Lib lttng-ctl data available command support
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <sys/mman.h>
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
32 #include <sys/stat.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <urcu/uatomic.h>
36 #include <unistd.h>
37 #include <config.h>
38
39 #include <common/common.h>
40 #include <common/compat/poll.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
47
48 #include "lttng-sessiond.h"
49 #include "channel.h"
50 #include "cmd.h"
51 #include "consumer.h"
52 #include "context.h"
53 #include "event.h"
54 #include "kernel.h"
55 #include "kernel-consumer.h"
56 #include "modprobe.h"
57 #include "shm.h"
58 #include "ust-ctl.h"
59 #include "ust-consumer.h"
60 #include "utils.h"
61 #include "fd-limit.h"
62 #include "filter.h"
63 #include "health.h"
64 #include "testpoint.h"
65
66 #define CONSUMERD_FILE "lttng-consumerd"
67
68 /* Const values */
69 const char default_home_dir[] = DEFAULT_HOME_DIR;
70 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
71 const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
72 const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
73
74 const char *progname;
75 const char *opt_tracing_group;
76 static int opt_sig_parent;
77 static int opt_verbose_consumer;
78 static int opt_daemon;
79 static int opt_no_kernel;
80 static int is_root; /* Set to 1 if the daemon is running as root */
81 static pid_t ppid; /* Parent PID for --sig-parent option */
82 static char *rundir;
83
84 /*
85 * Consumer daemon specific control data. Every value not initialized here is
86 * set to 0 by the static definition.
87 */
88 static struct consumer_data kconsumer_data = {
89 .type = LTTNG_CONSUMER_KERNEL,
90 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
91 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
92 .err_sock = -1,
93 .cmd_sock = -1,
94 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
95 .lock = PTHREAD_MUTEX_INITIALIZER,
96 .cond = PTHREAD_COND_INITIALIZER,
97 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
98 };
99 static struct consumer_data ustconsumer64_data = {
100 .type = LTTNG_CONSUMER64_UST,
101 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
109 };
110 static struct consumer_data ustconsumer32_data = {
111 .type = LTTNG_CONSUMER32_UST,
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
114 .err_sock = -1,
115 .cmd_sock = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121
122 /* Shared between threads */
123 static int dispatch_thread_exit;
124
125 /* Global application Unix socket path */
126 static char apps_unix_sock_path[PATH_MAX];
127 /* Global client Unix socket path */
128 static char client_unix_sock_path[PATH_MAX];
129 /* global wait shm path for UST */
130 static char wait_shm_path[PATH_MAX];
131 /* Global health check unix path */
132 static char health_unix_sock_path[PATH_MAX];
133
134 /* Sockets and FDs */
135 static int client_sock = -1;
136 static int apps_sock = -1;
137 int kernel_tracer_fd = -1;
138 static int kernel_poll_pipe[2] = { -1, -1 };
139
140 /*
141 * Quit pipe for all threads. This permits a single cancellation point
142 * for all threads when receiving an event on the pipe.
143 */
144 static int thread_quit_pipe[2] = { -1, -1 };
145
146 /*
147 * This pipe is used to inform the thread managing application communication
148 * that a command is queued and ready to be processed.
149 */
150 static int apps_cmd_pipe[2] = { -1, -1 };
151
152 /* Pthread, Mutexes and Semaphores */
153 static pthread_t apps_thread;
154 static pthread_t reg_apps_thread;
155 static pthread_t client_thread;
156 static pthread_t kernel_thread;
157 static pthread_t dispatch_thread;
158 static pthread_t health_thread;
159
160 /*
161 * UST registration command queue. This queue is tied with a futex and uses a N
162 * wakers / 1 waiter implemented and detailed in futex.c/.h
163 *
164 * The thread_manage_apps and thread_dispatch_ust_registration interact with
165 * this queue and the wait/wake scheme.
166 */
167 static struct ust_cmd_queue ust_cmd_queue;
168
169 /*
170 * Pointer initialized before thread creation.
171 *
172 * This points to the tracing session list containing the session count and a
173 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
174 * MUST NOT be taken if you call a public function in session.c.
175 *
176 * The lock is nested inside the structure: session_list_ptr->lock. Please use
177 * session_lock_list and session_unlock_list for lock acquisition.
178 */
179 static struct ltt_session_list *session_list_ptr;
180
181 int ust_consumerd64_fd = -1;
182 int ust_consumerd32_fd = -1;
183
184 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
185 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
186 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
187 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
188
189 static const char *module_proc_lttng = "/proc/lttng";
190
191 /*
192 * Consumer daemon state which is changed when spawning it, killing it or in
193 * case of a fatal error.
194 */
195 enum consumerd_state {
196 CONSUMER_STARTED = 1,
197 CONSUMER_STOPPED = 2,
198 CONSUMER_ERROR = 3,
199 };
200
201 /*
202 * This consumer daemon state is used to validate if a client command will be
203 * able to reach the consumer. If not, the client is informed. For instance,
204 * doing a "lttng start" when the consumer state is set to ERROR will return an
205 * error to the client.
206 *
207 * The following example shows a possible race condition of this scheme:
208 *
209 * consumer thread error happens
210 * client cmd arrives
211 * client cmd checks state -> still OK
212 * consumer thread exit, sets error
213 * client cmd try to talk to consumer
214 * ...
215 *
216 * However, since the consumer is a different daemon, we have no way of making
217 * sure the command will reach it safely even with this state flag. This is why
218 * we consider that up to the state validation during command processing, the
219 * command is safe. After that, we can not guarantee the correctness of the
220 * client request vis-a-vis the consumer.
221 */
222 static enum consumerd_state ust_consumerd_state;
223 static enum consumerd_state kernel_consumerd_state;
224
225 /* Used for the health monitoring of the session daemon. See health.h */
226 struct health_state health_thread_cmd;
227 struct health_state health_thread_app_manage;
228 struct health_state health_thread_app_reg;
229 struct health_state health_thread_kernel;
230
231 static
232 void setup_consumerd_path(void)
233 {
234 const char *bin, *libdir;
235
236 /*
237 * Allow INSTALL_BIN_PATH to be used as a target path for the
238 * native architecture size consumer if CONFIG_CONSUMER*_PATH
239 * has not been defined.
240 */
241 #if (CAA_BITS_PER_LONG == 32)
242 if (!consumerd32_bin[0]) {
243 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
244 }
245 if (!consumerd32_libdir[0]) {
246 consumerd32_libdir = INSTALL_LIB_PATH;
247 }
248 #elif (CAA_BITS_PER_LONG == 64)
249 if (!consumerd64_bin[0]) {
250 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
251 }
252 if (!consumerd64_libdir[0]) {
253 consumerd64_libdir = INSTALL_LIB_PATH;
254 }
255 #else
256 #error "Unknown bitness"
257 #endif
258
259 /*
260 * runtime env. var. overrides the build default.
261 */
262 bin = getenv("LTTNG_CONSUMERD32_BIN");
263 if (bin) {
264 consumerd32_bin = bin;
265 }
266 bin = getenv("LTTNG_CONSUMERD64_BIN");
267 if (bin) {
268 consumerd64_bin = bin;
269 }
270 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
271 if (libdir) {
272 consumerd32_libdir = libdir;
273 }
274 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
275 if (libdir) {
276 consumerd64_libdir = libdir;
277 }
278 }
279
280 /*
281 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
282 */
283 static int create_thread_poll_set(struct lttng_poll_event *events,
284 unsigned int size)
285 {
286 int ret;
287
288 if (events == NULL || size == 0) {
289 ret = -1;
290 goto error;
291 }
292
293 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
294 if (ret < 0) {
295 goto error;
296 }
297
298 /* Add quit pipe */
299 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
300 if (ret < 0) {
301 goto error;
302 }
303
304 return 0;
305
306 error:
307 return ret;
308 }
309
310 /*
311 * Check if the thread quit pipe was triggered.
312 *
313 * Return 1 if it was triggered else 0;
314 */
315 static int check_thread_quit_pipe(int fd, uint32_t events)
316 {
317 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
318 return 1;
319 }
320
321 return 0;
322 }
323
324 /*
325 * Return group ID of the tracing group or -1 if not found.
326 */
327 static gid_t allowed_group(void)
328 {
329 struct group *grp;
330
331 if (opt_tracing_group) {
332 grp = getgrnam(opt_tracing_group);
333 } else {
334 grp = getgrnam(default_tracing_group);
335 }
336 if (!grp) {
337 return -1;
338 } else {
339 return grp->gr_gid;
340 }
341 }
342
343 /*
344 * Init thread quit pipe.
345 *
346 * Return -1 on error or 0 if all pipes are created.
347 */
348 static int init_thread_quit_pipe(void)
349 {
350 int ret, i;
351
352 ret = pipe(thread_quit_pipe);
353 if (ret < 0) {
354 PERROR("thread quit pipe");
355 goto error;
356 }
357
358 for (i = 0; i < 2; i++) {
359 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
360 if (ret < 0) {
361 PERROR("fcntl");
362 goto error;
363 }
364 }
365
366 error:
367 return ret;
368 }
369
370 /*
371 * Stop all threads by closing the thread quit pipe.
372 */
373 static void stop_threads(void)
374 {
375 int ret;
376
377 /* Stopping all threads */
378 DBG("Terminating all threads");
379 ret = notify_thread_pipe(thread_quit_pipe[1]);
380 if (ret < 0) {
381 ERR("write error on thread quit pipe");
382 }
383
384 /* Dispatch thread */
385 CMM_STORE_SHARED(dispatch_thread_exit, 1);
386 futex_nto1_wake(&ust_cmd_queue.futex);
387 }
388
389 /*
390 * Cleanup the daemon
391 */
392 static void cleanup(void)
393 {
394 int ret;
395 char *cmd;
396 struct ltt_session *sess, *stmp;
397
398 DBG("Cleaning up");
399
400 /* First thing first, stop all threads */
401 utils_close_pipe(thread_quit_pipe);
402
403 DBG("Removing %s directory", rundir);
404 ret = asprintf(&cmd, "rm -rf %s", rundir);
405 if (ret < 0) {
406 ERR("asprintf failed. Something is really wrong!");
407 }
408
409 /* Remove lttng run directory */
410 ret = system(cmd);
411 if (ret < 0) {
412 ERR("Unable to clean %s", rundir);
413 }
414 free(cmd);
415
416 DBG("Cleaning up all sessions");
417
418 /* Destroy session list mutex */
419 if (session_list_ptr != NULL) {
420 pthread_mutex_destroy(&session_list_ptr->lock);
421
422 /* Cleanup ALL session */
423 cds_list_for_each_entry_safe(sess, stmp,
424 &session_list_ptr->head, list) {
425 cmd_destroy_session(sess, kernel_poll_pipe[1]);
426 }
427 }
428
429 DBG("Closing all UST sockets");
430 ust_app_clean_list();
431
432 if (is_root && !opt_no_kernel) {
433 DBG2("Closing kernel fd");
434 if (kernel_tracer_fd >= 0) {
435 ret = close(kernel_tracer_fd);
436 if (ret) {
437 PERROR("close");
438 }
439 }
440 DBG("Unloading kernel modules");
441 modprobe_remove_lttng_all();
442 }
443
444 utils_close_pipe(kernel_poll_pipe);
445 utils_close_pipe(apps_cmd_pipe);
446
447 /* <fun> */
448 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
449 "Matthew, BEET driven development works!%c[%dm",
450 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
451 /* </fun> */
452 }
453
454 /*
455 * Send data on a unix socket using the liblttsessiondcomm API.
456 *
457 * Return lttcomm error code.
458 */
459 static int send_unix_sock(int sock, void *buf, size_t len)
460 {
461 /* Check valid length */
462 if (len <= 0) {
463 return -1;
464 }
465
466 return lttcomm_send_unix_sock(sock, buf, len);
467 }
468
469 /*
470 * Free memory of a command context structure.
471 */
472 static void clean_command_ctx(struct command_ctx **cmd_ctx)
473 {
474 DBG("Clean command context structure");
475 if (*cmd_ctx) {
476 if ((*cmd_ctx)->llm) {
477 free((*cmd_ctx)->llm);
478 }
479 if ((*cmd_ctx)->lsm) {
480 free((*cmd_ctx)->lsm);
481 }
482 free(*cmd_ctx);
483 *cmd_ctx = NULL;
484 }
485 }
486
487 /*
488 * Notify UST applications using the shm mmap futex.
489 */
490 static int notify_ust_apps(int active)
491 {
492 char *wait_shm_mmap;
493
494 DBG("Notifying applications of session daemon state: %d", active);
495
496 /* See shm.c for this call implying mmap, shm and futex calls */
497 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
498 if (wait_shm_mmap == NULL) {
499 goto error;
500 }
501
502 /* Wake waiting process */
503 futex_wait_update((int32_t *) wait_shm_mmap, active);
504
505 /* Apps notified successfully */
506 return 0;
507
508 error:
509 return -1;
510 }
511
512 /*
513 * Setup the outgoing data buffer for the response (llm) by allocating the
514 * right amount of memory and copying the original information from the lsm
515 * structure.
516 *
517 * Return total size of the buffer pointed by buf.
518 */
519 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
520 {
521 int ret, buf_size;
522
523 buf_size = size;
524
525 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
526 if (cmd_ctx->llm == NULL) {
527 PERROR("zmalloc");
528 ret = -ENOMEM;
529 goto error;
530 }
531
532 /* Copy common data */
533 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
534 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
535
536 cmd_ctx->llm->data_size = size;
537 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
538
539 return buf_size;
540
541 error:
542 return ret;
543 }
544
545 /*
546 * Update the kernel poll set of all channel fd available over all tracing
547 * session. Add the wakeup pipe at the end of the set.
