Fix: Typo from a previous patch in an assert()
[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 /*
232 * Socket timeout for receiving and sending in seconds.
233 */
234 static int app_socket_timeout;
235
236 static
237 void setup_consumerd_path(void)
238 {
239 const char *bin, *libdir;
240
241 /*
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
245 */
246 #if (CAA_BITS_PER_LONG == 32)
247 if (!consumerd32_bin[0]) {
248 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
249 }
250 if (!consumerd32_libdir[0]) {
251 consumerd32_libdir = INSTALL_LIB_PATH;
252 }
253 #elif (CAA_BITS_PER_LONG == 64)
254 if (!consumerd64_bin[0]) {
255 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
256 }
257 if (!consumerd64_libdir[0]) {
258 consumerd64_libdir = INSTALL_LIB_PATH;
259 }
260 #else
261 #error "Unknown bitness"
262 #endif
263
264 /*
265 * runtime env. var. overrides the build default.
266 */
267 bin = getenv("LTTNG_CONSUMERD32_BIN");
268 if (bin) {
269 consumerd32_bin = bin;
270 }
271 bin = getenv("LTTNG_CONSUMERD64_BIN");
272 if (bin) {
273 consumerd64_bin = bin;
274 }
275 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
276 if (libdir) {
277 consumerd32_libdir = libdir;
278 }
279 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
280 if (libdir) {
281 consumerd64_libdir = libdir;
282 }
283 }
284
285 /*
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
287 */
288 static int create_thread_poll_set(struct lttng_poll_event *events,
289 unsigned int size)
290 {
291 int ret;
292
293 if (events == NULL || size == 0) {
294 ret = -1;
295 goto error;
296 }
297
298 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
299 if (ret < 0) {
300 goto error;
301 }
302
303 /* Add quit pipe */
304 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
305 if (ret < 0) {
306 goto error;
307 }
308
309 return 0;
310
311 error:
312 return ret;
313 }
314
315 /*
316 * Check if the thread quit pipe was triggered.
317 *
318 * Return 1 if it was triggered else 0;
319 */
320 static int check_thread_quit_pipe(int fd, uint32_t events)
321 {
322 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
323 return 1;
324 }
325
326 return 0;
327 }
328
329 /*
330 * Return group ID of the tracing group or -1 if not found.
331 */
332 static gid_t allowed_group(void)
333 {
334 struct group *grp;
335
336 if (opt_tracing_group) {
337 grp = getgrnam(opt_tracing_group);
338 } else {
339 grp = getgrnam(default_tracing_group);
340 }
341 if (!grp) {
342 return -1;
343 } else {
344 return grp->gr_gid;
345 }
346 }
347
348 /*
349 * Init thread quit pipe.
350 *
351 * Return -1 on error or 0 if all pipes are created.
352 */
353 static int init_thread_quit_pipe(void)
354 {
355 int ret, i;
356
357 ret = pipe(thread_quit_pipe);
358 if (ret < 0) {
359 PERROR("thread quit pipe");
360 goto error;
361 }
362
363 for (i = 0; i < 2; i++) {
364 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
365 if (ret < 0) {
366 PERROR("fcntl");
367 goto error;
368 }
369 }
370
371 error:
372 return ret;
373 }
374
375 /*
376 * Stop all threads by closing the thread quit pipe.
377 */
378 static void stop_threads(void)
379 {
380 int ret;
381
382 /* Stopping all threads */
383 DBG("Terminating all threads");
384 ret = notify_thread_pipe(thread_quit_pipe[1]);
385 if (ret < 0) {
386 ERR("write error on thread quit pipe");
387 }
388
389 /* Dispatch thread */
390 CMM_STORE_SHARED(dispatch_thread_exit, 1);
391 futex_nto1_wake(&ust_cmd_queue.futex);
392 }
393
394 /*
395 * Cleanup the daemon
396 */
397 static void cleanup(void)
398 {
399 int ret;
400 char *cmd = NULL;
401 struct ltt_session *sess, *stmp;
402
403 DBG("Cleaning up");
404
405 /* First thing first, stop all threads */
406 utils_close_pipe(thread_quit_pipe);
407
408 DBG("Removing %s directory", rundir);
409 ret = asprintf(&cmd, "rm -rf %s", rundir);
410 if (ret < 0) {
411 ERR("asprintf failed. Something is really wrong!");
412 }
413
414 /* Remove lttng run directory */
415 ret = system(cmd);
416 if (ret < 0) {
417 ERR("Unable to clean %s", rundir);
418 }
419 free(cmd);
420 free(rundir);
421
422 DBG("Cleaning up all sessions");
423
424 /* Destroy session list mutex */
425 if (session_list_ptr != NULL) {
426 pthread_mutex_destroy(&session_list_ptr->lock);
427
428 /* Cleanup ALL session */
429 cds_list_for_each_entry_safe(sess, stmp,
430 &session_list_ptr->head, list) {
431 cmd_destroy_session(sess, kernel_poll_pipe[1]);
432 }
433 }
434
435 DBG("Closing all UST sockets");
436 ust_app_clean_list();
437
438 if (is_root && !opt_no_kernel) {
439 DBG2("Closing kernel fd");
440 if (kernel_tracer_fd >= 0) {
441 ret = close(kernel_tracer_fd);
442 if (ret) {
443 PERROR("close");
444 }
445 }
446 DBG("Unloading kernel modules");
447 modprobe_remove_lttng_all();
448 }
449
450 /* <fun> */
451 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
452 "Matthew, BEET driven development works!%c[%dm",
453 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
454 /* </fun> */
455 }
456
457 /*
458 * Send data on a unix socket using the liblttsessiondcomm API.
459 *
460 * Return lttcomm error code.
461 */
462 static int send_unix_sock(int sock, void *buf, size_t len)
463 {
464 /* Check valid length */
465 if (len <= 0) {
466 return -1;
467 }
468
469 return lttcomm_send_unix_sock(sock, buf, len);
470 }
471
472 /*
473 * Free memory of a command context structure.
474 */
475 static void clean_command_ctx(struct command_ctx **cmd_ctx)
476 {
477 DBG("Clean command context structure");
478 if (*cmd_ctx) {
479 if ((*cmd_ctx)->llm) {
480 free((*cmd_ctx)->llm);
481 }
482 if ((*cmd_ctx)->lsm) {
483 free((*cmd_ctx)->lsm);
484 }
485 free(*cmd_ctx);
486 *cmd_ctx = NULL;
487 }
488 }
489
490 /*
491 * Notify UST applications using the shm mmap futex.
492 */
493 static int notify_ust_apps(int active)
494 {
495 char *wait_shm_mmap;
496
497 DBG("Notifying applications of session daemon state: %d", active);
498
499 /* See shm.c for this call implying mmap, shm and futex calls */
500 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
501 if (wait_shm_mmap == NULL) {
502 goto error;
503 }
504
505 /* Wake waiting process */
506 futex_wait_update((int32_t *) wait_shm_mmap, active);
507
508 /* Apps notified successfully */
509 return 0;
510
511 error:
512 return -1;
513 }
514
515 /*
516 * Setup the outgoing data buffer for the response (llm) by allocating the
517 * right amount of memory and copying the original information from the lsm
518 * structure.
519 *
520 * Return total size of the buffer pointed by buf.
521 */
522 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
523 {
524 int ret, buf_size;
525
526 buf_size = size;
527
528 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
529 if (cmd_ctx->llm == NULL) {
530 PERROR("zmalloc");
531 ret = -ENOMEM;
532 goto error;
533 }
534
535 /* Copy common data */
536 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
537 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
538
539 cmd_ctx->llm->data_size = size;
540 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
541
542 return buf_size;
543
544 error:
545 return ret;
546 }
547
548 /*
549 * Update the kernel poll set of all channel fd available over all tracing
550 * session. Add the wakeup pipe at the end of the set.
551 */
552 static int update_kernel_poll(struct lttng_poll_event *events)
553 {
554 int ret;
555 struct ltt_session *session;
556 struct ltt_kernel_channel *channel;
557
558 DBG("Updating kernel poll set");
559
560 session_lock_list();
561 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
562 session_lock(session);
563 if (session->kernel_session == NULL) {
564 session_unlock(session);
565 continue;
566 }
567
568 cds_list_for_each_entry(channel,
569 &session->kernel_session->channel_list.head, list) {
570 /* Add channel fd to the kernel poll set */
571 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
572 if (ret < 0) {
573 session_unlock(session);
574 goto error;
575 }
576 DBG("Channel fd %d added to kernel set", channel->fd);
577 }
578 session_unlock(session);
579 }
580 session_unlock_list();
581
582 return 0;
583
584 error:
585 session_unlock_list();
586 return -1;
587 }
588
589 /*
590 * Find the channel fd from 'fd' over all tracing session. When found, check
591 * for new channel stream and send those stream fds to the kernel consumer.
592 *
593 * Useful for CPU hotplug feature.
594 */
595 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
596 {
597 int ret = 0;
598 struct ltt_session *session;
599 struct ltt_kernel_session *ksess;
600 struct ltt_kernel_channel *channel;
601
602 DBG("Updating kernel streams for channel fd %d", fd);
603
604 session_lock_list();
605 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
606 session_lock(session);
607 if (session->kernel_session == NULL) {
608 session_unlock(session);
609 continue;
610 }
611 ksess = session->kernel_session;
612
613 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
614 if (channel->fd == fd) {
615 DBG("Channel found, updating kernel streams");
616 ret = kernel_open_channel_stream(channel);
617 if (ret < 0) {
618 goto error;
619 }
620
621 /*
622 * Have we already sent fds to the consumer? If yes, it means
623 * that tracing is started so it is safe to send our updated
624 * stream fds.
625 */
626 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
627 struct lttng_ht_iter iter;
628 struct consumer_socket *socket;
629
630
631 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
632 &iter.iter, socket, node.node) {
633 /* Code flow error */
634 assert(socket->fd >= 0);
635
636 pthread_mutex_lock(socket->lock);
637 ret = kernel_consumer_send_channel_stream(socket->fd,
638 channel, ksess);
639 pthread_mutex_unlock(socket->lock);
640 if (ret < 0) {
641 goto error;
642 }
643 }
644 }
645 goto error;
646 }
647 }
648 session_unlock(session);
649 }
650 session_unlock_list();
651 return ret;
652
653 error:
654 session_unlock(session);
655 session_unlock_list();
656 return ret;
657 }
658
659 /*
660 * For each tracing session, update newly registered apps.
661 */
662 static void update_ust_app(int app_sock)
663 {
664 struct ltt_session *sess, *stmp;
665
666 session_lock_list();
667
668 /* For all tracing session(s) */
669 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
670 session_lock(sess);
671 if (sess->ust_session) {
672 ust_app_global_update(sess->ust_session, app_sock);
673 }
674 session_unlock(sess);
675 }
676
677 session_unlock_list();
678 }
679
680 /*
681 * This thread manage event coming from the kernel.
682 *
683 * Features supported in this thread:
684 * -) CPU Hotplug
685 */
686 static void *thread_manage_kernel(void *data)
687 {
688 int ret, i, pollfd, update_poll_flag = 1, err = -1;
689 uint32_t revents, nb_fd;
690 char tmp;
691 struct lttng_poll_event events;
692
693 DBG("Thread manage kernel started");
694
695 testpoint(thread_manage_kernel);
696
697 health_code_update(&health_thread_kernel);
698
699 testpoint(thread_manage_kernel_before_loop);
700
701 ret = create_thread_poll_set(&events, 2);
702 if (ret < 0) {
703 goto error_poll_create;
704 }
705
706 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
707 if (ret < 0) {
708 goto error;
709 }
710
711 while (1) {
712 health_code_update(&health_thread_kernel);
713
714 if (update_poll_flag == 1) {
715 /*
716 * Reset number of fd in the poll set. Always 2 since there is the thread
717 * quit pipe and the kernel pipe.
718 */
719 events.nb_fd = 2;
720
721 ret = update_kernel_poll(&events);
722 if (ret < 0) {
723 goto error;
724 }
725 update_poll_flag = 0;
726 }
727
728 nb_fd = LTTNG_POLL_GETNB(&events);
729
730 DBG("Thread kernel polling on %d fds", nb_fd);
731
732 /* Zeroed the poll events */
733 lttng_poll_reset(&events);
734
735 /* Poll infinite value of time */
736 restart:
737 health_poll_update(&health_thread_kernel);
738 ret = lttng_poll_wait(&events, -1);
739 health_poll_update(&health_thread_kernel);
740 if (ret < 0) {
741 /*
742 * Restart interrupted system call.
743 */
744 if (errno == EINTR) {
745 goto restart;
746 }
747 goto error;
748 } else if (ret == 0) {
749 /* Should not happen since timeout is infinite */
750 ERR("Return value of poll is 0 with an infinite timeout.\n"
751 "This should not have happened! Continuing...");
752 continue;
753 }
754
755 for (i = 0; i < nb_fd; i++) {
756 /* Fetch once the poll data */
757 revents = LTTNG_POLL_GETEV(&events, i);
758 pollfd = LTTNG_POLL_GETFD(&events, i);
759
760 health_code_update(&health_thread_kernel);
761
762 /* Thread quit pipe has been closed. Killing thread. */
763 ret = check_thread_quit_pipe(pollfd, revents);
764 if (ret) {
765 err = 0;
766 goto exit;
767 }
768
769 /* Check for data on kernel pipe */
770 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
771 ret = read(kernel_poll_pipe[0], &tmp, 1);
772 update_poll_flag = 1;
773 continue;
774 } else {
775 /*
776 * New CPU detected by the kernel. Adding kernel stream to
777 * kernel session and updating the kernel consumer
778 */
779 if (revents & LPOLLIN) {
780 ret = update_kernel_stream(&kconsumer_data, pollfd);
781 if (ret < 0) {
782 continue;
783 }
784 break;
785 /*
786 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
787 * and unregister kernel stream at this point.
