b4755b0b8cce0d7eba70d60d4a474345ba2bbd1c
[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 free(rundir);
416
417 DBG("Cleaning up all sessions");
418
419 /* Destroy session list mutex */
420 if (session_list_ptr != NULL) {
421 pthread_mutex_destroy(&session_list_ptr->lock);
422
423 /* Cleanup ALL session */
424 cds_list_for_each_entry_safe(sess, stmp,
425 &session_list_ptr->head, list) {
426 cmd_destroy_session(sess, kernel_poll_pipe[1]);
427 }
428 }
429
430 DBG("Closing all UST sockets");
431 ust_app_clean_list();
432
433 if (is_root && !opt_no_kernel) {
434 DBG2("Closing kernel fd");
435 if (kernel_tracer_fd >= 0) {
436 ret = close(kernel_tracer_fd);
437 if (ret) {
438 PERROR("close");
439 }
440 }
441 DBG("Unloading kernel modules");
442 modprobe_remove_lttng_all();
443 }
444
445 utils_close_pipe(kernel_poll_pipe);
446 utils_close_pipe(apps_cmd_pipe);
447
448 /* <fun> */
449 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
450 "Matthew, BEET driven development works!%c[%dm",
451 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
452 /* </fun> */
453 }
454
455 /*
456 * Send data on a unix socket using the liblttsessiondcomm API.
457 *
458 * Return lttcomm error code.
459 */
460 static int send_unix_sock(int sock, void *buf, size_t len)
461 {
462 /* Check valid length */
463 if (len <= 0) {
464 return -1;
465 }
466
467 return lttcomm_send_unix_sock(sock, buf, len);
468 }
469
470 /*
471 * Free memory of a command context structure.
472 */
473 static void clean_command_ctx(struct command_ctx **cmd_ctx)
474 {
475 DBG("Clean command context structure");
476 if (*cmd_ctx) {
477 if ((*cmd_ctx)->llm) {
478 free((*cmd_ctx)->llm);
479 }
480 if ((*cmd_ctx)->lsm) {
481 free((*cmd_ctx)->lsm);
482 }
483 free(*cmd_ctx);
484 *cmd_ctx = NULL;
485 }
486 }
487
488 /*
489 * Notify UST applications using the shm mmap futex.
490 */
491 static int notify_ust_apps(int active)
492 {
493 char *wait_shm_mmap;
494
495 DBG("Notifying applications of session daemon state: %d", active);
496
497 /* See shm.c for this call implying mmap, shm and futex calls */
498 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
499 if (wait_shm_mmap == NULL) {
500 goto error;
501 }
502
503 /* Wake waiting process */
504 futex_wait_update((int32_t *) wait_shm_mmap, active);
505
506 /* Apps notified successfully */
507 return 0;
508
509 error:
510 return -1;
511 }
512
513 /*
514 * Setup the outgoing data buffer for the response (llm) by allocating the
515 * right amount of memory and copying the original information from the lsm
516 * structure.
517 *
518 * Return total size of the buffer pointed by buf.
519 */
520 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
521 {
522 int ret, buf_size;
523
524 buf_size = size;
525
526 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
527 if (cmd_ctx->llm == NULL) {
528 PERROR("zmalloc");
529 ret = -ENOMEM;
530 goto error;
531 }
532
533 /* Copy common data */
534 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
535 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
536
537 cmd_ctx->llm->data_size = size;
538 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
539
540 return buf_size;
541
542 error:
543 return ret;
544 }
545
546 /*
547 * Update the kernel poll set of all channel fd available over all tracing
548 * session. Add the wakeup pipe at the end of the set.
549 */
550 static int update_kernel_poll(struct lttng_poll_event *events)
551 {
552 int ret;
553 struct ltt_session *session;
554 struct ltt_kernel_channel *channel;
555
556 DBG("Updating kernel poll set");
557
558 session_lock_list();
559 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
560 session_lock(session);
561 if (session->kernel_session == NULL) {
562 session_unlock(session);
563 continue;
564 }
565
566 cds_list_for_each_entry(channel,
567 &session->kernel_session->channel_list.head, list) {
568 /* Add channel fd to the kernel poll set */
569 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
570 if (ret < 0) {
571 session_unlock(session);
572 goto error;
573 }
574 DBG("Channel fd %d added to kernel set", channel->fd);
575 }
576 session_unlock(session);
577 }
578 session_unlock_list();
579
580 return 0;
581
582 error:
583 session_unlock_list();
584 return -1;
585 }
586
587 /*
588 * Find the channel fd from 'fd' over all tracing session. When found, check
589 * for new channel stream and send those stream fds to the kernel consumer.
590 *
591 * Useful for CPU hotplug feature.
592 */
593 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
594 {
595 int ret = 0;
596 struct ltt_session *session;
597 struct ltt_kernel_session *ksess;
598 struct ltt_kernel_channel *channel;
599
600 DBG("Updating kernel streams for channel fd %d", fd);
601
602 session_lock_list();
603 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
604 session_lock(session);
605 if (session->kernel_session == NULL) {
606 session_unlock(session);
607 continue;
608 }
609 ksess = session->kernel_session;
610
611 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
612 if (channel->fd == fd) {
613 DBG("Channel found, updating kernel streams");
614 ret = kernel_open_channel_stream(channel);
615 if (ret < 0) {
616 goto error;
617 }
618
619 /*
620 * Have we already sent fds to the consumer? If yes, it means
621 * that tracing is started so it is safe to send our updated
622 * stream fds.
623 */
624 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
625 struct lttng_ht_iter iter;
626 struct consumer_socket *socket;
627
628
629 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
630 &iter.iter, socket, node.node) {
631 /* Code flow error */
632 assert(socket->fd >= 0);
633
634 pthread_mutex_lock(socket->lock);
635 ret = kernel_consumer_send_channel_stream(socket->fd,
636 channel, ksess);
637 pthread_mutex_unlock(socket->lock);
638 if (ret < 0) {
639 goto error;
640 }
641 }
642 }
643 goto error;
644 }
645 }
646 session_unlock(session);
647 }
648 session_unlock_list();
649 return ret;
650
651 error:
652 session_unlock(session);
653 session_unlock_list();
654 return ret;
655 }
656
657 /*
658 * For each tracing session, update newly registered apps.
659 */
660 static void update_ust_app(int app_sock)
661 {
662 struct ltt_session *sess, *stmp;
663
664 session_lock_list();
665
666 /* For all tracing session(s) */
667 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
668 session_lock(sess);
669 if (sess->ust_session) {
670 ust_app_global_update(sess->ust_session, app_sock);
671 }
672 session_unlock(sess);
673 }
674
675 session_unlock_list();
676 }
677
678 /*
679 * This thread manage event coming from the kernel.
680 *
681 * Features supported in this thread:
682 * -) CPU Hotplug
683 */
684 static void *thread_manage_kernel(void *data)
685 {
686 int ret, i, pollfd, update_poll_flag = 1, err = -1;
687 uint32_t revents, nb_fd;
688 char tmp;
689 struct lttng_poll_event events;
690
691 DBG("Thread manage kernel started");
692
693 testpoint(thread_manage_kernel);
694
695 health_code_update(&health_thread_kernel);
696
697 testpoint(thread_manage_kernel_before_loop);
698
699 ret = create_thread_poll_set(&events, 2);
700 if (ret < 0) {
701 goto error_poll_create;
702 }
703
704 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
705 if (ret < 0) {
706 goto error;
707 }
708
709 while (1) {
710 health_code_update(&health_thread_kernel);
711
712 if (update_poll_flag == 1) {
713 /*
714 * Reset number of fd in the poll set. Always 2 since there is the thread
715 * quit pipe and the kernel pipe.
716 */
717 events.nb_fd = 2;
718
719 ret = update_kernel_poll(&events);
720 if (ret < 0) {
721 goto error;
722 }
723 update_poll_flag = 0;
724 }
725
726 nb_fd = LTTNG_POLL_GETNB(&events);
727
728 DBG("Thread kernel polling on %d fds", nb_fd);
729
730 /* Zeroed the poll events */
731 lttng_poll_reset(&events);
732
733 /* Poll infinite value of time */
734 restart:
735 health_poll_update(&health_thread_kernel);
736 ret = lttng_poll_wait(&events, -1);
737 health_poll_update(&health_thread_kernel);
738 if (ret < 0) {
739 /*
740 * Restart interrupted system call.
741 */
742 if (errno == EINTR) {
743 goto restart;
744 }
745 goto error;
746 } else if (ret == 0) {
747 /* Should not happen since timeout is infinite */
748 ERR("Return value of poll is 0 with an infinite timeout.\n"
749 "This should not have happened! Continuing...");
750 continue;
751 }
752
753 for (i = 0; i < nb_fd; i++) {
754 /* Fetch once the poll data */
755 revents = LTTNG_POLL_GETEV(&events, i);
756 pollfd = LTTNG_POLL_GETFD(&events, i);
757
758 health_code_update(&health_thread_kernel);
759
760 /* Thread quit pipe has been closed. Killing thread. */
761 ret = check_thread_quit_pipe(pollfd, revents);
762 if (ret) {
763 err = 0;
764 goto exit;
765 }
766
767 /* Check for data on kernel pipe */
768 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
769 ret = read(kernel_poll_pipe[0], &tmp, 1);
770 update_poll_flag = 1;
771 continue;
772 } else {
773 /*
774 * New CPU detected by the kernel. Adding kernel stream to
775 * kernel session and updating the kernel consumer
776 */
777 if (revents & LPOLLIN) {
778 ret = update_kernel_stream(&kconsumer_data, pollfd);
779 if (ret < 0) {
780 continue;
781 }
782 break;
783 /*
784 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
785 * and unregister kernel stream at this point.
786 */
787 }
788 }
789 }
790 }
791
792 exit:
793 error:
794 lttng_poll_clean(&events);
795 error_poll_create:
796 if (err) {
797 health_error(&health_thread_kernel);
798 ERR("Health error occurred in %s", __func__);
799 }
800 health_exit(&health_thread_kernel);
801 DBG("Kernel thread dying");
802 return NULL;
803 }
804
805 /*
806 * Signal pthread condition of the consumer data that the thread.
807 */
808 static void signal_consumer_condition(struct consumer_data *data, int state)
809 {
810 pthread_mutex_lock(&data->cond_mutex);
811
812 /*
813 * The state is set before signaling. It can be any value, it's the waiter
814 * job to correctly interpret this condition variable associated to the
815 * consumer pthread_cond.
816 *
817 * A value of 0 means that the corresponding thread of the consumer data
818 * was not started. 1 indicates that the thread has started and is ready
819 * for action. A negative value means that there was an error during the
820 * thread bootstrap.
821 */
822 data->consumer_thread_is_ready = state;
823 (void) pthread_cond_signal(&data->cond);
824
825 pthread_mutex_unlock(&data->cond_mutex);
826 }
827
828 /*
829 * This thread manage the consumer error sent back to the session daemon.
830 */
831 static void *thread_manage_consumer(void *data)
832 {
833 int sock = -1, i, ret, pollfd, err = -1;
834 uint32_t revents, nb_fd;
835 enum lttcomm_return_code code;
836 struct lttng_poll_event events;
837 struct consumer_data *consumer_data = data;
838
839 DBG("[thread] Manage consumer started");
840
841 health_code_update(&consumer_data->health);
842
843 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
844 if (ret < 0) {
845 goto error_listen;
846 }
847
848 /*
849 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
850 * Nothing more will be added to this poll set.
851 */
852 ret = create_thread_poll_set(&events, 2);
853 if (ret < 0) {
854 goto error_poll;
855 }
856
857 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
858 if (ret < 0) {
859 goto error;
860 }
861
862 nb_fd = LTTNG_POLL_GETNB(&events);
863
864 health_code_update(&consumer_data->health);
865
866 /* Inifinite blocking call, waiting for transmission */
867 restart:
868 health_poll_update(&consumer_data->health);
869
870 testpoint(thread_manage_consumer);
871
872 ret = lttng_poll_wait(&events, -1);
873 health_poll_update(&consumer_data->health);
874 if (ret < 0) {
875 /*
876 * Restart interrupted system call.
