Add testpoints in lttng-sessiond for each threads
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <sys/mman.h>
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
32 #include <sys/stat.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <urcu/uatomic.h>
36 #include <unistd.h>
37 #include <config.h>
38
39 #include <common/common.h>
40 #include <common/compat/poll.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
47
48 #include "lttng-sessiond.h"
49 #include "channel.h"
50 #include "cmd.h"
51 #include "consumer.h"
52 #include "context.h"
53 #include "event.h"
54 #include "kernel.h"
55 #include "kernel-consumer.h"
56 #include "modprobe.h"
57 #include "shm.h"
58 #include "ust-ctl.h"
59 #include "ust-consumer.h"
60 #include "utils.h"
61 #include "fd-limit.h"
62 #include "filter.h"
63 #include "health.h"
64 #include "testpoint.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 /*
85 * Consumer daemon specific control data. Every value not initialized here is
86 * set to 0 by the static definition.
87 */
88 static struct consumer_data kconsumer_data = {
89 .type = LTTNG_CONSUMER_KERNEL,
90 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
91 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
92 .err_sock = -1,
93 .cmd_sock = -1,
94 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
95 .lock = PTHREAD_MUTEX_INITIALIZER,
96 .cond = PTHREAD_COND_INITIALIZER,
97 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
98 };
99 static struct consumer_data ustconsumer64_data = {
100 .type = LTTNG_CONSUMER64_UST,
101 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
109 };
110 static struct consumer_data ustconsumer32_data = {
111 .type = LTTNG_CONSUMER32_UST,
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
114 .err_sock = -1,
115 .cmd_sock = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121
122 /* Shared between threads */
123 static int dispatch_thread_exit;
124
125 /* Global application Unix socket path */
126 static char apps_unix_sock_path[PATH_MAX];
127 /* Global client Unix socket path */
128 static char client_unix_sock_path[PATH_MAX];
129 /* global wait shm path for UST */
130 static char wait_shm_path[PATH_MAX];
131 /* Global health check unix path */
132 static char health_unix_sock_path[PATH_MAX];
133
134 /* Sockets and FDs */
135 static int client_sock = -1;
136 static int apps_sock = -1;
137 int kernel_tracer_fd = -1;
138 static int kernel_poll_pipe[2] = { -1, -1 };
139
140 /*
141 * Quit pipe for all threads. This permits a single cancellation point
142 * for all threads when receiving an event on the pipe.
143 */
144 static int thread_quit_pipe[2] = { -1, -1 };
145
146 /*
147 * This pipe is used to inform the thread managing application communication
148 * that a command is queued and ready to be processed.
149 */
150 static int apps_cmd_pipe[2] = { -1, -1 };
151
152 /* Pthread, Mutexes and Semaphores */
153 static pthread_t apps_thread;
154 static pthread_t reg_apps_thread;
155 static pthread_t client_thread;
156 static pthread_t kernel_thread;
157 static pthread_t dispatch_thread;
158 static pthread_t health_thread;
159
160 /*
161 * UST registration command queue. This queue is tied with a futex and uses a N
162 * wakers / 1 waiter implemented and detailed in futex.c/.h
163 *
164 * The thread_manage_apps and thread_dispatch_ust_registration interact with
165 * this queue and the wait/wake scheme.
166 */
167 static struct ust_cmd_queue ust_cmd_queue;
168
169 /*
170 * Pointer initialized before thread creation.
171 *
172 * This points to the tracing session list containing the session count and a
173 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
174 * MUST NOT be taken if you call a public function in session.c.
175 *
176 * The lock is nested inside the structure: session_list_ptr->lock. Please use
177 * session_lock_list and session_unlock_list for lock acquisition.
178 */
179 static struct ltt_session_list *session_list_ptr;
180
181 int ust_consumerd64_fd = -1;
182 int ust_consumerd32_fd = -1;
183
184 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
185 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
186 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
187 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
188
189 static const char *module_proc_lttng = "/proc/lttng";
190
191 /*
192 * Consumer daemon state which is changed when spawning it, killing it or in
193 * case of a fatal error.
194 */
195 enum consumerd_state {
196 CONSUMER_STARTED = 1,
197 CONSUMER_STOPPED = 2,
198 CONSUMER_ERROR = 3,
199 };
200
201 /*
202 * This consumer daemon state is used to validate if a client command will be
203 * able to reach the consumer. If not, the client is informed. For instance,
204 * doing a "lttng start" when the consumer state is set to ERROR will return an
205 * error to the client.
206 *
207 * The following example shows a possible race condition of this scheme:
208 *
209 * consumer thread error happens
210 * client cmd arrives
211 * client cmd checks state -> still OK
212 * consumer thread exit, sets error
213 * client cmd try to talk to consumer
214 * ...
215 *
216 * However, since the consumer is a different daemon, we have no way of making
217 * sure the command will reach it safely even with this state flag. This is why
218 * we consider that up to the state validation during command processing, the
219 * command is safe. After that, we can not guarantee the correctness of the
220 * client request vis-a-vis the consumer.
221 */
222 static enum consumerd_state ust_consumerd_state;
223 static enum consumerd_state kernel_consumerd_state;
224
225 /* Used for the health monitoring of the session daemon. See health.h */
226 struct health_state health_thread_cmd;
227 struct health_state health_thread_app_manage;
228 struct health_state health_thread_app_reg;
229 struct health_state health_thread_kernel;
230
231 static
232 void setup_consumerd_path(void)
233 {
234 const char *bin, *libdir;
235
236 /*
237 * Allow INSTALL_BIN_PATH to be used as a target path for the
238 * native architecture size consumer if CONFIG_CONSUMER*_PATH
239 * has not been defined.
240 */
241 #if (CAA_BITS_PER_LONG == 32)
242 if (!consumerd32_bin[0]) {
243 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
244 }
245 if (!consumerd32_libdir[0]) {
246 consumerd32_libdir = INSTALL_LIB_PATH;
247 }
248 #elif (CAA_BITS_PER_LONG == 64)
249 if (!consumerd64_bin[0]) {
250 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
251 }
252 if (!consumerd64_libdir[0]) {
253 consumerd64_libdir = INSTALL_LIB_PATH;
254 }
255 #else
256 #error "Unknown bitness"
257 #endif
258
259 /*
260 * runtime env. var. overrides the build default.
261 */
262 bin = getenv("LTTNG_CONSUMERD32_BIN");
263 if (bin) {
264 consumerd32_bin = bin;
265 }
266 bin = getenv("LTTNG_CONSUMERD64_BIN");
267 if (bin) {
268 consumerd64_bin = bin;
269 }
270 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
271 if (libdir) {
272 consumerd32_libdir = libdir;
273 }
274 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
275 if (libdir) {
276 consumerd64_libdir = libdir;
277 }
278 }
279
280 /*
281 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
282 */
283 static int create_thread_poll_set(struct lttng_poll_event *events,
284 unsigned int size)
285 {
286 int ret;
287
288 if (events == NULL || size == 0) {
289 ret = -1;
290 goto error;
291 }
292
293 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
294 if (ret < 0) {
295 goto error;
296 }
297
298 /* Add quit pipe */
299 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
300 if (ret < 0) {
301 goto error;
302 }
303
304 return 0;
305
306 error:
307 return ret;
308 }
309
310 /*
311 * Check if the thread quit pipe was triggered.
312 *
313 * Return 1 if it was triggered else 0;
314 */
315 static int check_thread_quit_pipe(int fd, uint32_t events)
316 {
317 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
318 return 1;
319 }
320
321 return 0;
322 }
323
324 /*
325 * Return group ID of the tracing group or -1 if not found.
326 */
327 static gid_t allowed_group(void)
328 {
329 struct group *grp;
330
331 if (opt_tracing_group) {
332 grp = getgrnam(opt_tracing_group);
333 } else {
334 grp = getgrnam(default_tracing_group);
335 }
336 if (!grp) {
337 return -1;
338 } else {
339 return grp->gr_gid;
340 }
341 }
342
343 /*
344 * Init thread quit pipe.
345 *
346 * Return -1 on error or 0 if all pipes are created.
347 */
348 static int init_thread_quit_pipe(void)
349 {
350 int ret, i;
351
352 ret = pipe(thread_quit_pipe);
353 if (ret < 0) {
354 PERROR("thread quit pipe");
355 goto error;
356 }
357
358 for (i = 0; i < 2; i++) {
359 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
360 if (ret < 0) {
361 PERROR("fcntl");
362 goto error;
363 }
364 }
365
366 error:
367 return ret;
368 }
369
370 /*
371 * Stop all threads by closing the thread quit pipe.
372 */
373 static void stop_threads(void)
374 {
375 int ret;
376
377 /* Stopping all threads */
378 DBG("Terminating all threads");
379 ret = notify_thread_pipe(thread_quit_pipe[1]);
380 if (ret < 0) {
381 ERR("write error on thread quit pipe");
382 }
383
384 /* Dispatch thread */
385 CMM_STORE_SHARED(dispatch_thread_exit, 1);
386 futex_nto1_wake(&ust_cmd_queue.futex);
387 }
388
389 /*
390 * Cleanup the daemon
391 */
392 static void cleanup(void)
393 {
394 int ret;
395 char *cmd;
396 struct ltt_session *sess, *stmp;
397
398 DBG("Cleaning up");
399
400 /* First thing first, stop all threads */
401 utils_close_pipe(thread_quit_pipe);
402
403 DBG("Removing %s directory", rundir);
404 ret = asprintf(&cmd, "rm -rf %s", rundir);
405 if (ret < 0) {
406 ERR("asprintf failed. Something is really wrong!");
407 }
408
409 /* Remove lttng run directory */
410 ret = system(cmd);
411 if (ret < 0) {
412 ERR("Unable to clean %s", rundir);
413 }
414 free(cmd);
415
416 DBG("Cleaning up all sessions");
417
418 /* Destroy session list mutex */
419 if (session_list_ptr != NULL) {
420 pthread_mutex_destroy(&session_list_ptr->lock);
421
422 /* Cleanup ALL session */
423 cds_list_for_each_entry_safe(sess, stmp,
424 &session_list_ptr->head, list) {
425 cmd_destroy_session(sess, kernel_poll_pipe[1]);
426 }
427 }
428
429 DBG("Closing all UST sockets");
430 ust_app_clean_list();
431
432 if (is_root && !opt_no_kernel) {
433 DBG2("Closing kernel fd");
434 if (kernel_tracer_fd >= 0) {
435 ret = close(kernel_tracer_fd);
436 if (ret) {
437 PERROR("close");
438 }
439 }
440 DBG("Unloading kernel modules");
441 modprobe_remove_lttng_all();
442 }
443
444 utils_close_pipe(kernel_poll_pipe);
445 utils_close_pipe(apps_cmd_pipe);
446
447 /* <fun> */
448 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
449 "Matthew, BEET driven development works!%c[%dm",
450 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
451 /* </fun> */
452 }
453
454 /*
455 * Send data on a unix socket using the liblttsessiondcomm API.
456 *
457 * Return lttcomm error code.
458 */
459 static int send_unix_sock(int sock, void *buf, size_t len)
460 {
461 /* Check valid length */
462 if (len <= 0) {
463 return -1;
464 }
465
466 return lttcomm_send_unix_sock(sock, buf, len);
467 }
468
469 /*
470 * Free memory of a command context structure.
471 */
472 static void clean_command_ctx(struct command_ctx **cmd_ctx)
473 {
474 DBG("Clean command context structure");
475 if (*cmd_ctx) {
476 if ((*cmd_ctx)->llm) {
477 free((*cmd_ctx)->llm);
478 }
479 if ((*cmd_ctx)->lsm) {
480 free((*cmd_ctx)->lsm);
481 }
482 free(*cmd_ctx);
483 *cmd_ctx = NULL;
484 }
485 }
486
487 /*
488 * Notify UST applications using the shm mmap futex.
489 */
490 static int notify_ust_apps(int active)
491 {
492 char *wait_shm_mmap;
493
494 DBG("Notifying applications of session daemon state: %d", active);
495
496 /* See shm.c for this call implying mmap, shm and futex calls */
497 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
498 if (wait_shm_mmap == NULL) {
499 goto error;
500 }
501
502 /* Wake waiting process */
503 futex_wait_update((int32_t *) wait_shm_mmap, active);
504
505 /* Apps notified successfully */
506 return 0;
507
508 error:
509 return -1;
510 }
511
512 /*
513 * Setup the outgoing data buffer for the response (llm) by allocating the
514 * right amount of memory and copying the original information from the lsm
515 * structure.
516 *
517 * Return total size of the buffer pointed by buf.
518 */
519 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
520 {
521 int ret, buf_size;
522
523 buf_size = size;
524
525 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
526 if (cmd_ctx->llm == NULL) {
527 PERROR("zmalloc");
528 ret = -ENOMEM;
529 goto error;
530 }
531
532 /* Copy common data */
533 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
534 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
535
536 cmd_ctx->llm->data_size = size;
537 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
538
539 return buf_size;
540
541 error:
542 return ret;
543 }
544
545 /*
546 * Update the kernel poll set of all channel fd available over all tracing
547 * session. Add the wakeup pipe at the end of the set.
548 */
549 static int update_kernel_poll(struct lttng_poll_event *events)
550 {
551 int ret;
552 struct ltt_session *session;
553 struct ltt_kernel_channel *channel;
554
555 DBG("Updating kernel poll set");
556
557 session_lock_list();
558 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
559 session_lock(session);
560 if (session->kernel_session == NULL) {
561 session_unlock(session);
562 continue;
563 }
564
565 cds_list_for_each_entry(channel,
566 &session->kernel_session->channel_list.head, list) {
567 /* Add channel fd to the kernel poll set */
568 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
569 if (ret < 0) {
570 session_unlock(session);
571 goto error;
572 }
573 DBG("Channel fd %d added to kernel set", channel->fd);
574 }
575 session_unlock(session);
576 }
577 session_unlock_list();
578
579 return 0;
580
581 error:
582 session_unlock_list();
583 return -1;
584 }
585
586 /*
587 * Find the channel fd from 'fd' over all tracing session. When found, check
588 * for new channel stream and send those stream fds to the kernel consumer.
