76d1857004a415701ecb166f8ace70768bc8d472
[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 <semaphore.h>
25 #include <signal.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.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/poll.h>
42 #include <common/compat/socket.h>
43 #include <common/defaults.h>
44 #include <common/kernel-consumer/kernel-consumer.h>
45 #include <common/futex.h>
46 #include <common/relayd/relayd.h>
47 #include <common/utils.h>
48
49 #include "lttng-sessiond.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 "filter.h"
64 #include "health.h"
65
66 #define CONSUMERD_FILE "lttng-consumerd"
67
68 /* Const values */
69 const char default_home_dir[] = DEFAULT_HOME_DIR;
70 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
71 const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
72 const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
73
74 const char *progname;
75 const char *opt_tracing_group;
76 static int opt_sig_parent;
77 static int opt_verbose_consumer;
78 static int opt_daemon;
79 static int opt_no_kernel;
80 static int is_root; /* Set to 1 if the daemon is running as root */
81 static pid_t ppid; /* Parent PID for --sig-parent option */
82 static char *rundir;
83
84 /* Consumer daemon specific control data */
85 static struct consumer_data kconsumer_data = {
86 .type = LTTNG_CONSUMER_KERNEL,
87 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
88 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
89 .err_sock = -1,
90 .cmd_sock = -1,
91 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
92 .lock = PTHREAD_MUTEX_INITIALIZER,
93 };
94 static struct consumer_data ustconsumer64_data = {
95 .type = LTTNG_CONSUMER64_UST,
96 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
97 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
98 .err_sock = -1,
99 .cmd_sock = -1,
100 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
101 .lock = PTHREAD_MUTEX_INITIALIZER,
102 };
103 static struct consumer_data ustconsumer32_data = {
104 .type = LTTNG_CONSUMER32_UST,
105 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
106 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
107 .err_sock = -1,
108 .cmd_sock = -1,
109 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
110 .lock = PTHREAD_MUTEX_INITIALIZER,
111 };
112
113 /* Shared between threads */
114 static int dispatch_thread_exit;
115
116 /* Global application Unix socket path */
117 static char apps_unix_sock_path[PATH_MAX];
118 /* Global client Unix socket path */
119 static char client_unix_sock_path[PATH_MAX];
120 /* global wait shm path for UST */
121 static char wait_shm_path[PATH_MAX];
122 /* Global health check unix path */
123 static char health_unix_sock_path[PATH_MAX];
124
125 /* Sockets and FDs */
126 static int client_sock = -1;
127 static int apps_sock = -1;
128 int kernel_tracer_fd = -1;
129 static int kernel_poll_pipe[2] = { -1, -1 };
130
131 /*
132 * Quit pipe for all threads. This permits a single cancellation point
133 * for all threads when receiving an event on the pipe.
134 */
135 static int thread_quit_pipe[2] = { -1, -1 };
136
137 /*
138 * This pipe is used to inform the thread managing application communication
139 * that a command is queued and ready to be processed.
140 */
141 static int apps_cmd_pipe[2] = { -1, -1 };
142
143 /* Pthread, Mutexes and Semaphores */
144 static pthread_t apps_thread;
145 static pthread_t reg_apps_thread;
146 static pthread_t client_thread;
147 static pthread_t kernel_thread;
148 static pthread_t dispatch_thread;
149 static pthread_t health_thread;
150
151 /*
152 * UST registration command queue. This queue is tied with a futex and uses a N
153 * wakers / 1 waiter implemented and detailed in futex.c/.h
154 *
155 * The thread_manage_apps and thread_dispatch_ust_registration interact with
156 * this queue and the wait/wake scheme.
157 */
158 static struct ust_cmd_queue ust_cmd_queue;
159
160 /*
161 * Pointer initialized before thread creation.
162 *
163 * This points to the tracing session list containing the session count and a
164 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
165 * MUST NOT be taken if you call a public function in session.c.
166 *
167 * The lock is nested inside the structure: session_list_ptr->lock. Please use
168 * session_lock_list and session_unlock_list for lock acquisition.
169 */
170 static struct ltt_session_list *session_list_ptr;
171
172 int ust_consumerd64_fd = -1;
173 int ust_consumerd32_fd = -1;
174
175 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
176 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
177 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
178 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
179
180 static const char *module_proc_lttng = "/proc/lttng";
181
182 /*
183 * Consumer daemon state which is changed when spawning it, killing it or in
184 * case of a fatal error.
185 */
186 enum consumerd_state {
187 CONSUMER_STARTED = 1,
188 CONSUMER_STOPPED = 2,
189 CONSUMER_ERROR = 3,
190 };
191
192 /*
193 * This consumer daemon state is used to validate if a client command will be
194 * able to reach the consumer. If not, the client is informed. For instance,
195 * doing a "lttng start" when the consumer state is set to ERROR will return an
196 * error to the client.
197 *
198 * The following example shows a possible race condition of this scheme:
199 *
200 * consumer thread error happens
201 * client cmd arrives
202 * client cmd checks state -> still OK
203 * consumer thread exit, sets error
204 * client cmd try to talk to consumer
205 * ...
206 *
207 * However, since the consumer is a different daemon, we have no way of making
208 * sure the command will reach it safely even with this state flag. This is why
209 * we consider that up to the state validation during command processing, the
210 * command is safe. After that, we can not guarantee the correctness of the
211 * client request vis-a-vis the consumer.
212 */
213 static enum consumerd_state ust_consumerd_state;
214 static enum consumerd_state kernel_consumerd_state;
215
216 /* Used for the health monitoring of the session daemon. See health.h */
217 struct health_state health_thread_cmd;
218 struct health_state health_thread_app_manage;
219 struct health_state health_thread_app_reg;
220 struct health_state health_thread_kernel;
221
222 static
223 void setup_consumerd_path(void)
224 {
225 const char *bin, *libdir;
226
227 /*
228 * Allow INSTALL_BIN_PATH to be used as a target path for the
229 * native architecture size consumer if CONFIG_CONSUMER*_PATH
230 * has not been defined.
231 */
232 #if (CAA_BITS_PER_LONG == 32)
233 if (!consumerd32_bin[0]) {
234 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
235 }
236 if (!consumerd32_libdir[0]) {
237 consumerd32_libdir = INSTALL_LIB_PATH;
238 }
239 #elif (CAA_BITS_PER_LONG == 64)
240 if (!consumerd64_bin[0]) {
241 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
242 }
243 if (!consumerd64_libdir[0]) {
244 consumerd64_libdir = INSTALL_LIB_PATH;
245 }
246 #else
247 #error "Unknown bitness"
248 #endif
249
250 /*
251 * runtime env. var. overrides the build default.
252 */
253 bin = getenv("LTTNG_CONSUMERD32_BIN");
254 if (bin) {
255 consumerd32_bin = bin;
256 }
257 bin = getenv("LTTNG_CONSUMERD64_BIN");
258 if (bin) {
259 consumerd64_bin = bin;
260 }
261 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
262 if (libdir) {
263 consumerd32_libdir = libdir;
264 }
265 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
266 if (libdir) {
267 consumerd64_libdir = libdir;
268 }
269 }
270
271 /*
272 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
273 */
274 static int create_thread_poll_set(struct lttng_poll_event *events,
275 unsigned int size)
276 {
277 int ret;
278
279 if (events == NULL || size == 0) {
280 ret = -1;
281 goto error;
282 }
283
284 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
285 if (ret < 0) {
286 goto error;
287 }
288
289 /* Add quit pipe */
290 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
291 if (ret < 0) {
292 goto error;
293 }
294
295 return 0;
296
297 error:
298 return ret;
299 }
300
301 /*
302 * Check if the thread quit pipe was triggered.
303 *
304 * Return 1 if it was triggered else 0;
305 */
306 static int check_thread_quit_pipe(int fd, uint32_t events)
307 {
308 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
309 return 1;
310 }
311
312 return 0;
313 }
314
315 /*
316 * Return group ID of the tracing group or -1 if not found.
317 */
318 static gid_t allowed_group(void)
319 {
320 struct group *grp;
321
322 if (opt_tracing_group) {
323 grp = getgrnam(opt_tracing_group);
324 } else {
325 grp = getgrnam(default_tracing_group);
326 }
327 if (!grp) {
328 return -1;
329 } else {
330 return grp->gr_gid;
331 }
332 }
333
334 /*
335 * Init thread quit pipe.
336 *
337 * Return -1 on error or 0 if all pipes are created.
338 */
339 static int init_thread_quit_pipe(void)
340 {
341 int ret, i;
342
343 ret = pipe(thread_quit_pipe);
344 if (ret < 0) {
345 PERROR("thread quit pipe");
346 goto error;
347 }
348
349 for (i = 0; i < 2; i++) {
350 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
351 if (ret < 0) {
352 PERROR("fcntl");
353 goto error;
354 }
355 }
356
357 error:
358 return ret;
359 }
360
361 /*
362 * Stop all threads by closing the thread quit pipe.
363 */
364 static void stop_threads(void)
365 {
366 int ret;
367
368 /* Stopping all threads */
369 DBG("Terminating all threads");
370 ret = notify_thread_pipe(thread_quit_pipe[1]);
371 if (ret < 0) {
372 ERR("write error on thread quit pipe");
373 }
374
375 /* Dispatch thread */
376 CMM_STORE_SHARED(dispatch_thread_exit, 1);
377 futex_nto1_wake(&ust_cmd_queue.futex);
378 }
379
380 /*
381 * Cleanup the daemon
382 */
383 static void cleanup(void)
384 {
385 int ret;
386 char *cmd;
387 struct ltt_session *sess, *stmp;
388
389 DBG("Cleaning up");
390
391 /* First thing first, stop all threads */
392 utils_close_pipe(thread_quit_pipe);
393
394 DBG("Removing %s directory", rundir);
395 ret = asprintf(&cmd, "rm -rf %s", rundir);
396 if (ret < 0) {
397 ERR("asprintf failed. Something is really wrong!");
398 }
399
400 /* Remove lttng run directory */
401 ret = system(cmd);
402 if (ret < 0) {
403 ERR("Unable to clean %s", rundir);
404 }
405 free(cmd);
406
407 DBG("Cleaning up all sessions");
408
409 /* Destroy session list mutex */
410 if (session_list_ptr != NULL) {
411 pthread_mutex_destroy(&session_list_ptr->lock);
412
413 /* Cleanup ALL session */
414 cds_list_for_each_entry_safe(sess, stmp,
415 &session_list_ptr->head, list) {
416 cmd_destroy_session(sess, kernel_poll_pipe[1]);
417 }
418 }
419
420 DBG("Closing all UST sockets");
421 ust_app_clean_list();
422
423 if (is_root && !opt_no_kernel) {
424 DBG2("Closing kernel fd");
425 if (kernel_tracer_fd >= 0) {
426 ret = close(kernel_tracer_fd);
427 if (ret) {
428 PERROR("close");
429 }
430 }
431 DBG("Unloading kernel modules");
432 modprobe_remove_lttng_all();
433 }
434
435 utils_close_pipe(kernel_poll_pipe);
436 utils_close_pipe(apps_cmd_pipe);
437
438 /* <fun> */
439 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
440 "Matthew, BEET driven development works!%c[%dm",
441 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
442 /* </fun> */
443 }
444
445 /*
446 * Send data on a unix socket using the liblttsessiondcomm API.
447 *
448 * Return lttcomm error code.
449 */
450 static int send_unix_sock(int sock, void *buf, size_t len)
451 {
452 /* Check valid length */
453 if (len <= 0) {
454 return -1;
455 }
456
457 return lttcomm_send_unix_sock(sock, buf, len);
458 }
459
460 /*
461 * Free memory of a command context structure.
462 */
463 static void clean_command_ctx(struct command_ctx **cmd_ctx)
464 {
465 DBG("Clean command context structure");
466 if (*cmd_ctx) {
467 if ((*cmd_ctx)->llm) {
468 free((*cmd_ctx)->llm);
469 }
470 if ((*cmd_ctx)->lsm) {
471 free((*cmd_ctx)->lsm);
472 }
473 free(*cmd_ctx);
474 *cmd_ctx = NULL;
475 }
476 }
477
478 /*
479 * Notify UST applications using the shm mmap futex.
480 */
481 static int notify_ust_apps(int active)
482 {
483 char *wait_shm_mmap;
484
485 DBG("Notifying applications of session daemon state: %d", active);
486
487 /* See shm.c for this call implying mmap, shm and futex calls */
488 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
489 if (wait_shm_mmap == NULL) {
490 goto error;
491 }
492
493 /* Wake waiting process */
494 futex_wait_update((int32_t *) wait_shm_mmap, active);
495
496 /* Apps notified successfully */
497 return 0;
498
499 error:
500 return -1;
501 }
502
503 /*
504 * Setup the outgoing data buffer for the response (llm) by allocating the
505 * right amount of memory and copying the original information from the lsm
506 * structure.
507 *
508 * Return total size of the buffer pointed by buf.
509 */
510 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
511 {
512 int ret, buf_size;
513
514 buf_size = size;
515
516 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
517 if (cmd_ctx->llm == NULL) {
518 PERROR("zmalloc");
519 ret = -ENOMEM;
520 goto error;
521 }
522
523 /* Copy common data */
524 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
525 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
526
527 cmd_ctx->llm->data_size = size;
528 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
529
530 return buf_size;
531
532 error:
533 return ret;
534 }
535
536 /*
537 * Update the kernel poll set of all channel fd available over all tracing
538 * session. Add the wakeup pipe at the end of the set.