877 */
878 if (errno == EINTR) {
879 goto restart;
880 }
881 goto error;
882 }
883
884 for (i = 0; i < nb_fd; i++) {
885 /* Fetch once the poll data */
886 revents = LTTNG_POLL_GETEV(&events, i);
887 pollfd = LTTNG_POLL_GETFD(&events, i);
888
889 health_code_update(&consumer_data->health);
890
891 /* Thread quit pipe has been closed. Killing thread. */
892 ret = check_thread_quit_pipe(pollfd, revents);
893 if (ret) {
894 err = 0;
895 goto exit;
896 }
897
898 /* Event on the registration socket */
899 if (pollfd == consumer_data->err_sock) {
900 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
901 ERR("consumer err socket poll error");
902 goto error;
903 }
904 }
905 }
906
907 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
908 if (sock < 0) {
909 goto error;
910 }
911
912 health_code_update(&consumer_data->health);
913
914 DBG2("Receiving code from consumer err_sock");
915
916 /* Getting status code from kconsumerd */
917 ret = lttcomm_recv_unix_sock(sock, &code,
918 sizeof(enum lttcomm_return_code));
919 if (ret <= 0) {
920 goto error;
921 }
922
923 health_code_update(&consumer_data->health);
924
925 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
926 consumer_data->cmd_sock =
927 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
928 if (consumer_data->cmd_sock < 0) {
929 /* On error, signal condition and quit. */
930 signal_consumer_condition(consumer_data, -1);
931 PERROR("consumer connect");
932 goto error;
933 }
934 signal_consumer_condition(consumer_data, 1);
935 DBG("Consumer command socket ready");
936 } else {
937 ERR("consumer error when waiting for SOCK_READY : %s",
938 lttcomm_get_readable_code(-code));
939 goto error;
940 }
941
942 /* Remove the kconsumerd error sock since we've established a connexion */
943 ret = lttng_poll_del(&events, consumer_data->err_sock);
944 if (ret < 0) {
945 goto error;
946 }
947
948 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
949 if (ret < 0) {
950 goto error;
951 }
952
953 health_code_update(&consumer_data->health);
954
955 /* Update number of fd */
956 nb_fd = LTTNG_POLL_GETNB(&events);
957
958 /* Inifinite blocking call, waiting for transmission */
959 restart_poll:
960 health_poll_update(&consumer_data->health);
961 ret = lttng_poll_wait(&events, -1);
962 health_poll_update(&consumer_data->health);
963 if (ret < 0) {
964 /*
965 * Restart interrupted system call.
966 */
967 if (errno == EINTR) {
968 goto restart_poll;
969 }
970 goto error;
971 }
972
973 for (i = 0; i < nb_fd; i++) {
974 /* Fetch once the poll data */
975 revents = LTTNG_POLL_GETEV(&events, i);
976 pollfd = LTTNG_POLL_GETFD(&events, i);
977
978 health_code_update(&consumer_data->health);
979
980 /* Thread quit pipe has been closed. Killing thread. */
981 ret = check_thread_quit_pipe(pollfd, revents);
982 if (ret) {
983 err = 0;
984 goto exit;
985 }
986
987 /* Event on the kconsumerd socket */
988 if (pollfd == sock) {
989 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
990 ERR("consumer err socket second poll error");
991 goto error;
992 }
993 }
994 }
995
996 health_code_update(&consumer_data->health);
997
998 /* Wait for any kconsumerd error */
999 ret = lttcomm_recv_unix_sock(sock, &code,
1000 sizeof(enum lttcomm_return_code));
1001 if (ret <= 0) {
1002 ERR("consumer closed the command socket");
1003 goto error;
1004 }
1005
1006 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1007
1008 exit:
1009 error:
1010 /* Immediately set the consumerd state to stopped */
1011 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1012 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1013 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1014 consumer_data->type == LTTNG_CONSUMER32_UST) {
1015 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1016 } else {
1017 /* Code flow error... */
1018 assert(0);
1019 }
1020
1021 if (consumer_data->err_sock >= 0) {
1022 ret = close(consumer_data->err_sock);
1023 if (ret) {
1024 PERROR("close");
1025 }
1026 }
1027 if (consumer_data->cmd_sock >= 0) {
1028 ret = close(consumer_data->cmd_sock);
1029 if (ret) {
1030 PERROR("close");
1031 }
1032 }
1033 if (sock >= 0) {
1034 ret = close(sock);
1035 if (ret) {
1036 PERROR("close");
1037 }
1038 }
1039
1040 unlink(consumer_data->err_unix_sock_path);
1041 unlink(consumer_data->cmd_unix_sock_path);
1042 consumer_data->pid = 0;
1043
1044 lttng_poll_clean(&events);
1045 error_poll:
1046 error_listen:
1047 if (err) {
1048 health_error(&consumer_data->health);
1049 ERR("Health error occurred in %s", __func__);
1050 }
1051 health_exit(&consumer_data->health);
1052 DBG("consumer thread cleanup completed");
1053
1054 return NULL;
1055 }
1056
1057 /*
1058 * This thread manage application communication.
1059 */
1060 static void *thread_manage_apps(void *data)
1061 {
1062 int i, ret, pollfd, err = -1;
1063 uint32_t revents, nb_fd;
1064 struct ust_command ust_cmd;
1065 struct lttng_poll_event events;
1066
1067 DBG("[thread] Manage application started");
1068
1069 testpoint(thread_manage_apps);
1070
1071 rcu_register_thread();
1072 rcu_thread_online();
1073
1074 health_code_update(&health_thread_app_manage);
1075
1076 ret = create_thread_poll_set(&events, 2);
1077 if (ret < 0) {
1078 goto error_poll_create;
1079 }
1080
1081 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1082 if (ret < 0) {
1083 goto error;
1084 }
1085
1086 testpoint(thread_manage_apps_before_loop);
1087
1088 health_code_update(&health_thread_app_manage);
1089
1090 while (1) {
1091 /* Zeroed the events structure */
1092 lttng_poll_reset(&events);
1093
1094 nb_fd = LTTNG_POLL_GETNB(&events);
1095
1096 DBG("Apps thread polling on %d fds", nb_fd);
1097
1098 /* Inifinite blocking call, waiting for transmission */
1099 restart:
1100 health_poll_update(&health_thread_app_manage);
1101 ret = lttng_poll_wait(&events, -1);
1102 health_poll_update(&health_thread_app_manage);
1103 if (ret < 0) {
1104 /*
1105 * Restart interrupted system call.
1106 */
1107 if (errno == EINTR) {
1108 goto restart;
1109 }
1110 goto error;
1111 }
1112
1113 for (i = 0; i < nb_fd; i++) {
1114 /* Fetch once the poll data */
1115 revents = LTTNG_POLL_GETEV(&events, i);
1116 pollfd = LTTNG_POLL_GETFD(&events, i);
1117
1118 health_code_update(&health_thread_app_manage);
1119
1120 /* Thread quit pipe has been closed. Killing thread. */
1121 ret = check_thread_quit_pipe(pollfd, revents);
1122 if (ret) {
1123 err = 0;
1124 goto exit;
1125 }
1126
1127 /* Inspect the apps cmd pipe */
1128 if (pollfd == apps_cmd_pipe[0]) {
1129 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1130 ERR("Apps command pipe error");
1131 goto error;
1132 } else if (revents & LPOLLIN) {
1133 /* Empty pipe */
1134 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1135 if (ret < 0 || ret < sizeof(ust_cmd)) {
1136 PERROR("read apps cmd pipe");
1137 goto error;
1138 }
1139
1140 health_code_update(&health_thread_app_manage);
1141
1142 /* Register applicaton to the session daemon */
1143 ret = ust_app_register(&ust_cmd.reg_msg,
1144 ust_cmd.sock);
1145 if (ret == -ENOMEM) {
1146 goto error;
1147 } else if (ret < 0) {
1148 break;
1149 }
1150
1151 health_code_update(&health_thread_app_manage);
1152
1153 /*
1154 * Validate UST version compatibility.
1155 */
1156 ret = ust_app_validate_version(ust_cmd.sock);
1157 if (ret >= 0) {
1158 /*
1159 * Add channel(s) and event(s) to newly registered apps
1160 * from lttng global UST domain.
1161 */
1162 update_ust_app(ust_cmd.sock);
1163 }
1164
1165 health_code_update(&health_thread_app_manage);
1166
1167 ret = ust_app_register_done(ust_cmd.sock);
1168 if (ret < 0) {
1169 /*
1170 * If the registration is not possible, we simply
1171 * unregister the apps and continue
1172 */
1173 ust_app_unregister(ust_cmd.sock);
1174 } else {
1175 /*
1176 * We just need here to monitor the close of the UST
1177 * socket and poll set monitor those by default.
1178 * Listen on POLLIN (even if we never expect any
1179 * data) to ensure that hangup wakes us.
1180 */
1181 ret = lttng_poll_add(&events, ust_cmd.sock, LPOLLIN);
1182 if (ret < 0) {
1183 goto error;
1184 }
1185
1186 DBG("Apps with sock %d added to poll set",
1187 ust_cmd.sock);
1188 }
1189
1190 health_code_update(&health_thread_app_manage);
1191
1192 break;
1193 }
1194 } else {
1195 /*
1196 * At this point, we know that a registered application made
1197 * the event at poll_wait.
1198 */
1199 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1200 /* Removing from the poll set */
1201 ret = lttng_poll_del(&events, pollfd);
1202 if (ret < 0) {
1203 goto error;
1204 }
1205
1206 /* Socket closed on remote end. */
1207 ust_app_unregister(pollfd);
1208 break;
1209 }
1210 }
1211
1212 health_code_update(&health_thread_app_manage);
1213 }
1214 }
1215
1216 exit:
1217 error:
1218 lttng_poll_clean(&events);
1219 error_poll_create:
1220 if (err) {
1221 health_error(&health_thread_app_manage);
1222 ERR("Health error occurred in %s", __func__);
1223 }
1224 health_exit(&health_thread_app_manage);
1225 DBG("Application communication apps thread cleanup complete");
1226 rcu_thread_offline();
1227 rcu_unregister_thread();
1228 return NULL;
1229 }
1230
1231 /*
1232 * Dispatch request from the registration threads to the application
1233 * communication thread.
1234 */
1235 static void *thread_dispatch_ust_registration(void *data)
1236 {
1237 int ret;
1238 struct cds_wfq_node *node;
1239 struct ust_command *ust_cmd = NULL;
1240
1241 DBG("[thread] Dispatch UST command started");
1242
1243 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1244 /* Atomically prepare the queue futex */
1245 futex_nto1_prepare(&ust_cmd_queue.futex);
1246
1247 do {
1248 /* Dequeue command for registration */
1249 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1250 if (node == NULL) {
1251 DBG("Woken up but nothing in the UST command queue");
1252 /* Continue thread execution */
1253 break;
1254 }
1255
1256 ust_cmd = caa_container_of(node, struct ust_command, node);
1257
1258 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1259 " gid:%d sock:%d name:%s (version %d.%d)",
1260 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1261 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1262 ust_cmd->sock, ust_cmd->reg_msg.name,
1263 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1264 /*
1265 * Inform apps thread of the new application registration. This
1266 * call is blocking so we can be assured that the data will be read
1267 * at some point in time or wait to the end of the world :)
1268 */
1269 ret = write(apps_cmd_pipe[1], ust_cmd,
1270 sizeof(struct ust_command));
1271 if (ret < 0) {
1272 PERROR("write apps cmd pipe");
1273 if (errno == EBADF) {
1274 /*
1275 * We can't inform the application thread to process
1276 * registration. We will exit or else application
1277 * registration will not occur and tracing will never
1278 * start.
1279 */
1280 goto error;
1281 }
1282 }
1283 free(ust_cmd);
1284 } while (node != NULL);
1285
1286 /* Futex wait on queue. Blocking call on futex() */
1287 futex_nto1_wait(&ust_cmd_queue.futex);
1288 }
1289
1290 error:
1291 DBG("Dispatch thread dying");
1292 return NULL;
1293 }
1294
1295 /*
1296 * This thread manage application registration.
1297 */
1298 static void *thread_registration_apps(void *data)
1299 {
1300 int sock = -1, i, ret, pollfd, err = -1;
1301 uint32_t revents, nb_fd;
1302 struct lttng_poll_event events;
1303 /*
1304 * Get allocated in this thread, enqueued to a global queue, dequeued and
1305 * freed in the manage apps thread.
1306 */
1307 struct ust_command *ust_cmd = NULL;
1308
1309 DBG("[thread] Manage application registration started");
1310
1311 testpoint(thread_registration_apps);
1312
1313 ret = lttcomm_listen_unix_sock(apps_sock);
1314 if (ret < 0) {
1315 goto error_listen;
1316 }
1317
1318 /*
1319 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1320 * more will be added to this poll set.
1321 */
1322 ret = create_thread_poll_set(&events, 2);
1323 if (ret < 0) {
1324 goto error_create_poll;
1325 }
1326
1327 /* Add the application registration socket */
1328 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1329 if (ret < 0) {
1330 goto error_poll_add;
1331 }
1332
1333 /* Notify all applications to register */
1334 ret = notify_ust_apps(1);
1335 if (ret < 0) {
1336 ERR("Failed to notify applications or create the wait shared memory.\n"
1337 "Execution continues but there might be problem for already\n"
1338 "running applications that wishes to register.");
1339 }
1340
1341 while (1) {
1342 DBG("Accepting application registration");
1343
1344 nb_fd = LTTNG_POLL_GETNB(&events);
1345
1346 /* Inifinite blocking call, waiting for transmission */
1347 restart:
1348 health_poll_update(&health_thread_app_reg);
1349 ret = lttng_poll_wait(&events, -1);
1350 health_poll_update(&health_thread_app_reg);
1351 if (ret < 0) {
1352 /*
1353 * Restart interrupted system call.