548 */
549 static int update_kernel_poll(struct lttng_poll_event *events)
550 {
551 int ret;
552 struct ltt_session *session;
553 struct ltt_kernel_channel *channel;
554
555 DBG("Updating kernel poll set");
556
557 session_lock_list();
558 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
559 session_lock(session);
560 if (session->kernel_session == NULL) {
561 session_unlock(session);
562 continue;
563 }
564
565 cds_list_for_each_entry(channel,
566 &session->kernel_session->channel_list.head, list) {
567 /* Add channel fd to the kernel poll set */
568 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
569 if (ret < 0) {
570 session_unlock(session);
571 goto error;
572 }
573 DBG("Channel fd %d added to kernel set", channel->fd);
574 }
575 session_unlock(session);
576 }
577 session_unlock_list();
578
579 return 0;
580
581 error:
582 session_unlock_list();
583 return -1;
584 }
585
586 /*
587 * Find the channel fd from 'fd' over all tracing session. When found, check
588 * for new channel stream and send those stream fds to the kernel consumer.
589 *
590 * Useful for CPU hotplug feature.
591 */
592 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
593 {
594 int ret = 0;
595 struct ltt_session *session;
596 struct ltt_kernel_session *ksess;
597 struct ltt_kernel_channel *channel;
598
599 DBG("Updating kernel streams for channel fd %d", fd);
600
601 session_lock_list();
602 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
603 session_lock(session);
604 if (session->kernel_session == NULL) {
605 session_unlock(session);
606 continue;
607 }
608 ksess = session->kernel_session;
609
610 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
611 if (channel->fd == fd) {
612 DBG("Channel found, updating kernel streams");
613 ret = kernel_open_channel_stream(channel);
614 if (ret < 0) {
615 goto error;
616 }
617
618 /*
619 * Have we already sent fds to the consumer? If yes, it means
620 * that tracing is started so it is safe to send our updated
621 * stream fds.
622 */
623 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
624 struct lttng_ht_iter iter;
625 struct consumer_socket *socket;
626
627
628 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
629 &iter.iter, socket, node.node) {
630 /* Code flow error */
631 assert(socket->fd >= 0);
632
633 pthread_mutex_lock(socket->lock);
634 ret = kernel_consumer_send_channel_stream(socket->fd,
635 channel, ksess);
636 pthread_mutex_unlock(socket->lock);
637 if (ret < 0) {
638 goto error;
639 }
640 }
641 }
642 goto error;
643 }
644 }
645 session_unlock(session);
646 }
647 session_unlock_list();
648 return ret;
649
650 error:
651 session_unlock(session);
652 session_unlock_list();
653 return ret;
654 }
655
656 /*
657 * For each tracing session, update newly registered apps.
658 */
659 static void update_ust_app(int app_sock)
660 {
661 struct ltt_session *sess, *stmp;
662
663 session_lock_list();
664
665 /* For all tracing session(s) */
666 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
667 session_lock(sess);
668 if (sess->ust_session) {
669 ust_app_global_update(sess->ust_session, app_sock);
670 }
671 session_unlock(sess);
672 }
673
674 session_unlock_list();
675 }
676
677 /*
678 * This thread manage event coming from the kernel.
679 *
680 * Features supported in this thread:
681 * -) CPU Hotplug
682 */
683 static void *thread_manage_kernel(void *data)
684 {
685 int ret, i, pollfd, update_poll_flag = 1, err = -1;
686 uint32_t revents, nb_fd;
687 char tmp;
688 struct lttng_poll_event events;
689
690 DBG("Thread manage kernel started");
691
692 testpoint(thread_manage_kernel);
693
694 health_code_update(&health_thread_kernel);
695
696 testpoint(thread_manage_kernel_before_loop);
697
698 ret = create_thread_poll_set(&events, 2);
699 if (ret < 0) {
700 goto error_poll_create;
701 }
702
703 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
704 if (ret < 0) {
705 goto error;
706 }
707
708 while (1) {
709 health_code_update(&health_thread_kernel);
710
711 if (update_poll_flag == 1) {
712 /*
713 * Reset number of fd in the poll set. Always 2 since there is the thread
714 * quit pipe and the kernel pipe.
715 */
716 events.nb_fd = 2;
717
718 ret = update_kernel_poll(&events);
719 if (ret < 0) {
720 goto error;
721 }
722 update_poll_flag = 0;
723 }
724
725 nb_fd = LTTNG_POLL_GETNB(&events);
726
727 DBG("Thread kernel polling on %d fds", nb_fd);
728
729 /* Zeroed the poll events */
730 lttng_poll_reset(&events);
731
732 /* Poll infinite value of time */
733 restart:
734 health_poll_update(&health_thread_kernel);
735 ret = lttng_poll_wait(&events, -1);
736 health_poll_update(&health_thread_kernel);
737 if (ret < 0) {
738 /*
739 * Restart interrupted system call.
740 */
741 if (errno == EINTR) {
742 goto restart;
743 }
744 goto error;
745 } else if (ret == 0) {
746 /* Should not happen since timeout is infinite */
747 ERR("Return value of poll is 0 with an infinite timeout.\n"
748 "This should not have happened! Continuing...");
749 continue;
750 }
751
752 for (i = 0; i < nb_fd; i++) {
753 /* Fetch once the poll data */
754 revents = LTTNG_POLL_GETEV(&events, i);
755 pollfd = LTTNG_POLL_GETFD(&events, i);
756
757 health_code_update(&health_thread_kernel);
758
759 /* Thread quit pipe has been closed. Killing thread. */
760 ret = check_thread_quit_pipe(pollfd, revents);
761 if (ret) {
762 err = 0;
763 goto exit;
764 }
765
766 /* Check for data on kernel pipe */
767 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
768 ret = read(kernel_poll_pipe[0], &tmp, 1);
769 update_poll_flag = 1;
770 continue;
771 } else {
772 /*
773 * New CPU detected by the kernel. Adding kernel stream to
774 * kernel session and updating the kernel consumer
775 */
776 if (revents & LPOLLIN) {
777 ret = update_kernel_stream(&kconsumer_data, pollfd);
778 if (ret < 0) {
779 continue;
780 }
781 break;
782 /*
783 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
784 * and unregister kernel stream at this point.
785 */
786 }
787 }
788 }
789 }
790
791 exit:
792 error:
793 lttng_poll_clean(&events);
794 error_poll_create:
795 if (err) {
796 health_error(&health_thread_kernel);
797 ERR("Health error occurred in %s", __func__);
798 }
799 health_exit(&health_thread_kernel);
800 DBG("Kernel thread dying");
801 return NULL;
802 }
803
804 /*
805 * Signal pthread condition of the consumer data that the thread.
806 */
807 static void signal_consumer_condition(struct consumer_data *data, int state)
808 {
809 pthread_mutex_lock(&data->cond_mutex);
810
811 /*
812 * The state is set before signaling. It can be any value, it's the waiter
813 * job to correctly interpret this condition variable associated to the
814 * consumer pthread_cond.
815 *
816 * A value of 0 means that the corresponding thread of the consumer data
817 * was not started. 1 indicates that the thread has started and is ready
818 * for action. A negative value means that there was an error during the
819 * thread bootstrap.
820 */
821 data->consumer_thread_is_ready = state;
822 (void) pthread_cond_signal(&data->cond);
823
824 pthread_mutex_unlock(&data->cond_mutex);
825 }
826
827 /*
828 * This thread manage the consumer error sent back to the session daemon.
829 */
830 static void *thread_manage_consumer(void *data)
831 {
832 int sock = -1, i, ret, pollfd, err = -1;
833 uint32_t revents, nb_fd;
834 enum lttcomm_return_code code;
835 struct lttng_poll_event events;
836 struct consumer_data *consumer_data = data;
837
838 DBG("[thread] Manage consumer started");
839
840 health_code_update(&consumer_data->health);
841
842 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
843 if (ret < 0) {
844 goto error_listen;
845 }
846
847 /*
848 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
849 * Nothing more will be added to this poll set.
850 */
851 ret = create_thread_poll_set(&events, 2);
852 if (ret < 0) {
853 goto error_poll;
854 }
855
856 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
857 if (ret < 0) {
858 goto error;
859 }
860
861 nb_fd = LTTNG_POLL_GETNB(&events);
862
863 health_code_update(&consumer_data->health);
864
865 /* Inifinite blocking call, waiting for transmission */
866 restart:
867 health_poll_update(&consumer_data->health);
868
869 testpoint(thread_manage_consumer);
870
871 ret = lttng_poll_wait(&events, -1);
872 health_poll_update(&consumer_data->health);
873 if (ret < 0) {
874 /*
875 * Restart interrupted system call.
876 */
877 if (errno == EINTR) {
878 goto restart;
879 }
880 goto error;
881 }
882
883 for (i = 0; i < nb_fd; i++) {
884 /* Fetch once the poll data */
885 revents = LTTNG_POLL_GETEV(&events, i);
886 pollfd = LTTNG_POLL_GETFD(&events, i);
887
888 health_code_update(&consumer_data->health);
889
890 /* Thread quit pipe has been closed. Killing thread. */
891 ret = check_thread_quit_pipe(pollfd, revents);
892 if (ret) {
893 err = 0;
894 goto exit;
895 }
896
897 /* Event on the registration socket */
898 if (pollfd == consumer_data->err_sock) {
899 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
900 ERR("consumer err socket poll error");
901 goto error;
902 }
903 }
904 }
905
906 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
907 if (sock < 0) {
908 goto error;
909 }
910
911 health_code_update(&consumer_data->health);
912
913 DBG2("Receiving code from consumer err_sock");
914
915 /* Getting status code from kconsumerd */
916 ret = lttcomm_recv_unix_sock(sock, &code,
917 sizeof(enum lttcomm_return_code));
918 if (ret <= 0) {
919 goto error;
920 }
921
922 health_code_update(&consumer_data->health);
923
924 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
925 consumer_data->cmd_sock =
926 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
927 if (consumer_data->cmd_sock < 0) {
928 /* On error, signal condition and quit. */
929 signal_consumer_condition(consumer_data, -1);
930 PERROR("consumer connect");
931 goto error;
932 }
933 signal_consumer_condition(consumer_data, 1);
934 DBG("Consumer command socket ready");
935 } else {
936 ERR("consumer error when waiting for SOCK_READY : %s",
937 lttcomm_get_readable_code(-code));
938 goto error;
939 }
940
941 /* Remove the kconsumerd error sock since we've established a connexion */
942 ret = lttng_poll_del(&events, consumer_data->err_sock);
943 if (ret < 0) {
944 goto error;
945 }
946
947 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
948 if (ret < 0) {
949 goto error;
950 }
951
952 health_code_update(&consumer_data->health);
953
954 /* Update number of fd */
955 nb_fd = LTTNG_POLL_GETNB(&events);
956
957 /* Inifinite blocking call, waiting for transmission */
958 restart_poll:
959 health_poll_update(&consumer_data->health);
960 ret = lttng_poll_wait(&events, -1);
961 health_poll_update(&consumer_data->health);
962 if (ret < 0) {
963 /*
964 * Restart interrupted system call.
965 */
966 if (errno == EINTR) {
967 goto restart_poll;
968 }
969 goto error;
970 }
971
972 for (i = 0; i < nb_fd; i++) {
973 /* Fetch once the poll data */
974 revents = LTTNG_POLL_GETEV(&events, i);
975 pollfd = LTTNG_POLL_GETFD(&events, i);
976
977 health_code_update(&consumer_data->health);
978
979 /* Thread quit pipe has been closed. Killing thread. */
980 ret = check_thread_quit_pipe(pollfd, revents);
981 if (ret) {
982 err = 0;
983 goto exit;
984 }
985
986 /* Event on the kconsumerd socket */
987 if (pollfd == sock) {
988 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
989 ERR("consumer err socket second poll error");
990 goto error;
991 }
992 }
993 }
994
995 health_code_update(&consumer_data->health);
996
997 /* Wait for any kconsumerd error */
998 ret = lttcomm_recv_unix_sock(sock, &code,
999 sizeof(enum lttcomm_return_code));
1000 if (ret <= 0) {
1001 ERR("consumer closed the command socket");
1002 goto error;
1003 }
1004
1005 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1006
1007 exit:
1008 error:
1009 /* Immediately set the consumerd state to stopped */
1010 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1011 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1012 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1013 consumer_data->type == LTTNG_CONSUMER32_UST) {
1014 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1015 } else {
1016 /* Code flow error... */
1017 assert(0);
1018 }
1019
1020 if (consumer_data->err_sock >= 0) {
1021 ret = close(consumer_data->err_sock);
1022 if (ret) {
1023 PERROR("close");
1024 }
1025 }
1026 if (consumer_data->cmd_sock >= 0) {
1027 ret = close(consumer_data->cmd_sock);
1028 if (ret) {
1029 PERROR("close");
1030 }
1031 }
1032 if (sock >= 0) {
1033 ret = close(sock);
1034 if (ret) {
1035 PERROR("close");
1036 }
1037 }
1038
1039 unlink(consumer_data->err_unix_sock_path);
1040 unlink(consumer_data->cmd_unix_sock_path);
1041 consumer_data->pid = 0;
1042
1043 lttng_poll_clean(&events);
1044 error_poll:
1045 error_listen:
1046 if (err) {
1047 health_error(&consumer_data->health);
1048 ERR("Health error occurred in %s", __func__);
1049 }
1050 health_exit(&consumer_data->health);
1051 DBG("consumer thread cleanup completed");
1052
1053 return NULL;
1054 }
1055
1056 /*
1057 * This thread manage application communication.
1058 */
1059 static void *thread_manage_apps(void *data)
1060 {
1061 int i, ret, pollfd, err = -1;
1062 uint32_t revents, nb_fd;
1063 struct ust_command ust_cmd;
1064 struct lttng_poll_event events;
1065
1066 DBG("[thread] Manage application started");
1067
1068 testpoint(thread_manage_apps);
1069
1070 rcu_register_thread();
1071 rcu_thread_online();
1072
1073 health_code_update(&health_thread_app_manage);
1074
1075 ret = create_thread_poll_set(&events, 2);
1076 if (ret < 0) {
1077 goto error_poll_create;
1078 }
1079
1080 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1081 if (ret < 0) {
1082 goto error;
1083 }
1084
1085 testpoint(thread_manage_apps_before_loop);
1086
1087 health_code_update(&health_thread_app_manage);
1088
1089 while (1) {
1090 /* Zeroed the events structure */
1091 lttng_poll_reset(&events);
1092
1093 nb_fd = LTTNG_POLL_GETNB(&events);
1094
1095 DBG("Apps thread polling on %d fds", nb_fd);
1096
1097 /* Inifinite blocking call, waiting for transmission */
1098 restart:
1099 health_poll_update(&health_thread_app_manage);
1100 ret = lttng_poll_wait(&events, -1);
1101 health_poll_update(&health_thread_app_manage);
1102 if (ret < 0) {
1103 /*
1104 * Restart interrupted system call.