539 */
540 static int update_kernel_poll(struct lttng_poll_event *events)
541 {
542 int ret;
543 struct ltt_session *session;
544 struct ltt_kernel_channel *channel;
545
546 DBG("Updating kernel poll set");
547
548 session_lock_list();
549 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
550 session_lock(session);
551 if (session->kernel_session == NULL) {
552 session_unlock(session);
553 continue;
554 }
555
556 cds_list_for_each_entry(channel,
557 &session->kernel_session->channel_list.head, list) {
558 /* Add channel fd to the kernel poll set */
559 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
560 if (ret < 0) {
561 session_unlock(session);
562 goto error;
563 }
564 DBG("Channel fd %d added to kernel set", channel->fd);
565 }
566 session_unlock(session);
567 }
568 session_unlock_list();
569
570 return 0;
571
572 error:
573 session_unlock_list();
574 return -1;
575 }
576
577 /*
578 * Find the channel fd from 'fd' over all tracing session. When found, check
579 * for new channel stream and send those stream fds to the kernel consumer.
580 *
581 * Useful for CPU hotplug feature.
582 */
583 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
584 {
585 int ret = 0;
586 struct ltt_session *session;
587 struct ltt_kernel_session *ksess;
588 struct ltt_kernel_channel *channel;
589
590 DBG("Updating kernel streams for channel fd %d", fd);
591
592 session_lock_list();
593 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
594 session_lock(session);
595 if (session->kernel_session == NULL) {
596 session_unlock(session);
597 continue;
598 }
599 ksess = session->kernel_session;
600
601 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
602 if (channel->fd == fd) {
603 DBG("Channel found, updating kernel streams");
604 ret = kernel_open_channel_stream(channel);
605 if (ret < 0) {
606 goto error;
607 }
608
609 /*
610 * Have we already sent fds to the consumer? If yes, it means
611 * that tracing is started so it is safe to send our updated
612 * stream fds.
613 */
614 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
615 struct lttng_ht_iter iter;
616 struct consumer_socket *socket;
617
618
619 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
620 &iter.iter, socket, node.node) {
621 /* Code flow error */
622 assert(socket->fd >= 0);
623
624 pthread_mutex_lock(socket->lock);
625 ret = kernel_consumer_send_channel_stream(socket->fd,
626 channel, ksess);
627 pthread_mutex_unlock(socket->lock);
628 if (ret < 0) {
629 goto error;
630 }
631 }
632 }
633 goto error;
634 }
635 }
636 session_unlock(session);
637 }
638 session_unlock_list();
639 return ret;
640
641 error:
642 session_unlock(session);
643 session_unlock_list();
644 return ret;
645 }
646
647 /*
648 * For each tracing session, update newly registered apps.
649 */
650 static void update_ust_app(int app_sock)
651 {
652 struct ltt_session *sess, *stmp;
653
654 session_lock_list();
655
656 /* For all tracing session(s) */
657 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
658 session_lock(sess);
659 if (sess->ust_session) {
660 ust_app_global_update(sess->ust_session, app_sock);
661 }
662 session_unlock(sess);
663 }
664
665 session_unlock_list();
666 }
667
668 /*
669 * This thread manage event coming from the kernel.
670 *
671 * Features supported in this thread:
672 * -) CPU Hotplug
673 */
674 static void *thread_manage_kernel(void *data)
675 {
676 int ret, i, pollfd, update_poll_flag = 1, err = -1;
677 uint32_t revents, nb_fd;
678 char tmp;
679 struct lttng_poll_event events;
680
681 DBG("Thread manage kernel started");
682
683 health_code_update(&health_thread_kernel);
684
685 ret = create_thread_poll_set(&events, 2);
686 if (ret < 0) {
687 goto error_poll_create;
688 }
689
690 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
691 if (ret < 0) {
692 goto error;
693 }
694
695 while (1) {
696 health_code_update(&health_thread_kernel);
697
698 if (update_poll_flag == 1) {
699 /*
700 * Reset number of fd in the poll set. Always 2 since there is the thread
701 * quit pipe and the kernel pipe.
702 */
703 events.nb_fd = 2;
704
705 ret = update_kernel_poll(&events);
706 if (ret < 0) {
707 goto error;
708 }
709 update_poll_flag = 0;
710 }
711
712 nb_fd = LTTNG_POLL_GETNB(&events);
713
714 DBG("Thread kernel polling on %d fds", nb_fd);
715
716 /* Zeroed the poll events */
717 lttng_poll_reset(&events);
718
719 /* Poll infinite value of time */
720 restart:
721 health_poll_update(&health_thread_kernel);
722 ret = lttng_poll_wait(&events, -1);
723 health_poll_update(&health_thread_kernel);
724 if (ret < 0) {
725 /*
726 * Restart interrupted system call.
727 */
728 if (errno == EINTR) {
729 goto restart;
730 }
731 goto error;
732 } else if (ret == 0) {
733 /* Should not happen since timeout is infinite */
734 ERR("Return value of poll is 0 with an infinite timeout.\n"
735 "This should not have happened! Continuing...");
736 continue;
737 }
738
739 for (i = 0; i < nb_fd; i++) {
740 /* Fetch once the poll data */
741 revents = LTTNG_POLL_GETEV(&events, i);
742 pollfd = LTTNG_POLL_GETFD(&events, i);
743
744 health_code_update(&health_thread_kernel);
745
746 /* Thread quit pipe has been closed. Killing thread. */
747 ret = check_thread_quit_pipe(pollfd, revents);
748 if (ret) {
749 err = 0;
750 goto exit;
751 }
752
753 /* Check for data on kernel pipe */
754 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
755 ret = read(kernel_poll_pipe[0], &tmp, 1);
756 update_poll_flag = 1;
757 continue;
758 } else {
759 /*
760 * New CPU detected by the kernel. Adding kernel stream to
761 * kernel session and updating the kernel consumer
762 */
763 if (revents & LPOLLIN) {
764 ret = update_kernel_stream(&kconsumer_data, pollfd);
765 if (ret < 0) {
766 continue;
767 }
768 break;
769 /*
770 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
771 * and unregister kernel stream at this point.
772 */
773 }
774 }
775 }
776 }
777
778 exit:
779 error:
780 lttng_poll_clean(&events);
781 error_poll_create:
782 if (err) {
783 health_error(&health_thread_kernel);
784 ERR("Health error occurred in %s", __func__);
785 }
786 health_exit(&health_thread_kernel);
787 DBG("Kernel thread dying");
788 return NULL;
789 }
790
791 /*
792 * This thread manage the consumer error sent back to the session daemon.
793 */
794 static void *thread_manage_consumer(void *data)
795 {
796 int sock = -1, i, ret, pollfd, err = -1;
797 uint32_t revents, nb_fd;
798 enum lttcomm_return_code code;
799 struct lttng_poll_event events;
800 struct consumer_data *consumer_data = data;
801
802 DBG("[thread] Manage consumer started");
803
804 health_code_update(&consumer_data->health);
805
806 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
807 if (ret < 0) {
808 goto error_listen;
809 }
810
811 /*
812 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
813 * Nothing more will be added to this poll set.
814 */
815 ret = create_thread_poll_set(&events, 2);
816 if (ret < 0) {
817 goto error_poll;
818 }
819
820 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
821 if (ret < 0) {
822 goto error;
823 }
824
825 nb_fd = LTTNG_POLL_GETNB(&events);
826
827 health_code_update(&consumer_data->health);
828
829 /* Inifinite blocking call, waiting for transmission */
830 restart:
831 health_poll_update(&consumer_data->health);
832 ret = lttng_poll_wait(&events, -1);
833 health_poll_update(&consumer_data->health);
834 if (ret < 0) {
835 /*
836 * Restart interrupted system call.
837 */
838 if (errno == EINTR) {
839 goto restart;
840 }
841 goto error;
842 }
843
844 for (i = 0; i < nb_fd; i++) {
845 /* Fetch once the poll data */
846 revents = LTTNG_POLL_GETEV(&events, i);
847 pollfd = LTTNG_POLL_GETFD(&events, i);
848
849 health_code_update(&consumer_data->health);
850
851 /* Thread quit pipe has been closed. Killing thread. */
852 ret = check_thread_quit_pipe(pollfd, revents);
853 if (ret) {
854 err = 0;
855 goto exit;
856 }
857
858 /* Event on the registration socket */
859 if (pollfd == consumer_data->err_sock) {
860 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
861 ERR("consumer err socket poll error");
862 goto error;
863 }
864 }
865 }
866
867 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
868 if (sock < 0) {
869 goto error;
870 }
871
872 health_code_update(&consumer_data->health);
873
874 DBG2("Receiving code from consumer err_sock");
875
876 /* Getting status code from kconsumerd */
877 ret = lttcomm_recv_unix_sock(sock, &code,
878 sizeof(enum lttcomm_return_code));
879 if (ret <= 0) {
880 goto error;
881 }
882
883 health_code_update(&consumer_data->health);
884
885 if (code == CONSUMERD_COMMAND_SOCK_READY) {
886 consumer_data->cmd_sock =
887 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
888 if (consumer_data->cmd_sock < 0) {
889 sem_post(&consumer_data->sem);
890 PERROR("consumer connect");
891 goto error;
892 }
893 /* Signal condition to tell that the kconsumerd is ready */
894 sem_post(&consumer_data->sem);
895 DBG("consumer command socket ready");
896 } else {
897 ERR("consumer error when waiting for SOCK_READY : %s",
898 lttcomm_get_readable_code(-code));
899 goto error;
900 }
901
902 /* Remove the kconsumerd error sock since we've established a connexion */
903 ret = lttng_poll_del(&events, consumer_data->err_sock);
904 if (ret < 0) {
905 goto error;
906 }
907
908 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
909 if (ret < 0) {
910 goto error;
911 }
912
913 health_code_update(&consumer_data->health);
914
915 /* Update number of fd */
916 nb_fd = LTTNG_POLL_GETNB(&events);
917
918 /* Inifinite blocking call, waiting for transmission */
919 restart_poll:
920 health_poll_update(&consumer_data->health);
921 ret = lttng_poll_wait(&events, -1);
922 health_poll_update(&consumer_data->health);
923 if (ret < 0) {
924 /*
925 * Restart interrupted system call.
926 */
927 if (errno == EINTR) {
928 goto restart_poll;
929 }
930 goto error;
931 }
932
933 for (i = 0; i < nb_fd; i++) {
934 /* Fetch once the poll data */
935 revents = LTTNG_POLL_GETEV(&events, i);
936 pollfd = LTTNG_POLL_GETFD(&events, i);
937
938 health_code_update(&consumer_data->health);
939
940 /* Thread quit pipe has been closed. Killing thread. */
941 ret = check_thread_quit_pipe(pollfd, revents);
942 if (ret) {
943 err = 0;
944 goto exit;
945 }
946
947 /* Event on the kconsumerd socket */
948 if (pollfd == sock) {
949 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
950 ERR("consumer err socket second poll error");
951 goto error;
952 }
953 }
954 }
955
956 health_code_update(&consumer_data->health);
957
958 /* Wait for any kconsumerd error */
959 ret = lttcomm_recv_unix_sock(sock, &code,
960 sizeof(enum lttcomm_return_code));
961 if (ret <= 0) {
962 ERR("consumer closed the command socket");
963 goto error;
964 }
965
966 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
967
968 exit:
969 error:
970 /* Immediately set the consumerd state to stopped */
971 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
972 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
973 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
974 consumer_data->type == LTTNG_CONSUMER32_UST) {
975 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
976 } else {
977 /* Code flow error... */
978 assert(0);
979 }
980
981 if (consumer_data->err_sock >= 0) {
982 ret = close(consumer_data->err_sock);
983 if (ret) {
984 PERROR("close");
985 }
986 }
987 if (consumer_data->cmd_sock >= 0) {
988 ret = close(consumer_data->cmd_sock);
989 if (ret) {
990 PERROR("close");
991 }
992 }
993 if (sock >= 0) {
994 ret = close(sock);
995 if (ret) {
996 PERROR("close");
997 }
998 }
999
1000 unlink(consumer_data->err_unix_sock_path);
1001 unlink(consumer_data->cmd_unix_sock_path);
1002 consumer_data->pid = 0;
1003
1004 lttng_poll_clean(&events);
1005 error_poll:
1006 error_listen:
1007 if (err) {
1008 health_error(&consumer_data->health);
1009 ERR("Health error occurred in %s", __func__);
1010 }
1011 health_exit(&consumer_data->health);
1012 DBG("consumer thread cleanup completed");
1013
1014 return NULL;
1015 }
1016
1017 /*
1018 * This thread manage application communication.
1019 */
1020 static void *thread_manage_apps(void *data)
1021 {
1022 int i, ret, pollfd, err = -1;
1023 uint32_t revents, nb_fd;
1024 struct ust_command ust_cmd;
1025 struct lttng_poll_event events;
1026
1027 DBG("[thread] Manage application started");
1028
1029 rcu_register_thread();
1030 rcu_thread_online();
1031
1032 health_code_update(&health_thread_app_manage);
1033
1034 ret = create_thread_poll_set(&events, 2);
1035 if (ret < 0) {
1036 goto error_poll_create;
1037 }
1038
1039 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1040 if (ret < 0) {
1041 goto error;
1042 }
1043
1044 health_code_update(&health_thread_app_manage);
1045
1046 while (1) {
1047 /* Zeroed the events structure */
1048 lttng_poll_reset(&events);
1049
1050 nb_fd = LTTNG_POLL_GETNB(&events);
1051
1052 DBG("Apps thread polling on %d fds", nb_fd);
1053
1054 /* Inifinite blocking call, waiting for transmission */
1055 restart:
1056 health_poll_update(&health_thread_app_manage);
1057 ret = lttng_poll_wait(&events, -1);
1058 health_poll_update(&health_thread_app_manage);
1059 if (ret < 0) {
1060 /*
1061 * Restart interrupted system call.