589 *
590 * Useful for CPU hotplug feature.
591 */
592 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
593 {
594 int ret = 0;
595 struct ltt_session *session;
596 struct ltt_kernel_session *ksess;
597 struct ltt_kernel_channel *channel;
598
599 DBG("Updating kernel streams for channel fd %d", fd);
600
601 session_lock_list();
602 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
603 session_lock(session);
604 if (session->kernel_session == NULL) {
605 session_unlock(session);
606 continue;
607 }
608 ksess = session->kernel_session;
609
610 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
611 if (channel->fd == fd) {
612 DBG("Channel found, updating kernel streams");
613 ret = kernel_open_channel_stream(channel);
614 if (ret < 0) {
615 goto error;
616 }
617
618 /*
619 * Have we already sent fds to the consumer? If yes, it means
620 * that tracing is started so it is safe to send our updated
621 * stream fds.
622 */
623 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
624 struct lttng_ht_iter iter;
625 struct consumer_socket *socket;
626
627
628 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
629 &iter.iter, socket, node.node) {
630 /* Code flow error */
631 assert(socket->fd >= 0);
632
633 pthread_mutex_lock(socket->lock);
634 ret = kernel_consumer_send_channel_stream(socket->fd,
635 channel, ksess);
636 pthread_mutex_unlock(socket->lock);
637 if (ret < 0) {
638 goto error;
639 }
640 }
641 }
642 goto error;
643 }
644 }
645 session_unlock(session);
646 }
647 session_unlock_list();
648 return ret;
649
650 error:
651 session_unlock(session);
652 session_unlock_list();
653 return ret;
654 }
655
656 /*
657 * For each tracing session, update newly registered apps.
658 */
659 static void update_ust_app(int app_sock)
660 {
661 struct ltt_session *sess, *stmp;
662
663 session_lock_list();
664
665 /* For all tracing session(s) */
666 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
667 session_lock(sess);
668 if (sess->ust_session) {
669 ust_app_global_update(sess->ust_session, app_sock);
670 }
671 session_unlock(sess);
672 }
673
674 session_unlock_list();
675 }
676
677 /*
678 * This thread manage event coming from the kernel.
679 *
680 * Features supported in this thread:
681 * -) CPU Hotplug
682 */
683 static void *thread_manage_kernel(void *data)
684 {
685 int ret, i, pollfd, update_poll_flag = 1, err = -1;
686 uint32_t revents, nb_fd;
687 char tmp;
688 struct lttng_poll_event events;
689
690 DBG("Thread manage kernel started");
691
692 testpoint(thread_manage_kernel);
693
694 health_code_update(&health_thread_kernel);
695
696 testpoint(thread_manage_kernel_before_loop);
697
698 ret = create_thread_poll_set(&events, 2);
699 if (ret < 0) {
700 goto error_poll_create;
701 }
702
703 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
704 if (ret < 0) {
705 goto error;
706 }
707
708 while (1) {
709 health_code_update(&health_thread_kernel);
710
711 if (update_poll_flag == 1) {
712 /*
713 * Reset number of fd in the poll set. Always 2 since there is the thread
714 * quit pipe and the kernel pipe.
715 */
716 events.nb_fd = 2;
717
718 ret = update_kernel_poll(&events);
719 if (ret < 0) {
720 goto error;
721 }
722 update_poll_flag = 0;
723 }
724
725 nb_fd = LTTNG_POLL_GETNB(&events);
726
727 DBG("Thread kernel polling on %d fds", nb_fd);
728
729 /* Zeroed the poll events */
730 lttng_poll_reset(&events);
731
732 /* Poll infinite value of time */
733 restart:
734 health_poll_update(&health_thread_kernel);
735 ret = lttng_poll_wait(&events, -1);
736 health_poll_update(&health_thread_kernel);
737 if (ret < 0) {
738 /*
739 * Restart interrupted system call.
740 */
741 if (errno == EINTR) {
742 goto restart;
743 }
744 goto error;
745 } else if (ret == 0) {
746 /* Should not happen since timeout is infinite */
747 ERR("Return value of poll is 0 with an infinite timeout.\n"
748 "This should not have happened! Continuing...");
749 continue;
750 }
751
752 for (i = 0; i < nb_fd; i++) {
753 /* Fetch once the poll data */
754 revents = LTTNG_POLL_GETEV(&events, i);
755 pollfd = LTTNG_POLL_GETFD(&events, i);
756
757 health_code_update(&health_thread_kernel);
758
759 /* Thread quit pipe has been closed. Killing thread. */
760 ret = check_thread_quit_pipe(pollfd, revents);
761 if (ret) {
762 err = 0;
763 goto exit;
764 }
765
766 /* Check for data on kernel pipe */
767 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
768 ret = read(kernel_poll_pipe[0], &tmp, 1);
769 update_poll_flag = 1;
770 continue;
771 } else {
772 /*
773 * New CPU detected by the kernel. Adding kernel stream to
774 * kernel session and updating the kernel consumer
775 */
776 if (revents & LPOLLIN) {
777 ret = update_kernel_stream(&kconsumer_data, pollfd);
778 if (ret < 0) {
779 continue;
780 }
781 break;
782 /*
783 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
784 * and unregister kernel stream at this point.
785 */
786 }
787 }
788 }
789 }
790
791 exit:
792 error:
793 lttng_poll_clean(&events);
794 error_poll_create:
795 if (err) {
796 health_error(&health_thread_kernel);
797 ERR("Health error occurred in %s", __func__);
798 }
799 health_exit(&health_thread_kernel);
800 DBG("Kernel thread dying");
801 return NULL;
802 }
803
804 /*
805 * Signal pthread condition of the consumer data that the thread.
806 */
807 static void signal_consumer_condition(struct consumer_data *data, int state)
808 {
809 pthread_mutex_lock(&data->cond_mutex);
810
811 /*
812 * The state is set before signaling. It can be any value, it's the waiter
813 * job to correctly interpret this condition variable associated to the
814 * consumer pthread_cond.
815 *
816 * A value of 0 means that the corresponding thread of the consumer data
817 * was not started. 1 indicates that the thread has started and is ready
818 * for action. A negative value means that there was an error during the
819 * thread bootstrap.
820 */
821 data->consumer_thread_is_ready = state;
822 (void) pthread_cond_signal(&data->cond);
823
824 pthread_mutex_unlock(&data->cond_mutex);
825 }
826
827 /*
828 * This thread manage the consumer error sent back to the session daemon.
829 */
830 static void *thread_manage_consumer(void *data)
831 {
832 int sock = -1, i, ret, pollfd, err = -1;
833 uint32_t revents, nb_fd;
834 enum lttcomm_return_code code;
835 struct lttng_poll_event events;
836 struct consumer_data *consumer_data = data;
837
838 DBG("[thread] Manage consumer started");
839
840 health_code_update(&consumer_data->health);
841
842 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
843 if (ret < 0) {
844 goto error_listen;
845 }
846
847 /*
848 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
849 * Nothing more will be added to this poll set.
850 */
851 ret = create_thread_poll_set(&events, 2);
852 if (ret < 0) {
853 goto error_poll;
854 }
855
856 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
857 if (ret < 0) {
858 goto error;
859 }
860
861 nb_fd = LTTNG_POLL_GETNB(&events);
862
863 health_code_update(&consumer_data->health);
864
865 /* Inifinite blocking call, waiting for transmission */
866 restart:
867 health_poll_update(&consumer_data->health);
868
869 testpoint(thread_manage_consumer);
870
871 ret = lttng_poll_wait(&events, -1);
872 health_poll_update(&consumer_data->health);
873 if (ret < 0) {
874 /*
875 * Restart interrupted system call.
876 */
877 if (errno == EINTR) {
878 goto restart;
879 }
880 goto error;
881 }
882
883 for (i = 0; i < nb_fd; i++) {
884 /* Fetch once the poll data */
885 revents = LTTNG_POLL_GETEV(&events, i);
886 pollfd = LTTNG_POLL_GETFD(&events, i);
887
888 health_code_update(&consumer_data->health);
889
890 /* Thread quit pipe has been closed. Killing thread. */
891 ret = check_thread_quit_pipe(pollfd, revents);
892 if (ret) {
893 err = 0;
894 goto exit;
895 }
896
897 /* Event on the registration socket */
898 if (pollfd == consumer_data->err_sock) {
899 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
900 ERR("consumer err socket poll error");
901 goto error;
902 }
903 }
904 }
905
906 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
907 if (sock < 0) {
908 goto error;
909 }
910
911 health_code_update(&consumer_data->health);
912
913 DBG2("Receiving code from consumer err_sock");
914
915 /* Getting status code from kconsumerd */
916 ret = lttcomm_recv_unix_sock(sock, &code,
917 sizeof(enum lttcomm_return_code));
918 if (ret <= 0) {
919 goto error;
920 }
921
922 health_code_update(&consumer_data->health);
923
924 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
925 consumer_data->cmd_sock =
926 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
927 if (consumer_data->cmd_sock < 0) {
928 /* On error, signal condition and quit. */
929 signal_consumer_condition(consumer_data, -1);
930 PERROR("consumer connect");
931 goto error;
932 }
933 signal_consumer_condition(consumer_data, 1);
934 DBG("Consumer command socket ready");
935 } else {
936 ERR("consumer error when waiting for SOCK_READY : %s",
937 lttcomm_get_readable_code(-code));
938 goto error;
939 }
940
941 /* Remove the kconsumerd error sock since we've established a connexion */
942 ret = lttng_poll_del(&events, consumer_data->err_sock);
943 if (ret < 0) {
944 goto error;
945 }
946
947 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
948 if (ret < 0) {
949 goto error;
950 }
951
952 health_code_update(&consumer_data->health);
953
954 /* Update number of fd */
955 nb_fd = LTTNG_POLL_GETNB(&events);
956
957 /* Inifinite blocking call, waiting for transmission */
958 restart_poll:
959 health_poll_update(&consumer_data->health);
960 ret = lttng_poll_wait(&events, -1);
961 health_poll_update(&consumer_data->health);
962 if (ret < 0) {
963 /*
964 * Restart interrupted system call.
965 */
966 if (errno == EINTR) {
967 goto restart_poll;
968 }
969 goto error;
970 }
971
972 for (i = 0; i < nb_fd; i++) {
973 /* Fetch once the poll data */
974 revents = LTTNG_POLL_GETEV(&events, i);
975 pollfd = LTTNG_POLL_GETFD(&events, i);
976
977 health_code_update(&consumer_data->health);
978
979 /* Thread quit pipe has been closed. Killing thread. */
980 ret = check_thread_quit_pipe(pollfd, revents);
981 if (ret) {
982 err = 0;
983 goto exit;
984 }
985
986 /* Event on the kconsumerd socket */
987 if (pollfd == sock) {
988 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
989 ERR("consumer err socket second poll error");
990 goto error;
991 }
992 }
993 }
994
995 health_code_update(&consumer_data->health);
996
997 /* Wait for any kconsumerd error */
998 ret = lttcomm_recv_unix_sock(sock, &code,
999 sizeof(enum lttcomm_return_code));
1000 if (ret <= 0) {
1001 ERR("consumer closed the command socket");
1002 goto error;
1003 }
1004
1005 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1006
1007 exit:
1008 error:
1009 /* Immediately set the consumerd state to stopped */
1010 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1011 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1012 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1013 consumer_data->type == LTTNG_CONSUMER32_UST) {
1014 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1015 } else {
1016 /* Code flow error... */
1017 assert(0);
1018 }
1019
1020 if (consumer_data->err_sock >= 0) {
1021 ret = close(consumer_data->err_sock);
1022 if (ret) {
1023 PERROR("close");
1024 }
1025 }
1026 if (consumer_data->cmd_sock >= 0) {
1027 ret = close(consumer_data->cmd_sock);
1028 if (ret) {
1029 PERROR("close");
1030 }
1031 }
1032 if (sock >= 0) {
1033 ret = close(sock);
1034 if (ret) {
1035 PERROR("close");
1036 }
1037 }
1038
1039 unlink(consumer_data->err_unix_sock_path);
1040 unlink(consumer_data->cmd_unix_sock_path);
1041 consumer_data->pid = 0;
1042
1043 lttng_poll_clean(&events);
1044 error_poll:
1045 error_listen:
1046 if (err) {
1047 health_error(&consumer_data->health);
1048 ERR("Health error occurred in %s", __func__);
1049 }
1050 health_exit(&consumer_data->health);
1051 DBG("consumer thread cleanup completed");
1052
1053 return NULL;
1054 }
1055
1056 /*
1057 * This thread manage application communication.
1058 */
1059 static void *thread_manage_apps(void *data)
1060 {
1061 int i, ret, pollfd, err = -1;
1062 uint32_t revents, nb_fd;
1063 struct ust_command ust_cmd;
1064 struct lttng_poll_event events;
1065
1066 DBG("[thread] Manage application started");
1067
1068 testpoint(thread_manage_apps);
1069
1070 rcu_register_thread();
1071 rcu_thread_online();
1072
1073 health_code_update(&health_thread_app_manage);
1074
1075 ret = create_thread_poll_set(&events, 2);
1076 if (ret < 0) {
1077 goto error_poll_create;
1078 }
1079
1080 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1081 if (ret < 0) {
1082 goto error;
1083 }
1084
1085 testpoint(thread_manage_apps_before_loop);
1086
1087 health_code_update(&health_thread_app_manage);
1088
1089 while (1) {
1090 /* Zeroed the events structure */
1091 lttng_poll_reset(&events);
1092
1093 nb_fd = LTTNG_POLL_GETNB(&events);
1094
1095 DBG("Apps thread polling on %d fds", nb_fd);
1096
1097 /* Inifinite blocking call, waiting for transmission */
1098 restart:
1099 health_poll_update(&health_thread_app_manage);
1100 ret = lttng_poll_wait(&events, -1);
1101 health_poll_update(&health_thread_app_manage);
1102 if (ret < 0) {
1103 /*
1104 * Restart interrupted system call.
