Fix: remove wrong doing asserts in sessiond
[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 <inttypes.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <urcu/uatomic.h>
37 #include <unistd.h>
38 #include <config.h>
39
40 #include <common/common.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 "buffer-registry.h"
50 #include "channel.h"
51 #include "cmd.h"
52 #include "consumer.h"
53 #include "context.h"
54 #include "event.h"
55 #include "kernel.h"
56 #include "kernel-consumer.h"
57 #include "modprobe.h"
58 #include "shm.h"
59 #include "ust-ctl.h"
60 #include "ust-consumer.h"
61 #include "utils.h"
62 #include "fd-limit.h"
63 #include "health.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
66
67 #define CONSUMERD_FILE "lttng-consumerd"
68
69 /* Const values */
70 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
71
72 const char *progname;
73 const char *opt_tracing_group;
74 static const char *opt_pidfile;
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 .metadata_sock.fd = -1,
94 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
95 .lock = PTHREAD_MUTEX_INITIALIZER,
96 .cond = PTHREAD_COND_INITIALIZER,
97 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
98 };
99 static struct consumer_data ustconsumer64_data = {
100 .type = LTTNG_CONSUMER64_UST,
101 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .metadata_sock.fd = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer32_data = {
112 .type = LTTNG_CONSUMER32_UST,
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .metadata_sock.fd = -1,
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
122 };
123
124 /* Shared between threads */
125 static int dispatch_thread_exit;
126
127 /* Global application Unix socket path */
128 static char apps_unix_sock_path[PATH_MAX];
129 /* Global client Unix socket path */
130 static char client_unix_sock_path[PATH_MAX];
131 /* global wait shm path for UST */
132 static char wait_shm_path[PATH_MAX];
133 /* Global health check unix path */
134 static char health_unix_sock_path[PATH_MAX];
135
136 /* Sockets and FDs */
137 static int client_sock = -1;
138 static int apps_sock = -1;
139 int kernel_tracer_fd = -1;
140 static int kernel_poll_pipe[2] = { -1, -1 };
141
142 /*
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
145 */
146 static int thread_quit_pipe[2] = { -1, -1 };
147
148 /*
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
151 */
152 static int apps_cmd_pipe[2] = { -1, -1 };
153
154 int apps_cmd_notify_pipe[2] = { -1, -1 };
155
156 /* Pthread, Mutexes and Semaphores */
157 static pthread_t apps_thread;
158 static pthread_t apps_notify_thread;
159 static pthread_t reg_apps_thread;
160 static pthread_t client_thread;
161 static pthread_t kernel_thread;
162 static pthread_t dispatch_thread;
163 static pthread_t health_thread;
164 static pthread_t ht_cleanup_thread;
165
166 /*
167 * UST registration command queue. This queue is tied with a futex and uses a N
168 * wakers / 1 waiter implemented and detailed in futex.c/.h
169 *
170 * The thread_manage_apps and thread_dispatch_ust_registration interact with
171 * this queue and the wait/wake scheme.
172 */
173 static struct ust_cmd_queue ust_cmd_queue;
174
175 /*
176 * Pointer initialized before thread creation.
177 *
178 * This points to the tracing session list containing the session count and a
179 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
180 * MUST NOT be taken if you call a public function in session.c.
181 *
182 * The lock is nested inside the structure: session_list_ptr->lock. Please use
183 * session_lock_list and session_unlock_list for lock acquisition.
184 */
185 static struct ltt_session_list *session_list_ptr;
186
187 int ust_consumerd64_fd = -1;
188 int ust_consumerd32_fd = -1;
189
190 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
191 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
192 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
193 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
194
195 static const char *module_proc_lttng = "/proc/lttng";
196
197 /*
198 * Consumer daemon state which is changed when spawning it, killing it or in
199 * case of a fatal error.
200 */
201 enum consumerd_state {
202 CONSUMER_STARTED = 1,
203 CONSUMER_STOPPED = 2,
204 CONSUMER_ERROR = 3,
205 };
206
207 /*
208 * This consumer daemon state is used to validate if a client command will be
209 * able to reach the consumer. If not, the client is informed. For instance,
210 * doing a "lttng start" when the consumer state is set to ERROR will return an
211 * error to the client.
212 *
213 * The following example shows a possible race condition of this scheme:
214 *
215 * consumer thread error happens
216 * client cmd arrives
217 * client cmd checks state -> still OK
218 * consumer thread exit, sets error
219 * client cmd try to talk to consumer
220 * ...
221 *
222 * However, since the consumer is a different daemon, we have no way of making
223 * sure the command will reach it safely even with this state flag. This is why
224 * we consider that up to the state validation during command processing, the
225 * command is safe. After that, we can not guarantee the correctness of the
226 * client request vis-a-vis the consumer.
227 */
228 static enum consumerd_state ust_consumerd_state;
229 static enum consumerd_state kernel_consumerd_state;
230
231 /*
232 * Socket timeout for receiving and sending in seconds.
233 */
234 static int app_socket_timeout;
235
236 /* Set in main() with the current page size. */
237 long page_size;
238
239 static
240 void setup_consumerd_path(void)
241 {
242 const char *bin, *libdir;
243
244 /*
245 * Allow INSTALL_BIN_PATH to be used as a target path for the
246 * native architecture size consumer if CONFIG_CONSUMER*_PATH
247 * has not been defined.
248 */
249 #if (CAA_BITS_PER_LONG == 32)
250 if (!consumerd32_bin[0]) {
251 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
252 }
253 if (!consumerd32_libdir[0]) {
254 consumerd32_libdir = INSTALL_LIB_PATH;
255 }
256 #elif (CAA_BITS_PER_LONG == 64)
257 if (!consumerd64_bin[0]) {
258 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
259 }
260 if (!consumerd64_libdir[0]) {
261 consumerd64_libdir = INSTALL_LIB_PATH;
262 }
263 #else
264 #error "Unknown bitness"
265 #endif
266
267 /*
268 * runtime env. var. overrides the build default.
269 */
270 bin = getenv("LTTNG_CONSUMERD32_BIN");
271 if (bin) {
272 consumerd32_bin = bin;
273 }
274 bin = getenv("LTTNG_CONSUMERD64_BIN");
275 if (bin) {
276 consumerd64_bin = bin;
277 }
278 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
279 if (libdir) {
280 consumerd32_libdir = libdir;
281 }
282 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
283 if (libdir) {
284 consumerd64_libdir = libdir;
285 }
286 }
287
288 /*
289 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
290 */
291 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
292 {
293 int ret;
294
295 assert(events);
296
297 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
298 if (ret < 0) {
299 goto error;
300 }
301
302 /* Add quit pipe */
303 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
304 if (ret < 0) {
305 goto error;
306 }
307
308 return 0;
309
310 error:
311 return ret;
312 }
313
314 /*
315 * Check if the thread quit pipe was triggered.
316 *
317 * Return 1 if it was triggered else 0;
318 */
319 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
320 {
321 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
322 return 1;
323 }
324
325 return 0;
326 }
327
328 /*
329 * Return group ID of the tracing group or -1 if not found.
330 */
331 static gid_t allowed_group(void)
332 {
333 struct group *grp;
334
335 if (opt_tracing_group) {
336 grp = getgrnam(opt_tracing_group);
337 } else {
338 grp = getgrnam(default_tracing_group);
339 }
340 if (!grp) {
341 return -1;
342 } else {
343 return grp->gr_gid;
344 }
345 }
346
347 /*
348 * Init thread quit pipe.
349 *
350 * Return -1 on error or 0 if all pipes are created.
351 */
352 static int init_thread_quit_pipe(void)
353 {
354 int ret, i;
355
356 ret = pipe(thread_quit_pipe);
357 if (ret < 0) {
358 PERROR("thread quit pipe");
359 goto error;
360 }
361
362 for (i = 0; i < 2; i++) {
363 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
364 if (ret < 0) {
365 PERROR("fcntl");
366 goto error;
367 }
368 }
369
370 error:
371 return ret;
372 }
373
374 /*
375 * Stop all threads by closing the thread quit pipe.
376 */
377 static void stop_threads(void)
378 {
379 int ret;
380
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret = notify_thread_pipe(thread_quit_pipe[1]);
384 if (ret < 0) {
385 ERR("write error on thread quit pipe");
386 }
387
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit, 1);
390 futex_nto1_wake(&ust_cmd_queue.futex);
391 }
392
393 /*
394 * Close every consumer sockets.
395 */
396 static void close_consumer_sockets(void)
397 {
398 int ret;
399
400 if (kconsumer_data.err_sock >= 0) {
401 ret = close(kconsumer_data.err_sock);
402 if (ret < 0) {
403 PERROR("kernel consumer err_sock close");
404 }
405 }
406 if (ustconsumer32_data.err_sock >= 0) {
407 ret = close(ustconsumer32_data.err_sock);
408 if (ret < 0) {
409 PERROR("UST consumerd32 err_sock close");
410 }
411 }
412 if (ustconsumer64_data.err_sock >= 0) {
413 ret = close(ustconsumer64_data.err_sock);
414 if (ret < 0) {
415 PERROR("UST consumerd64 err_sock close");
416 }
417 }
418 if (kconsumer_data.cmd_sock >= 0) {
419 ret = close(kconsumer_data.cmd_sock);
420 if (ret < 0) {
421 PERROR("kernel consumer cmd_sock close");
422 }
423 }
424 if (ustconsumer32_data.cmd_sock >= 0) {
425 ret = close(ustconsumer32_data.cmd_sock);
426 if (ret < 0) {
427 PERROR("UST consumerd32 cmd_sock close");
428 }
429 }
430 if (ustconsumer64_data.cmd_sock >= 0) {
431 ret = close(ustconsumer64_data.cmd_sock);
432 if (ret < 0) {
433 PERROR("UST consumerd64 cmd_sock close");
434 }
435 }
436 }
437
438 /*
439 * Cleanup the daemon
440 */
441 static void cleanup(void)
442 {
443 int ret;
444 char *cmd = NULL;
445 struct ltt_session *sess, *stmp;
446
447 DBG("Cleaning up");
448
449 /* First thing first, stop all threads */
450 utils_close_pipe(thread_quit_pipe);
451
452 /*
453 * If opt_pidfile is undefined, the default file will be wiped when
454 * removing the rundir.
455 */
456 if (opt_pidfile) {
457 ret = remove(opt_pidfile);
458 if (ret < 0) {
459 PERROR("remove pidfile %s", opt_pidfile);
460 }
461 }
462
463 DBG("Removing %s directory", rundir);
464 ret = asprintf(&cmd, "rm -rf %s", rundir);
465 if (ret < 0) {
466 ERR("asprintf failed. Something is really wrong!");
467 }
468
469 /* Remove lttng run directory */
470 ret = system(cmd);
471 if (ret < 0) {
472 ERR("Unable to clean %s", rundir);
473 }
474 free(cmd);
475 free(rundir);
476
477 DBG("Cleaning up all sessions");
478
479 /* Destroy session list mutex */
480 if (session_list_ptr != NULL) {
481 pthread_mutex_destroy(&session_list_ptr->lock);
482
483 /* Cleanup ALL session */
484 cds_list_for_each_entry_safe(sess, stmp,
485 &session_list_ptr->head, list) {
486 cmd_destroy_session(sess, kernel_poll_pipe[1]);
487 }
488 }
489
490 DBG("Closing all UST sockets");
491 ust_app_clean_list();
492 buffer_reg_destroy_registries();
493
494 if (is_root && !opt_no_kernel) {
495 DBG2("Closing kernel fd");
496 if (kernel_tracer_fd >= 0) {
497 ret = close(kernel_tracer_fd);
498 if (ret) {
499 PERROR("close");
500 }
501 }
502 DBG("Unloading kernel modules");
503 modprobe_remove_lttng_all();
504 }
505
506 close_consumer_sockets();
507
508 /* <fun> */
509 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
510 "Matthew, BEET driven development works!%c[%dm",
511 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
512 /* </fun> */
513 }
514
515 /*
516 * Send data on a unix socket using the liblttsessiondcomm API.
517 *
518 * Return lttcomm error code.
519 */
520 static int send_unix_sock(int sock, void *buf, size_t len)
521 {
522 /* Check valid length */
523 if (len == 0) {
524 return -1;
525 }
526
527 return lttcomm_send_unix_sock(sock, buf, len);
528 }
529
530 /*
531 * Free memory of a command context structure.
532 */
533 static void clean_command_ctx(struct command_ctx **cmd_ctx)
534 {
535 DBG("Clean command context structure");
536 if (*cmd_ctx) {
537 if ((*cmd_ctx)->llm) {
538 free((*cmd_ctx)->llm);
539 }
540 if ((*cmd_ctx)->lsm) {
541 free((*cmd_ctx)->lsm);
542 }
543 free(*cmd_ctx);
544 *cmd_ctx = NULL;
545 }
546 }
547
548 /*
549 * Notify UST applications using the shm mmap futex.
550 */
551 static int notify_ust_apps(int active)
552 {
553 char *wait_shm_mmap;
554
555 DBG("Notifying applications of session daemon state: %d", active);
556
557 /* See shm.c for this call implying mmap, shm and futex calls */
558 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
559 if (wait_shm_mmap == NULL) {
560 goto error;
561 }
562
563 /* Wake waiting process */
564 futex_wait_update((int32_t *) wait_shm_mmap, active);
565
566 /* Apps notified successfully */
567 return 0;
568
569 error:
570 return -1;
571 }
572
573 /*
574 * Setup the outgoing data buffer for the response (llm) by allocating the
575 * right amount of memory and copying the original information from the lsm
576 * structure.
577 *
578 * Return total size of the buffer pointed by buf.
579 */
580 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
581 {
582 int ret, buf_size;
583
584 buf_size = size;
585
586 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
587 if (cmd_ctx->llm == NULL) {
588 PERROR("zmalloc");
589 ret = -ENOMEM;
590 goto error;
591 }
592
593 /* Copy common data */
594 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
595 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
596
597 cmd_ctx->llm->data_size = size;
598 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
599
600 return buf_size;
601
602 error:
603 return ret;
604 }
605
606 /*
607 * Update the kernel poll set of all channel fd available over all tracing
608 * session. Add the wakeup pipe at the end of the set.
609 */
610 static int update_kernel_poll(struct lttng_poll_event *events)
611 {
612 int ret;
613 struct ltt_session *session;
614 struct ltt_kernel_channel *channel;
615
616 DBG("Updating kernel poll set");
617
618 session_lock_list();
619 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
620 session_lock(session);
621 if (session->kernel_session == NULL) {
622 session_unlock(session);
623 continue;
624 }
625
626 cds_list_for_each_entry(channel,
627 &session->kernel_session->channel_list.head, list) {
628 /* Add channel fd to the kernel poll set */
629 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
630 if (ret < 0) {
631 session_unlock(session);
632 goto error;
633 }
634 DBG("Channel fd %d added to kernel set", channel->fd);
635 }
636 session_unlock(session);
637 }
638 session_unlock_list();
639
640 return 0;
641
642 error:
643 session_unlock_list();
644 return -1;
645 }
646
647 /*
648 * Find the channel fd from 'fd' over all tracing session. When found, check
649 * for new channel stream and send those stream fds to the kernel consumer.
650 *
651 * Useful for CPU hotplug feature.
652 */
653 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
654 {
655 int ret = 0;
656 struct ltt_session *session;
657 struct ltt_kernel_session *ksess;
658 struct ltt_kernel_channel *channel;
659
660 DBG("Updating kernel streams for channel fd %d", fd);
661
662 session_lock_list();
663 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
664 session_lock(session);
665 if (session->kernel_session == NULL) {
666 session_unlock(session);
667 continue;
668 }
669 ksess = session->kernel_session;
670
671 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
672 if (channel->fd == fd) {
673 DBG("Channel found, updating kernel streams");
674 ret = kernel_open_channel_stream(channel);
675 if (ret < 0) {
676 goto error;
677 }
678
679 /*
680 * Have we already sent fds to the consumer? If yes, it means
681 * that tracing is started so it is safe to send our updated
682 * stream fds.