1354 */
1355 if (errno == EINTR) {
1356 goto restart;
1357 }
1358 goto error;
1359 }
1360
1361 for (i = 0; i < nb_fd; i++) {
1362 health_code_update(&health_thread_app_reg);
1363
1364 /* Fetch once the poll data */
1365 revents = LTTNG_POLL_GETEV(&events, i);
1366 pollfd = LTTNG_POLL_GETFD(&events, i);
1367
1368 /* Thread quit pipe has been closed. Killing thread. */
1369 ret = check_thread_quit_pipe(pollfd, revents);
1370 if (ret) {
1371 err = 0;
1372 goto exit;
1373 }
1374
1375 /* Event on the registration socket */
1376 if (pollfd == apps_sock) {
1377 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1378 ERR("Register apps socket poll error");
1379 goto error;
1380 } else if (revents & LPOLLIN) {
1381 sock = lttcomm_accept_unix_sock(apps_sock);
1382 if (sock < 0) {
1383 goto error;
1384 }
1385
1386 /* Create UST registration command for enqueuing */
1387 ust_cmd = zmalloc(sizeof(struct ust_command));
1388 if (ust_cmd == NULL) {
1389 PERROR("ust command zmalloc");
1390 goto error;
1391 }
1392
1393 /*
1394 * Using message-based transmissions to ensure we don't
1395 * have to deal with partially received messages.
1396 */
1397 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1398 if (ret < 0) {
1399 ERR("Exhausted file descriptors allowed for applications.");
1400 free(ust_cmd);
1401 ret = close(sock);
1402 if (ret) {
1403 PERROR("close");
1404 }
1405 sock = -1;
1406 continue;
1407 }
1408 health_code_update(&health_thread_app_reg);
1409 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1410 sizeof(struct ust_register_msg));
1411 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1412 if (ret < 0) {
1413 PERROR("lttcomm_recv_unix_sock register apps");
1414 } else {
1415 ERR("Wrong size received on apps register");
1416 }
1417 free(ust_cmd);
1418 ret = close(sock);
1419 if (ret) {
1420 PERROR("close");
1421 }
1422 lttng_fd_put(LTTNG_FD_APPS, 1);
1423 sock = -1;
1424 continue;
1425 }
1426 health_code_update(&health_thread_app_reg);
1427
1428 ust_cmd->sock = sock;
1429 sock = -1;
1430
1431 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1432 " gid:%d sock:%d name:%s (version %d.%d)",
1433 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1434 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1435 ust_cmd->sock, ust_cmd->reg_msg.name,
1436 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1437
1438 /*
1439 * Lock free enqueue the registration request. The red pill
1440 * has been taken! This apps will be part of the *system*.
1441 */
1442 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1443
1444 /*
1445 * Wake the registration queue futex. Implicit memory
1446 * barrier with the exchange in cds_wfq_enqueue.
1447 */
1448 futex_nto1_wake(&ust_cmd_queue.futex);
1449 }
1450 }
1451 }
1452 }
1453
1454 exit:
1455 error:
1456 if (err) {
1457 health_error(&health_thread_app_reg);
1458 ERR("Health error occurred in %s", __func__);
1459 }
1460 health_exit(&health_thread_app_reg);
1461
1462 /* Notify that the registration thread is gone */
1463 notify_ust_apps(0);
1464
1465 if (apps_sock >= 0) {
1466 ret = close(apps_sock);
1467 if (ret) {
1468 PERROR("close");
1469 }
1470 }
1471 if (sock >= 0) {
1472 ret = close(sock);
1473 if (ret) {
1474 PERROR("close");
1475 }
1476 lttng_fd_put(LTTNG_FD_APPS, 1);
1477 }
1478 unlink(apps_unix_sock_path);
1479
1480 error_poll_add:
1481 lttng_poll_clean(&events);
1482 error_listen:
1483 error_create_poll:
1484 DBG("UST Registration thread cleanup complete");
1485
1486 return NULL;
1487 }
1488
1489 /*
1490 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1491 * exec or it will fails.
1492 */
1493 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1494 {
1495 int ret, clock_ret;
1496 struct timespec timeout;
1497
1498 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1499 consumer_data->consumer_thread_is_ready = 0;
1500
1501 /* Setup pthread condition */
1502 ret = pthread_condattr_init(&consumer_data->condattr);
1503 if (ret != 0) {
1504 errno = ret;
1505 PERROR("pthread_condattr_init consumer data");
1506 goto error;
1507 }
1508
1509 /*
1510 * Set the monotonic clock in order to make sure we DO NOT jump in time
1511 * between the clock_gettime() call and the timedwait call. See bug #324
1512 * for a more details and how we noticed it.
1513 */
1514 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1515 if (ret != 0) {
1516 errno = ret;
1517 PERROR("pthread_condattr_setclock consumer data");
1518 goto error;
1519 }
1520
1521 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1522 if (ret != 0) {
1523 errno = ret;
1524 PERROR("pthread_cond_init consumer data");
1525 goto error;
1526 }
1527
1528 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1529 consumer_data);
1530 if (ret != 0) {
1531 PERROR("pthread_create consumer");
1532 ret = -1;
1533 goto error;
1534 }
1535
1536 /* We are about to wait on a pthread condition */
1537 pthread_mutex_lock(&consumer_data->cond_mutex);
1538
1539 /* Get time for sem_timedwait absolute timeout */
1540 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1541 /*
1542 * Set the timeout for the condition timed wait even if the clock gettime
1543 * call fails since we might loop on that call and we want to avoid to
1544 * increment the timeout too many times.
1545 */
1546 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1547
1548 /*
1549 * The following loop COULD be skipped in some conditions so this is why we
1550 * set ret to 0 in order to make sure at least one round of the loop is
1551 * done.
1552 */
1553 ret = 0;
1554
1555 /*
1556 * Loop until the condition is reached or when a timeout is reached. Note
1557 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1558 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1559 * possible. This loop does not take any chances and works with both of
1560 * them.
1561 */
1562 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1563 if (clock_ret < 0) {
1564 PERROR("clock_gettime spawn consumer");
1565 /* Infinite wait for the consumerd thread to be ready */
1566 ret = pthread_cond_wait(&consumer_data->cond,
1567 &consumer_data->cond_mutex);
1568 } else {
1569 ret = pthread_cond_timedwait(&consumer_data->cond,
1570 &consumer_data->cond_mutex, &timeout);
1571 }
1572 }
1573
1574 /* Release the pthread condition */
1575 pthread_mutex_unlock(&consumer_data->cond_mutex);
1576
1577 if (ret != 0) {
1578 errno = ret;
1579 if (ret == ETIMEDOUT) {
1580 /*
1581 * Call has timed out so we kill the kconsumerd_thread and return
1582 * an error.
1583 */
1584 ERR("Condition timed out. The consumer thread was never ready."
1585 " Killing it");
1586 ret = pthread_cancel(consumer_data->thread);
1587 if (ret < 0) {
1588 PERROR("pthread_cancel consumer thread");
1589 }
1590 } else {
1591 PERROR("pthread_cond_wait failed consumer thread");
1592 }
1593 goto error;
1594 }
1595
1596 pthread_mutex_lock(&consumer_data->pid_mutex);
1597 if (consumer_data->pid == 0) {
1598 ERR("Consumerd did not start");
1599 pthread_mutex_unlock(&consumer_data->pid_mutex);
1600 goto error;
1601 }
1602 pthread_mutex_unlock(&consumer_data->pid_mutex);
1603
1604 return 0;
1605
1606 error:
1607 return ret;
1608 }
1609
1610 /*
1611 * Join consumer thread
1612 */
1613 static int join_consumer_thread(struct consumer_data *consumer_data)
1614 {
1615 void *status;
1616 int ret;
1617
1618 /* Consumer pid must be a real one. */
1619 if (consumer_data->pid > 0) {
1620 ret = kill(consumer_data->pid, SIGTERM);
1621 if (ret) {
1622 ERR("Error killing consumer daemon");
1623 return ret;
1624 }
1625 return pthread_join(consumer_data->thread, &status);
1626 } else {
1627 return 0;
1628 }
1629 }
1630
1631 /*
1632 * Fork and exec a consumer daemon (consumerd).
1633 *
1634 * Return pid if successful else -1.
1635 */
1636 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1637 {
1638 int ret;
1639 pid_t pid;
1640 const char *consumer_to_use;
1641 const char *verbosity;
1642 struct stat st;
1643
1644 DBG("Spawning consumerd");
1645
1646 pid = fork();
1647 if (pid == 0) {
1648 /*
1649 * Exec consumerd.
1650 */
1651 if (opt_verbose_consumer) {
1652 verbosity = "--verbose";
1653 } else {
1654 verbosity = "--quiet";
1655 }
1656 switch (consumer_data->type) {
1657 case LTTNG_CONSUMER_KERNEL:
1658 /*
1659 * Find out which consumerd to execute. We will first try the
1660 * 64-bit path, then the sessiond's installation directory, and
1661 * fallback on the 32-bit one,
1662 */
1663 DBG3("Looking for a kernel consumer at these locations:");
1664 DBG3(" 1) %s", consumerd64_bin);
1665 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1666 DBG3(" 3) %s", consumerd32_bin);
1667 if (stat(consumerd64_bin, &st) == 0) {
1668 DBG3("Found location #1");
1669 consumer_to_use = consumerd64_bin;
1670 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1671 DBG3("Found location #2");
1672 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1673 } else if (stat(consumerd32_bin, &st) == 0) {
1674 DBG3("Found location #3");
1675 consumer_to_use = consumerd32_bin;
1676 } else {
1677 DBG("Could not find any valid consumerd executable");
1678 break;
1679 }
1680 DBG("Using kernel consumer at: %s", consumer_to_use);
1681 execl(consumer_to_use,
1682 "lttng-consumerd", verbosity, "-k",
1683 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1684 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1685 NULL);
1686 break;
1687 case LTTNG_CONSUMER64_UST:
1688 {
1689 char *tmpnew = NULL;
1690
1691 if (consumerd64_libdir[0] != '\0') {
1692 char *tmp;
1693 size_t tmplen;
1694
1695 tmp = getenv("LD_LIBRARY_PATH");
1696 if (!tmp) {
1697 tmp = "";
1698 }
1699 tmplen = strlen("LD_LIBRARY_PATH=")
1700 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1701 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1702 if (!tmpnew) {
1703 ret = -ENOMEM;
1704 goto error;
1705 }
1706 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1707 strcat(tmpnew, consumerd64_libdir);
1708 if (tmp[0] != '\0') {
1709 strcat(tmpnew, ":");
1710 strcat(tmpnew, tmp);
1711 }
1712 ret = putenv(tmpnew);
1713 if (ret) {
1714 ret = -errno;
1715 goto error;
1716 }
1717 }
1718 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1719 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1720 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1721 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1722 NULL);
1723 if (consumerd64_libdir[0] != '\0') {
1724 free(tmpnew);
1725 }
1726 if (ret) {
1727 goto error;
1728 }
1729 break;
1730 }
1731 case LTTNG_CONSUMER32_UST:
1732 {
1733 char *tmpnew = NULL;
1734
1735 if (consumerd32_libdir[0] != '\0') {
1736 char *tmp;
1737 size_t tmplen;
1738
1739 tmp = getenv("LD_LIBRARY_PATH");
1740 if (!tmp) {
1741 tmp = "";
1742 }
1743 tmplen = strlen("LD_LIBRARY_PATH=")
1744 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1745 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1746 if (!tmpnew) {
1747 ret = -ENOMEM;
1748 goto error;
1749 }
1750 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1751 strcat(tmpnew, consumerd32_libdir);
1752 if (tmp[0] != '\0') {
1753 strcat(tmpnew, ":");
1754 strcat(tmpnew, tmp);
1755 }
1756 ret = putenv(tmpnew);
1757 if (ret) {
1758 ret = -errno;
1759 goto error;
1760 }
1761 }
1762 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1763 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1764 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1765 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1766 NULL);
1767 if (consumerd32_libdir[0] != '\0') {
1768 free(tmpnew);
1769 }
1770 if (ret) {
1771 goto error;
1772 }
1773 break;
1774 }
1775 default:
1776 PERROR("unknown consumer type");
1777 exit(EXIT_FAILURE);
1778 }
1779 if (errno != 0) {
1780 PERROR("kernel start consumer exec");
1781 }
1782 exit(EXIT_FAILURE);
1783 } else if (pid > 0) {
1784 ret = pid;
1785 } else {
1786 PERROR("start consumer fork");
1787 ret = -errno;
1788 }
1789 error:
1790 return ret;
1791 }
1792
1793 /*
1794 * Spawn the consumerd daemon and session daemon thread.