788 */
789 }
790 }
791 }
792 }
793
794 exit:
795 error:
796 lttng_poll_clean(&events);
797 error_poll_create:
798 utils_close_pipe(kernel_poll_pipe);
799 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
800 if (err) {
801 health_error(&health_thread_kernel);
802 ERR("Health error occurred in %s", __func__);
803 WARN("Kernel thread died unexpectedly. "
804 "Kernel tracing can continue but CPU hotplug is disabled.");
805 }
806 health_exit(&health_thread_kernel);
807 DBG("Kernel thread dying");
808 return NULL;
809 }
810
811 /*
812 * Signal pthread condition of the consumer data that the thread.
813 */
814 static void signal_consumer_condition(struct consumer_data *data, int state)
815 {
816 pthread_mutex_lock(&data->cond_mutex);
817
818 /*
819 * The state is set before signaling. It can be any value, it's the waiter
820 * job to correctly interpret this condition variable associated to the
821 * consumer pthread_cond.
822 *
823 * A value of 0 means that the corresponding thread of the consumer data
824 * was not started. 1 indicates that the thread has started and is ready
825 * for action. A negative value means that there was an error during the
826 * thread bootstrap.
827 */
828 data->consumer_thread_is_ready = state;
829 (void) pthread_cond_signal(&data->cond);
830
831 pthread_mutex_unlock(&data->cond_mutex);
832 }
833
834 /*
835 * This thread manage the consumer error sent back to the session daemon.
836 */
837 static void *thread_manage_consumer(void *data)
838 {
839 int sock = -1, i, ret, pollfd, err = -1;
840 uint32_t revents, nb_fd;
841 enum lttcomm_return_code code;
842 struct lttng_poll_event events;
843 struct consumer_data *consumer_data = data;
844
845 DBG("[thread] Manage consumer started");
846
847 /*
848 * Since the consumer thread can be spawned at any moment in time, we init
849 * the health to a poll status (1, which is a valid health over time).
850 * When the thread starts, we update here the health to a "code" path being
851 * an even value so this thread, when reaching a poll wait, does not
852 * trigger an error with an even value.
853 *
854 * Here is the use case we avoid.
855 *
856 * +1: the first poll update during initialization (main())
857 * +2 * x: multiple code update once in this thread.
858 * +1: poll wait in this thread (being a good health state).
859 * == even number which after the wait period shows as a bad health.
860 *
861 * In a nutshell, the following poll update to the health state brings back
862 * the state to an even value meaning a code path.
863 */
864 health_poll_update(&consumer_data->health);
865
866 /*
867 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
868 * Nothing more will be added to this poll set.
869 */
870 ret = create_thread_poll_set(&events, 2);
871 if (ret < 0) {
872 goto error_poll;
873 }
874
875 /*
876 * The error socket here is already in a listening state which was done
877 * just before spawning this thread to avoid a race between the consumer
878 * daemon exec trying to connect and the listen() call.
879 */
880 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
881 if (ret < 0) {
882 goto error;
883 }
884
885 nb_fd = LTTNG_POLL_GETNB(&events);
886
887 health_code_update(&consumer_data->health);
888
889 /* Inifinite blocking call, waiting for transmission */
890 restart:
891 health_poll_update(&consumer_data->health);
892
893 testpoint(thread_manage_consumer);
894
895 ret = lttng_poll_wait(&events, -1);
896 health_poll_update(&consumer_data->health);
897 if (ret < 0) {
898 /*
899 * Restart interrupted system call.
900 */
901 if (errno == EINTR) {
902 goto restart;
903 }
904 goto error;
905 }
906
907 for (i = 0; i < nb_fd; i++) {
908 /* Fetch once the poll data */
909 revents = LTTNG_POLL_GETEV(&events, i);
910 pollfd = LTTNG_POLL_GETFD(&events, i);
911
912 health_code_update(&consumer_data->health);
913
914 /* Thread quit pipe has been closed. Killing thread. */
915 ret = check_thread_quit_pipe(pollfd, revents);
916 if (ret) {
917 err = 0;
918 goto exit;
919 }
920
921 /* Event on the registration socket */
922 if (pollfd == consumer_data->err_sock) {
923 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
924 ERR("consumer err socket poll error");
925 goto error;
926 }
927 }
928 }
929
930 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
931 if (sock < 0) {
932 goto error;
933 }
934
935 /*
936 * Set the CLOEXEC flag. Return code is useless because either way, the
937 * show must go on.
938 */
939 (void) utils_set_fd_cloexec(sock);
940
941 health_code_update(&consumer_data->health);
942
943 DBG2("Receiving code from consumer err_sock");
944
945 /* Getting status code from kconsumerd */
946 ret = lttcomm_recv_unix_sock(sock, &code,
947 sizeof(enum lttcomm_return_code));
948 if (ret <= 0) {
949 goto error;
950 }
951
952 health_code_update(&consumer_data->health);
953
954 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
955 consumer_data->cmd_sock =
956 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
957 if (consumer_data->cmd_sock < 0) {
958 /* On error, signal condition and quit. */
959 signal_consumer_condition(consumer_data, -1);
960 PERROR("consumer connect");
961 goto error;
962 }
963 signal_consumer_condition(consumer_data, 1);
964 DBG("Consumer command socket ready");
965 } else {
966 ERR("consumer error when waiting for SOCK_READY : %s",
967 lttcomm_get_readable_code(-code));
968 goto error;
969 }
970
971 /* Remove the kconsumerd error sock since we've established a connexion */
972 ret = lttng_poll_del(&events, consumer_data->err_sock);
973 if (ret < 0) {
974 goto error;
975 }
976
977 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
978 if (ret < 0) {
979 goto error;
980 }
981
982 health_code_update(&consumer_data->health);
983
984 /* Update number of fd */
985 nb_fd = LTTNG_POLL_GETNB(&events);
986
987 /* Inifinite blocking call, waiting for transmission */
988 restart_poll:
989 health_poll_update(&consumer_data->health);
990 ret = lttng_poll_wait(&events, -1);
991 health_poll_update(&consumer_data->health);
992 if (ret < 0) {
993 /*
994 * Restart interrupted system call.
995 */
996 if (errno == EINTR) {
997 goto restart_poll;
998 }
999 goto error;
1000 }
1001
1002 for (i = 0; i < nb_fd; i++) {
1003 /* Fetch once the poll data */
1004 revents = LTTNG_POLL_GETEV(&events, i);
1005 pollfd = LTTNG_POLL_GETFD(&events, i);
1006
1007 health_code_update(&consumer_data->health);
1008
1009 /* Thread quit pipe has been closed. Killing thread. */
1010 ret = check_thread_quit_pipe(pollfd, revents);
1011 if (ret) {
1012 err = 0;
1013 goto exit;
1014 }
1015
1016 /* Event on the kconsumerd socket */
1017 if (pollfd == sock) {
1018 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1019 ERR("consumer err socket second poll error");
1020 goto error;
1021 }
1022 }
1023 }
1024
1025 health_code_update(&consumer_data->health);
1026
1027 /* Wait for any kconsumerd error */
1028 ret = lttcomm_recv_unix_sock(sock, &code,
1029 sizeof(enum lttcomm_return_code));
1030 if (ret <= 0) {
1031 ERR("consumer closed the command socket");
1032 goto error;
1033 }
1034
1035 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1036
1037 exit:
1038 error:
1039 /* Immediately set the consumerd state to stopped */
1040 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1041 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1042 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1043 consumer_data->type == LTTNG_CONSUMER32_UST) {
1044 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1045 } else {
1046 /* Code flow error... */
1047 assert(0);
1048 }
1049
1050 if (consumer_data->err_sock >= 0) {
1051 ret = close(consumer_data->err_sock);
1052 if (ret) {
1053 PERROR("close");
1054 }
1055 }
1056 if (consumer_data->cmd_sock >= 0) {
1057 ret = close(consumer_data->cmd_sock);
1058 if (ret) {
1059 PERROR("close");
1060 }
1061 }
1062 if (sock >= 0) {
1063 ret = close(sock);
1064 if (ret) {
1065 PERROR("close");
1066 }
1067 }
1068
1069 unlink(consumer_data->err_unix_sock_path);
1070 unlink(consumer_data->cmd_unix_sock_path);
1071 consumer_data->pid = 0;
1072
1073 lttng_poll_clean(&events);
1074 error_poll:
1075 if (err) {
1076 health_error(&consumer_data->health);
1077 ERR("Health error occurred in %s", __func__);
1078 }
1079 health_exit(&consumer_data->health);
1080 DBG("consumer thread cleanup completed");
1081
1082 return NULL;
1083 }
1084
1085 /*
1086 * This thread manage application communication.
1087 */
1088 static void *thread_manage_apps(void *data)
1089 {
1090 int i, ret, pollfd, err = -1;
1091 uint32_t revents, nb_fd;
1092 struct ust_command ust_cmd;
1093 struct lttng_poll_event events;
1094
1095 DBG("[thread] Manage application started");
1096
1097 testpoint(thread_manage_apps);
1098
1099 rcu_register_thread();
1100 rcu_thread_online();
1101
1102 health_code_update(&health_thread_app_manage);
1103
1104 ret = create_thread_poll_set(&events, 2);
1105 if (ret < 0) {
1106 goto error_poll_create;
1107 }
1108
1109 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1110 if (ret < 0) {
1111 goto error;
1112 }
1113
1114 testpoint(thread_manage_apps_before_loop);
1115
1116 health_code_update(&health_thread_app_manage);
1117
1118 while (1) {
1119 /* Zeroed the events structure */
1120 lttng_poll_reset(&events);
1121
1122 nb_fd = LTTNG_POLL_GETNB(&events);
1123
1124 DBG("Apps thread polling on %d fds", nb_fd);
1125
1126 /* Inifinite blocking call, waiting for transmission */
1127 restart:
1128 health_poll_update(&health_thread_app_manage);
1129 ret = lttng_poll_wait(&events, -1);
1130 health_poll_update(&health_thread_app_manage);
1131 if (ret < 0) {
1132 /*
1133 * Restart interrupted system call.
1134 */
1135 if (errno == EINTR) {
1136 goto restart;
1137 }
1138 goto error;
1139 }
1140
1141 for (i = 0; i < nb_fd; i++) {
1142 /* Fetch once the poll data */
1143 revents = LTTNG_POLL_GETEV(&events, i);
1144 pollfd = LTTNG_POLL_GETFD(&events, i);
1145
1146 health_code_update(&health_thread_app_manage);
1147
1148 /* Thread quit pipe has been closed. Killing thread. */
1149 ret = check_thread_quit_pipe(pollfd, revents);
1150 if (ret) {
1151 err = 0;
1152 goto exit;
1153 }
1154
1155 /* Inspect the apps cmd pipe */
1156 if (pollfd == apps_cmd_pipe[0]) {
1157 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1158 ERR("Apps command pipe error");
1159 goto error;
1160 } else if (revents & LPOLLIN) {
1161 /* Empty pipe */
1162 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1163 if (ret < 0 || ret < sizeof(ust_cmd)) {
1164 PERROR("read apps cmd pipe");
1165 goto error;
1166 }
1167
1168 health_code_update(&health_thread_app_manage);
1169
1170 /* Register applicaton to the session daemon */
1171 ret = ust_app_register(&ust_cmd.reg_msg,
1172 ust_cmd.sock);
1173 if (ret == -ENOMEM) {
1174 goto error;
1175 } else if (ret < 0) {
1176 break;
1177 }
1178
1179 health_code_update(&health_thread_app_manage);
1180
1181 /*
1182 * Validate UST version compatibility.
1183 */
1184 ret = ust_app_validate_version(ust_cmd.sock);
1185 if (ret >= 0) {
1186 /*
1187 * Add channel(s) and event(s) to newly registered apps
1188 * from lttng global UST domain.
1189 */
1190 update_ust_app(ust_cmd.sock);
1191 }
1192
1193 health_code_update(&health_thread_app_manage);
1194
1195 ret = ust_app_register_done(ust_cmd.sock);
1196 if (ret < 0) {
1197 /*
1198 * If the registration is not possible, we simply
1199 * unregister the apps and continue
1200 */
1201 ust_app_unregister(ust_cmd.sock);
1202 } else {
1203 /*
1204 * We only monitor the error events of the socket. This
1205 * thread does not handle any incoming data from UST
1206 * (POLLIN).
1207 */
1208 ret = lttng_poll_add(&events, ust_cmd.sock,
1209 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1210 if (ret < 0) {
1211 goto error;
1212 }
1213
1214 /* Set socket timeout for both receiving and ending */
1215 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1216 app_socket_timeout);
1217 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1218 app_socket_timeout);
1219
1220 DBG("Apps with sock %d added to poll set",
1221 ust_cmd.sock);
1222 }
1223
1224 health_code_update(&health_thread_app_manage);
1225
1226 break;
1227 }
1228 } else {
1229 /*
1230 * At this point, we know that a registered application made
1231 * the event at poll_wait.
1232 */
1233 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1234 /* Removing from the poll set */
1235 ret = lttng_poll_del(&events, pollfd);
1236 if (ret < 0) {
1237 goto error;
1238 }
1239
1240 /* Socket closed on remote end. */
1241 ust_app_unregister(pollfd);
1242 break;
1243 }
1244 }
1245
1246 health_code_update(&health_thread_app_manage);
1247 }
1248 }
1249
1250 exit:
1251 error:
1252 lttng_poll_clean(&events);
1253 error_poll_create:
1254 utils_close_pipe(apps_cmd_pipe);
1255 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1256
1257 /*
1258 * We don't clean the UST app hash table here since already registered
1259 * applications can still be controlled so let them be until the session
1260 * daemon dies or the applications stop.
1261 */
1262
1263 if (err) {
1264 health_error(&health_thread_app_manage);
1265 ERR("Health error occurred in %s", __func__);
1266 }
1267 health_exit(&health_thread_app_manage);
1268 DBG("Application communication apps thread cleanup complete");
1269 rcu_thread_offline();
1270 rcu_unregister_thread();
1271 return NULL;
1272 }
1273
1274 /*
1275 * Dispatch request from the registration threads to the application
1276 * communication thread.