1062 */
1063 if (errno == EINTR) {
1064 goto restart;
1065 }
1066 goto error;
1067 }
1068
1069 for (i = 0; i < nb_fd; i++) {
1070 /* Fetch once the poll data */
1071 revents = LTTNG_POLL_GETEV(&events, i);
1072 pollfd = LTTNG_POLL_GETFD(&events, i);
1073
1074 health_code_update(&health_thread_app_manage);
1075
1076 /* Thread quit pipe has been closed. Killing thread. */
1077 ret = check_thread_quit_pipe(pollfd, revents);
1078 if (ret) {
1079 err = 0;
1080 goto exit;
1081 }
1082
1083 /* Inspect the apps cmd pipe */
1084 if (pollfd == apps_cmd_pipe[0]) {
1085 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1086 ERR("Apps command pipe error");
1087 goto error;
1088 } else if (revents & LPOLLIN) {
1089 /* Empty pipe */
1090 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1091 if (ret < 0 || ret < sizeof(ust_cmd)) {
1092 PERROR("read apps cmd pipe");
1093 goto error;
1094 }
1095
1096 health_code_update(&health_thread_app_manage);
1097
1098 /* Register applicaton to the session daemon */
1099 ret = ust_app_register(&ust_cmd.reg_msg,
1100 ust_cmd.sock);
1101 if (ret == -ENOMEM) {
1102 goto error;
1103 } else if (ret < 0) {
1104 break;
1105 }
1106
1107 health_code_update(&health_thread_app_manage);
1108
1109 /*
1110 * Validate UST version compatibility.
1111 */
1112 ret = ust_app_validate_version(ust_cmd.sock);
1113 if (ret >= 0) {
1114 /*
1115 * Add channel(s) and event(s) to newly registered apps
1116 * from lttng global UST domain.
1117 */
1118 update_ust_app(ust_cmd.sock);
1119 }
1120
1121 health_code_update(&health_thread_app_manage);
1122
1123 ret = ust_app_register_done(ust_cmd.sock);
1124 if (ret < 0) {
1125 /*
1126 * If the registration is not possible, we simply
1127 * unregister the apps and continue
1128 */
1129 ust_app_unregister(ust_cmd.sock);
1130 } else {
1131 /*
1132 * We just need here to monitor the close of the UST
1133 * socket and poll set monitor those by default.
1134 * Listen on POLLIN (even if we never expect any
1135 * data) to ensure that hangup wakes us.
1136 */
1137 ret = lttng_poll_add(&events, ust_cmd.sock, LPOLLIN);
1138 if (ret < 0) {
1139 goto error;
1140 }
1141
1142 DBG("Apps with sock %d added to poll set",
1143 ust_cmd.sock);
1144 }
1145
1146 health_code_update(&health_thread_app_manage);
1147
1148 break;
1149 }
1150 } else {
1151 /*
1152 * At this point, we know that a registered application made
1153 * the event at poll_wait.
1154 */
1155 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1156 /* Removing from the poll set */
1157 ret = lttng_poll_del(&events, pollfd);
1158 if (ret < 0) {
1159 goto error;
1160 }
1161
1162 /* Socket closed on remote end. */
1163 ust_app_unregister(pollfd);
1164 break;
1165 }
1166 }
1167
1168 health_code_update(&health_thread_app_manage);
1169 }
1170 }
1171
1172 exit:
1173 error:
1174 lttng_poll_clean(&events);
1175 error_poll_create:
1176 if (err) {
1177 health_error(&health_thread_app_manage);
1178 ERR("Health error occurred in %s", __func__);
1179 }
1180 health_exit(&health_thread_app_manage);
1181 DBG("Application communication apps thread cleanup complete");
1182 rcu_thread_offline();
1183 rcu_unregister_thread();
1184 return NULL;
1185 }
1186
1187 /*
1188 * Dispatch request from the registration threads to the application
1189 * communication thread.
1190 */
1191 static void *thread_dispatch_ust_registration(void *data)
1192 {
1193 int ret;
1194 struct cds_wfq_node *node;
1195 struct ust_command *ust_cmd = NULL;
1196
1197 DBG("[thread] Dispatch UST command started");
1198
1199 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1200 /* Atomically prepare the queue futex */
1201 futex_nto1_prepare(&ust_cmd_queue.futex);
1202
1203 do {
1204 /* Dequeue command for registration */
1205 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1206 if (node == NULL) {
1207 DBG("Woken up but nothing in the UST command queue");
1208 /* Continue thread execution */
1209 break;
1210 }
1211
1212 ust_cmd = caa_container_of(node, struct ust_command, node);
1213
1214 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1215 " gid:%d sock:%d name:%s (version %d.%d)",
1216 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1217 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1218 ust_cmd->sock, ust_cmd->reg_msg.name,
1219 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1220 /*
1221 * Inform apps thread of the new application registration. This
1222 * call is blocking so we can be assured that the data will be read
1223 * at some point in time or wait to the end of the world :)
1224 */
1225 ret = write(apps_cmd_pipe[1], ust_cmd,
1226 sizeof(struct ust_command));
1227 if (ret < 0) {
1228 PERROR("write apps cmd pipe");
1229 if (errno == EBADF) {
1230 /*
1231 * We can't inform the application thread to process
1232 * registration. We will exit or else application
1233 * registration will not occur and tracing will never
1234 * start.
1235 */
1236 goto error;
1237 }
1238 }
1239 free(ust_cmd);
1240 } while (node != NULL);
1241
1242 /* Futex wait on queue. Blocking call on futex() */
1243 futex_nto1_wait(&ust_cmd_queue.futex);
1244 }
1245
1246 error:
1247 DBG("Dispatch thread dying");
1248 return NULL;
1249 }
1250
1251 /*
1252 * This thread manage application registration.
1253 */
1254 static void *thread_registration_apps(void *data)
1255 {
1256 int sock = -1, i, ret, pollfd, err = -1;
1257 uint32_t revents, nb_fd;
1258 struct lttng_poll_event events;
1259 /*
1260 * Get allocated in this thread, enqueued to a global queue, dequeued and
1261 * freed in the manage apps thread.
1262 */
1263 struct ust_command *ust_cmd = NULL;
1264
1265 DBG("[thread] Manage application registration started");
1266
1267 ret = lttcomm_listen_unix_sock(apps_sock);
1268 if (ret < 0) {
1269 goto error_listen;
1270 }
1271
1272 /*
1273 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1274 * more will be added to this poll set.
1275 */
1276 ret = create_thread_poll_set(&events, 2);
1277 if (ret < 0) {
1278 goto error_create_poll;
1279 }
1280
1281 /* Add the application registration socket */
1282 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1283 if (ret < 0) {
1284 goto error_poll_add;
1285 }
1286
1287 /* Notify all applications to register */
1288 ret = notify_ust_apps(1);
1289 if (ret < 0) {
1290 ERR("Failed to notify applications or create the wait shared memory.\n"
1291 "Execution continues but there might be problem for already\n"
1292 "running applications that wishes to register.");
1293 }
1294
1295 while (1) {
1296 DBG("Accepting application registration");
1297
1298 nb_fd = LTTNG_POLL_GETNB(&events);
1299
1300 /* Inifinite blocking call, waiting for transmission */
1301 restart:
1302 health_poll_update(&health_thread_app_reg);
1303 ret = lttng_poll_wait(&events, -1);
1304 health_poll_update(&health_thread_app_reg);
1305 if (ret < 0) {
1306 /*
1307 * Restart interrupted system call.
1308 */
1309 if (errno == EINTR) {
1310 goto restart;
1311 }
1312 goto error;
1313 }
1314
1315 for (i = 0; i < nb_fd; i++) {
1316 health_code_update(&health_thread_app_reg);
1317
1318 /* Fetch once the poll data */
1319 revents = LTTNG_POLL_GETEV(&events, i);
1320 pollfd = LTTNG_POLL_GETFD(&events, i);
1321
1322 /* Thread quit pipe has been closed. Killing thread. */
1323 ret = check_thread_quit_pipe(pollfd, revents);
1324 if (ret) {
1325 err = 0;
1326 goto exit;
1327 }
1328
1329 /* Event on the registration socket */
1330 if (pollfd == apps_sock) {
1331 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1332 ERR("Register apps socket poll error");
1333 goto error;
1334 } else if (revents & LPOLLIN) {
1335 sock = lttcomm_accept_unix_sock(apps_sock);
1336 if (sock < 0) {
1337 goto error;
1338 }
1339
1340 /* Create UST registration command for enqueuing */
1341 ust_cmd = zmalloc(sizeof(struct ust_command));
1342 if (ust_cmd == NULL) {
1343 PERROR("ust command zmalloc");
1344 goto error;
1345 }
1346
1347 /*
1348 * Using message-based transmissions to ensure we don't
1349 * have to deal with partially received messages.
1350 */
1351 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1352 if (ret < 0) {
1353 ERR("Exhausted file descriptors allowed for applications.");
1354 free(ust_cmd);
1355 ret = close(sock);
1356 if (ret) {
1357 PERROR("close");
1358 }
1359 sock = -1;
1360 continue;
1361 }
1362 health_code_update(&health_thread_app_reg);
1363 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1364 sizeof(struct ust_register_msg));
1365 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1366 if (ret < 0) {
1367 PERROR("lttcomm_recv_unix_sock register apps");
1368 } else {
1369 ERR("Wrong size received on apps register");
1370 }
1371 free(ust_cmd);
1372 ret = close(sock);
1373 if (ret) {
1374 PERROR("close");
1375 }
1376 lttng_fd_put(LTTNG_FD_APPS, 1);
1377 sock = -1;
1378 continue;
1379 }
1380 health_code_update(&health_thread_app_reg);
1381
1382 ust_cmd->sock = sock;
1383 sock = -1;
1384
1385 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1386 " gid:%d sock:%d name:%s (version %d.%d)",
1387 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1388 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1389 ust_cmd->sock, ust_cmd->reg_msg.name,
1390 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1391
1392 /*
1393 * Lock free enqueue the registration request. The red pill
1394 * has been taken! This apps will be part of the *system*.
1395 */
1396 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1397
1398 /*
1399 * Wake the registration queue futex. Implicit memory
1400 * barrier with the exchange in cds_wfq_enqueue.
1401 */
1402 futex_nto1_wake(&ust_cmd_queue.futex);
1403 }
1404 }
1405 }
1406 }
1407
1408 exit:
1409 error:
1410 if (err) {
1411 health_error(&health_thread_app_reg);
1412 ERR("Health error occurred in %s", __func__);
1413 }
1414 health_exit(&health_thread_app_reg);
1415
1416 /* Notify that the registration thread is gone */
1417 notify_ust_apps(0);
1418
1419 if (apps_sock >= 0) {
1420 ret = close(apps_sock);
1421 if (ret) {
1422 PERROR("close");
1423 }
1424 }
1425 if (sock >= 0) {
1426 ret = close(sock);
1427 if (ret) {
1428 PERROR("close");
1429 }
1430 lttng_fd_put(LTTNG_FD_APPS, 1);
1431 }
1432 unlink(apps_unix_sock_path);
1433
1434 error_poll_add:
1435 lttng_poll_clean(&events);
1436 error_listen:
1437 error_create_poll:
1438 DBG("UST Registration thread cleanup complete");
1439
1440 return NULL;
1441 }
1442
1443 /*
1444 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1445 * exec or it will fails.
1446 */
1447 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1448 {
1449 int ret;
1450 struct timespec timeout;
1451
1452 timeout.tv_sec = DEFAULT_SEM_WAIT_TIMEOUT;
1453 timeout.tv_nsec = 0;
1454
1455 /* Setup semaphore */
1456 ret = sem_init(&consumer_data->sem, 0, 0);
1457 if (ret < 0) {
1458 PERROR("sem_init consumer semaphore");
1459 goto error;
1460 }
1461
1462 ret = pthread_create(&consumer_data->thread, NULL,
1463 thread_manage_consumer, consumer_data);
1464 if (ret != 0) {
1465 PERROR("pthread_create consumer");
1466 ret = -1;
1467 goto error;
1468 }
1469
1470 /* Get time for sem_timedwait absolute timeout */
1471 ret = clock_gettime(CLOCK_REALTIME, &timeout);
1472 if (ret < 0) {
1473 PERROR("clock_gettime spawn consumer");
1474 /* Infinite wait for the kconsumerd thread to be ready */
1475 ret = sem_wait(&consumer_data->sem);
1476 } else {
1477 /* Normal timeout if the gettime was successful */
1478 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1479 ret = sem_timedwait(&consumer_data->sem, &timeout);
1480 }
1481
1482 if (ret < 0) {
1483 if (errno == ETIMEDOUT) {
1484 /*
1485 * Call has timed out so we kill the kconsumerd_thread and return
1486 * an error.
1487 */
1488 ERR("The consumer thread was never ready. Killing it");
1489 ret = pthread_cancel(consumer_data->thread);
1490 if (ret < 0) {
1491 PERROR("pthread_cancel consumer thread");
1492 }
1493 } else {
1494 PERROR("semaphore wait failed consumer thread");
1495 }
1496 goto error;
1497 }
1498
1499 pthread_mutex_lock(&consumer_data->pid_mutex);
1500 if (consumer_data->pid == 0) {
1501 ERR("Kconsumerd did not start");
1502 pthread_mutex_unlock(&consumer_data->pid_mutex);
1503 goto error;
1504 }
1505 pthread_mutex_unlock(&consumer_data->pid_mutex);
1506
1507 return 0;
1508
1509 error:
1510 return ret;
1511 }
1512
1513 /*
1514 * Join consumer thread
1515 */
1516 static int join_consumer_thread(struct consumer_data *consumer_data)
1517 {
1518 void *status;
1519 int ret;
1520
1521 /* Consumer pid must be a real one. */
1522 if (consumer_data->pid > 0) {
1523 ret = kill(consumer_data->pid, SIGTERM);
1524 if (ret) {
1525 ERR("Error killing consumer daemon");
1526 return ret;
1527 }
1528 return pthread_join(consumer_data->thread, &status);
1529 } else {
1530 return 0;
1531 }
1532 }
1533
1534 /*
1535 * Fork and exec a consumer daemon (consumerd).