1795 */
1796 static int start_consumerd(struct consumer_data *consumer_data)
1797 {
1798 int ret;
1799
1800 pthread_mutex_lock(&consumer_data->pid_mutex);
1801 if (consumer_data->pid != 0) {
1802 pthread_mutex_unlock(&consumer_data->pid_mutex);
1803 goto end;
1804 }
1805
1806 ret = spawn_consumerd(consumer_data);
1807 if (ret < 0) {
1808 ERR("Spawning consumerd failed");
1809 pthread_mutex_unlock(&consumer_data->pid_mutex);
1810 goto error;
1811 }
1812
1813 /* Setting up the consumer_data pid */
1814 consumer_data->pid = ret;
1815 DBG2("Consumer pid %d", consumer_data->pid);
1816 pthread_mutex_unlock(&consumer_data->pid_mutex);
1817
1818 DBG2("Spawning consumer control thread");
1819 ret = spawn_consumer_thread(consumer_data);
1820 if (ret < 0) {
1821 ERR("Fatal error spawning consumer control thread");
1822 goto error;
1823 }
1824
1825 end:
1826 return 0;
1827
1828 error:
1829 return ret;
1830 }
1831
1832 /*
1833 * Compute health status of each consumer. If one of them is zero (bad
1834 * state), we return 0.
1835 */
1836 static int check_consumer_health(void)
1837 {
1838 int ret;
1839
1840 ret = health_check_state(&kconsumer_data.health) &&
1841 health_check_state(&ustconsumer32_data.health) &&
1842 health_check_state(&ustconsumer64_data.health);
1843
1844 DBG3("Health consumer check %d", ret);
1845
1846 return ret;
1847 }
1848
1849 /*
1850 * Setup necessary data for kernel tracer action.
1851 */
1852 static int init_kernel_tracer(void)
1853 {
1854 int ret;
1855
1856 /* Modprobe lttng kernel modules */
1857 ret = modprobe_lttng_control();
1858 if (ret < 0) {
1859 goto error;
1860 }
1861
1862 /* Open debugfs lttng */
1863 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1864 if (kernel_tracer_fd < 0) {
1865 DBG("Failed to open %s", module_proc_lttng);
1866 ret = -1;
1867 goto error_open;
1868 }
1869
1870 /* Validate kernel version */
1871 ret = kernel_validate_version(kernel_tracer_fd);
1872 if (ret < 0) {
1873 goto error_version;
1874 }
1875
1876 ret = modprobe_lttng_data();
1877 if (ret < 0) {
1878 goto error_modules;
1879 }
1880
1881 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1882 return 0;
1883
1884 error_version:
1885 modprobe_remove_lttng_control();
1886 ret = close(kernel_tracer_fd);
1887 if (ret) {
1888 PERROR("close");
1889 }
1890 kernel_tracer_fd = -1;
1891 return LTTNG_ERR_KERN_VERSION;
1892
1893 error_modules:
1894 ret = close(kernel_tracer_fd);
1895 if (ret) {
1896 PERROR("close");
1897 }
1898
1899 error_open:
1900 modprobe_remove_lttng_control();
1901
1902 error:
1903 WARN("No kernel tracer available");
1904 kernel_tracer_fd = -1;
1905 if (!is_root) {
1906 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1907 } else {
1908 return LTTNG_ERR_KERN_NA;
1909 }
1910 }
1911
1912
1913 /*
1914 * Copy consumer output from the tracing session to the domain session. The
1915 * function also applies the right modification on a per domain basis for the
1916 * trace files destination directory.
1917 */
1918 static int copy_session_consumer(int domain, struct ltt_session *session)
1919 {
1920 int ret;
1921 const char *dir_name;
1922 struct consumer_output *consumer;
1923
1924 assert(session);
1925 assert(session->consumer);
1926
1927 switch (domain) {
1928 case LTTNG_DOMAIN_KERNEL:
1929 DBG3("Copying tracing session consumer output in kernel session");
1930 /*
1931 * XXX: We should audit the session creation and what this function
1932 * does "extra" in order to avoid a destroy since this function is used
1933 * in the domain session creation (kernel and ust) only. Same for UST
1934 * domain.
1935 */
1936 if (session->kernel_session->consumer) {
1937 consumer_destroy_output(session->kernel_session->consumer);
1938 }
1939 session->kernel_session->consumer =
1940 consumer_copy_output(session->consumer);
1941 /* Ease our life a bit for the next part */
1942 consumer = session->kernel_session->consumer;
1943 dir_name = DEFAULT_KERNEL_TRACE_DIR;
1944 break;
1945 case LTTNG_DOMAIN_UST:
1946 DBG3("Copying tracing session consumer output in UST session");
1947 if (session->ust_session->consumer) {
1948 consumer_destroy_output(session->ust_session->consumer);
1949 }
1950 session->ust_session->consumer =
1951 consumer_copy_output(session->consumer);
1952 /* Ease our life a bit for the next part */
1953 consumer = session->ust_session->consumer;
1954 dir_name = DEFAULT_UST_TRACE_DIR;
1955 break;
1956 default:
1957 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
1958 goto error;
1959 }
1960
1961 /* Append correct directory to subdir */
1962 strncat(consumer->subdir, dir_name,
1963 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
1964 DBG3("Copy session consumer subdir %s", consumer->subdir);
1965
1966 ret = LTTNG_OK;
1967
1968 error:
1969 return ret;
1970 }
1971
1972 /*
1973 * Create an UST session and add it to the session ust list.
1974 */
1975 static int create_ust_session(struct ltt_session *session,
1976 struct lttng_domain *domain)
1977 {
1978 int ret;
1979 struct ltt_ust_session *lus = NULL;
1980
1981 assert(session);
1982 assert(domain);
1983 assert(session->consumer);
1984
1985 switch (domain->type) {
1986 case LTTNG_DOMAIN_UST:
1987 break;
1988 default:
1989 ERR("Unknown UST domain on create session %d", domain->type);
1990 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
1991 goto error;
1992 }
1993
1994 DBG("Creating UST session");
1995
1996 lus = trace_ust_create_session(session->path, session->id, domain);
1997 if (lus == NULL) {
1998 ret = LTTNG_ERR_UST_SESS_FAIL;
1999 goto error;
2000 }
2001
2002 lus->uid = session->uid;
2003 lus->gid = session->gid;
2004 session->ust_session = lus;
2005
2006 /* Copy session output to the newly created UST session */
2007 ret = copy_session_consumer(domain->type, session);
2008 if (ret != LTTNG_OK) {
2009 goto error;
2010 }
2011
2012 return LTTNG_OK;
2013
2014 error:
2015 free(lus);
2016 session->ust_session = NULL;
2017 return ret;
2018 }
2019
2020 /*
2021 * Create a kernel tracer session then create the default channel.
2022 */
2023 static int create_kernel_session(struct ltt_session *session)
2024 {
2025 int ret;
2026
2027 DBG("Creating kernel session");
2028
2029 ret = kernel_create_session(session, kernel_tracer_fd);
2030 if (ret < 0) {
2031 ret = LTTNG_ERR_KERN_SESS_FAIL;
2032 goto error;
2033 }
2034
2035 /* Code flow safety */
2036 assert(session->kernel_session);
2037
2038 /* Copy session output to the newly created Kernel session */
2039 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2040 if (ret != LTTNG_OK) {
2041 goto error;
2042 }
2043
2044 /* Create directory(ies) on local filesystem. */
2045 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2046 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2047 ret = run_as_mkdir_recursive(
2048 session->kernel_session->consumer->dst.trace_path,
2049 S_IRWXU | S_IRWXG, session->uid, session->gid);
2050 if (ret < 0) {
2051 if (ret != -EEXIST) {
2052 ERR("Trace directory creation error");
2053 goto error;
2054 }
2055 }
2056 }
2057
2058 session->kernel_session->uid = session->uid;
2059 session->kernel_session->gid = session->gid;
2060
2061 return LTTNG_OK;
2062
2063 error:
2064 trace_kernel_destroy_session(session->kernel_session);
2065 session->kernel_session = NULL;
2066 return ret;
2067 }
2068
2069 /*
2070 * Count number of session permitted by uid/gid.
2071 */
2072 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2073 {
2074 unsigned int i = 0;
2075 struct ltt_session *session;
2076
2077 DBG("Counting number of available session for UID %d GID %d",
2078 uid, gid);
2079 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2080 /*
2081 * Only list the sessions the user can control.
2082 */
2083 if (!session_access_ok(session, uid, gid)) {
2084 continue;
2085 }
2086 i++;
2087 }
2088 return i;
2089 }
2090
2091 /*
2092 * Process the command requested by the lttng client within the command
2093 * context structure. This function make sure that the return structure (llm)
2094 * is set and ready for transmission before returning.
2095 *
2096 * Return any error encountered or 0 for success.
2097 *
2098 * "sock" is only used for special-case var. len data.
2099 */
2100 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2101 int *sock_error)
2102 {
2103 int ret = LTTNG_OK;
2104 int need_tracing_session = 1;
2105 int need_domain;
2106
2107 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2108
2109 *sock_error = 0;
2110
2111 switch (cmd_ctx->lsm->cmd_type) {
2112 case LTTNG_CREATE_SESSION:
2113 case LTTNG_DESTROY_SESSION:
2114 case LTTNG_LIST_SESSIONS:
2115 case LTTNG_LIST_DOMAINS:
2116 case LTTNG_START_TRACE:
2117 case LTTNG_STOP_TRACE:
2118 case LTTNG_DATA_AVAILABLE:
2119 need_domain = 0;
2120 break;
2121 default:
2122 need_domain = 1;
2123 }
2124
2125 if (opt_no_kernel && need_domain
2126 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2127 if (!is_root) {
2128 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2129 } else {
2130 ret = LTTNG_ERR_KERN_NA;
2131 }
2132 goto error;
2133 }
2134
2135 /* Deny register consumer if we already have a spawned consumer. */
2136 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2137 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2138 if (kconsumer_data.pid > 0) {
2139 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2140 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2141 goto error;
2142 }
2143 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2144 }
2145
2146 /*
2147 * Check for command that don't needs to allocate a returned payload. We do
2148 * this here so we don't have to make the call for no payload at each
2149 * command.
2150 */
2151 switch(cmd_ctx->lsm->cmd_type) {
2152 case LTTNG_LIST_SESSIONS:
2153 case LTTNG_LIST_TRACEPOINTS:
2154 case LTTNG_LIST_TRACEPOINT_FIELDS:
2155 case LTTNG_LIST_DOMAINS:
2156 case LTTNG_LIST_CHANNELS:
2157 case LTTNG_LIST_EVENTS:
2158 break;
2159 default:
2160 /* Setup lttng message with no payload */
2161 ret = setup_lttng_msg(cmd_ctx, 0);
2162 if (ret < 0) {
2163 /* This label does not try to unlock the session */
2164 goto init_setup_error;
2165 }
2166 }
2167
2168 /* Commands that DO NOT need a session. */
2169 switch (cmd_ctx->lsm->cmd_type) {
2170 case LTTNG_CREATE_SESSION:
2171 case LTTNG_CALIBRATE:
2172 case LTTNG_LIST_SESSIONS:
2173 case LTTNG_LIST_TRACEPOINTS:
2174 case LTTNG_LIST_TRACEPOINT_FIELDS:
2175 need_tracing_session = 0;
2176 break;
2177 default:
2178 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2179 /*
2180 * We keep the session list lock across _all_ commands
2181 * for now, because the per-session lock does not
2182 * handle teardown properly.
2183 */
2184 session_lock_list();
2185 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2186 if (cmd_ctx->session == NULL) {
2187 if (cmd_ctx->lsm->session.name != NULL) {
2188 ret = LTTNG_ERR_SESS_NOT_FOUND;
2189 } else {
2190 /* If no session name specified */
2191 ret = LTTNG_ERR_SELECT_SESS;
2192 }
2193 goto error;
2194 } else {
2195 /* Acquire lock for the session */
2196 session_lock(cmd_ctx->session);
2197 }
2198 break;
2199 }
2200
2201 if (!need_domain) {
2202 goto skip_domain;
2203 }
2204
2205 /*
2206 * Check domain type for specific "pre-action".
2207 */
2208 switch (cmd_ctx->lsm->domain.type) {
2209 case LTTNG_DOMAIN_KERNEL:
2210 if (!is_root) {
2211 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2212 goto error;
2213 }
2214
2215 /* Kernel tracer check */
2216 if (kernel_tracer_fd == -1) {
2217 /* Basically, load kernel tracer modules */
2218 ret = init_kernel_tracer();
2219 if (ret != 0) {
2220 goto error;
2221 }
2222 }
2223
2224 /* Consumer is in an ERROR state. Report back to client */
2225 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2226 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2227 goto error;
2228 }
2229
2230 /* Need a session for kernel command */
2231 if (need_tracing_session) {
2232 if (cmd_ctx->session->kernel_session == NULL) {
2233 ret = create_kernel_session(cmd_ctx->session);
2234 if (ret < 0) {
2235 ret = LTTNG_ERR_KERN_SESS_FAIL;
2236 goto error;
2237 }
2238 }
2239
2240 /* Start the kernel consumer daemon */
2241 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2242 if (kconsumer_data.pid == 0 &&
2243 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2244 cmd_ctx->session->start_consumer) {
2245 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2246 ret = start_consumerd(&kconsumer_data);
2247 if (ret < 0) {
2248 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2249 goto error;
2250 }
2251 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2252 } else {
2253 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2254 }
2255
2256 /*
2257 * The consumer was just spawned so we need to add the socket to
2258 * the consumer output of the session if exist.