1277 */
1278 static void *thread_dispatch_ust_registration(void *data)
1279 {
1280 int ret;
1281 struct cds_wfq_node *node;
1282 struct ust_command *ust_cmd = NULL;
1283
1284 DBG("[thread] Dispatch UST command started");
1285
1286 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1287 /* Atomically prepare the queue futex */
1288 futex_nto1_prepare(&ust_cmd_queue.futex);
1289
1290 do {
1291 /* Dequeue command for registration */
1292 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1293 if (node == NULL) {
1294 DBG("Woken up but nothing in the UST command queue");
1295 /* Continue thread execution */
1296 break;
1297 }
1298
1299 ust_cmd = caa_container_of(node, struct ust_command, node);
1300
1301 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1302 " gid:%d sock:%d name:%s (version %d.%d)",
1303 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1304 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1305 ust_cmd->sock, ust_cmd->reg_msg.name,
1306 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1307 /*
1308 * Inform apps thread of the new application registration. This
1309 * call is blocking so we can be assured that the data will be read
1310 * at some point in time or wait to the end of the world :)
1311 */
1312 if (apps_cmd_pipe[1] >= 0) {
1313 ret = write(apps_cmd_pipe[1], ust_cmd,
1314 sizeof(struct ust_command));
1315 if (ret < 0) {
1316 PERROR("write apps cmd pipe");
1317 if (errno == EBADF) {
1318 /*
1319 * We can't inform the application thread to process
1320 * registration. We will exit or else application
1321 * registration will not occur and tracing will never
1322 * start.
1323 */
1324 goto error;
1325 }
1326 }
1327 } else {
1328 /* Application manager thread is not available. */
1329 ret = close(ust_cmd->sock);
1330 if (ret < 0) {
1331 PERROR("close ust_cmd sock");
1332 }
1333 }
1334 free(ust_cmd);
1335 } while (node != NULL);
1336
1337 /* Futex wait on queue. Blocking call on futex() */
1338 futex_nto1_wait(&ust_cmd_queue.futex);
1339 }
1340
1341 error:
1342 DBG("Dispatch thread dying");
1343 return NULL;
1344 }
1345
1346 /*
1347 * This thread manage application registration.
1348 */
1349 static void *thread_registration_apps(void *data)
1350 {
1351 int sock = -1, i, ret, pollfd, err = -1;
1352 uint32_t revents, nb_fd;
1353 struct lttng_poll_event events;
1354 /*
1355 * Get allocated in this thread, enqueued to a global queue, dequeued and
1356 * freed in the manage apps thread.
1357 */
1358 struct ust_command *ust_cmd = NULL;
1359
1360 DBG("[thread] Manage application registration started");
1361
1362 testpoint(thread_registration_apps);
1363
1364 ret = lttcomm_listen_unix_sock(apps_sock);
1365 if (ret < 0) {
1366 goto error_listen;
1367 }
1368
1369 /*
1370 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1371 * more will be added to this poll set.
1372 */
1373 ret = create_thread_poll_set(&events, 2);
1374 if (ret < 0) {
1375 goto error_create_poll;
1376 }
1377
1378 /* Add the application registration socket */
1379 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1380 if (ret < 0) {
1381 goto error_poll_add;
1382 }
1383
1384 /* Notify all applications to register */
1385 ret = notify_ust_apps(1);
1386 if (ret < 0) {
1387 ERR("Failed to notify applications or create the wait shared memory.\n"
1388 "Execution continues but there might be problem for already\n"
1389 "running applications that wishes to register.");
1390 }
1391
1392 while (1) {
1393 DBG("Accepting application registration");
1394
1395 nb_fd = LTTNG_POLL_GETNB(&events);
1396
1397 /* Inifinite blocking call, waiting for transmission */
1398 restart:
1399 health_poll_update(&health_thread_app_reg);
1400 ret = lttng_poll_wait(&events, -1);
1401 health_poll_update(&health_thread_app_reg);
1402 if (ret < 0) {
1403 /*
1404 * Restart interrupted system call.
1405 */
1406 if (errno == EINTR) {
1407 goto restart;
1408 }
1409 goto error;
1410 }
1411
1412 for (i = 0; i < nb_fd; i++) {
1413 health_code_update(&health_thread_app_reg);
1414
1415 /* Fetch once the poll data */
1416 revents = LTTNG_POLL_GETEV(&events, i);
1417 pollfd = LTTNG_POLL_GETFD(&events, i);
1418
1419 /* Thread quit pipe has been closed. Killing thread. */
1420 ret = check_thread_quit_pipe(pollfd, revents);
1421 if (ret) {
1422 err = 0;
1423 goto exit;
1424 }
1425
1426 /* Event on the registration socket */
1427 if (pollfd == apps_sock) {
1428 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1429 ERR("Register apps socket poll error");
1430 goto error;
1431 } else if (revents & LPOLLIN) {
1432 sock = lttcomm_accept_unix_sock(apps_sock);
1433 if (sock < 0) {
1434 goto error;
1435 }
1436
1437 /*
1438 * Set the CLOEXEC flag. Return code is useless because
1439 * either way, the show must go on.
1440 */
1441 (void) utils_set_fd_cloexec(sock);
1442
1443 /* Create UST registration command for enqueuing */
1444 ust_cmd = zmalloc(sizeof(struct ust_command));
1445 if (ust_cmd == NULL) {
1446 PERROR("ust command zmalloc");
1447 goto error;
1448 }
1449
1450 /*
1451 * Using message-based transmissions to ensure we don't
1452 * have to deal with partially received messages.
1453 */
1454 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1455 if (ret < 0) {
1456 ERR("Exhausted file descriptors allowed for applications.");
1457 free(ust_cmd);
1458 ret = close(sock);
1459 if (ret) {
1460 PERROR("close");
1461 }
1462 sock = -1;
1463 continue;
1464 }
1465 health_code_update(&health_thread_app_reg);
1466 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1467 sizeof(struct ust_register_msg));
1468 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1469 if (ret < 0) {
1470 PERROR("lttcomm_recv_unix_sock register apps");
1471 } else {
1472 ERR("Wrong size received on apps register");
1473 }
1474 free(ust_cmd);
1475 ret = close(sock);
1476 if (ret) {
1477 PERROR("close");
1478 }
1479 lttng_fd_put(LTTNG_FD_APPS, 1);
1480 sock = -1;
1481 continue;
1482 }
1483 health_code_update(&health_thread_app_reg);
1484
1485 ust_cmd->sock = sock;
1486 sock = -1;
1487
1488 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1489 " gid:%d sock:%d name:%s (version %d.%d)",
1490 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1491 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1492 ust_cmd->sock, ust_cmd->reg_msg.name,
1493 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1494
1495 /*
1496 * Lock free enqueue the registration request. The red pill
1497 * has been taken! This apps will be part of the *system*.
1498 */
1499 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1500
1501 /*
1502 * Wake the registration queue futex. Implicit memory
1503 * barrier with the exchange in cds_wfq_enqueue.
1504 */
1505 futex_nto1_wake(&ust_cmd_queue.futex);
1506 }
1507 }
1508 }
1509 }
1510
1511 exit:
1512 error:
1513 if (err) {
1514 health_error(&health_thread_app_reg);
1515 ERR("Health error occurred in %s", __func__);
1516 }
1517 health_exit(&health_thread_app_reg);
1518
1519 /* Notify that the registration thread is gone */
1520 notify_ust_apps(0);
1521
1522 if (apps_sock >= 0) {
1523 ret = close(apps_sock);
1524 if (ret) {
1525 PERROR("close");
1526 }
1527 }
1528 if (sock >= 0) {
1529 ret = close(sock);
1530 if (ret) {
1531 PERROR("close");
1532 }
1533 lttng_fd_put(LTTNG_FD_APPS, 1);
1534 }
1535 unlink(apps_unix_sock_path);
1536
1537 error_poll_add:
1538 lttng_poll_clean(&events);
1539 error_listen:
1540 error_create_poll:
1541 DBG("UST Registration thread cleanup complete");
1542
1543 return NULL;
1544 }
1545
1546 /*
1547 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1548 * exec or it will fails.
1549 */
1550 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1551 {
1552 int ret, clock_ret;
1553 struct timespec timeout;
1554
1555 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1556 consumer_data->consumer_thread_is_ready = 0;
1557
1558 /* Setup pthread condition */
1559 ret = pthread_condattr_init(&consumer_data->condattr);
1560 if (ret != 0) {
1561 errno = ret;
1562 PERROR("pthread_condattr_init consumer data");
1563 goto error;
1564 }
1565
1566 /*
1567 * Set the monotonic clock in order to make sure we DO NOT jump in time
1568 * between the clock_gettime() call and the timedwait call. See bug #324
1569 * for a more details and how we noticed it.
1570 */
1571 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1572 if (ret != 0) {
1573 errno = ret;
1574 PERROR("pthread_condattr_setclock consumer data");
1575 goto error;
1576 }
1577
1578 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1579 if (ret != 0) {
1580 errno = ret;
1581 PERROR("pthread_cond_init consumer data");
1582 goto error;
1583 }
1584
1585 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1586 consumer_data);
1587 if (ret != 0) {
1588 PERROR("pthread_create consumer");
1589 ret = -1;
1590 goto error;
1591 }
1592
1593 /* We are about to wait on a pthread condition */
1594 pthread_mutex_lock(&consumer_data->cond_mutex);
1595
1596 /* Get time for sem_timedwait absolute timeout */
1597 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1598 /*
1599 * Set the timeout for the condition timed wait even if the clock gettime
1600 * call fails since we might loop on that call and we want to avoid to
1601 * increment the timeout too many times.
1602 */
1603 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1604
1605 /*
1606 * The following loop COULD be skipped in some conditions so this is why we
1607 * set ret to 0 in order to make sure at least one round of the loop is
1608 * done.
1609 */
1610 ret = 0;
1611
1612 /*
1613 * Loop until the condition is reached or when a timeout is reached. Note
1614 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1615 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1616 * possible. This loop does not take any chances and works with both of
1617 * them.
1618 */
1619 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1620 if (clock_ret < 0) {
1621 PERROR("clock_gettime spawn consumer");
1622 /* Infinite wait for the consumerd thread to be ready */
1623 ret = pthread_cond_wait(&consumer_data->cond,
1624 &consumer_data->cond_mutex);
1625 } else {
1626 ret = pthread_cond_timedwait(&consumer_data->cond,
1627 &consumer_data->cond_mutex, &timeout);
1628 }
1629 }
1630
1631 /* Release the pthread condition */
1632 pthread_mutex_unlock(&consumer_data->cond_mutex);
1633
1634 if (ret != 0) {
1635 errno = ret;
1636 if (ret == ETIMEDOUT) {
1637 /*
1638 * Call has timed out so we kill the kconsumerd_thread and return
1639 * an error.
1640 */
1641 ERR("Condition timed out. The consumer thread was never ready."
1642 " Killing it");
1643 ret = pthread_cancel(consumer_data->thread);
1644 if (ret < 0) {
1645 PERROR("pthread_cancel consumer thread");
1646 }
1647 } else {
1648 PERROR("pthread_cond_wait failed consumer thread");
1649 }
1650 goto error;
1651 }
1652
1653 pthread_mutex_lock(&consumer_data->pid_mutex);
1654 if (consumer_data->pid == 0) {
1655 ERR("Consumerd did not start");
1656 pthread_mutex_unlock(&consumer_data->pid_mutex);
1657 goto error;
1658 }
1659 pthread_mutex_unlock(&consumer_data->pid_mutex);
1660
1661 return 0;
1662
1663 error:
1664 return ret;
1665 }
1666
1667 /*
1668 * Join consumer thread
1669 */
1670 static int join_consumer_thread(struct consumer_data *consumer_data)
1671 {
1672 void *status;
1673 int ret;
1674
1675 /* Consumer pid must be a real one. */
1676 if (consumer_data->pid > 0) {
1677 ret = kill(consumer_data->pid, SIGTERM);
1678 if (ret) {
1679 ERR("Error killing consumer daemon");
1680 return ret;
1681 }
1682 return pthread_join(consumer_data->thread, &status);
1683 } else {
1684 return 0;
1685 }
1686 }
1687
1688 /*
1689 * Fork and exec a consumer daemon (consumerd).
1690 *
1691 * Return pid if successful else -1.
1692 */
1693 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1694 {
1695 int ret;
1696 pid_t pid;
1697 const char *consumer_to_use;
1698 const char *verbosity;
1699 struct stat st;
1700
1701 DBG("Spawning consumerd");
1702
1703 pid = fork();
1704 if (pid == 0) {
1705 /*
1706 * Exec consumerd.
