Cleanup unused health state reference
[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 "health.h"
63 #include "testpoint.h"
64
65 #define CONSUMERD_FILE "lttng-consumerd"
66
67 /* Const values */
68 const char default_home_dir[] = DEFAULT_HOME_DIR;
69 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
70 const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
71 const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
72
73 const char *progname;
74 const char *opt_tracing_group;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static char *rundir;
82
83 /*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
87 static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
91 .err_sock = -1,
92 .cmd_sock = -1,
93 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
94 .lock = PTHREAD_MUTEX_INITIALIZER,
95 .cond = PTHREAD_COND_INITIALIZER,
96 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
97 };
98 static struct consumer_data ustconsumer64_data = {
99 .type = LTTNG_CONSUMER64_UST,
100 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
101 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
102 .err_sock = -1,
103 .cmd_sock = -1,
104 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
105 .lock = PTHREAD_MUTEX_INITIALIZER,
106 .cond = PTHREAD_COND_INITIALIZER,
107 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
108 };
109 static struct consumer_data ustconsumer32_data = {
110 .type = LTTNG_CONSUMER32_UST,
111 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
112 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
116 .lock = PTHREAD_MUTEX_INITIALIZER,
117 .cond = PTHREAD_COND_INITIALIZER,
118 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
119 };
120
121 /* Shared between threads */
122 static int dispatch_thread_exit;
123
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path[PATH_MAX];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path[PATH_MAX];
128 /* global wait shm path for UST */
129 static char wait_shm_path[PATH_MAX];
130 /* Global health check unix path */
131 static char health_unix_sock_path[PATH_MAX];
132
133 /* Sockets and FDs */
134 static int client_sock = -1;
135 static int apps_sock = -1;
136 int kernel_tracer_fd = -1;
137 static int kernel_poll_pipe[2] = { -1, -1 };
138
139 /*
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
142 */
143 static int thread_quit_pipe[2] = { -1, -1 };
144
145 /*
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
148 */
149 static int apps_cmd_pipe[2] = { -1, -1 };
150
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread;
153 static pthread_t reg_apps_thread;
154 static pthread_t client_thread;
155 static pthread_t kernel_thread;
156 static pthread_t dispatch_thread;
157 static pthread_t health_thread;
158
159 /*
160 * UST registration command queue. This queue is tied with a futex and uses a N
161 * wakers / 1 waiter implemented and detailed in futex.c/.h
162 *
163 * The thread_manage_apps and thread_dispatch_ust_registration interact with
164 * this queue and the wait/wake scheme.
165 */
166 static struct ust_cmd_queue ust_cmd_queue;
167
168 /*
169 * Pointer initialized before thread creation.
170 *
171 * This points to the tracing session list containing the session count and a
172 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
173 * MUST NOT be taken if you call a public function in session.c.
174 *
175 * The lock is nested inside the structure: session_list_ptr->lock. Please use
176 * session_lock_list and session_unlock_list for lock acquisition.
177 */
178 static struct ltt_session_list *session_list_ptr;
179
180 int ust_consumerd64_fd = -1;
181 int ust_consumerd32_fd = -1;
182
183 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
184 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
185 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
186 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
187
188 static const char *module_proc_lttng = "/proc/lttng";
189
190 /*
191 * Consumer daemon state which is changed when spawning it, killing it or in
192 * case of a fatal error.
193 */
194 enum consumerd_state {
195 CONSUMER_STARTED = 1,
196 CONSUMER_STOPPED = 2,
197 CONSUMER_ERROR = 3,
198 };
199
200 /*
201 * This consumer daemon state is used to validate if a client command will be
202 * able to reach the consumer. If not, the client is informed. For instance,
203 * doing a "lttng start" when the consumer state is set to ERROR will return an
204 * error to the client.
205 *
206 * The following example shows a possible race condition of this scheme:
207 *
208 * consumer thread error happens
209 * client cmd arrives
210 * client cmd checks state -> still OK
211 * consumer thread exit, sets error
212 * client cmd try to talk to consumer
213 * ...
214 *
215 * However, since the consumer is a different daemon, we have no way of making
216 * sure the command will reach it safely even with this state flag. This is why
217 * we consider that up to the state validation during command processing, the
218 * command is safe. After that, we can not guarantee the correctness of the
219 * client request vis-a-vis the consumer.
220 */
221 static enum consumerd_state ust_consumerd_state;
222 static enum consumerd_state kernel_consumerd_state;
223
224 /*
225 * Socket timeout for receiving and sending in seconds.
226 */
227 static int app_socket_timeout;
228
229 static
230 void setup_consumerd_path(void)
231 {
232 const char *bin, *libdir;
233
234 /*
235 * Allow INSTALL_BIN_PATH to be used as a target path for the
236 * native architecture size consumer if CONFIG_CONSUMER*_PATH
237 * has not been defined.
238 */
239 #if (CAA_BITS_PER_LONG == 32)
240 if (!consumerd32_bin[0]) {
241 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
242 }
243 if (!consumerd32_libdir[0]) {
244 consumerd32_libdir = INSTALL_LIB_PATH;
245 }
246 #elif (CAA_BITS_PER_LONG == 64)
247 if (!consumerd64_bin[0]) {
248 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
249 }
250 if (!consumerd64_libdir[0]) {
251 consumerd64_libdir = INSTALL_LIB_PATH;
252 }
253 #else
254 #error "Unknown bitness"
255 #endif
256
257 /*
258 * runtime env. var. overrides the build default.
259 */
260 bin = getenv("LTTNG_CONSUMERD32_BIN");
261 if (bin) {
262 consumerd32_bin = bin;
263 }
264 bin = getenv("LTTNG_CONSUMERD64_BIN");
265 if (bin) {
266 consumerd64_bin = bin;
267 }
268 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
269 if (libdir) {
270 consumerd32_libdir = libdir;
271 }
272 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
273 if (libdir) {
274 consumerd64_libdir = libdir;
275 }
276 }
277
278 /*
279 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
280 */
281 static int create_thread_poll_set(struct lttng_poll_event *events,
282 unsigned int size)
283 {
284 int ret;
285
286 if (events == NULL || size == 0) {
287 ret = -1;
288 goto error;
289 }
290
291 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
292 if (ret < 0) {
293 goto error;
294 }
295
296 /* Add quit pipe */
297 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
298 if (ret < 0) {
299 goto error;
300 }
301
302 return 0;
303
304 error:
305 return ret;
306 }
307
308 /*
309 * Check if the thread quit pipe was triggered.
310 *
311 * Return 1 if it was triggered else 0;
312 */
313 static int check_thread_quit_pipe(int fd, uint32_t events)
314 {
315 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
316 return 1;
317 }
318
319 return 0;
320 }
321
322 /*
323 * Return group ID of the tracing group or -1 if not found.
324 */
325 static gid_t allowed_group(void)
326 {
327 struct group *grp;
328
329 if (opt_tracing_group) {
330 grp = getgrnam(opt_tracing_group);
331 } else {
332 grp = getgrnam(default_tracing_group);
333 }
334 if (!grp) {
335 return -1;
336 } else {
337 return grp->gr_gid;
338 }
339 }
340
341 /*
342 * Init thread quit pipe.
343 *
344 * Return -1 on error or 0 if all pipes are created.
345 */
346 static int init_thread_quit_pipe(void)
347 {
348 int ret, i;
349
350 ret = pipe(thread_quit_pipe);
351 if (ret < 0) {
352 PERROR("thread quit pipe");
353 goto error;
354 }
355
356 for (i = 0; i < 2; i++) {
357 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
358 if (ret < 0) {
359 PERROR("fcntl");
360 goto error;
361 }
362 }
363
364 error:
365 return ret;
366 }
367
368 /*
369 * Stop all threads by closing the thread quit pipe.
370 */
371 static void stop_threads(void)
372 {
373 int ret;
374
375 /* Stopping all threads */
376 DBG("Terminating all threads");
377 ret = notify_thread_pipe(thread_quit_pipe[1]);
378 if (ret < 0) {
379 ERR("write error on thread quit pipe");
380 }
381
382 /* Dispatch thread */
383 CMM_STORE_SHARED(dispatch_thread_exit, 1);
384 futex_nto1_wake(&ust_cmd_queue.futex);
385 }
386
387 /*
388 * Cleanup the daemon
389 */
390 static void cleanup(void)
391 {
392 int ret;
393 char *cmd = NULL;
394 struct ltt_session *sess, *stmp;
395
396 DBG("Cleaning up");
397
398 /* First thing first, stop all threads */
399 utils_close_pipe(thread_quit_pipe);
400
401 DBG("Removing %s directory", rundir);
402 ret = asprintf(&cmd, "rm -rf %s", rundir);
403 if (ret < 0) {
404 ERR("asprintf failed. Something is really wrong!");
405 }
406
407 /* Remove lttng run directory */
408 ret = system(cmd);
409 if (ret < 0) {
410 ERR("Unable to clean %s", rundir);
411 }
412 free(cmd);
413 free(rundir);
414
415 DBG("Cleaning up all sessions");
416
417 /* Destroy session list mutex */
418 if (session_list_ptr != NULL) {
419 pthread_mutex_destroy(&session_list_ptr->lock);
420
421 /* Cleanup ALL session */
422 cds_list_for_each_entry_safe(sess, stmp,
423 &session_list_ptr->head, list) {
424 cmd_destroy_session(sess, kernel_poll_pipe[1]);
425 }
426 }
427
428 DBG("Closing all UST sockets");
429 ust_app_clean_list();
430
431 if (is_root && !opt_no_kernel) {
432 DBG2("Closing kernel fd");
433 if (kernel_tracer_fd >= 0) {
434 ret = close(kernel_tracer_fd);
435 if (ret) {
436 PERROR("close");
437 }
438 }
439 DBG("Unloading kernel modules");
440 modprobe_remove_lttng_all();
441 }
442
443 /* <fun> */
444 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
445 "Matthew, BEET driven development works!%c[%dm",
446 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
447 /* </fun> */
448 }
449
450 /*
451 * Send data on a unix socket using the liblttsessiondcomm API.
452 *
453 * Return lttcomm error code.
454 */
455 static int send_unix_sock(int sock, void *buf, size_t len)
456 {
457 /* Check valid length */
458 if (len == 0) {
459 return -1;
460 }
461
462 return lttcomm_send_unix_sock(sock, buf, len);
463 }
464
465 /*
466 * Free memory of a command context structure.
467 */
468 static void clean_command_ctx(struct command_ctx **cmd_ctx)
469 {
470 DBG("Clean command context structure");
471 if (*cmd_ctx) {
472 if ((*cmd_ctx)->llm) {
473 free((*cmd_ctx)->llm);
474 }
475 if ((*cmd_ctx)->lsm) {
476 free((*cmd_ctx)->lsm);
477 }
478 free(*cmd_ctx);
479 *cmd_ctx = NULL;
480 }
481 }
482
483 /*
484 * Notify UST applications using the shm mmap futex.
485 */
486 static int notify_ust_apps(int active)
487 {
488 char *wait_shm_mmap;
489
490 DBG("Notifying applications of session daemon state: %d", active);
491
492 /* See shm.c for this call implying mmap, shm and futex calls */
493 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
494 if (wait_shm_mmap == NULL) {
495 goto error;
496 }
497
498 /* Wake waiting process */
499 futex_wait_update((int32_t *) wait_shm_mmap, active);
500
501 /* Apps notified successfully */
502 return 0;
503
504 error:
505 return -1;
506 }
507
508 /*
509 * Setup the outgoing data buffer for the response (llm) by allocating the
510 * right amount of memory and copying the original information from the lsm
511 * structure.
512 *
513 * Return total size of the buffer pointed by buf.
514 */
515 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
516 {
517 int ret, buf_size;
518
519 buf_size = size;
520
521 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
522 if (cmd_ctx->llm == NULL) {
523 PERROR("zmalloc");
524 ret = -ENOMEM;
525 goto error;
526 }
527
528 /* Copy common data */
529 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
530 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
531
532 cmd_ctx->llm->data_size = size;
533 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
534
535 return buf_size;
536
537 error:
538 return ret;
539 }
540
541 /*
542 * Update the kernel poll set of all channel fd available over all tracing
543 * session. Add the wakeup pipe at the end of the set.
544 */
545 static int update_kernel_poll(struct lttng_poll_event *events)
546 {
547 int ret;
548 struct ltt_session *session;
549 struct ltt_kernel_channel *channel;
550
551 DBG("Updating kernel poll set");
552
553 session_lock_list();
554 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
555 session_lock(session);
556 if (session->kernel_session == NULL) {
557 session_unlock(session);
558 continue;
559 }
560
561 cds_list_for_each_entry(channel,
562 &session->kernel_session->channel_list.head, list) {
563 /* Add channel fd to the kernel poll set */
564 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
565 if (ret < 0) {
566 session_unlock(session);
567 goto error;
568 }
569 DBG("Channel fd %d added to kernel set", channel->fd);
570 }
571 session_unlock(session);
572 }
573 session_unlock_list();
574
575 return 0;
576
577 error:
578 session_unlock_list();
579 return -1;
580 }
581
582 /*
583 * Find the channel fd from 'fd' over all tracing session. When found, check
584 * for new channel stream and send those stream fds to the kernel consumer.
585 *
586 * Useful for CPU hotplug feature.
587 */
588 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
589 {
590 int ret = 0;
591 struct ltt_session *session;
592 struct ltt_kernel_session *ksess;
593 struct ltt_kernel_channel *channel;
594
595 DBG("Updating kernel streams for channel fd %d", fd);
596
597 session_lock_list();
598 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
599 session_lock(session);
600 if (session->kernel_session == NULL) {
601 session_unlock(session);
602 continue;
603 }
604 ksess = session->kernel_session;
605
606 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
607 if (channel->fd == fd) {
608 DBG("Channel found, updating kernel streams");
609 ret = kernel_open_channel_stream(channel);
610 if (ret < 0) {
611 goto error;
612 }
613
614 /*
615 * Have we already sent fds to the consumer? If yes, it means
616 * that tracing is started so it is safe to send our updated
617 * stream fds.