1105 */
1106 if (errno == EINTR) {
1107 goto restart;
1108 }
1109 goto error;
1110 }
1111
1112 for (i = 0; i < nb_fd; i++) {
1113 /* Fetch once the poll data */
1114 revents = LTTNG_POLL_GETEV(&events, i);
1115 pollfd = LTTNG_POLL_GETFD(&events, i);
1116
1117 health_code_update(&health_thread_app_manage);
1118
1119 /* Thread quit pipe has been closed. Killing thread. */
1120 ret = check_thread_quit_pipe(pollfd, revents);
1121 if (ret) {
1122 err = 0;
1123 goto exit;
1124 }
1125
1126 /* Inspect the apps cmd pipe */
1127 if (pollfd == apps_cmd_pipe[0]) {
1128 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1129 ERR("Apps command pipe error");
1130 goto error;
1131 } else if (revents & LPOLLIN) {
1132 /* Empty pipe */
1133 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1134 if (ret < 0 || ret < sizeof(ust_cmd)) {
1135 PERROR("read apps cmd pipe");
1136 goto error;
1137 }
1138
1139 health_code_update(&health_thread_app_manage);
1140
1141 /* Register applicaton to the session daemon */
1142 ret = ust_app_register(&ust_cmd.reg_msg,
1143 ust_cmd.sock);
1144 if (ret == -ENOMEM) {
1145 goto error;
1146 } else if (ret < 0) {
1147 break;
1148 }
1149
1150 health_code_update(&health_thread_app_manage);
1151
1152 /*
1153 * Validate UST version compatibility.
1154 */
1155 ret = ust_app_validate_version(ust_cmd.sock);
1156 if (ret >= 0) {
1157 /*
1158 * Add channel(s) and event(s) to newly registered apps
1159 * from lttng global UST domain.
1160 */
1161 update_ust_app(ust_cmd.sock);
1162 }
1163
1164 health_code_update(&health_thread_app_manage);
1165
1166 ret = ust_app_register_done(ust_cmd.sock);
1167 if (ret < 0) {
1168 /*
1169 * If the registration is not possible, we simply
1170 * unregister the apps and continue
1171 */
1172 ust_app_unregister(ust_cmd.sock);
1173 } else {
1174 /*
1175 * We just need here to monitor the close of the UST
1176 * socket and poll set monitor those by default.
1177 * Listen on POLLIN (even if we never expect any
1178 * data) to ensure that hangup wakes us.
1179 */
1180 ret = lttng_poll_add(&events, ust_cmd.sock, LPOLLIN);
1181 if (ret < 0) {
1182 goto error;
1183 }
1184
1185 DBG("Apps with sock %d added to poll set",
1186 ust_cmd.sock);
1187 }
1188
1189 health_code_update(&health_thread_app_manage);
1190
1191 break;
1192 }
1193 } else {
1194 /*
1195 * At this point, we know that a registered application made
1196 * the event at poll_wait.
1197 */
1198 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1199 /* Removing from the poll set */
1200 ret = lttng_poll_del(&events, pollfd);
1201 if (ret < 0) {
1202 goto error;
1203 }
1204
1205 /* Socket closed on remote end. */
1206 ust_app_unregister(pollfd);
1207 break;
1208 }
1209 }
1210
1211 health_code_update(&health_thread_app_manage);
1212 }
1213 }
1214
1215 exit:
1216 error:
1217 lttng_poll_clean(&events);
1218 error_poll_create:
1219 if (err) {
1220 health_error(&health_thread_app_manage);
1221 ERR("Health error occurred in %s", __func__);
1222 }
1223 health_exit(&health_thread_app_manage);
1224 DBG("Application communication apps thread cleanup complete");
1225 rcu_thread_offline();
1226 rcu_unregister_thread();
1227 return NULL;
1228 }
1229
1230 /*
1231 * Dispatch request from the registration threads to the application
1232 * communication thread.
1233 */
1234 static void *thread_dispatch_ust_registration(void *data)
1235 {
1236 int ret;
1237 struct cds_wfq_node *node;
1238 struct ust_command *ust_cmd = NULL;
1239
1240 DBG("[thread] Dispatch UST command started");
1241
1242 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1243 /* Atomically prepare the queue futex */
1244 futex_nto1_prepare(&ust_cmd_queue.futex);
1245
1246 do {
1247 /* Dequeue command for registration */
1248 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1249 if (node == NULL) {
1250 DBG("Woken up but nothing in the UST command queue");
1251 /* Continue thread execution */
1252 break;
1253 }
1254
1255 ust_cmd = caa_container_of(node, struct ust_command, node);
1256
1257 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1258 " gid:%d sock:%d name:%s (version %d.%d)",
1259 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1260 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1261 ust_cmd->sock, ust_cmd->reg_msg.name,
1262 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1263 /*
1264 * Inform apps thread of the new application registration. This
1265 * call is blocking so we can be assured that the data will be read
1266 * at some point in time or wait to the end of the world :)
1267 */
1268 ret = write(apps_cmd_pipe[1], ust_cmd,
1269 sizeof(struct ust_command));
1270 if (ret < 0) {
1271 PERROR("write apps cmd pipe");
1272 if (errno == EBADF) {
1273 /*
1274 * We can't inform the application thread to process
1275 * registration. We will exit or else application
1276 * registration will not occur and tracing will never
1277 * start.
1278 */
1279 goto error;
1280 }
1281 }
1282 free(ust_cmd);
1283 } while (node != NULL);
1284
1285 /* Futex wait on queue. Blocking call on futex() */
1286 futex_nto1_wait(&ust_cmd_queue.futex);
1287 }
1288
1289 error:
1290 DBG("Dispatch thread dying");
1291 return NULL;
1292 }
1293
1294 /*
1295 * This thread manage application registration.
1296 */
1297 static void *thread_registration_apps(void *data)
1298 {
1299 int sock = -1, i, ret, pollfd, err = -1;
1300 uint32_t revents, nb_fd;
1301 struct lttng_poll_event events;
1302 /*
1303 * Get allocated in this thread, enqueued to a global queue, dequeued and
1304 * freed in the manage apps thread.
1305 */
1306 struct ust_command *ust_cmd = NULL;
1307
1308 DBG("[thread] Manage application registration started");
1309
1310 testpoint(thread_registration_apps);
1311
1312 ret = lttcomm_listen_unix_sock(apps_sock);
1313 if (ret < 0) {
1314 goto error_listen;
1315 }
1316
1317 /*
1318 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1319 * more will be added to this poll set.
1320 */
1321 ret = create_thread_poll_set(&events, 2);
1322 if (ret < 0) {
1323 goto error_create_poll;
1324 }
1325
1326 /* Add the application registration socket */
1327 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1328 if (ret < 0) {
1329 goto error_poll_add;
1330 }
1331
1332 /* Notify all applications to register */
1333 ret = notify_ust_apps(1);
1334 if (ret < 0) {
1335 ERR("Failed to notify applications or create the wait shared memory.\n"
1336 "Execution continues but there might be problem for already\n"
1337 "running applications that wishes to register.");
1338 }
1339
1340 while (1) {
1341 DBG("Accepting application registration");
1342
1343 nb_fd = LTTNG_POLL_GETNB(&events);
1344
1345 /* Inifinite blocking call, waiting for transmission */
1346 restart:
1347 health_poll_update(&health_thread_app_reg);
1348 ret = lttng_poll_wait(&events, -1);
1349 health_poll_update(&health_thread_app_reg);
1350 if (ret < 0) {
1351 /*
1352 * Restart interrupted system call.
1353 */
1354 if (errno == EINTR) {
1355 goto restart;
1356 }
1357 goto error;
1358 }
1359
1360 for (i = 0; i < nb_fd; i++) {
1361 health_code_update(&health_thread_app_reg);
1362
1363 /* Fetch once the poll data */
1364 revents = LTTNG_POLL_GETEV(&events, i);
1365 pollfd = LTTNG_POLL_GETFD(&events, i);
1366
1367 /* Thread quit pipe has been closed. Killing thread. */
1368 ret = check_thread_quit_pipe(pollfd, revents);
1369 if (ret) {
1370 err = 0;
1371 goto exit;
1372 }
1373
1374 /* Event on the registration socket */
1375 if (pollfd == apps_sock) {
1376 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1377 ERR("Register apps socket poll error");
1378 goto error;
1379 } else if (revents & LPOLLIN) {
1380 sock = lttcomm_accept_unix_sock(apps_sock);
1381 if (sock < 0) {
1382 goto error;
1383 }
1384
1385 /* Create UST registration command for enqueuing */
1386 ust_cmd = zmalloc(sizeof(struct ust_command));
1387 if (ust_cmd == NULL) {
1388 PERROR("ust command zmalloc");
1389 goto error;
1390 }
1391
1392 /*
1393 * Using message-based transmissions to ensure we don't
1394 * have to deal with partially received messages.
1395 */
1396 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1397 if (ret < 0) {
1398 ERR("Exhausted file descriptors allowed for applications.");
1399 free(ust_cmd);
1400 ret = close(sock);
1401 if (ret) {
1402 PERROR("close");
1403 }
1404 sock = -1;
1405 continue;
1406 }
1407 health_code_update(&health_thread_app_reg);
1408 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1409 sizeof(struct ust_register_msg));
1410 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1411 if (ret < 0) {
1412 PERROR("lttcomm_recv_unix_sock register apps");
1413 } else {
1414 ERR("Wrong size received on apps register");
1415 }
1416 free(ust_cmd);
1417 ret = close(sock);
1418 if (ret) {
1419 PERROR("close");
1420 }
1421 lttng_fd_put(LTTNG_FD_APPS, 1);
1422 sock = -1;
1423 continue;
1424 }
1425 health_code_update(&health_thread_app_reg);
1426
1427 ust_cmd->sock = sock;
1428 sock = -1;
1429
1430 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1431 " gid:%d sock:%d name:%s (version %d.%d)",
1432 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1433 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1434 ust_cmd->sock, ust_cmd->reg_msg.name,
1435 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1436
1437 /*
1438 * Lock free enqueue the registration request. The red pill
1439 * has been taken! This apps will be part of the *system*.
1440 */
1441 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1442
1443 /*
1444 * Wake the registration queue futex. Implicit memory
1445 * barrier with the exchange in cds_wfq_enqueue.
1446 */
1447 futex_nto1_wake(&ust_cmd_queue.futex);
1448 }
1449 }
1450 }
1451 }
1452
1453 exit:
1454 error:
1455 if (err) {
1456 health_error(&health_thread_app_reg);
1457 ERR("Health error occurred in %s", __func__);
1458 }
1459 health_exit(&health_thread_app_reg);
1460
1461 /* Notify that the registration thread is gone */
1462 notify_ust_apps(0);
1463
1464 if (apps_sock >= 0) {
1465 ret = close(apps_sock);
1466 if (ret) {
1467 PERROR("close");
1468 }
1469 }
1470 if (sock >= 0) {
1471 ret = close(sock);
1472 if (ret) {
1473 PERROR("close");
1474 }
1475 lttng_fd_put(LTTNG_FD_APPS, 1);
1476 }
1477 unlink(apps_unix_sock_path);
1478
1479 error_poll_add:
1480 lttng_poll_clean(&events);
1481 error_listen:
1482 error_create_poll:
1483 DBG("UST Registration thread cleanup complete");
1484
1485 return NULL;
1486 }
1487
1488 /*
1489 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1490 * exec or it will fails.
1491 */
1492 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1493 {
1494 int ret, clock_ret;
1495 struct timespec timeout;
1496
1497 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1498 consumer_data->consumer_thread_is_ready = 0;
1499
1500 /* Setup pthread condition */
1501 ret = pthread_condattr_init(&consumer_data->condattr);
1502 if (ret != 0) {
1503 errno = ret;
1504 PERROR("pthread_condattr_init consumer data");
1505 goto error;
1506 }
1507
1508 /*
1509 * Set the monotonic clock in order to make sure we DO NOT jump in time
1510 * between the clock_gettime() call and the timedwait call. See bug #324
1511 * for a more details and how we noticed it.
1512 */
1513 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1514 if (ret != 0) {
1515 errno = ret;
1516 PERROR("pthread_condattr_setclock consumer data");
1517 goto error;
1518 }
1519
1520 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1521 if (ret != 0) {
1522 errno = ret;
1523 PERROR("pthread_cond_init consumer data");
1524 goto error;
1525 }
1526
1527 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1528 consumer_data);
1529 if (ret != 0) {
1530 PERROR("pthread_create consumer");
1531 ret = -1;
1532 goto error;
1533 }
1534
1535 /* We are about to wait on a pthread condition */
1536 pthread_mutex_lock(&consumer_data->cond_mutex);
1537
1538 /* Get time for sem_timedwait absolute timeout */
1539 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1540 /*
1541 * Set the timeout for the condition timed wait even if the clock gettime
1542 * call fails since we might loop on that call and we want to avoid to
1543 * increment the timeout too many times.
1544 */
1545 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1546
1547 /*
1548 * The following loop COULD be skipped in some conditions so this is why we
1549 * set ret to 0 in order to make sure at least one round of the loop is
1550 * done.
1551 */
1552 ret = 0;
1553
1554 /*
1555 * Loop until the condition is reached or when a timeout is reached. Note
1556 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1557 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1558 * possible. This loop does not take any chances and works with both of
1559 * them.
1560 */
1561 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1562 if (clock_ret < 0) {
1563 PERROR("clock_gettime spawn consumer");
1564 /* Infinite wait for the consumerd thread to be ready */
1565 ret = pthread_cond_wait(&consumer_data->cond,
1566 &consumer_data->cond_mutex);
1567 } else {
1568 ret = pthread_cond_timedwait(&consumer_data->cond,
1569 &consumer_data->cond_mutex, &timeout);
1570 }
1571 }
1572
1573 /* Release the pthread condition */
1574 pthread_mutex_unlock(&consumer_data->cond_mutex);
1575
1576 if (ret != 0) {
1577 errno = ret;
1578 if (ret == ETIMEDOUT) {
1579 /*
1580 * Call has timed out so we kill the kconsumerd_thread and return
1581 * an error.