683 */
684 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
685 struct lttng_ht_iter iter;
686 struct consumer_socket *socket;
687
688 rcu_read_lock();
689 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
690 &iter.iter, socket, node.node) {
691 /* Code flow error */
692 assert(socket->fd >= 0);
693
694 pthread_mutex_lock(socket->lock);
695 ret = kernel_consumer_send_channel_stream(socket,
696 channel, ksess);
697 pthread_mutex_unlock(socket->lock);
698 if (ret < 0) {
699 rcu_read_unlock();
700 goto error;
701 }
702 }
703 rcu_read_unlock();
704 }
705 goto error;
706 }
707 }
708 session_unlock(session);
709 }
710 session_unlock_list();
711 return ret;
712
713 error:
714 session_unlock(session);
715 session_unlock_list();
716 return ret;
717 }
718
719 /*
720 * For each tracing session, update newly registered apps. The session list
721 * lock MUST be acquired before calling this.
722 */
723 static void update_ust_app(int app_sock)
724 {
725 struct ltt_session *sess, *stmp;
726
727 /* For all tracing session(s) */
728 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
729 session_lock(sess);
730 if (sess->ust_session) {
731 ust_app_global_update(sess->ust_session, app_sock);
732 }
733 session_unlock(sess);
734 }
735 }
736
737 /*
738 * This thread manage event coming from the kernel.
739 *
740 * Features supported in this thread:
741 * -) CPU Hotplug
742 */
743 static void *thread_manage_kernel(void *data)
744 {
745 int ret, i, pollfd, update_poll_flag = 1, err = -1;
746 uint32_t revents, nb_fd;
747 char tmp;
748 struct lttng_poll_event events;
749
750 DBG("[thread] Thread manage kernel started");
751
752 health_register(HEALTH_TYPE_KERNEL);
753
754 /*
755 * This first step of the while is to clean this structure which could free
756 * non NULL pointers so initialize it before the loop.
757 */
758 lttng_poll_init(&events);
759
760 if (testpoint(thread_manage_kernel)) {
761 goto error_testpoint;
762 }
763
764 health_code_update();
765
766 if (testpoint(thread_manage_kernel_before_loop)) {
767 goto error_testpoint;
768 }
769
770 while (1) {
771 health_code_update();
772
773 if (update_poll_flag == 1) {
774 /* Clean events object. We are about to populate it again. */
775 lttng_poll_clean(&events);
776
777 ret = sessiond_set_thread_pollset(&events, 2);
778 if (ret < 0) {
779 goto error_poll_create;
780 }
781
782 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
783 if (ret < 0) {
784 goto error;
785 }
786
787 /* This will add the available kernel channel if any. */
788 ret = update_kernel_poll(&events);
789 if (ret < 0) {
790 goto error;
791 }
792 update_poll_flag = 0;
793 }
794
795 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
796
797 /* Poll infinite value of time */
798 restart:
799 health_poll_entry();
800 ret = lttng_poll_wait(&events, -1);
801 health_poll_exit();
802 if (ret < 0) {
803 /*
804 * Restart interrupted system call.
805 */
806 if (errno == EINTR) {
807 goto restart;
808 }
809 goto error;
810 } else if (ret == 0) {
811 /* Should not happen since timeout is infinite */
812 ERR("Return value of poll is 0 with an infinite timeout.\n"
813 "This should not have happened! Continuing...");
814 continue;
815 }
816
817 nb_fd = ret;
818
819 for (i = 0; i < nb_fd; i++) {
820 /* Fetch once the poll data */
821 revents = LTTNG_POLL_GETEV(&events, i);
822 pollfd = LTTNG_POLL_GETFD(&events, i);
823
824 health_code_update();
825
826 /* Thread quit pipe has been closed. Killing thread. */
827 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
828 if (ret) {
829 err = 0;
830 goto exit;
831 }
832
833 /* Check for data on kernel pipe */
834 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
835 do {
836 ret = read(kernel_poll_pipe[0], &tmp, 1);
837 } while (ret < 0 && errno == EINTR);
838 /*
839 * Ret value is useless here, if this pipe gets any actions an
840 * update is required anyway.
841 */
842 update_poll_flag = 1;
843 continue;
844 } else {
845 /*
846 * New CPU detected by the kernel. Adding kernel stream to
847 * kernel session and updating the kernel consumer
848 */
849 if (revents & LPOLLIN) {
850 ret = update_kernel_stream(&kconsumer_data, pollfd);
851 if (ret < 0) {
852 continue;
853 }
854 break;
855 /*
856 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
857 * and unregister kernel stream at this point.
858 */
859 }
860 }
861 }
862 }
863
864 exit:
865 error:
866 lttng_poll_clean(&events);
867 error_poll_create:
868 error_testpoint:
869 utils_close_pipe(kernel_poll_pipe);
870 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
871 if (err) {
872 health_error();
873 ERR("Health error occurred in %s", __func__);
874 WARN("Kernel thread died unexpectedly. "
875 "Kernel tracing can continue but CPU hotplug is disabled.");
876 }
877 health_unregister();
878 DBG("Kernel thread dying");
879 return NULL;
880 }
881
882 /*
883 * Signal pthread condition of the consumer data that the thread.
884 */
885 static void signal_consumer_condition(struct consumer_data *data, int state)
886 {
887 pthread_mutex_lock(&data->cond_mutex);
888
889 /*
890 * The state is set before signaling. It can be any value, it's the waiter
891 * job to correctly interpret this condition variable associated to the
892 * consumer pthread_cond.
893 *
894 * A value of 0 means that the corresponding thread of the consumer data
895 * was not started. 1 indicates that the thread has started and is ready
896 * for action. A negative value means that there was an error during the
897 * thread bootstrap.
898 */
899 data->consumer_thread_is_ready = state;
900 (void) pthread_cond_signal(&data->cond);
901
902 pthread_mutex_unlock(&data->cond_mutex);
903 }
904
905 /*
906 * This thread manage the consumer error sent back to the session daemon.
907 */
908 static void *thread_manage_consumer(void *data)
909 {
910 int sock = -1, i, ret, pollfd, err = -1;
911 uint32_t revents, nb_fd;
912 enum lttcomm_return_code code;
913 struct lttng_poll_event events;
914 struct consumer_data *consumer_data = data;
915
916 DBG("[thread] Manage consumer started");
917
918 health_register(HEALTH_TYPE_CONSUMER);
919
920 health_code_update();
921
922 /*
923 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
924 * metadata_sock. Nothing more will be added to this poll set.
925 */
926 ret = sessiond_set_thread_pollset(&events, 3);
927 if (ret < 0) {
928 goto error_poll;
929 }
930
931 /*
932 * The error socket here is already in a listening state which was done
933 * just before spawning this thread to avoid a race between the consumer
934 * daemon exec trying to connect and the listen() call.
935 */
936 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
937 if (ret < 0) {
938 goto error;
939 }
940
941 health_code_update();
942
943 /* Infinite blocking call, waiting for transmission */
944 restart:
945 health_poll_entry();
946
947 if (testpoint(thread_manage_consumer)) {
948 goto error;
949 }
950
951 ret = lttng_poll_wait(&events, -1);
952 health_poll_exit();
953 if (ret < 0) {
954 /*
955 * Restart interrupted system call.
956 */
957 if (errno == EINTR) {
958 goto restart;
959 }
960 goto error;
961 }
962
963 nb_fd = ret;
964
965 for (i = 0; i < nb_fd; i++) {
966 /* Fetch once the poll data */
967 revents = LTTNG_POLL_GETEV(&events, i);
968 pollfd = LTTNG_POLL_GETFD(&events, i);
969
970 health_code_update();
971
972 /* Thread quit pipe has been closed. Killing thread. */
973 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
974 if (ret) {
975 err = 0;
976 goto exit;
977 }
978
979 /* Event on the registration socket */
980 if (pollfd == consumer_data->err_sock) {
981 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
982 ERR("consumer err socket poll error");
983 goto error;
984 }
985 }
986 }
987
988 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
989 if (sock < 0) {
990 goto error;
991 }
992
993 /*
994 * Set the CLOEXEC flag. Return code is useless because either way, the
995 * show must go on.
996 */
997 (void) utils_set_fd_cloexec(sock);
998
999 health_code_update();
1000
1001 DBG2("Receiving code from consumer err_sock");
1002
1003 /* Getting status code from kconsumerd */
1004 ret = lttcomm_recv_unix_sock(sock, &code,
1005 sizeof(enum lttcomm_return_code));
1006 if (ret <= 0) {
1007 goto error;
1008 }
1009
1010 health_code_update();
1011
1012 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1013 /* Connect both socket, command and metadata. */
1014 consumer_data->cmd_sock =
1015 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1016 consumer_data->metadata_sock.fd =
1017 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1018 if (consumer_data->cmd_sock < 0 ||
1019 consumer_data->metadata_sock.fd < 0) {
1020 PERROR("consumer connect cmd socket");
1021 /* On error, signal condition and quit. */
1022 signal_consumer_condition(consumer_data, -1);
1023 goto error;
1024 }
1025 /* Create metadata socket lock. */
1026 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1027 if (consumer_data->metadata_sock.lock == NULL) {
1028 PERROR("zmalloc pthread mutex");
1029 ret = -1;
1030 goto error;
1031 }
1032 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1033
1034 signal_consumer_condition(consumer_data, 1);
1035 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1036 DBG("Consumer metadata socket ready (fd: %d)",
1037 consumer_data->metadata_sock.fd);
1038 } else {
1039 ERR("consumer error when waiting for SOCK_READY : %s",
1040 lttcomm_get_readable_code(-code));
1041 goto error;
1042 }
1043
1044 /* Remove the consumerd error sock since we've established a connexion */
1045 ret = lttng_poll_del(&events, consumer_data->err_sock);
1046 if (ret < 0) {
1047 goto error;
1048 }
1049
1050 /* Add new accepted error socket. */
1051 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1052 if (ret < 0) {
1053 goto error;
1054 }
1055
1056 /* Add metadata socket that is successfully connected. */
1057 ret = lttng_poll_add(&events, consumer_data->metadata_sock.fd,
1058 LPOLLIN | LPOLLRDHUP);
1059 if (ret < 0) {
1060 goto error;
1061 }
1062
1063 health_code_update();
1064
1065 /* Infinite blocking call, waiting for transmission */
1066 restart_poll:
1067 while (1) {
1068 health_poll_entry();
1069 ret = lttng_poll_wait(&events, -1);
1070 health_poll_exit();
1071 if (ret < 0) {
1072 /*
1073 * Restart interrupted system call.
1074 */
1075 if (errno == EINTR) {
1076 goto restart_poll;
1077 }
1078 goto error;
1079 }
1080
1081 nb_fd = ret;
1082
1083 for (i = 0; i < nb_fd; i++) {
1084 /* Fetch once the poll data */
1085 revents = LTTNG_POLL_GETEV(&events, i);
1086 pollfd = LTTNG_POLL_GETFD(&events, i);
1087
1088 health_code_update();
1089
1090 /* Thread quit pipe has been closed. Killing thread. */
1091 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1092 if (ret) {
1093 err = 0;
1094 goto exit;
1095 }
1096
1097 if (pollfd == sock) {
1098 /* Event on the consumerd socket */
1099 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1100 ERR("consumer err socket second poll error");
1101 goto error;
1102 }
1103 health_code_update();
1104 /* Wait for any kconsumerd error */
1105 ret = lttcomm_recv_unix_sock(sock, &code,
1106 sizeof(enum lttcomm_return_code));
1107 if (ret <= 0) {
1108 ERR("consumer closed the command socket");
1109 goto error;
1110 }
1111
1112 ERR("consumer return code : %s",
1113 lttcomm_get_readable_code(-code));
1114
1115 goto exit;
1116 } else if (pollfd == consumer_data->metadata_sock.fd) {
1117 /* UST metadata requests */
1118 ret = ust_consumer_metadata_request(
1119 &consumer_data->metadata_sock);
1120 if (ret < 0) {
1121 ERR("Handling metadata request");
1122 goto error;
1123 }
1124 break;
1125 } else {
1126 ERR("Unknown pollfd");
1127 goto error;
1128 }
1129 }
1130 health_code_update();
1131 }
1132
1133 exit:
1134 error:
1135 /* Immediately set the consumerd state to stopped */
1136 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1137 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1138 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1139 consumer_data->type == LTTNG_CONSUMER32_UST) {
1140 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1141 } else {
1142 /* Code flow error... */
1143 assert(0);
1144 }
1145
1146 if (consumer_data->err_sock >= 0) {
1147 ret = close(consumer_data->err_sock);
1148 if (ret) {
1149 PERROR("close");
1150 }
1151 consumer_data->err_sock = -1;
1152 }
1153 if (consumer_data->cmd_sock >= 0) {
1154 ret = close(consumer_data->cmd_sock);
1155 if (ret) {
1156 PERROR("close");
1157 }
1158 consumer_data->cmd_sock = -1;
1159 }
1160 if (consumer_data->metadata_sock.fd >= 0) {
1161 ret = close(consumer_data->metadata_sock.fd);
1162 if (ret) {
1163 PERROR("close");
1164 }
1165 }
1166 /* Cleanup metadata socket mutex. */
1167 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1168 free(consumer_data->metadata_sock.lock);
1169
1170 if (sock >= 0) {
1171 ret = close(sock);
1172 if (ret) {
1173 PERROR("close");
1174 }
1175 }
1176
1177 unlink(consumer_data->err_unix_sock_path);
1178 unlink(consumer_data->cmd_unix_sock_path);
1179 consumer_data->pid = 0;
1180
1181 lttng_poll_clean(&events);
1182 error_poll:
1183 if (err) {
1184 health_error();
1185 ERR("Health error occurred in %s", __func__);
1186 }
1187 health_unregister();
1188 DBG("consumer thread cleanup completed");
1189
1190 return NULL;
1191 }
1192
1193 /*
1194 * This thread manage application communication.
1195 */
1196 static void *thread_manage_apps(void *data)
1197 {
1198 int i, ret, pollfd, err = -1;
1199 uint32_t revents, nb_fd;
1200 struct lttng_poll_event events;
1201
1202 DBG("[thread] Manage application started");
1203
1204 rcu_register_thread();
1205 rcu_thread_online();
1206
1207 health_register(HEALTH_TYPE_APP_MANAGE);
1208
1209 if (testpoint(thread_manage_apps)) {
1210 goto error_testpoint;
1211 }
1212
1213 health_code_update();
1214
1215 ret = sessiond_set_thread_pollset(&events, 2);
1216 if (ret < 0) {
1217 goto error_poll_create;
1218 }
1219
1220 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1221 if (ret < 0) {
1222 goto error;
1223 }
1224
1225 if (testpoint(thread_manage_apps_before_loop)) {
1226 goto error;
1227 }
1228
1229 health_code_update();
1230
1231 while (1) {
1232 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1233
1234 /* Inifinite blocking call, waiting for transmission */
1235 restart:
1236 health_poll_entry();
1237 ret = lttng_poll_wait(&events, -1);
1238 health_poll_exit();
1239 if (ret < 0) {
1240 /*
1241 * Restart interrupted system call.
1242 */
1243 if (errno == EINTR) {
1244 goto restart;
1245 }
1246 goto error;
1247 }
1248
1249 nb_fd = ret;
1250
1251 for (i = 0; i < nb_fd; i++) {
1252 /* Fetch once the poll data */
1253 revents = LTTNG_POLL_GETEV(&events, i);
1254 pollfd = LTTNG_POLL_GETFD(&events, i);
1255
1256 health_code_update();
1257
1258 /* Thread quit pipe has been closed. Killing thread. */
1259 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1260 if (ret) {
1261 err = 0;
1262 goto exit;
1263 }
1264
1265 /* Inspect the apps cmd pipe */
1266 if (pollfd == apps_cmd_pipe[0]) {
1267 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1268 ERR("Apps command pipe error");
1269 goto error;
1270 } else if (revents & LPOLLIN) {
1271 int sock;
1272
1273 /* Empty pipe */
1274 do {
1275 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1276 } while (ret < 0 && errno == EINTR);
1277 if (ret < 0 || ret < sizeof(sock)) {
1278 PERROR("read apps cmd pipe");
1279 goto error;
1280 }
1281
1282 health_code_update();
1283
1284 /*
1285 * We only monitor the error events of the socket. This
1286 * thread does not handle any incoming data from UST
1287 * (POLLIN).