1105 */
1106 if (errno == EINTR) {
1107 goto restart;
1108 }
1109 goto error;
1110 }
1111
1112 for (i = 0; i < nb_fd; i++) {
1113 /* Fetch once the poll data */
1114 revents = LTTNG_POLL_GETEV(&events, i);
1115 pollfd = LTTNG_POLL_GETFD(&events, i);
1116
1117 health_code_update(&health_thread_app_manage);
1118
1119 /* Thread quit pipe has been closed. Killing thread. */
1120 ret = check_thread_quit_pipe(pollfd, revents);
1121 if (ret) {
1122 err = 0;
1123 goto exit;
1124 }
1125
1126 /* Inspect the apps cmd pipe */
1127 if (pollfd == apps_cmd_pipe[0]) {
1128 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1129 ERR("Apps command pipe error");
1130 goto error;
1131 } else if (revents & LPOLLIN) {
1132 /* Empty pipe */
1133 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1134 if (ret < 0 || ret < sizeof(ust_cmd)) {
1135 PERROR("read apps cmd pipe");
1136 goto error;
1137 }
1138
1139 health_code_update(&health_thread_app_manage);
1140
1141 /* Register applicaton to the session daemon */
1142 ret = ust_app_register(&ust_cmd.reg_msg,
1143 ust_cmd.sock);
1144 if (ret == -ENOMEM) {
1145 goto error;
1146 } else if (ret < 0) {
1147 break;
1148 }
1149
1150 health_code_update(&health_thread_app_manage);
1151
1152 /*
1153 * Validate UST version compatibility.
1154 */
1155 ret = ust_app_validate_version(ust_cmd.sock);
1156 if (ret >= 0) {
1157 /*
1158 * Add channel(s) and event(s) to newly registered apps
1159 * from lttng global UST domain.
1160 */
1161 update_ust_app(ust_cmd.sock);
1162 }
1163
1164 health_code_update(&health_thread_app_manage);
1165
1166 ret = ust_app_register_done(ust_cmd.sock);
1167 if (ret < 0) {
1168 /*
1169 * If the registration is not possible, we simply
1170 * unregister the apps and continue
1171 */
1172 ust_app_unregister(ust_cmd.sock);
1173 } else {
1174 /*
1175 * We just need here to monitor the close of the UST
1176 * socket and poll set monitor those by default.
1177 * Listen on POLLIN (even if we never expect any
1178 * data) to ensure that hangup wakes us.
1179 */
1180 ret = lttng_poll_add(&events, ust_cmd.sock, LPOLLIN);
1181 if (ret < 0) {
1182 goto error;
1183 }
1184
1185 DBG("Apps with sock %d added to poll set",
1186 ust_cmd.sock);
1187 }
1188
1189 health_code_update(&health_thread_app_manage);
1190
1191 break;
1192 }
1193 } else {
1194 /*
1195 * At this point, we know that a registered application made
1196 * the event at poll_wait.
1197 */
1198 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1199 /* Removing from the poll set */
1200 ret = lttng_poll_del(&events, pollfd);
1201 if (ret < 0) {
1202 goto error;
1203 }
1204
1205 /* Socket closed on remote end. */
1206 ust_app_unregister(pollfd);
1207 break;
1208 }
1209 }
1210
1211 health_code_update(&health_thread_app_manage);
1212 }
1213 }
1214
1215 exit:
1216 error:
1217 lttng_poll_clean(&events);
1218 error_poll_create:
1219 if (err) {
1220 health_error(&health_thread_app_manage);
1221 ERR("Health error occurred in %s", __func__);
1222 }
1223 health_exit(&health_thread_app_manage);
1224 DBG("Application communication apps thread cleanup complete");
1225 rcu_thread_offline();
1226 rcu_unregister_thread();
1227 return NULL;
1228 }
1229
1230 /*
1231 * Dispatch request from the registration threads to the application
1232 * communication thread.
1233 */
1234 static void *thread_dispatch_ust_registration(void *data)
1235 {
1236 int ret;
1237 struct cds_wfq_node *node;
1238 struct ust_command *ust_cmd = NULL;
1239
1240 DBG("[thread] Dispatch UST command started");
1241
1242 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1243 /* Atomically prepare the queue futex */
1244 futex_nto1_prepare(&ust_cmd_queue.futex);
1245
1246 do {
1247 /* Dequeue command for registration */
1248 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1249 if (node == NULL) {
1250 DBG("Woken up but nothing in the UST command queue");
1251 /* Continue thread execution */
1252 break;
1253 }
1254
1255 ust_cmd = caa_container_of(node, struct ust_command, node);
1256
1257 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1258 " gid:%d sock:%d name:%s (version %d.%d)",
1259 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1260 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1261 ust_cmd->sock, ust_cmd->reg_msg.name,
1262 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1263 /*
1264 * Inform apps thread of the new application registration. This
1265 * call is blocking so we can be assured that the data will be read
1266 * at some point in time or wait to the end of the world :)
1267 */
1268 ret = write(apps_cmd_pipe[1], ust_cmd,
1269 sizeof(struct ust_command));
1270 if (ret < 0) {
1271 PERROR("write apps cmd pipe");
1272 if (errno == EBADF) {
1273 /*
1274 * We can't inform the application thread to process
1275 * registration. We will exit or else application
1276 * registration will not occur and tracing will never
1277 * start.
1278 */
1279 goto error;
1280 }
1281 }
1282 free(ust_cmd);
1283 } while (node != NULL);
1284
1285 /* Futex wait on queue. Blocking call on futex() */
1286 futex_nto1_wait(&ust_cmd_queue.futex);
1287 }
1288
1289 error:
1290 DBG("Dispatch thread dying");
1291 return NULL;
1292 }
1293
1294 /*
1295 * This thread manage application registration.
1296 */
1297 static void *thread_registration_apps(void *data)
1298 {
1299 int sock = -1, i, ret, pollfd, err = -1;
1300 uint32_t revents, nb_fd;
1301 struct lttng_poll_event events;
1302 /*
1303 * Get allocated in this thread, enqueued to a global queue, dequeued and
1304 * freed in the manage apps thread.
1305 */
1306 struct ust_command *ust_cmd = NULL;
1307
1308 DBG("[thread] Manage application registration started");
1309
1310 testpoint(thread_registration_apps);
1311
1312 ret = lttcomm_listen_unix_sock(apps_sock);
1313 if (ret < 0) {
1314 goto error_listen;
1315 }
1316
1317 /*
1318 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1319 * more will be added to this poll set.
1320 */
1321 ret = create_thread_poll_set(&events, 2);
1322 if (ret < 0) {
1323 goto error_create_poll;
1324 }
1325
1326 /* Add the application registration socket */
1327 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1328 if (ret < 0) {
1329 goto error_poll_add;
1330 }
1331
1332 /* Notify all applications to register */
1333 ret = notify_ust_apps(1);
1334 if (ret < 0) {
1335 ERR("Failed to notify applications or create the wait shared memory.\n"
1336 "Execution continues but there might be problem for already\n"
1337 "running applications that wishes to register.");
1338 }
1339
1340 while (1) {
1341 DBG("Accepting application registration");
1342
1343 nb_fd = LTTNG_POLL_GETNB(&events);
1344
1345 /* Inifinite blocking call, waiting for transmission */
1346 restart:
1347 health_poll_update(&health_thread_app_reg);
1348 ret = lttng_poll_wait(&events, -1);
1349 health_poll_update(&health_thread_app_reg);
1350 if (ret < 0) {
1351 /*
1352 * Restart interrupted system call.
1353 */
1354 if (errno == EINTR) {
1355 goto restart;
1356 }
1357 goto error;
1358 }
1359
1360 for (i = 0; i < nb_fd; i++) {
1361 health_code_update(&health_thread_app_reg);
1362
1363 /* Fetch once the poll data */
1364 revents = LTTNG_POLL_GETEV(&events, i);
1365 pollfd = LTTNG_POLL_GETFD(&events, i);
1366
1367 /* Thread quit pipe has been closed. Killing thread. */
1368 ret = check_thread_quit_pipe(pollfd, revents);
1369 if (ret) {
1370 err = 0;
1371 goto exit;
1372 }
1373
1374 /* Event on the registration socket */
1375 if (pollfd == apps_sock) {
1376 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1377 ERR("Register apps socket poll error");
1378 goto error;
1379 } else if (revents & LPOLLIN) {
1380 sock = lttcomm_accept_unix_sock(apps_sock);
1381 if (sock < 0) {
1382 goto error;
1383 }
1384
1385 /* Create UST registration command for enqueuing */
1386 ust_cmd = zmalloc(sizeof(struct ust_command));
1387 if (ust_cmd == NULL) {
1388 PERROR("ust command zmalloc");
1389 goto error;
1390 }
1391
1392 /*
1393 * Using message-based transmissions to ensure we don't
1394 * have to deal with partially received messages.
1395 */
1396 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1397 if (ret < 0) {
1398 ERR("Exhausted file descriptors allowed for applications.");
1399 free(ust_cmd);
1400 ret = close(sock);
1401 if (ret) {
1402 PERROR("close");
1403 }
1404 sock = -1;
1405 continue;
1406 }
1407 health_code_update(&health_thread_app_reg);
1408 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1409 sizeof(struct ust_register_msg));
1410 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1411 if (ret < 0) {
1412 PERROR("lttcomm_recv_unix_sock register apps");
1413 } else {
1414 ERR("Wrong size received on apps register");
1415 }
1416 free(ust_cmd);
1417 ret = close(sock);
1418 if (ret) {
1419 PERROR("close");
1420 }
1421 lttng_fd_put(LTTNG_FD_APPS, 1);
1422 sock = -1;
1423 continue;
1424 }
1425 health_code_update(&health_thread_app_reg);
1426
1427 ust_cmd->sock = sock;
1428 sock = -1;
1429
1430 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1431 " gid:%d sock:%d name:%s (version %d.%d)",
1432 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1433 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1434 ust_cmd->sock, ust_cmd->reg_msg.name,
1435 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1436
1437 /*
1438 * Lock free enqueue the registration request. The red pill
1439 * has been taken! This apps will be part of the *system*.
1440 */
1441 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1442
1443 /*
1444 * Wake the registration queue futex. Implicit memory
1445 * barrier with the exchange in cds_wfq_enqueue.
1446 */
1447 futex_nto1_wake(&ust_cmd_queue.futex);
1448 }
1449 }
1450 }
1451 }
1452
1453 exit:
1454 error:
1455 if (err) {
1456 health_error(&health_thread_app_reg);
1457 ERR("Health error occurred in %s", __func__);
1458 }
1459 health_exit(&health_thread_app_reg);
1460
1461 /* Notify that the registration thread is gone */
1462 notify_ust_apps(0);
1463
1464 if (apps_sock >= 0) {
1465 ret = close(apps_sock);
1466 if (ret) {
1467 PERROR("close");
1468 }
1469 }
1470 if (sock >= 0) {
1471 ret = close(sock);
1472 if (ret) {
1473 PERROR("close");
1474 }
1475 lttng_fd_put(LTTNG_FD_APPS, 1);
1476 }
1477 unlink(apps_unix_sock_path);
1478
1479 error_poll_add:
1480 lttng_poll_clean(&events);
1481 error_listen:
1482 error_create_poll:
1483 DBG("UST Registration thread cleanup complete");
1484
1485 return NULL;
1486 }
1487
1488 /*
1489 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1490 * exec or it will fails.
1491 */
1492 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1493 {
1494 int ret, clock_ret;
1495 struct timespec timeout;
1496
1497 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1498 consumer_data->consumer_thread_is_ready = 0;
1499
1500 /* Setup pthread condition */
1501 ret = pthread_condattr_init(&consumer_data->condattr);
1502 if (ret != 0) {
1503 errno = ret;
1504 PERROR("pthread_condattr_init consumer data");
1505 goto error;
1506 }
1507
1508 /*
1509 * Set the monotonic clock in order to make sure we DO NOT jump in time
1510 * between the clock_gettime() call and the timedwait call. See bug #324
1511 * for a more details and how we noticed it.
1512 */
1513 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1514 if (ret != 0) {
1515 errno = ret;
1516 PERROR("pthread_condattr_setclock consumer data");
1517 goto error;
1518 }
1519
1520 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1521 if (ret != 0) {
1522 errno = ret;
1523 PERROR("pthread_cond_init consumer data");
1524 goto error;
1525 }
1526
1527 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1528 consumer_data);
1529 if (ret != 0) {
1530 PERROR("pthread_create consumer");
1531 ret = -1;
1532 goto error;
1533 }
1534
1535 /* We are about to wait on a pthread condition */
1536 pthread_mutex_lock(&consumer_data->cond_mutex);
1537
1538 /* Get time for sem_timedwait absolute timeout */
1539 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1540 /*
1541 * Set the timeout for the condition timed wait even if the clock gettime
1542 * call fails since we might loop on that call and we want to avoid to
1543 * increment the timeout too many times.
1544 */
1545 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1546
1547 /*
1548 * The following loop COULD be skipped in some conditions so this is why we
1549 * set ret to 0 in order to make sure at least one round of the loop is
1550 * done.
1551 */
1552 ret = 0;
1553
1554 /*
1555 * Loop until the condition is reached or when a timeout is reached. Note
1556 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1557 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1558 * possible. This loop does not take any chances and works with both of
1559 * them.
1560 */
1561 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1562 if (clock_ret < 0) {
1563 PERROR("clock_gettime spawn consumer");
1564 /* Infinite wait for the consumerd thread to be ready */
1565 ret = pthread_cond_wait(&consumer_data->cond,
1566 &consumer_data->cond_mutex);
1567 } else {
1568 ret = pthread_cond_timedwait(&consumer_data->cond,
1569 &consumer_data->cond_mutex, &timeout);
1570 }
1571 }
1572
1573 /* Release the pthread condition */
1574 pthread_mutex_unlock(&consumer_data->cond_mutex);
1575
1576 if (ret != 0) {
1577 errno = ret;
1578 if (ret == ETIMEDOUT) {
1579 /*
1580 * Call has timed out so we kill the kconsumerd_thread and return
1581 * an error.