618 */
619 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
620 struct lttng_ht_iter iter;
621 struct consumer_socket *socket;
622
623 rcu_read_lock();
624 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
625 &iter.iter, socket, node.node) {
626 /* Code flow error */
627 assert(socket->fd >= 0);
628
629 pthread_mutex_lock(socket->lock);
630 ret = kernel_consumer_send_channel_stream(socket,
631 channel, ksess);
632 pthread_mutex_unlock(socket->lock);
633 if (ret < 0) {
634 rcu_read_unlock();
635 goto error;
636 }
637 }
638 rcu_read_unlock();
639 }
640 goto error;
641 }
642 }
643 session_unlock(session);
644 }
645 session_unlock_list();
646 return ret;
647
648 error:
649 session_unlock(session);
650 session_unlock_list();
651 return ret;
652 }
653
654 /*
655 * For each tracing session, update newly registered apps.
656 */
657 static void update_ust_app(int app_sock)
658 {
659 struct ltt_session *sess, *stmp;
660
661 session_lock_list();
662
663 /* For all tracing session(s) */
664 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
665 session_lock(sess);
666 if (sess->ust_session) {
667 ust_app_global_update(sess->ust_session, app_sock);
668 }
669 session_unlock(sess);
670 }
671
672 session_unlock_list();
673 }
674
675 /*
676 * This thread manage event coming from the kernel.
677 *
678 * Features supported in this thread:
679 * -) CPU Hotplug
680 */
681 static void *thread_manage_kernel(void *data)
682 {
683 int ret, i, pollfd, update_poll_flag = 1, err = -1;
684 uint32_t revents, nb_fd;
685 char tmp;
686 struct lttng_poll_event events;
687
688 DBG("[thread] Thread manage kernel started");
689
690 health_register(HEALTH_TYPE_KERNEL);
691
692 /*
693 * This first step of the while is to clean this structure which could free
694 * non NULL pointers so zero it before the loop.
695 */
696 memset(&events, 0, sizeof(events));
697
698 if (testpoint(thread_manage_kernel)) {
699 goto error_testpoint;
700 }
701
702 health_code_update();
703
704 if (testpoint(thread_manage_kernel_before_loop)) {
705 goto error_testpoint;
706 }
707
708 while (1) {
709 health_code_update();
710
711 if (update_poll_flag == 1) {
712 /* Clean events object. We are about to populate it again. */
713 lttng_poll_clean(&events);
714
715 ret = create_thread_poll_set(&events, 2);
716 if (ret < 0) {
717 goto error_poll_create;
718 }
719
720 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
721 if (ret < 0) {
722 goto error;
723 }
724
725 /* This will add the available kernel channel if any. */
726 ret = update_kernel_poll(&events);
727 if (ret < 0) {
728 goto error;
729 }
730 update_poll_flag = 0;
731 }
732
733 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
734
735 /* Poll infinite value of time */
736 restart:
737 health_poll_update();
738 ret = lttng_poll_wait(&events, -1);
739 health_poll_update();
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 nb_fd = ret;
756
757 for (i = 0; i < nb_fd; i++) {
758 /* Fetch once the poll data */
759 revents = LTTNG_POLL_GETEV(&events, i);
760 pollfd = LTTNG_POLL_GETFD(&events, i);
761
762 health_code_update();
763
764 /* Thread quit pipe has been closed. Killing thread. */
765 ret = check_thread_quit_pipe(pollfd, revents);
766 if (ret) {
767 err = 0;
768 goto exit;
769 }
770
771 /* Check for data on kernel pipe */
772 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
773 do {
774 ret = read(kernel_poll_pipe[0], &tmp, 1);
775 } while (ret < 0 && errno == EINTR);
776 /*
777 * Ret value is useless here, if this pipe gets any actions an
778 * update is required anyway.
779 */
780 update_poll_flag = 1;
781 continue;
782 } else {
783 /*
784 * New CPU detected by the kernel. Adding kernel stream to
785 * kernel session and updating the kernel consumer
786 */
787 if (revents & LPOLLIN) {
788 ret = update_kernel_stream(&kconsumer_data, pollfd);
789 if (ret < 0) {
790 continue;
791 }
792 break;
793 /*
794 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
795 * and unregister kernel stream at this point.
796 */
797 }
798 }
799 }
800 }
801
802 exit:
803 error:
804 lttng_poll_clean(&events);
805 error_poll_create:
806 error_testpoint:
807 utils_close_pipe(kernel_poll_pipe);
808 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
809 if (err) {
810 health_error();
811 ERR("Health error occurred in %s", __func__);
812 WARN("Kernel thread died unexpectedly. "
813 "Kernel tracing can continue but CPU hotplug is disabled.");
814 }
815 health_unregister();
816 DBG("Kernel thread dying");
817 return NULL;
818 }
819
820 /*
821 * Signal pthread condition of the consumer data that the thread.
822 */
823 static void signal_consumer_condition(struct consumer_data *data, int state)
824 {
825 pthread_mutex_lock(&data->cond_mutex);
826
827 /*
828 * The state is set before signaling. It can be any value, it's the waiter
829 * job to correctly interpret this condition variable associated to the
830 * consumer pthread_cond.
831 *
832 * A value of 0 means that the corresponding thread of the consumer data
833 * was not started. 1 indicates that the thread has started and is ready
834 * for action. A negative value means that there was an error during the
835 * thread bootstrap.
836 */
837 data->consumer_thread_is_ready = state;
838 (void) pthread_cond_signal(&data->cond);
839
840 pthread_mutex_unlock(&data->cond_mutex);
841 }
842
843 /*
844 * This thread manage the consumer error sent back to the session daemon.
845 */
846 static void *thread_manage_consumer(void *data)
847 {
848 int sock = -1, i, ret, pollfd, err = -1;
849 uint32_t revents, nb_fd;
850 enum lttcomm_return_code code;
851 struct lttng_poll_event events;
852 struct consumer_data *consumer_data = data;
853
854 DBG("[thread] Manage consumer started");
855
856 health_register(HEALTH_TYPE_CONSUMER);
857
858 /*
859 * Since the consumer thread can be spawned at any moment in time, we init
860 * the health to a poll status (1, which is a valid health over time).
861 * When the thread starts, we update here the health to a "code" path being
862 * an even value so this thread, when reaching a poll wait, does not
863 * trigger an error with an even value.
864 *
865 * Here is the use case we avoid.
866 *
867 * +1: the first poll update during initialization (main())
868 * +2 * x: multiple code update once in this thread.
869 * +1: poll wait in this thread (being a good health state).
870 * == even number which after the wait period shows as a bad health.
871 *
872 * In a nutshell, the following poll update to the health state brings back
873 * the state to an even value meaning a code path.
874 */
875 health_poll_update();
876
877 /*
878 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
879 * Nothing more will be added to this poll set.
880 */
881 ret = create_thread_poll_set(&events, 2);
882 if (ret < 0) {
883 goto error_poll;
884 }
885
886 /*
887 * The error socket here is already in a listening state which was done
888 * just before spawning this thread to avoid a race between the consumer
889 * daemon exec trying to connect and the listen() call.
890 */
891 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
892 if (ret < 0) {
893 goto error;
894 }
895
896 health_code_update();
897
898 /* Inifinite blocking call, waiting for transmission */
899 restart:
900 health_poll_update();
901
902 if (testpoint(thread_manage_consumer)) {
903 goto error;
904 }
905
906 ret = lttng_poll_wait(&events, -1);
907 health_poll_update();
908 if (ret < 0) {
909 /*
910 * Restart interrupted system call.
911 */
912 if (errno == EINTR) {
913 goto restart;
914 }
915 goto error;
916 }
917
918 nb_fd = ret;
919
920 for (i = 0; i < nb_fd; i++) {
921 /* Fetch once the poll data */
922 revents = LTTNG_POLL_GETEV(&events, i);
923 pollfd = LTTNG_POLL_GETFD(&events, i);
924
925 health_code_update();
926
927 /* Thread quit pipe has been closed. Killing thread. */
928 ret = check_thread_quit_pipe(pollfd, revents);
929 if (ret) {
930 err = 0;
931 goto exit;
932 }
933
934 /* Event on the registration socket */
935 if (pollfd == consumer_data->err_sock) {
936 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
937 ERR("consumer err socket poll error");
938 goto error;
939 }
940 }
941 }
942
943 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
944 if (sock < 0) {
945 goto error;
946 }
947
948 /*
949 * Set the CLOEXEC flag. Return code is useless because either way, the
950 * show must go on.
951 */
952 (void) utils_set_fd_cloexec(sock);
953
954 health_code_update();
955
956 DBG2("Receiving code from consumer err_sock");
957
958 /* Getting status code from kconsumerd */
959 ret = lttcomm_recv_unix_sock(sock, &code,
960 sizeof(enum lttcomm_return_code));
961 if (ret <= 0) {
962 goto error;
963 }
964
965 health_code_update();
966
967 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
968 consumer_data->cmd_sock =
969 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
970 if (consumer_data->cmd_sock < 0) {
971 /* On error, signal condition and quit. */
972 signal_consumer_condition(consumer_data, -1);
973 PERROR("consumer connect");
974 goto error;
975 }
976 signal_consumer_condition(consumer_data, 1);
977 DBG("Consumer command socket ready");
978 } else {
979 ERR("consumer error when waiting for SOCK_READY : %s",
980 lttcomm_get_readable_code(-code));
981 goto error;
982 }
983
984 /* Remove the kconsumerd error sock since we've established a connexion */
985 ret = lttng_poll_del(&events, consumer_data->err_sock);
986 if (ret < 0) {
987 goto error;
988 }
989
990 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
991 if (ret < 0) {
992 goto error;
993 }
994
995 health_code_update();
996
997 /* Inifinite blocking call, waiting for transmission */
998 restart_poll:
999 health_poll_update();
1000 ret = lttng_poll_wait(&events, -1);
1001 health_poll_update();
1002 if (ret < 0) {
1003 /*
1004 * Restart interrupted system call.
1005 */
1006 if (errno == EINTR) {
1007 goto restart_poll;
1008 }
1009 goto error;
1010 }
1011
1012 nb_fd = ret;
1013
1014 for (i = 0; i < nb_fd; i++) {
1015 /* Fetch once the poll data */
1016 revents = LTTNG_POLL_GETEV(&events, i);
1017 pollfd = LTTNG_POLL_GETFD(&events, i);
1018
1019 health_code_update();
1020
1021 /* Thread quit pipe has been closed. Killing thread. */
1022 ret = check_thread_quit_pipe(pollfd, revents);
1023 if (ret) {
1024 err = 0;
1025 goto exit;
1026 }
1027
1028 /* Event on the kconsumerd socket */
1029 if (pollfd == sock) {
1030 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1031 ERR("consumer err socket second poll error");
1032 goto error;
1033 }
1034 }
1035 }
1036
1037 health_code_update();
1038
1039 /* Wait for any kconsumerd error */
1040 ret = lttcomm_recv_unix_sock(sock, &code,
1041 sizeof(enum lttcomm_return_code));
1042 if (ret <= 0) {
1043 ERR("consumer closed the command socket");
1044 goto error;
1045 }
1046
1047 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1048
1049 exit:
1050 error:
1051 /* Immediately set the consumerd state to stopped */
1052 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1053 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1054 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1055 consumer_data->type == LTTNG_CONSUMER32_UST) {
1056 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1057 } else {
1058 /* Code flow error... */
1059 assert(0);
1060 }
1061
1062 if (consumer_data->err_sock >= 0) {
1063 ret = close(consumer_data->err_sock);
1064 if (ret) {
1065 PERROR("close");
1066 }
1067 }
1068 if (consumer_data->cmd_sock >= 0) {
1069 ret = close(consumer_data->cmd_sock);
1070 if (ret) {
1071 PERROR("close");
1072 }
1073 }
1074 if (sock >= 0) {
1075 ret = close(sock);
1076 if (ret) {
1077 PERROR("close");
1078 }
1079 }
1080
1081 unlink(consumer_data->err_unix_sock_path);
1082 unlink(consumer_data->cmd_unix_sock_path);
1083 consumer_data->pid = 0;
1084
1085 lttng_poll_clean(&events);
1086 error_poll:
1087 if (err) {
1088 health_error();
1089 ERR("Health error occurred in %s", __func__);
1090 }
1091 health_unregister();
1092 DBG("consumer thread cleanup completed");
1093
1094 return NULL;
1095 }
1096
1097 /*
1098 * This thread manage application communication.
1099 */
1100 static void *thread_manage_apps(void *data)
1101 {
1102 int i, ret, pollfd, err = -1;
1103 uint32_t revents, nb_fd;
1104 struct ust_command ust_cmd;
1105 struct lttng_poll_event events;
1106
1107 DBG("[thread] Manage application started");
1108
1109 rcu_register_thread();
1110 rcu_thread_online();
1111
1112 health_register(HEALTH_TYPE_APP_MANAGE);
1113
1114 if (testpoint(thread_manage_apps)) {
1115 goto error_testpoint;
1116 }
1117
1118 health_code_update();
1119
1120 ret = create_thread_poll_set(&events, 2);
1121 if (ret < 0) {
1122 goto error_poll_create;
1123 }
1124
1125 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1126 if (ret < 0) {
1127 goto error;
1128 }
1129
1130 if (testpoint(thread_manage_apps_before_loop)) {
1131 goto error;
1132 }
1133
1134 health_code_update();
1135
1136 while (1) {
1137 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1138
1139 /* Inifinite blocking call, waiting for transmission */
1140 restart:
1141 health_poll_update();
1142 ret = lttng_poll_wait(&events, -1);
1143 health_poll_update();
1144 if (ret < 0) {
1145 /*
1146 * Restart interrupted system call.
1147 */
1148 if (errno == EINTR) {
1149 goto restart;
1150 }
1151 goto error;
1152 }
1153
1154 nb_fd = ret;
1155
1156 for (i = 0; i < nb_fd; i++) {
1157 /* Fetch once the poll data */
1158 revents = LTTNG_POLL_GETEV(&events, i);
1159 pollfd = LTTNG_POLL_GETFD(&events, i);
1160
1161 health_code_update();
1162
1163 /* Thread quit pipe has been closed. Killing thread. */
1164 ret = check_thread_quit_pipe(pollfd, revents);
1165 if (ret) {
1166 err = 0;
1167 goto exit;
1168 }
1169
1170 /* Inspect the apps cmd pipe */
1171 if (pollfd == apps_cmd_pipe[0]) {
1172 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1173 ERR("Apps command pipe error");
1174 goto error;
1175 } else if (revents & LPOLLIN) {
1176 /* Empty pipe */
1177 do {
1178 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1179 } while (ret < 0 && errno == EINTR);
1180 if (ret < 0 || ret < sizeof(ust_cmd)) {
1181 PERROR("read apps cmd pipe");
1182 goto error;
1183 }
1184
1185 health_code_update();
1186
1187 /* Register applicaton to the session daemon */
1188 ret = ust_app_register(&ust_cmd.reg_msg,
1189 ust_cmd.sock);
1190 if (ret == -ENOMEM) {
1191 goto error;
1192 } else if (ret < 0) {
1193 break;
1194 }
1195
1196 health_code_update();
1197
1198 /*
1199 * Validate UST version compatibility.