1288 */
1289 ret = lttng_poll_add(&events, sock,
1290 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1291 if (ret < 0) {
1292 goto error;
1293 }
1294
1295 /* Set socket timeout for both receiving and ending */
1296 (void) lttcomm_setsockopt_rcv_timeout(sock,
1297 app_socket_timeout);
1298 (void) lttcomm_setsockopt_snd_timeout(sock,
1299 app_socket_timeout);
1300
1301 DBG("Apps with sock %d added to poll set", sock);
1302
1303 health_code_update();
1304
1305 break;
1306 }
1307 } else {
1308 /*
1309 * At this point, we know that a registered application made
1310 * the event at poll_wait.
1311 */
1312 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1313 /* Removing from the poll set */
1314 ret = lttng_poll_del(&events, pollfd);
1315 if (ret < 0) {
1316 goto error;
1317 }
1318
1319 /* Socket closed on remote end. */
1320 ust_app_unregister(pollfd);
1321 break;
1322 }
1323 }
1324
1325 health_code_update();
1326 }
1327 }
1328
1329 exit:
1330 error:
1331 lttng_poll_clean(&events);
1332 error_poll_create:
1333 error_testpoint:
1334 utils_close_pipe(apps_cmd_pipe);
1335 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1336
1337 /*
1338 * We don't clean the UST app hash table here since already registered
1339 * applications can still be controlled so let them be until the session
1340 * daemon dies or the applications stop.
1341 */
1342
1343 if (err) {
1344 health_error();
1345 ERR("Health error occurred in %s", __func__);
1346 }
1347 health_unregister();
1348 DBG("Application communication apps thread cleanup complete");
1349 rcu_thread_offline();
1350 rcu_unregister_thread();
1351 return NULL;
1352 }
1353
1354 /*
1355 * Send a socket to a thread This is called from the dispatch UST registration
1356 * thread once all sockets are set for the application.
1357 *
1358 * The sock value can be invalid, we don't really care, the thread will handle
1359 * it and make the necessary cleanup if so.
1360 *
1361 * On success, return 0 else a negative value being the errno message of the
1362 * write().
1363 */
1364 static int send_socket_to_thread(int fd, int sock)
1365 {
1366 int ret;
1367
1368 /*
1369 * It's possible that the FD is set as invalid with -1 concurrently just
1370 * before calling this function being a shutdown state of the thread.
1371 */
1372 if (fd < 0) {
1373 ret = -EBADF;
1374 goto error;
1375 }
1376
1377 do {
1378 ret = write(fd, &sock, sizeof(sock));
1379 } while (ret < 0 && errno == EINTR);
1380 if (ret < 0 || ret != sizeof(sock)) {
1381 PERROR("write apps pipe %d", fd);
1382 if (ret < 0) {
1383 ret = -errno;
1384 }
1385 goto error;
1386 }
1387
1388 /* All good. Don't send back the write positive ret value. */
1389 ret = 0;
1390 error:
1391 return ret;
1392 }
1393
1394 /*
1395 * Sanitize the wait queue of the dispatch registration thread meaning removing
1396 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1397 * notify socket is never received.
1398 */
1399 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1400 {
1401 int ret, nb_fd = 0, i;
1402 unsigned int fd_added = 0;
1403 struct lttng_poll_event events;
1404 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1405
1406 assert(wait_queue);
1407
1408 lttng_poll_init(&events);
1409
1410 /* Just skip everything for an empty queue. */
1411 if (!wait_queue->count) {
1412 goto end;
1413 }
1414
1415 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1416 if (ret < 0) {
1417 goto error_create;
1418 }
1419
1420 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1421 &wait_queue->head, head) {
1422 assert(wait_node->app);
1423 ret = lttng_poll_add(&events, wait_node->app->sock,
1424 LPOLLHUP | LPOLLERR);
1425 if (ret < 0) {
1426 goto error;
1427 }
1428
1429 fd_added = 1;
1430 }
1431
1432 if (!fd_added) {
1433 goto end;
1434 }
1435
1436 /*
1437 * Poll but don't block so we can quickly identify the faulty events and
1438 * clean them afterwards from the wait queue.
1439 */
1440 ret = lttng_poll_wait(&events, 0);
1441 if (ret < 0) {
1442 goto error;
1443 }
1444 nb_fd = ret;
1445
1446 for (i = 0; i < nb_fd; i++) {
1447 /* Get faulty FD. */
1448 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1449 int pollfd = LTTNG_POLL_GETFD(&events, i);
1450
1451 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1452 &wait_queue->head, head) {
1453 if (pollfd == wait_node->app->sock &&
1454 (revents & (LPOLLHUP | LPOLLERR))) {
1455 cds_list_del(&wait_node->head);
1456 wait_queue->count--;
1457 ust_app_destroy(wait_node->app);
1458 free(wait_node);
1459 break;
1460 }
1461 }
1462 }
1463
1464 if (nb_fd > 0) {
1465 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1466 }
1467
1468 end:
1469 lttng_poll_clean(&events);
1470 return;
1471
1472 error:
1473 lttng_poll_clean(&events);
1474 error_create:
1475 ERR("Unable to sanitize wait queue");
1476 return;
1477 }
1478
1479 /*
1480 * Dispatch request from the registration threads to the application
1481 * communication thread.
1482 */
1483 static void *thread_dispatch_ust_registration(void *data)
1484 {
1485 int ret, err = -1;
1486 struct cds_wfq_node *node;
1487 struct ust_command *ust_cmd = NULL;
1488 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1489 struct ust_reg_wait_queue wait_queue = {
1490 .count = 0,
1491 };
1492
1493 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1494
1495 health_code_update();
1496
1497 CDS_INIT_LIST_HEAD(&wait_queue.head);
1498
1499 DBG("[thread] Dispatch UST command started");
1500
1501 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1502 health_code_update();
1503
1504 /* Atomically prepare the queue futex */
1505 futex_nto1_prepare(&ust_cmd_queue.futex);
1506
1507 do {
1508 struct ust_app *app = NULL;
1509 ust_cmd = NULL;
1510
1511 /*
1512 * Make sure we don't have node(s) that have hung up before receiving
1513 * the notify socket. This is to clean the list in order to avoid
1514 * memory leaks from notify socket that are never seen.
1515 */
1516 sanitize_wait_queue(&wait_queue);
1517
1518 health_code_update();
1519 /* Dequeue command for registration */
1520 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1521 if (node == NULL) {
1522 DBG("Woken up but nothing in the UST command queue");
1523 /* Continue thread execution */
1524 break;
1525 }
1526
1527 ust_cmd = caa_container_of(node, struct ust_command, node);
1528
1529 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1530 " gid:%d sock:%d name:%s (version %d.%d)",
1531 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1532 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1533 ust_cmd->sock, ust_cmd->reg_msg.name,
1534 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1535
1536 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1537 wait_node = zmalloc(sizeof(*wait_node));
1538 if (!wait_node) {
1539 PERROR("zmalloc wait_node dispatch");
1540 ret = close(ust_cmd->sock);
1541 if (ret < 0) {
1542 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1543 }
1544 lttng_fd_put(1, LTTNG_FD_APPS);
1545 free(ust_cmd);
1546 goto error;
1547 }
1548 CDS_INIT_LIST_HEAD(&wait_node->head);
1549
1550 /* Create application object if socket is CMD. */
1551 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1552 ust_cmd->sock);
1553 if (!wait_node->app) {
1554 ret = close(ust_cmd->sock);
1555 if (ret < 0) {
1556 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1557 }
1558 lttng_fd_put(1, LTTNG_FD_APPS);
1559 free(wait_node);
1560 free(ust_cmd);
1561 continue;
1562 }
1563 /*
1564 * Add application to the wait queue so we can set the notify
1565 * socket before putting this object in the global ht.
1566 */
1567 cds_list_add(&wait_node->head, &wait_queue.head);
1568 wait_queue.count++;
1569
1570 free(ust_cmd);
1571 /*
1572 * We have to continue here since we don't have the notify
1573 * socket and the application MUST be added to the hash table
1574 * only at that moment.
1575 */
1576 continue;
1577 } else {
1578 /*
1579 * Look for the application in the local wait queue and set the
1580 * notify socket if found.
1581 */
1582 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1583 &wait_queue.head, head) {
1584 health_code_update();
1585 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1586 wait_node->app->notify_sock = ust_cmd->sock;
1587 cds_list_del(&wait_node->head);
1588 wait_queue.count--;
1589 app = wait_node->app;
1590 free(wait_node);
1591 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1592 break;
1593 }
1594 }
1595
1596 /*
1597 * With no application at this stage the received socket is
1598 * basically useless so close it before we free the cmd data
1599 * structure for good.
1600 */
1601 if (!app) {
1602 ret = close(ust_cmd->sock);
1603 if (ret < 0) {
1604 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1605 }
1606 lttng_fd_put(1, LTTNG_FD_APPS);
1607 }
1608 free(ust_cmd);
1609 }
1610
1611 if (app) {
1612 /*
1613 * @session_lock_list
1614 *
1615 * Lock the global session list so from the register up to the
1616 * registration done message, no thread can see the application
1617 * and change its state.
1618 */
1619 session_lock_list();
1620 rcu_read_lock();
1621
1622 /*
1623 * Add application to the global hash table. This needs to be
1624 * done before the update to the UST registry can locate the
1625 * application.
1626 */
1627 ust_app_add(app);
1628
1629 /* Set app version. This call will print an error if needed. */
1630 (void) ust_app_version(app);
1631
1632 /* Send notify socket through the notify pipe. */
1633 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1634 app->notify_sock);
1635 if (ret < 0) {
1636 rcu_read_unlock();
1637 session_unlock_list();
1638 /*
1639 * No notify thread, stop the UST tracing. However, this is
1640 * not an internal error of the this thread thus setting
1641 * the health error code to a normal exit.
1642 */
1643 err = 0;
1644 goto error;
1645 }
1646
1647 /*
1648 * Update newly registered application with the tracing
1649 * registry info already enabled information.
1650 */
1651 update_ust_app(app->sock);
1652
1653 /*
1654 * Don't care about return value. Let the manage apps threads
1655 * handle app unregistration upon socket close.
1656 */
1657 (void) ust_app_register_done(app->sock);
1658
1659 /*
1660 * Even if the application socket has been closed, send the app
1661 * to the thread and unregistration will take place at that
1662 * place.
1663 */
1664 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1665 if (ret < 0) {
1666 rcu_read_unlock();
1667 session_unlock_list();
1668 /*
1669 * No apps. thread, stop the UST tracing. However, this is
1670 * not an internal error of the this thread thus setting
1671 * the health error code to a normal exit.
1672 */
1673 err = 0;
1674 goto error;
1675 }
1676
1677 rcu_read_unlock();
1678 session_unlock_list();
1679 }
1680 } while (node != NULL);
1681
1682 health_poll_entry();
1683 /* Futex wait on queue. Blocking call on futex() */
1684 futex_nto1_wait(&ust_cmd_queue.futex);
1685 health_poll_exit();
1686 }
1687 /* Normal exit, no error */
1688 err = 0;
1689
1690 error:
1691 /* Clean up wait queue. */
1692 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1693 &wait_queue.head, head) {
1694 cds_list_del(&wait_node->head);
1695 wait_queue.count--;
1696 free(wait_node);
1697 }
1698
1699 DBG("Dispatch thread dying");
1700 if (err) {
1701 health_error();
1702 ERR("Health error occurred in %s", __func__);
1703 }
1704 health_unregister();
1705 return NULL;
1706 }
1707
1708 /*
1709 * This thread manage application registration.
1710 */
1711 static void *thread_registration_apps(void *data)
1712 {
1713 int sock = -1, i, ret, pollfd, err = -1;
1714 uint32_t revents, nb_fd;
1715 struct lttng_poll_event events;
1716 /*
1717 * Get allocated in this thread, enqueued to a global queue, dequeued and
1718 * freed in the manage apps thread.
1719 */
1720 struct ust_command *ust_cmd = NULL;
1721
1722 DBG("[thread] Manage application registration started");
1723
1724 health_register(HEALTH_TYPE_APP_REG);
1725
1726 if (testpoint(thread_registration_apps)) {
1727 goto error_testpoint;
1728 }
1729
1730 ret = lttcomm_listen_unix_sock(apps_sock);
1731 if (ret < 0) {
1732 goto error_listen;
1733 }
1734
1735 /*
1736 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1737 * more will be added to this poll set.
1738 */
1739 ret = sessiond_set_thread_pollset(&events, 2);
1740 if (ret < 0) {
1741 goto error_create_poll;
1742 }
1743
1744 /* Add the application registration socket */
1745 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1746 if (ret < 0) {
1747 goto error_poll_add;
1748 }
1749
1750 /* Notify all applications to register */
1751 ret = notify_ust_apps(1);
1752 if (ret < 0) {
1753 ERR("Failed to notify applications or create the wait shared memory.\n"
1754 "Execution continues but there might be problem for already\n"
1755 "running applications that wishes to register.");
1756 }
1757
1758 while (1) {
1759 DBG("Accepting application registration");
1760
1761 /* Inifinite blocking call, waiting for transmission */
1762 restart:
1763 health_poll_entry();
1764 ret = lttng_poll_wait(&events, -1);
1765 health_poll_exit();
1766 if (ret < 0) {
1767 /*
1768 * Restart interrupted system call.
1769 */
1770 if (errno == EINTR) {
1771 goto restart;
1772 }
1773 goto error;
1774 }
1775
1776 nb_fd = ret;
1777
1778 for (i = 0; i < nb_fd; i++) {
1779 health_code_update();
1780
1781 /* Fetch once the poll data */
1782 revents = LTTNG_POLL_GETEV(&events, i);
1783 pollfd = LTTNG_POLL_GETFD(&events, i);
1784
1785 /* Thread quit pipe has been closed. Killing thread. */
1786 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1787 if (ret) {
1788 err = 0;
1789 goto exit;
1790 }
1791
1792 /* Event on the registration socket */
1793 if (pollfd == apps_sock) {
1794 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1795 ERR("Register apps socket poll error");
1796 goto error;
1797 } else if (revents & LPOLLIN) {
1798 sock = lttcomm_accept_unix_sock(apps_sock);
1799 if (sock < 0) {
1800 goto error;
1801 }
1802
1803 /*
1804 * Set the CLOEXEC flag. Return code is useless because
1805 * either way, the show must go on.
1806 */
1807 (void) utils_set_fd_cloexec(sock);
1808
1809 /* Create UST registration command for enqueuing */
1810 ust_cmd = zmalloc(sizeof(struct ust_command));
1811 if (ust_cmd == NULL) {
1812 PERROR("ust command zmalloc");
1813 goto error;
1814 }
1815
1816 /*
1817 * Using message-based transmissions to ensure we don't
1818 * have to deal with partially received messages.