2259 */
2260 ret = consumer_create_socket(&kconsumer_data,
2261 cmd_ctx->session->kernel_session->consumer);
2262 if (ret < 0) {
2263 goto error;
2264 }
2265 }
2266
2267 break;
2268 case LTTNG_DOMAIN_UST:
2269 {
2270 /* Consumer is in an ERROR state. Report back to client */
2271 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2272 ret = LTTNG_ERR_NO_USTCONSUMERD;
2273 goto error;
2274 }
2275
2276 if (need_tracing_session) {
2277 /* Create UST session if none exist. */
2278 if (cmd_ctx->session->ust_session == NULL) {
2279 ret = create_ust_session(cmd_ctx->session,
2280 &cmd_ctx->lsm->domain);
2281 if (ret != LTTNG_OK) {
2282 goto error;
2283 }
2284 }
2285
2286 /* Start the UST consumer daemons */
2287 /* 64-bit */
2288 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2289 if (consumerd64_bin[0] != '\0' &&
2290 ustconsumer64_data.pid == 0 &&
2291 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2292 cmd_ctx->session->start_consumer) {
2293 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2294 ret = start_consumerd(&ustconsumer64_data);
2295 if (ret < 0) {
2296 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2297 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2298 goto error;
2299 }
2300
2301 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2302 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2303 } else {
2304 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2305 }
2306
2307 /*
2308 * Setup socket for consumer 64 bit. No need for atomic access
2309 * since it was set above and can ONLY be set in this thread.
2310 */
2311 ret = consumer_create_socket(&ustconsumer64_data,
2312 cmd_ctx->session->ust_session->consumer);
2313 if (ret < 0) {
2314 goto error;
2315 }
2316
2317 /* 32-bit */
2318 if (consumerd32_bin[0] != '\0' &&
2319 ustconsumer32_data.pid == 0 &&
2320 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2321 cmd_ctx->session->start_consumer) {
2322 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2323 ret = start_consumerd(&ustconsumer32_data);
2324 if (ret < 0) {
2325 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2326 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2327 goto error;
2328 }
2329
2330 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2331 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2332 } else {
2333 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2334 }
2335
2336 /*
2337 * Setup socket for consumer 64 bit. No need for atomic access
2338 * since it was set above and can ONLY be set in this thread.
2339 */
2340 ret = consumer_create_socket(&ustconsumer32_data,
2341 cmd_ctx->session->ust_session->consumer);
2342 if (ret < 0) {
2343 goto error;
2344 }
2345 }
2346 break;
2347 }
2348 default:
2349 break;
2350 }
2351 skip_domain:
2352
2353 /* Validate consumer daemon state when start/stop trace command */
2354 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2355 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2356 switch (cmd_ctx->lsm->domain.type) {
2357 case LTTNG_DOMAIN_UST:
2358 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2359 ret = LTTNG_ERR_NO_USTCONSUMERD;
2360 goto error;
2361 }
2362 break;
2363 case LTTNG_DOMAIN_KERNEL:
2364 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2365 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2366 goto error;
2367 }
2368 break;
2369 }
2370 }
2371
2372 /*
2373 * Check that the UID or GID match that of the tracing session.
2374 * The root user can interact with all sessions.
2375 */
2376 if (need_tracing_session) {
2377 if (!session_access_ok(cmd_ctx->session,
2378 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2379 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2380 ret = LTTNG_ERR_EPERM;
2381 goto error;
2382 }
2383 }
2384
2385 /* Process by command type */
2386 switch (cmd_ctx->lsm->cmd_type) {
2387 case LTTNG_ADD_CONTEXT:
2388 {
2389 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2390 cmd_ctx->lsm->u.context.channel_name,
2391 cmd_ctx->lsm->u.context.event_name,
2392 &cmd_ctx->lsm->u.context.ctx);
2393 break;
2394 }
2395 case LTTNG_DISABLE_CHANNEL:
2396 {
2397 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2398 cmd_ctx->lsm->u.disable.channel_name);
2399 break;
2400 }
2401 case LTTNG_DISABLE_EVENT:
2402 {
2403 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2404 cmd_ctx->lsm->u.disable.channel_name,
2405 cmd_ctx->lsm->u.disable.name);
2406 break;
2407 }
2408 case LTTNG_DISABLE_ALL_EVENT:
2409 {
2410 DBG("Disabling all events");
2411
2412 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2413 cmd_ctx->lsm->u.disable.channel_name);
2414 break;
2415 }
2416 case LTTNG_DISABLE_CONSUMER:
2417 {
2418 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2419 break;
2420 }
2421 case LTTNG_ENABLE_CHANNEL:
2422 {
2423 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2424 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2425 break;
2426 }
2427 case LTTNG_ENABLE_CONSUMER:
2428 {
2429 /*
2430 * XXX: 0 means that this URI should be applied on the session. Should
2431 * be a DOMAIN enuam.
2432 */
2433 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2434 if (ret != LTTNG_OK) {
2435 goto error;
2436 }
2437
2438 if (cmd_ctx->lsm->domain.type == 0) {
2439 /* Add the URI for the UST session if a consumer is present. */
2440 if (cmd_ctx->session->ust_session &&
2441 cmd_ctx->session->ust_session->consumer) {
2442 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2443 } else if (cmd_ctx->session->kernel_session &&
2444 cmd_ctx->session->kernel_session->consumer) {
2445 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2446 cmd_ctx->session);
2447 }
2448 }
2449 break;
2450 }
2451 case LTTNG_ENABLE_EVENT:
2452 {
2453 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2454 cmd_ctx->lsm->u.enable.channel_name,
2455 &cmd_ctx->lsm->u.enable.event, kernel_poll_pipe[1]);
2456 break;
2457 }
2458 case LTTNG_ENABLE_ALL_EVENT:
2459 {
2460 DBG("Enabling all events");
2461
2462 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2463 cmd_ctx->lsm->u.enable.channel_name,
2464 cmd_ctx->lsm->u.enable.event.type, kernel_poll_pipe[1]);
2465 break;
2466 }
2467 case LTTNG_LIST_TRACEPOINTS:
2468 {
2469 struct lttng_event *events;
2470 ssize_t nb_events;
2471
2472 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2473 if (nb_events < 0) {
2474 /* Return value is a negative lttng_error_code. */
2475 ret = -nb_events;
2476 goto error;
2477 }
2478
2479 /*
2480 * Setup lttng message with payload size set to the event list size in
2481 * bytes and then copy list into the llm payload.
2482 */
2483 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2484 if (ret < 0) {
2485 free(events);
2486 goto setup_error;
2487 }
2488
2489 /* Copy event list into message payload */
2490 memcpy(cmd_ctx->llm->payload, events,
2491 sizeof(struct lttng_event) * nb_events);
2492
2493 free(events);
2494
2495 ret = LTTNG_OK;
2496 break;
2497 }
2498 case LTTNG_LIST_TRACEPOINT_FIELDS:
2499 {
2500 struct lttng_event_field *fields;
2501 ssize_t nb_fields;
2502
2503 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2504 &fields);
2505 if (nb_fields < 0) {
2506 /* Return value is a negative lttng_error_code. */
2507 ret = -nb_fields;
2508 goto error;
2509 }
2510
2511 /*
2512 * Setup lttng message with payload size set to the event list size in
2513 * bytes and then copy list into the llm payload.
2514 */
2515 ret = setup_lttng_msg(cmd_ctx,
2516 sizeof(struct lttng_event_field) * nb_fields);
2517 if (ret < 0) {
2518 free(fields);
2519 goto setup_error;
2520 }
2521
2522 /* Copy event list into message payload */
2523 memcpy(cmd_ctx->llm->payload, fields,
2524 sizeof(struct lttng_event_field) * nb_fields);
2525
2526 free(fields);
2527
2528 ret = LTTNG_OK;
2529 break;
2530 }
2531 case LTTNG_SET_CONSUMER_URI:
2532 {
2533 size_t nb_uri, len;
2534 struct lttng_uri *uris;
2535
2536 nb_uri = cmd_ctx->lsm->u.uri.size;
2537 len = nb_uri * sizeof(struct lttng_uri);
2538
2539 if (nb_uri == 0) {
2540 ret = LTTNG_ERR_INVALID;
2541 goto error;
2542 }
2543
2544 uris = zmalloc(len);
2545 if (uris == NULL) {
2546 ret = LTTNG_ERR_FATAL;
2547 goto error;
2548 }
2549
2550 /* Receive variable len data */
2551 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2552 ret = lttcomm_recv_unix_sock(sock, uris, len);
2553 if (ret <= 0) {
2554 DBG("No URIs received from client... continuing");
2555 *sock_error = 1;
2556 ret = LTTNG_ERR_SESSION_FAIL;
2557 free(uris);
2558 goto error;
2559 }
2560
2561 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2562 nb_uri, uris);
2563 if (ret != LTTNG_OK) {
2564 free(uris);
2565 goto error;
2566 }
2567
2568 /*
2569 * XXX: 0 means that this URI should be applied on the session. Should
2570 * be a DOMAIN enuam.
2571 */
2572 if (cmd_ctx->lsm->domain.type == 0) {
2573 /* Add the URI for the UST session if a consumer is present. */
2574 if (cmd_ctx->session->ust_session &&
2575 cmd_ctx->session->ust_session->consumer) {
2576 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2577 nb_uri, uris);
2578 } else if (cmd_ctx->session->kernel_session &&
2579 cmd_ctx->session->kernel_session->consumer) {
2580 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2581 cmd_ctx->session, nb_uri, uris);
2582 }
2583 }
2584
2585 free(uris);
2586
2587 break;
2588 }
2589 case LTTNG_START_TRACE:
2590 {
2591 ret = cmd_start_trace(cmd_ctx->session);
2592 break;
2593 }
2594 case LTTNG_STOP_TRACE:
2595 {
2596 ret = cmd_stop_trace(cmd_ctx->session);
2597 break;
2598 }
2599 case LTTNG_CREATE_SESSION:
2600 {
2601 size_t nb_uri, len;
2602 struct lttng_uri *uris = NULL;
2603
2604 nb_uri = cmd_ctx->lsm->u.uri.size;
2605 len = nb_uri * sizeof(struct lttng_uri);
2606
2607 if (nb_uri > 0) {
2608 uris = zmalloc(len);
2609 if (uris == NULL) {
2610 ret = LTTNG_ERR_FATAL;
2611 goto error;
2612 }
2613
2614 /* Receive variable len data */
2615 DBG("Waiting for %zu URIs from client ...", nb_uri);
2616 ret = lttcomm_recv_unix_sock(sock, uris, len);
2617 if (ret <= 0) {
2618 DBG("No URIs received from client... continuing");
2619 *sock_error = 1;
2620 ret = LTTNG_ERR_SESSION_FAIL;
2621 free(uris);
2622 goto error;
2623 }
2624
2625 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2626 DBG("Creating session with ONE network URI is a bad call");
2627 ret = LTTNG_ERR_SESSION_FAIL;
2628 free(uris);
2629 goto error;
2630 }
2631 }
2632
2633 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2634 &cmd_ctx->creds);
2635
2636 free(uris);
2637
2638 break;
2639 }
2640 case LTTNG_DESTROY_SESSION:
2641 {
2642 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2643
2644 /* Set session to NULL so we do not unlock it after free. */
2645 cmd_ctx->session = NULL;
2646 break;
2647 }
2648 case LTTNG_LIST_DOMAINS:
2649 {
2650 ssize_t nb_dom;
2651 struct lttng_domain *domains;
2652
2653 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2654 if (nb_dom < 0) {
2655 /* Return value is a negative lttng_error_code. */
2656 ret = -nb_dom;
2657 goto error;
2658 }
2659
2660 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2661 if (ret < 0) {
2662 goto setup_error;
2663 }
2664
2665 /* Copy event list into message payload */
2666 memcpy(cmd_ctx->llm->payload, domains,
2667 nb_dom * sizeof(struct lttng_domain));
2668
2669 free(domains);
2670
2671 ret = LTTNG_OK;
2672 break;
2673 }
2674 case LTTNG_LIST_CHANNELS:
2675 {
2676 int nb_chan;
2677 struct lttng_channel *channels;
2678
2679 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2680 cmd_ctx->session, &channels);
2681 if (nb_chan < 0) {
2682 /* Return value is a negative lttng_error_code. */
2683 ret = -nb_chan;
2684 goto error;
2685 }
2686
2687 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2688 if (ret < 0) {
2689 goto setup_error;
2690 }
2691
2692 /* Copy event list into message payload */
2693 memcpy(cmd_ctx->llm->payload, channels,
2694 nb_chan * sizeof(struct lttng_channel));
2695
2696 free(channels);
2697
2698 ret = LTTNG_OK;
2699 break;
2700 }
2701 case LTTNG_LIST_EVENTS:
2702 {
2703 ssize_t nb_event;
2704 struct lttng_event *events = NULL;
2705
2706 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2707 cmd_ctx->lsm->u.list.channel_name, &events);
2708 if (nb_event < 0) {
2709 /* Return value is a negative lttng_error_code. */
2710 ret = -nb_event;
2711 goto error;
2712 }
2713
2714 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2715 if (ret < 0) {
2716 goto setup_error;
2717 }
2718
2719 /* Copy event list into message payload */
2720 memcpy(cmd_ctx->llm->payload, events,
2721 nb_event * sizeof(struct lttng_event));
2722
2723 free(events);
2724
2725 ret = LTTNG_OK;
2726 break;
2727 }
2728 case LTTNG_LIST_SESSIONS:
2729 {
2730 unsigned int nr_sessions;
2731
2732 session_lock_list();
2733 nr_sessions = lttng_sessions_count(
2734 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2735 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2736
2737 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2738 if (ret < 0) {
2739 session_unlock_list();
2740 goto setup_error;
2741 }
2742
2743 /* Filled the session array */
2744 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2745 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2746 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2747
2748 session_unlock_list();
2749
2750 ret = LTTNG_OK;
2751 break;
2752 }
2753 case LTTNG_CALIBRATE:
2754 {
2755 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2756 &cmd_ctx->lsm->u.calibrate);
2757 break;
2758 }
2759 case LTTNG_REGISTER_CONSUMER:
2760 {
2761 struct consumer_data *cdata;
2762
2763 switch (cmd_ctx->lsm->domain.type) {
2764 case LTTNG_DOMAIN_KERNEL:
2765 cdata = &kconsumer_data;
2766 break;
2767 default:
2768 ret = LTTNG_ERR_UND;
2769 goto error;
2770 }
2771
2772 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2773 cmd_ctx->lsm->u.reg.path, cdata);
2774 break;
2775 }
2776 case LTTNG_SET_FILTER:
2777 {
2778 struct lttng_filter_bytecode *bytecode;
2779
2780 if (cmd_ctx->lsm->u.filter.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2781 ret = LTTNG_ERR_FILTER_INVAL;
2782 goto error;
2783 }
2784 bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len);
2785 if (!bytecode) {
2786 ret = LTTNG_ERR_FILTER_NOMEM;
2787 goto error;
2788 }
2789 /* Receive var. len. data */
2790 DBG("Receiving var len data from client ...");
2791 ret = lttcomm_recv_unix_sock(sock, bytecode,
2792 cmd_ctx->lsm->u.filter.bytecode_len);
2793 if (ret <= 0) {
2794 DBG("Nothing recv() from client var len data... continuing");
2795 *sock_error = 1;
2796 ret = LTTNG_ERR_FILTER_INVAL;
2797 goto error;
2798 }
2799
2800 if (bytecode->len + sizeof(*bytecode)
2801 != cmd_ctx->lsm->u.filter.bytecode_len) {
2802 free(bytecode);
2803 ret = LTTNG_ERR_FILTER_INVAL;
2804 goto error;
2805 }
2806
2807 ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2808 cmd_ctx->lsm->u.filter.channel_name,
2809 cmd_ctx->lsm->u.filter.event_name,
2810 bytecode);
2811 break;
2812 }
2813 case LTTNG_DATA_AVAILABLE:
2814 {
2815 ret = cmd_data_available(cmd_ctx->session);
2816 break;
2817 }
2818 default:
2819 ret = LTTNG_ERR_UND;
2820 break;
2821 }
2822
2823 error:
2824 if (cmd_ctx->llm == NULL) {
2825 DBG("Missing llm structure. Allocating one.");
2826 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2827 goto setup_error;
2828 }
2829 }
2830 /* Set return code */
2831 cmd_ctx->llm->ret_code = ret;
2832 setup_error:
2833 if (cmd_ctx->session) {
2834 session_unlock(cmd_ctx->session);
2835 }
2836 if (need_tracing_session) {
2837 session_unlock_list();
2838 }
2839 init_setup_error:
2840 return ret;
2841 }
2842
2843 /*
2844 * Thread managing health check socket.