1582 */
1583 ERR("Condition timed out. The consumer thread was never ready."
1584 " Killing it");
1585 ret = pthread_cancel(consumer_data->thread);
1586 if (ret < 0) {
1587 PERROR("pthread_cancel consumer thread");
1588 }
1589 } else {
1590 PERROR("pthread_cond_wait failed consumer thread");
1591 }
1592 goto error;
1593 }
1594
1595 pthread_mutex_lock(&consumer_data->pid_mutex);
1596 if (consumer_data->pid == 0) {
1597 ERR("Consumerd did not start");
1598 pthread_mutex_unlock(&consumer_data->pid_mutex);
1599 goto error;
1600 }
1601 pthread_mutex_unlock(&consumer_data->pid_mutex);
1602
1603 return 0;
1604
1605 error:
1606 return ret;
1607 }
1608
1609 /*
1610 * Join consumer thread
1611 */
1612 static int join_consumer_thread(struct consumer_data *consumer_data)
1613 {
1614 void *status;
1615 int ret;
1616
1617 /* Consumer pid must be a real one. */
1618 if (consumer_data->pid > 0) {
1619 ret = kill(consumer_data->pid, SIGTERM);
1620 if (ret) {
1621 ERR("Error killing consumer daemon");
1622 return ret;
1623 }
1624 return pthread_join(consumer_data->thread, &status);
1625 } else {
1626 return 0;
1627 }
1628 }
1629
1630 /*
1631 * Fork and exec a consumer daemon (consumerd).
1632 *
1633 * Return pid if successful else -1.
1634 */
1635 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1636 {
1637 int ret;
1638 pid_t pid;
1639 const char *consumer_to_use;
1640 const char *verbosity;
1641 struct stat st;
1642
1643 DBG("Spawning consumerd");
1644
1645 pid = fork();
1646 if (pid == 0) {
1647 /*
1648 * Exec consumerd.
1649 */
1650 if (opt_verbose_consumer) {
1651 verbosity = "--verbose";
1652 } else {
1653 verbosity = "--quiet";
1654 }
1655 switch (consumer_data->type) {
1656 case LTTNG_CONSUMER_KERNEL:
1657 /*
1658 * Find out which consumerd to execute. We will first try the
1659 * 64-bit path, then the sessiond's installation directory, and
1660 * fallback on the 32-bit one,
1661 */
1662 DBG3("Looking for a kernel consumer at these locations:");
1663 DBG3(" 1) %s", consumerd64_bin);
1664 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1665 DBG3(" 3) %s", consumerd32_bin);
1666 if (stat(consumerd64_bin, &st) == 0) {
1667 DBG3("Found location #1");
1668 consumer_to_use = consumerd64_bin;
1669 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1670 DBG3("Found location #2");
1671 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1672 } else if (stat(consumerd32_bin, &st) == 0) {
1673 DBG3("Found location #3");
1674 consumer_to_use = consumerd32_bin;
1675 } else {
1676 DBG("Could not find any valid consumerd executable");
1677 break;
1678 }
1679 DBG("Using kernel consumer at: %s", consumer_to_use);
1680 execl(consumer_to_use,
1681 "lttng-consumerd", verbosity, "-k",
1682 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1683 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1684 NULL);
1685 break;
1686 case LTTNG_CONSUMER64_UST:
1687 {
1688 char *tmpnew = NULL;
1689
1690 if (consumerd64_libdir[0] != '\0') {
1691 char *tmp;
1692 size_t tmplen;
1693
1694 tmp = getenv("LD_LIBRARY_PATH");
1695 if (!tmp) {
1696 tmp = "";
1697 }
1698 tmplen = strlen("LD_LIBRARY_PATH=")
1699 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1700 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1701 if (!tmpnew) {
1702 ret = -ENOMEM;
1703 goto error;
1704 }
1705 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1706 strcat(tmpnew, consumerd64_libdir);
1707 if (tmp[0] != '\0') {
1708 strcat(tmpnew, ":");
1709 strcat(tmpnew, tmp);
1710 }
1711 ret = putenv(tmpnew);
1712 if (ret) {
1713 ret = -errno;
1714 goto error;
1715 }
1716 }
1717 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1718 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1719 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1720 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1721 NULL);
1722 if (consumerd64_libdir[0] != '\0') {
1723 free(tmpnew);
1724 }
1725 if (ret) {
1726 goto error;
1727 }
1728 break;
1729 }
1730 case LTTNG_CONSUMER32_UST:
1731 {
1732 char *tmpnew = NULL;
1733
1734 if (consumerd32_libdir[0] != '\0') {
1735 char *tmp;
1736 size_t tmplen;
1737
1738 tmp = getenv("LD_LIBRARY_PATH");
1739 if (!tmp) {
1740 tmp = "";
1741 }
1742 tmplen = strlen("LD_LIBRARY_PATH=")
1743 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1744 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1745 if (!tmpnew) {
1746 ret = -ENOMEM;
1747 goto error;
1748 }
1749 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1750 strcat(tmpnew, consumerd32_libdir);
1751 if (tmp[0] != '\0') {
1752 strcat(tmpnew, ":");
1753 strcat(tmpnew, tmp);
1754 }
1755 ret = putenv(tmpnew);
1756 if (ret) {
1757 ret = -errno;
1758 goto error;
1759 }
1760 }
1761 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1762 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1763 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1764 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1765 NULL);
1766 if (consumerd32_libdir[0] != '\0') {
1767 free(tmpnew);
1768 }
1769 if (ret) {
1770 goto error;
1771 }
1772 break;
1773 }
1774 default:
1775 PERROR("unknown consumer type");
1776 exit(EXIT_FAILURE);
1777 }
1778 if (errno != 0) {
1779 PERROR("kernel start consumer exec");
1780 }
1781 exit(EXIT_FAILURE);
1782 } else if (pid > 0) {
1783 ret = pid;
1784 } else {
1785 PERROR("start consumer fork");
1786 ret = -errno;
1787 }
1788 error:
1789 return ret;
1790 }
1791
1792 /*
1793 * Spawn the consumerd daemon and session daemon thread.
1794 */
1795 static int start_consumerd(struct consumer_data *consumer_data)
1796 {
1797 int ret;
1798
1799 pthread_mutex_lock(&consumer_data->pid_mutex);
1800 if (consumer_data->pid != 0) {
1801 pthread_mutex_unlock(&consumer_data->pid_mutex);
1802 goto end;
1803 }
1804
1805 ret = spawn_consumerd(consumer_data);
1806 if (ret < 0) {
1807 ERR("Spawning consumerd failed");
1808 pthread_mutex_unlock(&consumer_data->pid_mutex);
1809 goto error;
1810 }
1811
1812 /* Setting up the consumer_data pid */
1813 consumer_data->pid = ret;
1814 DBG2("Consumer pid %d", consumer_data->pid);
1815 pthread_mutex_unlock(&consumer_data->pid_mutex);
1816
1817 DBG2("Spawning consumer control thread");
1818 ret = spawn_consumer_thread(consumer_data);
1819 if (ret < 0) {
1820 ERR("Fatal error spawning consumer control thread");
1821 goto error;
1822 }
1823
1824 end:
1825 return 0;
1826
1827 error:
1828 return ret;
1829 }
1830
1831 /*
1832 * Compute health status of each consumer. If one of them is zero (bad
1833 * state), we return 0.
1834 */
1835 static int check_consumer_health(void)
1836 {
1837 int ret;
1838
1839 ret = health_check_state(&kconsumer_data.health) &&
1840 health_check_state(&ustconsumer32_data.health) &&
1841 health_check_state(&ustconsumer64_data.health);
1842
1843 DBG3("Health consumer check %d", ret);
1844
1845 return ret;
1846 }
1847
1848 /*
1849 * Setup necessary data for kernel tracer action.
1850 */
1851 static int init_kernel_tracer(void)
1852 {
1853 int ret;
1854
1855 /* Modprobe lttng kernel modules */
1856 ret = modprobe_lttng_control();
1857 if (ret < 0) {
1858 goto error;
1859 }
1860
1861 /* Open debugfs lttng */
1862 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1863 if (kernel_tracer_fd < 0) {
1864 DBG("Failed to open %s", module_proc_lttng);
1865 ret = -1;
1866 goto error_open;
1867 }
1868
1869 /* Validate kernel version */
1870 ret = kernel_validate_version(kernel_tracer_fd);
1871 if (ret < 0) {
1872 goto error_version;
1873 }
1874
1875 ret = modprobe_lttng_data();
1876 if (ret < 0) {
1877 goto error_modules;
1878 }
1879
1880 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1881 return 0;
1882
1883 error_version:
1884 modprobe_remove_lttng_control();
1885 ret = close(kernel_tracer_fd);
1886 if (ret) {
1887 PERROR("close");
1888 }
1889 kernel_tracer_fd = -1;
1890 return LTTNG_ERR_KERN_VERSION;
1891
1892 error_modules:
1893 ret = close(kernel_tracer_fd);
1894 if (ret) {
1895 PERROR("close");
1896 }
1897
1898 error_open:
1899 modprobe_remove_lttng_control();
1900
1901 error:
1902 WARN("No kernel tracer available");
1903 kernel_tracer_fd = -1;
1904 if (!is_root) {
1905 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1906 } else {
1907 return LTTNG_ERR_KERN_NA;
1908 }
1909 }
1910
1911
1912 /*
1913 * Copy consumer output from the tracing session to the domain session. The
1914 * function also applies the right modification on a per domain basis for the
1915 * trace files destination directory.
1916 */
1917 static int copy_session_consumer(int domain, struct ltt_session *session)
1918 {
1919 int ret;
1920 const char *dir_name;
1921 struct consumer_output *consumer;
1922
1923 assert(session);
1924 assert(session->consumer);
1925
1926 switch (domain) {
1927 case LTTNG_DOMAIN_KERNEL:
1928 DBG3("Copying tracing session consumer output in kernel session");
1929 session->kernel_session->consumer =
1930 consumer_copy_output(session->consumer);
1931 /* Ease our life a bit for the next part */
1932 consumer = session->kernel_session->consumer;
1933 dir_name = DEFAULT_KERNEL_TRACE_DIR;
1934 break;
1935 case LTTNG_DOMAIN_UST:
1936 DBG3("Copying tracing session consumer output in UST session");
1937 session->ust_session->consumer =
1938 consumer_copy_output(session->consumer);
1939 /* Ease our life a bit for the next part */
1940 consumer = session->ust_session->consumer;
1941 dir_name = DEFAULT_UST_TRACE_DIR;
1942 break;
1943 default:
1944 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
1945 goto error;
1946 }
1947
1948 /* Append correct directory to subdir */
1949 strncat(consumer->subdir, dir_name,
1950 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
1951 DBG3("Copy session consumer subdir %s", consumer->subdir);
1952
1953 ret = LTTNG_OK;
1954
1955 error:
1956 return ret;
1957 }
1958
1959 /*
1960 * Create an UST session and add it to the session ust list.
1961 */
1962 static int create_ust_session(struct ltt_session *session,
1963 struct lttng_domain *domain)
1964 {
1965 int ret;
1966 struct ltt_ust_session *lus = NULL;
1967
1968 assert(session);
1969 assert(domain);
1970 assert(session->consumer);
1971
1972 switch (domain->type) {
1973 case LTTNG_DOMAIN_UST:
1974 break;
1975 default:
1976 ERR("Unknown UST domain on create session %d", domain->type);
1977 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
1978 goto error;
1979 }
1980
1981 DBG("Creating UST session");
1982
1983 lus = trace_ust_create_session(session->path, session->id, domain);
1984 if (lus == NULL) {
1985 ret = LTTNG_ERR_UST_SESS_FAIL;
1986 goto error;
1987 }
1988
1989 lus->uid = session->uid;
1990 lus->gid = session->gid;
1991 session->ust_session = lus;
1992
1993 /* Copy session output to the newly created UST session */
1994 ret = copy_session_consumer(domain->type, session);
1995 if (ret != LTTNG_OK) {
1996 goto error;
1997 }
1998
1999 return LTTNG_OK;
2000
2001 error:
2002 free(lus);
2003 session->ust_session = NULL;
2004 return ret;
2005 }
2006
2007 /*
2008 * Create a kernel tracer session then create the default channel.
2009 */
2010 static int create_kernel_session(struct ltt_session *session)
2011 {
2012 int ret;
2013
2014 DBG("Creating kernel session");
2015
2016 ret = kernel_create_session(session, kernel_tracer_fd);
2017 if (ret < 0) {
2018 ret = LTTNG_ERR_KERN_SESS_FAIL;
2019 goto error;
2020 }
2021
2022 /* Code flow safety */
2023 assert(session->kernel_session);
2024
2025 /* Copy session output to the newly created Kernel session */
2026 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2027 if (ret != LTTNG_OK) {
2028 goto error;
2029 }
2030
2031 /* Create directory(ies) on local filesystem. */
2032 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2033 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2034 ret = run_as_mkdir_recursive(
2035 session->kernel_session->consumer->dst.trace_path,
2036 S_IRWXU | S_IRWXG, session->uid, session->gid);
2037 if (ret < 0) {
2038 if (ret != -EEXIST) {
2039 ERR("Trace directory creation error");
2040 goto error;
2041 }
2042 }
2043 }
2044
2045 session->kernel_session->uid = session->uid;
2046 session->kernel_session->gid = session->gid;
2047
2048 return LTTNG_OK;
2049
2050 error:
2051 trace_kernel_destroy_session(session->kernel_session);
2052 session->kernel_session = NULL;
2053 return ret;
2054 }
2055
2056 /*
2057 * Count number of session permitted by uid/gid.
2058 */
2059 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2060 {
2061 unsigned int i = 0;
2062 struct ltt_session *session;
2063
2064 DBG("Counting number of available session for UID %d GID %d",
2065 uid, gid);
2066 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2067 /*
2068 * Only list the sessions the user can control.
2069 */
2070 if (!session_access_ok(session, uid, gid)) {
2071 continue;
2072 }
2073 i++;
2074 }
2075 return i;
2076 }
2077
2078 /*
2079 * Process the command requested by the lttng client within the command
2080 * context structure. This function make sure that the return structure (llm)
2081 * is set and ready for transmission before returning.