1819 */
1820 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1821 if (ret < 0) {
1822 ERR("Exhausted file descriptors allowed for applications.");
1823 free(ust_cmd);
1824 ret = close(sock);
1825 if (ret) {
1826 PERROR("close");
1827 }
1828 sock = -1;
1829 continue;
1830 }
1831
1832 health_code_update();
1833 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1834 if (ret < 0) {
1835 free(ust_cmd);
1836 /* Close socket of the application. */
1837 ret = close(sock);
1838 if (ret) {
1839 PERROR("close");
1840 }
1841 lttng_fd_put(LTTNG_FD_APPS, 1);
1842 sock = -1;
1843 continue;
1844 }
1845 health_code_update();
1846
1847 ust_cmd->sock = sock;
1848 sock = -1;
1849
1850 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1851 " gid:%d sock:%d name:%s (version %d.%d)",
1852 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1853 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1854 ust_cmd->sock, ust_cmd->reg_msg.name,
1855 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1856
1857 /*
1858 * Lock free enqueue the registration request. The red pill
1859 * has been taken! This apps will be part of the *system*.
1860 */
1861 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1862
1863 /*
1864 * Wake the registration queue futex. Implicit memory
1865 * barrier with the exchange in cds_wfq_enqueue.
1866 */
1867 futex_nto1_wake(&ust_cmd_queue.futex);
1868 }
1869 }
1870 }
1871 }
1872
1873 exit:
1874 error:
1875 if (err) {
1876 health_error();
1877 ERR("Health error occurred in %s", __func__);
1878 }
1879
1880 /* Notify that the registration thread is gone */
1881 notify_ust_apps(0);
1882
1883 if (apps_sock >= 0) {
1884 ret = close(apps_sock);
1885 if (ret) {
1886 PERROR("close");
1887 }
1888 }
1889 if (sock >= 0) {
1890 ret = close(sock);
1891 if (ret) {
1892 PERROR("close");
1893 }
1894 lttng_fd_put(LTTNG_FD_APPS, 1);
1895 }
1896 unlink(apps_unix_sock_path);
1897
1898 error_poll_add:
1899 lttng_poll_clean(&events);
1900 error_listen:
1901 error_create_poll:
1902 error_testpoint:
1903 DBG("UST Registration thread cleanup complete");
1904 health_unregister();
1905
1906 return NULL;
1907 }
1908
1909 /*
1910 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1911 * exec or it will fails.
1912 */
1913 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1914 {
1915 int ret, clock_ret;
1916 struct timespec timeout;
1917
1918 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1919 consumer_data->consumer_thread_is_ready = 0;
1920
1921 /* Setup pthread condition */
1922 ret = pthread_condattr_init(&consumer_data->condattr);
1923 if (ret != 0) {
1924 errno = ret;
1925 PERROR("pthread_condattr_init consumer data");
1926 goto error;
1927 }
1928
1929 /*
1930 * Set the monotonic clock in order to make sure we DO NOT jump in time
1931 * between the clock_gettime() call and the timedwait call. See bug #324
1932 * for a more details and how we noticed it.
1933 */
1934 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1935 if (ret != 0) {
1936 errno = ret;
1937 PERROR("pthread_condattr_setclock consumer data");
1938 goto error;
1939 }
1940
1941 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1942 if (ret != 0) {
1943 errno = ret;
1944 PERROR("pthread_cond_init consumer data");
1945 goto error;
1946 }
1947
1948 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1949 consumer_data);
1950 if (ret != 0) {
1951 PERROR("pthread_create consumer");
1952 ret = -1;
1953 goto error;
1954 }
1955
1956 /* We are about to wait on a pthread condition */
1957 pthread_mutex_lock(&consumer_data->cond_mutex);
1958
1959 /* Get time for sem_timedwait absolute timeout */
1960 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1961 /*
1962 * Set the timeout for the condition timed wait even if the clock gettime
1963 * call fails since we might loop on that call and we want to avoid to
1964 * increment the timeout too many times.
1965 */
1966 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1967
1968 /*
1969 * The following loop COULD be skipped in some conditions so this is why we
1970 * set ret to 0 in order to make sure at least one round of the loop is
1971 * done.
1972 */
1973 ret = 0;
1974
1975 /*
1976 * Loop until the condition is reached or when a timeout is reached. Note
1977 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1978 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1979 * possible. This loop does not take any chances and works with both of
1980 * them.
1981 */
1982 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1983 if (clock_ret < 0) {
1984 PERROR("clock_gettime spawn consumer");
1985 /* Infinite wait for the consumerd thread to be ready */
1986 ret = pthread_cond_wait(&consumer_data->cond,
1987 &consumer_data->cond_mutex);
1988 } else {
1989 ret = pthread_cond_timedwait(&consumer_data->cond,
1990 &consumer_data->cond_mutex, &timeout);
1991 }
1992 }
1993
1994 /* Release the pthread condition */
1995 pthread_mutex_unlock(&consumer_data->cond_mutex);
1996
1997 if (ret != 0) {
1998 errno = ret;
1999 if (ret == ETIMEDOUT) {
2000 /*
2001 * Call has timed out so we kill the kconsumerd_thread and return
2002 * an error.
2003 */
2004 ERR("Condition timed out. The consumer thread was never ready."
2005 " Killing it");
2006 ret = pthread_cancel(consumer_data->thread);
2007 if (ret < 0) {
2008 PERROR("pthread_cancel consumer thread");
2009 }
2010 } else {
2011 PERROR("pthread_cond_wait failed consumer thread");
2012 }
2013 goto error;
2014 }
2015
2016 pthread_mutex_lock(&consumer_data->pid_mutex);
2017 if (consumer_data->pid == 0) {
2018 ERR("Consumerd did not start");
2019 pthread_mutex_unlock(&consumer_data->pid_mutex);
2020 goto error;
2021 }
2022 pthread_mutex_unlock(&consumer_data->pid_mutex);
2023
2024 return 0;
2025
2026 error:
2027 return ret;
2028 }
2029
2030 /*
2031 * Join consumer thread
2032 */
2033 static int join_consumer_thread(struct consumer_data *consumer_data)
2034 {
2035 void *status;
2036
2037 /* Consumer pid must be a real one. */
2038 if (consumer_data->pid > 0) {
2039 int ret;
2040 ret = kill(consumer_data->pid, SIGTERM);
2041 if (ret) {
2042 ERR("Error killing consumer daemon");
2043 return ret;
2044 }
2045 return pthread_join(consumer_data->thread, &status);
2046 } else {
2047 return 0;
2048 }
2049 }
2050
2051 /*
2052 * Fork and exec a consumer daemon (consumerd).
2053 *
2054 * Return pid if successful else -1.
2055 */
2056 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2057 {
2058 int ret;
2059 pid_t pid;
2060 const char *consumer_to_use;
2061 const char *verbosity;
2062 struct stat st;
2063
2064 DBG("Spawning consumerd");
2065
2066 pid = fork();
2067 if (pid == 0) {
2068 /*
2069 * Exec consumerd.
2070 */
2071 if (opt_verbose_consumer) {
2072 verbosity = "--verbose";
2073 } else {
2074 verbosity = "--quiet";
2075 }
2076 switch (consumer_data->type) {
2077 case LTTNG_CONSUMER_KERNEL:
2078 /*
2079 * Find out which consumerd to execute. We will first try the
2080 * 64-bit path, then the sessiond's installation directory, and
2081 * fallback on the 32-bit one,
2082 */
2083 DBG3("Looking for a kernel consumer at these locations:");
2084 DBG3(" 1) %s", consumerd64_bin);
2085 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2086 DBG3(" 3) %s", consumerd32_bin);
2087 if (stat(consumerd64_bin, &st) == 0) {
2088 DBG3("Found location #1");
2089 consumer_to_use = consumerd64_bin;
2090 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2091 DBG3("Found location #2");
2092 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2093 } else if (stat(consumerd32_bin, &st) == 0) {
2094 DBG3("Found location #3");
2095 consumer_to_use = consumerd32_bin;
2096 } else {
2097 DBG("Could not find any valid consumerd executable");
2098 break;
2099 }
2100 DBG("Using kernel consumer at: %s", consumer_to_use);
2101 execl(consumer_to_use,
2102 "lttng-consumerd", verbosity, "-k",
2103 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2104 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2105 NULL);
2106 break;
2107 case LTTNG_CONSUMER64_UST:
2108 {
2109 char *tmpnew = NULL;
2110
2111 if (consumerd64_libdir[0] != '\0') {
2112 char *tmp;
2113 size_t tmplen;
2114
2115 tmp = getenv("LD_LIBRARY_PATH");
2116 if (!tmp) {
2117 tmp = "";
2118 }
2119 tmplen = strlen("LD_LIBRARY_PATH=")
2120 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2121 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2122 if (!tmpnew) {
2123 ret = -ENOMEM;
2124 goto error;
2125 }
2126 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2127 strcat(tmpnew, consumerd64_libdir);
2128 if (tmp[0] != '\0') {
2129 strcat(tmpnew, ":");
2130 strcat(tmpnew, tmp);
2131 }
2132 ret = putenv(tmpnew);
2133 if (ret) {
2134 ret = -errno;
2135 free(tmpnew);
2136 goto error;
2137 }
2138 }
2139 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2140 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2141 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2142 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2143 NULL);
2144 if (consumerd64_libdir[0] != '\0') {
2145 free(tmpnew);
2146 }
2147 if (ret) {
2148 goto error;
2149 }
2150 break;
2151 }
2152 case LTTNG_CONSUMER32_UST:
2153 {
2154 char *tmpnew = NULL;
2155
2156 if (consumerd32_libdir[0] != '\0') {
2157 char *tmp;
2158 size_t tmplen;
2159
2160 tmp = getenv("LD_LIBRARY_PATH");
2161 if (!tmp) {
2162 tmp = "";
2163 }
2164 tmplen = strlen("LD_LIBRARY_PATH=")
2165 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2166 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2167 if (!tmpnew) {
2168 ret = -ENOMEM;
2169 goto error;
2170 }
2171 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2172 strcat(tmpnew, consumerd32_libdir);
2173 if (tmp[0] != '\0') {
2174 strcat(tmpnew, ":");
2175 strcat(tmpnew, tmp);
2176 }
2177 ret = putenv(tmpnew);
2178 if (ret) {
2179 ret = -errno;
2180 free(tmpnew);
2181 goto error;
2182 }
2183 }
2184 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2185 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2186 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2187 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2188 NULL);
2189 if (consumerd32_libdir[0] != '\0') {
2190 free(tmpnew);
2191 }
2192 if (ret) {
2193 goto error;
2194 }
2195 break;
2196 }
2197 default:
2198 PERROR("unknown consumer type");
2199 exit(EXIT_FAILURE);
2200 }
2201 if (errno != 0) {
2202 PERROR("kernel start consumer exec");
2203 }
2204 exit(EXIT_FAILURE);
2205 } else if (pid > 0) {
2206 ret = pid;
2207 } else {
2208 PERROR("start consumer fork");
2209 ret = -errno;
2210 }
2211 error:
2212 return ret;
2213 }
2214
2215 /*
2216 * Spawn the consumerd daemon and session daemon thread.
2217 */
2218 static int start_consumerd(struct consumer_data *consumer_data)
2219 {
2220 int ret;
2221
2222 /*
2223 * Set the listen() state on the socket since there is a possible race
2224 * between the exec() of the consumer daemon and this call if place in the
2225 * consumer thread. See bug #366 for more details.
2226 */
2227 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2228 if (ret < 0) {
2229 goto error;
2230 }
2231
2232 pthread_mutex_lock(&consumer_data->pid_mutex);
2233 if (consumer_data->pid != 0) {
2234 pthread_mutex_unlock(&consumer_data->pid_mutex);
2235 goto end;
2236 }
2237
2238 ret = spawn_consumerd(consumer_data);
2239 if (ret < 0) {
2240 ERR("Spawning consumerd failed");
2241 pthread_mutex_unlock(&consumer_data->pid_mutex);
2242 goto error;
2243 }
2244
2245 /* Setting up the consumer_data pid */
2246 consumer_data->pid = ret;
2247 DBG2("Consumer pid %d", consumer_data->pid);
2248 pthread_mutex_unlock(&consumer_data->pid_mutex);
2249
2250 DBG2("Spawning consumer control thread");
2251 ret = spawn_consumer_thread(consumer_data);
2252 if (ret < 0) {
2253 ERR("Fatal error spawning consumer control thread");
2254 goto error;
2255 }
2256
2257 end:
2258 return 0;
2259
2260 error:
2261 /* Cleanup already created sockets on error. */
2262 if (consumer_data->err_sock >= 0) {
2263 int err;
2264
2265 err = close(consumer_data->err_sock);
2266 if (err < 0) {
2267 PERROR("close consumer data error socket");
2268 }
2269 }
2270 return ret;
2271 }
2272
2273 /*
2274 * Compute health status of each consumer. If one of them is zero (bad
2275 * state), we return 0.
2276 */
2277 static int check_consumer_health(void)
2278 {
2279 int ret;
2280
2281 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2282
2283 DBG3("Health consumer check %d", ret);
2284
2285 return ret;
2286 }
2287
2288 /*
2289 * Setup necessary data for kernel tracer action.
2290 */
2291 static int init_kernel_tracer(void)
2292 {
2293 int ret;
2294
2295 /* Modprobe lttng kernel modules */
2296 ret = modprobe_lttng_control();
2297 if (ret < 0) {
2298 goto error;
2299 }
2300
2301 /* Open debugfs lttng */
2302 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2303 if (kernel_tracer_fd < 0) {
2304 DBG("Failed to open %s", module_proc_lttng);
2305 ret = -1;
2306 goto error_open;
2307 }
2308
2309 /* Validate kernel version */
2310 ret = kernel_validate_version(kernel_tracer_fd);
2311 if (ret < 0) {
2312 goto error_version;
2313 }
2314
2315 ret = modprobe_lttng_data();
2316 if (ret < 0) {
2317 goto error_modules;
2318 }
2319
2320 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2321 return 0;
2322
2323 error_version:
2324 modprobe_remove_lttng_control();
2325 ret = close(kernel_tracer_fd);
2326 if (ret) {
2327 PERROR("close");
2328 }
2329 kernel_tracer_fd = -1;
2330 return LTTNG_ERR_KERN_VERSION;
2331
2332 error_modules:
2333 ret = close(kernel_tracer_fd);
2334 if (ret) {
2335 PERROR("close");
2336 }
2337
2338 error_open:
2339 modprobe_remove_lttng_control();
2340
2341 error:
2342 WARN("No kernel tracer available");
2343 kernel_tracer_fd = -1;
2344 if (!is_root) {
2345 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2346 } else {
2347 return LTTNG_ERR_KERN_NA;
2348 }
2349 }
2350
2351
2352 /*
2353 * Copy consumer output from the tracing session to the domain session. The
2354 * function also applies the right modification on a per domain basis for the
2355 * trace files destination directory.
2356 *
2357 * Should *NOT* be called with RCU read-side lock held.
2358 */
2359 static int copy_session_consumer(int domain, struct ltt_session *session)
2360 {
2361 int ret;
2362 const char *dir_name;
2363 struct consumer_output *consumer;
2364
2365 assert(session);
2366 assert(session->consumer);
2367
2368 switch (domain) {
2369 case LTTNG_DOMAIN_KERNEL:
2370 DBG3("Copying tracing session consumer output in kernel session");
2371 /*
2372 * XXX: We should audit the session creation and what this function
2373 * does "extra" in order to avoid a destroy since this function is used
2374 * in the domain session creation (kernel and ust) only. Same for UST
2375 * domain.
2376 */
2377 if (session->kernel_session->consumer) {
2378 consumer_destroy_output(session->kernel_session->consumer);
2379 }
2380 session->kernel_session->consumer =
2381 consumer_copy_output(session->consumer);
2382 /* Ease our life a bit for the next part */
2383 consumer = session->kernel_session->consumer;
2384 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2385 break;
2386 case LTTNG_DOMAIN_UST:
2387 DBG3("Copying tracing session consumer output in UST session");
2388 if (session->ust_session->consumer) {
2389 consumer_destroy_output(session->ust_session->consumer);
2390 }
2391 session->ust_session->consumer =
2392 consumer_copy_output(session->consumer);
2393 /* Ease our life a bit for the next part */
2394 consumer = session->ust_session->consumer;
2395 dir_name = DEFAULT_UST_TRACE_DIR;
2396 break;
2397 default:
2398 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2399 goto error;
2400 }
2401
2402 /* Append correct directory to subdir */
2403 strncat(consumer->subdir, dir_name,
2404 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2405 DBG3("Copy session consumer subdir %s", consumer->subdir);
2406
2407 ret = LTTNG_OK;
2408
2409 error:
2410 return ret;
2411 }
2412
2413 /*
2414 * Create an UST session and add it to the session ust list.