1200 */
1201 ret = ust_app_validate_version(ust_cmd.sock);
1202 if (ret >= 0) {
1203 /*
1204 * Add channel(s) and event(s) to newly registered apps
1205 * from lttng global UST domain.
1206 */
1207 update_ust_app(ust_cmd.sock);
1208 }
1209
1210 health_code_update();
1211
1212 ret = ust_app_register_done(ust_cmd.sock);
1213 if (ret < 0) {
1214 /*
1215 * If the registration is not possible, we simply
1216 * unregister the apps and continue
1217 */
1218 ust_app_unregister(ust_cmd.sock);
1219 } else {
1220 /*
1221 * We only monitor the error events of the socket. This
1222 * thread does not handle any incoming data from UST
1223 * (POLLIN).
1224 */
1225 ret = lttng_poll_add(&events, ust_cmd.sock,
1226 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1227 if (ret < 0) {
1228 goto error;
1229 }
1230
1231 /* Set socket timeout for both receiving and ending */
1232 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1233 app_socket_timeout);
1234 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1235 app_socket_timeout);
1236
1237 DBG("Apps with sock %d added to poll set",
1238 ust_cmd.sock);
1239 }
1240
1241 health_code_update();
1242
1243 break;
1244 }
1245 } else {
1246 /*
1247 * At this point, we know that a registered application made
1248 * the event at poll_wait.
1249 */
1250 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1251 /* Removing from the poll set */
1252 ret = lttng_poll_del(&events, pollfd);
1253 if (ret < 0) {
1254 goto error;
1255 }
1256
1257 /* Socket closed on remote end. */
1258 ust_app_unregister(pollfd);
1259 break;
1260 }
1261 }
1262
1263 health_code_update();
1264 }
1265 }
1266
1267 exit:
1268 error:
1269 lttng_poll_clean(&events);
1270 error_poll_create:
1271 error_testpoint:
1272 utils_close_pipe(apps_cmd_pipe);
1273 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1274
1275 /*
1276 * We don't clean the UST app hash table here since already registered
1277 * applications can still be controlled so let them be until the session
1278 * daemon dies or the applications stop.
1279 */
1280
1281 if (err) {
1282 health_error();
1283 ERR("Health error occurred in %s", __func__);
1284 }
1285 health_unregister();
1286 DBG("Application communication apps thread cleanup complete");
1287 rcu_thread_offline();
1288 rcu_unregister_thread();
1289 return NULL;
1290 }
1291
1292 /*
1293 * Dispatch request from the registration threads to the application
1294 * communication thread.
1295 */
1296 static void *thread_dispatch_ust_registration(void *data)
1297 {
1298 int ret;
1299 struct cds_wfq_node *node;
1300 struct ust_command *ust_cmd = NULL;
1301
1302 DBG("[thread] Dispatch UST command started");
1303
1304 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1305 /* Atomically prepare the queue futex */
1306 futex_nto1_prepare(&ust_cmd_queue.futex);
1307
1308 do {
1309 /* Dequeue command for registration */
1310 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1311 if (node == NULL) {
1312 DBG("Woken up but nothing in the UST command queue");
1313 /* Continue thread execution */
1314 break;
1315 }
1316
1317 ust_cmd = caa_container_of(node, struct ust_command, node);
1318
1319 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1320 " gid:%d sock:%d name:%s (version %d.%d)",
1321 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1322 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1323 ust_cmd->sock, ust_cmd->reg_msg.name,
1324 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1325 /*
1326 * Inform apps thread of the new application registration. This
1327 * call is blocking so we can be assured that the data will be read
1328 * at some point in time or wait to the end of the world :)
1329 */
1330 if (apps_cmd_pipe[1] >= 0) {
1331 do {
1332 ret = write(apps_cmd_pipe[1], ust_cmd,
1333 sizeof(struct ust_command));
1334 } while (ret < 0 && errno == EINTR);
1335 if (ret < 0 || ret != sizeof(struct ust_command)) {
1336 PERROR("write apps cmd pipe");
1337 if (errno == EBADF) {
1338 /*
1339 * We can't inform the application thread to process
1340 * registration. We will exit or else application
1341 * registration will not occur and tracing will never
1342 * start.
1343 */
1344 goto error;
1345 }
1346 }
1347 } else {
1348 /* Application manager thread is not available. */
1349 ret = close(ust_cmd->sock);
1350 if (ret < 0) {
1351 PERROR("close ust_cmd sock");
1352 }
1353 }
1354 free(ust_cmd);
1355 } while (node != NULL);
1356
1357 /* Futex wait on queue. Blocking call on futex() */
1358 futex_nto1_wait(&ust_cmd_queue.futex);
1359 }
1360
1361 error:
1362 DBG("Dispatch thread dying");
1363 return NULL;
1364 }
1365
1366 /*
1367 * This thread manage application registration.
1368 */
1369 static void *thread_registration_apps(void *data)
1370 {
1371 int sock = -1, i, ret, pollfd, err = -1;
1372 uint32_t revents, nb_fd;
1373 struct lttng_poll_event events;
1374 /*
1375 * Get allocated in this thread, enqueued to a global queue, dequeued and
1376 * freed in the manage apps thread.
1377 */
1378 struct ust_command *ust_cmd = NULL;
1379
1380 DBG("[thread] Manage application registration started");
1381
1382 health_register(HEALTH_TYPE_APP_REG);
1383
1384 if (testpoint(thread_registration_apps)) {
1385 goto error_testpoint;
1386 }
1387
1388 ret = lttcomm_listen_unix_sock(apps_sock);
1389 if (ret < 0) {
1390 goto error_listen;
1391 }
1392
1393 /*
1394 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1395 * more will be added to this poll set.
1396 */
1397 ret = create_thread_poll_set(&events, 2);
1398 if (ret < 0) {
1399 goto error_create_poll;
1400 }
1401
1402 /* Add the application registration socket */
1403 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1404 if (ret < 0) {
1405 goto error_poll_add;
1406 }
1407
1408 /* Notify all applications to register */
1409 ret = notify_ust_apps(1);
1410 if (ret < 0) {
1411 ERR("Failed to notify applications or create the wait shared memory.\n"
1412 "Execution continues but there might be problem for already\n"
1413 "running applications that wishes to register.");
1414 }
1415
1416 while (1) {
1417 DBG("Accepting application registration");
1418
1419 /* Inifinite blocking call, waiting for transmission */
1420 restart:
1421 health_poll_update();
1422 ret = lttng_poll_wait(&events, -1);
1423 health_poll_update();
1424 if (ret < 0) {
1425 /*
1426 * Restart interrupted system call.
1427 */
1428 if (errno == EINTR) {
1429 goto restart;
1430 }
1431 goto error;
1432 }
1433
1434 nb_fd = ret;
1435
1436 for (i = 0; i < nb_fd; i++) {
1437 health_code_update();
1438
1439 /* Fetch once the poll data */
1440 revents = LTTNG_POLL_GETEV(&events, i);
1441 pollfd = LTTNG_POLL_GETFD(&events, i);
1442
1443 /* Thread quit pipe has been closed. Killing thread. */
1444 ret = check_thread_quit_pipe(pollfd, revents);
1445 if (ret) {
1446 err = 0;
1447 goto exit;
1448 }
1449
1450 /* Event on the registration socket */
1451 if (pollfd == apps_sock) {
1452 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1453 ERR("Register apps socket poll error");
1454 goto error;
1455 } else if (revents & LPOLLIN) {
1456 sock = lttcomm_accept_unix_sock(apps_sock);
1457 if (sock < 0) {
1458 goto error;
1459 }
1460
1461 /*
1462 * Set the CLOEXEC flag. Return code is useless because
1463 * either way, the show must go on.
1464 */
1465 (void) utils_set_fd_cloexec(sock);
1466
1467 /* Create UST registration command for enqueuing */
1468 ust_cmd = zmalloc(sizeof(struct ust_command));
1469 if (ust_cmd == NULL) {
1470 PERROR("ust command zmalloc");
1471 goto error;
1472 }
1473
1474 /*
1475 * Using message-based transmissions to ensure we don't
1476 * have to deal with partially received messages.
1477 */
1478 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1479 if (ret < 0) {
1480 ERR("Exhausted file descriptors allowed for applications.");
1481 free(ust_cmd);
1482 ret = close(sock);
1483 if (ret) {
1484 PERROR("close");
1485 }
1486 sock = -1;
1487 continue;
1488 }
1489 health_code_update();
1490 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1491 sizeof(struct ust_register_msg));
1492 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1493 if (ret < 0) {
1494 PERROR("lttcomm_recv_unix_sock register apps");
1495 } else {
1496 ERR("Wrong size received on apps register");
1497 }
1498 free(ust_cmd);
1499 ret = close(sock);
1500 if (ret) {
1501 PERROR("close");
1502 }
1503 lttng_fd_put(LTTNG_FD_APPS, 1);
1504 sock = -1;
1505 continue;
1506 }
1507 health_code_update();
1508
1509 ust_cmd->sock = sock;
1510 sock = -1;
1511
1512 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1513 " gid:%d sock:%d name:%s (version %d.%d)",
1514 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1515 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1516 ust_cmd->sock, ust_cmd->reg_msg.name,
1517 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1518
1519 /*
1520 * Lock free enqueue the registration request. The red pill
1521 * has been taken! This apps will be part of the *system*.
1522 */
1523 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1524
1525 /*
1526 * Wake the registration queue futex. Implicit memory
1527 * barrier with the exchange in cds_wfq_enqueue.
1528 */
1529 futex_nto1_wake(&ust_cmd_queue.futex);
1530 }
1531 }
1532 }
1533 }
1534
1535 exit:
1536 error:
1537 if (err) {
1538 health_error();
1539 ERR("Health error occurred in %s", __func__);
1540 }
1541
1542 /* Notify that the registration thread is gone */
1543 notify_ust_apps(0);
1544
1545 if (apps_sock >= 0) {
1546 ret = close(apps_sock);
1547 if (ret) {
1548 PERROR("close");
1549 }
1550 }
1551 if (sock >= 0) {
1552 ret = close(sock);
1553 if (ret) {
1554 PERROR("close");
1555 }
1556 lttng_fd_put(LTTNG_FD_APPS, 1);
1557 }
1558 unlink(apps_unix_sock_path);
1559
1560 error_poll_add:
1561 lttng_poll_clean(&events);
1562 error_listen:
1563 error_create_poll:
1564 error_testpoint:
1565 DBG("UST Registration thread cleanup complete");
1566 health_unregister();
1567
1568 return NULL;
1569 }
1570
1571 /*
1572 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1573 * exec or it will fails.
1574 */
1575 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1576 {
1577 int ret, clock_ret;
1578 struct timespec timeout;
1579
1580 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1581 consumer_data->consumer_thread_is_ready = 0;
1582
1583 /* Setup pthread condition */
1584 ret = pthread_condattr_init(&consumer_data->condattr);
1585 if (ret != 0) {
1586 errno = ret;
1587 PERROR("pthread_condattr_init consumer data");
1588 goto error;
1589 }
1590
1591 /*
1592 * Set the monotonic clock in order to make sure we DO NOT jump in time
1593 * between the clock_gettime() call and the timedwait call. See bug #324
1594 * for a more details and how we noticed it.
1595 */
1596 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1597 if (ret != 0) {
1598 errno = ret;
1599 PERROR("pthread_condattr_setclock consumer data");
1600 goto error;
1601 }
1602
1603 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1604 if (ret != 0) {
1605 errno = ret;
1606 PERROR("pthread_cond_init consumer data");
1607 goto error;
1608 }
1609
1610 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1611 consumer_data);
1612 if (ret != 0) {
1613 PERROR("pthread_create consumer");
1614 ret = -1;
1615 goto error;
1616 }
1617
1618 /* We are about to wait on a pthread condition */
1619 pthread_mutex_lock(&consumer_data->cond_mutex);
1620
1621 /* Get time for sem_timedwait absolute timeout */
1622 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1623 /*
1624 * Set the timeout for the condition timed wait even if the clock gettime
1625 * call fails since we might loop on that call and we want to avoid to
1626 * increment the timeout too many times.
1627 */
1628 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1629
1630 /*
1631 * The following loop COULD be skipped in some conditions so this is why we
1632 * set ret to 0 in order to make sure at least one round of the loop is
1633 * done.
1634 */
1635 ret = 0;
1636
1637 /*
1638 * Loop until the condition is reached or when a timeout is reached. Note
1639 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1640 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1641 * possible. This loop does not take any chances and works with both of
1642 * them.
1643 */
1644 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1645 if (clock_ret < 0) {
1646 PERROR("clock_gettime spawn consumer");
1647 /* Infinite wait for the consumerd thread to be ready */
1648 ret = pthread_cond_wait(&consumer_data->cond,
1649 &consumer_data->cond_mutex);
1650 } else {
1651 ret = pthread_cond_timedwait(&consumer_data->cond,
1652 &consumer_data->cond_mutex, &timeout);
1653 }
1654 }
1655
1656 /* Release the pthread condition */
1657 pthread_mutex_unlock(&consumer_data->cond_mutex);
1658
1659 if (ret != 0) {
1660 errno = ret;
1661 if (ret == ETIMEDOUT) {
1662 /*
1663 * Call has timed out so we kill the kconsumerd_thread and return
1664 * an error.