1582 */
1583 ERR("Condition timed out. The consumer thread was never ready."
1584 " Killing it");
1585 ret = pthread_cancel(consumer_data->thread);
1586 if (ret < 0) {
1587 PERROR("pthread_cancel consumer thread");
1588 }
1589 } else {
1590 PERROR("pthread_cond_wait failed consumer thread");
1591 }
1592 goto error;
1593 }
1594
1595 pthread_mutex_lock(&consumer_data->pid_mutex);
1596 if (consumer_data->pid == 0) {
1597 ERR("Consumerd did not start");
1598 pthread_mutex_unlock(&consumer_data->pid_mutex);
1599 goto error;
1600 }
1601 pthread_mutex_unlock(&consumer_data->pid_mutex);
1602
1603 return 0;
1604
1605 error:
1606 return ret;
1607 }
1608
1609 /*
1610 * Join consumer thread
1611 */
1612 static int join_consumer_thread(struct consumer_data *consumer_data)
1613 {
1614 void *status;
1615 int ret;
1616
1617 /* Consumer pid must be a real one. */
1618 if (consumer_data->pid > 0) {
1619 ret = kill(consumer_data->pid, SIGTERM);
1620 if (ret) {
1621 ERR("Error killing consumer daemon");
1622 return ret;
1623 }
1624 return pthread_join(consumer_data->thread, &status);
1625 } else {
1626 return 0;
1627 }
1628 }
1629
1630 /*
1631 * Fork and exec a consumer daemon (consumerd).
1632 *
1633 * Return pid if successful else -1.
1634 */
1635 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1636 {
1637 int ret;
1638 pid_t pid;
1639 const char *consumer_to_use;
1640 const char *verbosity;
1641 struct stat st;
1642
1643 DBG("Spawning consumerd");
1644
1645 pid = fork();
1646 if (pid == 0) {
1647 /*
1648 * Exec consumerd.
1649 */
1650 if (opt_verbose_consumer) {
1651 verbosity = "--verbose";
1652 } else {
1653 verbosity = "--quiet";
1654 }
1655 switch (consumer_data->type) {
1656 case LTTNG_CONSUMER_KERNEL:
1657 /*
1658 * Find out which consumerd to execute. We will first try the
1659 * 64-bit path, then the sessiond's installation directory, and
1660 * fallback on the 32-bit one,
1661 */
1662 DBG3("Looking for a kernel consumer at these locations:");
1663 DBG3(" 1) %s", consumerd64_bin);
1664 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1665 DBG3(" 3) %s", consumerd32_bin);
1666 if (stat(consumerd64_bin, &st) == 0) {
1667 DBG3("Found location #1");
1668 consumer_to_use = consumerd64_bin;
1669 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1670 DBG3("Found location #2");
1671 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1672 } else if (stat(consumerd32_bin, &st) == 0) {
1673 DBG3("Found location #3");
1674 consumer_to_use = consumerd32_bin;
1675 } else {
1676 DBG("Could not find any valid consumerd executable");
1677 break;
1678 }
1679 DBG("Using kernel consumer at: %s", consumer_to_use);
1680 execl(consumer_to_use,
1681 "lttng-consumerd", verbosity, "-k",
1682 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1683 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1684 NULL);
1685 break;
1686 case LTTNG_CONSUMER64_UST:
1687 {
1688 char *tmpnew = NULL;
1689
1690 if (consumerd64_libdir[0] != '\0') {
1691 char *tmp;
1692 size_t tmplen;
1693
1694 tmp = getenv("LD_LIBRARY_PATH");
1695 if (!tmp) {
1696 tmp = "";
1697 }
1698 tmplen = strlen("LD_LIBRARY_PATH=")
1699 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1700 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1701 if (!tmpnew) {
1702 ret = -ENOMEM;
1703 goto error;
1704 }
1705 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1706 strcat(tmpnew, consumerd64_libdir);
1707 if (tmp[0] != '\0') {
1708 strcat(tmpnew, ":");
1709 strcat(tmpnew, tmp);
1710 }
1711 ret = putenv(tmpnew);
1712 if (ret) {
1713 ret = -errno;
1714 goto error;
1715 }
1716 }
1717 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1718 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1719 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1720 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1721 NULL);
1722 if (consumerd64_libdir[0] != '\0') {
1723 free(tmpnew);
1724 }
1725 if (ret) {
1726 goto error;
1727 }
1728 break;
1729 }
1730 case LTTNG_CONSUMER32_UST:
1731 {
1732 char *tmpnew = NULL;
1733
1734 if (consumerd32_libdir[0] != '\0') {
1735 char *tmp;
1736 size_t tmplen;
1737
1738 tmp = getenv("LD_LIBRARY_PATH");
1739 if (!tmp) {
1740 tmp = "";
1741 }
1742 tmplen = strlen("LD_LIBRARY_PATH=")
1743 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1744 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1745 if (!tmpnew) {
1746 ret = -ENOMEM;
1747 goto error;
1748 }
1749 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1750 strcat(tmpnew, consumerd32_libdir);
1751 if (tmp[0] != '\0') {
1752 strcat(tmpnew, ":");
1753 strcat(tmpnew, tmp);
1754 }
1755 ret = putenv(tmpnew);
1756 if (ret) {
1757 ret = -errno;
1758 goto error;
1759 }
1760 }
1761 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1762 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1763 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1764 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1765 NULL);
1766 if (consumerd32_libdir[0] != '\0') {
1767 free(tmpnew);
1768 }
1769 if (ret) {
1770 goto error;
1771 }
1772 break;
1773 }
1774 default:
1775 PERROR("unknown consumer type");
1776 exit(EXIT_FAILURE);
1777 }
1778 if (errno != 0) {
1779 PERROR("kernel start consumer exec");
1780 }
1781 exit(EXIT_FAILURE);
1782 } else if (pid > 0) {
1783 ret = pid;
1784 } else {
1785 PERROR("start consumer fork");
1786 ret = -errno;
1787 }
1788 error:
1789 return ret;
1790 }
1791
1792 /*
1793 * Spawn the consumerd daemon and session daemon thread.
1794 */
1795 static int start_consumerd(struct consumer_data *consumer_data)
1796 {
1797 int ret;
1798
1799 pthread_mutex_lock(&consumer_data->pid_mutex);
1800 if (consumer_data->pid != 0) {
1801 pthread_mutex_unlock(&consumer_data->pid_mutex);
1802 goto end;
1803 }
1804
1805 ret = spawn_consumerd(consumer_data);
1806 if (ret < 0) {
1807 ERR("Spawning consumerd failed");
1808 pthread_mutex_unlock(&consumer_data->pid_mutex);
1809 goto error;
1810 }
1811
1812 /* Setting up the consumer_data pid */
1813 consumer_data->pid = ret;
1814 DBG2("Consumer pid %d", consumer_data->pid);
1815 pthread_mutex_unlock(&consumer_data->pid_mutex);
1816
1817 DBG2("Spawning consumer control thread");
1818 ret = spawn_consumer_thread(consumer_data);
1819 if (ret < 0) {
1820 ERR("Fatal error spawning consumer control thread");
1821 goto error;
1822 }
1823
1824 end:
1825 return 0;
1826
1827 error:
1828 return ret;
1829 }
1830
1831 /*
1832 * Compute health status of each consumer. If one of them is zero (bad
1833 * state), we return 0.
1834 */
1835 static int check_consumer_health(void)
1836 {
1837 int ret;
1838
1839 ret = health_check_state(&kconsumer_data.health) &&
1840 health_check_state(&ustconsumer32_data.health) &&
1841 health_check_state(&ustconsumer64_data.health);
1842
1843 DBG3("Health consumer check %d", ret);
1844
1845 return ret;
1846 }
1847
1848 /*
1849 * Setup necessary data for kernel tracer action.
1850 */
1851 static int init_kernel_tracer(void)
1852 {
1853 int ret;
1854
1855 /* Modprobe lttng kernel modules */
1856 ret = modprobe_lttng_control();
1857 if (ret < 0) {
1858 goto error;
1859 }
1860
1861 /* Open debugfs lttng */
1862 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1863 if (kernel_tracer_fd < 0) {
1864 DBG("Failed to open %s", module_proc_lttng);
1865 ret = -1;
1866 goto error_open;
1867 }
1868
1869 /* Validate kernel version */
1870 ret = kernel_validate_version(kernel_tracer_fd);
1871 if (ret < 0) {
1872 goto error_version;
1873 }
1874
1875 ret = modprobe_lttng_data();
1876 if (ret < 0) {
1877 goto error_modules;
1878 }
1879
1880 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1881 return 0;
1882
1883 error_version:
1884 modprobe_remove_lttng_control();
1885 ret = close(kernel_tracer_fd);
1886 if (ret) {
1887 PERROR("close");
1888 }
1889 kernel_tracer_fd = -1;
1890 return LTTNG_ERR_KERN_VERSION;
1891
1892 error_modules:
1893 ret = close(kernel_tracer_fd);
1894 if (ret) {
1895 PERROR("close");
1896 }
1897
1898 error_open:
1899 modprobe_remove_lttng_control();
1900
1901 error:
1902 WARN("No kernel tracer available");
1903 kernel_tracer_fd = -1;
1904 if (!is_root) {
1905 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1906 } else {
1907 return LTTNG_ERR_KERN_NA;
1908 }
1909 }
1910
1911
1912 /*
1913 * Copy consumer output from the tracing session to the domain session. The
1914 * function also applies the right modification on a per domain basis for the
1915 * trace files destination directory.
1916 */
1917 static int copy_session_consumer(int domain, struct ltt_session *session)
1918 {
1919 int ret;
1920 const char *dir_name;
1921 struct consumer_output *consumer;
1922
1923 assert(session);
1924 assert(session->consumer);
1925
1926 switch (domain) {
1927 case LTTNG_DOMAIN_KERNEL:
1928 DBG3("Copying tracing session consumer output in kernel session");
1929 session->kernel_session->consumer =
1930 consumer_copy_output(session->consumer);
1931 /* Ease our life a bit for the next part */
1932 consumer = session->kernel_session->consumer;
1933 dir_name = DEFAULT_KERNEL_TRACE_DIR;
1934 break;
1935 case LTTNG_DOMAIN_UST:
1936 DBG3("Copying tracing session consumer output in UST session");
1937 session->ust_session->consumer =
1938 consumer_copy_output(session->consumer);
1939 /* Ease our life a bit for the next part */
1940 consumer = session->ust_session->consumer;
1941 dir_name = DEFAULT_UST_TRACE_DIR;
1942 break;
1943 default:
1944 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
1945 goto error;
1946 }
1947
1948 /* Append correct directory to subdir */
1949 strncat(consumer->subdir, dir_name,
1950 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
1951 DBG3("Copy session consumer subdir %s", consumer->subdir);
1952
1953 ret = LTTNG_OK;
1954
1955 error:
1956 return ret;
1957 }
1958
1959 /*
1960 * Create an UST session and add it to the session ust list.
1961 */
1962 static int create_ust_session(struct ltt_session *session,
1963 struct lttng_domain *domain)
1964 {
1965 int ret;
1966 struct ltt_ust_session *lus = NULL;
1967
1968 assert(session);
1969 assert(domain);
1970 assert(session->consumer);
1971
1972 switch (domain->type) {
1973 case LTTNG_DOMAIN_UST:
1974 break;
1975 default:
1976 ERR("Unknown UST domain on create session %d", domain->type);
1977 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
1978 goto error;
1979 }
1980
1981 DBG("Creating UST session");
1982
1983 lus = trace_ust_create_session(session->path, session->id, domain);
1984 if (lus == NULL) {
1985 ret = LTTNG_ERR_UST_SESS_FAIL;
1986 goto error;
1987 }
1988
1989 lus->uid = session->uid;
1990 lus->gid = session->gid;
1991 session->ust_session = lus;
1992
1993 /* Copy session output to the newly created UST session */
1994 ret = copy_session_consumer(domain->type, session);
1995 if (ret != LTTNG_OK) {
1996 goto error;
1997 }
1998
1999 return LTTNG_OK;
2000
2001 error:
2002 free(lus);
2003 session->ust_session = NULL;
2004 return ret;
2005 }
2006
2007 /*
2008 * Create a kernel tracer session then create the default channel.
2009 */
2010 static int create_kernel_session(struct ltt_session *session)
2011 {
2012 int ret;
2013
2014 DBG("Creating kernel session");
2015
2016 ret = kernel_create_session(session, kernel_tracer_fd);
2017 if (ret < 0) {
2018 ret = LTTNG_ERR_KERN_SESS_FAIL;
2019 goto error;
2020 }
2021
2022 /* Code flow safety */
2023 assert(session->kernel_session);
2024
2025 /* Copy session output to the newly created Kernel session */
2026 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2027 if (ret != LTTNG_OK) {
2028 goto error;
2029 }
2030
2031 /* Create directory(ies) on local filesystem. */
2032 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2033 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2034 ret = run_as_mkdir_recursive(
2035 session->kernel_session->consumer->dst.trace_path,
2036 S_IRWXU | S_IRWXG, session->uid, session->gid);
2037 if (ret < 0) {
2038 if (ret != -EEXIST) {
2039 ERR("Trace directory creation error");
2040 goto error;
2041 }
2042 }
2043 }
2044
2045 session->kernel_session->uid = session->uid;
2046 session->kernel_session->gid = session->gid;
2047
2048 return LTTNG_OK;
2049
2050 error:
2051 trace_kernel_destroy_session(session->kernel_session);
2052 session->kernel_session = NULL;
2053 return ret;
2054 }
2055
2056 /*
2057 * Count number of session permitted by uid/gid.
2058 */
2059 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2060 {
2061 unsigned int i = 0;
2062 struct ltt_session *session;
2063
2064 DBG("Counting number of available session for UID %d GID %d",
2065 uid, gid);
2066 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2067 /*
2068 * Only list the sessions the user can control.