2415 *
2416 * Should *NOT* be called with RCU read-side lock held.
2417 */
2418 static int create_ust_session(struct ltt_session *session,
2419 struct lttng_domain *domain)
2420 {
2421 int ret;
2422 struct ltt_ust_session *lus = NULL;
2423
2424 assert(session);
2425 assert(domain);
2426 assert(session->consumer);
2427
2428 switch (domain->type) {
2429 case LTTNG_DOMAIN_UST:
2430 break;
2431 default:
2432 ERR("Unknown UST domain on create session %d", domain->type);
2433 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2434 goto error;
2435 }
2436
2437 DBG("Creating UST session");
2438
2439 lus = trace_ust_create_session(session->id);
2440 if (lus == NULL) {
2441 ret = LTTNG_ERR_UST_SESS_FAIL;
2442 goto error;
2443 }
2444
2445 lus->uid = session->uid;
2446 lus->gid = session->gid;
2447 session->ust_session = lus;
2448
2449 /* Copy session output to the newly created UST session */
2450 ret = copy_session_consumer(domain->type, session);
2451 if (ret != LTTNG_OK) {
2452 goto error;
2453 }
2454
2455 return LTTNG_OK;
2456
2457 error:
2458 free(lus);
2459 session->ust_session = NULL;
2460 return ret;
2461 }
2462
2463 /*
2464 * Create a kernel tracer session then create the default channel.
2465 */
2466 static int create_kernel_session(struct ltt_session *session)
2467 {
2468 int ret;
2469
2470 DBG("Creating kernel session");
2471
2472 ret = kernel_create_session(session, kernel_tracer_fd);
2473 if (ret < 0) {
2474 ret = LTTNG_ERR_KERN_SESS_FAIL;
2475 goto error;
2476 }
2477
2478 /* Code flow safety */
2479 assert(session->kernel_session);
2480
2481 /* Copy session output to the newly created Kernel session */
2482 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2483 if (ret != LTTNG_OK) {
2484 goto error;
2485 }
2486
2487 /* Create directory(ies) on local filesystem. */
2488 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2489 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2490 ret = run_as_mkdir_recursive(
2491 session->kernel_session->consumer->dst.trace_path,
2492 S_IRWXU | S_IRWXG, session->uid, session->gid);
2493 if (ret < 0) {
2494 if (ret != -EEXIST) {
2495 ERR("Trace directory creation error");
2496 goto error;
2497 }
2498 }
2499 }
2500
2501 session->kernel_session->uid = session->uid;
2502 session->kernel_session->gid = session->gid;
2503
2504 return LTTNG_OK;
2505
2506 error:
2507 trace_kernel_destroy_session(session->kernel_session);
2508 session->kernel_session = NULL;
2509 return ret;
2510 }
2511
2512 /*
2513 * Count number of session permitted by uid/gid.
2514 */
2515 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2516 {
2517 unsigned int i = 0;
2518 struct ltt_session *session;
2519
2520 DBG("Counting number of available session for UID %d GID %d",
2521 uid, gid);
2522 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2523 /*
2524 * Only list the sessions the user can control.
2525 */
2526 if (!session_access_ok(session, uid, gid)) {
2527 continue;
2528 }
2529 i++;
2530 }
2531 return i;
2532 }
2533
2534 /*
2535 * Process the command requested by the lttng client within the command
2536 * context structure. This function make sure that the return structure (llm)
2537 * is set and ready for transmission before returning.
2538 *
2539 * Return any error encountered or 0 for success.
2540 *
2541 * "sock" is only used for special-case var. len data.
2542 *
2543 * Should *NOT* be called with RCU read-side lock held.
2544 */
2545 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2546 int *sock_error)
2547 {
2548 int ret = LTTNG_OK;
2549 int need_tracing_session = 1;
2550 int need_domain;
2551
2552 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2553
2554 *sock_error = 0;
2555
2556 switch (cmd_ctx->lsm->cmd_type) {
2557 case LTTNG_CREATE_SESSION:
2558 case LTTNG_DESTROY_SESSION:
2559 case LTTNG_LIST_SESSIONS:
2560 case LTTNG_LIST_DOMAINS:
2561 case LTTNG_START_TRACE:
2562 case LTTNG_STOP_TRACE:
2563 case LTTNG_DATA_PENDING:
2564 need_domain = 0;
2565 break;
2566 default:
2567 need_domain = 1;
2568 }
2569
2570 if (opt_no_kernel && need_domain
2571 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2572 if (!is_root) {
2573 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2574 } else {
2575 ret = LTTNG_ERR_KERN_NA;
2576 }
2577 goto error;
2578 }
2579
2580 /* Deny register consumer if we already have a spawned consumer. */
2581 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2582 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2583 if (kconsumer_data.pid > 0) {
2584 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2585 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2586 goto error;
2587 }
2588 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2589 }
2590
2591 /*
2592 * Check for command that don't needs to allocate a returned payload. We do
2593 * this here so we don't have to make the call for no payload at each
2594 * command.
2595 */
2596 switch(cmd_ctx->lsm->cmd_type) {
2597 case LTTNG_LIST_SESSIONS:
2598 case LTTNG_LIST_TRACEPOINTS:
2599 case LTTNG_LIST_TRACEPOINT_FIELDS:
2600 case LTTNG_LIST_DOMAINS:
2601 case LTTNG_LIST_CHANNELS:
2602 case LTTNG_LIST_EVENTS:
2603 break;
2604 default:
2605 /* Setup lttng message with no payload */
2606 ret = setup_lttng_msg(cmd_ctx, 0);
2607 if (ret < 0) {
2608 /* This label does not try to unlock the session */
2609 goto init_setup_error;
2610 }
2611 }
2612
2613 /* Commands that DO NOT need a session. */
2614 switch (cmd_ctx->lsm->cmd_type) {
2615 case LTTNG_CREATE_SESSION:
2616 case LTTNG_CALIBRATE:
2617 case LTTNG_LIST_SESSIONS:
2618 case LTTNG_LIST_TRACEPOINTS:
2619 case LTTNG_LIST_TRACEPOINT_FIELDS:
2620 need_tracing_session = 0;
2621 break;
2622 default:
2623 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2624 /*
2625 * We keep the session list lock across _all_ commands
2626 * for now, because the per-session lock does not
2627 * handle teardown properly.
2628 */
2629 session_lock_list();
2630 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2631 if (cmd_ctx->session == NULL) {
2632 ret = LTTNG_ERR_SESS_NOT_FOUND;
2633 goto error;
2634 } else {
2635 /* Acquire lock for the session */
2636 session_lock(cmd_ctx->session);
2637 }
2638 break;
2639 }
2640
2641 if (!need_domain) {
2642 goto skip_domain;
2643 }
2644
2645 /*
2646 * Check domain type for specific "pre-action".
2647 */
2648 switch (cmd_ctx->lsm->domain.type) {
2649 case LTTNG_DOMAIN_KERNEL:
2650 if (!is_root) {
2651 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2652 goto error;
2653 }
2654
2655 /* Kernel tracer check */
2656 if (kernel_tracer_fd == -1) {
2657 /* Basically, load kernel tracer modules */
2658 ret = init_kernel_tracer();
2659 if (ret != 0) {
2660 goto error;
2661 }
2662 }
2663
2664 /* Consumer is in an ERROR state. Report back to client */
2665 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2666 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2667 goto error;
2668 }
2669
2670 /* Need a session for kernel command */
2671 if (need_tracing_session) {
2672 if (cmd_ctx->session->kernel_session == NULL) {
2673 ret = create_kernel_session(cmd_ctx->session);
2674 if (ret < 0) {
2675 ret = LTTNG_ERR_KERN_SESS_FAIL;
2676 goto error;
2677 }
2678 }
2679
2680 /* Start the kernel consumer daemon */
2681 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2682 if (kconsumer_data.pid == 0 &&
2683 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2684 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2685 ret = start_consumerd(&kconsumer_data);
2686 if (ret < 0) {
2687 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2688 goto error;
2689 }
2690 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2691 } else {
2692 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2693 }
2694
2695 /*
2696 * The consumer was just spawned so we need to add the socket to
2697 * the consumer output of the session if exist.
2698 */
2699 ret = consumer_create_socket(&kconsumer_data,
2700 cmd_ctx->session->kernel_session->consumer);
2701 if (ret < 0) {
2702 goto error;
2703 }
2704 }
2705
2706 break;
2707 case LTTNG_DOMAIN_UST:
2708 {
2709 if (!ust_app_supported()) {
2710 ret = LTTNG_ERR_NO_UST;
2711 goto error;
2712 }
2713 /* Consumer is in an ERROR state. Report back to client */
2714 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2715 ret = LTTNG_ERR_NO_USTCONSUMERD;
2716 goto error;
2717 }
2718
2719 if (need_tracing_session) {
2720 /* Create UST session if none exist. */
2721 if (cmd_ctx->session->ust_session == NULL) {
2722 ret = create_ust_session(cmd_ctx->session,
2723 &cmd_ctx->lsm->domain);
2724 if (ret != LTTNG_OK) {
2725 goto error;
2726 }
2727 }
2728
2729 /* Start the UST consumer daemons */
2730 /* 64-bit */
2731 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2732 if (consumerd64_bin[0] != '\0' &&
2733 ustconsumer64_data.pid == 0 &&
2734 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2735 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2736 ret = start_consumerd(&ustconsumer64_data);
2737 if (ret < 0) {
2738 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2739 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2740 goto error;
2741 }
2742
2743 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2744 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2745 } else {
2746 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2747 }
2748
2749 /*
2750 * Setup socket for consumer 64 bit. No need for atomic access
2751 * since it was set above and can ONLY be set in this thread.
2752 */
2753 ret = consumer_create_socket(&ustconsumer64_data,
2754 cmd_ctx->session->ust_session->consumer);
2755 if (ret < 0) {
2756 goto error;
2757 }
2758
2759 /* 32-bit */
2760 if (consumerd32_bin[0] != '\0' &&
2761 ustconsumer32_data.pid == 0 &&
2762 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2763 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2764 ret = start_consumerd(&ustconsumer32_data);
2765 if (ret < 0) {
2766 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2767 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2768 goto error;
2769 }
2770
2771 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2772 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2773 } else {
2774 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2775 }
2776
2777 /*
2778 * Setup socket for consumer 64 bit. No need for atomic access
2779 * since it was set above and can ONLY be set in this thread.
2780 */
2781 ret = consumer_create_socket(&ustconsumer32_data,
2782 cmd_ctx->session->ust_session->consumer);
2783 if (ret < 0) {
2784 goto error;
2785 }
2786 }
2787 break;
2788 }
2789 default:
2790 break;
2791 }
2792 skip_domain:
2793
2794 /* Validate consumer daemon state when start/stop trace command */
2795 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2796 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2797 switch (cmd_ctx->lsm->domain.type) {
2798 case LTTNG_DOMAIN_UST:
2799 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2800 ret = LTTNG_ERR_NO_USTCONSUMERD;
2801 goto error;
2802 }
2803 break;
2804 case LTTNG_DOMAIN_KERNEL:
2805 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2806 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2807 goto error;
2808 }
2809 break;
2810 }
2811 }
2812
2813 /*
2814 * Check that the UID or GID match that of the tracing session.
2815 * The root user can interact with all sessions.
2816 */
2817 if (need_tracing_session) {
2818 if (!session_access_ok(cmd_ctx->session,
2819 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2820 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2821 ret = LTTNG_ERR_EPERM;
2822 goto error;
2823 }
2824 }
2825
2826 /*
2827 * Send relayd information to consumer as soon as we have a domain and a
2828 * session defined.
2829 */
2830 if (cmd_ctx->session && need_domain) {
2831 /*
2832 * Setup relayd if not done yet. If the relayd information was already
2833 * sent to the consumer, this call will gracefully return.
2834 */
2835 ret = cmd_setup_relayd(cmd_ctx->session);
2836 if (ret != LTTNG_OK) {
2837 goto error;
2838 }
2839 }
2840
2841 /* Process by command type */
2842 switch (cmd_ctx->lsm->cmd_type) {
2843 case LTTNG_ADD_CONTEXT:
2844 {
2845 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2846 cmd_ctx->lsm->u.context.channel_name,
2847 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2848 break;
2849 }
2850 case LTTNG_DISABLE_CHANNEL:
2851 {
2852 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2853 cmd_ctx->lsm->u.disable.channel_name);
2854 break;
2855 }
2856 case LTTNG_DISABLE_EVENT:
2857 {
2858 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2859 cmd_ctx->lsm->u.disable.channel_name,
2860 cmd_ctx->lsm->u.disable.name);
2861 break;
2862 }
2863 case LTTNG_DISABLE_ALL_EVENT:
2864 {
2865 DBG("Disabling all events");
2866
2867 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2868 cmd_ctx->lsm->u.disable.channel_name);
2869 break;
2870 }
2871 case LTTNG_ENABLE_CHANNEL:
2872 {
2873 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2874 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2875 break;
2876 }
2877 case LTTNG_ENABLE_EVENT:
2878 {
2879 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2880 cmd_ctx->lsm->u.enable.channel_name,
2881 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2882 break;
2883 }
2884 case LTTNG_ENABLE_ALL_EVENT:
2885 {
2886 DBG("Enabling all events");
2887
2888 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2889 cmd_ctx->lsm->u.enable.channel_name,
2890 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2891 break;
2892 }
2893 case LTTNG_LIST_TRACEPOINTS:
2894 {
2895 struct lttng_event *events;
2896 ssize_t nb_events;
2897
2898 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2899 if (nb_events < 0) {
2900 /* Return value is a negative lttng_error_code. */
2901 ret = -nb_events;
2902 goto error;
2903 }
2904
2905 /*
2906 * Setup lttng message with payload size set to the event list size in
2907 * bytes and then copy list into the llm payload.
2908 */
2909 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2910 if (ret < 0) {
2911 free(events);
2912 goto setup_error;
2913 }
2914
2915 /* Copy event list into message payload */
2916 memcpy(cmd_ctx->llm->payload, events,
2917 sizeof(struct lttng_event) * nb_events);
2918
2919 free(events);
2920
2921 ret = LTTNG_OK;
2922 break;
2923 }
2924 case LTTNG_LIST_TRACEPOINT_FIELDS:
2925 {
2926 struct lttng_event_field *fields;
2927 ssize_t nb_fields;
2928
2929 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2930 &fields);
2931 if (nb_fields < 0) {
2932 /* Return value is a negative lttng_error_code. */
2933 ret = -nb_fields;
2934 goto error;
2935 }
2936
2937 /*
2938 * Setup lttng message with payload size set to the event list size in
2939 * bytes and then copy list into the llm payload.