1707 */
1708 if (opt_verbose_consumer) {
1709 verbosity = "--verbose";
1710 } else {
1711 verbosity = "--quiet";
1712 }
1713 switch (consumer_data->type) {
1714 case LTTNG_CONSUMER_KERNEL:
1715 /*
1716 * Find out which consumerd to execute. We will first try the
1717 * 64-bit path, then the sessiond's installation directory, and
1718 * fallback on the 32-bit one,
1719 */
1720 DBG3("Looking for a kernel consumer at these locations:");
1721 DBG3(" 1) %s", consumerd64_bin);
1722 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1723 DBG3(" 3) %s", consumerd32_bin);
1724 if (stat(consumerd64_bin, &st) == 0) {
1725 DBG3("Found location #1");
1726 consumer_to_use = consumerd64_bin;
1727 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1728 DBG3("Found location #2");
1729 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1730 } else if (stat(consumerd32_bin, &st) == 0) {
1731 DBG3("Found location #3");
1732 consumer_to_use = consumerd32_bin;
1733 } else {
1734 DBG("Could not find any valid consumerd executable");
1735 break;
1736 }
1737 DBG("Using kernel consumer at: %s", consumer_to_use);
1738 execl(consumer_to_use,
1739 "lttng-consumerd", verbosity, "-k",
1740 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1741 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1742 NULL);
1743 break;
1744 case LTTNG_CONSUMER64_UST:
1745 {
1746 char *tmpnew = NULL;
1747
1748 if (consumerd64_libdir[0] != '\0') {
1749 char *tmp;
1750 size_t tmplen;
1751
1752 tmp = getenv("LD_LIBRARY_PATH");
1753 if (!tmp) {
1754 tmp = "";
1755 }
1756 tmplen = strlen("LD_LIBRARY_PATH=")
1757 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1758 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1759 if (!tmpnew) {
1760 ret = -ENOMEM;
1761 goto error;
1762 }
1763 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1764 strcat(tmpnew, consumerd64_libdir);
1765 if (tmp[0] != '\0') {
1766 strcat(tmpnew, ":");
1767 strcat(tmpnew, tmp);
1768 }
1769 ret = putenv(tmpnew);
1770 if (ret) {
1771 ret = -errno;
1772 goto error;
1773 }
1774 }
1775 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1776 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1777 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1778 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1779 NULL);
1780 if (consumerd64_libdir[0] != '\0') {
1781 free(tmpnew);
1782 }
1783 if (ret) {
1784 goto error;
1785 }
1786 break;
1787 }
1788 case LTTNG_CONSUMER32_UST:
1789 {
1790 char *tmpnew = NULL;
1791
1792 if (consumerd32_libdir[0] != '\0') {
1793 char *tmp;
1794 size_t tmplen;
1795
1796 tmp = getenv("LD_LIBRARY_PATH");
1797 if (!tmp) {
1798 tmp = "";
1799 }
1800 tmplen = strlen("LD_LIBRARY_PATH=")
1801 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1802 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1803 if (!tmpnew) {
1804 ret = -ENOMEM;
1805 goto error;
1806 }
1807 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1808 strcat(tmpnew, consumerd32_libdir);
1809 if (tmp[0] != '\0') {
1810 strcat(tmpnew, ":");
1811 strcat(tmpnew, tmp);
1812 }
1813 ret = putenv(tmpnew);
1814 if (ret) {
1815 ret = -errno;
1816 goto error;
1817 }
1818 }
1819 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1820 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1821 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1822 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1823 NULL);
1824 if (consumerd32_libdir[0] != '\0') {
1825 free(tmpnew);
1826 }
1827 if (ret) {
1828 goto error;
1829 }
1830 break;
1831 }
1832 default:
1833 PERROR("unknown consumer type");
1834 exit(EXIT_FAILURE);
1835 }
1836 if (errno != 0) {
1837 PERROR("kernel start consumer exec");
1838 }
1839 exit(EXIT_FAILURE);
1840 } else if (pid > 0) {
1841 ret = pid;
1842 } else {
1843 PERROR("start consumer fork");
1844 ret = -errno;
1845 }
1846 error:
1847 return ret;
1848 }
1849
1850 /*
1851 * Spawn the consumerd daemon and session daemon thread.
1852 */
1853 static int start_consumerd(struct consumer_data *consumer_data)
1854 {
1855 int ret, err;
1856
1857 /*
1858 * Set the listen() state on the socket since there is a possible race
1859 * between the exec() of the consumer daemon and this call if place in the
1860 * consumer thread. See bug #366 for more details.
1861 */
1862 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1863 if (ret < 0) {
1864 goto error;
1865 }
1866
1867 pthread_mutex_lock(&consumer_data->pid_mutex);
1868 if (consumer_data->pid != 0) {
1869 pthread_mutex_unlock(&consumer_data->pid_mutex);
1870 goto end;
1871 }
1872
1873 ret = spawn_consumerd(consumer_data);
1874 if (ret < 0) {
1875 ERR("Spawning consumerd failed");
1876 pthread_mutex_unlock(&consumer_data->pid_mutex);
1877 goto error;
1878 }
1879
1880 /* Setting up the consumer_data pid */
1881 consumer_data->pid = ret;
1882 DBG2("Consumer pid %d", consumer_data->pid);
1883 pthread_mutex_unlock(&consumer_data->pid_mutex);
1884
1885 DBG2("Spawning consumer control thread");
1886 ret = spawn_consumer_thread(consumer_data);
1887 if (ret < 0) {
1888 ERR("Fatal error spawning consumer control thread");
1889 goto error;
1890 }
1891
1892 end:
1893 return 0;
1894
1895 error:
1896 /* Cleanup already created socket on error. */
1897 if (consumer_data->err_sock >= 0) {
1898 err = close(consumer_data->err_sock);
1899 if (err < 0) {
1900 PERROR("close consumer data error socket");
1901 }
1902 }
1903 return ret;
1904 }
1905
1906 /*
1907 * Compute health status of each consumer. If one of them is zero (bad
1908 * state), we return 0.
1909 */
1910 static int check_consumer_health(void)
1911 {
1912 int ret;
1913
1914 ret = health_check_state(&kconsumer_data.health) &&
1915 health_check_state(&ustconsumer32_data.health) &&
1916 health_check_state(&ustconsumer64_data.health);
1917
1918 DBG3("Health consumer check %d", ret);
1919
1920 return ret;
1921 }
1922
1923 /*
1924 * Setup necessary data for kernel tracer action.
1925 */
1926 static int init_kernel_tracer(void)
1927 {
1928 int ret;
1929
1930 /* Modprobe lttng kernel modules */
1931 ret = modprobe_lttng_control();
1932 if (ret < 0) {
1933 goto error;
1934 }
1935
1936 /* Open debugfs lttng */
1937 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1938 if (kernel_tracer_fd < 0) {
1939 DBG("Failed to open %s", module_proc_lttng);
1940 ret = -1;
1941 goto error_open;
1942 }
1943
1944 /* Validate kernel version */
1945 ret = kernel_validate_version(kernel_tracer_fd);
1946 if (ret < 0) {
1947 goto error_version;
1948 }
1949
1950 ret = modprobe_lttng_data();
1951 if (ret < 0) {
1952 goto error_modules;
1953 }
1954
1955 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1956 return 0;
1957
1958 error_version:
1959 modprobe_remove_lttng_control();
1960 ret = close(kernel_tracer_fd);
1961 if (ret) {
1962 PERROR("close");
1963 }
1964 kernel_tracer_fd = -1;
1965 return LTTNG_ERR_KERN_VERSION;
1966
1967 error_modules:
1968 ret = close(kernel_tracer_fd);
1969 if (ret) {
1970 PERROR("close");
1971 }
1972
1973 error_open:
1974 modprobe_remove_lttng_control();
1975
1976 error:
1977 WARN("No kernel tracer available");
1978 kernel_tracer_fd = -1;
1979 if (!is_root) {
1980 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1981 } else {
1982 return LTTNG_ERR_KERN_NA;
1983 }
1984 }
1985
1986
1987 /*
1988 * Copy consumer output from the tracing session to the domain session. The
1989 * function also applies the right modification on a per domain basis for the
1990 * trace files destination directory.
1991 */
1992 static int copy_session_consumer(int domain, struct ltt_session *session)
1993 {
1994 int ret;
1995 const char *dir_name;
1996 struct consumer_output *consumer;
1997
1998 assert(session);
1999 assert(session->consumer);
2000
2001 switch (domain) {
2002 case LTTNG_DOMAIN_KERNEL:
2003 DBG3("Copying tracing session consumer output in kernel session");
2004 /*
2005 * XXX: We should audit the session creation and what this function
2006 * does "extra" in order to avoid a destroy since this function is used
2007 * in the domain session creation (kernel and ust) only. Same for UST
2008 * domain.
2009 */
2010 if (session->kernel_session->consumer) {
2011 consumer_destroy_output(session->kernel_session->consumer);
2012 }
2013 session->kernel_session->consumer =
2014 consumer_copy_output(session->consumer);
2015 /* Ease our life a bit for the next part */
2016 consumer = session->kernel_session->consumer;
2017 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2018 break;
2019 case LTTNG_DOMAIN_UST:
2020 DBG3("Copying tracing session consumer output in UST session");
2021 if (session->ust_session->consumer) {
2022 consumer_destroy_output(session->ust_session->consumer);
2023 }
2024 session->ust_session->consumer =
2025 consumer_copy_output(session->consumer);
2026 /* Ease our life a bit for the next part */
2027 consumer = session->ust_session->consumer;
2028 dir_name = DEFAULT_UST_TRACE_DIR;
2029 break;
2030 default:
2031 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2032 goto error;
2033 }
2034
2035 /* Append correct directory to subdir */
2036 strncat(consumer->subdir, dir_name,
2037 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2038 DBG3("Copy session consumer subdir %s", consumer->subdir);
2039
2040 ret = LTTNG_OK;
2041
2042 error:
2043 return ret;
2044 }
2045
2046 /*
2047 * Create an UST session and add it to the session ust list.
2048 */
2049 static int create_ust_session(struct ltt_session *session,
2050 struct lttng_domain *domain)
2051 {
2052 int ret;
2053 struct ltt_ust_session *lus = NULL;
2054
2055 assert(session);
2056 assert(domain);
2057 assert(session->consumer);
2058
2059 switch (domain->type) {
2060 case LTTNG_DOMAIN_UST:
2061 break;
2062 default:
2063 ERR("Unknown UST domain on create session %d", domain->type);
2064 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2065 goto error;
2066 }
2067
2068 DBG("Creating UST session");
2069
2070 lus = trace_ust_create_session(session->path, session->id, domain);
2071 if (lus == NULL) {
2072 ret = LTTNG_ERR_UST_SESS_FAIL;
2073 goto error;
2074 }
2075
2076 lus->uid = session->uid;
2077 lus->gid = session->gid;
2078 session->ust_session = lus;
2079
2080 /* Copy session output to the newly created UST session */
2081 ret = copy_session_consumer(domain->type, session);
2082 if (ret != LTTNG_OK) {
2083 goto error;
2084 }
2085
2086 return LTTNG_OK;
2087
2088 error:
2089 free(lus);
2090 session->ust_session = NULL;
2091 return ret;
2092 }
2093
2094 /*
2095 * Create a kernel tracer session then create the default channel.
2096 */
2097 static int create_kernel_session(struct ltt_session *session)
2098 {
2099 int ret;
2100
2101 DBG("Creating kernel session");
2102
2103 ret = kernel_create_session(session, kernel_tracer_fd);
2104 if (ret < 0) {
2105 ret = LTTNG_ERR_KERN_SESS_FAIL;
2106 goto error;
2107 }
2108
2109 /* Code flow safety */
2110 assert(session->kernel_session);
2111
2112 /* Copy session output to the newly created Kernel session */
2113 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2114 if (ret != LTTNG_OK) {
2115 goto error;
2116 }
2117
2118 /* Create directory(ies) on local filesystem. */
2119 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2120 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2121 ret = run_as_mkdir_recursive(
2122 session->kernel_session->consumer->dst.trace_path,
2123 S_IRWXU | S_IRWXG, session->uid, session->gid);
2124 if (ret < 0) {
2125 if (ret != -EEXIST) {
2126 ERR("Trace directory creation error");
2127 goto error;
2128 }
2129 }
2130 }
2131
2132 session->kernel_session->uid = session->uid;
2133 session->kernel_session->gid = session->gid;
2134
2135 return LTTNG_OK;
2136
2137 error:
2138 trace_kernel_destroy_session(session->kernel_session);
2139 session->kernel_session = NULL;
2140 return ret;
2141 }
2142
2143 /*
2144 * Count number of session permitted by uid/gid.
2145 */
2146 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2147 {
2148 unsigned int i = 0;
2149 struct ltt_session *session;
2150
2151 DBG("Counting number of available session for UID %d GID %d",
2152 uid, gid);
2153 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2154 /*
2155 * Only list the sessions the user can control.
2156 */
2157 if (!session_access_ok(session, uid, gid)) {
2158 continue;
2159 }
2160 i++;
2161 }
2162 return i;
2163 }
2164
2165 /*
2166 * Process the command requested by the lttng client within the command
2167 * context structure. This function make sure that the return structure (llm)
2168 * is set and ready for transmission before returning.
2169 *
2170 * Return any error encountered or 0 for success.
2171 *
2172 * "sock" is only used for special-case var. len data.
2173 */
2174 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2175 int *sock_error)
2176 {
2177 int ret = LTTNG_OK;
2178 int need_tracing_session = 1;
2179 int need_domain;
2180
2181 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2182
2183 *sock_error = 0;
2184
2185 switch (cmd_ctx->lsm->cmd_type) {
2186 case LTTNG_CREATE_SESSION:
2187 case LTTNG_DESTROY_SESSION:
2188 case LTTNG_LIST_SESSIONS:
2189 case LTTNG_LIST_DOMAINS:
2190 case LTTNG_START_TRACE:
2191 case LTTNG_STOP_TRACE:
2192 case LTTNG_DATA_PENDING:
2193 need_domain = 0;
2194 break;
2195 default:
2196 need_domain = 1;
2197 }
2198
2199 if (opt_no_kernel && need_domain
2200 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2201 if (!is_root) {
2202 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2203 } else {
2204 ret = LTTNG_ERR_KERN_NA;
2205 }
2206 goto error;
2207 }
2208
2209 /* Deny register consumer if we already have a spawned consumer. */
2210 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2211 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2212 if (kconsumer_data.pid > 0) {
2213 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2214 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2215 goto error;
2216 }
2217 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2218 }
2219
2220 /*
2221 * Check for command that don't needs to allocate a returned payload. We do
2222 * this here so we don't have to make the call for no payload at each
2223 * command.
2224 */
2225 switch(cmd_ctx->lsm->cmd_type) {
2226 case LTTNG_LIST_SESSIONS:
2227 case LTTNG_LIST_TRACEPOINTS:
2228 case LTTNG_LIST_TRACEPOINT_FIELDS:
2229 case LTTNG_LIST_DOMAINS:
2230 case LTTNG_LIST_CHANNELS:
2231 case LTTNG_LIST_EVENTS:
2232 break;
2233 default:
2234 /* Setup lttng message with no payload */
2235 ret = setup_lttng_msg(cmd_ctx, 0);
2236 if (ret < 0) {
2237 /* This label does not try to unlock the session */
2238 goto init_setup_error;
2239 }
2240 }
2241
2242 /* Commands that DO NOT need a session. */
2243 switch (cmd_ctx->lsm->cmd_type) {
2244 case LTTNG_CREATE_SESSION:
2245 case LTTNG_CALIBRATE:
2246 case LTTNG_LIST_SESSIONS:
2247 case LTTNG_LIST_TRACEPOINTS:
2248 case LTTNG_LIST_TRACEPOINT_FIELDS:
2249 need_tracing_session = 0;
2250 break;
2251 default:
2252 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2253 /*
2254 * We keep the session list lock across _all_ commands
2255 * for now, because the per-session lock does not
2256 * handle teardown properly.