2069 */
2070 if (!session_access_ok(session, uid, gid)) {
2071 continue;
2072 }
2073 i++;
2074 }
2075 return i;
2076 }
2077
2078 /*
2079 * Process the command requested by the lttng client within the command
2080 * context structure. This function make sure that the return structure (llm)
2081 * is set and ready for transmission before returning.
2082 *
2083 * Return any error encountered or 0 for success.
2084 *
2085 * "sock" is only used for special-case var. len data.
2086 */
2087 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2088 int *sock_error)
2089 {
2090 int ret = LTTNG_OK;
2091 int need_tracing_session = 1;
2092 int need_domain;
2093
2094 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2095
2096 *sock_error = 0;
2097
2098 switch (cmd_ctx->lsm->cmd_type) {
2099 case LTTNG_CREATE_SESSION:
2100 case LTTNG_DESTROY_SESSION:
2101 case LTTNG_LIST_SESSIONS:
2102 case LTTNG_LIST_DOMAINS:
2103 case LTTNG_START_TRACE:
2104 case LTTNG_STOP_TRACE:
2105 case LTTNG_DATA_AVAILABLE:
2106 need_domain = 0;
2107 break;
2108 default:
2109 need_domain = 1;
2110 }
2111
2112 if (opt_no_kernel && need_domain
2113 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2114 if (!is_root) {
2115 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2116 } else {
2117 ret = LTTNG_ERR_KERN_NA;
2118 }
2119 goto error;
2120 }
2121
2122 /* Deny register consumer if we already have a spawned consumer. */
2123 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2124 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2125 if (kconsumer_data.pid > 0) {
2126 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2127 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2128 goto error;
2129 }
2130 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2131 }
2132
2133 /*
2134 * Check for command that don't needs to allocate a returned payload. We do
2135 * this here so we don't have to make the call for no payload at each
2136 * command.
2137 */
2138 switch(cmd_ctx->lsm->cmd_type) {
2139 case LTTNG_LIST_SESSIONS:
2140 case LTTNG_LIST_TRACEPOINTS:
2141 case LTTNG_LIST_TRACEPOINT_FIELDS:
2142 case LTTNG_LIST_DOMAINS:
2143 case LTTNG_LIST_CHANNELS:
2144 case LTTNG_LIST_EVENTS:
2145 break;
2146 default:
2147 /* Setup lttng message with no payload */
2148 ret = setup_lttng_msg(cmd_ctx, 0);
2149 if (ret < 0) {
2150 /* This label does not try to unlock the session */
2151 goto init_setup_error;
2152 }
2153 }
2154
2155 /* Commands that DO NOT need a session. */
2156 switch (cmd_ctx->lsm->cmd_type) {
2157 case LTTNG_CREATE_SESSION:
2158 case LTTNG_CALIBRATE:
2159 case LTTNG_LIST_SESSIONS:
2160 case LTTNG_LIST_TRACEPOINTS:
2161 case LTTNG_LIST_TRACEPOINT_FIELDS:
2162 need_tracing_session = 0;
2163 break;
2164 default:
2165 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2166 /*
2167 * We keep the session list lock across _all_ commands
2168 * for now, because the per-session lock does not
2169 * handle teardown properly.
2170 */
2171 session_lock_list();
2172 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2173 if (cmd_ctx->session == NULL) {
2174 if (cmd_ctx->lsm->session.name != NULL) {
2175 ret = LTTNG_ERR_SESS_NOT_FOUND;
2176 } else {
2177 /* If no session name specified */
2178 ret = LTTNG_ERR_SELECT_SESS;
2179 }
2180 goto error;
2181 } else {
2182 /* Acquire lock for the session */
2183 session_lock(cmd_ctx->session);
2184 }
2185 break;
2186 }
2187
2188 if (!need_domain) {
2189 goto skip_domain;
2190 }
2191
2192 /*
2193 * Check domain type for specific "pre-action".
2194 */
2195 switch (cmd_ctx->lsm->domain.type) {
2196 case LTTNG_DOMAIN_KERNEL:
2197 if (!is_root) {
2198 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2199 goto error;
2200 }
2201
2202 /* Kernel tracer check */
2203 if (kernel_tracer_fd == -1) {
2204 /* Basically, load kernel tracer modules */
2205 ret = init_kernel_tracer();
2206 if (ret != 0) {
2207 goto error;
2208 }
2209 }
2210
2211 /* Consumer is in an ERROR state. Report back to client */
2212 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2213 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2214 goto error;
2215 }
2216
2217 /* Need a session for kernel command */
2218 if (need_tracing_session) {
2219 if (cmd_ctx->session->kernel_session == NULL) {
2220 ret = create_kernel_session(cmd_ctx->session);
2221 if (ret < 0) {
2222 ret = LTTNG_ERR_KERN_SESS_FAIL;
2223 goto error;
2224 }
2225 }
2226
2227 /* Start the kernel consumer daemon */
2228 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2229 if (kconsumer_data.pid == 0 &&
2230 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2231 cmd_ctx->session->start_consumer) {
2232 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2233 ret = start_consumerd(&kconsumer_data);
2234 if (ret < 0) {
2235 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2236 goto error;
2237 }
2238 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2239 } else {
2240 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2241 }
2242
2243 /*
2244 * The consumer was just spawned so we need to add the socket to
2245 * the consumer output of the session if exist.
2246 */
2247 ret = consumer_create_socket(&kconsumer_data,
2248 cmd_ctx->session->kernel_session->consumer);
2249 if (ret < 0) {
2250 goto error;
2251 }
2252 }
2253
2254 break;
2255 case LTTNG_DOMAIN_UST:
2256 {
2257 /* Consumer is in an ERROR state. Report back to client */
2258 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2259 ret = LTTNG_ERR_NO_USTCONSUMERD;
2260 goto error;
2261 }
2262
2263 if (need_tracing_session) {
2264 /* Create UST session if none exist. */
2265 if (cmd_ctx->session->ust_session == NULL) {
2266 ret = create_ust_session(cmd_ctx->session,
2267 &cmd_ctx->lsm->domain);
2268 if (ret != LTTNG_OK) {
2269 goto error;
2270 }
2271 }
2272
2273 /* Start the UST consumer daemons */
2274 /* 64-bit */
2275 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2276 if (consumerd64_bin[0] != '\0' &&
2277 ustconsumer64_data.pid == 0 &&
2278 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2279 cmd_ctx->session->start_consumer) {
2280 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2281 ret = start_consumerd(&ustconsumer64_data);
2282 if (ret < 0) {
2283 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2284 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2285 goto error;
2286 }
2287
2288 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2289 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2290 } else {
2291 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2292 }
2293
2294 /*
2295 * Setup socket for consumer 64 bit. No need for atomic access
2296 * since it was set above and can ONLY be set in this thread.
2297 */
2298 ret = consumer_create_socket(&ustconsumer64_data,
2299 cmd_ctx->session->ust_session->consumer);
2300 if (ret < 0) {
2301 goto error;
2302 }
2303
2304 /* 32-bit */
2305 if (consumerd32_bin[0] != '\0' &&
2306 ustconsumer32_data.pid == 0 &&
2307 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2308 cmd_ctx->session->start_consumer) {
2309 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2310 ret = start_consumerd(&ustconsumer32_data);
2311 if (ret < 0) {
2312 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2313 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2314 goto error;
2315 }
2316
2317 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2318 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2319 } else {
2320 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2321 }
2322
2323 /*
2324 * Setup socket for consumer 64 bit. No need for atomic access
2325 * since it was set above and can ONLY be set in this thread.
2326 */
2327 ret = consumer_create_socket(&ustconsumer32_data,
2328 cmd_ctx->session->ust_session->consumer);
2329 if (ret < 0) {
2330 goto error;
2331 }
2332 }
2333 break;
2334 }
2335 default:
2336 break;
2337 }
2338 skip_domain:
2339
2340 /* Validate consumer daemon state when start/stop trace command */
2341 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2342 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2343 switch (cmd_ctx->lsm->domain.type) {
2344 case LTTNG_DOMAIN_UST:
2345 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2346 ret = LTTNG_ERR_NO_USTCONSUMERD;
2347 goto error;
2348 }
2349 break;
2350 case LTTNG_DOMAIN_KERNEL:
2351 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2352 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2353 goto error;
2354 }
2355 break;
2356 }
2357 }
2358
2359 /*
2360 * Check that the UID or GID match that of the tracing session.
2361 * The root user can interact with all sessions.
2362 */
2363 if (need_tracing_session) {
2364 if (!session_access_ok(cmd_ctx->session,
2365 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2366 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2367 ret = LTTNG_ERR_EPERM;
2368 goto error;
2369 }
2370 }
2371
2372 /* Process by command type */
2373 switch (cmd_ctx->lsm->cmd_type) {
2374 case LTTNG_ADD_CONTEXT:
2375 {
2376 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2377 cmd_ctx->lsm->u.context.channel_name,
2378 cmd_ctx->lsm->u.context.event_name,
2379 &cmd_ctx->lsm->u.context.ctx);
2380 break;
2381 }
2382 case LTTNG_DISABLE_CHANNEL:
2383 {
2384 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2385 cmd_ctx->lsm->u.disable.channel_name);
2386 break;
2387 }
2388 case LTTNG_DISABLE_EVENT:
2389 {
2390 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2391 cmd_ctx->lsm->u.disable.channel_name,
2392 cmd_ctx->lsm->u.disable.name);
2393 break;
2394 }
2395 case LTTNG_DISABLE_ALL_EVENT:
2396 {
2397 DBG("Disabling all events");
2398
2399 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2400 cmd_ctx->lsm->u.disable.channel_name);
2401 break;
2402 }
2403 case LTTNG_DISABLE_CONSUMER:
2404 {
2405 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2406 break;
2407 }
2408 case LTTNG_ENABLE_CHANNEL:
2409 {
2410 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2411 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2412 break;
2413 }
2414 case LTTNG_ENABLE_CONSUMER:
2415 {
2416 /*
2417 * XXX: 0 means that this URI should be applied on the session. Should
2418 * be a DOMAIN enuam.
2419 */
2420 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2421 if (ret != LTTNG_OK) {
2422 goto error;
2423 }
2424
2425 if (cmd_ctx->lsm->domain.type == 0) {
2426 /* Add the URI for the UST session if a consumer is present. */
2427 if (cmd_ctx->session->ust_session &&
2428 cmd_ctx->session->ust_session->consumer) {
2429 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2430 } else if (cmd_ctx->session->kernel_session &&
2431 cmd_ctx->session->kernel_session->consumer) {
2432 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2433 cmd_ctx->session);
2434 }
2435 }
2436 break;
2437 }
2438 case LTTNG_ENABLE_EVENT:
2439 {
2440 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2441 cmd_ctx->lsm->u.enable.channel_name,
2442 &cmd_ctx->lsm->u.enable.event, kernel_poll_pipe[1]);
2443 break;
2444 }
2445 case LTTNG_ENABLE_ALL_EVENT:
2446 {
2447 DBG("Enabling all events");
2448
2449 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2450 cmd_ctx->lsm->u.enable.channel_name,
2451 cmd_ctx->lsm->u.enable.event.type, kernel_poll_pipe[1]);
2452 break;
2453 }
2454 case LTTNG_LIST_TRACEPOINTS:
2455 {
2456 struct lttng_event *events;
2457 ssize_t nb_events;
2458
2459 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2460 if (nb_events < 0) {
2461 /* Return value is a negative lttng_error_code. */
2462 ret = -nb_events;
2463 goto error;
2464 }
2465
2466 /*
2467 * Setup lttng message with payload size set to the event list size in
2468 * bytes and then copy list into the llm payload.
2469 */
2470 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2471 if (ret < 0) {
2472 free(events);
2473 goto setup_error;
2474 }
2475
2476 /* Copy event list into message payload */
2477 memcpy(cmd_ctx->llm->payload, events,
2478 sizeof(struct lttng_event) * nb_events);
2479
2480 free(events);
2481
2482 ret = LTTNG_OK;
2483 break;
2484 }
2485 case LTTNG_LIST_TRACEPOINT_FIELDS:
2486 {
2487 struct lttng_event_field *fields;
2488 ssize_t nb_fields;
2489
2490 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2491 &fields);
2492 if (nb_fields < 0) {
2493 /* Return value is a negative lttng_error_code. */
2494 ret = -nb_fields;
2495 goto error;
2496 }
2497
2498 /*
2499 * Setup lttng message with payload size set to the event list size in
2500 * bytes and then copy list into the llm payload.
2501 */
2502 ret = setup_lttng_msg(cmd_ctx,
2503 sizeof(struct lttng_event_field) * nb_fields);
2504 if (ret < 0) {
2505 free(fields);
2506 goto setup_error;
2507 }
2508
2509 /* Copy event list into message payload */
2510 memcpy(cmd_ctx->llm->payload, fields,
2511 sizeof(struct lttng_event_field) * nb_fields);
2512
2513 free(fields);
2514
2515 ret = LTTNG_OK;
2516 break;
2517 }
2518 case LTTNG_SET_CONSUMER_URI:
2519 {
2520 size_t nb_uri, len;
2521 struct lttng_uri *uris;
2522
2523 nb_uri = cmd_ctx->lsm->u.uri.size;
2524 len = nb_uri * sizeof(struct lttng_uri);
2525
2526 if (nb_uri == 0) {
2527 ret = LTTNG_ERR_INVALID;
2528 goto error;
2529 }
2530
2531 uris = zmalloc(len);
2532 if (uris == NULL) {
2533 ret = LTTNG_ERR_FATAL;
2534 goto error;
2535 }
2536
2537 /* Receive variable len data */
2538 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2539 ret = lttcomm_recv_unix_sock(sock, uris, len);
2540 if (ret <= 0) {
2541 DBG("No URIs received from client... continuing");
2542 *sock_error = 1;
2543 ret = LTTNG_ERR_SESSION_FAIL;
2544 goto error;
2545 }
2546
2547 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2548 nb_uri, uris);
2549 if (ret != LTTNG_OK) {
2550 goto error;
2551 }
2552
2553 /*
2554 * XXX: 0 means that this URI should be applied on the session. Should
2555 * be a DOMAIN enuam.