2940 */
2941 ret = setup_lttng_msg(cmd_ctx,
2942 sizeof(struct lttng_event_field) * nb_fields);
2943 if (ret < 0) {
2944 free(fields);
2945 goto setup_error;
2946 }
2947
2948 /* Copy event list into message payload */
2949 memcpy(cmd_ctx->llm->payload, fields,
2950 sizeof(struct lttng_event_field) * nb_fields);
2951
2952 free(fields);
2953
2954 ret = LTTNG_OK;
2955 break;
2956 }
2957 case LTTNG_SET_CONSUMER_URI:
2958 {
2959 size_t nb_uri, len;
2960 struct lttng_uri *uris;
2961
2962 nb_uri = cmd_ctx->lsm->u.uri.size;
2963 len = nb_uri * sizeof(struct lttng_uri);
2964
2965 if (nb_uri == 0) {
2966 ret = LTTNG_ERR_INVALID;
2967 goto error;
2968 }
2969
2970 uris = zmalloc(len);
2971 if (uris == NULL) {
2972 ret = LTTNG_ERR_FATAL;
2973 goto error;
2974 }
2975
2976 /* Receive variable len data */
2977 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2978 ret = lttcomm_recv_unix_sock(sock, uris, len);
2979 if (ret <= 0) {
2980 DBG("No URIs received from client... continuing");
2981 *sock_error = 1;
2982 ret = LTTNG_ERR_SESSION_FAIL;
2983 free(uris);
2984 goto error;
2985 }
2986
2987 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2988 nb_uri, uris);
2989 if (ret != LTTNG_OK) {
2990 free(uris);
2991 goto error;
2992 }
2993
2994 /*
2995 * XXX: 0 means that this URI should be applied on the session. Should
2996 * be a DOMAIN enuam.
2997 */
2998 if (cmd_ctx->lsm->domain.type == 0) {
2999 /* Add the URI for the UST session if a consumer is present. */
3000 if (cmd_ctx->session->ust_session &&
3001 cmd_ctx->session->ust_session->consumer) {
3002 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3003 nb_uri, uris);
3004 } else if (cmd_ctx->session->kernel_session &&
3005 cmd_ctx->session->kernel_session->consumer) {
3006 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3007 cmd_ctx->session, nb_uri, uris);
3008 }
3009 }
3010
3011 free(uris);
3012
3013 break;
3014 }
3015 case LTTNG_START_TRACE:
3016 {
3017 ret = cmd_start_trace(cmd_ctx->session);
3018 break;
3019 }
3020 case LTTNG_STOP_TRACE:
3021 {
3022 ret = cmd_stop_trace(cmd_ctx->session);
3023 break;
3024 }
3025 case LTTNG_CREATE_SESSION:
3026 {
3027 size_t nb_uri, len;
3028 struct lttng_uri *uris = NULL;
3029
3030 nb_uri = cmd_ctx->lsm->u.uri.size;
3031 len = nb_uri * sizeof(struct lttng_uri);
3032
3033 if (nb_uri > 0) {
3034 uris = zmalloc(len);
3035 if (uris == NULL) {
3036 ret = LTTNG_ERR_FATAL;
3037 goto error;
3038 }
3039
3040 /* Receive variable len data */
3041 DBG("Waiting for %zu URIs from client ...", nb_uri);
3042 ret = lttcomm_recv_unix_sock(sock, uris, len);
3043 if (ret <= 0) {
3044 DBG("No URIs received from client... continuing");
3045 *sock_error = 1;
3046 ret = LTTNG_ERR_SESSION_FAIL;
3047 free(uris);
3048 goto error;
3049 }
3050
3051 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3052 DBG("Creating session with ONE network URI is a bad call");
3053 ret = LTTNG_ERR_SESSION_FAIL;
3054 free(uris);
3055 goto error;
3056 }
3057 }
3058
3059 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3060 &cmd_ctx->creds);
3061
3062 free(uris);
3063
3064 break;
3065 }
3066 case LTTNG_DESTROY_SESSION:
3067 {
3068 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3069
3070 /* Set session to NULL so we do not unlock it after free. */
3071 cmd_ctx->session = NULL;
3072 break;
3073 }
3074 case LTTNG_LIST_DOMAINS:
3075 {
3076 ssize_t nb_dom;
3077 struct lttng_domain *domains;
3078
3079 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3080 if (nb_dom < 0) {
3081 /* Return value is a negative lttng_error_code. */
3082 ret = -nb_dom;
3083 goto error;
3084 }
3085
3086 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3087 if (ret < 0) {
3088 free(domains);
3089 goto setup_error;
3090 }
3091
3092 /* Copy event list into message payload */
3093 memcpy(cmd_ctx->llm->payload, domains,
3094 nb_dom * sizeof(struct lttng_domain));
3095
3096 free(domains);
3097
3098 ret = LTTNG_OK;
3099 break;
3100 }
3101 case LTTNG_LIST_CHANNELS:
3102 {
3103 int nb_chan;
3104 struct lttng_channel *channels;
3105
3106 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3107 cmd_ctx->session, &channels);
3108 if (nb_chan < 0) {
3109 /* Return value is a negative lttng_error_code. */
3110 ret = -nb_chan;
3111 goto error;
3112 }
3113
3114 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3115 if (ret < 0) {
3116 free(channels);
3117 goto setup_error;
3118 }
3119
3120 /* Copy event list into message payload */
3121 memcpy(cmd_ctx->llm->payload, channels,
3122 nb_chan * sizeof(struct lttng_channel));
3123
3124 free(channels);
3125
3126 ret = LTTNG_OK;
3127 break;
3128 }
3129 case LTTNG_LIST_EVENTS:
3130 {
3131 ssize_t nb_event;
3132 struct lttng_event *events = NULL;
3133
3134 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3135 cmd_ctx->lsm->u.list.channel_name, &events);
3136 if (nb_event < 0) {
3137 /* Return value is a negative lttng_error_code. */
3138 ret = -nb_event;
3139 goto error;
3140 }
3141
3142 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3143 if (ret < 0) {
3144 free(events);
3145 goto setup_error;
3146 }
3147
3148 /* Copy event list into message payload */
3149 memcpy(cmd_ctx->llm->payload, events,
3150 nb_event * sizeof(struct lttng_event));
3151
3152 free(events);
3153
3154 ret = LTTNG_OK;
3155 break;
3156 }
3157 case LTTNG_LIST_SESSIONS:
3158 {
3159 unsigned int nr_sessions;
3160
3161 session_lock_list();
3162 nr_sessions = lttng_sessions_count(
3163 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3164 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3165
3166 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3167 if (ret < 0) {
3168 session_unlock_list();
3169 goto setup_error;
3170 }
3171
3172 /* Filled the session array */
3173 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3174 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3175 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3176
3177 session_unlock_list();
3178
3179 ret = LTTNG_OK;
3180 break;
3181 }
3182 case LTTNG_CALIBRATE:
3183 {
3184 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3185 &cmd_ctx->lsm->u.calibrate);
3186 break;
3187 }
3188 case LTTNG_REGISTER_CONSUMER:
3189 {
3190 struct consumer_data *cdata;
3191
3192 switch (cmd_ctx->lsm->domain.type) {
3193 case LTTNG_DOMAIN_KERNEL:
3194 cdata = &kconsumer_data;
3195 break;
3196 default:
3197 ret = LTTNG_ERR_UND;
3198 goto error;
3199 }
3200
3201 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3202 cmd_ctx->lsm->u.reg.path, cdata);
3203 break;
3204 }
3205 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3206 {
3207 struct lttng_filter_bytecode *bytecode;
3208
3209 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3210 ret = LTTNG_ERR_FILTER_INVAL;
3211 goto error;
3212 }
3213 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3214 ret = LTTNG_ERR_FILTER_INVAL;
3215 goto error;
3216 }
3217 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3218 if (!bytecode) {
3219 ret = LTTNG_ERR_FILTER_NOMEM;
3220 goto error;
3221 }
3222 /* Receive var. len. data */
3223 DBG("Receiving var len data from client ...");
3224 ret = lttcomm_recv_unix_sock(sock, bytecode,
3225 cmd_ctx->lsm->u.enable.bytecode_len);
3226 if (ret <= 0) {
3227 DBG("Nothing recv() from client var len data... continuing");
3228 *sock_error = 1;
3229 ret = LTTNG_ERR_FILTER_INVAL;
3230 goto error;
3231 }
3232
3233 if (bytecode->len + sizeof(*bytecode)
3234 != cmd_ctx->lsm->u.enable.bytecode_len) {
3235 free(bytecode);
3236 ret = LTTNG_ERR_FILTER_INVAL;
3237 goto error;
3238 }
3239
3240 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3241 cmd_ctx->lsm->u.enable.channel_name,
3242 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3243 break;
3244 }
3245 case LTTNG_DATA_PENDING:
3246 {
3247 ret = cmd_data_pending(cmd_ctx->session);
3248 break;
3249 }
3250 default:
3251 ret = LTTNG_ERR_UND;
3252 break;
3253 }
3254
3255 error:
3256 if (cmd_ctx->llm == NULL) {
3257 DBG("Missing llm structure. Allocating one.");
3258 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3259 goto setup_error;
3260 }
3261 }
3262 /* Set return code */
3263 cmd_ctx->llm->ret_code = ret;
3264 setup_error:
3265 if (cmd_ctx->session) {
3266 session_unlock(cmd_ctx->session);
3267 }
3268 if (need_tracing_session) {
3269 session_unlock_list();
3270 }
3271 init_setup_error:
3272 return ret;
3273 }
3274
3275 /*
3276 * Thread managing health check socket.
3277 */
3278 static void *thread_manage_health(void *data)
3279 {
3280 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3281 uint32_t revents, nb_fd;
3282 struct lttng_poll_event events;
3283 struct lttcomm_health_msg msg;
3284 struct lttcomm_health_data reply;
3285
3286 DBG("[thread] Manage health check started");
3287
3288 rcu_register_thread();
3289
3290 /* We might hit an error path before this is created. */
3291 lttng_poll_init(&events);
3292
3293 /* Create unix socket */
3294 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3295 if (sock < 0) {
3296 ERR("Unable to create health check Unix socket");
3297 ret = -1;
3298 goto error;
3299 }
3300
3301 /*
3302 * Set the CLOEXEC flag. Return code is useless because either way, the
3303 * show must go on.
3304 */
3305 (void) utils_set_fd_cloexec(sock);
3306
3307 ret = lttcomm_listen_unix_sock(sock);
3308 if (ret < 0) {
3309 goto error;
3310 }
3311
3312 /*
3313 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3314 * more will be added to this poll set.
3315 */
3316 ret = sessiond_set_thread_pollset(&events, 2);
3317 if (ret < 0) {
3318 goto error;
3319 }
3320
3321 /* Add the application registration socket */
3322 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3323 if (ret < 0) {
3324 goto error;
3325 }
3326
3327 while (1) {
3328 DBG("Health check ready");
3329
3330 /* Inifinite blocking call, waiting for transmission */
3331 restart:
3332 ret = lttng_poll_wait(&events, -1);
3333 if (ret < 0) {
3334 /*
3335 * Restart interrupted system call.
3336 */
3337 if (errno == EINTR) {
3338 goto restart;
3339 }
3340 goto error;
3341 }
3342
3343 nb_fd = ret;
3344
3345 for (i = 0; i < nb_fd; i++) {
3346 /* Fetch once the poll data */
3347 revents = LTTNG_POLL_GETEV(&events, i);
3348 pollfd = LTTNG_POLL_GETFD(&events, i);
3349
3350 /* Thread quit pipe has been closed. Killing thread. */
3351 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3352 if (ret) {
3353 err = 0;
3354 goto exit;
3355 }
3356
3357 /* Event on the registration socket */
3358 if (pollfd == sock) {
3359 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3360 ERR("Health socket poll error");
3361 goto error;
3362 }
3363 }
3364 }
3365
3366 new_sock = lttcomm_accept_unix_sock(sock);
3367 if (new_sock < 0) {
3368 goto error;
3369 }
3370
3371 /*
3372 * Set the CLOEXEC flag. Return code is useless because either way, the
3373 * show must go on.
3374 */
3375 (void) utils_set_fd_cloexec(new_sock);
3376
3377 DBG("Receiving data from client for health...");
3378 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3379 if (ret <= 0) {
3380 DBG("Nothing recv() from client... continuing");
3381 ret = close(new_sock);
3382 if (ret) {
3383 PERROR("close");
3384 }
3385 new_sock = -1;
3386 continue;
3387 }
3388
3389 rcu_thread_online();
3390
3391 switch (msg.component) {
3392 case LTTNG_HEALTH_CMD:
3393 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3394 break;
3395 case LTTNG_HEALTH_APP_MANAGE:
3396 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3397 break;
3398 case LTTNG_HEALTH_APP_REG:
3399 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3400 break;
3401 case LTTNG_HEALTH_KERNEL:
3402 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3403 break;
3404 case LTTNG_HEALTH_CONSUMER:
3405 reply.ret_code = check_consumer_health();
3406 break;
3407 case LTTNG_HEALTH_HT_CLEANUP:
3408 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3409 break;
3410 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3411 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3412 break;
3413 case LTTNG_HEALTH_APP_REG_DISPATCH:
3414 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3415 break;
3416 case LTTNG_HEALTH_ALL:
3417 reply.ret_code =
3418 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3419 health_check_state(HEALTH_TYPE_APP_REG) &&
3420 health_check_state(HEALTH_TYPE_CMD) &&
3421 health_check_state(HEALTH_TYPE_KERNEL) &&
3422 check_consumer_health() &&
3423 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
3424 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3425 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3426 break;
3427 default:
3428 reply.ret_code = LTTNG_ERR_UND;
3429 break;
3430 }
3431
3432 /*
3433 * Flip ret value since 0 is a success and 1 indicates a bad health for
3434 * the client where in the sessiond it is the opposite. Again, this is
3435 * just to make things easier for us poor developer which enjoy a lot
3436 * lazyness.
3437 */
3438 if (reply.ret_code == 0 || reply.ret_code == 1) {
3439 reply.ret_code = !reply.ret_code;
3440 }
3441
3442 DBG2("Health check return value %d", reply.ret_code);
3443
3444 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3445 if (ret < 0) {
3446 ERR("Failed to send health data back to client");
3447 }
3448
3449 /* End of transmission */
3450 ret = close(new_sock);
3451 if (ret) {
3452 PERROR("close");
3453 }
3454 new_sock = -1;
3455 }
3456
3457 exit:
3458 error:
3459 if (err) {
3460 ERR("Health error occurred in %s", __func__);
3461 }
3462 DBG("Health check thread dying");
3463 unlink(health_unix_sock_path);
3464 if (sock >= 0) {
3465 ret = close(sock);
3466 if (ret) {
3467 PERROR("close");
3468 }
3469 }
3470
3471 lttng_poll_clean(&events);
3472
3473 rcu_unregister_thread();
3474 return NULL;
3475 }
3476
3477 /*
3478 * This thread manage all clients request using the unix client socket for
3479 * communication.
3480 */
3481 static void *thread_manage_clients(void *data)
3482 {
3483 int sock = -1, ret, i, pollfd, err = -1;
3484 int sock_error;
3485 uint32_t revents, nb_fd;
3486 struct command_ctx *cmd_ctx = NULL;
3487 struct lttng_poll_event events;
3488
3489 DBG("[thread] Manage client started");
3490
3491 rcu_register_thread();
3492
3493 health_register(HEALTH_TYPE_CMD);
3494
3495 if (testpoint(thread_manage_clients)) {
3496 goto error_testpoint;
3497 }
3498
3499 health_code_update();
3500
3501 ret = lttcomm_listen_unix_sock(client_sock);
3502 if (ret < 0) {
3503 goto error_listen;
3504 }
3505
3506 /*
3507 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3508 * more will be added to this poll set.
3509 */
3510 ret = sessiond_set_thread_pollset(&events, 2);
3511 if (ret < 0) {
3512 goto error_create_poll;
3513 }
3514
3515 /* Add the application registration socket */
3516 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3517 if (ret < 0) {
3518 goto error;
3519 }
3520
3521 /*
3522 * Notify parent pid that we are ready to accept command for client side.
3523 */
3524 if (opt_sig_parent) {
3525 kill(ppid, SIGUSR1);
3526 }
3527
3528 if (testpoint(thread_manage_clients_before_loop)) {
3529 goto error;
3530 }
3531
3532 health_code_update();
3533
3534 while (1) {
3535 DBG("Accepting client command ...");
3536
3537 /* Inifinite blocking call, waiting for transmission */
3538 restart:
3539 health_poll_entry();
3540 ret = lttng_poll_wait(&events, -1);
3541 health_poll_exit();
3542 if (ret < 0) {
3543 /*
3544 * Restart interrupted system call.