2845 */
2846 static void *thread_manage_health(void *data)
2847 {
2848 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2849 uint32_t revents, nb_fd;
2850 struct lttng_poll_event events;
2851 struct lttcomm_health_msg msg;
2852 struct lttcomm_health_data reply;
2853
2854 DBG("[thread] Manage health check started");
2855
2856 rcu_register_thread();
2857
2858 /* Create unix socket */
2859 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2860 if (sock < 0) {
2861 ERR("Unable to create health check Unix socket");
2862 ret = -1;
2863 goto error;
2864 }
2865
2866 ret = lttcomm_listen_unix_sock(sock);
2867 if (ret < 0) {
2868 goto error;
2869 }
2870
2871 /*
2872 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2873 * more will be added to this poll set.
2874 */
2875 ret = create_thread_poll_set(&events, 2);
2876 if (ret < 0) {
2877 goto error;
2878 }
2879
2880 /* Add the application registration socket */
2881 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2882 if (ret < 0) {
2883 goto error;
2884 }
2885
2886 while (1) {
2887 DBG("Health check ready");
2888
2889 nb_fd = LTTNG_POLL_GETNB(&events);
2890
2891 /* Inifinite blocking call, waiting for transmission */
2892 restart:
2893 ret = lttng_poll_wait(&events, -1);
2894 if (ret < 0) {
2895 /*
2896 * Restart interrupted system call.
2897 */
2898 if (errno == EINTR) {
2899 goto restart;
2900 }
2901 goto error;
2902 }
2903
2904 for (i = 0; i < nb_fd; i++) {
2905 /* Fetch once the poll data */
2906 revents = LTTNG_POLL_GETEV(&events, i);
2907 pollfd = LTTNG_POLL_GETFD(&events, i);
2908
2909 /* Thread quit pipe has been closed. Killing thread. */
2910 ret = check_thread_quit_pipe(pollfd, revents);
2911 if (ret) {
2912 err = 0;
2913 goto exit;
2914 }
2915
2916 /* Event on the registration socket */
2917 if (pollfd == sock) {
2918 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2919 ERR("Health socket poll error");
2920 goto error;
2921 }
2922 }
2923 }
2924
2925 new_sock = lttcomm_accept_unix_sock(sock);
2926 if (new_sock < 0) {
2927 goto error;
2928 }
2929
2930 DBG("Receiving data from client for health...");
2931 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
2932 if (ret <= 0) {
2933 DBG("Nothing recv() from client... continuing");
2934 ret = close(new_sock);
2935 if (ret) {
2936 PERROR("close");
2937 }
2938 new_sock = -1;
2939 continue;
2940 }
2941
2942 rcu_thread_online();
2943
2944 switch (msg.component) {
2945 case LTTNG_HEALTH_CMD:
2946 reply.ret_code = health_check_state(&health_thread_cmd);
2947 break;
2948 case LTTNG_HEALTH_APP_MANAGE:
2949 reply.ret_code = health_check_state(&health_thread_app_manage);
2950 break;
2951 case LTTNG_HEALTH_APP_REG:
2952 reply.ret_code = health_check_state(&health_thread_app_reg);
2953 break;
2954 case LTTNG_HEALTH_KERNEL:
2955 reply.ret_code = health_check_state(&health_thread_kernel);
2956 break;
2957 case LTTNG_HEALTH_CONSUMER:
2958 reply.ret_code = check_consumer_health();
2959 break;
2960 case LTTNG_HEALTH_ALL:
2961 reply.ret_code =
2962 health_check_state(&health_thread_app_manage) &&
2963 health_check_state(&health_thread_app_reg) &&
2964 health_check_state(&health_thread_cmd) &&
2965 health_check_state(&health_thread_kernel) &&
2966 check_consumer_health();
2967 break;
2968 default:
2969 reply.ret_code = LTTNG_ERR_UND;
2970 break;
2971 }
2972
2973 /*
2974 * Flip ret value since 0 is a success and 1 indicates a bad health for
2975 * the client where in the sessiond it is the opposite. Again, this is
2976 * just to make things easier for us poor developer which enjoy a lot
2977 * lazyness.
2978 */
2979 if (reply.ret_code == 0 || reply.ret_code == 1) {
2980 reply.ret_code = !reply.ret_code;
2981 }
2982
2983 DBG2("Health check return value %d", reply.ret_code);
2984
2985 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
2986 if (ret < 0) {
2987 ERR("Failed to send health data back to client");
2988 }
2989
2990 /* End of transmission */
2991 ret = close(new_sock);
2992 if (ret) {
2993 PERROR("close");
2994 }
2995 new_sock = -1;
2996 }
2997
2998 exit:
2999 error:
3000 if (err) {
3001 ERR("Health error occurred in %s", __func__);
3002 }
3003 DBG("Health check thread dying");
3004 unlink(health_unix_sock_path);
3005 if (sock >= 0) {
3006 ret = close(sock);
3007 if (ret) {
3008 PERROR("close");
3009 }
3010 }
3011 if (new_sock >= 0) {
3012 ret = close(new_sock);
3013 if (ret) {
3014 PERROR("close");
3015 }
3016 }
3017
3018 lttng_poll_clean(&events);
3019
3020 rcu_unregister_thread();
3021 return NULL;
3022 }
3023
3024 /*
3025 * This thread manage all clients request using the unix client socket for
3026 * communication.
3027 */
3028 static void *thread_manage_clients(void *data)
3029 {
3030 int sock = -1, ret, i, pollfd, err = -1;
3031 int sock_error;
3032 uint32_t revents, nb_fd;
3033 struct command_ctx *cmd_ctx = NULL;
3034 struct lttng_poll_event events;
3035
3036 DBG("[thread] Manage client started");
3037
3038 testpoint(thread_manage_clients);
3039
3040 rcu_register_thread();
3041
3042 health_code_update(&health_thread_cmd);
3043
3044 ret = lttcomm_listen_unix_sock(client_sock);
3045 if (ret < 0) {
3046 goto error;
3047 }
3048
3049 /*
3050 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3051 * more will be added to this poll set.
3052 */
3053 ret = create_thread_poll_set(&events, 2);
3054 if (ret < 0) {
3055 goto error;
3056 }
3057
3058 /* Add the application registration socket */
3059 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3060 if (ret < 0) {
3061 goto error;
3062 }
3063
3064 /*
3065 * Notify parent pid that we are ready to accept command for client side.
3066 */
3067 if (opt_sig_parent) {
3068 kill(ppid, SIGUSR1);
3069 }
3070
3071 testpoint(thread_manage_clients_before_loop);
3072
3073 health_code_update(&health_thread_cmd);
3074
3075 while (1) {
3076 DBG("Accepting client command ...");
3077
3078 nb_fd = LTTNG_POLL_GETNB(&events);
3079
3080 /* Inifinite blocking call, waiting for transmission */
3081 restart:
3082 health_poll_update(&health_thread_cmd);
3083 ret = lttng_poll_wait(&events, -1);
3084 health_poll_update(&health_thread_cmd);
3085 if (ret < 0) {
3086 /*
3087 * Restart interrupted system call.
3088 */
3089 if (errno == EINTR) {
3090 goto restart;
3091 }
3092 goto error;
3093 }
3094
3095 for (i = 0; i < nb_fd; i++) {
3096 /* Fetch once the poll data */
3097 revents = LTTNG_POLL_GETEV(&events, i);
3098 pollfd = LTTNG_POLL_GETFD(&events, i);
3099
3100 health_code_update(&health_thread_cmd);
3101
3102 /* Thread quit pipe has been closed. Killing thread. */
3103 ret = check_thread_quit_pipe(pollfd, revents);
3104 if (ret) {
3105 err = 0;
3106 goto exit;
3107 }
3108
3109 /* Event on the registration socket */
3110 if (pollfd == client_sock) {
3111 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3112 ERR("Client socket poll error");
3113 goto error;
3114 }
3115 }
3116 }
3117
3118 DBG("Wait for client response");
3119
3120 health_code_update(&health_thread_cmd);
3121
3122 sock = lttcomm_accept_unix_sock(client_sock);
3123 if (sock < 0) {
3124 goto error;
3125 }
3126
3127 /* Set socket option for credentials retrieval */
3128 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3129 if (ret < 0) {
3130 goto error;
3131 }
3132
3133 /* Allocate context command to process the client request */
3134 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3135 if (cmd_ctx == NULL) {
3136 PERROR("zmalloc cmd_ctx");
3137 goto error;
3138 }
3139
3140 /* Allocate data buffer for reception */
3141 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3142 if (cmd_ctx->lsm == NULL) {
3143 PERROR("zmalloc cmd_ctx->lsm");
3144 goto error;
3145 }
3146
3147 cmd_ctx->llm = NULL;
3148 cmd_ctx->session = NULL;
3149
3150 health_code_update(&health_thread_cmd);
3151
3152 /*
3153 * Data is received from the lttng client. The struct
3154 * lttcomm_session_msg (lsm) contains the command and data request of
3155 * the client.
3156 */
3157 DBG("Receiving data from client ...");
3158 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3159 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3160 if (ret <= 0) {
3161 DBG("Nothing recv() from client... continuing");
3162 ret = close(sock);
3163 if (ret) {
3164 PERROR("close");
3165 }
3166 sock = -1;
3167 clean_command_ctx(&cmd_ctx);
3168 continue;
3169 }
3170
3171 health_code_update(&health_thread_cmd);
3172
3173 // TODO: Validate cmd_ctx including sanity check for
3174 // security purpose.