2556 */
2557 if (cmd_ctx->lsm->domain.type == 0) {
2558 /* Add the URI for the UST session if a consumer is present. */
2559 if (cmd_ctx->session->ust_session &&
2560 cmd_ctx->session->ust_session->consumer) {
2561 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2562 nb_uri, uris);
2563 } else if (cmd_ctx->session->kernel_session &&
2564 cmd_ctx->session->kernel_session->consumer) {
2565 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2566 cmd_ctx->session, nb_uri, uris);
2567 }
2568 }
2569
2570 break;
2571 }
2572 case LTTNG_START_TRACE:
2573 {
2574 ret = cmd_start_trace(cmd_ctx->session);
2575 break;
2576 }
2577 case LTTNG_STOP_TRACE:
2578 {
2579 ret = cmd_stop_trace(cmd_ctx->session);
2580 break;
2581 }
2582 case LTTNG_CREATE_SESSION:
2583 {
2584 size_t nb_uri, len;
2585 struct lttng_uri *uris = NULL;
2586
2587 nb_uri = cmd_ctx->lsm->u.uri.size;
2588 len = nb_uri * sizeof(struct lttng_uri);
2589
2590 if (nb_uri > 0) {
2591 uris = zmalloc(len);
2592 if (uris == NULL) {
2593 ret = LTTNG_ERR_FATAL;
2594 goto error;
2595 }
2596
2597 /* Receive variable len data */
2598 DBG("Waiting for %zu URIs from client ...", nb_uri);
2599 ret = lttcomm_recv_unix_sock(sock, uris, len);
2600 if (ret <= 0) {
2601 DBG("No URIs received from client... continuing");
2602 *sock_error = 1;
2603 ret = LTTNG_ERR_SESSION_FAIL;
2604 goto error;
2605 }
2606
2607 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2608 DBG("Creating session with ONE network URI is a bad call");
2609 ret = LTTNG_ERR_SESSION_FAIL;
2610 goto error;
2611 }
2612 }
2613
2614 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2615 &cmd_ctx->creds);
2616
2617 break;
2618 }
2619 case LTTNG_DESTROY_SESSION:
2620 {
2621 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2622
2623 /* Set session to NULL so we do not unlock it after free. */
2624 cmd_ctx->session = NULL;
2625 break;
2626 }
2627 case LTTNG_LIST_DOMAINS:
2628 {
2629 ssize_t nb_dom;
2630 struct lttng_domain *domains;
2631
2632 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2633 if (nb_dom < 0) {
2634 /* Return value is a negative lttng_error_code. */
2635 ret = -nb_dom;
2636 goto error;
2637 }
2638
2639 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2640 if (ret < 0) {
2641 goto setup_error;
2642 }
2643
2644 /* Copy event list into message payload */
2645 memcpy(cmd_ctx->llm->payload, domains,
2646 nb_dom * sizeof(struct lttng_domain));
2647
2648 free(domains);
2649
2650 ret = LTTNG_OK;
2651 break;
2652 }
2653 case LTTNG_LIST_CHANNELS:
2654 {
2655 int nb_chan;
2656 struct lttng_channel *channels;
2657
2658 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2659 cmd_ctx->session, &channels);
2660 if (nb_chan < 0) {
2661 /* Return value is a negative lttng_error_code. */
2662 ret = -nb_chan;
2663 goto error;
2664 }
2665
2666 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2667 if (ret < 0) {
2668 goto setup_error;
2669 }
2670
2671 /* Copy event list into message payload */
2672 memcpy(cmd_ctx->llm->payload, channels,
2673 nb_chan * sizeof(struct lttng_channel));
2674
2675 free(channels);
2676
2677 ret = LTTNG_OK;
2678 break;
2679 }
2680 case LTTNG_LIST_EVENTS:
2681 {
2682 ssize_t nb_event;
2683 struct lttng_event *events = NULL;
2684
2685 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2686 cmd_ctx->lsm->u.list.channel_name, &events);
2687 if (nb_event < 0) {
2688 /* Return value is a negative lttng_error_code. */
2689 ret = -nb_event;
2690 goto error;
2691 }
2692
2693 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2694 if (ret < 0) {
2695 goto setup_error;
2696 }
2697
2698 /* Copy event list into message payload */
2699 memcpy(cmd_ctx->llm->payload, events,
2700 nb_event * sizeof(struct lttng_event));
2701
2702 free(events);
2703
2704 ret = LTTNG_OK;
2705 break;
2706 }
2707 case LTTNG_LIST_SESSIONS:
2708 {
2709 unsigned int nr_sessions;
2710
2711 session_lock_list();
2712 nr_sessions = lttng_sessions_count(
2713 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2714 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2715
2716 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2717 if (ret < 0) {
2718 session_unlock_list();
2719 goto setup_error;
2720 }
2721
2722 /* Filled the session array */
2723 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2724 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2725 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2726
2727 session_unlock_list();
2728
2729 ret = LTTNG_OK;
2730 break;
2731 }
2732 case LTTNG_CALIBRATE:
2733 {
2734 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2735 &cmd_ctx->lsm->u.calibrate);
2736 break;
2737 }
2738 case LTTNG_REGISTER_CONSUMER:
2739 {
2740 struct consumer_data *cdata;
2741
2742 switch (cmd_ctx->lsm->domain.type) {
2743 case LTTNG_DOMAIN_KERNEL:
2744 cdata = &kconsumer_data;
2745 break;
2746 default:
2747 ret = LTTNG_ERR_UND;
2748 goto error;
2749 }
2750
2751 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2752 cmd_ctx->lsm->u.reg.path, cdata);
2753 break;
2754 }
2755 case LTTNG_SET_FILTER:
2756 {
2757 struct lttng_filter_bytecode *bytecode;
2758
2759 if (cmd_ctx->lsm->u.filter.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2760 ret = LTTNG_ERR_FILTER_INVAL;
2761 goto error;
2762 }
2763 bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len);
2764 if (!bytecode) {
2765 ret = LTTNG_ERR_FILTER_NOMEM;
2766 goto error;
2767 }
2768 /* Receive var. len. data */
2769 DBG("Receiving var len data from client ...");
2770 ret = lttcomm_recv_unix_sock(sock, bytecode,
2771 cmd_ctx->lsm->u.filter.bytecode_len);
2772 if (ret <= 0) {
2773 DBG("Nothing recv() from client var len data... continuing");
2774 *sock_error = 1;
2775 ret = LTTNG_ERR_FILTER_INVAL;
2776 goto error;
2777 }
2778
2779 if (bytecode->len + sizeof(*bytecode)
2780 != cmd_ctx->lsm->u.filter.bytecode_len) {
2781 free(bytecode);
2782 ret = LTTNG_ERR_FILTER_INVAL;
2783 goto error;
2784 }
2785
2786 ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2787 cmd_ctx->lsm->u.filter.channel_name,
2788 cmd_ctx->lsm->u.filter.event_name,
2789 bytecode);
2790 break;
2791 }
2792 case LTTNG_DATA_AVAILABLE:
2793 {
2794 ret = cmd_data_available(cmd_ctx->session);
2795 break;
2796 }
2797 default:
2798 ret = LTTNG_ERR_UND;
2799 break;
2800 }
2801
2802 error:
2803 if (cmd_ctx->llm == NULL) {
2804 DBG("Missing llm structure. Allocating one.");
2805 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2806 goto setup_error;
2807 }
2808 }
2809 /* Set return code */
2810 cmd_ctx->llm->ret_code = ret;
2811 setup_error:
2812 if (cmd_ctx->session) {
2813 session_unlock(cmd_ctx->session);
2814 }
2815 if (need_tracing_session) {
2816 session_unlock_list();
2817 }
2818 init_setup_error:
2819 return ret;
2820 }
2821
2822 /*
2823 * Thread managing health check socket.
2824 */
2825 static void *thread_manage_health(void *data)
2826 {
2827 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2828 uint32_t revents, nb_fd;
2829 struct lttng_poll_event events;
2830 struct lttcomm_health_msg msg;
2831 struct lttcomm_health_data reply;
2832
2833 DBG("[thread] Manage health check started");
2834
2835 rcu_register_thread();
2836
2837 /* Create unix socket */
2838 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2839 if (sock < 0) {
2840 ERR("Unable to create health check Unix socket");
2841 ret = -1;
2842 goto error;
2843 }
2844
2845 ret = lttcomm_listen_unix_sock(sock);
2846 if (ret < 0) {
2847 goto error;
2848 }
2849
2850 /*
2851 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2852 * more will be added to this poll set.
2853 */
2854 ret = create_thread_poll_set(&events, 2);
2855 if (ret < 0) {
2856 goto error;
2857 }
2858
2859 /* Add the application registration socket */
2860 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2861 if (ret < 0) {
2862 goto error;
2863 }
2864
2865 while (1) {
2866 DBG("Health check ready");
2867
2868 nb_fd = LTTNG_POLL_GETNB(&events);
2869
2870 /* Inifinite blocking call, waiting for transmission */
2871 restart:
2872 ret = lttng_poll_wait(&events, -1);
2873 if (ret < 0) {
2874 /*
2875 * Restart interrupted system call.
2876 */
2877 if (errno == EINTR) {
2878 goto restart;
2879 }
2880 goto error;
2881 }
2882
2883 for (i = 0; i < nb_fd; i++) {
2884 /* Fetch once the poll data */
2885 revents = LTTNG_POLL_GETEV(&events, i);
2886 pollfd = LTTNG_POLL_GETFD(&events, i);
2887
2888 /* Thread quit pipe has been closed. Killing thread. */
2889 ret = check_thread_quit_pipe(pollfd, revents);
2890 if (ret) {
2891 err = 0;
2892 goto exit;
2893 }
2894
2895 /* Event on the registration socket */
2896 if (pollfd == sock) {
2897 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2898 ERR("Health socket poll error");
2899 goto error;
2900 }
2901 }
2902 }
2903
2904 new_sock = lttcomm_accept_unix_sock(sock);
2905 if (new_sock < 0) {
2906 goto error;
2907 }
2908
2909 DBG("Receiving data from client for health...");
2910 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
2911 if (ret <= 0) {
2912 DBG("Nothing recv() from client... continuing");
2913 ret = close(new_sock);
2914 if (ret) {
2915 PERROR("close");
2916 }
2917 new_sock = -1;
2918 continue;
2919 }
2920
2921 rcu_thread_online();
2922
2923 switch (msg.component) {
2924 case LTTNG_HEALTH_CMD:
2925 reply.ret_code = health_check_state(&health_thread_cmd);
2926 break;
2927 case LTTNG_HEALTH_APP_MANAGE:
2928 reply.ret_code = health_check_state(&health_thread_app_manage);
2929 break;
2930 case LTTNG_HEALTH_APP_REG:
2931 reply.ret_code = health_check_state(&health_thread_app_reg);
2932 break;
2933 case LTTNG_HEALTH_KERNEL:
2934 reply.ret_code = health_check_state(&health_thread_kernel);
2935 break;
2936 case LTTNG_HEALTH_CONSUMER:
2937 reply.ret_code = check_consumer_health();
2938 break;
2939 case LTTNG_HEALTH_ALL:
2940 reply.ret_code =
2941 health_check_state(&health_thread_app_manage) &&
2942 health_check_state(&health_thread_app_reg) &&
2943 health_check_state(&health_thread_cmd) &&
2944 health_check_state(&health_thread_kernel) &&
2945 check_consumer_health();
2946 break;
2947 default:
2948 reply.ret_code = LTTNG_ERR_UND;
2949 break;
2950 }
2951
2952 /*
2953 * Flip ret value since 0 is a success and 1 indicates a bad health for
2954 * the client where in the sessiond it is the opposite. Again, this is
2955 * just to make things easier for us poor developer which enjoy a lot
2956 * lazyness.
2957 */
2958 if (reply.ret_code == 0 || reply.ret_code == 1) {
2959 reply.ret_code = !reply.ret_code;
2960 }
2961
2962 DBG2("Health check return value %d", reply.ret_code);
2963
2964 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
2965 if (ret < 0) {
2966 ERR("Failed to send health data back to client");
2967 }
2968
2969 /* End of transmission */
2970 ret = close(new_sock);
2971 if (ret) {
2972 PERROR("close");
2973 }
2974 new_sock = -1;
2975 }
2976
2977 exit:
2978 error:
2979 if (err) {
2980 ERR("Health error occurred in %s", __func__);
2981 }
2982 DBG("Health check thread dying");
2983 unlink(health_unix_sock_path);
2984 if (sock >= 0) {
2985 ret = close(sock);
2986 if (ret) {
2987 PERROR("close");
2988 }
2989 }
2990 if (new_sock >= 0) {
2991 ret = close(new_sock);
2992 if (ret) {
2993 PERROR("close");
2994 }
2995 }
2996
2997 lttng_poll_clean(&events);
2998
2999 rcu_unregister_thread();
3000 return NULL;
3001 }
3002
3003 /*
3004 * This thread manage all clients request using the unix client socket for
3005 * communication.
3006 */
3007 static void *thread_manage_clients(void *data)
3008 {
3009 int sock = -1, ret, i, pollfd, err = -1;
3010 int sock_error;
3011 uint32_t revents, nb_fd;
3012 struct command_ctx *cmd_ctx = NULL;
3013 struct lttng_poll_event events;
3014
3015 DBG("[thread] Manage client started");
3016
3017 testpoint(thread_manage_clients);
3018
3019 rcu_register_thread();
3020
3021 health_code_update(&health_thread_cmd);
3022
3023 ret = lttcomm_listen_unix_sock(client_sock);
3024 if (ret < 0) {
3025 goto error;
3026 }
3027
3028 /*
3029 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3030 * more will be added to this poll set.
3031 */
3032 ret = create_thread_poll_set(&events, 2);
3033 if (ret < 0) {
3034 goto error;
3035 }
3036
3037 /* Add the application registration socket */
3038 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3039 if (ret < 0) {
3040 goto error;
3041 }
3042
3043 /*
3044 * Notify parent pid that we are ready to accept command for client side.