1536 *
1537 * Return pid if successful else -1.
1538 */
1539 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1540 {
1541 int ret;
1542 pid_t pid;
1543 const char *consumer_to_use;
1544 const char *verbosity;
1545 struct stat st;
1546
1547 DBG("Spawning consumerd");
1548
1549 pid = fork();
1550 if (pid == 0) {
1551 /*
1552 * Exec consumerd.
1553 */
1554 if (opt_verbose_consumer) {
1555 verbosity = "--verbose";
1556 } else {
1557 verbosity = "--quiet";
1558 }
1559 switch (consumer_data->type) {
1560 case LTTNG_CONSUMER_KERNEL:
1561 /*
1562 * Find out which consumerd to execute. We will first try the
1563 * 64-bit path, then the sessiond's installation directory, and
1564 * fallback on the 32-bit one,
1565 */
1566 DBG3("Looking for a kernel consumer at these locations:");
1567 DBG3(" 1) %s", consumerd64_bin);
1568 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1569 DBG3(" 3) %s", consumerd32_bin);
1570 if (stat(consumerd64_bin, &st) == 0) {
1571 DBG3("Found location #1");
1572 consumer_to_use = consumerd64_bin;
1573 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1574 DBG3("Found location #2");
1575 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1576 } else if (stat(consumerd32_bin, &st) == 0) {
1577 DBG3("Found location #3");
1578 consumer_to_use = consumerd32_bin;
1579 } else {
1580 DBG("Could not find any valid consumerd executable");
1581 break;
1582 }
1583 DBG("Using kernel consumer at: %s", consumer_to_use);
1584 execl(consumer_to_use,
1585 "lttng-consumerd", verbosity, "-k",
1586 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1587 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1588 NULL);
1589 break;
1590 case LTTNG_CONSUMER64_UST:
1591 {
1592 char *tmpnew = NULL;
1593
1594 if (consumerd64_libdir[0] != '\0') {
1595 char *tmp;
1596 size_t tmplen;
1597
1598 tmp = getenv("LD_LIBRARY_PATH");
1599 if (!tmp) {
1600 tmp = "";
1601 }
1602 tmplen = strlen("LD_LIBRARY_PATH=")
1603 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1604 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1605 if (!tmpnew) {
1606 ret = -ENOMEM;
1607 goto error;
1608 }
1609 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1610 strcat(tmpnew, consumerd64_libdir);
1611 if (tmp[0] != '\0') {
1612 strcat(tmpnew, ":");
1613 strcat(tmpnew, tmp);
1614 }
1615 ret = putenv(tmpnew);
1616 if (ret) {
1617 ret = -errno;
1618 goto error;
1619 }
1620 }
1621 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1622 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1623 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1624 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1625 NULL);
1626 if (consumerd64_libdir[0] != '\0') {
1627 free(tmpnew);
1628 }
1629 if (ret) {
1630 goto error;
1631 }
1632 break;
1633 }
1634 case LTTNG_CONSUMER32_UST:
1635 {
1636 char *tmpnew = NULL;
1637
1638 if (consumerd32_libdir[0] != '\0') {
1639 char *tmp;
1640 size_t tmplen;
1641
1642 tmp = getenv("LD_LIBRARY_PATH");
1643 if (!tmp) {
1644 tmp = "";
1645 }
1646 tmplen = strlen("LD_LIBRARY_PATH=")
1647 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1648 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1649 if (!tmpnew) {
1650 ret = -ENOMEM;
1651 goto error;
1652 }
1653 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1654 strcat(tmpnew, consumerd32_libdir);
1655 if (tmp[0] != '\0') {
1656 strcat(tmpnew, ":");
1657 strcat(tmpnew, tmp);
1658 }
1659 ret = putenv(tmpnew);
1660 if (ret) {
1661 ret = -errno;
1662 goto error;
1663 }
1664 }
1665 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1666 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1667 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1668 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1669 NULL);
1670 if (consumerd32_libdir[0] != '\0') {
1671 free(tmpnew);
1672 }
1673 if (ret) {
1674 goto error;
1675 }
1676 break;
1677 }
1678 default:
1679 PERROR("unknown consumer type");
1680 exit(EXIT_FAILURE);
1681 }
1682 if (errno != 0) {
1683 PERROR("kernel start consumer exec");
1684 }
1685 exit(EXIT_FAILURE);
1686 } else if (pid > 0) {
1687 ret = pid;
1688 } else {
1689 PERROR("start consumer fork");
1690 ret = -errno;
1691 }
1692 error:
1693 return ret;
1694 }
1695
1696 /*
1697 * Spawn the consumerd daemon and session daemon thread.
1698 */
1699 static int start_consumerd(struct consumer_data *consumer_data)
1700 {
1701 int ret;
1702
1703 pthread_mutex_lock(&consumer_data->pid_mutex);
1704 if (consumer_data->pid != 0) {
1705 pthread_mutex_unlock(&consumer_data->pid_mutex);
1706 goto end;
1707 }
1708
1709 ret = spawn_consumerd(consumer_data);
1710 if (ret < 0) {
1711 ERR("Spawning consumerd failed");
1712 pthread_mutex_unlock(&consumer_data->pid_mutex);
1713 goto error;
1714 }
1715
1716 /* Setting up the consumer_data pid */
1717 consumer_data->pid = ret;
1718 DBG2("Consumer pid %d", consumer_data->pid);
1719 pthread_mutex_unlock(&consumer_data->pid_mutex);
1720
1721 DBG2("Spawning consumer control thread");
1722 ret = spawn_consumer_thread(consumer_data);
1723 if (ret < 0) {
1724 ERR("Fatal error spawning consumer control thread");
1725 goto error;
1726 }
1727
1728 end:
1729 return 0;
1730
1731 error:
1732 return ret;
1733 }
1734
1735 /*
1736 * Compute health status of each consumer. If one of them is zero (bad
1737 * state), we return 0.
1738 */
1739 static int check_consumer_health(void)
1740 {
1741 int ret;
1742
1743 ret = health_check_state(&kconsumer_data.health) &&
1744 health_check_state(&ustconsumer32_data.health) &&
1745 health_check_state(&ustconsumer64_data.health);
1746
1747 DBG3("Health consumer check %d", ret);
1748
1749 return ret;
1750 }
1751
1752 /*
1753 * Setup necessary data for kernel tracer action.
1754 */
1755 static int init_kernel_tracer(void)
1756 {
1757 int ret;
1758
1759 /* Modprobe lttng kernel modules */
1760 ret = modprobe_lttng_control();
1761 if (ret < 0) {
1762 goto error;
1763 }
1764
1765 /* Open debugfs lttng */
1766 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1767 if (kernel_tracer_fd < 0) {
1768 DBG("Failed to open %s", module_proc_lttng);
1769 ret = -1;
1770 goto error_open;
1771 }
1772
1773 /* Validate kernel version */
1774 ret = kernel_validate_version(kernel_tracer_fd);
1775 if (ret < 0) {
1776 goto error_version;
1777 }
1778
1779 ret = modprobe_lttng_data();
1780 if (ret < 0) {
1781 goto error_modules;
1782 }
1783
1784 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1785 return 0;
1786
1787 error_version:
1788 modprobe_remove_lttng_control();
1789 ret = close(kernel_tracer_fd);
1790 if (ret) {
1791 PERROR("close");
1792 }
1793 kernel_tracer_fd = -1;
1794 return LTTCOMM_KERN_VERSION;
1795
1796 error_modules:
1797 ret = close(kernel_tracer_fd);
1798 if (ret) {
1799 PERROR("close");
1800 }
1801
1802 error_open:
1803 modprobe_remove_lttng_control();
1804
1805 error:
1806 WARN("No kernel tracer available");
1807 kernel_tracer_fd = -1;
1808 if (!is_root) {
1809 return LTTCOMM_NEED_ROOT_SESSIOND;
1810 } else {
1811 return LTTCOMM_KERN_NA;
1812 }
1813 }
1814
1815
1816 /*
1817 * Copy consumer output from the tracing session to the domain session. The
1818 * function also applies the right modification on a per domain basis for the
1819 * trace files destination directory.
1820 */
1821 static int copy_session_consumer(int domain, struct ltt_session *session)
1822 {
1823 int ret;
1824 const char *dir_name;
1825 struct consumer_output *consumer;
1826
1827 assert(session);
1828 assert(session->consumer);
1829
1830 switch (domain) {
1831 case LTTNG_DOMAIN_KERNEL:
1832 DBG3("Copying tracing session consumer output in kernel session");
1833 session->kernel_session->consumer =
1834 consumer_copy_output(session->consumer);
1835 /* Ease our life a bit for the next part */
1836 consumer = session->kernel_session->consumer;
1837 dir_name = DEFAULT_KERNEL_TRACE_DIR;
1838 break;
1839 case LTTNG_DOMAIN_UST:
1840 DBG3("Copying tracing session consumer output in UST session");
1841 session->ust_session->consumer =
1842 consumer_copy_output(session->consumer);
1843 /* Ease our life a bit for the next part */
1844 consumer = session->ust_session->consumer;
1845 dir_name = DEFAULT_UST_TRACE_DIR;
1846 break;
1847 default:
1848 ret = LTTCOMM_UNKNOWN_DOMAIN;
1849 goto error;
1850 }
1851
1852 ret = consumer_set_subdir(session->consumer, session->name);
1853 if (ret < 0) {
1854 ret = LTTCOMM_FATAL;
1855 goto error;
1856 }
1857
1858 /* Append correct directory to subdir */
1859 strncat(consumer->subdir, dir_name,
1860 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
1861 DBG3("Copy session consumer subdir %s", consumer->subdir);
1862
1863 ret = LTTCOMM_OK;
1864
1865 error:
1866 return ret;
1867 }
1868
1869 /*
1870 * Create an UST session and add it to the session ust list.
1871 */
1872 static int create_ust_session(struct ltt_session *session,
1873 struct lttng_domain *domain)
1874 {
1875 int ret;
1876 struct ltt_ust_session *lus = NULL;
1877
1878 assert(session);
1879 assert(domain);
1880 assert(session->consumer);
1881
1882 switch (domain->type) {
1883 case LTTNG_DOMAIN_UST:
1884 break;
1885 default:
1886 ERR("Unknown UST domain on create session %d", domain->type);
1887 ret = LTTCOMM_UNKNOWN_DOMAIN;
1888 goto error;
1889 }
1890
1891 DBG("Creating UST session");
1892
1893 lus = trace_ust_create_session(session->path, session->id, domain);
1894 if (lus == NULL) {
1895 ret = LTTCOMM_UST_SESS_FAIL;
1896 goto error;
1897 }
1898
1899 lus->uid = session->uid;
1900 lus->gid = session->gid;
1901 session->ust_session = lus;
1902
1903 /* Copy session output to the newly created UST session */
1904 ret = copy_session_consumer(domain->type, session);
1905 if (ret != LTTCOMM_OK) {
1906 goto error;
1907 }
1908
1909 return LTTCOMM_OK;
1910
1911 error:
1912 free(lus);
1913 session->ust_session = NULL;
1914 return ret;
1915 }
1916
1917 /*
1918 * Create a kernel tracer session then create the default channel.
1919 */
1920 static int create_kernel_session(struct ltt_session *session)
1921 {
1922 int ret;
1923
1924 DBG("Creating kernel session");
1925
1926 ret = kernel_create_session(session, kernel_tracer_fd);
1927 if (ret < 0) {
1928 ret = LTTCOMM_KERN_SESS_FAIL;
1929 goto error;
1930 }
1931
1932 /* Code flow safety */
1933 assert(session->kernel_session);
1934
1935 /* Copy session output to the newly created Kernel session */
1936 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
1937 if (ret != LTTCOMM_OK) {
1938 goto error;
1939 }
1940
1941 /* Create directory(ies) on local filesystem. */
1942 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
1943 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
1944 ret = run_as_mkdir_recursive(
1945 session->kernel_session->consumer->dst.trace_path,
1946 S_IRWXU | S_IRWXG, session->uid, session->gid);
1947 if (ret < 0) {
1948 if (ret != -EEXIST) {
1949 ERR("Trace directory creation error");
1950 goto error;
1951 }
1952 }
1953 }
1954
1955 session->kernel_session->uid = session->uid;
1956 session->kernel_session->gid = session->gid;
1957
1958 return LTTCOMM_OK;
1959
1960 error:
1961 trace_kernel_destroy_session(session->kernel_session);
1962 session->kernel_session = NULL;
1963 return ret;
1964 }
1965
1966 /*
1967 * Count number of session permitted by uid/gid.
1968 */
1969 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
1970 {
1971 unsigned int i = 0;
1972 struct ltt_session *session;
1973
1974 DBG("Counting number of available session for UID %d GID %d",
1975 uid, gid);
1976 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
1977 /*
1978 * Only list the sessions the user can control.
1979 */
1980 if (!session_access_ok(session, uid, gid)) {
1981 continue;
1982 }
1983 i++;
1984 }
1985 return i;
1986 }
1987
1988 /*
1989 * Process the command requested by the lttng client within the command
1990 * context structure. This function make sure that the return structure (llm)
1991 * is set and ready for transmission before returning.
1992 *
1993 * Return any error encountered or 0 for success.
1994 *
1995 * "sock" is only used for special-case var. len data.