2257 */
2258 session_lock_list();
2259 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2260 if (cmd_ctx->session == NULL) {
2261 if (cmd_ctx->lsm->session.name != NULL) {
2262 ret = LTTNG_ERR_SESS_NOT_FOUND;
2263 } else {
2264 /* If no session name specified */
2265 ret = LTTNG_ERR_SELECT_SESS;
2266 }
2267 goto error;
2268 } else {
2269 /* Acquire lock for the session */
2270 session_lock(cmd_ctx->session);
2271 }
2272 break;
2273 }
2274
2275 if (!need_domain) {
2276 goto skip_domain;
2277 }
2278
2279 /*
2280 * Check domain type for specific "pre-action".
2281 */
2282 switch (cmd_ctx->lsm->domain.type) {
2283 case LTTNG_DOMAIN_KERNEL:
2284 if (!is_root) {
2285 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2286 goto error;
2287 }
2288
2289 /* Kernel tracer check */
2290 if (kernel_tracer_fd == -1) {
2291 /* Basically, load kernel tracer modules */
2292 ret = init_kernel_tracer();
2293 if (ret != 0) {
2294 goto error;
2295 }
2296 }
2297
2298 /* Consumer is in an ERROR state. Report back to client */
2299 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2300 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2301 goto error;
2302 }
2303
2304 /* Need a session for kernel command */
2305 if (need_tracing_session) {
2306 if (cmd_ctx->session->kernel_session == NULL) {
2307 ret = create_kernel_session(cmd_ctx->session);
2308 if (ret < 0) {
2309 ret = LTTNG_ERR_KERN_SESS_FAIL;
2310 goto error;
2311 }
2312 }
2313
2314 /* Start the kernel consumer daemon */
2315 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2316 if (kconsumer_data.pid == 0 &&
2317 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2318 cmd_ctx->session->start_consumer) {
2319 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2320 ret = start_consumerd(&kconsumer_data);
2321 if (ret < 0) {
2322 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2323 goto error;
2324 }
2325 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2326 } else {
2327 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2328 }
2329
2330 /*
2331 * The consumer was just spawned so we need to add the socket to
2332 * the consumer output of the session if exist.
2333 */
2334 ret = consumer_create_socket(&kconsumer_data,
2335 cmd_ctx->session->kernel_session->consumer);
2336 if (ret < 0) {
2337 goto error;
2338 }
2339 }
2340
2341 break;
2342 case LTTNG_DOMAIN_UST:
2343 {
2344 /* Consumer is in an ERROR state. Report back to client */
2345 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2346 ret = LTTNG_ERR_NO_USTCONSUMERD;
2347 goto error;
2348 }
2349
2350 if (need_tracing_session) {
2351 /* Create UST session if none exist. */
2352 if (cmd_ctx->session->ust_session == NULL) {
2353 ret = create_ust_session(cmd_ctx->session,
2354 &cmd_ctx->lsm->domain);
2355 if (ret != LTTNG_OK) {
2356 goto error;
2357 }
2358 }
2359
2360 /* Start the UST consumer daemons */
2361 /* 64-bit */
2362 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2363 if (consumerd64_bin[0] != '\0' &&
2364 ustconsumer64_data.pid == 0 &&
2365 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2366 cmd_ctx->session->start_consumer) {
2367 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2368 ret = start_consumerd(&ustconsumer64_data);
2369 if (ret < 0) {
2370 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2371 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2372 goto error;
2373 }
2374
2375 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2376 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2377 } else {
2378 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2379 }
2380
2381 /*
2382 * Setup socket for consumer 64 bit. No need for atomic access
2383 * since it was set above and can ONLY be set in this thread.
2384 */
2385 ret = consumer_create_socket(&ustconsumer64_data,
2386 cmd_ctx->session->ust_session->consumer);
2387 if (ret < 0) {
2388 goto error;
2389 }
2390
2391 /* 32-bit */
2392 if (consumerd32_bin[0] != '\0' &&
2393 ustconsumer32_data.pid == 0 &&
2394 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2395 cmd_ctx->session->start_consumer) {
2396 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2397 ret = start_consumerd(&ustconsumer32_data);
2398 if (ret < 0) {
2399 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2400 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2401 goto error;
2402 }
2403
2404 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2405 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2406 } else {
2407 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2408 }
2409
2410 /*
2411 * Setup socket for consumer 64 bit. No need for atomic access
2412 * since it was set above and can ONLY be set in this thread.
2413 */
2414 ret = consumer_create_socket(&ustconsumer32_data,
2415 cmd_ctx->session->ust_session->consumer);
2416 if (ret < 0) {
2417 goto error;
2418 }
2419 }
2420 break;
2421 }
2422 default:
2423 break;
2424 }
2425 skip_domain:
2426
2427 /* Validate consumer daemon state when start/stop trace command */
2428 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2429 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2430 switch (cmd_ctx->lsm->domain.type) {
2431 case LTTNG_DOMAIN_UST:
2432 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2433 ret = LTTNG_ERR_NO_USTCONSUMERD;
2434 goto error;
2435 }
2436 break;
2437 case LTTNG_DOMAIN_KERNEL:
2438 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2439 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2440 goto error;
2441 }
2442 break;
2443 }
2444 }
2445
2446 /*
2447 * Check that the UID or GID match that of the tracing session.
2448 * The root user can interact with all sessions.
2449 */
2450 if (need_tracing_session) {
2451 if (!session_access_ok(cmd_ctx->session,
2452 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2453 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2454 ret = LTTNG_ERR_EPERM;
2455 goto error;
2456 }
2457 }
2458
2459 /* Process by command type */
2460 switch (cmd_ctx->lsm->cmd_type) {
2461 case LTTNG_ADD_CONTEXT:
2462 {
2463 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2464 cmd_ctx->lsm->u.context.channel_name,
2465 cmd_ctx->lsm->u.context.event_name,
2466 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2467 break;
2468 }
2469 case LTTNG_DISABLE_CHANNEL:
2470 {
2471 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2472 cmd_ctx->lsm->u.disable.channel_name);
2473 break;
2474 }
2475 case LTTNG_DISABLE_EVENT:
2476 {
2477 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2478 cmd_ctx->lsm->u.disable.channel_name,
2479 cmd_ctx->lsm->u.disable.name);
2480 break;
2481 }
2482 case LTTNG_DISABLE_ALL_EVENT:
2483 {
2484 DBG("Disabling all events");
2485
2486 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2487 cmd_ctx->lsm->u.disable.channel_name);
2488 break;
2489 }
2490 case LTTNG_DISABLE_CONSUMER:
2491 {
2492 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2493 break;
2494 }
2495 case LTTNG_ENABLE_CHANNEL:
2496 {
2497 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2498 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2499 break;
2500 }
2501 case LTTNG_ENABLE_CONSUMER:
2502 {
2503 /*
2504 * XXX: 0 means that this URI should be applied on the session. Should
2505 * be a DOMAIN enuam.
2506 */
2507 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2508 if (ret != LTTNG_OK) {
2509 goto error;
2510 }
2511
2512 if (cmd_ctx->lsm->domain.type == 0) {
2513 /* Add the URI for the UST session if a consumer is present. */
2514 if (cmd_ctx->session->ust_session &&
2515 cmd_ctx->session->ust_session->consumer) {
2516 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2517 } else if (cmd_ctx->session->kernel_session &&
2518 cmd_ctx->session->kernel_session->consumer) {
2519 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2520 cmd_ctx->session);
2521 }
2522 }
2523 break;
2524 }
2525 case LTTNG_ENABLE_EVENT:
2526 {
2527 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2528 cmd_ctx->lsm->u.enable.channel_name,
2529 &cmd_ctx->lsm->u.enable.event, kernel_poll_pipe[1]);
2530 break;
2531 }
2532 case LTTNG_ENABLE_ALL_EVENT:
2533 {
2534 DBG("Enabling all events");
2535
2536 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2537 cmd_ctx->lsm->u.enable.channel_name,
2538 cmd_ctx->lsm->u.enable.event.type, kernel_poll_pipe[1]);
2539 break;
2540 }
2541 case LTTNG_LIST_TRACEPOINTS:
2542 {
2543 struct lttng_event *events;
2544 ssize_t nb_events;
2545
2546 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2547 if (nb_events < 0) {
2548 /* Return value is a negative lttng_error_code. */
2549 ret = -nb_events;
2550 goto error;
2551 }
2552
2553 /*
2554 * Setup lttng message with payload size set to the event list size in
2555 * bytes and then copy list into the llm payload.
2556 */
2557 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2558 if (ret < 0) {
2559 free(events);
2560 goto setup_error;
2561 }
2562
2563 /* Copy event list into message payload */
2564 memcpy(cmd_ctx->llm->payload, events,
2565 sizeof(struct lttng_event) * nb_events);
2566
2567 free(events);
2568
2569 ret = LTTNG_OK;
2570 break;
2571 }
2572 case LTTNG_LIST_TRACEPOINT_FIELDS:
2573 {
2574 struct lttng_event_field *fields;
2575 ssize_t nb_fields;
2576
2577 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2578 &fields);
2579 if (nb_fields < 0) {
2580 /* Return value is a negative lttng_error_code. */
2581 ret = -nb_fields;
2582 goto error;
2583 }
2584
2585 /*
2586 * Setup lttng message with payload size set to the event list size in
2587 * bytes and then copy list into the llm payload.
2588 */
2589 ret = setup_lttng_msg(cmd_ctx,
2590 sizeof(struct lttng_event_field) * nb_fields);
2591 if (ret < 0) {
2592 free(fields);
2593 goto setup_error;
2594 }
2595
2596 /* Copy event list into message payload */
2597 memcpy(cmd_ctx->llm->payload, fields,
2598 sizeof(struct lttng_event_field) * nb_fields);
2599
2600 free(fields);
2601
2602 ret = LTTNG_OK;
2603 break;
2604 }
2605 case LTTNG_SET_CONSUMER_URI:
2606 {
2607 size_t nb_uri, len;
2608 struct lttng_uri *uris;
2609
2610 nb_uri = cmd_ctx->lsm->u.uri.size;
2611 len = nb_uri * sizeof(struct lttng_uri);
2612
2613 if (nb_uri == 0) {
2614 ret = LTTNG_ERR_INVALID;
2615 goto error;
2616 }
2617
2618 uris = zmalloc(len);
2619 if (uris == NULL) {
2620 ret = LTTNG_ERR_FATAL;
2621 goto error;
2622 }
2623
2624 /* Receive variable len data */
2625 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2626 ret = lttcomm_recv_unix_sock(sock, uris, len);
2627 if (ret <= 0) {
2628 DBG("No URIs received from client... continuing");
2629 *sock_error = 1;
2630 ret = LTTNG_ERR_SESSION_FAIL;
2631 free(uris);
2632 goto error;
2633 }
2634
2635 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2636 nb_uri, uris);
2637 if (ret != LTTNG_OK) {
2638 free(uris);
2639 goto error;
2640 }
2641
2642 /*
2643 * XXX: 0 means that this URI should be applied on the session. Should
2644 * be a DOMAIN enuam.