2082 *
2083 * Return any error encountered or 0 for success.
2084 *
2085 * "sock" is only used for special-case var. len data.
2086 */
2087 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2088 int *sock_error)
2089 {
2090 int ret = LTTNG_OK;
2091 int need_tracing_session = 1;
2092 int need_domain;
2093
2094 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2095
2096 *sock_error = 0;
2097
2098 switch (cmd_ctx->lsm->cmd_type) {
2099 case LTTNG_CREATE_SESSION:
2100 case LTTNG_DESTROY_SESSION:
2101 case LTTNG_LIST_SESSIONS:
2102 case LTTNG_LIST_DOMAINS:
2103 case LTTNG_START_TRACE:
2104 case LTTNG_STOP_TRACE:
2105 need_domain = 0;
2106 break;
2107 default:
2108 need_domain = 1;
2109 }
2110
2111 if (opt_no_kernel && need_domain
2112 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2113 if (!is_root) {
2114 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2115 } else {
2116 ret = LTTNG_ERR_KERN_NA;
2117 }
2118 goto error;
2119 }
2120
2121 /* Deny register consumer if we already have a spawned consumer. */
2122 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2123 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2124 if (kconsumer_data.pid > 0) {
2125 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2126 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2127 goto error;
2128 }
2129 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2130 }
2131
2132 /*
2133 * Check for command that don't needs to allocate a returned payload. We do
2134 * this here so we don't have to make the call for no payload at each
2135 * command.
2136 */
2137 switch(cmd_ctx->lsm->cmd_type) {
2138 case LTTNG_LIST_SESSIONS:
2139 case LTTNG_LIST_TRACEPOINTS:
2140 case LTTNG_LIST_TRACEPOINT_FIELDS:
2141 case LTTNG_LIST_DOMAINS:
2142 case LTTNG_LIST_CHANNELS:
2143 case LTTNG_LIST_EVENTS:
2144 break;
2145 default:
2146 /* Setup lttng message with no payload */
2147 ret = setup_lttng_msg(cmd_ctx, 0);
2148 if (ret < 0) {
2149 /* This label does not try to unlock the session */
2150 goto init_setup_error;
2151 }
2152 }
2153
2154 /* Commands that DO NOT need a session. */
2155 switch (cmd_ctx->lsm->cmd_type) {
2156 case LTTNG_CREATE_SESSION:
2157 case LTTNG_CALIBRATE:
2158 case LTTNG_LIST_SESSIONS:
2159 case LTTNG_LIST_TRACEPOINTS:
2160 case LTTNG_LIST_TRACEPOINT_FIELDS:
2161 need_tracing_session = 0;
2162 break;
2163 default:
2164 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2165 /*
2166 * We keep the session list lock across _all_ commands
2167 * for now, because the per-session lock does not
2168 * handle teardown properly.
2169 */
2170 session_lock_list();
2171 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2172 if (cmd_ctx->session == NULL) {
2173 if (cmd_ctx->lsm->session.name != NULL) {
2174 ret = LTTNG_ERR_SESS_NOT_FOUND;
2175 } else {
2176 /* If no session name specified */
2177 ret = LTTNG_ERR_SELECT_SESS;
2178 }
2179 goto error;
2180 } else {
2181 /* Acquire lock for the session */
2182 session_lock(cmd_ctx->session);
2183 }
2184 break;
2185 }
2186
2187 if (!need_domain) {
2188 goto skip_domain;
2189 }
2190
2191 /*
2192 * Check domain type for specific "pre-action".
2193 */
2194 switch (cmd_ctx->lsm->domain.type) {
2195 case LTTNG_DOMAIN_KERNEL:
2196 if (!is_root) {
2197 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2198 goto error;
2199 }
2200
2201 /* Kernel tracer check */
2202 if (kernel_tracer_fd == -1) {
2203 /* Basically, load kernel tracer modules */
2204 ret = init_kernel_tracer();
2205 if (ret != 0) {
2206 goto error;
2207 }
2208 }
2209
2210 /* Consumer is in an ERROR state. Report back to client */
2211 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2212 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2213 goto error;
2214 }
2215
2216 /* Need a session for kernel command */
2217 if (need_tracing_session) {
2218 if (cmd_ctx->session->kernel_session == NULL) {
2219 ret = create_kernel_session(cmd_ctx->session);
2220 if (ret < 0) {
2221 ret = LTTNG_ERR_KERN_SESS_FAIL;
2222 goto error;
2223 }
2224 }
2225
2226 /* Start the kernel consumer daemon */
2227 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2228 if (kconsumer_data.pid == 0 &&
2229 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2230 cmd_ctx->session->start_consumer) {
2231 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2232 ret = start_consumerd(&kconsumer_data);
2233 if (ret < 0) {
2234 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2235 goto error;
2236 }
2237 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2238 } else {
2239 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2240 }
2241
2242 /*
2243 * The consumer was just spawned so we need to add the socket to
2244 * the consumer output of the session if exist.
2245 */
2246 ret = consumer_create_socket(&kconsumer_data,
2247 cmd_ctx->session->kernel_session->consumer);
2248 if (ret < 0) {
2249 goto error;
2250 }
2251 }
2252
2253 break;
2254 case LTTNG_DOMAIN_UST:
2255 {
2256 /* Consumer is in an ERROR state. Report back to client */
2257 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2258 ret = LTTNG_ERR_NO_USTCONSUMERD;
2259 goto error;
2260 }
2261
2262 if (need_tracing_session) {
2263 /* Create UST session if none exist. */
2264 if (cmd_ctx->session->ust_session == NULL) {
2265 ret = create_ust_session(cmd_ctx->session,
2266 &cmd_ctx->lsm->domain);
2267 if (ret != LTTNG_OK) {
2268 goto error;
2269 }
2270 }
2271
2272 /* Start the UST consumer daemons */
2273 /* 64-bit */
2274 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2275 if (consumerd64_bin[0] != '\0' &&
2276 ustconsumer64_data.pid == 0 &&
2277 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2278 cmd_ctx->session->start_consumer) {
2279 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2280 ret = start_consumerd(&ustconsumer64_data);
2281 if (ret < 0) {
2282 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2283 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2284 goto error;
2285 }
2286
2287 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2288 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2289 } else {
2290 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2291 }
2292
2293 /*
2294 * Setup socket for consumer 64 bit. No need for atomic access
2295 * since it was set above and can ONLY be set in this thread.
2296 */
2297 ret = consumer_create_socket(&ustconsumer64_data,
2298 cmd_ctx->session->ust_session->consumer);
2299 if (ret < 0) {
2300 goto error;
2301 }
2302
2303 /* 32-bit */
2304 if (consumerd32_bin[0] != '\0' &&
2305 ustconsumer32_data.pid == 0 &&
2306 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2307 cmd_ctx->session->start_consumer) {
2308 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2309 ret = start_consumerd(&ustconsumer32_data);
2310 if (ret < 0) {
2311 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2312 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2313 goto error;
2314 }
2315
2316 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2317 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2318 } else {
2319 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2320 }
2321
2322 /*
2323 * Setup socket for consumer 64 bit. No need for atomic access
2324 * since it was set above and can ONLY be set in this thread.
2325 */
2326 ret = consumer_create_socket(&ustconsumer32_data,
2327 cmd_ctx->session->ust_session->consumer);
2328 if (ret < 0) {
2329 goto error;
2330 }
2331 }
2332 break;
2333 }
2334 default:
2335 break;
2336 }
2337 skip_domain:
2338
2339 /* Validate consumer daemon state when start/stop trace command */
2340 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2341 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2342 switch (cmd_ctx->lsm->domain.type) {
2343 case LTTNG_DOMAIN_UST:
2344 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2345 ret = LTTNG_ERR_NO_USTCONSUMERD;
2346 goto error;
2347 }
2348 break;
2349 case LTTNG_DOMAIN_KERNEL:
2350 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2351 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2352 goto error;
2353 }
2354 break;
2355 }
2356 }
2357
2358 /*
2359 * Check that the UID or GID match that of the tracing session.
2360 * The root user can interact with all sessions.
2361 */
2362 if (need_tracing_session) {
2363 if (!session_access_ok(cmd_ctx->session,
2364 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2365 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2366 ret = LTTNG_ERR_EPERM;
2367 goto error;
2368 }
2369 }
2370
2371 /* Process by command type */
2372 switch (cmd_ctx->lsm->cmd_type) {
2373 case LTTNG_ADD_CONTEXT:
2374 {
2375 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2376 cmd_ctx->lsm->u.context.channel_name,
2377 cmd_ctx->lsm->u.context.event_name,
2378 &cmd_ctx->lsm->u.context.ctx);
2379 break;
2380 }
2381 case LTTNG_DISABLE_CHANNEL:
2382 {
2383 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2384 cmd_ctx->lsm->u.disable.channel_name);
2385 break;
2386 }
2387 case LTTNG_DISABLE_EVENT:
2388 {
2389 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2390 cmd_ctx->lsm->u.disable.channel_name,
2391 cmd_ctx->lsm->u.disable.name);
2392 break;
2393 }
2394 case LTTNG_DISABLE_ALL_EVENT:
2395 {
2396 DBG("Disabling all events");
2397
2398 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2399 cmd_ctx->lsm->u.disable.channel_name);
2400 break;
2401 }
2402 case LTTNG_DISABLE_CONSUMER:
2403 {
2404 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2405 break;
2406 }
2407 case LTTNG_ENABLE_CHANNEL:
2408 {
2409 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2410 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2411 break;
2412 }
2413 case LTTNG_ENABLE_CONSUMER:
2414 {
2415 /*
2416 * XXX: 0 means that this URI should be applied on the session. Should
2417 * be a DOMAIN enuam.
2418 */
2419 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2420 if (ret != LTTNG_OK) {
2421 goto error;
2422 }
2423
2424 if (cmd_ctx->lsm->domain.type == 0) {
2425 /* Add the URI for the UST session if a consumer is present. */
2426 if (cmd_ctx->session->ust_session &&
2427 cmd_ctx->session->ust_session->consumer) {
2428 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2429 } else if (cmd_ctx->session->kernel_session &&
2430 cmd_ctx->session->kernel_session->consumer) {
2431 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2432 cmd_ctx->session);
2433 }
2434 }
2435 break;
2436 }
2437 case LTTNG_ENABLE_EVENT:
2438 {
2439 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2440 cmd_ctx->lsm->u.enable.channel_name,
2441 &cmd_ctx->lsm->u.enable.event, kernel_poll_pipe[1]);
2442 break;
2443 }
2444 case LTTNG_ENABLE_ALL_EVENT:
2445 {
2446 DBG("Enabling all events");
2447
2448 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2449 cmd_ctx->lsm->u.enable.channel_name,
2450 cmd_ctx->lsm->u.enable.event.type, kernel_poll_pipe[1]);
2451 break;
2452 }
2453 case LTTNG_LIST_TRACEPOINTS:
2454 {
2455 struct lttng_event *events;
2456 ssize_t nb_events;
2457
2458 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2459 if (nb_events < 0) {
2460 /* Return value is a negative lttng_error_code. */
2461 ret = -nb_events;
2462 goto error;
2463 }
2464
2465 /*
2466 * Setup lttng message with payload size set to the event list size in
2467 * bytes and then copy list into the llm payload.
2468 */
2469 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2470 if (ret < 0) {
2471 free(events);
2472 goto setup_error;
2473 }
2474
2475 /* Copy event list into message payload */
2476 memcpy(cmd_ctx->llm->payload, events,
2477 sizeof(struct lttng_event) * nb_events);
2478
2479 free(events);
2480
2481 ret = LTTNG_OK;
2482 break;
2483 }
2484 case LTTNG_LIST_TRACEPOINT_FIELDS:
2485 {
2486 struct lttng_event_field *fields;
2487 ssize_t nb_fields;
2488
2489 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2490 &fields);
2491 if (nb_fields < 0) {
2492 /* Return value is a negative lttng_error_code. */
2493 ret = -nb_fields;
2494 goto error;
2495 }
2496
2497 /*
2498 * Setup lttng message with payload size set to the event list size in
2499 * bytes and then copy list into the llm payload.
2500 */
2501 ret = setup_lttng_msg(cmd_ctx,
2502 sizeof(struct lttng_event_field) * nb_fields);
2503 if (ret < 0) {
2504 free(fields);
2505 goto setup_error;
2506 }
2507
2508 /* Copy event list into message payload */
2509 memcpy(cmd_ctx->llm->payload, fields,
2510 sizeof(struct lttng_event_field) * nb_fields);
2511
2512 free(fields);
2513
2514 ret = LTTNG_OK;
2515 break;
2516 }
2517 case LTTNG_SET_CONSUMER_URI:
2518 {
2519 size_t nb_uri, len;
2520 struct lttng_uri *uris;
2521
2522 nb_uri = cmd_ctx->lsm->u.uri.size;
2523 len = nb_uri * sizeof(struct lttng_uri);
2524
2525 if (nb_uri == 0) {
2526 ret = LTTNG_ERR_INVALID;
2527 goto error;
2528 }
2529
2530 uris = zmalloc(len);
2531 if (uris == NULL) {
2532 ret = LTTNG_ERR_FATAL;
2533 goto error;
2534 }
2535
2536 /* Receive variable len data */
2537 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2538 ret = lttcomm_recv_unix_sock(sock, uris, len);
2539 if (ret <= 0) {
2540 DBG("No URIs received from client... continuing");
2541 *sock_error = 1;
2542 ret = LTTNG_ERR_SESSION_FAIL;
2543 goto error;
2544 }
2545
2546 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2547 nb_uri, uris);
2548 if (ret != LTTNG_OK) {
2549 goto error;
2550 }
2551
2552 /*
2553 * XXX: 0 means that this URI should be applied on the session. Should
2554 * be a DOMAIN enuam.