1996 */
1997 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
1998 int *sock_error)
1999 {
2000 int ret = LTTCOMM_OK;
2001 int need_tracing_session = 1;
2002 int need_domain;
2003
2004 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2005
2006 *sock_error = 0;
2007
2008 switch (cmd_ctx->lsm->cmd_type) {
2009 case LTTNG_CREATE_SESSION:
2010 case LTTNG_DESTROY_SESSION:
2011 case LTTNG_LIST_SESSIONS:
2012 case LTTNG_LIST_DOMAINS:
2013 case LTTNG_START_TRACE:
2014 case LTTNG_STOP_TRACE:
2015 need_domain = 0;
2016 break;
2017 default:
2018 need_domain = 1;
2019 }
2020
2021 if (opt_no_kernel && need_domain
2022 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2023 if (!is_root) {
2024 ret = LTTCOMM_NEED_ROOT_SESSIOND;
2025 } else {
2026 ret = LTTCOMM_KERN_NA;
2027 }
2028 goto error;
2029 }
2030
2031 /* Deny register consumer if we already have a spawned consumer. */
2032 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2033 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2034 if (kconsumer_data.pid > 0) {
2035 ret = LTTCOMM_KERN_CONSUMER_FAIL;
2036 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2037 goto error;
2038 }
2039 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2040 }
2041
2042 /*
2043 * Check for command that don't needs to allocate a returned payload. We do
2044 * this here so we don't have to make the call for no payload at each
2045 * command.
2046 */
2047 switch(cmd_ctx->lsm->cmd_type) {
2048 case LTTNG_LIST_SESSIONS:
2049 case LTTNG_LIST_TRACEPOINTS:
2050 case LTTNG_LIST_TRACEPOINT_FIELDS:
2051 case LTTNG_LIST_DOMAINS:
2052 case LTTNG_LIST_CHANNELS:
2053 case LTTNG_LIST_EVENTS:
2054 break;
2055 default:
2056 /* Setup lttng message with no payload */
2057 ret = setup_lttng_msg(cmd_ctx, 0);
2058 if (ret < 0) {
2059 /* This label does not try to unlock the session */
2060 goto init_setup_error;
2061 }
2062 }
2063
2064 /* Commands that DO NOT need a session. */
2065 switch (cmd_ctx->lsm->cmd_type) {
2066 case LTTNG_CREATE_SESSION:
2067 case LTTNG_CALIBRATE:
2068 case LTTNG_LIST_SESSIONS:
2069 case LTTNG_LIST_TRACEPOINTS:
2070 case LTTNG_LIST_TRACEPOINT_FIELDS:
2071 need_tracing_session = 0;
2072 break;
2073 default:
2074 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2075 /*
2076 * We keep the session list lock across _all_ commands
2077 * for now, because the per-session lock does not
2078 * handle teardown properly.
2079 */
2080 session_lock_list();
2081 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2082 if (cmd_ctx->session == NULL) {
2083 if (cmd_ctx->lsm->session.name != NULL) {
2084 ret = LTTCOMM_SESS_NOT_FOUND;
2085 } else {
2086 /* If no session name specified */
2087 ret = LTTCOMM_SELECT_SESS;
2088 }
2089 goto error;
2090 } else {
2091 /* Acquire lock for the session */
2092 session_lock(cmd_ctx->session);
2093 }
2094 break;
2095 }
2096
2097 if (!need_domain) {
2098 goto skip_domain;
2099 }
2100
2101 /*
2102 * Check domain type for specific "pre-action".
2103 */
2104 switch (cmd_ctx->lsm->domain.type) {
2105 case LTTNG_DOMAIN_KERNEL:
2106 if (!is_root) {
2107 ret = LTTCOMM_NEED_ROOT_SESSIOND;
2108 goto error;
2109 }
2110
2111 /* Kernel tracer check */
2112 if (kernel_tracer_fd == -1) {
2113 /* Basically, load kernel tracer modules */
2114 ret = init_kernel_tracer();
2115 if (ret != 0) {
2116 goto error;
2117 }
2118 }
2119
2120 /* Consumer is in an ERROR state. Report back to client */
2121 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2122 ret = LTTCOMM_NO_KERNCONSUMERD;
2123 goto error;
2124 }
2125
2126 /* Need a session for kernel command */
2127 if (need_tracing_session) {
2128 if (cmd_ctx->session->kernel_session == NULL) {
2129 ret = create_kernel_session(cmd_ctx->session);
2130 if (ret < 0) {
2131 ret = LTTCOMM_KERN_SESS_FAIL;
2132 goto error;
2133 }
2134 }
2135
2136 /* Start the kernel consumer daemon */
2137 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2138 if (kconsumer_data.pid == 0 &&
2139 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2140 cmd_ctx->session->start_consumer) {
2141 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2142 ret = start_consumerd(&kconsumer_data);
2143 if (ret < 0) {
2144 ret = LTTCOMM_KERN_CONSUMER_FAIL;
2145 goto error;
2146 }
2147 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2148 } else {
2149 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2150 }
2151
2152 /*
2153 * The consumer was just spawned so we need to add the socket to
2154 * the consumer output of the session if exist.
2155 */
2156 ret = consumer_create_socket(&kconsumer_data,
2157 cmd_ctx->session->kernel_session->consumer);
2158 if (ret < 0) {
2159 goto error;
2160 }
2161 }
2162
2163 break;
2164 case LTTNG_DOMAIN_UST:
2165 {
2166 /* Consumer is in an ERROR state. Report back to client */
2167 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2168 ret = LTTCOMM_NO_USTCONSUMERD;
2169 goto error;
2170 }
2171
2172 if (need_tracing_session) {
2173 /* Create UST session if none exist. */
2174 if (cmd_ctx->session->ust_session == NULL) {
2175 ret = create_ust_session(cmd_ctx->session,
2176 &cmd_ctx->lsm->domain);
2177 if (ret != LTTCOMM_OK) {
2178 goto error;
2179 }
2180 }
2181
2182 /* Start the UST consumer daemons */
2183 /* 64-bit */
2184 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2185 if (consumerd64_bin[0] != '\0' &&
2186 ustconsumer64_data.pid == 0 &&
2187 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2188 cmd_ctx->session->start_consumer) {
2189 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2190 ret = start_consumerd(&ustconsumer64_data);
2191 if (ret < 0) {
2192 ret = LTTCOMM_UST_CONSUMER64_FAIL;
2193 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2194 goto error;
2195 }
2196
2197 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2198 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2199 } else {
2200 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2201 }
2202
2203 /*
2204 * Setup socket for consumer 64 bit. No need for atomic access
2205 * since it was set above and can ONLY be set in this thread.
2206 */
2207 ret = consumer_create_socket(&ustconsumer64_data,
2208 cmd_ctx->session->ust_session->consumer);
2209 if (ret < 0) {
2210 goto error;
2211 }
2212
2213 /* 32-bit */
2214 if (consumerd32_bin[0] != '\0' &&
2215 ustconsumer32_data.pid == 0 &&
2216 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2217 cmd_ctx->session->start_consumer) {
2218 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2219 ret = start_consumerd(&ustconsumer32_data);
2220 if (ret < 0) {
2221 ret = LTTCOMM_UST_CONSUMER32_FAIL;
2222 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2223 goto error;
2224 }
2225
2226 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2227 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2228 } else {
2229 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2230 }
2231
2232 /*
2233 * Setup socket for consumer 64 bit. No need for atomic access
2234 * since it was set above and can ONLY be set in this thread.
2235 */
2236 ret = consumer_create_socket(&ustconsumer32_data,
2237 cmd_ctx->session->ust_session->consumer);
2238 if (ret < 0) {
2239 goto error;
2240 }
2241 }
2242 break;
2243 }
2244 default:
2245 break;
2246 }
2247 skip_domain:
2248
2249 /* Validate consumer daemon state when start/stop trace command */
2250 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2251 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2252 switch (cmd_ctx->lsm->domain.type) {
2253 case LTTNG_DOMAIN_UST:
2254 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2255 ret = LTTCOMM_NO_USTCONSUMERD;
2256 goto error;
2257 }
2258 break;
2259 case LTTNG_DOMAIN_KERNEL:
2260 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2261 ret = LTTCOMM_NO_KERNCONSUMERD;
2262 goto error;
2263 }
2264 break;
2265 }
2266 }
2267
2268 /*
2269 * Check that the UID or GID match that of the tracing session.
2270 * The root user can interact with all sessions.
2271 */
2272 if (need_tracing_session) {
2273 if (!session_access_ok(cmd_ctx->session,
2274 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2275 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2276 ret = LTTCOMM_EPERM;
2277 goto error;
2278 }
2279 }
2280
2281 /* Process by command type */
2282 switch (cmd_ctx->lsm->cmd_type) {
2283 case LTTNG_ADD_CONTEXT:
2284 {
2285 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2286 cmd_ctx->lsm->u.context.channel_name,
2287 cmd_ctx->lsm->u.context.event_name,
2288 &cmd_ctx->lsm->u.context.ctx);
2289 break;
2290 }
2291 case LTTNG_DISABLE_CHANNEL:
2292 {
2293 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2294 cmd_ctx->lsm->u.disable.channel_name);
2295 break;
2296 }
2297 case LTTNG_DISABLE_EVENT:
2298 {
2299 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2300 cmd_ctx->lsm->u.disable.channel_name,
2301 cmd_ctx->lsm->u.disable.name);
2302 break;
2303 }
2304 case LTTNG_DISABLE_ALL_EVENT:
2305 {
2306 DBG("Disabling all events");
2307
2308 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2309 cmd_ctx->lsm->u.disable.channel_name);
2310 break;
2311 }
2312 case LTTNG_DISABLE_CONSUMER:
2313 {
2314 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2315 break;
2316 }
2317 case LTTNG_ENABLE_CHANNEL:
2318 {
2319 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2320 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2321 break;
2322 }
2323 case LTTNG_ENABLE_CONSUMER:
2324 {
2325 /*
2326 * XXX: 0 means that this URI should be applied on the session. Should
2327 * be a DOMAIN enuam.
2328 */
2329 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2330 if (ret != LTTCOMM_OK) {
2331 goto error;
2332 }
2333
2334 if (cmd_ctx->lsm->domain.type == 0) {
2335 /* Add the URI for the UST session if a consumer is present. */
2336 if (cmd_ctx->session->ust_session &&
2337 cmd_ctx->session->ust_session->consumer) {
2338 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2339 } else if (cmd_ctx->session->kernel_session &&
2340 cmd_ctx->session->kernel_session->consumer) {
2341 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2342 cmd_ctx->session);
2343 }
2344 }
2345 break;
2346 }
2347 case LTTNG_ENABLE_EVENT:
2348 {
2349 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2350 cmd_ctx->lsm->u.enable.channel_name,
2351 &cmd_ctx->lsm->u.enable.event, kernel_poll_pipe[1]);
2352 break;
2353 }
2354 case LTTNG_ENABLE_ALL_EVENT:
2355 {
2356 DBG("Enabling all events");
2357
2358 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2359 cmd_ctx->lsm->u.enable.channel_name,
2360 cmd_ctx->lsm->u.enable.event.type, kernel_poll_pipe[1]);
2361 break;
2362 }
2363 case LTTNG_LIST_TRACEPOINTS:
2364 {
2365 struct lttng_event *events;
2366 ssize_t nb_events;
2367
2368 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2369 if (nb_events < 0) {
2370 ret = -nb_events;
2371 goto error;
2372 }
2373
2374 /*
2375 * Setup lttng message with payload size set to the event list size in
2376 * bytes and then copy list into the llm payload.
2377 */
2378 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2379 if (ret < 0) {
2380 free(events);
2381 goto setup_error;
2382 }
2383
2384 /* Copy event list into message payload */
2385 memcpy(cmd_ctx->llm->payload, events,
2386 sizeof(struct lttng_event) * nb_events);
2387
2388 free(events);
2389
2390 ret = LTTCOMM_OK;
2391 break;
2392 }
2393 case LTTNG_LIST_TRACEPOINT_FIELDS:
2394 {
2395 struct lttng_event_field *fields;
2396 ssize_t nb_fields;
2397
2398 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2399 &fields);
2400 if (nb_fields < 0) {
2401 ret = -nb_fields;
2402 goto error;
2403 }
2404
2405 /*
2406 * Setup lttng message with payload size set to the event list size in
2407 * bytes and then copy list into the llm payload.
2408 */
2409 ret = setup_lttng_msg(cmd_ctx,
2410 sizeof(struct lttng_event_field) * nb_fields);
2411 if (ret < 0) {
2412 free(fields);
2413 goto setup_error;
2414 }
2415
2416 /* Copy event list into message payload */
2417 memcpy(cmd_ctx->llm->payload, fields,
2418 sizeof(struct lttng_event_field) * nb_fields);
2419
2420 free(fields);
2421
2422 ret = LTTCOMM_OK;
2423 break;
2424 }
2425 case LTTNG_SET_CONSUMER_URI:
2426 {
2427 size_t nb_uri, len;
2428 struct lttng_uri *uris;
2429
2430 nb_uri = cmd_ctx->lsm->u.uri.size;
2431 len = nb_uri * sizeof(struct lttng_uri);
2432
2433 if (nb_uri == 0) {
2434 ret = LTTCOMM_INVALID;
2435 goto error;
2436 }
2437
2438 uris = zmalloc(len);
2439 if (uris == NULL) {
2440 ret = LTTCOMM_FATAL;
2441 goto error;
2442 }
2443
2444 /* Receive variable len data */
2445 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2446 ret = lttcomm_recv_unix_sock(sock, uris, len);
2447 if (ret <= 0) {
2448 DBG("No URIs received from client... continuing");
2449 *sock_error = 1;
2450 ret = LTTCOMM_SESSION_FAIL;
2451 goto error;
2452 }
2453
2454 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2455 nb_uri, uris);
2456 if (ret != LTTCOMM_OK) {
2457 goto error;
2458 }
2459
2460 /*
2461 * XXX: 0 means that this URI should be applied on the session. Should
2462 * be a DOMAIN enuam.