3045 */
3046 if (opt_sig_parent) {
3047 kill(ppid, SIGUSR1);
3048 }
3049
3050 testpoint(thread_manage_clients_before_loop);
3051
3052 health_code_update(&health_thread_cmd);
3053
3054 while (1) {
3055 DBG("Accepting client command ...");
3056
3057 nb_fd = LTTNG_POLL_GETNB(&events);
3058
3059 /* Inifinite blocking call, waiting for transmission */
3060 restart:
3061 health_poll_update(&health_thread_cmd);
3062 ret = lttng_poll_wait(&events, -1);
3063 health_poll_update(&health_thread_cmd);
3064 if (ret < 0) {
3065 /*
3066 * Restart interrupted system call.
3067 */
3068 if (errno == EINTR) {
3069 goto restart;
3070 }
3071 goto error;
3072 }
3073
3074 for (i = 0; i < nb_fd; i++) {
3075 /* Fetch once the poll data */
3076 revents = LTTNG_POLL_GETEV(&events, i);
3077 pollfd = LTTNG_POLL_GETFD(&events, i);
3078
3079 health_code_update(&health_thread_cmd);
3080
3081 /* Thread quit pipe has been closed. Killing thread. */
3082 ret = check_thread_quit_pipe(pollfd, revents);
3083 if (ret) {
3084 err = 0;
3085 goto exit;
3086 }
3087
3088 /* Event on the registration socket */
3089 if (pollfd == client_sock) {
3090 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3091 ERR("Client socket poll error");
3092 goto error;
3093 }
3094 }
3095 }
3096
3097 DBG("Wait for client response");
3098
3099 health_code_update(&health_thread_cmd);
3100
3101 sock = lttcomm_accept_unix_sock(client_sock);
3102 if (sock < 0) {
3103 goto error;
3104 }
3105
3106 /* Set socket option for credentials retrieval */
3107 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3108 if (ret < 0) {
3109 goto error;
3110 }
3111
3112 /* Allocate context command to process the client request */
3113 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3114 if (cmd_ctx == NULL) {
3115 PERROR("zmalloc cmd_ctx");
3116 goto error;
3117 }
3118
3119 /* Allocate data buffer for reception */
3120 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3121 if (cmd_ctx->lsm == NULL) {
3122 PERROR("zmalloc cmd_ctx->lsm");
3123 goto error;
3124 }
3125
3126 cmd_ctx->llm = NULL;
3127 cmd_ctx->session = NULL;
3128
3129 health_code_update(&health_thread_cmd);
3130
3131 /*
3132 * Data is received from the lttng client. The struct
3133 * lttcomm_session_msg (lsm) contains the command and data request of
3134 * the client.
3135 */
3136 DBG("Receiving data from client ...");
3137 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3138 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3139 if (ret <= 0) {
3140 DBG("Nothing recv() from client... continuing");
3141 ret = close(sock);
3142 if (ret) {
3143 PERROR("close");
3144 }
3145 sock = -1;
3146 clean_command_ctx(&cmd_ctx);
3147 continue;
3148 }
3149
3150 health_code_update(&health_thread_cmd);
3151
3152 // TODO: Validate cmd_ctx including sanity check for
3153 // security purpose.
3154
3155 rcu_thread_online();
3156 /*
3157 * This function dispatch the work to the kernel or userspace tracer
3158 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3159 * informations for the client. The command context struct contains
3160 * everything this function may needs.
3161 */
3162 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3163 rcu_thread_offline();
3164 if (ret < 0) {
3165 if (sock_error) {
3166 ret = close(sock);
3167 if (ret) {
3168 PERROR("close");
3169 }
3170 sock = -1;
3171 }
3172 /*
3173 * TODO: Inform client somehow of the fatal error. At
3174 * this point, ret < 0 means that a zmalloc failed
3175 * (ENOMEM). Error detected but still accept
3176 * command, unless a socket error has been
3177 * detected.
3178 */
3179 clean_command_ctx(&cmd_ctx);
3180 continue;
3181 }
3182
3183 health_code_update(&health_thread_cmd);
3184
3185 DBG("Sending response (size: %d, retcode: %s)",
3186 cmd_ctx->lttng_msg_size,
3187 lttng_strerror(-cmd_ctx->llm->ret_code));
3188 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3189 if (ret < 0) {
3190 ERR("Failed to send data back to client");
3191 }
3192
3193 /* End of transmission */
3194 ret = close(sock);
3195 if (ret) {
3196 PERROR("close");
3197 }
3198 sock = -1;
3199
3200 clean_command_ctx(&cmd_ctx);
3201
3202 health_code_update(&health_thread_cmd);
3203 }
3204
3205 exit:
3206 error:
3207 if (err) {
3208 health_error(&health_thread_cmd);
3209 ERR("Health error occurred in %s", __func__);
3210 }
3211 health_exit(&health_thread_cmd);
3212
3213 DBG("Client thread dying");
3214 unlink(client_unix_sock_path);
3215 if (client_sock >= 0) {
3216 ret = close(client_sock);
3217 if (ret) {
3218 PERROR("close");
3219 }
3220 }
3221 if (sock >= 0) {
3222 ret = close(sock);
3223 if (ret) {
3224 PERROR("close");
3225 }
3226 }
3227
3228 lttng_poll_clean(&events);
3229 clean_command_ctx(&cmd_ctx);
3230
3231 rcu_unregister_thread();
3232 return NULL;
3233 }
3234
3235
3236 /*
3237 * usage function on stderr
3238 */
3239 static void usage(void)
3240 {
3241 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3242 fprintf(stderr, " -h, --help Display this usage.\n");
3243 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3244 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3245 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3246 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3247 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3248 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3249 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3250 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3251 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3252 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3253 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3254 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3255 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3256 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3257 fprintf(stderr, " -V, --version Show version number.\n");
3258 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3259 fprintf(stderr, " -q, --quiet No output at all.\n");
3260 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3261 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3262 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3263 }
3264
3265 /*
3266 * daemon argument parsing
3267 */
3268 static int parse_args(int argc, char **argv)
3269 {
3270 int c;
3271
3272 static struct option long_options[] = {
3273 { "client-sock", 1, 0, 'c' },
3274 { "apps-sock", 1, 0, 'a' },
3275 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3276 { "kconsumerd-err-sock", 1, 0, 'E' },
3277 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3278 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3279 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3280 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3281 { "consumerd32-path", 1, 0, 'u' },
3282 { "consumerd32-libdir", 1, 0, 'U' },
3283 { "consumerd64-path", 1, 0, 't' },
3284 { "consumerd64-libdir", 1, 0, 'T' },
3285 { "daemonize", 0, 0, 'd' },
3286 { "sig-parent", 0, 0, 'S' },
3287 { "help", 0, 0, 'h' },
3288 { "group", 1, 0, 'g' },
3289 { "version", 0, 0, 'V' },
3290 { "quiet", 0, 0, 'q' },
3291 { "verbose", 0, 0, 'v' },
3292 { "verbose-consumer", 0, 0, 'Z' },
3293 { "no-kernel", 0, 0, 'N' },
3294 { NULL, 0, 0, 0 }
3295 };
3296
3297 while (1) {
3298 int option_index = 0;
3299 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3300 long_options, &option_index);
3301 if (c == -1) {
3302 break;
3303 }
3304
3305 switch (c) {
3306 case 0:
3307 fprintf(stderr, "option %s", long_options[option_index].name);
3308 if (optarg) {
3309 fprintf(stderr, " with arg %s\n", optarg);
3310 }
3311 break;
3312 case 'c':
3313 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3314 break;
3315 case 'a':
3316 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3317 break;
3318 case 'd':
3319 opt_daemon = 1;
3320 break;
3321 case 'g':
3322 opt_tracing_group = optarg;
3323 break;
3324 case 'h':
3325 usage();
3326 exit(EXIT_FAILURE);
3327 case 'V':
3328 fprintf(stdout, "%s\n", VERSION);
3329 exit(EXIT_SUCCESS);
3330 case 'S':
3331 opt_sig_parent = 1;
3332 break;
3333 case 'E':
3334 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3335 break;
3336 case 'C':
3337 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3338 break;
3339 case 'F':
3340 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3341 break;
3342 case 'D':
3343 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3344 break;
3345 case 'H':
3346 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3347 break;
3348 case 'G':
3349 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3350 break;
3351 case 'N':
3352 opt_no_kernel = 1;
3353 break;
3354 case 'q':
3355 lttng_opt_quiet = 1;
3356 break;
3357 case 'v':
3358 /* Verbose level can increase using multiple -v */
3359 lttng_opt_verbose += 1;
3360 break;
3361 case 'Z':
3362 opt_verbose_consumer += 1;
3363 break;
3364 case 'u':
3365 consumerd32_bin= optarg;
3366 break;
3367 case 'U':
3368 consumerd32_libdir = optarg;
3369 break;
3370 case 't':
3371 consumerd64_bin = optarg;
3372 break;
3373 case 'T':
3374 consumerd64_libdir = optarg;
3375 break;
3376 default:
3377 /* Unknown option or other error.
3378 * Error is printed by getopt, just return */
3379 return -1;
3380 }
3381 }
3382
3383 return 0;
3384 }
3385
3386 /*
3387 * Creates the two needed socket by the daemon.
3388 * apps_sock - The communication socket for all UST apps.
3389 * client_sock - The communication of the cli tool (lttng).
3390 */
3391 static int init_daemon_socket(void)
3392 {
3393 int ret = 0;
3394 mode_t old_umask;
3395
3396 old_umask = umask(0);
3397
3398 /* Create client tool unix socket */
3399 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3400 if (client_sock < 0) {
3401 ERR("Create unix sock failed: %s", client_unix_sock_path);
3402 ret = -1;
3403 goto end;
3404 }
3405
3406 /* File permission MUST be 660 */
3407 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3408 if (ret < 0) {
3409 ERR("Set file permissions failed: %s", client_unix_sock_path);
3410 PERROR("chmod");
3411 goto end;
3412 }
3413
3414 /* Create the application unix socket */
3415 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3416 if (apps_sock < 0) {
3417 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3418 ret = -1;
3419 goto end;
3420 }
3421
3422 /* File permission MUST be 666 */
3423 ret = chmod(apps_unix_sock_path,
3424 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3425 if (ret < 0) {
3426 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3427 PERROR("chmod");
3428 goto end;
3429 }
3430
3431 end:
3432 umask(old_umask);
3433 return ret;
3434 }
3435
3436 /*
3437 * Check if the global socket is available, and if a daemon is answering at the
3438 * other side. If yes, error is returned.
3439 */
3440 static int check_existing_daemon(void)
3441 {
3442 /* Is there anybody out there ? */
3443 if (lttng_session_daemon_alive()) {
3444 return -EEXIST;
3445 }
3446
3447 return 0;
3448 }
3449
3450 /*
3451 * Set the tracing group gid onto the client socket.
3452 *
3453 * Race window between mkdir and chown is OK because we are going from more
3454 * permissive (root.root) to less permissive (root.tracing).
3455 */
3456 static int set_permissions(char *rundir)
3457 {
3458 int ret;
3459 gid_t gid;
3460
3461 ret = allowed_group();
3462 if (ret < 0) {
3463 WARN("No tracing group detected");
3464 ret = 0;
3465 goto end;
3466 }
3467
3468 gid = ret;
3469
3470 /* Set lttng run dir */
3471 ret = chown(rundir, 0, gid);
3472 if (ret < 0) {
3473 ERR("Unable to set group on %s", rundir);
3474 PERROR("chown");
3475 }
3476
3477 /* Ensure tracing group can search the run dir */
3478 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3479 if (ret < 0) {
3480 ERR("Unable to set permissions on %s", rundir);
3481 PERROR("chmod");
3482 }
3483
3484 /* lttng client socket path */
3485 ret = chown(client_unix_sock_path, 0, gid);
3486 if (ret < 0) {
3487 ERR("Unable to set group on %s", client_unix_sock_path);
3488 PERROR("chown");
3489 }
3490
3491 /* kconsumer error socket path */
3492 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3493 if (ret < 0) {
3494 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3495 PERROR("chown");
3496 }
3497
3498 /* 64-bit ustconsumer error socket path */
3499 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3500 if (ret < 0) {
3501 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3502 PERROR("chown");
3503 }
3504
3505 /* 32-bit ustconsumer compat32 error socket path */
3506 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3507 if (ret < 0) {
3508 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3509 PERROR("chown");
3510 }
3511
3512 DBG("All permissions are set");
3513
3514 end:
3515 return ret;
3516 }
3517
3518 /*
3519 * Create the lttng run directory needed for all global sockets and pipe.
3520 */
3521 static int create_lttng_rundir(const char *rundir)
3522 {
3523 int ret;
3524
3525 DBG3("Creating LTTng run directory: %s", rundir);
3526
3527 ret = mkdir(rundir, S_IRWXU);
3528 if (ret < 0) {
3529 if (errno != EEXIST) {
3530 ERR("Unable to create %s", rundir);
3531 goto error;
3532 } else {
3533 ret = 0;
3534 }
3535 }
3536
3537 error:
3538 return ret;
3539 }
3540
3541 /*
3542 * Setup sockets and directory needed by the kconsumerd communication with the
3543 * session daemon.
3544 */
3545 static int set_consumer_sockets(struct consumer_data *consumer_data,
3546 const char *rundir)
3547 {
3548 int ret;
3549 char path[PATH_MAX];
3550
3551 switch (consumer_data->type) {
3552 case LTTNG_CONSUMER_KERNEL:
3553 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3554 break;
3555 case LTTNG_CONSUMER64_UST:
3556 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3557 break;
3558 case LTTNG_CONSUMER32_UST:
3559 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3560 break;
3561 default:
3562 ERR("Consumer type unknown");
3563 ret = -EINVAL;
3564 goto error;
3565 }
3566
3567 DBG2("Creating consumer directory: %s", path);
3568
3569 ret = mkdir(path, S_IRWXU);
3570 if (ret < 0) {
3571 if (errno != EEXIST) {
3572 PERROR("mkdir");
3573 ERR("Failed to create %s", path);
3574 goto error;
3575 }
3576 ret = -1;
3577 }
3578
3579 /* Create the kconsumerd error unix socket */
3580 consumer_data->err_sock =
3581 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3582 if (consumer_data->err_sock < 0) {
3583 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3584 ret = -1;
3585 goto error;
3586 }
3587
3588 /* File permission MUST be 660 */
3589 ret = chmod(consumer_data->err_unix_sock_path,
3590 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3591 if (ret < 0) {
3592 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3593 PERROR("chmod");
3594 goto error;
3595 }
3596
3597 error:
3598 return ret;
3599 }
3600
3601 /*
3602 * Signal handler for the daemon
3603 *
3604 * Simply stop all worker threads, leaving main() return gracefully after
3605 * joining all threads and calling cleanup().