2645 */
2646 if (cmd_ctx->lsm->domain.type == 0) {
2647 /* Add the URI for the UST session if a consumer is present. */
2648 if (cmd_ctx->session->ust_session &&
2649 cmd_ctx->session->ust_session->consumer) {
2650 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2651 nb_uri, uris);
2652 } else if (cmd_ctx->session->kernel_session &&
2653 cmd_ctx->session->kernel_session->consumer) {
2654 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2655 cmd_ctx->session, nb_uri, uris);
2656 }
2657 }
2658
2659 free(uris);
2660
2661 break;
2662 }
2663 case LTTNG_START_TRACE:
2664 {
2665 ret = cmd_start_trace(cmd_ctx->session);
2666 break;
2667 }
2668 case LTTNG_STOP_TRACE:
2669 {
2670 ret = cmd_stop_trace(cmd_ctx->session);
2671 break;
2672 }
2673 case LTTNG_CREATE_SESSION:
2674 {
2675 size_t nb_uri, len;
2676 struct lttng_uri *uris = NULL;
2677
2678 nb_uri = cmd_ctx->lsm->u.uri.size;
2679 len = nb_uri * sizeof(struct lttng_uri);
2680
2681 if (nb_uri > 0) {
2682 uris = zmalloc(len);
2683 if (uris == NULL) {
2684 ret = LTTNG_ERR_FATAL;
2685 goto error;
2686 }
2687
2688 /* Receive variable len data */
2689 DBG("Waiting for %zu URIs from client ...", nb_uri);
2690 ret = lttcomm_recv_unix_sock(sock, uris, len);
2691 if (ret <= 0) {
2692 DBG("No URIs received from client... continuing");
2693 *sock_error = 1;
2694 ret = LTTNG_ERR_SESSION_FAIL;
2695 free(uris);
2696 goto error;
2697 }
2698
2699 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2700 DBG("Creating session with ONE network URI is a bad call");
2701 ret = LTTNG_ERR_SESSION_FAIL;
2702 free(uris);
2703 goto error;
2704 }
2705 }
2706
2707 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2708 &cmd_ctx->creds);
2709
2710 free(uris);
2711
2712 break;
2713 }
2714 case LTTNG_DESTROY_SESSION:
2715 {
2716 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2717
2718 /* Set session to NULL so we do not unlock it after free. */
2719 cmd_ctx->session = NULL;
2720 break;
2721 }
2722 case LTTNG_LIST_DOMAINS:
2723 {
2724 ssize_t nb_dom;
2725 struct lttng_domain *domains;
2726
2727 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2728 if (nb_dom < 0) {
2729 /* Return value is a negative lttng_error_code. */
2730 ret = -nb_dom;
2731 goto error;
2732 }
2733
2734 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2735 if (ret < 0) {
2736 goto setup_error;
2737 }
2738
2739 /* Copy event list into message payload */
2740 memcpy(cmd_ctx->llm->payload, domains,
2741 nb_dom * sizeof(struct lttng_domain));
2742
2743 free(domains);
2744
2745 ret = LTTNG_OK;
2746 break;
2747 }
2748 case LTTNG_LIST_CHANNELS:
2749 {
2750 int nb_chan;
2751 struct lttng_channel *channels;
2752
2753 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2754 cmd_ctx->session, &channels);
2755 if (nb_chan < 0) {
2756 /* Return value is a negative lttng_error_code. */
2757 ret = -nb_chan;
2758 goto error;
2759 }
2760
2761 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2762 if (ret < 0) {
2763 goto setup_error;
2764 }
2765
2766 /* Copy event list into message payload */
2767 memcpy(cmd_ctx->llm->payload, channels,
2768 nb_chan * sizeof(struct lttng_channel));
2769
2770 free(channels);
2771
2772 ret = LTTNG_OK;
2773 break;
2774 }
2775 case LTTNG_LIST_EVENTS:
2776 {
2777 ssize_t nb_event;
2778 struct lttng_event *events = NULL;
2779
2780 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2781 cmd_ctx->lsm->u.list.channel_name, &events);
2782 if (nb_event < 0) {
2783 /* Return value is a negative lttng_error_code. */
2784 ret = -nb_event;
2785 goto error;
2786 }
2787
2788 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2789 if (ret < 0) {
2790 goto setup_error;
2791 }
2792
2793 /* Copy event list into message payload */
2794 memcpy(cmd_ctx->llm->payload, events,
2795 nb_event * sizeof(struct lttng_event));
2796
2797 free(events);
2798
2799 ret = LTTNG_OK;
2800 break;
2801 }
2802 case LTTNG_LIST_SESSIONS:
2803 {
2804 unsigned int nr_sessions;
2805
2806 session_lock_list();
2807 nr_sessions = lttng_sessions_count(
2808 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2809 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2810
2811 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2812 if (ret < 0) {
2813 session_unlock_list();
2814 goto setup_error;
2815 }
2816
2817 /* Filled the session array */
2818 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2819 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2820 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2821
2822 session_unlock_list();
2823
2824 ret = LTTNG_OK;
2825 break;
2826 }
2827 case LTTNG_CALIBRATE:
2828 {
2829 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2830 &cmd_ctx->lsm->u.calibrate);
2831 break;
2832 }
2833 case LTTNG_REGISTER_CONSUMER:
2834 {
2835 struct consumer_data *cdata;
2836
2837 switch (cmd_ctx->lsm->domain.type) {
2838 case LTTNG_DOMAIN_KERNEL:
2839 cdata = &kconsumer_data;
2840 break;
2841 default:
2842 ret = LTTNG_ERR_UND;
2843 goto error;
2844 }
2845
2846 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2847 cmd_ctx->lsm->u.reg.path, cdata);
2848 break;
2849 }
2850 case LTTNG_SET_FILTER:
2851 {
2852 struct lttng_filter_bytecode *bytecode;
2853
2854 if (cmd_ctx->lsm->u.filter.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2855 ret = LTTNG_ERR_FILTER_INVAL;
2856 goto error;
2857 }
2858 bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len);
2859 if (!bytecode) {
2860 ret = LTTNG_ERR_FILTER_NOMEM;
2861 goto error;
2862 }
2863 /* Receive var. len. data */
2864 DBG("Receiving var len data from client ...");
2865 ret = lttcomm_recv_unix_sock(sock, bytecode,
2866 cmd_ctx->lsm->u.filter.bytecode_len);
2867 if (ret <= 0) {
2868 DBG("Nothing recv() from client var len data... continuing");
2869 *sock_error = 1;
2870 ret = LTTNG_ERR_FILTER_INVAL;
2871 goto error;
2872 }
2873
2874 if (bytecode->len + sizeof(*bytecode)
2875 != cmd_ctx->lsm->u.filter.bytecode_len) {
2876 free(bytecode);
2877 ret = LTTNG_ERR_FILTER_INVAL;
2878 goto error;
2879 }
2880
2881 ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2882 cmd_ctx->lsm->u.filter.channel_name,
2883 cmd_ctx->lsm->u.filter.event_name,
2884 bytecode);
2885 break;
2886 }
2887 case LTTNG_DATA_PENDING:
2888 {
2889 ret = cmd_data_pending(cmd_ctx->session);
2890 break;
2891 }
2892 default:
2893 ret = LTTNG_ERR_UND;
2894 break;
2895 }
2896
2897 error:
2898 if (cmd_ctx->llm == NULL) {
2899 DBG("Missing llm structure. Allocating one.");
2900 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2901 goto setup_error;
2902 }
2903 }
2904 /* Set return code */
2905 cmd_ctx->llm->ret_code = ret;
2906 setup_error:
2907 if (cmd_ctx->session) {
2908 session_unlock(cmd_ctx->session);
2909 }
2910 if (need_tracing_session) {
2911 session_unlock_list();
2912 }
2913 init_setup_error:
2914 return ret;
2915 }
2916
2917 /*
2918 * Thread managing health check socket.
2919 */
2920 static void *thread_manage_health(void *data)
2921 {
2922 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2923 uint32_t revents, nb_fd;
2924 struct lttng_poll_event events;
2925 struct lttcomm_health_msg msg;
2926 struct lttcomm_health_data reply;
2927
2928 DBG("[thread] Manage health check started");
2929
2930 rcu_register_thread();
2931
2932 /* Create unix socket */
2933 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2934 if (sock < 0) {
2935 ERR("Unable to create health check Unix socket");
2936 ret = -1;
2937 goto error;
2938 }
2939
2940 /*
2941 * Set the CLOEXEC flag. Return code is useless because either way, the
2942 * show must go on.
2943 */
2944 (void) utils_set_fd_cloexec(sock);
2945
2946 ret = lttcomm_listen_unix_sock(sock);
2947 if (ret < 0) {
2948 goto error;
2949 }
2950
2951 /*
2952 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2953 * more will be added to this poll set.
2954 */
2955 ret = create_thread_poll_set(&events, 2);
2956 if (ret < 0) {
2957 goto error;
2958 }
2959
2960 /* Add the application registration socket */
2961 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2962 if (ret < 0) {
2963 goto error;
2964 }
2965
2966 while (1) {
2967 DBG("Health check ready");
2968
2969 nb_fd = LTTNG_POLL_GETNB(&events);
2970
2971 /* Inifinite blocking call, waiting for transmission */
2972 restart:
2973 ret = lttng_poll_wait(&events, -1);
2974 if (ret < 0) {
2975 /*
2976 * Restart interrupted system call.
2977 */
2978 if (errno == EINTR) {
2979 goto restart;
2980 }
2981 goto error;
2982 }
2983
2984 for (i = 0; i < nb_fd; i++) {
2985 /* Fetch once the poll data */
2986 revents = LTTNG_POLL_GETEV(&events, i);
2987 pollfd = LTTNG_POLL_GETFD(&events, i);
2988
2989 /* Thread quit pipe has been closed. Killing thread. */
2990 ret = check_thread_quit_pipe(pollfd, revents);
2991 if (ret) {
2992 err = 0;
2993 goto exit;
2994 }
2995
2996 /* Event on the registration socket */
2997 if (pollfd == sock) {
2998 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2999 ERR("Health socket poll error");
3000 goto error;
3001 }
3002 }
3003 }
3004
3005 new_sock = lttcomm_accept_unix_sock(sock);
3006 if (new_sock < 0) {
3007 goto error;
3008 }
3009
3010 /*
3011 * Set the CLOEXEC flag. Return code is useless because either way, the
3012 * show must go on.
3013 */
3014 (void) utils_set_fd_cloexec(new_sock);
3015
3016 DBG("Receiving data from client for health...");
3017 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3018 if (ret <= 0) {
3019 DBG("Nothing recv() from client... continuing");
3020 ret = close(new_sock);
3021 if (ret) {
3022 PERROR("close");
3023 }
3024 new_sock = -1;
3025 continue;
3026 }
3027
3028 rcu_thread_online();
3029
3030 switch (msg.component) {
3031 case LTTNG_HEALTH_CMD:
3032 reply.ret_code = health_check_state(&health_thread_cmd);
3033 break;
3034 case LTTNG_HEALTH_APP_MANAGE:
3035 reply.ret_code = health_check_state(&health_thread_app_manage);
3036 break;
3037 case LTTNG_HEALTH_APP_REG:
3038 reply.ret_code = health_check_state(&health_thread_app_reg);
3039 break;
3040 case LTTNG_HEALTH_KERNEL:
3041 reply.ret_code = health_check_state(&health_thread_kernel);
3042 break;
3043 case LTTNG_HEALTH_CONSUMER:
3044 reply.ret_code = check_consumer_health();
3045 break;
3046 case LTTNG_HEALTH_ALL:
3047 reply.ret_code =
3048 health_check_state(&health_thread_app_manage) &&
3049 health_check_state(&health_thread_app_reg) &&
3050 health_check_state(&health_thread_cmd) &&
3051 health_check_state(&health_thread_kernel) &&
3052 check_consumer_health();
3053 break;
3054 default:
3055 reply.ret_code = LTTNG_ERR_UND;
3056 break;
3057 }
3058
3059 /*
3060 * Flip ret value since 0 is a success and 1 indicates a bad health for
3061 * the client where in the sessiond it is the opposite. Again, this is
3062 * just to make things easier for us poor developer which enjoy a lot
3063 * lazyness.
3064 */
3065 if (reply.ret_code == 0 || reply.ret_code == 1) {
3066 reply.ret_code = !reply.ret_code;
3067 }
3068
3069 DBG2("Health check return value %d", reply.ret_code);
3070
3071 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3072 if (ret < 0) {
3073 ERR("Failed to send health data back to client");
3074 }
3075
3076 /* End of transmission */
3077 ret = close(new_sock);
3078 if (ret) {
3079 PERROR("close");
3080 }
3081 new_sock = -1;
3082 }
3083
3084 exit:
3085 error:
3086 if (err) {
3087 ERR("Health error occurred in %s", __func__);
3088 }
3089 DBG("Health check thread dying");
3090 unlink(health_unix_sock_path);
3091 if (sock >= 0) {
3092 ret = close(sock);
3093 if (ret) {
3094 PERROR("close");
3095 }
3096 }
3097 if (new_sock >= 0) {
3098 ret = close(new_sock);
3099 if (ret) {
3100 PERROR("close");
3101 }
3102 }
3103
3104 lttng_poll_clean(&events);
3105
3106 rcu_unregister_thread();
3107 return NULL;
3108 }
3109
3110 /*
3111 * This thread manage all clients request using the unix client socket for
3112 * communication.
3113 */
3114 static void *thread_manage_clients(void *data)
3115 {
3116 int sock = -1, ret, i, pollfd, err = -1;
3117 int sock_error;
3118 uint32_t revents, nb_fd;
3119 struct command_ctx *cmd_ctx = NULL;
3120 struct lttng_poll_event events;
3121
3122 DBG("[thread] Manage client started");
3123
3124 testpoint(thread_manage_clients);
3125
3126 rcu_register_thread();
3127
3128 health_code_update(&health_thread_cmd);
3129
3130 ret = lttcomm_listen_unix_sock(client_sock);
3131 if (ret < 0) {
3132 goto error_listen;
3133 }
3134
3135 /*
3136 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3137 * more will be added to this poll set.
3138 */
3139 ret = create_thread_poll_set(&events, 2);
3140 if (ret < 0) {
3141 goto error_create_poll;
3142 }
3143
3144 /* Add the application registration socket */
3145 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3146 if (ret < 0) {
3147 goto error;
3148 }
3149
3150 /*
3151 * Notify parent pid that we are ready to accept command for client side.
3152 */
3153 if (opt_sig_parent) {
3154 kill(ppid, SIGUSR1);
3155 }
3156
3157 testpoint(thread_manage_clients_before_loop);
3158
3159 health_code_update(&health_thread_cmd);
3160
3161 while (1) {
3162 DBG("Accepting client command ...");
3163
3164 nb_fd = LTTNG_POLL_GETNB(&events);
3165
3166 /* Inifinite blocking call, waiting for transmission */
3167 restart:
3168 health_poll_update(&health_thread_cmd);
3169 ret = lttng_poll_wait(&events, -1);
3170 health_poll_update(&health_thread_cmd);
3171 if (ret < 0) {
3172 /*
3173 * Restart interrupted system call.
3174 */
3175 if (errno == EINTR) {
3176 goto restart;
3177 }
3178 goto error;
3179 }
3180
3181 for (i = 0; i < nb_fd; i++) {
3182 /* Fetch once the poll data */
3183 revents = LTTNG_POLL_GETEV(&events, i);
3184 pollfd = LTTNG_POLL_GETFD(&events, i);
3185
3186 health_code_update(&health_thread_cmd);
3187
3188 /* Thread quit pipe has been closed. Killing thread. */
3189 ret = check_thread_quit_pipe(pollfd, revents);
3190 if (ret) {
3191 err = 0;
3192 goto exit;
3193 }
3194
3195 /* Event on the registration socket */
3196 if (pollfd == client_sock) {
3197 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3198 ERR("Client socket poll error");
3199 goto error;
3200 }
3201 }
3202 }
3203
3204 DBG("Wait for client response");
3205
3206 health_code_update(&health_thread_cmd);
3207
3208 sock = lttcomm_accept_unix_sock(client_sock);
3209 if (sock < 0) {
3210 goto error;
3211 }
3212
3213 /*
3214 * Set the CLOEXEC flag. Return code is useless because either way, the
3215 * show must go on.