2555 */
2556 if (cmd_ctx->lsm->domain.type == 0) {
2557 /* Add the URI for the UST session if a consumer is present. */
2558 if (cmd_ctx->session->ust_session &&
2559 cmd_ctx->session->ust_session->consumer) {
2560 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2561 nb_uri, uris);
2562 } else if (cmd_ctx->session->kernel_session &&
2563 cmd_ctx->session->kernel_session->consumer) {
2564 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2565 cmd_ctx->session, nb_uri, uris);
2566 }
2567 }
2568
2569 break;
2570 }
2571 case LTTNG_START_TRACE:
2572 {
2573 ret = cmd_start_trace(cmd_ctx->session);
2574 break;
2575 }
2576 case LTTNG_STOP_TRACE:
2577 {
2578 ret = cmd_stop_trace(cmd_ctx->session);
2579 break;
2580 }
2581 case LTTNG_CREATE_SESSION:
2582 {
2583 size_t nb_uri, len;
2584 struct lttng_uri *uris = NULL;
2585
2586 nb_uri = cmd_ctx->lsm->u.uri.size;
2587 len = nb_uri * sizeof(struct lttng_uri);
2588
2589 if (nb_uri > 0) {
2590 uris = zmalloc(len);
2591 if (uris == NULL) {
2592 ret = LTTNG_ERR_FATAL;
2593 goto error;
2594 }
2595
2596 /* Receive variable len data */
2597 DBG("Waiting for %zu URIs from client ...", nb_uri);
2598 ret = lttcomm_recv_unix_sock(sock, uris, len);
2599 if (ret <= 0) {
2600 DBG("No URIs received from client... continuing");
2601 *sock_error = 1;
2602 ret = LTTNG_ERR_SESSION_FAIL;
2603 goto error;
2604 }
2605
2606 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2607 DBG("Creating session with ONE network URI is a bad call");
2608 ret = LTTNG_ERR_SESSION_FAIL;
2609 goto error;
2610 }
2611 }
2612
2613 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2614 &cmd_ctx->creds);
2615
2616 break;
2617 }
2618 case LTTNG_DESTROY_SESSION:
2619 {
2620 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2621
2622 /* Set session to NULL so we do not unlock it after free. */
2623 cmd_ctx->session = NULL;
2624 break;
2625 }
2626 case LTTNG_LIST_DOMAINS:
2627 {
2628 ssize_t nb_dom;
2629 struct lttng_domain *domains;
2630
2631 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2632 if (nb_dom < 0) {
2633 /* Return value is a negative lttng_error_code. */
2634 ret = -nb_dom;
2635 goto error;
2636 }
2637
2638 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2639 if (ret < 0) {
2640 goto setup_error;
2641 }
2642
2643 /* Copy event list into message payload */
2644 memcpy(cmd_ctx->llm->payload, domains,
2645 nb_dom * sizeof(struct lttng_domain));
2646
2647 free(domains);
2648
2649 ret = LTTNG_OK;
2650 break;
2651 }
2652 case LTTNG_LIST_CHANNELS:
2653 {
2654 int nb_chan;
2655 struct lttng_channel *channels;
2656
2657 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2658 cmd_ctx->session, &channels);
2659 if (nb_chan < 0) {
2660 /* Return value is a negative lttng_error_code. */
2661 ret = -nb_chan;
2662 goto error;
2663 }
2664
2665 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2666 if (ret < 0) {
2667 goto setup_error;
2668 }
2669
2670 /* Copy event list into message payload */
2671 memcpy(cmd_ctx->llm->payload, channels,
2672 nb_chan * sizeof(struct lttng_channel));
2673
2674 free(channels);
2675
2676 ret = LTTNG_OK;
2677 break;
2678 }
2679 case LTTNG_LIST_EVENTS:
2680 {
2681 ssize_t nb_event;
2682 struct lttng_event *events = NULL;
2683
2684 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2685 cmd_ctx->lsm->u.list.channel_name, &events);
2686 if (nb_event < 0) {
2687 /* Return value is a negative lttng_error_code. */
2688 ret = -nb_event;
2689 goto error;
2690 }
2691
2692 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2693 if (ret < 0) {
2694 goto setup_error;
2695 }
2696
2697 /* Copy event list into message payload */
2698 memcpy(cmd_ctx->llm->payload, events,
2699 nb_event * sizeof(struct lttng_event));
2700
2701 free(events);
2702
2703 ret = LTTNG_OK;
2704 break;
2705 }
2706 case LTTNG_LIST_SESSIONS:
2707 {
2708 unsigned int nr_sessions;
2709
2710 session_lock_list();
2711 nr_sessions = lttng_sessions_count(
2712 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2713 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2714
2715 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2716 if (ret < 0) {
2717 session_unlock_list();
2718 goto setup_error;
2719 }
2720
2721 /* Filled the session array */
2722 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2723 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2724 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2725
2726 session_unlock_list();
2727
2728 ret = LTTNG_OK;
2729 break;
2730 }
2731 case LTTNG_CALIBRATE:
2732 {
2733 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2734 &cmd_ctx->lsm->u.calibrate);
2735 break;
2736 }
2737 case LTTNG_REGISTER_CONSUMER:
2738 {
2739 struct consumer_data *cdata;
2740
2741 switch (cmd_ctx->lsm->domain.type) {
2742 case LTTNG_DOMAIN_KERNEL:
2743 cdata = &kconsumer_data;
2744 break;
2745 default:
2746 ret = LTTNG_ERR_UND;
2747 goto error;
2748 }
2749
2750 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2751 cmd_ctx->lsm->u.reg.path, cdata);
2752 break;
2753 }
2754 case LTTNG_SET_FILTER:
2755 {
2756 struct lttng_filter_bytecode *bytecode;
2757
2758 if (cmd_ctx->lsm->u.filter.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2759 ret = LTTNG_ERR_FILTER_INVAL;
2760 goto error;
2761 }
2762 bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len);
2763 if (!bytecode) {
2764 ret = LTTNG_ERR_FILTER_NOMEM;
2765 goto error;
2766 }
2767 /* Receive var. len. data */
2768 DBG("Receiving var len data from client ...");
2769 ret = lttcomm_recv_unix_sock(sock, bytecode,
2770 cmd_ctx->lsm->u.filter.bytecode_len);
2771 if (ret <= 0) {
2772 DBG("Nothing recv() from client var len data... continuing");
2773 *sock_error = 1;
2774 ret = LTTNG_ERR_FILTER_INVAL;
2775 goto error;
2776 }
2777
2778 if (bytecode->len + sizeof(*bytecode)
2779 != cmd_ctx->lsm->u.filter.bytecode_len) {
2780 free(bytecode);
2781 ret = LTTNG_ERR_FILTER_INVAL;
2782 goto error;
2783 }
2784
2785 ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2786 cmd_ctx->lsm->u.filter.channel_name,
2787 cmd_ctx->lsm->u.filter.event_name,
2788 bytecode);
2789 break;
2790 }
2791 default:
2792 ret = LTTNG_ERR_UND;
2793 break;
2794 }
2795
2796 error:
2797 if (cmd_ctx->llm == NULL) {
2798 DBG("Missing llm structure. Allocating one.");
2799 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2800 goto setup_error;
2801 }
2802 }
2803 /* Set return code */
2804 cmd_ctx->llm->ret_code = ret;
2805 setup_error:
2806 if (cmd_ctx->session) {
2807 session_unlock(cmd_ctx->session);
2808 }
2809 if (need_tracing_session) {
2810 session_unlock_list();
2811 }
2812 init_setup_error:
2813 return ret;
2814 }
2815
2816 /*
2817 * Thread managing health check socket.
2818 */
2819 static void *thread_manage_health(void *data)
2820 {
2821 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2822 uint32_t revents, nb_fd;
2823 struct lttng_poll_event events;
2824 struct lttcomm_health_msg msg;
2825 struct lttcomm_health_data reply;
2826
2827 DBG("[thread] Manage health check started");
2828
2829 rcu_register_thread();
2830
2831 /* Create unix socket */
2832 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2833 if (sock < 0) {
2834 ERR("Unable to create health check Unix socket");
2835 ret = -1;
2836 goto error;
2837 }
2838
2839 ret = lttcomm_listen_unix_sock(sock);
2840 if (ret < 0) {
2841 goto error;
2842 }
2843
2844 /*
2845 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2846 * more will be added to this poll set.
2847 */
2848 ret = create_thread_poll_set(&events, 2);
2849 if (ret < 0) {
2850 goto error;
2851 }
2852
2853 /* Add the application registration socket */
2854 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2855 if (ret < 0) {
2856 goto error;
2857 }
2858
2859 while (1) {
2860 DBG("Health check ready");
2861
2862 nb_fd = LTTNG_POLL_GETNB(&events);
2863
2864 /* Inifinite blocking call, waiting for transmission */
2865 restart:
2866 ret = lttng_poll_wait(&events, -1);
2867 if (ret < 0) {
2868 /*
2869 * Restart interrupted system call.
2870 */
2871 if (errno == EINTR) {
2872 goto restart;
2873 }
2874 goto error;
2875 }
2876
2877 for (i = 0; i < nb_fd; i++) {
2878 /* Fetch once the poll data */
2879 revents = LTTNG_POLL_GETEV(&events, i);
2880 pollfd = LTTNG_POLL_GETFD(&events, i);
2881
2882 /* Thread quit pipe has been closed. Killing thread. */
2883 ret = check_thread_quit_pipe(pollfd, revents);
2884 if (ret) {
2885 err = 0;
2886 goto exit;
2887 }
2888
2889 /* Event on the registration socket */
2890 if (pollfd == sock) {
2891 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2892 ERR("Health socket poll error");
2893 goto error;
2894 }
2895 }
2896 }
2897
2898 new_sock = lttcomm_accept_unix_sock(sock);
2899 if (new_sock < 0) {
2900 goto error;
2901 }
2902
2903 DBG("Receiving data from client for health...");
2904 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
2905 if (ret <= 0) {
2906 DBG("Nothing recv() from client... continuing");
2907 ret = close(new_sock);
2908 if (ret) {
2909 PERROR("close");
2910 }
2911 new_sock = -1;
2912 continue;
2913 }
2914
2915 rcu_thread_online();
2916
2917 switch (msg.component) {
2918 case LTTNG_HEALTH_CMD:
2919 reply.ret_code = health_check_state(&health_thread_cmd);
2920 break;
2921 case LTTNG_HEALTH_APP_MANAGE:
2922 reply.ret_code = health_check_state(&health_thread_app_manage);
2923 break;
2924 case LTTNG_HEALTH_APP_REG:
2925 reply.ret_code = health_check_state(&health_thread_app_reg);
2926 break;
2927 case LTTNG_HEALTH_KERNEL:
2928 reply.ret_code = health_check_state(&health_thread_kernel);
2929 break;
2930 case LTTNG_HEALTH_CONSUMER:
2931 reply.ret_code = check_consumer_health();
2932 break;
2933 case LTTNG_HEALTH_ALL:
2934 reply.ret_code =
2935 health_check_state(&health_thread_app_manage) &&
2936 health_check_state(&health_thread_app_reg) &&
2937 health_check_state(&health_thread_cmd) &&
2938 health_check_state(&health_thread_kernel) &&
2939 check_consumer_health();
2940 break;
2941 default:
2942 reply.ret_code = LTTNG_ERR_UND;
2943 break;
2944 }
2945
2946 /*
2947 * Flip ret value since 0 is a success and 1 indicates a bad health for
2948 * the client where in the sessiond it is the opposite. Again, this is
2949 * just to make things easier for us poor developer which enjoy a lot
2950 * lazyness.
2951 */
2952 if (reply.ret_code == 0 || reply.ret_code == 1) {
2953 reply.ret_code = !reply.ret_code;
2954 }
2955
2956 DBG2("Health check return value %d", reply.ret_code);
2957
2958 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
2959 if (ret < 0) {
2960 ERR("Failed to send health data back to client");
2961 }
2962
2963 /* End of transmission */
2964 ret = close(new_sock);
2965 if (ret) {
2966 PERROR("close");
2967 }
2968 new_sock = -1;
2969 }
2970
2971 exit:
2972 error:
2973 if (err) {
2974 ERR("Health error occurred in %s", __func__);
2975 }
2976 DBG("Health check thread dying");
2977 unlink(health_unix_sock_path);
2978 if (sock >= 0) {
2979 ret = close(sock);
2980 if (ret) {
2981 PERROR("close");
2982 }
2983 }
2984 if (new_sock >= 0) {
2985 ret = close(new_sock);
2986 if (ret) {
2987 PERROR("close");
2988 }
2989 }
2990
2991 lttng_poll_clean(&events);
2992
2993 rcu_unregister_thread();
2994 return NULL;
2995 }
2996
2997 /*
2998 * This thread manage all clients request using the unix client socket for
2999 * communication.
3000 */
3001 static void *thread_manage_clients(void *data)
3002 {
3003 int sock = -1, ret, i, pollfd, err = -1;
3004 int sock_error;
3005 uint32_t revents, nb_fd;
3006 struct command_ctx *cmd_ctx = NULL;
3007 struct lttng_poll_event events;
3008
3009 DBG("[thread] Manage client started");
3010
3011 testpoint(thread_manage_clients);
3012
3013 rcu_register_thread();
3014
3015 health_code_update(&health_thread_cmd);
3016
3017 ret = lttcomm_listen_unix_sock(client_sock);
3018 if (ret < 0) {
3019 goto error;
3020 }
3021
3022 /*
3023 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3024 * more will be added to this poll set.
3025 */
3026 ret = create_thread_poll_set(&events, 2);
3027 if (ret < 0) {
3028 goto error;
3029 }
3030
3031 /* Add the application registration socket */
3032 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3033 if (ret < 0) {
3034 goto error;
3035 }
3036
3037 /*
3038 * Notify parent pid that we are ready to accept command for client side.