1665 */
1666 ERR("Condition timed out. The consumer thread was never ready."
1667 " Killing it");
1668 ret = pthread_cancel(consumer_data->thread);
1669 if (ret < 0) {
1670 PERROR("pthread_cancel consumer thread");
1671 }
1672 } else {
1673 PERROR("pthread_cond_wait failed consumer thread");
1674 }
1675 goto error;
1676 }
1677
1678 pthread_mutex_lock(&consumer_data->pid_mutex);
1679 if (consumer_data->pid == 0) {
1680 ERR("Consumerd did not start");
1681 pthread_mutex_unlock(&consumer_data->pid_mutex);
1682 goto error;
1683 }
1684 pthread_mutex_unlock(&consumer_data->pid_mutex);
1685
1686 return 0;
1687
1688 error:
1689 return ret;
1690 }
1691
1692 /*
1693 * Join consumer thread
1694 */
1695 static int join_consumer_thread(struct consumer_data *consumer_data)
1696 {
1697 void *status;
1698
1699 /* Consumer pid must be a real one. */
1700 if (consumer_data->pid > 0) {
1701 int ret;
1702 ret = kill(consumer_data->pid, SIGTERM);
1703 if (ret) {
1704 ERR("Error killing consumer daemon");
1705 return ret;
1706 }
1707 return pthread_join(consumer_data->thread, &status);
1708 } else {
1709 return 0;
1710 }
1711 }
1712
1713 /*
1714 * Fork and exec a consumer daemon (consumerd).
1715 *
1716 * Return pid if successful else -1.
1717 */
1718 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1719 {
1720 int ret;
1721 pid_t pid;
1722 const char *consumer_to_use;
1723 const char *verbosity;
1724 struct stat st;
1725
1726 DBG("Spawning consumerd");
1727
1728 pid = fork();
1729 if (pid == 0) {
1730 /*
1731 * Exec consumerd.
1732 */
1733 if (opt_verbose_consumer) {
1734 verbosity = "--verbose";
1735 } else {
1736 verbosity = "--quiet";
1737 }
1738 switch (consumer_data->type) {
1739 case LTTNG_CONSUMER_KERNEL:
1740 /*
1741 * Find out which consumerd to execute. We will first try the
1742 * 64-bit path, then the sessiond's installation directory, and
1743 * fallback on the 32-bit one,
1744 */
1745 DBG3("Looking for a kernel consumer at these locations:");
1746 DBG3(" 1) %s", consumerd64_bin);
1747 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1748 DBG3(" 3) %s", consumerd32_bin);
1749 if (stat(consumerd64_bin, &st) == 0) {
1750 DBG3("Found location #1");
1751 consumer_to_use = consumerd64_bin;
1752 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1753 DBG3("Found location #2");
1754 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1755 } else if (stat(consumerd32_bin, &st) == 0) {
1756 DBG3("Found location #3");
1757 consumer_to_use = consumerd32_bin;
1758 } else {
1759 DBG("Could not find any valid consumerd executable");
1760 break;
1761 }
1762 DBG("Using kernel consumer at: %s", consumer_to_use);
1763 execl(consumer_to_use,
1764 "lttng-consumerd", verbosity, "-k",
1765 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1766 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1767 NULL);
1768 break;
1769 case LTTNG_CONSUMER64_UST:
1770 {
1771 char *tmpnew = NULL;
1772
1773 if (consumerd64_libdir[0] != '\0') {
1774 char *tmp;
1775 size_t tmplen;
1776
1777 tmp = getenv("LD_LIBRARY_PATH");
1778 if (!tmp) {
1779 tmp = "";
1780 }
1781 tmplen = strlen("LD_LIBRARY_PATH=")
1782 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1783 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1784 if (!tmpnew) {
1785 ret = -ENOMEM;
1786 goto error;
1787 }
1788 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1789 strcat(tmpnew, consumerd64_libdir);
1790 if (tmp[0] != '\0') {
1791 strcat(tmpnew, ":");
1792 strcat(tmpnew, tmp);
1793 }
1794 ret = putenv(tmpnew);
1795 if (ret) {
1796 ret = -errno;
1797 goto error;
1798 }
1799 }
1800 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1801 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1802 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1803 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1804 NULL);
1805 if (consumerd64_libdir[0] != '\0') {
1806 free(tmpnew);
1807 }
1808 if (ret) {
1809 goto error;
1810 }
1811 break;
1812 }
1813 case LTTNG_CONSUMER32_UST:
1814 {
1815 char *tmpnew = NULL;
1816
1817 if (consumerd32_libdir[0] != '\0') {
1818 char *tmp;
1819 size_t tmplen;
1820
1821 tmp = getenv("LD_LIBRARY_PATH");
1822 if (!tmp) {
1823 tmp = "";
1824 }
1825 tmplen = strlen("LD_LIBRARY_PATH=")
1826 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1827 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1828 if (!tmpnew) {
1829 ret = -ENOMEM;
1830 goto error;
1831 }
1832 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1833 strcat(tmpnew, consumerd32_libdir);
1834 if (tmp[0] != '\0') {
1835 strcat(tmpnew, ":");
1836 strcat(tmpnew, tmp);
1837 }
1838 ret = putenv(tmpnew);
1839 if (ret) {
1840 ret = -errno;
1841 goto error;
1842 }
1843 }
1844 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1845 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1846 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1847 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1848 NULL);
1849 if (consumerd32_libdir[0] != '\0') {
1850 free(tmpnew);
1851 }
1852 if (ret) {
1853 goto error;
1854 }
1855 break;
1856 }
1857 default:
1858 PERROR("unknown consumer type");
1859 exit(EXIT_FAILURE);
1860 }
1861 if (errno != 0) {
1862 PERROR("kernel start consumer exec");
1863 }
1864 exit(EXIT_FAILURE);
1865 } else if (pid > 0) {
1866 ret = pid;
1867 } else {
1868 PERROR("start consumer fork");
1869 ret = -errno;
1870 }
1871 error:
1872 return ret;
1873 }
1874
1875 /*
1876 * Spawn the consumerd daemon and session daemon thread.
1877 */
1878 static int start_consumerd(struct consumer_data *consumer_data)
1879 {
1880 int ret;
1881
1882 /*
1883 * Set the listen() state on the socket since there is a possible race
1884 * between the exec() of the consumer daemon and this call if place in the
1885 * consumer thread. See bug #366 for more details.
1886 */
1887 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1888 if (ret < 0) {
1889 goto error;
1890 }
1891
1892 pthread_mutex_lock(&consumer_data->pid_mutex);
1893 if (consumer_data->pid != 0) {
1894 pthread_mutex_unlock(&consumer_data->pid_mutex);
1895 goto end;
1896 }
1897
1898 ret = spawn_consumerd(consumer_data);
1899 if (ret < 0) {
1900 ERR("Spawning consumerd failed");
1901 pthread_mutex_unlock(&consumer_data->pid_mutex);
1902 goto error;
1903 }
1904
1905 /* Setting up the consumer_data pid */
1906 consumer_data->pid = ret;
1907 DBG2("Consumer pid %d", consumer_data->pid);
1908 pthread_mutex_unlock(&consumer_data->pid_mutex);
1909
1910 DBG2("Spawning consumer control thread");
1911 ret = spawn_consumer_thread(consumer_data);
1912 if (ret < 0) {
1913 ERR("Fatal error spawning consumer control thread");
1914 goto error;
1915 }
1916
1917 end:
1918 return 0;
1919
1920 error:
1921 /* Cleanup already created socket on error. */
1922 if (consumer_data->err_sock >= 0) {
1923 int err;
1924
1925 err = close(consumer_data->err_sock);
1926 if (err < 0) {
1927 PERROR("close consumer data error socket");
1928 }
1929 }
1930 return ret;
1931 }
1932
1933 /*
1934 * Compute health status of each consumer. If one of them is zero (bad
1935 * state), we return 0.
1936 */
1937 static int check_consumer_health(void)
1938 {
1939 int ret;
1940
1941 ret = health_check_state(HEALTH_TYPE_CONSUMER);
1942
1943 DBG3("Health consumer check %d", ret);
1944
1945 return ret;
1946 }
1947
1948 /*
1949 * Setup necessary data for kernel tracer action.
1950 */
1951 static int init_kernel_tracer(void)
1952 {
1953 int ret;
1954
1955 /* Modprobe lttng kernel modules */
1956 ret = modprobe_lttng_control();
1957 if (ret < 0) {
1958 goto error;
1959 }
1960
1961 /* Open debugfs lttng */
1962 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1963 if (kernel_tracer_fd < 0) {
1964 DBG("Failed to open %s", module_proc_lttng);
1965 ret = -1;
1966 goto error_open;
1967 }
1968
1969 /* Validate kernel version */
1970 ret = kernel_validate_version(kernel_tracer_fd);
1971 if (ret < 0) {
1972 goto error_version;
1973 }
1974
1975 ret = modprobe_lttng_data();
1976 if (ret < 0) {
1977 goto error_modules;
1978 }
1979
1980 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1981 return 0;
1982
1983 error_version:
1984 modprobe_remove_lttng_control();
1985 ret = close(kernel_tracer_fd);
1986 if (ret) {
1987 PERROR("close");
1988 }
1989 kernel_tracer_fd = -1;
1990 return LTTNG_ERR_KERN_VERSION;
1991
1992 error_modules:
1993 ret = close(kernel_tracer_fd);
1994 if (ret) {
1995 PERROR("close");
1996 }
1997
1998 error_open:
1999 modprobe_remove_lttng_control();
2000
2001 error:
2002 WARN("No kernel tracer available");
2003 kernel_tracer_fd = -1;
2004 if (!is_root) {
2005 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2006 } else {
2007 return LTTNG_ERR_KERN_NA;
2008 }
2009 }
2010
2011
2012 /*
2013 * Copy consumer output from the tracing session to the domain session. The
2014 * function also applies the right modification on a per domain basis for the
2015 * trace files destination directory.
2016 */
2017 static int copy_session_consumer(int domain, struct ltt_session *session)
2018 {
2019 int ret;
2020 const char *dir_name;
2021 struct consumer_output *consumer;
2022
2023 assert(session);
2024 assert(session->consumer);
2025
2026 switch (domain) {
2027 case LTTNG_DOMAIN_KERNEL:
2028 DBG3("Copying tracing session consumer output in kernel session");
2029 /*
2030 * XXX: We should audit the session creation and what this function
2031 * does "extra" in order to avoid a destroy since this function is used
2032 * in the domain session creation (kernel and ust) only. Same for UST
2033 * domain.
2034 */
2035 if (session->kernel_session->consumer) {
2036 consumer_destroy_output(session->kernel_session->consumer);
2037 }
2038 session->kernel_session->consumer =
2039 consumer_copy_output(session->consumer);
2040 /* Ease our life a bit for the next part */
2041 consumer = session->kernel_session->consumer;
2042 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2043 break;
2044 case LTTNG_DOMAIN_UST:
2045 DBG3("Copying tracing session consumer output in UST session");
2046 if (session->ust_session->consumer) {
2047 consumer_destroy_output(session->ust_session->consumer);
2048 }
2049 session->ust_session->consumer =
2050 consumer_copy_output(session->consumer);
2051 /* Ease our life a bit for the next part */
2052 consumer = session->ust_session->consumer;
2053 dir_name = DEFAULT_UST_TRACE_DIR;
2054 break;
2055 default:
2056 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2057 goto error;
2058 }
2059
2060 /* Append correct directory to subdir */
2061 strncat(consumer->subdir, dir_name,
2062 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2063 DBG3("Copy session consumer subdir %s", consumer->subdir);
2064
2065 ret = LTTNG_OK;
2066
2067 error:
2068 return ret;
2069 }
2070
2071 /*
2072 * Create an UST session and add it to the session ust list.
2073 */
2074 static int create_ust_session(struct ltt_session *session,
2075 struct lttng_domain *domain)
2076 {
2077 int ret;
2078 struct ltt_ust_session *lus = NULL;
2079
2080 assert(session);
2081 assert(domain);
2082 assert(session->consumer);
2083
2084 switch (domain->type) {
2085 case LTTNG_DOMAIN_UST:
2086 break;
2087 default:
2088 ERR("Unknown UST domain on create session %d", domain->type);
2089 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2090 goto error;
2091 }
2092
2093 DBG("Creating UST session");
2094
2095 lus = trace_ust_create_session(session->path, session->id, domain);
2096 if (lus == NULL) {
2097 ret = LTTNG_ERR_UST_SESS_FAIL;
2098 goto error;
2099 }
2100
2101 lus->uid = session->uid;
2102 lus->gid = session->gid;
2103 session->ust_session = lus;
2104
2105 /* Copy session output to the newly created UST session */
2106 ret = copy_session_consumer(domain->type, session);
2107 if (ret != LTTNG_OK) {
2108 goto error;
2109 }
2110
2111 return LTTNG_OK;
2112
2113 error:
2114 free(lus);
2115 session->ust_session = NULL;
2116 return ret;
2117 }
2118
2119 /*
2120 * Create a kernel tracer session then create the default channel.
2121 */
2122 static int create_kernel_session(struct ltt_session *session)
2123 {
2124 int ret;
2125
2126 DBG("Creating kernel session");
2127
2128 ret = kernel_create_session(session, kernel_tracer_fd);
2129 if (ret < 0) {
2130 ret = LTTNG_ERR_KERN_SESS_FAIL;
2131 goto error;
2132 }
2133
2134 /* Code flow safety */
2135 assert(session->kernel_session);
2136
2137 /* Copy session output to the newly created Kernel session */
2138 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2139 if (ret != LTTNG_OK) {
2140 goto error;
2141 }
2142
2143 /* Create directory(ies) on local filesystem. */
2144 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2145 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2146 ret = run_as_mkdir_recursive(
2147 session->kernel_session->consumer->dst.trace_path,
2148 S_IRWXU | S_IRWXG, session->uid, session->gid);
2149 if (ret < 0) {
2150 if (ret != -EEXIST) {
2151 ERR("Trace directory creation error");
2152 goto error;
2153 }
2154 }
2155 }
2156
2157 session->kernel_session->uid = session->uid;
2158 session->kernel_session->gid = session->gid;
2159
2160 return LTTNG_OK;
2161
2162 error:
2163 trace_kernel_destroy_session(session->kernel_session);
2164 session->kernel_session = NULL;
2165 return ret;
2166 }
2167
2168 /*
2169 * Count number of session permitted by uid/gid.
2170 */
2171 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2172 {
2173 unsigned int i = 0;
2174 struct ltt_session *session;
2175
2176 DBG("Counting number of available session for UID %d GID %d",
2177 uid, gid);
2178 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2179 /*
2180 * Only list the sessions the user can control.