3216 */
3217 (void) utils_set_fd_cloexec(sock);
3218
3219 /* Set socket option for credentials retrieval */
3220 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3221 if (ret < 0) {
3222 goto error;
3223 }
3224
3225 /* Allocate context command to process the client request */
3226 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3227 if (cmd_ctx == NULL) {
3228 PERROR("zmalloc cmd_ctx");
3229 goto error;
3230 }
3231
3232 /* Allocate data buffer for reception */
3233 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3234 if (cmd_ctx->lsm == NULL) {
3235 PERROR("zmalloc cmd_ctx->lsm");
3236 goto error;
3237 }
3238
3239 cmd_ctx->llm = NULL;
3240 cmd_ctx->session = NULL;
3241
3242 health_code_update(&health_thread_cmd);
3243
3244 /*
3245 * Data is received from the lttng client. The struct
3246 * lttcomm_session_msg (lsm) contains the command and data request of
3247 * the client.
3248 */
3249 DBG("Receiving data from client ...");
3250 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3251 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3252 if (ret <= 0) {
3253 DBG("Nothing recv() from client... continuing");
3254 ret = close(sock);
3255 if (ret) {
3256 PERROR("close");
3257 }
3258 sock = -1;
3259 clean_command_ctx(&cmd_ctx);
3260 continue;
3261 }
3262
3263 health_code_update(&health_thread_cmd);
3264
3265 // TODO: Validate cmd_ctx including sanity check for
3266 // security purpose.
3267
3268 rcu_thread_online();
3269 /*
3270 * This function dispatch the work to the kernel or userspace tracer
3271 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3272 * informations for the client. The command context struct contains
3273 * everything this function may needs.
3274 */
3275 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3276 rcu_thread_offline();
3277 if (ret < 0) {
3278 if (sock_error) {
3279 ret = close(sock);
3280 if (ret) {
3281 PERROR("close");
3282 }
3283 sock = -1;
3284 }
3285 /*
3286 * TODO: Inform client somehow of the fatal error. At
3287 * this point, ret < 0 means that a zmalloc failed
3288 * (ENOMEM). Error detected but still accept
3289 * command, unless a socket error has been
3290 * detected.
3291 */
3292 clean_command_ctx(&cmd_ctx);
3293 continue;
3294 }
3295
3296 health_code_update(&health_thread_cmd);
3297
3298 DBG("Sending response (size: %d, retcode: %s)",
3299 cmd_ctx->lttng_msg_size,
3300 lttng_strerror(-cmd_ctx->llm->ret_code));
3301 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3302 if (ret < 0) {
3303 ERR("Failed to send data back to client");
3304 }
3305
3306 /* End of transmission */
3307 ret = close(sock);
3308 if (ret) {
3309 PERROR("close");
3310 }
3311 sock = -1;
3312
3313 clean_command_ctx(&cmd_ctx);
3314
3315 health_code_update(&health_thread_cmd);
3316 }
3317
3318 exit:
3319 error:
3320 if (sock >= 0) {
3321 ret = close(sock);
3322 if (ret) {
3323 PERROR("close");
3324 }
3325 }
3326
3327 lttng_poll_clean(&events);
3328 clean_command_ctx(&cmd_ctx);
3329
3330 error_listen:
3331 error_create_poll:
3332 unlink(client_unix_sock_path);
3333 if (client_sock >= 0) {
3334 ret = close(client_sock);
3335 if (ret) {
3336 PERROR("close");
3337 }
3338 }
3339
3340 if (err) {
3341 health_error(&health_thread_cmd);
3342 ERR("Health error occurred in %s", __func__);
3343 }
3344
3345 health_exit(&health_thread_cmd);
3346
3347 DBG("Client thread dying");
3348
3349 rcu_unregister_thread();
3350 return NULL;
3351 }
3352
3353
3354 /*
3355 * usage function on stderr
3356 */
3357 static void usage(void)
3358 {
3359 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3360 fprintf(stderr, " -h, --help Display this usage.\n");
3361 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3362 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3363 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3364 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3365 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3366 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3367 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3368 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3369 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3370 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3371 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3372 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3373 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3374 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3375 fprintf(stderr, " -V, --version Show version number.\n");
3376 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3377 fprintf(stderr, " -q, --quiet No output at all.\n");
3378 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3379 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3380 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3381 }
3382
3383 /*
3384 * daemon argument parsing
3385 */
3386 static int parse_args(int argc, char **argv)
3387 {
3388 int c;
3389
3390 static struct option long_options[] = {
3391 { "client-sock", 1, 0, 'c' },
3392 { "apps-sock", 1, 0, 'a' },
3393 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3394 { "kconsumerd-err-sock", 1, 0, 'E' },
3395 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3396 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3397 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3398 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3399 { "consumerd32-path", 1, 0, 'u' },
3400 { "consumerd32-libdir", 1, 0, 'U' },
3401 { "consumerd64-path", 1, 0, 't' },
3402 { "consumerd64-libdir", 1, 0, 'T' },
3403 { "daemonize", 0, 0, 'd' },
3404 { "sig-parent", 0, 0, 'S' },
3405 { "help", 0, 0, 'h' },
3406 { "group", 1, 0, 'g' },
3407 { "version", 0, 0, 'V' },
3408 { "quiet", 0, 0, 'q' },
3409 { "verbose", 0, 0, 'v' },
3410 { "verbose-consumer", 0, 0, 'Z' },
3411 { "no-kernel", 0, 0, 'N' },
3412 { NULL, 0, 0, 0 }
3413 };
3414
3415 while (1) {
3416 int option_index = 0;
3417 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3418 long_options, &option_index);
3419 if (c == -1) {
3420 break;
3421 }
3422
3423 switch (c) {
3424 case 0:
3425 fprintf(stderr, "option %s", long_options[option_index].name);
3426 if (optarg) {
3427 fprintf(stderr, " with arg %s\n", optarg);
3428 }
3429 break;
3430 case 'c':
3431 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3432 break;
3433 case 'a':
3434 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3435 break;
3436 case 'd':
3437 opt_daemon = 1;
3438 break;
3439 case 'g':
3440 opt_tracing_group = optarg;
3441 break;
3442 case 'h':
3443 usage();
3444 exit(EXIT_FAILURE);
3445 case 'V':
3446 fprintf(stdout, "%s\n", VERSION);
3447 exit(EXIT_SUCCESS);
3448 case 'S':
3449 opt_sig_parent = 1;
3450 break;
3451 case 'E':
3452 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3453 break;
3454 case 'C':
3455 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3456 break;
3457 case 'F':
3458 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3459 break;
3460 case 'D':
3461 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3462 break;
3463 case 'H':
3464 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3465 break;
3466 case 'G':
3467 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3468 break;
3469 case 'N':
3470 opt_no_kernel = 1;
3471 break;
3472 case 'q':
3473 lttng_opt_quiet = 1;
3474 break;
3475 case 'v':
3476 /* Verbose level can increase using multiple -v */
3477 lttng_opt_verbose += 1;
3478 break;
3479 case 'Z':
3480 opt_verbose_consumer += 1;
3481 break;
3482 case 'u':
3483 consumerd32_bin= optarg;
3484 break;
3485 case 'U':
3486 consumerd32_libdir = optarg;
3487 break;
3488 case 't':
3489 consumerd64_bin = optarg;
3490 break;
3491 case 'T':
3492 consumerd64_libdir = optarg;
3493 break;
3494 default:
3495 /* Unknown option or other error.
3496 * Error is printed by getopt, just return */
3497 return -1;
3498 }
3499 }
3500
3501 return 0;
3502 }
3503
3504 /*
3505 * Creates the two needed socket by the daemon.
3506 * apps_sock - The communication socket for all UST apps.
3507 * client_sock - The communication of the cli tool (lttng).
3508 */
3509 static int init_daemon_socket(void)
3510 {
3511 int ret = 0;
3512 mode_t old_umask;
3513
3514 old_umask = umask(0);
3515
3516 /* Create client tool unix socket */
3517 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3518 if (client_sock < 0) {
3519 ERR("Create unix sock failed: %s", client_unix_sock_path);
3520 ret = -1;
3521 goto end;
3522 }
3523
3524 /* Set the cloexec flag */
3525 ret = utils_set_fd_cloexec(client_sock);
3526 if (ret < 0) {
3527 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3528 "Continuing but note that the consumer daemon will have a "
3529 "reference to this socket on exec()", client_sock);
3530 }
3531
3532 /* File permission MUST be 660 */
3533 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3534 if (ret < 0) {
3535 ERR("Set file permissions failed: %s", client_unix_sock_path);
3536 PERROR("chmod");
3537 goto end;
3538 }
3539
3540 /* Create the application unix socket */
3541 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3542 if (apps_sock < 0) {
3543 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3544 ret = -1;
3545 goto end;
3546 }
3547
3548 /* Set the cloexec flag */
3549 ret = utils_set_fd_cloexec(apps_sock);
3550 if (ret < 0) {
3551 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3552 "Continuing but note that the consumer daemon will have a "
3553 "reference to this socket on exec()", apps_sock);
3554 }
3555
3556 /* File permission MUST be 666 */
3557 ret = chmod(apps_unix_sock_path,
3558 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3559 if (ret < 0) {
3560 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3561 PERROR("chmod");
3562 goto end;
3563 }
3564
3565 DBG3("Session daemon client socket %d and application socket %d created",
3566 client_sock, apps_sock);
3567
3568 end:
3569 umask(old_umask);
3570 return ret;
3571 }
3572
3573 /*
3574 * Check if the global socket is available, and if a daemon is answering at the
3575 * other side. If yes, error is returned.
3576 */
3577 static int check_existing_daemon(void)
3578 {
3579 /* Is there anybody out there ? */
3580 if (lttng_session_daemon_alive()) {
3581 return -EEXIST;
3582 }
3583
3584 return 0;
3585 }
3586
3587 /*
3588 * Set the tracing group gid onto the client socket.
3589 *
3590 * Race window between mkdir and chown is OK because we are going from more
3591 * permissive (root.root) to less permissive (root.tracing).
3592 */
3593 static int set_permissions(char *rundir)
3594 {
3595 int ret;
3596 gid_t gid;
3597
3598 ret = allowed_group();
3599 if (ret < 0) {
3600 WARN("No tracing group detected");
3601 ret = 0;
3602 goto end;
3603 }
3604
3605 gid = ret;
3606
3607 /* Set lttng run dir */
3608 ret = chown(rundir, 0, gid);
3609 if (ret < 0) {
3610 ERR("Unable to set group on %s", rundir);
3611 PERROR("chown");
3612 }
3613
3614 /* Ensure tracing group can search the run dir */
3615 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3616 if (ret < 0) {
3617 ERR("Unable to set permissions on %s", rundir);
3618 PERROR("chmod");
3619 }
3620
3621 /* lttng client socket path */
3622 ret = chown(client_unix_sock_path, 0, gid);
3623 if (ret < 0) {
3624 ERR("Unable to set group on %s", client_unix_sock_path);
3625 PERROR("chown");
3626 }
3627
3628 /* kconsumer error socket path */
3629 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3630 if (ret < 0) {
3631 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3632 PERROR("chown");
3633 }
3634
3635 /* 64-bit ustconsumer error socket path */
3636 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3637 if (ret < 0) {
3638 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3639 PERROR("chown");
3640 }
3641
3642 /* 32-bit ustconsumer compat32 error socket path */
3643 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3644 if (ret < 0) {
3645 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3646 PERROR("chown");
3647 }
3648
3649 DBG("All permissions are set");
3650
3651 end:
3652 return ret;
3653 }
3654
3655 /*
3656 * Create the lttng run directory needed for all global sockets and pipe.
3657 */
3658 static int create_lttng_rundir(const char *rundir)
3659 {
3660 int ret;
3661
3662 DBG3("Creating LTTng run directory: %s", rundir);
3663
3664 ret = mkdir(rundir, S_IRWXU);
3665 if (ret < 0) {
3666 if (errno != EEXIST) {
3667 ERR("Unable to create %s", rundir);
3668 goto error;
3669 } else {
3670 ret = 0;
3671 }
3672 }
3673
3674 error:
3675 return ret;
3676 }
3677
3678 /*
3679 * Setup sockets and directory needed by the kconsumerd communication with the
3680 * session daemon.
3681 */
3682 static int set_consumer_sockets(struct consumer_data *consumer_data,
3683 const char *rundir)
3684 {
3685 int ret;
3686 char path[PATH_MAX];
3687
3688 switch (consumer_data->type) {
3689 case LTTNG_CONSUMER_KERNEL:
3690 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3691 break;
3692 case LTTNG_CONSUMER64_UST:
3693 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3694 break;
3695 case LTTNG_CONSUMER32_UST:
3696 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3697 break;
3698 default:
3699 ERR("Consumer type unknown");
3700 ret = -EINVAL;
3701 goto error;
3702 }
3703
3704 DBG2("Creating consumer directory: %s", path);
3705
3706 ret = mkdir(path, S_IRWXU);
3707 if (ret < 0) {
3708 if (errno != EEXIST) {
3709 PERROR("mkdir");
3710 ERR("Failed to create %s", path);
3711 goto error;
3712 }
3713 ret = -1;
3714 }
3715
3716 /* Create the kconsumerd error unix socket */
3717 consumer_data->err_sock =
3718 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3719 if (consumer_data->err_sock < 0) {
3720 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3721 ret = -1;
3722 goto error;
3723 }
3724
3725 /* File permission MUST be 660 */
3726 ret = chmod(consumer_data->err_unix_sock_path,
3727 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3728 if (ret < 0) {
3729 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3730 PERROR("chmod");
3731 goto error;
3732 }
3733
3734 error:
3735 return ret;
3736 }
3737
3738 /*
3739 * Signal handler for the daemon
3740 *
3741 * Simply stop all worker threads, leaving main() return gracefully after
3742 * joining all threads and calling cleanup().