3039 */
3040 if (opt_sig_parent) {
3041 kill(ppid, SIGUSR1);
3042 }
3043
3044 testpoint(thread_manage_clients_before_loop);
3045
3046 health_code_update(&health_thread_cmd);
3047
3048 while (1) {
3049 DBG("Accepting client command ...");
3050
3051 nb_fd = LTTNG_POLL_GETNB(&events);
3052
3053 /* Inifinite blocking call, waiting for transmission */
3054 restart:
3055 health_poll_update(&health_thread_cmd);
3056 ret = lttng_poll_wait(&events, -1);
3057 health_poll_update(&health_thread_cmd);
3058 if (ret < 0) {
3059 /*
3060 * Restart interrupted system call.
3061 */
3062 if (errno == EINTR) {
3063 goto restart;
3064 }
3065 goto error;
3066 }
3067
3068 for (i = 0; i < nb_fd; i++) {
3069 /* Fetch once the poll data */
3070 revents = LTTNG_POLL_GETEV(&events, i);
3071 pollfd = LTTNG_POLL_GETFD(&events, i);
3072
3073 health_code_update(&health_thread_cmd);
3074
3075 /* Thread quit pipe has been closed. Killing thread. */
3076 ret = check_thread_quit_pipe(pollfd, revents);
3077 if (ret) {
3078 err = 0;
3079 goto exit;
3080 }
3081
3082 /* Event on the registration socket */
3083 if (pollfd == client_sock) {
3084 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3085 ERR("Client socket poll error");
3086 goto error;
3087 }
3088 }
3089 }
3090
3091 DBG("Wait for client response");
3092
3093 health_code_update(&health_thread_cmd);
3094
3095 sock = lttcomm_accept_unix_sock(client_sock);
3096 if (sock < 0) {
3097 goto error;
3098 }
3099
3100 /* Set socket option for credentials retrieval */
3101 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3102 if (ret < 0) {
3103 goto error;
3104 }
3105
3106 /* Allocate context command to process the client request */
3107 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3108 if (cmd_ctx == NULL) {
3109 PERROR("zmalloc cmd_ctx");
3110 goto error;
3111 }
3112
3113 /* Allocate data buffer for reception */
3114 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3115 if (cmd_ctx->lsm == NULL) {
3116 PERROR("zmalloc cmd_ctx->lsm");
3117 goto error;
3118 }
3119
3120 cmd_ctx->llm = NULL;
3121 cmd_ctx->session = NULL;
3122
3123 health_code_update(&health_thread_cmd);
3124
3125 /*
3126 * Data is received from the lttng client. The struct
3127 * lttcomm_session_msg (lsm) contains the command and data request of
3128 * the client.
3129 */
3130 DBG("Receiving data from client ...");
3131 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3132 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3133 if (ret <= 0) {
3134 DBG("Nothing recv() from client... continuing");
3135 ret = close(sock);
3136 if (ret) {
3137 PERROR("close");
3138 }
3139 sock = -1;
3140 clean_command_ctx(&cmd_ctx);
3141 continue;
3142 }
3143
3144 health_code_update(&health_thread_cmd);
3145
3146 // TODO: Validate cmd_ctx including sanity check for
3147 // security purpose.
3148
3149 rcu_thread_online();
3150 /*
3151 * This function dispatch the work to the kernel or userspace tracer
3152 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3153 * informations for the client. The command context struct contains
3154 * everything this function may needs.
3155 */
3156 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3157 rcu_thread_offline();
3158 if (ret < 0) {
3159 if (sock_error) {
3160 ret = close(sock);
3161 if (ret) {
3162 PERROR("close");
3163 }
3164 sock = -1;
3165 }
3166 /*
3167 * TODO: Inform client somehow of the fatal error. At
3168 * this point, ret < 0 means that a zmalloc failed
3169 * (ENOMEM). Error detected but still accept
3170 * command, unless a socket error has been
3171 * detected.
3172 */
3173 clean_command_ctx(&cmd_ctx);
3174 continue;
3175 }
3176
3177 health_code_update(&health_thread_cmd);
3178
3179 DBG("Sending response (size: %d, retcode: %s)",
3180 cmd_ctx->lttng_msg_size,
3181 lttng_strerror(-cmd_ctx->llm->ret_code));
3182 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3183 if (ret < 0) {
3184 ERR("Failed to send data back to client");
3185 }
3186
3187 /* End of transmission */
3188 ret = close(sock);
3189 if (ret) {
3190 PERROR("close");
3191 }
3192 sock = -1;
3193
3194 clean_command_ctx(&cmd_ctx);
3195
3196 health_code_update(&health_thread_cmd);
3197 }
3198
3199 exit:
3200 error:
3201 if (err) {
3202 health_error(&health_thread_cmd);
3203 ERR("Health error occurred in %s", __func__);
3204 }
3205 health_exit(&health_thread_cmd);
3206
3207 DBG("Client thread dying");
3208 unlink(client_unix_sock_path);
3209 if (client_sock >= 0) {
3210 ret = close(client_sock);
3211 if (ret) {
3212 PERROR("close");
3213 }
3214 }
3215 if (sock >= 0) {
3216 ret = close(sock);
3217 if (ret) {
3218 PERROR("close");
3219 }
3220 }
3221
3222 lttng_poll_clean(&events);
3223 clean_command_ctx(&cmd_ctx);
3224
3225 rcu_unregister_thread();
3226 return NULL;
3227 }
3228
3229
3230 /*
3231 * usage function on stderr
3232 */
3233 static void usage(void)
3234 {
3235 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3236 fprintf(stderr, " -h, --help Display this usage.\n");
3237 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3238 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3239 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3240 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3241 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3242 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3243 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3244 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3245 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3246 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3247 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3248 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3249 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3250 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3251 fprintf(stderr, " -V, --version Show version number.\n");
3252 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3253 fprintf(stderr, " -q, --quiet No output at all.\n");
3254 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3255 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3256 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3257 }
3258
3259 /*
3260 * daemon argument parsing
3261 */
3262 static int parse_args(int argc, char **argv)
3263 {
3264 int c;
3265
3266 static struct option long_options[] = {
3267 { "client-sock", 1, 0, 'c' },
3268 { "apps-sock", 1, 0, 'a' },
3269 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3270 { "kconsumerd-err-sock", 1, 0, 'E' },
3271 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3272 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3273 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3274 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3275 { "consumerd32-path", 1, 0, 'u' },
3276 { "consumerd32-libdir", 1, 0, 'U' },
3277 { "consumerd64-path", 1, 0, 't' },
3278 { "consumerd64-libdir", 1, 0, 'T' },
3279 { "daemonize", 0, 0, 'd' },
3280 { "sig-parent", 0, 0, 'S' },
3281 { "help", 0, 0, 'h' },
3282 { "group", 1, 0, 'g' },
3283 { "version", 0, 0, 'V' },
3284 { "quiet", 0, 0, 'q' },
3285 { "verbose", 0, 0, 'v' },
3286 { "verbose-consumer", 0, 0, 'Z' },
3287 { "no-kernel", 0, 0, 'N' },
3288 { NULL, 0, 0, 0 }
3289 };
3290
3291 while (1) {
3292 int option_index = 0;
3293 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3294 long_options, &option_index);
3295 if (c == -1) {
3296 break;
3297 }
3298
3299 switch (c) {
3300 case 0:
3301 fprintf(stderr, "option %s", long_options[option_index].name);
3302 if (optarg) {
3303 fprintf(stderr, " with arg %s\n", optarg);
3304 }
3305 break;
3306 case 'c':
3307 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3308 break;
3309 case 'a':
3310 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3311 break;
3312 case 'd':
3313 opt_daemon = 1;
3314 break;
3315 case 'g':
3316 opt_tracing_group = optarg;
3317 break;
3318 case 'h':
3319 usage();
3320 exit(EXIT_FAILURE);
3321 case 'V':
3322 fprintf(stdout, "%s\n", VERSION);
3323 exit(EXIT_SUCCESS);
3324 case 'S':
3325 opt_sig_parent = 1;
3326 break;
3327 case 'E':
3328 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3329 break;
3330 case 'C':
3331 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3332 break;
3333 case 'F':
3334 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3335 break;
3336 case 'D':
3337 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3338 break;
3339 case 'H':
3340 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3341 break;
3342 case 'G':
3343 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3344 break;
3345 case 'N':
3346 opt_no_kernel = 1;
3347 break;
3348 case 'q':
3349 lttng_opt_quiet = 1;
3350 break;
3351 case 'v':
3352 /* Verbose level can increase using multiple -v */
3353 lttng_opt_verbose += 1;
3354 break;
3355 case 'Z':
3356 opt_verbose_consumer += 1;
3357 break;
3358 case 'u':
3359 consumerd32_bin= optarg;
3360 break;
3361 case 'U':
3362 consumerd32_libdir = optarg;
3363 break;
3364 case 't':
3365 consumerd64_bin = optarg;
3366 break;
3367 case 'T':
3368 consumerd64_libdir = optarg;
3369 break;
3370 default:
3371 /* Unknown option or other error.
3372 * Error is printed by getopt, just return */
3373 return -1;
3374 }
3375 }
3376
3377 return 0;
3378 }
3379
3380 /*
3381 * Creates the two needed socket by the daemon.
3382 * apps_sock - The communication socket for all UST apps.
3383 * client_sock - The communication of the cli tool (lttng).
3384 */
3385 static int init_daemon_socket(void)
3386 {
3387 int ret = 0;
3388 mode_t old_umask;
3389
3390 old_umask = umask(0);
3391
3392 /* Create client tool unix socket */
3393 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3394 if (client_sock < 0) {
3395 ERR("Create unix sock failed: %s", client_unix_sock_path);
3396 ret = -1;
3397 goto end;
3398 }
3399
3400 /* File permission MUST be 660 */
3401 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3402 if (ret < 0) {
3403 ERR("Set file permissions failed: %s", client_unix_sock_path);
3404 PERROR("chmod");
3405 goto end;
3406 }
3407
3408 /* Create the application unix socket */
3409 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3410 if (apps_sock < 0) {
3411 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3412 ret = -1;
3413 goto end;
3414 }
3415
3416 /* File permission MUST be 666 */
3417 ret = chmod(apps_unix_sock_path,
3418 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3419 if (ret < 0) {
3420 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3421 PERROR("chmod");
3422 goto end;
3423 }
3424
3425 end:
3426 umask(old_umask);
3427 return ret;
3428 }
3429
3430 /*
3431 * Check if the global socket is available, and if a daemon is answering at the
3432 * other side. If yes, error is returned.
3433 */
3434 static int check_existing_daemon(void)
3435 {
3436 /* Is there anybody out there ? */
3437 if (lttng_session_daemon_alive()) {
3438 return -EEXIST;
3439 }
3440
3441 return 0;
3442 }
3443
3444 /*
3445 * Set the tracing group gid onto the client socket.
3446 *
3447 * Race window between mkdir and chown is OK because we are going from more
3448 * permissive (root.root) to less permissive (root.tracing).
3449 */
3450 static int set_permissions(char *rundir)
3451 {
3452 int ret;
3453 gid_t gid;
3454
3455 ret = allowed_group();
3456 if (ret < 0) {
3457 WARN("No tracing group detected");
3458 ret = 0;
3459 goto end;
3460 }
3461
3462 gid = ret;
3463
3464 /* Set lttng run dir */
3465 ret = chown(rundir, 0, gid);
3466 if (ret < 0) {
3467 ERR("Unable to set group on %s", rundir);
3468 PERROR("chown");
3469 }
3470
3471 /* Ensure tracing group can search the run dir */
3472 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3473 if (ret < 0) {
3474 ERR("Unable to set permissions on %s", rundir);
3475 PERROR("chmod");
3476 }
3477
3478 /* lttng client socket path */
3479 ret = chown(client_unix_sock_path, 0, gid);
3480 if (ret < 0) {
3481 ERR("Unable to set group on %s", client_unix_sock_path);
3482 PERROR("chown");
3483 }
3484
3485 /* kconsumer error socket path */
3486 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3487 if (ret < 0) {
3488 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3489 PERROR("chown");
3490 }
3491
3492 /* 64-bit ustconsumer error socket path */
3493 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3494 if (ret < 0) {
3495 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3496 PERROR("chown");
3497 }
3498
3499 /* 32-bit ustconsumer compat32 error socket path */
3500 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3501 if (ret < 0) {
3502 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3503 PERROR("chown");
3504 }
3505
3506 DBG("All permissions are set");
3507
3508 end:
3509 return ret;
3510 }
3511
3512 /*
3513 * Create the lttng run directory needed for all global sockets and pipe.
3514 */
3515 static int create_lttng_rundir(const char *rundir)
3516 {
3517 int ret;
3518
3519 DBG3("Creating LTTng run directory: %s", rundir);
3520
3521 ret = mkdir(rundir, S_IRWXU);
3522 if (ret < 0) {
3523 if (errno != EEXIST) {
3524 ERR("Unable to create %s", rundir);
3525 goto error;
3526 } else {
3527 ret = 0;
3528 }
3529 }
3530
3531 error:
3532 return ret;
3533 }
3534
3535 /*
3536 * Setup sockets and directory needed by the kconsumerd communication with the
3537 * session daemon.
3538 */
3539 static int set_consumer_sockets(struct consumer_data *consumer_data,
3540 const char *rundir)
3541 {
3542 int ret;
3543 char path[PATH_MAX];
3544
3545 switch (consumer_data->type) {
3546 case LTTNG_CONSUMER_KERNEL:
3547 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3548 break;
3549 case LTTNG_CONSUMER64_UST:
3550 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3551 break;
3552 case LTTNG_CONSUMER32_UST:
3553 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3554 break;
3555 default:
3556 ERR("Consumer type unknown");
3557 ret = -EINVAL;
3558 goto error;
3559 }
3560
3561 DBG2("Creating consumer directory: %s", path);
3562
3563 ret = mkdir(path, S_IRWXU);
3564 if (ret < 0) {
3565 if (errno != EEXIST) {
3566 PERROR("mkdir");
3567 ERR("Failed to create %s", path);
3568 goto error;
3569 }
3570 ret = -1;
3571 }
3572
3573 /* Create the kconsumerd error unix socket */
3574 consumer_data->err_sock =
3575 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3576 if (consumer_data->err_sock < 0) {
3577 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3578 ret = -1;
3579 goto error;
3580 }
3581
3582 /* File permission MUST be 660 */
3583 ret = chmod(consumer_data->err_unix_sock_path,
3584 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3585 if (ret < 0) {
3586 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3587 PERROR("chmod");
3588 goto error;
3589 }
3590
3591 error:
3592 return ret;
3593 }
3594
3595 /*
3596 * Signal handler for the daemon
3597 *
3598 * Simply stop all worker threads, leaving main() return gracefully after
3599 * joining all threads and calling cleanup().