2181 */
2182 if (!session_access_ok(session, uid, gid)) {
2183 continue;
2184 }
2185 i++;
2186 }
2187 return i;
2188 }
2189
2190 /*
2191 * Process the command requested by the lttng client within the command
2192 * context structure. This function make sure that the return structure (llm)
2193 * is set and ready for transmission before returning.
2194 *
2195 * Return any error encountered or 0 for success.
2196 *
2197 * "sock" is only used for special-case var. len data.
2198 */
2199 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2200 int *sock_error)
2201 {
2202 int ret = LTTNG_OK;
2203 int need_tracing_session = 1;
2204 int need_domain;
2205
2206 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2207
2208 *sock_error = 0;
2209
2210 switch (cmd_ctx->lsm->cmd_type) {
2211 case LTTNG_CREATE_SESSION:
2212 case LTTNG_DESTROY_SESSION:
2213 case LTTNG_LIST_SESSIONS:
2214 case LTTNG_LIST_DOMAINS:
2215 case LTTNG_START_TRACE:
2216 case LTTNG_STOP_TRACE:
2217 case LTTNG_DATA_PENDING:
2218 need_domain = 0;
2219 break;
2220 default:
2221 need_domain = 1;
2222 }
2223
2224 if (opt_no_kernel && need_domain
2225 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2226 if (!is_root) {
2227 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2228 } else {
2229 ret = LTTNG_ERR_KERN_NA;
2230 }
2231 goto error;
2232 }
2233
2234 /* Deny register consumer if we already have a spawned consumer. */
2235 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2236 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2237 if (kconsumer_data.pid > 0) {
2238 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2239 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2240 goto error;
2241 }
2242 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2243 }
2244
2245 /*
2246 * Check for command that don't needs to allocate a returned payload. We do
2247 * this here so we don't have to make the call for no payload at each
2248 * command.
2249 */
2250 switch(cmd_ctx->lsm->cmd_type) {
2251 case LTTNG_LIST_SESSIONS:
2252 case LTTNG_LIST_TRACEPOINTS:
2253 case LTTNG_LIST_TRACEPOINT_FIELDS:
2254 case LTTNG_LIST_DOMAINS:
2255 case LTTNG_LIST_CHANNELS:
2256 case LTTNG_LIST_EVENTS:
2257 break;
2258 default:
2259 /* Setup lttng message with no payload */
2260 ret = setup_lttng_msg(cmd_ctx, 0);
2261 if (ret < 0) {
2262 /* This label does not try to unlock the session */
2263 goto init_setup_error;
2264 }
2265 }
2266
2267 /* Commands that DO NOT need a session. */
2268 switch (cmd_ctx->lsm->cmd_type) {
2269 case LTTNG_CREATE_SESSION:
2270 case LTTNG_CALIBRATE:
2271 case LTTNG_LIST_SESSIONS:
2272 case LTTNG_LIST_TRACEPOINTS:
2273 case LTTNG_LIST_TRACEPOINT_FIELDS:
2274 need_tracing_session = 0;
2275 break;
2276 default:
2277 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2278 /*
2279 * We keep the session list lock across _all_ commands
2280 * for now, because the per-session lock does not
2281 * handle teardown properly.
2282 */
2283 session_lock_list();
2284 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2285 if (cmd_ctx->session == NULL) {
2286 if (cmd_ctx->lsm->session.name != NULL) {
2287 ret = LTTNG_ERR_SESS_NOT_FOUND;
2288 } else {
2289 /* If no session name specified */
2290 ret = LTTNG_ERR_SELECT_SESS;
2291 }
2292 goto error;
2293 } else {
2294 /* Acquire lock for the session */
2295 session_lock(cmd_ctx->session);
2296 }
2297 break;
2298 }
2299
2300 if (!need_domain) {
2301 goto skip_domain;
2302 }
2303
2304 /*
2305 * Check domain type for specific "pre-action".
2306 */
2307 switch (cmd_ctx->lsm->domain.type) {
2308 case LTTNG_DOMAIN_KERNEL:
2309 if (!is_root) {
2310 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2311 goto error;
2312 }
2313
2314 /* Kernel tracer check */
2315 if (kernel_tracer_fd == -1) {
2316 /* Basically, load kernel tracer modules */
2317 ret = init_kernel_tracer();
2318 if (ret != 0) {
2319 goto error;
2320 }
2321 }
2322
2323 /* Consumer is in an ERROR state. Report back to client */
2324 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2325 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2326 goto error;
2327 }
2328
2329 /* Need a session for kernel command */
2330 if (need_tracing_session) {
2331 if (cmd_ctx->session->kernel_session == NULL) {
2332 ret = create_kernel_session(cmd_ctx->session);
2333 if (ret < 0) {
2334 ret = LTTNG_ERR_KERN_SESS_FAIL;
2335 goto error;
2336 }
2337 }
2338
2339 /* Start the kernel consumer daemon */
2340 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2341 if (kconsumer_data.pid == 0 &&
2342 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2343 cmd_ctx->session->start_consumer) {
2344 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2345 ret = start_consumerd(&kconsumer_data);
2346 if (ret < 0) {
2347 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2348 goto error;
2349 }
2350 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2351 } else {
2352 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2353 }
2354
2355 /*
2356 * The consumer was just spawned so we need to add the socket to
2357 * the consumer output of the session if exist.
2358 */
2359 ret = consumer_create_socket(&kconsumer_data,
2360 cmd_ctx->session->kernel_session->consumer);
2361 if (ret < 0) {
2362 goto error;
2363 }
2364 }
2365
2366 break;
2367 case LTTNG_DOMAIN_UST:
2368 {
2369 /* Consumer is in an ERROR state. Report back to client */
2370 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2371 ret = LTTNG_ERR_NO_USTCONSUMERD;
2372 goto error;
2373 }
2374
2375 if (need_tracing_session) {
2376 /* Create UST session if none exist. */
2377 if (cmd_ctx->session->ust_session == NULL) {
2378 ret = create_ust_session(cmd_ctx->session,
2379 &cmd_ctx->lsm->domain);
2380 if (ret != LTTNG_OK) {
2381 goto error;
2382 }
2383 }
2384
2385 /* Start the UST consumer daemons */
2386 /* 64-bit */
2387 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2388 if (consumerd64_bin[0] != '\0' &&
2389 ustconsumer64_data.pid == 0 &&
2390 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2391 cmd_ctx->session->start_consumer) {
2392 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2393 ret = start_consumerd(&ustconsumer64_data);
2394 if (ret < 0) {
2395 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2396 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2397 goto error;
2398 }
2399
2400 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2401 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2402 } else {
2403 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2404 }
2405
2406 /*
2407 * Setup socket for consumer 64 bit. No need for atomic access
2408 * since it was set above and can ONLY be set in this thread.
2409 */
2410 ret = consumer_create_socket(&ustconsumer64_data,
2411 cmd_ctx->session->ust_session->consumer);
2412 if (ret < 0) {
2413 goto error;
2414 }
2415
2416 /* 32-bit */
2417 if (consumerd32_bin[0] != '\0' &&
2418 ustconsumer32_data.pid == 0 &&
2419 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2420 cmd_ctx->session->start_consumer) {
2421 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2422 ret = start_consumerd(&ustconsumer32_data);
2423 if (ret < 0) {
2424 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2425 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2426 goto error;
2427 }
2428
2429 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2430 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2431 } else {
2432 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2433 }
2434
2435 /*
2436 * Setup socket for consumer 64 bit. No need for atomic access
2437 * since it was set above and can ONLY be set in this thread.
2438 */
2439 ret = consumer_create_socket(&ustconsumer32_data,
2440 cmd_ctx->session->ust_session->consumer);
2441 if (ret < 0) {
2442 goto error;
2443 }
2444 }
2445 break;
2446 }
2447 default:
2448 break;
2449 }
2450 skip_domain:
2451
2452 /* Validate consumer daemon state when start/stop trace command */
2453 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2454 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2455 switch (cmd_ctx->lsm->domain.type) {
2456 case LTTNG_DOMAIN_UST:
2457 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2458 ret = LTTNG_ERR_NO_USTCONSUMERD;
2459 goto error;
2460 }
2461 break;
2462 case LTTNG_DOMAIN_KERNEL:
2463 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2464 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2465 goto error;
2466 }
2467 break;
2468 }
2469 }
2470
2471 /*
2472 * Check that the UID or GID match that of the tracing session.
2473 * The root user can interact with all sessions.
2474 */
2475 if (need_tracing_session) {
2476 if (!session_access_ok(cmd_ctx->session,
2477 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2478 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2479 ret = LTTNG_ERR_EPERM;
2480 goto error;
2481 }
2482 }
2483
2484 /* Process by command type */
2485 switch (cmd_ctx->lsm->cmd_type) {
2486 case LTTNG_ADD_CONTEXT:
2487 {
2488 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2489 cmd_ctx->lsm->u.context.channel_name,
2490 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2491 break;
2492 }
2493 case LTTNG_DISABLE_CHANNEL:
2494 {
2495 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2496 cmd_ctx->lsm->u.disable.channel_name);
2497 break;
2498 }
2499 case LTTNG_DISABLE_EVENT:
2500 {
2501 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2502 cmd_ctx->lsm->u.disable.channel_name,
2503 cmd_ctx->lsm->u.disable.name);
2504 break;
2505 }
2506 case LTTNG_DISABLE_ALL_EVENT:
2507 {
2508 DBG("Disabling all events");
2509
2510 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2511 cmd_ctx->lsm->u.disable.channel_name);
2512 break;
2513 }
2514 case LTTNG_DISABLE_CONSUMER:
2515 {
2516 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2517 break;
2518 }
2519 case LTTNG_ENABLE_CHANNEL:
2520 {
2521 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2522 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2523 break;
2524 }
2525 case LTTNG_ENABLE_CONSUMER:
2526 {
2527 /*
2528 * XXX: 0 means that this URI should be applied on the session. Should
2529 * be a DOMAIN enuam.
2530 */
2531 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2532 if (ret != LTTNG_OK) {
2533 goto error;
2534 }
2535
2536 if (cmd_ctx->lsm->domain.type == 0) {
2537 /* Add the URI for the UST session if a consumer is present. */
2538 if (cmd_ctx->session->ust_session &&
2539 cmd_ctx->session->ust_session->consumer) {
2540 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2541 } else if (cmd_ctx->session->kernel_session &&
2542 cmd_ctx->session->kernel_session->consumer) {
2543 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2544 cmd_ctx->session);
2545 }
2546 }
2547 break;
2548 }
2549 case LTTNG_ENABLE_EVENT:
2550 {
2551 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2552 cmd_ctx->lsm->u.enable.channel_name,
2553 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2554 break;
2555 }
2556 case LTTNG_ENABLE_ALL_EVENT:
2557 {
2558 DBG("Enabling all events");
2559
2560 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2561 cmd_ctx->lsm->u.enable.channel_name,
2562 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2563 break;
2564 }
2565 case LTTNG_LIST_TRACEPOINTS:
2566 {
2567 struct lttng_event *events;
2568 ssize_t nb_events;
2569
2570 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2571 if (nb_events < 0) {
2572 /* Return value is a negative lttng_error_code. */
2573 ret = -nb_events;
2574 goto error;
2575 }
2576
2577 /*
2578 * Setup lttng message with payload size set to the event list size in
2579 * bytes and then copy list into the llm payload.
2580 */
2581 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2582 if (ret < 0) {
2583 free(events);
2584 goto setup_error;
2585 }
2586
2587 /* Copy event list into message payload */
2588 memcpy(cmd_ctx->llm->payload, events,
2589 sizeof(struct lttng_event) * nb_events);
2590
2591 free(events);
2592
2593 ret = LTTNG_OK;
2594 break;
2595 }
2596 case LTTNG_LIST_TRACEPOINT_FIELDS:
2597 {
2598 struct lttng_event_field *fields;
2599 ssize_t nb_fields;
2600
2601 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2602 &fields);
2603 if (nb_fields < 0) {
2604 /* Return value is a negative lttng_error_code. */
2605 ret = -nb_fields;
2606 goto error;
2607 }
2608
2609 /*
2610 * Setup lttng message with payload size set to the event list size in
2611 * bytes and then copy list into the llm payload.
2612 */
2613 ret = setup_lttng_msg(cmd_ctx,
2614 sizeof(struct lttng_event_field) * nb_fields);
2615 if (ret < 0) {
2616 free(fields);
2617 goto setup_error;
2618 }
2619
2620 /* Copy event list into message payload */
2621 memcpy(cmd_ctx->llm->payload, fields,
2622 sizeof(struct lttng_event_field) * nb_fields);
2623
2624 free(fields);
2625
2626 ret = LTTNG_OK;
2627 break;
2628 }
2629 case LTTNG_SET_CONSUMER_URI:
2630 {
2631 size_t nb_uri, len;
2632 struct lttng_uri *uris;
2633
2634 nb_uri = cmd_ctx->lsm->u.uri.size;
2635 len = nb_uri * sizeof(struct lttng_uri);
2636
2637 if (nb_uri == 0) {
2638 ret = LTTNG_ERR_INVALID;
2639 goto error;
2640 }
2641
2642 uris = zmalloc(len);
2643 if (uris == NULL) {
2644 ret = LTTNG_ERR_FATAL;
2645 goto error;
2646 }
2647
2648 /* Receive variable len data */
2649 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2650 ret = lttcomm_recv_unix_sock(sock, uris, len);
2651 if (ret <= 0) {
2652 DBG("No URIs received from client... continuing");
2653 *sock_error = 1;
2654 ret = LTTNG_ERR_SESSION_FAIL;
2655 free(uris);
2656 goto error;
2657 }
2658
2659 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2660 nb_uri, uris);
2661 if (ret != LTTNG_OK) {
2662 free(uris);
2663 goto error;
2664 }
2665
2666 /*
2667 * XXX: 0 means that this URI should be applied on the session. Should
2668 * be a DOMAIN enuam.
