket()) < 0) {
4455 goto exit;
4456 }
4457
4458 /* Set credentials to socket */
4459 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4460 goto exit;
4461 }
4462
4463 /* Get parent pid if -S, --sig-parent is specified. */
4464 if (opt_sig_parent) {
4465 ppid = getppid();
4466 }
4467
4468 /* Setup the kernel pipe for waking up the kernel thread */
4469 if (is_root && !opt_no_kernel) {
4470 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4471 goto exit;
4472 }
4473 }
4474
4475 /* Setup the thread ht_cleanup communication pipe. */
4476 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4477 goto exit;
4478 }
4479
4480 /* Setup the thread apps communication pipe. */
4481 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4482 goto exit;
4483 }
4484
4485 /* Setup the thread apps notify communication pipe. */
4486 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4487 goto exit;
4488 }
4489
4490 /* Initialize global buffer per UID and PID registry. */
4491 buffer_reg_init_uid_registry();
4492 buffer_reg_init_pid_registry();
4493
4494 /* Init UST command queue. */
4495 cds_wfq_init(&ust_cmd_queue.queue);
4496
4497 /*
4498 * Get session list pointer. This pointer MUST NOT be free(). This list is
4499 * statically declared in session.c
4500 */
4501 session_list_ptr = session_get_list();
4502
4503 /* Set up max poll set size */
4504 lttng_poll_set_max_size();
4505
4506 cmd_init();
4507
4508 /* Check for the application socket timeout env variable. */
4509 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4510 if (env_app_timeout) {
4511 app_socket_timeout = atoi(env_app_timeout);
4512 } else {
4513 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4514 }
4515
4516 write_pidfile();
4517
4518 /* Create thread to manage the client socket */
4519 ret = pthread_create(&ht_cleanup_thread, NULL,
4520 thread_ht_cleanup, (void *) NULL);
4521 if (ret != 0) {
4522 PERROR("pthread_create ht_cleanup");
4523 goto exit_ht_cleanup;
4524 }
4525
4526 /* Create thread to manage the client socket */
4527 ret = pthread_create(&health_thread, NULL,
4528 thread_manage_health, (void *) NULL);
4529 if (ret != 0) {
4530 PERROR("pthread_create health");
4531 goto exit_health;
4532 }
4533
4534 /* Create thread to manage the client socket */
4535 ret = pthread_create(&client_thread, NULL,
4536 thread_manage_clients, (void *) NULL);
4537 if (ret != 0) {
4538 PERROR("pthread_create clients");
4539 goto exit_client;
4540 }
4541
4542 /* Create thread to dispatch registration */
4543 ret = pthread_create(&dispatch_thread, NULL,
4544 thread_dispatch_ust_registration, (void *) NULL);
4545 if (ret != 0) {
4546 PERROR("pthread_create dispatch");
4547 goto exit_dispatch;
4548 }
4549
4550 /* Create thread to manage application registration. */
4551 ret = pthread_create(&reg_apps_thread, NULL,
4552 thread_registration_apps, (void *) NULL);
4553 if (ret != 0) {
4554 PERROR("pthread_create registration");
4555 goto exit_reg_apps;
4556 }
4557
4558 /* Create thread to manage application socket */
4559 ret = pthread_create(&apps_thread, NULL,
4560 thread_manage_apps, (void *) NULL);
4561 if (ret != 0) {
4562 PERROR("pthread_create apps");
4563 goto exit_apps;
4564 }
4565
4566 /* Create thread to manage application notify socket */
4567 ret = pthread_create(&apps_notify_thread, NULL,
4568 ust_thread_manage_notify, (void *) NULL);
4569 if (ret != 0) {
4570 PERROR("pthread_create apps");
4571 goto exit_apps;
4572 }
4573
4574 /* Don't start this thread if kernel tracing is not requested nor root */
4575 if (is_root && !opt_no_kernel) {
4576 /* Create kernel thread to manage kernel event */
4577 ret = pthread_create(&kernel_thread, NULL,
4578 thread_manage_kernel, (void *) NULL);
4579 if (ret != 0) {
4580 PERROR("pthread_create kernel");
4581 goto exit_kernel;
4582 }
4583
4584 ret = pthread_join(kernel_thread, &status);
4585 if (ret != 0) {
4586 PERROR("pthread_join");
4587 goto error; /* join error, exit without cleanup */
4588 }
4589 }
4590
4591 exit_kernel:
4592 ret = pthread_join(apps_thread, &status);
4593 if (ret != 0) {
4594 PERROR("pthread_join");
4595 goto error; /* join error, exit without cleanup */
4596 }
4597
4598 exit_apps:
4599 ret = pthread_join(reg_apps_thread, &status);
4600 if (ret != 0) {
4601 PERROR("pthread_join");
4602 goto error; /* join error, exit without cleanup */
4603 }
4604
4605 exit_reg_apps:
4606 ret = pthread_join(dispatch_thread, &status);
4607 if (ret != 0) {
4608 PERROR("pthread_join");
4609 goto error; /* join error, exit without cleanup */
4610 }
4611
4612 exit_dispatch:
4613 ret = pthread_join(client_thread, &status);
4614 if (ret != 0) {
4615 PERROR("pthread_join");
4616 goto error; /* join error, exit without cleanup */
4617 }
4618
4619 ret = join_consumer_thread(&kconsumer_data);
4620 if (ret != 0) {
4621 PERROR("join_consumer");
4622 goto error; /* join error, exit without cleanup */
4623 }
4624
4625 ret = join_consumer_thread(&ustconsumer32_data);
4626 if (ret != 0) {
4627 PERROR("join_consumer ust32");
4628 goto error; /* join error, exit without cleanup */
4629 }
4630
4631 ret = join_consumer_thread(&ustconsumer64_data);
4632 if (ret != 0) {
4633 PERROR("join_consumer ust64");
4634 goto error; /* join error, exit without cleanup */
4635 }
4636
4637 exit_client:
4638 ret = pthread_join(health_thread, &status);
4639 if (ret != 0) {
4640 PERROR("pthread_join health thread");
4641 goto error; /* join error, exit without cleanup */
4642 }
4643
4644 exit_health:
4645 ret = pthread_join(ht_cleanup_thread, &status);
4646 if (ret != 0) {
4647 PERROR("pthread_join ht cleanup thread");
4648 goto error; /* join error, exit without cleanup */
4649 }
4650 exit_ht_cleanup:
4651 exit:
4652 /*
4653 * cleanup() is called when no other thread is running.
4654 */
4655 rcu_thread_online();
4656 cleanup();
4657 rcu_thread_offline();
4658 rcu_unregister_thread();
4659 if (!ret) {
4660 exit(EXIT_SUCCESS);
4661 }
4662 error:
4663 exit(EXIT_FAILURE);
4664 }
This page took 0.13477 seconds and 4 git commands to generate.