3600 */
3601 static void sighandler(int sig)
3602 {
3603 switch (sig) {
3604 case SIGPIPE:
3605 DBG("SIGPIPE caught");
3606 return;
3607 case SIGINT:
3608 DBG("SIGINT caught");
3609 stop_threads();
3610 break;
3611 case SIGTERM:
3612 DBG("SIGTERM caught");
3613 stop_threads();
3614 break;
3615 default:
3616 break;
3617 }
3618 }
3619
3620 /*
3621 * Setup signal handler for :
3622 * SIGINT, SIGTERM, SIGPIPE
3623 */
3624 static int set_signal_handler(void)
3625 {
3626 int ret = 0;
3627 struct sigaction sa;
3628 sigset_t sigset;
3629
3630 if ((ret = sigemptyset(&sigset)) < 0) {
3631 PERROR("sigemptyset");
3632 return ret;
3633 }
3634
3635 sa.sa_handler = sighandler;
3636 sa.sa_mask = sigset;
3637 sa.sa_flags = 0;
3638 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3639 PERROR("sigaction");
3640 return ret;
3641 }
3642
3643 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3644 PERROR("sigaction");
3645 return ret;
3646 }
3647
3648 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3649 PERROR("sigaction");
3650 return ret;
3651 }
3652
3653 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3654
3655 return ret;
3656 }
3657
3658 /*
3659 * Set open files limit to unlimited. This daemon can open a large number of
3660 * file descriptors in order to consumer multiple kernel traces.
3661 */
3662 static void set_ulimit(void)
3663 {
3664 int ret;
3665 struct rlimit lim;
3666
3667 /* The kernel does not allowed an infinite limit for open files */
3668 lim.rlim_cur = 65535;
3669 lim.rlim_max = 65535;
3670
3671 ret = setrlimit(RLIMIT_NOFILE, &lim);
3672 if (ret < 0) {
3673 PERROR("failed to set open files limit");
3674 }
3675 }
3676
3677 /*
3678 * main
3679 */
3680 int main(int argc, char **argv)
3681 {
3682 int ret = 0;
3683 void *status;
3684 const char *home_path;
3685
3686 init_kernel_workarounds();
3687
3688 rcu_register_thread();
3689
3690 setup_consumerd_path();
3691
3692 /* Parse arguments */
3693 progname = argv[0];
3694 if ((ret = parse_args(argc, argv) < 0)) {
3695 goto error;
3696 }
3697
3698 /* Daemonize */
3699 if (opt_daemon) {
3700 int i;
3701
3702 /*
3703 * fork
3704 * child: setsid, close FD 0, 1, 2, chdir /
3705 * parent: exit (if fork is successful)
3706 */
3707 ret = daemon(0, 0);
3708 if (ret < 0) {
3709 PERROR("daemon");
3710 goto error;
3711 }
3712 /*
3713 * We are in the child. Make sure all other file
3714 * descriptors are closed, in case we are called with
3715 * more opened file descriptors than the standard ones.
3716 */
3717 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3718 (void) close(i);
3719 }
3720 }
3721
3722 /* Create thread quit pipe */
3723 if ((ret = init_thread_quit_pipe()) < 0) {
3724 goto error;
3725 }
3726
3727 /* Check if daemon is UID = 0 */
3728 is_root = !getuid();
3729
3730 if (is_root) {
3731 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3732
3733 /* Create global run dir with root access */
3734 ret = create_lttng_rundir(rundir);
3735 if (ret < 0) {
3736 goto error;
3737 }
3738
3739 if (strlen(apps_unix_sock_path) == 0) {
3740 snprintf(apps_unix_sock_path, PATH_MAX,
3741 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3742 }
3743
3744 if (strlen(client_unix_sock_path) == 0) {
3745 snprintf(client_unix_sock_path, PATH_MAX,
3746 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3747 }
3748
3749 /* Set global SHM for ust */
3750 if (strlen(wait_shm_path) == 0) {
3751 snprintf(wait_shm_path, PATH_MAX,
3752 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3753 }
3754
3755 if (strlen(health_unix_sock_path) == 0) {
3756 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3757 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3758 }
3759
3760 /* Setup kernel consumerd path */
3761 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3762 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3763 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3764 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3765
3766 DBG2("Kernel consumer err path: %s",
3767 kconsumer_data.err_unix_sock_path);
3768 DBG2("Kernel consumer cmd path: %s",
3769 kconsumer_data.cmd_unix_sock_path);
3770 } else {
3771 home_path = get_home_dir();
3772 if (home_path == NULL) {
3773 /* TODO: Add --socket PATH option */
3774 ERR("Can't get HOME directory for sockets creation.");
3775 ret = -EPERM;
3776 goto error;
3777 }
3778
3779 /*
3780 * Create rundir from home path. This will create something like
3781 * $HOME/.lttng
3782 */
3783 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3784 if (ret < 0) {
3785 ret = -ENOMEM;
3786 goto error;
3787 }
3788
3789 ret = create_lttng_rundir(rundir);
3790 if (ret < 0) {
3791 goto error;
3792 }
3793
3794 if (strlen(apps_unix_sock_path) == 0) {
3795 snprintf(apps_unix_sock_path, PATH_MAX,
3796 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3797 }
3798
3799 /* Set the cli tool unix socket path */
3800 if (strlen(client_unix_sock_path) == 0) {
3801 snprintf(client_unix_sock_path, PATH_MAX,
3802 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3803 }
3804
3805 /* Set global SHM for ust */
3806 if (strlen(wait_shm_path) == 0) {
3807 snprintf(wait_shm_path, PATH_MAX,
3808 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3809 }
3810
3811 /* Set health check Unix path */
3812 if (strlen(health_unix_sock_path) == 0) {
3813 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3814 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3815 }
3816 }
3817
3818 /* Set consumer initial state */
3819 kernel_consumerd_state = CONSUMER_STOPPED;
3820 ust_consumerd_state = CONSUMER_STOPPED;
3821
3822 DBG("Client socket path %s", client_unix_sock_path);
3823 DBG("Application socket path %s", apps_unix_sock_path);
3824 DBG("LTTng run directory path: %s", rundir);
3825
3826 /* 32 bits consumerd path setup */
3827 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
3828 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
3829 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
3830 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
3831
3832 DBG2("UST consumer 32 bits err path: %s",
3833 ustconsumer32_data.err_unix_sock_path);
3834 DBG2("UST consumer 32 bits cmd path: %s",
3835 ustconsumer32_data.cmd_unix_sock_path);
3836
3837 /* 64 bits consumerd path setup */
3838 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
3839 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
3840 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
3841 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
3842
3843 DBG2("UST consumer 64 bits err path: %s",
3844 ustconsumer64_data.err_unix_sock_path);
3845 DBG2("UST consumer 64 bits cmd path: %s",
3846 ustconsumer64_data.cmd_unix_sock_path);
3847
3848 /*
3849 * See if daemon already exist.
3850 */
3851 if ((ret = check_existing_daemon()) < 0) {
3852 ERR("Already running daemon.\n");
3853 /*
3854 * We do not goto exit because we must not cleanup()
3855 * because a daemon is already running.
3856 */
3857 goto error;
3858 }
3859
3860 /*
3861 * Init UST app hash table. Alloc hash table before this point since
3862 * cleanup() can get called after that point.
3863 */
3864 ust_app_ht_alloc();
3865
3866 /* After this point, we can safely call cleanup() with "goto exit" */
3867
3868 /*
3869 * These actions must be executed as root. We do that *after* setting up
3870 * the sockets path because we MUST make the check for another daemon using
3871 * those paths *before* trying to set the kernel consumer sockets and init
3872 * kernel tracer.
3873 */
3874 if (is_root) {
3875 ret = set_consumer_sockets(&kconsumer_data, rundir);
3876 if (ret < 0) {
3877 goto exit;
3878 }
3879
3880 /* Setup kernel tracer */
3881 if (!opt_no_kernel) {
3882 init_kernel_tracer();
3883 }
3884
3885 /* Set ulimit for open files */
3886 set_ulimit();
3887 }
3888 /* init lttng_fd tracking must be done after set_ulimit. */
3889 lttng_fd_init();
3890
3891 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
3892 if (ret < 0) {
3893 goto exit;
3894 }
3895
3896 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
3897 if (ret < 0) {
3898 goto exit;
3899 }
3900
3901 if ((ret = set_signal_handler()) < 0) {
3902 goto exit;
3903 }
3904
3905 /* Setup the needed unix socket */
3906 if ((ret = init_daemon_socket()) < 0) {
3907 goto exit;
3908 }
3909
3910 /* Set credentials to socket */
3911 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
3912 goto exit;
3913 }
3914
3915 /* Get parent pid if -S, --sig-parent is specified. */
3916 if (opt_sig_parent) {
3917 ppid = getppid();
3918 }
3919
3920 /* Setup the kernel pipe for waking up the kernel thread */
3921 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
3922 goto exit;
3923 }
3924
3925 /* Setup the thread apps communication pipe. */
3926 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
3927 goto exit;
3928 }
3929
3930 /* Init UST command queue. */
3931 cds_wfq_init(&ust_cmd_queue.queue);
3932
3933 /*
3934 * Get session list pointer. This pointer MUST NOT be free(). This list is
3935 * statically declared in session.c
3936 */
3937 session_list_ptr = session_get_list();
3938
3939 /* Set up max poll set size */
3940 lttng_poll_set_max_size();
3941
3942 cmd_init();
3943
3944 /* Init all health thread counters. */
3945 health_init(&health_thread_cmd);
3946 health_init(&health_thread_kernel);
3947 health_init(&health_thread_app_manage);
3948 health_init(&health_thread_app_reg);
3949
3950 /*
3951 * Init health counters of the consumer thread. We do a quick hack here to
3952 * the state of the consumer health is fine even if the thread is not
3953 * started. This is simply to ease our life and has no cost what so ever.
3954 */
3955 health_init(&kconsumer_data.health);
3956 health_poll_update(&kconsumer_data.health);
3957 health_init(&ustconsumer32_data.health);
3958 health_poll_update(&ustconsumer32_data.health);
3959 health_init(&ustconsumer64_data.health);
3960 health_poll_update(&ustconsumer64_data.health);
3961
3962 /* Create thread to manage the client socket */
3963 ret = pthread_create(&health_thread, NULL,
3964 thread_manage_health, (void *) NULL);
3965 if (ret != 0) {
3966 PERROR("pthread_create health");
3967 goto exit_health;
3968 }
3969
3970 /* Create thread to manage the client socket */
3971 ret = pthread_create(&client_thread, NULL,
3972 thread_manage_clients, (void *) NULL);
3973 if (ret != 0) {
3974 PERROR("pthread_create clients");
3975 goto exit_client;
3976 }
3977
3978 /* Create thread to dispatch registration */
3979 ret = pthread_create(&dispatch_thread, NULL,
3980 thread_dispatch_ust_registration, (void *) NULL);
3981 if (ret != 0) {
3982 PERROR("pthread_create dispatch");
3983 goto exit_dispatch;
3984 }
3985
3986 /* Create thread to manage application registration. */
3987 ret = pthread_create(&reg_apps_thread, NULL,
3988 thread_registration_apps, (void *) NULL);
3989 if (ret != 0) {
3990 PERROR("pthread_create registration");
3991 goto exit_reg_apps;
3992 }
3993
3994 /* Create thread to manage application socket */
3995 ret = pthread_create(&apps_thread, NULL,
3996 thread_manage_apps, (void *) NULL);
3997 if (ret != 0) {
3998 PERROR("pthread_create apps");
3999 goto exit_apps;
4000 }
4001
4002 /* Create kernel thread to manage kernel event */
4003 ret = pthread_create(&kernel_thread, NULL,
4004 thread_manage_kernel, (void *) NULL);
4005 if (ret != 0) {
4006 PERROR("pthread_create kernel");
4007 goto exit_kernel;
4008 }
4009
4010 ret = pthread_join(kernel_thread, &status);
4011 if (ret != 0) {
4012 PERROR("pthread_join");
4013 goto error; /* join error, exit without cleanup */
4014 }
4015
4016 exit_kernel:
4017 ret = pthread_join(apps_thread, &status);
4018 if (ret != 0) {
4019 PERROR("pthread_join");
4020 goto error; /* join error, exit without cleanup */
4021 }
4022
4023 exit_apps:
4024 ret = pthread_join(reg_apps_thread, &status);
4025 if (ret != 0) {
4026 PERROR("pthread_join");
4027 goto error; /* join error, exit without cleanup */
4028 }
4029
4030 exit_reg_apps:
4031 ret = pthread_join(dispatch_thread, &status);
4032 if (ret != 0) {
4033 PERROR("pthread_join");
4034 goto error; /* join error, exit without cleanup */
4035 }
4036
4037 exit_dispatch:
4038 ret = pthread_join(client_thread, &status);
4039 if (ret != 0) {
4040 PERROR("pthread_join");
4041 goto error; /* join error, exit without cleanup */
4042 }
4043
4044 ret = join_consumer_thread(&kconsumer_data);
4045 if (ret != 0) {
4046 PERROR("join_consumer");
4047 goto error; /* join error, exit without cleanup */
4048 }
4049
4050 exit_client:
4051 exit_health:
4052 exit:
4053 /*
4054 * cleanup() is called when no other thread is running.
4055 */
4056 rcu_thread_online();
4057 cleanup();
4058 rcu_thread_offline();
4059 rcu_unregister_thread();
4060 if (!ret) {
4061 exit(EXIT_SUCCESS);
4062 }
4063 error:
4064 exit(EXIT_FAILURE);
4065 }
This page took 0.223192 seconds and 6 git commands to generate.