2669 */
2670 if (cmd_ctx->lsm->domain.type == 0) {
2671 /* Add the URI for the UST session if a consumer is present. */
2672 if (cmd_ctx->session->ust_session &&
2673 cmd_ctx->session->ust_session->consumer) {
2674 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2675 nb_uri, uris);
2676 } else if (cmd_ctx->session->kernel_session &&
2677 cmd_ctx->session->kernel_session->consumer) {
2678 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2679 cmd_ctx->session, nb_uri, uris);
2680 }
2681 }
2682
2683 free(uris);
2684
2685 break;
2686 }
2687 case LTTNG_START_TRACE:
2688 {
2689 ret = cmd_start_trace(cmd_ctx->session);
2690 break;
2691 }
2692 case LTTNG_STOP_TRACE:
2693 {
2694 ret = cmd_stop_trace(cmd_ctx->session);
2695 break;
2696 }
2697 case LTTNG_CREATE_SESSION:
2698 {
2699 size_t nb_uri, len;
2700 struct lttng_uri *uris = NULL;
2701
2702 nb_uri = cmd_ctx->lsm->u.uri.size;
2703 len = nb_uri * sizeof(struct lttng_uri);
2704
2705 if (nb_uri > 0) {
2706 uris = zmalloc(len);
2707 if (uris == NULL) {
2708 ret = LTTNG_ERR_FATAL;
2709 goto error;
2710 }
2711
2712 /* Receive variable len data */
2713 DBG("Waiting for %zu URIs from client ...", nb_uri);
2714 ret = lttcomm_recv_unix_sock(sock, uris, len);
2715 if (ret <= 0) {
2716 DBG("No URIs received from client... continuing");
2717 *sock_error = 1;
2718 ret = LTTNG_ERR_SESSION_FAIL;
2719 free(uris);
2720 goto error;
2721 }
2722
2723 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2724 DBG("Creating session with ONE network URI is a bad call");
2725 ret = LTTNG_ERR_SESSION_FAIL;
2726 free(uris);
2727 goto error;
2728 }
2729 }
2730
2731 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2732 &cmd_ctx->creds);
2733
2734 free(uris);
2735
2736 break;
2737 }
2738 case LTTNG_DESTROY_SESSION:
2739 {
2740 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2741
2742 /* Set session to NULL so we do not unlock it after free. */
2743 cmd_ctx->session = NULL;
2744 break;
2745 }
2746 case LTTNG_LIST_DOMAINS:
2747 {
2748 ssize_t nb_dom;
2749 struct lttng_domain *domains;
2750
2751 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2752 if (nb_dom < 0) {
2753 /* Return value is a negative lttng_error_code. */
2754 ret = -nb_dom;
2755 goto error;
2756 }
2757
2758 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2759 if (ret < 0) {
2760 goto setup_error;
2761 }
2762
2763 /* Copy event list into message payload */
2764 memcpy(cmd_ctx->llm->payload, domains,
2765 nb_dom * sizeof(struct lttng_domain));
2766
2767 free(domains);
2768
2769 ret = LTTNG_OK;
2770 break;
2771 }
2772 case LTTNG_LIST_CHANNELS:
2773 {
2774 int nb_chan;
2775 struct lttng_channel *channels;
2776
2777 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2778 cmd_ctx->session, &channels);
2779 if (nb_chan < 0) {
2780 /* Return value is a negative lttng_error_code. */
2781 ret = -nb_chan;
2782 goto error;
2783 }
2784
2785 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2786 if (ret < 0) {
2787 goto setup_error;
2788 }
2789
2790 /* Copy event list into message payload */
2791 memcpy(cmd_ctx->llm->payload, channels,
2792 nb_chan * sizeof(struct lttng_channel));
2793
2794 free(channels);
2795
2796 ret = LTTNG_OK;
2797 break;
2798 }
2799 case LTTNG_LIST_EVENTS:
2800 {
2801 ssize_t nb_event;
2802 struct lttng_event *events = NULL;
2803
2804 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2805 cmd_ctx->lsm->u.list.channel_name, &events);
2806 if (nb_event < 0) {
2807 /* Return value is a negative lttng_error_code. */
2808 ret = -nb_event;
2809 goto error;
2810 }
2811
2812 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2813 if (ret < 0) {
2814 goto setup_error;
2815 }
2816
2817 /* Copy event list into message payload */
2818 memcpy(cmd_ctx->llm->payload, events,
2819 nb_event * sizeof(struct lttng_event));
2820
2821 free(events);
2822
2823 ret = LTTNG_OK;
2824 break;
2825 }
2826 case LTTNG_LIST_SESSIONS:
2827 {
2828 unsigned int nr_sessions;
2829
2830 session_lock_list();
2831 nr_sessions = lttng_sessions_count(
2832 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2833 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2834
2835 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2836 if (ret < 0) {
2837 session_unlock_list();
2838 goto setup_error;
2839 }
2840
2841 /* Filled the session array */
2842 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2843 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2844 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2845
2846 session_unlock_list();
2847
2848 ret = LTTNG_OK;
2849 break;
2850 }
2851 case LTTNG_CALIBRATE:
2852 {
2853 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2854 &cmd_ctx->lsm->u.calibrate);
2855 break;
2856 }
2857 case LTTNG_REGISTER_CONSUMER:
2858 {
2859 struct consumer_data *cdata;
2860
2861 switch (cmd_ctx->lsm->domain.type) {
2862 case LTTNG_DOMAIN_KERNEL:
2863 cdata = &kconsumer_data;
2864 break;
2865 default:
2866 ret = LTTNG_ERR_UND;
2867 goto error;
2868 }
2869
2870 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2871 cmd_ctx->lsm->u.reg.path, cdata);
2872 break;
2873 }
2874 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2875 {
2876 struct lttng_filter_bytecode *bytecode;
2877
2878 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2879 ret = LTTNG_ERR_FILTER_INVAL;
2880 goto error;
2881 }
2882 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2883 ret = LTTNG_ERR_FILTER_INVAL;
2884 goto error;
2885 }
2886 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2887 if (!bytecode) {
2888 ret = LTTNG_ERR_FILTER_NOMEM;
2889 goto error;
2890 }
2891 /* Receive var. len. data */
2892 DBG("Receiving var len data from client ...");
2893 ret = lttcomm_recv_unix_sock(sock, bytecode,
2894 cmd_ctx->lsm->u.enable.bytecode_len);
2895 if (ret <= 0) {
2896 DBG("Nothing recv() from client var len data... continuing");
2897 *sock_error = 1;
2898 ret = LTTNG_ERR_FILTER_INVAL;
2899 goto error;
2900 }
2901
2902 if (bytecode->len + sizeof(*bytecode)
2903 != cmd_ctx->lsm->u.enable.bytecode_len) {
2904 free(bytecode);
2905 ret = LTTNG_ERR_FILTER_INVAL;
2906 goto error;
2907 }
2908
2909 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2910 cmd_ctx->lsm->u.enable.channel_name,
2911 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2912 break;
2913 }
2914 case LTTNG_DATA_PENDING:
2915 {
2916 ret = cmd_data_pending(cmd_ctx->session);
2917 break;
2918 }
2919 default:
2920 ret = LTTNG_ERR_UND;
2921 break;
2922 }
2923
2924 error:
2925 if (cmd_ctx->llm == NULL) {
2926 DBG("Missing llm structure. Allocating one.");
2927 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2928 goto setup_error;
2929 }
2930 }
2931 /* Set return code */
2932 cmd_ctx->llm->ret_code = ret;
2933 setup_error:
2934 if (cmd_ctx->session) {
2935 session_unlock(cmd_ctx->session);
2936 }
2937 if (need_tracing_session) {
2938 session_unlock_list();
2939 }
2940 init_setup_error:
2941 return ret;
2942 }
2943
2944 /*
2945 * Thread managing health check socket.
2946 */
2947 static void *thread_manage_health(void *data)
2948 {
2949 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2950 uint32_t revents, nb_fd;
2951 struct lttng_poll_event events;
2952 struct lttcomm_health_msg msg;
2953 struct lttcomm_health_data reply;
2954
2955 DBG("[thread] Manage health check started");
2956
2957 rcu_register_thread();
2958
2959 /* Create unix socket */
2960 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2961 if (sock < 0) {
2962 ERR("Unable to create health check Unix socket");
2963 ret = -1;
2964 goto error;
2965 }
2966
2967 /*
2968 * Set the CLOEXEC flag. Return code is useless because either way, the
2969 * show must go on.
2970 */
2971 (void) utils_set_fd_cloexec(sock);
2972
2973 ret = lttcomm_listen_unix_sock(sock);
2974 if (ret < 0) {
2975 goto error;
2976 }
2977
2978 /*
2979 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2980 * more will be added to this poll set.
2981 */
2982 ret = create_thread_poll_set(&events, 2);
2983 if (ret < 0) {
2984 goto error;
2985 }
2986
2987 /* Add the application registration socket */
2988 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2989 if (ret < 0) {
2990 goto error;
2991 }
2992
2993 while (1) {
2994 DBG("Health check ready");
2995
2996 /* Inifinite blocking call, waiting for transmission */
2997 restart:
2998 ret = lttng_poll_wait(&events, -1);
2999 if (ret < 0) {
3000 /*
3001 * Restart interrupted system call.
3002 */
3003 if (errno == EINTR) {
3004 goto restart;
3005 }
3006 goto error;
3007 }
3008
3009 nb_fd = ret;
3010
3011 for (i = 0; i < nb_fd; i++) {
3012 /* Fetch once the poll data */
3013 revents = LTTNG_POLL_GETEV(&events, i);
3014 pollfd = LTTNG_POLL_GETFD(&events, i);
3015
3016 /* Thread quit pipe has been closed. Killing thread. */
3017 ret = check_thread_quit_pipe(pollfd, revents);
3018 if (ret) {
3019 err = 0;
3020 goto exit;
3021 }
3022
3023 /* Event on the registration socket */
3024 if (pollfd == sock) {
3025 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3026 ERR("Health socket poll error");
3027 goto error;
3028 }
3029 }
3030 }
3031
3032 new_sock = lttcomm_accept_unix_sock(sock);
3033 if (new_sock < 0) {
3034 goto error;
3035 }
3036
3037 /*
3038 * Set the CLOEXEC flag. Return code is useless because either way, the
3039 * show must go on.
3040 */
3041 (void) utils_set_fd_cloexec(new_sock);
3042
3043 DBG("Receiving data from client for health...");
3044 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3045 if (ret <= 0) {
3046 DBG("Nothing recv() from client... continuing");
3047 ret = close(new_sock);
3048 if (ret) {
3049 PERROR("close");
3050 }
3051 new_sock = -1;
3052 continue;
3053 }
3054
3055 rcu_thread_online();
3056
3057 switch (msg.component) {
3058 case LTTNG_HEALTH_CMD:
3059 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3060 break;
3061 case LTTNG_HEALTH_APP_MANAGE:
3062 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3063 break;
3064 case LTTNG_HEALTH_APP_REG:
3065 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3066 break;
3067 case LTTNG_HEALTH_KERNEL:
3068 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3069 break;
3070 case LTTNG_HEALTH_CONSUMER:
3071 reply.ret_code = check_consumer_health();
3072 break;
3073 case LTTNG_HEALTH_ALL:
3074 reply.ret_code =
3075 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3076 health_check_state(HEALTH_TYPE_APP_REG) &&
3077 health_check_state(HEALTH_TYPE_CMD) &&
3078 health_check_state(HEALTH_TYPE_KERNEL) &&
3079 check_consumer_health();
3080 break;
3081 default:
3082 reply.ret_code = LTTNG_ERR_UND;
3083 break;
3084 }
3085
3086 /*
3087 * Flip ret value since 0 is a success and 1 indicates a bad health for
3088 * the client where in the sessiond it is the opposite. Again, this is
3089 * just to make things easier for us poor developer which enjoy a lot
3090 * lazyness.
3091 */
3092 if (reply.ret_code == 0 || reply.ret_code == 1) {
3093 reply.ret_code = !reply.ret_code;
3094 }
3095
3096 DBG2("Health check return value %d", reply.ret_code);
3097
3098 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3099 if (ret < 0) {
3100 ERR("Failed to send health data back to client");
3101 }
3102
3103 /* End of transmission */
3104 ret = close(new_sock);
3105 if (ret) {
3106 PERROR("close");
3107 }
3108 new_sock = -1;
3109 }
3110
3111 exit:
3112 error:
3113 if (err) {
3114 ERR("Health error occurred in %s", __func__);
3115 }
3116 DBG("Health check thread dying");
3117 unlink(health_unix_sock_path);
3118 if (sock >= 0) {
3119 ret = close(sock);
3120 if (ret) {
3121 PERROR("close");
3122 }
3123 }
3124 if (new_sock >= 0) {
3125 ret = close(new_sock);
3126 if (ret) {
3127 PERROR("close");
3128 }
3129 }
3130
3131 lttng_poll_clean(&events);
3132
3133 rcu_unregister_thread();
3134 return NULL;
3135 }
3136
3137 /*
3138 * This thread manage all clients request using the unix client socket for
3139 * communication.
3140 */
3141 static void *thread_manage_clients(void *data)
3142 {
3143 int sock = -1, ret, i, pollfd, err = -1;
3144 int sock_error;
3145 uint32_t revents, nb_fd;
3146 struct command_ctx *cmd_ctx = NULL;
3147 struct lttng_poll_event events;
3148
3149 DBG("[thread] Manage client started");
3150
3151 rcu_register_thread();
3152
3153 health_register(HEALTH_TYPE_CMD);
3154
3155 if (testpoint(thread_manage_clients)) {
3156 goto error_testpoint;
3157 }
3158
3159 health_code_update();
3160
3161 ret = lttcomm_listen_unix_sock(client_sock);
3162 if (ret < 0) {
3163 goto error_listen;
3164 }
3165
3166 /*
3167 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3168 * more will be added to this poll set.
3169 */
3170 ret = create_thread_poll_set(&events, 2);
3171 if (ret < 0) {
3172 goto error_create_poll;
3173 }
3174
3175 /* Add the application registration socket */
3176 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3177 if (ret < 0) {
3178 goto error;
3179 }
3180
3181 /*
3182 * Notify parent pid that we are ready to accept command for client side.
3183 */
3184 if (opt_sig_parent) {
3185 kill(ppid, SIGUSR1);
3186 }
3187
3188 if (testpoint(thread_manage_clients_before_loop)) {
3189 goto error;
3190 }
3191
3192 health_code_update();
3193
3194 while (1) {
3195 DBG("Accepting client command ...");
3196
3197 /* Inifinite blocking call, waiting for transmission */
3198 restart:
3199 health_poll_update();
3200 ret = lttng_poll_wait(&events, -1);
3201 health_poll_update();
3202 if (ret < 0) {
3203 /*
3204 * Restart interrupted system call.