2463 */
2464 if (cmd_ctx->lsm->domain.type == 0) {
2465 /* Add the URI for the UST session if a consumer is present. */
2466 if (cmd_ctx->session->ust_session &&
2467 cmd_ctx->session->ust_session->consumer) {
2468 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2469 nb_uri, uris);
2470 } else if (cmd_ctx->session->kernel_session &&
2471 cmd_ctx->session->kernel_session->consumer) {
2472 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2473 cmd_ctx->session, nb_uri, uris);
2474 }
2475 }
2476
2477 break;
2478 }
2479 case LTTNG_START_TRACE:
2480 {
2481 ret = cmd_start_trace(cmd_ctx->session);
2482 break;
2483 }
2484 case LTTNG_STOP_TRACE:
2485 {
2486 ret = cmd_stop_trace(cmd_ctx->session);
2487 break;
2488 }
2489 case LTTNG_CREATE_SESSION:
2490 {
2491 size_t nb_uri, len;
2492 struct lttng_uri *uris = NULL;
2493
2494 nb_uri = cmd_ctx->lsm->u.uri.size;
2495 len = nb_uri * sizeof(struct lttng_uri);
2496
2497 if (nb_uri > 0) {
2498 uris = zmalloc(len);
2499 if (uris == NULL) {
2500 ret = LTTCOMM_FATAL;
2501 goto error;
2502 }
2503
2504 /* Receive variable len data */
2505 DBG("Waiting for %zu URIs from client ...", nb_uri);
2506 ret = lttcomm_recv_unix_sock(sock, uris, len);
2507 if (ret <= 0) {
2508 DBG("No URIs received from client... continuing");
2509 *sock_error = 1;
2510 ret = LTTCOMM_SESSION_FAIL;
2511 goto error;
2512 }
2513
2514 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2515 DBG("Creating session with ONE network URI is a bad call");
2516 ret = LTTCOMM_SESSION_FAIL;
2517 goto error;
2518 }
2519 }
2520
2521 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2522 &cmd_ctx->creds);
2523
2524 break;
2525 }
2526 case LTTNG_DESTROY_SESSION:
2527 {
2528 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2529
2530 /* Set session to NULL so we do not unlock it after free. */
2531 cmd_ctx->session = NULL;
2532 break;
2533 }
2534 case LTTNG_LIST_DOMAINS:
2535 {
2536 ssize_t nb_dom;
2537 struct lttng_domain *domains;
2538
2539 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2540 if (nb_dom < 0) {
2541 ret = -nb_dom;
2542 goto error;
2543 }
2544
2545 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2546 if (ret < 0) {
2547 goto setup_error;
2548 }
2549
2550 /* Copy event list into message payload */
2551 memcpy(cmd_ctx->llm->payload, domains,
2552 nb_dom * sizeof(struct lttng_domain));
2553
2554 free(domains);
2555
2556 ret = LTTCOMM_OK;
2557 break;
2558 }
2559 case LTTNG_LIST_CHANNELS:
2560 {
2561 int nb_chan;
2562 struct lttng_channel *channels;
2563
2564 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2565 cmd_ctx->session, &channels);
2566 if (nb_chan < 0) {
2567 ret = -nb_chan;
2568 goto error;
2569 }
2570
2571 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2572 if (ret < 0) {
2573 goto setup_error;
2574 }
2575
2576 /* Copy event list into message payload */
2577 memcpy(cmd_ctx->llm->payload, channels,
2578 nb_chan * sizeof(struct lttng_channel));
2579
2580 free(channels);
2581
2582 ret = LTTCOMM_OK;
2583 break;
2584 }
2585 case LTTNG_LIST_EVENTS:
2586 {
2587 ssize_t nb_event;
2588 struct lttng_event *events = NULL;
2589
2590 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2591 cmd_ctx->lsm->u.list.channel_name, &events);
2592 if (nb_event < 0) {
2593 ret = -nb_event;
2594 goto error;
2595 }
2596
2597 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2598 if (ret < 0) {
2599 goto setup_error;
2600 }
2601
2602 /* Copy event list into message payload */
2603 memcpy(cmd_ctx->llm->payload, events,
2604 nb_event * sizeof(struct lttng_event));
2605
2606 free(events);
2607
2608 ret = LTTCOMM_OK;
2609 break;
2610 }
2611 case LTTNG_LIST_SESSIONS:
2612 {
2613 unsigned int nr_sessions;
2614
2615 session_lock_list();
2616 nr_sessions = lttng_sessions_count(
2617 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2618 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2619
2620 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2621 if (ret < 0) {
2622 session_unlock_list();
2623 goto setup_error;
2624 }
2625
2626 /* Filled the session array */
2627 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2628 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2629 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2630
2631 session_unlock_list();
2632
2633 ret = LTTCOMM_OK;
2634 break;
2635 }
2636 case LTTNG_CALIBRATE:
2637 {
2638 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2639 &cmd_ctx->lsm->u.calibrate);
2640 break;
2641 }
2642 case LTTNG_REGISTER_CONSUMER:
2643 {
2644 struct consumer_data *cdata;
2645
2646 switch (cmd_ctx->lsm->domain.type) {
2647 case LTTNG_DOMAIN_KERNEL:
2648 cdata = &kconsumer_data;
2649 break;
2650 default:
2651 ret = LTTCOMM_UND;
2652 goto error;
2653 }
2654
2655 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2656 cmd_ctx->lsm->u.reg.path, cdata);
2657 break;
2658 }
2659 case LTTNG_SET_FILTER:
2660 {
2661 struct lttng_filter_bytecode *bytecode;
2662
2663 if (cmd_ctx->lsm->u.filter.bytecode_len > 65336) {
2664 ret = LTTCOMM_FILTER_INVAL;
2665 goto error;
2666 }
2667 bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len);
2668 if (!bytecode) {
2669 ret = LTTCOMM_FILTER_NOMEM;
2670 goto error;
2671 }
2672 /* Receive var. len. data */
2673 DBG("Receiving var len data from client ...");
2674 ret = lttcomm_recv_unix_sock(sock, bytecode,
2675 cmd_ctx->lsm->u.filter.bytecode_len);
2676 if (ret <= 0) {
2677 DBG("Nothing recv() from client var len data... continuing");
2678 *sock_error = 1;
2679 ret = LTTCOMM_FILTER_INVAL;
2680 goto error;
2681 }
2682
2683 if (bytecode->len + sizeof(*bytecode)
2684 != cmd_ctx->lsm->u.filter.bytecode_len) {
2685 free(bytecode);
2686 ret = LTTCOMM_FILTER_INVAL;
2687 goto error;
2688 }
2689
2690 ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2691 cmd_ctx->lsm->u.filter.channel_name,
2692 cmd_ctx->lsm->u.filter.event_name,
2693 bytecode);
2694 break;
2695 }
2696 default:
2697 ret = LTTCOMM_UND;
2698 break;
2699 }
2700
2701 error:
2702 if (cmd_ctx->llm == NULL) {
2703 DBG("Missing llm structure. Allocating one.");
2704 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2705 goto setup_error;
2706 }
2707 }
2708 /* Set return code */
2709 cmd_ctx->llm->ret_code = ret;
2710 setup_error:
2711 if (cmd_ctx->session) {
2712 session_unlock(cmd_ctx->session);
2713 }
2714 if (need_tracing_session) {
2715 session_unlock_list();
2716 }
2717 init_setup_error:
2718 return ret;
2719 }
2720
2721 /*
2722 * Thread managing health check socket.
2723 */
2724 static void *thread_manage_health(void *data)
2725 {
2726 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2727 uint32_t revents, nb_fd;
2728 struct lttng_poll_event events;
2729 struct lttcomm_health_msg msg;
2730 struct lttcomm_health_data reply;
2731
2732 DBG("[thread] Manage health check started");
2733
2734 rcu_register_thread();
2735
2736 /* Create unix socket */
2737 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2738 if (sock < 0) {
2739 ERR("Unable to create health check Unix socket");
2740 ret = -1;
2741 goto error;
2742 }
2743
2744 ret = lttcomm_listen_unix_sock(sock);
2745 if (ret < 0) {
2746 goto error;
2747 }
2748
2749 /*
2750 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2751 * more will be added to this poll set.
2752 */
2753 ret = create_thread_poll_set(&events, 2);
2754 if (ret < 0) {
2755 goto error;
2756 }
2757
2758 /* Add the application registration socket */
2759 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2760 if (ret < 0) {
2761 goto error;
2762 }
2763
2764 while (1) {
2765 DBG("Health check ready");
2766
2767 nb_fd = LTTNG_POLL_GETNB(&events);
2768
2769 /* Inifinite blocking call, waiting for transmission */
2770 restart:
2771 ret = lttng_poll_wait(&events, -1);
2772 if (ret < 0) {
2773 /*
2774 * Restart interrupted system call.
2775 */
2776 if (errno == EINTR) {
2777 goto restart;
2778 }
2779 goto error;
2780 }
2781
2782 for (i = 0; i < nb_fd; i++) {
2783 /* Fetch once the poll data */
2784 revents = LTTNG_POLL_GETEV(&events, i);
2785 pollfd = LTTNG_POLL_GETFD(&events, i);
2786
2787 /* Thread quit pipe has been closed. Killing thread. */
2788 ret = check_thread_quit_pipe(pollfd, revents);
2789 if (ret) {
2790 err = 0;
2791 goto exit;
2792 }
2793
2794 /* Event on the registration socket */
2795 if (pollfd == sock) {
2796 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2797 ERR("Health socket poll error");
2798 goto error;
2799 }
2800 }
2801 }
2802
2803 new_sock = lttcomm_accept_unix_sock(sock);
2804 if (new_sock < 0) {
2805 goto error;
2806 }
2807
2808 DBG("Receiving data from client for health...");
2809 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
2810 if (ret <= 0) {
2811 DBG("Nothing recv() from client... continuing");
2812 ret = close(new_sock);
2813 if (ret) {
2814 PERROR("close");
2815 }
2816 new_sock = -1;
2817 continue;
2818 }
2819
2820 rcu_thread_online();
2821
2822 switch (msg.component) {
2823 case LTTNG_HEALTH_CMD:
2824 reply.ret_code = health_check_state(&health_thread_cmd);
2825 break;
2826 case LTTNG_HEALTH_APP_MANAGE:
2827 reply.ret_code = health_check_state(&health_thread_app_manage);
2828 break;
2829 case LTTNG_HEALTH_APP_REG:
2830 reply.ret_code = health_check_state(&health_thread_app_reg);
2831 break;
2832 case LTTNG_HEALTH_KERNEL:
2833 reply.ret_code = health_check_state(&health_thread_kernel);
2834 break;
2835 case LTTNG_HEALTH_CONSUMER:
2836 reply.ret_code = check_consumer_health();
2837 break;
2838 case LTTNG_HEALTH_ALL:
2839 reply.ret_code =
2840 health_check_state(&health_thread_app_manage) &&
2841 health_check_state(&health_thread_app_reg) &&
2842 health_check_state(&health_thread_cmd) &&
2843 health_check_state(&health_thread_kernel) &&
2844 check_consumer_health();
2845 break;
2846 default:
2847 reply.ret_code = LTTCOMM_UND;
2848 break;
2849 }
2850
2851 /*
2852 * Flip ret value since 0 is a success and 1 indicates a bad health for
2853 * the client where in the sessiond it is the opposite. Again, this is
2854 * just to make things easier for us poor developer which enjoy a lot
2855 * lazyness.
2856 */
2857 if (reply.ret_code == 0 || reply.ret_code == 1) {
2858 reply.ret_code = !reply.ret_code;
2859 }
2860
2861 DBG2("Health check return value %d", reply.ret_code);
2862
2863 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
2864 if (ret < 0) {
2865 ERR("Failed to send health data back to client");
2866 }
2867
2868 /* End of transmission */
2869 ret = close(new_sock);
2870 if (ret) {
2871 PERROR("close");
2872 }
2873 new_sock = -1;
2874 }
2875
2876 exit:
2877 error:
2878 if (err) {
2879 ERR("Health error occurred in %s", __func__);
2880 }
2881 DBG("Health check thread dying");
2882 unlink(health_unix_sock_path);
2883 if (sock >= 0) {
2884 ret = close(sock);
2885 if (ret) {
2886 PERROR("close");
2887 }
2888 }
2889 if (new_sock >= 0) {
2890 ret = close(new_sock);
2891 if (ret) {
2892 PERROR("close");
2893 }
2894 }
2895
2896 lttng_poll_clean(&events);
2897
2898 rcu_unregister_thread();
2899 return NULL;
2900 }
2901
2902 /*
2903 * This thread manage all clients request using the unix client socket for
2904 * communication.
2905 */
2906 static void *thread_manage_clients(void *data)
2907 {
2908 int sock = -1, ret, i, pollfd, err = -1;
2909 int sock_error;
2910 uint32_t revents, nb_fd;
2911 struct command_ctx *cmd_ctx = NULL;
2912 struct lttng_poll_event events;
2913
2914 DBG("[thread] Manage client started");
2915
2916 rcu_register_thread();
2917
2918 health_code_update(&health_thread_cmd);
2919
2920 ret = lttcomm_listen_unix_sock(client_sock);
2921 if (ret < 0) {
2922 goto error;
2923 }
2924
2925 /*
2926 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2927 * more will be added to this poll set.
2928 */
2929 ret = create_thread_poll_set(&events, 2);
2930 if (ret < 0) {
2931 goto error;
2932 }
2933
2934 /* Add the application registration socket */
2935 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
2936 if (ret < 0) {
2937 goto error;
2938 }
2939
2940 /*
2941 * Notify parent pid that we are ready to accept command for client side.
2942 */
2943 if (opt_sig_parent) {
2944 kill(ppid, SIGUSR1);
2945 }
2946
2947 health_code_update(&health_thread_cmd);
2948
2949 while (1) {
2950 DBG("Accepting client command ...");
2951
2952 nb_fd = LTTNG_POLL_GETNB(&events);
2953
2954 /* Inifinite blocking call, waiting for transmission */
2955 restart:
2956 health_poll_update(&health_thread_cmd);
2957 ret = lttng_poll_wait(&events, -1);
2958 health_poll_update(&health_thread_cmd);
2959 if (ret < 0) {
2960 /*
2961 * Restart interrupted system call.