3606 */
3607 static void sighandler(int sig)
3608 {
3609 switch (sig) {
3610 case SIGPIPE:
3611 DBG("SIGPIPE caught");
3612 return;
3613 case SIGINT:
3614 DBG("SIGINT caught");
3615 stop_threads();
3616 break;
3617 case SIGTERM:
3618 DBG("SIGTERM caught");
3619 stop_threads();
3620 break;
3621 default:
3622 break;
3623 }
3624 }
3625
3626 /*
3627 * Setup signal handler for :
3628 * SIGINT, SIGTERM, SIGPIPE
3629 */
3630 static int set_signal_handler(void)
3631 {
3632 int ret = 0;
3633 struct sigaction sa;
3634 sigset_t sigset;
3635
3636 if ((ret = sigemptyset(&sigset)) < 0) {
3637 PERROR("sigemptyset");
3638 return ret;
3639 }
3640
3641 sa.sa_handler = sighandler;
3642 sa.sa_mask = sigset;
3643 sa.sa_flags = 0;
3644 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3645 PERROR("sigaction");
3646 return ret;
3647 }
3648
3649 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3650 PERROR("sigaction");
3651 return ret;
3652 }
3653
3654 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3655 PERROR("sigaction");
3656 return ret;
3657 }
3658
3659 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3660
3661 return ret;
3662 }
3663
3664 /*
3665 * Set open files limit to unlimited. This daemon can open a large number of
3666 * file descriptors in order to consumer multiple kernel traces.
3667 */
3668 static void set_ulimit(void)
3669 {
3670 int ret;
3671 struct rlimit lim;
3672
3673 /* The kernel does not allowed an infinite limit for open files */
3674 lim.rlim_cur = 65535;
3675 lim.rlim_max = 65535;
3676
3677 ret = setrlimit(RLIMIT_NOFILE, &lim);
3678 if (ret < 0) {
3679 PERROR("failed to set open files limit");
3680 }
3681 }
3682
3683 /*
3684 * main
3685 */
3686 int main(int argc, char **argv)
3687 {
3688 int ret = 0;
3689 void *status;
3690 const char *home_path;
3691
3692 init_kernel_workarounds();
3693
3694 rcu_register_thread();
3695
3696 setup_consumerd_path();
3697
3698 /* Parse arguments */
3699 progname = argv[0];
3700 if ((ret = parse_args(argc, argv) < 0)) {
3701 goto error;
3702 }
3703
3704 /* Daemonize */
3705 if (opt_daemon) {
3706 int i;
3707
3708 /*
3709 * fork
3710 * child: setsid, close FD 0, 1, 2, chdir /
3711 * parent: exit (if fork is successful)
3712 */
3713 ret = daemon(0, 0);
3714 if (ret < 0) {
3715 PERROR("daemon");
3716 goto error;
3717 }
3718 /*
3719 * We are in the child. Make sure all other file
3720 * descriptors are closed, in case we are called with
3721 * more opened file descriptors than the standard ones.
3722 */
3723 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3724 (void) close(i);
3725 }
3726 }
3727
3728 /* Create thread quit pipe */
3729 if ((ret = init_thread_quit_pipe()) < 0) {
3730 goto error;
3731 }
3732
3733 /* Check if daemon is UID = 0 */
3734 is_root = !getuid();
3735
3736 if (is_root) {
3737 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3738
3739 /* Create global run dir with root access */
3740 ret = create_lttng_rundir(rundir);
3741 if (ret < 0) {
3742 goto error;
3743 }
3744
3745 if (strlen(apps_unix_sock_path) == 0) {
3746 snprintf(apps_unix_sock_path, PATH_MAX,
3747 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3748 }
3749
3750 if (strlen(client_unix_sock_path) == 0) {
3751 snprintf(client_unix_sock_path, PATH_MAX,
3752 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3753 }
3754
3755 /* Set global SHM for ust */
3756 if (strlen(wait_shm_path) == 0) {
3757 snprintf(wait_shm_path, PATH_MAX,
3758 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3759 }
3760
3761 if (strlen(health_unix_sock_path) == 0) {
3762 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3763 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3764 }
3765
3766 /* Setup kernel consumerd path */
3767 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3768 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3769 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3770 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3771
3772 DBG2("Kernel consumer err path: %s",
3773 kconsumer_data.err_unix_sock_path);
3774 DBG2("Kernel consumer cmd path: %s",
3775 kconsumer_data.cmd_unix_sock_path);
3776 } else {
3777 home_path = get_home_dir();
3778 if (home_path == NULL) {
3779 /* TODO: Add --socket PATH option */
3780 ERR("Can't get HOME directory for sockets creation.");
3781 ret = -EPERM;
3782 goto error;
3783 }
3784
3785 /*
3786 * Create rundir from home path. This will create something like
3787 * $HOME/.lttng
3788 */
3789 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3790 if (ret < 0) {
3791 ret = -ENOMEM;
3792 goto error;
3793 }
3794
3795 ret = create_lttng_rundir(rundir);
3796 if (ret < 0) {
3797 goto error;
3798 }
3799
3800 if (strlen(apps_unix_sock_path) == 0) {
3801 snprintf(apps_unix_sock_path, PATH_MAX,
3802 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3803 }
3804
3805 /* Set the cli tool unix socket path */
3806 if (strlen(client_unix_sock_path) == 0) {
3807 snprintf(client_unix_sock_path, PATH_MAX,
3808 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3809 }
3810
3811 /* Set global SHM for ust */
3812 if (strlen(wait_shm_path) == 0) {
3813 snprintf(wait_shm_path, PATH_MAX,
3814 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3815 }
3816
3817 /* Set health check Unix path */
3818 if (strlen(health_unix_sock_path) == 0) {
3819 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3820 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3821 }
3822 }
3823
3824 /* Set consumer initial state */
3825 kernel_consumerd_state = CONSUMER_STOPPED;
3826 ust_consumerd_state = CONSUMER_STOPPED;
3827
3828 DBG("Client socket path %s", client_unix_sock_path);
3829 DBG("Application socket path %s", apps_unix_sock_path);
3830 DBG("LTTng run directory path: %s", rundir);
3831
3832 /* 32 bits consumerd path setup */
3833 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
3834 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
3835 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
3836 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
3837
3838 DBG2("UST consumer 32 bits err path: %s",
3839 ustconsumer32_data.err_unix_sock_path);
3840 DBG2("UST consumer 32 bits cmd path: %s",
3841 ustconsumer32_data.cmd_unix_sock_path);
3842
3843 /* 64 bits consumerd path setup */
3844 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
3845 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
3846 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
3847 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
3848
3849 DBG2("UST consumer 64 bits err path: %s",
3850 ustconsumer64_data.err_unix_sock_path);
3851 DBG2("UST consumer 64 bits cmd path: %s",
3852 ustconsumer64_data.cmd_unix_sock_path);
3853
3854 /*
3855 * See if daemon already exist.
3856 */
3857 if ((ret = check_existing_daemon()) < 0) {
3858 ERR("Already running daemon.\n");
3859 /*
3860 * We do not goto exit because we must not cleanup()
3861 * because a daemon is already running.
3862 */
3863 goto error;
3864 }
3865
3866 /*
3867 * Init UST app hash table. Alloc hash table before this point since
3868 * cleanup() can get called after that point.
3869 */
3870 ust_app_ht_alloc();
3871
3872 /* After this point, we can safely call cleanup() with "goto exit" */
3873
3874 /*
3875 * These actions must be executed as root. We do that *after* setting up
3876 * the sockets path because we MUST make the check for another daemon using
3877 * those paths *before* trying to set the kernel consumer sockets and init
3878 * kernel tracer.
3879 */
3880 if (is_root) {
3881 ret = set_consumer_sockets(&kconsumer_data, rundir);
3882 if (ret < 0) {
3883 goto exit;
3884 }
3885
3886 /* Setup kernel tracer */
3887 if (!opt_no_kernel) {
3888 init_kernel_tracer();
3889 }
3890
3891 /* Set ulimit for open files */
3892 set_ulimit();
3893 }
3894 /* init lttng_fd tracking must be done after set_ulimit. */
3895 lttng_fd_init();
3896
3897 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
3898 if (ret < 0) {
3899 goto exit;
3900 }
3901
3902 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
3903 if (ret < 0) {
3904 goto exit;
3905 }
3906
3907 if ((ret = set_signal_handler()) < 0) {
3908 goto exit;
3909 }
3910
3911 /* Setup the needed unix socket */
3912 if ((ret = init_daemon_socket()) < 0) {
3913 goto exit;
3914 }
3915
3916 /* Set credentials to socket */
3917 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
3918 goto exit;
3919 }
3920
3921 /* Get parent pid if -S, --sig-parent is specified. */
3922 if (opt_sig_parent) {
3923 ppid = getppid();
3924 }
3925
3926 /* Setup the kernel pipe for waking up the kernel thread */
3927 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
3928 goto exit;
3929 }
3930
3931 /* Setup the thread apps communication pipe. */
3932 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
3933 goto exit;
3934 }
3935
3936 /* Init UST command queue. */
3937 cds_wfq_init(&ust_cmd_queue.queue);
3938
3939 /*
3940 * Get session list pointer. This pointer MUST NOT be free(). This list is
3941 * statically declared in session.c
3942 */
3943 session_list_ptr = session_get_list();
3944
3945 /* Set up max poll set size */
3946 lttng_poll_set_max_size();
3947
3948 cmd_init();
3949
3950 /* Init all health thread counters. */
3951 health_init(&health_thread_cmd);
3952 health_init(&health_thread_kernel);
3953 health_init(&health_thread_app_manage);
3954 health_init(&health_thread_app_reg);
3955
3956 /*
3957 * Init health counters of the consumer thread. We do a quick hack here to
3958 * the state of the consumer health is fine even if the thread is not
3959 * started. This is simply to ease our life and has no cost what so ever.
3960 */
3961 health_init(&kconsumer_data.health);
3962 health_poll_update(&kconsumer_data.health);
3963 health_init(&ustconsumer32_data.health);
3964 health_poll_update(&ustconsumer32_data.health);
3965 health_init(&ustconsumer64_data.health);
3966 health_poll_update(&ustconsumer64_data.health);
3967
3968 /* Create thread to manage the client socket */
3969 ret = pthread_create(&health_thread, NULL,
3970 thread_manage_health, (void *) NULL);
3971 if (ret != 0) {
3972 PERROR("pthread_create health");
3973 goto exit_health;
3974 }
3975
3976 /* Create thread to manage the client socket */
3977 ret = pthread_create(&client_thread, NULL,
3978 thread_manage_clients, (void *) NULL);
3979 if (ret != 0) {
3980 PERROR("pthread_create clients");
3981 goto exit_client;
3982 }
3983
3984 /* Create thread to dispatch registration */
3985 ret = pthread_create(&dispatch_thread, NULL,
3986 thread_dispatch_ust_registration, (void *) NULL);
3987 if (ret != 0) {
3988 PERROR("pthread_create dispatch");
3989 goto exit_dispatch;
3990 }
3991
3992 /* Create thread to manage application registration. */
3993 ret = pthread_create(&reg_apps_thread, NULL,
3994 thread_registration_apps, (void *) NULL);
3995 if (ret != 0) {
3996 PERROR("pthread_create registration");
3997 goto exit_reg_apps;
3998 }
3999
4000 /* Create thread to manage application socket */
4001 ret = pthread_create(&apps_thread, NULL,
4002 thread_manage_apps, (void *) NULL);
4003 if (ret != 0) {
4004 PERROR("pthread_create apps");
4005 goto exit_apps;
4006 }
4007
4008 /* Create kernel thread to manage kernel event */
4009 ret = pthread_create(&kernel_thread, NULL,
4010 thread_manage_kernel, (void *) NULL);
4011 if (ret != 0) {
4012 PERROR("pthread_create kernel");
4013 goto exit_kernel;
4014 }
4015
4016 ret = pthread_join(kernel_thread, &status);
4017 if (ret != 0) {
4018 PERROR("pthread_join");
4019 goto error; /* join error, exit without cleanup */
4020 }
4021
4022 exit_kernel:
4023 ret = pthread_join(apps_thread, &status);
4024 if (ret != 0) {
4025 PERROR("pthread_join");
4026 goto error; /* join error, exit without cleanup */
4027 }
4028
4029 exit_apps:
4030 ret = pthread_join(reg_apps_thread, &status);
4031 if (ret != 0) {
4032 PERROR("pthread_join");
4033 goto error; /* join error, exit without cleanup */
4034 }
4035
4036 exit_reg_apps:
4037 ret = pthread_join(dispatch_thread, &status);
4038 if (ret != 0) {
4039 PERROR("pthread_join");
4040 goto error; /* join error, exit without cleanup */
4041 }
4042
4043 exit_dispatch:
4044 ret = pthread_join(client_thread, &status);
4045 if (ret != 0) {
4046 PERROR("pthread_join");
4047 goto error; /* join error, exit without cleanup */
4048 }
4049
4050 ret = join_consumer_thread(&kconsumer_data);
4051 if (ret != 0) {
4052 PERROR("join_consumer");
4053 goto error; /* join error, exit without cleanup */
4054 }
4055
4056 exit_client:
4057 exit_health:
4058 exit:
4059 /*
4060 * cleanup() is called when no other thread is running.
4061 */
4062 rcu_thread_online();
4063 cleanup();
4064 rcu_thread_offline();
4065 rcu_unregister_thread();
4066 if (!ret) {
4067 exit(EXIT_SUCCESS);
4068 }
4069 error:
4070 exit(EXIT_FAILURE);
4071 }
This page took 0.208053 seconds and 4 git commands to generate.