3175
3176 rcu_thread_online();
3177 /*
3178 * This function dispatch the work to the kernel or userspace tracer
3179 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3180 * informations for the client. The command context struct contains
3181 * everything this function may needs.
3182 */
3183 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3184 rcu_thread_offline();
3185 if (ret < 0) {
3186 if (sock_error) {
3187 ret = close(sock);
3188 if (ret) {
3189 PERROR("close");
3190 }
3191 sock = -1;
3192 }
3193 /*
3194 * TODO: Inform client somehow of the fatal error. At
3195 * this point, ret < 0 means that a zmalloc failed
3196 * (ENOMEM). Error detected but still accept
3197 * command, unless a socket error has been
3198 * detected.
3199 */
3200 clean_command_ctx(&cmd_ctx);
3201 continue;
3202 }
3203
3204 health_code_update(&health_thread_cmd);
3205
3206 DBG("Sending response (size: %d, retcode: %s)",
3207 cmd_ctx->lttng_msg_size,
3208 lttng_strerror(-cmd_ctx->llm->ret_code));
3209 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3210 if (ret < 0) {
3211 ERR("Failed to send data back to client");
3212 }
3213
3214 /* End of transmission */
3215 ret = close(sock);
3216 if (ret) {
3217 PERROR("close");
3218 }
3219 sock = -1;
3220
3221 clean_command_ctx(&cmd_ctx);
3222
3223 health_code_update(&health_thread_cmd);
3224 }
3225
3226 exit:
3227 error:
3228 if (err) {
3229 health_error(&health_thread_cmd);
3230 ERR("Health error occurred in %s", __func__);
3231 }
3232 health_exit(&health_thread_cmd);
3233
3234 DBG("Client thread dying");
3235 unlink(client_unix_sock_path);
3236 if (client_sock >= 0) {
3237 ret = close(client_sock);
3238 if (ret) {
3239 PERROR("close");
3240 }
3241 }
3242 if (sock >= 0) {
3243 ret = close(sock);
3244 if (ret) {
3245 PERROR("close");
3246 }
3247 }
3248
3249 lttng_poll_clean(&events);
3250 clean_command_ctx(&cmd_ctx);
3251
3252 rcu_unregister_thread();
3253 return NULL;
3254 }
3255
3256
3257 /*
3258 * usage function on stderr
3259 */
3260 static void usage(void)
3261 {
3262 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3263 fprintf(stderr, " -h, --help Display this usage.\n");
3264 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3265 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3266 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3267 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3268 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3269 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3270 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3271 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3272 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3273 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3274 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3275 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3276 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3277 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3278 fprintf(stderr, " -V, --version Show version number.\n");
3279 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3280 fprintf(stderr, " -q, --quiet No output at all.\n");
3281 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3282 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3283 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3284 }
3285
3286 /*
3287 * daemon argument parsing
3288 */
3289 static int parse_args(int argc, char **argv)
3290 {
3291 int c;
3292
3293 static struct option long_options[] = {
3294 { "client-sock", 1, 0, 'c' },
3295 { "apps-sock", 1, 0, 'a' },
3296 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3297 { "kconsumerd-err-sock", 1, 0, 'E' },
3298 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3299 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3300 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3301 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3302 { "consumerd32-path", 1, 0, 'u' },
3303 { "consumerd32-libdir", 1, 0, 'U' },
3304 { "consumerd64-path", 1, 0, 't' },
3305 { "consumerd64-libdir", 1, 0, 'T' },
3306 { "daemonize", 0, 0, 'd' },
3307 { "sig-parent", 0, 0, 'S' },
3308 { "help", 0, 0, 'h' },
3309 { "group", 1, 0, 'g' },
3310 { "version", 0, 0, 'V' },
3311 { "quiet", 0, 0, 'q' },
3312 { "verbose", 0, 0, 'v' },
3313 { "verbose-consumer", 0, 0, 'Z' },
3314 { "no-kernel", 0, 0, 'N' },
3315 { NULL, 0, 0, 0 }
3316 };
3317
3318 while (1) {
3319 int option_index = 0;
3320 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3321 long_options, &option_index);
3322 if (c == -1) {
3323 break;
3324 }
3325
3326 switch (c) {
3327 case 0:
3328 fprintf(stderr, "option %s", long_options[option_index].name);
3329 if (optarg) {
3330 fprintf(stderr, " with arg %s\n", optarg);
3331 }
3332 break;
3333 case 'c':
3334 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3335 break;
3336 case 'a':
3337 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3338 break;
3339 case 'd':
3340 opt_daemon = 1;
3341 break;
3342 case 'g':
3343 opt_tracing_group = optarg;
3344 break;
3345 case 'h':
3346 usage();
3347 exit(EXIT_FAILURE);
3348 case 'V':
3349 fprintf(stdout, "%s\n", VERSION);
3350 exit(EXIT_SUCCESS);
3351 case 'S':
3352 opt_sig_parent = 1;
3353 break;
3354 case 'E':
3355 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3356 break;
3357 case 'C':
3358 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3359 break;
3360 case 'F':
3361 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3362 break;
3363 case 'D':
3364 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3365 break;
3366 case 'H':
3367 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3368 break;
3369 case 'G':
3370 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3371 break;
3372 case 'N':
3373 opt_no_kernel = 1;
3374 break;
3375 case 'q':
3376 lttng_opt_quiet = 1;
3377 break;
3378 case 'v':
3379 /* Verbose level can increase using multiple -v */
3380 lttng_opt_verbose += 1;
3381 break;
3382 case 'Z':
3383 opt_verbose_consumer += 1;
3384 break;
3385 case 'u':
3386 consumerd32_bin= optarg;
3387 break;
3388 case 'U':
3389 consumerd32_libdir = optarg;
3390 break;
3391 case 't':
3392 consumerd64_bin = optarg;
3393 break;
3394 case 'T':
3395 consumerd64_libdir = optarg;
3396 break;
3397 default:
3398 /* Unknown option or other error.
3399 * Error is printed by getopt, just return */
3400 return -1;
3401 }
3402 }
3403
3404 return 0;
3405 }
3406
3407 /*
3408 * Creates the two needed socket by the daemon.
3409 * apps_sock - The communication socket for all UST apps.
3410 * client_sock - The communication of the cli tool (lttng).
3411 */
3412 static int init_daemon_socket(void)
3413 {
3414 int ret = 0;
3415 mode_t old_umask;
3416
3417 old_umask = umask(0);
3418
3419 /* Create client tool unix socket */
3420 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3421 if (client_sock < 0) {
3422 ERR("Create unix sock failed: %s", client_unix_sock_path);
3423 ret = -1;
3424 goto end;
3425 }
3426
3427 /* File permission MUST be 660 */
3428 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3429 if (ret < 0) {
3430 ERR("Set file permissions failed: %s", client_unix_sock_path);
3431 PERROR("chmod");
3432 goto end;
3433 }
3434
3435 /* Create the application unix socket */
3436 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3437 if (apps_sock < 0) {
3438 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3439 ret = -1;
3440 goto end;
3441 }
3442
3443 /* File permission MUST be 666 */
3444 ret = chmod(apps_unix_sock_path,
3445 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3446 if (ret < 0) {
3447 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3448 PERROR("chmod");
3449 goto end;
3450 }
3451
3452 end:
3453 umask(old_umask);
3454 return ret;
3455 }
3456
3457 /*
3458 * Check if the global socket is available, and if a daemon is answering at the
3459 * other side. If yes, error is returned.
3460 */
3461 static int check_existing_daemon(void)
3462 {
3463 /* Is there anybody out there ? */
3464 if (lttng_session_daemon_alive()) {
3465 return -EEXIST;
3466 }
3467
3468 return 0;
3469 }
3470
3471 /*
3472 * Set the tracing group gid onto the client socket.
3473 *
3474 * Race window between mkdir and chown is OK because we are going from more
3475 * permissive (root.root) to less permissive (root.tracing).
3476 */
3477 static int set_permissions(char *rundir)
3478 {
3479 int ret;
3480 gid_t gid;
3481
3482 ret = allowed_group();
3483 if (ret < 0) {
3484 WARN("No tracing group detected");
3485 ret = 0;
3486 goto end;
3487 }
3488
3489 gid = ret;
3490
3491 /* Set lttng run dir */
3492 ret = chown(rundir, 0, gid);
3493 if (ret < 0) {
3494 ERR("Unable to set group on %s", rundir);
3495 PERROR("chown");
3496 }
3497
3498 /* Ensure tracing group can search the run dir */
3499 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3500 if (ret < 0) {
3501 ERR("Unable to set permissions on %s", rundir);
3502 PERROR("chmod");
3503 }
3504
3505 /* lttng client socket path */
3506 ret = chown(client_unix_sock_path, 0, gid);
3507 if (ret < 0) {
3508 ERR("Unable to set group on %s", client_unix_sock_path);
3509 PERROR("chown");
3510 }
3511
3512 /* kconsumer error socket path */
3513 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3514 if (ret < 0) {
3515 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3516 PERROR("chown");
3517 }
3518
3519 /* 64-bit ustconsumer error socket path */
3520 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3521 if (ret < 0) {
3522 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3523 PERROR("chown");
3524 }
3525
3526 /* 32-bit ustconsumer compat32 error socket path */
3527 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3528 if (ret < 0) {
3529 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3530 PERROR("chown");
3531 }
3532
3533 DBG("All permissions are set");
3534
3535 end:
3536 return ret;
3537 }
3538
3539 /*
3540 * Create the lttng run directory needed for all global sockets and pipe.
3541 */
3542 static int create_lttng_rundir(const char *rundir)
3543 {
3544 int ret;
3545
3546 DBG3("Creating LTTng run directory: %s", rundir);
3547
3548 ret = mkdir(rundir, S_IRWXU);
3549 if (ret < 0) {
3550 if (errno != EEXIST) {
3551 ERR("Unable to create %s", rundir);
3552 goto error;
3553 } else {
3554 ret = 0;
3555 }
3556 }
3557
3558 error:
3559 return ret;
3560 }
3561
3562 /*
3563 * Setup sockets and directory needed by the kconsumerd communication with the
3564 * session daemon.
3565 */
3566 static int set_consumer_sockets(struct consumer_data *consumer_data,
3567 const char *rundir)
3568 {
3569 int ret;
3570 char path[PATH_MAX];
3571
3572 switch (consumer_data->type) {
3573 case LTTNG_CONSUMER_KERNEL:
3574 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3575 break;
3576 case LTTNG_CONSUMER64_UST:
3577 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3578 break;
3579 case LTTNG_CONSUMER32_UST:
3580 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3581 break;
3582 default:
3583 ERR("Consumer type unknown");
3584 ret = -EINVAL;
3585 goto error;
3586 }
3587
3588 DBG2("Creating consumer directory: %s", path);
3589
3590 ret = mkdir(path, S_IRWXU);
3591 if (ret < 0) {
3592 if (errno != EEXIST) {
3593 PERROR("mkdir");
3594 ERR("Failed to create %s", path);
3595 goto error;
3596 }
3597 ret = -1;
3598 }
3599
3600 /* Create the kconsumerd error unix socket */
3601 consumer_data->err_sock =
3602 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3603 if (consumer_data->err_sock < 0) {
3604 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3605 ret = -1;
3606 goto error;
3607 }
3608
3609 /* File permission MUST be 660 */
3610 ret = chmod(consumer_data->err_unix_sock_path,
3611 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3612 if (ret < 0) {
3613 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3614 PERROR("chmod");
3615 goto error;
3616 }
3617
3618 error:
3619 return ret;
3620 }
3621
3622 /*
3623 * Signal handler for the daemon
3624 *
3625 * Simply stop all worker threads, leaving main() return gracefully after
3626 * joining all threads and calling cleanup().
3627 */
3628 static void sighandler(int sig)
3629 {
3630 switch (sig) {
3631 case SIGPIPE:
3632 DBG("SIGPIPE caught");
3633 return;
3634 case SIGINT:
3635 DBG("SIGINT caught");
3636 stop_threads();
3637 break;
3638 case SIGTERM:
3639 DBG("SIGTERM caught");
3640 stop_threads();
3641 break;
3642 default:
3643 break;
3644 }
3645 }
3646
3647 /*
3648 * Setup signal handler for :
3649 * SIGINT, SIGTERM, SIGPIPE
3650 */
3651 static int set_signal_handler(void)
3652 {
3653 int ret = 0;
3654 struct sigaction sa;
3655 sigset_t sigset;
3656
3657 if ((ret = sigemptyset(&sigset)) < 0) {
3658 PERROR("sigemptyset");
3659 return ret;
3660 }
3661
3662 sa.sa_handler = sighandler;
3663 sa.sa_mask = sigset;
3664 sa.sa_flags = 0;
3665 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3666 PERROR("sigaction");
3667 return ret;
3668 }
3669
3670 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3671 PERROR("sigaction");
3672 return ret;
3673 }
3674
3675 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3676 PERROR("sigaction");
3677 return ret;
3678 }
3679
3680 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3681
3682 return ret;
3683 }
3684
3685 /*
3686 * Set open files limit to unlimited. This daemon can open a large number of
3687 * file descriptors in order to consumer multiple kernel traces.