2962 */
2963 if (errno == EINTR) {
2964 goto restart;
2965 }
2966 goto error;
2967 }
2968
2969 for (i = 0; i < nb_fd; i++) {
2970 /* Fetch once the poll data */
2971 revents = LTTNG_POLL_GETEV(&events, i);
2972 pollfd = LTTNG_POLL_GETFD(&events, i);
2973
2974 health_code_update(&health_thread_cmd);
2975
2976 /* Thread quit pipe has been closed. Killing thread. */
2977 ret = check_thread_quit_pipe(pollfd, revents);
2978 if (ret) {
2979 err = 0;
2980 goto exit;
2981 }
2982
2983 /* Event on the registration socket */
2984 if (pollfd == client_sock) {
2985 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2986 ERR("Client socket poll error");
2987 goto error;
2988 }
2989 }
2990 }
2991
2992 DBG("Wait for client response");
2993
2994 health_code_update(&health_thread_cmd);
2995
2996 sock = lttcomm_accept_unix_sock(client_sock);
2997 if (sock < 0) {
2998 goto error;
2999 }
3000
3001 /* Set socket option for credentials retrieval */
3002 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3003 if (ret < 0) {
3004 goto error;
3005 }
3006
3007 /* Allocate context command to process the client request */
3008 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3009 if (cmd_ctx == NULL) {
3010 PERROR("zmalloc cmd_ctx");
3011 goto error;
3012 }
3013
3014 /* Allocate data buffer for reception */
3015 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3016 if (cmd_ctx->lsm == NULL) {
3017 PERROR("zmalloc cmd_ctx->lsm");
3018 goto error;
3019 }
3020
3021 cmd_ctx->llm = NULL;
3022 cmd_ctx->session = NULL;
3023
3024 health_code_update(&health_thread_cmd);
3025
3026 /*
3027 * Data is received from the lttng client. The struct
3028 * lttcomm_session_msg (lsm) contains the command and data request of
3029 * the client.
3030 */
3031 DBG("Receiving data from client ...");
3032 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3033 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3034 if (ret <= 0) {
3035 DBG("Nothing recv() from client... continuing");
3036 ret = close(sock);
3037 if (ret) {
3038 PERROR("close");
3039 }
3040 sock = -1;
3041 clean_command_ctx(&cmd_ctx);
3042 continue;
3043 }
3044
3045 health_code_update(&health_thread_cmd);
3046
3047 // TODO: Validate cmd_ctx including sanity check for
3048 // security purpose.
3049
3050 rcu_thread_online();
3051 /*
3052 * This function dispatch the work to the kernel or userspace tracer
3053 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3054 * informations for the client. The command context struct contains
3055 * everything this function may needs.
3056 */
3057 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3058 rcu_thread_offline();
3059 if (ret < 0) {
3060 if (sock_error) {
3061 ret = close(sock);
3062 if (ret) {
3063 PERROR("close");
3064 }
3065 sock = -1;
3066 }
3067 /*
3068 * TODO: Inform client somehow of the fatal error. At
3069 * this point, ret < 0 means that a zmalloc failed
3070 * (ENOMEM). Error detected but still accept
3071 * command, unless a socket error has been
3072 * detected.
3073 */
3074 clean_command_ctx(&cmd_ctx);
3075 continue;
3076 }
3077
3078 health_code_update(&health_thread_cmd);
3079
3080 DBG("Sending response (size: %d, retcode: %s)",
3081 cmd_ctx->lttng_msg_size,
3082 lttng_strerror(-cmd_ctx->llm->ret_code));
3083 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3084 if (ret < 0) {
3085 ERR("Failed to send data back to client");
3086 }
3087
3088 /* End of transmission */
3089 ret = close(sock);
3090 if (ret) {
3091 PERROR("close");
3092 }
3093 sock = -1;
3094
3095 clean_command_ctx(&cmd_ctx);
3096
3097 health_code_update(&health_thread_cmd);
3098 }
3099
3100 exit:
3101 error:
3102 if (err) {
3103 health_error(&health_thread_cmd);
3104 ERR("Health error occurred in %s", __func__);
3105 }
3106 health_exit(&health_thread_cmd);
3107
3108 DBG("Client thread dying");
3109 unlink(client_unix_sock_path);
3110 if (client_sock >= 0) {
3111 ret = close(client_sock);
3112 if (ret) {
3113 PERROR("close");
3114 }
3115 }
3116 if (sock >= 0) {
3117 ret = close(sock);
3118 if (ret) {
3119 PERROR("close");
3120 }
3121 }
3122
3123 lttng_poll_clean(&events);
3124 clean_command_ctx(&cmd_ctx);
3125
3126 rcu_unregister_thread();
3127 return NULL;
3128 }
3129
3130
3131 /*
3132 * usage function on stderr
3133 */
3134 static void usage(void)
3135 {
3136 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3137 fprintf(stderr, " -h, --help Display this usage.\n");
3138 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3139 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3140 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3141 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3142 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3143 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3144 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3145 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3146 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3147 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3148 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3149 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3150 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3151 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3152 fprintf(stderr, " -V, --version Show version number.\n");
3153 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3154 fprintf(stderr, " -q, --quiet No output at all.\n");
3155 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3156 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3157 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3158 }
3159
3160 /*
3161 * daemon argument parsing
3162 */
3163 static int parse_args(int argc, char **argv)
3164 {
3165 int c;
3166
3167 static struct option long_options[] = {
3168 { "client-sock", 1, 0, 'c' },
3169 { "apps-sock", 1, 0, 'a' },
3170 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3171 { "kconsumerd-err-sock", 1, 0, 'E' },
3172 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3173 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3174 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3175 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3176 { "consumerd32-path", 1, 0, 'u' },
3177 { "consumerd32-libdir", 1, 0, 'U' },
3178 { "consumerd64-path", 1, 0, 't' },
3179 { "consumerd64-libdir", 1, 0, 'T' },
3180 { "daemonize", 0, 0, 'd' },
3181 { "sig-parent", 0, 0, 'S' },
3182 { "help", 0, 0, 'h' },
3183 { "group", 1, 0, 'g' },
3184 { "version", 0, 0, 'V' },
3185 { "quiet", 0, 0, 'q' },
3186 { "verbose", 0, 0, 'v' },
3187 { "verbose-consumer", 0, 0, 'Z' },
3188 { "no-kernel", 0, 0, 'N' },
3189 { NULL, 0, 0, 0 }
3190 };
3191
3192 while (1) {
3193 int option_index = 0;
3194 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3195 long_options, &option_index);
3196 if (c == -1) {
3197 break;
3198 }
3199
3200 switch (c) {
3201 case 0:
3202 fprintf(stderr, "option %s", long_options[option_index].name);
3203 if (optarg) {
3204 fprintf(stderr, " with arg %s\n", optarg);
3205 }
3206 break;
3207 case 'c':
3208 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3209 break;
3210 case 'a':
3211 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3212 break;
3213 case 'd':
3214 opt_daemon = 1;
3215 break;
3216 case 'g':
3217 opt_tracing_group = optarg;
3218 break;
3219 case 'h':
3220 usage();
3221 exit(EXIT_FAILURE);
3222 case 'V':
3223 fprintf(stdout, "%s\n", VERSION);
3224 exit(EXIT_SUCCESS);
3225 case 'S':
3226 opt_sig_parent = 1;
3227 break;
3228 case 'E':
3229 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3230 break;
3231 case 'C':
3232 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3233 break;
3234 case 'F':
3235 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3236 break;
3237 case 'D':
3238 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3239 break;
3240 case 'H':
3241 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3242 break;
3243 case 'G':
3244 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3245 break;
3246 case 'N':
3247 opt_no_kernel = 1;
3248 break;
3249 case 'q':
3250 lttng_opt_quiet = 1;
3251 break;
3252 case 'v':
3253 /* Verbose level can increase using multiple -v */
3254 lttng_opt_verbose += 1;
3255 break;
3256 case 'Z':
3257 opt_verbose_consumer += 1;
3258 break;
3259 case 'u':
3260 consumerd32_bin= optarg;
3261 break;
3262 case 'U':
3263 consumerd32_libdir = optarg;
3264 break;
3265 case 't':
3266 consumerd64_bin = optarg;
3267 break;
3268 case 'T':
3269 consumerd64_libdir = optarg;
3270 break;
3271 default:
3272 /* Unknown option or other error.
3273 * Error is printed by getopt, just return */
3274 return -1;
3275 }
3276 }
3277
3278 return 0;
3279 }
3280
3281 /*
3282 * Creates the two needed socket by the daemon.
3283 * apps_sock - The communication socket for all UST apps.
3284 * client_sock - The communication of the cli tool (lttng).
3285 */
3286 static int init_daemon_socket(void)
3287 {
3288 int ret = 0;
3289 mode_t old_umask;
3290
3291 old_umask = umask(0);
3292
3293 /* Create client tool unix socket */
3294 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3295 if (client_sock < 0) {
3296 ERR("Create unix sock failed: %s", client_unix_sock_path);
3297 ret = -1;
3298 goto end;
3299 }
3300
3301 /* File permission MUST be 660 */
3302 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3303 if (ret < 0) {
3304 ERR("Set file permissions failed: %s", client_unix_sock_path);
3305 PERROR("chmod");
3306 goto end;
3307 }
3308
3309 /* Create the application unix socket */
3310 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3311 if (apps_sock < 0) {
3312 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3313 ret = -1;
3314 goto end;
3315 }
3316
3317 /* File permission MUST be 666 */
3318 ret = chmod(apps_unix_sock_path,
3319 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3320 if (ret < 0) {
3321 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3322 PERROR("chmod");
3323 goto end;
3324 }
3325
3326 end:
3327 umask(old_umask);
3328 return ret;
3329 }
3330
3331 /*
3332 * Check if the global socket is available, and if a daemon is answering at the
3333 * other side. If yes, error is returned.
3334 */
3335 static int check_existing_daemon(void)
3336 {
3337 /* Is there anybody out there ? */
3338 if (lttng_session_daemon_alive()) {
3339 return -EEXIST;
3340 }
3341
3342 return 0;
3343 }
3344
3345 /*
3346 * Set the tracing group gid onto the client socket.
3347 *
3348 * Race window between mkdir and chown is OK because we are going from more
3349 * permissive (root.root) to less permissive (root.tracing).
3350 */
3351 static int set_permissions(char *rundir)
3352 {
3353 int ret;
3354 gid_t gid;
3355
3356 ret = allowed_group();
3357 if (ret < 0) {
3358 WARN("No tracing group detected");
3359 ret = 0;
3360 goto end;
3361 }
3362
3363 gid = ret;
3364
3365 /* Set lttng run dir */
3366 ret = chown(rundir, 0, gid);
3367 if (ret < 0) {
3368 ERR("Unable to set group on %s", rundir);
3369 PERROR("chown");
3370 }
3371
3372 /* Ensure tracing group can search the run dir */
3373 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3374 if (ret < 0) {
3375 ERR("Unable to set permissions on %s", rundir);
3376 PERROR("chmod");
3377 }
3378
3379 /* lttng client socket path */
3380 ret = chown(client_unix_sock_path, 0, gid);
3381 if (ret < 0) {
3382 ERR("Unable to set group on %s", client_unix_sock_path);
3383 PERROR("chown");
3384 }
3385
3386 /* kconsumer error socket path */
3387 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3388 if (ret < 0) {
3389 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3390 PERROR("chown");
3391 }
3392
3393 /* 64-bit ustconsumer error socket path */
3394 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3395 if (ret < 0) {
3396 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3397 PERROR("chown");
3398 }
3399
3400 /* 32-bit ustconsumer compat32 error socket path */
3401 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3402 if (ret < 0) {
3403 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3404 PERROR("chown");
3405 }
3406
3407 DBG("All permissions are set");
3408
3409 end:
3410 return ret;
3411 }
3412
3413 /*
3414 * Create the lttng run directory needed for all global sockets and pipe.
3415 */
3416 static int create_lttng_rundir(const char *rundir)
3417 {
3418 int ret;
3419
3420 DBG3("Creating LTTng run directory: %s", rundir);
3421
3422 ret = mkdir(rundir, S_IRWXU);
3423 if (ret < 0) {
3424 if (errno != EEXIST) {
3425 ERR("Unable to create %s", rundir);
3426 goto error;
3427 } else {
3428 ret = 0;
3429 }
3430 }
3431
3432 error:
3433 return ret;
3434 }
3435
3436 /*
3437 * Setup sockets and directory needed by the kconsumerd communication with the
3438 * session daemon.
3439 */
3440 static int set_consumer_sockets(struct consumer_data *consumer_data,
3441 const char *rundir)
3442 {
3443 int ret;
3444 char path[PATH_MAX];
3445
3446 switch (consumer_data->type) {
3447 case LTTNG_CONSUMER_KERNEL:
3448 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3449 break;
3450 case LTTNG_CONSUMER64_UST:
3451 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3452 break;
3453 case LTTNG_CONSUMER32_UST:
3454 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3455 break;
3456 default:
3457 ERR("Consumer type unknown");
3458 ret = -EINVAL;
3459 goto error;
3460 }
3461
3462 DBG2("Creating consumer directory: %s", path);
3463
3464 ret = mkdir(path, S_IRWXU);
3465 if (ret < 0) {
3466 if (errno != EEXIST) {
3467 PERROR("mkdir");
3468 ERR("Failed to create %s", path);
3469 goto error;
3470 }
3471 ret = -1;
3472 }
3473
3474 /* Create the kconsumerd error unix socket */
3475 consumer_data->err_sock =
3476 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3477 if (consumer_data->err_sock < 0) {
3478 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3479 ret = -1;
3480 goto error;
3481 }
3482
3483 /* File permission MUST be 660 */
3484 ret = chmod(consumer_data->err_unix_sock_path,
3485 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3486 if (ret < 0) {
3487 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3488 PERROR("chmod");
3489 goto error;
3490 }
3491
3492 error:
3493 return ret;
3494 }
3495
3496 /*
3497 * Signal handler for the daemon
3498 *
3499 * Simply stop all worker threads, leaving main() return gracefully after
3500 * joining all threads and calling cleanup().