3545 */
3546 if (errno == EINTR) {
3547 goto restart;
3548 }
3549 goto error;
3550 }
3551
3552 nb_fd = ret;
3553
3554 for (i = 0; i < nb_fd; i++) {
3555 /* Fetch once the poll data */
3556 revents = LTTNG_POLL_GETEV(&events, i);
3557 pollfd = LTTNG_POLL_GETFD(&events, i);
3558
3559 health_code_update();
3560
3561 /* Thread quit pipe has been closed. Killing thread. */
3562 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3563 if (ret) {
3564 err = 0;
3565 goto exit;
3566 }
3567
3568 /* Event on the registration socket */
3569 if (pollfd == client_sock) {
3570 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3571 ERR("Client socket poll error");
3572 goto error;
3573 }
3574 }
3575 }
3576
3577 DBG("Wait for client response");
3578
3579 health_code_update();
3580
3581 sock = lttcomm_accept_unix_sock(client_sock);
3582 if (sock < 0) {
3583 goto error;
3584 }
3585
3586 /*
3587 * Set the CLOEXEC flag. Return code is useless because either way, the
3588 * show must go on.
3589 */
3590 (void) utils_set_fd_cloexec(sock);
3591
3592 /* Set socket option for credentials retrieval */
3593 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3594 if (ret < 0) {
3595 goto error;
3596 }
3597
3598 /* Allocate context command to process the client request */
3599 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3600 if (cmd_ctx == NULL) {
3601 PERROR("zmalloc cmd_ctx");
3602 goto error;
3603 }
3604
3605 /* Allocate data buffer for reception */
3606 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3607 if (cmd_ctx->lsm == NULL) {
3608 PERROR("zmalloc cmd_ctx->lsm");
3609 goto error;
3610 }
3611
3612 cmd_ctx->llm = NULL;
3613 cmd_ctx->session = NULL;
3614
3615 health_code_update();
3616
3617 /*
3618 * Data is received from the lttng client. The struct
3619 * lttcomm_session_msg (lsm) contains the command and data request of
3620 * the client.
3621 */
3622 DBG("Receiving data from client ...");
3623 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3624 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3625 if (ret <= 0) {
3626 DBG("Nothing recv() from client... continuing");
3627 ret = close(sock);
3628 if (ret) {
3629 PERROR("close");
3630 }
3631 sock = -1;
3632 clean_command_ctx(&cmd_ctx);
3633 continue;
3634 }
3635
3636 health_code_update();
3637
3638 // TODO: Validate cmd_ctx including sanity check for
3639 // security purpose.
3640
3641 rcu_thread_online();
3642 /*
3643 * This function dispatch the work to the kernel or userspace tracer
3644 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3645 * informations for the client. The command context struct contains
3646 * everything this function may needs.
3647 */
3648 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3649 rcu_thread_offline();
3650 if (ret < 0) {
3651 ret = close(sock);
3652 if (ret) {
3653 PERROR("close");
3654 }
3655 sock = -1;
3656 /*
3657 * TODO: Inform client somehow of the fatal error. At
3658 * this point, ret < 0 means that a zmalloc failed
3659 * (ENOMEM). Error detected but still accept
3660 * command, unless a socket error has been
3661 * detected.
3662 */
3663 clean_command_ctx(&cmd_ctx);
3664 continue;
3665 }
3666
3667 health_code_update();
3668
3669 DBG("Sending response (size: %d, retcode: %s)",
3670 cmd_ctx->lttng_msg_size,
3671 lttng_strerror(-cmd_ctx->llm->ret_code));
3672 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3673 if (ret < 0) {
3674 ERR("Failed to send data back to client");
3675 }
3676
3677 /* End of transmission */
3678 ret = close(sock);
3679 if (ret) {
3680 PERROR("close");
3681 }
3682 sock = -1;
3683
3684 clean_command_ctx(&cmd_ctx);
3685
3686 health_code_update();
3687 }
3688
3689 exit:
3690 error:
3691 if (sock >= 0) {
3692 ret = close(sock);
3693 if (ret) {
3694 PERROR("close");
3695 }
3696 }
3697
3698 lttng_poll_clean(&events);
3699 clean_command_ctx(&cmd_ctx);
3700
3701 error_listen:
3702 error_create_poll:
3703 error_testpoint:
3704 unlink(client_unix_sock_path);
3705 if (client_sock >= 0) {
3706 ret = close(client_sock);
3707 if (ret) {
3708 PERROR("close");
3709 }
3710 }
3711
3712 if (err) {
3713 health_error();
3714 ERR("Health error occurred in %s", __func__);
3715 }
3716
3717 health_unregister();
3718
3719 DBG("Client thread dying");
3720
3721 rcu_unregister_thread();
3722 return NULL;
3723 }
3724
3725
3726 /*
3727 * usage function on stderr
3728 */
3729 static void usage(void)
3730 {
3731 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3732 fprintf(stderr, " -h, --help Display this usage.\n");
3733 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3734 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3735 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3736 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3737 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3738 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3739 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3740 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3741 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3742 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3743 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3744 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3745 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3746 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3747 fprintf(stderr, " -V, --version Show version number.\n");
3748 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3749 fprintf(stderr, " -q, --quiet No output at all.\n");
3750 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3751 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3752 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3753 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3754 }
3755
3756 /*
3757 * daemon argument parsing
3758 */
3759 static int parse_args(int argc, char **argv)
3760 {
3761 int c;
3762
3763 static struct option long_options[] = {
3764 { "client-sock", 1, 0, 'c' },
3765 { "apps-sock", 1, 0, 'a' },
3766 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3767 { "kconsumerd-err-sock", 1, 0, 'E' },
3768 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3769 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3770 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3771 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3772 { "consumerd32-path", 1, 0, 'u' },
3773 { "consumerd32-libdir", 1, 0, 'U' },
3774 { "consumerd64-path", 1, 0, 't' },
3775 { "consumerd64-libdir", 1, 0, 'T' },
3776 { "daemonize", 0, 0, 'd' },
3777 { "sig-parent", 0, 0, 'S' },
3778 { "help", 0, 0, 'h' },
3779 { "group", 1, 0, 'g' },
3780 { "version", 0, 0, 'V' },
3781 { "quiet", 0, 0, 'q' },
3782 { "verbose", 0, 0, 'v' },
3783 { "verbose-consumer", 0, 0, 'Z' },
3784 { "no-kernel", 0, 0, 'N' },
3785 { "pidfile", 1, 0, 'p' },
3786 { NULL, 0, 0, 0 }
3787 };
3788
3789 while (1) {
3790 int option_index = 0;
3791 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3792 long_options, &option_index);
3793 if (c == -1) {
3794 break;
3795 }
3796
3797 switch (c) {
3798 case 0:
3799 fprintf(stderr, "option %s", long_options[option_index].name);
3800 if (optarg) {
3801 fprintf(stderr, " with arg %s\n", optarg);
3802 }
3803 break;
3804 case 'c':
3805 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3806 break;
3807 case 'a':
3808 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3809 break;
3810 case 'd':
3811 opt_daemon = 1;
3812 break;
3813 case 'g':
3814 opt_tracing_group = optarg;
3815 break;
3816 case 'h':
3817 usage();
3818 exit(EXIT_FAILURE);
3819 case 'V':
3820 fprintf(stdout, "%s\n", VERSION);
3821 exit(EXIT_SUCCESS);
3822 case 'S':
3823 opt_sig_parent = 1;
3824 break;
3825 case 'E':
3826 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3827 break;
3828 case 'C':
3829 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3830 break;
3831 case 'F':
3832 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3833 break;
3834 case 'D':
3835 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3836 break;
3837 case 'H':
3838 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3839 break;
3840 case 'G':
3841 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3842 break;
3843 case 'N':
3844 opt_no_kernel = 1;
3845 break;
3846 case 'q':
3847 lttng_opt_quiet = 1;
3848 break;
3849 case 'v':
3850 /* Verbose level can increase using multiple -v */
3851 lttng_opt_verbose += 1;
3852 break;
3853 case 'Z':
3854 opt_verbose_consumer += 1;
3855 break;
3856 case 'u':
3857 consumerd32_bin= optarg;
3858 break;
3859 case 'U':
3860 consumerd32_libdir = optarg;
3861 break;
3862 case 't':
3863 consumerd64_bin = optarg;
3864 break;
3865 case 'T':
3866 consumerd64_libdir = optarg;
3867 break;
3868 case 'p':
3869 opt_pidfile = optarg;
3870 break;
3871 default:
3872 /* Unknown option or other error.
3873 * Error is printed by getopt, just return */
3874 return -1;
3875 }
3876 }
3877
3878 return 0;
3879 }
3880
3881 /*
3882 * Creates the two needed socket by the daemon.
3883 * apps_sock - The communication socket for all UST apps.
3884 * client_sock - The communication of the cli tool (lttng).
3885 */
3886 static int init_daemon_socket(void)
3887 {
3888 int ret = 0;
3889 mode_t old_umask;
3890
3891 old_umask = umask(0);
3892
3893 /* Create client tool unix socket */
3894 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3895 if (client_sock < 0) {
3896 ERR("Create unix sock failed: %s", client_unix_sock_path);
3897 ret = -1;
3898 goto end;
3899 }
3900
3901 /* Set the cloexec flag */
3902 ret = utils_set_fd_cloexec(client_sock);
3903 if (ret < 0) {
3904 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3905 "Continuing but note that the consumer daemon will have a "
3906 "reference to this socket on exec()", client_sock);
3907 }
3908
3909 /* File permission MUST be 660 */
3910 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3911 if (ret < 0) {
3912 ERR("Set file permissions failed: %s", client_unix_sock_path);
3913 PERROR("chmod");
3914 goto end;
3915 }
3916
3917 /* Create the application unix socket */
3918 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3919 if (apps_sock < 0) {
3920 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3921 ret = -1;
3922 goto end;
3923 }
3924
3925 /* Set the cloexec flag */
3926 ret = utils_set_fd_cloexec(apps_sock);
3927 if (ret < 0) {
3928 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3929 "Continuing but note that the consumer daemon will have a "
3930 "reference to this socket on exec()", apps_sock);
3931 }
3932
3933 /* File permission MUST be 666 */
3934 ret = chmod(apps_unix_sock_path,
3935 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3936 if (ret < 0) {
3937 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3938 PERROR("chmod");
3939 goto end;
3940 }
3941
3942 DBG3("Session daemon client socket %d and application socket %d created",
3943 client_sock, apps_sock);
3944
3945 end:
3946 umask(old_umask);
3947 return ret;
3948 }
3949
3950 /*
3951 * Check if the global socket is available, and if a daemon is answering at the
3952 * other side. If yes, error is returned.
3953 */
3954 static int check_existing_daemon(void)
3955 {
3956 /* Is there anybody out there ? */
3957 if (lttng_session_daemon_alive()) {
3958 return -EEXIST;
3959 }
3960
3961 return 0;
3962 }
3963
3964 /*
3965 * Set the tracing group gid onto the client socket.
3966 *
3967 * Race window between mkdir and chown is OK because we are going from more
3968 * permissive (root.root) to less permissive (root.tracing).
3969 */
3970 static int set_permissions(char *rundir)
3971 {
3972 int ret;
3973 gid_t gid;
3974
3975 ret = allowed_group();
3976 if (ret < 0) {
3977 WARN("No tracing group detected");
3978 /* Setting gid to 0 if no tracing group is found */
3979 gid = 0;
3980 } else {
3981 gid = ret;
3982 }
3983
3984 /* Set lttng run dir */
3985 ret = chown(rundir, 0, gid);
3986 if (ret < 0) {
3987 ERR("Unable to set group on %s", rundir);
3988 PERROR("chown");
3989 }
3990
3991 /* Ensure all applications and tracing group can search the run dir */
3992 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3993 if (ret < 0) {
3994 ERR("Unable to set permissions on %s", rundir);
3995 PERROR("chmod");
3996 }
3997
3998 /* lttng client socket path */
3999 ret = chown(client_unix_sock_path, 0, gid);
4000 if (ret < 0) {
4001 ERR("Unable to set group on %s", client_unix_sock_path);
4002 PERROR("chown");
4003 }
4004
4005 /* kconsumer error socket path */
4006 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
4007 if (ret < 0) {
4008 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4009 PERROR("chown");
4010 }
4011
4012 /* 64-bit ustconsumer error socket path */
4013 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
4014 if (ret < 0) {
4015 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4016 PERROR("chown");
4017 }
4018
4019 /* 32-bit ustconsumer compat32 error socket path */
4020 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
4021 if (ret < 0) {
4022 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4023 PERROR("chown");
4024 }
4025
4026 DBG("All permissions are set");
4027
4028 return ret;
4029 }
4030
4031 /*
4032 * Create the lttng run directory needed for all global sockets and pipe.
4033 */
4034 static int create_lttng_rundir(const char *rundir)
4035 {
4036 int ret;
4037
4038 DBG3("Creating LTTng run directory: %s", rundir);
4039
4040 ret = mkdir(rundir, S_IRWXU);
4041 if (ret < 0) {
4042 if (errno != EEXIST) {
4043 ERR("Unable to create %s", rundir);
4044 goto error;
4045 } else {
4046 ret = 0;
4047 }
4048 }
4049
4050 error:
4051 return ret;
4052 }
4053
4054 /*
4055 * Setup sockets and directory needed by the kconsumerd communication with the
4056 * session daemon.
4057 */
4058 static int set_consumer_sockets(struct consumer_data *consumer_data,
4059 const char *rundir)
4060 {
4061 int ret;
4062 char path[PATH_MAX];
4063
4064 switch (consumer_data->type) {
4065 case LTTNG_CONSUMER_KERNEL:
4066 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4067 break;
4068 case LTTNG_CONSUMER64_UST:
4069 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4070 break;
4071 case LTTNG_CONSUMER32_UST:
4072 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4073 break;
4074 default:
4075 ERR("Consumer type unknown");
4076 ret = -EINVAL;
4077 goto error;
4078 }
4079
4080 DBG2("Creating consumer directory: %s", path);
4081
4082 ret = mkdir(path, S_IRWXU);
4083 if (ret < 0) {
4084 if (errno != EEXIST) {
4085 PERROR("mkdir");
4086 ERR("Failed to create %s", path);
4087 goto error;
4088 }
4089 ret = -1;
4090 }
4091
4092 /* Create the kconsumerd error unix socket */
4093 consumer_data->err_sock =
4094 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4095 if (consumer_data->err_sock < 0) {
4096 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4097 ret = -1;
4098 goto error;
4099 }
4100
4101 /*
4102 * Set the CLOEXEC flag. Return code is useless because either way, the
4103 * show must go on.
4104 */
4105 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4106 if (ret < 0) {
4107 PERROR("utils_set_fd_cloexec");
4108 /* continue anyway */
4109 }
4110
4111 /* File permission MUST be 660 */
4112 ret = chmod(consumer_data->err_unix_sock_path,
4113 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4114 if (ret < 0) {
4115 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4116 PERROR("chmod");
4117 goto error;
4118 }
4119
4120 error:
4121 return ret;
4122 }
4123
4124 /*
4125 * Signal handler for the daemon
4126 *
4127 * Simply stop all worker threads, leaving main() return gracefully after
4128 * joining all threads and calling cleanup().