3688 */
3689 static void set_ulimit(void)
3690 {
3691 int ret;
3692 struct rlimit lim;
3693
3694 /* The kernel does not allowed an infinite limit for open files */
3695 lim.rlim_cur = 65535;
3696 lim.rlim_max = 65535;
3697
3698 ret = setrlimit(RLIMIT_NOFILE, &lim);
3699 if (ret < 0) {
3700 PERROR("failed to set open files limit");
3701 }
3702 }
3703
3704 /*
3705 * main
3706 */
3707 int main(int argc, char **argv)
3708 {
3709 int ret = 0;
3710 void *status;
3711 const char *home_path;
3712
3713 init_kernel_workarounds();
3714
3715 rcu_register_thread();
3716
3717 setup_consumerd_path();
3718
3719 /* Parse arguments */
3720 progname = argv[0];
3721 if ((ret = parse_args(argc, argv) < 0)) {
3722 goto error;
3723 }
3724
3725 /* Daemonize */
3726 if (opt_daemon) {
3727 int i;
3728
3729 /*
3730 * fork
3731 * child: setsid, close FD 0, 1, 2, chdir /
3732 * parent: exit (if fork is successful)
3733 */
3734 ret = daemon(0, 0);
3735 if (ret < 0) {
3736 PERROR("daemon");
3737 goto error;
3738 }
3739 /*
3740 * We are in the child. Make sure all other file
3741 * descriptors are closed, in case we are called with
3742 * more opened file descriptors than the standard ones.
3743 */
3744 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3745 (void) close(i);
3746 }
3747 }
3748
3749 /* Create thread quit pipe */
3750 if ((ret = init_thread_quit_pipe()) < 0) {
3751 goto error;
3752 }
3753
3754 /* Check if daemon is UID = 0 */
3755 is_root = !getuid();
3756
3757 if (is_root) {
3758 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3759
3760 /* Create global run dir with root access */
3761 ret = create_lttng_rundir(rundir);
3762 if (ret < 0) {
3763 goto error;
3764 }
3765
3766 if (strlen(apps_unix_sock_path) == 0) {
3767 snprintf(apps_unix_sock_path, PATH_MAX,
3768 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3769 }
3770
3771 if (strlen(client_unix_sock_path) == 0) {
3772 snprintf(client_unix_sock_path, PATH_MAX,
3773 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3774 }
3775
3776 /* Set global SHM for ust */
3777 if (strlen(wait_shm_path) == 0) {
3778 snprintf(wait_shm_path, PATH_MAX,
3779 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3780 }
3781
3782 if (strlen(health_unix_sock_path) == 0) {
3783 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3784 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3785 }
3786
3787 /* Setup kernel consumerd path */
3788 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3789 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3790 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3791 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3792
3793 DBG2("Kernel consumer err path: %s",
3794 kconsumer_data.err_unix_sock_path);
3795 DBG2("Kernel consumer cmd path: %s",
3796 kconsumer_data.cmd_unix_sock_path);
3797 } else {
3798 home_path = get_home_dir();
3799 if (home_path == NULL) {
3800 /* TODO: Add --socket PATH option */
3801 ERR("Can't get HOME directory for sockets creation.");
3802 ret = -EPERM;
3803 goto error;
3804 }
3805
3806 /*
3807 * Create rundir from home path. This will create something like
3808 * $HOME/.lttng
3809 */
3810 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3811 if (ret < 0) {
3812 ret = -ENOMEM;
3813 goto error;
3814 }
3815
3816 ret = create_lttng_rundir(rundir);
3817 if (ret < 0) {
3818 goto error;
3819 }
3820
3821 if (strlen(apps_unix_sock_path) == 0) {
3822 snprintf(apps_unix_sock_path, PATH_MAX,
3823 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3824 }
3825
3826 /* Set the cli tool unix socket path */
3827 if (strlen(client_unix_sock_path) == 0) {
3828 snprintf(client_unix_sock_path, PATH_MAX,
3829 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3830 }
3831
3832 /* Set global SHM for ust */
3833 if (strlen(wait_shm_path) == 0) {
3834 snprintf(wait_shm_path, PATH_MAX,
3835 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3836 }
3837
3838 /* Set health check Unix path */
3839 if (strlen(health_unix_sock_path) == 0) {
3840 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3841 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3842 }
3843 }
3844
3845 /* Set consumer initial state */
3846 kernel_consumerd_state = CONSUMER_STOPPED;
3847 ust_consumerd_state = CONSUMER_STOPPED;
3848
3849 DBG("Client socket path %s", client_unix_sock_path);
3850 DBG("Application socket path %s", apps_unix_sock_path);
3851 DBG("LTTng run directory path: %s", rundir);
3852
3853 /* 32 bits consumerd path setup */
3854 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
3855 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
3856 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
3857 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
3858
3859 DBG2("UST consumer 32 bits err path: %s",
3860 ustconsumer32_data.err_unix_sock_path);
3861 DBG2("UST consumer 32 bits cmd path: %s",
3862 ustconsumer32_data.cmd_unix_sock_path);
3863
3864 /* 64 bits consumerd path setup */
3865 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
3866 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
3867 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
3868 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
3869
3870 DBG2("UST consumer 64 bits err path: %s",
3871 ustconsumer64_data.err_unix_sock_path);
3872 DBG2("UST consumer 64 bits cmd path: %s",
3873 ustconsumer64_data.cmd_unix_sock_path);
3874
3875 /*
3876 * See if daemon already exist.
3877 */
3878 if ((ret = check_existing_daemon()) < 0) {
3879 ERR("Already running daemon.\n");
3880 /*
3881 * We do not goto exit because we must not cleanup()
3882 * because a daemon is already running.
3883 */
3884 goto error;
3885 }
3886
3887 /*
3888 * Init UST app hash table. Alloc hash table before this point since
3889 * cleanup() can get called after that point.
3890 */
3891 ust_app_ht_alloc();
3892
3893 /* After this point, we can safely call cleanup() with "goto exit" */
3894
3895 /*
3896 * These actions must be executed as root. We do that *after* setting up
3897 * the sockets path because we MUST make the check for another daemon using
3898 * those paths *before* trying to set the kernel consumer sockets and init
3899 * kernel tracer.
3900 */
3901 if (is_root) {
3902 ret = set_consumer_sockets(&kconsumer_data, rundir);
3903 if (ret < 0) {
3904 goto exit;
3905 }
3906
3907 /* Setup kernel tracer */
3908 if (!opt_no_kernel) {
3909 init_kernel_tracer();
3910 }
3911
3912 /* Set ulimit for open files */
3913 set_ulimit();
3914 }
3915 /* init lttng_fd tracking must be done after set_ulimit. */
3916 lttng_fd_init();
3917
3918 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
3919 if (ret < 0) {
3920 goto exit;
3921 }
3922
3923 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
3924 if (ret < 0) {
3925 goto exit;
3926 }
3927
3928 if ((ret = set_signal_handler()) < 0) {
3929 goto exit;
3930 }
3931
3932 /* Setup the needed unix socket */
3933 if ((ret = init_daemon_socket()) < 0) {
3934 goto exit;
3935 }
3936
3937 /* Set credentials to socket */
3938 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
3939 goto exit;
3940 }
3941
3942 /* Get parent pid if -S, --sig-parent is specified. */
3943 if (opt_sig_parent) {
3944 ppid = getppid();
3945 }
3946
3947 /* Setup the kernel pipe for waking up the kernel thread */
3948 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
3949 goto exit;
3950 }
3951
3952 /* Setup the thread apps communication pipe. */
3953 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
3954 goto exit;
3955 }
3956
3957 /* Init UST command queue. */
3958 cds_wfq_init(&ust_cmd_queue.queue);
3959
3960 /*
3961 * Get session list pointer. This pointer MUST NOT be free(). This list is
3962 * statically declared in session.c
3963 */
3964 session_list_ptr = session_get_list();
3965
3966 /* Set up max poll set size */
3967 lttng_poll_set_max_size();
3968
3969 cmd_init();
3970
3971 /* Init all health thread counters. */
3972 health_init(&health_thread_cmd);
3973 health_init(&health_thread_kernel);
3974 health_init(&health_thread_app_manage);
3975 health_init(&health_thread_app_reg);
3976
3977 /*
3978 * Init health counters of the consumer thread. We do a quick hack here to
3979 * the state of the consumer health is fine even if the thread is not
3980 * started. This is simply to ease our life and has no cost what so ever.
3981 */
3982 health_init(&kconsumer_data.health);
3983 health_poll_update(&kconsumer_data.health);
3984 health_init(&ustconsumer32_data.health);
3985 health_poll_update(&ustconsumer32_data.health);
3986 health_init(&ustconsumer64_data.health);
3987 health_poll_update(&ustconsumer64_data.health);
3988
3989 /* Create thread to manage the client socket */
3990 ret = pthread_create(&health_thread, NULL,
3991 thread_manage_health, (void *) NULL);
3992 if (ret != 0) {
3993 PERROR("pthread_create health");
3994 goto exit_health;
3995 }
3996
3997 /* Create thread to manage the client socket */
3998 ret = pthread_create(&client_thread, NULL,
3999 thread_manage_clients, (void *) NULL);
4000 if (ret != 0) {
4001 PERROR("pthread_create clients");
4002 goto exit_client;
4003 }
4004
4005 /* Create thread to dispatch registration */
4006 ret = pthread_create(&dispatch_thread, NULL,
4007 thread_dispatch_ust_registration, (void *) NULL);
4008 if (ret != 0) {
4009 PERROR("pthread_create dispatch");
4010 goto exit_dispatch;
4011 }
4012
4013 /* Create thread to manage application registration. */
4014 ret = pthread_create(&reg_apps_thread, NULL,
4015 thread_registration_apps, (void *) NULL);
4016 if (ret != 0) {
4017 PERROR("pthread_create registration");
4018 goto exit_reg_apps;
4019 }
4020
4021 /* Create thread to manage application socket */
4022 ret = pthread_create(&apps_thread, NULL,
4023 thread_manage_apps, (void *) NULL);
4024 if (ret != 0) {
4025 PERROR("pthread_create apps");
4026 goto exit_apps;
4027 }
4028
4029 /* Create kernel thread to manage kernel event */
4030 ret = pthread_create(&kernel_thread, NULL,
4031 thread_manage_kernel, (void *) NULL);
4032 if (ret != 0) {
4033 PERROR("pthread_create kernel");
4034 goto exit_kernel;
4035 }
4036
4037 ret = pthread_join(kernel_thread, &status);
4038 if (ret != 0) {
4039 PERROR("pthread_join");
4040 goto error; /* join error, exit without cleanup */
4041 }
4042
4043 exit_kernel:
4044 ret = pthread_join(apps_thread, &status);
4045 if (ret != 0) {
4046 PERROR("pthread_join");
4047 goto error; /* join error, exit without cleanup */
4048 }
4049
4050 exit_apps:
4051 ret = pthread_join(reg_apps_thread, &status);
4052 if (ret != 0) {
4053 PERROR("pthread_join");
4054 goto error; /* join error, exit without cleanup */
4055 }
4056
4057 exit_reg_apps:
4058 ret = pthread_join(dispatch_thread, &status);
4059 if (ret != 0) {
4060 PERROR("pthread_join");
4061 goto error; /* join error, exit without cleanup */
4062 }
4063
4064 exit_dispatch:
4065 ret = pthread_join(client_thread, &status);
4066 if (ret != 0) {
4067 PERROR("pthread_join");
4068 goto error; /* join error, exit without cleanup */
4069 }
4070
4071 ret = join_consumer_thread(&kconsumer_data);
4072 if (ret != 0) {
4073 PERROR("join_consumer");
4074 goto error; /* join error, exit without cleanup */
4075 }
4076
4077 ret = join_consumer_thread(&ustconsumer32_data);
4078 if (ret != 0) {
4079 PERROR("join_consumer ust32");
4080 goto error; /* join error, exit without cleanup */
4081 }
4082
4083 ret = join_consumer_thread(&ustconsumer64_data);
4084 if (ret != 0) {
4085 PERROR("join_consumer ust64");
4086 goto error; /* join error, exit without cleanup */
4087 }
4088
4089 exit_client:
4090 ret = pthread_join(health_thread, &status);
4091 if (ret != 0) {
4092 PERROR("pthread_join health thread");
4093 goto error; /* join error, exit without cleanup */
4094 }
4095
4096 exit_health:
4097 exit:
4098 /*
4099 * cleanup() is called when no other thread is running.
4100 */
4101 rcu_thread_online();
4102 cleanup();
4103 rcu_thread_offline();
4104 rcu_unregister_thread();
4105 if (!ret) {
4106 exit(EXIT_SUCCESS);
4107 }
4108 error:
4109 exit(EXIT_FAILURE);
4110 }
This page took 0.192501 seconds and 4 git commands to generate.