3501 */
3502 static void sighandler(int sig)
3503 {
3504 switch (sig) {
3505 case SIGPIPE:
3506 DBG("SIGPIPE caught");
3507 return;
3508 case SIGINT:
3509 DBG("SIGINT caught");
3510 stop_threads();
3511 break;
3512 case SIGTERM:
3513 DBG("SIGTERM caught");
3514 stop_threads();
3515 break;
3516 default:
3517 break;
3518 }
3519 }
3520
3521 /*
3522 * Setup signal handler for :
3523 * SIGINT, SIGTERM, SIGPIPE
3524 */
3525 static int set_signal_handler(void)
3526 {
3527 int ret = 0;
3528 struct sigaction sa;
3529 sigset_t sigset;
3530
3531 if ((ret = sigemptyset(&sigset)) < 0) {
3532 PERROR("sigemptyset");
3533 return ret;
3534 }
3535
3536 sa.sa_handler = sighandler;
3537 sa.sa_mask = sigset;
3538 sa.sa_flags = 0;
3539 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3540 PERROR("sigaction");
3541 return ret;
3542 }
3543
3544 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3545 PERROR("sigaction");
3546 return ret;
3547 }
3548
3549 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3550 PERROR("sigaction");
3551 return ret;
3552 }
3553
3554 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3555
3556 return ret;
3557 }
3558
3559 /*
3560 * Set open files limit to unlimited. This daemon can open a large number of
3561 * file descriptors in order to consumer multiple kernel traces.
3562 */
3563 static void set_ulimit(void)
3564 {
3565 int ret;
3566 struct rlimit lim;
3567
3568 /* The kernel does not allowed an infinite limit for open files */
3569 lim.rlim_cur = 65535;
3570 lim.rlim_max = 65535;
3571
3572 ret = setrlimit(RLIMIT_NOFILE, &lim);
3573 if (ret < 0) {
3574 PERROR("failed to set open files limit");
3575 }
3576 }
3577
3578 /*
3579 * main
3580 */
3581 int main(int argc, char **argv)
3582 {
3583 int ret = 0;
3584 void *status;
3585 const char *home_path;
3586
3587 init_kernel_workarounds();
3588
3589 rcu_register_thread();
3590
3591 setup_consumerd_path();
3592
3593 /* Parse arguments */
3594 progname = argv[0];
3595 if ((ret = parse_args(argc, argv) < 0)) {
3596 goto error;
3597 }
3598
3599 /* Daemonize */
3600 if (opt_daemon) {
3601 int i;
3602
3603 /*
3604 * fork
3605 * child: setsid, close FD 0, 1, 2, chdir /
3606 * parent: exit (if fork is successful)
3607 */
3608 ret = daemon(0, 0);
3609 if (ret < 0) {
3610 PERROR("daemon");
3611 goto error;
3612 }
3613 /*
3614 * We are in the child. Make sure all other file
3615 * descriptors are closed, in case we are called with
3616 * more opened file descriptors than the standard ones.
3617 */
3618 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3619 (void) close(i);
3620 }
3621 }
3622
3623 /* Create thread quit pipe */
3624 if ((ret = init_thread_quit_pipe()) < 0) {
3625 goto error;
3626 }
3627
3628 /* Check if daemon is UID = 0 */
3629 is_root = !getuid();
3630
3631 if (is_root) {
3632 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3633
3634 /* Create global run dir with root access */
3635 ret = create_lttng_rundir(rundir);
3636 if (ret < 0) {
3637 goto error;
3638 }
3639
3640 if (strlen(apps_unix_sock_path) == 0) {
3641 snprintf(apps_unix_sock_path, PATH_MAX,
3642 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3643 }
3644
3645 if (strlen(client_unix_sock_path) == 0) {
3646 snprintf(client_unix_sock_path, PATH_MAX,
3647 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3648 }
3649
3650 /* Set global SHM for ust */
3651 if (strlen(wait_shm_path) == 0) {
3652 snprintf(wait_shm_path, PATH_MAX,
3653 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3654 }
3655
3656 if (strlen(health_unix_sock_path) == 0) {
3657 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3658 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3659 }
3660
3661 /* Setup kernel consumerd path */
3662 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3663 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3664 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3665 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3666
3667 DBG2("Kernel consumer err path: %s",
3668 kconsumer_data.err_unix_sock_path);
3669 DBG2("Kernel consumer cmd path: %s",
3670 kconsumer_data.cmd_unix_sock_path);
3671 } else {
3672 home_path = get_home_dir();
3673 if (home_path == NULL) {
3674 /* TODO: Add --socket PATH option */
3675 ERR("Can't get HOME directory for sockets creation.");
3676 ret = -EPERM;
3677 goto error;
3678 }
3679
3680 /*
3681 * Create rundir from home path. This will create something like
3682 * $HOME/.lttng
3683 */
3684 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3685 if (ret < 0) {
3686 ret = -ENOMEM;
3687 goto error;
3688 }
3689
3690 ret = create_lttng_rundir(rundir);
3691 if (ret < 0) {
3692 goto error;
3693 }
3694
3695 if (strlen(apps_unix_sock_path) == 0) {
3696 snprintf(apps_unix_sock_path, PATH_MAX,
3697 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3698 }
3699
3700 /* Set the cli tool unix socket path */
3701 if (strlen(client_unix_sock_path) == 0) {
3702 snprintf(client_unix_sock_path, PATH_MAX,
3703 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3704 }
3705
3706 /* Set global SHM for ust */
3707 if (strlen(wait_shm_path) == 0) {
3708 snprintf(wait_shm_path, PATH_MAX,
3709 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3710 }
3711
3712 /* Set health check Unix path */
3713 if (strlen(health_unix_sock_path) == 0) {
3714 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3715 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3716 }
3717 }
3718
3719 /* Set consumer initial state */
3720 kernel_consumerd_state = CONSUMER_STOPPED;
3721 ust_consumerd_state = CONSUMER_STOPPED;
3722
3723 DBG("Client socket path %s", client_unix_sock_path);
3724 DBG("Application socket path %s", apps_unix_sock_path);
3725 DBG("LTTng run directory path: %s", rundir);
3726
3727 /* 32 bits consumerd path setup */
3728 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
3729 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
3730 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
3731 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
3732
3733 DBG2("UST consumer 32 bits err path: %s",
3734 ustconsumer32_data.err_unix_sock_path);
3735 DBG2("UST consumer 32 bits cmd path: %s",
3736 ustconsumer32_data.cmd_unix_sock_path);
3737
3738 /* 64 bits consumerd path setup */
3739 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
3740 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
3741 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
3742 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
3743
3744 DBG2("UST consumer 64 bits err path: %s",
3745 ustconsumer64_data.err_unix_sock_path);
3746 DBG2("UST consumer 64 bits cmd path: %s",
3747 ustconsumer64_data.cmd_unix_sock_path);
3748
3749 /*
3750 * See if daemon already exist.
3751 */
3752 if ((ret = check_existing_daemon()) < 0) {
3753 ERR("Already running daemon.\n");
3754 /*
3755 * We do not goto exit because we must not cleanup()
3756 * because a daemon is already running.
3757 */
3758 goto error;
3759 }
3760
3761 /*
3762 * Init UST app hash table. Alloc hash table before this point since
3763 * cleanup() can get called after that point.
3764 */
3765 ust_app_ht_alloc();
3766
3767 /* After this point, we can safely call cleanup() with "goto exit" */
3768
3769 /*
3770 * These actions must be executed as root. We do that *after* setting up
3771 * the sockets path because we MUST make the check for another daemon using
3772 * those paths *before* trying to set the kernel consumer sockets and init
3773 * kernel tracer.
3774 */
3775 if (is_root) {
3776 ret = set_consumer_sockets(&kconsumer_data, rundir);
3777 if (ret < 0) {
3778 goto exit;
3779 }
3780
3781 /* Setup kernel tracer */
3782 if (!opt_no_kernel) {
3783 init_kernel_tracer();
3784 }
3785
3786 /* Set ulimit for open files */
3787 set_ulimit();
3788 }
3789 /* init lttng_fd tracking must be done after set_ulimit. */
3790 lttng_fd_init();
3791
3792 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
3793 if (ret < 0) {
3794 goto exit;
3795 }
3796
3797 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
3798 if (ret < 0) {
3799 goto exit;
3800 }
3801
3802 if ((ret = set_signal_handler()) < 0) {
3803 goto exit;
3804 }
3805
3806 /* Setup the needed unix socket */
3807 if ((ret = init_daemon_socket()) < 0) {
3808 goto exit;
3809 }
3810
3811 /* Set credentials to socket */
3812 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
3813 goto exit;
3814 }
3815
3816 /* Get parent pid if -S, --sig-parent is specified. */
3817 if (opt_sig_parent) {
3818 ppid = getppid();
3819 }
3820
3821 /* Setup the kernel pipe for waking up the kernel thread */
3822 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
3823 goto exit;
3824 }
3825
3826 /* Setup the thread apps communication pipe. */
3827 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
3828 goto exit;
3829 }
3830
3831 /* Init UST command queue. */
3832 cds_wfq_init(&ust_cmd_queue.queue);
3833
3834 /*
3835 * Get session list pointer. This pointer MUST NOT be free(). This list is
3836 * statically declared in session.c
3837 */
3838 session_list_ptr = session_get_list();
3839
3840 /* Set up max poll set size */
3841 lttng_poll_set_max_size();
3842
3843 cmd_init();
3844
3845 /* Init all health thread counters. */
3846 health_init(&health_thread_cmd);
3847 health_init(&health_thread_kernel);
3848 health_init(&health_thread_app_manage);
3849 health_init(&health_thread_app_reg);
3850
3851 /*
3852 * Init health counters of the consumer thread. We do a quick hack here to
3853 * the state of the consumer health is fine even if the thread is not
3854 * started. This is simply to ease our life and has no cost what so ever.
3855 */
3856 health_init(&kconsumer_data.health);
3857 health_poll_update(&kconsumer_data.health);
3858 health_init(&ustconsumer32_data.health);
3859 health_poll_update(&ustconsumer32_data.health);
3860 health_init(&ustconsumer64_data.health);
3861 health_poll_update(&ustconsumer64_data.health);
3862
3863 /* Create thread to manage the client socket */
3864 ret = pthread_create(&health_thread, NULL,
3865 thread_manage_health, (void *) NULL);
3866 if (ret != 0) {
3867 PERROR("pthread_create health");
3868 goto exit_health;
3869 }
3870
3871 /* Create thread to manage the client socket */
3872 ret = pthread_create(&client_thread, NULL,
3873 thread_manage_clients, (void *) NULL);
3874 if (ret != 0) {
3875 PERROR("pthread_create clients");
3876 goto exit_client;
3877 }
3878
3879 /* Create thread to dispatch registration */
3880 ret = pthread_create(&dispatch_thread, NULL,
3881 thread_dispatch_ust_registration, (void *) NULL);
3882 if (ret != 0) {
3883 PERROR("pthread_create dispatch");
3884 goto exit_dispatch;
3885 }
3886
3887 /* Create thread to manage application registration. */
3888 ret = pthread_create(&reg_apps_thread, NULL,
3889 thread_registration_apps, (void *) NULL);
3890 if (ret != 0) {
3891 PERROR("pthread_create registration");
3892 goto exit_reg_apps;
3893 }
3894
3895 /* Create thread to manage application socket */
3896 ret = pthread_create(&apps_thread, NULL,
3897 thread_manage_apps, (void *) NULL);
3898 if (ret != 0) {
3899 PERROR("pthread_create apps");
3900 goto exit_apps;
3901 }
3902
3903 /* Create kernel thread to manage kernel event */
3904 ret = pthread_create(&kernel_thread, NULL,
3905 thread_manage_kernel, (void *) NULL);
3906 if (ret != 0) {
3907 PERROR("pthread_create kernel");
3908 goto exit_kernel;
3909 }
3910
3911 ret = pthread_join(kernel_thread, &status);
3912 if (ret != 0) {
3913 PERROR("pthread_join");
3914 goto error; /* join error, exit without cleanup */
3915 }
3916
3917 exit_kernel:
3918 ret = pthread_join(apps_thread, &status);
3919 if (ret != 0) {
3920 PERROR("pthread_join");
3921 goto error; /* join error, exit without cleanup */
3922 }
3923
3924 exit_apps:
3925 ret = pthread_join(reg_apps_thread, &status);
3926 if (ret != 0) {
3927 PERROR("pthread_join");
3928 goto error; /* join error, exit without cleanup */
3929 }
3930
3931 exit_reg_apps:
3932 ret = pthread_join(dispatch_thread, &status);
3933 if (ret != 0) {
3934 PERROR("pthread_join");
3935 goto error; /* join error, exit without cleanup */
3936 }
3937
3938 exit_dispatch:
3939 ret = pthread_join(client_thread, &status);
3940 if (ret != 0) {
3941 PERROR("pthread_join");
3942 goto error; /* join error, exit without cleanup */
3943 }
3944
3945 ret = join_consumer_thread(&kconsumer_data);
3946 if (ret != 0) {
3947 PERROR("join_consumer");
3948 goto error; /* join error, exit without cleanup */
3949 }
3950
3951 exit_client:
3952 exit_health:
3953 exit:
3954 /*
3955 * cleanup() is called when no other thread is running.
3956 */
3957 rcu_thread_online();
3958 cleanup();
3959 rcu_thread_offline();
3960 rcu_unregister_thread();
3961 if (!ret) {
3962 exit(EXIT_SUCCESS);
3963 }
3964 error:
3965 exit(EXIT_FAILURE);
3966 }
This page took 0.184899 seconds and 3 git commands to generate.