4129 */
4130 static void sighandler(int sig)
4131 {
4132 switch (sig) {
4133 case SIGPIPE:
4134 DBG("SIGPIPE caught");
4135 return;
4136 case SIGINT:
4137 DBG("SIGINT caught");
4138 stop_threads();
4139 break;
4140 case SIGTERM:
4141 DBG("SIGTERM caught");
4142 stop_threads();
4143 break;
4144 default:
4145 break;
4146 }
4147 }
4148
4149 /*
4150 * Setup signal handler for :
4151 * SIGINT, SIGTERM, SIGPIPE
4152 */
4153 static int set_signal_handler(void)
4154 {
4155 int ret = 0;
4156 struct sigaction sa;
4157 sigset_t sigset;
4158
4159 if ((ret = sigemptyset(&sigset)) < 0) {
4160 PERROR("sigemptyset");
4161 return ret;
4162 }
4163
4164 sa.sa_handler = sighandler;
4165 sa.sa_mask = sigset;
4166 sa.sa_flags = 0;
4167 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4168 PERROR("sigaction");
4169 return ret;
4170 }
4171
4172 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4173 PERROR("sigaction");
4174 return ret;
4175 }
4176
4177 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4178 PERROR("sigaction");
4179 return ret;
4180 }
4181
4182 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4183
4184 return ret;
4185 }
4186
4187 /*
4188 * Set open files limit to unlimited. This daemon can open a large number of
4189 * file descriptors in order to consumer multiple kernel traces.
4190 */
4191 static void set_ulimit(void)
4192 {
4193 int ret;
4194 struct rlimit lim;
4195
4196 /* The kernel does not allowed an infinite limit for open files */
4197 lim.rlim_cur = 65535;
4198 lim.rlim_max = 65535;
4199
4200 ret = setrlimit(RLIMIT_NOFILE, &lim);
4201 if (ret < 0) {
4202 PERROR("failed to set open files limit");
4203 }
4204 }
4205
4206 /*
4207 * Write pidfile using the rundir and opt_pidfile.
4208 */
4209 static void write_pidfile(void)
4210 {
4211 int ret;
4212 char pidfile_path[PATH_MAX];
4213
4214 assert(rundir);
4215
4216 if (opt_pidfile) {
4217 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4218 } else {
4219 /* Build pidfile path from rundir and opt_pidfile. */
4220 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4221 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4222 if (ret < 0) {
4223 PERROR("snprintf pidfile path");
4224 goto error;
4225 }
4226 }
4227
4228 /*
4229 * Create pid file in rundir. Return value is of no importance. The
4230 * execution will continue even though we are not able to write the file.
4231 */
4232 (void) utils_create_pid_file(getpid(), pidfile_path);
4233
4234 error:
4235 return;
4236 }
4237
4238 /*
4239 * main
4240 */
4241 int main(int argc, char **argv)
4242 {
4243 int ret = 0;
4244 void *status;
4245 const char *home_path, *env_app_timeout;
4246
4247 init_kernel_workarounds();
4248
4249 rcu_register_thread();
4250
4251 setup_consumerd_path();
4252
4253 page_size = sysconf(_SC_PAGESIZE);
4254 if (page_size < 0) {
4255 PERROR("sysconf _SC_PAGESIZE");
4256 page_size = LONG_MAX;
4257 WARN("Fallback page size to %ld", page_size);
4258 }
4259
4260 /* Parse arguments */
4261 progname = argv[0];
4262 if ((ret = parse_args(argc, argv)) < 0) {
4263 goto error;
4264 }
4265
4266 /* Daemonize */
4267 if (opt_daemon) {
4268 int i;
4269
4270 /*
4271 * fork
4272 * child: setsid, close FD 0, 1, 2, chdir /
4273 * parent: exit (if fork is successful)
4274 */
4275 ret = daemon(0, 0);
4276 if (ret < 0) {
4277 PERROR("daemon");
4278 goto error;
4279 }
4280 /*
4281 * We are in the child. Make sure all other file
4282 * descriptors are closed, in case we are called with
4283 * more opened file descriptors than the standard ones.
4284 */
4285 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4286 (void) close(i);
4287 }
4288 }
4289
4290 /* Create thread quit pipe */
4291 if ((ret = init_thread_quit_pipe()) < 0) {
4292 goto error;
4293 }
4294
4295 /* Check if daemon is UID = 0 */
4296 is_root = !getuid();
4297
4298 if (is_root) {
4299 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4300
4301 /* Create global run dir with root access */
4302 ret = create_lttng_rundir(rundir);
4303 if (ret < 0) {
4304 goto error;
4305 }
4306
4307 if (strlen(apps_unix_sock_path) == 0) {
4308 snprintf(apps_unix_sock_path, PATH_MAX,
4309 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4310 }
4311
4312 if (strlen(client_unix_sock_path) == 0) {
4313 snprintf(client_unix_sock_path, PATH_MAX,
4314 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4315 }
4316
4317 /* Set global SHM for ust */
4318 if (strlen(wait_shm_path) == 0) {
4319 snprintf(wait_shm_path, PATH_MAX,
4320 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4321 }
4322
4323 if (strlen(health_unix_sock_path) == 0) {
4324 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4325 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4326 }
4327
4328 /* Setup kernel consumerd path */
4329 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4330 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4331 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4332 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4333
4334 DBG2("Kernel consumer err path: %s",
4335 kconsumer_data.err_unix_sock_path);
4336 DBG2("Kernel consumer cmd path: %s",
4337 kconsumer_data.cmd_unix_sock_path);
4338 } else {
4339 home_path = utils_get_home_dir();
4340 if (home_path == NULL) {
4341 /* TODO: Add --socket PATH option */
4342 ERR("Can't get HOME directory for sockets creation.");
4343 ret = -EPERM;
4344 goto error;
4345 }
4346
4347 /*
4348 * Create rundir from home path. This will create something like
4349 * $HOME/.lttng
4350 */
4351 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4352 if (ret < 0) {
4353 ret = -ENOMEM;
4354 goto error;
4355 }
4356
4357 ret = create_lttng_rundir(rundir);
4358 if (ret < 0) {
4359 goto error;
4360 }
4361
4362 if (strlen(apps_unix_sock_path) == 0) {
4363 snprintf(apps_unix_sock_path, PATH_MAX,
4364 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4365 }
4366
4367 /* Set the cli tool unix socket path */
4368 if (strlen(client_unix_sock_path) == 0) {
4369 snprintf(client_unix_sock_path, PATH_MAX,
4370 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4371 }
4372
4373 /* Set global SHM for ust */
4374 if (strlen(wait_shm_path) == 0) {
4375 snprintf(wait_shm_path, PATH_MAX,
4376 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4377 }
4378
4379 /* Set health check Unix path */
4380 if (strlen(health_unix_sock_path) == 0) {
4381 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4382 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4383 }
4384 }
4385
4386 /* Set consumer initial state */
4387 kernel_consumerd_state = CONSUMER_STOPPED;
4388 ust_consumerd_state = CONSUMER_STOPPED;
4389
4390 DBG("Client socket path %s", client_unix_sock_path);
4391 DBG("Application socket path %s", apps_unix_sock_path);
4392 DBG("Application wait path %s", wait_shm_path);
4393 DBG("LTTng run directory path: %s", rundir);
4394
4395 /* 32 bits consumerd path setup */
4396 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4397 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4398 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4399 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4400
4401 DBG2("UST consumer 32 bits err path: %s",
4402 ustconsumer32_data.err_unix_sock_path);
4403 DBG2("UST consumer 32 bits cmd path: %s",
4404 ustconsumer32_data.cmd_unix_sock_path);
4405
4406 /* 64 bits consumerd path setup */
4407 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4408 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4409 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4410 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4411
4412 DBG2("UST consumer 64 bits err path: %s",
4413 ustconsumer64_data.err_unix_sock_path);
4414 DBG2("UST consumer 64 bits cmd path: %s",
4415 ustconsumer64_data.cmd_unix_sock_path);
4416
4417 /*
4418 * See if daemon already exist.
4419 */
4420 if ((ret = check_existing_daemon()) < 0) {
4421 ERR("Already running daemon.\n");
4422 /*
4423 * We do not goto exit because we must not cleanup()
4424 * because a daemon is already running.
4425 */
4426 goto error;
4427 }
4428
4429 /*
4430 * Init UST app hash table. Alloc hash table before this point since
4431 * cleanup() can get called after that point.
4432 */
4433 ust_app_ht_alloc();
4434
4435 /* After this point, we can safely call cleanup() with "goto exit" */
4436
4437 /*
4438 * These actions must be executed as root. We do that *after* setting up
4439 * the sockets path because we MUST make the check for another daemon using
4440 * those paths *before* trying to set the kernel consumer sockets and init
4441 * kernel tracer.
4442 */
4443 if (is_root) {
4444 ret = set_consumer_sockets(&kconsumer_data, rundir);
4445 if (ret < 0) {
4446 goto exit;
4447 }
4448
4449 /* Setup kernel tracer */
4450 if (!opt_no_kernel) {
4451 init_kernel_tracer();
4452 }
4453
4454 /* Set ulimit for open files */
4455 set_ulimit();
4456 }
4457 /* init lttng_fd tracking must be done after set_ulimit. */
4458 lttng_fd_init();
4459
4460 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4461 if (ret < 0) {
4462 goto exit;
4463 }
4464
4465 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4466 if (ret < 0) {
4467 goto exit;
4468 }
4469
4470 if ((ret = set_signal_handler()) < 0) {
4471 goto exit;
4472 }
4473
4474 /* Setup the needed unix socket */
4475 if ((ret = init_daemon_socket()) < 0) {
4476 goto exit;
4477 }
4478
4479 /* Set credentials to socket */
4480 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4481 goto exit;
4482 }
4483
4484 /* Get parent pid if -S, --sig-parent is specified. */
4485 if (opt_sig_parent) {
4486 ppid = getppid();
4487 }
4488
4489 /* Setup the kernel pipe for waking up the kernel thread */
4490 if (is_root && !opt_no_kernel) {
4491 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4492 goto exit;
4493 }
4494 }
4495
4496 /* Setup the thread ht_cleanup communication pipe. */
4497 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4498 goto exit;
4499 }
4500
4501 /* Setup the thread apps communication pipe. */
4502 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4503 goto exit;
4504 }
4505
4506 /* Setup the thread apps notify communication pipe. */
4507 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4508 goto exit;
4509 }
4510
4511 /* Initialize global buffer per UID and PID registry. */
4512 buffer_reg_init_uid_registry();
4513 buffer_reg_init_pid_registry();
4514
4515 /* Init UST command queue. */
4516 cds_wfq_init(&ust_cmd_queue.queue);
4517
4518 /*
4519 * Get session list pointer. This pointer MUST NOT be free(). This list is
4520 * statically declared in session.c
4521 */
4522 session_list_ptr = session_get_list();
4523
4524 /* Set up max poll set size */
4525 lttng_poll_set_max_size();
4526
4527 cmd_init();
4528
4529 /* Check for the application socket timeout env variable. */
4530 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4531 if (env_app_timeout) {
4532 app_socket_timeout = atoi(env_app_timeout);
4533 } else {
4534 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4535 }
4536
4537 write_pidfile();
4538
4539 /* This is to get the TCP timeout value. */
4540 lttcomm_inet_init();
4541
4542 /*
4543 * Initialize the health check subsystem. This call should set the
4544 * appropriate time values.
4545 */
4546 health_init();
4547
4548 /* Create thread to manage the client socket */
4549 ret = pthread_create(&ht_cleanup_thread, NULL,
4550 thread_ht_cleanup, (void *) NULL);
4551 if (ret != 0) {
4552 PERROR("pthread_create ht_cleanup");
4553 goto exit_ht_cleanup;
4554 }
4555
4556 /* Create thread to manage the client socket */
4557 ret = pthread_create(&health_thread, NULL,
4558 thread_manage_health, (void *) NULL);
4559 if (ret != 0) {
4560 PERROR("pthread_create health");
4561 goto exit_health;
4562 }
4563
4564 /* Create thread to manage the client socket */
4565 ret = pthread_create(&client_thread, NULL,
4566 thread_manage_clients, (void *) NULL);
4567 if (ret != 0) {
4568 PERROR("pthread_create clients");
4569 goto exit_client;
4570 }
4571
4572 /* Create thread to dispatch registration */
4573 ret = pthread_create(&dispatch_thread, NULL,
4574 thread_dispatch_ust_registration, (void *) NULL);
4575 if (ret != 0) {
4576 PERROR("pthread_create dispatch");
4577 goto exit_dispatch;
4578 }
4579
4580 /* Create thread to manage application registration. */
4581 ret = pthread_create(&reg_apps_thread, NULL,
4582 thread_registration_apps, (void *) NULL);
4583 if (ret != 0) {
4584 PERROR("pthread_create registration");
4585 goto exit_reg_apps;
4586 }
4587
4588 /* Create thread to manage application socket */
4589 ret = pthread_create(&apps_thread, NULL,
4590 thread_manage_apps, (void *) NULL);
4591 if (ret != 0) {
4592 PERROR("pthread_create apps");
4593 goto exit_apps;
4594 }
4595
4596 /* Create thread to manage application notify socket */
4597 ret = pthread_create(&apps_notify_thread, NULL,
4598 ust_thread_manage_notify, (void *) NULL);
4599 if (ret != 0) {
4600 PERROR("pthread_create apps");
4601 goto exit_apps;
4602 }
4603
4604 /* Don't start this thread if kernel tracing is not requested nor root */
4605 if (is_root && !opt_no_kernel) {
4606 /* Create kernel thread to manage kernel event */
4607 ret = pthread_create(&kernel_thread, NULL,
4608 thread_manage_kernel, (void *) NULL);
4609 if (ret != 0) {
4610 PERROR("pthread_create kernel");
4611 goto exit_kernel;
4612 }
4613
4614 ret = pthread_join(kernel_thread, &status);
4615 if (ret != 0) {
4616 PERROR("pthread_join");
4617 goto error; /* join error, exit without cleanup */
4618 }
4619 }
4620
4621 exit_kernel:
4622 ret = pthread_join(apps_thread, &status);
4623 if (ret != 0) {
4624 PERROR("pthread_join");
4625 goto error; /* join error, exit without cleanup */
4626 }
4627
4628 exit_apps:
4629 ret = pthread_join(reg_apps_thread, &status);
4630 if (ret != 0) {
4631 PERROR("pthread_join");
4632 goto error; /* join error, exit without cleanup */
4633 }
4634
4635 exit_reg_apps:
4636 ret = pthread_join(dispatch_thread, &status);
4637 if (ret != 0) {
4638 PERROR("pthread_join");
4639 goto error; /* join error, exit without cleanup */
4640 }
4641
4642 exit_dispatch:
4643 ret = pthread_join(client_thread, &status);
4644 if (ret != 0) {
4645 PERROR("pthread_join");
4646 goto error; /* join error, exit without cleanup */
4647 }
4648
4649 ret = join_consumer_thread(&kconsumer_data);
4650 if (ret != 0) {
4651 PERROR("join_consumer");
4652 goto error; /* join error, exit without cleanup */
4653 }
4654
4655 ret = join_consumer_thread(&ustconsumer32_data);
4656 if (ret != 0) {
4657 PERROR("join_consumer ust32");
4658 goto error; /* join error, exit without cleanup */
4659 }
4660
4661 ret = join_consumer_thread(&ustconsumer64_data);
4662 if (ret != 0) {
4663 PERROR("join_consumer ust64");
4664 goto error; /* join error, exit without cleanup */
4665 }
4666
4667 exit_client:
4668 ret = pthread_join(health_thread, &status);
4669 if (ret != 0) {
4670 PERROR("pthread_join health thread");
4671 goto error; /* join error, exit without cleanup */
4672 }
4673
4674 exit_health:
4675 ret = pthread_join(ht_cleanup_thread, &status);
4676 if (ret != 0) {
4677 PERROR("pthread_join ht cleanup thread");
4678 goto error; /* join error, exit without cleanup */
4679 }
4680 exit_ht_cleanup:
4681 exit:
4682 /*
4683 * cleanup() is called when no other thread is running.
4684 */
4685 rcu_thread_online();
4686 cleanup();
4687 rcu_thread_offline();
4688 rcu_unregister_thread();
4689 if (!ret) {
4690 exit(EXIT_SUCCESS);
4691 }
4692 error:
4693 exit(EXIT_FAILURE);
4694 }
This page took 0.125482 seconds and 4 git commands to generate.