3205 */
3206 if (errno == EINTR) {
3207 goto restart;
3208 }
3209 goto error;
3210 }
3211
3212 nb_fd = ret;
3213
3214 for (i = 0; i < nb_fd; i++) {
3215 /* Fetch once the poll data */
3216 revents = LTTNG_POLL_GETEV(&events, i);
3217 pollfd = LTTNG_POLL_GETFD(&events, i);
3218
3219 health_code_update();
3220
3221 /* Thread quit pipe has been closed. Killing thread. */
3222 ret = check_thread_quit_pipe(pollfd, revents);
3223 if (ret) {
3224 err = 0;
3225 goto exit;
3226 }
3227
3228 /* Event on the registration socket */
3229 if (pollfd == client_sock) {
3230 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3231 ERR("Client socket poll error");
3232 goto error;
3233 }
3234 }
3235 }
3236
3237 DBG("Wait for client response");
3238
3239 health_code_update();
3240
3241 sock = lttcomm_accept_unix_sock(client_sock);
3242 if (sock < 0) {
3243 goto error;
3244 }
3245
3246 /*
3247 * Set the CLOEXEC flag. Return code is useless because either way, the
3248 * show must go on.
3249 */
3250 (void) utils_set_fd_cloexec(sock);
3251
3252 /* Set socket option for credentials retrieval */
3253 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3254 if (ret < 0) {
3255 goto error;
3256 }
3257
3258 /* Allocate context command to process the client request */
3259 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3260 if (cmd_ctx == NULL) {
3261 PERROR("zmalloc cmd_ctx");
3262 goto error;
3263 }
3264
3265 /* Allocate data buffer for reception */
3266 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3267 if (cmd_ctx->lsm == NULL) {
3268 PERROR("zmalloc cmd_ctx->lsm");
3269 goto error;
3270 }
3271
3272 cmd_ctx->llm = NULL;
3273 cmd_ctx->session = NULL;
3274
3275 health_code_update();
3276
3277 /*
3278 * Data is received from the lttng client. The struct
3279 * lttcomm_session_msg (lsm) contains the command and data request of
3280 * the client.
3281 */
3282 DBG("Receiving data from client ...");
3283 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3284 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3285 if (ret <= 0) {
3286 DBG("Nothing recv() from client... continuing");
3287 ret = close(sock);
3288 if (ret) {
3289 PERROR("close");
3290 }
3291 sock = -1;
3292 clean_command_ctx(&cmd_ctx);
3293 continue;
3294 }
3295
3296 health_code_update();
3297
3298 // TODO: Validate cmd_ctx including sanity check for
3299 // security purpose.
3300
3301 rcu_thread_online();
3302 /*
3303 * This function dispatch the work to the kernel or userspace tracer
3304 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3305 * informations for the client. The command context struct contains
3306 * everything this function may needs.
3307 */
3308 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3309 rcu_thread_offline();
3310 if (ret < 0) {
3311 if (sock_error) {
3312 ret = close(sock);
3313 if (ret) {
3314 PERROR("close");
3315 }
3316 sock = -1;
3317 }
3318 /*
3319 * TODO: Inform client somehow of the fatal error. At
3320 * this point, ret < 0 means that a zmalloc failed
3321 * (ENOMEM). Error detected but still accept
3322 * command, unless a socket error has been
3323 * detected.
3324 */
3325 clean_command_ctx(&cmd_ctx);
3326 continue;
3327 }
3328
3329 health_code_update();
3330
3331 DBG("Sending response (size: %d, retcode: %s)",
3332 cmd_ctx->lttng_msg_size,
3333 lttng_strerror(-cmd_ctx->llm->ret_code));
3334 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3335 if (ret < 0) {
3336 ERR("Failed to send data back to client");
3337 }
3338
3339 /* End of transmission */
3340 ret = close(sock);
3341 if (ret) {
3342 PERROR("close");
3343 }
3344 sock = -1;
3345
3346 clean_command_ctx(&cmd_ctx);
3347
3348 health_code_update();
3349 }
3350
3351 exit:
3352 error:
3353 if (sock >= 0) {
3354 ret = close(sock);
3355 if (ret) {
3356 PERROR("close");
3357 }
3358 }
3359
3360 lttng_poll_clean(&events);
3361 clean_command_ctx(&cmd_ctx);
3362
3363 error_listen:
3364 error_create_poll:
3365 error_testpoint:
3366 unlink(client_unix_sock_path);
3367 if (client_sock >= 0) {
3368 ret = close(client_sock);
3369 if (ret) {
3370 PERROR("close");
3371 }
3372 }
3373
3374 if (err) {
3375 health_error();
3376 ERR("Health error occurred in %s", __func__);
3377 }
3378
3379 health_unregister();
3380
3381 DBG("Client thread dying");
3382
3383 rcu_unregister_thread();
3384 return NULL;
3385 }
3386
3387
3388 /*
3389 * usage function on stderr
3390 */
3391 static void usage(void)
3392 {
3393 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3394 fprintf(stderr, " -h, --help Display this usage.\n");
3395 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3396 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3397 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3398 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3399 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3400 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3401 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3402 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3403 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3404 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3405 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3406 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3407 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3408 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3409 fprintf(stderr, " -V, --version Show version number.\n");
3410 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3411 fprintf(stderr, " -q, --quiet No output at all.\n");
3412 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3413 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3414 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3415 }
3416
3417 /*
3418 * daemon argument parsing
3419 */
3420 static int parse_args(int argc, char **argv)
3421 {
3422 int c;
3423
3424 static struct option long_options[] = {
3425 { "client-sock", 1, 0, 'c' },
3426 { "apps-sock", 1, 0, 'a' },
3427 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3428 { "kconsumerd-err-sock", 1, 0, 'E' },
3429 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3430 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3431 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3432 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3433 { "consumerd32-path", 1, 0, 'u' },
3434 { "consumerd32-libdir", 1, 0, 'U' },
3435 { "consumerd64-path", 1, 0, 't' },
3436 { "consumerd64-libdir", 1, 0, 'T' },
3437 { "daemonize", 0, 0, 'd' },
3438 { "sig-parent", 0, 0, 'S' },
3439 { "help", 0, 0, 'h' },
3440 { "group", 1, 0, 'g' },
3441 { "version", 0, 0, 'V' },
3442 { "quiet", 0, 0, 'q' },
3443 { "verbose", 0, 0, 'v' },
3444 { "verbose-consumer", 0, 0, 'Z' },
3445 { "no-kernel", 0, 0, 'N' },
3446 { NULL, 0, 0, 0 }
3447 };
3448
3449 while (1) {
3450 int option_index = 0;
3451 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3452 long_options, &option_index);
3453 if (c == -1) {
3454 break;
3455 }
3456
3457 switch (c) {
3458 case 0:
3459 fprintf(stderr, "option %s", long_options[option_index].name);
3460 if (optarg) {
3461 fprintf(stderr, " with arg %s\n", optarg);
3462 }
3463 break;
3464 case 'c':
3465 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3466 break;
3467 case 'a':
3468 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3469 break;
3470 case 'd':
3471 opt_daemon = 1;
3472 break;
3473 case 'g':
3474 opt_tracing_group = optarg;
3475 break;
3476 case 'h':
3477 usage();
3478 exit(EXIT_FAILURE);
3479 case 'V':
3480 fprintf(stdout, "%s\n", VERSION);
3481 exit(EXIT_SUCCESS);
3482 case 'S':
3483 opt_sig_parent = 1;
3484 break;
3485 case 'E':
3486 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3487 break;
3488 case 'C':
3489 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3490 break;
3491 case 'F':
3492 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3493 break;
3494 case 'D':
3495 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3496 break;
3497 case 'H':
3498 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3499 break;
3500 case 'G':
3501 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3502 break;
3503 case 'N':
3504 opt_no_kernel = 1;
3505 break;
3506 case 'q':
3507 lttng_opt_quiet = 1;
3508 break;
3509 case 'v':
3510 /* Verbose level can increase using multiple -v */
3511 lttng_opt_verbose += 1;
3512 break;
3513 case 'Z':
3514 opt_verbose_consumer += 1;
3515 break;
3516 case 'u':
3517 consumerd32_bin= optarg;
3518 break;
3519 case 'U':
3520 consumerd32_libdir = optarg;
3521 break;
3522 case 't':
3523 consumerd64_bin = optarg;
3524 break;
3525 case 'T':
3526 consumerd64_libdir = optarg;
3527 break;
3528 default:
3529 /* Unknown option or other error.
3530 * Error is printed by getopt, just return */
3531 return -1;
3532 }
3533 }
3534
3535 return 0;
3536 }
3537
3538 /*
3539 * Creates the two needed socket by the daemon.
3540 * apps_sock - The communication socket for all UST apps.
3541 * client_sock - The communication of the cli tool (lttng).
3542 */
3543 static int init_daemon_socket(void)
3544 {
3545 int ret = 0;
3546 mode_t old_umask;
3547
3548 old_umask = umask(0);
3549
3550 /* Create client tool unix socket */
3551 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3552 if (client_sock < 0) {
3553 ERR("Create unix sock failed: %s", client_unix_sock_path);
3554 ret = -1;
3555 goto end;
3556 }
3557
3558 /* Set the cloexec flag */
3559 ret = utils_set_fd_cloexec(client_sock);
3560 if (ret < 0) {
3561 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3562 "Continuing but note that the consumer daemon will have a "
3563 "reference to this socket on exec()", client_sock);
3564 }
3565
3566 /* File permission MUST be 660 */
3567 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3568 if (ret < 0) {
3569 ERR("Set file permissions failed: %s", client_unix_sock_path);
3570 PERROR("chmod");
3571 goto end;
3572 }
3573
3574 /* Create the application unix socket */
3575 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3576 if (apps_sock < 0) {
3577 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3578 ret = -1;
3579 goto end;
3580 }
3581
3582 /* Set the cloexec flag */
3583 ret = utils_set_fd_cloexec(apps_sock);
3584 if (ret < 0) {
3585 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3586 "Continuing but note that the consumer daemon will have a "
3587 "reference to this socket on exec()", apps_sock);
3588 }
3589
3590 /* File permission MUST be 666 */
3591 ret = chmod(apps_unix_sock_path,
3592 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3593 if (ret < 0) {
3594 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3595 PERROR("chmod");
3596 goto end;
3597 }
3598
3599 DBG3("Session daemon client socket %d and application socket %d created",
3600 client_sock, apps_sock);
3601
3602 end:
3603 umask(old_umask);
3604 return ret;
3605 }
3606
3607 /*
3608 * Check if the global socket is available, and if a daemon is answering at the
3609 * other side. If yes, error is returned.
3610 */
3611 static int check_existing_daemon(void)
3612 {
3613 /* Is there anybody out there ? */
3614 if (lttng_session_daemon_alive()) {
3615 return -EEXIST;
3616 }
3617
3618 return 0;
3619 }
3620
3621 /*
3622 * Set the tracing group gid onto the client socket.
3623 *
3624 * Race window between mkdir and chown is OK because we are going from more
3625 * permissive (root.root) to less permissive (root.tracing).
3626 */
3627 static int set_permissions(char *rundir)
3628 {
3629 int ret;
3630 gid_t gid;
3631
3632 ret = allowed_group();
3633 if (ret < 0) {
3634 WARN("No tracing group detected");
3635 ret = 0;
3636 goto end;
3637 }
3638
3639 gid = ret;
3640
3641 /* Set lttng run dir */
3642 ret = chown(rundir, 0, gid);
3643 if (ret < 0) {
3644 ERR("Unable to set group on %s", rundir);
3645 PERROR("chown");
3646 }
3647
3648 /* Ensure tracing group can search the run dir */
3649 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3650 if (ret < 0) {
3651 ERR("Unable to set permissions on %s", rundir);
3652 PERROR("chmod");
3653 }
3654
3655 /* lttng client socket path */
3656 ret = chown(client_unix_sock_path, 0, gid);
3657 if (ret < 0) {
3658 ERR("Unable to set group on %s", client_unix_sock_path);
3659 PERROR("chown");
3660 }
3661
3662 /* kconsumer error socket path */
3663 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3664 if (ret < 0) {
3665 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3666 PERROR("chown");
3667 }
3668
3669 /* 64-bit ustconsumer error socket path */
3670 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3671 if (ret < 0) {
3672 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3673 PERROR("chown");
3674 }
3675
3676 /* 32-bit ustconsumer compat32 error socket path */
3677 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3678 if (ret < 0) {
3679 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3680 PERROR("chown");
3681 }
3682
3683 DBG("All permissions are set");
3684
3685 end:
3686 return ret;
3687 }
3688
3689 /*
3690 * Create the lttng run directory needed for all global sockets and pipe.
3691 */
3692 static int create_lttng_rundir(const char *rundir)
3693 {
3694 int ret;
3695
3696 DBG3("Creating LTTng run directory: %s", rundir);
3697
3698 ret = mkdir(rundir, S_IRWXU);
3699 if (ret < 0) {
3700 if (errno != EEXIST) {
3701 ERR("Unable to create %s", rundir);
3702 goto error;
3703 } else {
3704 ret = 0;
3705 }
3706 }
3707
3708 error:
3709 return ret;
3710 }
3711
3712 /*
3713 * Setup sockets and directory needed by the kconsumerd communication with the
3714 * session daemon.
3715 */
3716 static int set_consumer_sockets(struct consumer_data *consumer_data,
3717 const char *rundir)
3718 {
3719 int ret;
3720 char path[PATH_MAX];
3721
3722 switch (consumer_data->type) {
3723 case LTTNG_CONSUMER_KERNEL:
3724 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3725 break;
3726 case LTTNG_CONSUMER64_UST:
3727 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3728 break;
3729 case LTTNG_CONSUMER32_UST:
3730 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3731 break;
3732 default:
3733 ERR("Consumer type unknown");
3734 ret = -EINVAL;
3735 goto error;
3736 }
3737
3738 DBG2("Creating consumer directory: %s", path);
3739
3740 ret = mkdir(path, S_IRWXU);
3741 if (ret < 0) {
3742 if (errno != EEXIST) {
3743 PERROR("mkdir");
3744 ERR("Failed to create %s", path);
3745 goto error;
3746 }
3747 ret = -1;
3748 }
3749
3750 /* Create the kconsumerd error unix socket */
3751 consumer_data->err_sock =
3752 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3753 if (consumer_data->err_sock < 0) {
3754 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3755 ret = -1;
3756 goto error;
3757 }
3758
3759 /* File permission MUST be 660 */
3760 ret = chmod(consumer_data->err_unix_sock_path,
3761 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3762 if (ret < 0) {
3763 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3764 PERROR("chmod");
3765 goto error;
3766 }
3767
3768 error:
3769 return ret;
3770 }
3771
3772 /*
3773 * Signal handler for the daemon
3774 *
3775 * Simply stop all worker threads, leaving main() return gracefully after
3776 * joining all threads and calling cleanup().