3743 */
3744 static void sighandler(int sig)
3745 {
3746 switch (sig) {
3747 case SIGPIPE:
3748 DBG("SIGPIPE caught");
3749 return;
3750 case SIGINT:
3751 DBG("SIGINT caught");
3752 stop_threads();
3753 break;
3754 case SIGTERM:
3755 DBG("SIGTERM caught");
3756 stop_threads();
3757 break;
3758 default:
3759 break;
3760 }
3761 }
3762
3763 /*
3764 * Setup signal handler for :
3765 * SIGINT, SIGTERM, SIGPIPE
3766 */
3767 static int set_signal_handler(void)
3768 {
3769 int ret = 0;
3770 struct sigaction sa;
3771 sigset_t sigset;
3772
3773 if ((ret = sigemptyset(&sigset)) < 0) {
3774 PERROR("sigemptyset");
3775 return ret;
3776 }
3777
3778 sa.sa_handler = sighandler;
3779 sa.sa_mask = sigset;
3780 sa.sa_flags = 0;
3781 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3782 PERROR("sigaction");
3783 return ret;
3784 }
3785
3786 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3787 PERROR("sigaction");
3788 return ret;
3789 }
3790
3791 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3792 PERROR("sigaction");
3793 return ret;
3794 }
3795
3796 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3797
3798 return ret;
3799 }
3800
3801 /*
3802 * Set open files limit to unlimited. This daemon can open a large number of
3803 * file descriptors in order to consumer multiple kernel traces.
3804 */
3805 static void set_ulimit(void)
3806 {
3807 int ret;
3808 struct rlimit lim;
3809
3810 /* The kernel does not allowed an infinite limit for open files */
3811 lim.rlim_cur = 65535;
3812 lim.rlim_max = 65535;
3813
3814 ret = setrlimit(RLIMIT_NOFILE, &lim);
3815 if (ret < 0) {
3816 PERROR("failed to set open files limit");
3817 }
3818 }
3819
3820 /*
3821 * main
3822 */
3823 int main(int argc, char **argv)
3824 {
3825 int ret = 0;
3826 void *status;
3827 const char *home_path, *env_app_timeout;
3828
3829 init_kernel_workarounds();
3830
3831 rcu_register_thread();
3832
3833 setup_consumerd_path();
3834
3835 /* Parse arguments */
3836 progname = argv[0];
3837 if ((ret = parse_args(argc, argv) < 0)) {
3838 goto error;
3839 }
3840
3841 /* Daemonize */
3842 if (opt_daemon) {
3843 int i;
3844
3845 /*
3846 * fork
3847 * child: setsid, close FD 0, 1, 2, chdir /
3848 * parent: exit (if fork is successful)
3849 */
3850 ret = daemon(0, 0);
3851 if (ret < 0) {
3852 PERROR("daemon");
3853 goto error;
3854 }
3855 /*
3856 * We are in the child. Make sure all other file
3857 * descriptors are closed, in case we are called with
3858 * more opened file descriptors than the standard ones.
3859 */
3860 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3861 (void) close(i);
3862 }
3863 }
3864
3865 /* Create thread quit pipe */
3866 if ((ret = init_thread_quit_pipe()) < 0) {
3867 goto error;
3868 }
3869
3870 /* Check if daemon is UID = 0 */
3871 is_root = !getuid();
3872
3873 if (is_root) {
3874 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3875
3876 /* Create global run dir with root access */
3877 ret = create_lttng_rundir(rundir);
3878 if (ret < 0) {
3879 goto error;
3880 }
3881
3882 if (strlen(apps_unix_sock_path) == 0) {
3883 snprintf(apps_unix_sock_path, PATH_MAX,
3884 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3885 }
3886
3887 if (strlen(client_unix_sock_path) == 0) {
3888 snprintf(client_unix_sock_path, PATH_MAX,
3889 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3890 }
3891
3892 /* Set global SHM for ust */
3893 if (strlen(wait_shm_path) == 0) {
3894 snprintf(wait_shm_path, PATH_MAX,
3895 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3896 }
3897
3898 if (strlen(health_unix_sock_path) == 0) {
3899 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3900 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3901 }
3902
3903 /* Setup kernel consumerd path */
3904 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3905 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3906 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3907 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3908
3909 DBG2("Kernel consumer err path: %s",
3910 kconsumer_data.err_unix_sock_path);
3911 DBG2("Kernel consumer cmd path: %s",
3912 kconsumer_data.cmd_unix_sock_path);
3913 } else {
3914 home_path = get_home_dir();
3915 if (home_path == NULL) {
3916 /* TODO: Add --socket PATH option */
3917 ERR("Can't get HOME directory for sockets creation.");
3918 ret = -EPERM;
3919 goto error;
3920 }
3921
3922 /*
3923 * Create rundir from home path. This will create something like
3924 * $HOME/.lttng
3925 */
3926 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3927 if (ret < 0) {
3928 ret = -ENOMEM;
3929 goto error;
3930 }
3931
3932 ret = create_lttng_rundir(rundir);
3933 if (ret < 0) {
3934 goto error;
3935 }
3936
3937 if (strlen(apps_unix_sock_path) == 0) {
3938 snprintf(apps_unix_sock_path, PATH_MAX,
3939 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3940 }
3941
3942 /* Set the cli tool unix socket path */
3943 if (strlen(client_unix_sock_path) == 0) {
3944 snprintf(client_unix_sock_path, PATH_MAX,
3945 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3946 }
3947
3948 /* Set global SHM for ust */
3949 if (strlen(wait_shm_path) == 0) {
3950 snprintf(wait_shm_path, PATH_MAX,
3951 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3952 }
3953
3954 /* Set health check Unix path */
3955 if (strlen(health_unix_sock_path) == 0) {
3956 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3957 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3958 }
3959 }
3960
3961 /* Set consumer initial state */
3962 kernel_consumerd_state = CONSUMER_STOPPED;
3963 ust_consumerd_state = CONSUMER_STOPPED;
3964
3965 DBG("Client socket path %s", client_unix_sock_path);
3966 DBG("Application socket path %s", apps_unix_sock_path);
3967 DBG("LTTng run directory path: %s", rundir);
3968
3969 /* 32 bits consumerd path setup */
3970 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
3971 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
3972 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
3973 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
3974
3975 DBG2("UST consumer 32 bits err path: %s",
3976 ustconsumer32_data.err_unix_sock_path);
3977 DBG2("UST consumer 32 bits cmd path: %s",
3978 ustconsumer32_data.cmd_unix_sock_path);
3979
3980 /* 64 bits consumerd path setup */
3981 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
3982 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
3983 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
3984 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
3985
3986 DBG2("UST consumer 64 bits err path: %s",
3987 ustconsumer64_data.err_unix_sock_path);
3988 DBG2("UST consumer 64 bits cmd path: %s",
3989 ustconsumer64_data.cmd_unix_sock_path);
3990
3991 /*
3992 * See if daemon already exist.
3993 */
3994 if ((ret = check_existing_daemon()) < 0) {
3995 ERR("Already running daemon.\n");
3996 /*
3997 * We do not goto exit because we must not cleanup()
3998 * because a daemon is already running.
3999 */
4000 goto error;
4001 }
4002
4003 /*
4004 * Init UST app hash table. Alloc hash table before this point since
4005 * cleanup() can get called after that point.
4006 */
4007 ust_app_ht_alloc();
4008
4009 /* After this point, we can safely call cleanup() with "goto exit" */
4010
4011 /*
4012 * These actions must be executed as root. We do that *after* setting up
4013 * the sockets path because we MUST make the check for another daemon using
4014 * those paths *before* trying to set the kernel consumer sockets and init
4015 * kernel tracer.
4016 */
4017 if (is_root) {
4018 ret = set_consumer_sockets(&kconsumer_data, rundir);
4019 if (ret < 0) {
4020 goto exit;
4021 }
4022
4023 /* Setup kernel tracer */
4024 if (!opt_no_kernel) {
4025 init_kernel_tracer();
4026 }
4027
4028 /* Set ulimit for open files */
4029 set_ulimit();
4030 }
4031 /* init lttng_fd tracking must be done after set_ulimit. */
4032 lttng_fd_init();
4033
4034 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4035 if (ret < 0) {
4036 goto exit;
4037 }
4038
4039 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4040 if (ret < 0) {
4041 goto exit;
4042 }
4043
4044 if ((ret = set_signal_handler()) < 0) {
4045 goto exit;
4046 }
4047
4048 /* Setup the needed unix socket */
4049 if ((ret = init_daemon_socket()) < 0) {
4050 goto exit;
4051 }
4052
4053 /* Set credentials to socket */
4054 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4055 goto exit;
4056 }
4057
4058 /* Get parent pid if -S, --sig-parent is specified. */
4059 if (opt_sig_parent) {
4060 ppid = getppid();
4061 }
4062
4063 /* Setup the kernel pipe for waking up the kernel thread */
4064 if (is_root && !opt_no_kernel) {
4065 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4066 goto exit;
4067 }
4068 }
4069
4070 /* Setup the thread apps communication pipe. */
4071 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4072 goto exit;
4073 }
4074
4075 /* Init UST command queue. */
4076 cds_wfq_init(&ust_cmd_queue.queue);
4077
4078 /*
4079 * Get session list pointer. This pointer MUST NOT be free(). This list is
4080 * statically declared in session.c
4081 */
4082 session_list_ptr = session_get_list();
4083
4084 /* Set up max poll set size */
4085 lttng_poll_set_max_size();
4086
4087 cmd_init();
4088
4089 /* Init all health thread counters. */
4090 health_init(&health_thread_cmd);
4091 health_init(&health_thread_kernel);
4092 health_init(&health_thread_app_manage);
4093 health_init(&health_thread_app_reg);
4094
4095 /*
4096 * Init health counters of the consumer thread. We do a quick hack here to
4097 * the state of the consumer health is fine even if the thread is not
4098 * started. Once the thread starts, the health state is updated with a poll
4099 * value to set a health code path. This is simply to ease our life and has
4100 * no cost what so ever.
4101 */
4102 health_init(&kconsumer_data.health);
4103 health_poll_update(&kconsumer_data.health);
4104 health_init(&ustconsumer32_data.health);
4105 health_poll_update(&ustconsumer32_data.health);
4106 health_init(&ustconsumer64_data.health);
4107 health_poll_update(&ustconsumer64_data.health);
4108
4109 /* Check for the application socket timeout env variable. */
4110 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4111 if (env_app_timeout) {
4112 app_socket_timeout = atoi(env_app_timeout);
4113 } else {
4114 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4115 }
4116
4117 /* Create thread to manage the client socket */
4118 ret = pthread_create(&health_thread, NULL,
4119 thread_manage_health, (void *) NULL);
4120 if (ret != 0) {
4121 PERROR("pthread_create health");
4122 goto exit_health;
4123 }
4124
4125 /* Create thread to manage the client socket */
4126 ret = pthread_create(&client_thread, NULL,
4127 thread_manage_clients, (void *) NULL);
4128 if (ret != 0) {
4129 PERROR("pthread_create clients");
4130 goto exit_client;
4131 }
4132
4133 /* Create thread to dispatch registration */
4134 ret = pthread_create(&dispatch_thread, NULL,
4135 thread_dispatch_ust_registration, (void *) NULL);
4136 if (ret != 0) {
4137 PERROR("pthread_create dispatch");
4138 goto exit_dispatch;
4139 }
4140
4141 /* Create thread to manage application registration. */
4142 ret = pthread_create(&reg_apps_thread, NULL,
4143 thread_registration_apps, (void *) NULL);
4144 if (ret != 0) {
4145 PERROR("pthread_create registration");
4146 goto exit_reg_apps;
4147 }
4148
4149 /* Create thread to manage application socket */
4150 ret = pthread_create(&apps_thread, NULL,
4151 thread_manage_apps, (void *) NULL);
4152 if (ret != 0) {
4153 PERROR("pthread_create apps");
4154 goto exit_apps;
4155 }
4156
4157 /* Don't start this thread if kernel tracing is not requested nor root */
4158 if (is_root && !opt_no_kernel) {
4159 /* Create kernel thread to manage kernel event */
4160 ret = pthread_create(&kernel_thread, NULL,
4161 thread_manage_kernel, (void *) NULL);
4162 if (ret != 0) {
4163 PERROR("pthread_create kernel");
4164 goto exit_kernel;
4165 }
4166
4167 ret = pthread_join(kernel_thread, &status);
4168 if (ret != 0) {
4169 PERROR("pthread_join");
4170 goto error; /* join error, exit without cleanup */
4171 }
4172 }
4173
4174 exit_kernel:
4175 ret = pthread_join(apps_thread, &status);
4176 if (ret != 0) {
4177 PERROR("pthread_join");
4178 goto error; /* join error, exit without cleanup */
4179 }
4180
4181 exit_apps:
4182 ret = pthread_join(reg_apps_thread, &status);
4183 if (ret != 0) {
4184 PERROR("pthread_join");
4185 goto error; /* join error, exit without cleanup */
4186 }
4187
4188 exit_reg_apps:
4189 ret = pthread_join(dispatch_thread, &status);
4190 if (ret != 0) {
4191 PERROR("pthread_join");
4192 goto error; /* join error, exit without cleanup */
4193 }
4194
4195 exit_dispatch:
4196 ret = pthread_join(client_thread, &status);
4197 if (ret != 0) {
4198 PERROR("pthread_join");
4199 goto error; /* join error, exit without cleanup */
4200 }
4201
4202 ret = join_consumer_thread(&kconsumer_data);
4203 if (ret != 0) {
4204 PERROR("join_consumer");
4205 goto error; /* join error, exit without cleanup */
4206 }
4207
4208 ret = join_consumer_thread(&ustconsumer32_data);
4209 if (ret != 0) {
4210 PERROR("join_consumer ust32");
4211 goto error; /* join error, exit without cleanup */
4212 }
4213
4214 ret = join_consumer_thread(&ustconsumer64_data);
4215 if (ret != 0) {
4216 PERROR("join_consumer ust64");
4217 goto error; /* join error, exit without cleanup */
4218 }
4219
4220 exit_client:
4221 ret = pthread_join(health_thread, &status);
4222 if (ret != 0) {
4223 PERROR("pthread_join health thread");
4224 goto error; /* join error, exit without cleanup */
4225 }
4226
4227 exit_health:
4228 exit:
4229 /*
4230 * cleanup() is called when no other thread is running.
4231 */
4232 rcu_thread_online();
4233 cleanup();
4234 rcu_thread_offline();
4235 rcu_unregister_thread();
4236 if (!ret) {
4237 exit(EXIT_SUCCESS);
4238 }
4239 error:
4240 exit(EXIT_FAILURE);
4241 }
This page took 0.17366 seconds and 4 git commands to generate.