3777 */
3778 static void sighandler(int sig)
3779 {
3780 switch (sig) {
3781 case SIGPIPE:
3782 DBG("SIGPIPE caught");
3783 return;
3784 case SIGINT:
3785 DBG("SIGINT caught");
3786 stop_threads();
3787 break;
3788 case SIGTERM:
3789 DBG("SIGTERM caught");
3790 stop_threads();
3791 break;
3792 default:
3793 break;
3794 }
3795 }
3796
3797 /*
3798 * Setup signal handler for :
3799 * SIGINT, SIGTERM, SIGPIPE
3800 */
3801 static int set_signal_handler(void)
3802 {
3803 int ret = 0;
3804 struct sigaction sa;
3805 sigset_t sigset;
3806
3807 if ((ret = sigemptyset(&sigset)) < 0) {
3808 PERROR("sigemptyset");
3809 return ret;
3810 }
3811
3812 sa.sa_handler = sighandler;
3813 sa.sa_mask = sigset;
3814 sa.sa_flags = 0;
3815 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3816 PERROR("sigaction");
3817 return ret;
3818 }
3819
3820 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3821 PERROR("sigaction");
3822 return ret;
3823 }
3824
3825 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3826 PERROR("sigaction");
3827 return ret;
3828 }
3829
3830 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3831
3832 return ret;
3833 }
3834
3835 /*
3836 * Set open files limit to unlimited. This daemon can open a large number of
3837 * file descriptors in order to consumer multiple kernel traces.
3838 */
3839 static void set_ulimit(void)
3840 {
3841 int ret;
3842 struct rlimit lim;
3843
3844 /* The kernel does not allowed an infinite limit for open files */
3845 lim.rlim_cur = 65535;
3846 lim.rlim_max = 65535;
3847
3848 ret = setrlimit(RLIMIT_NOFILE, &lim);
3849 if (ret < 0) {
3850 PERROR("failed to set open files limit");
3851 }
3852 }
3853
3854 /*
3855 * main
3856 */
3857 int main(int argc, char **argv)
3858 {
3859 int ret = 0;
3860 void *status;
3861 const char *home_path, *env_app_timeout;
3862
3863 init_kernel_workarounds();
3864
3865 rcu_register_thread();
3866
3867 setup_consumerd_path();
3868
3869 /* Parse arguments */
3870 progname = argv[0];
3871 if ((ret = parse_args(argc, argv)) < 0) {
3872 goto error;
3873 }
3874
3875 /* Daemonize */
3876 if (opt_daemon) {
3877 int i;
3878
3879 /*
3880 * fork
3881 * child: setsid, close FD 0, 1, 2, chdir /
3882 * parent: exit (if fork is successful)
3883 */
3884 ret = daemon(0, 0);
3885 if (ret < 0) {
3886 PERROR("daemon");
3887 goto error;
3888 }
3889 /*
3890 * We are in the child. Make sure all other file
3891 * descriptors are closed, in case we are called with
3892 * more opened file descriptors than the standard ones.
3893 */
3894 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3895 (void) close(i);
3896 }
3897 }
3898
3899 /* Create thread quit pipe */
3900 if ((ret = init_thread_quit_pipe()) < 0) {
3901 goto error;
3902 }
3903
3904 /* Check if daemon is UID = 0 */
3905 is_root = !getuid();
3906
3907 if (is_root) {
3908 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3909
3910 /* Create global run dir with root access */
3911 ret = create_lttng_rundir(rundir);
3912 if (ret < 0) {
3913 goto error;
3914 }
3915
3916 if (strlen(apps_unix_sock_path) == 0) {
3917 snprintf(apps_unix_sock_path, PATH_MAX,
3918 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3919 }
3920
3921 if (strlen(client_unix_sock_path) == 0) {
3922 snprintf(client_unix_sock_path, PATH_MAX,
3923 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3924 }
3925
3926 /* Set global SHM for ust */
3927 if (strlen(wait_shm_path) == 0) {
3928 snprintf(wait_shm_path, PATH_MAX,
3929 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3930 }
3931
3932 if (strlen(health_unix_sock_path) == 0) {
3933 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3934 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3935 }
3936
3937 /* Setup kernel consumerd path */
3938 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3939 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3940 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3941 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3942
3943 DBG2("Kernel consumer err path: %s",
3944 kconsumer_data.err_unix_sock_path);
3945 DBG2("Kernel consumer cmd path: %s",
3946 kconsumer_data.cmd_unix_sock_path);
3947 } else {
3948 home_path = get_home_dir();
3949 if (home_path == NULL) {
3950 /* TODO: Add --socket PATH option */
3951 ERR("Can't get HOME directory for sockets creation.");
3952 ret = -EPERM;
3953 goto error;
3954 }
3955
3956 /*
3957 * Create rundir from home path. This will create something like
3958 * $HOME/.lttng
3959 */
3960 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3961 if (ret < 0) {
3962 ret = -ENOMEM;
3963 goto error;
3964 }
3965
3966 ret = create_lttng_rundir(rundir);
3967 if (ret < 0) {
3968 goto error;
3969 }
3970
3971 if (strlen(apps_unix_sock_path) == 0) {
3972 snprintf(apps_unix_sock_path, PATH_MAX,
3973 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3974 }
3975
3976 /* Set the cli tool unix socket path */
3977 if (strlen(client_unix_sock_path) == 0) {
3978 snprintf(client_unix_sock_path, PATH_MAX,
3979 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3980 }
3981
3982 /* Set global SHM for ust */
3983 if (strlen(wait_shm_path) == 0) {
3984 snprintf(wait_shm_path, PATH_MAX,
3985 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3986 }
3987
3988 /* Set health check Unix path */
3989 if (strlen(health_unix_sock_path) == 0) {
3990 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3991 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3992 }
3993 }
3994
3995 /* Set consumer initial state */
3996 kernel_consumerd_state = CONSUMER_STOPPED;
3997 ust_consumerd_state = CONSUMER_STOPPED;
3998
3999 DBG("Client socket path %s", client_unix_sock_path);
4000 DBG("Application socket path %s", apps_unix_sock_path);
4001 DBG("LTTng run directory path: %s", rundir);
4002
4003 /* 32 bits consumerd path setup */
4004 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4005 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4006 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4007 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4008
4009 DBG2("UST consumer 32 bits err path: %s",
4010 ustconsumer32_data.err_unix_sock_path);
4011 DBG2("UST consumer 32 bits cmd path: %s",
4012 ustconsumer32_data.cmd_unix_sock_path);
4013
4014 /* 64 bits consumerd path setup */
4015 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4016 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4017 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4018 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4019
4020 DBG2("UST consumer 64 bits err path: %s",
4021 ustconsumer64_data.err_unix_sock_path);
4022 DBG2("UST consumer 64 bits cmd path: %s",
4023 ustconsumer64_data.cmd_unix_sock_path);
4024
4025 /*
4026 * See if daemon already exist.
4027 */
4028 if ((ret = check_existing_daemon()) < 0) {
4029 ERR("Already running daemon.\n");
4030 /*
4031 * We do not goto exit because we must not cleanup()
4032 * because a daemon is already running.
4033 */
4034 goto error;
4035 }
4036
4037 /*
4038 * Init UST app hash table. Alloc hash table before this point since
4039 * cleanup() can get called after that point.
4040 */
4041 ust_app_ht_alloc();
4042
4043 /* After this point, we can safely call cleanup() with "goto exit" */
4044
4045 /*
4046 * These actions must be executed as root. We do that *after* setting up
4047 * the sockets path because we MUST make the check for another daemon using
4048 * those paths *before* trying to set the kernel consumer sockets and init
4049 * kernel tracer.
4050 */
4051 if (is_root) {
4052 ret = set_consumer_sockets(&kconsumer_data, rundir);
4053 if (ret < 0) {
4054 goto exit;
4055 }
4056
4057 /* Setup kernel tracer */
4058 if (!opt_no_kernel) {
4059 init_kernel_tracer();
4060 }
4061
4062 /* Set ulimit for open files */
4063 set_ulimit();
4064 }
4065 /* init lttng_fd tracking must be done after set_ulimit. */
4066 lttng_fd_init();
4067
4068 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4069 if (ret < 0) {
4070 goto exit;
4071 }
4072
4073 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4074 if (ret < 0) {
4075 goto exit;
4076 }
4077
4078 if ((ret = set_signal_handler()) < 0) {
4079 goto exit;
4080 }
4081
4082 /* Setup the needed unix socket */
4083 if ((ret = init_daemon_socket()) < 0) {
4084 goto exit;
4085 }
4086
4087 /* Set credentials to socket */
4088 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4089 goto exit;
4090 }
4091
4092 /* Get parent pid if -S, --sig-parent is specified. */
4093 if (opt_sig_parent) {
4094 ppid = getppid();
4095 }
4096
4097 /* Setup the kernel pipe for waking up the kernel thread */
4098 if (is_root && !opt_no_kernel) {
4099 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4100 goto exit;
4101 }
4102 }
4103
4104 /* Setup the thread apps communication pipe. */
4105 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4106 goto exit;
4107 }
4108
4109 /* Init UST command queue. */
4110 cds_wfq_init(&ust_cmd_queue.queue);
4111
4112 /*
4113 * Get session list pointer. This pointer MUST NOT be free(). This list is
4114 * statically declared in session.c
4115 */
4116 session_list_ptr = session_get_list();
4117
4118 /* Set up max poll set size */
4119 lttng_poll_set_max_size();
4120
4121 cmd_init();
4122
4123 /* Check for the application socket timeout env variable. */
4124 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4125 if (env_app_timeout) {
4126 app_socket_timeout = atoi(env_app_timeout);
4127 } else {
4128 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4129 }
4130
4131 /* Create thread to manage the client socket */
4132 ret = pthread_create(&health_thread, NULL,
4133 thread_manage_health, (void *) NULL);
4134 if (ret != 0) {
4135 PERROR("pthread_create health");
4136 goto exit_health;
4137 }
4138
4139 /* Create thread to manage the client socket */
4140 ret = pthread_create(&client_thread, NULL,
4141 thread_manage_clients, (void *) NULL);
4142 if (ret != 0) {
4143 PERROR("pthread_create clients");
4144 goto exit_client;
4145 }
4146
4147 /* Create thread to dispatch registration */
4148 ret = pthread_create(&dispatch_thread, NULL,
4149 thread_dispatch_ust_registration, (void *) NULL);
4150 if (ret != 0) {
4151 PERROR("pthread_create dispatch");
4152 goto exit_dispatch;
4153 }
4154
4155 /* Create thread to manage application registration. */
4156 ret = pthread_create(&reg_apps_thread, NULL,
4157 thread_registration_apps, (void *) NULL);
4158 if (ret != 0) {
4159 PERROR("pthread_create registration");
4160 goto exit_reg_apps;
4161 }
4162
4163 /* Create thread to manage application socket */
4164 ret = pthread_create(&apps_thread, NULL,
4165 thread_manage_apps, (void *) NULL);
4166 if (ret != 0) {
4167 PERROR("pthread_create apps");
4168 goto exit_apps;
4169 }
4170
4171 /* Don't start this thread if kernel tracing is not requested nor root */
4172 if (is_root && !opt_no_kernel) {
4173 /* Create kernel thread to manage kernel event */
4174 ret = pthread_create(&kernel_thread, NULL,
4175 thread_manage_kernel, (void *) NULL);
4176 if (ret != 0) {
4177 PERROR("pthread_create kernel");
4178 goto exit_kernel;
4179 }
4180
4181 ret = pthread_join(kernel_thread, &status);
4182 if (ret != 0) {
4183 PERROR("pthread_join");
4184 goto error; /* join error, exit without cleanup */
4185 }
4186 }
4187
4188 exit_kernel:
4189 ret = pthread_join(apps_thread, &status);
4190 if (ret != 0) {
4191 PERROR("pthread_join");
4192 goto error; /* join error, exit without cleanup */
4193 }
4194
4195 exit_apps:
4196 ret = pthread_join(reg_apps_thread, &status);
4197 if (ret != 0) {
4198 PERROR("pthread_join");
4199 goto error; /* join error, exit without cleanup */
4200 }
4201
4202 exit_reg_apps:
4203 ret = pthread_join(dispatch_thread, &status);
4204 if (ret != 0) {
4205 PERROR("pthread_join");
4206 goto error; /* join error, exit without cleanup */
4207 }
4208
4209 exit_dispatch:
4210 ret = pthread_join(client_thread, &status);
4211 if (ret != 0) {
4212 PERROR("pthread_join");
4213 goto error; /* join error, exit without cleanup */
4214 }
4215
4216 ret = join_consumer_thread(&kconsumer_data);
4217 if (ret != 0) {
4218 PERROR("join_consumer");
4219 goto error; /* join error, exit without cleanup */
4220 }
4221
4222 ret = join_consumer_thread(&ustconsumer32_data);
4223 if (ret != 0) {
4224 PERROR("join_consumer ust32");
4225 goto error; /* join error, exit without cleanup */
4226 }
4227
4228 ret = join_consumer_thread(&ustconsumer64_data);
4229 if (ret != 0) {
4230 PERROR("join_consumer ust64");
4231 goto error; /* join error, exit without cleanup */
4232 }
4233
4234 exit_client:
4235 ret = pthread_join(health_thread, &status);
4236 if (ret != 0) {
4237 PERROR("pthread_join health thread");
4238 goto error; /* join error, exit without cleanup */
4239 }
4240
4241 exit_health:
4242 exit:
4243 /*
4244 * cleanup() is called when no other thread is running.
4245 */
4246 rcu_thread_online();
4247 cleanup();
4248 rcu_thread_offline();
4249 rcu_unregister_thread();
4250 if (!ret) {
4251 exit(EXIT_SUCCESS);
4252 }
4253 error:
4254 exit(EXIT_FAILURE);
4255 }
This page took 0.117355 seconds and 4 git commands to generate.